Академический Документы
Профессиональный Документы
Культура Документы
Volume 126
Series editors
Wolfgang Schröder, Lehrstuhl für Strömungslehre und Aerodynamisches Institut,
Aachen, Germany
e-mail: office@aia.rwth-aachen.de
Bendiks Jan Boersma, Delft University of Technology, CA Delft, The Netherlands
e-mail: b.j.boersma@tudelft.nl
Kozo Fujii, The Institute of Space and Astronautical Science, Kanagawa, Japan
e-mail: fujii@flab.eng.isas.jaxa.jp
Werner Haase, Imperial College of Science Technology and Medicine,
Hohenbrunn, Germany
e-mail: whac@haa.se
Ernst Heinrich Hirschel, Zorneding, Germany
e-mail: e.h.hirschel@t-online.de
Michael A. Leschziner, Imperial College of Science Technology and Medicine,
London, UK
e-mail: mike.leschziner@imperial.ac.uk
Jacques Periaux, Paris, France
e-mail: jperiaux@free.fr
Sergio Pirozzoli, Università di Roma “La Sapienza”, Roma, Italy
e-mail: sergio.pirozzoli@uniroma1.it
Arthur Rizzi, KTH Royal Institute of Technology, Stockholm, Sweden
e-mail: rizzi@aero.kth.se
Bernard Roux, Technopole de Chateau-Gombert, Marseille Cedex, France
e-mail: broux@l3m.univ-mrs.fr
Yurii I. Shokin, Siberian Branch of the Russian Academy of Sciences,
Novosibirsk, Russia
e-mail: shokin@ict.nsc.ru
About this Series
ABC
Editors
Jens C.O. Nielsen David Thompson
Department of Applied University of Southampton Inst. Sound and
Mechanics/CHARMEC Vibration
Chalmers University of Technology Southampton
Göteborg, Sweden United Kingdom
James T. Nelson
Wilson, Ihrig and Associates
Emeryville, USA
This volume contains the peer reviewed contributions to the 11th International
Workshop on Railway Noise (IWRN11), which took place in Uddevalla, Sweden, on
September 9–13, 2013. The workshop was organised by the Competence Centre in Rail-
way Mechanics (CHARMEC) and the Departments of Applied Mechanics and Applied
Acoustics at Chalmers University of Technology in Gothenburg, Sweden. It was sup-
ported by Bombardier Transportation, voestalpine Schienen, Lucchini and Chalmers /
CHARMEC.
The workshop was attended by 160 delegates from 19 countries around the world:
Sweden (33 delegates), Germany (18), United Kingdom (17), France (12), The Nether-
lands (11), China (8), Australia (7), Austria (7), Belgium (7), Czech Republic (7), Den-
mark (7), Switzerland (6), Japan (5), United States (4), Norway (3), South Korea (3),
Spain (3), Finland (1) and Hong Kong (1).
Railway traffic is, in comparison with other modes of transportation, safe and envi-
ronmentally friendly and is generally described as the most sustainable mode for re-
gional and international transports. According to the White Paper on Transport, issued
by the European Commission in 2011, one of the key goals by 2050 is a 50 % shift of
medium distance intercity passenger and freight journeys from road to rail and water-
borne transport. This will contribute to a 60 % reduction in carbon emissions by the
middle of the century. To promote the shift from road to rail, the environmental impact
induced by the railway in terms of noise and vibration needs to be further reduced.
Since the first IWRN in 1976, held in Derby (UK) with some 35 delegates, the work-
shop series has been established as a regular event that every three years brings together
the leading researchers and engineers in all fields related to railway noise and vibration.
The workshops have to a great extent contributed to the understanding and solution of
many problems in railway noise and vibration, building a scientific foundation for re-
ducing the environmental impact by air-borne, ground-borne and structure-borne noise
and vibration.
Following the tradition from previous workshops, the scientific programme of
IWRN11 was held as a single-session event (no parallel sessions) over three and a half
days. The programme contained 55 oral presentations and 36 poster presentations, the
latter including a three-minute oral presentation to introduce each poster. The present
VI Preface
volume contains the peer reviewed papers from 84 of these presentations, including 2
state-of-the papers on ground-borne vibration due to railway traffic and on railway noise
generated by high-speed trains. IWRN11 covered 9 different topics of railway noise
and vibration: 1. Prospects, legal regulation and perception, 2. Wheel and rail noise, 3.
Prediction, measurements and monitoring, 4. Ground-borne vibration, 5. Squeal noise
and structure-borne noise, 6. Aerodynamic noise generated by high-speed trains, 7. Re-
silient track forms, 8. Grinding, corrugation and roughness, and 9. Interior noise and
sound barriers.
There is no formal organisation behind the IWRN but rather an informal, commit-
ted International Committee. It supports the chairman during the preparation process
with the experience and expertise of its members. Assistance is given to formulate the
scientific programme by reviewing the submitted abstracts, to act as session chairmen,
and to act as peer review group and editors of the IWRN proceedings published in this
volume.
The International Committee is grateful to Anders Frid, Wolfgang Kropp, Roger
Lundén, Astrid Pieringer and Peter Torstensson of the local committee for their great
commitment and care in organising the workshop. Special thanks to Pernilla Appelgren
Johansson, Christian Johansson and Sara Nielsen for their work related to the admin-
istration, communication and graphic design of material for the Workshop, and to the
staff of Bohusgården Hotel & Conference Centre.
The editors of this volume are grateful to Professor Wolfgang Schröder as the general
editor of the “Notes on Numerical Fluid Mechanics and Multidisciplinary Design” and
also to the staff of the Springer Verlag (in particular Dr Leontina Di Cecco) for the
opportunity to publish the proceedings of the IWRN11 workshop in this series. Note
that previous workshop proceedings have also been published in this series (IWRN9 in
volume 99 and IWRN10 in volume 118).
We hope that this volume will be used as a “state-of-the-art” reference by scientists
and engineers involved in solving noise and vibration problems related to railway traffic.
J. Oertli
1 Introduction
the noise emitted from freight trains. This paper summarizes the main railway noise
activities in Europe in terms of policy, economics, technical possibilities as well as the
strategy of the railways to deal with noise issues. It concludes with some problematic
trends.
The author is chairman of the UIC (International Union of Railways) Network Noise
and Vibration. This article is based on the information gained through the work and
contacts of this network.
National legislation differs throughout Europe. Many countries have reception limits
for new and significantly altered lines, however only Switzerland, Italy and Norway
have limits for existing lines. Usually limit values apply to the façade of buildings but
in some cases (e.g. Norway) they apply to indoor areas. In The Netherlands, Germany
and Switzerland noise differentiated track access charges are in effect. Germany,
France and Austria spend considerable amounts on existing lines, even though noise
Railway Noise Control in Europe: Current Status 3
abatement is not stipulated by the legislation. Finally, there is a Swiss plan to ban cast
iron brake blocks by 2020.
Most European countries have national incentives and policies promoting the
implementation of retrofitting the rolling stock with silent brake blocks. The most
prominent examples are Switzerland, where all Swiss rolling stock is in the process of
being retrofitted with composite brake blocks. This programme is financed by the
government, which in turn receives the funds mostly from the road sector. Switzerland
has also introduced noise differentiated track access charges (see chapter 6). The
Netherlands are also very active in promoting retrofitting. Some of the activities
include the launching of numerous studies and pilot projects to test composite brake
blocks and the introduction of noise differentiated track access charging. Also, in
Germany, noise differentiated track access charges have come into effect. Additionally
Germany has strongly supported the development of LL-brake blocks (see chapter 5).
A summary of national initiatives and legislation is given in the 2010 UIC state of the
art report on railway noise [2].
Numerous studies (e.g. the STAIRRS project [3]) have considered the economics of
railway noise control, comparing the costs and benefits of different noise control
possibilities and combinations thereof. In general, noise barriers, especially high ones
have a poor cost-benefit ratio, while retrofitting the freight fleet has a beneficial
cost-benefit ratio. To date network wide cost benefit studies for other measures (e.g.
rail dampers, rail grinding) however are lacking.
The railways have adopted the following noise control strategy: 1) Reduce the noise of
all new freight vehicles by introducing and adopting TSI limit values. 2) Promote the
retrofitting of existing freight vehicles with composite brake blocks. 3) Build noise
barriers and install insulated windows. 4) Pursue further solutions in special cases.
Such further solutions include rail dampers, acoustic rail grinding, solutions for trains
parked in depots and stations, solutions against curve squeal, measures on steel bridges,
and to improve the noise situation in railway freight yards. The current situation in
terms of retrofitting and noise barriers is as follows:
Retrofitting: Railway rolling noise is caused by rough wheels on rough rails, significant
noise reduction can be achieved by replacing cast-iron brake blocks with composite
brake blocks. Two types of composite brake blocks are available: 1) The K-blocks are
homologated however require adapting the wheel set due to the different braking
characteristics. This makes retrofitting a fairly expensive option. 2) The LL brake block
has similar braking characteristics to the cast iron brake block. This makes retrofitting
less expensive than with K-blocks. The EuropeTrain project tested the LL brake block
on 200’000 km throughout Europe. Homologation of this brake block was approved
4 J. Oertli
mid 2013. Production of these brake blocks must be started before large scale
retrofitting can be undertaken. It must be noted that retrofitted freight wagons both with
K- and with LL-blocks incur higher life cycle costs due to a greater wheel wear. This
increase in life cycle costs has, however, not been adequately quantified to date.
Noise barriers: A large amount of noise barrier construction has taken place in the past
years. Until 2007 a UIC survey [4] indicated that at least 1000 km of noise barriers were
built along railway lines in Europe. A total of € 150 – 200 Million are spent every year
in Europe on noise barriers to mitigate railway noise.
Noise differentiated track access charges (NDTAC) have been proposed as the main
incentive for retrofitting the rolling stock by the EU Commission and several European
countries such as Switzerland, The Netherlands, and Germany. The recast of the first
railway package (Directive 2012/34/EU), adopted in November 2012, foresees an
optional introduction of noise-differentiated track access charges. To date there has
been little effect from this incentive; however the effect may increase with larger
participation. Major risks of this incentive are: 1) the wagon owners who must
undertake the investment are often different from the operators who stand to gain from
the reduced track access charges and 2) to date it is unclear if the proposed levels of
NDTAC will cover the additional life cycle costs incurred by retrofitting.
7 Problematic Trends
Although the railways have made considerable progress in railway noise reduction,
several problematic trends may be observed:
1) There is a beginning tendency by government agencies towards protecting capital
instead of people. For example Van Praag and Ferrer-i-Carbonell [5] propose noise
compensations based on income, where richer people are entitled to higher
compensation in absolute money terms. Another example is Switzerland, where
legislation is being discussed to compensate home owners based on the value of
their property. The compensations would have to be paid for by the railway
infrastructure manager and the exact modalities are still being developed. These
ideas would serve as an incentive to undertake more noise control in areas with
high incomes or high property values, rather than where noise levels are highest or
the largest number of people is affected.
2) Whole system optimizations are rare. For example noise barriers are often
preferred over retrofitting the rolling stock, even though in most cases a mixture of
both measures would provide for the best cost-benefit ratio. Also, an optimization
of all construction parameters is rare. This may lead to the introduction of
measures against vibrations, which, might lead to more noise being generated.
Another example it the introduction of soft rail pads for the purpose of
infrastructure maintenance without considering the effects on noise. Optimization
Railway Noise Control in Europe: Current Status 5
of entire systems would probably lead to lower overall costs and higher benefits for
all involved parties.
3) There are exaggerated expectations from many new technologies. Often these are
tested in unclear circumstances resulting in high effectiveness. The railways are
then under pressure from authorities to implement these technologies, even though
they do not perform well under real circumstances or may have security or
maintenance problems. Examples, where this has happened include rail dampers
and low height noise barriers close to the track.
4) The effects on the split between traffic modes are not sufficiently considered.
Since the railways are a sustainable means of transport, it is important that the costs
for noise control measures do not change the modal split of transport in favor of
other modes (in the process also increasing the noise emissions of other modes).
This risk must be considered when discussing mitigation options, since the
railways operate in a very competitive market. It is therefore in the interest of
society as a whole to finance railway noise control from outside of the railway
system.
5) Simplifications may lead to wrong conclusions. For example the effectiveness of a
certain noise mitigation method often depends on local conditions and
generalizations are not possible. Results from one community or project usually
cannot be extrapolated to entire networks.
6) Failure to use proper experimental designs: When testing new technologies, the
inherent variability of noise measurements, local conditions etc. is often not
considered, so that the tests do not have an appropriate number of replicates and no
correct statistical analyses are undertaken. Nonetheless, the results from these tests
are used to make policy decisions which may involve many millions of Euros.
8 Outlook
In sum the railways have become quieter and will continue to do so. Retrofitting freight
rolling stock with composite brake blocks is the most effective measure and efforts
must and will continue in this regard. The main incentive for retrofitting will be noise
differentiated track access charging. However, more noise barriers will be built while
additional measures such as rail dampers or acoustic rail grinding will prove useful in
hot spots. When considering noise mitigation options, it will become increasingly
important to look at entire systems, be it on the level of track construction or the traffic
policy level.
References
[1] EU Commission: Effective reduction of noise generated by rail freight wagons in the
European Union 2013, Roadmap of DG Move (2013)
[2] Oertli, J., Hübner, P.: Railway Noise in Europe, A 2010 report on the state of the art. UIC,
International Union of Railways (2010)
6 J. Oertli
[3] Oertli, J.: The STAIRRS project, work package 1: a cost-effectiveness analysis of railway
noise reduction on a European scale. Journal of Sound and Vibration 267, 431–437 (2003)
[4] Hübner, P.: Noise Reduction in European Railway Infrastructure, Status Report 2007. UIC,
International Union of Railways (2007)
[5] Van Praag, B.M.S., Ferrer-i-Carbonell, A.: Happiness Quantified, A Satisfaction Calculus
Approach, Revised Edition. Oxford University Press (2007)
Novel Legislation for Railway Lines and Motorways
in The Netherlands
P.H. de Vos
1 Introduction
On the occasion of the TSI-Noise for conventional speed and the TSI-High Speed
coming into force, an attempt has been made to limit the noise production of railway
vehicles, at least for vehicles entering on the market and operating in international
traffic. In due time, the existing freight fleet will be either retrofitted or phased out,
either through effective incentives or through the inclusion of limits for existing
wagons into the TSI-Noise. Thus, substantial progress has been made since these
options were first suggested, e.g. in the 6th IWRN in Voss, Norway. It is however not
only the noisy part of the fleet, but also the increasing traffic intensity on the main
railway lines that causes public concern and reactions. Since 1987, the national policy
in The Netherlands and several other countries has been to impose a standstill on the
noise exposure from existing railway lines. In fact, the text of the original Dutch
regulation suggests, that any change of the railway line causing a significant increase of
the noise exposure would require the infrastructure manager to take mitigation
measures ensuring this stand still (at least in as far as the exposure levels would exceed
the preferred limit values). The shortcoming was that a possible increase of noise
exposure levels would only be identified on the occasion of a spatial planning
procedure, i.e. in the course of a planned extension of the railway under concern. The
mere increase of traffic on the line, i.e. without planning procedure, would indeed result
in an increase of the noise exposure, but the responsibility to mitigate the effect of this
increase was never assigned and moreover the increase would remain unobserved. In
the decade to follow this legislation coming into force there was a significant increase
of rail traffic, both freight and passenger, on many lines. Residents would apply for
noise protection, even in court, but their applications were never successful. This
affected the credibility of the state as a protector of its residents’ rights and well being.
It also caused severe opposition against railway transport in general from residents and
their political representatives.
The situation as described above is more or less identical in several other European
countries with strict noise legislation. These countries would encounter the same
credibility problem and might be tempted to solve it in the same way as The
Netherlands intended.
18 years later, a proposed revision of the Noise Control Act was adopted in Dutch
Parliament, implementing Noise Production Ceilings for main motorways (i.e. roads
managed by the state road authority) and main railways (i.e. managed by the national
infrastructure manager ProRail). The current paper introduces the new legislation and
illustrates the possible benefits and drawbacks of the ceilings.
The ceiling is derived from the calculated noise level at a large number of virtual
reference points along the track, representing the traffic and track situation as it was
assessed during the three years preceding the year when the ceiling was set. A margin
of 1.5 dB is added to this calculated noise level in order to arrive at the ceiling value.
Once the ceiling has been officially set, it is the responsibility of the rail infrastructure
manager that it is never exceeded.
Novel Legislation for Railway Lines and Motorways in The Netherlands 9
Noiseproduction
Noise production measure
warning measure
warning
Ceiling
Ceiling
Current
Current
Mitigation Realnoise
Real noise
Mitigation
measure production
production
measure
Introduction
Introduction time
time
ofceilings
of ceilings
When the new legislation was proposed, the main issues raised were the ceiling level at
introduction of the system and the margin necessary to allow a reasonable traffic
growth without mitigation. A comprehensive impact assessment study was carried out,
in assignment of the former Dutch Ministry of Housing, Spatial Planning and
Environment, to assess consequences of various scenarios, both for road and rail [2]. In
case of an expected future excess of the ceiling value, the system should allow
sufficient time for planning and preparation of mitigation measures that would avoid
that excess. This case was simulated using various economic growth scenarios, with the
corresponding transport developments for road and rail. For freight it was assumed that
80% of the fleet would have been retrofitted in the study’s reference year 2020. In
addition, the number of sites exceeding the ceilings should be limited such that the
system would be manageable for the infrastructure manager, both in terms of work load
and in terms of costs. At the same time, the effects on the overall noise exposure of
residents along main roads and tracks would have to be limited and certainly better than
under the current legislation.
The study concluded that the yearly fluctuations in traffic intensity are far more
significant for rail than for road, which led to the conclusion that the starting point for
the ceiling for railway noise would have to be based on a three years average. A range
of scenarios was used for different values of the margin, viz. 0.5 dB, 1.0 dB, 1.5 dB,
2.0 dB and 2.5 dB. It was found, that the 1.5 dB represents that optimum both in terms
of economic and managerial criteria and in terms of annoyance and health effects. The
overall, quantitative conclusion is summarized in the following Table 1.
Table 1. Overall results of the impact assessment study comparing four different options for the
margin to be included in the ceiling values
Working margin
0.5 dB 1.0 dB 1.5 dB 2.0 dB
Feasibility
- - - -
Implementation
Feasibility
- - + +
Amount
Cost - - + +
Annoyance/Health + + + +
A grid of approximately 60,000 virtual reference points, each 100 m apart and 50 m
from either side of the track, at 4 meters above local ground, was defined around the
3,000 km of railway track in The Netherlands. In each point, the calculated noise level
in dB Lden was averaged over the three years 2006, 2007 and 2008. The resulting value
is increased by the working margin of 1.5 dB, allowing some 40% growth of traffic. All
the reference points and the corresponding ceiling values are kept in a public register,
published under the responsibility of the Minister of Infrastructure and Environment.
Novel Legislation for Railway Lines and Motorways in The Netherlands 11
The infrastructure manager must report, on an annual basis, the compliance of the
actual calculated noise level at each reference point with the ceiling levels. If an excess
of the ceiling is expected, the infrastructure manager must take appropriate mitigation
measures with a noise reducing effect which he can then include into the calculated
level. No legal procedure is required for such measures, unless the selected measure
represents a noise barrier. In that case a formal revision of the ceiling must be applied
for (the Minister of Transport is responsible to allow such a revision). In the case of
such a revision and where noise sensitive dwellings are present, the noise exposure at
the façade(s) has to be assessed and compared to legal limits.
What if the noise exposure level already exceeds the maximum legal exposure limit?
Then the infrastructure manager will have to assess this exposure level, and offer noise
mitigation measures resulting in a reduction of this exposure. Once these measures
have been implemented – a substantial state budget is foreseen to achieve that
throughout the whole country by 2022 – the ceilings will have to be reduced thus
reflecting the situation after implementation.
Ceilings may be revised, i.e. increased or decreased, but any such change will
require the Ministerial formal approval. Municipalities may request a decrease of the
ceiling enabling them to realize urban development along the track within the legal
limits. Such a decrease of the ceiling may be viable by the city erecting a noise barrier
along the track, at its own cost, or else paying for alternative mitigation. An increase of
the ceiling could be requested by the infrastructure manager, in order to avoid costly
measures, for instance at locations without residents.
Finally, an overall reduction of all ceilings could be requested by the Minister of
Environment, for instance the case of a technical improvement of the rolling stock that
would reduce its noise production.
The new system has some clear advantages, compared to the previous legislation:
• Residents are better protected against increasing noise exposure due to traffic
growth,
• There is more transparency for the residents,
• If someone intends to develop a plan for sensitive buildings along the track,
there would be no more discussions about the future noise exposure, as this
would be strictly defined by the situation where the ceiling is entirely used,
• The usual discussions with residents on the validity of traffic predictions can
be avoided, thus reducing the risk of legal procedures,
• Some track modifications or track renewals can be carried out much faster,
because there is no longer a need for a noise impact assessment as long as the
effect remains within the ceiling
• The new legislation is likely to enhance a wider application of rail absorbers
and has most probably contributed to set off the retrofitting of freight wagons
on Dutch track.
12 P.H. de Vos
References
[1] Decree of 4 April 2012, relating to revision of the Decree on Noise Annoyance and some
other decrees in relation to the implementation of Noise Production Ceilings. State Journal
2012 no. 164 (2012)
[2] van der Stap, P., Bos, S., de Vos, P.H.: Optimal working margin for railways when
implementing noise production ceilings, DHV report C1942.01.001, in assignment of the
Ministry of Housing, Spatial Planning and Environment (October 27, 2009)
Bearable Railway Noise Limits in Europe
Summary. The question ‘What are bearable limits for environmental railway
noise?’ is discussed regularly in different forums on a national scale and on a
European level. A systematic evaluation of all aspects in what ‘bearable’ could
consist of was always missing. The UIC research Project ‘Bearable limits and
emission ceilings’ [1, 2] has brought UIC in the position to propose for the first
time a well-balanced limit for noise reception. This noise reception limit is a
trade-off between the disturbing impact of noise for line side residents and
realistic possibilities for viable railways. Findings are based on an extensive
study that was commissioned by the UIC and carried out by dBvision in the
Netherlands.
A bearable value of noise reception limits for the night (Lnight) is not lower
than around 55 dB. More stringent limit values are not effective because:
- For values above 55 dB railway noise is the dominant source for sleep
disturbed persons in urban areas near railway lines. For values lower than
55 dB, it is more effective to spend money on measures for road traffic noise.
This will generally result in more reduction of the overall sleep disturbance.
- Below 50 dB, results show a large increase of cost. Noise limits up to 55 dB
are cost-effective.
Results are based on a 202 km railway line sample Rotterdam – Venlo and
extrapolation to the ERTMS corridors. These ERTMS corridors are defined in
the European Rail Infrastructure Masterplan as the main freight corridors (see
Fig. 1).
1 Developments
Different developments put pressure on noise limits. The four most important
developments are:
Fig. 1. The UIC ERIM network of international rail corridors. This network is mainly for freight
on which a European Rail Infrastructure Masterplan could be built (UIC Atlas 2008 of
Infrastructure in the ERIM Network).
The next two paragraphs describe items number 1 and 2 in more detail.
Despite its environmentally friendly image, rail transport encounters substantial public
opposition to noise in some European regions. The Commission believes that “if no
remedial action is taken, this could lead to restrictions in rail freight traffic along the
most important European rail corridors. A possible modal shift from rail to road on
these corridors would lead to increasing environmental impacts.” [1].
Bearable Railway Noise Limits in Europe 15
Retrofitting 370 000 freight wagons is the main objective to avoid this scenario. This
objective should be achieved by a combination of three policy instruments [3]:
1. Noise-differentiated track access charges (NDTAC);
2. Noise emission ceilings;
3. Voluntary commitments (railway undertakings could pass NDTAC benefits to
wagon owners, rail sector could start individual retrofitting programmes).
The noise emission ceilings are proposed as a second step to prevent an increase of
noise, after the initial retrofitting programmes have been completed.
In 2009 the WHO published a report, Night Noise Guidelines for Europe [4], was
published in 2009 to serve as guidance for action plans under the Environmental Noise
Directive. WHO proposes to adopt the Night Noise Guideline (NNG, 40 dB) as a limit
for new projects (road/rail/residential areas), while the Interim Target (IT, 55 dB) can
be used for existing cases. The Interim Target, however, is not based on health
considerations but on feasibility. Therefore, the NNG should be used as long-term goal.
The WHO realizes that implementing the recommended noise targets takes time and
money:
- Governments should adopt the health guidelines for community noise as targets to
be achieved in the long-term.
- Cost-effectiveness and cost-benefit analyses should be considered as potential
instruments when making management decisions.
2 Findings
In this study the focus has been on a 202 km freight railway line. The results are
extrapolated to the most relevant part of the UIC European Railway Infrastructure
Masterplan (ERIM) network. The ERIM project focuses on a high-level infrastructure
of six major international rail (ERMTS) corridors within and between 32 countries.
These corridors are mainly used for freight traffic. The ERIM network has a route
length of 50 000 km. These ERTMS corridors have a route length of 15 000 km. For
this extrapolation corrections are made for cost per km based on different traffic
volumes, sleep disturbed persons based on different traffic volumes, route length and
average density of buildings per country.
This study proposes that limits for noise reception should be combined with a noise
emission monitoring system (see Fig. 2). This monitoring system follows the noise
emission on a periodical (i.e. yearly) basis and compares it to a pre-set ceiling level.
The system stimulates the railways to take measures at the source (retrofitting) to avoid
that freight traffic noise exceeding the ceiling.
16 F.B.J. Elbers and E. Verheijen
Fig. 2. System with noise emission ceiling and noise reception limits. CBC method: cost-benefit
criterion.
A bearable value of noise reception limits for the night (Lnight) is not lower than
around 55 dB. More stringent limit values are not effective because:
- For lower values than 55 dB, it is more effective to spend money on measures for
road traffic noise. This is because if railway noise is lower than 55 dB, it is
generally not the dominant source of sleep disturbance in urban areas.
- Below 50 dB, results show a large increase of cost and a small increase of benefits
(see Fig. 3).
Fig. 3. The impact on costs and benefits (reduction of sleep disturbed persons) of different Lnight
limit values for 15 000 km main freight corridors
Self-reported annoyance studies show that 10% of the persons are sleep disturbed if
exposed to Lnight values of 55 dB for rail [7]. In a European perspective a 55 dB limit on
the six main freight corridors with a total length of 15 000 km would cost € 10.8
billion1. These freight corridors transport 43 % of the total European freight. The € 10.8
billion costs are for noise barriers up to a height of 6 m and for rail dampers. A
reduction to € 8.6 billion is possible if noise measures are mainly placed in densely
populated areas, where they are more cost-effective. This can be achieved by a decision
support method called a CBC method (cost-benefit criterion).
1
1 000 000 000 = a billion (one thousand million)
Bearable Railway Noise Limits in Europe 17
More stringent Lnight noise reception limits will significantly increase costs. The
additional benefit of a 5 dB more stringent noise limit becomes less and less, while
additional costs increase (see Fig. 3). Therefore more stringent noise reception limits
become less efficient.
With this CBC method it is possible to avoid noise measures in situations where
costs are unacceptably high, relative to the number of dwellings that benefit. With the
CBC method the focus on additional noise measures is in urban areas. Fig. 3 shows the
impact of variations of the limit values.
Sleep disturbance is generally dominated by road traffic in situations where the Lnight
for railway noise is 55 dB or less. Therefore, a reduction of railway noise is only
effective until a certain limit. In order to benefit from low noise limits for railway noise,
additional measures against urban road traffic noise should be taken first. Without
measures for urban road traffic noise, a further reduction of railway traffic noise does
not contribute to a reduction of sleep disturbed persons.
Several countries have different noise legislation limits for road and rail traffic noise
[8]. If one would include the dose-effect responses of rail noise and road noise, an
advantage for rail is found of 8-12 dB for sleep disturbance (Lnight) and 6-8 dB for
annoyance (Lden), see Fig. 4. This difference is sometimes called ‘noise annoyance rail
correction factor’. If this health-related correction factor is applied, the costs of
infrastructure noise measures would be reduced by 14 – 27 % for the 40 dB limit value
and around 89% for the 70 dB limit value.
Fig. 4. The correction factor (railbonus) derived from the difference between railway and road
dose response relationships
18 F.B.J. Elbers and E. Verheijen
For a 100 % retrofitted freight wagon fleet, the cost of infrastructure noise measures
would be reduced by 14 – 27 % for the 40 dB limit value and around 89 % for the 70 dB
limit value. Unlike the local effects for barriers and rail dampers, noise reduction by
retrofitting is everywhere along the railway line.
2.4 Use of Cost Benefit Criterion Gives Possible Cost Reduction of € 2.2 Billion
With almost equal effects on reduction of sleep disturbed persons, the use of a cost
benefit criterion (CBC) makes it possible to reduce costs between 7 and 37 %. Or,
alternatively, it can be calculated that with equal costs a more stringent Lnight noise limit
can be established. For example: With the same budget a limit value of about 56 dB is
possible without a cost benefit criterion and a more stringent limit value of 53 dB with a
cost benefit criterion. This reduction is obtained due to the fact that the CBC method
focuses on urban areas.
The combination of the noise annoyance rail correction factor, 100 % retrofitting
and cost benefit criterion will reduce costs for infrastructure noise measures by € 7.7
billion.
for noise reception levels and noise abatement programmes. Then it provides a better
protection against unacceptable noise exposure than legislation that solely relies on
reception limits.
Switzerland and the Netherlands have developed quite similar systems of emission
ceilings [9]. The main difference lies in the definition of the source: whether or not to
include barriers in the noise emission level (Fig. 5). The Swiss and Dutch ceilings have
in common that they are backed up by a legal framework of noise reception limits,
which existed already in these countries long before the ceilings were established. In
both countries, different ceilings apply to different railway lines.
References
[1] Elbers, F.B.J., Verheijen, E.: Exploring bearable noise limits and emission ceilings for the
railways Part II: Cost and benefit study for different noise limits (UIC Project ‘Bearable
limits and emission ceilings’), dBvision (November 26, 2011)
[2] Verheijen, E., Elbers, F.B.J.: Exploring bearable noise limits and emission ceilings for the
railways Part I: National and European legislation and analysis of different noise limit
systems (UIC Project ‘Bearable limits and emission ceilings’), dBvision (October 25, 2011)
[3] Rail noise abatement measures addressing the existing fleet, COM, 432 final (2008)
[4] Night noise guidelines for Europe, World Health Organisation (2009)
[5] Oertli, J., Hübner, P.: Railway noise in Europe – A 2010 report on the state of the art, UIC
(September 2010)
[6] Perkuszewska, A., Ebenberger, A.: European Rail Infrastructure Masterplan (ERIM) – 2007
Final Report, UIC (January 2008)
[7] Miedema, H., Passchier-Vermeer, W., Vos, H.: Elements for a position paper on night-time
transportation noise and sleep disturbance, TNO Inro (January 2003)
[8] de Vos, P.: The railway noise bonus; discussion paper on the noise annoyance correction
factor – Final report, UIC (November 2010)
[9] Verheijen, E., Elbers, F.B.J.: Future European noise emission ceilings: Threat or solution? A
review based on Swiss and Dutch ceilings. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E.,
Iida, M., Nelson, J.T., Thompson, D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and
Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 126, pp. 71–78.
Springer, Heidelberg (2015)
State-of-the-Art of the Noise Emission of Railway Cars
Summary. The current full revision of the TSI Noise includes consideration of a
second step of noise limit values for railway cars, taking into account the
developments in technology.
The state-of-the-art of noise emissions of European rail vehicles was
determined on the basis of a collection of TSI Noise data of railway cars. From
these data the state-of-the-art of the noise emission of European railway cars was
derived.
The results indicate that the most of the current limit values for stationary
noise and starting noise are significantly above the state-of-the-art noise emission
performance of European railway cars. With regard to pass-by noise, the
state-of-the-art and the current limit values are closer together. However, in the
case of freight wagons and electric multiple units, the current limit values are
approximately 5 dB above the state-of-the-art.
It would therefore be reasonable to adjust the limit values to the
state-of-the-art within the scope of the TSI Noise revision.
As a result of the revision process, the scope of the TSI is extended to a much
wider network, a continuous limit curve for freight wagons is defined in the TSI
and new requirements for stationary noise were introduced, which shall help to
handle problems with stationary vehicles. For most vehicle categories the
redefined limit values are still significantly above the state-of-the-art.
1 Introduction
The scheduled TSI Noise revision [8, 9], which started in May 2011 has to deal with
different topics as defined in the existing TSI Noise [1]. They are part of the working
program of the European Rail Agency (ERA) or are requested by members of the
working party on TSI Noise (WP NOI). The main topics are:
• Merging the requirements for high-speed and conventional rail in one document.
• Scope extension to networks not included in the Trans-European Networks (TEN),
the so called OFF-TEN.
• Reduce the verification burden by permitting more flexibility in assessment.
• Extend the requirements for stationary noise in order to solve problems with parking
vehicles, which occurred in different member states.
• Consider the introduction of new “Basic Parameters”: Braking noise and curve
squeal.
• Define a continuous limit curve for freight wagons depending on their
axles-to-length-ratio (APL).
• Set new pass-by noise and starting noise limit values.
The last point is the most important and controversial – to solve it, a sufficiently
large database on real noise emission is needed. The state-of-the-art of the noise
emission of railway cars, however, is not known as statistically reliable data about the
noise emission is not available.
Thus, the Federal Environment Agency of Germany (UBA) initiated a research
project (Ufoplan FKZ 3709 54 145) to give a representative overview of the noise
emissions of European rail vehicles measured according to TSI Noise and to determine
the state-of-the-art of noise emissions of European rail vehicles [6, 7]. The research
project was conducted by Müller-BBM. Results are presented in the next sections.
The core of the research project was to collect acoustical data of newly homologated
railway vehicles. The data collection was done at all parties involved in the approval
process such as railway operators, notified bodies or manufacturers as well as railway
State-of-the-Art of the Noise Emission of Railway Cars 23
In addition to the vehicle data details on rail roughness and the decay rate were
collected.
The response rate of the survey was very good overall. In total 378 data sets could be
collected. It can be assumed that this number is representative according to
EN ISO 11689 [8].
Prior to recording the data in a data base several checks were performed in order to
guarantee a high quality of the stored data:
1. Check of datasets with regard to duplicate files.
2. Compliance with the reference track properties of the test track.
- If compliance with the track conditions (rail roughness and decay rate)
can be assumed the data of the pass-by measurement will be deemed to be
trustworthy (assurance class I).
- Otherwise the data are considered to be comparable only to a limited
degree (assurance class II).
24 S. Lutzenberger, C. Gutmann, and U. Reichart
For assessing the pass-by measurements only data of assurance class I are
considered.
3. Furthermore, the data were checked for plausibility:
- With plausibility checks in the questionnaires it was possible to identify
contradictory or false information in the questionnaires.
- The collected data were compared with values from type tests conducted
by Müller-BBM.
In the event of any doubt as for plausibility the data were not considered
in the evaluation.
4. The remaining data were grouped by sources (notified bodies, manufacturer,
operator) and the individual results were compared to each other (mean
value and standard deviation). Thus, it could be guaranteed that no filtered
data (particularly quiet or noisy vehicles) were passed on with the
forwarding of data from the individual sources.
The collected data nearly exclusively originate from type tests according to the
TSI Noise 2006.
There is a strong scatter of the results for stationary and starting noise. Some of the
emission parameters of the railway vehicles are clearly below the limit values. The
scatter of results for pass-by noise is at a significantly lower level. Only for freight
wagons larger scatters can be observed. All pass-by levels are clearly closer to the
existing limit values than the stationary and starting noise values.
The result of the data collection is listed in the following Table 1:
Table 1. Results of the data collection for the operating conditions and railway vehicle categories
of the TSI Noise
Average
Number
TSI Standard- of the Lower Upper
Average Median of data
Category Noise limit deviation most quartile quartile
dB(A) dB(A) (overall
dB(A) dB quiet dB(A) dB(A)
378)
33 %
Table 1. (continued)
Noise emission data for machinery and equipment can be assessed and evaluated on the
basis of ISO 11689 [10]. This standard was adopted as a guideline for the analysis of
the noise emission data collected for the current study.
26 S. Lutzenberger, C. Gutmann, and U. Reichart
ISO 11689 [10] e.g. defines the low-noise control performance level L1
(high-noise emission machines) and the high-noise control performance level L2
(low-noise emission machines) as follows:
• High-noise emission machines L1 (low acoustic level):
L1 is the highest integer value achieved or exceeded by the noisiest 17.5 % of the
investigated vehicles.
• Low-noise emission machines L2 (high acoustic level):
L2 is the smallest integer value that the quietest 10 % of the investigated railway
vehicles achieve or fall below.
The state-of-the-art of noise data can be evaluated according to ISO 11689 [10], ISO
12100 [11] and [12], [13] as follows:
A machine is in compliance with the state-of-the-art when its noise emission value is
below that of 50 % – 75 % of the other machines in its group.
Two values are defined for the state-of-the-art:
• Moderate state-of-the-art: Median of noise emission data.
• Ambitious state-of-the-art: Lower quantile of noise emission data.
With this methodology noise emissions are determined that comply with the
state-of-the-art.
The following table presents the determined state-of-the-art of railway cars together
with the current limit values. As far as the data showed significant differences for
individual construction types (e.g. diesel-electric and diesel-hydraulic engines) this was
considered in the evaluation.
Table 2. Limit values according to the TSI, moderate state-of-the-art and ambitious
state-of-the-art
It is apparent that a number of the current limit values for stationary noise and
starting noise are significantly above the state-of-the-art. In these cases, the limit values
should be adjusted to the state-of-the-art within the scope of the TSI Noise revision.
1
Proposed new category
2
P ≥ 500 kW or P < 500 kW.
3
Normalised to the reference APL [14]
State-of-the-Art of the Noise Emission of Railway Cars 27
The sound emission of the vehicle aggregates can well be influenced by means of
acoustic measures in such a way that very low-noise vehicles can be achieved.
With regard to pass-by noise, the state-of-the-art and the current limit values are
closer together. However, also in this case reductions of the current limit values of
approximately 5 dB for freight wagons and electric multiple units are required in order
to adjust the limit values to the state-of-the-art.
For the determination of limit values it must be considered that measurement results
depend on various factors apart from a railway vehicle's noise emission [15]. The
measurement uncertainty is e.g. influenced by
• inherent uncertainty of sound level meters and calibrators,
• measurement methods,
• spread of noise emissions within a series of railway vehicles,
The measured values for railway vehicles thus scatter within a certain range.
Stochastic contributions are however averaged in the statistic parameters of a dataset,
such as the median. The measurement uncertainties have to be taken into account when
formulating limit values [6].
The revised TSI Noise [5] will include extended requirements on stationary noise. The
recent TSI Noise defines a limit for the average SPL of all measuring positions (LpAeq,T).
In addition to that, the revised TSI Noise will focus on single sound source (SPL at each
measured position) as well as on intermittent and impulsive sounds from air
compressors and blow-off valves. Details and limit values are still under discussion.
New limit values for starting noise were defined, which can be found in [5].
Similarly, the European Railway Agency (ERA) stated that there is no mandate from
the Commission or the Member States to include braking noise in the revised TSI
Noise. As long as no national rules or limit values exist there is no need to harmonize
for interoperability. Therefore, the revised TSI Noise will not define requirements for
braking noise or curve squeal.
5.5 Conclusion
Most of the goals of the revision were achieved: The merging of requirements on
high-speed and conventional rail, the simplification and the scope extension. New
requirements for stationary noise, which shall help to handle problems with parking
vehicles, and a continuous limit curve for freight wagons were introduced. On the other
hand, braking noise and curve squeal will not be addressed in the revised TSI Noise.
From the point of view of the Federal Environment Agency of Germany the new limit
values, esp. for pass-by noise, are not satisfactory. Bearing in mind that the vehicles,
which will be type-approved under this TSI Noise, will be in service until 2050 or even
longer, we should have done better.
References
[1] TSI-Noise: Commission Decision of 4 April 2011 concerning the technical specifications
of interoperability relating to the subsystem ‘rolling stock – noise’ of the trans-European
conventional rail system (2011/229/EU)
[2] TSI High Speed Rail Systems: Technical specification for the interoperability (TSI)
relating to the subsystem “vehicles” of the trans-European high speed rail system
(2008/232/EG) (February 21, 2008)
[3] Directive 2008/57/EC of the European Parliament and of the Council of 17 June 2008 on
the interoperability of the rail system within the Community (2008)
[4] EN ISO 3095: Railway applications –Acoustics – Measurement of noise emitted by
railbound vehicles (November 2005)
[5] TSI Noise Draft 1.0, ERA
[6] Lutzenberger, S., Gutmann, C.: UFOPLAN FKZ 3709 54 145: Ermittlung des Standes
der Technik der Geräuschemissionen europäischer Schienenfahrzeuge und deren
Lärmminderungspotenzial mit Darstellung von Best-Practice-Beispielen (2013),
http://www.umweltbundesamt.de/uba-infomedien/4441.html
[7] Lutzenberger, S., Gutmann, C.: UFOPLAN FKZ 3709 54 145: Noise emission of
European railway cars and their noise reduction potential: data collection, evaluation and
examples of best-practice railway cars, summary (2013),
http://www.umweltdaten.de/publikationen/fpdf-k/k4441.pdf
[8] Reichart, U., Weinandy, R.: TSI Noise – Große Revision, Fachtagung Bahnakustik,
Infrastruktur, Fahrzeuge, Betrieb (2011)
[9] Reichart, U., Weinandy, R., Lutzenberger, S., Gutmann, C.: Determination of the state of
the art of the noise emissions of European rail vehicles & TSI noise revision, DAGA
(2013)
State-of-the-Art of the Noise Emission of Railway Cars 29
[10] EN IS0 11689: Acoustics – Procedure for the comparison of noise-emission data for
machinery and equipment (ISO 11689: 1996)
[11] ISO 12001: Acoustics – Noise emitted by machinery and equipment – Rules for the
drafting and presentation of a noise test code (1996)
[12] Deutsche Gesetzliche Unfallversicherung DGVZ: Leise Maschinen, Auswahl und
Beschaffung, Fachausschuss-Informationsblatt (August 2010)
[13] Fritz, K., et al.: Geräuschdatenblatt für die Beschaffung von Maschinen,
Forschungsergebnisse für die Praxis, Lärmminderung – Geräuschdatenblatt,
Arbeitswissenschaftliche Erkenntnisse Nr. 85 (2. überarbeitete Auflage). Herausgeber:
Bundesanstalt für Arbeitsschutz und Ar¬beitsmedizin, Dortmund (1998) ISSN
0720-1699
[14] Meunier, N., Fodiman, P.: Revision of TSI Noi – Proposal for the new requirements to
freight wagons, version 2, on-going discussion (October 04, 2011)
[15] Lutzenberger, S., Gralla, G.: Ermittlung der Unsicherheiten bei bahnakustischen
Messungen – Erläuterungen zum Anhang E der prEN ISO 3095, Fachtagung Bahnakustik
(2011)
On Separation of Vehicle Noise for Limit Setting
in Future Legislation
Summary. This paper will propose to take a next step to find a consensus method
and separation of vehicle noise from track noise to be used for legislative
purposes and this method may in addition reduce or eliminate the need for
detailed track quality characterization in TSI certification of rolling stock.
1 Introduction
During the last decade there has been a considerable progress in the area of railway
noise characterization. Not so long ago track was characterized, even in rolling stock
contracts, as “normal good track quality“, without further details. Looking back this is
very surprising, since when rolling noise is dominating, the same vehicle running at the
same speed can give 10-15 dB difference in pass by noise levels depending on the
quality of the track.
Today we have a European legislation in the TSI NOI (Technical Specification for
the Interoperability NOIse) [1] where the quality of a defined reference track is defined
from an acoustics point of view and measured in third octave bands according to ISO
standards. Even a parameter like the lateral spatial decay rate is measured each time to
define the track quality even though there is no evidence that this parameter has any
significant influence on the resulting noise levels on the tracks of today [2]. The
specifications for track roughness are so detailed that it was considered necessary in a
limited revision of the TSI NOI during 2011 to introduce a so called “small deviation
method”, to allow a minimum of variance. The pendulum is going from one extreme to
another and hopefully we will now approach a new phase with a more optimized
solution in the middle.
2 Background
Pass-by noise limits are an essential part of the Noise TSI with the purpose of assuring
interoperability. This purpose must be compatible with maintaining the competiveness
of the railway sector and with improving the environmental protection. Improving the
3 Separation
One option to focus on the vehicle part of the rolling noise would be to reduce the track
noise by tightening the reference track definition considerably. This option leads
however to even lower accessibility to test track and increases the complexity and cost.
Time and cost for certifying rolling stock according to TSI NOI is a big issue for the
industries involved and the sector as a whole.
Another option would be to use a vehicle-track separation technique. It is important
to have a consensus around such a method. Taking into account the cost aspect, this
separation technique shall be simple so that it does not lead to a more complicated
procedure than today.
Different approaches of track-vehicle noise separation methods of different
complexity have been presented in the past. The quest for this Holy Grail is to find a
method, which is able to separate the vehicle-track noise contributions reasonably well
while maintaining a minimum of experimental effort and complexity. Essentially the
desired separation method should not be more complicated to carry out than a standard
pass-by noise test of today.
On Separation of Vehicle Noise for Limit Setting in Future Legislation 33
The STAIRRS project [6] came up with three tools that are used for characterization
and separation. Within the present investigation the PBA method was used. The PBA is
a prediction method for pass-by noise, described in detail in [4], [5]. Based on
measured total wheel/rail roughness spectra the pass-by level can be calculated by
using vibro-acoustic transfer functions. The PBA also defines measurement methods
for the estimation of the total roughness, track decay rate and total transfer functions.
The total transfer function can be split into a track transfer function and a vehicle
transfer function. This allows the separated calculation of track and vehicle noise
contributions.
The general approach is to calculate the noise from the track Lpeq,tr,i and the vehicle
Lpeq,veh,i separately:
To perform these analyses the transfer functions of the track LHp,tr,i and of the vehicle
LHp,veh,i must be known. Also the total combined effective roughness Lr,tot,i and the axle
density N/L must be known.
By measuring the pass-by level and the total roughness of a reference vehicle the
track transfer function can be estimated. Once this transfer function is known, the
vehicle transfer function of the vehicle/train under test can be calculated by measuring
the total transfer function, illustrated in Fig. 1 case b).
There is no restriction on roughness levels of the vehicle or the track, although very
poor surface conditions and unwelded rail joints must be avoided. The method is most
suited for microphone distances of 7.5 m.
a) b)
Characterizing of test track Performing Noise test
LHp,veh,i
LHp,tot,i
Lpeq,ref,i + Lpeq,tot,i
LHp,tr,i
LHp,tr,i
Fig. 1. Noise contribution a) for reference vehicle/train and b) for the "normal" train under
test, characterized by the transfer functions; indicates energetic sum
Lref
Fig. 2. Section of a typical reference vehicle EMU used to measure the track transfer function
4 Results
assumed that the measured total transfer functions are similar to the track transfer
functions of this test site.
For frequencies higher than 2000 Hz a higher uncertainty must be accepted because:
an influence of the vehicle is expected and the measurement of the rail foot
accelerations caused some uncertainties in this frequency range. [5]
95
Transfer Function amplitude [dB re 20 Pa/m½]
90
85
80
75
70
Train type 1, run 1 Train type 1, run 2
65 Train type 1, run 3 Train type 2, run 1
1250
1600
2000
2500
3150
4000
5000
6300
8000
100
125
160
200
250
315
400
500
630
800
50
63
80
Frequency [Hz]
Fig. 3. Track transfer functions measured at Test Site A, derived from three different reference
trains, three pass-bys for each train type at 80 kph
If the track transfer function is known, the vehicle transfer function of any other train
can be calculated by measuring the pass-by level according to ISO 3095 and the total
combined effective roughness by using the PBA or directly measured roughness. The
total pass-by level can be split into a vehicle and track part by using the track and
vehicle transfer functions.
In Table 1 the measured and calculated total pass-by levels of EMUs of the same
family at three different test sites are shown. Based on the track and vehicle transfer
functions the track- and vehicle part of the total pass-by level is presented. The vehicle
emitted noise is equal at both TSI compliant test sites A and B. The separation was
based on the measured track transfer function at test site A from reference vehicle type
3 (passenger coaches).
In Table 2 the results of the separation of pass-bys from different trains at test site A
are shown. The separation was based on the averaged track transfer function from the
reference vehicles 1 and 2 (EMUs).
36 T. Thron, S. Leth, and B. Stegemann
The total pass-by level can be predicted with a high accuracy at all test sites. The part
emitted by the track dominates the total pass-by level for the EMU at lower speed. As
shown in Table 2 the vehicle emitted noise dominates for EMUs with higher speed as
well as for locomotives and freight wagons. For the TSI compliant test sites the results
do not vary, because the rail roughness and track decay rate is very similar. If the noise
sources and their characteristics don’t change, the vehicle transfer function is a
describing parameter for the vehicle noise part.
The results for the non TSI compliant test site C differ from the other sites, caused by
a non compliant rail roughness and track decay rate. It is shown that the vehicle transfer
function from one test site can be used at any other test site. If the vehicle transfer
function is known, the vehicle emitted noise part can be calculated on other track
properties.
Table 1. Comparison of three measured and predicted LpAeq pass-by levels at three different test
sites for different types of the same EMU family based on track transfer function of reference
vehicle type 3 at test site A (± indicates spread between three measurements)
Table 2. Comparison of measured and predicted LpAeq pass-by levels at test site A for different
trains, based on the averaged track transfer function of reference vehicle types 1 and 2
For an EMU it shall be checked, whether the results of TWINS calculations are
comparable to the measured track and vehicle part of the rolling noise. The pass-by
levels derived from the separation method were calculated by using the averaged track
On Separation of Vehicle Noise for Limit Setting in Future Legislation 37
transfer function at test site A and the average vehicle transfer function of all pass-bys
from this EMU at the test site A. According to the TWINS calculation the TSI rail
roughness and the ISO 3095 wheel roughness for a typical disc braked wheel was used
to calculate the total roughness.
The results are shown in Fig. 4 for the total pass-by level, the track and vehicle part
in 3rd octave bands. The total pass-by level of the TWINS calculation and the pass-by
level of the separation method show a good agreement. For frequencies over 3000 Hz
the TWINS calculated pass-by levels are higher than the results of the separation
method.
a) TWINS calculaons vs. Separaon Methods b) TWINS calculaons vs. Separaon Methods
measurements - Total pass-by level measurements - Track part pass-by level
105 105
100 100
95 95
Sound power level [dB(A)]
85 85
80 80
75 75
70 70
65 65
90
85
80
75
70
65
TWINS calculaon vehicle part
60
Sep. Methods measurement vehicle part
55
100 160 250 400 630 1000 1600 2500 4000 6300
Frequency [Hz]
Fig. 4. TWINS calculations and results of the separation method for an EMU with 160 kph for
the a) total pass-by level, b) the track and c) vehicle part of the pass-by level
For the vehicle part of the pass-by noise the results are equal only in the frequency
range between 1000 and 2000 Hz. For higher frequencies the TWINS results are up to
10 dB(A) higher. For lower frequencies the results of the separation method are higher.
This can be explained by taking the track part into account. Especially for the lower
frequencies the track part dominates the total pass-by level. The vehicle part is much
38 T. Thron, S. Leth, and B. Stegemann
lower, up to 20 dB(A) in the TWINS calculation. This means that the vehicle part can’t
be measured in correctly. The mistake in the total pass-by level however is negligible.
Based on the validation exercises it was shown, that a typical track transfer function for
a typical TSI NOI test site can be defined. This transfer function is a describing
parameter of a test site like rail roughness and track decay rate. Once the track transfer
function is known, the vehicle part of the pass-by level can be estimated. It was shown,
that the vehicle part of the pass-by level is equal at two different TSI NOI test sites. The
comparison with TWINS calculations shows a good agreement over a wide frequency
range to the measured vehicle and track part by using the proposed separation method.
By taking these results into account, it may be possible in the future to define a limit
value for the vehicle noise for legislative purposes but a considerable amount of work is
needed to find consensus and enough accuracy for such a method. Furthermore the
measured vehicle transfer function can be used for calculations on any arbitrary track
type, e.g. in noise prediction schemes. For an ongoing development process of the
proposed separation method the following points should be clarified:
• define the A2 filter for pad stiffness of a typical TSI NOI track type in the PBA
• test an alternative method to the reference vehicle method to increase the accuracy
especially for higher frequencies
• test a “small deviation method” for the influence of the track decay rate on the transfer
function, as proposed in [8]
• perform further benchmarks to validate the proposed method based on the technical
report [5]
References
[1] 2011/229/EU. Commission Decision of 4 April 2011 concerning the technical specification
for interoperability relating to the subsystem ‘rolling stock – noise’ of the trans-European
conventional rail system (2011)
[2] Leth, S., Frid, A.: Track quality extrapolation for railway vehicle pass-by noise. In: ICSV,
vol. 16 (2009)
[3] ISO 3095:2013. Acoustics – Railway applications – Measurement of noise emitted by
railbound vehicles
[4] Dittrich, M.G., Janssens, M.H.A.: Improved measurement methods for railway rolling
noise. Journal of Sound and Vibration 231, 595–609 (2000)
[5] prCEN/TR 1-1:2013.14. Railway applications – Noise emission – Measurement method for
combined roughness, track decay rates and transfer functions
[6] STAIRRS – Final Technical Report (December 2003)
[7] METARAIL Project – Final Report for Publication (December 1999)
[8] Létourneaux, F., Meunier, N., Fodiman, P.: Small deviations procedure: a new way to
introduce flexibility in the conformity assessment of reference tracks for pass-by acoustic
tests. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T.,
Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 118, pp. 523–531. Springer, Heidelberg (2012)
Estimating the Performance of Wheel Dampers Using
Laboratory Methods and a Prediction Tool
Summary. Wheel and rail dampers are well known mitigation measures against
rolling noise. A combination of laboratory measurements and computations
seems the most efficient way to determine their effect. The DEUFRAKO project
STARDAMP had the aim of supporting the transfer of wheel and rail dampers
from the research phase to their regular application. One goal of the project was
the development of a prediction tool that is dedicated to the estimation of the
efficiency of wheel and rail dampers. The input data relies on laboratory
measurements that are relatively easy to perform. This paper focuses on the
wheel damper part. Rail dampers are addressed in a companion paper.
1 Introduction
Rolling noise is the predominant railway noise source within a relatively wide speed
range. As rolling noise is generated by the roughness of rail and wheel, a common
mitigation measure consists in reducing the roughness, e.g. by rail grinding, and the
use of brake systems that avoid wheel corrugation. To obtain further reductions of
noise emission at the source, the dynamic response of wheel and track to roughness
excitation can be minimised by the use of rail and wheel dampers. This paper
focuses on the wheel damper part while rail dampers are addressed in a companion
paper [1].
Several measurement campaigns have been performed in recent years to quantify the
effect of rail and wheel dampers (e.g. within the SILENCE [2] and LzarG projects [3]).
However, such field tests are generally costly and time consuming. Moreover, field
tests only deliver information about a few very specific cases (in terms of track, rolling
stock, roughness, etc) which are difficult to transpose to different situations. On the
other hand, damping is very difficult to assess using purely computational approaches.
For these reasons, a combination of laboratory measurements and computations seems
the most efficient way to determine the effect of rail or wheel dampers. A software tool
called STARDAMP has been developed to process laboratory measurements of wheel
and rail damping to estimate the performance of dampers. This uses the same
theoretical models as in the TWINS software [4] but is much easier to use. Indeed, a
main goal of the STARDAMP project was to provide an easy-to-use tool for
infrastructure managers and public authorities to help the decision making process
regarding railway noise mitigation measures.
Regarding the wheel, a modal basis corresponding to a wheel both with and without
dampers is required. The wheel geometry is too complicated to allow using simple
analytical models, therefore the (bare) wheel is described in terms of Finite Element
(FE) computed natural frequencies and mode shapes at the contact point and at a
limited number of positions on the external face. This information is stored in a text file
which is loaded in the STARDAMP tool; wheel mobilities are then calculated through
modal summation.
Bare wheel FE models are very reliable and do not need any experimental tuning. In
the case of the ‘damped wheel’ the modal basis is updated with experimental data. A
general introduction to this procedure can be found in reference [5]. This paper
describes the calculation and measurement procedure for wheels and the use of the data
with the STARDAMP prediction tool for rolling noise.
As mentioned above, the undamped wheels are modelled with FEA. The FE model
can be built in 2D (axi-symmetric model) or 3D; the 2D model is preferable because
of simpler post-processing of the data. Generally, a single wheel clamped at the wheel
hub is used. This represents a sufficient approximation of the real boundary condition
because the modes that mainly contribute to rolling noise are hardly influenced by the
axle (having 2 or more nodal diameters). All wheels are assumed to be axi-symmetric,
even if a 3D FE-model is used for computations. The STARDAMP software uses a
wheel modal basis defined over a wheel section, assuming a cos (n ) dependence of
all modes in circumferential direction (n being the number of nodal diameters and
being the circumferential angle of the wheel).
Bare, axi-symmetric monobloc wheel models do not need any tuning with
experimental data because calculated mode shapes and natural frequencies are
generally very reliable provided that suitable precautions are taken. The modal
damping of a given wheel mode mainly depends on the number of nodal diameters of
this mode. It is virtually independent of the wheel design. The values to be used are
therefore well known [6]. Importantly, for the most important modes the damping that
is introduced by wheel-rail interaction (so called ‘rolling damping’, due to the
behaviour of the coupled rail as an equivalent damper) is much higher than the modal
damping inherent to the wheelset itself. This is also the reason why a direct comparison
between the behaviour of a free bare wheel and a free wheel equipped with dampers
does not give any useful information about rolling noise reduction. The simulation of
wheel-rail interaction is necessary in order to obtain such information.
Estimating the Performance of Wheel Dampers Using Laboratory Methods 41
The wheel modal basis of the ‘damped wheel’ is constructed by updating the
(axi-symmetric) bare wheel model with measured natural frequencies and modal
damping. A full experimental analysis is often not necessary and the modal damping
can be extracted from a set of measured receptances by using the “half power
bandwidth” method. However, in the case of efficiently working absorbers, this
simplified method may lead to errors. These can be overcome by comparing the
recalculated receptance (by the use of the prediction tool) with the directly measured
receptance. This procedure is illustrated in section 3.1.2 The methodology proposed
in the prediction tool does not account for any modification that occurs in the wheel
mode shapes due to the presence of the dampers, nor shielding, although in principle
changes in the mode shapes could be included in the updating procedure. A full
wheelset should be used for measurements. This permits the clamped wheel FE model
to be tuned directly with measured data. If a wheelset is not available, measurements
can be performed on a single wheel. Both a free wheel and a clamped wheel FE
model are then built and a correction factor between the natural frequencies of the
free and the clamped model is determined for each mode. The clamped wheel model
is then tuned according to measurements, taking into account the previously
determined correction factor for natural frequencies. Experience shows that for certain
types of absorbers the mounting conditions influence their efficiency. It is therefore
highly recommended to repeat measurements on several samples or to repeat
measurements after dismantling and remounting the absorbers.
Non-axi-symmetric wheels and resilient wheels are not included in the scope of the
prediction tool. However, their modelling is feasible to a certain extent. Similarly to
TWINS [4], special care is necessary to ensure that the simplification of a wheel model
still leads to valid results. The modelling of wheels equipped with brake discs can
generally be performed analogously to damped wheels because the main effect of brake
discs is added damping.
The prediction tool is split into two modules. A first module named mp-Editor permits
FEA results to be imported and allows the wheel modal basis to be updated with
experimental data. Additionally, it is possible to compute wheel receptances and
compare them with measured ones.
The second module is called ‘STARDAMP’. It implements TWINS-like predictions
of rolling noise in a user-friendly way. Compared with TWINS [4] the number of
options is limited in order to allow more ready access by non-expert users.
42 B. Betgen et al.
The structural dynamics of the wheel is calculated from a FE model and stored in a text
file which contains modal frequencies, modal masses, modal damping and mode
shapes. The description of mode shapes is limited to 3 degrees of freedom at the
nominal contact point plus the axial displacement of a small subset of nodes on the
external side of the wheel. This information is only needed over one single wheel
section, with an additional identifier n indicating the number of nodal diameters of each
mode. The reason for this is that the wheel is considered as axi-symmetric, i.e. each
mode (with a given deformation of the wheel section) has a cos (nθ) dependency over
its circumference. In TWINS [4], this text file has the extension ‘mp’ for ‘modal
parameters’ file. In STARDAMP, a new file type is introduced that namely includes a
fuller identification of all modes, i.e. a description of the type of wheel section
deformation.1 Also the radial coordinates of the response nodes are indicated, which
previously had to be entered by the user. The mp-Editor permits the wheel modal data
to be loaded either directly from FE analysis output files or from an existing mp-file.
The data is then organised in tables that allow natural frequencies and damping to be
edited easily. Finally, the correct tuning of the wheel modal basis can be checked by
comparing calculated and measured receptances. These steps are explained in the
subsections below.
1
Generally the most important modes in terms of rolling noise are radial modes (termed ‘R, n’)
and axial modes with one nodal circle (termed 1 L, n) of order n between 2 and 4.
Estimating the Performance of Wheel Dampers Using Laboratory Methods 43
Frequency Response Functions can therefore be loaded into the mp-Editor. To control
not only radial but also 1-axial modes, radial receptances at the outer side of the tread as
well as at the centre should be checked. The mp-Editor therefore permits an offset from
the nominal contact point to be defined. Similarly, axial receptances should be checked.
Fig. 1 compares radial receptances of a freight wheel equipped with dampers
measured and computed at mid-tread. A certain number of modes in this model have
not yet been tuned in the left hand plot, because the simplified modal analysis
performed did not permit all modes to be identified (notably 2-axial modes). This
‘initial guess’ can be readily improved through an iterative updating process whose
result is shown on the right hand plot of Fig. 1.
-140 -140
measurement measurement
calculation calculation
Recpetance, dB re 1 m/N
Recpetance, dB re 1 m/N
-160 -160
-180 -180
-200 -200
-220 -220
-240 -240
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Frequency, kHz Frequency, kHz
Fig. 1. Comparison of measured and calculated radial receptances at contact point before (left)
and after (right) tuning
3.2 STARDAMP
elastic support has a single layer only. For ballasted track the sleeper can be monobloc
(concrete or wooden) or bibloc. The contact stiffness in the vertical direction is
computed by linearizing the relationship between wheel-rail approach distance and
applied load as formulated for example in [7]; contact vertical receptance is readily
derived from stiffness. Lateral receptance is obtained in a similar manner but in this
case the effect of creepages (relative velocity divided by mean rolling velocity) is also
considered.
Default traffic types can be chosen to specify the static axle load, number of axles
per length, speed and roughness. However, all parameters can also be set manually.
Pre-defined traffic types include ‘freight’, ‘passenger’ and ‘high speed’ traffic. A
default wheel model for each traffic type is provided with the tool. The user can also
include their own. When assessing wheel damper effectiveness, specific new modal
parameters files have to be prepared as outlined above; the undamped and damped
wheels must correspond to each other apart from the increase in modal damping. For
the assessment of rail dampers, the default wheel model is sufficient as the wheel has
very low influence on track behaviour.
Estimating the Performance of Wheel Dampers Using Laboratory Methods 45
The track model is defined in terms of rail pad stiffness and damping as well as
sleeper type. Again, default settings are proposed, but the parameters can be set freely.
As mentioned above, ballasted as well as slab track models are available. Track decay
rates can be calculated from the analytical model or loaded from external files. The
latter is necessary for the assessment of rail dampers, which is explained in reference
[1]. Measured track decay rates are not needed for the assessment of wheel dampers,
although the track properties have an impact on the effect of wheel dampers on the
overall level. Indeed, the same reduction of wheel radiated sound power will lead to a
higher overall reduction on a quieter track.
90 90
total rail
SPL, dB re 20x10 -6 Pa
SPL, dB re 20x10 -6 Pa
80 80
70 70
60 60
sleeper w heel
50 50
100 200 400 800 1.6k 3.15k 100 200 400 800 1.6k 3.15k
Frequency, Hz Frequency, Hz
4 Conclusion
References
[1] Toward, M.G.R., Squicciarini, G., Thompson, D.J., Gao, Y.: Estimating the performance of
rail dampers using laboratory methods and a prediction tool. In: Nielsen, J.C.O., Anderson,
D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D., Tielkes, T., Towers, D.A., de Vos,
P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 126,
pp. 47–54. Springer, Heidelberg (2015)
[2] SILENCE project (2008), http://www.silence-ip.org/site/
[3] LZarG project, see documentation on, http://www.lzarg.de/
[4] Thompson, D.J., Hemsworth, B., Vincent, N.: Experimental validation of the TWINS
prediction program for rolling noise, part 1: description of the model and method. Journal of
Sound and Vibration 193(1), 123–135 (1996)
[5] Betgen, B., Bouvet, P., Thompson, D.J., Demilly, F., Gerlach, T.: Assessment of the
efficiency of railway wheel dampers using laboratory methods within the STARDAMP
project. In: Proceedings of Acoustics 2012, Nantes (2012)
[6] Jones, C.J.C., Thompson, D.J.: Rolling noise generated by wheels with visco-elastic layers.
Journal of Sound and Vibration 231(3), 779–790 (2000)
[7] Timoshenko, S.P., Goodier, J.N.: Theory of elasticity, 3rd edn. McGraw-Hill (1982)
[8] Thompson, D.J., Fodiman, P., Mahé, H.: Experimental validation of the TWINS prediction
program for rolling noise, part 2: Results. Journal of Sound and Vibration 193(1), 137–147
(1996)
[9] Jones, C.J.C., Thompson, D.J.: Extended validation of a theoretical model for railway
rolling noise using novel wheel and track designs. Journal of Sound and Vibration 267(3),
509–522 (2003)
Estimating the Performance of Rail Dampers
Using Laboratory Methods and Software Predictions
Summary. Rail dampers are designed to reduce the rail component of rolling
noise by increasing the attenuation with distance along the rail (decay rate, DR).
There is no standardized method to assess the performance of rail dampers. The
method described here, developed during the Franco-German STARDAMP
project, uses laboratory tests and computer simulation to avoid the need for
expensive and time-consuming field trials. The premise of the method is that the
DRs of a damped track can be found from summing the DRs of a short-section of
damped ‘freely supported’ rail and the DRs of an undamped track. Reasonable
predictions of the decay rates of a test track have been made using this method.
Software has been produced that implements TWINS-like predictions of rolling
noise with and without rail dampers to predict the damper effect. The effect of
rail pad stiffness on the effectiveness of rail dampers has been considered for
track constructions typical in the UK and a regional train travelling at 120 km/h.
For track fitted with ‘soft’ 120 MN/m rail pads, the dampers are predicted to
reduce the total level by 2.5 dB(A) while with the ‘stiff’ 800 MN/m pads a
0.7 dB(A) reduction is expected.
1 Introduction
The noise radiated by the rail is usually the dominant source of rolling noise between
0.5 and 2 kHz and often in terms of overall level [1]. Rail dampers are now
commercially available that are designed to reduce the rail component of noise by
increasing the attenuation with distance along the rail (decay rate, DR) and hence
reduce the radiating length of the rail. These dampers tend to be bolted or clipped onto
the rail between sleepers and work on the principle of tuned mass dampers [2-4]. There
is no standardized method to assess the performance of rail dampers. Railways are
often obliged to undertake line testing, which can be expensive and may lead to results
that are ambiguous or difficult to generalise.
Two methods for determining damped track DRs were tested in the STARDAMP
project. With both methods, the damped track DRs are found by summing the DRs of
an undamped track on which the dampers are intended to be fitted and the DRs of a
section of freely supported damped rail [5]. With the first method, the damped free-rail
DRs are determined for either a 4 m or 6 m length of damped rail at low frequency from
the modal properties of the rail, and at high frequencies directly from point and transfer
frequency response functions (FRFs) at either end of the rail [2,5]. In the second
method, the damped free-rail DRs are determined from FRFs measured at intervals
along a longer (e.g. 32 m) rail using a method similar to the track decay rate
measurement standard EN15461:2008 [6]. The two methods for determining DRs of
damped ‘freely supported’ rails showed reasonable agreement between 300 Hz and
5 kHz. The modal method for determining DRs on the ‘short’ rail was restricted to low
frequencies (< 300 Hz) and resulted in much lower rates than those measured on the
‘long’ 32 m rail. With dampers designed for conventional track, below 400 Hz the
damper DRs are relatively low and tend to have little influence on overall track DRs.
Consequently, the direct short-rail method, yielding plausible measurements down to
300 Hz, is considered to be sufficient for many applications. The method is
summarized in Section 2; further details can be found in [5,7].
The in-situ performance of dampers will depend not only on their effect on the track
DRs but also on the relative contributions of the wheels and individual track
components to the radiated noise. These contributions might be predicted (e.g. using
TWINS [8]). However, currently available software require a large number of input
parameters and considerable expertise of the user. An aim of the Franco-German
STARDAMP project is to develop a more user-friendly method to predict the acoustic
performance of rail and wheel dampers. The method described in Section 3 uses
laboratory tests and computer simulation and avoids the need for expensive and
time-consuming field trials. The application to wheel dampers is described in a
comparison paper [9]
The premise of the 6 m rail method developed within STARDAMP is that the DRs of a
damped track can be found by summing the DRs of a damped ‘freely supported rail and
the DRs of an undamped track. The damped free-rail DRs are derived from the
attenuation measured along a 6 m length of rail.
The proposed damped free rail test procedure is outlined below. To demonstrate the
method, example results are given for Schrey and Veit (S&V) rail dampers mounted on
UIC 60 rail. Each S&V damper consists of two 7.0 kg laminated rubber and steel
construction absorber masses bolted on to the rail web via a solid steel base plate
(2.8 kg). The total mass of each damper assembly is 18.6 kg. Other dampers were tested
within the project with broadly similar results [7].
With the method, dampers are installed symmetrically over the whole length of a
6 m rail at a centre-to-centre spacing representative of the intended track installation (in
this case UIC 60 rail and 0.6 m spacing see Fig. 1). The rail should be ‘freely
Estimating the Performance of Rail Dampers Using Laboratory Methods 49
suspended’ at either end on a foundation that is soft enough so that the bounce mode
has a natural frequency less than 30 Hz. Here, 12 rubber rail pads were used at either
end of the rail, giving a bounce mode ≈ 20 Hz. Miniature accelerometers are rigidly
attached as close as possible to either end of the rail, attached either at the centre of the
rail head for vertical measurements or on the side of the rail head for lateral
measurements. Here, they were attached 5 mm from the rail end using beeswax. A
small instrumented hammer, with a hard tip, is used to excite the rail with a force of
approximately 400 N. This was adequate to ensure that the force spectrum is flat up to
high frequencies, dropping by less than 20 dB by 7 kHz.
For both lateral and vertical measurements, a point FRF at one end and a transfer
FRF from one end to the other end is measured. The rail temperature should be
controlled between 18 and 25°C during the tests. Further measurements are
recommended at temperatures encompassing the in-situ temperature range. It is also
recommended to measure more than one sample of rail fitted with a given type of rail
damper in order to check variability.
In each one-third octave band, the DR is determined as the decibel difference of the
transfer FRF to the point FRF divided by the rail length. With low DRs, the % error in
the DR for a given dB error in the FRFs is large and therefore in practice the lower
threshold for reliable measurements is found to be ~1.0 dB/m.
An instrumented hammer was used to excite the rail at each of the measurement
points in turn. The response was measured with an accelerometer mounted at the
reference point.
DRs in each ⅓ octave band up to 5 kHz were calculated in dB/m from the point
frequency response function (FRF) at the reference point, A( x0 ) , and the transfer
FRFs, A( xn ) , between the reference position and the other points on the measurement
grid, xn , using:
nmax 2
A( x n )
DR = 4.343 ∑ A( x )
n =0
2
Δx n (2.1)
0
To calculate improvements to the overall sound level, predictions are required of the
contributions of the individual track components, with and without the dampers, for
which the software described in the next section is intended.
3 STARDAMP Software
The tool, developed within the STARDAMP project, is based on the same theoretical
models used in the TWINS software [8]. It implements an analytical description of the
wheel-rail interaction where the contact forces are calculated as the ratio between the
wheel-rail roughness spectrum and the sum of rail, wheel and contact mobilities. Both
vertical and lateral degrees of freedom at the contact are considered. From the contact
forces, wheel, rail and sleeper responses are calculated and the sound power levels
estimated through radiation efficiencies. If rail dampers are to be included, their effect
is accounted for by replacing analytically calculated rail DRs with measured DRs.
Finally a simple model for acoustic propagation above a partially absorptive flat ground
gives the sound pressure levels at specific field positions.
Vertical and lateral rail mobilities are calculated with a model of a Timoshenko
beam [11] on a double layer continuous elastic support, which represents pads, sleepers
and ballast. Coupling between vertical and lateral motion is modelled by a constant
factor (normally between -7 dB and -12 dB). To define the track, several combinations
of track types, sleeper types, rail types and pad stiffness and damping values can be
selected. Most importantly, the track can be ballasted or slab-track, in this second case
the continuous elastic support has a single layer only. For ballasted track the sleepers
can be monobloc (concrete or wooden), modelled as beams, or bibloc, modelled as
lumped masses. The software can determine DRs analytically from the track’s
theoretical response or use measured values. When measured DRs are used, all the
other wheel and track-related quantities (e.g. mobilities and contact forces) are retained
from analytical calculations and are assumed not to be modified by the presence of
dampers.
52 M.G.R. Toward et al.
The wheel is described with a Finite Element (FE) model that is used to compute
natural frequencies and mode shapes at the contact point and at a limited number of
positions on the external face. This information is stored in an external text file (modal
parameters file) that is loaded in the software; wheel mobilities are then calculated
through modal summation and modal damping ratios can be added, either adopting
standard values or after measurements. Modal models of three typical undamped
wheels of freight, regional, and high-speed trains are implemented in the software. The
user can also include their own.
Typical roughness spectra corresponding to wheels with cast-iron brake blocks,
K-block brakes and disc brakes are supplied; again measured values can be loaded by
the user. Generally, the number of accessible options is reduced with respect to TWINS
in order to simplify use. These are presented through a simple Graphical User
Interface. Lastly, to increase reliability, the final results shown are an average over
three contact positions: the nominal one (70 mm from flange back) and ± 10 mm from
this.
The software permits the direct assessment of rail dampers, wheel dampers, or a
combination of both. In this paper only the application of rail dampers is discussed;
wheel dampers are discussed in [9]. When the software is used for assessing dampers, it
first computes pass-by noise levels for a baseline model without dampers then it
estimates noise levels considering the dampers. The effectiveness can be then
visualised by comparing damped versus non-damped sound pressure spectra and
overall levels.
To illustrate the STARDAMP software, the effect of dampers on noise from a train
pass-by has been predicted for two different track conditions typical in the UK. For the
first case, ‘soft’ 120 MN/m rail pads are assumed, while in the second case, stiffer
800 MN/m pads are assumed. Other track parameters were selected to be consistent with
the test track (see Section 2.2). For both cases, a regional train travelling at 120 km/h with
roughness representative for disc brakes has been assumed. Rail roughness measured
on a ‘typical’ ballasted UK track with 800 MN/m pads was assumed for both cases. The
decay rates measured on the short rail (Figs 2 and 3) have been used as input to the
software, along with measured track decay rates applicable to each pad stiffness.
Fig. 4 gives the predicted noise levels for a receiver at 7.5 m from the centre of the
track fitted with soft rail pads. The noise contribution of the rail is dominant in the mid
frequency region, wheel noise is the main source at high frequency, while the
contribution of the sleepers is at a much lower level. There is a substantial reduction in
the rail contribution after introducing the rail dampers (solid lines), giving an overall
reduction of 6 dB(A) in this component. There is also some reduction in the sleeper
noise but this component is relatively low compared to the others and has minimal
effect on the overall level. There is no reduction predicted in the wheel component of
noise because the contact forces in the model are not modified by the introduction of
dampers on the track. The overall noise is reduced by about 2.5 dB(A).
Estimating the Performance of Rail Dampers Using Laboratory Methods 53
Fig. 5 gives the predicted noise levels for the track fitted with stiff pads. The higher
stiffness of the rail pads decreases the rail component of rolling noise but conversely
increases the noise radiated from the sleeper due to the increased coupling (compare
Fig. 4 and Fig. 5). As the decay rates are initially higher, the damper only reduces the
rail component by about 3.5 dB(A). As a result of the lower rail contribution, the wheel
noise dominates the overall noise level and hence the effect of the damper on the
overall noise is relatively small at 0.7 dB(A).
85
Tota l ; 90.2 dB(A)
80
Pa
Ra i l ; 87.8 dB(A)
-6
85
80 Tota l; 87.9 dB(A)
Pa
75
Wheel; 86.4 dB(A)
SPL, dB ref 20x10
4 Conclusions
References
[1] Thompson, D.J.: Railway Noise and Vibration - Mechanisms Modelling and Means of
Control. Elsevier (2009)
[2] Thompson, D.J., et al.: A tuned damping device for reducing noise from railway track.
Applied Acoustics 68, 43–57 (2007)
[3] Van Haaren, E., van Keulen, G.A.: New rail dampers at the railway link
Roosendaal-Vlissingen tested within the Dutch Innovation Programa. In:
Schulte-Werning, B., et al. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 378–383. Springer, Heidelberg (2008)
[4] Asmussen, B., et al.: Reducing the noise emission by increasing the damping of the rail.
Results of a Field Test. In: Schulte-Werning, B., et al. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 99, pp. 229–235. Springer,
Heidelberg (2008)
[5] Toward, M.G.R., Thompson, D.J.: Laboratory methods for testing the performance of
acoustic rail dampers. In: Proceedings of Acoustics 2012, Nantes, France (2012)
[6] EN 15461:2008: Railway applications – Noise emissions, Characterization of the dynamic
properties of track sections for pass by noise measurements (2008)
[7] STARDAMP Standardisation of damping technologies for the reduction of rolling noise –
Final Report. DeuFrako Projekt STARDAMP (2013)
[8] Thompson, D.J., et al.: Experimental Validation of the TWINS prediction programme for
rolling noise, PART 1: Description of the model and method. Journal of Sound and
Vibration 193, 123–135 (1996)
[9] Betgen, B., Bouvet, P., Squicciarini, G., Thompson, D.J., Jones, C.J.C.: Estimating the
performance of wheel dampers using laboratory methods and a prediction tool. In: Nielsen,
J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D., Tielkes, T.,
Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 126, pp. 39–46. Springer, Heidelberg (2015)
[10] Broadbent, R.A., et al.: Evaluation of the effects of temperature on rail pad properties, rail
decay rates and noise radiation. In: 16th International Congress on Sound and Vibration,
Krakow (July 2009)
[11] Timoshenko, S.P., Goodier, J.N.: Theory of elasticity, 3rd edn. McGraw-Hill (1982)
Experimental and Theoretical Studies on Impact Noise
Generation due to Rail Joints
Summary. When a train passes over discontinuities on a rail, e.g. rail joints,
impact noise due to the discontinuities is generated impulsively. In this paper, an
attempt is made to understand the impact noise generation mechanism experi-
mentally and theoretically. By performing static and running tests, vibratory
properties of the track and wheel are investigated first. Second, on the basis of the
measurements in the tests, a theoretical model to predict impact noise is devel-
oped.
Through the static test for the track, it is found that two facing rails at a rail
joint no longer move together above 1000 Hz. In the running tests, for both the
wheel and rail vibrations, the measured A-weighted levels at three rail joints
show an increase of about 9 dB for a doubling of train speed. Also, by using the
measured results, the separate contribution of noise from wheel, rail and sleeper
to the total impact noise at one rail joint is estimated quantitatively. The results
show that the sleeper has the greater contribution below 630 Hz, whilst the wheel
is predominant above 2000 Hz. By using the prediction model, the overall trends
in noise are well predicted. Also, the model gives an estimate of the contributions
of wheel, rail and sleeper to the total impact noise at the rail joint, and the pre-
dictions show good agreement with the results estimated by the measurements.
1 Introduction
When a railway train passes over discontinuities on a rail, e.g. rail joints, the discon-
tinuities induce strong interaction forces on both wheel and track. Consequently, im-
pact noise is generated impulsively and strongly compared with rolling noise. As a
way of reducing impact noise, continuously welded rail has been widely used. How-
ever, many rail joints still remain at crossing, track-circuit breaks and so on. There-
fore, in order to make effective measures for reducing impact noise, a better under-
standing of its generation is required.
Comprehensive studies have been carried out to investigate impact noise genera-
tion due to rail joints using theoretical models and scale model experiments [1, 2]. In
[2], Wu and Thompson developed a hybrid approach, where an equivalent roughness
spectrum due to a wheel passing over a rail joint is derived by a simplified non-linear
model of the wheel and track, and, by applying the spectrum in the TWINS model,
the noise radiation from both wheel and track is predicted.
In this paper, an attempt is made to understand the impact noise generation mecha-
nism both experimentally and theoretically. By using measurements obtained in both
static and running tests, the contributions of wheel, rail and sleeper to the total impact
noise are estimated. Also, the model developed by Wu and Thompson is validated and
modified on the basis of the measurements.
Both static and running tests were carried out at the test site at the Railway Technical
Research Institute. The track was of ballasted construction, and wooden sleepers were
used. This is located on a narrow gauge line, with track gauge 1.067 m. The geometries of
the rail joints are given in Table 1. These joints were located at the mid-point between
two neighbouring sleepers. Fig. 1 shows the setup of the vibration and sound measure-
ments for the track. The measurements were made at three positions for each rail joint:
- two accelerometers under the rail and one accelerometer on sleeper,
- one microphone at a distance of 2 m from the centre of the track.
For the wheel, vibration measurements were also carried out. The measurements
were made at three positions on the surface of the wheel (two on the web, one on the
tyre). In these static tests, the vibratory behaviours of the track and wheel were inves-
tigated by using a hammer excitation. The excitation was applied on the upper and
lower rail, respectively. Especially, for the track, the influence of a rail joint on rail
vibration was examined carefully at rail-joint A. In the running tests, the sound and
vibration measurements were performed at the three rail joints in the site, and simul-
taneously the vibrations of the wheel were measured when the wheel rolled over the
joints. The same train ran back and forth on the same track in the site during the running
tests.
Rail-
0.4m
Height Width
joint Fishplate
2.2 Vibratory Behaviour of the Track and Wheel in the Static Tests
Fig. 2(a) shows the vertical point accelerances at rail-joint A. In Fig. 2, the results
measured at the mid-point between two rail-joints are also shown as a reference where
no influence of the joints is expected. It is found that, above 1000 Hz, the result at
rail-joint A is greater than that at the mid-point. This could be due to the fact that the
vibratory behaviour of the rail at the rail-joints is similar to that of a cantilever beam at
Experimental and Theoretical Studies on Impact Noise Generation due to Rail Joints 57
free end. In the results measured at the mid-point, the resonance behaviour responsible
for the relative motion of the rail and sleeper cannot be seen clearly. This is probably
because the rail was fastened on lightweight wooden sleepers directly by track spikes
and the track in the site was not kept in good condition. Fig. 2(b) shows the difference
between rail vibrations induced on two facing rails across rail-joint A when one of the
rails is excited by a hammer. Results are shown only for vertical direction. It is found
that the results are close to zero below 800 Hz, which means that both the rails move
together. However, above 1000 Hz, differences of up to about 10 dB are found. This
suggests that, above 1000 Hz, the two rails are no longer coupled. Also, this indicates
that above-mentioned vibratory behaviour at the joint should be introduced in the
prediction model.
1 15
(a) (b)
2
0 10
10
-1 5
10
-2 0
10
-3 -5
10
100 1000 63 125 250 500 1k 2k 4k
Frequency (Hz) Frequency (Hz)
Fig. 2. Vibratory behaviour of the rail at rail-joint A. (a) Point accelerance. ―, Measurement at
rail joint, - - -, measurement at mid-point, ―, prediction (see Section 3.1). (b) Level difference
between vibrations on two facing rails. ―, Measurement (upper rail), - - -, measurement (lower
rail), ○, prediction (see Section 3.1).
Fig. 3 shows the measured vibrations of the wheel and rail plotted against train speed.
For both the wheel and rail vibrations, the results measured at the three rail joints
increase at a rate of approximately 30log10V, where V is the train speed. In addition, it is
found that the overall vibration levels depend on rail joint geometry. At one rail-joint,
the rail vibration for a step-up joint is slightly greater than that for a step-down joint,
whereas the results of the wheel vibration are approximately equivalent for both joints.
120 110
(a) (b)
A-weighted vibration
A-weighted vibration
velocity level (dB)
110 100
100 90
90 80
10 20 30 40 50 60 10 20 30 40 50 60
Train speed (km/h) Train speed (km/h)
Fig. 3. Rail and wheel vibrations (dB ref 5·10-8 m/s). ○, Step-down joint at A, ●, step-up joint at
A, ∆, step-down joint at B, ▲, step-up joint at B, □, step-down joint at C, ■, step-up joint at C. (a)
Vertical rail vibration, (b) radial wheel vibration.
58 T. Kitagawa et al.
110 110
A-weighted VL (dB)
A-weighted VL (dB)
(a) (b)
100 100
90 90
80 80
63 125 250 500 1000 2000 4000 63 125 250 500 1000 2000 4000
Frequency (Hz) Frequency (Hz)
Fig. 4. Rail vibration at rail-joint A (dB ref 5·10-8 m/s). ―, Upper rail at 20 km/h, - - -, lower rail
at 20 km/h, ―, upper rail at 35 km/h, - - -, lower rail at 35 km/h. (a) Step-up joint, (b) step-down
joint.
2
100 10
(a) (b)
A-weighted VL (dB)
2
80 0
10
70
-1
10
60
-2
50 10
-3
40 10
2 3 2 3
10 10 10 10
Frequency (Hz)
Frequency (Hz)
Fig. 5. (a) Wheel radial vibration at rail-joint A (dB ref 5·10-8 m/s) at 35 km/h. ―, Step-down
joint, ―, step-up joint. (b) Wheel point accelerance ―, Radial direction, ―, axial direction.
Fig. 4 shows the vertical rail vibrations induced on the ends of the two rails at
rail-joint A. For the step-up joint, the results at the upper rail are greater, and the dif-
ference between the results at the two rail ends is similar to those obtained in the static
test (see Fig. 2(b)). This suggests that, for a step-up joint, the upper rail is primarily
excited by the wheel. For the step-down joint, similar trends can be seen, but the results
at the lower rail are greater. This also indicates that, for a step-down joint, the excitation
due to the wheel mainly occurs at the lower rail. Fig. 5(a) shows the wheel vibration in
the radial direction at rail-joint A for a speed of 35 km/h. A series of peaks appear in the
results, which are closely associated with the resonant frequencies of the wheel (see
Fig. 5(b)). Also, at most of these
peaks, the results are equivalent in
amplitude for both joints. 100
A-weighted SPL (dB)
90
2.4 Contributions of Wheel, 80
Rail and Sleeper to Impact
70
Noise
60
The separate contributions of noise 50
from wheel, rail and sleeper to the
40
total impact noise at rail-joint A 63 125 250 500 1000 2000 4000 OA
during a train pass-by are estimated Frequency (Hz)
by combining the measured wheel Fig. 6. Results estimated on the basis of measured
and rail vibrations with the radia- rail and wheel vibrations at rail-joint A for a
tion model used in the TWINS step-down joint at 35 km/h. −●−, measurement, −●−,
model. Fig. 6 shows the results total (estimated), -·-, sleeper, ···, rail, - -, wheel.
Experimental and Theoretical Studies on Impact Noise Generation due to Rail Joints 59
estimated on the basis of the measurements in the form of absolute spectra. The results
show good agreement with the measurements above 200 Hz. It can be seen that the
sleeper is the most important source below 630 Hz, whilst the wheel is the predominant
source above 2000 Hz. In the middle frequencies, the rail component of noise becomes
dominant. Also, the results estimated at rail-joints B and C have similar trends (not
shown here).
The predictions in term of noise and vibration are performed basically according to the
hybrid approach [2] by the following procedure:
(1) The trajectory of the centre of a wheel rolling over a rail joint on a rigid track is
estimated from the geometry of the two facing rail at the joint. For simplicity, the
rails are not dipped.
(2) Fig. 7 shows the wheel/track
W
interaction model. The wheel
wheel
consists of its mass with a
Mass
damped spring. The spring is x
tuned to match the main trough
Spring Damper
around 300 Hz in the measure- z
F
ments for the wheel accerelance, Spring
track F
and the damper is used to reduce zR=0 Beam
zL=0
the sharpness of this trough. The Spring
track is modelled as a pair of Mass
Spring
semi-infinite Timoshenko beams
on a continuous support, which is Fig. 7. Wheel/track interaction model
composed of a resilient layer (the
rail pads), a mass layer (the sleepers) and a second resilient layer (the ballast). In
order to consider the vibratory behaviour at the joint seen in Section 2.2, the two rails
are connected by a complex spring, which includes stiffness in vertical displacement.
In Fig. 2(b), the result predicted with the track model is also shown. The prediction
shows good agreement with the measurements.
(3) The vertical wheel/rail interaction model in the time domain in (2) is expressed in a
state-space form. The track and wheel are coupled through a non-linear Hertzian
contact force. The contact force induced by the wheel/rail impact process is nu-
merically estimated by using the Runge-Kutta method.
(4) By using the results obtained in (3), the contact force in the time domain is trans-
formed into the frequency domain. An equivalent roughness spectrum is derived
from the force spectrum by using a linear wheel/rail interaction model, in which the
contact spring is linear and there is no loss of contact between the wheel and rail.
(5) The equivalent roughness spectra are used as inputs into TWINS to predict noise
radiation from the wheel and rail.
60 T. Kitagawa et al.
110
(a) (b)
100
A-weighted SPL (dB)
Prediction (dB)
100
90
90
80
80
80 90 100 10 20 30 40 50 60
Measurement (dB) Train speed (km/h)
Fig. 9. (a) Predicted noise plotted against measured noise for all cases. ○, Step-down joint at A, ●,
step-up joint at A, ∆, step-down joint at B, ▲, step-up joint at B, □, step-down joint at C, ■, step-up
joint at C (b) Overall noise levels plotted against train speed at rail-joint A. ○, Measurements for a
step-down joint, □, measurements for a step-up joint , ―, prediction for a step-down joint, - - -,
prediction for a step-up joint.
Experimental and Theoretical Studies on Impact Noise Generation due to Rail Joints 61
at rail-joint A. The measurements are also shown for comparison. The results are well
predicted. The predictions are almost the same for both step-up and step-down joints.
This is because the equivalent roughnesses are almost the same for step-up/step-down
joints at a given rail joint below 60km/h.
In order to consider the spectral variation, the difference between predicted and
measured noise spectra is constructed for each of six rail-joint conditions in 40 meas-
urements. Fig. 10 shows the spectral differences as the mean for all cases. The average
difference is -1.5 dB in the frequency range 63-4000 Hz (over-predicted). The results
can be seen to be over-predicted below 250 Hz, whereas the results show an un-
der-prediction in the frequency range 630-1250 Hz. This over-prediction is closely
related to the over-estimation of the sleeper vibration. It is also found that these trends
are similar for both step-up and step-down joints. This is again due to approximately
the same equivalent roughness for both joints. Fig. 11 gives an estimate of the contri-
butions of wheel, rail and sleeper to the total impact noise for a step-down joint at
rail-joint A. The sleeper is the important source below 630 Hz, and the wheel compo-
nent of noise is predominant above 2000 Hz. The contribution of each noise component
is similar to that seen in Fig. 6 in the whole frequency range. Also, the results at
rail-joints B and C had similar trends to those seen in Fig. 11 (not shown here). This
suggests that the model is reasonable to predict impact noise at rail joints.
100
Measurement - Prediction (dB)
15
A-weighted SPL (dB)
90
10
80
5
0 70
-5 60
-10 50
-15
63 125 250 500 1000 2000 4000 40
Frequency (Hz) 63 125 250 500 1000 2000 4000 OA
Frequency (Hz)
Fig. 10. Average differences between predicted and Fig. 11. Predicted noise and measured noise at
measured noise spectra for all cases. ―, All cases, rail-joint A for a step-down joint. Train ran
- - -, step-down joints, ―, step-up joints. at 35 km/h. −●−, Measurement, −●−, total
(prediction), - - -, sleeper, ······, rail, -·-, wheel.
Fig. 12(a) shows the predicted vertical rail vibration velocity levels plotted against the
measured level in terms of overall A-weighted levels. The individual points represent
one of the six rail-joint conditions available. The overall trends are under-predicted,
and the mean difference between the measurements and predictions is 3.5 dB. Fig.
12(b) shows the overall wheel vibration in the axial direction plotted against the
measured results in A-weighted levels. It is found that the results vary widely, but the
mean difference is nearly equal to zero.
62 T. Kitagawa et al.
Prediction (dB)
100 90
90 80
4 Conclusion
References
[1] Ver, I.L., Ventres, C.S., Myles, M.M.: Wheel/rail noise, Part III: Impact noise generation by
wheel and rail discontinuities. Journal of Sound and Vibration 46, 395–417 (1976)
[2] Wu, T.X., Thompson, D.J.: On the impact noise generation due to a wheel passing over rail
joints. Journal of Sound and Vibration 267, 485–496 (2003)
An Explicit Integration Finite Element Method
for Impact Noise Generation at a Squat
Section of Road and Railway Engineering, Faculty of Civil Engineering and Geosciences,
Delft University of Technology,
Stevinweg 1, 2628 CN, Delft, The Netherlands
Z.Li@tudelft.nl
Summary. This paper presents a full finite element (FE) interaction model of
wheel-track to study the wheel-rail impact noise caused by a squat. The wheel,
the rail and some other track components are modeled with finite elements in
three dimensions, where necessary and appropriate. Realistic contact geometry,
including geometric irregularity (squat) in the contact surfaces is considered. The
integration is performed in the time domain with an explicit central difference
scheme. For convergence, the Courant time step condition is enforced, which,
together with the detailed modeling of the structure and continuum of the
wheel-track system, effectively guarantees that vibration frequency of 10 kHz or
higher is reproduced. By making use of the calculated velocities and pressures on
the vibrating surfaces, the boundary element method (BEM) based on Helmholtz
equation is adopted to transform the vibrations of the track into acoustic signals.
1 Introduction
2 Method
As shown in Fig. 1, a 3D finite element model was set up for the time domain
simulation of a wheel rolling over a squat with the speed v. A half of a wheelset was
modeled with a section of railway track 9.14 meters long. In [8] it was shown that the
length of track model 10 m, 20 m and 40 m may affect the vibration amplitude to some
extent but has little influence on frequency contents. A track length of 40 m is enough
to avoid effect of the end reflection on the numerical results. In this first attempt to
apply the explicit integration FE method to predict impact noise, we choose 15 sleeper
spans (9.14 m) to reduce the model size. The wheel geometry corresponds to a
passenger car wheel of the Dutch railway network with a radius of 0.46 meter; the rail is
modeled as UIC60 with an inclination of 1:40. The squat position is 0.45 m away from
the initial wheel position, and just over the edge of the sleeper where the support
stiffness varies, which has been considered as an important contributing factor for squat
development [7].
The wheel and the rail were modeled with 8-node elements. In order to achieve a
high accuracy of the solution with a reasonable model size, non-uniform meshing is
used. Fig. 2 shows two meshing schemes applied to the cross-section of the rail. The
number of elements in meshes 1 and 2 are respectively 293 and 383. The results
calculated by the two different meshes will be compared in the later analyses. The other
track parameters used in the model were taken from [9].
To apply a squat to the originally smooth surface of a rail for the simulation,
geometry of a typical severe squat was measured with RAILPROF as [8]. Since the
RAILPROF measures the longitudinal-vertical rail profile along the center line of the
rail, while the maximum deviation of the vertical irregularity probably occurs off the
center line, and cracks beneath the squat may affect the accuracy of the measurement,
modification is made for a more realistic shape of squat in the model. Fig. 3 shows the
surfaces of the rail with or without a squat, whose estimated length and depth are
respectively 40 mm and 0.3 mm.
66 Z. Yang, Z. Li, and R.P.B.J. Dollevoet
Fig. 3. Surfaces of the rail model without (left) or with (right) a squat
The Boundary Element method is employed to transform the vibration of the wheel and
rail into acoustic radiation [10]. In this case, a boundary surface enclosing a volume
(rail or wheel) is surrounded by an ideal and homogeneous fluid medium (air). In the
frequency domain, the acoustic wave propagation in an ideal fluid is governed by
Helmholtz equation, which can be discretized into a linear equation system by meshing
the vibration surface of rail and wheel. Because of the non-linear character of impact
problem, the transient vibration response of track structural is computed in the time
domain. Through the Fast Fourier Transform, velocities on the surface of the structure
are converted into a frequency response, which can be taken as boundary conditions for
the BEM. This frequency-domain method allows the calculation of sound pressure at
any observation point of the acoustic domain.
Relying on the FE Model and the noise calculation method mentioned above, the
contact force, vibration and noise emission of the modeled wheel-rail system were
obtained.
calculated by mesh 1 model. Here comparison is made in Fig. 4(a) and (b) between squat
and wheel flat because of their close similarity: they both cause impact at wheel-rail
contact due to the irregularities in the contact surfaces. The only difference is that a squat
is in rail surface and a wheel flat is in wheel surface. According to Newton’s law, the
resulting dynamic contact forces on wheel and rail are equal.
In the wavelet power spectrum (WPS) in Fig. 5, it is shown that the 1 kHz frequency
components mainly occur at the position of the squat (0.45-0.53 m) and this
phenomenon is more obvious for the contact force calculated by mesh 2. It can be
deduced that the rail mesh affects less the position of the main frequency of the contact
force but has stronger influence on the frequency content. A dense mesh is more able
to simulate high frequency components.
Fig. 5. Time history and wavelet power spectra of contact force calculated by Mesh 1 (left) and
Mesh 2 (right)
good agreement with the measured results in the time domain, especially for the case of
mesh 2. In the frequency domain, the one third octave band spectra in Fig. 6(b) indicate
that the wheel acceleration calculated by mesh 2 is in accordance with the measured
results in the frequency range below 4 kHz. Two main frequency bands around 250 Hz
and 1 kHz are accurately calculated and the acceleration levels are more or less the
same as the measurement. As for the model with mesh 1, the two main frequencies of
wheel accelerations are successfully predicted, but there exists a larger deviation in
values of acceleration level.
(a) Original (above) and filtered (below) time-histories (b) One-third octave band spectra
Fig. 6. Comparison of simulated and measured axle box acceleration
On the basis of the BEM, the vibrating rail and wheel can be regarded as the
sources of acoustic emission. The acoustic results are calculated by the normal
vibration velocities and pressures on the surface of the rail and the wheel to investigate
their noise contributions. The near-field acoustic signal radiated by the rail is plotted in
An Explicit Integration Finite Element Method for Impact Noise Generation at a Squat 69
Fig. 8 to present the wave shape of the simulated acoustic signals. The peak values of
both time-history signal in Fig. 8(a) and power spectrum in Fig. 8(b) occur at the
impact point. The sound pressure amplitude reaches up to 124 Pa; The main
frequencies of impact noise concentrate on 600 Hz and 1600-2500 Hz and its high
frequency components can reach 6 kHz or even higher.
Fig. 8. Time-history (left) and wavelet power spectrum (right) of acoustic signal collected by
receiver at 0 m
One third octave spectra of the acoustic signals collected by each receiver
contributed by the rail and the wheel are shown in Fig. 9(a) and 9(b) respectively. For
the noise radiated by the rail, the main frequency bands of the near-field signal are
around 500 Hz and 1600 Hz, corresponding to the results in Fig. 8, and the main
frequency bands of the far-field signals are around 315 Hz, 800 Hz and 1600 Hz. For
the noise contributed by the wheel, the common main frequencies of both near-field
signal and far-field signals are around 315 Hz and 3150 Hz. The one-third octave
curves of sound emitted by the wheel also present two peaks at 6.3 kHz and 10 kHz.
Since previous studies on impact noise never extended the analyzing frequency range
to more than 5 kHz, measurements are required for validation of these results.
Fig. 9. One-third octave spectra of predicted impact noise radiated by the rail (left) and the wheel
(right)
The wheel-rail contact force and the vibration of the axle box simulated by the FE
model were verified respectively by comparison with the literature and field
measurements. The good agreements of the results qualify the model for simulating the
70 Z. Yang, Z. Li, and R.P.B.J. Dollevoet
noise generated by wheel-rail impact over the squat. The influence of the rail mesh was
investigated and it indicates that the results calculated by the model with the denser
mesh has a better agreement with test results.
The BEM based on the Helmholtz equation was employed to transform the
vibrations simulated by the FE model into acoustic signals. The wavelet power spectra
of acoustic signals show that the frequency of impact noise can reach up to 6 kHz and
its main frequency ranges are concentrated on 600 Hz and 1300-2500 Hz. The
one-third octave band curves indicate that the main frequencies of noise contributed by
the rail and wheel are respectively 1500 Hz and 3150 Hz or even higher, which need to
be proved by measurements.
This is the first attempt to apply this explicit integration finite element method to
simulate impact noise generation. Due to the track access restriction, field tests could
not yet be performed. The vibration of the rail together with the impact noise
propagation need to be measured in the future to verify the simulation results. Taking
the measurements as reference, mesh optimization of the model should be further
studied. Refinement of wheel mesh is also necessary for accurate prediction. The model
will be further extended to examine the effects of the short track length on the contact
force and noise generation. Welds and insulated joints as more common discontinuity
scenarios for impact noise generation will be modeled and their impact noise emissions
will be studied on the basis of this squat impact noise simulation in future work.
References
[1] Thompson, D.J., Jones, C.J.C.: A review of the modelling of wheel/rail noise generation.
Journal of Sound and Vibration 231(3), 519–536 (2000)
[2] Vér, I.L., Ventres, C.S., Myles, M.M.: Wheel/rail noise—part III: impact noise generation
by wheel and rail discontinuities. Journal of Sound and Vibration 46(3), 395–417 (1976)
[3] Remington, P.J.: Wheel/rail squeal and impact noise: What do we know? What don’t we
know? Where do we go from here? Journal of Sound and Vibration 116(2), 339–353
(1987)
[4] Newton, S.G., Clark, R.A.: An investigation into the dynamic effects on the track of
wheelflats on railway vehicles. Journal of Mechanical Engineering Science 21(4), 287–297
(1979)
[5] Wu, T.X., Thompson, D.J.: A hybrid model for the noise generation due to railway wheel
flats. Journal of Sound and Vibration 251(1), 115–139 (2002)
[6] Pieringer, A., Kropp, W., Thompson, D.J.: Investigation of the dynamic contact filter effect
in vertical wheel/rail interaction using a 2D and a 3D non-Hertzian contact model.
Wear 271(1-2), 328–338 (2011)
[7] Li, Z., Zhao, X., Dollevoet, R.P.B.J., Molodova, M.: Differential wear and plastic
deformation as causes of squat at track local stiffness change combined with other track
short defects. Vehicle System Dynamics 46, 237–246 (2008)
[8] Molodova, M., Li, Z., Dollevoet, R.P.B.J.: Axle box acceleration: Measurement and
simulation for detection of short track defects. Wear 271(1-2), 349–356 (2011)
[9] Zhao, X., Li, Z., Liu, J.: Wheel-rail impact and the dynamic forces at discrete supports of
rails in the presence of singular rail surface defects. Proc. Instn Mech. Engrs, Part F:
Journal of Rail and Rapid Transit. 226, 124–139 (2011)
[10] Wu, T.W.: Boundary Element Acoustics: Fundamentals and computer codes. Advances in
boundary elements. WIT Press, Southampton (2000)
Future European Noise Emission Ceilings: Threat
or Solution? A Review Based on Swiss and Dutch Ceilings
1 Introduction
Despite its environmentally friendly image, rail transport encounters substantial public
opposition to noise in some European regions. The European Commission (EC)
believes that “if no remedial action is taken, this could lead to restrictions in rail freight
traffic along the most important European rail corridors” [1]. A threefold European
policy strategy has been unfolded in 2008:
1. Noise-differentiated track access charges (NDTAC);
2. Noise emission ceilings;
3. Voluntary commitments of Member States.
While the first instrument (NDTAC) received top priority, the noise emissions
ceilings are proposed as a second step. Ceilings are required “in order to maintain the
noise reduction achieved by retrofitting”, because ortherwise the desired growth of rail
freight transport would lead to more noise.
During 2013 a renewed set of rail freight noise policy options was presented by the
Directorate-General for Mobility and Transport (DG MOVE) [2, 3]. These options
cover NDTAC, subsidies for retrofitting, maximum reception levels, TSI limits for
existing rolling stock, and track maintenance requirements. Two other options, the
‘TEN-T approach’ and the ‘density approach’, make use of emission ceilings. While
the TEN-T approach restricts the ceilings to the TEN-T network of freight corridors
only, the density approach concentrates on areas with high population densities.
possible to detect locations where noise has been growing and action may be required.
According to the EC, it would be a task for the rail sector “to find optimal solutions” to
avoid exceeding of the ceiling. Thus, a system of noise emission ceilings is a trigger for
action and an incentive for (possible) source measures. It is not meant as an alternative
for the existing national noise legislation.
90 dB 68 dB 54 dB 47 dB
3.1 Switzerland
The Swiss abatement programme, started in 2000 is based on emission ceilings for the
railway network, called Emission Plan 2015. Distinct ceilings are defined at about
6 600 track sections of variable length (between 1 meter and 20 km) along the entire
railway network (3700 km). The Swiss ceilings are monitored by a combination of
measurements and calculations [8]. The measurements, carried out continuously with
six fixed systems and a few mobile systems, see Fig. 3, mainly aim at monitoring the
progress of the vehicle retrofitting programme. Additional calculations are required to
check if the actual noise emission on the whole network is still compliant with
Emission Plan 2015. This task has not yet been completed. A huge noise abatement
programme, with façade insulation and noise barriers, was based on the predicted noise
situation of 2015. This programme is nearly finalized.
If the ceilings are exceeded in the long term, the infrastructure manager is
responsible for further noise abatement. The track section under consideration is then
treated as a track being upgraded and (possible) noise measures are taken near
dwellings in accordance with the legal cost-benefit scheme. The legal framework for
this intervention is laid down in article 37a of the Swiss Ordinance on Noise Protection
[9]. At present, the ceilings of Emission Plan 2015 are not restrictive for railway
operations. On most locations the measured levels are a few decibels below the ceiling
levels, providing headroom for traffic growth.
74 E. Verheijen and F.B.J. Elbers
Fixed system
Mobile system
Fig. 3. Fixed and mobile measurement sites on the Swiss railway network
Dutch
calculation point
50 m
1m
When the ceilings were adopted in legislation on July 1st, 2012, the ceilings were
fixed at 1.5 dB above the current1 noise level at each reference point. In some cases the
ceilings were fixed at the situation expected in the near future. The actual noise impact
at these points is monitored every year by calculation. If ceilings are exceeded or if a
permanent capacity increase is expected, the track section under consideration is
treated as a track being upgraded and (possible) noise measures are taken near
dwellings in accordance with the (already existing) national cost-benefit scheme.
Source measures like rail dampers and acoustical grinding are stimulated by the fact
that the procedural workload is less heavy compared to noise barriers.
1
The ‘current noise level’ has been defined as the average level over 2006, 2007 and 2008.
Future European Noise Emission Ceilings: Threat or Solution? 75
Fig. 5. Reference points at 50 m from the outer tracks near Utrecht station (NL)
The initial height of the ceilings, featuring a headroom of 1.5 dB, has been chosen
after a number of assessment studies between 2005 and 2008. The assessments
concentrated on long-term and short-term bearability of different headroom settings.
The long-term effects refer to financial aspects for measures and incentives for
retrofitting. The costs of the ceiling system in the long run are compared to the base
system (no change of legislation). The short-term effects relate to operational
consequences (network capacity, annual time-table).
Both the noise ceiling level and the actual annual level are calculated in a
computation model where buildings and reflective structures other than noise barriers
are left out deliberately. This is done for technical and administrative reasons. For the
purpose of triggering noise growth this simplification is allowed.
The ceilings are not inflexible. If traffic grows, the local ceiling level can be
increased, in case noise reduction measures are less cost-effective than façade
insulation, or decreased, after noise barriers have been installed. Naturally, the
exceedance of ceilings in areas without any dwellings does not require noise measures
and is simply taken into account by increasing the ceiling height.
The development of the ceiling system in the Netherlands took about 15 years from
initial idea to legislation. After the decision was made to implement the system (around
2006), the remaining time was used to acquire more accurate traffic and superstructure
data, to develop software tools, to set up a ceiling administration, to implement
independent inspection and to reform the infrastructure manager’s process of traffic
capacity allocation.
3.3 Comparison
The Swiss ceilings are defined in a fairly simple way. Ceilings are guarded using
measurement stations, but the main purpose of these stations is to obtain an overview of
the progress on retrofitting noisy wagons. The Dutch ceilings are very detailed and
serve as a warning system: they ensure that local reception limits will not be exceeded
without notice.
76 E. Verheijen and F.B.J. Elbers
5 Conclusions
Two countries have developed a concept of noise emission ceilings. These countries do
so in a very different way, but there are also similarities. The EC idea of noise emission
ceilings has been based on these examples. The Swiss ceilings are defined in a fairly
simple way. Ceilings are guarded using measurement stations, but the main purpose of
these stations is to obtain an overview of the progress on retrofitting noisy wagons. The
Dutch ceilings are very detailed and serve as a warning system: they ensure that local
reception limits will not be exceeded without notice.
The European policy instrument of noise emission ceilings for the railways does not
necessarily interfere with existing national noise legislation. Because the concept is not
yet very specific, it is currently not possible to take a position. However, if ceilings are
introduced as an incentive for retrofitting freight wagons, there is a short term need for
implementation.
A system of noise emission ceilings can be supportive to keep railway traffic viable
and competitive in the future, but the Dutch case shows that the workload to set up a
very detailed system can be considerable. The Swiss case shows that if ceilings are
combined with noise abatement, the cost can be huge. The advantage for the line side
residents is that ceilings prevent railway noise from growing continuously. The
78 E. Verheijen and F.B.J. Elbers
advantage for the railways is that they should be able to (partly) use the additional noise
capacity that is released by retrofitting with LL blocks, for future growth. The
advantage for the government could be a reduction of cost for track-side noise
measures. For this to come true, the ingredients that make up the ceiling system must be
balanced properly.
This article is based on the findings of an assessment study on ‘Bearable Noise
Limits’ commission by the International Union of Railways (UIC), conducted in 2010
and 2011 [11].
References
[1] European Commission, Rail noise abatement measures addressing the existing fleet,
COM, 432 final (2008)
[2] Effective reduction of noise generated by rail freight wagons in the European Union,
Roadmap of DG MOVE (April 2013)
[3] Public and Stakeholder Consultation by DG MOVE (July-October 2013)
[4] Elbers, F.B.J., Verheijen, E.: Bearable railway noise limits in Europe. In: Nielsen, J.C.O.,
Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D., Tielkes, T., Towers,
D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems.
NNFM, vol. 126, pp. 13–19. Springer, Heidelberg (2015)
[5] Guidelines for Community Noise. World Health Organization, Geneva (1999)
[6] Night Noise Guidelines for Europe. World Health Organization, Geneva (2009)
[7] Final Report on Task 3, Impact Assessment and Proposal of Action Plan, Study on the
implementation of the END, Milieu-RPA-TNO (May 2010)
[8] Attinger, R.: Monitoring railway noise in Switzerland from 2003 to 2010. European Rail
Technology Review 50(3) (September 2010)
[9] Lärmschutz-Verordnung, SR 814.14/LSV
[10] Lärmsanierung der Eisenbahnen, Standbericht, Bundesamt für Verkehr (2012)
[11] Exploring bearable noise limits and emission ceilings for the railways - Part I, ref. no.
UIC001-01-016 (October 25, 2011), http://www.uic.org/IMG/
pdf/p250_bearable_noise_limits_part_i_final.pdf
Comparison between Road and Rail Noise Cost
per Transported Ton of Cargo
1 Introduction
The social cost of noise is normally used when performing benefit-cost analysis (BCA)
while planning infrastructure projects or noise mitigation measures. There are many
national models for determining the yearly cost of being exposed at home to a certain
noise level, and they typically assign a valuation that gives the monetary effect as a
function of the equivalent noise level in dB at the exposed façade of the dwelling. In
this study we have used the official noise values in use in Sweden for road and railway
traffic (ASEK 5) [1] together with the official noise prediction methods [2-3] to
estimate the acoustic and monetary impact of transporting 1 ton of cargo through two
example transportation corridors.
2 Method
The method employed in this paper is described in detail in [4], but the basic steps are
to calculate the noise level at each residence using the standardized Nordic prediction
methods for road and railway traffic noise [2-3], and then calculate the marginal
increase in noise level one additional train/truck passage causes. The short run marginal
cost (SRMC) can then be calculated using the official valuation functions by adding up
the contribution for each residence along the road or railway segment under study.
The main difficulty with applying the method in practice is the demand for accurate
input data, not only on the location of all residences close to the road or railway line and
the total traffic flows including speeds, but also on all potentially screening objects
such as terrain, buildings, noise barriers and so on. In this study we have chosen to
simplify the input data as described in Table 1. The data on the infrastructure was
supplied by the Swedish Traffic Administration (Trafikverket) and the Swedish
National Road Database (NVDB). The population data was supplied by Statistics
Sweden (SCB).
For each population square (with 250 m side) the population was assumed to be
uniformly distributed over the whole area of the square, except that no population was
assumed closer to the source than 25 m. No detailed data on ground attenuation, noise
barriers and screening by terrain was available so instead a simplified method was used
where the inhabitants in each square were assumed to be distributed over both directly
exposed positions and other screened positions. The distribution function was modeled
Fig. 1. Official Swedish valuation functions for traffic noise (from ASEK 5 [1])
Comparison between Road and Rail Noise Cost per Transported Ton of Cargo 81
after detailed calculations in more than 1,200 receiver positions in five different
railway exposed areas and three road traffic noise exposed areas in Sweden [5]. The
accuracy of the simplification is hard to estimate, but a crude estimation using two
reference areas not used for the fitting of the model with approximately 100 receivers
gives an uncertainty of ±35 % on the calculated SRMC.
The reference train was assumed to be a 500 m long electrically powered train set
with 25 wagons and a total load of 1,500 metric tons. The reference truck was assumed
to be a diesel powered vehicle with a total load of 42 metric tons. Both were assumed to
travel at 90 km/h where possible, and at the maximum allowable speed if it was lower
than 90 km/h.
The valuation functions from ASEK 5 [1] give the yearly cost in SEK (1 SEK is
approximately 0.15 USD or 0.12 EUR, June 2013) per person exposed as a function of
the equivalent 24h noise level on the façade of the dwelling. The valuation functions for
road and railway traffic noise are shown in Fig. 1. For lower equivalent levels the rail
function gives lower values, but the relation is reversed above 69 dB. Note that
normally many more inhabitants are exposed to lower levels since the higher levels
only can be found very close to the road or railway.
3 Results
The calculated SRMC in SEK (price level 2010) is illustrated for each km between
Gothenburg and Stockholm in the map in Fig. 2. The area of each circle is proportional
Fig. 2. Map of the marginal cost of transporting 1 ton of cargo between Gothenburg and
Stockholm on railway (upper line, light gray) or truck (lower line, dark gray). The area of the
circle is proportional to the SRMC.
82 M. Ögren, H. Andersson, and T. Jerson
to the SRMC or transporting 1 ton of cargo for a 1 km section of the road or railway.
The total distance is 3 % longer for the road than for the railway, see Table 2. Note that
the SRMC is close to zero in rural areas, and it increases sharply where the population
density is higher. For areas where no inhabitants live closer to the source than 1 km the
SRMC is identical to zero.
The results are also given in Table 2 and Fig. 3. The average marginal cost is
determined where the cost is high, i.e. more densely populated areas, which gives a
skewed distribution where a big part of the road/railway network gives almost no
contribution. The number of inhabitants within 1 km of the source is about equal for
both corridors, but note that more inhabitants live closer to the railway. This is probably
due to the central location of the railway in many of the urban areas it is passing
through, which is in contrast to the road which often circles around the urban centers.
Length
471 458
[km]
Average traffic
27 500 122
24h
Population
409 000 400 000
within 1km
Population
61 000 88 000
within 250 m
Noise emission
SEL at 25 m
71.9 75.1
and 90 km/h
[dB/ton]
SRMC
1 vehicle 0.13 10
[SEK/km]
SRMC
1 ton 0.0031 0.0069
[SEK/tonkm]
SRMC 1 ton
median (max) 0.0002 (0.08) 0.0005 (0.20)
[SEK/tonkm]
SRMC low
noise 1 ton 0.00098 0.00069
[SEK/tonkm]
The last line in Table 2 gives the result for a new calculation assuming a low noise
variant of the vehicles. The freight train is assumed to be retrofitted with composite
brake blocks (K-blocks) for an overall improvement of 10 dB, see [6]. The truck is
Comparison between Road and Rail Noise Cost per Transported Ton of Cargo 83
assumed to be equipped with low noise mufflers and tires, giving an improvement of
5 dB according to [7]. For these low noise variants the railway transport has less
marginal cost than road transport, showing the important potential for improvement of
noise emission of freight wagons.
Fig. 3. SRMC for transporting 1 ton of cargo between Gothenburg and Stockholm. Upper picture is
cost in SEK/tonkm, lower is accumulated cost in SEK/ton when starting from Gothenburg.
4 Discussion
The average marginal cost is approximately twice as high for rail transport of 1 ton of
cargo. This is perhaps surprising considering that the valuation function gives a lower
value for railway noise, but it turns out that the higher noise emission per transported
ton together with the higher number of inhabitants close to the railway as compared to
the road more than compensates for the lower valuation. The potential for reduction is
demonstrated by the fact that railway transport gives lower marginal costs per ton if
84 M. Ögren, H. Andersson, and T. Jerson
K-blocks are used. If a railway noise charge scheme is implemented where brake type
is taken into account it would be a strong incentive for freight vehicle operators to
retrofit their train sets using low noise technology. The same is of course true for truck
transport.
References
[1] Norlin, E., Grudemo, S., et al.: Samhällsekonomiska principer och kalkylvärden för
transportsektorn: ASEK 5 (Swedish Transport Administration) (2012)
[2] Ringheim, M., et al.: Railway Traffic Noise – Nordic Prediction Method. TemaNord
1996:524, Nordic Council of Ministers, Copenhagen, Denmark (1996) ISBN
92-9120-837-X
[3] Jonasson, H., Kragh, J., et al.: Road traffic noise – Nordic Prediction method. TemaNord
1996:525, Nordic Council of Ministers, Copenhagen, Denmark (1996) ISBN
92-9120-836-1
[4] Andersson, H., Ögren, M.: Charging the polluters, a pricing model for road and railway
noise. Journal of Transport Economics and Policy 47(2) (2013)
[5] Ögren, M., Jerson, T.: Mätning och beräkning av buller och vibrationer från tågtrafik inom
TVANE-projektet
[6] Oertli, J., Hübner, P.: Railway noise in Europe – a 2010 report on the state of the art,
International Union of Railways (UIC) (2010)
[7] Sandberg, U., Ejsmont, J.: Tyre/road noise reference book. Informex (2002) ISBN
91-631-2610-9
A Survey of Freight Locomotive Passby Noise Emissions
1 Introduction
Noise from rail freight is a particular concern for communities adjacent to rail freight
lines in New South Wales (NSW). Noisy locomotives are often identified in
community complaints about rail noise.
In NSW the noise emissions from locomotives with existing approval to operate on
the network are not currently restricted. The noise emissions from locomotives vary
widely across the operating fleet, which includes many different classes of
locomotives.
2.1 Objectives
The primary objective of this survey was to improve the understanding of the
environmental noise impacts of freight locomotives, in particular the statistical spread
of locomotive noise, and the relative noise emissions under different operating
conditions. A further aim was to conduct a subjective noise assessment to investigate
whether the presence of particular sound characteristics might be as important as
maximum noise level parameters in identifying locomotives that may be considered to
be particularly noisy or annoying.
2.2 Measurements
Five passby measurement locations were identified on the RailCorp network (Table 1).
The locations were selected to include locations near straight track on relatively steep
grades, as well as straight and level track. Curves were excluded. Audio data was
recorded at each location for a minimum of a week for post-processing, with
microphones located at 7.5 m from the near track centreline, at a height of 1.5 m above
top of rail.
Subjective Description
Score
1 Locomotive noise mostly inaudible below wheel-rail rolling noise
2 Locomotive audible at low levels
3 Locomotive noise noticeably above rolling noise
4 High locomotive noise, but no particularly annoying characteristics
5 High levels of locomotive noise with annoying characteristic
6+ Outlier event, noticeably subjectively worse than the majority
3 Results
A total of 607 passby events were examined. 31 % of all identified events were
mixed-class consists. The two most common individual locomotive classes made up
88 B.E. Croft et al.
13 % and 12 % of all events. Two further individual classes made up 6 % and 5 % of the
total. The remaining events were other less numerous locomotive classes, or
locomotives that could not be identified. Overall more than 95 % of events on the main
lines were identified.
Table 3. Number of measurements, speeds and subjective score by location and track
The survey results indicate that the 95th percentile LAmax noise level parameter does
not necessarily increase with increasing uphill grade. At L5 (a level location), this
parameter was similar to or greater than the level recorded in the uphill direction (Down
track) at L1, reflecting the dependence of A-weighted maximum noise levels on the
speed of the passby. The mix of locomotive classes at any particular location is also a
factor.
At the three locations on steep grades, the 50th percentile LCmax levels in the uphill
direction were within 1 dB. This parameter was 5 dB to 6 dB lower at the two relatively
level measurement locations. More variation was observed in the 95th percentile LCmax
levels at uphill locations.
As an example of the distribution of noise levels, Fig. 1 shows the measured LAmax
noise levels at L1 and Fig. 2 shows the measured LCmax noise levels at the same location.
It can be seen that the LAmax parameter does not correlate with the subjectively assessed
“outlier” passbys. The LCmax parameter is a much better indicator of outlier events at
L1. The LCmax parameter was observed to correspond reasonably well with subjectively
assessed outlier events, in both the uphill and downhill directions at all locations.
4 Concluding Remarks
Across all the passby measurement locations and directions, the measurements
displayed a range of at least 10 dB in each of the parameters examined. A range of more
than 20 dB in a noise parameter was observed in many cases. This range of results
reflects the variation in noise emissions between locomotive classes, operating at a
range of speeds and in different notch settings. Even at uphill locations where freight
locomotives are presumed to be generally under maximum load, the variation between
different events can be as much as 10 dB.
This range of measured levels from freight passbys is important to consider in
undertaking measurements at a particular location, for example with a view to
validating noise models. A small sample size is potentially not sufficient to characterise
the locomotive noise impacts at any one location.
The investigation of perceived subjective noise impacts indicates that subjective
outliers do not always correlate well to the A-weighted maximum noise level. This
parameter is sometimes elevated by short duration events such as flanging or wheel
flats. If the locomotive is travelling fast in a low notch setting the LAmax level increases
with speed, whereas if the locomotive is travelling slower or in high notch settings the
speed is less of a factor.
Subjective outlier events are better correlated to the C-weighted maximum noise
level. This parameter was found to be a reasonably good indicator of passby events that
would be classified as “subjective outliers”. However, even this parameter does not
capture all subjective outliers. Some locomotive passby events were observed to have a
particularly noticeable low frequency, “throaty” character that stands out from other
passbys. These events can generally be attributable to specific locomotive classes, and
are noticeable even in cases when the A-weighted and C-weighted maximum noise
levels during the passby are relatively low. Further work is required to investigate
sound quality metrics that could be used to better capture the subjective response of the
community to locomotive noise.
References
[1] Walters, C. et al.: A guide to Australian locomotion: a guide book to diesel and electric
locomotives running in commercial service today (2009 ed). Australian Railway Historical
Society (New South Wales Division), Redfern, NSW (2008)
[2] Environment Protection Authority NSW - EPL12208 (Licence version date: July 6, 2012)
[3] Environment Protection Authority NSW - EPL3142 (Licence version date: October 25,
2012)
On the Importance of Accuracy of Geographic Model
Data for Noise Impact Studies
Summary. Since 1 July 2012, ProRail has to work with a new legal system of
noise production ceilings (NPCs) to control the noise impact of the railway traffic
on its network. The new system requires that ProRail computes the noise impact
of the yearly traffic on about 60 000 reference points. In addition, ProRail has to
demonstrate that the noise does not exceed the NPC at any point.
ProRail has to compute the noise level at each reference point using the Dutch
computation method SRMII (which is the recommended European interim
computation method for railway noise). This model has to be updated yearly to
reflect all changes in the network and surroundings. The data for the model
comes from different sources. It is possible that small changes in the model are
introduced by small inaccuracies in the data collection process. A source of small
changes is the collection of geographic data with photogrammetric
measurements with its inherent measurement and processing (in)accuracy.
To investigate the sensitivity of the calculations to small data collection
inaccuracies, we have done an extensive parameter study. We performed
parameter variations in the national-scale model and analyzed the resulting
change in noise level on a statistical basis. This paper presents the results of this
study and shows that the accuracy of the height information is crucial.
1 Introduction
Since 1 July 2012, the Dutch noise legislation has changed conceptually. There is a
new law that enforces that the infrastructure managers have to prove that they comply
with the legal limits, not only when they build or change the infrastructure but also,
they have to demonstrate, on a yearly basis, that the actual noise production is within
legal limits [1].
The legal limit is called the noise production ceiling (NPC) and is enforced at a
collection of evaluation points (known as reference points) along the infrastructure.
The actual noise level is not controlled by measurements but determined by noise
calculations using the yearly averaged traffic and actual state (with inherent acoustic
properties) of the infrastructure. The systematics of NPCs is based on calculations,
because noise calculations are considered more robust and cost-efficient than noise
measurements. As a consequence, the infrastructure manager needs to have an accurate
calculation model for its entire network and it needs to maintain and update this model
to be able to check the NPCs yearly.
The necessity to build and maintain a calculation model for several years brings new
challenges. One of them is how to deal with model updates if the source data for the
existing infrastructure changes. When the model of existing infrastructure or
surroundings is updated, the calculated noise levels should not change because nothing
has changed in the real world. In practice, there will always be small changes due to
inaccuracies in the data collection process. The question is: what accuracy for the input
data is required to ensure that the noise level at the reference points does not change due
to modeling artifacts? To find an answer to this question we have done a
comprehensive study on the sensitivity of the calculated noise levels due to small
changes in the input data. We have limited the study to geographical data of
infrastructure and surroundings since preliminary studies showed that those data were
likely to have largest influence.
For railways, ProRail is the responsible organization to ensure that the noise emission
on all national railways remains within the legal limits enforced by the NPCs. This
implicates the need of a model of their complete Dutch network (over 3 000 km of
track) and close surroundings (without buildings). Furthermore, calculation software
is required, based on the SRMII calculation method, to be able to check the noise
level on the network of about 60 000 reference points.
The calculation model for the NPC calculations consists of three parts: 1) the
geographical data for track and infrastructure elements (bridges, tunnels, noise screens
etc.) 2) the traffic data (trains, routes, timetable, speed profile etc.) and 3) track
properties (track type, rail roughness, and steel bridge emissions). The data on which
the model is based are maintained by ProRail.
2.2 Updating
The fact that ProRail has to check the NPCs has led to new requirements for model
accuracy that were not relevant before. In the past, an acoustic consultant would make
a model to the best of his or her knowledge and would calculate the noise immission
and noise control measures using that model. Then, the noise control measures would
be implemented and the model would cease to be relevant. The noise control measure
is what would remain for the future.
In the ‘NPC world’, not only the noise measures remain, but also the NPC limit
values that are enforced by law. And this is where difficulties may arise. The law
requires ProRail to update the model every year to reflect the changes that have been
On the Importance of Accuracy of Geographic Model Data for Noise Impact Studies 95
made to the track infrastructure in real life. These changes may of course lead to
changes in the noise immision at the reference points and may require that new noise
reduction measures have to be taken.
However, there is another source of changes to the model: each year, the geographic
information department of ProRail updates (part of) their GIS model of the track and its
surroundings. Each year, this GIS model might be slightly different due to the fact that
the geographic data acquisition has a limited accuracy (in the order of 0.1-0.5 meters
for photogrammetric measurements). This means that, although nothing has changed in
the outside world, the model used to check the NPCs can have ‘small’ changes which
may lead to changes in the computed noise immission values at the reference points. Of
course these modeling uncertainties should not lead to the necessity to take new noise
reduction measures because nothing has changed in practice.
Clearly, ProRail wants to know to what extent the yearly model updating process
would lead to unwanted changes in the noise immision on the reference points. We
therefore devised a parameter study to investigate the impact of model changes to the
computed noise immission.
The impact of small changes in the model data is different for the various input data
types:
• Discrete data (such as train type, track type, speed profile, routes) may have a large
influence but there is no inherent inaccuracy in the acquisition of these data: when
they are stored correctly and do not change during the year, then they have no
impact on the calculations from year to year;
• Proportional data (such as number of vehicles, rail roughness) have a proportional
influence: a small change of these data leads to a proportionally small change in the
calculated level at the reference points (e.g. a 10 % error in the number of vehicles
will typically lead to a 0.4 dB change in noise immission). But, as for discrete data,
there is no inherent inaccuracy in the acquisition of these data;
• Geographic (such as position of track and all infrastructure elements) have a
non-proportional influence on the noise calculations: a small change of e.g. 10 cm
noise barrier height can have an impact on the calculated noise level between 0 and
say 5 dB, depending on the relative position of track, noise screen and receiver. So
small geographic changes due to the acquisition process may lead to a fictitious
transgression of the NPC.
This means that for ProRail, the first priority is to control the accuracy of the
geographic input data.
To investigate the impact of geographic changes we have made a parameter study. For
this study, a typical approach would be to define a limited number of typical model
configurations and vary the geographic position of the model components (e.g. tracks
96 A.H.W.M. Kuijpers, M.S. Roovers, and G.H. Groenveld
position in X, Y, Z, noise barrier height and distance from track, small variations in the
terrain model, horizontal and vertical position of bridges etc.). This provides a
well-controlled environment for the parameter study, but has some disadvantages:
1. It requires a lot of model building and separate calculations for each parameter
variation;
2. A lot of care has to be taken to ensure model consistency when individual model
components are varied. For example by changing the position of the bridge surface
with respect to the track, the track may be off or under the bridge surface, which is
physically impossible. Or by changing a noise screen position, it may get
unrealistically close to the track. So these studies may lead to false conclusions.
These consistency rules imply that the parameter variation models need to be
checked by hand. A time-consuming task;
3. Even with consistent models, the question arises: to what extent are these ‘typical’
model configurations representative for the actual model covering the whole of the
Netherlands? Do we have the worst cases or are we underestimating the problem?
To overcome these disadvantages, we chose a different approach: we chose not to vary
the model components but to vary the position of the reference points on which the
immision level is computed. Here we use a kind of ‘geographic reciprocity’ principle.
You can bring the track closer to the receiver or you can bring the receiver closer to the
track. Both will lead to approximately the same change in noise immision level. The big
advantage of changing the receiver and not the source is that the model consistency
remains unaffected. And since no manual check is necessary, the variations can
automated so it takes less time to do the parameter studies. The automation makes it
possible to do the parameter study on the complete model of the Netherlands, so
representativity is guaranteed.
The first step in the study was to vary this position in three orthogonal directions:
perpendicular (R) to the track, parallel to the track (P) and in height (Z) (see Fig. 1). The
reference points for the NPC calculations are placed at 50 m from the nearest track, at a
height of 4 m above the ground level. For each of the reference points we calculated the
change in computed sound level compared with the original computed value.
3 Results
R-100cm R+100cm
60 60
percentage [%]
percentage [%]
40 40
20 20
0 0
-1 0 1 -1 0 1
level difference [dB] level difference [dB]
Fig. 2. Computed noise level difference for a distance change minus/plus 100 cm
Changing the position of noise screens, platforms and bridges along the track is
equivalent to changing the position of the reference points parallel to the track. The
reference points are positioned at 50 m from the track and have a mutual distance of
100 m. We changed this position by plus and minus 1 m. This showed a symmetric
noise level change around the average value of 0 dB and a sensitivity of 0.0018 dB/m
(see Fig. 3).
P-100cm P+100cm
80 80
percentage [%]
percentage [%]
60 60
40 40
20 20
0 0
-1 0 1 -1 0 1
level difference [dB] level difference [dB]
Fig. 3. Computed noise level difference for a parallel position change minus/plus 100 cm
98 A.H.W.M. Kuijpers, M.S. Roovers, and G.H. Groenveld
Changing the position of the track and the height of the noise screening objects (noise
barriers, platform walls, bridges etc.) can be simulated by changing the vertical
position of the reference points. But a change of 1 m in the position of a reference
point may or may not be equivalent to a 1 m change of track position. It all depends
on the relative position of source and receiver and whether or not there are objects in
the line of sight from source to receiver. In general a displacement of the receiver of
<x> m in vertical direction is equivalent to a change of track vertical position of
<x> m when there are no objects in the line of sight, or the objects close to the track
move in vertical direction with the same amount as the track (e.g. a screen will be
defined with a height relative to the track, so if the track is raised, the top of the
screen will be raised with the same amount in acoustic models).
To assess the sensitivity of the computed noise levels to vertical position, we have
done four parameter variations: Z-100 cm, Z+100 cm, Z+50 cm, and Z+10 cm. The
histograms of the level differences are displayed in Fig. 4. The parameter study shows
that the noise level on average will change by 0.7 dB/m.
Z-100cm Z+100cm
60 60
percentage [%]
percentage [%]
40 40
20 20
0 0
-1 0 1 -1 0 1
level difference [dB] level difference [dB]
Z+50cm Z+10cm
60 60
percentage [%]
percentage [%]
40 40
20 20
0 0
-1 0 1 -1 0 1
level difference [dB] level difference [dB]
In the analyses above we make no distinction between reference points that are
behind noise screens (about 9% of the total) and reference points that have no screen
between track and reference point. If we separate these two groups and determine the
sensitivity we see that the reference points behind screens are more affected by the
vertical position change: 1.2 dB/m for reference points behind screens and 0.64 dB/m
for reference points without a screen immediately between source and receiver.
On the Importance of Accuracy of Geographic Model Data for Noise Impact Studies 99
percentage [%]
percentage [%]
40 40
20 20
0 0
-2 -1 0 1 2 -2 -1 0 1 2
level difference [dB] level difference [dB]
Fig. 5. Computed noise level difference for receivers for vertical position change of 1 m without
and with noise screen in between track and receiver
To further investigate the influence of screens on the sensitivity, we have made another
parameter study where we varied the screen height with 10, 20 and 50 cm and analyzed
the computed noise level change. Of course, in most cases, there will be no sound level
difference because in 91% of the cases, there is no screen between track and reference
point. If we leave out the category with a 0 dB change, we get the histograms of Fig. 6.
Fig. 6. Computed noise level difference with an increase of screen height of 50, 20 and 10 cm
We found that the sensitivity is rather large and does not resemble a normal
distribution. When we separately assess the sensitivity for reference points behind
screens we find a value of: +3.1 dB/m. This implies that an ‘artificial’ exceeding of
the NPC is most likely at reference points where the screen height is acquired
independently of the source (track) height. The recommended solution is to always
relate the screen height relative to the track height. The track and screen couple will
move as a rigid system with respect to the reference point. In that case, the sensitivity
is reduced to +1.2 dB/m (see section 3.3). This solution implies that absolute screen
heights need to be recalculated every time that the track height is changed in the
model updating.
We have used the GUM-method [4] to obtain a value for the combined sensitivity due
to geographic uncertainty, where we have assumed that the positional uncertainty is
100 A.H.W.M. Kuijpers, M.S. Roovers, and G.H. Groenveld
independent in r, p, and z direction and that the screen height uncertainty is eliminated
by relating the screen to the track height. In that case the total sensitivity is a defined as:
2
⎛ ∂f ⎞ ⎛ ∂f ⎞ ⎛ ∂f ⎞
2 2
ut2otal = ⎜ ⎟ u 2 ( r ) + ⎜ ⎟ u 2 ( p ) + ⎜ ⎟ u 2 ( z ) (3.1)
∂
⎝ ⎠r ∂
⎝ ⎠p ⎝ ∂z ⎠
If we assume a standard uncertainty for the positional data of 0.1 m, then we get a
combined uncertainty as shown in Table 1.
4 Conclusion
Computed noise levels at the reference points for NPCs are not very sensitive to small
variations of the geographic position in the horizontal plane. However, the computed
noise levels are rather sensitive to the accuracy of the vertical position of the model
components, especially for noise screen height and/or the position of the track close to
noise screens. The high sensitivity of the results for changes in the screen height can
be eliminated by always relating the top of the screen to the actual track height, when
building the computation model.
Using this approach we have estimated the standard uncertainty in the
computations to be 0.065 or 0.12 dB for reference points without and with screen
respectively.
References
[1] Wet Milieubeheer: article 11.20 (2012),
http://wetten.overheid.nl/BWBR0003245/Hoofdstuk11/Titel113/Af
deling1132/11321/Artikel1120
[2] Dutch ministry of infrastructure and environment (I&M): Dutch computation method 2012
(2012), http://wetten.overheid.nl/BWBR0031722
[3] Kuijpers, A.H.W.M.: Sensitivity of NPC calculations for geographic inaccuracy, report
M+P.RAIL.13.01.2 (July 2013)
[4] JCGM, Evaluation of measurement data — Guide to the expression of uncertainty in
measurement (GUM), JCGM 100:2008 (2008),
http://www.bipm.org/utils/common/documents/
jcgm/JCGM_100_2008_E.pdf
The Efficiency of Noise Reduction Measures on Railway
Infrastructure in Normal Operating Conditions -
NOVIBRAIL
1 Introduction
The NOVIBRAIL project “Noise and Vibration in the Railway system" was
established with the support of the TAČR - Technology Agency of the Czech Republic
and is registered under number TA 1031267. The program TAČR - Support Applied
Research and Experimental Development ALFA, PP3 - Sustainable development of
transport, C31-Reducing the negative impacts of transport on the environment,
including increased efficiency drives the development of new propulsion systems and
fuels, C33-Reducing the negative impacts of infrastructure projects on the
environment.
One focus was mainly on the working life of rail absorbers (chapter 2) like
continuation former project, the noise emissions from the rail brakes (chapter 3)
exceeded on the more types of the rail brakes and different types of the protection. The
possibility of the noise reduction on platforms (chapter 4) is a completely new solution
[1].
A test section was retrofitted, with new UIC 60 rail with spring fasteners on the original
concrete sleepers. Rail absorbers were installed on the rails (different types on each
track) and the track geometry was repaired.
The measurement campaign was realized on the line Kolín – Poděbrady where two
types of rail absorbers were installed (see Fig. 1). The measurement positions were
according to EN ISO 3095 [2].
In Tables 1 and 2 the test results from the measurement campaigns are shown, before
retrofitting, after retrofitting, after implementation of the rail absorbers [3], [4] and
after 1.5 years of operation.
The results are energetic averages from all trains of each category normalized to
80 km/h. All trains were divided into the following categories.
Category 1 - Passenger and express trains equipped cast iron brake blocs only
Category 2 - Passenger and express trains equipped with tread and disc brake
Category 3 - EMU equipped with disc brake only (design series 471)
Category 4 - Freight trains
Category 5 - Motor Cars
The results show nearly the same reduction of noise emission (6.5 dB average
together with retrofitting of the track) 2.5 dB for absorbers only immediately after
installation.
Table 1. Test results from the first track (rail absorbers type a)
After
Before After 1.5 year
Category of train After retrofitting implementation
retrofitting operation
absorbers
Cat. 1 and 2 93.5 90.4 88.9 90.5
Cat. 3 87.1 82.9 81.2 81.8
Cat. 4 95.2 91.5 91.1 89.6
Cat. 5 95.4 94.0 91.6 93.6
The rail absorbers type a lost some of their damping properties, mainly for category
1.2 and 5 after one and a half year operation.
The Efficiency of Noise Reduction Measures on Railway Infrastructure 103
Table 2. Test results from the second track (rail absorbers type b)
After
Before After 1.5 year
Category of train After retrofitting implementation
retrofitting operation
absorbers
Cat. 1 and 2 92.8 88.9 87.8 85.3
Cat. 3 87.7 80.1 78.3 77.5
Cat. 4 97.1 93.3 90.8 90.8
Cat. 5 97.5 93.0 92.8 88.5
Rail absorbers type b did not lose their damping properties. The final results after
one-and-a-half-years of operation were even lower than immediately after installation.
The reason for this can probably be attributed to an increasing amount of vehicles
equipped with disc brakes and a proper run-in rail head surface.
Fig. 2 shows the level of acoustic roughness on both tracks on the measuring site
measured after 1.5 year operation.
The track contribution from the acoustic roughness point of view is nearly the same
on the both tracks and is nearly in the reference test section quality.
The noise from marshalling yards is the main issue in recent years. Especially noise
from the rail brakes is very annoying, because includes very strong tonal components
(rail brake squeal). In the frame of this project an attempt was made to map different
types of hump yard and the efficiency of some implementations against noise emission.
104 J. Hlaváček, L. Hejzlar, and R. Kolmačka
It was anticipated that the noise reduction equipment might reduce the rail brake squeal
by up to 60 dB by special lubrication on the wheel thread. Three measuring campaigns
were realized: before and after installation of the equipment and finally after complete
renovation of rail brakes [5, 6].
In Fig. 3 a simple drawing of the rail brakes with the measurement positions is given.
In Table 3 the test results from the measurement campaigns are given, before
reconstruction, with and without lubrication and after reconstruction, with and without
lubrication.
Results are the energetic mean value in dB(A) from all wagons for the transit time
from the relevant rail brake (7.5 m measurement distance) [2].
Before After
Before After
Measurement point reconstruction + reconstruction +
reconstruction reconstruction
lubrication lubrication
M1 113.3 111.3 113.0 107.8
M2 119.2 113.0 117.6 110.8
The spring-hydraulic rail brakes use hydraulic pumps for the braking power. The
compressor station is no longer necessary and this additional noise source is absent.
The braking squeal is present too.
In Fig. 4 a simple drawing of the rail brakes is shown with the measurement
positions.
In this case a low noise barrier of the height 1.2 m (above rail surface) with
absorbing layer was used between the rail brakes and measurement position M2.
Results are the energetic mean value in dB(A) from all wagons for the transit time
from the rail brake (7.5 m measurement distance) [2].
Measurement point without low noise barrier with low noise barrier
M1 112.0 -
M2 111.7 97.2
This kind of the rail brake works on the principle of a self-locking braking mechanism
and the brake with self-regulation, optimal contact pressure on the wheel brakes of the
car in direct proportion to the wheel pressure.
In Fig. 5 a simple drawing is shown of the rail brakes with the measurement
positions.
106 J. Hlaváček, L. Hejzlar, and R. Kolmačka
In this case the provisional lubrication was used between the rail brakes and
measurement position M1 and M2.
Results are an energetic mean value in dB(A) from all wagons for the transit time
from the rail brake (7.5 m measurement distance) [2].
3.4 Conclusion
All types of the rail brakes have similar noise emissions without noise control measures
(for the similar input and output speed). The pneumatic brakes are slightly louder due to
additional sources (compressors). By lubrication it is possible to reach more than 10 dB
overall noise reduction. The low noise barrier achieve up to 15 dB reduction.
The combined measures seem to be more effective, but are not in all cases sufficient.
The aim of this activity is to assess the effectiveness of prefabricated platform edge
H130 with rubber blocks for noise emissions reduction from rail traffic. This noise
reduction measure is applied for the first time in the Czech Republic on a single
track-line.
In Fig. 6 a simple drawing of this solution is shown.
The Efficiency of Noise Reduction Measures on Railway Infrastructure 107
The microphones positions were 1.2 m above top of the rail and 7.5 m far from the
center of the track [2].
In Table 4.1 the test results from the measurement campaigns are given, before
retrofitting, after retrofitting only and after implementation of the platform edge H130
with absorbing layer.
The results are the energetic average from all trains of the each category from the
same trains measured on the side without platform and with platform. All trains were
divided into the two following categories only.
Category 2 - Passenger and express trains equipped with tread and disc brake
Category 4 - Freight trains
Category of train Side without platform Side with platform edge H130
Cat. 2 88.5 85.8
Cat. 4 89.2 87.0
The results show a reduction of the total noise emission of 2.2 dB (for freight trains)
and 2.7 dB (for passenger trains). The difference comes from higher speed at the
passenger trains.
References
[1] Hlaváček, J., Hejzlar, L., et al.: Expert report on the progress and achievements of the
solution in 2012, from the project Nr: TA01031267 “Noise and vibration in the railway
system” (2012)
[2] EN ISO 3095:2006 “Railway applications – Acoustics – Measurement of noise emitted by
railbound vehicles” (2006)
[3] Hlaváček, J.: Technical report TZ 003/2010 “Measurement Noise and Ground vibration on
the retrofitted section of the track Kolín – Poděbrady”
108 J. Hlaváček, L. Hejzlar, and R. Kolmačka
1 Introduction
Impacts at railway crossings and due to wheel defects such as wheel flats have been a
source of growing concern and can be quite severe. In addition to potential damage that
may be caused, impacts at wheel and rail discontinuities generate high noise and
vibration. Substantial improvements in impact noise mitigation have been achieved by
using welded rails, but wheel flats, dipped welds, switches and crossings still remain
important sources of disturbance. Compared with railway rolling noise, time-domain
methods have to be employed for impact noise modelling, and this leads to much
greater computational demands.
In [1], Ver et al. categorised the various irregularities leading to impact noise as a
‘smooth irregularity’, ‘level joint’, ‘step-up joint’, ‘step-down joint’ and ‘flat wheel’,
and their characteristics were tabulated accordingly. Remington [2] extended the work
from Ver et al. [1] and estimated an ‘equivalent’ roughness spectrum for various wheel
and rail discontinuities. Wu and Thompson [3] developed a hybrid method in which an
equivalent roughness spectrum was obtained from a simplified time-domain
calculation. This spectrum was then used as input to a more detailed frequency-domain
model, Track-Wheel-Interaction- Noise-Software (TWINS) [4], to predict the impact
noise from wheel flats.
A simple mass-spring-damper equivalent track model, with only three
degrees-of-freedom (3-DOF), is developed here to gain insight into the impact vibration
induced by a wheel flat. Using this very efficient time-domain wheel/rail interaction
model, the characteristics of impact vibration due to wheel flats are studied and the
effects of vehicle speed and track support stiffness are investigated as examples. The
same model could also be used to investigate rail joints.
The frequency response of an infinite track system below about 1 kHz is similar to a
two degree-of-freedom (2-DOF) system. Therefore proposed to study the
characteristics of impact vibration using an equivalent mass-spring system in place of a
wheel/rail system. Compared with Wu and Thompson [3] who used a rational
polynomial approximation to the track, this model has the advantage that the masses
and springs can be identified with physical components of the track. The masses M1 and
M2 can be loosely associated with the rail and sleepers and the springs K1 and K2 with
the pad and ballast.
KT = 2 2 ( EI r )
1/4 3/4
s (2.1)
where s is the support stiffness per unit length. For a track system with a two layer
support, the low frequency asymptote of the frequency dependent support stiffness can
be used, where
sp sb
s= (2.2)
sp + sb
where sp and sb are the rail pad and ballast stiffness per unit length. The equivalent
length of track that deflects under a point load, L = KT / s = 2 2 4 EI r / s can be used as
a scale factor to convert values of mass and stiffness per unit length into discrete mass
and stiffness values to calculate the response of the track, i.e. K1=sp/L.
The mobility of the 2-DOF mass-spring system is compared with an infinite
Timoshenko beam track model (‘Rodel’ as used in TWINS [4]) in Fig. 2. The stiffness
values sp = 920 MN/m2 and sb = 67 MN/m2 are used in this case. The track mobility is
well represented by the simple 2-DOF mass-spring system below about 1 kHz with the
resonances and anti-resonances well preserved.
Characterizing Wheel Flat Impact Noise with an Efficient Time Domain Model 111
uw Mw
F
Kc
M0 M0
u0
F K 0, C 0
u1 M1 M1 M1
K1, C1
u2 M2 M2 M2
K2, C2
Track (Rodel)
Mobility magnitude [m/s/N]
-4
10 2-DOF
3-DOF
-5
10
-6
10
-7
10 1 2 3 4
10 10 10 10
2
Phase [radians]
-1
-2 1 2 3 4
10 10 10 10
Frequency [Hz]
Fig. 2. Comparison of rail mobility between 2-DOF/3-DOF mass-spring equivalent system and
infinite Timoshenko track system (Rodel)
112 J. Yang, D.J. Thompson, and Y. Takano
At high frequencies, the track mobility is damping controlled as can be seen from
Fig. 2 with a phase close to 0. This is due to the fact that for an infinite track, energy is
carried away from the driving point due to free wave propagation [6]. Therefore, the
2-DOF mass-spring system will be improved in this section to take account of the high
frequency behaviour.
To match the high frequency behaviour of the track model, an additional damper C0
can be added in series with the 2-DOF mass/spring model, as shown in Fig. 1(b). The
value of the damper can be set to the high frequency asymptote of the point mobility of
an infinite Timoshenko beam [7] :
where A is the shear area, G is the shear modulus and ρ is the density of the rail.
M0 M0 M0
M0
K0
M1 M1
M1 M1
K1
M2
M2 M2 M2
K2 original
state 96Hz 797Hz 8187Hz
Fig. 3. Mode shapes of 3-DOF mass-spring system
As the added damper will also affect the low frequency behaviour which is
unwanted, a spring K0 can be used in parallel with the damper C0, as shown in
Fig. 1(b). The spring K0 is added to avoid dominance of the damping effect at low
frequencies. Moreover, a small mass M0=0.5 kg is added to avoid difficulties of
coupling the model to the contact spring in a time domain model. The mobility of this
3-DOF mass-spring system is also shown in Fig. 2, where M0=0.5 kg, M1=74 kg,
M2=188 kg; K0=1.3x109 N/m, K1=1.3x109 N/m, K2=9.6x107 N/m, C0=2.0x105 Ns/m,
C1=2.2x103 Ns/m and C2=1.1x105 Ns/m. The improved system represents the track
very well for the frequency range considered.
The mode shapes of this 3-DOF mass-spring system are shown in Fig. 3. The
original state is also shown. The axes are arbitrary, as is the amplitude of the mode
shapes, but the relative amplitudes are preserved. The horizontal lines indicate the
positions of the different masses in their un-deformed state. It can be seen that the final
resonance due to the addition of M0 is beyond the frequency range of interest. The
relative motion between masses M0 and M1 are very small for the first two modes. The
mode shape ratios of the three modes are [1.02:1.02:1], [1:0.99:-0.39] and [1:0:0]
respectively.
Characterizing Wheel Flat Impact Noise with an Efficient Time Domain Model 113
To study the characteristics of the coupled system, a mass Mw, of 600 kg, representing
the wheel is coupled to the track through a non-linear contact spring, as shown in Fig.
1 (c). The contact spring has a variable stiffness Kc depending on relative deflection of
wheel and rail (uw–u0). A static load W of 33 kN is applied, representing the weight of
the sprung mass above the wheel. The equations of motion for this system are
M wuw + CH ( u w − u0 − r )
1.5
=W
M 0u0 + C0 ( u0 − u1 ) -CH ( u w − u0 − r ) + K 0 ( u0 − u1 ) = 0
1.5
(3.1)
M 1u1 + K 0 ( u1 − u0 ) + C0 ( u1 − u0 ) + C1 ( u1 − u2 ) + K1 ( u1 − u2 ) = 0
M 2u2 + C1 ( u2 − u1 ) + C2u2 + K1 ( u2 − u1 ) + K 2u2 = 0
where a non-linear Hertzian contact spring is assumed and loss of contact is allowed,
i.e. CH = 0 if uw-u0-r<0. r represents the ‘roughness’ input between the wheel and the
rail which can include the wheel flat geometry. CH is the coefficient of the Hertzian
contact spring for which the 93.7x109 Nm-3/2 is used throughout the calculations.
The equations of motion can be arranged in a state space matrix form as:
Ag = Bg + C (3.2)
where
{ }
T
g = uw uw u0 u0 u1 u1 u2 u2 Fˆc (3.3)
A = diag (1 M w 1 M 0 1 M 1 1 M 2 1) (3.4)
⎡0 1 0 0 0 0 0 0 0⎤
⎢ − Kc 0 Kc 0 0 0 0 0 0⎥
⎢0 0 0 1 0 0 0 0 0⎥
⎢ Kc 0 − Kc − K0 −C0 K0 C0 0 0 0⎥
B = ⎢0 0 0 0 0 1 0 0 0⎥ (3.5)
⎢0 0 K0 C0 − K 0 − K1 −C0 − C1 K1 C1 0⎥
⎢0 0 0 0 0 0 0 1 0⎥
⎢0 0 0 0 K1 C1 − K1 − K 2 −C1 − C2 0 ⎥⎥
⎢
⎣ Kc 0 − Kc 0 0 0 0 0 0⎦
C = {0 W + K c r 0 − K c r 0 0 0 0 − K c r}
T
(3.6)
where the force impulse F̂c at each contact position is the integral of the interaction
force Fc over the length of the time step, as used by Nielsen and Igeland [8]. The state
space equation (3.2) can be solved using a time stepping routine; in the present work a
fourth order Runge-Kutta method is used.
114 J. Yang, D.J. Thompson, and Y. Takano
4 Validation
The wheel/rail interaction calculation using the mass-spring model is validated against a
detailed FE track model based on [8]. The calculations are performed in the time domain
with a wheel flat irregularity excitation. The irregularity excitation is obtained from a
measured roughness profile of a wheel with a flat spot [9]. In order to take account of
the finite wheel/rail contact area which is known to introduce a contact filter effect for
noise generation, a 3D contact filter [10] was applied to the measured profile prior to
carrying out the time domain calculations. This 3D contact filter considers the complete
profile of wheel and rail roughness using a 3D non-Hertzian contact model.
x 10
4 Train speed =120km/h
Contact force [N]
8 m−k
6 FE
4
0
0.56 0.565 0.57 0.575 0.58 0.585 0.59 0.595 0.6
Time [s]
80
Contact force
(dB re 1N)
70
60
50
40
125 250 500 1000 1600 2500
1/3 Oct Centre Frequency (Hz)
Fig. 4. Contact force comparison between mass-spring (m-k) system and FE system in terms of
time domain wheel/rail interaction. The spectra are shown at the bottom. The velocity of moving
wheel (or moving speed of roughness for m-k system) is 120 km/h.
One example is considered here, for a speed of 120 km/h (33.3 m/s) as shown in Fig.
4. The contact force time histories are shown at the top while the spectra are displayed
at the bottom of Fig. 4. The mass-spring model (labelled m-k) agrees very well with
the FE track model. Importantly, it takes only about one-tenth of the time to obtain
results for the mass-spring system for each run compared to the FE model for the same
moving distance of the wheel on the same computer. This is a significant saving of time
allowing parameter studies to investigate the characteristics of wheel/rail interaction.
5 Results
Two examples, variation with respect to the train speed v and rail ‘pad’ stiffness K1,
are considered in this section.
The first one occurs when the wheel cannot stay in contact with the rail due to the
geometric change of the irregularity. This can be seen at speeds of 60 km/h and higher.
Because of loss of contact, an impact is generated between wheel and rail when contact
is re-established, leading to a subsequent dynamic loss of contact. Fig. 5 indicates that
the force peak is turning into a double peak as train speed increases. Moreover, the
double peak becomes more apparent with increasing speed.
4 4
] x 10 30km/h x 10 60km/h
N
[
e
10 10
c
r
o
f
t
5 5
c
a
p
m 0 0
I 2.7 2.8 2.9 3 3.1 2.7 2.8 2.9 3 3.1
4 4
] x 10 120km/h x 10 180km/h
N
[
e
10 10
c
r
o
f
t
5 5
c
a
p
m 0 0
I 2.7 2.8 2.9 3 3.1 2.7 2.8 2.9 3 3.1
Distance [m] Distance [m]
Fig. 5. Impact force history at different speeds
4
x 10
50km/h K1/2 4
x 10
120km/h
Impact force [N]
Impact force [N]
10
K1
8
10
K1*2 6
5 4
0 0
0.2 0.205 0.21 0.215 0.22 0.083 0.084 0.085 0.086 0.087 0.088 0.089 0.09 0.091
Time [s] Time [s]
80
Contact force
Contact force
70
(dB re 1N)
(dB re 1N)
70
60
60
50
50
Fig. 6. Contact force time histories and spectra at vehicle speeds of 50 km/h and 120 km/h for
three different stiffness values of K1
It can be seen that at 50 km/h, a higher value of K1 leads to a larger peak impact force
as one would expect. However, at 120 km/h, varying K1 relative to the reference value,
regardless of whether it is bigger or smaller, leads to a higher peak contact force for the
116 J. Yang, D.J. Thompson, and Y. Takano
parameter combination used. In terms of spectra of the contact forces, Fig. 6 indicates
that larger values of K1 lead to a higher force of the first peak at around 250 Hz. The
second peak of the contact force, at around 1 kHz for the original K1, occurs at a higher
frequency for larger K1. The second peak is at a lower level than the first peak but the
difference gets smaller with a higher speed or smaller K1.
6 Conclusions
A 3-DOF mass-spring model has been developed to represent the track system. This
model is validated by comparison with a Timoshenko beam track model. An
advantage of this model is that the masses and springs are physically linked to the
track parameters. Coupled with the wheel mass and contact spring, a very efficient
time-domain wheel/rail interaction model is formed. This can be used to study the
characteristics of wheel flat impact vibration and noise and examples of results for
speed and support stiffness variation have been given.
References
[1] Vér, I.L., Ventres, C.S., Myles, M.M.: Wheel/rail noise–Part III: Impact noise generation
by wheel and rail discontinuities. Journal of Sound and Vibration 46(3), 395–417 (1976)
[2] Remington, P.J.: Wheel/rail squeal and impact noise: What do we know? What don’t we
know? Where do we go from here? Journal of Sound and Vibration 116(2), 339–353
(1987)
[3] Wu, T.X., Thompson, D.J.: A hybrid model for the noise generation due to railway wheel
flats. Journal of Sound and Vibration 251(1), 115–139 (2002)
[4] Thompson, D.J., Janssens, M.H.A., de Beer, F.G.: TWINS: Track-Wheel Interaction Noise
Software, theoretical manual (version 3), TNO report HAG-PRT-9902111999, Delft
[5] Thompson, D.J.: Railway Noise and Vibration, 506 p. Elsevier Science, Oxford (2008)
[6] Graff, K.F.: Wave Motion in Elastic Solids. Dover, New York (1975)
[7] Ruge, P., Birk, C.: A comparison of infinite Timoshenko and Euler–Bernoulli beam
models on Winkler foundation in the frequency and time domain. Journal of Sound and
Vibration 304(3-5), 932–947 (2007)
[8] Nielsen, J.C.O., Igeland, A.: Vertical dynamic interaction between train and track influence
of wheel and track imperfections. Journal of Sound and Vibration 187(5), 825–839 (1995)
[9] Yang, J.: Time domain models of rail/wheel interaction - taking account of surface defects.
PhD thesis, University of Southampton (2012)
[10] Pieringer, A., Kropp, W., Thompson, D.J.: Investigation of the dynamic contact filter effect
in vertical wheel/rail interaction using a 2D and a 3D non-Hertzian contact model.
Wear 271(1-2), 328–338 (2011)
[11] Steenbergen, M.J.M.: Modelling of wheels and rail discontinuities in dynamic wheel-rail
contact. Vehicle System Dynamics 44, 763–787 (2006)
Study on the Sound Radiation Directivity of a
Railway Wheel and the Relationship between Directivity
and Mode Shape
J. Han, X.B. Xiao, R.Q. Wang, X. Zhao, G.T. Zhao, and X.S. Jin
Summary. Due to the complex geometry of a railway wheel, the shape of its
modes may have to be properly considered to understand and predict the sound
radiation directivity around it. A straight-web metro wheel is studied in this paper
with a radial excitation. First, the near-field sound radiation at the natural
frequencies is measured in experiments performed in a semi-anechoic room. The
natural frequencies here were determined by modal test. The corresponding
mode shape is calculated from a finite element (FE) model. A good agreement is
found between measurements and simulations in terms of natural frequencies.
Further, the relationship between mode shape and sound field near the wheel is
determined. Second, a rotating semi-circular frame of radius 2 m centered at the
center of the wheel is employed to measure the directivity pattern. The
directivities of A-weighted total level, one-third octave bands and modes are
recorded at a total of 667 measuring points. A radial mode (r, 2) and an axial
mode (1, 2) are taken as examples to derive empirical equations of directivity
pattern. A model developed by the boundary element (BE) method is found to
simulate the directivity pattern more precisely than the derived equations.
1 Introduction
In the literature, studies on the directivity of wheel radiation can only be found in a
few reports, in which simple simulation models were usually employed. In 1976,
Remington [1] measured sound radiation directivity of a wheel suspended on a rail.
Two microphones with a fixed height were used to record the Sound Pressure Level
(SPL) at different positions. Six measuring points that were 3.1 m away from the
wheel were selected in the test, and a constant angle of 15° was kept between the
neighboring measuring points. The wheel was regarded as a monopole source to
within ± 5 dB in all directions. It should be noted that in the test of Remington the
SPL in some representative directions might not be captured because of the limited
measuring points. In 1983, Wolde and van Ruiten [2] declared that results of their
directivity measurement of wheel radiation were different from that of Remington.
However, they did not publish their data. In 2002, Thompson [3, 4] measured the
As shown in Fig. 1(a), a one-meter long rotating linear microphone array is used to
record the near-field sound radiation during a test. The microphone array covers 11
measuring points which are 30 cm away from the surface of the wheel. The array
rotates once every 12° (until 180°) to scan the entire plane.
Microphones
90°
Rotating every 12° Falling ball
Interval 5°
Falling ball device
90°
0°
0
Interval 10 cm Rotating every 5°
Hitting point
3 Numerical Model
4 Results Analysis
The relationship between vibration and sound near the wheel is first studied in section
4.1 on the basis of measurements of near-field sound radiation and results of modal
analysis. Afterwards, the directivity pattern is further examined in section 4.2 by
comparing the experimental and numerical results.
Fig. 3 shows two mode shape of the wheel calculated by the FE model, being at
1762 Hz and 2228 Hz, respectively. The mode shape at 1762 Hz is a radial mode (r, 2)
with 2-nodal-diameters, while the one at 2228 Hz is an axial mode (1, 2) with
one-nodal-circle and 2-nodal-diameters. Fig. 4 shows the near-field sound radiation at
natural frequencies of 1761 Hz and 2208 Hz, which are measured from the outside of
the wheel in the semi-anechoic room.
Note that the modes shown in Fig. 3 are similar to the near-field sound radiation in
Fig. 4. By comparison it is found that the frequency difference between the simulation
and the experiment is only 0.05 % and 0.90 %, respectively. For other modes, the
differences are no more than 2 %. From this it can be concluded that the developed FE
model can accurately predict the experiment.
Fig. 5 shows the quarter-sphere directivity contour of overall A-weighted SPL. The
surface contains 667 acoustic measuring points, and the analyzed frequency range is 0 -
6.4 kHz. It is clearly seen that the distribution of sound radiation is non-uniform on the
sphere surface, and the overall level differences are up to about 6 dB.
Fig. 6 shows the quarter-sphere directivity contour at two central frequencies of 1/3
octave bands, i.e. at 1600 Hz (1410-1780 Hz) and 2000 Hz (1780-2240 Hz) that are
important for wheel-rail noise. Fig. 7 shows the quarter-sphere directivity contour of
two dominant modes at 1761 Hz and 2208 Hz. It is found that the directivity pattern is
similar in Figs. 6 and 7. This is can be explained as follows: The modes in Fig. 7
dominate the sound in the corresponding 1/3 octave bands in Fig. 6.
For clarity, the vertical plane containing the wheel axis (see Fig. 8) is chosen for
further analysis of the relationship between directivity and mode shape. The measured
directivity in the plane is plotted in Fig. 8, in which the angular coordinate from 0° to
180° is labeled. Note that the measurement interval is 5° as explained above.
It is seen from Fig. 8 that the SPLs have a minimum on or close to the axis of the
wheel (0°) for the radial mode (r, 2) and the axial mode (1, 2). The directivity is
approximately symmetric, and several side lobes occur at certain angles. For the radial
mode (r, 2) the SPL at 90° is close to those at other angles, while a local minimum
occurs at this position for the one-nodal-circle axial mode (1, 2).
Based on observing the vibration shape of the radial mode, the main trend of
directivity of the radial mode can be expressed by |cosθ|. According to the experimental
directivity results, when the number of nodal diameters is 2, there are 6 side lobes in the
directivity from 0° to 180°. And when the number of nodal diameters increases by 1, the
number of side lobs increases by 4. Thus side lobes of directivity can be expressed by
sin[(4n-2)θ]. n represents the number of nodal diameters. Coefficients of 0.3 and 0.7 in
equation (4.1) are used as weights. Taking n=2 as the example, the directivity for radial
mode is supposed by empirical equation (4.1), as shown in Fig. 9(a).
Based on the one-nodal-circle axial mode shape, there are four parts on the web
vibrating strongly. The four parts can be regarded as two pairs of dipoles symmetrical
about the wheel center. The two pairs of dipoles vibrate conversely, and each dipole
includes two sources vibrating in the same direction. Taking n=2 as the example, the
directivity for one-nodal-circle axial mode can be supposed by empirical equation
(4.2), as shown in Fig. 9(b). SPLs in the axial direction have a minimum due to
cancellation by the contributions of two opposite dipoles. In equation (4.2), the distance
of two sources l=0.54 m, wave number k=2πf/c, frequency f=2208 Hz, speed of sound
c=344 m/s.
Study on the Sound Radiation Directivity of a Railway Wheel and the Relationship 123
sin(2 k Δ ) l
D (θ ) = ,Δ = sin θ (4.2)
2 sin( k Δ ) 2
5 Conclusions
Based on the results shown above, the following conclusions are drawn.
1) By comparing the two natural frequencies of simulation and experiment, the
difference is small. On both mode shapes and near-field sound radiation of a wheel, a
good agreement is found between the developed model and experiments.
2) Around the wheel, the overall SPL deviation in a quarter-sphere is found to be up
to 6 dB. By comparing the directivities between 1/3 octave band and dominant modes
124 J. Han et al.
in the band, it is found that the directivity pattern between them is similar. The
directivity of 1/3 octave band is determined by that of dominant modes.
3) Based on the experimental directivity data and the modal analysis, the relation
between directivity pattern and mode shape are linked by the empirical equations. And
then the directivity patterns for radial modes and axial modes can be simulated by two
empirical equations approximately.
4) For the radial mode (r, 2) and the axial mode (1, 2) as the example, the SPLs have
a minimum on or close to the axis of the wheel (0°). The directivity is approximately
symmetrical about the plane containing the wheel. As the angle of directivity changes,
several side lobes appear. For the radial mode (r, 2) the difference of SPLs between the
direction of 90° and others is not significant, while for the one-nodal-circle axial mode
(1, 2), the SPLs near or on the direction of 90° have a local minimum.
5) A FE-BE model is developed to calculate the directivity pattern, and is validated
by comparing to the experiments.
Acknowledgements. The present work was supported by the National Science and
Technology Support Program of China (2009BAG12A01-B06), the National High
Technology Research and Development Program of China (863 Program)
(2011AA11A103-2-2), the Funds for Innovation Research Team of Ministry of
Education of China (IRT1178, SWJTU12ZT01), and the Fundamental Research Funds
for the Central Universities and the Science (SWJTU12ZT01).
References
[1] Remington, P.J.: Wheel/rail noise Part I: characterization of the wheel/rail dynamic system.
Journal of Sound and Vibration 46(3), 359–379 (1976)
[2] Wolde, T., van Ruiten, C.J.M.: Sources and mechanisms of wheel/rail noise: state-of-art and
recent research. Journal of Sound and Vibration 87(2), 147–160 (1983)
[3] Thompson, D.J.: Sound radiation from a vibrating railway wheel. Journal of Sound and
Vibration 253(2), 401–419 (2002)
[4] Thompson, D.J.: Railway Noise and Vibration: Mechanisms, Modelling and Means of
Control. Elsevier (2009)
[5] Zhang, X., Jonasson, H.: Directivity of railway noise sources. Journal of Sound and
Vibration 293(2), 995–1006 (2006)
[6] Zhang, X.: Directivity of railway rolling noise. In: Schulte-Werning, B., Thompson, D.,
Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson, J., Maeda, T., de vos, P. (eds.) Noise and
Vibr. Mit. Rail Trans. Sys. NNFM, vol. 99, pp. 426–432. Springer, Heidelberg (2008)
[7] Zhang, X.: Modeling the directivity of wheel/Rail radiation using a circular/straight line of
perpendicular dipole pairs. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B.,
Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 143–150. Springer,
Heidelberg (2012)
[8] Zhang, X.: Applicable directivity description of railway noise sources. PhD thesis. Chalmers
University of Technology, Göteborg Sweden (2010)
[9] Fang, R., Xiao, X.B., Jin, X.S., Wen, Z.F.: Effect of shapes on sound radiation from railway
wheel. In: Proceedings of International Conference on Mechanical Engineering and
Mechanics 2007, Wuxi, China (2007)
Empirical Modeling of Railway Aerodynamic Noise Using
One Microphone Pass-By Recording
X. Zhang
1 Introduction
For high-speed trains there are two important noise types: rolling noise and
aerodynamic noise. For the former, modeling has been successful: one can use the
powerful calculation software, the Track Wheel Interaction Software (TWINS), to
calculate the noise emission from the wheels and the rails/track [1, 2], or (probably
more often used in practice), the indirect roughness method (IRM) to predict the total
noise emission from the wheel-rail interaction [3]. For aerodynamic noise, after more
than 20 years intensive research, the different origins of it as well as the relative
strengths of the component sources have been defined, and respective noise generation
mechanisms are well understood [2]. The modeling of simple configurations, such as
pantograph and pantograph recess, has produced promising results [4, 5]. As a
component of the total noise level, aerodynamic noise becomes comparable to rolling
noise above about 350 km/h, while its influence is already measurable above about
220 km/h [2, 6].
In 2010, the SP Technical Research Institute of Sweden (SP) was consulted about
predicting railway aerodynamic noise, in order to evaluate noise impact from
high-speed trains (HSTs) on Swedish tracks in the near future [7]. As stated in [2],
quantification of the strengths of aero-noise components remains difficult and the
modeling has been used only within the research environment, mostly limited to very
simple configurations. Therefore, for handling the aero-component of the noise impact
from HSTs, one has to work out some empirical method. For the above mentioned task,
X2 trains were selected. This train type is currently popular in Sweden, and has a
nominal top speed 200 km/h. Moreover, in the 1990s X2 trains ran frequently at a
higher top speed between 250 and 270 km/h [8]. With the advantage of having the SEL
data collected in 1994 by SP, which cover a wide speed range from 70 to 270 km/h [8],
and the total roughness and total transfer function which were determined during the
IMAGINE project (2004-2007) [9], X2 trains were suited for making an empirical
prediction of the aerodynamic noise.
In Section 2, for the X2 train type, the modeling of rolling noise will be verified by
comparing the model-predicted sound exposure level (SEL) of the rolling noise with
the collected SEL data (in octave bands from 63 Hz to 4000 Hz). Next, in Section 3, an
empirical description of the aerodynamic noise will be presented. Further discussion of
this empirical modeling will be given in Section 4.
One microphone recording of train pass-by noise contains the contributions of all noise
sources, i.e. traction noise, rolling noise, aerodynamic noise, and less important sources
such as train body vibration noise. As rolling noise can now be predicted accurately and
traction noise is negligible at high speed, one microphone recordings of high-speed
train noise become useful for empirically modeling railway aerodynamic noise.
The railway tracks of concern are ballasted, and equipped with mono-block
concrete sleepers and BV50 rails (which are similar to UIC54 rails). The total
roughness and the total transfer function (over one of or over the three tracks) were
determined based on the data collected by SP during 2004-2005 [9]. The SEL data
collected by SP in 1994 [8] were measured at 25 m from the respective track centre
line. For each 10 km/h interval between 130 and 270 km/h, also including 70 km/h,
the representative data were obtained by averaging over a number of pass-by
recordings at different sites and normalized to 10 m from the track centre line. The
standard deviation of the A-weighted SEL data is about 1-1.5 dBA. (There are some
exceptions: (1) no pass-by around 150 km/h was recorded; (2) only one pass-by was
recorded for speeds around 70, 160, 170 and 270 km/h.)
According to the IRM, the equivalent sound pressure level when rolling noise
predominates, L p ,tot (V , f ) , measured in the standard position (7.5 m from the track
centre line and 1.2 m above the railhead) over the time interval for the vehicle at speed
Empirical Modeling of Railway Aerodynamic Noise Using One Microphone 127
V passing the measurement position ‘from buffer to buffer’ [3], is related to the total
roughness and the total transfer function as follows
L p ,tot (V , f ) = 10 log10 (N axle / Lwagon ) + LH ,tot ( f ) + Lr ,tot (V / f ) (2.1)
where Lr ,tot (V / f ) is the total roughness level, LH ,tot ( f ) is the total transfer
function (1 axle per meter), N axle is the number of axles per wagon, Lwagon is the
wagon length, V is the train speed (in m/s) and f is the frequency (in Hz).
Then, as proposed in [10], the sound power level (LW) per meter of train can be
obtained using a simple practical method:
where Aline , propagation describes the relevant propagation attenuation of the sound. For
convenience, this quantity was pre-calculated using the Nord2000 propagation
model [11] and provided as tabular constants (in one-third octave bands from 25 Hz to
10 000 Hz) which are suitable for most practical situations [10].
Moreover, by using the vehicle and track transfer functions, the rail and wheel
radiation can further be separated. With i = 1 for rail radiation and 2 for wheel radiation,
the component sound power levels are:
Next, the SEL due to the rolling noise can be determined as follows [7]:
⎡ l ⎤ ⎧⎪ϕ max ⎫⎪
SEL = LWA (1 m, 0 ) + 10 log10 ⎢ train ⎥ + 10 log10 ⎨ ∫ 10 [ΔL (ϕ )− Aexcess (ϕ )] / 10 dϕ ⎬ (2.4)
⎣ 4πdV ⎦ ⎪⎩ϕ min ⎪⎭
Fig. 1. A comparison of the SEL values predicted by the IRM (rolling noise only) ( ) and SP
1994 data (also containing aerodynamic noise) ( ) [8]
For high-speed train noise, only rolling noise and aerodynamic noise are important. As
rolling noise is well predicted by the IRM, the LW of the aerodynamic noise at a high
speed, e.g. 250 km/h or higher, can then be determined by subtracting the rolling noise
contribution from the measured total. Moreover, by imposing a speed exponent of 6
due to the dipole characteristic of the aero-noise sources [2], it is then possible to
predict the aerodynamic noise at other speeds. Furthermore, for aerodynamic noise the
spectrum also varies with speed: shifting to higher frequencies with increasing train
speed as follows [13]:
f = f 0 * V / V0 (3.1)
However, it was found that, at 250 Hz and below, much better prediction results (by
fitting the data) were obtained when using the following relation:
LW , aero ( f , V ) = LW , aero ( f * V0 / V , V0 ) + 40 log10 (V / V0 ) (3.3)
X2 trains. (Note: 260 km/h was taken as the reference speed because seven pass-bys at
260 km/h were recorded compared with one at 270 km/h [8].)
Eqs (3.1) to (3.3) together with the LW value at a typical high speed, e.g. as listed in
Table 1, constitute the empirical modeling for aerodynamic noise of X2 trains. As can
be seen from Fig. 2, the prediction of the combined model, which consists of the IRM
for rolling noise and the empirical modeling for aerodynamic noise, agrees well with
the measurements within the entire speed range.
Table 1. The LW of aerodynamic noise for X2 train type at 260 km/h [7]
Fig. 2. A comparison of the SEL values predicted by the combined model ( ) and SP 1994 data
( )
130 X. Zhang
4 Discussion
In this paper, an empirical description of railway aerodynamic noise was presented.
The model prediction of the SEL values of the train pass-by noise, with or without the
contribution of aerodynamic noise being considered, was compared with the SEL data.
A good agreement between the predicted SEL values and the measured SEL data
suggests that this empirical description of railway aerodynamic noise is useful in
engineering applications. Therefore, this empirical description together with the
prediction method for rolling noise can be used to evaluate high-speed train noise.
For the speed dependence, the speed exponent for rolling noise is typically around 3.
In the IRM modeling, the speed dependence is automatically introduced by the
description of roughness and transfer function. The IRM calculations confirm this
understanding, while also showing small variations in speed exponent between train
types and for different speed ranges (see Table 2), that are probably mainly due to a
difference in roughness spectrum and the spectrum shift (with train speed) in the
frequency domain.
For the total aerodynamic noise, the equivalent speed exponent was found to be
around 6.8 (see Table 2), differing from the value of 6 that was used in the modeling.
The cause of this difference is due to two factors: (1) the introduced spectrum shift
given by Eq. (3.1) and (2) the A-weighting effect. In other words, if the spectrum were
flat, or if linear weighting were used, the speed exponent would be the same as what
was used in the modeling.
It was also found that, for low frequency components (< 315 Hz) of the aero-noise, a
speed exponent 4 instead of 6 should be used in the modeling – this seems to suggest
that at low frequencies a monopole source dominates the aerodynamic noise. This
potential monopole source may originate from the pressure rise caused by the
outward-pushed air (from the track) when a train approaches, and the pressure drop
caused by the inward-dragged air when the train recedes.
Moreover, during the European Imagine project [14] this question was raised: how
to use those one microphone data? As the IRM requires one microphone data for the
train pass-by noise as well as data for the rail vertical vibration, the huge quantity of one
microphone data might be considered less useful for an advanced prediction method
such as the IRM. However, this is not the case. One microphone data contains
contributions from all the sub-sources. If the contributions of important sub-sources,
e.g. rolling noise and aerodynamic noise for HST trains, can properly be separated then
such one microphone data can be of value. In this paper, our calculations showed the
Empirical Modeling of Railway Aerodynamic Noise Using One Microphone 131
way to use such one microphone data: (1) for a chosen train type, the rolling noise shall
be modeled using the indirect roughness method; (2) when the IRM modeling is
completed, one microphone data for the same train type become useful for modeling
the aerodynamic noise. One should of course be careful about the track type and the
total roughness. For the data used in the IRM modeling and for the one microphone
data used for modeling the aerodynamic noise, the tracks should be of the same type
and the total roughness should be similar. (Notes: (1) Although different track types can
in principle be handled by respective track transfer functions, using data for a different
track type may introduce extra uncertainty in the relevant spectrum components. (2)
The similarity of two sets of total roughness data can be judged, e.g. by comparing the
resulting overall A-weighted SELs at different speeds, as shown in Fig. 1.)
References
[1] Mahé, H., Thompson, D.J., Zach, A., Hölzl, G.: Experimental validation of the prediction
model TWINS for rolling noise. In: Proc. of Inter-Noise 1993, Leuven, Belgium, pp.
1459–1462 (1993)
[2] Thompson, D.: Railway Noise and Vibration: Mechanisms, Modelling and Means of
Control. Elsevier (2009)
[3] Janssens, M.H.A., Dittrich, M.G., de Beer, F.G., Jones, C.J.C.: Railway noise
measurement method for pass-by noise, total effective roughness, transfer functions and
track special decay. Journal of Sound and Vibration 293, 1007–1028 (2006)
[4] Takaishi, T., et al.: Numerical analysis of aero-dynamic noise emitted from a pantograph
based on non-compact greens function. Journal of Environment and Engineering 5,
84–96 (2010)
[5] Gloerfelt, X., et al.: Direct computation of the noise radiated by a subsonic cavity flow
and application of integral methods. Journal of Sound and Vibration 266, 119–146 (2003)
[6] Block, J.: UK aerodynamic noise measurements and analysis, IMA6TR-
051223-AEATUK (the IMAGINE technical report)
[7] Zhang, X.: Prediction of high speed train noise on Swedish tracks, SP Report 75 (2010)
[8] Göransson, C., Ström, T.: Externt buller från svenska tågtyper – Nya indata till den
nordiska beräkningsmodellen, SP Rapport 25 (1994) (in Swedish)
[9] Zhang, X.: Determination of Swedish Input for the Harmonoise Source Model for
Railway Noise, SP Report 50 (2005)
[10] Zhang, X.: A practical method to determine the sound power of railway noise using
one-microphone recordings. In: Proc. of Forum Acusticum 2005, (paper 579-0),
Budapest, pp. 1225–1228 (2005)
[11] Plovsing, B., Kragh, J.: Nord2000. Comprehensive Outdoor Sound Propagation Model,
DELTA Acoustics & Vibration Report AV 1851/00, Lyngby (2000)
[12] Zhang, X.: The directivity of railway noise at different speeds. Journal of Sound and
Vibration 329, 5273–5288 (2010)
[13] Stiebel, D.: Aerodynamic Noise – Frequency Dependency and Transferability of
Parameters, HAR12-MO-030918-DB01
[14] The Imagine project,
http://ec.europa.eu/research/fp6/ssp/imagine_en.htm
Localizing Noise Sources on a Rail Vehicle during Pass-By
Summary. This paper describes a pass-by measurement technique that has been
developed for localization and visualization of noise sources on moving rail
vehicles using beamforming. Based on measurements with an array of
microphones, while also measuring the position of the vehicle, the technique
calculates the contribution of noise, and visualizes it as a contour plot on top of a
picture of the train. Deconvolution is applied in addition to traditional
beamforming in order to get an improved spatial resolution in the noise map. A
set of measurements was made on two different types of regional trains on the
Danish railway: the Oresund trains and the IR4 trains. The speed of all the trains
was approximately 120 km/h. The results show that deconvolution is efficient for
identifying wind noise on the pantograph of the Oresund trains. The IR4 trains
turned out to have a strong source at the very front part of the train for
frequencies around 600 Hz - 800 Hz with a radiated sound power that was
approximately 5 dB above the noise radiated by the noisiest bogies. The cause of
this noise is yet unknown, but a potential explanation could be an aerodynamic
phenomenon at the front.
1 Introduction
Delay-And-Sum (DAS) beamforming is a powerful method for understanding the
sound radiation from acoustic noise sources. Often the method is used for localizing
stationary (fixed) sources, but it can also be applied on moving sources such as for
aircraft fly-over [1-3] or rotating blades on wind turbines [4, 5]. In [6], beamforming is
applied on a high speed train in a wind tunnel to analyze the noise radiation from
aerodynamic sources around the train, and in [7,8], beamforming is applied under real
pass-by conditions to analyze the wheel and rail induced noise.
During recent years, deconvolution techniques have been introduced as a
post-processing after DAS to improve the resolution and reduce the level of ghost
sources in the calculated noise maps [9]. The purpose of the paper is to describe a
commercially available system that combines moving source DAS with deconvolution
for applications on trains.
2 Theory
measured signals in such a way that the delayed signals add up coherently if sound is
approaching the array from that focus point. That is, for the jth focus point on the source
the beamformed signal is calculated as:
M
⎛ rmj (t ) ⎞
b j (t ) =
1
∑p m
⎜⎜ t +
c ⎟⎠
⎟, (2.1)
M m =1 ⎝
where c is the speed of sound and rmj/c is the time delay that ensures the coherent
summation. Next, the same time signals are delayed differently corresponding to a new
focus point to get the contribution from that point. Typically, the focus points are at
fixed global positions, but for pass-by applications they are a function of time, and
being fixed in the local coordinate system of the vehicle. As shown in [2] this tracking
automatically performs a de-dopplerization thereby providing the frequency content on
the source rather than at the receiver position.
Once the beamformed time signals have been computed, averaging of autopower
spectra (using FFT) is performed for each focus point in the time interval at which the
point was inside the applied covering angle of the array (for example up to ±35º
off-axis). The fact that the focus points follows the movement of the source means that
the averaging process will not smear out the resulting map of the sound field in contrast
to the case with stationary focus points.
Beamforming maps will include both the sources as well as their reflections, which
can for instance be introduced by the ground below the array. If the ground is flat and
rigid, and if the position of the microphones with respect to the ground is known
accurately, mirror ground conditions can be assumed. That is, the sources can be
assumed to have a mirror source below the ground, and the array can be mirrored as
well. For mirror ground conditions the array is often shaped as a half-wheel placed on
the ground, which means that the effective array will be a full wheel. This gives a
higher vertical spatial resolution in the maps than the half-wheel alone.
2.2 Deconvolution
The resolution of the DAS maps can be further improved by applying deconvolution
[9]. For deconvolution it is assumed that the source can be represented as a set of
incoherent monopole sources on the mapping surface. This is a reasonable assumption
when dealing with aerodynamic noise due to the uncorrelated nature of turbulence
excitation.
Ideally, DAS should represent each monopole as a delta function in the map, with
the level of the peak representing the pressure contribution from that monopole. In
practice, however, the beamformer will have a limited resolution, meaning that the
ideal delta function will be spatially smeared. This can be seen a 2D spatial impulse
response (also called the Point Spread Function, PSF) introduced by the beamformer.
The PSF can be predicted for a given array configuration and source position. The idea
of deconvolution is to compute the PSF, and deconvolve it with the DAS map to get
back to the real sources as illustrated in Fig.1.
Localizing Noise Sources on a Rail Vehicle during Pass-By 135
where DAS is the Delay-And-Sum map and PSF0 is the PSF for a monopole on-axis in
front of the array. The output from deconvolution basically is a map of the strengths, A,
of the monopoles, whereas DAS yields the pressure contribution from the source at the
focus point. There is a wide range of deconvolution techniques, but this paper considers
only the Non-Negative Least Squares (NNLS) method based on Fast Fourier Transform
(FFT) (see Ref. [9] for further details).
Fig. 1. Illustration of the idea behind deconvolution. The DAS map is assumed to consist of a
linear combination of Point-Spread-Functions, PSIi with individual amplitudes, Ai.
3 Measurement Results
In April 2013 array measurements were carried out at Skodsborg Station North of
Copenhagen in Denmark on a set of regional trains. Four measurements were made:
two passages of the Danish IR4 trains with four carriages, and two passages of the
Swedish/Danish Oresund trains with six carriages. Both types of trains are electrical.
Only trains driving north were considered, and they were all passing from left to right
(seen from the array) at approximately 120 km/h. The platform is assumed to be acting
as a mirror ground, which is why a half-wheel array was selected for the measurements
(see Fig. 2).
Fig. 2. Half-wheel microphone array (with 42 channels) on the platform at Skodsborg Station,
Denmark. Two photo-cells were used for position calculation (indicated with arrows).
Fig. 3 shows the calculated pressure contribution using DAS on one of the Oresund
trains for the 448 Hz - 1088 Hz frequency range. Note that the lowest 55 cm of the train
is not included in the calculation, since this part was below the platform. Although the
platform damps the sound radiation from the wheel/rails, high pressure contribution is
still clearly seen at the locations of the wheels. This is due to the fact that the upper part
of the wheels was above the ground, and also, the radiation from the wheel-rail contact
point is likely to be transmitted due to reflections on the train. All bogies are seen to
contribute more or less equally (with a pressure level of around 70 dB).
Although the wheels are dominating the maps in Fig. 3, sound is also generated at
the pantographs despite the relatively low train speed (120 km/h). Fig. 4 shows a
zoomed view on carriage 2 and 3 using DAS and deconvolution respectively. Higher
resolution is clearly obtained around the wheels when applying deconvolution, even if
the method is assuming the sound is emitted by a grid of incoherent point sources,
which may not be the best assumption for vibrations in the wheels. The deconvolution
maps also give better results than DAS around the pantograph, where the ghost images
are suppressed when compared with the DAS result. For this frequency band it is easy
to see that noise comes from the top part of the unfolded pantograph as well as from the
folded pantograph, which is exposed to a lot of wind because it is in front of the
unfolded pantograph. Fig. 5 shows a picture of the pantograph.
Localizing Noise Sources on a Rail Vehicle during Pass-By 137
Fig. 3. A-weighted DAS pressure contribution on an Oresund train driving to the right. The
carriage numbers are indicated on the map. Display range: 15 dB. Max. level: 70 dB. 448 Hz -
1088 Hz.
Fig. 4. A-weighted pressure contribution level from wheels and pantograph for an Oresund
train at 864Hz-1120 Hz. The train is travelling to the right. Display range: 15 dB. Upper: DAS
pressure contribution, max. level 62 dB. Lower: Deconvolution pressure contribution density,
max. level 68 dB.
138 J. Gomes, J. Hald, and B. Ginn
Fig. 5. Picture of pantograph on Oresund train. The train is driving to the right, and the front
pantograph is folded.
(a)
(b)
Fig. 6. DAS pressure contribution (A-weighted) on an IR4 train driving to the right. The carriage
numbers are indicated on the map. Colors of the trains do not resemble the actual colors. Display
range: 15 dB. (a) 448 Hz - 1088 Hz, max. level: 82 dB. (b) 1088 Hz - 2496 Hz, max. level: 79 dB.
Fig. 7. Sound power (based on intensity maps) as a function of frequency (64 Hz bands) for
different areas on the IR4 train. A peak is seen around 600 Hz from the front area of the train.
140 J. Gomes, J. Hald, and B. Ginn
4 Conclusions
This paper describes a measurement system for mapping of noise sources on rail
vehicles during pass-by. The system uses an array of microphones together with the
measured position of the train as a function of time to calculate and visualize the sound
field on the surface of the vehicle. The data is processed using a moving source
Delay-And-sum approach, and deconvolution for improving the resolution even
further. The presented results are based on a measurements made on Danish regional
electric trains (Oresund trains and IR4 trains) driving at about 120 km/h. Noise from the
pantographs was clearly identified despite the relatively low speed of the train. In
particular, deconvolution was efficient at pinpointing the position of those aerodynamic
sources on the Oresund trains. It was seen that the IR4 trains had a strong noise source
at the very front of the train at around 600 Hz - 800 Hz. The reason for this is yet
unknown. However, since the engines and traction are not located at the front bogie,
and since the sound did not appear at the wheel location, a potential cause could be an
aerodynamic phenomenon.
References
[1] Michel, U., Barsikow, B., Helbig, J., Hellmig, M., Schüttpelz, M.: Flyover noise
measurements on landing aircraft with a microphone array, AIAA Paper 98-2336
[2] Hald, J., Ishii, Y., Ishii, T., Oinuma, H., Nagai, K., Yokokawa, Y., Yamamoto, K.:
High-resolution fly-over beamforming using a small practical array. AIAA Paper (2012)
[3] Sijtsma, P., Stoker, R.: Determination of absolute contributions of aircraft noise components
using fly-over array measurements. AIAA Paper 2004-2958
[4] Oerlemans, S., Sijtsma, P., López, B.M.: Location and quantification of noise sources on a
wind turbine. Journal of Sound and Vibration 299, 869–883 (2007)
[5] Gomes, J.: Noise source identification with blade tracking on a wind turbine. In:
Proceedings of Inter-Noise 2012, New York City (2012)
[6] Lauterbach, A., Ehrenfried, K., Kröber, S., Ahlefeldt, T., Loose, S.: Microphone array
measurements on high-speed trains in wind tunnels. In: Berlin Beamforming Conference,
BeBeC (2010)
[7] Le Courtois, F., Thomas, J.-H., Poisson, F., Pascal, J.-C.: Identification of the rail radiation
using beamforming and a 2 D array. In: Proceedings of Acoustics 2012, Nantes (2012)
[8] Mellet, C., Letourneaux, F., Poisson, F., Talotte, C.: High speed train noise emission: Latest
investigation of the aerodynamic/rolling noise contribution. Journal of Sound and
Vibration 293, 535–546 (2006)
[9] Ehrenfried, K., Koop, L.: A comparison of iterative deconvolution algorithms for the
mapping of acoustic sources, AIAA Paper 2006-2711
Experimental Characterization of the Vibro-Acoustic
Behaviour of a Switch
Summary. The work presented in this paper is part of the SNCF project BPS
(noise from singular points), in which the noise emitted by steel bridges, joints
and switches are investigated. Switches and crossings are numerous in urban
inhabited areas. From a dynamic point of view, the switch is a complex track
component where many wheel/track interaction phenomena are combined. In
this complex physical context, the objective of the measurement campaign
presented in this paper is to identify and characterize the sources distributed
along a specific switch that contribute to the radiated noise. The measurements
were performed in December 2012 on a switch (and also on a reference section)
crossed by mixed traffic, equipped with accelerometers (for track and
ground-borne vibration) and microphones. Compared to the reference section,
the track characterisation by impact hammer excitation shows a lower stiffness
under the crossing nose and a higher stiffness in the 150-500 Hz frequency range
under the switch rail. The amplification due to the wooden sleepers is responsible
for a rolling noise increase in the same frequency range. The crossing nose and
the insulated joint are impulsive sources with a given power. For the pass-bys
generating a high noise level at the reference site, their respective noise
contributions are negligible.
1 Introduction
For conventional speeds between 50 km/h and 300 km/h, rolling noise due to rail and
wheel roughness interaction has been identified as the main railway noise source [1].
However, in urban areas where residents are the most exposed to railway induced
annoyance (including noise and ground-borne vibrations), rolling noise may not be
the most annoying noise source anymore. In those areas, specific singular points are
noted: level crossings, curves, joints, welds, crossings and switches.
The BPS project (standing for Noise from Singular Points) started by SNCF in
2010 aims for a better understanding of the phenomena involved in the noise
generated by steel bridges, joints and switches. In the early stage of the project,
bridges and turnouts were counted and classified in the region of Ile de France (Paris
and close suburbs) depending on the acoustic characteristics of the track section they
belong to. This classification is given in Table 1.
Table 1. Listing and classification of railway steel bridges and turnouts in Ile de France
More than half of the turnouts listed in inhabited areas are located in very noisy
track sections according to the acoustic maps of the French network. The problem of
steel bridges has been studied in previous European projects and mitigation measures
such as rail dampers were proven to be efficient [2].
The noise generated by railway joints was studied by Wu and Thompson in [3].
They assessed the noise radiated by a wheel crossing a joint using the TWINS
software and an equivalent roughness spectrum for the wheel/rail vertical interaction
as an input. They found an increase of the impact noise with speed at a rate of
20 log10 V. Depending on the joint geometry, the impact noise was found to be up to
10 dB higher than the rolling noise.
The first measurements performed within the BPS project were also dedicated to
joints. Considering mixed traffic over a given joint, they showed that the joint could
be more considered as source with a given power, whose influence on the overall
noise mainly depends on the rolling noise itself.
For the switch issue, analytical models for the track dynamics simulation are not
available. Most existing models for turnouts are time domain models based on Finite
Elements Methods for the track and on multi-body systems for the vehicles, and are
limited to low frequencies [4]. They are principally used for maintenance purposes, in
order to predict the impact of mitigation measures on the wheel/rail interaction force
(crossing nose with optimized geometry, stiffness adjustment, etc.). These models
indicated a large variety of complex phenomena involved in the wheel/rail interaction
force generation:
The tests were performed in December 2012 in the north east of France
(Saint-Hilaire-au-Temple) on a switch of 80 m length in a straight track. All
measurements performed for the track section with the switch were also performed for
a reference section in alignment, hundreds of meters away on the same track.
In the switch, several zones were identified as potential sources for noise and
vibration:
• Changes in the fastening system (from bi-bloc sleepers apart from the switch
to wooden sleepers along the switch),
• Changes in the rail type (UIC 33 apart, UIC 50 along the switch except UIC
60 along the switch panel),
• Clearance at crossing nose, contact jump from switch rail to stock rail,
• Numerous welds and an insulated joint.
In order to appraise the influence of these potential sources in the noise emitted at
pass-by, the dynamic behaviour of the track was first investigated. For each of these
points, the track receptance was measured with impact hammer excitations and the
transfer function between the rail (force) and the ground vibration was also measured at
8 m, 16 m and 32 m from the track. The soil was characterized by a Multichannel
Analysis of Surface Waves. The rail geometry was also measured in terms of roughness
at both sections and rail transverse profile along the switch/stock rail transition and
along the crossing.
At the same time, pass-by measurements were performed. Acoustic levels were
measured at 7.5 m and 25 m from the track (respectively at 1.2 m and 3.5 m height) in
front of specific ‘hot-spots’. Ground-borne vibration was measured in front of the same
hot-spots, at the same distances as mentioned above. Both reference and switch
sections were crossed by the same train. The traffic was mixed and always in the same
direction.
The following results focus on the noise measured at the crossing nose, at the switch
rail and at the insulated joint. The corresponding receptances are given in Section 3.1
and the acoustic results are presented in Section 3.2. A schematic view of the switch is
given in Fig. 1.
144 B. Faure et al.
3 Results
The vertical track receptance at mid span is given in Fig. 2 for the 10-250 Hz frequency
range (heavy hammer excitation) as well as for the 40-3150 Hz frequency range (light
hammer excitation).
-7 -7
10 10
Receptance modulus (m/N)
-8
10
-8 -9
10 10
-10
Reference track 10
Crossing Nose Reference track
Stock Rail Crossing zone
-9
Switch Rail -11
Switch panel
10 10
10 20 40 80 160 50 100 200 400 800 1600 3150
Frequency (Hz) Frequency (Hz)
Fig. 2. Track receptance at different positions over the switch section and at the reference section.
Left: Low frequency excitation. Right: High frequency excitation.
The low frequency spectra show a low track stiffness under the crossing noise with a
sharp resonance around 75 Hz. This softening is generally planned at track design
stages to reduce the impact loads experienced by the crossing nose when wheels pass
over the gap. At the left hand side of Fig. 2, the stiffness under the stock rail is
equivalent to the stiffness of the reference section, both being slightly lower than the
stiffness under the switch rail. As the wheels are crossing from the switch rail to the
stock rail in our case, this might also reduce the impact loads due to the contact jumps.
The high frequency spectra are more related to the rail vibration (and therefore
radiation) under a roughness-type excitation. Usual track resonances can be identified for
the reference section: ballast resonance around 100 Hz and pinned-pinned resonance
Experimental Characterization of the Vibro-Acoustic Behaviour of a Switch 145
around 900 Hz (UIC 50 rail). The rail pad resonance at 800 Hz is almost merged with
the pinned-pinned resonance due to the high stiffness of the rail pads (estimated at
750 MN/m with a continuously supported beam model). The legends Crossing zone and
Switch panel of Fig. 2 (right) refer to the rail sections opposite to the crossing nose and
the switch rail. The high receptance values between 150 Hz and 500 Hz for the switch and
the crossing are due to the wooden sleepers. The rail pad resonance frequency is moved
closer to the pinned-pinned frequency and the resonance of the sleepers between the
ballast and the rail pad stiffness occurs at 700 Hz. This resonance of the sleepers causes a
natural damping of the rail vibration around this frequency.
10 10
TGV TGV 120
8 8
-4 -4 60
-6 -6 50
85 90 95 100 80 85 90 95
L eq, Tp at reference section (dB) L eq, Tp at reference section (dB)
Crossing at 7,5 m Crossing at 25 m
14 14
140
FREIGHT FREIGHT
12 12
CORAIL CORAIL 130
MU TRAIN MU TRAIN
Δ L eq, Tp at switch section (dB)
10 10
TGV TGV 120
8 8
Train speed (km/h)
110
6 6
100
4 4
90
2 2
80
0 0
70
-2 -2
-4 -4 60
-6 -6 50
85 90 95 100 80 85 90 95
L eq, Tp at reference section (dB) L eq, Tp at reference section (dB)
Fig. 3. Extra noise level due to the crossing nose on both sides of the track
146 B. Faure et al.
For the large majority of the trains, the switch leads to higher Leq,Tp values, whatever
the switch part considered. The extreme values on the left hand side of the figures have
to be interpreted with care as the train speed was more than 10 % higher on the switch
section than on the reference section.
The influence of the crossing nose clearly decreases with increasing train noise at the
reference section, until a certain limit after which it almost vanishes (Fig. 3). At 7.5 m
from the track, the extra noise level is about 2 dB higher on the side of the crossing
compared to the opposite side. At 25 m, the extra noise level is more or less the same at
both sides of the crossing.
In front of the switch rail, the extra noise level seems to be more independent of the
train reference noise (Fig. 4). This statement cannot be fully supported by the curves as
there is a lack of data with a reference noise between the fastest MU Train group and
the CORAIL group. However, the analysis of the available data suggests that the noise
in front of the switch rail is more related to extra rolling noise due to the track dynamics
along the switch. This is supported by the pass-by spectra of Fig. 5 showing a strong
increase in the frequency bands where the switch receptance is higher than the
receptance of the reference track (see Fig. 2).
10 10
TGV TGV 120
8 8
-4 -4 60
-6 -6 50
85 90 95 100 80 85 90 95
L eq, Tp at reference section (dB) L eq, Tp at reference section (dB)
Switch rail at 7,5 m Switch rail at 25 m
14 14
140
FREIGHT FREIGHT
12 12
CORAIL CORAIL 130
MU TRAIN MU TRAIN
Δ L eq, Tp at switch section (dB)
10 10
TGV TGV 120
8 8
Train speed (km/h)
110
6 6
100
4 4
90
2 2
80
0 0
70
-2 -2
-4 -4 60
-6 -6 50
85 90 95 100 80 85 90 95
L eq, Tp at reference section (dB) L eq, Tp at reference section (dB)
Fig. 4. Extra noise level due to the switch panel at both sides of the track
Experimental Characterization of the Vibro-Acoustic Behaviour of a Switch 147
100 100
Reference section Reference section
90 Switch at 7.5 m 90 Switch at 7.5 m (opposite side)
80 80
PSD (dB/Hz)
70 70
60 60
50 50
40 40
30 30
50 100 200 400 800 1600 3150 50 100 200 400 800 1600 3150
Frequency (Hz) Frequency (Hz)
Fig. 5. Pass-by spectrum at 7.5 m in front of the switch rail compared to the reference section.
Left: switch rail side. Right: opposite side.
The insulated joint behaves more like the crossing, i.e. as a proper source with a
given power (Fig. 6 left). This power depends on the geometrical parameters of the
joint and also on speed. It was proved in [3] from simulations but also observed in the
“joint” part of the BPS project. The results presented in Fig. 6 right are SNCF
measurements performed in two different campaigns. The + label corresponds to a
specific joint with mixed traffic (J2), the other labels correspond to another joint (J1)
crossed by a test train with different speeds and two crossing directions ( ∇ for the
downward crossing and Δ for the upward crossing). The negative slope of the line with
increasing reference noise depends primarily on the joint geometrical characteristics
and secondarily on the speed.
The pass-by spectra measured in front of the crossing nose and in front of the joint
(not presented here) show an increase of the acoustic power at almost all frequencies,
which is characteristic of impulsive sources (reduced time leads to a broadband
frequency spectrum).
Joint at 7,5 m (insulated joint) Two classical joints (J1 & J2) at 7.5 m
14 14
140
FREIGHT Test train upward (J1)
12 12
CORAIL Test train downward (J1) 130
MU TRAIN Mixed Traffic (J2)
Δ L eq, Tp at switch section (dB)
10 10
TGV 120
8 8
Train speed (km/h)
110
6 6
100
4 4
90
2 2
80
0 0
70
-2 -2
-4 -4 60
-6 -6 50
85 90 95 100 85 90 95 100
L eq, Tp at reference section (dB) L eq, Tp at reference section (dB)
Fig. 6. Extra noise due to rail joints. Left: insulated joint within the switch (see Fig. 1). Right:
previous measurements performed on two classical joints.
148 B. Faure et al.
4 Conclusion
In this paper, the noise radiated by a switch was investigated in terms of equivalent
noise levels at pass-by, combined with track stiffness measurements. Among the
multiple sources that are suspected to be responsible for noise and vibration generation,
attention was focussed on the crossing nose, on the contact jump between the switch
rail and the stock rail, and on an insulated joint located at the end of the switch. Only the
direct route and the trailing move were considered. By comparing stiffness
measurements with equivalent noise levels at pass-by, the following results were
obtained:
• The crossing nose and the joint (being insulated or not) are impulsive sources
with a given power. Their influence on the overall noise is limited in the case
of noisy trains.
• The contact jump between the switch rail and the stock rail is not an
impulsive source.
• The specific track fastening under the switch (especially the wooden
sleepers) leads to track dynamics which tends to increase the rolling noise in
the 150-500 Hz frequency range (for the switch investigated).
In order to better understand the mechanisms involved in the switch, the data
obtained from this measurement campaign must be processed in a different way. Other
data collected should also be investigated in the near future. These include roughness
measurements and rail vibration measurements during pass-by at different locations
along the switch, as well as some sleeper vibration. Signal processing indicators such as
correlation functions might be also useful to quantify the contribution of different
sources to the overall noise.
References
[1] Mellet, C., et al.: High speed train noise emission: Latest investigation of the
aerodynamic/rolling noise contribution. Journal of Sound and Vibration 293(3-5), 535–546
(2006)
[2] Poisson, F., Margiocchi, F.: The use of dynamic dampers on the rail to reduce the noise of
steel railway bridges. Journal of Sound and Vibration 293(3-5), 944–952 (2006)
[3] Wu, T.X., Thompson, D.J.: On the impact noise generation due to a wheel passing over rail
joints. Journal of Sound and Vibration 267(3), 485–496 (2003)
[4] Kassa, E., Nielsen, J.C.O.: Dynamic interaction between train and railway turnout: full-scale
field test and validation of simulation models. Vehicle System Dynamics 46(1), 521–534
(2008)
[5] Pålsson, B., Nielsen, J.C.O.: Wheel-rail interaction and damage in switches and crossings.
Vehicle System Dynamics 50(1), 43–58 (2012)
Experimental Comparison of Maximum Length Sequence
(MLS) and Impact Hammer Methods to Evaluate
Vibration Transfer Functions in Soil
Summary. The purpose of this paper is to test the validity of a pseudo binary
sequence based method which is spectrally flat, and which is called MLS (for
Maximum Length Sequences) to assess the differences in the level of vibration
between different accelerometers, and evaluate key parameters that can be
changed to improve the results. Before checking the validity of the MLS method,
a description of the site where the test took place and a comparison of three
different methods to setup accelerometers in soil are given. Measurements are
also used to compute the improvement of signal to noise ratio as a function of the
number of hits. Comparison between the impact hammer and MLS method is
then analyzed. The main aspects of the signal processing used to analyze MLS
and impact hammer signals are presented. Some information on the type of
device used in the study and the primary characteristics of measured impulse
responses are also presented. A comparison of mitigation at different distances
between the MLS method (fixed parameters) and impact hammer method is then
given. A parametric study is then performed on each MLS key parameter: MLS
order, number of means, number of impulse responses analyzed, influence of
background noise and linear behavior of the electrodynamic shaker. Finally, the
variation of soil properties measured with the MLS and impact hammer methods
over five days is given.
data have a poor signal to noise ratio for large distances. Work performed by Singleton
[6] showed that MLS can be used to compute transfer functions, improving the signal to
noise ratio and the frequency range. This paper investigates the possibility of applying
the MLS method to railway infrastructures and particularly quantifies uncertainties
related to this type of method (key parameters, metrology, etc.). The first part of this
paper is dedicated to comparison between different accelerometers in soil and
improvement of signal to noise ratio with number of hits in the case of an impact
source. The second part is dedicated to comparisons of the MLS, impact hammer and
falling mass methods.
Gypsum Soil
blocks 10cm
adhesive 40cm
Steel spike
signals. For the impact hammer, each accelerance (power spectral density ratio between
acceleration and force) was computed. Mean values and coherences were then
computed over all hits. For the calculation of signal to noise ratio at hit n°n, equation
2.1 is used. Ssignal(i,t) is the time evolution of signal hit n°i and Snoise(j,t) is the time
evolution of noise of hit n°j.
⎛ n ⎞ ⎛ n ⎞
SNR(n) = RMS⎜ ∑ S signal (i, t ) ⎟ RMS⎜⎜ ∑ Snoise ( j , t ) ⎟⎟ (2.1)
⎝ i =1 ⎠ ⎝ j =1 ⎠
The MLS method is commonly used in acoustics to evaluate impulse response (IR) of
rooms. It is based on binary pseudorandom sequences that have an autocorrelation
function similar to a Dirac impulse. MLS sequences can be generated by a linear
feedback shift register and have a length of n = 2m -1, m being the order of the sequence
and n the number of samples of the sequence. For the improvement of the signal to
noise ratio, the MLS sequences are averaged. There are limits to the use of the MLS
method. For example, if the sequence is long and the sampling frequency is low
(3200 Hz), the measurement can be quite long (about ten minutes). There is another
problem: the impulse response measured by the system takes into account impulse
responses of the filters, shaker, cables, and amplifier so the measurement can be
sensitive to the system variations. More details on the MLS method and associated
computations can be found in [4, 6, 7].
3.1 Comparison
Fig. 2 shows differences between three mountings in terms of one-third octave band
acceleration levels (reference 10-6 m/s²) at 1, 2, 4, 8, 16 and 24 m from the concrete
block for the impact hammer source. It shows that below 5 Hz, levels are strictly
different. For all distances, between 5 and 20 Hz, levels of configuration (a) are the
same as the other ones. Above this frequency and up to 100 Hz, levels for the (a)
configuration are higher than for the other ones. Differences between configuration (b)
and (c) appear between 63 and 100 Hz for distances between 1 and 8 m. Beyond this
distance, results for configurations (b) and (c) are very similar.
just after 41 hits as opposed to 90 hits at 8 m. The improvement of signal to noise ratio
for accelerometers at 24 and 32 m is more chaotic but it is monotonic with the number
of hits. More hits can probably increase this signal to noise ratio.
Fig. 2. : Comparison of three different transducer mountings. (..) steel cube (a), (--) adhesive (b),
(-) spikes (c).
Fig. 3. Improvement of signal to noise ratio with number of hits. (-) 1 m, (--) 2 m, (..-) 4 m, (-.)
8 m, (o) 16 m, (^) 24 m, (*) 32 m.
Experimental Comparison of MLS and Impact Hammer Methods 153
The device used in this study is the same as the one used in [6]. The moving mass is a
1.48 kg ferromagnetic core driven by a car power amplifier. The total weight of the
device is 5 kg, and can be easily moved from one site to another. More information on
this device can be found in [4, 7].
A parametric study has been performed on key parameters of the MLS method: MLS
average (2, 4, 16 and 32 measures), MLS order (order 13/14/15), number of Impulse
Responses (IR) needed to compute mitigation between accelerometers (5/10 IRs). Two
other tests were performed on perturbation of the measurement by background noise
(rolling stock pass-by during measurement) and saturation of the electrodynamic
device [4, 7]. Results are given in Table 1. It can be seen that background noise has no
influence on the global vibration value, 5 means can be enough to ensure good
measurement, great differences can be found in terms of level generated, 16 means are
enough to ensure similar global values to 32 means, and an order of 15 is needed to
ensure good measurements. Results are given in Table 1.
Table 1. Differences in dB, global and max values in third octave band, for different
modifications of MLS parameters, over frequency range 4 – 315 Hz
4.4 Comparison of Impact Hammer, Falling Mass and MLS Methods with
Optimized Parameters
A study of MLS parameters lead to an optimization of order and number of means
within the MLS computation. Final measurements were performed with these
optimized parameters (Table 2) and were compared to measurements with an impact
hammer and a falling mass. In this experiment, the shaker is used in linear mode. Fig. 5
presents results of this comparison. Vibration mitigation graphs are in accordance with
those obtained using impulse methods. Differences in the 1 – 10 Hz frequency range
are due to bad coherence values [4]. At 16 m, for the 1-160 Hz frequency range,
mitigations for the MLS and impact hammer methods are less than 2 dB. For four
distances, when coherence is above 0.9 [4], results given by the three methods are
similar. This last measurement shows that MLS with optimized parameters gives same
results as the falling mass and impact hammer methods.
Distance [m] 4 8 16 24
MLS order 14 14 15 15
Number of means 16
Experimental Comparison of MLS and Impact Hammer Methods 155
Fig. 5. Attenuation comparison (-) falling mass, (--) MLS, (-o) impact hammer
Fig. 6. Measurement reproducibility. (-) MLS 05/25/2012, (--) MLS 05/20/2012, (-.) impact
hammer 05/20/2012, (-o) impact hammer 05/20/2012.
156 G. Coquel and A.K. Kengang
The two methods give very similar results, and two measurements performed over
five days are similar between 10 and 100 Hz.
5 Conclusion
This paper investigated a way to use the MLS method to evaluate vibration mitigation
in soils. It showed that acceleration measurements performed on a steel spike or in a
hole in the soil filled with gypsum block glue lead to the same results. It showed that
signal to noise ratio can be improved for long distances with more than 128 hits. A
parametric study led to an optimization of MLS parameters and a comparison of this
method with an impact hammer method and a falling method showed very similar
results. A final comparison of mitigation between accelerometers for the impact
hammer method and the MLS method over five days also showed very similar results.
The next studies will focus on the evaluation of force injected by an electrodynamic
shaker to improve signal to noise ratio for long distances and low frequency coherence.
References
[1] International Standard Organisation, ISO 14837-1:2005: Mechanical vibration -
Ground-borne noise and vibration arising from rail systems. Part 1: General guidance (2005)
[2] Houbrechts, J., Schevenels, M., Lombaert, G., Degrande, G., Rücker, W., Cuellar, V.,
Smekal, A.: Test procedures for determination of dynamic soil characteristics. RIVAS
(Railway Induced Vibration Abatement Solutions Collaborative) project – Deliverable 1.1
(2011), http://www.rivas-project.eu/
[3] Madshus, C.: Test procedures for determination of dynamic soil characteristics, ISO
committee TC108/SC2/WG8, ISO 14837 – 3.1 (project submitted to international survey)
(2013)
[4] Kengni-Kengang, A.: Détermination métrologique des fonctions de transfert “infra-bâti”
basé sur une source de vibration entretenue. Master degree thesis, Ecole Nationale
Supérieure d’Ingénieurs du Mans (2012) (in French)
[5] Scilab 5.3.0 – Open and free software released under CECILL license and developed by the
Scilab consortium - Digiteo, http://www.scilab.org
[6] Singleton Jr., H.L.: Vibration transfer mobility measurement using maximum length
sequences. In: Proc. 150th Meeting of the Acoustical Society of America/NOISE-CON
2005, Minneapolis, Minnesota (2005)
[7] Coquel, G., Kengni-Kengang, A.: Comparison of impact hammer and maximum length
sequence method to measure vibration transfer functions in soils. In: Proceedings of
Internoise 2013, Noise Control for Quality of Life, Innsbruck, Austria, September 15-18
(2013)
sonRAIL Web Tool – A New Web Application of the Swiss
Method on Railway Noise Calculation Released in 2013
Summary. The sonRAIL web tool represents a public application using the
Swiss method of railway noise emission calculation for vehicles and track
sections. It is intended to be used by engineers, planners and administrators for
the evaluation of specific situations or noise mitigation measures as well as for
exchange and enlargement of knowledge about railway noise in the community.
The sonRAIL method is the property of the Federal Office of Environment of
Switzerland (FOEN) and is hosted at https://sonrail.empa.ch. Access to the web
tool can be requested at sonrail@empa.ch.
1 Introduction
In Switzerland, railway noise emission has been calculated according to SEMIBEL [1]
method since 1990. In 2010, a new state of the art calculation method called sonRAIL
[2] was presented. The sonRAIL method is based on extensive measurements of
vehicles and tracks. A high accuracy in the results can be obtained by taking additional
parameters into account such as transfer functions of vehicles and tracks, individual site
parameters (e.g. rail roughness) and track types. Modular tools for sound emission,
propagation and environmental noise rating are available. The noise emission module
[3] and also the sound propagation module [4] are designed to simulate the fundamental
physical processes with high accuracy in the results. Thereby current situations as well
as the effects of different noise mitigation measures can be investigated. The entire
method allows precise calculation of noise impact levels by taking into account
relevant environmental conditions, track parameters and vehicle properties. The
emission levels are determined as sound power spectra for different source heights.
The sonRAIL Emission module has been released in 2013 as a free web application:
the ‘web tool’ [5]. This web tool grants access to the full functionality of the sonRAIL
Emission model and calculation method. The web application has been developed by
LCC Consulting in collaboration with PROSE and EMPA [6]. It is accessible after a
log-in procedure at https://sonrail.empa.ch.
Two different types of calculations can be performed with the sonRAIL web tool: on
the one hand detailed emission data of individual vehicles can be gathered and on the
other hand entire track sections with numerous vehicles and trains can be calculated.
The noise emission of entire track sections depends on the track properties such as rail
roughness and track decay rate, on the traffic mix and on train speeds. Trains can be
arranged individually. Alternatively predefined compositions such as TGV or ICE
trains can be used. Vehicles, which can be single locomotives, wagons of multiple
units, coaches or freight wagons, can either be taken from the data base or created from
scratch. Such extensions of the database are primarily only accessible to the data
owner. However they can be made available to other users too, after a verification
process.
Rail Roughness
Wheel Roughness
Transferfunction
track
Transferfunction
vehicle
Travel speed
Fig. 1. sonRAIL input parameters (left) and sound source calculation heights (right)
The purpose of the sonRAIL web tool is the calculation of vehicle and track noise
emission taking the relevant parameters into account (Fig. 1). The general calculation is
based on one-third octave spectral data of roughness, transfer functions and sound
power of sources (see Fig. 2). The results are sound spectra and sound levels at different
heights for vehicles according to the Swiss noise protection guideline (SLV) for track
sections. At the moment the focus is directed on pass-by operation only, standstill and
starting noise is optional.
In the menu “vehicles” the pass-by noise emission of rail vehicles can be
investigated according to EN ISO 3095 [7]. The noise emission of track sections with
side parameters and certain traffic mixes can be calculated in the menu “track section”.
Within a session a user can take vehicles from the data base or create it from scratch.
sonRAIL Web Tool – A New Web Application of the Swiss Method 159
Fig. 2. sonRAIL input scheme for primary and secondary noise sources [3]
Taking into account that default track parameters of the web tool represent typical
rail roughness spectra and track decay rates of the Swiss network, the results may differ
from similar track types in other countries. Railway tracks in Switzerland are
commonly equipped with stiff rail pads which lead to highly damped rails and less track
sound contribution compared to tracks with soft rail pads.
Sound emission of vehicles can be calculated for different travel speeds of conventional
and high speed railways. Relevant sound sources according to their contribution within
the corresponding speed range such as auxiliary noise, engine noise, rolling noise and
aerodynamic noise can be modelled. Rolling noise is calculated using the roughness
spectra of the wheel and rail. For every vehicle from the data base, a transfer function is
defined (Fig. 3). Different track types are characterized by their specific track decay
rate and track transfer function. For calculations either default or measured transfer
functions can be used.
The sonRAIL web tool provides a data base of more than 100 vehicle types mainly
from the Swiss operators, but also includes common trains from interoperable traffic.
Each vehicle is characterized by a median wheel roughness depending on the brake
160 C. Czolbe, J.-M. Wunderli, and F. Fischer
system, a vehicle transfer function and sound sources from auxiliary aggregates. The
vehicles are described by a short note and are linked to Wikipedia for further
information.
For a pass-by noise calculation the track type, roughness spectra and speed have to
be selected. Additionally, special situations such as bridges, curves and switches can be
investigated. Internal results are sound power spectra, which are provided as sound
level spectra at different heights (for sonRAIL calculation, see Fig. 1) or spectra at the
EN ISO 3095 measuring point 7.5 m from the centre of the track. As an example, Fig. 4
shows the calculated pass-by noise of a locomotive at 160 km/h. Results can be copied
to the clipboard for further use in spread sheets.
New vehicle types can be created from available acoustic data such as rolling noise,
transfer functions and sound source information. To derive sufficient input data from
measurement results an acoustic test specification [8] is available at the FOEN.
Designed vehicles can be stored within the user environment for further calculations or
train configurations. Furthermore validated acoustic data for new vehicle types can be
added to the default data base to make them available to the community.
sonRAIL Web Tool – A New Web Application of the Swiss Method 161
85
80
75
70
65
60
55
50
45
40
100
125
160
200
250
315
400
500
630
800
1000
1250
1600
2000
2500
3150
4000
5000
6300
8000
1/3 octave frequency [Hz]
Fig. 4. sonRAIL web tool vehicle noise emission spectra result Re460 loco at 160 km/h on track
type with UIC 60 rails and monobloc concrete sleepers [6]
The track calculation is the standard application for the web tool. If the traffic mix
consists of vehicles from the data base, only the track type is needed to run the
calculation. A track type is defined by rail type and sleeper type. The rail roughness can
be set to rough, average or smooth. The track decay rate depends on the rail sleeper
configuration. To investigate railway routes or multiple track sections with different
track properties or traffic mix a batch calculation can be done.
Track type and roughness condition have to be selected first. Trains can be
configured from the vehicle data base. The speed, train configuration and ratio of wheel
flats can be adjusted for each train separately. Results are in the form of sound emission
level spectra at different heights (for sonRAIL calculation) or spectra at the EN ISO
3095 measuring point 7.5 m from the centre of the track. To calculate railway routes a
list calculation will be implemented soon. In principle, sound emission levels according
to other railway noise calculation methods can be derived using a transfer function.
Hence the comparison to noise emission results of CNOSSOS [9], Schall03 [10] or
HARMONOISE [11] can be done if requested.
Results are presented as A-weighted pass-by levels and spectra at a distance of 7.5 m
distance or alternatively normalized to a reference distance of 1 m. A complete result
table with background data for calculation can be created by means of an Excel Export
button.
162 C. Czolbe, J.-M. Wunderli, and F. Fischer
Four categories describing the accuracy of the emission data are introduced. For
accurate measurements where all input parameters are known a standard deviation of
0.5 dB(A) is assigned, according to the uncertainty derived from the sonRAIL
measurement campaign. A second category with an uncertainty of 1.0 dB(A) is
introduced for high-quality calculations, for example performed with TWINS [12]. For
measurements and calculations with reduced quality, uncertainties of 2.0 and
3.0 dB(A), respectively, can be used. Every sound source in the emission database is
assigned to one of the four categories. For a pass-by of an entire train or even several
trains the resulting uncertainty is calculated according to the laws of error propagation
as shown in Equation (5.1), where s denotes the standard deviation of the resulting sum
of all sound sources, sn is the standard deviation of source n and Ln is the sound level of
source n [13].
s=
∑ (s ⋅ 10
n
0.1⋅ Ln
)2
(5.1)
∑10 0.1⋅ Ln
All calculation results in the web tool are categorized according to their accuracy.
Results with a standard deviation s ≤ 1 dB(A) are highlighted in green, results with a
standard deviation of more than 2 dB(A) are shown in red, indicating a reduced
reliability of the calculation results. Standard deviations between 1 and 2 dB(A) are
shown in orange (see Fig. 5 for vehicle calculation).
By use of the sonRAIL web tool free application a calculation method for railways
can be applied for different purposes such as:
1. Railway track emission calculation based on available data base
(currently 100 vehicles, 20 tracks)
2. Batch calculation of different track situations according to rail roughness or rail
dampers
3. Calculation of noise emission on bridges, at curves (non squeal), with wheel
flats and switches
4. Vehicle noise behaviour according to different track parameters, effectiveness
of noise mitigation measures
5. Vehicle sound emission according to available sound source properties and
transfer functions
Within a sonRAIL web tool user account (see Fig. 5) the vehicles and track types can
be obtained from the data base. Results can be calculated for individually parameter
variations and copied to the clipboard. For track section noise calculation trains have to
sonRAIL Web Tool – A New Web Application of the Swiss Method 163
be configured according to the given traffic mix and speed range. The documentation of
the results is contained in contents sound level and spectra tables and configuration text
files. Results are not stored after a session is closed. The calculation of individual
designed vehicles and track types by using specific input files will be possible very
soon.
Fig. 5. sonRAIL web tool user interface - vehicle noise emission calculation SBB Re 460 loco at
160 km/h on concrete monobloc sleepers and UIC 60 rails [6]
References
[1] SEMIBEL, Schriftenreihe Umweltschutz 116 (1990)
[2] Federal Office of Environment Switzerland: sonRAIL Project Documentation (2010),
http://www.empa.ch/plugin/template/empa/102455
[3] Thron, T., Hecht, M.: The sonRAIL emission model for railway noise in Switzerland. Acta
Acustica 96, 873–883 (2010)
[4] Hecht, M., Wunderli, J.M., Thron, T., Sehu, D.: sonRAIL - The new Swiss calculation
model for railway noise. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B.,
Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 515–522. Springer,
Heidelberg (2012)
[5] Czolbe, C., Koepfli, M.: AIA-DAGA, sonRAIL Webtool application (2013)
[6] sonRAIL web tool application hosted at EMPA, https://sonrail.empa.ch
[7] Standard EN ISO 3095: Railway Applications – Acoustics - Measurement of noise emitted
by railbound vehicles, prEN (2011)
[8] Federal Office of Environment Switzerland: Acoustic test specification to determine
sonRAIL input data, report (2012), http://www.bafu.admin.ch
[9] Kephalopoulos, S., Paviotti, M., Anfosso, F.-L.: Common Noise Assessment Methods in
Europe (CNOSSOS-EU) EUR 25379 EN Luxembourg Publications Office Of the
European Union, 180 pp. EC Joint Research Centre Institute For Health And Consumer
Protection TP 281 21027-Ispra VA, Italy (2012) ISBN 978-92-79-25281-5 (pdf)
[10] Schall03 1990, Richtlinie zur Berechnung der Schallimmissionen von Schienenwegen,
Guideline of BMVBS Germany, 1st Revision Draft Schall03 2006: 21.12.2006, 2nd
Revision Draft Schall03 2013: 12.04.2013 (1990)
[11] van Beek, A., Verheijen, E.: HARMONOISE Harmonised Accurate and Reliable Methods
for the EU Directive on the Assessment and Management of Environmental Noise (2003)
[12] Thompson, D.J., Janssens, M.H.A., de Beer, F.G.: TWINS: Track Wheel Interaction Noise
Software, theoretical manual (version 3.0). TNO report HAG-RPT-990211, Delft (1999)
[13] Probst, W., Donner, U.: Die Unsicherheit des Beurteilungspegels bei der
Immissionsprognose. Zeitschrift für Lärmbekämpfung 49(#3) (2002)
Concept for Measuring Aeroacoustic Noise Transmission
in Trains Derived from Experience Gained in Aircraft
Testing
1 Introduction
Cost-saving is an important factor in the transport sector. This includes
production/acquisition and total lifecycle costs. Low energy consumption is one of the
areas of cost-saving. In particular the use of new materials and lightweight
constructions is an interesting possibility for reducing the energy consumption of trains
and aircraft. However, these new construction concepts need new acoustical concepts
to ensure that passenger comfort is maintained and/or improved.
Various sources and source mechanisms play a part in the overall acoustical
performance of trains (see Fig. 1). In the high-speed sector (above 300 km/h [2])
aeroacoustic source excitation is one of the most important sources. Nowadays exterior
aeroacoustic noise can be and is measured in wind tunnels. Many different cases can be
studied easily and in most cases at low costs with scaled models in wind tunnels. But,
depending on the operating range of the real train, the conditions (pressure, temperature)
for the fluid in the wind tunnel have to be chosen carefully, with the consequence that
high demands are placed on the acoustic measurement equipment [3, 4, 5].
Fig. 1. Acoustic sources of trains (side view of Next Generation Train (NGT), source: DLR
(CC-BY 3.0))
However, for the measurement of transmission from aeroacoustic noise sources into
the passenger cabin, scaled models are difficult to implement, since not only the shape
but also the material characteristics of the scaled model need to be adapted to study
transmission effects. Mostly, this is impossible without there being some differences in
the acoustical behaviour of the train and the model. Therefore in-situ measurement
techniques are essential.
The aim of in-situ measurements is to:
- understand the source mechanisms;
- separate different sources from each other and
- understand transfer paths.
With this knowledge appropriate numerical models can be developed and/or adapted to
calculate the interior noise as early as possible in the design and construction process.
The concept for measuring aeroacoustic noise transmission in an aircraft will show
how to determine the source, the response of the fuselage and the noise radiation into
the cabin. It has already been used on a DLR flight-test campaign [1] and can be applied
to high-speed trains, as well.
As with an aircraft, the body shell of a high-speed train is excited by turbulent
boundary layer pressure fluctuations and responds to it with vibrations. The vibrational
energy is transmitted through the structure and results in noise radiation which depends
on the wavelength and the structure (see Fig. 2). This fluid-body interaction and its
energy transmission into the compartment are useful to study in order to separate this
excitation from the other main vibrational source: the basic rail-wheel interaction. The
physical differences between sources and the variation of operational conditions from
measurement to measurement influence the interior noise, and can be used to quantify the
amount of noise from different sources. The separation of the sources is the basis for
tightly focussed measurements for an effective reduction of acoustic transmissions.
Another advantage is derived from knowledge of the pressure fluctuations: the possibility
to estimate the acoustical behaviour of new materials used in the body shell of the train
during the design process without the need for a prototype to be built at this early stage.
Concept for Measuring Aeroacoustic Noise Transmission in Trains 167
With the in-situ measurements explained below, a detailed analysis of the acoustical
and hydrodynamic pressure fluctuations is possible [6].
Fig. 3. Windows with sensor layout for pinhole mounted pressure transducers [6]
168 J. Galuba and C. Spehr
To obtain the transfer path of the vibro-acoustic energy, accelerometers were placed on
the aluminium skin, the stringer and the frames (i.e. the support structure) of the
aircraft. In addition accelerometers were mounted on the sound-radiating lining inside
the cabin. To apply this concept to trains, accelerometers would have to be placed on
the structure and the lining at a specified cross section.
By varying the operational conditions the influence of different sources on the
vibrational behaviour of the train can be studied, as was done in the flight test
campaign. With these data a systematic analysis of the response of the carriage is
possible, this being not limited only to aeroacoustic noise.
The results from a performance test in the DLR Do728 cabin test facility are shown
in the annex.
Fig. 6. 2D-Coherence extracted from flight test at f = 1349 Hz, Ma = 0.78, FL350 [6]
An example of model evaluation for an aircraft TBL is shown in Fig. 7 [6]. The
extracted coherence lengths over frequency from the measurements are compared to
the Efimtsov and Corcos model. Especially in the low frequency range both models
disagree with the measured data. Therefore a new model was developed to describe the
TBL (DLR flight test in Fig.7).
Models for the description of excitation and transmission can be used for the case
where there is systematic mismatching of the acoustic parameters to reduce the level of
the transmitted acoustic energy. Measuring the transmission from the aeroacoustic
excitation to the passengers’ ears provides knowledge about main transfer paths and
sound radiation of the lining.
Concept for Measuring Aeroacoustic Noise Transmission in Trains 171
4 Conclusion
References
[1] Spehr, C., Hennings, H., Buchholz, H., Bouhaj, M., Haxter, S., Hebler, A.: In-flight sound
measurements: a first overview. AIAA Paper 2012-2208 (2012)
[2] Krylov, V.V.: Noise and vibration from high-speed trains. Thomas Telford (2001)
[3] Ahlefeldt, T., Koop, L.: Microphone array measurements in a cryogenic wind tunnel.
AIAA Journal 48(7), 1470–1479 (2010)
[4] Ahlefeldt, T.: Aeroacoustic measurements of a scaled half-model at high Reynolds
numbers. AIAA Journal (in press, accepted for publication May 6, 2013)
[5] Ahlefeldt, T., Quest, J.: High-Reynolds number aeroacoustic testing under pressurised
cryogenic conditions in PETW. In: 50th AIAA Aerospace Sciences Meeting including the
New Horizons Forum and Aerospace Exposition (2012)
[6] Haxter, S., Spehr, C.: Two-dimensional evaluation of turbulent boundary layer pressure
fluctuations at cruise flight conditions. AIAA Paper 2012-2139 (2012)
[7] Schewe, G.: On the structure and resolution of wall-pressure fluctuations associated with
turbulent boundary-layer flow. Journal of Fluid Mechanics 134, 311–328 (1983)
[8] Kokavecz, J., Spehr, C., Seemann, L.: Microphone array applications in cabins. In:
Internoise 2012, New York City, USA, August 19-22 (2012)
[9] Kokavecz, J., Spehr, C.: Microphone array technology for enhanced sound source
localisation in cabins. AIA-DAGA, Merano, Italy (2013)
[10] Castellini, P., Sassaroli, A.: Acoustic source localisation in a reverberant environment by
average beamforming. Mechanical Systems and Signal Processing 24, 796–808 (2010)
[11] Ehrenfried, K., Koop, L.: Experimental study of pressure fluctuations beneath a
compressible turbulent boundary layer. AIAA Paper 2008-2800 (2008)
[12] Haxter, S., Spehr, C.: Examination of the influence of flight altitude and speed on the
Efimtsov model parameters. AIAA Paper 2013-2028 (2013)
[13] Efimtsov, M.: Characteristics of the field of turbulent pressures at the wall of a boundary
layer. Sov. Phys. Acoust 28(4), 289–292 (1982)
[14] Corcos, G.M.: Resolution of pressure in turbulence. The Journal of the Acoustical Society
of America 35(2), 192–199 (1963)
172 J. Galuba and C. Spehr
Fig. 8. Results from conventional beamforming in the reverberant environment of the Do728
cabin test facility for three different array angles and a frequency of 793 Hz
Fig. 9. Results from average beamforming method for a loudspeaker on a seat at 769 Hz (left)
and a radiating panel as source at 793 Hz (right)
Innovative Measures for Reducing Noise Radiation
from Track
DB Systemtechnik GmbH,
Völckerstr.5, 80939 München, Germany
Manfred.Beier@deutschebahn.com
1 Introduction
In the years 2009-2011 a project was carried out within the German economic stimulus
program Konjunkturprogramm II (KPII) with the objective to investigate and
demonstrate the effect of innovative noise reduction technologies [1]. The focus was
mainly put on retrofit, infrastructure technologies that reduce the track emission
directly at or close to the source. The Federal Ministry of Transport, Building and
Urban Affairs (BMVBS) provided funding for performing acoustic measurements on
13 new and innovative technologies that were tested both individually and in
combinations. Altogether more than 80 measures using these technologies were
installed and tested.
The different technologies that were applied are shown in Fig. 1. The results are
presented in the following sections (with the exception of measures against bridge
noise, which are presented in a separate publication [2]).
Fig. 1. Noise mitigation measures and their primary locations investigated within the German
economic stimulus program: 1) Rail dampers 2) Rail lubricators against curve squeal 3)
High-speed grinding 4) Measures against bridge noise 5) Under-ballast mats 6) Foamed ballast 7)
Under-sleeper pads 8) Coated rails 9) Photovoltaic systems 10) Gabion walls 11) Sound barrier
additions 12) Low noise barriers 13) Friction modifier for rail brakes [1]
For noise abatement and noise prevention, the newly defined set of calculation rules
for quantifying the railway noise (Schall 03 neu [3]) will come into force soon in
Germany. In order for a new technology to be formally accepted as a mitigation
measure its noise reduction effect has to be assessed and a way for how it should be
accounted for in the noise calculation has to be established. The objective of KPII was
to support this step and for many of the tested noise reduction measures it meant
determining a correction value that can be subtracted from the predicted railway noise
when the installation of a measure is planned. This correction value should ideally be
applicable for all kinds of situations and installation conditions.
Given the nature of the economic stimulus program, the measures had to render a
benefit to as many residents as possible. Since this often implies densely built-up, urban
areas these requirements constituted a major challenge given the sometimes difficult
measurement conditions. In order to determine the effect of the installations, the
acoustic department at DB Systemtechnik developed a new measurement and
evaluation concept [4] and was entrusted with the supervision of the measurement
activities and reviewing of the results.
2 Measurement Concept
The measurement concept is based on a comparison of the noise emissions of the track
at the test section before installation of the measure with the situation after installation.
Additional measurements are performed in parallel at reference sections without
measures, which improve the comparability. This concept is illustrated in Fig 2. For
investigations focusing on the amelioration of the vibration situation, where the
Innovative Measures for Reducing Noise Radiation from Track 175
installation requires a complete rebuild of the track structure and therefore also a
change of excitation, the effect was determined only from comparing the results of the
test section with those of the reference section.
3 Results
Fig. 3. Influence of track decay rate before installation (averaged from 500 Hz to 2 kHz) on the
effect of rail dampers
176 M. Beier, T. Lölgen, and M. Starnberg
Fig. 4 shows a box plot diagram of the measured noise reduction for four of the rail
damper types. The data that is included in this analysis is the average result of the
different train categories that have been measured per test track. This has been done as
an attempt to avoid an uneven representation of the results since varying number of
train categories was measured at different sites and it is evidently the track condition
that has the largest influence on the results. The fifth rail damper was only measured at
one site and the results are not shown here.
Fig. 4. The measured noise reduction of four of the rail dampers. The box plots show the median
and the first and third quartiles. The whiskers correspond to the min and max values.
Based on these results it was determined that the correction values for the
installation of rail dampers or rail shielding when predicting railway noise should be
-2 dB and -3 dB respectively.
On two occasions, the influence of a coating of the rail web with a thickness of 5 mm
foaming was investigated. Such a coating is expected to reduce the rail vibration and as
a consequence the noise emission. Based on these two measurement results no noise
reduction effect could be established.
Rail lubrication reduces curve squeal and is also expected to reduce the growth of
certain forms of corrugation. In a curved track section, a friction modifier is applied to
the rail head of the inner rail and the inner rail flange of the outer rail. The new
calculation rules require a correction value for curves with small radius that increases
the predicted railway noise in order to compensate for curve squeal. For curves with
radius, 300 m < R < 500 m this correction value is +3 dB [3].
The purpose of the investigations was to assess whether this added correction value
could be ignored in the case that a rail lubrication system is installed. In the investigated
cases, the curve radius R was in 9 cases 300 m < R < 500 m and less than 300 m in one
case. Not all the investigations have been finished yet, because of pending necessary
optimizations of some of the installations.
The measurements consisted of airborne noise measurements before and after
activation of the systems. The results are displayed in the form of histograms of the
maximum levels of each train pass-by. An example is shown in Fig. 5.
Innovative Measures for Reducing Noise Radiation from Track 177
Fig. 5. Distribution of level occurrence before and after activation of the rail lubrication to reduce
curve squeal (Berliner Ringbahn, v>80 km/h, frequency range 2 kHz – 10 kHz) [1]
As can be seen in Fig. 5, the maximum levels before activation of the rail lubrication
were up to 108 dB(A). After activation the maximum level was reduced to only
84 dB(A). It was concluded that by installing rail lubrication no additional correction
for the compensation of curve squeal is required for the noise prediction of curves in the
range of radius that was measured.
Fig. 6. (left) Photo showing one tested low noise barrier type [1]. (right) Noise reduction effect
of low noise barriers of 55 cm height for different receiver positions and train categories.
The noise emitted from the nearest track was reduced by 2-3 dB, the noise from the
far track was reduced by 0-2 dB. The effect of low noise barriers strongly depends on
the height of the receiver point and which track is considered (next to the barrier or at a
greater distance). The best effects can be achieved by placing low noise barriers at each
side of the track, additionally - in case of a double track line - between the two tracks. In
this case the effect can reach values of 5 dB for freight trains and 7 dB for disk braked
trains (h=0.74 m, receiver at 25 m distance and 3.5 m height).
Besides low noise barriers, the following forms of measures on the propagation path
were investigated:
178 M. Beier, T. Lölgen, and M. Starnberg
Gabion walls increase the applications of noise barriers in urban areas since they
may be very well fitted within the urban area. Their effect is identical to the effect of
standard noise barriers. Benefits are, that they may be used as a biotope for small
animals and that they are unattractive to graffiti-abuse
Noise barriers with attached spoiler or diffraction aid may improve the shielding
effect of the sound barrier. This was measured at two locations in KPII. The results
showed that such extensions roughly had the same effect as a noise barrier of the same
height. Still the easily attachable system can be useful if for any reason the height of
existing barriers needs to be increased.
Photovoltaic systems were tested positive for feasibility on sound barriers. For the
given configurations no significant acoustic effect (positive or negative) could be
detected.
2 0,6 0,8
-0,6 -0,5
0 -1,2
-2,4 -2,0
-2 -0,7 -0,8
-1,8 -2,1
-4 -2,7
-3,9
-6
T.TVI 32( 3) Sh, creat ed
-8
20.08 .10
06.09 .10
04.10 .10
13.09 .10
06.12 .10
02.05 .11
09.05 .11
10.06 .11
04.10 .11
22.-25.11 .10
21.04 .11
28.04 .11
29.04 .11
12.05 .11
13.09 .11
22.09 .11
21.10 .11
09.-11.11 .11
03.12 .11
Date of measurement
Fig. 7. Result of HSG in relation to “specially monitored track” determined with the DB sound
monitoring car. Grinding date is marked with triangles [1].
Innovative Measures for Reducing Noise Radiation from Track 179
Fig. 8. Histogram and cumulative frequency distribution of beam rail brake before activation of
friction modifier (left) and after activation (right) [1]
Fig. 9. Insertion loss ΔLv of an under-ballast mat with a modulus cstat= 0.03 N/mm³ for regional
and freight trains (measuring distance 9.5 m from track center) [1]
180 M. Beier, T. Lölgen, and M. Starnberg
Under-sleeper pads reduce the contact forces between sleeper and ballast; thereby
they increase the elasticity of the track and reduce the emitted vibrations. In KPII
padded sleepers with a static stiffness of cstat = 0.25 N/mm3 (“hard”) and
cstat = 0.15 N/mm3 (“medium”) were tested. The measurements were carried out with
artificial excitation (shaker, dropping weight) as well as during train pass-by. The
results were different according to the method of excitation, see Fig.10.
Fig. 10. Third octave spectrum of the insertion loss of under sleeper pads (cstat = 0.15 N/mm3) for
artificial excitation (thick line) and for train pass-by (other curves) [1]
The artificial excitation leads to a positive insertion loss of 2.5 dB between 80 and
250 Hz for medium stiffness pads and 2.5 to 5 dB between 160 and 350 Hz for hard
pads. For train pass-by, insertion losses of 2.5 to 5 dB between 25 Hz and 125 Hz have
been measured for medium pads.
4 Conclusions
The technologies presented in this paper considerably expand the stock of available
conventional track related noise mitigation technologies. Future technologies, after
being formally recognized, can be designed more efficiently and resident-friendly for
the use in noise prevention and noise rehabilitation projects. The potential is
demonstrated for further development of the technologies concerning their acoustic
effects and the costs. The innovation process will continue.
References
[1] DB Netz AG: Innovative Maßnahmen zum Lärm- und Erschütterungsschutz am Fahrweg.
Schlussbericht (2012)
[2] Stiebel, D., Lölgen, T., Gerbig, C.: Innovative measures for reducing noise radiation from steel
railway bridges. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T.,
Thompson, D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 126, pp. 409–416. Springer, Heidelberg (2015)
[3] SCHALL 03 (2012) Richtlinie zur Berechnung der Schallimmissionen von Eisenbahnen
und Straßenbahnen; Entwurf, Stand 21.12.2006, Überarbeitung 2012, Stand 08.06.2012
[4] DB Systemtechnik: Mindestanforderungen an Nachweismessungen zur quantitativen
Bewertung von infrastrukturbasierten Innovationen zur Minderung des Schienenlärms.
Bericht (2010)
Innovative Noise Mitigation Measures in the Framework
of “Konjunkturprogramm II” in Germany
1 Introduction
The technologies have been installed in 82 test sites on existing railway lines; some
technologies were tested in combination. The measurements were conducted and
documented by different certified measuring institutes. The data was provided by each
institute as data sheets.
The results of the rail dampers/rail shields and the low noise barriers are presented as
example in the following.
2 Methods
The summarizing of the data for each technology was done in order to add the results to
the new German calculation method for railway noise (Schall 03 (2012) [5]). Therefore
the noise values had to be converted to octave bands from 63 Hz to 8 kHz.
The range of different train speeds (measurements between 80 and 160 km/h were
considered) at different test sites made it necessary to convert the noise levels to a standard
velocity of 100 km/h. After correction of the speed influence noise levels were averaged
over the number of usable pass-bys separate for each test site, track and type of train.
The effectiveness of a technology is calculated from the difference of the
measurement results of a test section before and after installing the technology.
Simultaneous measurements at a nearby reference section without installing measures
Innovative Noise Mitigation Measures in the Framework of “Konjunkturprogramm II” 183
were used for correction of other influences. According to the method of data
summarization the effectiveness of a technology was given for each type of train at
each test site.
After plausibility checks the effectiveness of a technology was averaged over the
different test sites. In case of different train models on the various test sites a
standardized emission spectrum for the train types was used for the calculation of the
A-weighted overall noise reduction effect.
Numerous test sites were equipped with rail dampers or rail shields. The scheme of
evaluating the thereby obtained measurement data is given in the following section:
Regarding the subsequent application of Schall 03 (2012) [5], in the first step the
noise levels listed in the measurement reports, given as TEL in third-octave bands, are
summarized in the octave bands by adding up the energy values of three third-octave
bands.
In order to allow a comparison between different test sites the spectra have been
converted to the reference speed of v 0 = 100 km/h. The conversion has been done in
accordance with Schall 03 (2012), where there is a frequency dependent velocity
coefficient b :
⎛ v train ⎞
Lv 0 = Lv ,train − b ⋅ lg ⎜⎜ ⎟⎟ (3.1)
⎝ 100km / h ⎠
The calculation method Schall 03 (2012) [5] uses different velocity coefficients b for
different types of sound sources (e.g. propulsion noise, aerodynamic noise). While
calculating the conversion, it is assumed in accordance with the measured speed range
(esp. no high-speed traffic), that the rolling noise predominantly determines the
acoustic level compared to other noise sources. The velocity coefficients b for rolling
noise are given in Table 1:
In the next step, the arithmetic mean values in octave bands are calculated separately
for each train type and the measurement before, LV and after installation, L N . Both
calculations were done for the test section (track section in which the measure has been
installed) and for the reference section (track section without the measure being
installed) in accordance with the test specification [1]. For this purpose, the associated
standard deviations are calculated in accordance with equation 11 a of DIN V EN V
13005 [2] sV and sN.
184 U. Moehler, M. Liepert, and A. Martens
u = sV2 + s N2 (3.2)
Table 2. Calculated length related sound power level in dB(A) of different train types at reference
speed of 100 km/h in octave bands, in accordance with Schall 03 (2012); these spectra are used for
calculation of an overall A-weighted noise reduction of a mitigation measure on each train type
Table 3. Overall noise reduction of different models of rail dampers and rail shields in dB(A)
Type of train rail rail rail rail rail rail
dampers dampers dampers dampers dampers shields
Model Model Model Model Model
No 1 No 2 No 3 No 4 No 5
ICE-Train -2 -1 -2 - - -
Intercity -1 -1 -2 -1 - -2
Passenger -1 -2 -2 -1 0 -3
Train
Suburban -3 - -1 -1 -1 -4
Train
Freight Train -1 -2 -3 -2 - -1
For the low noise barriers the reductions in average noise level were determined for
each test site of the KP II (cf.[8]). At each test site a different type of low noise barriers
was installed and tested. Next, based on Schall 03 (2012) [5] and the model of the
terrain at the respective measurement location, comparative calculations are made
between the measured and calculated noise reduction resulting from the low noise
barrier, in order to quantify the overestimation mentioned above.
The comparative calculations, see Fig. 1, established that by reducing the effective
height of the low noise barrier by approx. 30 % in the calculation method, this
overestimation can be compensated in a simple manner and based on the system of
Schall 03 (2012) [5].
Fig. 1. Sample comparison between values obtained by measurement and level reduction of the
low noise barrier calculated by taking a height value reduced effectively by 30 %
Innovative Noise Mitigation Measures in the Framework of “Konjunkturprogramm II” 187
The low noise barrier can thus be considered in Schall 03 (2012) [5] to be like a
conventional noise barrier taking two aspects into consideration:
- Approach of the inflection edge in the real distance to the track axis,
- Approach of an effective height over the top edge of the rail reduced by 30% with
respect to the real height.
5 Conclusions
The analyses of the measurements shows as result, that in the case of rail dampers, a
noise mitigation of about 2 dB(A) dependent on the type of train is possible. The effect
of rail shields showed a results of about 3 dB(A) in summary. The effect of low noise
barriers was modeled with regard to the German calculation method with the position
of the noise source of rolling noise on the top of the rail. In this case, the shielding -
effect of the low noise barrier is reduced comparable with a reduction of the height of
the barrier of 30 %. compared to conventional noise protection walls must be taken into
consideration.
References
[1] 08-P6835-TTZ112, Grundlegende Anforderungen an Nachweismessungen zur
quantitativen Bewertung von infrastrukturbasierten Innovationen zur Minderung des
Schienenlärms, DB Systemtechnik, Akustik und Erschütterungen - TTZ112 (July 22, 2010)
[2] DIN EN ISO 3095, Railway applications – Acoustics – Measurement of noise emitted by
rail-bound vehicles (November 2005)
[3] EN 15610; Railway applications - Noise emission - Rail roughness measurement related to
rolling noise generation (June 2009)
[4] EN 15461; Railway applications - Noise emission - Characterisation of the dynamic
properties of track sections for pass by noise measurements (January 2011)
[5] SCHALL 03: Richtlinie zur Berechnung der Schallimmissionen von Eisenbahnen und
Straßenbahnen; Entwurf, Stand: 21.12.2006, Überarbeitung 2012, Stand 08.06.2012 (2012)
[6] DIN V EN V 13005, Guide to the expression of uncertainty in measurement (June 1999)
[7] SCHALL 03 (1990), Akustik 03: Richtlinie zur Berechnung der Schallimmissionen von
Schienenwegen (Schall 03), Ausgabe (1990)
[8] Schlussbericht zum KP II Teil 2: Innovative Maßnahmen zum Lärm- und
Erschütterungsschutz am Fahrweg – Schlussbericht TEIL 2 – Maßnahmensteckbriefe, DB
Netz AG – I.N (June 15, 2012)
Indirect Method of Rail Roughness
Measurement – VUKV Implementation
and Initial Results
VÚKV a.s.,
Bucharova 8, Prague, Czech Republic
phamova@vukv.cz
Summary. The paper deals with one of the indirect methods of rail roughness
measurement – the axle-box acceleration measurement. This method has been
developed by British Rail Research in the 1980s [1]. VUKV is adapting this
method to the VUKV test car and to Czech railway conditions. First of all, the
influence of out-of-roundness of the axle bearing on the axle-box acceleration
was examined. After that, acceleration measurements were performed on
different types of tracks. The acceleration was converted to deflection and
compared with the direct measurement of acoustic roughness.
1 Introduction
Rail acoustic roughness can be measured by direct or indirect methods. Both have their
advantages and disadvantages. The first is more accurate but slow and suitable only for
short distance measurements. The second is not as accurate but quick and suitable for
long distance measurements and without the necessity of any exclusions. One of the
indirect methods is the axle-box acceleration measurement. Approximate rail acoustic
roughness is obtained from the measured axle-box acceleration.
This method has been developed by British Rail Research. Prose extended it by
including wheel roughness in 2010 [2]. The goal of this research is the adapting the
indirect method based on the axle-box acceleration measurement to the VUKV test car.
This is equipped with GP200 bogies and disc brakes. Its maximum velocity is 200 km/h.
Since acceleration sensor is mounted on the axle-box, authors wanted to know whether
and how the axle bearing out-of-roundness influences the measured acceleration. This
influence was examined. Roundness of new and old axle bearings (similar to the type of
bearings used in the measuring car) was measured. Each bearing was measured three
times with almost identical results. In Fig. 1, the measured out-of-roundness results are
shown for the new and old axle bearings. Also depicted is the acoustic roughness limit
for reference track for noise measurement according to TSI “noise” (TSI limit), [3].
A simplified mathematical model of the measuring car was made, using the
measured out-of-roundness as the input excitation. The model comprises rigid bodies
connected by kinematic and force linkages. Wheel-rail contact was considered as
elastic.
Results of computations predicted that out-of-roundness would influence measured
acceleration at higher speeds.
out-of-roundness [dB], ref 1e-6 m
20
-20
new
-40 old
TSI limit
-60
3 2 1
10 10 10
wavelength [mm]
3 Experimental Procedure
− the axle-box acceleration with the acceleration sensor KS80D (MMF, [5])
always fixed on the axle-box of the VUKV measuring car;
− the instantaneous velocity by GPS (Global Positioning System);
− the yaw angle between bogie and carbody by a wire sensor. This served as
rough information whether the car was located on straight track or in curve.
The measured acceleration was converted to deflection assuming that the signal is a
sum of sine functions. It was then compared with a direct measurement of acoustic
roughness on both rails.
Measurements were performed on tracks with the different roughness – a very fine
track (test ring), a main track and a corrugated track. Measurements on the main and the
corrugated track were performed during one day, unfortunately a half year after the
measurement on the very fine track. On the other hand, wheel roughness changes very
slowly, especially on the measuring car, and wheel profiles have been newly turned one
year before the first measuring day. During mentioned half year, the test car covered
approximately 5,000 km on test rings and 2,500 km on standard tracks. Wheel profile
surfaces were smooth to look and touch in both cases.
The axle-box acceleration was measured on a test ring. The testing train consisted of a
locomotive, the measuring car, a coach and another locomotive. An acceleration sensor
was fastened on the last wheelset of the measuring car.
The measurement was made at three different speeds – 60 km/h, 80 km/h and
140 km/h. In Fig. 2, the indirect and direct methods are compared. The direct
measurement is depicted in one-third octave bands. The results indicate that while the
agreement is not very good the amplitudes are very low.
1e-6*effective deflection [m]
5 5 5
indirect e
4 4 4
direct e
3 3 3 direct i
2 2 2
1 1 1
0 0 0
2 1 0 2 1 0 2 1 0
10 10 10 10 10 10 10 10 10
wavelength [mm]
a b c
Fig. 2. Comparison of direct and indirect method for very fine track: a) 60 km/h, b) 80 km/h,
c) 140 km/h; e means external rail, i means inner rail
The measurement was again made at three different speeds – 80 km/h, 120 km/h and
140 km/h. The indirect and direct methods were compared with almost the same result
as in a previous case.
The axle-box acceleration was measured on the corrugated track near Chocen town
(150 km from Prague). The corrugated track consists mainly of curves with small radii.
The testing train consisted of a locomotive, the VUKV measuring car and four coaches.
An acceleration sensor was fastened on the first wheelset of the measuring car.
Measurements were made at three different speeds – 20 km/h, 40 km/h and 80 km/h.
Three measurement positions close to each other were chosen (within 500 m). A
short, 6 m long, track section without corrugation surrounded by highly corrugated
sections was found during running of the train and was monitored with special care. A
subsequent inspection detected a renewed rail joint. Fig. 3 shows the time-dependent
acceleration signal during passage of the test train over the whole chosen track.
Comparison of both direct and indirect methods is depicted in Fig. 4. Acoustic
roughness measured by the direct method is given in one-third octave bands. The
wavelength of the corrugation was always between 100 and 150 mm and axle-box
acceleration revealed it very well. As one would expect, better agreement is indicated
for lower vehicle speed.
80 80 80
20 km/h 20 km/h 20 km/h
60 60 60
40 40 40
20 20 20
0 0 0
2 1 0 2 1 0 2 1 0
10 10 10 10 10 10 10 10 10
1e-6*eff. deflection [m]
80 80 80
40 km/h 40 km/h 40 km/h
60 60 60
40 40 40
20 20 20
0 0 0
2 1 0 2 1 0 2 1 0
10 10 10 10 10 10 10 10 10
80 80 80
80 km/h 80 km/h 80 km/h
60 60 60
40 40 40
20 20 20
0 0 0
2 1 0 2 1 0 2 1 0
10 10 10 10 10 10 10 10 10
wavelength [mm]
a b c
Fig. 4. Comparison of direct (--) and indirect method (-) for corrugated track: a) before “silent”
section, b) “silent” section, c) after “silent” section
80 80
20 km/h
40 km/h
1e-6*eff. deflection [m]
60 60 80 km/h
direct method
40 40
20 20
0 0
2 1 2 1
10 10 10 10
wavelength [mm] wavelength [mm]
a b
Fig. 5. Comparison of direct and indirect method (20 km/h, 40 km/h, 80 km/h) for corrugated
track and different speeds: a) before “silent” section, b) after “silent” section
80 80
1e-6*eff. deflection [m]
deflection
60 60 exponencial function
40 40
20 20
63.86*exp(-0.0199*v) 65.41*exp(-0.0176*v)
0 0
0 25 50 75 100 0 25 50 75 100
train speed [km/h] train speed [km/h]
a b
Fig. 6. Estimated roughness amplitude for wavelength of 132 mm and corresponding peak
values: a) before “silent” section, b) after “silent” section
Comparative measurements were carried out on a ballasted track and on a slab track.
The distance between measured positions was 300 m. The ballasted track has B91S
sleepers spaced 600 mm apart and pads with static stiffness of 55-75 kN/mm. Rails on
the slab track are connected to the slabs at regular intervals of 650 mm by pads with
static stiffness of 22.5 kN/mm.
The axle-box acceleration, the rail acceleration and the sound pressure level were
measured at both positions. The train speed was 120 km/h. Rail acoustic roughness was
low and the same for both cases.
One accelerometer was fixed on the measuring car axle-box; two accelerometers
were fastened on the rail bottoms (one for the ballasted track, another for the slab
track). In Fig. 7, measured accelerations are shown.
Moreover, the sound pressure measurement was performed according to ISO EN
3095 [7]. The monitored value was LAeq for the measuring car (from the front to the rear
Indirect Method of Rail Roughness Measurement – VUKV Implementation 195
buffer). For the ballasted track, LAeq was 91.2 dB(A) and for the slab track it was
100.8 dB(A), almost 10 dB higher than for the ballasted track.
Since the corresponding axle-box deflections for both cases were the same and low,
it can be concluded that the considered two types of track do not influence deflection
although other evaluated quantities do.
effective acceleration [m/s2]
15
axle-box
rail
10
0
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
frequency [kHz] frequency [Hz]
a b
Fig. 7. Effective acceleration: a) ballasted track, b) slab track
6 Conclusions
The axle-box measurements were conducted on tracks with different quality – a very
fine track (test ring), a main track and a corrugated track. Tracks with low acoustic
roughness give low amplitudes of accelerations and vice versa. The measured
corrugated track has corrugation with wavelengths of 100 – 150 mm – it is a typical
wavelength on the Czech tracks. The axle-box acceleration reveals this corrugation
very well even without knowledge of the wheel roughness.
The influence of out-of-roundness of the axle bearing was not determined during the
acceleration measurements, it will not be considered in future work.
The influence of type of track (ballasted, slab) on the axle-box deflection was not
significant; it will not be included into next evaluation.
7 Further Work
Since all existing evaluations and interpretations were done without knowledge of
wheel roughness, the use of special wheelsets and wheel roughness will be taken into
account in the follow-up work. Acceleration sensors will be fixed on each axle-box.
Measurements on tracks with different roughness wavelengths are also planned.
However, further work is mainly aimed at finding transfer function “rail roughness –
axle-box deflection”. To do this, measurement of the wheel point mobility using the
stand for measuring wheelset is intended. The measurement of rail mobility on the test
ring is also planned.
196 L. Phamová et al.
Acknowledgement. This research has been conducted with the support of the
Technology Agency of the Czech Republic, the project TA01031267 Railway noise
emission and vibration - “Novibrail”. This support is gratefully acknowledged.
References
[1] Lewis, R.B., Richard, A.W.: A new method for the routine measurement of rail head
corrugations. Rail International 17(2), 37–41 (1986)
[2] Czolbe, C., Bühler, S.: Erfassung der Schienenrauheiten durch indirekte Messung –
Ergebnisse vom Eisenbahnnetz der Schweiz. ZEVrail, pp. 32–39 (2010)
[3] European Union. COMMISSION DECISION of 23 December 2005: technical specification
for interoperability relating to the subsystem ‘rolling stock — noise’ of the trans-European
conventional rail system. In: 2005. 2006/66/EC, http://eur-lex.europa.eu/
LexUriServ/site/en/oj/2006/l_037/l_03720060208en00010049.pdf
[4] Thompson, D.: Railway Noise and Vibration, Mechanisms, Modelling and Means of
Control. Elsevier, Southampton (2009) ISBN 978-0-08-045147-3
[5] Beschleunigungsaufnehmer und Schwingungsmesstechnik von Metra. MMF. (online).
MMF. (cit. 1.5.2011), http://www.mmf.de
[6] MEAS Prog., MEAS Prog. (cit. May 1, 2011), http://www.measprog.com/
[7] ISO EN 3095. Acoustics – Railway applications – Measurement of noise emitted by
railbound vehicles (2013)
[8] Projekt Novibrail,
http://www.isvav.cz/projectDetail.do?rowId=TA01031267
Background for a New Standard on Pass-By
Measurement of Combined Roughness, Track Decay Rate
and Vibroacoustic Transfer Functions
1 Introduction
A measurement method for combined roughness, track decay rates and transfer
functions derived from rail vibration during a train pass-by was initially developed in
the late nineties [1]. This method was later implemented in software tools [2]
and applied in several countries for various purposes [3, 4, 5]. Due to the broad
application potential and the need for a common approach, standardisation work
has been undertaken since 2011 within the framework of CEN /TC256/WG3. This
resulted in a draft CEN Technical Report [6] in 2013, describing the method and
providing background information on benchmarks that could serve as a basis for a new
standard.
In this paper, the scope, the main elements of the method and some benchmark
results are presented. An energy iteration method is described which is used to
determine track decay rates, which are a key input to the method in addition to vertical
rail vibration.
The method is used to determine combined wheel-rail roughness and track decay rates
from rail vibration during the pass-by of a train. By combining vibration measurements
with sound pressure measurements from the same pass-by, a vibroacoustic transfer
function for rolling noise can be determined.
Combined roughness is a quantity that determines the level of excitation of
wheel-rail rolling noise. The transfer function can be used to characterise the
vibroacoustic behaviour of the vehicle-track system for a given roughness excitation in
relation to rolling noise. Besides measuring the main quantities of combined roughness,
track decay rates and transfer functions, the method can be used to characterise the
effect of noise control measures, to monitor wheel roughness and to separate rolling
noise from other sources, and for other purposes.
The method is not intended for approval of sections of reference track as specified in
the TSI [7], and has not yet been validated for non-standard wheel types such as small
wheels or resilient tram wheels, or for non-standard track types such as embedded rail
or grooved rail.
The main measured quantities are vertical railhead acceleration a(t) in [m/s2], sound
pressure level p(t) in [Pa] at a distance of 7.5 m from the track centreline, train transit
time tp [s], train length ℓ [m], number of axles Nax and train speed v in [m/s]. One third
octave spectra of the equivalent sound pressure level Lpeq,tp(fc) and railhead vertical
vibration Laeq,tp(fc) over the transit time tp and in frequency bands fc [Hz] are
determined.
The method derives vertical track decay rate Dz(fc) in [dB/m], combined effective
roughness LRtot(fc,v) or LRtot(λ) in dB re 10-6 [m] as a function of wavelength λ. A
procedure for converting from frequency to wavelength spectrum is described in the
report, using the relation λ = v/fc and correctly distributing spectral energy into standard
one third octave wavelength bands. The total effective roughness can be related to the
summated wheel and rail direct roughness, together with the contact filter.
The decay rate is determined according to the energy iteration method set out in the
following section. The lateral track decay rate can also be determined if lateral railhead
acceleration is also measured.
The method also specifies an acoustic pressure to roughness transfer function
LHpR,tot,nl(fc) in dB re 20 [Pa/m1/2]. This is theoretically independent from train speed,
number of axles and train length. The transfer function can also be expressed in terms
of sound power if required, which is proportional to the train length.
The effective combined roughness is derived from the vertical railhead vibration
Laeq,tp(fc) and vertical track decay rate according to
LRtot(fc,v)=Laeq,tp(fc,v)+10lg(Dz(fc)/(8.686Nax/ℓ))-A1(fc)–A2(fc)–40lg(2πfc) (2.1)
Background for a New Standard on Pass-By Measurement of Combined Roughness 199
The Nax/ℓ term normalises the equivalent vibration level to the number of axles and
train length. A1(fc) is the difference between the vibration spectrum measured at the
actual measurement position and at the railhead. It may be neglected below around
4 kHz. A2(fc) is the difference between the combined roughness and the contact point
displacement. It is tabulated in [6] for a range of wheel/rail combinations, and can also
be calculated from the wheel, rail and contact point receptances if required.
The total transfer function LHpR,tot,nl(fc) is determined from
Track decay rates can be estimated from pass-by rail vibrations in different ways, such
as slope estimation of the filtered level history or energy methods, see [6]. The energy
iteration method takes the contributions of other wheels into account and is less
sensitive to manual inputs as long as the wheel positions are correctly specified. This
method is analogous to the hammer impact method described in EN 15461 [8], with the
difference that the moving wheel provides the excitation instead of a hammer, and the
track has a real load.
The rail vibration amplitude due to a single wheel positioned at x=0 is assumed to be
described by an exponential function
where x is the position away from the contact point along the rail, A(x) is the vibration
amplitude along the rail, A(0) is the vibration amplitude at the position of the wheel
contact point and β is a decay exponent. The vertical decay rate Dz in dB/m can be
given as
This decay rate Dz is derived from the evaluation of the ratio R(fc) of the integrated
vibration energy over a length L2, potentially including the whole train pass-by
( AΣ2L2 ( f c ) ) versus the integrated vibration energy over a short length L1 directly
around the wheels ( AΣ2L1 ( f c ) ). L1 is taken as the shortest axle distance in the train (or
part of the train). A common minimum wheel distance is 1.8 m, in which case the
analysis length L1 extends from -0.9 m to +0.9 m around each axle position. The axle
position is determined by a wheel trigger signal or manually from the acceleration
signal.
AΣ2L1 ( f c )
R( f c ) = ≈ 1 − e − βL1 (3.3)
AΣ2L2 ( f c )
200 M.G. Dittrich, F. Létourneaux, and H. Dupuis
where:
1 − e βL1
∫ (A ( x ) e )
N N L1 / 2 N
A 2
ΣL1 =∑ A 2
=∑ − β | x − xn | 2
dx = ∑ A (x ) 2
(3.4)
β
n , L1 n n n n
n =1 n =1 − L1 / 2 n =1
( ) 1 − e − βL2
N N N
1
AΣ2L2 = ∑ ∫ An ( xn ) e ∑ A (x ) ≈ β ∑ A (x )
− β | x − x n| | 2
dx = 2 2
(3.5)
β
n n n n
n =1 L2 n =1 n =1
The approximation at the right hand side in equations (3.3) and (3.5) above is valid
for sufficiently large L2, e.g. a train length or the length of a (group of) vehicle(s). The
quantities A2ΣL1 and A2ΣL2 can be determined straightforwardly in each frequency band
from filtered measured acceleration signals (see Fig. 1). The time interval Tx which
corresponds to L1 contains mainly energy from the single wheel, but also contributions
from other wheels, particularly the nearby ones. Tx should be chosen slightly less than
the smallest distance between wheels over the whole train, to avoid overlap in energy
summation. From equations (3.2), (3.4) and (3.5) the vibration decay rate Dz is
determined from:
ln(1 − R( f c ))
8.686
Dz ( f c ) = − (3.6)
L1
The measured vertical rail acceleration level potentially contains contributions from
all wheels. So the pass-by slopes of the vibration level are also affected by these
contributions and need to be adjusted accordingly. This can be done by the following
iteration procedure:
1500
1500
Tx T T T 1000
x x x
Acceleration [m/s ]
500
2
1000 0
-500
-1000
Acceleration [m/s ]
500 T
2
-1500
0 1 2 3 4 5 6 7 8 9
Time [s]
-500
-1000
t t2 t t
-1500 1 3 4
2.4 2.5 2.6 2.7 2.8 2.9
Time [s]
Fig. 1. Top right inset: Unfiltered time signal of vertical rail vibration of whole pass-by with total
integration time T (corresponding to length L2) indicated. Main figure: selected part of time
signal indicated in top figure showing integration times Tx corresponding to lengths L1 and times
t1 to t4 corresponding to the positions of wheels.
Background for a New Standard on Pass-By Measurement of Combined Roughness 201
• A formula for estimating the decay exponent β(fc) at each frequency band is based
on equations (3.2), (3.3) and equation (3.6). The vibration energy ratio R(fc) is
determined for the whole train pass-by or selected part of it as in eq. (3.5), and
multiplied by N/wk(fc), where N is the number of axles and wk(fc) a weighting
coefficient.
ln(1 − R ( f c ) N / wk ( f c ) )
βk ( fc ) = − (3.7)
L1
A starting estimate for the initial decay exponent β1(fc) is obtained with w1=2N. This
initial condition w1=2N assumes that half the vibration energy at each wheel is from
other wheels.
• The subsequent iterations are then calculated until the decay exponent βi(fc) is stable
to within 0.5 dB :
N N
wk ( f c ) = ∑∑ e
− 2 β k −1 x j − xi
(3.8)
j =1 i =1
where xj-xi is the distance between the current wheel j and another wheel i. The
weighting coefficient wk represents a sum of the squared contributions from all wheels,
viewed from each wheel and then summated over all wheels. If the decay exponent is
large, the effect of adjacent wheels is small and wk quickly converges to wk=N. If the
decay is small then wk becomes larger. Sufficient convergence is often achieved in
around 5 steps. A flow diagram is provided in [6].
5 Benchmark Tests
As part of the preparation of the CEN report [6], a series of benchmark tests was
performed on pass-by data of different trains on different tracks from Germany, France,
Switzerland, Austria and the Netherlands. Repeatability, reproducibility,
representativeness (comparison with direct measurement), averaging and uncertainty
were analysed and sensitivity for inputs such as train speed, train type, accelerometer
position and user inputs were investigated and reported. An example of spread in
vertical decay rates from pass-by measurements of different trains at varying speeds,
and a comparison with a hammer measurement is shown in Fig. 2. Despite the
difference in loading conditions of the track, the results show a good correspondence
between the direct and indirect methods. Variation of combined roughness due to
accelerometer position on the rail is shown in Fig. 3, together with the spread at one
position for a series of pass-bys. The spread (see envelope curves T3- and T3+) is less
than +/- 3 dB for wavelengths above 1 cm. A similar trend is found for measurements at
different train speeds.
Fig. 4 illustrates transfer functions for different pass-bys of a test train at different
speeds. The spread in these results is less than +/- 2 dB over a wide frequency range.
Given the spread found, averaging over at least 3 pass-bys is recommended to obtain a
representative mean value, for decay rates, combined roughness and transfer functions.
202 M.G. Dittrich, F. Létourneaux, and H. Dupuis
Fig. 2. Vertical track decay rates determined from multiple train pass-bys, compared with
directly measured track decay rate (AEF Impact) and TSI vertical limit (lower curve). Data from
French track, different trains at varying speeds.
Fig. 3. Combined roughness determined for different measurement positions on the rail, for a
single mixed test train pass-by at 80 km/h on a German ballast track with monobloc sleepers
Background for a New Standard on Pass-By Measurement of Combined Roughness 203
Fig. 4. Total transfer function for several pass-bys and two speeds, mixed test train on German
ballast track with monobloc sleepers. Legend: pass-by code, speed in km/h, meas. Position.
6 Conclusions
Main elements for a new standard for measurement of combined roughness, track
decay rates and transfer functions from train pass-bys have been presented and
documented in a draft CEN report [6]. The processing steps to determine each
of the quantities and recommendations for dealing with uncertainty have been
set out. Measurement uncertainty in the combined roughness spectrum can be
estimated at +/- 3 dB in third-octave bands, assuming averaging over at least
3 pass-bys. The method allows the determination of the main acoustical wheel/track
characteristics of normal service traffic using simple measurement equipment.
References
[1] Dittrich, M.G., Janssens, M.H.A.: Improved measurement methods for railway rolling
noise. Journal of Sound and Vibration 231(3), 595–609 (2000)
[2] Janssens, M.H.A., Dittrich, M.G., de Beer, F.G., Jones, C.J.C.: Railway noise measurement
method for pass-by noise, total effective roughness, transfer functions and track spatial
decay. Journal of Sound and Vibration 293, 1007–1028 (2006)
204 M.G. Dittrich, F. Létourneaux, and H. Dupuis
[3] Dittrich, M.G.: Track decay rate measurements using the PBA technique. In: Proceedings of
Euronoise 2006 (2006)
[4] Jansen, H.W., Janssens, M.H.A., Dittrich, M.G.: Practical applications of total wheel/rail
roughness measurements during train pass-bys. In: 9th IWRN, Munich, Germany
(September 2007)
[5] Jansen, H.W., Dittrich, M.G.: Separation of rolling noise and aerodynamic noise by
in-service measurement of combined roughness and transfer functions on a high speed slab
track. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T.,
Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 118, pp. 479–486. Springer, Heidelberg (2012)
[6] Dittrich, M.G., et al.: CEN TC256/WG3 Work Item 256538: Measurement method for
combined roughness, track decay rates and transfer functions. Draft CEN Technical Report
(April 2013)
[7] 2011/229/EU, Commission Decision of 4 April 2011 concerning the technical specifications
of interoperability relating to the subsystem ‘rolling stock – noise’ of the trans-European
conventional rail system (2011)
[8] EN 15461:2008+A1:2010: Railway applications - Noise emission - Characterisation of the
dynamic properties of track sections for pass by noise measurements
Monitoring Rail Condition Based on Sound and Vibration
Sensors Installed on an Operational Train
1 Introduction
The interaction between wheel and rail is one of the main causes of noise generation
when a train is passing by. A railway track typically consists of various components
such as insulated joints, level crossings etc. Proper maintenance of the track, including
its various components is important from a safety and reliability point of view. This is
important as noise emission can be reduced significantly by an optimized maintenance
strategy. Typically, defect detection related inspection and maintenance work is carried
out in a routine manner (and as-per-need basis) periodically using a measurement train.
This approach is costly and damages can occur in between measurement campaigns.
Banedanmark’s noise mitigation approach focuses on reducing the noise at the
source. To investigate how the noise depends on the track components and their
condition, a system capable of collecting and analyzing noise and vibration data
measured from a train in running operation is desirable. Such system can also be used to
detect critical wear and defects of the track and thus support the planning of
maintenance work and development of new tracks with lower noise emission.
Unlike the periodic inspection that is currenly state of the art; in recent times
researchers have suggested monitoring of railway tracks using sensors installed on
in-service trains [1, 2]. Work presented in this paper presents a new strategy for
continuously monitoring the railway track using acoustic and vibration sensors (along
with a GPS) installed on an operational train. The main idea behind this project is that
each track component (insulated joints, level crossings etc.) generates a specific sound or
vibration signature characteristic of its own, which might change in case of deterioration
or unusual conditions. Similarly, a damaged section of the track, e.g. due to corrugation,
will also have a characteristic signature different from the track under normal conditions.
Thus by monitoring the track continuously, a change in the signature over time can be
analysed for defect initiation and other such purposes. In this paper the following aspects
of the project are presented: railway track monitoring strategy, measurement campaign
and data acquisition, data analysis and results achieved so far.
2 Measurement Campaign
One of the major challenges in this project is to design and install a sensor system that is
sensitive to the changes occurring in the railway track and also sufficiently robust to
withstand the operating environment. This section describes the measurement
campaign designed on the basis of these considerations.
For the measurement campaign, an InterCity passenger train is rented and one of its
middle bogies is instrumented with various sensors. There are two surface microphones
(B&K Type 4949) underneath the floor. Four hydrophones (B&K Type 8104) are
mounted on the special bearings near the contact point between wheel and rail. A
tri-axial accelerometer is mounted on each of the four axleboxes of the bogie and two
more uniaxial accelerometers are mounted close to the center of the axle shafts.
Additionally, there is an outdoor antenna for the GPS system mounted on the roof. The
reason for the choice of so many sensors, and their installation locations, is to ensure
that all necessary information regarding the acoustic and vibration phenomenon, which
can be useful for the purpose of continuous monitoring of rails, is captured sufficiently.
The ultimate goal is to design a system having an optimized number of sensors and it is
hoped that this study will help in achieving this aim.
The data from the 22 sensors are acquired at a sampling rate of 50.1 kHz using a
National Instrument data acquisition system. Fig. 1 shows the measurement setup along
with the sensor location. The instrumented train is run multiple times (and on multiple
3 Method
The monitoring task in this project is divided into three major tasks: Data management,
event detection and event classification.
For each run of the measurement campaign suggested above, a huge amount of data
is collected. The first task is to manage this huge amount of data. This is followed by
the event detection phase. Note that not all of the data collected are of interest, only the
data segments that are important from monitoring perspective are of interest. Such
segments are e.g. ones specific to rail track elements that are being monitored and
segments specific to defects in the rail (such as corrugation). The event detection task
involves filtering these segments of interest from the collected data.
The goal of the event detection task is only to identify whether a data segment is of
interest or not, it does not label the detected segments (e.g. whether the detected event
segment pertains to a crossing or a switch or a corrugated section). This task is carried
out in the third stage; Classification stage. These three stages are now explained in
more detail in the following sections.
In the data management task of the project, the data are arranged for easy accessibility.
This requires, amongst other tasks, converting the data to a common format, arranging
it in a database using the associated metadata (run number, train speed, sensor groups
etc). This task also involves preparing training and validation datasets to be used in the
event detection and classification stages. At the current state of the project, efforts are
devoted to examining the acoustic signals. The reason behind this choice relates to the
human capability to recognize sound events by listening to them (not all but still many
events can be identified and related to rail track based on listening). Thus, the signals
from a surface microphone are used to build the training and validation sets. Different
portions of the signals have been selected and then classified by audio testing. The
considered categories are: ‘Non Event’, ‘Shock’, ‘Crossing’, Squeal’, ‘Other’.
Fig. 2 shows typical time-frequency representation of data segments pertaining to
these categories. Squeal noise is not so difficult for humans to distinguish from other
events because of its high frequency content. Shocks are defined here as a short impulse
in signal. It is difficult to relate this event to a specific track component, because it
seems that many track components can generate this type of sound, like isolator joints
or a bad welding for example. Crossings generate a sound very close to shocks, but
audio training helps to distinguish these two types of events in most cases. The
208 T. Jensen et al.
short-term Fourier transforms displayed in Fig. 2 show that the crossing noise has a
slightly broader frequency range than the shock noise. In general, this is what has been
noted for other crossings and shocks, however with a few exceptions.
Fig. 2. Examples of short-term Fourier transforms for each event type: from left to right: a
crossing event, a shock event, a squeal event, a ‘no event’ case. Speed for all is around 60 km/h.
The training and validation datasets based on the above mentioned five categories
are prepared for a subjective evaluation. Five listeners classified the chosen data
segments as belonging to any of the five categories in a listening experiment. A data
segment is assigned a certain category if four out of five listeners agree on that
particular category; otherwise the data segment is excluded from the training and
validation databases. This is done to ensure that the training and validation sets are as
accurate as possible. However, this rule is relaxed for crossings because their
localization in the signals is accurately known (from GPS based alignment). The GPS
based alignment method for localization of other events was not precise and hence the
jury based subjective evaluation method was relied upon for preparing the training and
validation datasets.
The purpose of the event detector is to detect any transient events or abnormalities, but
without expressing what type of event it is. This problem can be approached as a 2-class
classification case: one class for the events, the other for the non-events. Different
techniques exist for this approach [3]. The technique presented here is based on
short-term Fourier transforms (STFT).
The STFT plots displayed in Fig. 2 show that the transient events can be visibly
detected; these events appear through a sudden burst of acoustic energy, spreading over
a wide range of frequencies. The proposed event detection technique exploits this
sudden increase of energy, visible in spectra, to detect events [4]. For this purpose, the
difference in the spectra at two consecutive instants is considered and the difference
vector is summed up over frequencies to obtain a new function G as a function of time.
Notice that only positive differences are considered, in order to detect the start of the
event. The process is illustrated in Fig. 3. Function G provides positions of possible
events, as illustrated below.
Monitoring Rail Condition Based on Sound and Vibration Sensors 209
3.3 Classification
The events evaluated through the subjective evaluation were classified by Fisher’s
Linear Discriminant (FLD) [6]. FLD maximizes the distance between the classes, and
minimizes the size of the inter-class scatter during its training. It is a linear classifier
and designed to distinguish two classes originally. For multi-class classification
problems, two approaches can be used, i.e. one-versus-all and one-versus-one. The
one-versus-all strategy constructs a FLD for each class, which distinguishes the target
class from remaining classes. In contrast, the one-versus-one strategy makes a FLD for
each pair of classes, which distinguishes samples in the target class from the ones in the
other class in the pair. Mediante [7] showed that the one-versus-all strategy performs
better than one-versus-one strategy for the classification of traffic noises including
noise from trains, and therefore the one-versus-all strategy has been used for the study.
In [7], a number of features were compared in terms of their performance on
classification, and concluded that Mel Frequency Cepstral Coefficients (MFCC) results
in the best classification rate for traffic noises. In Fig. 2 the spectrogram of the squeal
210 T. Jensen et al.
event shows at higher sound pressure level (SPL) at high frequencies compared to low
frequencies. The opposite happens for the shock event and the crossing event. For this
reason, the first four MFCC coefficients, out of the ten calculated, are used as features
together with the low-frequency SPL and the high-frequency SPL. The cut-off
frequency of the calculated SPLs is 1 kHz. The selected six features are calculated for
each event under investigation.
The dataset having the shock and the squeal events is further divided into two groups
for training and validation. In one condition, the 120 km/h (maximum operating speed)
dataset is used for training and the 60 km/h dataset is used for validation. Later, the two
datasets are swapped. For the crossing events, an equal number of events is used for
training and validation. The 60 km/h dataset has 32 shock events and 8 squeal events,
and the 120 km/h dataset has 55 shock events and 8 squeal events. There are 110
crossing events divided into two sets for the 60 km/h and the 120 km/h dataset.
As described in section 3.2, two classes are considered for the event detector: ‘Event’ –
‘No Event’. ‘Event class’ has segments containing shock, crossing and squeals.
Segments with no audio event are classified as ‘No Event’.
The data are split in two parts defining the training (80 % of the samples) and the test
validation sets (20 %). To get a good estimate of the event detector performances, the
split is performed randomly many times and for each of these training sets, and a ROC
curve is built. Fig. 4 (left) shows good consistency of the ROC curves for all the
different training sets. For a 0 % False Positive Rate (FPR) [5], about 80 % True
Positive Rate (TPR) [5] is obtained. Similarly, a TPR of about 90 % is obtained for a
10 % FPR. It is also observed that for a set target of 10 % FPR (for example), the
threshold for the function G is consistent over the different training sets (Fig. 4, right).
This demonstrates the robustness of the technique.
The performance of the event detector is now tested on the validation datasets. For
this purpose, based on the training dataset, a threshold value corresponding to a certain
FPR and TPR value is chosen as the event detector. The criterion for judging the
performance of this detector is based on the fact that it should result in same values of
TPR and FPR for the validation dataset as is the case with training dataset. In other
words, performance of the detector (corresponding to chosen threshold value for 10 %
FPR and 90 % TPR in training dataset) is good if it returns a 10 % FPR and 90 % TPR
on the validation dataset as well.
It is observed that a threshold value of 0.023 (corresponding to 10 % FPR and 90 %
TPR in training data) results in almost similar values of TPR and FPR for the validation
data, hence illustrating its good performance as an event detector. Further, as indicated
by dark dots on Fig. 4(Left), different test validation datasets show values close to 10 %
FPR and 90 % TPR, highlighting the good (and robust) performance of the chosen
threshold and hence the event detector technique.
Monitoring Rail Condition Based on Sound and Vibration Sensors 211
Fig. 4. Left: ROC curves for 60 different training datasets (grey curves). The black dots
correspond to FPR and TPR obtained on testing sets for a threshold of around 0.023. Right:
thresholds (around 0.023) obtained from ROC curves for a FPR of 10 %.
4.2 Classification
When using the 120 km/h dataset for training, 88 % of the squeal events, 84 % of the
shock events, and 84 % of the crossing events are correctly classified for the 60 km/h
dataset. However, when the 60 km/h dataset is used for training, the classification rate
decreases significantly to 50 % of the squeal events, 38 % of shock events, and 84 % of
the crossing events. This may be caused by the fact that the 120 km/h dataset contains
all operating conditions while the 60 km/h dataset does not. Notice that both 60 km/h
and 120 km/h are the maximum operating speed. This means that assuming the training
dataset contains all operating conditions for the validation dataset, the designed
combination of FLD and the features are able to classify the train track noises, i.e.
shock, squeal, and crossing, with a classification rate higher than 84 %.
During preliminary investigation, other techniques like Gaussian Mixture Model
(GMM) are also investigated for the classification, but the results are comparatively
inferior. They seem to be sensitive to unequal number of samples for the three event
types. For GMM, samples for each class should be possible to be modeled as a
Gaussian distribution and the number of samples for the squeal and shock events may
not be enough to fulfill the requirement. Therefore, in the next measurement campaign,
a reasonable number of events for each class will be recorded, and the comparison
among the classifiers will be evaluated again.
5 Conclusion
To investigate how the condition and superstructure of a rail track impacts the
wheel/rail contact noise, a system consisting of various acoustic and vibration sensors
has been mounted on an InterCity train and the sensor signals have been acquired along
with position data from a GPS-system. To analyze the huge amount of data a two step
classification approach has been developed. The purpose of the first step is to detect
events of interest e.g. when the train passes a crossing and etc. The second step
classifies the events in various classes. The classification step is based on Mel
Frequency Cepstral Coefficients as features and Fishers Linear Discriminant as
classifier, and this combination results in a classification rate higher than 84 %.
212 T. Jensen et al.
Thus, it can be concluded that the system is able to extract information about a
railway track from rail-wheel contact noise recorded with sensors on an InterCity
passenger train. Only acoustic signals from surface microphones are studied and
signals from the other sensors are to be analyzed in future work. By optimizing the
setup for the next measurement campaign and with further development of the data
analysis tools it is believed that more information about the track component condition
and its impacts on wheel/rail contact noise can be obtained.
References
[1] Lee, J.S., Choi, S., Kim, S.S., Park, C., Kim, Y.G.: A mixed filtering approach for track
condition monitoring using accelerometers on the axle box and bogie. IEEE Transactions on
Instrumentation and Measurement 61, 749–758 (2012)
[2] Mori, H., Tsunashima, H., Kojima, T., Matsumoto, A., Mizuma, T.: Condition monitoring
of railway track using in-service vehicle. Journal of Mechanical Systems for Transportation
and Logistics 3, 154–165 (2010)
[3] Lozano-Angulo, J.A.: Detection and one class classification of transient events in train track
noise. Master’s Thesis, Technical University of Denmark, Denmark (2012)
[4] Masri, P.: Computed modelling of sound for transformation and synthesis of musical signal.
Ph.D. dissertation, University of Bristol, UK (1996)
[5] Fawcett, T.: An introduction to ROC analysis. Pattern Recognition Letters 27, 861–874
(2006)
[6] Ye, Q., Zhao, C.X., Zhang, H.F., Chen, X.B.: Recursive ‘concave–convex’ Fisher Linear
Discriminant with applications to face, handwritten digit and terrain recognition. Pattern
Recognition 45(1), 54–65 (2012)
[7] Mediante, E.C.: Sound recognition techniques: application to city noise. Bachelor thesis,
Technical University of Denmark, Denmark (2012)
Transposition of Noise Type Test Data
for Tracks and Vehicles
Summary. New railway vehicles in Europe have to comply with noise limits as
defined in the Technical Specification for Interoperability (TSI). The pass-by test
has to be carried out at a site that fulfills the TSI requirements. The need has
arisen to be able to perform TSI pass-by type testing at non-compliant sites. In
the ACOUTRAIN project, procedures have been proposed for the transposition
of pass-by measurement results for vehicle and track. This allows for
transposition of pass-by data obtained at a non-compliant site to a TSI compliant
track, or to another vehicle on the same track. Transposition of track or vehicle
requires source separation of the rolling noise contributions of vehicle and track
to the total pass-by level. Several methods will be discussed for this purpose,
based on combinations of measurement results and calculations. A transposition
method is illustrated and validated by a practical case, in which both the
reference and the transposed situations have actually been measured.
1 Introduction
New railway vehicles in Europe have to comply with noise limits as defined in the
Technical Specification for Interoperability, TSI Noise [1]. The pass-by test has to be
carried out at a site that fulfills requirements in terms of track decay rate and rail
roughness. The need has arisen to be able to perform TSI pass-by type testing at
non-compliant sites, as it can sometimes be difficult or costly to access or maintain a
track that fulfills the TSI requirements. The roughness levels of a TSI track might have
increased over time, resulting in exceeding the TSI limit for rail roughness at certain
wavelength bands. The small deviation method in the TSI Noise 2011 allows for
compensation for small roughness deviations, which cause a difference in sound
pressure level of up to 1 dB (see appendix B of [1]). Larger deviations in roughness
levels, or any deviations in track decay rates are not allowed. Such effects can be
accounted for by transposition.
The objective of the ACOUTRAIN project is to simplify and improve the acoustic
certification process of new rolling stock, by including some elements of virtual testing.
Procedures have been proposed for transposition of pass-by measurement results for
vehicle and track [2]. On the track side, pass-by data obtained at a non-compliant site
need to be transposed to a TSI compliant track, resulting in a virtual TSI test. Similar
procedures can be used to determine source terms for environmental noise prediction
from TSI measurements. Also, non-TSI compliant tracks can be used as target tracks in
order to assess the increase of noise due to, for example, slab track. On the vehicle side,
the outcome of tests on one vehicle is to be translated to another on the same track. The
wheels may have a different design or be fitted with dampers.
2 Transposition Method
Fig. 1 presents five quantities that break down of pass-by noise into quantities
determined by wheel and track [3]. The pass-by rolling noise of trains can be
decomposed into a vehicle transfer function level LHpR,veh and a track transfer function
level LHpR,tr in dB re 20 Pa/√m (both normalized to the number of axles N per unit
vehicle length L), with effective wheel roughness level LR,veh in dB re 1 μm, effective
rail roughness level LR,tr in dB re 1 μm and train speed v in m/s, as excitation functions
(see Fig. 1). Wheel roughness, rail roughness and train speed characterise the excitation
of rolling noise at the wheel-rail contact. The ratios of the resulting trackside wheel and
track noise components to the roughness excitation are expressed as transfer functions,
which capture every aspect of vibro-acoustic transmission of the vehicle or track
implicitly: the wheel and track vibration response to the excitation, its radiation, and
the sound transmission to the trackside. All transfer functions depend only on the
respective track or vehicle design, and are independent of the train speed and the
roughness levels.
The inherent assumption is that (moderate) changes in the wheel transfer
function will have negligible effect on the track contribution, so the dynamic
wheel-rail interaction can be neglected. This assumption is valid for the range of
common track and wheel types [3] and not, for example, for very small wheels or
resilient wheels. This would require more knowledge about the ratio of wheel, rail
and contact-spring mobilities in the contact patch [4]. In the present case it is
assumed a vehicle can be characterized independently from the track using its
transfer function and roughness. Similarly, the track is characterized by its own
roughness and transfer function.
The total pass-by level Lp in dB re 20 μPa can be reconstructed from its key
quantities:
⎛N⎞
L p (v, f to ) = LR ,tot (λ (v, f to )) + LHpR,tot ( f to ) + 10 log10 ⎜ ⎟ = (2.1)
⎝L⎠
with ⊕ denoting the energy sum, fto 1/3 octave frequency bands, λ wavelength in m and
v train speed in m/s.
Transposition of Noise Type Test Data for Tracks and Vehicles 215
Fig. 1. Measurement quantities for rolling noise, separated into excitation and transfer functions
and vehicle and track components. The transfer functions are normalized to the number of axles
per unit length and the roughness is ‘filtered’ by the contact.
The transfer functions and roughnesses for track and rolling stock of Fig. 1 form a
set of ‘building blocks’ which allow the prediction of rolling noise spectra at any speed
for different combinations of vehicles and tracks.
Transposition
The pass-by level of vehicle X on track A, LptotXA in dB re 20 μPa, can be transposed to
the level on a target track B, LptotXB by:
The pass-by level of vehicle X on track A can be transposed to the level of vehicle Y on
the same track, LptotYA by:
where Ytrans,AB and Ytrans,XY denote frequency and speed-dependent transposition factors
in dB to be applied on the measured pass-by level of vehicle X at reference site A for
respectively transposition of track and vehicle.
Separation
Transposition from one site to another requires source separation: vehicle and track
rolling noise contributions to the total pass-by level are to be separated. For Vehicle X
on site A the separate pass-by levels of track and vehicle can be calculated from:
L p ,vehXA (v, f to ) = [LR ,vehX (λ (v, f to )) ⊕ LR ,trA (λ (v, f to ))] + LHpR,vehX ( f to ) + 10 log10 ⎜ ⎟
⎛N⎞ (2.4)
⎝L⎠
L p ,trXA (v, f to ) = [LR ,vehX (λ (v, f to )) ⊕ LR ,trA (λ (v, f to ))] + LHpR,trA ( f to ) + 10 log10 ⎜ ⎟
⎛N⎞ (2.5)
⎝L⎠
Ytrans , AB = ⎣⎡ LR ,totXB − LR ,totXA ⎦⎤ + ⎡⎣ LHpR ,vehX ⊕ LHpR ,trB ⎤⎦ − ⎡⎣ LHpR ,vehX ⊕ LHpR ,trA ⎤⎦ (2.6)
216 H.W. Jansen et al.
which could also be written as: ∆ Lp,tot = ∆ LR,tot + ∆ LHpR,tot. Differences in roughness
levels ∆LR,tot are speed dependent, differences in transfer functions ∆LHpR,tot are not.
Similarly, the transposition factor of Eq. (2.3) for the vehicle can be derived:
Ytrans , XY = ⎡⎣ LR ,totYA − LR ,totXA ⎤⎦ + ⎡⎣ LHpR ,vehY ⊕ LHpR ,trA ⎤⎦ − ⎡⎣ LHpR ,vehX ⊕ LHpR ,trA ⎤⎦ (2.7)
The effects of noise control measures like rail or wheel dampers are included in the
transfer functions. It is preferable if rail vibration is available, as this can be used to
assess combined roughness levels (including contact filter) and combined transfer
functions [3] or to tune a calculation model to measured decay rates and pass-by levels.
2.1 Measurement
Within ACOUTRAIN several ways to assess the quantities of equations (2.6) and (2.7)
have been considered. First of all, all quantities can be measured. Roughness levels
could be measured directly or indirectly (via rail vibrations [3]). However,
experimental assessment of the track and vehicle transfer functions is not trivial. It
requires the measurement of the pass-by of a vehicle with relatively small wheels,
allowing the radiation of the track to be determined, or static measurements of the
tracks. This has been investigated in the STAIRRS project, in which a fully
experimental transposition has been conducted successfully [3, 5] within an accuracy
of ±4 dB for individual one-third octave bands and, on average, ±1 dB in terms of
A-weighted overall level. Also the combined transfer function could be assessed by
combining pass-by levels and measured vertical rail vibration [3]. If the track transfer
function is known from additional measurement or calculation (see next section), the
vehicle transfer function can be derived.
2.2 Calculation
Secondly, the transfer functions of the vehicle and the track can be assessed by a rolling
noise calculation model (e.g. TWINS [4, 6-8]), with the possibility of including
intermediate measurement results like wheel damping and track decay rates. The
excitation of the TWINS models is set to a nominal roughness amplitude of 1 m in each
1/3 octave band. Transposition is carried out with modified track or vehicle properties.
Other changes concerning pad stiffness, roughness or site geometry should also be
considered for this purpose.
In principle, if the noise spectrum of the reference track can be predicted correctly with
TWINS, the separate results for wheel and track can be used to infer their relative
contributions. However this does rely on the separate components also being correctly
predicted. If the noise spectrum is not correctly predicted some adjustment is needed to
the predictions, to account for differences in the roughness and vibro-acoustic
transmission effects. Adjustments are then made to all components of the TWINS results
(wheel, rail and sleeper) to ensure agreement with the measured results. Clearly such
adjustments should be applied with care and should not be too large. In the absence of
other information it is assumed that the sound pressure components remain in the same
proportion as the sound power components in a given one-third octave band.
Transposition of Noise Type Test Data for Tracks and Vehicles 217
2.3 Database
Another option is a variation on the calculation approach, in which only the vehicle
transfer function is to be calculated by the user and the track transfer function is taken
from a database, with available calculated or measured track contributions for a
selection of standard track types, such as ISO [9] or TSI [1] compliant, with typical
sleeper and pad types and track decay rates. For each transposition case an adequate
track transfer function has to be chosen. The track transfer function can be modified in
order to predict the SPL for the target track. If track B has a different decay rate than
track A, a different track transfer function from the database will be selected.
Since manufacturers often already have calculated vehicle transfer functions
available for their rolling stock, they can be combined with different pre-calculated or
measured track transfer functions without the need of dedicated software. The
applicability of the database approach mainly depends on the question whether the
most relevant European tracks can be represented in the database. For non-standard
tracks, such as slab tracks, the corresponding transfer functions will have to be added to
the database.
2.4 Applicability
For all approaches the reference pass-by level (vehicle X on reference track A) can be
calculated by eq. (2.4) and (2.5) and compared to the actual measured pass-by level.
The extent to which the pass-by level on the reference site can be predicted by the
applied roughness and transfer functions can be used as a quality indicator for the input
parameters. It is proposed to allow a maximum absolute value of third octave band
differences between baseline pass-by level measurement and estimation of 4 dB.
Further, the difference between total pass-by levels is limited to 1 dB(A), otherwise the
transposition is not recommended.
3 Validation
In the ACOUTRAIN project several measurement datasets will be used for validation
of the transposition procedure, ranging from largely deviating track types, with large
differences in pass-by level, to standard ISO [9] and TSI [1] compliant sites. In this
paper the focus is on the results of the STARDAMP campaign with VB2N rolling stock
[10] which were used to validate the transposition calculation method. For this
campaign pass-by measurements were conducted on two sections of an identical track,
with and without rail dampers. This allows for a transposition of the track. Some of the
wheels on the train were equipped with wheel dampers, which are used for a
transposition of the vehicle. The track consisted of standard ballasted track with
monobloc concrete sleepers. It is noted that the ground sloped away from the track at
the test site. The effect of ground height on the ground dip in the transfer functions was
found to be important. The database approach, see section 2.3, could therefore not be
applied for this campaign due to a lack of accurate transfer functions. However, it will
be tested with other data on standard tracks, which is not discussed in this paper.
218 H.W. Jansen et al.
All available acoustic data of the track and vehicle are shown in Fig. 2. Since
roughness levels of the target track and vehicle are similar to those of the reference
case, hardly any roughness corrections are required. Besides this information, the
damping of the damped wheel was provided as a TWINS modal parameter file by the
STARDAMP consortium, constructed according to the methodology developed in the
project [10].
Table 1 shows the results of the transpositions of track and vehicle (TWINS
calculations), together with the measurement results of the pass-by level at target track
and vehicle. The corresponding spectra are shown in Fig. 3 for a pass-by speed of
110 km/h. For transposition of the track the transposed pass-by levels are
underestimated by 2-3 dB. For transposition of the vehicle the transposed pass-by
levels are overestimated by at most 0.5 dB.
Fig. 4 shows the frequency dependent transposition factors obtained directly from
measurements (difference between reference and target track/vehicle) and from
calculations. The calculated transposition factors show the same trend as the measured
ones.
Fig. 2. Input parameters for the transposition case, average (contact filtered) roughness levels of
damped and reference wheels and damped and reference tracks (left), measured pass-by levels at
7.5m on site A (right), track decay rates (TDR) for both tracks (centre) determined from rail
vibrations during pass-by, all in 1/3 octave bands
Table 1. Transposed and measured Lp,eq,tp at 7.5 m for vehicle X on site B (transposition of the
track) and vehicle Y on site A (transposition of the vehicle) and differences ΔLp between
transposed and measured results in dB(A) re 20 µPa
Fig. 3. Transposed pass-by levels (for track left, for vehicle right). Also the actual measured
pass-by levels of target vehicle at the target track are shown.
Fig. 4. Transposition factors (for track left, for vehicle right) according to eq. (2.2) and (2.3)
averaged over all pass-by speeds, which is allowed since changes in roughness are small for the
considered cases (see Fig. 2)
In the considered cases, the transposition of the vehicle worked out better than
transposition of the track. This was caused by some unexplained effects in the pass-by
measurements at the damped track. The pass-by level at 1250 Hz was increased by over
5 dB, which is not consistent with increased track decay rates. The effect is also
noticeable in Fig. 4 as a sharp peak. At other frequencies the calculations show the
correct trend.
By separation of track and vehicle contributions to the pass-by level, measurement results
can be transposed from one site to another or one vehicle to another. This can be done
both experimentally and by calculations. However, for calculations, measurements are
also required in order to set input parameters and build models.
Transposition can be applied to cases where roughness levels vary, track properties
are different, or both. The uncertainty in roughness corrections for transposition can be
expected to be small (about 1 dB) if the rail roughness does not vary for different
wheel/rail contact positions. The uncertainty in correcting transfer functions is higher.
Therefore it is proposed that the extent to which the pass-by level on the reference site
220 H.W. Jansen et al.
can be predicted by the applied roughness and transfer functions should be used as a
quality indicator. It is proposed to allow a maximum absolute value of third octave
band differences between baseline pass-by level measurement and estimation of 4 dB.
Further, the difference between total pass-by levels is limited to 1 dB(A), otherwise the
transposition is not recommended.
Transposition requires basic knowledge of railway acoustics. The user should be
capable of judging the usefulness of the input parameters of the procedures, as the
quality of the input data highly determines the accuracy that can be achieved.
References
[1] Commission Decision of 4 April 2011 concerning the technical specifications of
interoperability relating to the subsystem ‘rolling stock-noise’ of the trans-European
conventional rail system. (TSI Noise, 2011/229/EU)
[2] Jansen, H.W. et al.: Source separation and transposition techniques, ACOUTRAIN report,
ACT-T2_4-TNO-023-02 (2014)
[3] Janssens, M.H.A., et al.: Railway noise measurement method for pass-by noise, total
effective roughness, transfer functions and track spatial decay. Journal of Sound and
Vibration 293, 1007–1028 (2006)
[4] Thompson, D.J.: Railway noise and vibration: mechanisms, modelling and means of
control. Elsevier, Great Britain (2009)
[5] Jansen, H.W., de Beer, F.G.: Validation measurements for indirect roughness method,
STAIRRS Report, TNO-RPT-020078 (July 2002)
[6] Thompson, D.J., Hemsworth, B., Vincent, N.: Experimental validation of the TWINS
prediction program, part 1: method. Journal of Sound and Vibration 191, 123–135 (1996)
[7] Thompson, D.J., Fodiman, P., Mahé, H.: Experimental validation of the TWINS prediction
program, part 2: results. Journal of Sound and Vibration 193, 137–147 (1996)
[8] Jones, C.J.C., Thompson, D.J.: Extended validation of a theoretical model for railway
rolling noise using novel wheel and track designs. Journal of Sound and Vibration 267,
509–522 (2003)
[9] International standard: Railway applications – Acoustics – Measurements of noise emitted
by railbound vehicles, ISO 3095 (2005)
[10] STARDAMP Standardisation of damping technologies for the reduction of rolling noise,
Final Report (May 2013)
Virtual Testing within the TSI Noise: How to Introduce
Numerical Simulation into a Certification Process?
1 Introduction
In 2006 the TSI Noise (Technical Specifications for Interoperability) [1] came into
force introducing limiting values in noise emission for the certification of new or
upgraded rolling stock. The current TSI Noise certification process is based on real
tests such as pass-by noise measurements. This procedure is often seen as very
restricting considering that new rolling stock is often largely based on existing designs
and minor changes therefore need to be taken into account in terms of noise emission.
ACOUTRAIN is a research project within the 7th framework programme co-funded by
the European Commission with the goal of simplifying and improving the acoustic
certification process of new rolling stock. This simplification will mainly be based on
using virtual testing, i.e. numerical simulations to support real testing for the
assessment of global noise of rolling stock. The structure and links of the different
work-packages are presented on the webpage of the project [2].
2.1 Context
The TSI Noise draft is a mandate of ERA (European Railway Agency), who could
consult a group of experts for its revisions writing. End of 2013, a new revision of the
TSI Noise has been proposed by ERA. Within last years, the sector claims for
improving and simplifying some of the TSI Noise requirements, based on the research
and standardization activities that tend to demonstrate the reliability of virtual testing
approach. Within this frame, ACOUTRAIN aims at proposing some future
improvements within TSI and standardization.
The TSI Noise certification process requires real tests to be carried out in order to
assess the global noise levels for different running conditions: stationary, pass-by and
starting noise measurements for the exterior noise and the driver’s cabin interior noise.
The requirements for these measurements are largely based on the ISO3095 [3]
standard. This means that the measurement protocols used for the TSI Noise
certification have gained a wide recognition because they have been used all over
Europe during many years by a lot of different measurement teams. This has allowed
the repeatability and representativeness of the described protocols to be improved and
therefore it has ensured the reliability of the global process.
Nevertheless, some of the TSI Noise requirements in terms of measurement
protocols appear to be difficult to achieve. For example, pass-by measurement must be
carried out on a TSI compliant track (low roughness and high track decay rate, so that
the track contribution is minimized compared to the rolling stock contribution). It can
be very difficult to find such a track and to gain access to it. Moreover, it could be, for
some cases, meaningless to carry out the complete TSI Noise measurement process: for
example when a new vehicle’s design is largely based on a reference vehicle’s one,
which has already been certified.
Concerning the TSI compliant track issue, the small deviation method has been
proposed in the TSI 2011 revision [1] to relax this constraint a little. It allows
measurements to be performed on a non TSI track as long as it is proven that this
deviation from the reference measurement conditions has a minor impact (less than 1
dB) on the global pass-by noise.
With the objective of simplifying the TSI Noise process, the simplified approach has
been introduced by ERA in the TSI 2011 revision to allow, in specific cases, not
carrying out the required real measurements. Instead, some evidence has to be provided
that the new vehicle will be less noisy or no noisier than its reference vehicle, already
certified.
In such a context, it is considered that, for some cases and under certain conditions,
numerical simulations can substitute real measurement for the assessment of rolling
stock noise. If such a numerical approach has to be used for certification purposes, it
has to be demonstrated that:
• it is highly representative of the real measurement conditions;
• it has the same degree of reliability.
Virtual Testing within the TSI Noise 223
To do so, the equipment with the largest contributions in terms of noise are identified
and turned into equivalent point noise sources. The same process is carried out to
consider rolling noise when it is required (for pass-by simulation): wheels, rail and
sleepers are represented as equivalent point noise sources.
The equivalent point noise sources are generally defined with:
Considering that the real noise sources are generally extensive compared to the
distance of measurement (70 cm diameter for a common cooling fan and a distance of
measurement of 7.5 m), one real source is often represented by several equivalent point
sources.
In ACOUTRAIN, the acoustical definition of the rolling stock as a set of noise
sources is called the Virtual Vehicle. This virtual vehicle is defined in dedicated
software called simulation tool. Then, the global noise level emitted by the rolling stock
is generally assessed by an energy summation of the different equivalent noise source
contributions, at the reception point. The numerical tool handles the propagation of
noise contributions by taking into account environmental conditions such as ground
effect and Doppler effect.
224 E. Bongini and R. Cordero
Since the simulation of the propagation of the different noise contributions is based
on well known analytical formulations and is therefore well mastered, the key point of
the overall procedure is the source identification and characterization.
For the assessment of rolling noise equivalent sources, a numerical approach can be
used (for example TWINS software that allows computing rolling noise sources. The
way of using TWINS has been harmonized within ACOUTRAIN, reported in [4]). For
the other types of noise source, lab tests of dismounted equipment could be used to
define their equivalent noise sources. However, as the integration effect in the vehicle
has a large influence on the noise emitted by the different real sources, measurement on
the overall vehicle at standstill is typically required.
To allow a virtual procedure, such as the one described above, to be used in a
certification process and particularly in TSI Noise, each step has to be fully mastered. It
has to be demonstrated that virtual testing has the same degree of reliability and
representativeness as the real tests.
The TSI Noise certification could be supported by virtual testing in different ways.
Three main approaches of VT have been identified [5]:
• The full virtual testing approach: this implies that we are assessing the total
vehicle external noise level solely by assessing the individual noise sources. This
means that no real assembled vehicle is required for the certification. Only a
virtual vehicle will be used to compute all or some of the levels that are to be
compared with the limiting values of the TSI. We can also refer to this approach
as Virtual certification configuration, in which absolute levels can be obtained
with simulation only.
• The hybrid approach: in this case both conventional/physical tests and virtual
vehicle predictions can be used. We can say that simulation supports real tests and
vice versa. When virtual testing is implemented for the first time this approach is
preferred, because advantages from real tests and simulation are combined
without needing a reference virtual vehicle that is already validated (see
Extension to Approval method): it concatenates confidence in the well-known test
methods and the potential of the new virtual test method. It requires several
experimental tests on sub-systems as well as on the train itself.
• The extension of approval: this third approach basically differs from the previous
two in one aspect: it does not start from scratch but from a validated virtual
vehicle of a similar rolling stock family. From this validated virtual vehicle a
derivative model is defined through modifications of the input data of the original
model (e.g. changes in geometry, noise source positions, noise source
amplitudes), which are not expected to have a strong influence on the reliability of
the model. Hence the existing validation is still considered acceptable also for the
new derivative model. However, at the request of the Notified Body (independent
Virtual Testing within the TSI Noise 225
Even if numerical simulations have been widely used in vehicle noise assessment at
design and development phases of new rolling stock, a full virtual testing approach
cannot be foreseen for the next years. Beforehand, the reliability of the VT approach
needs to be proven, by supporting and comparing this VT approach with real
measurements, as described under the hybrid approach.
During the first years of VT use, which can be seen as a proof of concept phase, the
methodology called Verification and Validation allows structuring the demonstration
of Virtual Testing reliability. It defines which steps of the process have to be controlled
and how.
Fig. 2. Verification and validation methodology applied to virtual testing for TSI Noise
Fig. 2 illustrates the methodology built within the first year of ACOUTRAIN, based
on the V&V process but specifically designed for acoustic certification purpose, where
it is formally stipulated that virtual testing reliability relies on simulation tool
certification and virtual vehicle validation.
• the tool verification (step 1) which consists in verifying that it has no bugs and it
corresponds to the process requirements,
• the tool validation (step 2) that ensures that the analytical models are exactly
programmed
226 E. Bongini and R. Cordero
For the tool verification (step 1), ACOUTRAIN partners have defined a common
format of input and output of simulation tools. The input format corresponds to the
most common formats used in the existing simulation tools that are known to the
ACOUTRAIN consortium. The outputs correspond to the acoustic indicators used in
technical requirements of the TSI and some additional acoustic indicators such as
pass-by signature or third-octave band spectrum that allow checks of the numerical
process in the validation step.
For the tool validation (step 2), the ACOUTRAIN partners have proposed reference
cases that test how the simulation tool handles the propagation phenomena and the
energy summation at the reception points.
The corresponding reference output values have been computed with existing tools
already validated. Twenty-four reference cases have been defined, with various
combinations of source location, receptor location and ground properties, as well as
source pass-by speed. An example of a comparison between reference value and tested
values for one reference case is presented in Fig. 3.
Fig. 3. Comparison of 3 different simulation tools to the reference spetrum (Lp, R1 Analytic), for
a monopole at 0.5 m from a reflective ground – Receptor at [7.5 m, 1.2 m] for the track
Considering that most of the simulation tools will not produce exactly the reference
values for the different reference cases, corridors of validation have to be defined. They
correspond to the domain around the reference value in which tested output is
considered as “acceptable”. For the tool validation these corridors of validation will be
defined considering a maximum discrepancy level on the global level and maximum
discrepancy levels on the corresponding 1/3rd octave band spectrum levels, with a
narrower corridor around the frequency range [800 Hz, 4000 Hz] when pass-by
configurations are foreseen and constant corridors over the entire frequency range
when standstill configurations are foreseen.
In the hybrid approach, its validation is based on comparison with real test results
carried out on the same rolling stock, for various operating conditions: at standstill or
for pass-by at different speeds for example.
In this validation process, uncertainties have to be taken into account for two
reasons. First, the uncertainty level of the global output indicates the degree of accuracy
of the new process. In fact, the new methods based on virtual testing should be at least
as accurate as the established methods in order to gain confidence from the railway
sector. A way to evaluate the accuracy is to determine the uncertainty of a result.
Secondly, comparison between real measurement and virtual testing must consider the
level of uncertainties of both processes. Currently no quantification of the uncertainties
is required when certifying vehicles against the TSI Noise. The accuracy and the
comparability are ensured by performing the measurements according to well defined
established procedures and the requirement of a reference track. Additionally, the
pass-by noise measurement has to be repeated three times at each speed and the
discrepancy between each measurement should be less than 3 dB. However, for
validation of virtual vehicle, the real test measurement can be carried out in such a way
that the output yRT (generally, the global level) is defined with its statistical properties,
giving its corresponding random vector YRT.
For the VT process, the whole deterministic railway acoustic problem can be
expressed as a function of an R-transformation:
[
x → y VT = R(x) ] (3.1)
where x represents the acoustic power / acoustic pressure of the different noise sources
defining the rolling stock and the propagation parameters (temperature, wind speed,
ground parameters, site topology), yVT represents the equivalent noise level at [1.2m ;
7.5 m] or any other location, and R is the transformation from x to yVT (radiation and
propagation of the different noise sources, energy summation at the reception point of
the different noise source contributions and computation of the equivalent noise level).
In railway acoustics, R is fully deterministic and the stochastic dimension comes from
the input variabilities. The corresponding stochastic problem can be written as:
[
X → Y VT = R (x) ] (3.2)
4 Conclusion
This paper is a summary of the first requirements for applying a virtual testing
procedure in TSI Noise certification that have been defined in the first year of the
ACOUTRAIN project. To gain recognition among TSI stakeholders, this new
procedure has to be as reliable as the real testing process used up to now. It requires:
• a numerical tool that supports exactly the analytical equations representing noise
propagation in free-field environment,
• a virtual vehicle with a high degree of representativeness (compared with the real
rolling stock).
• for the tool certification, reference cases have been defined for which the tool will
be tested, by comparison against reference values;
• for the virtual vehicle validation, results of the virtual testing process will be
compared to real test results, taking into account uncertainty levels of both
processes.
Moreover, before coming into force, the overall VT approach will be tested in
ACOUTRAIN, for some specific cases called application cases, to verify its
applicability and its reliability.
References
[1] Commission 2006/66/EC: Commission decision of 23rd December 2011 concerning the
technical specification for interoperability relating to the subsystem "rolling stock -
noise“ of the trans-European conventional rail system. Official Journal of the European
Union L37 (49), 1–49 (2011)
[2] ACOUTRAIN web page, http://www.acoutrain.eu/
[3] ISO 3095:2005: Railway Applications - Acoustics - Measurement of noise emitted by
railbounds vehicles (2005)
[4] Thompson, D.J., et al.: ACOUTRAIN project, deliverable 2.1 public, User guide describing
procedures for modelling wheels and tracks (2012),
http://www.acoutrain.eu/?page_id=883
[5] Bongini, E., et al.: ACOUTRAIN project, deliverable 1.2, Recommendations for a future
certification process - Virtual testing implementation in the framework of the TSI Noise,
http://www.acoutrain.eu/?page_id=883
A Study of the Measurement Technology of Noise
Sources of High-Speed Trains
1 Introduction
All practical microphone array measurements results undergo deviations from results
obtained by an ideal array configuration, i.e. one with very small microphone spacings
and a very long array length. Such a system becomes expensive as well as impractical
to set up and use.
In this study the goal is to build a useful microphone array system for measurements
of high-speed trains with traveling speeds up to 400 km/h. A speed increase from
300 km/h to 421 km/h lead to an increase in the noise level from the train of nearly
10 dB: high-speed train noise is a complex matter, as noise sources are located all over
the train and are broad band in character. For the purpose of investigating the source
strengths in more detail, it is necessary to enhance the quality and the reliability of the
measurement system in terms of the resolution and dynamic range by means of an
efficient configuration of the positions of the microphones used.
Measured beam power values are also corrected through a comparison of the
measurement values with the simulated values. To this end, an arrangement for a
microphone array with an enhanced method of processing multiple channel signals is
studied in order to compare the acoustic strengths of the various noises sources of
high-speed trains traveling at speeds up to 400 km/h. Some results are shown and
discussed.
First simulations are conducted via simple delay-and-sum beamforming in the time
domain, with processing steps that include localization, quantification and spectrum
calculation for microphone pressure signals caused by a wheelset as it travels straight
ahead of the array. In Fig. 1, several microphone configurations are shown. The
irregular star-shaped array (Fig. 1(a)) consists of is 24 arms/beams in a star
configuration, it has a total of 88 microphones and is capable of separating two point
sources from 500 Hz and upwards. In the upper area of the noise sources aliasing
effects can be observed in the simulation results (Fig. 2(a)). An Archimedian spiral
(Fig. 1(b)) consists of 96 microphones and has a diameter of approximately 4 m. It is
capable of separating two point sources from 500 Hz and upwards (Fig. 2(b)), but it
shows aliasing effects in the upper area as well, and the Archimedian spiral may be
difficult to build in practice, as the microphones have mostly “irregular” positions.
The radial array, consisting of 144 microphones with a diameter of 6 m, is shown in
Fig. 1(c).
Multi-arm spirals can be generated with the help of the equation for Fermat’s
spiral: r=±θ1/2. Numerical simulations are conducted for different Fermat’s spiral
configurations. The array configuration is shown in Fig. 1(d) and the results are
shown in Fig. 2(d). This consists of 144 microphones with a diameter of 6 m. The
useful frequency region for this configuration is about 160 Hz – 3150 Hz. This Fermat
spiral with six arms produces images nearly as good as sound images.
In delay-and-sum beamforming [2], a point monopole sound source radiates into free
air, the wave travels a distance r at the speed of sound c to the observation point, and
it arrives at time instant delayed by r/c [1]. With an array of microphones, it is
possible to focus on a specific point source by summing up all microphone signals
while accounting for the propagation time delays between the source and the receivers
(microphones). If a propagating signal is present in an array’s aperture, the sensor
outputs, delayed by appropriate amounts and summed, reinforce the signal with
respect to noise or waves propagating in other directions.
The delays that reinforce the signal are directly related to the length of time it takes
for the signal to propagate between the source and the sensors.
The calculating program includes the following features: the set-up of an arbitrary
microphone configuration, the reading of the measurement data, beamforming
(stationary, moving), source strength (sound power) calculation and the directivity
calculations of moving sources.
A Study of the Measurement Technology of Noise Sources of High-Speed Trains 231
(a) (b)
(c) (d)
Fig. 1. Illustration of a microphone arrangement (in mm) for various types of microphone
arrays (a) Irregular star, (b) Archimedian spiral, (c) radial array, (d) Fermat’s spiral array
(a) (b)
(c) (d)
Fig. 2. Simulation of the measurements with microphone arrays for the sound sources
localizations (160 Hz ~ 3150 Hz) ; (a) Irregular star, (b) Archimedian spiral, (c) radial array,
(d) Fermat’s spiral array
A Study of the Measurement Technology of Noise Sources of High-Speed Trains 233
For the field test, a Fermat’s spiral array is constructed along a high-speed railway
line (Fig. 6). The train is an EMU, with each bogie motorized and with disk brakes. In
the middle, there is a coach; to the left and right of this coach there are two other
coaches. Each coach has a two-wheel bogie at each end. Noise radiation from the
wheels in the bogie is clearly observable. The rail radiates around at around 800 Hz ~
1600 Hz. Also, noise radiates from equipment-structure on the roofs and from the
inter-coach spacing (Fig. 7).
In Fig. 8, noise maps of the power car of a high-speed train are shown. The
high-speed train passed by at a speed of 288 km/h. Here, both the results of the
measurement and of analyses of 2011 and 2012 are shown. The performance of the
revised microphone array was improved in the low-frequency range (< 500 Hz) and in
the higher-frequency range (>1250 Hz).
A Study of the Measurement Technology of Noise Sources of High-Speed Trains 235
(a) (b)
(c) (d)
Fig. 7. Noise radiation from a high-speed train at 410 km/h travelling from right-to-left (a)
250 Hz, (b) 500 Hz, (c) 800 Hz, (d) 160 Hz ~ 3150 Hz
Fig. 8. Noise radiation from a power car of a high-speed train travelling at 288 km/h; measured
in 2011 with a microphone array (4 m, 96 channels) (top), measured in 2012 with a microphone
array (6 m, 144 channels) (bottom)
4 Concluding Remarks
In order to enhance the properties and ability to localize and quantify the source
strengths of high-speed train noise sources, simulations are conducted via simple
delay-and-sum beamforming in the time domain, with processing steps that include
localization, quantification and spectrum calculation for microphone pressure signals.
With a microphone array system using the ‘Fermat spiral with six arms’, satisfactory
images could be produced in the frequency region of 160 Hz – 3150 Hz. For the
calculation of the strengths of the sound sources, sound power is determined within a
selected region by the summation of power of multiple sound sources. A simulation
module is used to test the properties of different microphone configurations and to
calculate sound power correction factors.
Signal processing methods, such as side lobe filtering, are updating with the aim of
improving the performances of the system in the source identification for the high
frequency ranges.
References
[1] Morse, P.M., Ingard, K.U.: Theoretical acoustics. Princeton University Press (1968)
[2] Johnson, D.H., Dudgeon, D.E.: Array signal processing. Prentice Hall (1993)
[3] Nordborg, A., Koh, H.-I.: Microphone array measurements of high-speed train noise.
AIA-DAGA (2013)
Determination of Insertion Losses for Vibration
Mitigation Measures in Track by Artificial Vibration
Excitation
Summary. This paper describes free-field tests carried out to test a procedure for
the investigation of vibration mitigation measures. The test procedure consists of
(1) train passing measurements, (2) measurement of the subsoil transfer-mobility
by impact tests and (3) vibration measurement with an artificial, force-controlled
excitation by means of a newly-developed force-controlled harmonic shaker
including the procedure to simulate the sprung and un-sprung masses of trains.
It is shown that the mitigation effect based on the dynamic spring-mass effect
of vibration isolation for resilient elements can be well described by this method.
This component of mitigation effect can also be verified by means of the artificial
excitation.
On the other hand, the results of the train pass-by measurements show
additional reduction potential of under-sleeper pads (USP) due to the mechanical
effect of a homogenized track stiffness.
1 Introduction
effectiveness of mitigation for different track sections with configurations of standard and
heavy sleepers and under-sleeper-pads (USP) in different stiffness categories including a
slab-track configuration. These tests shall be carried out at full-scale in the field.
The paper describes the procedure and measurement campaign performed within
RIVAS.
where: De(fTn): insertion loss; Lv,1(fTn): third-octave spectrum of ground borne noise at
the reference track system, Lv,2(fTn) : third-octave spectrum of ground borne noise at the
track system with mitigation measure.
before mitigated
F2
non mitigated
F1
L V, 2
L V, 1
afterward
Fig. 1. Basic principle of determining the IL with the two procedures, the “before/afterwards”
method or procedure 2 on the left and the “right/left”-method or procedure 1 on the right side
Determing IL
To determine the transfer mobility function from the head of the rail to the point of
interest a stationary vibration excitation with specific requirements was needed:
o Force controlled vibration generator which is able to operate a vibration sweep
over a broad range of frequency
o Applicable to large scale test in test rig as well for investigations at
commercial tracks
o Quickly ready for use in track (< 1 h) to minimize the track possession
o Suitable for work at different track systems, under power cables and easy
access to tracks
o Simulation of different load conditions for the un-sprung and sprung masses
240 R. Garburg, D. Heiland, and M. Mistler
Fig. 4. The harmonic shaker BUTTERFLY® with the excavators as dead load (dynamically
decoupled mass), from top to down: Pos 1 = 304 kN, Pos 2 = 221 kN and Pos 3 = 5 kN of
unsprung dead-load
2012-11-04_05-04-07 ME5-Sweep6_Clip/K:1/S:1/Kraft
10000
[N]
Kraft / N
0
Force
-10000
100
Frequenz / Hz [Hz]
80
Frequency
60
40
20
0
0 50 100 150 200 250
2012-11-04_05-04-07 ME5-Sweep6_Clip/K:3/S:1/Fundament
20
[mm/s]
10
Geschwindigkeit / mm/s
Velocity 0
-10
-20
0 50 100 150 200 250
2012-11-04_05-04-07 ME5-Sweep6_Clip/K:4/S:1/ME5a(8m)
0,2
[mm/s]
0,1
Geschwindigkeit / mm/s
0,0
Velocity
-0,1
-0,2
0 50 100 150 200 250
2012-11-04_05-04-07 ME5-Sweep6_Clip/K:7/S:1/ME5d(16m)
0,50
[mm/s]
0,25
Geschwindigkeit / mm/s
0,00
Velocity
-0,25
-0,50
0 50 100 150 200 250
Fig. 5. Example of measured vibrations with hydraulic shaker on the track (from top to bottom
each as time history): input force from shaker, frequency, vibration velocity at the un-sprung
mass and ground vibration velocity as response at 8 m and 16 m from track
4 Measurements
three train types at the same track section. In general the freight trains caused the
highest vibration velocity level but in some specific frequency ranges the vibration
level of ICE category exceed these of freight trains.
90 90 90
22 pass bys 6 pass bys 33 pass bys
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
2
3,15
5
8
12,5
20
31,5
50
80
125
200
315
2
3,15
5
8
12,5
20
31,5
50
80
125
200
315
2
3,15
5
8
12,5
20
31,5
50
80
125
200
315
1/3 Octave band center frequency [Hz] 1/3 Octave band center frequency [Hz] 1/3 Octave band center frequency [Hz]
Fig. 6. Vibration velocities (third-octave band at 8 m from track) of train pass by at the reference
site (ME 1) of three different train-categories: electric rail-car (left), ICE-T (middle) and freight
trains (right)
To determine the transfer mobility from the rail head to the measuring points at 8 and
16 m from the track the shaker BUTTERFLY® as described was used.
Considering the subsoil correction the IL is determined as follows:
De(fTn) = (Lv,1(fTn) – F1(fTn) – A1(fTn)) – (Lv,2(fTn) – F2(fTn) – A2(fTn)) (4.1)
where: Fi(fTn): Excitation force spectrum at the reference section (i=1) and at the track
system with mitigation measure (i=2) respectively; Ai(fTn): Transfer mobility of the
subsoil at the reference section (i=1) and at the track system with mitigation measure
(i=2) respectively.
Determination of Insertion Losses for Vibration Mitigation Measures in Track 243
Fig. 7 shows the insertion losses determined by train passages (ICE trains) compared
to those by the artificial excitation at the 8m point from track using the soil correction
function. A more detailed analysis is given in [6] and [7].
The results for the IL by the shaker excitation shows clearly the expected results
compared to calculation methods based on dynamic models with mass-spring systems
or impedance models [3]. The results of IL by train passages show some additional
positive mitigation effects, particularly at low frequencies, which are not possible to
explain with these models. This phenomenon was recognized in several other
investigations as described in [8, 9] and by others.
20 20
10 10
Insertion Loss [dB]
0 0
4 5 6,3 8 1012,516 20 2531,540 50 63 80 100125160 5 6,3 8 10 12,5 16 20 25 31,5 40 50 63 80 100 125 160
-5 -5
-10 -10
-15 -15
1/3 Octave Band Center Frequency [Hz] 1/3 Octave Band Center Frequency [Hz]
Fig. 7. Comparison of IL determined by ICE train passages (left) and by artificial harmonic
excitation (right) for the two investigated different types of USP
This shows clearly that the USP have some additional positive side-effects on
vibration mitigation which cannot be described with simplified models as well as with the
stationary harmonic artificial excitation. This mitigation effect, based on mechanical
behaviour e.g. due to embedding the ballast stones in resilient material with an increase of
contact area together with a decrease of contact forces between the ballast and sleeper
[10], the better load-distribution in the longitudinal direction as well as the general better
track quality (e.g. avoiding of hanging sleepers), can`t be detected by stationary
excitation. Also the part of IL due to the change in the parametric excitation [11] cannot
be seen by a stationary artificial excitation. This also explains why the mitigation effect of
USP is often underestimated using dynamic prediction models.
The next step within RIVAS will be to measure, with the same configuration of
shaker and preload, the IL of different configurations of heavy sleepers and USP
stiffness including a slab-track configuration. These tests will be carried out in a
full-scale-field test by the RIVAS Partner Eiffage Rail GmbH.
Acknowledgment. The research leading to these results has received funding from the
European Union Seventh Framework Program under grant agreement no 265754.
References
[1] RIVAS, http://www.rivas-project.eu
[2] DIN SPEC 45673-3 (Working draft 2013): Mechanical vibration - Resilient elements used
in railway tracks - Part 3: Experimental evaluation of insertion loss from artificial
excitation of mounted track systems (in a test rig and in situ). In preparation as replacement
for DIN V 45673-3:2004-09
[3] DIN 45673-4: Mechanical vibration - Resilient elements used in railway tracks - Part 4:
Analytical evaluation of insertion loss of mounted track systems
[4] Stiebel, D.: Protocol for free field measurements of mitigation effects in the project RIVAS
for WP 2, 3, 4, 5, RIVAS, Deliverable D1.2 (2011)
[5] Behr, W.: Innovationsprojekt “Leiser Zug auf realem Gleis” (LZarG), Abschluss-Bericht,
Förderkennzeichen 19 U 7020 A) (2012)
[6] Mistler, M., Heiland, D.: Free field vibration measurements - Experimental investigation of
the Insertion Loss of under-sleeper pads within the framework of RIVAS, Measurement
Report ID 80-7781-B1 (2012)
[7] Garburg, R., et al.: Description of test procedures based on laboratory tests and field tests
including validation, RIVAS Deliverable D1.10 (2013)
[8] UIC International Union of Railways, UIC Project Under Sleeper Pads - Semelles sous
traverses - Schwellenbesohlungen - Summarising Report (2009)
[9] Loy, H.: Mitigation vibration using under sleeper pads. Railway Gazette International
(April 2012)
[10] Freudenstein, S.: Die Kontaktspannung zwischen elastisch besohlten Schwellen und
Schotter, ETR, Nr 05 (May 2011)
[11] Nélain, B., Vincent, N., Lombaert, G., Degrande, G.: Control of railway induced ground
vibrations: Influence of excitation mechanisms on the efficiency of resilient track layers.
In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D.,
Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 126, pp. 635–642. Springer, Heidelberg (2015)
The Prediction of Vibration Transfer for Railway Induced
Ground Vibration
1 Introduction
The project Railway Induced Vibration Abatement Solutions (RIVAS) [1] is carried
out under the EU’s Seventh Framework Programme for Research and aims at reducing
the environmental impact of ground-borne vibration from railway traffic. Within the
frame of the RIVAS project, abatement measures for ground-borne noise and vibration
for tracks at grade are studied. The aim is to develop innovative measures at the source
to reduce the annoyance of lineside residents. Mitigation measures studied apply to rail
vehicle design, rolling stock maintenance, track design, subgrade engineering, and the
transmission path within the railway infrastructure.
In work package 1 of the RIVAS project, test procedures have been established to
determine the performance of mitigation measures under realistic conditions where the
performance of mitigation measures is determined experimentally by measuring the
The site in El Realengo is located in the Segura river flood plain next to the railway line
between Murcia and Alicante, where a new high-speed railway is under construction.
At this site, the effectiveness of subgrade stiffening next to the track as a mitigation
measure will be assessed. Thus, a test section is selected adjacent to the existing
conventional line where the mitigation measure will be installed together with a
reference section along the same track with no mitigation.
The soil at the site in El Realengo consists of a desiccated upper crust, soft flood
plain soil layers and hard alluvial cone soil layers on top of triassic bedrock. The
dynamic soil properties have been evaluated by geophysical tests performed by
CEDEX in April 2012, including Spectral Analysis of Surface Waves tests, seismic
piezocone down-hole tests and seismic refraction tests, both at the test site and the
reference site. The identified layer thickness, shear wave velocity, dilatational wave
velocity and density for each layer of a horizontally layered soil profile are summarized
in Table 1. The estimated material damping ratio is subsequently determined more
accurately based on the measured free field transfer functions.
Table 1. Dynamic soil properties at the site in El Realengo (provisional values indicated by *)
h Cs Cp βs βp ρ
Layer
[m] [m/s] [m/s] [-] [-] [kg/m3]
1 0.30 270 560 0.025* 0.025* 1800
2 1.20 150 470 0.025* 0.025* 1750
3 8.50 150 1560 0.025* 0.025* 1750
4 10.00 475 1560 0.025* 0.025* 1900
5 ∞ 550 2030 0.025* 0.025* 1900
The Prediction of Vibration Transfer for Railway Induced Ground Vibration 247
The free field transfer functions are measured both at the reference section and the test
section by means of a falling weight deflectometer. Impacts are applied on a circular
metallic foundation installed at 12 m from the centerline of the track while the ground
velocity is measured using geophones installed on a straight measurement line
perpendicular to the centerline of the track at a distance of 2 m to 64 m from the source.
The transfer function between the applied impact and the measured velocity is
determined based on a number of impacts in order to improve the signal-to-noise ratio.
The measured transfer functions and the 95 % confidence interval are shown in Fig. 1
for both sections at 6 m, 12 m, 24 m and 48 m from the track.
50 50
6m 12m
Mobility [dB ref 10−8 m/s/N]
0 0
−25 −25
−50 −50
0 50 100 150 200 0 50 100 150 200
50 50
24m 48m
Mobility [dB ref 10−8 m/s/N]
25 25
0 0
−25 −25
−50 −50
0 50 100 150 200 0 50 100 150 200
Fig. 1. Transfer functions at (a) 6 m, (b) 12 m, (c) 24 m and (d) 48 m predicted with initial soil
profile (dashed grey line) and updated material damping parameters (dashed black line)
compared to experimental results for the test section (light grey line) and the reference section
(dark grey line). The grey regions indicates the 95 % confidence intervals.
in Fig. 2(a), based on the identified soil profile. The predicted transfer function is
compared to the experimental results (displayed in Fig. 1). A good approximation to the
experimental result is obtained for low frequencies. At frequencies above 25 Hz, the
measured transfer function is generally overestimated. The discrepancy between the
measured and predicted transfer function grows larger for increasing distance from the
source, indicating an underestimation of the material damping ratio.
(a) (b)
Fig. 2. Schematic view of the prediction model for the free field transfer functions
The results follow the general expectation that material damping decreases with
depth. The imposed lower bound value is reached for the fourth layer and the halfspace.
As the wave propagation is mainly concentrated in the top three layers in the frequency
range of interest, the significance of the material damping ratio in the bottom two layers
is found to be relatively small.
Fig. 3(b) shows that the agreement between the measured and predicted Arias
intensity after optimization is very good for both sections, even if the optimization
process only takes into account the results of the test section. Including the results of the
reference section leads to unrealistic values of the material damping ratio in the first
and fourth layer and the supporting halfspace, but hardly affects the resulting prediction
of the transfer functions, indicating again the smaller importance of the material
damping ratio in these layers.
The Prediction of Vibration Transfer for Railway Induced Ground Vibration 249
−11 −11
10 10
−12 −12
10 10
−14 −14
10 10
−15 −15
10 10
−16 −16
10 10
−17 −17
10 10
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Distance [m] Distance [m]
(a) (b)
Fig. 3. Predicted Arias intensity (black line) (a) before and (b) after optimization of the soil’s
material damping ratio, compared to the measured Arias intensity at the test section (light grey
line) and the reference section (dark grey line)
The conventional line at the site in El Realengo has a single ballasted track with a wide
rail gauge (1.668 m). The rails are of the type RN 45 (stiffness ErIr = 3.00×106 Nm2,
mass per unit length ρrAr = 44.8 kg/m) and are supported by bi-block reinforced
concrete sleepers (mass msl = 200 kg, rotational inertia ρslIsl = 102.4 kgm2). Rail pads
are located between the sleepers and the rails and are initially assumed to have a
stiffness krp = 300×106 N/m and viscous damping crp = 3.5×103 Ns/m. The ballast layer
of 0.50 m (shear wave velocity Cs = 250 m/s, density ρb = 1600 kg/m3, Poisson’s ratio υ
= 0.2) is supported by an embankment of 0.50 m (shear wave velocity Cs = 200 m/s,
density ρe = 1700 kg/m3, Poisson’s ratio υ = 0.35). The stiffness of the rail pad and the
shear wave velocity of the ballast are initial estimates and are updated in Section 3.3.
The track receptance has been measured in unloaded conditions where a hammer
impact is applied on the rail and the corresponding rail response is measured. The
dynamic track behavior is strongly affected, however, by the magnitude of the track
load. The receptance has therefore additionally been measured when the track is loaded
250 H. Verbraken et al.
by a train, by measuring the track deflection and the axle loads during the passage of a
train. This results in an estimation of 2×10-8 m/N for the static track receptance (also
shown in Fig. 4).
The track receptance is predicted by means of a 2.5D coupled finite element – boundary
element model [4], shown in Fig. 2(b), where the track and soil are modeled as invariant
in the longitudinal direction in order to reduce the computational cost.
Fig. 4 compares the receptance predicted based on the design values presented in
Section 3.1 with the measured receptance in loaded conditions. It is clear that the initial
track parameters lead to an underestimation of the track receptance, or an
overestimation of the track stiffness.
Fig. 4. Predicted track receptance before optimization (grey line) and after optimization (black
line) of the track parameters, compared to the track receptance measured in loaded condition
during train passages (dotted line)
The parameters of the resilient track components are updated by fitting the predicted
and measured results in a frequency range between 0 and 60 Hz, resulting in a ballast
shear wave velocity Cs = 150 m/s and a rail pad stiffness krp = 31×106 N/m.
In order to fit the measured track receptance, a decrease in track stiffness is required.
This is only achieved by a decrease in the stiffness of the ballast and the rail pad, as they
are the only resilient track components considered in the updating process, which
results in an unrealistically low value of the rail pad stiffness. Rather than physical
characteristics of the ballast and the rail pad, the updated parameters should therefore
be seen as input parameters for the present numerical model in order to obtain an
accurate prediction.
Fig. 4 shows that the updated track receptance agrees well with the measured
receptance between 0 and 60 Hz.
50 50
10C 16C
Mobility [dB ref 10−8 m/s/N]
0 0
−25 −25
−50 −50
0 50 100 150 200 0 50 100 150 200
50 50
24C 48C
Mobility [dB ref 10−8 m/s/N]
25 25
0 0
−25 −25
−50 −50
0 50 100 150 200 0 50 100 150 200
Fig. 5. Transfer functions at (a) 10m, (b) 16 m, (c) 24 m and (d) 48 m from the track predicted
with updated track parameters (dashed line) compared to experimental results for the test section
(light grey line) and the reference section (dark grey line). The grey regions indicates the 95 %
confidence intervals.
The free field transfer function is predicted numerically by means of the 2.5D coupled
finite element – boundary element model [4] (Fig. 2b). The transfer functions are
predicted based on the updated track parameters, presented in section 3.3. Fig. 5
compares the predicted transfer functions with the experimental result. The misfit
between the prediction and the experimental results is often smaller than the difference
observed between the transfer functions at both sections. Generally, good prediction
accuracy is obtained based on the identified soil and track characteristics.
252 H. Verbraken et al.
5 Conclusions
Benchmark tests carried out at a site in El Realengo (Spain) show that an accurate
prediction of the vibration transfer is obtained based on identified soil and track
characteristics. Updating of the soil’s material damping ratio improves the prediction
accuracy. Updating of the track parameters strongly affects the track receptance but
only has a small effect on the track – free field transfer functions.
Acknowledgment. The results presented in this paper have been obtained within the
frame of the EU FP7 project RIVAS (Railway Induced Vibration Abatement Solutions)
under grant agreement No. 265754. The first author is a doctoral fellow of the
Research Foundation Flanders (FWO). The financial support is gratefully
acknowledged.
References
[1] http://www.rivas-project.eu
[2] Schevenels, M., François, S., Degrande, G.: An ElastoDynamics Toolbox for MATLAB.
Computers & Geosciences 35(8), 1752–1754 (2009)
[3] Arias, A.: A measure of earthquake intensity. In: Hansen, R.J. (ed.) Seismic Design in
Nuclear Power Plants, pp. 438–483. MIT Press, Massachusetts (1970)
[4] François, F., Schevenels, M., Lombaert, G., Galvín, P., Degrande, G.: A 2.5D coupled
FE-BE methodology for the dynamic interaction between longitudinally invariant structures
and a layered halfspace. Computer Methods in Applied Mechanics and Engineering 199,
1536–1548 (2010)
Ground-Borne Vibration due to Railway Traffic:
A Review of Excitation Mechanisms, Prediction
Methods and Mitigation Measures
1 Introduction
Ground-borne noise and vibration due to railway traffic is a problem of large societal
and economic importance. This problem has gained considerable attention recently for a
number of reasons. Problems with congestion of road traffic in densely populated areas
have led to an increased demand for public transport, stimulating the construction of
light rail systems and underground railway lines. The development of high speed train
(HST) networks in Europe and Asia has led to a significant increase in the speed of
passenger trains and raised concerns on the corresponding environmental impact. The
desired shift of freight transport by road to rail will lead to higher axle loads and an
increased number of freight trains. Furthermore, railway noise and vibration also need
to be addressed for the development of land in urban areas adjacent to railway lines.
Railway-induced vibrations are generated by quasi-static and dynamic axle loads;
the latter are due to several mechanisms such as wheel and rail unevenness, impact
excitation due to rail joints and wheel flats, and parametric excitation due to sleeper pe-
riodicity [140]. These loads are transferred to the track, its supporting structure (ballast,
subgrade, slab or tunnel) and the soil, where vibrations propagate as elastic waves and
excite the foundations of nearby buildings (figure 1). In the frequency range between
1 and 80 Hz, building vibration is perceived as mechanical vibration of the human
body, whereas between 16 and 250 Hz, ground-borne vibrations can cause re-radiated
or structure-borne noise by vibrating walls and floors [40]. Low frequency feelable
ground-borne vibration is most often due to freight traffic or rolling stock with heavy un-
sprung masses travelling on tracks at grade in areas with soft soil. Problems of ground-
borne noise are typically encountered when the airborne noise is effectively shielded off,
i.e. in case of underground railway traffic or tracks at grade equipped with noise barri-
ers [140]. Norms and guidelines recognise discomfort to people [19, 30, 66, 67, 136],
malfunctioning of sensitive equipment [137] and damage to buildings [31, 135] as pos-
sible consequences of vibrations. Guidance on the prediction of ground-borne vibration
arising from rail systems is provided in a recent ISO standard [68].
Traffic-induced vibrations have been investigated since the beginning of the previous
century. In 1929, Hyde and Lintern refer to studies that were conducted in 1901 on the
vibration that was induced by the Central London Railway [65]. They state "there was
probably no subject before engineers in the country today, particularly those dealing
with road transport, that caused more trouble, or gave rise to more concern, than vi-
bration". Since then, a lot of research has been performed on transportation-induced
noise and vibration, e.g. for the development of prediction methods and mitigation
measures. The problem of railway-induced ground vibration is still of large concern
to various stakeholders and highly relevant in view of the rapid extension of high speed
[52, 97, 125, 139] and urban railway networks [25], the increased use of land for con-
struction of buildings near railways, and an increased public awareness of noise and
vibration.
The aim of this paper is to provide a comprehensive overview of the state of the art on
railway-induced ground vibration. The focus is on the governing physical mechanisms,
prediction, and mitigation of ground-borne vibration. The response of buildings [40, 44,
149], human perception of vibration and exposure-effect relation [49, 112, 115, 144],
and limit values in norms and guidelines are not discussed. Moreover, the discussion is
Ground-Borne Vibration due to Railway Traffic 255
mostly limited to ground-borne vibration for the case of railway traffic at grade. Many
of the elements discussed will also apply, however, to ground-borne vibration and noise
generated by underground railway traffic and trains running on viaducts [55, 139, 155].
For a more general overview on rolling noise and vibration due to railway traffic, the
reader is referred to the recent book of Thompson [140] whereas a discussion of ground
vibration for a wider range of excitation sources, including traffic, construction activi-
ties, and seismic events, can be found in the book of Semblat and Pecker [127].
The outline of this paper is as follows. In section 2, a brief recapitulation is made of
the relevant characteristics governing wave propagation in the soil and the wave fields
generated by loads at fixed and moving positions are discussed. These elements are
used to clarify quasi-static and dynamic excitation of railway-induced ground vibra-
tion in section 3. Next, it is discussed how track-soil interaction determines how the
vehicle loads are transferred to the soil. In section 4, an overview of methods for the
prediction of ground-borne vibration is given, making a distinction between numeri-
cal methods, based on physical or mechanical models, and empirical models derived
from measured data. The mitigation of railway-induced ground vibration is discussed
in section 5, where the focus is given to the mitigation measures at the source and
on the transmission path. Open points requiring further investigation are listed in the
conclusions.
From this it can be seen that elastic properties of the soil can be defined in terms of the
Lamé constants λ and μ, the Young’s modulus E and Poisson’s ratio ν or in terms of
the fundamental wave speeds Cp and Cs in each case in combination with the density ρ.
The ratio s of the dilatational and transversal wave velocities depends only on Poisson’s
ratio ν:
Cs 1 − 2ν
s= = (2.3)
Cp 2 − 2ν
Typical values for wave velocities in soft soil such as peat are Cs = 50 m/s, Cp =
357 m/s, whereas for a medium stiff soil typically Cs = 150 m/s, Cp = 300 m/s, and
for a stiff soil Cs = 400 m/s, Cp = 800 m/s. Even higher values are encountered in the
case of sandstone and rock materials.
At low frequencies, energy dissipation in the soil under cyclic excitation has been
found to be rate-independent [4]. Frequency-independent hysteretic material damping
in the soil can be modelled in the frequency domain by means of the correspondence
principle [117, 121] through the use of complex Lamé coefficients:
where βp and βs represent the frequency- independent hysteretic material damping ratios
for dilatational and shear waves, respectively. Alternatively, loss factors ηp = 2βp , ηs =
2βs are used to characterize energy dissipation.
At the free surface of a halfspace, interaction between dilatational and shear waves
results in a surface or Rayleigh wave [116]. In the case of a homogeneous halfspace, a
single Rayleigh wave mode exists with a phase velocity CR approximately equal to [1]:
0.862 + 1.14ν
CR ≈ Cs (2.6)
1+ν
For a realistic range of Poisson’s ratios, the Rayleigh wave velocity is in the range of
90% to 95% of the shear wave velocity. Equation (2.6) shows that in a homogeneous
halfspace, the Rayleigh wave velocity does not depend on the frequency, meaning that
the wave is non-dispersive. The Rayleigh wave propagates along the free surface of the
halfspace and has a limited penetration depth of approximately one wavelength.
Natural soils often have a horizontal stratification as their formation is generally gov-
erned by phenomena affecting large areas of land, such as erosion, sediment transport,
and weathering processes [39]. In this case, the soil can be modelled as a horizontally
layered elastic halfspace, where the material properties Cs , Cp , ρ, βs , and βp vary in the
vertical direction only. In a layered halfspace, multiple Rayleigh modes occur, which
are dispersive due to the variation with depth of the dynamic soil characteristics. This is
illustrated by results for a site at Lincent, Belgium, located next to the high speed line
between Brussels and Köln, where ground vibration measurements have been made for
validation of numerical models [91, 94]. Borings performed prior to the construction
of the high speed line show that the soil consists of a shallow quaterny top layer of silt
with a thickness of 1.2 m, followed by a layer of fine sand up to a depth of 3.2 m. Be-
tween 3.2 m and 7.5 m is a sequence of stiff layers of arenite (a sediment of a sandstone
Ground-Borne Vibration due to Railway Traffic 257
0 0
2 2
Depth [m]
Depth [m]
4 4
6 6
8 8
10 10
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
Displacement [−] Displacement [−]
Fig. 2. Rayleigh wave modes calculated for a layered halfspace (site at Lincent, Belgium): (a)
dispersion curves of the first three Rayleigh wave modes, (b) horizontal and (c) vertical compo-
nent of the fundamental Rayleigh wave mode as a function of depth at 10 Hz (solid line), 20 Hz
(dashed line), and 40 Hz (dotted line)
residue) embedded in clay. The small strain dynamic soil characteristics at the site have
been determined by means of two Spectral Analysis of Surface Wave tests and five Seis-
mic Cone Penetration Tests. These results show that, in the frequency range of interest
for railway-induced vibration, the soil can be represented by a single layer with a thick-
ness of 3.0 m and a shear wave velocity of about 150 m/s on top of a stiffer halfspace
with a shear wave velocity of 280 m/s. A summary of the dynamic soil characteristics
at this site as assumed in previous studies [91, 94] is shown in Table 1.
Layer d Cs Cp E ν ρ β
[m] [m/s] [m/s] [×106 N/m2 ] [-] [kg/m3 ] [-]
1 3 150 300 120 0.333 2000 0.03
2 ∞ 280 560 418 0.333 2000 0.03
Figure 2a shows the dispersion curve of the first three Rayleigh wave modes, com-
puted with the ElastoDynamics Toolbox (EDT) for MATLAB [123]. At low frequen-
cies, the fundamental Rayleigh wave has a phase velocity close to 261 m/s at low
frequencies, corresponding to the phase velocity of Rayleigh waves in the halfspace.
At high frequencies, the phase velocity converges asymptotically to a value of about
140 m/s, corresponding to the Rayleigh wave velocity in the top layer. The variation
with depth of the fundamental mode shape is shown in figures 2b and 2c. At low fre-
quencies, the Rayleigh wave reaches very deep, explaining why the phase velocity is
affected by the stiff deeper layers whereas at high frequencies, the motion is concen-
trated near the surface and dominated by the properties of the soft surface layer.
Higher order Rayleigh wave modes with a more complicated mode shape develop
at higher frequencies. The cut-on frequencies of the second and third Rayleigh wave
are observed at 20 Hz and 44 Hz, where the higher order waves have an initial phase
velocity of 280 m/s, corresponding to the shear wave velocity of the halfspace [145]. In
the Spectral Analysis of Surface Waves test, information on the dispersive behaviour of
258 G. Lombaert et al.
the Rayleigh wave modes is extracted from an in situ experiment and used for identify-
ing the variation with depth of the shear wave velocity and the corresponding material
damping ratio [101, 120]. The grey area in the background of figure 2a is a represen-
tation in the frequency - phase velocity domain of the Green’s function or fundamental
solution for the vertical displacement due to a vertical point load, both at the surface
of the layered halfspace. These results illustrate the relative importance of the different
surface wave modes in this particular problem.
In the presence of ground water, the pores between the solid skeleton may be com-
pletely saturated with water. Biot’s theory [15, 16] can be used to describe wave prop-
agation in saturated poroelastic media for the prediction of railway-induced ground
vibration [103]. According to Biot’s theory, two P-waves and a single S-wave are found
in a saturated poroelastic medium [122]. A characteristic frequency that is inversely
proportional to the permeability of the poroelastic medium, determines the importance
of the corresponding relative motion of the solid and fluid phase. For typical soils, the
characteristic frequency is in the order of several kHz, which is much higher than the
frequency range of interest here. At frequencies that are very low compared to the char-
acteristic frequency, no relative motion between the fluid and solid phase occurs for
the first type of P-wave and the S-wave. For the second type of P-wave, the fluid and
solid phase move perfectly out-of-phase. This type of wave, however, is characterized
by very high attenuation coefficients, meaning that wave propagation essentially occurs
through a single P-wave and S-wave. In the frequency range of interest for railway in-
duced ground vibration, the soil can therefore be modelled as an equivalent dry elastic
medium, provided that the density and incompressibility of the saturated soil layers are
accounted for by equivalent Lamé coefficients μeq and λeq [122]. The resulting P-wave
velocity is high as the presence of the pore fluid results in low compressibility while
the S-wave velocity of the orginal dry solid is only weakly affected due to the change
in density. Moderate seasonal variations of the ground water level were found not to
significantly affect peak values of ground vibrations produced by road traffic [122].
For a point load acting at a fixed position, the relation between the soil’s displacement
response û(r, ω) and the load amplitude fˆ(ω) can be written in the frequency domain as
follows because of the assumed linear elastic constitutive behaviour:
where Ĥ(r, ω) is the transfer function and r is the distance between the receiver and
the source. A hat above a variable denotes its representation in the frequency domain.
Figure 3a shows the modulus of the transfer function that represents the vertical dis-
placement response at 8, 16, 32 and 64 m from a vertical point source for the site at
Lincent. At very low frequencies, the attenuation with distance is proportional to 1/r.
A steep rise as a function of the frequency is found between 10 and 20 Hz. This occurs
at the onset of wave propagation in the surface layer, when one quarter of the Rayleigh
wavelength computed with the properties of the top soil material fits within the thick-
ness of this layer. A moderate peak is found in the transfer functions near the frequency
Ground-Borne Vibration due to Railway Traffic 259
at which half the Rayleigh wavelength fits within the surface layer [8]. From this fre-
quency on, the surface wave mainly propagates in the soft top layer (figures 2b and 2c).
The peaks and troughs observed for each of the transfer functions in figure 3a are due
to interference between different types of waves [122].
−9 −9
10 10
Displacement [m/N]
Displacement [m/N]
−10 −10
10 10
−11 −11
10 10
−12 −12
10 10
0 20 40 60 80 100 0 20 40 60 80 100
Frequency [Hz] Distance [m]
Fig. 3. Modulus of the transfer function Ĥ(r, ω) (a) as a function of the frequency f = ω/2π at
r = 8 m (solid line), r = 16 m (dashed line), r = 32 m (dotted line), and r = 64 m (dashed-dotted
line) from the source and (b) as a function of the distance r from the source at f = 10 Hz (solid
line), f = 20 Hz (dashed line), and f = 40 Hz (dotted line)
Figure 3b shows the modulus of the transfer function as a function of the distance
from the source at 10, 20, and 40 Hz. When the distance from the source is small com-
pared to the dominant wavelength in the soil, the receiver is located in the so-called near
field region where different wave components can generally not be distinguished [4]. At
larger distances, P-, S- and Rayleigh wave fronts have fully developed and their arrival
can be detected. The attenuation with distance is governed by geometric attenuation and
material damping. Geometric attenuation is caused by the expansion of the wavefronts,
resulting in the spreading of energy over an increasing area. In a homogeneous halfs-
pace, √geometric attenuation of Rayleigh waves generated by a point load is proportional
to 1/ r and frequency independent. This is no longer the case for a layered halfspace,
however, as multiple Rayleigh wave modes (figure 2a) will contribute to the response
at the soil’s surface. Material damping introduces an additional frequency dependent
attenuation with distance which is due to energy dissipation in the soil. The stronger
reduction with increasing frequency of the transfer functions at larger distances from
the source (figure 3) is mainly due to the effect of material damping. As a result, the
frequency at which the peak of the transfer function is found decreases with increasing
distance from the source. The observed attenuation with distance of the transfer func-
tions of a layered halfspace as shown in figure 3 is hard to capture in analytical form
and needs to be retrieved by numerical simulations, e.g. by means of the direct stiffness
method [78, 79] or a transfer matrix approach [54, 141].
load. Even when the load amplitude is constant in time, the motion of the load will lead
to a time-varying response at a fixed point. The response depends on the magnitude of
the load speed relative to the wave velocities in the soil. Figure 4 shows the calculated
wave field generated by a point load of constant unit amplitude for three load speeds:
v = 0, v = 100 m/s, and v = 200 m/s moving on the layered halfspace representing the
soil at the site of Lincent (table 1). These results have been obtained from the fundamen-
tal solution or Green’s function of the layered halfspace by applying the Betti-Rayleigh
reciprocity theorem and exploiting the invariance of the problem geometry in the di-
rection in which the load moves [92]. Comparison of figures 4a and 4b shows that the
wave field generated by a load moving at a speed of 100 m/s, which is below the lower
limit of the phase velocities of surface waves (figure 2a), is similar to the one of a load
at a fixed position. No propagating waves are emitted by the moving load in this case.
An observer at a fixed position in the free field will observe a time variant response,
however, as the deflection shape travels with the load at a speed of 100 m/s. When the
load speed exceeds the lower limit of the phase velocities of surface waves, the wave
field generated by the moving load changes drastically. In this case, propagating waves
are generated by the moving load, giving rise to a sharp discontinuity in the wave field,
and the formation of a Mach cone. The displacement amplitudes are significantly larger
in this case.
When considering railway traffic, the critical load speed at which large amplifications
of track displacements are expected, also depends on the properties of the track, but is
generally controlled by the Rayleigh wave velocities in the soil [97]. For high speed
−8 −8
x 10 x 10
1 1
Displacement [m]
Displacement [m]
0 0
−1 −1
−25 −25
0 25 0 25
0 0
25 −25 25 −25
x [m] y [m] x [m] y [m]
(a) (b)
−8
x 10
1
Displacement [m]
−1
−25
0 25
0
25 −25
(c) x [m] y [m]
Fig. 4. Wave field generated by a unit point load with constant amplitude as seen in a moving
frame: (a) at a fixed position, (b) moving at a speed of 100 m/s, and (c) moving at a speed of
200 m/s (peak amplitudes clipped). Calculated results for soil properties of Lincent, Belgium.
Ground-Borne Vibration due to Railway Traffic 261
trains (v ≥ 180 km/h) travelling on tracks supported by very soft soil (Cs ≤ 50 m/s),
the train speed can actually be close to or even larger than this critical load speed. In
the 1990’s, the problem of trains running at trans-critical speeds was addressed in a se-
ries of papers by Krylov [85, 86, 88] focussing on ground vibration and Dieterman and
Metrikine [32, 33] investigating track-soil interaction. These studies have indicated that
a significant amplification of vibration levels and track displacements is obtained com-
pared with the subcritical speed range, leading to problems of track stability and safety.
This was confirmed by field measurements at the site of Ledsgård along the West Coast
Line in Sweden where track displacements up to 10 mm have been measured during the
passage of the X2000 train [3]. This case has been considered in numerical simulations
by many authors, among which the first were Madshus and Kaynia [97]. Problems en-
countered with high speed trains running at trans-Rayleigh speeds are primarily of large
concern for reasons of track stability, however, necessitating subgrade stiffening or piled
track foundations, and will rarely be an issue for problems of environmental ground-
borne vibration. In the following, focus will therefore be on the subcritical range of
train speeds.
−8 −8
x 10 x 10
1 1
Displacement [m]
Displacement [m]
0 0
−1 −1
−25 −25
0 25 0 25
0 0
25 −25 25 −25
x [m] y [m] x [m] y [m]
(a) (b)
−8
x 10
1
Displacement [m]
−1
−25
0 25
0
25 −25
(c) x [m] y [m]
Fig. 5. Wave field generated by a unit point load with harmonic time variation ( f = 25 Hz) as
seen in a moving frame: (a) at a fixed position, (b) moving at a speed of 100 m/s, and (c) moving
at a speed of 200 m/s (peak amplitudes clipped). Calculated results for soil properties of Lincent,
Belgium.
Whereas a moving load with constant amplitude only generates propagating waves
at trans-Rayleigh load speeds, a moving load with time-varying amplitude will gener-
ate propagating waves, irrespective of the load speed, similarly to the case of a fixed
262 G. Lombaert et al.
position. This is illustrated in figure 5 which shows the wave field generated by a unit
point load with harmonic variation in time ( f = 25 Hz) for the same three load speeds.
Comparison of figures 5a and 5b clearly shows the Doppler effect in the case of a mov-
ing load. In front of the moving load (figure 5b), shorter wavelengths are observed, in
agreement with the higher frequency observed at a fixed receiver ahead of the source,
whereas the longer wavelengths behind the load correspond to the lower frequencies
observed at a point behind the source. The Doppler effect has been identified in field
measurements of railway-induced vibrations by Ditzel et al. [35]. When the load speed
exceeds the lowest Rayleigh wave velocity in the soil, the wave field displays a Mach
cone (figure 5c) as in the case of a load with constant amplitude (figure 4c), but without
the large amplification observed in the latter case.
Due to the load motion, equation (2.7) that relates the response to the load amplitude
for a load at a fixed position, is no longer valid. For train speeds that are low compared
with the wave velocities in the soil, however, the effect of the moving load and the
corresponding Doppler shift are small and equation (2.7) holds approximately at the
time when the load and receiver are at an intermediate distance r.
3 Excitation Mechanisms
3.1 Quasi-Static Excitation
The loads applied to the track by a running train can be decomposed into a static and
dynamic load component. Assuming a linear behaviour of the track and the support-
ing soil, the resulting ground vibration u(t) can be decomposed into the quasi-static
contribution us (t) and the dynamic contribution ud (t):
u(t) = us (t) + ud (t) (3.1)
When the train speed is situated in the subcritical range, the quasi-static response of
the soil resembles a sequence of bowl shaped deflections, each similar in form to the
one for an upward load shown in figure 4b, that travel with the train. The time vari-
ation of the response at a fixed point is therefore due to successive rising and falling
of the response at the passage of each axle. The repeated passage of axles leads to the
characteristic peaks and troughs in the narrow-band frequency spectrum of the response
that are determined by the axle and bogie passage frequencies [9, 10, 28]. This can be
understood by writing the quasi-static response as a superposition of the contributions
of different train axles:
na
us (t) = wk us0 (t − yk0 /v) (3.2)
k=1
where na is the number of axles of the train, wk is the weight carried by axle k, us0 (t)
is the response due to a moving load with unit magnitude and yk0 is the position of the
axle on the track at a reference time t = 0. In the frequency domain, this expression
becomes:
⎡ na y ⎤⎥
⎢⎢⎢ k0 ⎥
⎥⎥⎥
ûs (ω) = ûs0 (ω) ⎢⎢⎣ wk exp iω (3.3)
k=1
v ⎦
Ground-Borne Vibration due to Railway Traffic 263
where the bracketed term that depends on the distribution of the weight over the axles
and the train speed v gives rise to the characteristic shape of the narrow band response
spectrum [9, 10, 28].
This is now illustrated for the computed response of the sleeper due to the passage
of an InterCity train at the site of Lincent. For more details regarding the model and
input parameters, the reader is referred to Lombaert and Degrande [91]. Figures 6a and
6b show the time history and narrow-band spectrum of the sleeper velocity due to the
passage of the first axle of an InterCity train at a speed of 156 km/h (43.3 m/s), which is
well below the lower limit of the Rayleigh wave velocities in the soil. Due to the short
duration of the quasi-static sleeper response for a single axle (figure 6a), the passages
of individual axles do not overlap in time and are still observed in the time history
response for the entire train (figure 6c). The results show that the InterCity train is in
"pull mode" where the axles of the locomotive that carry the largest weight come first.
Comparison of the narrow-band spectra for a single axle (figure 6b) and the full train
(figure 6d) shows how the characteristic peaks and troughs appear by multiplication
with the bracketed term in equation (3.3).
−3
x 10
0.05
1
Velocity [m/s/Hz]
Velocity [m/s]
0.8
0 0.6
0.4
0.2
−0.05 0
−0.1 −0.05 0 0.05 0.1 0 20 40 60 80 100
Time [s] Frequency [Hz]
(a) (b)
0.05 0.012
0.01
Velocity [m/s/Hz]
Velocity [m/s]
0.008
0 0.006
0.004
0.002
−0.05 0
−5 0 5 0 20 40 60 80 100
Time [s] Frequency [Hz]
(c) (d)
Fig. 6. Quasi-static contribution to the sleeper response due to the passage of an InterCity train
at a speed of 156 km/h: (a) time history and (b) narrow-band spectrum of the velocity during the
passage of the first axle of the locomotive and (c) time history and (d) narrow-band spectrum of
the velocity during the passage of the entire train
In the subcritical speed range considered here, the narrow-band spectrum of the
quasi-static contribution to the free field response becomes increasingly concentrated
at low frequencies with increasing distance from the track. This is due to the charac-
teristic bowl shape of the deflection field for each axle (figure 4b). At a larger distance
264 G. Lombaert et al.
from the track, the arrival of a single axle is detected earlier and its passage lasts for
a longer time. The time scale at which the response rises and falls therefore increases
with the distance from the track, implying that its representation in the frequency do-
main gets more concentrated at lower frequencies. This is fundamentally different from
the aforementioned reduction in the frequency at which the peak value of the transfer
functions (figure 3) is found, as the latter is due to energy dissipation characterized by
the material damping ratio. Furthermore, as the distance from the track increases, con-
tributions from different axles and bogies coalesce and can no longer be identified in
the time history of the vibration response.
Wheel and Track Unevenness. In order to understand how parameters of the vehicle,
track, and supporting soil affect the dynamic load component, it is instructive to con-
sider how the dynamic load component can be computed from the combined wheel and
track unevenness [9, 94, 131]. First, it is assumed that a perfect contact exists between
the train and the track, implying that the following expression should hold:
where ua (ω) and ut (ω) collect the displacements of the axles and the track at the na mov-
ing contact points. The vector uw/r (ω) collects the combined wheel and rail unevenness
perceived by the axles.
Next, the displacements ua (ω) and ut (ω) are expressed in terms of the dynamic train
loads by means of the track and vehicle compliance matrices Ĉt (ω) and Ĉv (ω) that
relate the displacements at the multiple moving contact points to the vector ĝd (ω) of
dynamic train loads [94]:
Ĉt (ω) + Ĉv (ω) ĝd (ω) = −ûw/r (ω) (3.5)
An additional compliance matrix representing the contact spring is in some cases con-
sidered in equation (3.5) to account for Hertzian contact between the wheel and the rail
[130]. In the frequency range of interest for ground-borne vibration, the contact stiffness
is relatively high compared to the track and vehicle stiffness, so that the corresponding
compliance can generally be disregarded. Note that equation (3.5) does not allow ac-
counting for large deflections or loss of contact, e.g. in the presence of wheel flats [154];
this would require a non-linear model. Furthermore, it has been implicity assumed that
Ground-Borne Vibration due to Railway Traffic 265
the track dynamic properties are translationally invariant in the longitudinal direction
of the track. This assumption is also acceptable for discretely supported ballasted tracks
as the rail receptance at a sleeper and in between two sleepers are similar in the low fre-
quency range of interest for ground-borne vibration [82, 154]. Equation (3.5) therefore
provides a reasonable basis for investigating how vehicle-track interaction affects the
dynamic train loads.
Simplifying equation (3.5) for a single axle allows effects from vehicle-track inter-
action to be highlighted. If the train speed is relatively low, the corresponding track
compliance in a moving frame of reference is approximately equal to the track recep-
tance, and can roughly be represented as the inverse of the track stiffness 1/kt when
its imaginary part and frequency dependence are disregarded. At frequencies of more
than a few Hertz [82], the vehicle’s primary and secondary suspension isolate the bogie
and the body from the wheelset. At sufficiently high frequencies (above approximately
10 Hz), the vehicle’s unsprung mass Mu is therefore the only component that affects
the vertical dynamic loads and can be represented as a rigid body [82]. In this case,
the compliance of a single axle becomes −1/(Mu ω2 ). Introducing these expressions in
equation (3.5) leads to the following expression for the dynamic load at axle k:
kt Mu ω2
ĝdk (ω) = − ûk (ω) (3.6)
kt − Mu ω2 w/r
The denominator
√ on the right hand side of equation (3.6) becomes zero at the frequency
ω = kt /Mu where "resonance" of the unsprung mass on the track stiffness (sometimes
known as the P2 resonance) occurs. At this frequency, the narrow-band spectrum of the
dynamic vehicle load ĝdk (ω) displays a resonance peak. The peak is strongly damped
in reality as the track stiffness has a significant imaginary part which is mainly due to
radiation damping.
Disregarding wheel unevenness for the moment, the frequency domain representa-
tion of the unevenness ûkw/r (ω) experienced by axle k in equation (3.6) can be computed
from the wavenumber domain representation ũw/r (ky ) of the unevenness uw/r (y) along
the track:
1 ω y
k0
ûkw/r (ω) = ũkw/r − exp iω (3.7)
v v v
where yk0 is the position of the axle on the track at a reference time t = 0. The ge-
ometrical irregularities considered here include broad-band unevenness of a random
nature, typically modelled as a stationary random process characterized by its power
spectral density (PSD) [108], but can also represent localized irregularities such as rail
joints giving rise to impact excitation. Equation (3.7) indicates that, in the absence of
wheel unevenness, all axles experience the same excitation apart from a shift in time
accounted for by the phase shift exp (iωyk0 /v) in equation (3.7). When all train carriages
have similar characteristics and track unevenness is the dominant source of excitation,
the resulting dynamic loads will also be similar and the narrow-band spectrum of the
dynamic response contribution will have a similar shape as the one of the quasi-static
response contribution.
Equation (3.7) relates the spectral content of the unevenness in the wavenumber do-
main or, equivalently, in terms of its reciprocal, the wavelength λy = 2π/ky, to the
266 G. Lombaert et al.
Table 2. Relevant range of wavelengths λy for ground-borne vibration and noise as a function of
the train speed v
−3
x 10
2
200
Velocity [dBref 10−8 m/s]
Velocity [m/s]
1 150
0 100
50
−1
0
−2 0 1 2
−5 0 5 10 10 10
Time [s] Frequency [Hz]
(a) (b)
Fig. 7. Free field response at 16 m from the track due to the passage of an InterCity train at a
speed of 156 km/h: (a) time history and (b) one-third octave band spectra of the total response
(solid line), quasi-static (dashed line) and dynamic (dotted line) response contribution
Ground-Borne Vibration due to Railway Traffic 267
time history of the free field velocity at 16 m from the track. In contrast to what was
observed for the quasi-static contribution to the sleeper response in figure 6c, the pas-
sage of individual axles can no longer be identified. Even when the dynamic response
contribution was not accounted for, this would be the case as at a larger distance from
the track, the quasi-static response contributions from different axles and bogies coa-
lesce. Figure 7b shows the one-third octave band spectra of the free field velocity at
16 m from the track. The one-third octave band spectra are computed according to the
German standard DIN 45672-2 [29] for a reference period T 2 which is here considered
as the duration of the stationary part of the response. Comparing the one-third octave
band spectra of the total response, and the quasi-static and dynamic response contri-
butions in figure 7b shows that the latter dominates the total response. A significant
contribution of the quasi-static excitation to the total response is in this case only found
in the frequency range below 3 Hz.
The quasi-static contribution to the response generally remains important, however,
in the immediate vicinity of the track. No attempt can therefore be made to use equation
(2.7) for estimating the dynamic loads applied to the track from the response measured
close to the track and transfer functions Ĥ(r, ω) between the track and the free field
[143]. Furthermore, this also implies that the estimation of insertion loss values from
field tests requires measurements at a sufficiently large distance from the track. The
relative importance of quasi-static and dynamic excitation depends on the train speed,
the ratio of the static and dynamic axle loads, and the dynamic characteristics of the
track and the soil as indicated by Sheng et al. [130, 131], Auersch [10], Lombaert and
Degrande [91], and Triepaischajonsak et al. [142, 143].
where the coupling between the position xk (τ) of the k-th axle (k = 1, . . . , na ) and the
time history of the load gk (τ) through the time τ gives rise to the Doppler effect. Due to
Ground-Borne Vibration due to Railway Traffic 269
this coupling, the expression cannot be rewritten in the simple form of equation (2.7)
unless the train is assumed to be at a fixed position, i.e. xk (τ) ≈ xk0 , as for calculating
the stationary part of the vibration response [148].
The transfer function Hts (x, x , t) between the track and the free field in equation (4.1)
determines how the load applied at a position x on the track is transferred to the soil,
to provide response at a point x . The transfer is determined by track-soil interaction,
which will give rise to waves of the coupled system that propagate along the track.
The resulting load transfer could be compared with a combined filtering in the time
and space domain. Generally, a distribution of the load over a larger area, determined
by the track width and load spreading in the direction along the track, will result in
a reduction in the high frequency vibration transmitted to the free field. This is due
to destructive interference of waves transmitted by different parts of the contact area.
An accurate prediction of the stress distribution at the track-soil interface is therefore
essential for predicting ground-borne vibration due to railway traffic, in particular when
the wavelength in the soil is comparable or smaller than the characteristic dimension of
the stress distribution area [134].
General purpose 3D finite element (FE) methods offer the largest flexibility in mod-
elling, but require appropriate procedures to avoid spurious reflections at the bound-
aries of the finite volume of soil accounted for in the analysis [36, 37]. Alternatively,
3D coupled finite element - boundary element (BE) methods can be used [47, 107]. The
versatility of 3D FE and 3D FE/BE models comes at a very high computational cost,
however. Dedicated models have therefore been developed that exploit the (assumed)
regularity of the track and the underlying soil.
When the track and the soil are regarded as translationally invariant, a Fourier trans-
formation with respect to the coordinate y along the track leads to an efficient solution
in the frequency-wavenumber domain [7, 45]:
+∞
ũ(x, ky , z, ω) = û(x, y, z, ω) exp(iky y) dy
−∞
+∞
1
û(x, y, z, ω) = ũ(x, ky , z, ω) exp(−iky y) dky (4.2)
2π −∞
Notwithstanding the large recent progress in the development of numerical models for
railway-induced ground vibration, they are still mainly used for research. In contrast,
engineering practice mostly still makes use of empirical methods. The ISO 14837-1
standard that provides general guidance on ground-borne noise and vibration arising
from rail systems [68] indicates that requirements for absolute predictions change dur-
ing the various stages of development and distinguishes between scoping models (ear-
liest stage), environmental assessment models (planning process) and detailed design
models (part of construction and design). This categorization is seen in many of the
empirical models in use but, of course, also applies to numerical models.
Examples of empirical methods include the procedures developed by the Federal
Railroad Administration (FRA) and the Federal Transit Administration (FTA) of the
U.S. Department of Transportation [52, 53], the method developed by the Swiss Federal
Railways (SBB) [90], the method of Madshus et al. [96] which was based on measure-
ments in Norway and Sweden, and the method of Hood et al. [58] which was developed
within the frame of the Channel Tunnel Rail Link project in the UK. The procedures
developed by FRA and FTA distinguish between the three different levels of assess-
ment of the ISO 14837-1 standard [68]. The Detailed Vibration Assessment is based on
a prediction technique developed by Bovey [18] and Nelson and Saurenman [104] and
presents a more elaborate method for the prediction of ground-borne vibrations and re-
radiated noise in buildings. The method developed by SBB [90] distinguishes between
two prediction models, VIBRA-1 and VIBRA-2, where the latter is more detailed and
considers, for example, frequency-dependent attenuation models. The empirical meth-
ods by Madshus et al. [96] and Hood et al. [58] follow a similar structure as the one by
SBB, and additionally consider the issue of prediction uncertainty.
The aforementioned empirical methods aim at predicting the quasi-stationary re-
sponse during a train passage and can be cast in the following general form of the ISO
14837-1 standard [68]:
where A( f ) is the magnitude of ground vibration, typically a root mean square value in
one-third octave bands for detailed design situations, S ( f ) is the source strength, P( f )
characterizes the propagation path, and R( f ) the receiver. ISO 14837-1 [68] stipulates
that each of these terms should be further divided into relevant components, which in-
teract and can only be assumed uncoupled in some situations for simplified models.
Leaving out of consideration the receiver term R( f ), equation (4.3) has the same struc-
ture as the one in equation (2.7) for a dynamic load at a fixed position. The latter ex-
pression is strictly speaking not valid for the narrow-band spectrum of the response due
to a moving dynamic load because of the Doppler effect. A reasonable estimate for the
quasi-stationary response is still obtained, however, when the train speed is relatively
small compared to the wave velocities in the soil [148].
272 G. Lombaert et al.
The use of empirical methods is limited to those cases where a suitable characterization
of the source strength (force density, reference vibration level) and the vibration trans-
mission (line source transfer mobility, attenuation law) is available. In order to over-
come these limitations, empirical methods could be combined with numerical methods
in a hybrid or semi-empirical prediction procedure.
Hybrid prediction methods seem particularly appealing in the case of new tracks,
new rolling stock or modifications of existing tracks. Numerical models can here be
used to assess the influence of rolling stock characteristics and track design on ground
vibration, avoiding the need for in situ tests possibly requiring the construction of test
tracks. In the case of new tracks, a numerically predicted source strength can be intro-
duced in equation (4.3) and combined with an experimentally determined transfer func-
tion or attenuation law that inherently takes into account the particularities of vibration
transmission. Since, for the case of new track infrastructure, these transfer functions
are determined in the absence of the track, an additional correction may be needed.
If modifications are made to the rolling stock or track structure, numerical modelling
can be used to evaluate the change in source strength and vibration transfer in equation
(4.3), so as to correct the existing vibration level. Accurately predicting relative levels
of vibration due to changes in track or vehicle parameters is expected to be much more
reliable than making predictions in absolute terms [72].
Hybrid prediction methods may therefore allow the accuracy of numerical models to
be improved by an adequate characterization of the vibration transfer, while at the same
time providing the flexibility of numerical models to assess a wide range of rolling stock
and track parameters. In order to introduce results from numerical models in empirical
predictions, the terms in equation (4.3) or corrections to them could be computed by
direct simulation of experiments as e.g. prescribed in the Detailed Vibration Assessment
of the U.S. Department of Transportation or an analytical expression of these terms can
be derived [148].
Track Structure. A third way to intervene with the aim of reducing ground-borne noise
and vibration is by modification of the track structure. In order to isolate dynamically
part of the track superstructure, resilient elements can be included at different levels
in the track structure. Examples of resilient track elements include resilient fasteners
[102], sleeper pads [17, 51, 70, 126], ballast mats [21, 53, 102, 150], and slab mats in
floating slab tracks [102, 151]. As a general rule, measures providing resilience at a
Ground-Borne Vibration due to Railway Traffic 275
lower level in the track will dynamically isolate a larger part of the track mass and work
in a lower frequency range. Ballast mats and floating slab tracks therefore seem best
suited for treatment of ground-borne vibration whereas other measures are, at least at
first sight, mostly of benefit for reduction of ground-borne noise.
The working principle of resilient elements such as ballast mats [150] or floating
slab tracks [80, 151] is usually demonstrated by considering the force transmissibility,
i.e. the ratio of force applied to the system and transmitted force, of a single degree of
freedom (SDOF) system [140]. When considering a SDOF system consisting of a mass
including the vehicle’s unsprung mass and the track mass and a spring representing
the track stiffness, the force transmissibility is controlled by the coupled wheel/track
resonance frequency [140]. The resonance frequency and, therefore, the track stiffness
should be as low as possible for an effective vibration reduction. Resonance frequen-
cies between 8 and 16 Hz are reported for floating slab tracks in the USA and Canada
[102], the Singapore Mass Rapid Transit system [25] and the North-South high speed
train connection through the city of Antwerp [125]. In the following, the effect of in-
cluding additional resilience in the track is assessed through equations (3.5) and (4.1)
that determine the dynamic vehicle loads and the transfer of vibrations into the free
field, respectively. Geometric track unevenness is assumed to be the primary source of
excitation for the moment. Parametric excitation will be discussed at the end.
First, equation (3.5) shows that a reduction of the track stiffness will affect train-track
interaction and, therefore, the dynamic vehicle loads.√ Reduction of the track stiffness kt
will, for example, result in a lower frequency ω = kt /Mu at which resonance occurs of
the unsprung mass on the track stiffness. This leads to a shift of the frequency spectrum
of the dynamic vehicle load to lower frequencies and, in the case of unmodified wheel
and track unevenness, amplification of the dynamic vehicle load around the new reso-
nance frequency and a reduction at higher frequencies, in particular around the original
resonance frequency.
Second, the introduction of resilient elements will also modify the transfer of vibra-
tion from the track to the soil as characterized by the transfer function Hts (xk (τ), x , t−τ)
in equation (4.1). The introduction of a resilient layer will lead to a (reduced) cut-on
frequency of waves which predominantly travel in the upper part of the track on top
of the resilient layer. When the soil flexibility is disregarded, the cut-on frequencies of
propagating track waves are the natural frequencies of the 2D model of the track sec-
tion [94]. At sufficiently high frequencies, the vibration energy is effectively confined
in the upper track part and less energy is radiated into the soil. The actual reduction in
the free field is determined by the relation between the phase velocities of the waves
propagating in the track and Rayleigh waves in the soil [94]. In case of unmodified dy-
namic vehicle loads, an amplification of ground-borne vibration or negative insertion
loss values are found around the newly introduced cut-on frequency and a reduction
or positive insertion loss values at higher frequencies. For an effective vibration reduc-
tion, the cut-on frequency and, therefore, the stiffness of the resilient layer should be as
low as possible. The latter is limited, however, by the maximum allowable static track
deflection.
The combined effect of the modified dynamic load and the modification of the trans-
fer function determines the overall effect of the resilient element on the ground vibration
276 G. Lombaert et al.
during a train passage. For an accurate and detailed quantification of the vibration reduc-
tion, three-dimensional numerical models are needed that allow for interaction between
waves propagating in the track and in the soil [62, 95].
In the previous discussion, geometric track unevenness was assumed to be the main
source of excitation. Adding resilience in the track system may also lead to a reduction
of parametric excitation due to differences in support stiffness [60, 70]. The additional
resilience in the track support will to some extent smooth variations in support stiffness,
e.g. due to variation in ballast or subgrade properties. In this way, resilient rail fastening
systems, rail pads or sleeper pads will lead to a reduction of vibration at relatively low
frequencies [60], which is unexpected from the discussion above.
Track Subgrade. Stiffening of subgrade under the track is essentially performed for
improving the bearing capacity of soft soils and avoiding excessive track settlements,
but has also been shown to be effective in reducing ground vibration [2], in particular
when quasi-static excitation is important [111]. Various techniques for soil stiffening
have been developed, some of which also allow for the treatment of soil under existing
tracks. Subgrade stiffening is not considered as a practical option for vibration mitiga-
tion by railway operators because the works generally require interruption of railway
operation.
Mitigation measures in the transmission path aim at impeding the propagation of elastic
waves travelling in the soil from the railway track to nearby buildings. Examples include
open trenches, soft and stiff wave barriers, wave impeding barriers, and placement of
heavy masses next to the track.
An open trench (figure 8a) is effective in cutting off the propagation of surface or
Rayleigh waves if the depth of the trench is at least half the Rayleigh wavelength
[119, 153]. The benefit is lost beyond a certain distance, however, due to the fact that
diffraction can occur around the bottom of the barrier. For stability reasons, an open
trench is limited to shallow depths, and the presence of ground water compromises its
efficiency. This necessitates the use of soft or stiff in-fill materials (figure 8b). When
including soft in-fill materials, the stiffness of the material should be as low as possi-
ble to limit the transmission of vibration and approach the ideal case of an open trench
[46, 69]. Gas cushions [98], rubber chips [81], and polystyrene [27] have been used for
creating soft wave barriers. Design rules that have been determined using calculations
and experiments [14, 59] indicate that impractical depths of trench are mostly required
to achieve high attenuations of vibration in a homogeneous ground at low frequencies.
Often, however, a relatively thin upper layer of soil is important in determining the
nature of surface wave propagation, with increased propagation for frequencies where
surface waves can propagate predominantly in this upper layer. A trench may be ex-
pected to have a greater effect in such a layered ground than in the homogeneous case
[48, 69, 73].
A stiff wave barrier [23] is realized by a concrete slab, a row of steel or concrete
piles [77], a sheet pile wall [34] or a jet grouting wall [24]. In this case, the lateral
Ground-Borne Vibration due to Railway Traffic 277
stability of the screen is not an issue and the installation of the screen may be more
straightforward. A stiff wave barrier essentially behaves as a stiff beam embedded in a
softer material and the transmission of waves propagating in the soil is impeded when
the trace wavelength of the incident waves is smaller than the bending wavelength in
the stiffened block [24]. The mitigation measure will be more effective at sites with a
soft soil.
(d)
Fig. 8. Vibration mitigation measures on the transmission path: (a) open trech, (b) soft or stiff
wave barrier, (c) wave impeding block, and (d) heavy masses next to the track
Wave impeding blocks (figure 8c) are stiff inclusions placed under or next to the
railway track in order to modify the wave propagation in the soil [110, 138]. The idea
is based on the existence of a cut-off frequency in a soil. Below the cut-off frequency
(figure 3a), no propagating modes exist in the soft top layer. By installing a sufficiently
stiff wave impeding block in the soil [6, 133], an attempt is made to reduce the thickess
of the soft top layer so as to increase the cut-off frequency of the soil. In some cases
it has been found, however, that the vibration level is amplified at frequencies higher
than the cut-off frequency [110], so that appropriate care must be taken in the design.
Ideally, a rigid substratum with infinite lateral dimensions is created, in order to obtain
the desired cut-off of surface waves. In practice, however, only wave impeding blocks
with finite dimensions can be created. According to Takemiya and Fujiwara [138], the
width of the WIB should be sufficiently large compared with the dominant wavelength
in the soil.
A simple idea that has been proposed to reduce ground-borne vibration due to rail-
way traffic at grade is to place heavy masses such as gabion walls composed of stone
baskets or concrete blocks on the ground surface next to the track [71, 87]. These ir-
regularities are expected to cause a scattering of the incident surface waves, resulting
in a reduction of the transmitted wave field. When effective, these measures could be
designed in conjunction with noise barriers such as gabion walls in order to obtain an
integrated solution for ground-borne vibration and airborne noise. Up to now, few stud-
278 G. Lombaert et al.
ies have been carried out, however about the practical use of masses for mitigation of
railway-induced vibration.
Initial numerical analysis and preliminary experiments show that mitigation mea-
sures on the transmission path offer the prospect of obtaining vibration reduction at low
frequencies [140]. More research is currently needed, however, to confirm these initial
findings.
6 Conclusion
References
[1] Achenbach, J.: Wave propagation in elastic solids. North-Holland Series in Applied Math-
ematics and Mechanics, vol. 16. North-Holland, Amsterdam (1973)
[2] Adam, D., Vogel, A., Zimmermann, A.: Ground improvement techniques beneath existing
rail tracks. Ground Improvement 11(4), 229–235 (2007)
[3] Adolfsson, K., Andréasson, B., Bengtson, P.-E., Bodare, A., Madshus, C., Massarch, R.,
Wallmark, G., Zackrisson, P.: High speed lines on soft ground. Evaluation and analyses of
measurements from the West Coast Line. Technical report, Banverket, Sweden (1999)
[4] Aki, K., Richards, P.: Quantitative seismology, 2nd edn. University Science Books, Sausal-
ito (2002)
[5] Andersen, L., Jones, C.: Coupled boundary and finite element analysis of vibration from
railway tunnels – a comparison of two- and three-dimensional models. Journal of Sound
and Vibration 293(3-5), 611–625 (2006)
[6] Andersen, L., Nielsen, S.: Reduction of ground vibration by means of barriers or soil
improvement along a railway track. Soil Dynamics and Earthquake Engineering 25, 701–
716 (2005)
[7] Aubry, D., Clouteau, D., Bonnet, G.: Modelling of wave propagation due to fixed or mo-
bile dynamic sources. In: Chouw, N., Schmid, G. (eds.) Wave Propagation and Reduction
of Vibrations Workshop Wave 1994, pp. 109–121. Ruhr Universität Bochum, Germany
(1994)
[8] Auersch, L.: Wave propagation in layered soils: theoretical solution in wavenumber do-
main and experimental results of hammer and railway traffic excitation. Journal of Sound
and Vibration 173(2), 233–264 (1994)
[9] Auersch, L.: The excitation of ground vibration by rail traffic: theory of vehicle-track-soil
interaction and measurements on high-speed lines. Journal of Sound and Vibration 284(1-
2), 103–132 (2005)
[10] Auersch, L.: Ground vibration due to railway traffic – The calculation of the effects of mov-
ing static loads and their experimental verification. Journal of Sound and Vibration 293,
599–610 (2006)
[11] Bedford, A., Drumheller, D.: Introduction to elastic wave propagation. John Wiley and
Sons (1993)
[12] Berggren, E.: Railway Track Stiffness. Dynamic Measurements and Evaluation for Effi-
cient Maintenance. Ph.D. thesis, Royal Institute of Technology (KTH) (2009)
[13] Berggren, E., Kaynia, A., Dehlbom, B.: Identification of substructure properties of railway
tracks by dynamic stiffness measurements and simulations. Journal of Sound and Vibra-
tion 329(19), 3999–4016 (2010)
[14] Beskos, D., Dasgupta, B., Vardoulakis, I.: Vibration isolation using open or filled trenches.
Part I: 2-D homogeneous soil. Computational Mechanics 1, 43–63 (1986)
[15] Biot, M.: Theory of propagation of elastic waves in a fluid-saturated porous solid. I. low-
frequency range. Journal of the Acoustical Society of America 28(2), 168–178 (1956)
[16] Biot, M.: Theory of propagation of elastic waves in a fluid-saturated porous solid. II. high-
frequency range. Journal of the Acoustical Society of America 28(2), 179–191 (1956)
[17] Bongini, E., Lombaert, G., François, S., Degrande, G.: A parametric study of the impact
of mitigation measures on ground borne vibration due to railway traffic. In: De Roeck,
G., Degrande, G., Lombaert, G., Müller, G. (eds.) Proceedings of the 8th International
Conference on Structural Dynamics EURODYN 2011, pp. 663–670. CD-ROM, Leuven
(2011)
[18] Bovey, E.: Development of an impact method to determine the vibration transfer charac-
teristics of railway installations. Journal of Sound and Vibration 87(2), 357–370 (1983)
280 G. Lombaert et al.
[36] Ekevid, T., Lane, H., Wiberg, N.-E.: Adaptive solid wave propagation – influences of
boundary conditions in high-speed train applications. Computer Methods in Applied Me-
chanics and Engineering 195, 236–250 (2006)
[37] Ekevid, T., Wiberg, N.-E.: Wave propagation related to high-speed train. A scaled bound-
ary FE-approach for unbounded domains. Computer Methods in Applied Mechanics and
Engineering 191, 3947–3964 (2002)
[38] Esveld, C.: Modern railway track, 2nd edn. MRT-Productions, Zaltbommel (2001)
[39] Fenton, G.: Random field modeling of CPT data. Journal of Geotechnical and Geoenvi-
ronmental Engineering, Proceedings of the ASCE 125(6), 486–535 (1999)
[40] Fiala, P., Degrande, G., Augusztinovicz, F.: Numerical modelling of ground borne noise
and vibration in buildings due to surface rail traffic. Journal of Sound and Vibration 301(3-
5), 718–738 (2007)
[41] Fiala, P., Gupta, S., Degrande, G., Augusztinovicz, F.: A parametric study on counter-
measures to mitigate subway traffic induced vibration and noise in buildings. In: Sas, P.,
Bergen, B. (eds.) Proceedings of ISMA 2008 International Conference on Noise and Vi-
bration Engineering, Leuven, pp. 2751–2764 (September 2008)
[42] Forrest, J., Hunt, H.: Ground vibration generated by trains in underground tunnels. Journal
of Sound and Vibration 294, 706–736 (2006)
[43] Forrest, J., Hunt, H.: A three-dimensional tunnel model for calculation of train-induced
ground vibration. Journal of Sound and Vibration 294, 678–705 (2006)
[44] François, S., Pyl, L., Masoumi, H., Degrande, G.: The influence of dynamic soil-structure
interaction on traffic induced vibrations in buildings. Soil Dynamics and Earthquake En-
gineering 27(7), 655–674 (2007)
[45] François, S., Schevenels, M., Lombaert, G., Galvín, P., Degrande, G.: A 2.5D coupled
FE-BE methodology for the dynamic interaction between longitudinally invariant struc-
tures and a layered halfspace. Computer Methods in Applied Mechanics and Engineer-
ing 199(23-24), 1536–1548 (2010)
[46] François, S., Schevenels, M., Thyssen, B., Borgions, J., Degrande, G.: Design and effi-
ciency of a vibration isolating screen in the soil. Soil Dynamics and Earthquake Engineer-
ing 39, 113–127 (2012)
[47] Galvín, P., Romero, A., Domínguez, J.: Fully three-dimensional analysis of high-speed
train–track–soil–structure dynamic interaction. Journal of Sound and Vibration 329,
5147–5163 (2010)
[48] Garcia-Bennett, A., Jones, C.J.C., Thompson, D.J.: A numerical investigation of railway
ground vibration mitigation using a trench in a layered soil. In: Maeda, T., Gautier, P.-E.,
Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos,
P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118,
pp. 315–322. Springer, Heidelberg (2012)
[49] Gidlöf-Gunnarsson, A., Ögren, M., Terson, J., Öhrström, E.: Railway noise annoyance
and the importance of number of trains, ground vibration, and building situational factors.
Noise Health 14, 190–201 (2012)
[50] Grassie, S.: Rail irregularities, corrugation and acoustic roughness: characteristics, signifi-
cance and effects of reprofiling. Journal of Rail and Rapid Transit 226(5), 542–557 (2012)
[51] Guigou-Carter, C., Villot, M., Guillerme, B., Petit, C.: Analytical and experimental study
of sleeper SAT S312 in slab track SATEBA system. Journal of Sound and Vibration 293(3-
5), 878–887 (2006); Proceedings of the 8th International Workshop on Railway Noise
[52] Hanson, C., Towers, D., Meister, L.: High-speed ground transportation noise and vibration
impact assessment. HMMH Report 293630-4, U.S. Department of Transportation, Federal
Railroad Administration, Office of Railroad Development (October 2005)
282 G. Lombaert et al.
[53] Hanson, C., Towers, D., Meister, L.: Transit noise and vibration impact assessment. Report
FTA-VA-90-1003-06, U.S. Department of Transportation, Federal Transit Admininistra-
tion, Office of Planning and Environment (May 2006)
[54] Haskell, N.: The dispersion of surface waves on multilayered media. Bulletin of the Seis-
mological Society of America 73, 17–43 (1953)
[55] He, X., Kawatani, M., Nishiyama, S.: An analytical approach to train-induced site vibra-
tion around shinkansen viaducts. Structure and Infrastructure Engineering: Maintenance,
Management, Life-Cycle Design and Performance 6(6), 689–701 (2010)
[56] Heckl, M., Hauck, G., Wettschureck, R.: Structure-borne sound and vibration from rail
traffic. Journal of Sound and Vibration 193(1), 175–184 (1996)
[57] Hemsworth, B.: Reducing groundborne vibrations: state of the art study. Journal of Sound
and Vibration 231(3), 703–709 (2000)
[58] Hood, R., Greer, R., Breslin, M., Williams, P.: The calculation and assessment of ground-
borne noise and perceptible vibration from trains in tunnels. Journal of Sound and Vibra-
tion 193, 215–225 (1996)
[59] Hung, H., Yang, Y., Chang, D.: Wave barriers for reduction of train-induced vibrations in
soils. Journal of Geotechnical Engineering, Proceedings of the ASCE 130(12), 1283–1291
(2004)
[60] Hunt, H.: Modelling of rail roughness for the evaluation of vibration-isolation measures.
In: 12th International Congress on Sound and Vibration, Lisbon, Portugal (July 2005)
[61] Hunt, H.E.M.: Types of rail roughness and the selection of vibration isolation measures. In:
Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson,
J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 341–347. Springer, Heidelberg (2008)
[62] Hunt, H., Forrest, J.: Floating slab track for vibration reduction: why simple models don’t
work. In: 7th International Congress on Sound and Vibration, Garmisch-Partenkirchen,
Germany (July 2000)
[63] Hunt, H., Hussein, M.: Vibration from railways: can we achieve better than +/-10 dB pre-
diction accuracy? In: 14th International Congress on Sound and Vibration, Cairns, Aus-
tralia (July 2007)
[64] Hussein, M., Hunt, H.: A numerical model for calculating vibration from a railway tunnel
embedded in a full-space. Journal of Sound and Vibration 305, 401–431 (2007)
[65] Hyde, J., Lintern, H.: The vibrations of roads and structures. In: Minutes of Proceedings
of the Institution of Civil Engineers, pp. 187–243 (1929)
[66] International Organization for Standardization. ISO 2631-2:1997: Mechanical vibration
and shock - Evaluation of human exposure to whole-body vibration - Part 1: General re-
quirements, 2nd edn. (1997)
[67] International Organization for Standardization. ISO 2631-2:1999: Mechanical vibration
and shock - Evaluation of human exposure to whole-body vibration - Part 2: Vibration in
buildings (1 to 80 Hz) (1999)
[68] International Organization for Standardization. ISO 14837-1:2005 Mechanical vibration
- Ground-borne noise and vibration arising from rail systems - Part 1: General guidance
(2005)
[69] Jiang, J., Toward, M.G.R., Dijckmans, A., Thompson, D.J., Degrande, G., Lombaert, G.,
Ryue, J.: Reducing railway induced ground-borne vibration by using trenches and buried
soft barriers. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T.,
Thompson, D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitiga-
tion for Rail Transportation Systems. NNFM, vol. 126, pp. 555–562. Springer, Heidelberg
(2015)
Ground-Borne Vibration due to Railway Traffic 283
[70] Johansson, A., Nielsen, J., Bolmsvik, R., Karlstrom, A., Lunden, R.: Under sleeper pads -
Influence on dynamic train-track interaction. Wear 265, 1479–1487 (2008)
[71] Jones, C.: Use of numerical-models to determine the effectiveness of anti-vibration sys-
tems for railways. Proceedings of the Institution of Civil Engineers-Transport 105(1),
43–51 (1994)
[72] Jones, C., Block, J.: Prediction of ground vibration from freight trains. Journal of Sound
and Vibration 193(1), 205–213 (1996)
[73] Jones, C., Thompson, D., Andreu-Medina, J.: Initial theoretical study of reducing surface-
propagating vibration from trains using earthworks close to the track. In: Proceedings
of the 8th International Conference on Structural Dynamics, EURODYN 2011, Leuven,
Belgium, pp. 684–691 (July 2011)
[74] Jones, S., Hunt, H.: The effect of inclined soil layers on surface vibration from under-
ground railways using the thin layer method. ASCE Journal of Engineering Mechan-
ics 137(12), 887–900 (2011)
[75] Jones, S., Hunt, H.: Voids at the tunnel–soil interface for calculation of ground vibration
from underground railways. Journal of Sound and Vibration 330(2), 245–270 (2011)
[76] Karlström, A., Boström, A.: An analytical model for train induced ground vibrations from
railways. Journal of Sound and Vibration 292, 221–241 (2006)
[77] Kattis, S., Polyzos, D., Beskos, D.: Vibration isolation by a row of piles using a 3-D fre-
quency domain BEM. International Journal for Numerical Methods in Engineering 46,
713–728 (1999)
[78] Kausel, E.: Fundamental solutions in elastodynamics: a compendium. Cambridge Univer-
sity Press, New York (2006)
[79] Kausel, E., Roësset, J.: Stiffness matrices for layered soils. Bulletin of the Seismological
Society of America 71(6), 1743–1761 (1981)
[80] Kawaharazuka, T., Hiramatsu, T., Ohkawa, H., Koyasu, M.: Experimental study on vi-
bration reduction by isolated railway. In: Proceedings of InterNoise 1996, pp. 1549–1552
(1996)
[81] Kim, M., Lee, P., Kim, D., Kwon, H.: Vibration isolation using flexible rubber chip bar-
riers. In: Schmid, G., Chouw, N. (eds.) Proceedings of the International Workshop Wave
2000, Wave Propagation, Moving Load, Vibration Reduction, pp. 289–298. A.A Balkema,
Amsterdam (2000)
[82] Knothe, K., Grassie, S.: Modelling of railway track and vehicle/track interaction at high
frequencies. Vehicle Systems Dynamics 22, 209–262 (1993)
[83] Knothe, K., Wu, Y.: Receptance behaviour of railway track and subgrade. Archive of Ap-
plied Mechanics 68, 457–470 (1998)
[84] Kouroussis, G., Verlinden, O., Conti, C.: On the interest of integrating vehicle dynamics
for the ground propagation of vibrations: the case of urban railway traffic. Vehicle Systems
Dynamics 48(12), 1553–1571 (2010)
[85] Krylov, V.: On the theory of railway-induced ground vibrations. Journal de Physique
IV 4(C5), 769–772 (1994)
[86] Krylov, V.: Generation of ground vibrations by superfast trains. Applied Acoustics 44,
149–164 (1995)
[87] Krylov, V.: Scattering of rayleigh waves by heavy masses as method of protection against
traffic-induced ground vibrations. In: Takemiya, H. (ed.) Environmental Vibrations. Pre-
diction, Monitoring, Mitigation and Evaluation, pp. 393–398. Taylor and Francis Group,
London (2005)
[88] Krylov, V., Ferguson, C.: Recent progress in the theory of railway-generated ground vibra-
tions. Proceedings of the Institute of Acoustics 17(4), 55–68 (1995)
284 G. Lombaert et al.
[89] Kuo, K., Hunt, H., Hussein, M.: The effect of a twin tunnel on the propagation of ground-
borne vibration from an underground railway. Journal of Sound and Vibration 330(25),
6203–6222 (2011)
[90] Kuppelwieser, H., Ziegler, A.: A tool for predicting vibration and structure-borne noise
immissions caused by railways. Journal of Sound and Vibration 193, 261–267 (1996)
[91] Lombaert, G., Degrande, G.: Ground-borne vibration due to static and dynamic axle loads
of InterCity and high speed trains. Journal of Sound and Vibration 319(3-5), 1036–1066
(2009)
[92] Lombaert, G., Degrande, G., Clouteau, D.: Numerical modelling of free field traffic in-
duced vibrations. Soil Dynamics and Earthquake Engineering 19(7), 473–488 (2000)
[93] Lombaert, G., Degrande, G., Galvín, P., Bongini, E., Poisson, F.: A comparison of pre-
dicted and measured ground vibrations due to high speed, passenger, and freight trains. In:
Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning,
B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transporta-
tion Systems. NNFM, vol. 118, pp. 231–238. Springer, Heidelberg (2012)
[94] Lombaert, G., Degrande, G., Kogut, J., François, S.: The experimental validation of a
numerical model for the prediction of railway induced vibrations. Journal of Sound and
Vibration 297(3-5), 512–535 (2006)
[95] Lombaert, G., Degrande, G., Vanhauwere, B., Vandeborght, B., François, S.: The control
of ground borne vibrations from railway traffic by means of continuous floating slabs.
Journal of Sound and Vibration 297(3-5), 946–961 (2006)
[96] Madshus, C., Bessason, B., Hårvik, L.: Prediction model for low frequency vibration from
high speed railways on soft ground. Journal of Sound and Vibration 193(1), 195–203
(1996)
[97] Madshus, C., Kaynia, A.: High-speed railway lines on soft ground: dynamic behaviour at
critical train speed. Journal of Sound and Vibration 231(3), 689–701 (2000)
[98] Massarsch, K.: Vibration isolation using gas-filled cushions. In: Proceedings of the Geo-
Frontiers 2005 Congress. American Society of Civil Engineers, Austin (2005)
[99] Mirza, A.A., Frid, A., Nielsen, J.C.O., Jones, C.J.C.: Ground vibrations induced by railway
traffic - the influence of vehicle parameters. In: Maeda, T., Gautier, P.-E., Hanson, C.E.,
Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise
and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 259–266.
Springer, Heidelberg (2012)
[100] Müller, K., Grundmann, H., Lenz, S.: Nonlinear interaction between a moving vehicle
and a plate elastically mounted on a tunnel. Journal of Sound and Vibration 310, 558–586
(2008)
[101] Nazarian, S., Desai, M.: Automated surface wave method: field testing. Journal of
Geotechnical Engineering, Proceedings of the ASCE 119(7), 1094–1111 (1993)
[102] Nelson, J.: Recent developments in ground-borne noise and vibration control. Journal of
Sound and Vibration 193(1), 367–376 (1996)
[103] Nelson, J.: Prediction of ground vibrations from train using seismic reflectivity methods
for a porous soil. Journal of Sound and Vibration 231(3), 727–737 (2000)
[104] Nelson, J., Saurenman, H.: A prediction procedure for rail transportation groundborne
noise and vibration. Transportation Research Record 1143, 26–35 (1987)
[105] Nielsen, J., Igeland, A.: Vertical dynamic interaction between train and track-influence of
wheel and rail imperfections. Journal of Sound and Vibration 187(5), 825–839 (1995)
[106] Nielsen, J., Oscarsson, J.: Simulation of dynamic train-track interaction with state-
dependent track properties. Journal of Sound and Vibration 275, 515–532 (2004)
[107] O’Brien, J., Rizos, D.: A 3D FEM-BEM methodology for simulation of high speed train
induced vibrations. Soil Dynamics and Earthquake Engineering 25, 289–301 (2005)
Ground-Borne Vibration due to Railway Traffic 285
[108] ORE. Question C116: Wechselwirkung zwischen Fahrzeugen und gleis, Bericht Nr. 1:
Spektrale Dichte der Unregelmässigkeiten in der Gleislage. Technical report, Office for
Research and Experiments of the International Union of Railways, Utrecht, NL (1971)
[109] Oscarsson, J.: Simulation of train-track interaction with stochastic track properties. Vehicle
Systems Dynamics 37(6), 449–469 (2002)
[110] Peplow, A., Jones, C., Petyt, M.: Surface vibration propagation over a layered halfspace
with an inclusion. Applied Acoustics 56, 283–296 (1999)
[111] Peplow, A., Kaynia, A.: Prediction and validation of traffic vibration reduction due to
cement column stabilization. Soil Dynamics and Earthquake Engineering 27, 793–802
(2007)
[112] Peris, E., Woodcock, J., Sica, G., Moorhouse, A., Waddington, D.: Annoyance due to
railway vibration at different times of the day. Journal of the Acoustical Society of Amer-
ica 131(2), 191–196 (2012)
[113] Phillips, J.E., Nelson, J.T.: Analysis and design of new floating slab track for special track-
work using finite element analysis (FEA). In: Maeda, T., Gautier, P.-E., Hanson, C.E.,
Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise
and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 275–282.
Springer, Heidelberg (2012)
[114] Pyl, L., Degrande, G., Clouteau, D.: Validation of a source-receiver model for road traffic
induced vibrations in buildings. II: Receiver model. ASCE Journal of Engineering Me-
chanics 130(12), 1394–1406 (2004)
[115] Klæboe, R., Hårvik, L., Turunen-Rise, I.H., Madshus, C.: Vibration in dwellings from
road and rail traffic-part ii: exposure-effect relationships based on ordinal logit and logistic
regression models. Applied Acoustics 64, 89–109 (2003)
[116] Rayleigh, J.: On waves propagated along the plane surface of an elastic solid. Proceedings
of the London Mathematical Society 17, 4–11 (1887)
[117] Read, W.: Stress analysis for compressible viscoelastic materials. Journal of Applied
Physics 21, 671–674 (1950)
[118] Rhayma, N., Bressolette, P., Breul, P., Fogli, M., Saussine, G.: A probabilistic approach for
estimating the behavior of railway tracks. Engineering Structures 33, 2120–2133 (2011)
[119] Richart, F., Hall, J., Woods, R.: Vibrations of soils and foundations. Prentice-Hall, Engle-
wood Cliffs (1970)
[120] Rix, G., Lai, C., Spang Jr., A.: In situ measurement of damping ratio using surface
waves. Journal of Geotechnical and Geoenvironmental Engineering, Proceedings of the
ASCE 126(5), 472–480 (2000)
[121] Rizzo, F., Shippy, D.: An application of the correspondence principle of linear viscoelas-
ticity theory. SIAM Journal on Applied Mathematics 21(2), 321–330 (1971)
[122] Schevenels, M., Degrande, G., Lombaert, G.: The influence of the depth of the ground
water table on free field road traffic induced vibrations. International Journal for Numerical
and Analytical Methods in Geomechanics 28(5), 395–419 (2004)
[123] Schevenels, M., François, S., Degrande, G.E.: An ElastoDynamics Toolbox for MATLAB.
Computers & Geosciences 35(8), 1752–1754 (2009)
[124] Schevenels, M., Lombaert, G., Degrande, G., François, S.: A probabilistic assessment of
resolution in the SASW test and its impact on the prediction of ground vibrations. Geo-
physical Journal International 172(1), 262–275 (2008)
[125] Schillemans, L.: Impact of sound and vibration of the north-south high-speed railway
connection through the city of Antwerp, Belgium. Journal of Sound and Vibration 267,
637–649 (2003)
[126] Schulte-Werning, B., Asmussen, B., Behr, W., Degen, K.G., Garburg, R.: Advancements
in noise and vibration abatement to support the noise reduction policy strategy of deutsche
286 G. Lombaert et al.
bahn. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-
Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 118, pp. 9–16. Springer, Heidelberg (2012)
[127] Semblat, J.-F., Pecker, A.: Waves and vibrations in soils: earthquakes, traffic, shocks, con-
struction works. IUSS Press, Pavia (2009)
[128] Sheng, X., Jones, C., Petyt, M.: Ground vibration generated by a harmonic load acting on
a railway track. Journal of Sound and Vibration 225(1), 3–28 (1999)
[129] Sheng, X., Jones, C., Petyt, M.: Ground vibration generated by a load moving along a
railway track. Journal of Sound and Vibration 228(1), 129–156 (1999)
[130] Sheng, X., Jones, C., Thompson, D.: A comparison of a theoretical model for quasi-
statically and dynamically induced environmental vibration from trains with measure-
ments. Journal of Sound and Vibration 267(3), 621–635 (2003)
[131] Sheng, X., Jones, C., Thompson, D.: A theoretical model for ground vibration from trains
generated by vertical track irregularities. Journal of Sound and Vibration 272(3-5), 937–
965 (2004)
[132] Sheng, X., Jones, C., Thompson, D.: Modelling ground vibrations from railways using
wavenumber finite- and boundary-element methods. Proceedings of the Royal Society A -
Mathematical, Physical and Engineering Sciences 461, 2043–2070 (2005)
[133] Sheng, X., Jones, C., Thompson, D.: Prediction of ground vibration from trains using the
wavenumber finite and boundary element methods. Journal of Sound and Vibration 293,
575–586 (2006)
[134] Steenbergen, M., Metrikine, A.: The effect of the interface conditions on the dynamic
response of a beam on a half-space to a moving load. European Journal of Mechanics,
A/Solids 26, 33–54 (2007)
[135] Stichting Bouwresearch. SBR deel A: Schade aan gebouwen door trillingen: meet- en
beoordelingsrichtlijn (2002)
[136] Stichting Bouwresearch. SBR deel B: Hinder voor personen in gebouwen door trillingen:
meet- en beoordelingsrichtlijn (2002)
[137] Stichting Bouwresearch. SBR deel C: Storing aan apparatuur door trillingen: meet- en
beoordelingsrichtlijn (2002)
[138] Takemiya, H., Fujiwara, A.: Wave propagation/impediment in a stratum and wave imped-
ing block (WIB) measured for SSI response reduction. Soil Dynamics and Earthquake
Engineering 13, 49–61 (1994)
[139] Takemiya, H., Bian, X.C.: Shinkansen high-speed train induced ground vibrations in view
of viaduct-ground interaction. Soil Dynamics and Earthquake Engineering 27, 506–520
(2007)
[140] Thompson, D.: Railway noise and vibration: mechanisms, modelling, and means of con-
trol. Elsevier, Oxford (2009)
[141] Thomson, W.: Transmission of elastic waves through a stratified solid medium. Journal of
Applied Physics 21, 89–93 (1950)
[142] Triepaischajonsak, N.: The influence of various excitation mechanisms on ground vibra-
tion from trains. Ph.D. thesis, University of Southampton (2011)
[143] Triepaischajonsak, N., Thompson, D.J., Jones, C.J.C., Ryue, J.: Track-based control mea-
sures for ground vibration. The influence of Quasi-static Loads and Dynamic Excitation.
In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning,
B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 118, pp. 249–257. Springer, Heidelberg (2012)
[144] Turunen-Rise, I., Brekke, A., Hårvik, L., Madshus, C., Klæboe, R.: Vibration in dwellings
from road and rail traffic-part i: a new norwegian measurement standard and classification
system. Applied Acoustics 64, 71–87 (2003)
Ground-Borne Vibration due to Railway Traffic 287
[145] Udías, A.: Principles of seismology. Cambridge University Press, Cambridge (1999)
[146] Varandas, J., Hölscher, P., Silva, M.: Dynamic behaviour of railway tracks on transitions
zones. Computers and Structures 89, 1468–1479 (2011)
[147] Verbraken, H., Eysermans, H., Dechief, E., François, S., Lombaert, G., Degrande, G.: Ver-
ification of an empirical prediction method for railway induced vibration. In: Maeda, T.,
Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thomp-
son, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems.
NNFM, vol. 118, pp. 239–247. Springer, Heidelberg (2012)
[148] Verbraken, H., Lombaert, G., Degrande, G.: Verification of an empirical prediction method
for railway induced vibrations by means of numerical simulations. Journal of Sound and
Vibration 330(8), 1692–1703 (2011)
[149] Villot, M., Ropars, P., Jean, P., Bongini, E., Poisson, F.: Modeling the influence of struc-
tural modifications on the response of a building to railway vibration. Noise Control Engi-
neering Journal 59(6), 641–651 (2011)
[150] Wettschureck, R., Kurze, U.: Einfügungsdämmass von Unterschottermatten. Acustica 58,
177–182 (1985)
[151] Wilson, G., Saurenman, H., Nelson, J.: Control of ground-borne noise and vibration. Jour-
nal of Sound and Vibration 87(2), 339–350 (1983)
[152] With, C., Metrikine, A., Bodare, A.: Identification of effective properties of the railway
substructure in the low-frequency range using a heavy oscillating unit on the track. Archive
of Applied Mechanics 80, 959–968 (2010)
[153] Woods, R.: Screening of surface waves in soils. Journal of the Soil Mechanics and Foun-
dation Division, Proceedings of the ASCE 94(SM4), 951–979 (1968)
[154] Wu, T., Thompson, D.: On the parametric excitation of the wheel/track system. Journal of
Sound and Vibration 278(4-5), 725–747 (2004)
[155] Xia, H., Zhang, N., Cao, Y.: Experimental study of train-induced vibrations of environ-
ments and buildings. Journal of Sound and Vibration 280, 1017–1029 (2005)
[156] Yang, Y., Hung, H.: A 2.5D finite-infinite element approach for modelling visco-elastic
bodies subjected to moving loads. International Journal for Numerical Methods in Engi-
neering 51, 1317–1336 (2001)
[157] Yang, Y., Hung, H.: Soil vibrations caused by underground moving trains. Journal of
Geotechnical and Geoenvironmental Engineering, Proceedings of the ASCE 134(11),
1633–1644 (2008)
Prediction of Railway Induced Vibration and Ground
Borne Noise Exposure in Building and Associated
Annoyance
Summary. This paper presents the prediction model used in the European
project RIVAS to evaluate the performances of the mitigation measures in
reducing railway induced vibration and ground borne noise inside buildings. A
robust empirical model (Vibra-2 from SBB) has been used, with the help of (i) a
ground structure calculation model (MEFISSTO from CSTB) to estimate the
effect of ground and building foundation changes on vibration immission, and
(ii) building acoustics theory to estimate ground borne noise from floor vibration.
The paper also presents the exposure descriptors and the associated
exposure-response relationships chosen to evaluate the mitigation measures
inside buildings in terms of attenuation of vibration and ground borne noise
exposure and corresponding decrease of annoyance. Four exposure descriptors
have been used: maximum values and equivalent values of both Wm-weighted
vibration and A-weighted ground borne noise; an idealized exposure-response
curve, the same for all descriptors has been retained, with target values deduced
from existing exposure-response curves.
1 Introduction
The European project RIVAS (2011-2013) [1] aims at developing mitigation
measures at the source level (train/track/ground) to decrease railway induced ground
vibration, particularly for freight trains; the mitigation measure performances are
determined as an insertion loss calculated from ground vibration near tracks. To
evaluate the effects of these mitigation measures on people in buildings, CSTB, as a
RIVAS partner, has defined a prediction model to translate the mitigation measure
performances in terms of attenuation of vibration and ground borne noise exposure in
building and decrease of annoyance. This paper describes the models that have been
used to estimate vibration and ground borne noise in buildings for typical situations,
and defines the exposure descriptors used for evaluating the results in terms of
annoyance. With no EC funding in RIVAS for developing new prediction models or
defining new exposure descriptors, only existing models and data as well as existing
exposure descriptors and associated exposure-annoyance relationships have been
identified and used. A simple calculation procedure (MATLAB tool) was developed
to both estimate the vibration and ground borne noise in building before and after
The ground borne vibration transmission path is decomposed into four transfer
functions as shown in Fig. 1 and as proposed in [3, 4]: TF1 from the ground (near the
tracks, 8 m in RIVAS) to the (free field) ground near the building, TF2 from the ground
(near building) to building foundations, TF3 from building foundations to the floor and
TF4 from floor vibration to room ground borne noise. They are expressed as 1/3 octave
response spectra over the frequency range 4 Hz to 250 Hz to include both vibration and
ground borne noise.
A 2.5 D BEM ground propagation model (ISVR model, [5]) was used to pre-calculate
six transfer functions (8 to 12 m, 8 to 16 m, 8 to 20 m, 8 to 24 m, 8 to 28 m and 8 to
32 m) for each soil types with input soil parameters obtained from on-site
measurements. Three reference soils (sites) have been considered in RIVAS [6]:
Horstwalde (Germany), “normal” homogeneous (cs ~ 250 m/s); Lincent (Belgium),
layered with increasing stiffness (from cs ~ 130 m/s to cs ~ 350 m/s); and Furet
(Sweden), soft (cs ~ 130 m/s).
Prediction of Railway Induced Vibration and Ground Borne Noise Exposure 291
Fig. 2. Ground to building foundation transfer function; SBB empirical model (multifamily
building category)
through the building façade; (ii) this low frequency noise (often below 100 Hz) should
not be measured using one microphone in the room center (room mode shapes show
higher sound levels close to walls or in room corners). Consequently ground borne
noise should preferably be calculated from floor vibration using floor average radiation
efficiencies commonly used in building acoustics [2]. However the input parameter is
the floor space average velocity and not the velocity at mid span mentioned above.
Assuming (i) the floor space averaged velocity is a few dB lower than the velocity at
mid span and (ii) the space averaged room noise is a few dB higher than the noise at the
room center, the following transfer function TF4 was used for concrete floors, relating
per 1/3 octave band, floor vibration levels estimated at mid span to room space
averaged sound levels, including both floor and ceiling radiation and neglecting walls
(thinner and therefore radiating less):
Lp(av) ~ Lv(meas) + 7 dB (Lv in dB ref. 5 10-8 m/s) (2.1)
This relationship lies within the 60 % confidence interval of the empirical models
proposed by SBB, is close to data from DB [10], and is consistent with the relationship
proposed by the US Federal Railroad Administration (but using another dB reference
for velocity levels) [4]
A similar radiation behavior has been experimentally found for rather heavy
(loaded) wood floors in Switzerland [7] and Germany [10].
Fig. 3. Building foundation to floor transfer function; SBB empirical model (concrete floor category)
Existing descriptors were reviewed by RIVAS [11]. The review showed that for both
vibration and noise, two descriptor types were relevant: maximum values (of running
RMS quantities), related to disturbance (short term annoyance) and traffic oriented
equivalent values (RMS or VDV), related to (long term) annoyance; so a total of four
types of descriptors have been used.
Prediction of Railway Induced Vibration and Ground Borne Noise Exposure 293
Fig. 4. Vibration exposure-response relationship (USA) for maximum values, transposed into
velocity levels in dB ref. 5 10-8 m/s
sound descriptor); EEA also suggests to deduce indoor sound levels from a mean 30 dB
façade insulation. Knowing that façades usually act as a low pass filter for airborne
noise, transmitted airborne noise and ground borne noise were supposed to have similar
low frequency spectra, allowing the use of the EEA curve for ground borne noise in the
RIVAS simplified evaluation tool .
Fig. 6. Ground borne noise exposure-response relationship (USA) for maximum values (slow
response)
All the above exposure-response curves have similar slopes (at least within the
confidence intervals given); so the same idealized curve was used for the four
descriptors, with the main features given in Fig. 7. For each descriptor, target values
Prediction of Railway Induced Vibration and Ground Borne Noise Exposure 295
4 Conclusion
In this paper, the simplified statistical model used in the RIVAS project for estimating
the vibration levels and the vibration induced sound levels generated inside buildings
by railway lines has been discussed, as well as the exposure descriptors used to estimate
annoyance. These statistical model and exposure descriptors were used in RIVAS to
evaluate and compare the impact inside buildings of the mitigation measures (improved
tracks or vehicles, track or vehicle maintenance, ground barriers, etc) developed in the
project. In this simplified evaluation tool, only mean responses of buildings on
reference grounds are estimated and the existing exposure descriptors used are not
perfect: the Wm and A weightings, single number quantities and airborne noise curves
used for GBN could certainly be further discussed; nevertheless, these descriptors have
at least been associated with exposure-annoyance curves determined from large scale
296 M. Villot et al.
surveys. Two sets of target values have been chosen, corresponding to respectively 5 %
and 10 % of HA people on the existing exposure-annoyance curves chosen. All the
evaluation results of RIVAS can be found in [6].
References
[1] RIVAS: FP7 project (2011-2013) Railway Induced Vibration Abatement Solutions, under
grant agreement 265754
[2] RIVAS Deliverable 1.6: Procedures to predict exposure in buildings and estimate
annoyance
[3] VDI 3837: Ground borne vibration in the vicinity of at-grade rail systems; spectral
prediction method (2006)
[4] US Department of Transportation, Federal Railroad Administration: High-speed ground
transportation noise and vibration impact assessment (2012)
[5] Sheng, X., Jones, C.J.C., Thompson, D.J.: Prediction of ground vibration from trains using
the wavenumber finite and boundary element methods. Journal of Sound and
Vibration 293(3-5), 575–586 (2006)
[6] RIVAS Deliverable 1.9: Evaluating the mitigation measures developed in WP2-5 in terms
of decrease of exposure and annoyance
[7] RENVIB II Final Report Phase 1 and Phase 2; ERRI Utrecht (2000)
[8] Jean, P., Guigou, C., Villot, M.: 2D ½ BEM model of ground structure interaction. Journal
of Building Acoustics 11(3), 157–173 (2004)
[9] Auersch, L., Rücker, W.: A user-friendly prediction tool for railway induced ground
vibration: Emission – Transmission – Immission. In: Schulte- Werning, B., Thompson, D.,
Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise
and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 99, pp. 129–135.
Springer, Heidelberg (2008)
[10] Said, A., et al.: Ermittlung des sekundären luftschalls aus dem Schienenverkehr. Zeitschrift
für Lärmnbekämpfung 53, Nr.1 (2006)
[11] RIVAS Deliverable 1.4 (access on web site): Review of existing standards, regulations and
guidelines as well as laboratory and field studies concerning human exposure to vibration
[12] Zapfe, J.A., et al.: Transit Cooperative Research Program (TCRP) D-12, Final report
(2009)
[13] Peris, E., et al.: Annoyance due to railway vibration at different times of the day. J. Acoust.
Soc. Am. 131(2), Express Letters (2012)
[14] European Environment Agency Technical report N° 11/2010: Good practice guide on
noise exposure and potential health effects
Attenuation of Railway Noise and Vibration in Two
Concrete Frame Multi-storey Buildings
1 Introduction
Many uncertainties exist when predicting the exposure of a building’s occupants to
groundborne noise and vibration from a railway. This is due to the number and
complexity of components in the transmission path including: vehicle and track
dynamics; ground composition; foundation type; building structure, layout and
geometry; room surface finishes, dimensions and layout. As such, uncertainties of ten
decibels or so are often cited (e.g. [1]). In addition, existing empirical prediction
methods (e.g. [2, 3]) do not give building or railway design engineers the opportunity to
explore the effect that design choices might have on the groundborne noise and
vibration exposure of building occupants.
FE modelling of buildings is considered for parametric study, with the results used
to develop more complete empirical predictions. For validation, FE models have been
developed for two multi-storey concrete frame buildings in London, for which
measured vibration data are available.
2 Empirical Guidance
The primary reference for predicting groundborne railway noise and vibration in
buildings is Chapter 16 of the Transportation Noise Reference Book (TNRB) [2]. This
advises that the range of attenuation per storey in a building is frequency dependent,
typically 2 to 6 dB (16 to 250 Hz), whereas mid-span floor amplification may be up to
around 14 dB.
The Federal Transit Administration (FTA) guidelines for noise and vibration impact
assessment [3] advises that 2 dB of vibration attenuation should be expected per storey
for the first five floors, and 1 dB of attenuation for upper storeys. Amplification of
around 6 dB may be expected near the resonance frequencies for building elements
such as floors and walls.
3 Measurements
110 110
TCR KX
Mid-span max. velocity (dB)
100 100
90 90
80 80
70 70
60 60
Basement Ground floor Basement Ground floor
50 1st floor 2nd floor 50 1st floor 2nd floor
3rd floor 5th floor 4th floor 6th floor
40 40
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
Fig. 1. Average LSmax mid-span train pass-by vibration levels; TCR (left) and KX (right)
4 FE Models
The excitation consists of an arbitrary unit force distributed over the external faces
of the basement walls and floor, with a vertical magnitude twice that in the lateral
directions. The appropriate input force direction was determined from recent site
vibration measurements at TCR that are not detailed here for brevity.
To provide sufficient data resolution, results were obtained at 1301 frequency points
between 3 and 200 Hz. the frequency spacing ranges from 0.02 Hz at low frequencies
to 0.5 Hz at the upper end of the frequency range considered, with around five
frequency points within the half-power bandwidth of each mode.
The TCR model has 0.44M degrees of freedom, taking around 35 seconds to
calculate a single frequency point, and around 13 hours for the complete calculation.
The KX model has 2.63M degrees of freedom, taking around 30 minutes to calculate a
single frequency point, and around 29 days for the complete calculation. The computer
system utilised a 3.1 GHz Intel i7 quad-core processor, with 16 GB RAM.
5 Results
The FE model solutions were evaluated in terms of vertical velocity at every mid-span
and near every column on each storey. This gives 218 discrete points for the TCR
model, and 848 for the KX model. Statistical analyses were performed on the resulting
vibration levels (in decibels). Considerable variation is observed with location on a
given floor level, with a standard deviation at mid-span locations of 2-7 dB for TCR
and 3-6 dB for KX. The floor plan of KX is more irregular than TCR, but TCR has a
small number of different floor span configurations. Therefore, there exist differences
between the buildings for variation in vibration level over given storeys.
For the TCR development, Fig. 3 shows the mean mid-span floor vibration relative
to the mean basement vibration level, for the ground, 1st, 3rd and 5th floors. The
plus/minus one standard deviation range is also presented as well as the average
measured value. Mid-span floor vibration values relative to those of the basement are
shown rather than column values, for conciseness and compatibility with other
literature (e.g. [5]).
Fig. 4 shows the mid-span vibration for the ground and 4th floors relative to the mean
vibration near the columns for those storeys.
The overall linear and A-weighted vibration levels at mid-span locations relative to
the mean basement vibration levels are shown in Fig. 5. The overall levels have been
Attenuation of Railway Noise and Vibration in Two Concrete Frame Multi-storey 301
Vibration relative to
basement (dB)
basement (dB)
10 10
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
Vibration relative to
Vibration relative to
basement (dB)
basement (dB)
10 10
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
Fig. 3. TCR mid-span vibration relative to basement: 1/3 octave band levels
10 10
to column (dB)
to column (dB)
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
Fig. 4. TCR mid-span vibration relative to column: 1/3 octave band levels
calculated using the average measured basement vibration spectrum (see Fig. 1), with
the appropriate corrections applied for each storey.
Utilising the same approach as for the TCR building, Fig. 6 shows floor vibration in
the KX building at various floor levels, relative to the mean basement level.
15 15
Linear A-weighted
Vibration relative to
Vibration relative to
10 10
basement (dBA)
basement (dB)
5 5
0 0
-5 -5
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-15 -15
-1 0 1 2 3 4 5 -1 0 1 2 3 4 5
Floor level Floor level
Fig. 5. TCR mid-span vibration relative to basement: Overall linear and A-weighted levels
302 D.E.J. Lurcock, D.J. Thompson, and O.G. Bewes
20
Ground floor 20
2nd floor
Vibration relative to
Vibration relative to
basement (dB)
basement (dB)
10 10
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
Vibration relative to
basement (dB)
basement (dB)
10 10
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
The mean mid-span amplification for the KX ground and 4th floors is shown in
Fig. 7. The overall linear and A-weighted vibration levels at mid-span locations relative
to the mean basement vibration are shown in Fig. 8.
10 10
to column (dB)
to column (dB)
0 0
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-20 -20
4 8 16 31.5 63 125 4 8 16 31.5 63 125
Third octave band centre frequency (Hz) Third octave band centre frequency (Hz)
15 15
Linear A-weighted
Vibration relative to
Vibration relative to
10 10
basement (dBA)
basement (dB)
5 5
0 0
-5 -5
-10 -10
Mean ± σ Mean Meas. Mean ± σ Mean Meas.
-15 -15
-1 0 1 2 3 4 5 6 7 -1 0 1 2 3 4 5 6 7
Floor level Floor level
Fig. 8. KX mid-span vibration relative to basement: Overall linear and A-weighted levels
Attenuation of Railway Noise and Vibration in Two Concrete Frame Multi-storey 303
6 Discussion
The results show a broad agreement between measured and predicted data in terms of
relative values and spectrum shapes, although the measured data do not always lie
within the plus/minus one standard deviation range for the predicted data. The TCR
building exhibited less agreement than the KX building, which is understood to be
mainly due to inaccuracies in the model geometry due to limited knowledge of the
detailed structural layout. It is also anticipated that errors in material parameters and
measurements further reduce the level of agreement.
The maximum amplification in both buildings occurred between 8-16 Hz, with
values up to around 15 dB relative to basement levels for the predicted data (20 dB for
measured data). Most frequency bands in this region undergo slight inter-storey
amplification (~1 dB per storey) but with attenuation for the peaks, resulting in a
broadening and shift in the spectral peak toward the lower frequencies at the upper
storeys. At higher frequencies (31.5 to 200 Hz), the KX building exhibited attenuation
of up to 1-2 dB per storey; the TCR building showed less than 1 dB per storey.
Mid-span amplification is greatest at lower storeys with mean values of up to around
10 dB for the KX building, and around 15 dB for the more regular TCR building. The
reduced amplification at upper storeys is not due to less vibration at mid-spans, but
rather to increased column vibration in the low frequency region.
Overall A-weighted values showed maximum attenuation of 2 dB for the TCR
building and 7 dB for the KX building, relative to basement vibration. There is slight
inter-storey amplification noted for the uppermost storeys.
Contrary to the guidance values in [2, 3], the predicted and measured data for both
the KX and TCR buildings do not support the assertions of inter-storey attenuation, for
third-octave or overall levels. Mid-span amplification is of a similar order to the
guidance in [2], and to a lesser extent the FTA guidelines [3].
15 15
Linear A-weighted
Vibration relative to
10
Vibration relative to
10 Num. 2 4 6*
basement (dBA)
basement (dB)
storeys 8 10 12
5 5
0 0
-5 -5
-10 Num. 2 4 6*
storeys -10
8 10 12
-15 -15
-1 0 1 2 3 4 5 6 7 8 9 10 11 -1 0 1 2 3 4 5 6 7 8 9 10 11
Floor level Floor level
Fig. 9. Generic building study, mid-span vibration relative to basement: Overall levels
8 Conclusion
FE models have been developed for two multi-storey bare concrete frame buildings
subject to vibration from nearby underground railways. Predictions of inter-storey and
mid-span amplification show some agreement with measured data, but neither the
measured nor predicted results support advice given in conventional empirical
guidance documents.
The FE model approach is considered suitable for an extensive parametric study into
the effect of building parameters on noise and vibration levels to inform improved
empirical prediction methods.
Preliminary studies on a generic building model show a dependence of inter-storey
level difference on the total number of building storeys, but with overall A-weighted
levels undergoing attenuation by around 1 dB per storey. Future work will investigate
other structural parameters, as well as the influence of damping introduced to the
building through non-structural components.
References
[1] Jones, K., Kuo, K., Hussein, M., Hunt, H.: Prediction uncertainties and inaccuracies
resulting from common assumptions in modelling vibration from underground railways.
Proc. Instn Mech. Engrs, Part F: Journal of Rail and Rapid Transit 226, 501–512 (2012)
[2] Remington, P.J., Kurzweil, L.G., Towers, D.A.: Low-Frequency Noise and Vibration from
Trains. In: Nelson, P.M. (ed.) Transportation Noise Reference Book. Butterworth & Co.,
London (1987)
[3] Federal Transit Agency. Transit Noise and Vibration Impact Assessment,
FTA-VA-90-1003-06, US Dept of Transportation (2006)
[4] Anderson, D.: Engineering prediction of railway vibration transmission in buildings. In:
Proceedings of Euronoise 1992, London (1992)
[5] Villot, M., et al.: Procedures to predict exposure in buildings and estimate annoyance.
RIVAS Deliverable D1.6 (2012), http://www.rivas-project.eu
Developing a Good Practice Guide on the Evaluation
of Human Response to Vibration from Railways
in Residential Environments
University of Salford,
Salford, Greater Manchester, M5 4WT, UK
j.s.woodcock@salford.ac.uk
Summary. The adverse effects that noise and vibration from railway systems in
residential environments can have on people are key obstacles for the
development of new rail systems and the operation of existing lines. Recent years
have seen an increase in public sensitivity towards noise and vibration from rail
systems and the success of legislation to control noise levels around railway lines
has resulted in an increase in the number of complaints about railway-induced
vibration. Costly mitigation measures coupled with unclear or non-existent
assessment methods mean that there is a need in industry and consultancy for
clear guidance on the assessment of groundborne vibration from rail systems
with respect to human response.
The current EU FP7 project CargoVibes is to publish a good practice guide on
the assessment of the human response to railway induced vibration in residential
environments. The aim of the guidance will be to provide end users with a set of
practical tools to assess the human impact of “steady state” railway vibration
primarily in terms of annoyance and sleep disturbance. Encompassing the current
state of knowledge regarding the human response to vibration in residential
environments alongside the practical outputs of the CargoVibes project,
this document is intended to promote policy and standard development in this
field.
The current paper will present the preliminary contents of the guidance, which
have been shaped by a workshop held at the University of Salford. This paper is
intended to promote debate and enable contributions from the IWRN community
to ensure that the guidance is relevant to the current needs of legislators, rail and
infrastructure operators, consultants, and local authorities.
Preface. This paper was presented to the IWRN11 in Uddevalla, Sweden in 2013
as a forum for discussion on the contents of a good practice guide for the
evaluation of railway vibration that was being developed as part of the FP7
project CargoVibes. Since that time, the guidance document has been completed
and can be downloaded at http://usir.salford.ac.uk/30855/.
Groundborne vibration that propagates into buildings may be felt by residents or heard
as groundborne noise or vibration induced rattle. For surface rail, these phenomena are
generally also accompanied by airborne noise. Exposure to these phenomena can result
in adverse effects such as annoyance and sleep disturbance. The human response to
vibration and noise is also strongly influenced by factors that are not directly related to
the exposure such as people’s attitude or emotional response towards the source. With a
projected increase in the market share of freight traffic on rail from 8 % in 2001 to 15 %
in 2020, a clear understanding of the human response to noise and vibration is needed if
this shift isn’t to be accompanied by a corresponding increase in adverse impacts on
residents living in close vicinity of freight railway lines.
Despite the recent attention given to the human response to groundborne vibration
(for example, the current CargoVibes project and [1]), compared to the human response
to environmental noise it has been a relatively neglected area of research. Guidance
available for the assessment of the human response to vibration varies from country to
country in the form of national and international standards. There are over 18 such
standards (for example [2, 3, 4]) and currently no consensus as to the most appropriate
single figure descriptor of vibration exposure or the appropriate criteria to prevent
adverse effects. There is therefore a need in industry and consultancy for clear guidance
on the assessment of vibration that is based on the current best available scientific
evidence.
The aim of the current EU FP7 project CargoVibes is the development and
validation of measures to ensure acceptable levels of vibration for residents living in
the vicinity of freight railway lines with a view to the expansion of freight traffic on
rail. One of the main objectives of this project is the development of criteria for the
evaluation of the adverse effects of vibration in residential environments which are
envisaged as extensions of current standards and guidelines. This objective will be
realized in the form of a good practice guide that will deliver guidance for the
evaluation of the human response to vibration from railways in residential
environments.
The good practice guide will contain the most up-to-date information and guidance
on the assessment of vibration in residential environments along with additional field
studies, meta-analytic exposure-response curves, and laboratory evaluations of sleep
disturbance performed under the CargoVibes project. This information will be
presented in a harmonized manner and compared to current limits and guidance.
Although the focus of the CargoVibes project is the evaluation of vibration from freight
railway operations, the good practice guide will address all sources of railway
vibration. This is due in part to the lack of experimental data on the human response to
freight vibration.
On 14th May 2013, a workshop was held at the University of Salford that gathered
international experts in the field of railway vibration from industry, consultancy, and
academia. The aim of this workshop was to discuss key aspects and challenges of the
evaluation of vibration in residential environments with respect to human response.
Developing a Good Practice Guide on the Evaluation of Human Response to Vibration 307
The outcomes of this workshop have been used to shape and inform the contents of the
guide.
The good practice guide will be organized into three main sections: annoyance, sleep
disturbance, and non-exposure factors. The remainder of this paper will give an
overview of the contents of each of these three sections.
2 Annoyance
Annoyance is a concept that is widely used for the evaluation of the negative impact an
environmental stressor has on a population. Despite lacking a theoretical underpinning,
annoyance is the measure most widely used in setting limits for environmental noise.
Annoyance is a broad concept which describes the negative evaluation of
environmental conditions and may encompass activity disturbance, emotional
responses, and attitudinal responses towards the source. The measurement of
annoyance due to environmental noise is standardized in ISO/TS 15666:2003 [5] which
sets out a specific wording of a question which can be used in social surveys to ask
about noise annoyance. This standard also recommends the use of two response scales
which can be used to record responses to the standard noise annoyance question.
On the other hand, the measurement of annoyance due to environmental vibration
has not been standardized at an international level, though there is guidance available in
the form of the Nordtest method [6]. Although the overall aim of the measurement of
annoyance due to environmental vibration is similar to the measurement of annoyance
due to environmental noise, there are some important differences to be taken into
account in the measurement approach mainly arising from the fact there are numerous
modalities through which vibration can be perceived. Environmental vibration may be
perceived kinesthetically via forces and movements within the body, viscerally through
receptors in the abdomen, tactilely through mechanoreceptors in the skin, visually
through the relative change of the position of objects, and aurally through re-radiated
noise and vibration-induced rattle. For example, the social survey described in [5]
asked 932 subjects through which surfaces they were able to perceive railway induced
vibration. 70.5 % of the respondents were able to feel vibration through the floor,
63.6 % through the bed, 51.3 % through a chair, 29.4 % through touching surfaces with
their hands, and 11.7 % through other sources. Coupled with the fact that people are
less adept at talking about experiencing vibration than they are about the perception of
sound, the wording of a question to measure self-reported vibration is not a trivial
matter. Therefore, it is expected that the good practice guide will provide suggested
wordings for questions to be used in socio-vibration surveys.
Limit values for environmental noise are derived from exposure-response
relationships calculated by combining data from many previous field studies (i.e.
meta-analytic) that describe the proportion of respondents expected to express
annoyance above a given threshold for a given level of noise exposure. There have been
a number of field studies in the past decade investigating the relationship between
annoyance and vibration exposure in residential environments. These studies have
produced exposure-response relationships for annoyance due to vibration (see, for
308 J.S. Woodcock et al.
example, [5, 6, 7, 8, 9]) but the relationships are difficult to compare due to differences
in the way vibration exposure has been assessed in the different studies. These
differences have hampered the development of internationally recognized limits and
guidance. Through the CargoVibes project, the data from these studies have been
collected and combined into a single meta-analytic exposure-response relationship that
will be reproduced in the good practice guide.
The two major considerations for the formulation of exposure-response
relationships are the thresholds of annoyance that are reported and the choice of single
figure descriptor. For consistency with existing international guidance for noise, the
thresholds of annoyance used in exposure-response relationships for environmental
noise [10] will be used for the meta-analytic curves reported in the good practice guide,
namely the proportion of respondents expressing annoyance in the upper 28 %
(“Percent Highly Annoyed”), upper 50 % (“Percent Annoyed”), and upper 72 %
(“Percent Slightly Annoyed”) of the annoyance response scale. The choice of single
figure descriptor of vibration exposure however is less straight forward.
Vibration exposure descriptors for the assessment of annoyance differ between
international and national standards and guidance. The recommended descriptors
generally fall into one of three categories: maximum running root-mean-square (RMS)
usually with a time constant of 1 s or 0.125 s, energy equivalent root-mean-square over
a defined time period, or a cumulative dose. Differences are also found between
national guidelines as to whether these descriptors are assessed from measurements of
acceleration or velocity. Numerous different frequency weightings can be applied to
the assessment of vibration exposure that are intended to reflect the frequency
dependency of the human perception of whole body vibration. Studies that have looked
into the correlation between different descriptors of vibration exposure and annoyance
have generally found that the different descriptors all have a similar correlation with
self-reported annoyance [5, 7]. Based on current evidence, it therefore may not be
possible at this time to advocate one descriptor over another. The exposure-response
relationships in the good practice guide will be presented in terms of a number of
different vibration exposure descriptors: a maximum running r.m.s. velocity, a
weighted 24-hour r.m.s. acceleration, and a weighted 24-hour Vibration Dose Value.
A number of EU member states have vibration exposure limits in place, often
without indication as to their basis. The good practice guide will not suggest limits per
se, however limits in place in different member states will be compared to the
meta-analytic exposure-response relationships and the expected prevalence of
annoyance for the different limits will be reported.
Perceptible vibration is rarely unaccompanied by noise; this noise could be
re-radiated groundborne noise, airborne noise, or vibration induced rattle. Laboratory
and field studies suggest that there is a synergistic relationship between noise and
vibration and annoyance [5, 6, 11, 12, 13]. As this field of research is still in its relative
infancy, it is difficult to suggest appropriate criteria for the prevention of annoyance
based on current evidence. What evidence there is available will be presented in the
guide.
The modeling of groundborne vibration is difficult mainly due to variations in local
ground conditions. Therefore, vibration in residential environments is often assessed
Developing a Good Practice Guide on the Evaluation of Human Response to Vibration 309
3 Sleep Disturbance
Sleep is an important biological function that is necessary for the wellbeing of people [14].
Sufficient evidence exists of a relationship between exposure to environmental noise and
sleep disturbance that limit values have been set to protect against noise induced sleep
disturbance [14]. There is however less experimental data regarding the effect of whole
body vibration on sleep. The question therefore arises as to whether the limited available
data on the subject is sufficient to set criteria for night time vibration limits.
There are a number of different ways, both objective and subjective, by which sleep
disturbance can be measured and quantified. Self-reported sleep disturbance is
measured like annoyance via a questionnaire. Unlike annoyance however the question
posed and the scale to measure self-reported sleep disturbance are not standardized.
Compared with annoyance due to railway induced vibration, there is relatively little
experimental data relating railway induced vibration to sleep disturbance. A study in
the United Kingdom has produced an exposure-response curve indicating the
proportion of respondents reporting sleep disturbance for a given vibration exposure
(N = 752) [5].
Within the CargoVibes project, sleep disturbance due to vibration and noise induced
by freight trains has been studied in a laboratory setting [15, 16]. This study has shown
that noise and vibration from freight trains has a negative impact on both self-reported
and objective measures (e.g. heart rate) of sleep quality and that these negative effects
are related to vibration amplitude. The study also investigated the effect of the number
of freight trains on sleep quality with inconclusive results; an increase in the number of
trains per night was found to have a detrimental effect on sleep quality at moderate
vibration amplitudes but was found to have no effect at higher vibration amplitudes.
The results generated by this study represent the best data presently available on the
subject of vibration induced sleep disturbance. However, further studies are needed
before robust exposure-response relationships, thresholds, and criteria can be
recommended.
4 Non-exposure Factors
Factors that are not directly attributable to vibration exposure can have a strong
influence on the human response to vibration [17]. An understanding of these factors
can help interpret the observed annoyance responses and can also have important
operational implications.
There is evidence that the annoyance response to vibration is different at different
times of day with annoyance at night being greater than that in the evening, and
annoyance in the evening being greater than that at the daytime for the same magnitude
of vibration exposure [18]. It should be borne in mind that night time annoyance is not
310 J.S. Woodcock et al.
necessarily the same as sleep disturbance. This suggests that a nighttime vibration
indicator or a day-evening-night weighted 24-hour descriptor may be appropriate. That
the annoyance response is greater during the nighttime period than the daytime has
important implications for freight traffic. In operational terms, this implies that
annoyance can be mitigated by controlling the times of day at which trains run.
Concern that vibration is causing damage to property has been found to be a
mediating factor in the annoyance response towards railway vibration [5]. That the
levels of vibration generated by railway traffic in residential environments is unlikely
to cause even cosmetic damage [19] suggests that better community engagement could
help manage the annoyance response.
Based on current evidence, annoyance due to vibration is source dependent. For
example, annoyance due to construction activities has been found to be greater for the
same magnitude of vibration exposure as railway [5]. Even within railway, differences
in annoyance response to both noise and vibration have been found between passenger
trains, freight trains, and maintenance activities [20]. This suggests that these
differences should be considered if assessing annoyance due to mixed rail operations.
In general, the evidence available regarding the human response to vibration in
residential environments has been collected under steady-state conditions using
cross-sectional surveys. It must be borne in mind when implementing the results
presented in the guide that the response to a change in vibration exposure may be
underestimated using current relationships, although this has not been found to be the
case for reductions in traffic noise exposure [21]. This and other limitations will be
clearly discussed in the guide.
5 Conclusions
This paper has provided an overview of the proposed contents of a good practice guide
on the evaluation of human response to vibration from railways in residential
environments. The guide will address three broad areas relating to the human response
to vibration: annoyance, sleep disturbance, and non-exposure factors. The guide will
present the current state of knowledge on each of these topics and discuss how this
relates to current standards, guidelines, and limit values. This will result in a set of
practical tools that can be used across different EU member states for the assessment of
vibration from railways with respect to human response. The good practice guide is due
for completion in September 2013.
Acknowledgements. This paper was funded by the EU FP7 through the CargoVibes
project. Within the CargoVibes project, the meta-analysis work was conducted by TNO
and the sleep disturbance work was conducted by the University of Gothenburg.
References
[1] RIVAS – Rail Induced Vibration Abatement Solutions,
http://www.rivas-project.eu (accessed February 19, 2014)
[2] International Organisation for Standardization. ISO 2631-1:1997 Mechanical vibration
and shock - Evaluation of human exposure to whole-body vibration - Part 1: General
requirements
Developing a Good Practice Guide on the Evaluation of Human Response to Vibration 311
Summary. Sound Transit is constructing a light rail transit system with twin
bored tunnels through the University of Washington (UW) campus in Seattle,
Washington, USA. Ground vibration impacts on the UW campus were
extensively studied with long range vibration propagation modeling and testing,
vehicle force density measurements, and dynamic modeling of the vehicle and
track isolation system. The vibration level design criteria were specified in third
octave band velocities ranging from 500 nm/s to 2,500 nm/s at frequencies
ranging from 3.16 to 100 Hz, placing substantial demands on vibration isolation
design. Vehicle vibration force density levels were measured at ballasted track
and resilient direct fixation fastener track. Vibration reductions of the order of 10
to 15 dB at 40 to 80 Hz were observed after profile grinding of the rails at direct
fixation track. Vibration isolation provisions include discontinuous floating slabs
with design resonance frequency of 5 Hz and high compliance direct fixation
fasteners. Limits for rail undulation were specified to control low frequency
vibration. Rail grinding and wheel truing combined with vibration monitoring
and wheel flat detection complete the vibration control design. A major design
complication is the provision of a direct current traction power cable beneath the
slabs to control magnetic fields induced on campus by the trains.
1 Introduction
Sound Transit is constructing extensions to its light rail system in the metropolitan area
of Seattle, Washington State, USA. Sound Transit currently operates the South Link,
extending from Seattle to the international airport (SeaTac). Segments under design
and construction include the University Link, Northgate Link, Lynwood Link, East
Link, and South Link Extension. The University Link is currently under construction,
involving several kilometers of twin bored tunnels extending from downtown Seattle to
the University of Washington (UW), including two stations. The North Link would
include several kilometers of twin bored tunnel and a short section of at-grade slab
track and aerial structure at the northern end. The North Link includes three stations:
the University District, Roosevelt, and Northgate stations. Sections of the University
Link and North Link alignments pass through the UW campus in approximately one
kilometer of twin bored tunnels. This paper concerns the prediction and control of
ground vibration on the UW campus and related measurements.
2 Vibration Criteria
The Sound Transit entered into a memorandum of agreement that prescribed individual
vibration limits for twenty-four buildings containing vibration sensitive laboratories,
based on ambient vibration data collected by the UW. The vibration limits varied
considerably from building to building. The maximum and the minimum of the various
one-third octave band vibration limits are indicated in Fig. 1. The preservation of the
existing ambient vibration environment for the foreseeable future is a principal
requirement of the University of Washington.
50
1/3 OCTAVE VELOCITY LEVEL - DB
40
RE 5E-6CM/S
30
20
10
MAXIMUM
MINIMUM
0
3.15 6.3 12.5 25 50 100
FREQUENCY - HZ
3 Geology
The geological materials at the University of Washington campus and Seattle area
include over-consolidated glacial tills, sands, and gravels of thicknesses ranging from
30 m to 150 m overlying marine sediments. Seismic shear and compression wave
velocity profiles are shown in Fig. 2, obtained by in-situ down-hole velocity surveys.
10
COMPRESSION
DEPTH - M
20
SHEAR
30
40
The glacial tills, sands, and gravels lie over marine sediments with slightly lower shear
wave velocities that increase with increasing depth. The geological materials are
generally very stiff, and their response to ground vibration is generally lower than that
of alluvial soils. However, the quality factor (Q) of the material is very high.
4 Vibration Prediction
70
60
FDL - DB RE 1N/M1/2
50
40
30
OA 8 16 31.5 63 125
FREQUENCY - HZ
The tests at direct fixation track with resilient fasteners were conducted in the Beacon
Hill tunnel at both curved and tangent sections of track. The data shown in Fig. 3 indicate
that the track resonance is shifted downward from about 80 Hz for ballasted track to
50 Hz for resilient direct fixation track. The wheels are resilient wheels, and the
resonance frequencies of the system consisting of axle and wheel centers, resilient wheel
elastomeric springs, wheel tires, and resiliently supported rail are also in this range. Initial
tests at direct fixation track in the Beacon Hill tunnel were conducted in 2009, and again
in 2012 to improve estimates of the force density level at low frequency third octave
bands extending down to 3.16 Hz, using shakers and cross-spectral techniques. During
theses tests, the third octave FDLs at frequencies above 40 Hz were observed to be
substantially lower than those measured in 2009 on DF track. The reduction was
attributed to profile grinding conducted in December of 2009 after the initial tests.
The peak at 20 Hz is attributed to the center bogie of the low floor vehicle. The bogie
is unpowered, and has independently rotating wheels, in contrast to the powered bogies
with solid axles. The 20 Hz peak is produced by a fraction of the fleet of otherwise
identical vehicles.
The peak at 8 and 10 Hz for 56 km/h operation is related to wheel set rotation and
perhaps undulation of the rail. The frequency of this peak increases proportionately
with train speed as may be expected for geometric roughness, and the amplitude
increases as the wheel-set rotation frequency moves up on the resonance curve of the
primary suspension. The primary suspension resonance frequency appears to be about
10 Hz. Similar results were obtained at several other speeds.
In general, these data indicate that rail profile grinding in combination with wheel
truing yield low dynamic forces at roughness wavelengths of the order of 0.25 m to
0.35 m, important for ground-borne noise control. Little reduction of vibration below
20 Hz by profile grinding was apparent.
layers, using the shear and compression wave velocity profiles shown in Fig. 2.
Responses at distances as great as 600 m were computed for various source depths,
assuming that the soil vertical heterogeneity was constant over the propagation distance.
Results were converted to third octave band point source responses (PSR) and
energy-summed over the design maximum train length to obtain theoretical LSR's. The
LSR's were then adjusted to match the measured LSR’s at close range and the results of
long range testing with the vibratory roller. The predictions beyond 200 m were adjusted
upward by five decibels, based on the long range test results with vibratory roller.
5 Vibration Control
Vibration control provisions include low-undulation rail, high-compliance resilient
direct fixation fasteners, low resonance frequency floating slab track, wheel truing and
monitoring, rail grinding, and wayside vibration monitoring.
10
RELATIVE ROUGHNESS -
-10
-20
DB
-30
-40
-50
1 2 4 8 16 31.6 63 125
FREQUENCY - HZ
k = 36 MN/m/m
k = 13 MN/m/m
k = 3.4MN/m/m (FLOATING SLAB)
Fig. 4. Effect of rail support modulus on invert roughness for speed 56 km/h
318 J.T. Nelson and D.L. Watry
Floating slab track isolation is planned for the North Link tunnels on the UW campus.
The floating slab system consists of discontinuous pre-cast slabs positioned at 1.219 m
pitch, each separated by 50.8 mm. This pitch was selected so that the leading axle of the
bogie would pass over the center of one slab while the trailing axle would pass over a
gap between slabs. Thus, dynamic forces at the slab-passage frequency would be out of
phase and radiate vibration inefficiently. The rail fastener pitch would be 610 mm,
giving a pin-pin mode of about 1,000 to 1,200 Hz for the 115 RE rail section.
Each pre-cast slab will consist of heavy-weight concrete of density 3,400 kg/m3,
and, combined with track, will weigh about 5,500 kg. Each slab will be supported on
four natural rubber main support pads and laterally stabilized with two natural rubber
side pads. The diameter of the circular main support pads will be 318 mm and the
unloaded thickness will be 191 mm. The side pads will maintain the lateral position of
the track. The dimensions of the natural rubber side pads are planned to be 203 mm
high by 381 mm long by 114 mm thick (unloaded). The side pads will be
pre-compressed between the slab and tunnel side wall by about ten percent. Two
natural rubber pads will separate the slabs from each other.
The theoretical vibration response of the proposed floating slab, based on finite
element models of the floating slab track and vehicle and a standard direct-fixation
track and vehicle, is illustrated in Fig. 5.
10
0
INSERTION GAIN - DB
-10
-20
-30
-40
-50
-60
OA 4 8 16 31.5 63 125
FREQUENCY - HZ
5HZ FLOATING SLAB TRACK
HIGH COMPLIANCE FASTENERS
frequency, which reduction would likely be due to the reduction of the effect of invert
roughness by the floating slab isolators.
A prototype floating slab is planned for the University Link tunnels on a curve and
grade similar to that planned for the University of Washington campus tunnels. The
purpose of the prototype is to verify performance and identify construction issues prior
to finalizing design for the North Link alignment on the University of Washington
campus.
5.3 Rail
Tunnel wall vibration will be monitored continuously to detect vibration level increases
due to vehicle and track wear. Data will be transmitted to central train control where it
will be automatically processed into 1/3 octave bands, compared with the UW vibration
limits, and stored for archival reference. A major goal is to allow researchers to
compare experimental anomalies with archived ground vibration data to determine if
the anomaly might have been due to trains.
6 Conclusion
The vibration control effort at the University of Washington campus in Seattle,
Washington, presents unique challenges for vibration prediction and control. Other
transit systems in the United States are experiencing similar demands from other
vibration sensitive institutions and commercial entities. Examples include the
University of Minnesota, concert halls, recording studios, and educational facilities.
The prediction models, performance data, vibration isolation provisions, and
monitoring programs are tools that can be applied to unique problems of this nature.
References
[1] Transit Noise and Vibration Impact Assessment, U.S. Department of Transportation,
Federal Transit Administration (2006)
[2] Nelson, J.T.: Prediction of ground vibration from trains using seismic reflectivity methods
for a porous soil. Journal of Sound and Vibration 231(3) (2000)
[3] Nelson, J.T., Saurenman, H.J.: State-of-the-art review: prediction and control of
groundborne noise and vibration from rail transit trains, Wilson, Ihrig & Associates, Final
Report US DOT/TSC pg5-18, Fig. 5.6 (1983)
[4] Hunt, H.E.M.: Types of rail roughness and the selection of vibration isolation measures. In:
Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson, J.,
Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 341–347. Springer, Heidelberg (2008)
Recent Developments in the Pipe-in-Pipe Model
for Underground-Railway Vibration Predictions
1 Introduction
The Pipe-in-Pipe model (PiP) [1] is well established as a tool for predicting vibration
from underground railways. The model consists of an inner pipe representing the tunnel,
and an outer pipe with infinite external radius, representing the soil. A train of infinite
length is represented by an infinite number of axle masses with a constant spacing
moving on a track with random, white-noise roughness excitation. Floating slab track,
soil layers, rigid bedrock and a free surface can all be incorporated into a PiP prediction.
The advantage of PiP over other more detailed models, such as those using coupled
boundary element and finite element methods, is the reduced processing time. PiP was
first launched in 1999, and a number of updated versions have been released since.
All railway-vibration prediction models contain simplifying assumptions, ranging
from the global scale (e.g. ignoring the presence of piled foundations) to the local scale
(e.g. assuming soil homogeneity). These assumptions introduce inaccuracy and
uncertainty into railway-vibration predictions. To improve the understanding of the
limitations of the PiP model, the authors have investigated a number of commonly
-160 -200
25Hz 160Hz
-165
rms Velocity (dB, ref 1 m/s)
-175 -220
-180
-230
-185
-240
-190
-195 -250
-50 -25 0 25 50 -50 -25 0 25 50
Surface Location (m) Surface Location (m)
Fig. 1. Vertical surface velocity response due to vertical rail load applied within a tunnel buried 15 m
below surface. Bold line: homogeneous response; fine lines: 90 % confidence interval. Soil properties:
mean elastic modulus 0.55 GPa; elastic modulus variability +/- 0.17 GPa; density 2000 kg/m3;
Poisson’s ratio 0.44; dilation and shear damping coefficient 0.05 [3].
When included, soil layers are incorporated into PiP and other railway-vibration
models as horizontal slices, whereas real soil stratification tends to vary in inclination,
commonly up to five degrees. A 2D thin-layer model [4] was used to model a layer
inclined at 1°, 3° and 5°. A loading spectrum spanning 20–200 Hz was applied to a
buried tunnel 15 m below the surface, simulating railway induced vibrations. The
variation in RMS velocity for the 20-200 Hz frequency band ranges from ±1 dB to
±7 dB with increasing inclination, an example of which is shown in Fig. 2 where insertion
gain (IG) is measured relative to a horizontal layer. Surface motion due to lower
10
rms Velocity IG (dB, ref 1 m/s)
-5
-10
-50 0 50
Surface Distance (m)
Fig. 2. Surface velocity insertion gains for a 5m layer inclined atop a homogeneous halfspace
(Elayer=2Ehalfspace) solid: 1°; dashed: 3°; dash-dotted: 5°. Soil properties: elastic modulus 550 MPa;
density 2000 kg/m3; Poisson’s ratio 0.44; dilation and shear damping coefficient 0.05 [4].
324 K.A. Kuo et al.
15
5
10
y-direction (m)
0 0
-5
-10
-15 -5
-15 -10 -5 0 5 10 15
x-direction (m)
Fig. 3. Insertion gain for a 4 m x 90° void at the 50 Hz third-octave passband. Model and soil properties
listed in [5].
For ease of installation, floating slab track (FST) can be manufactured in discrete
pre-cast sections rather than cast in-situ as a continuous slab. Periodic-infinite structure
theory was used to incorporate discontinuous slab track into the PiP model [6]. The
vibration response differs by ±10 dB from the continuous FST at 150 Hz, and to a lesser
extent at other frequencies between 1 and 250 Hz. Floating-slab tracks with
discontinuous slabs generate more vibration at the free-free slab natural frequencies
and less vibration at frequencies above those natural frequencies. For a 6m slab, the
first free-free slab resonance occurs at 63 Hz, considerably higher than a typical FST
resonance frequency of 10-30 Hz. A decoupled model, where forces calculated from a
track on a rigid foundation are used as input to a separate tunnel-soil model, can be used
to calculate vibration in the far-field. However, for improved accuracy and for cases
when the stiffness of track is large, a coupled track-tunnel-soil system is needed for the
calculation [6].
Recent Developments in the PiP Model for Underground-Railway Vibration Predictions 325
In areas of weak soil, piled foundations can be found in close proximity to railway
tunnels. The presence of piles is often ignored in modelling simulations. A
semi-analytical model [7] for pile vibration was included in the PiP model using a weak
coupling method. For a four-pile group, vibration attenuation of the order of 10 dB is
seen at 50-80 Hz. At some localised frequencies coinciding with strong antiresonances
in the railway vibration field, the presence of the piles can result in vibration level
increases of up to 8 dB.
Fig. 4. Vertical-displacement field insertion gain (dBref [1 m]) produced at 60 Hz by a unit, vertical
force applied to the tunnel invert. Model and soil properties listed in [8]. Darkness indicates
attenuation, and lightness indicates amplification.
vibration countermeasures changes not only the track characteristics, but also the
roughness perceived by the wheels by -20 dB to +15 dB. As an example, a rail that is
manufactured perfectly straight will be perceived as “rough” if it is fixed to a track bed
that is uneven. Introducing resilience will release the rail from the track bed, thereby
reducing perceived roughness. Conversely, a “bent rail” (with residual curvature after
manufacture) will be best fixed to a high-quality slab in order to reduce roughness.
Resilience inserted between the rail and slab will cause perceived roughness to
increase. As a rule of thumb, if resilient elements are added above the source of
roughness, there will most likely be a reduction in perceived roughness, especially at
short wavelengths. However, when the resilient elements are added below the source of
roughness, the perceived roughness will most likely increase for all wavelengths. Thus
the effectiveness of in-track countermeasures can only be properly evaluated when the
types of roughness present are known.
Due to the complexity of the underground environment, models seldom include both a
railway and a building within the same computation (‘full coupling’), but instead the
vibration field produced by the railway is used as an input in a building model, with
varying degrees of interaction between the two systems. The degree of interaction may
range from ‘no coupling’, which involves using the railway vibration field as a direct
input to the building model, to ‘weak coupling’ where the railway and building models
are joined at a number of discrete points.
The PiP model has been used to produce a railway-induced vibration field, to
investigate the difference between an uncoupled railway-pile building model and a
weakly coupled system that satisfies force equilibrium and displacement compatibility at
a series of discrete ‘shared’ nodes [10]. The building and the piles are modelled as a 2D
frame using the dynamic stiffness matrix method. In the weakly coupled system, the
joining of subsystems method is used to couple the ground and the foundation together,
using the thin-layer method for calculation of the relevant transfer functions. The piles are
10m long, with the nearest pile-head located 15 m vertically and 20m horizontally from
the tunnel centre. The weakly coupled system was found to exhibit vibration levels up to
20 dB higher than the uncoupled system in the frequency range of 10-80 Hz.
The work involving soil variations (i.e. inhomogeneous soils, layer inclination, and
voids) has shown that this is an important modelling feature that would be beneficial to
future releases of PiP. Incorporating the thin-layer method employed in the models
briefly alluded to above would increase PiP runtimes beyond the current two-minute
limit. Thus, more computationally efficient methods for simulating soil variation are
under investigation. A study using Legendre polynomials to solve the weak-form
governing equations of thick-layer media with spatially varying properties is hoped to
resolve this issue. The semi-analytic approach is predicted to produce significantly
lower runtimes as an order of magnitude fewer polynomials should be required
vis-à-vis thin-layer elements. The weak-form solution can be used to generate Green’s
functions for soils with varied parameters that can be incorporated directly into the
current PiP formulation or other boundary-element models. This will provide a
parameter evaluation that can be used to assess prediction uncertainty due to soil
variation.
The work of piled foundations and second tunnels is being continued as part of the
‘Modelling of Train Induced Vibration (MOTIV)’ project, involving the University of
Cambridge and ISVR, University of Southampton. This project began in January 2013
and has three objectives: firstly, the effect of pre-load and non-linear behaviour of
resilient elements of tracks will be investigated using excitation models (i.e. tracks on
elastic foundations); secondly, to develop a detailed Periodic Boundary Element model
and couple that to the relevant structural elements to account for the interaction
between twin-tunnels, soil and pile foundations; and thirdly, to further develop PiP to
consider calculating vibration transmitted to buildings on pile-foundations from surface
tracks and from tracks in tunnels embedded in multi-layered ground.
References
[1] Hussein, M.F.M., Hunt, H.E.M.: A numerical model for calculating vibration from a
railway tunnel embedded in a full-space. Journal of Sound and Vibration 305(3), 401–431
(2007)
[2] Phoon, K., Kulhawy, F.: Characterization of geotechnical variability. Canadian Geotechnical
Journal 36, 612–624 (1999)
328 K.A. Kuo et al.
[3] Jones, S., Hunt, H.E.M.: Predicting surface vibration from underground railways through
inhomogeneous soil using the thin-layer method. Journal of Sound and Vibration 331(9),
2055–2069 (2012)
[4] Jones, S., Hunt, H.E.M.: The effect of inclined soil layers on surface vibration from
underground railways using the thin-layer method. Journal of Engineering
Mechanics 137(12), 887–900 (2011)
[5] Jones, S., Hunt, H.E.M.: Voids at the tunnel-soil interface for calculation of ground
vibration from underground railways. Journal of Sound and Vibration 331(12), 245–270
(2011)
[6] Hussein, M.F.M., Hunt, H.E.M.: A numerical model for calculating vibration due to a
harmonic moving load on a floating-slab with discontinuous slabs in an underground
railway tunnel. Journal of Sound and Vibration 321(1-2), 363–374 (2009)
[7] Kuo, K.A., Hunt, H.E.M.: An efficient model for the dynamic behaviour of a single pile in
viscoelastic soil. Journal of Sound and Vibration 332(10), 2549–2561 (2013)
[8] Kuo, K.A., et al.: The effect of a twin tunnel on the propagation of ground-borne vibration
from an underground railway. Journal of Sound and Vibration 330(25), 6203–6222 (2011)
[9] Verachtert, R.: The effect of track roughness and train speed on railway-induced ground
vibration. M.Sc. thesis, Katholieke Universiteit Leuven (2013)
[10] Hussein, M.F.M., et al.: The dynamic effect of a piled-foundation building on an incident
vibration field from an underground railway tunnel. In: Proceedings of the 20th
International Conference on Sound and Vibration 2013, Bangkok (2013)
[11] Gatzwiller, K.B.: Differences between superstructure and substructure resilience:
indicative ground-borne vibration predictions based on the Pipe-in-Pipe model from
Cambridge University. In: Proceedings of Railway Engineering – 2013, London (2013)
Prediction of Railway-Induced Ground
Vibrations: The Use of Minimal Coordinate
Method for Vehicle Modelling
Summary. This paper examines the combination of finite element method and
multibody modelling to simulate the generation and propagation of ground
vibration in the vicinity of railway networks. Based on the assumption that the
source of vibrations lies at the wheel/rail contact, a multibody model of the
vehicle is built using minimal coordinates, which leads to a system of pure
ordinary differential equations, without constraint equations. Track and
foundation dynamic equations are coupled to the vehicle's equation of motion,
using non-linear Hertzian theory. From these results, the ballast reaction on the
subgrade is used in a second subproblem where free field ground response is
computed using the finite element software ABAQUS.
1 Introduction
In [8] the authors have validated a prediction model based on a two-stage approach,
taking into account the vehicle dynamics influence on the track deflection and therefore
on the ground wave propagation (Fig. 1). Since the problem is complex, and in order to
take into account all the phenomena, a split model was preferred. Track/soil interaction
is taken into account using a simplified condensed form of soil like a simple Winkler
foundation or a coupled lumped mass model [9]. To the author's knowledge, this
approach is one of the most promising ways of focusing on some specific mechanisms
of vibration generation and modelling capabilities (wheel/rail interaction, vehicle
influence, three-dimensional soil model). The limitation lies in the tedious definition of
the equations of motion governing the vehicle. They are indeed manually copied into
the model.
typical application is presented, based on the case of a tramway where the results
obtained with the improved approach are compared with experimental results. The
paper is ended with some conclusions.
2 Modelling Strategy
The minimal (or generalized) coordinate approach, developed by Hiller [10], is very
efficient for open loop systems such as train models. Compared to classical approaches
– Cartesian or relatives approaches – used in software packages (Adams/Rail,
Simpack), this method has the advantage to automatically eliminate all the forces and to
arrive at a system of pure differential equations using the virtual power principle
( being the number of degrees of freedom)
(2.1)
with, for each of the bodies, and the corresponding mass and central
inertia tensor, and the resultant force and moment, the acceleration of the
centre of gravity, the partial contributions of in the velocity
(2.2)
(2.3)
The position and orientation of each body (car body, bogie, wheelset, or other
inertial component, with a role in the vehicle dynamics) is expressed by means of a
homogeneous transformation matrix, which gives the situation of the associated local
frame with respect to another frame as a function of the configuration parameters .
This block 4 x 4 matrix has the general form
(2.4)
where is the coordinate vector of frame with respect to frame , and is the
rotation tensor describing the orientation of frame with respect to frame . For any
complex mechanical system, the motion of each body can be decomposed into a
succession of elementary motions defined as successive multiplications of simpler
332 G. Kouroussis et al.
The proposed method was implemented in a C++ object-oriented program, using the
EasyDyn library [11]. An application based on the MuPad/Xcas platform generates
symbolic kinematic expressions. It creates a C++ code directly compilable against the
EasyDyn library which is completed by the applied forces (suspensions, wheel/rail
contact) and the link with the track model (already established and only depending on
the site parameters). In particular, the wheel/rail contact is defined using the non-linear
Hertz theory.
Prediction of Railway-Induced Ground Vibrations 333
The simulation of the vehicle/track subsystem provides the time history of the ground
forces, defined as the visco-elastic reaction of the ballast on the subgrade. These forces
are used in a three-dimensional soil model to compute the ground wave propagation
and consequently the ground vibration at several distances from the track. In the present
study, the finite element software ABAQUS is used, using specific boundary conditions
(viscous boundary and infinite elements) to mimic the infinite extension of the soil
[12-14]. A time domain integration scheme is preferred, able to accurately represent the
ground wave propagation with respect to the chosen meshing guidelines.
The T2000 light railway vehicle circulating in Brussels is a three-car vehicle that uses
advanced technology such as independently rotating wheels and low floor design.
Motors are directly mounted inside the wheels and, to reduce the effect of the unsprung
masses, resilient wheels are used for the motor wheels only. The vehicle model
considers the bounce and the pitch motions of each component (wheelset, bogie,
carbody). The tram case is a perfect example of dynamic vehicle/track interaction
where the quasi-static deflection has a small effect on the ground vibrations due to the
low speed of the vehicle.
Fig. 3 presents the kinematic layout of each part (leading and central carriages) in the
T2000 tram model. The number of configuration parameters is fixed to 19,
according to the selected pitch and bounce motions of each body. From the general
tram dimensions (Fig. 3(a)) and the vehicle speed (constant or time-dependant), the
homogeneous transformation matrices of each body is expressed with the help
of elementary motions as a displacement or a rotation about
the -axis (Fig. 3(b)). Translational and rotational velocities and accelerations are
derived from the symbolic tool and the external forces (primary and secondary
suspensions, gravity and wheel/rail contact) are defined, in order to obtain numerically
the equations of motion, as described by Eq. (2.1). Vehicle, track and soil dynamic
parameters are presented in a previous work [15] corresponding to the configuration of
a ballasted site in Brussels.
Figs. 4 and 5 present the soil surface velocity results obtained with the proposed
method when the tram travels at constant speed = 30 km/h over a single rail surface
defect. The latter is defined as a plate placed on the rail head with a shape sufficiently
similar to that of a rail joint, rail crossing or an event such as gear switching, all
commonly encountered in tramway networks. This situation provides an interesting
context where the vehicle dynamics strongly affect the ground responses [15]. The two
distances (2 and 8 m) from the track, where experimental data are also available, are
retained to monitor ground vibrations. Related results were presented in [15], obtained
from linear equations of motions which were established manually and coupled to
EasyDyn solver. In both data sets, each wheel/defect impact is well emphasized and
334 G. Kouroussis et al.
dominates the velocity vibration level. On each curve, the contact phases are indicated
in Fig. 4 and 5 and show a slight time-shift between the results at 2 and at 8 m, which is
again similar to [15]. This confirms the previous validation provided by the authors and
demonstrates that the modified methodology has an acceptable level of accuracy.
Fig. 4. Vertical ground velocity at 2 m from the track during the passing of the T2000 tram at
speed = 30 km/h on a singular rail surface defect (left: measured, right: predicted)
Fig. 5. Vertical ground velocity at 8 m from the track during the passing of the T2000 tram at
speed = 30 km/h on a singular rail surface defect (left: measured, right: predicted)
4 Conclusion
In this paper the use of minimal coordinates was proposed as a methodology tool in
vehicle/track dynamics with ground vibration evaluation. This method is not
commonly used in railway dynamics software packages: relative or Cartesian
coordinate methods are more frequently used and impose a set of differential and
algebraic equations. The approach with minimal coordinates leads to a minimal number
of equations, but setting up the latter approach is a long procedure as it requires a
dedicated resolution of kinematics. This is easily solved with a symbolic tool. The
numerical model was developed in search of a user-friendly application where the user
has to include a minimum amount of information about the vehicle configuration. The
theory and formulation of the vehicle models are simple and can be coupled, without
difficulty, to standard dynamics equations governing the track and the subgrade, the
latter being defined by a coupled lumped mass model. The strategy proposed here
demonstrates all the potential of the EasyDyn framework to address practical issues that
can be encountered in applications of track analysis or ground vibrations evaluation.
References
[1] Sheng, X., Jones, C.J.C., Thompson, D.J.: Prediction of ground vibration from trains using
the wavenumber, finite and boundary element methods. Journal of Sound and
Vibration 293(3-5), 575–586 (2006)
336 G. Kouroussis et al.
[2] Chebli, H., Othman, R., Clouteau, D., Arnst, M., Degrande, G.: 3D periodic BE-FE model
for various transportation structures interacting with soil. Computers and
Geotechnics 35(1), 22–32 (2008)
[3] Costa, P.A., Calçada, R., Silva Cardoso, A.: Vibrations induced by railway traffic:
influence of the mechanical properties of the train on the dynamic excitation mechanism.
In: Proceedings of 8th European Conference on Structural Dynamics: Eurodyn 2011,
Leuven, Belgium, pp. 804–811 (2011)
[4] Auersch, L.: Train induced ground vibrations: different amplitude–speed relations for two
layered soils. Proc. IMechE, Part F: J. Rail and Rapid Transit. 226(5), 469–488 (2012)
[5] Ekevid, T., Lane, H., Wiberg, N.-E.: Adaptive solid wave propagation - influences of
boundary conditions in high-speed train applications. Computer Methods in Applied
Mechanics and Engineering 195(4-6), 236–250 (2006)
[6] Kaynia, A.M., Madshus, C., Zackrisson, P.: Ground vibration from high-speed trains:
prediction and countermeasure. Journal of Geotechnical and Geoenvironmental
Engineering 126(6), 531–537 (2000)
[7] Takemiya, H., Bian, X.: Substructure simulation of inhomogeneous track and layered
ground dynamic interaction under train passage. Journal of Engineering Mechanics 131(7),
699–711 (2005)
[8] Kouroussis, G., Verlinden, O., Conti, C.: A two-step time simulation of ground vibrations
induced by the railway traffic. Journal of Mechanical Engineering Science 226(2),
454–472 (2012)
[9] Kouroussis, G., Gazetas, G., Anastasopoulos, I., Conti, C., Verlinden, O.: Discrete
modelling of vertical track-soil coupling for vehicle-track dynamics. Soil Dynamics and
Earthquake Engineering 31(12), 1711–1723 (2011)
[10] Hiller, M.: Dynamics of multibody system with minimal coordinates. In: Pereira, M.F.O.,
Ambrosio, J.C. (eds.) Computer-Aided Analysis of Rigid and Flexible Mechanical
Systems - Proceedings of the NATO Advanced Study Institute 1993, Tròia, Portugal,
pp. 119–163 (1993)
[11] Verlinden, O., Ben Fekih, L., Kouroussis, G.: Symbolic generation of the kinematics of
multibody systems in EasyDyn: from MuPAD to Xcas/Giac. Theoretical & Applied
Mechanics Letters 3(1), 130012 (2013)
[12] Kouroussis, G., Verlinden, O., Conti, C.: Ground propagation of vibrations from railway
vehicles using a finite/infinite-element model of the soil. Proc. IMechE, Part F: J. Rail and
Rapid Transit. 223(4), 405–413 (2009)
[13] Lysmer, J., Kuhlemeyer, R.L.: Finite dynamic model for infinite media. Journal of the
Engineering Mechanics Division, Proceedings of the ASCE 95(EM4), 859–877 (1969)
[14] Kouroussis, G., Verlinden, O., Conti, C.: Finite-dynamic model for infinite media:
corrected solution of viscous boundary efficiency. Journal of Engineering
Mechanics 137(7), 509–511 (2011)
[15] Kouroussis, G., Verlinden, O., Conti, C.: Efficiency of resilient wheels on the alleviation of
railway ground vibrations. Proc. IMechE, Part F: J. Rail and Rapid Transit. 226(4),
381–396 (2012)
Transfer Path Analysis on a Siemens Combino-Plus Tram
in Almada – Seixal (Lisbon)
Siemens AG Österreich,
Eggenbergerstrasse 31, A 8020 Graz, Austria
gerald.schleinzer@siemens.com
1 Introduction
Transfer path analysis (TPA) is a well-established method to estimate and rank the
contribution of each studied noise transfer path of a noise source [1, 2]. For a railway
bogie the usual noise sources are the rail-wheel contact and additionally traction noise
in the case of a powered bogie. In general, each degree of freedom (DOF) acting at an
interface point is a possible transfer path. In practice not all DOF are used, based on the
experience with the system. For example since the rail-wheel contact acts mainly in a
vertical direction, the vertical transfer paths are often the most important ones. TPA can
also be performed on a computational basis with multi-body-simulation or
finite-element-modeling. Furthermore, the TPA provides acoustical interface forces
which can be used in a customer-supplier relationship to specify the acoustical
properties of a component. In this case the TPA was used to estimate the interior noise
during the design phase of a new vehicle.
2 Mathematical Background
For structure-borne noise, each interface force induces one part of the interior noise
pressure. The sum of all existing paths gives the total interior noise pi, see Eq. (2.1), at
location i inside if all transfer paths j have been considered:
pi = ∑S
paths j
ij Fj (2.1)
where pi is the sound pressure spectrum at the microphone i, Sij is the frequency
response function named “car-body sensitivity” of the receiving system and Fj, Eq.
(2.2), is the interface force during operation at position j. In our case the operating
forces were determined indirectly using the input “impedance” at the DOF interface.
F j = ∑ m jk ak (2.2)
k
where mjk is the dynamic mass of the receiving system and ak are the interface
accelerations when the vehicle is running. Again j is the position of the interface force
and k is the position of the interface acceleration. In the general case, we consider cross
talk where j and k are not the same.
The vehicle on which the measurements were carried out was a Siemens tram type
“Combino Plus” as shown in Fig. 1 below.
The track was a recently (2007) built line south of Lisbon, shown in the system map
in Fig. 2 below.
Transfer Path Analysis on a Siemens Combino-Plus Tram in Almada – Seixal (Lisbon) 339
4 Measurement at Standstill
4.1 Preparation
Sij
mjj aj
F
Fig. 3 above shows the principle of this transfer path analysis. At standstill, the force Fj
was applied with an impact hammer to measure the dynamic mass at the interfaces and
the car-body sensitivity. During running, the accelerations aj were measured which
allowed an indirect evaluation of the in-service forces.
340 G. Schleinzer and T. Kern
Finally, 52 channels were set up to get an overall picture of the structure borne noise
transmitted into the car-body using the analyser shown above in Fig. 4.
4.2 Measurement of the Dynamic Mass of the Interfaces between the Vehicle
Body and Bogie Coupling Elements
Accelerometers were mounted on the car-body at all important interfaces. In some
cases, e.g. for the yaw dampers, triaxial accelerometers were used. Hence, the
accelerations and subsequently the forces in all three directions could be determined.
The measurement was made in a coupled situation which was possible since all
coupling elements had resilient elements at the connection point.
SD
YD
TR
The coupling points between bogie and car-body are the four secondary dampers
SD, the two yaw dampers YD and the traction rod TR as shown above in Fig. 5. The
secondary spring was neglected because of its low stiffness. An example of an
accelerometer mounted at a coupling point is given in Fig. 6 below.
The excitation was applied with an impact hammer as shown in Fig. 7. At places
which could not be reached with the hammer a force transducer mounted on an
all-thread rod was used.
The dynamic mass, Eq. (4.1), at interface j is given by:
1 (4.1)
m jj =
H jj
where
CPS a j F j (ω )
H jj (ω ) = H 1, jj (ω ) = (4.2)
APS F j F j (ω )
and Hjj, Eq. (4.2), is the point inertance estimated by using the H1 frequency response
estimator with CPS being the averaged cross power spectrum and APS being the
342 G. Schleinzer and T. Kern
averaged auto power spectrum, typically over five hits with the impact hammer. Since
the coupling points are well away from each other, cross talk Hij with i≠j was
disregarded and therefore the dynamic mass is the reciprocal of the inertance as in
equation (3.1). Otherwise, it would be the inverse of the matrix Hij. Fig. 8 gives one
example for the mobility of the point where the secondary damper is mounted to the
car-body (Fig. 6). Instead of the dynamic mass, the mobility is shown, simply because it
is more commonly used.
-80
-90
Amp in dB
re 1 m/Ns -100
-110
-120
16 31.5 63 125 250 500 1000
frequency in Hz
In the second step the vehicle was driven as if in normal operation over the track. Here,
the operational accelerations were measured for Eq. (3.2), which provided the coupling
forces. In addition to the data from the accelerometers and microphones. GPS
coordinates were recorded for a later selection of different track segments. Subjective
impressions or special events on the track were noted separately.
Transfer Path Analysis on a Siemens Combino-Plus Tram in Almada – Seixal (Lisbon) 343
6 Results of TPA
After selecting adequate track segments, the calculation scheme, as shown in Eqs. (2.1)
and (2.2), was used. Additionally, a 1/3rd-octave band analysis was applied.
60
50
Amp in
dB(A) re
2e-005 Pa 40
30
calculated TPA sumlevel
measured sumlevels
20
16 31,5 63 125 250 500 1000 2000 4000
frequency in Hz
Fig. 9 above shows the result for one track segment. The red and green lines are the
directly measured interior noise at two microphone positions in the bogie section of the
car-body. The blue line is the sum of the structure-borne paths and an important result
of the TPA. It is clearly visible that structure borne noise coming from the bogie
dominates the interior noise at frequencies up to 315 Hz. Above this frequency, the
influence of airborne noise coming through the floor and the windows grows
continuously. Finally, we saw the yaw damper in the lateral direction gives the highest
input to the interior noise. Also, the sum of all structure-borne paths is several dB lower
than the measured interior noise and therefore, the airborne paths still dominate the
interior noise.
7 Validation
To ensure that the structure-borne noise transfer has been estimated correctly, a
complementary direct measurement (Fig. 10) was performed on a yaw damper. As
shown below in Fig. 10, the damper was excited at one end with a shaker and the
reaction forces were measured directly at the other end with a triaxial force transducer.
This gave the dynamic transfer stiffness of the damper. This dynamic stiffness was then
multiplied with the measured displacement on the bogie-side damper interface and
gave an alternative (more direct) estimation of the acoustical interface forces.
344 G. Schleinzer and T. Kern
Fortunately, the result of this alternative measurement gave the same result as the
one provided by TPA within an uncertainty of ±2 dBA.
References
[1] Moorhouse, A.T., Elliott, A.S., Evans, T.A.: In situ measurement of the blocked force of
structure-borne sound sources. Journal of Sound and Vibration 325, 679–685 (2009)
[2] Plunt, J.: Strategy for transfer path analysis (TPA) applied to vibro-acoustic systems at
medium and high frequencies. Proc. ISMA 23, Belgium
Characteristics of Sound Insulation and Insertion
Loss of Different Deloading Sound
Barriers for High-Speed Railways
Summary. When a high-speed train quickly passes along the sound barriers
installed on the two sides of a viaduct, the fluctuating wind pressure acting on
the sound barriers causes an increase in the aerodynamic loading acting on the
viaduct. It not only affects the train operation safety adversely, but also speeds
up the sound barrier damage rate and shortens their service life. Thus,
high-speed railway noise barriers should not only satisfy the requirement that
they have a noise reduction function but also that they have a deloading
function due the fluctuating train air-load acting on them. Hence, the design and
use of deloading noise barriers serve this purpose. This paper investigates the
acoustic performance of different deloading sound barriers numerically. In the
investigation, the sound insulation calculation model is established and verified
by using an existing effective mass model. The propagation characteristics of
the sound transmitted through the barriers are analyzed and the sound diffracted
over the barrier top is considered according to the field tests. The results are
useful to the efficient acoustic optimization design of high-speed railway
deloading sound barriers.
1 Introduction
Track v Train
Noise Barrier
where m0 = ρ (l0 + 2Δl ) / σ is the effective mass per unit area and θ is the angle of
incidence. The value of A is 1, a constant proportional to the noise source power. k is
the wavenumber, r01and r21 are the distances from the source and receiver to the
elemental area ds on the screen, at angles θ01 and θ21 respectively to the normal of
the surface, ρ is the density of the air, σ is the open area ratio (0.5), l0 is the gap depth
(247 mm) andΔl is the gap width (20 mm).
Characteristics of Sound Insulation and IL of Different Deloading Sound Barriers 347
Gap
5m
Sealed
Wall Line source
Free field
Incident direction
Deloading barrier 4m
Source room
Sealed
Fig. 3 shows a comparison of the results calculated by the MLE method and the
BEM at frequencies of 200 Hz and 400 Hz.
In Fig. 3 the vertical ordinate represents the transmission loss, which is defined as
the difference of the incidence sound pressure and the field point sound pressure. The
abscissa represents the distance between the noise barrier and the field point. As
shown in Fig. 3, the maximum differences between the simulation results and the
theoretical results at the 2 frequencies are only 0.6 dB and 0.9 dB, respectively. It
follows that, using the BEM to calculate the transmission loss in this model at low
frequencies is reliable. But the BEM is suitable for all frequencies, so the present
calculation model is valid and reasonable.
348 B. He et al.
In order to study the sound insulation and propagation law of different barriers, three
barrier structures are considered as shown in Fig. 4. The structures are respectively
louver-shaped, cylinder-shaped and V-shaped. The size of each is marked in the
figures.
175mm 65 mm
20 mm
50 mm
20 mm 20 mm
20 mm 20 mm
(a) Louver-shaped (b) Cylinder-shaped (c) V-shaped
The transmission loss distributions of the three barriers are calculated based on the
model shown in Fig. 1. To understand the propagation of the sound transmitted
through the barrier, Fig. 5 gives the transmission loss nephogram of the louver-shaped
barrier at 500 Hz and 2500 Hz. The vertical axis represents the barrier length, and the
lateral axis indicates the distance from the barrier. From Fig. 5, it can be seen that the
transmission losses are symmetrically distributed about the center line of the 4m noise
barrier at any frequency. Behind the noise barrier, the near region is an interference
area. In the rear of the interference area, a ‘secondary source’ emerges. After the
source, the sound transmits normally. At 500 Hz, the interference area is about 0~9 m
from the barrier, but the area is about 0~40 m at 2500 Hz. It’s clear that, as the
frequency increases, the impact of the interference on the sound field behind the
barrier is greater.
Secondary source
Secondary source
Distance from noise barrier (m) Distance from noise barrier (m)
(a) Transmission loss nephogram at 500 Hz (b) Transmission loss nephogram at 2500 Hz
Fig. 5. Transmission loss nephogram at different frequencies (Louver-shaped)
From Fig. 6 it can be seen that, in a certain area behind the barrier, the transmission
loss fluctuates with distance. At 500 Hz, it fluctuates between 0 and 9 m. At 9 m away
from the barrier, it’s at its minimum and corresponds to the ‘secondary source’. At
more than 9 m, the value increases gradually. At 9 m, the transmission losses of
louver-shaped, cylinder-shaped and V-shaped barriers are, respectively, 1.4 dB,
3.0 dB and 3.7 dB. The difference between the transmission losses of the
louver-shaped and the cylinder-shaped barriers is 1.6 dB and the difference between
the cylinder-shaped and the V-shaped barriers is 0.7 dB. When the distance is more
than 9 m, the two differences are, respectively, 1.4 dB and 0.8 dB. So, the sound
propagation beyond the secondary source is invariant. The increase of transmission
loss is only caused by the distance due to air absorption. The sound spreading in this
region is mainly caused by the secondary source. When the distance from the noise
barrier to the field points doubles, the transmission loss increases by 2.9 dB. This
sound propagation is coincident with the line source propagation law. At 1000Hz, the
interference area is broader. The propagation follows the same law. When the
distance is doubled, the increase of transmission loss is 2.8 dB, which is coincident
with the line source propagation.
From the previous analysis, the sound field behind the secondary source is mainly
determined by this source. Furthermore, the difference between the maximum
minimum transmission losses of the interference zones is within 3dB. Hence, the
paper defines the transmission loss at the secondary source as the sound reduction
index (SRI). Fig. 7 shows the SRI of the three deloading barriers at the frequency of
500 Hz ~3150 Hz, which is the main frequency range of high-speed train noise.
At all frequencies, the sound insulation performance of the V-shaped barrier is the
best, for the three types of the barriers. In the different frequency bands, the
performances of the louver-shaped and the V-shaped barriers are different. At
500, 630, 2000 and 2500 Hz, the cylinder-shaped barrier has better sound insulation
performance; at other frequencies, the louver-shaped barrier is preferred. In order to
realistically evaluate the insertion losses of these barriers, the actual source spectra of
high-speed trains are required.
350 B. He et al.
The diffracted sound waves over the top of the barrier combine with the transmitted
sound waves to reduce the effectiveness of the deloading barriers.
If IT and ID are the sound pressure contribution at receiver due to transmitted
through the barrier and the sound pressure contribution due to the sound diffracted
over the barrier top respectively, compared to a solid barrier, the increased pressure
level in dB due to the deloading barriers is:
ΔLp=10log(1+IT/ID) (3.1)
Door area
Connection area
Fig. 8. Source identification nephogram of high-speed train (v=342 km/h)
The sources are extracted from the field test. In order to predict the sound barrier
insertion loss precisely, the noise sources of a high-speed train are obtained through
beamforming array measurement. In Fig. 8, the top left figure shows the
Beamforming used for source identification, the top right figure is a CRH high-speed
train of China and the lower figure is the source identification nephogram of the
high-speed train operating at speed of 342 km/h. According to the main sound source
Characteristics of Sound Insulation and IL of Different Deloading Sound Barriers 351
distribution, the sound sources are divided into four segments, which are the
wheel/rail area, door area, connection area and pantograph area, respectively. The
integral of the pressure in each area is carried out and then the total pressures of each
segment are averaged.
The average sound pressure levels are shown in Fig. 9. The wheel/rail area noise,
which includes wheel/rail rolling noise and aerodynamic noise in the wheel/rail area,
is much louder than noise from other areas. Fig. 10 gives the diffraction model that
considers the multiple reflections between the sound barrier and the vehicles, as well
as the viaduct. The four points indicate the four source types, respectively. The height
of the solid barrier is 2.15 m.
Sound source
Viaduct
The sound pressure levels behind the solid barrier are calculated on the basis of the
model shown in Fig. 10. The increased sound pressure levels are computed according
to equation (3.1). Thereby, the sound pressure levels behind the deloading barriers are
obtained. Fig. 11 gives the insertion losses at five typical field points. The insertion
loss of the V-shaped barrier is the greatest at all the field points. At the point 30 m
away from the track center line and at a height of -8.5 m (1.5 m above the ground),
the insertion loss is 2.6 dB. The louver-shaped barrier is better than the
cylinder-shaped barrier at the near field point, but worse at the far field point. Due to
the interference effect, some of the insertion losses are negative. Placing absorbing
material on the sound barriers or reducing the open gap areas in the barriers can
increase the sound insulation and improve the insertion loss.
4
Louver-shaped
Cylinder-shaped
2
Insertion loss (dB)
V-shaped
-2
-4
( 7.5m,1.2m) ( 7.5m,3.5m) ( 25m,3.5m) ( 30m,-8.5m) ( 30m,1.5m)
4 Conclusion
The paper has established a calculation model for sound insulation and insertion loss
of deloading barriers. The sound insulation model is verified by the MLE method in
the low frequency range. The sound insulations and insertion losses of three types of
deloading barriers are calculated. The results show that the sound insulation
calculation model presented in the paper is suitable for analyzing the effect on line
noise sources when the high-speed trains pass by. In the near area behind the barriers,
the sound field interference is strong. In the rear of interference area, there is a
‘secondary source’. In the far area behind the barrier, the sound spreads according to
the principle of line source radiation. As the frequency increases, the impact of the
interference on the sound field behind the barrier becomes greater. The sound
insulation performance of the V-shaped barrier is the best among the three kinds of
barriers. In different frequency bands, the sound insulation performances of
louver-shaped and V-shaped barriers differ.
The insertion loss calculation model is established based on source identification.
In the model, different sources and multiple reflections are considered. The insertion
loss of the V-shaped barrier is the greatest at all the field points, compared to the other
two kinds of barriers. The calculation model and related results provide scientific
guidance for the efficient acoustic optimization design of the sound barriers used on
high-speed railways.
Acknowledgement. The present work was supported by the National Science and
Technology Support Program of China (2009BAG12A01-B06), the 2011 Doctoral
Innovation Funds of Southwest Jiaotong University and the Fundamental Research
Funds for the Central Universities, the National High Technology Research and
Development Program of China (863 Program) (2011AA11A103-4-2).
References
[1] Chandler-wilde, S.N., Hothersall, D.C.: Sound propagation above an inhomogeneous
impedance plane. Journal of Sound and Vibration 98(4), 475–491 (1985)
[2] Matsumoto, T., Yamamoto, K., et al.: Scale model studies of new type highway noise
barriers. In: Proceedings of Inter Noise 1994, Yokohama, Japan (1994)
[3] Watts, G.R., Hothersall, D.C., Horoshenkov, K.V.: Measured and predicted acoustic
performance of vertically louvered noise barrier. Applied Acoustics 62, 1287–1311 (2001)
[4] Cremer, L., Müller, H., (translated by Schultz T.J.): Principles and applications of room
acoustics, vol. 2, pp. 181–190. Applied Science Publishers (1982)
Optimizing Capacity of Railroad Yards within Noise
Limits Using a Dynamic Noise Model
Summary. At railroad yards passenger trains are parked overnight and prepared
for the next day and/or freight trains are shunted. These activities are diverse and
commonly processed ad-hoc. Noise permits for railroad yards are in contrast
inflexible. More activities take place at the railroad yards as the intensity on the
tracks increases. Noise production is increased and the permit becomes more and
more restrictive.
ProRail is the railway infrastructure manager and responsible for complying
with the community noise limits from the environmental laws. ProRail is also
responsible for the yearly capacity allocation process. A noise propagation model
is used for the calculation of the noise levels. Building such a model is
time-intensive so the noise impact of only one mode of operation of the yard is
investigated, leaving other modes of operations to be unknown.
M+P developed a software program “Dynamic Noise Model” (DNM) on
behalf of ProRail. Many possible modes of operation can be evaluated with just a
push of a button with this calculation tool. This makes it easier to find alternative
modes of operations which produce less noise. Optimization of the use of the
yard within noise limits is possible. A DNM also can be used to make informed
decisions about taking noise measures or applying changes to the activities in the
yard.
The DNM uses a database with two datasets. The first set has the attenuations
between standardized processes and receivers. These attenuations are calculated
in advance with a propagation model. The second dataset contains all sound
power levels (SPL) by train type. This is combined with a list of all activities
taking place in the railroad yard, resulting in the total SPL for all receiver
positions.
ProRail is going to use the DNM for a large number of railroad yards for
testing for compliance with environmental laws during the capacity allocation
process.
1 Introduction
At railroad yards passenger trains are parked overnight and prepared for the next day.
Also freight wagons can be shunted at these yards. Typical noise sources are idling
diesel engines, compressors, ventilation, squeal noise from switches or curves, impact
noise from joints etc.
Noise originating from railroad yards is restricted by permits in the Netherlands.
Each yard has its own permit with local noise limits which differ for the three
evaluation periods day, evening and night. In the Netherlands there are about 100 of
these yards. A permit is usually updated every 5 to 10 years.
ProRail is the railway infrastructure manager and responsible for compliance with
the environmental noise limits. ProRail is also responsible for the yearly capacity
allocation process. As the intensity on the tracks increases, more activities are
performed in the railroad yards. The need increases for better and fast tests for
compliance with noise limits.
During the capacity allocation process the planned activities need to be tested
against the noise regulations. Noise propagation models with detailed information
about the planned activities were used for the calculation of the noise levels. It takes
about 2 to 4 weeks per yard to build these and every successive model. Successive
models are needed when a specific set of planned activities fails to fit in the noise
regulations. Building these models is done by specialists. This sequential process of
gathering the necessary information from the railway operators, the modeling of the
planned activities and the testing of the noise levels takes too long to fit in the capacity
allocation process.
A new approach of the problem has led to the development of a software program
“Dynamic Noise Model” (DNM) by M+P. The new approach requires less detailed
information of the activities. The time to perform a test is reduced from 2 to 3 weeks to
only 1 second. Acoustical knowledge is no longer required so that capacity planners of
ProRail can perform the calculations themselves.
In this article the principle of the DNM is explained. First, a short introduction of the
Dutch legislation concerning railroad yards is given. Then, the principle of the DNM is
described. Finally, the advantages and disadvantages of the DNM are evaluated.
The sequence of modeling and testing for compliance with noise limits using a DNM is
shown in Fig. 1.
combine this with the PWL per process and per Done each test.
train type With the DNM this can
be done in a second
calculate the total SPL per evaluation period and
check them against the levels in the permit
In Fig. 1 the last part is done by software, the DNM. These calculations can be done
quickly. The input is a list of activities which can be changed easily. Being able to
change the activities dynamically is what gives the software its name: “Dynamic Noise
Model” (DNM).
Each step shown in Fig. 1 is described in more detail in the following paragraphs.
process location 5
Fig. 2. Schematic layout of a railroad yard with process locations and driving lanes which connect
them
Optimizing Capacity of Railroad Yards within Noise Limits Using a DNM 357
The noise sources that can originate from a driving lane are:
• driving noise: wheel-rail interaction and traction;
• braking noise;
• squeal noise from diverging passages of switches;
• impact noise from joints in track, switches or crossings;
• noise from accelerating trains.
Depending on the distance from the receiver point a driving lane can consists of a
number of neighboring tracks.
Modeling of the standardized process locations and standardized driving lanes is done
with the following rules:
A driving lane consists of several tracks. Therefore switches and crossings within a
driving lane are not always passed. The calculation of the noise of the passing of these
switches and crossings is done in a statistical way, where the likelihood of passing is
used to calculate the expected use. This expected use is also stored in the database per
switches or crossings, per driving lane.
In Table 1 an overview is given of all sound sources and the way they are
represented in the model. Note that some have the same representation in the model,
which makes modeling easier.
358 S.N. Hoogzaad and M.S. Roovers
Table 1. Connection between the sound sources, the location and modeling
During the capacity allocation process, the railway operators give information about
the processes they want to execute. ProRail combines all requests for capacity and tries
to fit all requests within the physical, safety and environmental constrains. For the
railroad yards this results in a list of activities that different operators wants to execute.
These processes are split in three groups for the three separate evaluation periods.
A standard process on a process location is defined as one second of process time with
one wagon. Thus a process which takes one hour and is executed with a train consisting
of eight wagons is equal to 60*60*8 = 28800 standard processes. These calculations are
done by the DNM separately for the three evaluation periods.
Optimizing Capacity of Railroad Yards within Noise Limits Using a DNM 359
The driving noise is corrected for the driving speed. If a train drives twice as fast as
the standard process, traveling time is halved. The faster driving train produces a higher
PWL. Both corrections are made by the DNM.
The noise productions are different per train type and per type of activity. The PWL per
octave is known per train type and per operation mode and sorted in a database. To
enable some flexibility, some train types are grouped.
During the calculations of the total SPL per receiver, the database with PWL is
combined with the database of attenuations in the following way:
1. find the relevant PWL for an activity that is executed;
2. reduce the PWL by the attenuation per receiver.
3.7 Calculate the Total Sound Levels and Check with the Permit
The last step is to adjust by the logarithm of the number of times a standard process it is
executed. Then all SPL caused by all activities are summed. The resulting total SPL is
tested against the noise levels from the permit.
The DNM has information about all activities and resulting noise levels. From this the
DNM can determine the dominant activities and noise levels can be reduced by finding
alternatives for these activities which produces less noise.
Dominant noise source can also be identified by the DNM. Calculation of the effect
of source measures is possible within the DNM. For example the effect of measures to
reduce squeal or impact noise from switches can be calculated per switch. In this way
the number of source measures can be optimized, thereby reducing the costs.
Using the DNM has advantages and disadvantages when compared with the sound
propagation models which calculate attenuations between source and receivers using
transfer functions. Sound propagation models require a precise description of the sound
source. This makes these models inflexible if different activities have to be modeled
sequentially. The DNM is very good at handling different activities sequentially, but
has difficulties in handling changes in attenuations by for example changes in the
propagation path. The advantages of the DNM are:
• Fast calculation of the noise levels of various modes of operations.
• Easy identification of the processes which cause high noise levels.
• Optimization of the use of the yard within noise limits.
360 S.N. Hoogzaad and M.S. Roovers
6 Conclusions
Dynamic Noise Models can handle changes in modes of operation easily. This makes
Dynamic Noise Models useful during the capacity allocation process, when
calculations of the SPL of different modes of operation of a railroad yard have to be
done quickly. Also the communication between ProRail and the rail operators about the
effect of sound limits on the activities becomes easier. This makes it possible for
ProRail to be compliant with the environmental laws and optimize the use of the yard
during the capacity allocation process.
Modeling of Wheel-Track Interaction with Rail Vibration
Damper and Its Application for Suppressing Short Pitch
Rail Corrugation
Summary. Some kinds of short pitch rail corrugation are associated with the
standing waves of rail vibration. These standing waves can be suppressed by rail
vibration dampers, although they were originally developed to reduce the railway
rolling noise via attenuation of the wave propagation in the rail. In this study the
track models with rail dampers applied are introduced. Two kinds of wheel-track
interaction, the single and multiple wheel-rail interaction, are studied to
investigate the effects of the rail damper on the suppression of short pitch rail
corrugation from the point of view of wheel-rail interaction force.
1 Introduction
Rail vibration dampers were originally developed to reduce rail-radiated noise. A rail
damper introduces extra damping to the rail and increases the decay rate of wave
propagation in the rail. A compound track model combined with the rail damper is
required for simulation of the wheel-rail dynamics. Theoretical modeling and analysis
of track dynamics with the rail damper applied was conducted by Thompson [1] and
Wu [2].
It is believed that the formation of short pitch rail corrugation or “roaring rails” is
associated with the pinned-pinned resonance of rail vibration. At the pinned-pinned
resonance a standing wave mode appears in the rail and its half wavelength equals the
length of one sleeper span. Studies have shown that stiffer railpads lead to higher
wear rates and faster corrugation growth [3], and that the use of resilient railpads
delays the formation of rail corrugation [4].
Recently, short pitch rail corrugation has also been found to form consistently on
the tangent track sections in some metro lines in Beijing and Shanghai, where the high
compliance rail fasteners are used to reduce the ground-borne vibration of the tunnels.
The pinned-pinned resonance there is weak and the damping introduced by the rail
fastener is low, so that the decay rate of wave propagation in the rail is also low. This
seems beneficial to reduce short pitch rail corrugation rates. Due to the low vibration
decay rate, however, the vibration wave is reflected at the wheels and may form
standing waves in the rail between the wheels. These standing waves may lead to
large fluctuations in the wheel-rail interaction force. Combined with the traction force,
the fluctuating wheel-rail force may cause periodic and uneven wear on the rail head
and thus the formation of short pitch rail corrugation.
Both kinds of rail corrugation mentioned above are related to the standing waves in
the rail, although their generation is by different mechanisms. As the rail damper can
effectively attenuate bending wave propagation in the rail, it is possible to suppress
short pitch rail corrugation by reducing the intensity of the standing waves with the
rail damper. The effects of the rail damper on short pitch rail corrugation were studied
by Croft [5] and Wu [6]. They both found that rail corrugation associated with the
pinned-pinned resonance can be reduced with rail dampers.
In this study track models with the rail damper included are introduced. Two kinds
of wheel-rail interaction, the single and multiple wheel-rail interaction, are studied to
investigate rail corrugation growth. It is shown that the suppression of short pitch rail
corrugation by damping rail vibration with rail dampers is possible.
Two track models with the rail damper applied are used in this study. The first one is for
ballasted track which is modeled as an infinite Timoshenko beam discretely supported
by the railpad, sleeper and ballast in series, as shown in Fig. 1. Each rail damper
attached to the rail is represented by a discrete mass-spring system. This model is used
for analyzing the mechanism of short pitch rail corrugation associated with the
pinned-pinned resonance. The second model is of slab track, modeled as an infinite
Timoshenko beam on a continuous elastic foundation. The rail damper is represented
by a continuous mass-spring layer, shown as in Fig. 2. This model is used for multiple
wheel-rail interaction and for analysis of the mechanism of short pitch rail corrugation
due to the use of high compliance rail fasteners.
The equations of motion can be found in [2] for the two track models with the rail
damper applied. The parameters used in the models are listed in Table 1.
fc
md rail damper
k kd
rail
pad kp
sleeper ms
ballast kb
md
damper
kd
rail
kp
pad
mw V
xw
kH
xr r
Fig. 3. Single wheel-rail interaction
R
Fc = − (3.1)
αw +αc +αr
where αw = −1/mwω2 is the receptance of the mass wheel, αc = 1/kH is the receptance of
the contact spring, where kH is the linear contact stiffness, and αr is the point
364 T.X. Wu and Y.R. Wang
receptance of the rail at the excitation position. R is the spectrum of the combined
roughness on the wheel and rail tread. The sign convention adopted here for
roughness is positive for a rail dip and negative for an asperity.
-8
Receptance (m/N)
10
-10
10
2 3
10 10
Frequency (Hz)
Fig. 4. Receptance of rail vibration, — excitation at mid-span, − · − excitation at sleeper
Rail receptance is calculated using the track model in Fig. 1 (no rail damper) for
excitation at mid-span and at a sleeper. The rail receptance can be seen from Fig. 4 to be
similar for both cases in the frequency region up to 800 Hz. The pinned-pinned
resonance of rail vibration appears at about 1 kHz due to the wave reflections at the rail
supports. For other positions in a span the rail receptance varies between the two
extreme cases.
The wheel-rail interaction force caused by roughness excitation is calculated using
equation (3.1). In the calculations the wheel mass is chosen to be mw = 600 kg,
including the unsprung mass attached to the wheel, and the contact stiffness
kH =1.436 MN/mm. The results are shown in Fig. 5 for the wheel-rail interaction force
due to 1 μm roughness input.
104
Wheel rail force (N/um)
103
102
102 103
0
Phase (degrees)
-45
-90
-135
-180
102 103
Frequency (Hz)
The wheel-rail interaction forces are similar up to about 800 Hz for both cases of
the wheel at mid-span and at a sleeper. However, around the pinned-pinned resonance
frequency the wheel-rail interaction forces differ in both magnitude and phase for the
two wheel positions. Since the wheel-rail force varies at different positions of the
wheel in a sleeper span during the train passage, uneven wear may occur. As a result,
short pitch rail corrugation associated with the pinned-pinned resonance may form on
the rail head after the passage of millions of wheelsets.
In Fig. 5 the phase of the wheel-rail force (to the rail) is relative to the roughness
phase and has an important influence on the wear-type rail corrugation. It will grow if
the wheel-rail force reaches a maximum on a crest and a minimum in a trough of the
corrugation. This is because a lower wheel-rail force results in a larger slip zone and
more wear in the contact area, whereas a higher wheel-rail force causes less wear,
provided that the longitudinal traction force remains stable.
-8
Receptance (m/N)
10
-10
10
10 2 103
Frequency (Hz)
Fig. 6. Rail receptance with rail damper, keys as for Fig. 4
366 T.X. Wu and Y.R. Wang
103
102
102 103
0
Phase (degrees)
-45
-90
-135
-180
102 103
Frequency (Hz)
Fig. 7 Wheel-rail interaction force due to 1 μm roughness with rail damper, keys as for Fig. 4
For the slab track in the tunnel the high compliance rail fasteners are used to reduce
ground-borne vibration caused by the train passage. As the stiffness of the high
compliance fastener is very low, e.g. about 10 MN/m, the pinned-pinned resonance
mode will not occur. However, since the damping introduced by the rail fastener is low,
another type of standing wave may occur in the rail due to the reflections at the wheels.
This type of standing wave may lead to large fluctuation in the wheel-rail interaction
force and thus short pitch rail corrugation.
Fig. 8 shows the multiple wheel-rail interaction model, in which four wheels are
represented, corresponding to a pair of bogies of two adjacent cars. The interaction
force F1 (active) caused by the roughness at wheel 1 will generate an incident wave
propagating along the rail, which interacts with wheels 2, 3 and 4, leading to the
generation of ‘passive’ wheel-rail interaction forces P21, P31 and P41. These passive
wheel-rail forces also generate propagating waves of rail vibration.
z
wheel 1 wheel 2 wheel 3 wheel 4
u
F1 contact spring
P21 rail P31 P41
u1 u2 u3 u4
4
10
2
10
1
10
2 3
10 10
0
Phase (degrees)
-45
-90
-135
-180
2 3
10 10
Frequency (Hz)
Fig. 9. Active wheel-rail interaction force, — at wheel 1, ···· at wheel 2
4
10
Active force (N/um)
3
10
2
10
1
10
2 3
10 10
0
Phase (degrees)
-45
-90
-135
-180
2 3
10 10
Frequency (Hz)
Fig. 10. Active wheel-rail interaction force for rail damper applied, — at wheel 1, ···· at wheel 2
The active wheel-rail interaction forces are calculated for 1 μm roughness excitation
and the results are shown in Fig. 9 and 10 for the cases with and without rail dampers,
respectively. The parameters used here are mw = 840 kg, kH = 1.14 MN/mm,
md = 16 kg/m, kd = 227 MN/m2 and ηd = 0.25. The parameters for the rail damper
correspond to a natural frequency 600 Hz.
The active wheel-rail force shown in Fig. 9 has some high magnitude peaks at above
200 Hz. More peaks appear for wheel 2 than for wheel 1 because it receives wave
reflections from both sides. The phase of the wheel-rail force (to the rail) is relative to
the roughness; 0˚ indicates that the wheel-rail force reaches maximum at a roughness
dip, and −180˚ indicates that the wheel-rail force reaches a minimum at a dip. In the
latter case a larger slip zone and more wear occur in the contact patch, and thus the
wear-type rail corrugation is able to grow. When the rail damper is applied, the peaks
in the active wheel-rail force spectrum are suppressed at higher frequencies, because
the extra damping effectively attenuates the incident and reflected waves. Therefore,
the rail corrugation formation due to the use of high compliance rail fastener can be
suppressed by the rail damper.
Two types of standing waves exist in the rail vibration. One is the pinned-pinned
resonance mode appearing in the stiffly supported rail. The other occurs when the rail
fastener has very low stiffness and leads to resonances between the wheels. These
standing waves cause large variation in the wheel-rail forces and thus short pitch rail
corrugation. However, they can be suppressed by rail vibration dampers, and thus rail
corrugation associated with them may be avoided.
The current study has been supported by the Research Fund for the Doctoral
Program of Higher Education of China, Grant No. 20120073110047.
References
[1] Thompson, D.J.: A continuous damped vibration absorber to reduce broad-band wave
propagation in beams. Journal of Sound and Vibration 311, 824–842 (2008)
[2] Wu, T.X.: On the railway track dynamics with rail vibration absorber for noise reduction.
Journal of Sound and Vibration 309(3-5), 739–755 (2008)
[3] Ilias, H.: The influence of railpad stiffness on wheelset/track interaction and corrugation
growth. Journal of Sound and Vibration 277(5), 935–948 (1999)
[4] Hempelmann, K., Knothe, K.: An extended linear model for the prediction of short pitch
corrugation. Wear 191, 161–169 (1996)
[5] Croft, B.E., et al.: Modelling the effect of rail dampers on wheel–rail interaction forces and
rail roughness growth rates. Journal of Sound and Vibration 323, 17–32 (2009)
[6] Wu, T.X.: Effects on short pitch rail corrugation growth of a rail vibration absorber/damper.
Wear 271, 339–348 (2011)
Investigating the Effects of a Network-Wide Rail
Grinding Strategy on Wayside Noise Levels
1 Introduction
This paper presents work commissioned by Network Rail to demonstrate the reduction
in noise emissions achieved between the first and second rounds of noise mapping for
the Environmental Noise Directive (END) that were completed in 2007 and 2012
respectively. The full version of this article has been published by SAGE in the Journal
of Rail and Rapid Transit [1].
Rolling noise is the most important source of railway noise at the typical operating
speeds of trains on the GB network. Rolling noise is generated by the dynamic forces
created at the wheel-rail interface by the interaction of the roughness of the wheel and
rail. In GB, the Calculation of Railway Noise 1995 (CRN) [2] was used as the
methodology for the first and second rounds of END noise mapping. CRN is largely
empirical and focuses on rolling and traction noise. Curve and brake squeal are not
currently included in the mapping results. With operational running speeds limited to
200 km/h on the network, except for trains on GB’s High Speed 1 line, aerodynamic
noise has little relevance to the majority of the GB network [3]. For these reasons rail
roughness is a key parameter for the accurate prediction of rail noise for the END.
Between 2002 and 2004, Network Rail developed a new preventative maintenance rail
grinding strategy to address rolling contact fatigue, involving the purchase of three new
64-stone grinding machines. These machines work on a cyclic basis across the heavily
used parts of the network with the grinding trains travelling as ‘slow moving trains’
within revenue service traffic. This has enabled greater productivity by allowing
grinding to be undertaken outside of line closures. Currently, this strategy is applied to
lines carrying more than five million tonnes of traffic per year. From 2009 grinding was
carried out based on curvature and tonnage and originally was carried out at every 15
Equivalent Million Gross Tonnes (EMGT) on curves of less than 2500 m radius and
every 45 EMGT on curves of radius greater than 2500 m and straight track. Currently,
therefore, a typical section of main line track might be ground every one or two years
on straight sections and every six months on curves. No cyclic grinding was undertaken
on the network for the 10 year period prior to 2002. Grinding was limited to the use of
small machines on a site-specific basis. While the purpose of the grinding is not to
reduce noise, rail grinding affects wayside rolling noise by virtue of the smoother rails
that result [4]. The grinding strategy introduced between 2002 and 2004 is expected to
have an effect on wayside noise levels on main lines.
In 2004 a study was completed by AEA Technology Ltd on behalf of the Department
for Environment, Food and Rural Affairs (DEFRA) [5, 6] that considered the
implications on noise predictions of a level of rail quality different from that which was
implicit in the rolling stock source terms of the CRN methodology. The study used the
NoiseMon system to perform indirect measurements of rail ‘acoustic’ roughness. The
NoiseMon system consisted of a microphone mounted beneath the floor of British Rail
Mk 3 coaches together with a data acquisition system inside the coaches.
Measurements were taken for normal operations on Great Britain’s East Coast and
West Coast Mainlines in the early 2000s to determine statistically the variation in the
condition of the rail running surfaces on the GB network. The difference between the
wayside level that was produced by the Mk 3 coach carrying the NoiseMon system and
the level that would be predicted with CRN, at 25 m from the nearest rail at a given
location, was termed the Acoustic Track Quality (ATQ). Measuring the wayside noise
of train passbys at a statistically large-enough number of measurement locations was
not feasible to evaluate the ATQ for the whole network. Nor was it feasible to measure
directly the rail roughness of the whole network. A feasible alternative used in [5] was
to derive the distribution of noise levels on the GB Network by measuring under-floor
noise on large sections of track and normalizing the data to a single train speed;
estimating the corresponding distribution of wayside noise using an under-floor to
Investigating the Effects of a Network-Wide Rail Grinding Strategy 371
wayside transfer function measured during the passby of the Mk 3 coach carrying the
NoiseMon system; then comparing this to the reference single vehicle SEL for a Mk 3
coach given in CRN to derive the ATQ. A positive value of ATQ indicates that the level
produced by the Mk 3 coach would exceed the level that would be predicted by CRN at
that location, suggesting a higher contribution of noise caused by increased rail
roughness. The ATQ determined from the 2004 study presented in terms of its
distribution over a large section of the network is reproduced in Fig. 1. The mode (value
that appears most often) of the ATQ shown in Fig. 1 is +4 dB. The 2004 result indicates
that the noise produced by a Mk 3 coach on the GB Network would, on average, be
4 dB higher than the equivalent level that would be predicted by CRN.
Number in 1dB−wide Category (%)
12
10
0
−20 −10 0 10 20 30 40
Acoustic Track Quality (dB)
Fig. 1. Distribution of the Acoustic Track Quality on typical GB track assumed for 2004 –
reproduced from [5]
3.1 Summary
The objective of the current study was to determine an up to date value of ATQ for the
GB network and investigate any change attributable to the new grinding strategy. To
maintain consistency, a similar approach to that used in the 2004 study has been
adopted. Under-floor to wayside transfer functions were measured at three sites on the
East and West Coast Main Lines (ECML & WCML): Tring on the WCML; Rossington
on the ECML; and Burton Coggles on the ECML. Transfer functions were successfully
obtained for a total of 6 different tracks, 2 at each location. The transfer functions were
used to derive an up to-date value for ATQ from 1100 km of under-floor microphone
data gathered on the same lines in 2012. Fig. 2 shows the 2012 ATQ compared to the
findings of the 2004 study.
The key outcome is that the mode of the ATQ is negative. The mode of the ATQ for
the combined routes is -4 dB. When compared to the 2004 study, this Fig. suggests that
current rolling noise levels for smooth-wheeled trains on the main routes are, on
average, 8dB quieter than measured in 2004. This implies a significant reduction in rail
372 N.J. Craven et al.
head roughness. A study [7] carried out by DeltaRail for the Rail Safety and Standards
Board demonstrates that the ATQ for track that meets the limit levels for rail roughness
and track decay rate required for train noise tests as part of the Technical Specification
for Interoperability (TSI) [8] for noise from conventional railways is approximately
-1.7 dB. Therefore the current study suggests that the WCML and ECML have, on
average, rail roughness levels that are compliant with the roughness limit defined in the
Noise TSI. Another observation is that the shape of the distribution is narrower than
that obtained in the 2004 study. This change is consistent with a step change in rail
grinding, which eliminates sections of rail with very high roughness (for example
corrugated rails). This change in distribution shape was also predicted in [5].
Number in 1dB−wide Category (%)
15
10
0
−20 −15 −10 −5 0 5 10 15 20
Acoustic Track Quality (dB)
Fig. 2. Reduction in GB average ATQ and associated distributions. −−, 2012; − −, 2004.
The NoiseMon system is currently carried on a single Mk 3 coach that is part of the
Network Rail New Measurement Train (NMT). The NMT consists of a rake of 5 Mk 3
coaches and 2 Class 43 power cars. It is therefore necessary to “extract” the noise of
one Mk 3 coach from the wayside measurements to calculate the under-floor to
wayside transfer function from the measured data. For the purpose of this study, an
approach similar to that outlined in [5, 6] was used where a model of the noise from
each vehicle was fitted to the time history of the full train passby to determine the
contribution of the Mk 3 coach to the total passby SEL of the complete rake of the
NMT. Each bogie is modelled as a dipole source, such that the NMT can be modelled
as a distribution of 14 dipoles. A similar approach was first used by Peters [9]. This
model is used to simulate a time history of the passby of the NMT at a reference
distance of 25 m. Once a best fit is achieved, the SEL of one Mk 3 coach can be
obtained by integrating the area under the graph representing the noise from two
bogies. The scaling of SEL with speed was assumed to follow a 20 log V relationship,
where V is the train speed. Fig. 3 shows an example of the dipole model fitted to the
time history of the NMT passby at one of the measurement sites.
Investigating the Effects of a Network-Wide Rail Grinding Strategy 373
90
85
LAeq 64ms, dB
80
75
70
10 11 12 13 14 15 16
time, s
Fig. 3. Example of the dipole model fitted to the time history of the NMT passby. ····, Mk3 coach
bogies (predicted); −·−; power car bogies (predicted); − −, single Mk3 coach (predicted); −−, complete
train passby (predicted); −× −; complete train passby (measured).
The under-floor to wayside transfer function recorded on different weeks had changed
by nearly 5 dB within two weeks but was consistent to approximately 1 dB within the
same measurement day. This suggests that the gain of the NoiseMon system is
“drifting” over different days or long periods. It was not possible to access the
NoiseMon system within this study to investigate its calibration as the NMT is in
constant use. For the purposes of this study the drift in calibration of the NoiseMon
system has been assumed to be negligible during the same measurement day, although
this could not be confirmed. This has limited the ATQ derivation to data captured on
the ECML and WCML on a single day.
measurement site; and rail roughness levels are expected to be higher on the Up line
than on the Down line at the roughness measurement site. The rail roughness was
measured with a Rail Measurement Corrugation Analysis Trolley. Measurements were
undertaken in accordance with the methodology specified by EN 15610:2009 Railway
applications— Noise emission— Rail roughness measurement related to rolling noise
generation [10]. The measured average rail roughness levels for both Up and Down
lines are presented in Fig. 5.
90 90
Noise data (dB)
88 88
N R N R
86 86
84 84
150 150.5 151 150 150.5 151
Mileage Mileage
Fig. 4. NoiseMon data from the NMT at the noise and roughness measurements sites: Down line (left);
Up line (right); noise measurement location (N); roughness measurement location (R)
40
Rossington (Down London)
Rossington (Up London)
Roughness level [dB re 1 μ m]
30
Average Rail Roughness UK
20 TSI Rail Roughness
Typical wheel roughness
10
−10
−20
−1 −2
10 10
one−third octave band wavelength [m]
Fig. 5. Rail roughness at Rossington test site compared with the TSI limit values, the average GB
roughness measured in the 1990s [11] and a typical disc-braked wheel from [12]
The levels are compared with an average GB rail roughness derived from
measurements in the 1990s [11], and also the TSI rail roughness limit spectrum for
reference track for pass-by noise tests. On the Down line, rail roughness levels are
lower than the TSI limit values except for the 6.3 cm band where distinct corrugation
has been measured. The levels are also lower than the 1990’s average GB roughness in
all bands. Rail roughness levels are higher on the Up line than on the Down line. This is
consistent with data measured by the NoiseMon system and presented in Fig. 4. The
levels exceed the TSI limit for wavelengths greater than 3.15 cm. At shorter
wavelengths the measured levels are below the TSI limit. In the wavelength range 31.5
to 5 cm the rail roughness exceeds the 1990’s average GB roughness.
Investigating the Effects of a Network-Wide Rail Grinding Strategy 375
To understand if the measured levels are consistent with the noise data measured at
Rossington a prediction of the relative level of wayside noise that would be expected
for two rail roughness profiles was carried out. This was done by, firstly, calculating the
combined wheel-rail roughness spectrum for a typical disc-braked wheel (from [12])
with the rail roughness measured on the Down line at Rossington and a typical
disc-braked wheel with the TSI roughness limit level for passby noise measurement,
then calculating the frequency spectrum for the combined wheel-rail roughness for a
train speed of 182 km/h and, finally, calculating the overall A-weighted level of each of
the combinations. The results of the analysis are shown in Table 1 and are normalised
to the combination of the typical disc-braked wheel with the TSI limit values. The
wayside A-weighted noise levels at the Rossington site would be expected to be 1.6 dB
less than at a site with a rail roughness equal to the TSI limit values. In [7] a similar
approach was used to conclude that wayside noise levels on track with the rail
roughness implicit in CRN would be expected to be 1.7 dB higher than on a track with
‘typical’ wheels and rail roughness similar to the TSI limit values. Hence, wayside
noise levels at Rossington for smooth-wheeled stock are expected to be 3.3 dB less than
at a track with rail roughness similar to that which is implicit in CRN. At the
Rossington measurement site the SEL of the single Mk 3 coach was measured at
80.2 dB. The SEL of a single Mk 3 coach at 200 km/h predicted by CRN is 83.2 dB,
meaning our measured SEL was 3 dB less than would be predicted by CRN. Therefore
the measured roughness data and noise data were consistent.
Table 1. Summary of relative A-weighted roughness levels calculated for three combinations of
disc-braked wheel and rail roughness
4 Conclusion
This study provides strong evidence that the Network Rail grinding strategy introduced
from 2002 has resulted in a significant improvement in Acoustic Track Quality (ATQ)
across the GB network. A GB network average ATQ value of -4 dB has been derived
based on a survey of 1100 km of track on Great Britain’s East Coast and West Coast
Mainlines. This indicates a very large reduction of 8 dB relative to 2004. It also
suggests that the majority of GB’s East and West Coast Mainlines have rails with
roughness levels below the TSI roughness limit. Whilst these results are very
interesting, further work is considered necessary to better understand and quantify the
significance of the apparent drift in NoiseMon data. The results of the study show how
a maintenance rail grinding strategy can reduce wayside noise levels across large parts
of a railway network. While the measure of track quality used to evaluate the benefits of
376 N.J. Craven et al.
References
[1] Craven, N., et al.: Responding to the environmental noise directive by demonstrating the
benefits of rail grinding on the Great Britain’s railway network. Proc of the IMechE, Part F:
Journal of Rail and Rapid Transit 227(6), 668–676 (2013)
[2] The Department for Transport. Calculation of Railway Noise 1995. HMSO (1995)
[3] Thompson, D.J.: Railway noise and vibration: Mechanisms, modelling and means of
control. Elsevier Ltd. (2009) ISBN-13: 978-0-08-045147-3
[4] Schulte-Werning, B., et al.: Headed for the low-noise railway: the DB noise reduction
research programme. In: Proc. WCRR 2001 (2001)
[5] Hardy, A.E.J., Jones, R.R.K.: Rail and wheel roughness – implications for noise mapping
based on the Calculation of Railway Noise procedure. Defra, Report Number
AEATR-PC&E-2003-002 (2004)
[6] Hardy, A.E.J., Jones, R.R.K., Turner, S.: The influence of real-world rail head roughness
on railway noise prediction. JSV 293, 965–974 (2006)
[7] Jones, R.R.K.: Trends in GB rolling stock noise levels. Rail Safety and Standards Board,
Report Number DeltaRail-ES-2009-005 (2009)
[8] 2011/229/EU: Commission Decision of 4 April 2011 concerning the technical
specifications of interoperability relating to the subsystem ‘rolling stock noise’ of the
trans-European conventional rail system (notified under document C(2011) 658) (2011)
[9] Peters, S.: The prediction of railway noise profiles. Journal of Sound and Vibration 32(1),
87–99 (1974)
[10] EN 15610:2009 Railway applications— Noise emission— Rail roughness measurement
related to rolling noise generation
[11] Hardy, A.E.J.: Draft proposal for noise measurement standard for ERRI committee C163.
Report RR-SPS-97-012 of AEAT Rail Ltd. Published through European Rail Research
Institute (1997)
[12] Dings, P.C., Dittrich, M.G.: Roughness on Dutch railway wheels and rails. JSV 193,
103–112 (1996)
Acoustic and Dynamic Characteristics of a Complex
Urban Turnout Using Fibre-Reinforced Foamed
Urethane (FFU) Bearers
S. Kaewunruen
1 Introduction
A railway turnout is a special track system used to divert a train from a particular
direction or a particular track onto other directions or other tracks. A turnout is the
structural grillage system that consists of steel rails, points (or ‘switches’), crossings
(special track components, also known as a ‘frog’), steel plates, rubber pads, insulators,
fasteners, screw spikes, beam bearers (either timber or concrete), ballast and formation,
as shown in Fig. 1 [1]. Currently, the procurement for high-quality long timber bearers
used in complex turnout systems has been very difficult for construction and renewal
processes in Australia. Many problems with long timber turnout bearers (>4m) also
include localised weakness, large deformation, warping or unstable dimensions that
can easily cause obstructions during the turnout assembly resulting in a poor geometry
of new turnouts. Then, the wheel/rail interaction (see Fig. 2) over such poor short-pitch
irregularity induces impact forces and vibration that exacerbate the condition and
undermine the service life of turnout components and the integrity of the turnout
system as a whole [1-5].
Points
Closure Rails
Check Rail
Check Rail
Crossing
Fig. 1. Typical turnout geometry
Nose
Wing rail
Fig. 2. Transfer zone at crossing where a conical wheel traversing a v-crossing (white paint
showing the contact band) and running over a dip angle inducing impact force [5]
Acoustic and Dynamic Characteristics of a Complex Urban Turnout Using FFU Bearers 379
The difficulty to obtain high-quality timbers has led to two possible alternatives in
practice: first, to use the concrete long bearers with splice plates; second, to use
alternative materials (e.g. Fibre-reinforced Foamed Urethane or so-called FFU;
composite materials, plastic rubber materials, etc.) with the similar characteristics as
timber. A critical review has suggested a field trial of FFU material because of its
high-impact attenuation, high damping property, high UV resistance, and long service
life. As a result, a complex turnout junction with aged timber bearers at Hornsby NSW
Australia has been renewed in 2010 using FFU material. There were five stages of
construction: note that the first turnout was installed in October 2010 and the double
slips were in installed late June 2011. Also, due to the light weight of FFU bearers, a
special arrangement was designed to maintain lateral stability of the turnouts [6].
Table 1. Basic properties of FFU material in comparison with timber bearers [6-9]
Properties Timber Birch FFU bearers3 [8-9]
bearers1 bearers2 New After 10 After 15 After 30
years years years
Service life (years) 5-10 5-10 50 40 35 20
Density (kg/m3) 1050 - 750 740 740 740 740
1120
Bending strength 65 80 142 125 131 116
(MPa) > 70
Vertical compression 60 40 58 66 63 55
strength (MPa) > 40
Shear strength (MPa) 6.1 12 10 9.5 9.6 7
>7
Elastic modulus 16,000 7100 8100 8044 8788 8414
(MPa) > 6000
1
Timber bearer properties are derived from AS1720 Strength Group 2 [6]
2
Birch timber bearer properties are derived from the technical datasheet [6]
3
More detailed engineering properties are available in Ref [9]
This paper focuses on the acoustic and dynamic performance of the alternative FFU
material as a like-for-like replacement of timber bearers. This study involves the
inspection, train-track interaction, sound pressure and vibration measurements of the
double slips, and benchmarking with other available field data [6-9].
FFU bearers are made of continuous glass fibre reinforced rigid polyurethane foam.
The foam has advantages over plastic and wood, e.g. durability and corrosion
resistance, electrical insulation, heat resistance, light-weight and strength, and good
fabrication/assembly/coating. The high damping characteristic of FFU bearers would
be beneficial to the impact and vibration absorption in turnout crossings and supporting
components. Fundamental engineering properties of the FFU material are tabulated in
Table 1 [5, 8]. In the field installation, FFU bearers with the cross section of 250 mm x
180 mm have been utilised throughout.
380 S. Kaewunruen
3 Trial Site
Because of the complexity of the junction, a set of urban turnouts has been chosen for
this trial. The trial site is at Hornsby, New South Wales. The junction comprises of
three set of turnouts, two sets of single slips and a set of double slips. In this study, the
double slips as shown in Fig. 3 were chosen for acoustic and vibration measurements.
k-crossings
Up Down
Fig. 4 shows the test site at Hornsby (Northern line of RailCorp network). The diamond
crossings and heel joints create additional impacts. The track configuration consists of
60 kg/m rails, fastening system (e-Clip type), steel plates (no rubber pads), FFU
bearers, ballast bed and formation. The track supports mixed traffic (passenger and
freight trains). The operational speeds are 60 km/h for freight trains and 80 km/h for
passenger trains, although almost all the electric passenger trains stop at Hornsby
station. The axle load of freight trains could be up to 25 t.
4.2 Instrumentation
Accelerometers were installed at the rail web, base plates and bearers at a k-crossing
and at the interface between FFU bearers and concrete sleepers as shown in Fig. 5. A
high speed camera was used to read the dynamic displacement of the turnout. A sound
level meter was installed at 7.5 m from the nearest k-crossing of the diamond. A train
speed radar was installed at a nearby overhead wiring mast.
Acoustic and Dynamic Characteristics of a Complex Urban Turnout Using FFU Bearers 381
Scales for
displacement
Accelerometers
vibration over the double-slips section and Fig. 8 shows the benchmarking with other
field measurements. The results confirm that relatively the dynamic characteristics of
FFU bearers are equivalent to timber bearers [10].
a) at k-crossing
Rigid body
30
Undulation due to unsettlement of brand new turnout
20
15
10
5
Before: 03-06-11
After: 11-05-12
0
33.95 34 34.05 34.1
Distance (KM:m)
160
Hornsby V (FFU)
Flemington V1 (timber)
140 Bondi V (concrete slab)
Glenfield Swing nose (concrete)
Average Peak Vibration, g
80
60
40
20
0
0 20 40 60 80
Average train speed, km/h
35 14
30 12
25 y = 0.4861x y = 0.1942x
10
R2 = 0.8648 R2 = 0.8159
P1 Vibration, g
P2 Vibration, g
20 8
15 6
10 4
Hornsby V (FFU)
Hornsby V (FFU)
Flemington V1 (timber)
5 Bondi V (concrete slab) Flemington V1 (timber)
2 Bondi V (concrete slab)
Glenfield Swing nose (concrete)
Linear (Hornsby V (FFU)) Glenfield Swing nose (concrete)
0 Linear (Hornsby V (FFU))
0
0 20 40 60 80 0 20 40 60 80
Average train speed, km/h Average train speed, km/h
Fig. 8. Vibration data benchmarking (top: overall; bottom left: P1 – high frequency first peak of the
force spectrum; and bottom right: P2 – moderate-frequency second peak of the force spectrum)
6 Conclusions
Based on the condition inspection and vibration measurements, it is found that FFU
material has equivalent acoustic and dynamic performance to timber bearers while
lasting longer. Also, FFU bearers perform well in P1 region but not very well in P2
band. This is because the impact excitation in P1/P2 frequency band could excite the
384 S. Kaewunruen
References
[1] Andersson, C., Dahlberg, T.: Wheel/rail impacts at a railway turnout crossing. Proceedings
of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit 212F,
135–146 (2003)
[2] Remennikov, A.M., Kaewunruen, S.: A review of loading conditions for railway track
structures due to train and track vertical interaction. Structural Control and Health
Monitoring 15(2), 207–234 (2007)
[3] Kaewunruen, S., Remennikov, A.M.: Dynamic flexural influence on a railway concrete
sleeper in track system due to a single wheel impact. Eng Failure Analysis 16(3), 705–712
(2009)
[4] Kaewunruen, S.: Effectiveness of using elastomeric pads to mitigate impact vibration at an
urban turnout crossing. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B.,
Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 357–365. Springer,
Heidelberg (2012)
[5] Kaewunruen, S.: Review of alternative fibre-reinforced foamed urethane (FFU) material
for timber-replacement turnout bearers. Technical Report TR162, RailCorp – Track
Engineering, Sydney NSW, 14 p. (2009)
[6] Kaewunruen, S.: In-situ performance of alternative fibre-reinforced foamed urethane
(FFU) material for timber-replacement turnout bearers. Technical Report TR188, RailCorp
– Track Engineering, Sydney NSW, 86 p. (2011)
[7] Kaewunruen, S.: Vertical and lateral stability performance of alternative fibre-reinforced
foamed urethane (FFU) material for timber-replacement turnout bearers. Technical Report
TR197, RailCorp – Track Engineering, Sydney NSW, 40 p. (2012)
[8] Sekisui Co.: Engineering Properties of FFU materials, Tokyo Japan (2012)
[9] Kaewunruen, S.: Monitoring in-service performance of fibre-reinforced foamed urethane
sleepers/bearers in railway urban turnout systems. Structural Monitoring and
Maintenance 1(1), 131–157 (2014)
[10] RailCorp Rolling Stock Engineering: Minimum operating standards for rolling stock,
Sydney, Australia (2012)
Ensuring Acceptable Vibration Levels in Listed
Buildings by Means of Precise
Vibration Measurements and Highly-Efficient Floating
Slab Track
T. Jaquet
Summary. This paper describes studies associated with planning for a new
Metro Line in close vicinity to heritage buildings in the historic heart of Beijing
requiring strict compliance with vibration criteria. To minimize uncertainties,
extensive measurements were performed, including borehole tests to assess local
soil properties. Furthermore, the influence of several key design factors on the
final results is analyzed. The importance of long-term design and installation
experience of highly efficient FST is illustrated
1 Introduction
Traffic induced vibrations and their impact on listed building became an important topic
in the recent years. Especially for these buildings, there is a risk of micro damages which
could be dangerous over time. To prevent this damage scenario vibration surveys have
been conducted to assess the impact. Due to the various excitation sources the range of
frequencies of interest is very large. This report considers the planned alignment of Metro
line 6 and line 8 in Beijing. The planning envisages an alignment below historical
buildings in Beijing City Centre. The Chinese standard considers very strictly the risk of
micro damages for historical buildings. Therefore, for ancient buildings very severe
vibration criteria are defined in GB/T 50452-2008 “Technical specification for protection
of historic buildings against man-made vibration”. This was the motivation to perform
extensive vibration measurements along the planned alignment.
2 Vibration Requirements
Historical buildings (e.g. museums, temples) require low vibration levels to prevent
negative effects on building use. Several international standards define vibration
criteria for these types of buildings. The Chinese standard: GB/T 50452-2008,
Technical specification for protection of historic buildings against man-made vibration,
provides limits for permissible vibration depending on the structure and the cultural
importance. The limits will be strictly observed.
Vibrations are caused by the large forces between the wheel and the rail. These forces
fluctuate in response to wheel and rail roughness over a wide range of frequencies. The
vibrations propagate from the track through the soil. The wave energy becomes less due
firstly to the geometric dispersion and secondly to mitigation through soil damping
(energy absorption). The main frequency band of train excitation is between 10 Hz and
150 Hz with a maximum at approx. 50 Hz. In a building this frequency is noticeable as
a rumbling noise which is radiated from the walls and floors. This kind of noise is
called ground borne noise.
Fig. 1. Train excitation and the effect on the environment (Building Response)
The principle procedure of a vibration prognosis can be divided into three sections:
• Excitation source
• Transfer system (Propagation)
• Building response
First of all the planned alignment was checked regarding probable vibration impacts.
These were mainly areas where the planned tunnel will be directly below historical
buildings. For these situations a detailed vibration analysis was conducted. At these
sites soil tests were performed to check the local subsoil conditions. Most of the project
survey time was needed to get the permissions for drilling the holes.
Ensuring Acceptable Vibration Levels in Listed Buildings by Means of Precise Vibration 387
4 Analysed Alignment
First different tunnel structures with and without Floating slab tracks (FST) at the metro
line 4 and line 10 were surveyed to receive true isolation efficiencies of installed FST.
The support frequencies of these FST are f = 7.5 Hz and 6 Hz. The installation of these
FST was required due to close vicinity of institute buildings of a university and R+D
buildings for the micro technology. The distance T/R to the ground level was approx.
15 m. In addition to insertion loss measurements the structural dynamics of the FST
system itself were analysed. Also the impact of the tunnel structure on the vibration
level was considered. It is well known that tuning of the tunnel stiffness influences the
transfer mobility from the track to the environmental buildings. For this purpose
dynamic stiffness measurements of singular components were carried out.
( )=| ( )| = ( − ∗ (2 ) ) + (2 )² (4.1)
The insertion loss was calculated from the pass-by vibration level of two adjacent
sections with and without FST. The insertion loss curve shows also a minimum at 8 Hz,
which refers to the system frequency. Beyond 12.5 Hz the isolation rises up to 30 dB at
160 Hz. The mitigation of the FST was also revealed by measurements performed at the
ground above the tunnel. The vibration level on the ground with FST is mainly caused
by the road traffic. The section of line with standard slab track clearly shows that the
main vibration level was generated by the metro pass-by.
388 T. Jaquet
Fig. 4. Vibration level at the tunnel wall for the standard slab track section and FST
Metro
0,10
Metro Metro
0,1 Slab Track
0,05
Geschwindigkeit / mm/s
Geschwindigkeit / mm/s
0,00 0,0
-0,05
-0,1
Road Traffic
-0,10 FST
-0,2
-0,15
05:10:00 05:15:00 05:20:00 08:20:00 08:30:00 08:40:00
Zeit / s Zeit / s
5 Subway Testing
The local soil conditions have an important influence on the wave propagation and
therefore on the vibration transmission. A rough prediction of the transmission leads to
major uncertainties due to difficult propagation in a layered soil. The different wave
modes and interaction between different structures are presented in Fig. 6.
The test procedure basically consists of dropping a heavy weight (impact hammer)
in a borehole and recording the response at several distances from the impact. The
impact hammer has the advantage of a clear load introduction into the subsoil. The
impact produces a sharp peak which leads to vibration at a broad frequency
range (1-400 Hz). The impact was carried through a steel rod to the bottom of the hole.
A load cell, installed at the bottom of the borehole, delivers the impact force.
Simultaneously to the force measurement the vibration level is measured in the
environment. The ratio between the vibration level and the exciting force leads to
mobility Msoil( f ) = vi ( f ) . The principle application is shown in Fig.6.
Fi ( f )
Fig. 7. shows a typical result of the wave propagation. Clearly visible is the
mitigation of the mobility over length. Up to distances of 16 m and frequencies lower
than 100 Hz no fundamental mitigation was recorded. For greater distances a
mitigation of higher frequencies was measured. Lower frequencies show no
attenuation, which means that for distances up to 16 m the metro vibration were not
reduced. Mitigation measures were needed. The borehole tests need a low ambient
vibration level. Therefore, it is recommended to carry out these measurements at a time
of day with less road traffic. Also the coherence between excitation source and
vibration has to be checked.
8m 16m 32m
Sensor 4
Loadcell
Beijing, Line 8
Transfer Mobility, vertical
Excitation in the borehole No 1
1,0E-04
8m
1,0E-05
16 m
32 m
1,0E-06
mobility [m/s/N]
1,0E-07
1,0E-08
1,0E-09
1,0E-10
1,0E-11
1 10 100 1000
Frequency [Hz]
The force spectrum of the rolling stock was measured at two sites with two independent
procedures. Directly measuring of the spectrum by force transducers integrated in the
spring elements and an indirect method by measuring the stiffness and the vibration
level. Both methods have led to the same result.
2010-02-01_09-20-16 No34_SkaAdd/K:1/S:1/Force Sensor K1
30
20
Force / kN
10
-10
0 10 20 30
Zeit / s
7 Vibration Prediction
Thus a prediction with sufficient safety was conducted and therefore appropriate
mitigation to be within the strict limits could be guaranteed.
with
Ftrain(f) = Force Spectra / Force Density of the track vehicle
M(f) = Transfer Mobility from the track to the ground level
Clinesource(f) = Line source factor of the train
Cbuilding(f) = Adjustments to account for ground to building foundation
interaction
Beijing Line 8
Expected vibration level on the ground (z-direction)
( Source: Metro Line, Bored Tunnel incl. FST 7.5 Hz, Depth to t/r: 20 m)
1
0,1
velocity [mm/s]
0,01
vibration level at 0 m
0,001
vibration level at 8 m
vibration level at 16 m
0,0001
vibration level at 32 m
0,00001
1 10 100 1000
frequency [Hz]
The installation of a 7.5 Hz Floating Slab Track enables the compliance of the
criteria even for close distances to the listed buildings. An alignment in a greater depth
was not necessary, neither was a stiffer tunnel structure needed. The suggested bored
tunnel structure could be retained and is sufficient to accommodate the recommended
FST as shown in Fig 10.
References
[1] GB/T 50452-2008 Technical specification for protection of historic buildings against
man-made vibration, China
[2] Transit Noise and vibration impact assessment (May 2006)
[3] Wagner, H.-G., Jaquet, T.: Introduction of the GERB Floating Slab Track system applied to
Taipei Metro Xinji Line. In: MRT Conference, Taipei (2009)
[4] Jaquet, T., Garburg, R.: Measurements and investigations at the Floating-Track-Bed System
in the North-South Tunnel in Berlin. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E.,
Hanson, C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 99, pp. 150–157. Springer,
Heidelberg (2008)
An Assessment of the Effectiveness of Replacing
Slab Track to Control Groundborne Noise
and Vibration in Buildings above an Existing
Railway Tunnel
1 Introduction
Disturbance from noise from trains operating in tunnels can occur when vibration
generated by wheel-rail interaction is transmitted into the ground and then into
surrounding buildings where it is felt as perceptible vibration or re-radiated and heard
as groundborne noise [1]. The primary source of vibration generation at the wheel-rail
interface is the combined roughness of the wheels and rails. This means that poorly
maintained wheels or severely corrugated rails can lead to very high levels
groundborne noise. It is possible to reduce the vibration generated at the wheel rail
interface by reducing the roughness of the rails through grinding where corrugation has
occurred [1]. Without frequent grinding the rate of roughness growth at a site that is
prone to corrugation can mean that the benefit of rail grinding is short lived. There is
some evidence that the use of resilient elements in the track can reduce the rate of
roughness growth [2]. It is also possible to isolate the transmission of vibration into the
ground with the use of resilient elements in the track [3]. This means that mitigation
measures employed to isolate vibration from the ground, such as highly resilient
baseplates, may also reduce the rail roughness growth rate at a site that is prone to
corrugation. Modern heat treated rails have been shown to reduce the rate of
corrugation growth by one third and also change the wavelength of corrugation that
occurs [4]. It is also known that the occurrence of corrugation is highly dependent upon
the initial roughness profile of the rail [5]. Corrugation is not always removed from the
rail through grinding hence it is possible that the initial conditions required for
corrugation to develop remain following grinding works. Re-railing with new rails will
remove corrugation. For these reasons track replacement to control groundborne noise
through isolation of vibration will also reduce the vibration generated by rail roughness
and perhaps reduce or eliminate the growth of corrugation at the site.
2 Background
Thameslink is a project to increase capacity on the commuter rail route from north to
south through central London. The route runs through the King's Cross tunnel, above
which is situated the St Pancras Chambers, a grade 1 listed building completed in 1876.
It has recently been returned to its former use as a high quality hotel, after being derelict
for a number of years. Through the planning process for Thameslink, Network Rail
committed to a programme of track refurbishment to reduce the groundborne noise
experienced within the St Pancras Chambers. The railway tunnel is at a depth of
approximately 7 m, includes twin track on tight radii (185 m) and is structurally
coupled to the foundations of the Chambers. Prior to replacement, the existing paved
concrete tracks (PACT) provided little isolation of train-induced vibration and were
prone to the growth of corrugation. Consequently, rooms in the hotel were exposed to
groundborne noise levels of up to 55 dB LAmax,slow. A study carried out for Network Rail
[6] used models of the train-track and ground [7], and measured groundborne noise
spectra to predict the reduction in groundborne noise that could be achieved with the
installation of various proprietary track systems. The study predicted that the Pandrol
Vanguard fastening system would reduce the A-weighted groundborne noise level in
the hotel by 10-15 dB (depending on the dynamic stiffness of the selected system). A
similar study later in the design process [8] provided similar conclusions. Work was
also commissioned to investigate the effect of wheel and rail condition on groundborne
noise levels. The study [9] included measurement of rail roughness and groundborne
noise in the Chambers before and after rail grinding and found a reduction in the
A-weighted groundborne noise level of approximately 6dB. Between September 2010
and January 2011 a 320 m long section of the twin track was re-railed with micro head
hardened rail and the Pandrol Vanguard with a dynamic stiffness of approximately 8
MN/m per fastener installed. From the previous studies it was expected that the benefit
of this would be a reduction in groundborne noise of approximately 16-20 dB(A).
It was not possible to determine if the corrugation present at the site would return after
the works and hence the longevity of the benefits could not be quantified. This paper
describes measurements undertaken to quantify the overall benefit of the works and
their effectiveness over a period of 18 months.
An Assessment of the Effectiveness of Replacing Slab Track 395
Fig. 1. St Pancras Chambers (left); existing PACT system in the tunnel below the hotel (right)
3 Measurements
Measurements of rail roughness on a 300 m sample of both tracks were carried out prior
to re-railing and four weeks after re-railing. Following re-railing and the installation of
the Pandrol Vanguard rail fastening system, the rail roughness on the same section of
track were measured a further three times, to monitor the change in the railhead
condition over an 18 month period. The measurements and analysis of the rail
roughness data were carried out in accordance with the methodology specified by EN
15610:2009 [10].
Measurements of the groundborne noise and vibration were carried out in the hotel in
rooms closest to the tunnels over an extended period: immediately before re-railing;
following re-railing; following the installation of the Pandrol Vanguard track system
and intermittently for a period of 18 months after the works. Noise measurements were
made with a Brüel & Kjær 2260 sound level meter and a Rion DA-20 digital signal
recorder. Simultaneous vibration measurements were made with a PCB 356 B18
tri-axial accelerometer and Rion DA-20 digital signal recorder.
4 Results
between 0.125 m and 0.16 m. This corresponds to excitation in the frequency range 80
Hz to 100 Hz. Following re-railing roughness levels were reduced by 24 dB and 25 dB
in the 0.125 m and 0.16 m wavelength bands respectively. Also shown are the rail
roughness levels on the same track before and after rail grinding in 2006 from [5]. Prior
to re-railing in 2011 and prior to grinding in 2006 the rail was in a similar condition.
The reduction in the roughness level at the corrugation wavelengths following grinding
was similar to the reduction that was seen following re-railing, however, the
corrugation in the 0.1 m and 0.125 m wavelengths was not completely ground out in
2006.
Frequency at 13ms−1 (Hz)
12.5 25 63 125 250 630 1250 2500
50
40
Rail roughness (dB re 1µm)
30
20
10
−10
−20
0 −1 −2
10 10 10
Wavelength (m)
Fig. 2. Average rail roughness in King’s Cross Tunnel. −−, before re-railing (2010); ····, one
month after re-railing (2011); −·−, before rail grinding (2006); − −, after rail grinding (2006).
The average train event groundborne noise and A-weighted vibration levels are
presented in Table 1. Re-railing has resulted in a reduction in noise and A-weighted
vibration of about 10 dB. Some of this reduction may have resulted from the removal of
sections of the continuous rail pad to allow installation of the baseplates, thus creating
temporary discrete rail supports during the installation of the rail. This will have
reduced the stiffness of the rail support and increased the vibration isolation provided
by the track. In [9] a reduction of 6 dB in the A-weighted noise level was measured for
a similar reduction in roughness. Following the installation of the Pandrol Vanguard
system a further reduction of approximately 8 dB was seen in the A-weighted vibration
level. The apparent reduction in groundborne noise was only 3 dB as the train event
noise levels could not be measured above the ambient noise level in the hotel room.
Overall the works achieved a reduction of 13 dB and 19 dB in A-weighted groundborne
noise and vibration respectively. Also shown in Table 1 are measurements made in the
hotel in April 2013, 18 months after completion of the works. There has been no
significant change in the noise and vibration level measured in the hotel.
An Assessment of the Effectiveness of Replacing Slab Track 397
Table 1. The average train event groundborne noise and A-weighted vibration levels measured in
the hotel before and after the works
90
−6
80
Vibration level (dB re 1x10
70
60
50
40
1 2
10 10
One−third octave band centre frequency (Hz)
Fig. 3. Average Lmax,Slow vibration spectra measured in the hotel. ····, before re-railing; − −,
1 month after re-railing, −−, 1 month after installation of Pandrol Vanguard.
The one-third octave band vibration spectra measured in the hotel before and after
the works are presented in Fig. 3. The results are presented with frequency on the lower
axis and the corresponding roughness excitation wavelength on the upper axis. Before
re-railing the vibration is highest in the frequency range 50 Hz to 125 Hz. This
corresponds well to the corrugation wavelengths seen in Fig. 2. Following re-railing the
vibration level is reduced significantly at all frequencies above 50 Hz. The reduction is
not as large as would be expected at the corrugation wavelengths from the change in
roughness. It is possible that this is because the roughness of the wheels now dominates
the combined roughness spectrum, as was also suggested by the earlier study which
considered the effect of rail grinding [9]. Following the installation of the Pandrol
Vanguard system the vibration is reduced at all frequencies above 30 Hz.
One month after re-railing, a running band approximately 35 mm wide was observed on
all rails. The roughness was sampled on lines centered on 25 mm and 40 mm from
398 O.G. Bewes, L.J. Jakielaszek, and M.L. Richardson
gauge side respectively. The lateral sampling locations were retained for all further
measurements. The roughness measured on both rails of the down line at the two lateral
sampling locations, one, seven, eleven and nineteen months after re-railing are
presented in Fig. 4.
50 50
a b
40 40
30 30
20 20
10 10
0 0
−10 −10
−20 −20
0 −1 −2 0 −1 −2
10 10 10 10 10 10
Wavelength (m) Wavelength (m)
50 50
c d
40 40
Rail roughness (dB re 1µm)
30 30
20 20
10 10
0 0
−10 −10
−20 −20
0 −1 −2 0 −1 −2
10 10 10 10 10 10
Wavelength (m) Wavelength (m)
Fig. 4. Rail roughness following re-railing: (a) Low rail 20 mm lateral sample, (b) Low rail 40 mm
lateral sample, (c) High rail 20 mm lateral sample, (d) High rail 40mm lateral sample. −−, 1 month after
re-railing; − −, 7 months after re-railing; −·−, 11 months after re-railing; ····, 19 months after re-railing.
Looking first at the low rail (Fig. 4(a) and (b)), for wavelengths less than 0.06 m the
roughness for the 40 mm lateral sample was slightly higher than for the 20 mm sample
one month after re-railing. Following this measurement, the roughness at both sample
locations progressively reduced to approximately the same level. A reduction of up to
10 dB was measured in this wavelength range during the 19 month monitoring period.
Seven months after re-railing the highest levels of roughness were measured on the
40 mm sample. This corresponded to the observation of wear on large parts of the
measurement section. Because the new rails were not ground immediately following
installation, the apparent wear was thought to be the remnants of mill scale [9]. The
wear was not present 11 months after re-railing. Similar behavior was measured on the
high rail (Fig. 4(c) and (d)) for all wavelengths less than 0.1 m. Nineteen months after
re-railing, the roughness level in this range was significantly lower than 1 month after
re-railing. In the wavelength range 0.06 m to 0.3 m on the high rail, very little change in
roughness was measured during the monitoring period. In the same range on the low
An Assessment of the Effectiveness of Replacing Slab Track 399
rail, some corrugation was measured in the 0.125 m wavelength band seven and eleven
months after re-railing. The corrugation was not present in the measurements nineteen
months after re-railing. Of most importance to this study, it can be seen that on all
rails there has been no significant increase in rail roughness wavelengths in the range
0.125 m to 0.16 m where corrugation developed prior to the works. In the wavelength
range 0.3 m to 2 m a large variation in the roughness was measured. In the 2m band, the
change in roughness was up to approximately 20 dB. Seven months after re-railing, a
large reduction in the roughness was seen in this range. It is possible that this was a
result of the track re-alignment that was carried out four months after re-railing when
the Pandrol Vanguard had been fully installed. Three of the samples have shown an
increase in roughness in this range in subsequent measurements.
Table 2 shows the rail roughness levels at the corrugation wavelengths from Fig. 1
measured shortly after grinding in August 2006 [9] and prior to the works described in
this study in July 2011. Over this period of 60 months the roughness increased by
between 13 and 28 dB. Also shown in Table 2 is an estimated increase in roughness in
these wavelengths that would have occurred during the first 18 months after rail
grinding. An increase of 4 to 8 dB over 18 months is predicted, assuming a linear
increase in roughness level. Unfortunately there is no intermediate data available to
confirm the validity of this assumption at this site. It is possible that the corrugation
prior to the works did not develop until later in the 60 month period. However two other
studies at sites prone to corrugation have shown that the growth of corrugation is
obvious soon after rail grinding [11] or re-railing has occurred [2].
Table 2. The change in rail roughness level in the tunnels prior to track replacement at the
wavelengths of corrugation – From Fig. 1
5 Conclusion
Measurements of noise and vibration in the hotel have shown that the overall
immediate benefit of the works described here are a reduction in the average
A-weighted train event maximum vibration level of 19 dB. It is estimated that 6 dB of
400 O.G. Bewes, L.J. Jakielaszek, and M.L. Richardson
the reduction was as a result of the re-railing of the tracks in the tunnels. Noise and
vibration measurements taken 18 months after the works were completed showed no
significant change in the noise and vibration in the hotel. Measurements of rail
roughness have shown no significant change in rail roughness in the first 18 months
since the works were completed and no recurrence of corrugation that was seen in the
0.1 m to 0.16 m wavelengths prior to the works. The study therefore has shown that
replacement of the existing corrugated PACT track with new micro head hardened rails
and a highly resilient track fastening system has eliminated corrugation growth for 18
months following the works.
References
[1] Nelson, P.M.: Transportation Noise Reference Book. Butterworth & Co. Ltd. (1987)
[2] Vanhonacker, T.: Accurate quantification and follow up of rail corrugation on several rail
transit networks. In: Proc. of 9th International Workshop on Railway Noise (IWRN)
(2007)
[3] Nelson, J.T.: Recent developments in groundborne noise and vibration control. Journal of
Sound and Vibration 193(1), 367–376 (1996)
[4] Heyder, R., Girsch, G.: Testing of HSH rails in high-speed tracks to minimise rail damage.
Wear 258, 1014–1021 (2005)
[5] Schmid, F. (ed.): Wheel-Rail Best Practice Handbook. Birmingham University Press,
Birmingham (2010), A.N. Harris
[6] St Pancras Chambers TI-NV-06-02. Report for Network Rail prepared by Temple (June 07
2007)
[7] Jones, C.J.C.: Groundborne noise from new railway tunnels (Invited Paper). In:
Proceedings of Internoise, Liverpool, UK, Book 1, pp. 421–426 (1996)
[8] Acoustic evaluation of mitigated track systems, Report AAc/211645-03/R01 for Network
Rail prepared by Arup (April 2010)
[9] Assessment of potential benefits of railhead maintenance on groundborne noise levels.
Report for Network Rail prepared by Temple (February 16, 2007)
[10] EN 15610:2009 Railway applications— Noise emission— Rail roughness measurement
related to rolling noise generation
[11] Grassie, S.L.: Rail irregularities, corrugation and acoustic roughness: characteristics
significance and effects of reprofiling. Proc IMechE Part F: J Rail and Rapid
Transit 226(5), 542–557 (2012)
Mitigation Measures against Vibration
for Ballasted Tracks – Optimisation of Sleepers,
Sleeper Pads and the Substructure by Combined
Finite-Element Boundary-Element Calculations
1 Introduction
Under-sleeper pads as a measure to mitigate ground borne vibration have been
investigated in the European research project RIVAS (Railway induced vibration
abatement solutions). Two numerical studies [1, 2] have been prepared for ballasted
and slab tracks. The combined finite-element boundary-element method is described
shortly in this article, followed by the results on the un-isolated track and the isolated
track with under-sleeper pads.
2 Methods
2.1 Finite-Element Boundary-Element Method
The track-soil systems are calculated by the combined finite-element
boundary-element method [3, 4, 5]. The track including the rails, rail pads, sleepers,
under-sleeper pads, and ballast is modeled by the finite element method (Fig. 1)
whereas the homogeneous or layered soil is modeled by the boundary element method.
The dynamic stiffness matrix of the soil is established by using the Green’s functions of
an elastic layered half-space [5, 6]. All calculations (Green’s functions, boundary
matrix and finite element matrices) are performed in the frequency domain. Special
additional methods (within the FEBEM) have been developed for infinite tracks [7]
which are also applied to this study. The track is excited by a dynamic axle load (a pair
of vertical forces, which acts on the rails above the central sleeper).
The results of a FEBEM calculation are the compliance uR/FT of the track and the
mobilities vi/FT of the soil due to a wheelset load FT on the track.
The transfer function between the irregularities s and the force FT on the track can be
calculated as
FT K T KV
=− (2.1)
s K T + KV
FT KT
= (2.2)
FV KT + KV
if the force FV = -KVs is used which is generated by the vehicle and the irregularities in
case of a rigid track.
After the vehicle-track interaction is solved, the ground vibration can be calculated
by multiplying the vehicle-track transfer function and the mobilities of the soil
vi v F v KT
= i T = i (2.3)
FV FT FV FT KT + KV
At the end, two different tracks (isolated and un-isolated track) are compared by
comparing the vibration at the same soil point xi. The velocity ratio vi,I /vi,U of the
isolated and un-isolated track can be built by the transfer functions without specifying
the excitation
vi , I ⎛ vi KT ⎞
⎜⎜ ⎟⎟
vi , I FV ⎝ T T + KV ⎠, I
F K
= = . (2.4)
vi ,U vi ,U ⎛ vi KT ⎞
⎜⎜ ⎟⎟
FV ⎝ FT KT + KV ⎠,U
rails m’R = 2 x 60 kg/m, distance of the sleepers d = 0.6 m, stiffness of the rail pads
kP = 300 x 106 N/m, modulus of elasticity of the sleepers ES = 3 x 1010 N/m2, mass
density of the sleepers ρS = 2.5 x 103 kg/m3, length of the sleepers aS = 2.6 m, height of
the sleepers hS = 0.2 m, width of the sleepers bS = 0.26 m, shear modulus of the ballast
GB = 18 x 107 N/m2, width of the ballast aB = 3.6 - 5.6 m, height of the ballast
hB = 0.3 m, shear modulus of the soil G = 8 x 107 N/m2, shear wave velocity of the soil
vS = 200 m/s, mass density of the soil and ballast ρ = 2 x 103 kg/m3, Poisson’s ratio of
the soil and ballast ν = 0.33, hysteretic damping of the soil and ballast D = 2.5 %
(defined by G = G0(1+i2D)), hysteretic damping of the elastic elements DP = DS = 10 %.
At first, the compliance of the track and the mobilities of the soil for the reference track
are discussed which are shown in Fig. 2a (the triangles) together with the results of tracks
on different soils. The track compliance is almost constant at uR/FT = 3 x 10-9 m/N for all
frequencies. That means a displacement of 0.3 mm under a 100 kN axle load. The
mobilities of the soil (Fig. 2a right) show some characteristic changes in the frequency
range of interest. At first, the soil amplitudes increase with frequency up to 40 Hz.
At higher frequencies, there is a general reduction due to the material damping of the soil,
which has its strongest effect on the far-field points of the soil, and due to the distribution
of the excitation force across the track. The distribution of the force leads also to marked
minimum amplitudes at 63 and 80 Hz where one wavelength of the soil fits into the
sleeper length.
The stiffness of the soil has a strong influence on the compliance of the track for
frequencies up to 64 Hz (Fig. 2a left). A soft soil of G = 2 x 107 N/m2 or vS = 100 m/s
yields a track compliance of almost uR/FT = 10-8 m/N what means 1 mm displacement
under a 100 kN axle load. The influence on the ground vibration is even stronger
(Fig. 2a right). The ground vibration amplitudes are strictly proportional to the inverse
of the shear modulus G -1 in the regularly increasing low-frequency part of the soil
mobilities. At high frequency, the mobilities are rather constant and less regular. The
softest soil may have the lowest amplitudes due to the strong effects of the material
damping and the load distribution across the track.
a)
b)
c)
d)
Fig. 2. Track displacements (left) and ground vibration (ratios) at r = 8 m (right) a) of un-isolated
tracks on different soils of vS = 100, 150, 200, 300 500 m/s, isolated tracks with b)
different sleeper pad stiffnesses kS = 25, 50, 100, 200 kN/mm, c) different sleeper masses
mS = 75, 150, 300, 600 kg, d) different wheelset masses mW = 1000, 1500, 2000,
3000 kg
406 L. Auersch and W. Rücker
The softest under-sleeper pad yields a static track compliance higher than
0.01 mm/kN (1 mm per 100 kN axle load, Fig. 2b). Stiffer under-sleeper pads are below
this value. All under-sleeper pads have track-pad resonances below 160 Hz, starting at
40 Hz for the softest and ending at 100 Hz for the stiffest sleeper pad. Including the
wheelset mass yields a somewhat lower vehicle-track resonance between 25 and 80 Hz
(Fig. 2b right). Above the resonance, a strong decrease of the transfer functions is
observed. The softest under-sleeper pads yield the strongest reduction. The reduction is
quantified by the ground vibration ratios in Figure 2b right. Reductions of up to 0.1 or
even 0.01 are reached at 160 Hz. At 100 Hz, the reductions are between 0.5 and 0.1.
Below 100 Hz, only softer under-sleeper pads yield considerable reductions. Around
the vehicle-track resonance, all pads yield resonance amplifications, the highest
amplifications for the softest sleeper pads.
The sleeper mass has an influence on the high-frequency compliance of the track
(Fig. 2c left). The sleeper-on-pad resonance is shifted from 80 Hz, via 60 and 50 Hz to
less than 40 Hz if the mass is doubled three times. The strong reduction above this
resonance frequency can clearly be seen in the track compliances where the lowest
amplitudes are found for the lowest resonance frequency.
The ground vibration ratios are very similar for all masses. The vehicle-track
resonances are close together at 40 Hz for the light sleepers. Only the highest sleeper
mass shows a lower resonance frequency of 28 Hz and an additional reduction at most
frequencies above 40 Hz.
The wheelset mass has an influence on the vehicle-track interaction but not on the
compliance of the track. The vehicle-track resonance is shifted from 40 Hz for the
standard wheelset of 1500 kg to 32 and 25 Hz for heavier wheelsets of 2000 and 3000 kg
(Fig. 2d). Correspondingly, there is a reduction of the rail response at higher frequencies,
strongest for the heaviest wheelset (Fig. 2d left). That means that all results which are
presented for a wheelset mass of 1500 kg could be modified by using the heavy wheelset
of 3000 kg. The results for the ground vibration ratios are shown on Figure 2d right.
All effects of the heavier wheelsets can also be found in the ground vibration ratios up to
80 Hz. Then the reduction effects disappear, and at frequencies higher than 100 Hz the
reduction is better for lower wheelsets masses. This stems from the fact that the reference
system has considerably higher amplitudes for a low wheelset mass.
4.4 Variation of Other Parameters and Slab Tracks with Wide Sleepers
A similar parameter study has been performed for slab tracks with wide sleepers on
sleeper pads (Fig.3), [2]. Many parameters have been shown to have no influence on
the efficiency of the under-sleeper pads. The main parameter is once again the stiffness
of the under-sleeper pads. The mitigation effect is improved by the higher mass of the
wide sleeper.
Mitigation Measures against Vibration for Ballasted Tracks 407
Fig. 3. Finite element model of the slab track with wide sleepers on soft sleeper pads
5 Conclusion
The ground vibrations, which are generated by trains on different tracks, have been
calculated by finite-element boundary-element models. Tracks with under-sleeper pads
have been investigated in a wide parameter study. The main parameters that influence
the reduction of ground vibration are the stiffness of the under-sleeper pad, the mass
(the width) of the sleeper, and the mass of the wheelset. The softest sleeper pad yields
the best reduction of the ground vibration. As the standard isolated track, a track with
an under-sleeper pad of a stiffness of kS = 50 x 106 N/m has been chosen. The resonance
frequency for this pad stiffness is observed at 35 Hz. The reduction of the ground
vibration is about vi,I /vi,U = 0.15 at 100 Hz. The reduction can be improved by softer
sleeper pads and heavier sleepers.
The influence of other parameters has been examined: stiff or soft rail pads, the
bending stiffness of the rail, the sleeper and the track slab, the stiffness of the ballast,
the sub-soil, and the soil, and the layering of the soil. All these parameters show no or
only a minor influence on the mitigation effect of under-sleeper pads.
Acknowledgements. The BAM work for the RIVAS project has been accomplished by
L. Auersch (numerical studies), E. Knothe, R. Makris and E. Kretzschmar (laboratory
tests), S. Said (field tests), W. Rücker (project leading) and has received funding from
the European Union Seventh Framework Programme under grant agreement n°
265754. The work package 3 “Mitigation measures track” has been organized by E.
Bongini (SNCF), and B. Asmussen is the coordinator of the research project RIVAS.
References
[1] Auersch, L.: Mitigation measures for ballasted tracks - sleepers, sleeper pads and
substructure - Results from the finite-element boundary-element method. Report for
RIVAS Deliverable 3.2, BAM, Berlin, 2012. In: Faure, B., Bongini, E. (eds.) BAM, KUL:
Results of the Parameter Studies and Prioritization for Prototype Construction for Ballasted
Track. Deliverable 3.2 of the European research project RIVAS “Railway Induced
Vibration Abatement Solutions”. International Union of Railways (UIC), Paris (2012)
[2] Auersch, L.: Results of the parameter studies and prioritization for prototype construction
for slab track. Deliverable 3.3 of the European research project RIVAS “Railway induced
vibration abatement solutions”. BAM, Berlin (2012)
408 L. Auersch and W. Rücker
[3] Auersch, L., Schmid, G.: A simple boundary element formulation and its application to
wavefield excited soil-structure interaction. Earthquake Engineering and Structural
Dynamics 19, 931–947 (1990)
[4] Auersch, L.: Dynamics of the railway track and the underlying soil: the boundary-element
solution, theoretical results and their experimental verification. Vehicle System
Dynamics 43, 671–695 (2005)
[5] Auersch, L.: Wave propagation in the elastic half-space due to an interior load and its
application to ground vibration problems and buildings on pile foundations. Soil Dynamics
and Earthquake Engineering 30, 925–936 (2010)
[6] Auersch, L.: Wave propagation in layered soil: theoretical solution in wavenumber domain
and experimental results of hammer and railway traffic excitation. Journal of Sound and
Vibration 173, 233–264 (1994)
[7] Auersch, L.: Dynamic axle loads on tracks with and without ballast mats – numerical
results of three-dimensional vehicle-track-soil models. Journal of Rail and Rapid
Transit 220, 169–183 (2006)
[8] Auersch, L.: The excitation of ground vibration by rail traffic: Theory of vehicle-track-soil
interaction and measurements on high-speed lines. Journal of Sound and Vibration 284,
103–132 (2005)
[9] Auersch, L.: Zur Parametererregung des Rad-Schiene-Systems: Berechnung der
Fahrzeug-Fahrweg-Untergrund-Dynamik und experimentelle Verifikation am
Hochgeschwindigkeitszug Intercity Experimental. Ingenieur-Archiv. 60, 141–156 (1990)
[10] Rücker, W.: Dynamic interaction of a railroad-bed with the subsoil. In: Proc. Conf. Soil
Dynamics and Earthquake Engineering, Southampton, pp. 435–448 (1982)
[11] Rücker, W., Auersch, L., Baeßler, M., Knothe, K., Popp, K., Savidis, S., Schmid, G.: A
comparative study of results from numerical track-subsoil calculations. In: Popp, K.,
Schiehlen, W. (eds.) System Dynamics and Long-Term Behaviour of Railway Vehicles,
Track and Subgrade. LNAM, pp. 471–488. Springer, Berlin (2003)
Innovative Measures for Reducing Noise Radiation
from Steel Railway Bridges
DB Systemtechnik,
Völckerstr.5, 80939 München, Germany
dorothee.stiebel@deutschebahn.com
1 Introduction
Railway bridges produce increased levels of noise radiation because, aside from the
rolling noise, vibrations propagating into the bridge structure lead to a low frequency
noise emission, which people can find particularly annoying. In Fig. 1, the generation
mechanisms of bridge noise are illustrated and typical measurement results for a train
passing a bridge compared to the same train passing the adjacent plain track are
presented. Depending on the construction details and the excitation, annoyance due to
bridge noise can occur between 40 and 500 Hz.
Fig. 1. In the left figure, the noise components of bridge noise and their generation are illustrated;
the right diagramm shows a typical bridge noise spectrum compared to plain track
For noise abatement and noise prevention, the newly defined set of calculation rules
for quantifying the railway noise (“Schall 03 neu”) will come into force soon. As
shown in Fig. 1 right, bridges are characterized by the difference of their overall noise
level to the noise level of the adjacent plain track [1]. For a large number of bridges of
different types, these so called bridge correction factors KBR were evaluated. KBR
strongly depends on the construction details of the bridge and can therefore be very
different for different bridges of apparently the same type (Fig. 2 left). Average values
were determined for bridges of the same type and are to be used in the context of the
mandatory calculation rules (Fig. 2 right) [2].
Fig. 2. The left diagram shows the bridge correction factors KBR for 10 different steel bridges
with directly fastened track; the right figure illustrates the bridge categories and their KBR values
as implemented in the new calculation rules [2]
For steel bridges with directly fastened or ballasted track, the bridge correction
factors reach high values (12 dB and 6 dB respectively). Steel bridges on main lines
within densely populated areas represent special noise hotspots, for which innovative,
and preferably retrofit, mitigation measures are urgently needed.
Innovative Measures for Reducing Noise Radiation from Steel Railway Bridges 411
2 Measurement Concept
As mitigation measures for bridge noise, typically resilient elements in the track are
used which lead to a reduction of vibrations transferred to the bridge structure.
Furthermore, dampers can be installed at the noise-radiating bridge parts in order to
reduce vibrations. Both approaches were considered in the project.
To assess the efficiency of new measures, a reduction KLM has to be determined
which is defined as the difference of KBR before and after installation of the measure.
The measurement concept is illustrated in Fig. 3 left. Since KLM is dependent on a
number of parameters and is therefore subject to large variations, a mean value has to
be calculated. Therefore, mitigation measures should be tested at a variety of bridges.
Before starting the investigation, a new measurement concept [5], including
guidelines for planning, performing, evaluating and documenting measurements on
innovative measures to be installed at Deutsche Bahn AG was developed. Besides the
acoustical measurements (typically at a distance of 25 m / 7.5 m) to the track center and
3.5 m / 1.2 m above rail head), also measurements of structure-borne vibrations at rails,
sleepers, and the bridge structure were proposed (see Fig. 3 left).
An evaluation of existing measurements showed, that the insertion loss (the
difference in the spectral bridge correction factor before and after installation of a
measure) as determined from structure-borne vibration measurements under certain
circumstances was comparable to the insertion loss from measurements of airborne
noise (see Fig. 3 right) [6].
412 D. Stiebel, T. Lölgen, and C. Gerbig
Fig. 3. The left figure illustrates the noise measuring concept and the measuring locations of
vibrations; the right figure shows the comparison of insertion losses due to a resilient rail
fastening system determined from noise and vibration measurements at different locations [6]
It was concluded that the effect of a measure could also be derived exclusively from
the measurement of vibrations at the bridge. This is of great importance for bridges for
which measurements of airborne noise are not possible or difficult. However, the
corresponding reduction KLM has to be calculated. Since the insertion loss spectrum is
largely independent of the bridge parameters, measured insertion losses from different
bridges with the same installed type of measure can be averaged to reduce the
measurement uncertainty.
To determine KLM from an average insertion loss spectrum, the measure is applied
hypothetically to another bridge, for which measurements of airborne noise during the
passage of a train do exist (example shown in Fig. 4 from [7]).
90
Sound pressure level (dB)
80
70
Fig. 4. The left diagram shows the measured insertion loss of an under-ballast mat; the right
diagram shows the measured noise spectra next to a bridge and the adjacent plain track and the
calculated airborne noise spectrum after installation of the under-ballast mat [7]
Finally KLM is derived as the difference of the bridge correction factors KBR. By
repeating this procedure for several representative bridges, a correction factor
characterizing a mitigation measure suitable for formal recognition can be determined.
As an example, this procedure leads to an average reduction KLM of 4 dB for standard
under-ballast mats.
Innovative Measures for Reducing Noise Radiation from Steel Railway Bridges 413
3 Results
Fig. 5. The picture shows the resilient rail fastening system used; the diagram illustrates the
bridge correction factor KBR before and after installation. “Bridge #” is a bridge ID, with
subscripts referring to different directions; “Stiffness” refers to stiffness of the rail fastening.
With the exception of pads with a stiffness of 0.22 N/mm3, a significant effect can be
observed starting at frequencies above 63 Hz (see Fig. 6 right). For bridges with a noise
emission at relatively high frequencies (> 100 Hz), the use of under-sleeper pads
therefore shows a good effect. However, for bridges with a distinct noise emission at
lower frequencies, under-sleeper pads tend to show only small effects. Additionally the
increase in vibration levels at the sleepers indicates a higher level of rolling noise in the
frequency range of 200 to 250 Hz.
Fig. 6. The picture shows sleepers with pads before installation; the diagram shows the difference
in levels before/after installation of the under-sleeper pads for noise levels (“Air.”) and vibration
levels (“Vib.). “Bridge #” denotes the ID of the bridge, with subscripts for different directions,
“Type” refers to bridge types “T” trough shaped cross section, “H” Hat shaped cross section, “A”
Arch bridge.
Quantifying the effect of the under-sleeper pads in form of a reduction KLM (mainly
based on measured vibration levels) results in an average value of 1 dB. This means,
that the averaged effect of under-sleeper pads is less than the averaged effect of the
commonly used under-ballast mats (4 dB). Further effort is therefore needed to reduce
the resonance frequency of the under-sleeper pads.
Fig. 7. The picture shows bridge dampers mounted on web plates; the diagram compares level
differences before and after installation for noise (“Air.”) and for vibrations (“Vib.”) at different
locations. “Bridge #”:ID of the bridge, with subscripts for different directions; “Type” refers to
bridge types; “Damper loc.” to the dampers location.
Averaged over all steel bridges with ballasted track equipped rail dampers, the
vibration level was reduced by about 1 dB.
For steel bridge #4 with directly fastened track, rail dampers were installed
additionally to the resilient rail fastening (see chapter 3.1). This leads to a further
reduction of KBR of more than 1 dB. Since the measured vibration levels were reduced
similarly, this indicates that not only the rolling noise is affected, but also the bridge
noise. However, since this was already reduced by the measures installed before, its
effect is rather low.
Finally at bridge #4, the existing handrail was filled with absorbing material,
therefore acting as a sound barrier. This leads to an additional decrease of KBR due to a
reduction of the rolling noise in the range of 0.9 to 1.9 dB.
On two occasions, the existing covers (plates or steel grates) were replaced by
elastically supported covers. Contributions stemming from rattling plates were
eliminated effectively. This measure should be used in combination with other
measures against bridge noise.
4 Conclusion
The very extensive investigation performed in Germany allowed a quite general view
on the performance of different mitigation measures but also on possible problems
during planning, installation and operation. As a result, very efficient measures for steel
bridges with and without ballast could be proposed. In addition, it was shown that
reducing bridge noise and rolling noise in combination leads to bridges with a sound
radiation nearly comparable to the plain track. As a consequence, annoyance due to
bridge noise is avoidable even for existing bridges with an initial high noise radiation.
References
[1] Stiebel, D., Behr, W., Brandl, W., Degen, K.G.: Maßnahmen zur Reduzierung des
Brückendröhnens. In: DAGA 2005 (2005)
[2] Überarbeitung der Richtlinie zur Berechnung der Schallimmissionen von Schienenwegen –
Schall 03, draft version (2006)
[3] DB Netz AG: Innovative Maßnahmen zum Lärm- und Erschütterungsschutz am Fahrweg,
final report (2012)
[4] Beier, M., Lölgen, T., Starnberg, M.: Innovative measures for reducing noise radiation from
track. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson,
D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 126, pp. 173–180. Springer, Heidelberg (2015)
[5] DB Systemtechnik: Mindestanforderungen an Nachweismessungen zur quantitativen
Bewertung von infrastrukturbasierten Innovationen zur Minderung des Schienenlärms,
report (2010)
[6] BZA: Akustische Messungen an einer Hilfsbrücke in Saarlouis vor und nach dem Einbau
von elastischen Schienenlagern Typ Ioarg 314, report (1992)
[7] Stiebel, D.: Lärmminderung an Eisenbahnbrücken – Messkonzept zur Ermittlung von
Einfügedämmungen und Korrekturfaktoren. In: DAGA 2012 (2012)
Modelling of Railway Curve Squeal Including
Effects of Wheel Rotation
1 Introduction
wheel rotation in a squeal model has - to the knowledge of the authors - so far not been
investigated.
The rotation of the wheelset leads to a splitting of the resonance peaks in the
frequency response of the wheel into pairs of peaks [1]. The frequency shift of the
resonance peaks of the rotating wheel compared to the peaks of the stationary wheel is
a result of the inertial effects due to rotation, and consequently the shift increases with
the wheelset's angular velocity. Nevertheless for modes where the corresponding mode
shapes produce large deflections at big distance from the wheelset axle (e.g. wheel
modes in contrast to axle modes), the associated inertial effect can be important even at
small velocities. In high-speed applications, the wheelset rotation has been shown to be
significant for the calculation of dynamic train-track interaction, especially if the
excitation takes place at frequencies where two different eigenmodes being affected by
wheel rotation have coinciding resonance frequencies [8]. Wheel squeal arises
essentially at low speeds during the negotiation of tight curves. Nevertheless, a possible
influence of wheel rotation on squeal cannot be disregarded a priori. Squeal is known to
occur predominantly at frequencies corresponding to axial modes of the wheel with
zero nodal circles and several nodal diameters [1], which are affected by wheel rotation
also at lower speeds. In [6], the change in wheel diameter during the lifetime of a wheel
(leading to a change in the eigenfrequencies of the wheel) has been shown to
significantly influence the occurrence of curve squeal. The frequency shifts due to
wheel rotation are expected to be smaller than changes due to reprofiling of the wheel,
but due to the strongly non-linear character of curve squeal already minor changes in
parameters may lead to the sudden onset or disappearance of squeal.
The aim of the present study is two-fold. First, it will be investigated numerically
with a detailed wheel/rail interaction model [7] whether it is important to include wheel
rotation for the prediction of curve squeal. Results obtained with a stationary wheel will
be contrasted with results obtained with a rotating wheel. Second, two different models
of a rotating wheel will be compared: a realistic Eulerian approach including
gyroscopic effects and centrifugal stiffening [9] and a simplified approach, where the
rotating wheel is replaced by a stationary wheel with a moving load. Consequently, the
study will give indications about the model complexity needed for the prediction of
curve squeal.
The wheel/rail interaction model used in this study has been published earlier in [7] and
is extended to take account of the effects of wheel rotation. The time-domain model
includes the coupling between vertical and lateral dynamics and shows squeal in the
case of a constant friction curve. Track and wheel are described as linear systems using
pre-calculated impulse response (or Green's) functions that are derived from detailed
finite element models. The time-stepping procedure is carried out by convoluting the
Modelling of Railway Curve Squeal Including Effects of Wheel Rotation 419
wheel/rail contact forces with the Green's functions of wheel and track. This Green's
function approach leads to high computational efficiency. Measured wheel and rail
profiles are used and the wheel/rail contact position has been pre-calculated for an
assumed lateral displacement of the wheel on the rail [7]. Wheel and rail model have
been chosen based on the conditions in a curve in the network of Stockholm metro,
where severe corrugation and squeal occur [10].
The wheel considered is a C20 metro wheel (radius 0.39 m, mass 213 kg), which is
clamped at the centre of the hub. It has been modelled with three-dimensional 20-node
solid elements with a commercial finite-element (FE) software. The model contains a
total of 7920 elements and 37800 nodes. For the stationary wheel, the wheel is
represented by the vertical point receptance, the lateral point receptance and the
vertical/lateral cross receptance calculated with the FE-model at the pre-determined
contact point on the wheel tread. The eigenmodes of the wheel up to 7 kHz have been
included in the receptances. The Green's functions corresponding to the receptances are
obtained by inverse Fourier transform. The inclusion of the wheel rotation is described
in Section 3.
The track model consists of one continuously supported rail [7] of type BV50, which is
a common Swedish rail type. It is built with waveguide finite elements using the
software package WANDS [11]. In the interaction model, the track is represented by
moving Green's functions, which include the motion of the contact point along the
rail [7, 12]. For simplicity, the lateral motion of the contact point has been neglected.
The moving Green's functions give the displacement response of the rail due to an
impulse excitation of the rail at time t=0 and position x=x0 at a point moving along the
rail with train speed v away from the excitation point. The discrete version of the
moving Green's functions is constructed from ordinary Green's functions obtained from
a series of track transfer receptances calculated with the waveguide finite element
model of the rail. This procedure requires that space and time resolution are linked by
the train speed: Δx=v Δt. The dynamics of the track in vertical direction, in lateral
direction and the coupling between the vertical and lateral dynamics have been
included.
considers the contribution of the structural dynamics of wheel and track to the
creepages. The contact model is solved online in the time-stepping procedure.
Two different models of a rotating wheel are considered: a realistic Eulerian approach
and a simplified approach.
The realistic model of the rotating wheel is based on an Eulerian modal coordinate
approach [9] which uses the finite element methodology for obtaining the equation of
motion terms. The model adopts a modal approach where mode shapes are used as
basis functions in the displacement formulation. The model takes advantage of the solid
of revolution geometry since the mode shapes are implemented in non-rotating
coordinates. The Eulerian approach avoids the need to introduce a rotating frame of
reference, but the kinematics of the material particles of the solid are not explicit: the
coordinate system is associated with the displacement of a particle that at instant t
occupies an arbitrary spatial position. This coordinate system is more efficient because
the forces in the wheel-rail contact and the primary suspension are referred to in a
non-rotating frame. The resultant mathematical model is linear and its equation of
motion matrices are constant and can be pre-calculated. Based on the mathematical
model, receptances are calculated at the pre-determined contact point on the wheel
tread. In the interaction model, the rotating wheel is represented by the corresponding
(ordinary) Green's functions.
Fig. 1. Magnitude of the lateral wheel receptance for train speed v=50 km/h: stationary wheel
(----), stationary wheel with moving load ( , grey), rotating wheel ( ).
4 Simulation Results
Applying the three different sets of pre-calculated Green's functions for the wheel, the
dynamic wheel/rail interaction during curving has been calculated for a range of input
parameters. For a train speed of v=50 km/h, combinations of different values of lateral
creepage and friction coefficient were investigated. Similar as in [7], the resulting
lateral contact force F2 was expressed as a force level according to
F
LF2 = 20 log 2, rms . (4.1)
1N
The rms-value F2,rms of the transient part of the signal was obtained as
∫ ( F (t) − F )
1 t1 +T 2
F2, rms = 2 2 dt , (4.2)
t1
T
where F2 is the mean value of the force in the considered time interval. The rms-value
was calculated from the last 0.15 s of the force signal, which in total was 3.5 s long.
422 A. Pieringer, L. Baeza, and W. Kropp
Fig. 2. Simulation results for the stationary wheel: force level calculated according to
Equation (4.1)
a) b)
Fig. 3. Simulation results for the stationary wheel with moving force (a) and the rotating
wheel(b): difference in force level with regard to the stationary wheel (LF2-LF2,stat).
Modelling of Railway Curve Squeal Including Effects of Wheel Rotation 423
a)
20.5
Lateral force [kN]
20
19.5
0 0.5 1 1.5 2
Time [s]
b)
20.5
Lateral force [kN]
20
19.5
2.192 2.194 2.196 2.198 2.2
Time [s]
c)
Fig. 4. Lateral contact force for the case μ=0.3 and η=-1%; stationary wheel (----, dark grey),
stationary wheel with moving load ( , light grey), rotating wheel ( ); a) time series, b) zoom on the
time series, c) power spectrum.
5 Conclusions
Three different models for the wheel were included in an existing model for the
simulation of curve squeal: a model of a stationary wheel, a simplified approach for a
rotating wheel including only the moving load effect and a realistic approach for a
rotating wheel considering centrifugal stiffening and gyroscopic effects. The wheel
receptances obtained from both rotating wheel models were almost identical. In the
dynamic simulations, all three wheel models gave similar results for different
combinations of friction coefficient and lateral creepage in terms of the lateral contact
forces. On the one hand, these results indicate that the stationary wheel with moving
load is a valid approach to include the wheel rotation, at least for the train speed of
50 km/h considered in this study. On the other hand, it can be concluded that using a
424 A. Pieringer, L. Baeza, and W. Kropp
stationary wheel is sufficient for both capturing the tendency to squeal and predicting
the resulting lateral forces. Relevant differences occurred only during the build-up of
the stick/slip oscillation. The rotation of the wheel delays the build-up process, but it
does not reduce the final force amplitudes. However, the vibration pattern on the wheel
will change due to rotation. This is expected to influence the radiation from the wheel.
Therefore, the influence of wheel rotation on sound radiation needs to be investigated
further.
References
[1] Thompson, D.: Railway noise and vibration: Mechanisms, modelling and means of control.
Elsevier, Oxford (2009)
[2] Fingberg, U.: A model for wheel-rail squealing noise. J. Sound Vib., 365–377 (1990)
[3] Périard, F.J.: Wheel-rail noise generation: Curve Squealing by Trams. PhD thesis,
Technische Universiteit Delft (1998)
[4] Chiello, O., Ayasse, J.-B., Vincent, N., Koch, J.-R.: Curve squeal of urban rolling stock -
Part 3: Theoretical model. J. Sound Vib. 293, 710–727 (2006)
[5] Brunel, J.F., Dufrénoy, P., Naït, M., Muñoz, J.L., Demilly, F.: Transient model for curve
squeal noise. J. Sound Vib. 293, 758–765 (2006)
[6] Glocker, C., Cataldi-Spinola, E., Leine, R.I.: Curve squealing of trains: Measurement,
modelling and simulation. J. Sound Vib. 324, 365–386 (2009)
[7] Pieringer, A.: Time-domain modelling of high-frequency wheel/rail interaction. PhD
thesis, Chalmers University, Göteborg, Sweden (2011)
[8] Torstensson, P.T., Nielsen, J.C.O., Baeza, L.: Dynamic train-track interaction at high
vehicle speeds - Modelling of wheelset dynamics and wheel rotation. J. Sound Vib. 330,
5309–5321 (2011)
[9] Martínez-Casas, J., Mazzola, L., Baeza, L., Bruni, S.: Numerical estimation of stresses in
railway axles using train-track interaction model. Int. J. Fatigue 47, 18–30 (2013)
[10] Torstensson, P.T., Nielsen, J.C.O.: Monitoring of rail corrugation growth due to irregular
wear on a railway metro curve. Wear 267, 556–561 (2009)
[11] Nilsson, C.-M., Jones, C.J.C., Thompson, D.J., Ryue, J.: A waveguide finite element and
boundary element approach to calculating the sound radiated by railway and tram rails. J.
Sound Vib. 321, 813–836 (2009)
[12] Nordborg, A.: Wheel/rail noise generation due to nonlinear effects and parametric
excitation. J. Acoust. Soc. Am. 111(4), 1772–1781 (2002)
[13] Kalker, J.J.: Three-dimensional elastic bodies in rolling contact. Kluwer Academic
Publishers, Dordrecht (1990)
FASTSIM with Falling Friction and Friction Memory
E.A.H. Vollebregt
Summary. In this paper we show that falling friction poses a principal difficulty
for the FASTSIM algorithm. Using velocity-dependent μ leads to discontinuities
in the tractions, displacements and slip velocity. These discontinuities are due to
the local and instantaneous relations that are used. They are circumvented in
different, unsatisfactory ways in the extensions of FASTSIM that are presented
in the literature. We describe a new FASTSIM algorithm that deals with falling
friction properly, using the concept of friction memory.
1 Introduction
These points are addressed in the current paper. We expose the difficulties with the
straight-forward approach and propose a better way of dealing with them. Our approach
is termed “friction memory”: the friction needs time to adapt, and thereby has memory
of the previous history. We show results, including a brief comparison to CONTACT,
in which falling friction is also available [8, 9].
In this work we focus on the frictional, tangential contact problem under steady
conditions. The equations for this are well-known, see for instance [3, 10] or [4].
• The wheel is slipping slightly with respect to the rail. Ignoring the effects of
deformation, this implies relative velocity w called rigid slip, between the two
surfaces,
w(x) = [ ξ- φy, η+φx]T. (2.1)
Here ξ, η, φ are the longitudinal, lateral and spin creepages, and V is the rolling
velocity.
• Rigid slip invokes elastic deformation u at surface positions where the traction is
low, and actual slipping s at positions where the traction reached its maximum. The
slip equation says that these effects must add up:
s(x) = w(x) - V du(x)/ dx . (2.2)
• In the simplified theory, the traction p is assumed to be proportional to the tangential
displacement difference u, with L the so-called flexibility:
u(x) = L p(x) . (2.3)
• The friction law states that the traction cannot exceed the traction bound
g=μs pn, with pn the pressure and μs the friction coefficient. Where slip occurs its
direction is opposite to p:
in adhesion area H: sa = ||s|| = 0, ||p|| ≤ g , (2.4a)
in slip area S: sa > 0, p = -g s / sa . (2.4b)
• The essence of falling friction is that the coefficient μs is variable and depends on the
slip velocity. For instance using an exponential falling friction law [9]:
μs(sa) = μstat - (μstat - μkin) exp(-log(2) sa /shlf ) . (2.5)
Here μstat is the base value and μstat - μkin the maximum decrease due to slip velocity;
shlf is the velocity at which half of the decrease is obtained.
FASTSIM [11] is an ingenious algorithm that solves the traction distribution p(x) in
steady rolling situations. It tracks the evolution of a single surface point as it moves
through the contact area [10]. At the leading edge of the contact area the boundary
condition p=0 is used. Then, as long as the particle is in the adhesion area, s=0 by (2.4),
and the new traction is found using (2.2) and (2.3).
When the traction reaches the traction bound g slip sets in. The equations are then
solved by first computing the estimate padh, assuming s=0, and then scaling p=g,
FASTSIM with Falling Friction and Friction Memory 427
padh/||padh||. This exploits that p and s are collinear (equation (2.4b)), for a graphical
illustration see Giménez et.al. [6].
The first extension of this approach with a velocity dependent friction law that we are
aware of is that of Frédéric Périard [1]. This approach closely follows the original
FASTSIM algorithm. For each new point “i+1” the adhesion solution padhi+1 is
computed first in order to get the direction of pi+1 and -si+1. Then equation (2.4b) is used
with the current estimate of g i+1, equation (2.2) is used to evaluate the slip, and the
friction law (2.5) is used to get a new estimate for g i+1. These steps are iterated until
convergence is reached for the point. Unfortunately, not many results are presented for
this approach.
In 2000, Piotrowski and Kik presented a comparable approach at a Polish
conference. This approach is described in [5]. One difference to Périard’s approach is
that Piotrowski does not use equation (2.2), but instead uses an analysis of the frictional
power density instead to get the magnitude of the slip. This changes the equations but
does not change the results. Also, Piotrowski found a zigzag pattern in the traction
distribution (dashed lines in Fig. 1(a)) that he identified as “recuperation of static
friction” and that he considered non-physical. In order to circumvent this he uses
iteration only for the first point in the slip area. After that, he maintains the same
coefficient of friction for all consecutive points (drawn lines in Fig. 1(a)).
A third extension of FASTSIM with velocity-dependent friction was presented by
Giménez et.al. [6]. This introduces yet another approach for computing the slip. Two
contributions are distinguished: the radial component of the rigid slip w, and the
time-dependence of the traction bound g. When using velocity dependent friction, this
latter term also consists of two parts: one arising from the change of pressure pn, and
another from the change of μs due to the change of slip velocity. The paper presents the
approach for a piecewise linear friction law. In that case the equations for a new point
“i+1” are linear and can be solved in one go, no iteration is required. The approach can
be used for non-linear friction laws such as (2.5) too, using iteration, see Rovira et.al.
[7].
Fig. 1. (a): tractions and slip for Périard/Piotrowski’s methods applied to a testcase with pure
longitudinal creepage (see Section 5). (b): results for the method by Giménez et.al.
428 E.A.H. Vollebregt
Using their initial approach, Giménez et.al. found wildly varying slip velocities on a
fine discretization grid. They attribute these “stability problems” to the reduction of the
friction coefficient with increasing slip velocity. In order to circumvent the problems
they simply ignore one term from the equations. This amounts to using a fictitious slip
ŝa instead of the true slip velocity. They also object to the jump of the tractions at the
transition from the adhesion to the slip area. This is circumvented by changing the
traction bound, using a coefficient of friction computed with the fictitious slip. The
results of this approach are illustrated in Fig. 1(b), showing both the fictitious slip ŝa
and the true slip sa.
A complication exists in the equations that are used, (2.1)—(2.5), that is not recognized
in the earlier works and that spoils the results. The problem is that an instantaneous
friction law where μs depends on sa makes the slip s in equation (2.2) point back to itself
in an unphysical way.
• It is easy to see that slip s must jump from zero to something at the transition from
adhesion to the slip area. Namely, the rigid slip w is put into growth of u (equation
(2.2)) until p hits the traction bound. After that, further rigid slip is accounted for in s.
• In an instantaneous friction law, the friction coefficient μs responds immediately, i.e.
it makes a jump to a lower value.
• This immediately affects the tractions p, and by equation (2.3) we find a
discontinuity in the deformation u.
• This contradicts to a finite slip velocity. The only value that μs can go is to
μkin = lim sa→ ∞ μs(sa ) . (3.1)
The slip velocity cannot be infinite for longer than an infinitesimal time period. After
that the friction coefficient jumps to a higher value again. At the same time, the particle
must stop sliding and go back to the adhesion area. If it would continue slipping, the
slip would have the wrong sense, i.e. it would not be opposite to the traction p.
A basic premise in the analysis of rolling contact is that the contact area can be
subdivided in stick and slip areas. However, this is no longer true for the model that is
defined by equations (2.1) to (2.5). From our analysis we see that the model predicts
infinite slip on separate curves that dissect the contact area. “Immediate recuperation”
(zigzag line in Fig. 1(a)) must always occur. And the friction value on the curves is
always μkin, irrespective of the function μs(sa), the rigid slip w or the pressure pn. This is
not the behavior we want, and therefore the model is not appropriate for representing
the underlying physical phenomena.
The two solutions that are offered by Giménez et.al. and Piotrowski are in our view
both unsatisfactory. Giménez et.al. simply ignore part of the friction law and assign the
gift of prophecy to the surface particles. Note that in Fig. 1(b), sliding sets in before the
tractions reach the original traction bound μstat pn. Piotrowski on the other hand
maintains the first value of μs experienced in the slip area. This value is strongly
affected by whether the traction bound is reached just before or after reaching the next
FASTSIM with Falling Friction and Friction Memory 429
discretization point, and therefore behaves like a random variable. Further, this value
tends to μkin if finer and finer grids are used, thus beating the purpose of the functional
dependence (2.5).
The instabilities that are inherent to the model (2.1)—(2.5) arise because of the
instantaneous reaction of μ to sa in equation (2.5). The model is regularized by adjusting
μ gradually instead of instantaneously, by introducing memory of the previous state.
The friction coefficient μs of (2.5) is then interpreted as the steady state value that arises
at continuous sliding at velocity sa. It is distinguished from the actual coefficient of
friction μ(x). The tendency of μ to μs is made proportional to the slip velocity and to the
difference μ-μs:
-V dμ/dx = -max( sa(x), smin)/dc•(μ(x) - μs(sa(x))) . (4.1)
Here dc is a characteristic slip distance over which the relaxation occurs, and smin is a
minimum value, that allows adaptation of μ in the adhesion area. A nice thing of this
approach is that it is flexible with respect to the steady state behavior that is encoded in
μs(sa).
The concept of friction memory comes from the rate- and state-dependent friction
laws that are widely used in the earthquake community [12, 13]. The physical motivation
is that the actual contact takes place at the tips of asperities, that the strength of asperity
contacts increases over time due to plastic creep, and that the friction reduces by
breaking some contacts and replacing them by new ones. The amount of local sliding
that is needed to break the contact is of the order of the size of the asperities themselves.
Therefore typical values for the memory distance dc are in the order of micro-meters.
In the adhesion area, μ changes slowly and is needed only for checking the traction
bound. Therefore we restrict our attention to the evolution of the unknowns in the slip
area. The main equations are then (2.2) for s and (4.1) for μ. Analysis of the model is
simplified further by focusing on a situation where η=φ=0 and ξ>0. In that case
s=[sa,0]T such that we get a 2D problem in the unknowns sa and μ, independent of y. An
equation for sa is obtained by substitution of (2.4b) and (2.3) in (2.2):
In FASTSIM with Coulomb friction, the slip velocity jumps from zero to s-a upon
entering the slip area. It is tempting to view the equations as a 2x2 system of differential
equations and prescribe μ=μstat and s=s-a at the interface. But note that the system does
not contain any differentials of sa. In fact, equation (4.2) can be substituted in (4.1),
430 E.A.H. Vollebregt
leaving a single (nonlinear) first order differential equation for μ(x). Consequently only
a single initial condition is allowed.
Analysis of equations (4.1) and (4.2) goes by plotting the equations as curves of s=sa
and μ'=dμ/dx in the (s,μ') plane for different values of the parameters. This quickly
reveals the possible scenarios. Particularly, no solution can be found for small values of
dc. One way to avoid this is by always using sufficiently large dc. Another way forward
is to conjecture that the slip velocity temporarily rises to infinity such that the friction
coefficient drops from μ- before the jump to a lower value μ+ after the jump. The
appropriate value for μ+ is then the largest value for which the equations have a solution
again. This leads to another 2x2 system of equations that is analyzed in a similar way as
the equations for the regular case.
The extensions can be incorporated in the FASTSIM algorithm quickly. The overall
structure of the algorithm stays the same, starting at the leading edge and processing
one new point at a time. The main difference is in the computations for points in the slip
area. This needs an iteration process for solving the two equations simultaneously,
which is similar to the iterations used in the existing algorithms [5, 7].
5 Results
Results are presented for the test-case that was considered earlier in [8, 14]. This
concerns a locomotive wheel with radius 625mm and wheel load Fn=106.7kN. The
resulting contact patch has semi-axes a=6.3 and b=12.6mm. The rolling velocity is
V=10m/s. The parameters of the exponential friction law (2.5) are μstat=0.33,
μkin=0.140, shlf=1.25m/s. For the friction memory, the memory distance is
dc=0.003mm and smin=0.001m/s.
Detailed results for the new method are shown in Fig. 2 for a single case with pure
longitudinal creepage ξ=2%, η=φ=0. (The same scenario is considered in Fig. 1.) The
flexibility L is obtained from the original FASTSIM algorithm and then multiplied by
kA=0.15, to reduce the initial slope of the traction curve with respect to Kalker's linear
theory.
Fig. 2. (a): surface tractions -px and slip velocities sx for FASTSIM for different memory
distances dc for a scenario with longitudinal creep. (b): corresponding friction coefficients μ.
FASTSIM with Falling Friction and Friction Memory 431
Fig. 2(a) shows the tractions -px and slip sx along the centerline y=0 of the contact
area, for four different values of dc. The graph on the right shows the corresponding
values of the friction coefficient μ. From this we see that we can control the evolution of
the friction coefficient, from slow and smooth to fast with a kink. The transition from
adhesion to slip occurs slightly later than in the Giménez method, because we don't use
the gift of prophecy.
For small values of dc, our coefficient of friction μ approaches that of Giménez. This
appears to be a general tendency. An interesting compromise is then to take Giménez'
method, without the look-ahead feature, and use this as an approximation of friction
memory for small dc. This is particularly relevant because of the simplifications that are
made in FASTSIM anyhow, ignoring part of the elasticity of the material. Because of
these simplifications, the total forces Fx and Fy are typically accurate up to 20% [3].
Therefore it makes no sense to construct a very detailed model on the basis of the
FASTSIM approach.
Fig. 3 shows computed traction curves for different methods for two scenarios.
Fig. 3(a) concerns running on tangent track with pure longitudinal creepage; Fig. 3(b)
concerns steady curving with a contact angle of 20º and yaw angle ψ=12mrad. In both
cases, our new method is close to the Giménez algorithm again. The differences at low
creepages (Fig. 3(b)) are due to using one flexibility L instead of three flexibilities Lξ,
Lη, Lφ. Both algorithms are close to the extended CONTACT model of [8], and perform
better in this sense than the original Kalker theory and Polach’s algorithm.
Fig. 3. Computed traction curves for the Eurosprinter 127001 locomotive. (a): pure longitudinal
creepage, comparison with measurements. (b): steady curving, mixed creepages.
6 Conclusions
In this paper we considered the steady state rolling contact problem with falling
friction, and its approximation using the simplified theory. In this context, an
instantaneous velocity-dependent friction law leads to infinite slip velocity at discrete
instances in time, followed by immediate recuperation of static friction. In the earlier
extensions of FASTSIM with falling friction this is circumvented in somewhat ad-hoc
ways, which we feel are not satisfactory.
432 E.A.H. Vollebregt
Our approach for falling friction in FASTSIM relies on what is called friction
memory. A friction formula μs(s) is postulated that characterizes the friction under
steady state sliding at velocity s. The actual friction coefficient evolves towards this
steady state friction, at a rate proportional to the sliding velocity. This means that the
friction memory fades with sliding distance “s δt”.
Numerical results show that the new friction law behaves well and regularizes the
results of the earlier approaches. The approach is applied successfully to the
reproduction of the creep-force relation obtained in measurements. However, these
measurements involve only longitudinal creepage. In order to put FASTSIM and its
extensions with falling friction really to the test, measurements should be performed for
different combinations of longitudinal, lateral and spin creepage.
References
[1] Périard, F.J.: Wheel-rail noise generation: Curve squealing by trams. PhD thesis, Delft
University of Technology, The Netherlands (1998)
[2] Thompson, D.J.: Railway noise and vibration: Mechanisms, modelling and means of
control. Elsevier, Oxford (2008)
[3] Vollebregt, E.A.H., Iwnicki, S.D., Xie, G., Shackleton, P.: Assessing the accuracy of
different simplified frictional rolling contact algorithms. Vehicle System Dynamics 50(1),
1–17 (2012)
[4] Kalker, J.J.: Three-dimensional elastic bodies in rolling contact. Solid Mechanics and its
Applications. Kluwer Academic Publishers (1990)
[5] Piotrowski, J.: Kalker’s algorithm Fastsim solves tangential contact problems with
slip-dependent friction and friction anisotropy. Vehicle System Dynamics 48(7), 869–889
(2010)
[6] Giménez, J.G., Alonso, A., Gomez, E.: Introduction of a friction coefficient dependent on
the slip in the FASTSIM algorithm. Vehicle System Dynamics 43, 233–244 (2005)
[7] Rovira, A., Roda, A., Lewis, R., Marshall, M.B.: Application of Fastsim with variable
coefficient of friction using twin disc experimental measurements. Wear 274-275,
109–126 (2012)
[8] Vollebregt, E.A.H.: Numerical modeling of measured railway creep versus creep-force
curves with CONTACT. Wear (2013), doi:10.1016/j.wear.2013.11.030
[9] Vollebregt, E.A.H., Schuttelaars, H.M.: Quasi-static analysis of 2-dimensional rolling
contact with slip-velocity dependent friction. J. of Sound and Vibration 331(9), 2141–2155
(2012)
[10] Vollebregt, E.A.H., Wilders, P.: FASTSIM2: a second order accurate frictional rolling
contact algorithm. Comput. Mech. 47(1), 105–116 (2010)
[11] Kalker, J.J.: A fast algorithm for the simplified theory of rolling contact. Vehicle System
Dynamics 11, 1–13 (1982)
[12] Dieterich, J.H.: Modeling of rock friction 1. Experimental results and constitutive
equations. Journal of Geophysical Research 84, 2161–2168 (1979)
[13] Ruina, A.: Slip instability and state variable friction laws. Journal of Geophysical
Research 88, 10359–10370 (1983)
[14] Polach, O.: Creep forces in simulations of traction vehicles running on adhesion limit.
Wear 258, 992–1000 (2005)
Towards an Engineering Model for Curve Squeal
Summary. Curve squeal is a strong tonal noise that may arise when a railway
vehicle negotiates a curve. The wheel/rail contact model is the central part of
prediction models, describing the frictional instability occurring in the contact
during squeal. A previously developed time-domain squeal model considers the
wheel and rail dynamics, and the wheel/rail contact is solved using Kalker’s
nonlinear transient CONTACT algorithm with Coulomb friction. In this paper,
contact models with different degree of simplification are compared to
CONTACT within the previously developed squeal model in order to determine
a suitable contact algorithm for an engineering curve squeal model. Kalker’s
steady-state FASTSIM is evaluated, and, without further modification, shows
unsatisfying results. An alternative transient single-point contact algorithm
named SPOINT is formulated with the friction model derived from CONTACT.
Compared to the original model results, the SPOINT implementation results are
promising and similar to results from CONTACT.
1 Introduction
Squeal noise is a strong tonal noise that may occur when a railway vehicle negotiates a
relatively tight curve (R < 200 m [1]). The curve radius at which squeal is expected to
occur depends on the vehicle bogie wheelbase. The wheelbase and curve radius define
the angle of attack, which is considered the main kinematic parameter as it defines the
amount of lateral creepage that occurs between wheel and rail [2].
The wheel and rail excitation force originates from the frictional instability
occurring in the wheel/rail contact. This instability is caused by the slip velocity
dependent falling friction characteristic of the wheel/rail contact and/or the
vertical/lateral dynamic coupling of the wheel and rail (modes coupling) [3]. It is still
not clear to what extent each of these factors contributes to the development of squeal,
but cases where squeal is obtained either from simulations or experiments is reported
for each cause (c.f. [3, 4, 5]) separately. Additionally, it is not yet fully understood
which parameters affect the occurrence of curve squeal.
It is clear that the main part of any squeal model is the wheel/rail contact model.
From the many available rolling contact models, Kalker’s nonlinear transient
CONTACT [6] and his linear steady-state FASTSIM [6, 13] algorithms are widely
used. The recent time-domain squeal model developed by Pieringer [3], where
CONTACT with Coulomb friction is used and the wheel and rail dynamics are
included by means of Green’s functions, is one of the most detailed models able to
simulate curve squeal. However, CONTACT is computationally expensive and not
viable in everyday engineering practice. Other time-domain squeal models, mainly
using simplified contact algorithms, were developed e.g. by Fingberg [7], Périard [8],
Huang et al. [9] and Heckl [10]. However, in those models falling friction curves are
used, and the influence of different causes of curve squeal cannot be identified, nor can
be the influence of different rolling contact algorithms.
The intention of this paper is to investigate, within Pieringer’s squeal model [3], the
suitability of different simplified rolling contact algorithms for use in a computationally
efficient engineering model for curve squeal. At this point all contact algorithms apply
the Coulomb friction model, or a model derived from it.
where gij ( 0 ) is the first value of the Green’s function, which gives the local
instantaneous deformation of the body due to the excitation force in the current time
step. The gij ( 0 ) term can be treated as the dynamic flexibility coefficient, which
accounts for the dynamic effects of the system [11]. The second sum on the right-hand
side of equation (2.1) gives the dynamic response due to forces acting in previous time
steps. The indices i, j = 1, 2, 3 refer to the longitudinal, lateral and vertical directions of
the contact coordinate system.
3 Contact Models
Every rolling contact model consists of two main submodels: a normal and a tangential
contact model. The contact model couples the wheel and rail dynamics, and the
tangential contact model is crucial for simulating the frictional instability occurring in
curve squeal.
3.1 CONTACT
Kalker’s CONTACT [6] consists of the non-Hertzian normal contact algorithm NORM
and the transient tangential contact algorithm TANG. The first values of the wheel and
rail Green’s functions are neglected, which significantly simplifies the algorithm. Both
NORM and TANG algorithms are based on the elastic half-space assumption and the
Boussinesq-Cerruti equations. While NORM determines the elements in contact,
TANG determines the stick and slip regions of the contact area. The wheel and rail
dynamic contributions are included in the per-element rigid shift:
where γ x , γ y and γ ω are the longitudinal, lateral and spin creepages, Δx is the
discretization element length, and ( x I , yI ) the element center coordinates in the
contact coordinate system. The dynamics of the wheel and rail are included by means
of the current time step responses ξ 2W and ξ 2R , and the previous time step responses
ξ 2W ,t −1 and ξ 2R,t −1 .
i i
Both NORM and TANG are iterative active set algorithms [6]. The non-linear
system of equations in TANG is solved using the Newton-Raphson method.
436 I. Zenzerovic, A. Pieringer, and W. Kropp
3.2 FASTSIM
Kalker’s steady-state FASTSIM [6, 13] is based on the simplified theory where the
deformation at a point of the contact area depends only on the load at that point. The
normal contact problem is solved using the Hertz contact theory.
The effective, per-element, rigid slip is modified to include the dynamic response of
the wheel and rail:
γy +
(ξ R
2
− ξ 2R,ti −1 ) − (ξ 2W − ξ 2W ,ti −1 )
γx yI γ ω Δx x Iγ ω (3.2)
wI 1 = − , wI 2 = + ,
Lx Lω Ly Lω
where Lx , Ly and Lω are the flexibility parameters computed from the semi-axes
ratio a b of the contact ellipse and the creepage coefficients Cij . The creepage
coefficients are tabulated in [6].
3.3 SPOINT
The single-point normal and tangential contact problems are solved simultaneously.
The wheel and rail dynamics terms, that contain the first values of the Green’s
functions, are included in the contact point gross slip velocity:
⎛ u − uti −1 ⎞
s1 = ⎜γ x + 1 1 ⎟V,
⎝ Δx ⎠
(3.3)
⎡
s2 = γ y +
⎢
( ξ 2R − ξ 2R,ti −1 ) − (ξ 2W − ξ 2W ,ti −1 ) u2 − u2ti −1 ⎤
+ ⎥V,
⎢⎣ Δx Δx ⎥⎦
where u1 and u2 are the contact deformations in the current time step, u1ti −1 and u2ti −1
the contact deformations in the previous time step, and V the rolling velocity. The
system of equations defining the single-point contact is:
⎛ 32R G 2ξ 3 ⎞
0.5
s
Fτ + μ ( s) F3 τ = 0, F3 − ⎜⎜ 3 0 3 ⎟
= 0, τ = 1, 2, (3.4)
δ ( ν ) ⎟⎠
2
s ⎝ 3C
9 1−
where s = s12 + s22 is the absolute slip velocity, δ3C a constant depending on the ratio
a b , R0 the effective radius of curvature, and ξ 3 is the combined wheel/rail vertical
dynamics contribution. The contact deformation is computed as uτ = KFτ , with K
being the contact compliance obtained from the vertical contact stiffness linearized
around the vertical preload. According to [12], the lateral contact stiffness is 20%
higher than the vertical stiffness. The contact plane compliance is then K = K 3 1.2
with K 3 being the vertical contact compliance.
The resulting system of equations is solved using the Newton-Raphson method. In
its current formulation, SPOINT is unable to account for spin creepage.
Towards an Engineering Model for Curve Squeal 437
Fig. 1. Comparison of the friction model used in SPOINT and Coulomb friction for two values of
the friction coefficient 0.3 and 0.5. RF – regularized friction; C – Coulomb
The friction model has the form of a regularized friction (RF). With increasing
Coulomb friction in CONTACT the transition region of the RF (i.e. the region before
the curve converges to a constant value) is extended.
The results underline the importance to distinguish between local and global friction
models. The local friction model applies to a single contact element (or particle in
contact), while the global friction model applies to the complete contact. The
discretization of the contact and the elastic half-space in CONTACT, as a third body
between the wheel and rail, soften the friction curve for small slip velocities. By using
the CONTACT steady-state solution as input for the friction model in SPOINT, this
effect is accounted for in the simplified approach. The results also highlight the
question how an appropriate friction model for real cases should be formulated at all.
4 Results
The occurrence of stick-slip oscillations, signifying curve squeal, can be determined
from the time histories of the lateral contact force obtained from simulations with the
squeal model. Simulations were performed for a vehicle velocity of V=50 km/h and the
438 I. Zenzerovic, A. Pieringer, and W. Kropp
time step length follows from the kinematic relationship Δt = Δx V ( Δx = 0.5 mm).
The amplitude of the lateral force oscillations is quantified with the lateral force level
LF 2 based on the root mean square (RMS) value of the lateral force as:
∫ α ( F (α ) − F )
1 α1 +n 2
LF 2 = 20 log F2,rms , F2,rms = 2 2 dα , (4.1)
n 1
where n=1000 is the number of time samples considered, F2 the mean value of the
force in the considered time interval, and α is the discrete time step number.
4.1 FASTSIM
The steady-state FASTSIM, in its original formulation, seems not to be appropriate to
simulate curve squeal. Results were not physical in all analyzed cases. Obviously,
FASTSIM is not suitable for solving the rolling contact problem when the wheel and
rail dynamics are included directly in the rigid slip, c.f. equation (3.2). However, the
application of a transient algorithm based on FASTSIM (see e.g. [14]) could solve this
problem. The inclusion of the contact area deformation history or previous time step
tractions distribution enhances the simulation behavior and physical results can be
obtained. However, these additional terms alter the original FASTSIM algorithm
changing its properties and further investigations might be needed.
4.2 SPOINT
In Fig. 2 the CONTACT and SPOINT dynamic simulation results are shown in terms of
the lateral force levels. Simulations were performed for different combinations of
friction coefficient and lateral creepage values. Except for high friction values, very
good agreement is found between CONTACT and SPOINT results.
Fig. 2. CONTACT and SPOINT dynamic simulations results presented in terms of RMS values.
The case 5 is denoted with C5.
Towards an Engineering Model for Curve Squeal 439
Fig. 3. Lateral force time histories of case 5 ( γ y = −0.01, μ = 0.4 ) obtained with CONTACT
and SPOINT
In Fig. 3 and Fig. 4, the time histories of a single simulation and the details of the
occurring stick-slip lateral force oscillations are shown respectively. Very good
agreement is found in the stick-slip oscillation details. Deviations between CONTACT
and SPOINT results can be due to spatial and time discretization and the uncertainties
regarding the contact longitudinal and lateral stiffnesses, which are a required input
parameter in SPOINT.
Fig. 4. Details of the lateral contact force oscillations due to stick-slip occurring in the contact for
case 5
5 Conclusion
Two alternative contact formulations were compared to the results of CONTACT, as
used in [3], with respect to simulations of squealing. The contact deformation history,
neglected in steady state FASTSIM, was shown to be crucial to obtain reasonable
results from FASTSIM. The single-point contact showed very good agreement with
CONTACT. This however demands that the global friction model is derived from
CONTACT. However, in practice it might be easier to obtain such global friction
models (e.g. from traction curves) than the local friction model for CONTACT.
440 I. Zenzerovic, A. Pieringer, and W. Kropp
Acknowledgements. The presented work has been performed within the project
“Abatement of Curve Squeal Noise from Trains” (VB11) and is part of the activities in
the Centre of Excellence CHARMEC (CHAlmers Railway MEChanics).
References
[1] Thompson, D.J.: Railway noise and vibration: Mechanisms, modelling and means of
control. Elsevier, Oxford (2009)
[2] Vincent, N., et al.: Curve squeal of urban rolling stock - Part 1: State of art and field
measurements. Journal of Sound and Vibration 293, 691–700 (2006)
[3] Pieringer, A.: Time-domain modelling of high-frequency wheel/rail interaction. PhD
Thesis, Chalmers University of Technology (2011)
[4] Rudd, M.J.: Wheel/rail noise – Part II: Wheel squeal. Journal of Sound and
Vibration 46(3), 381–394 (1976)
[5] Koch, J.R., Vincent, N., Chollet, H., Chiello, O.: Curve squeal of urban rolling stock – Part
2: Parametric study on a study on a ¼ scale test rig. Journal of Sound and Vibration 293,
701–709 (2006)
[6] Kalker, J.J.: Three-dimensional elastic bodies in rolling contact. Kluwer Academic
Publishers, Dordrecht (1990)
[7] Fingberg, U.: A model of wheel-rail squealing noise. Journal of Sound and
Vibration 143(3), 365–377 (1990)
[8] Périard, F.J.: Wheel-Rail Noise Generation: Curve Squealing by Trams. PhD Thesis,
Technische Universiteit Delft (1998)
[9] Huang, Z.Y., Thompson, D.J., Jones, C.J.C.: Squeal prediction for a bogied vehicle in a
curve. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B.,
Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 99, pp. 313–319. Springer, Heidelberg (2008)
[10] Heckl, M.A., Abrahams, I.D.: Curve squeal of train wheels, Part 1: Mathematical model for
its generation. Journal of Sound and Vibration 229(3), 669–693 (2000)
[11] Andersson, P.: Modelling interfacial details in tyre/road contact – Adhesion forces and
non-linear contact stiffness. PhD Thesis, Chalmers University of Technology (2005)
[12] Wu, T.X., Thompson, D.J.: Wheel/rail non-linear interaction with coupling between
vertical and lateral directions. Vehicle System Dynamics 41(1), 27–49 (2004)
[13] Kalker, J.J.: A fast algorithm for the simplified theory of rolling contact. Vehicle System
Dynamics 11, 1–13 (1982)
[14] Guiral, A., Alonso, A., Baeza, L., Giménez, J.G.: Non-steady state modelling of wheel-rail
contact problem. Vehicle System Dynamics 51(1), 91–108 (2013)
An Investigation of the Influence of Track Dynamics
on Curve Noise
Summary. Experience has shown that curve noise issues can become more
severe when curved track is upgraded from timber to concrete sleepers. This
suggests that changing the dynamics of the track structure, for instance using a
softer rail fastening system and/or sleeper, could provide ways to reduce curve
squeal. The problem, however, is that the changes in dynamic characteristics
between timber and concrete sleepers that explain the difference in curve noise
behavior are not currently understood. Observations and anecdotes have hinted at
the effect of different rail pads, fasteners, rail dampers and gauge relief on curve
noise, but have failed to definitively identify either a successful mitigation
strategy or a quantitative relationship with curve noise generation. This paper
presents the results of field trials on curved track, conducted before and after an
upgrade from timber sleepers to concrete sleepers, aimed at improving this
understanding.
1 Introduction
Curve squeal is one of the greatest environmental impacts from rail operations in NSW.
Squeal leads to sleep disturbance and significant outcry from the community in the
vicinity of tight curves. Many residents complain that the problem has only emerged, or
has increased beyond acceptable limits, following upgrades from timber to concrete
sleepers. This pattern of track upgrade quickly followed by community outcry
regarding wheel squeal has been observed all over the network in the past decade.
During this time there have been a number of coincident changes, including new rolling
stock, changes to maintenance practices, relocation of gauge face lubrication units, and
adoption of head hardened rail. Any or all of these could also have an impact on curve
noise.
The perceived increase in curve noise impact has been confirmed [1] and various
hypotheses discussed. This has included ad hoc studies of curve noise on track with
resilient fasteners, and the trial of “spare” rail dampers that were designed for rolling
noise mitigation.
The replacement of timber sleepers on the main line is being driven by operational
considerations and it is not feasible to reverse this process. But the question remains as
to whether some of the characteristics of timber sleepers might help mitigate curve
noise, and if so, what are they? This paper presents the results of an investigation that
aimed to uncover these properties.
2 Methodology
Two separate sets of tests were performed: attended and unattended. The attended tests
included measurements of the Track Decay Rate (TDR) in both vertical and lateral
directions, the track mobilities, and the rail roughness. The unattended measurements
included recording noise and vibration from passing trains for a period of one week.
Both sets of tests were performed before and after the upgrade from timber to concrete
sleepers.
The test curve is bi-directional and carries a mixture of passenger and freight traffic.
The curve is of radius 600 m which is greater than would typically be associated with a
curve noise hotspot but within the range in which squeal would be anticipated. The
track is approximately level with typical speeds of around 30~40 km/h during the tests.
The rails were not replaced during the track upgrade. They are 60 kg/m, AS1085.1 head
hardened rail. The spacing of both timber and concrete sleepers are 600 mm. The rail
fasteners on the timber sleeper track were Pandrol e-clip on steel rail plates with dog
spikes; the rail fasteners on the concrete sleeper track are Pandrol fast-clip HDPE rail
pad.
In addition, measurements of track decay rate, rail mobility (including the cross
mobility between the two rails), rail profiles and rail roughness were performed, both
before and after the upgrade.
Fig. 1. Measurement layout at the test curve, showing 1) the wayside microphone located at
1300 mm from nearest the rail, 2) rail mounted accelerometers measuring both lateral and
vertical acceleration, 3) sleeper mounted accelerometers measuring vertical vibration, 4) string
pots measuring lateral displacement of the rail head, and 5) a wheel sensor measuring the passby
time of each wheelset
The recorded noise and vibration data of each train pass-by was processed to identify:
• The SEL and LAeq,125ms noise levels for the train passby; and,
• The SEL and LAeq,125ms of any squeal or flanging events that occurred, and
the side (Inner or Outer rail side) on which they originated.
Curve noise events were classified into three severity categories corresponding to
noise levels exceeding 100 dBA, 110 dBA and 120 dBA at the microphone. These
categories represent approximately 80 dBA, 90 dBA and 100 dBA at 15 m from the track.
The categories were weighted on an energy basis whereby an event in the highest
category was weighted by a factor of 1 and an event in the category below was weighted
444 J. Jiang et al.
by a factor of 0.1 (and so on). The weighted results were then normalized to the number
of freight train events. For each week of measurement, the weighted and normalized
results represent an equivalent number of severe curve noise events per freight train.
3 Results
3.1 Noise
The overall results from the test site, as shown in Fig. 2, reflect the pattern observed at
other sites on the network, i.e. that curve noise was present before the upgrade from
timber to concrete sleepers, but increased after the upgrade. This was obtained even
though the relatively modest curve radius of 600 m would not normally be associated
with a curve noise hot spot.
Fig. 2. Overall number of weighted curve noise events per freight train – before and after the
track upgrade, 92 freight trains for the Before and 46 freight trains for the After test
The track decay rates (TDRs) measured at the test site are shown in Fig. 3. The TDR in
the frequency region of most interest for squeal, 1500 Hz – 5k Hz, is much greater on
the timber sleeper trackform than on the concrete sleeper trackform, especially in the
lateral direction. This would result in more vibration energy being removed from the
rail in the squeal frequency region on the timber sleeper track. This may reduce the
tendency to squeal and/or the severity of any squeal that occurs.
An Investigation of the Influence of Track Dynamics on Curve Noise 445
Samples of the dynamic gauge measured under train pass-by are shown in Fig. 4. These
examples show that the dynamic gauge of the timber sleeper track is typically 3~4
times that of the concrete sleeper track, which indicates the concrete sleeper track is
stiffer generally than the timber sleeper track in the lateral direction.
Fig. 3. Track Decay Rates at the test site both before and after the sleeper upgrade – vertical (top) and
lateral (bottom)
446 J. Jiang et al.
Fig. 4. Dynamic gauge under a passenger train (a) and a freight train pass-by (b)
Each wheel exerts load on the track and causes both vertical and lateral rail
deflection. The vertical deflection can be calculated as [2]:
F0 β − β
u ( x) = e
x
(cos (β x ) + sin (β x )) (3.1)
2s
where F0 is the vertical load, β= (s/4EI)^1/4, s is the stiffness per unit length, EI is the
bending stiffness of the rail, x is the distance along the rail.
The lateral deflections from a range of passenger train pass-bys were used to
estimate the lateral stiffness by using equation 3.1 and assuming:
• Lateral rail bending stiffness EI =1.08 MN/m2;
• No rotation of the rail; and,
• Lateral loading = 20 kN
An Investigation of the Influence of Track Dynamics on Curve Noise 447
Based on these assumptions, the calculated lateral stiffness of the concrete sleeper
track was between 11 and 28 MN/m2 per unit length; whereas the timber sleeper track
was around 2 to 4 MN/m2 per unit length, i.e. a factor of 6 times increase in stiffness.
The driving point track mobilities (lateral) at the mid sleeper spacing are shown in
Fig. 5.
The mobilities are generally similar in the squeal frequency region, but the
resonances and anti-resonances are sharper in the concrete sleeper mobility suggesting
a lower level of damping.
Fig. 5. Driving point mobility at the mid sleeper spacing in the lateral direction
4 Discussion
The track decay rate and dynamic gauge exhibited the greatest difference between the
timber and concrete sleeper trackforms.
The relationship between gauge (static) and squeal has been examined previously
[3], sometimes with conflicting advice, and no definitive conclusions have been made.
The current investigation suggests that the lateral flexibility of the timber sleeper track
may support mitigation by allowing the gauge to widen under axle load.
The TDR follows the expected trend in the rolling noise frequency region, with the
concrete sleeper trackform providing much higher TDR than the wooden sleeper
trackform (less than 1000 Hz). In the squeal frequency region however, this
relationship was reversed. Just as this means lower rolling noise for concrete sleeper
trackforms as more vibration energy is transferred into the ground, it may be that
wooden sleeper trackforms are less susceptible to curve noise as vibration energy at
higher frequencies is more effectively removed from the wheel.
448 J. Jiang et al.
5 Further Work
The results obtained through this investigation are promising in the sense that both the
TDR and the dynamic gauge can be investigated separately in future tests. The dynamic
gauge can potentially be replicated by using more resilient fasteners, particularly “egg”
style fasteners that support a degree of “roll” of the rail under high axle loads. The static
gauge may be widened by e.g. removing the outer biscuits from the rail clips, at least
for the purposes of a short term trial.
The TDR in the squeal frequency region can be increased through the use of rail
dampers. These dampers are traditionally designed to mitigate rolling noise, but
purpose designed dampers which are active above 1500Hz could be trialed on tight
curves.
Alternatively, sleepers that offer the operational benefits of concrete yet mimic the
dynamic properties of timber, such as composite sleepers, may be considered.
Composite sleepers and transoms of various forms are already installed at isolated
locations on the NSW rail network. A trial of composite sleepers on a tight curve may
also be a possibility.
TfNSW Freight and Regional Development are currently in the process of
investigating these strategies as part of a wider effort to combat curve noise.
Acknowledgements. The authors would like to acknowledge Susan Kay of ARTC for
facilitating the access to the tracks and the CRC for Rail Innovation who developed test
methods and algorithms that were used in this investigation.
References
[1] Anderson, D.: Wheel squeal measurement, management and mitigation on the NSW rail
network. AAS (2004)
[2] Thompson, D.J., et al.: Railway noise and vibration: mechanisms, modelling and means of
control, p. 168. Elsevier (2009)
[3] Nelson, J.T.: TCRP REPORT 23: Wheel/rail noise control manual, p. 85. National Academy
Press (1997)
Field Trials of Gauge Face Lubrication and Top-of-Rail
Friction Modification for Curve Noise Mitigation
1 Introduction
RailCorp has been investigating the issue of curve noise for over 20 years, including
through the use of top-of-rail friction modifiers since the mid-1990s. Despite
substantial effort, wayside treatment of curve noise has given mixed results. It is likely
that this is due to the complex nature of the curve noise mechanisms (particularly wheel
squeal) which, in turn, meant that it had not been possible to fully understand and
control the underlying mechanisms at all affected sites.
3 Methodology
Each of the three monitoring locations was equipped with accelerometers mounted on
both rails, microphones at 1.2 m from each rail, an optical switch serving as a wheel
sensor, a data acquisition system mounted at the wayside, and solar power supply,
regulator and batteries. The data acquisition systems at each measurement location
were controlled by a wayside PC to capture and store recordings for each train pass-by.
The sampling rate was 51.2 kHz per channel with standard anti-aliasing filtering.
Lubrication equipment was fitted with remote monitoring to confirm operational status
and record product output quantities. A portable weather station was deployed
throughout the trial to record rainfall, relative humidity, temperature and wind speed at
half hourly intervals.
Two techniques were used to evaluate the rail lubrication condition throughout the
test area. These were: visual inspection accompanied by a "finger print" test; and
friction measurement using a hand-pushed tribometer. The tribometer required more
onerous safe working procedures and sometimes gave inconsistent results, so the visual
inspection and finger print method was used on most occasions.
Finally, data from the permanent wayside angle of attack system was available for
each of the trains monitored during the trial. This provided angle of attack (the angular
attitude of a wheel flange to the rail) and lateral position data for each axle as well as
information for each bogie (such as rotation).
3.5 Analysis
The recorded signals from each train pass-by event were analysed to determine the
overall noise levels for the train and the short-term energy average noise level
(LAeq,125ms) associated with each axle. Where curve noise was detected, additional
analysis was carried out to match the event to specific axle(s) and to compare rail
vibration levels to determine whether it originated from the inner or outer wheel [10].
Curve noise events were classified into three severity categories corresponding to
noise levels exceeding 100 dBA, 110 dBA and 120 dBA at the microphone. These
categories represent approximately 80 dBA, 90 dBA and 100 dBA at 15 m from the
track. The categories were weighted on an energy basis whereby an event in the
highest category was weighted by a factor of 1 and an event in the category below
was weighted by a factor of 0.1 (and so on). The weighted results were then
normalized to the number of freight train events. For each test scenario, the weighted
and normalized results represent an equivalent number of severe curve noise events
per freight train.
Field Trials of Gauge Face Lubrication and Top-of-Rail Friction Modification 453
4 Results
4.1 Overview
Fig. 1 shows an example time history of noise levels measured throughout the test
curve during a freight train pass-by during test 3 (with lubrication applied to the outer
rail). The characteristic increase in noise due to the diesel locomotives is evident at the
start of the pass-by at each location but, in this example, the severe curve noise peaks
observed at location 1 are not clearly visible downstream of the treatment at locations 2
and 3.
Wheel squeal
events at location 1
Noise level 1.2m from rail, dBA
Time, seconds
Fig. 1. Example noise time history for a freight train pass-by during test 3 (lubrication of the
outer rail). Location 1: , Location 2: , Location 3:
Fig. 2 shows the normalized and weighted results for freight trains throughout the
trials. In test 1, without lubrication or friction modifier treatment in place, severe curve
noise occurred at a rate of approximately 0.8 events per freight train near the beginning
of the curve and deteriorated to 1.4 events towards the end of the curve. Passenger
trains did not generate any severe squeal events in the highest category and were
approximately 100 times less likely to generate curve noise.
The normalized curve noise index at the start of the curve was generally around 0.4
events per freight train, or above, throughout Phase I, but showed fluctuation despite
being conducted under nominally identical (untreated) conditions.
454 D. Curley et al.
Test scenario
Fig. 2. Weighted and normalized curve noise levels generated by freight trains
Two lubrication scenarios (treatment of the outer rail only in test 3 and treatment of
both rails in test 4) gave substantial reductions in squeal noise. In contrast, treatment of
the inner rail only (test 2) gave negligible improvement.
The fluctuation observed at location 1 during the first phase of the trials changed
significantly when lubrication was introduced upstream in Phase II. Curve noise at the
start of the curve showed less fluctuation and was consistently below 0.2.
The additional lubricator deployed for Phase II is now a permanent installation
approximately 1km upstream of the test curve. Ongoing noise monitoring near the start
of the test curve over a period of approximately 6 months confirms the link between
curve noise and lubrication of the outer rail. When the lubricator is functioning curve
noise levels are consistent with those achieved in Phase II of the trials, but noise levels
increase significantly within 2 or 3 days if the lubricator is out of service. Track
inspections suggest that lubrication of the gauge corner region of the rail (rather than
just the gauge face) is important for noise reduction.
Significant curve noise reduction was also achieved when friction modifier was applied
to both rails in test 5, although not to the same extent as that achieved with lubrication.
When the dosage was reduced in test 6 the outcomes deteriorated towards the end of the
curve. When friction modifier was applied to the inner rail only (test 7), no benefit was
obtained.
Field Trials of Gauge Face Lubrication and Top-of-Rail Friction Modification 455
The two friction modifier products (one oil-based, the other water-based) showed
similar results under freight traffic but the oil-based product gave better results for
passenger trains (i.e. in the less severe curve noise categories).
Rolling stock angle of attack behavior during the trials was consistent with previous
experience at this curve [10], with the majority of bogies within the expected range of 5
to 7 mrad at the leading axle and approximately zero at the trailing axle. However, a
proportion of freight bogies show angle of attack values significantly outside the
expected range (20 mrad and above).
During Phase I, with no lubrication in place at the angle of attack monitoring
location, 12 % of axles with high angle of attack (20 mrad and above) generated severe
squeal noise (above 120 dBA at 1.2 m from the rail). This proportion reduced to
approximately 0.5 % when lubrication was introduced upstream. At the same time,
lubrication resulted in an increase in the average angle of attack of some freight
wagons, but this change was much less significant than the reduced probability of
squeal noise for a high angle of attack axle.
The results from the trials include two findings that conflict with the existing theory and
knowledge summarized in Section 2. The first was that lubrication, rather than top-of-rail
friction modifier treatment, had a profound effect in reducing wheel squeal. Visual
inspections of the rail showed that both the gauge corner and gauge face of the outer rail
were well lubricated during this test and that the rail head was generally dry. This
confirms that the improvement in squeal noise resulted from rail lubrication rather than
inadvertent contamination (and therefore friction modification) of the top of the rail.
The second finding was that friction modifier applied only to the inner rail gave no
benefit. This suggests that, contrary to established theory, squeal at this curve is not
caused by lateral creep and friction-instability at the inner rail.
Previous work at this curve [10] identified a high proportion of squeal noise events
at the outer rail. In the present trial, rail vibration data confirmed that the majority
(approximately 90 %) of curve noise events originated at the outer rail during Phase I,
but showed that this proportion reduced to less than 50 % when lubrication of the outer
rail was introduced upstream in Phase II. This shows that curve noise at this site was
initially dominated by wheel/rail interaction at the outer rail and explains why
treatments applied only to the inner rail appeared to provide no benefit.
The trials reported in this paper show that the curve noise investigation methods
developed by the rail Cooperative Research Project [10] provide excellent insights into
the underlying effects. The findings explain why friction modification had provided
mixed results and show that some curves are dominated by squeal noise arising at the
outer rail. This, in turn, can be controlled by conventional rail lubrication, provided it
achieves sufficient coverage of the gauge corner region of the rail.
456 D. Curley et al.
Acknowledgements. This work was carried out in collaboration with the CRC for Rail
Innovation Project R1-105 Improved Rail Noise Management. The CRC is established
and supported under the Australian Government's Cooperative Research Centres
program.
References
[1] Anderson, D., Wheatley, N.: Mitigation of wheel squeal and flanging noise on the
Australian rail network. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson,
C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation
for Rail Transportation Systems. NNFM, vol. 99, pp. 399–405. Springer, Heidelberg
(2008)
[2] Rudd, M.: Wheel/rail noise—part 2: wheel squeal. Journal of Sound and Vibration 46,
381–394 (1976)
[3] Bühler, S., Thallemer, B.: How to avoid squeal noise on railways - State of the art and
practical experience. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C.,
Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 99, pp. 406–411. Springer, Heidelberg (2008)
[4] Dwight, R., et al.: On-board wheel-rail noise monitoring for the direction of track
maintenance. In: Proceedings of the Conference on Railway Engineering 2006,
Melbourne, Australia, pp. 521–525 (2006)
[5] Powell, J.: Wheel squeal noise control at Queensland Rail. In: Proceedings of Advanced
Rail Management Wheel/Rail Interface Seminar, Chicago (2001)
[6] Tickell, C., et al.: Rail wheel squeal – some causes and a case study of freight-car wheel
squeal reduction. In: Proceedings of Acoustics 2004, Gold Coast, Australia (2004)
[7] Kerr, M., et al.: Squeal appeal - Addressing noise at the wheel/rail interface. In:
Proceedings of the Conference on Railway Engineering, Rockhampton, Australia (1998)
[8] Thompson, D.J., et al.: Project A3 – Railway noise: curve squeal, roughness growth,
friction and wear – start up study. Rail Research UK (June 2003)
[9] Truck Bogie Optical Geometry Inspection system, Wayside Inspection Devices,
http://www.wid.ca/tbogi.htm
[10] Jiang, J., Dwight, R., Anderson, D.: Field verification of curving noise mechanisms. In:
Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning,
B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 118, pp. 349–356. Springer, Heidelberg (2012)
Railway Noise Generated by High-Speed Trains
F. Poisson
Summary. High speed trains have been running for almost 50 years in many
countries. High speed railway systems differ from each other in terms of rolling
stock, track, commercial speed, operating conditions, maintenance. Then, the
environmental noise varies according to all these parameters. To improve the
insertion of high speed railway systems in the environment, noise mitigation
measures have been developed. Each one provides a noise reduction for the
considered system. Combinations of the noise reduction solutions are also very
efficient and cost effective.
A state of the art of pass-by noise measurement results from several
high-speed railway systems all around the world is presented and discussed. The
main noise sources of high speed railway system, the rolling noise, the
aerodynamic noise and the equipment noise, are described. The measurement
and calculation methods to characterise the sources are presented. Each
mitigation measure and its efficiency are described. As a conclusion, some
indications for future research topics in high speed railway noise are proposed.
1 Introduction
Passenger trains have been considered as high speed trains (HST) when their
commercial speed is higher than 250 km/h. In the TSI (Technical Specifications for
Interoperability [1]), the definition of high-speed railway is higher than 189 km/h. HST
are running since 1964 in Japan. Now, high speed railway systems are in operation in
15 countries (China, Japan, Spain, France, Germany, Italy, Turkey, South Korea,
United States, Taiwan, Belgium, Netherlands, United Kingdom, Austria, Switzerland)
and several new projects are planned (Saudi Arabia, Morocco, Portugal, Sweden,
Poland, Russia, Brazil, India). In all the countries, HSTs are running on a dedicated
infrastructure at a commercial speed between 250 and 320 km/h. Most of them are also
running on a classical infrastructure at lower speed.
Less than ten train manufacturers are sharing the market of rolling stock and then
different architectures are proposed: classical train set with conventional bogies vs.
articulated train set with Jacobs bogies, single floor vs. duplex, distributed power vs.
concentrated power, etc. The tracks are also a mixture of ballasted and slab tracks, with
different fastening systems. Then, these differences of high speed railway systems lead
to a difference of pass-by noise.
The pass-by noise measurement results of several HSTs are presented in Section 2
and the array measurement technique is introduced. The mains noise sources are
presented in Sections 3, 4 and 5 while Section 6 deals with the noise propagation. The
future of pass-by noise reduction is discussed in Section 7. Main outcomes are
summarized in the conclusion.
In order to prepare a state of the art of the pass-by noise levels of HST, countries which
are running high speed trains have been asked to provide pass-by noise sound pressure
levels of HST in commercial service. In most of the cases, sound pressure levels are
measured according to the standard ISO 3095:2005 [2]. Then, measurement results can
be, a priori, compared. Nevertheless, as track characteristics are not provided, the
pass-by noise levels of Fig. 1 represent the noise of the global system, the rolling stock
and the track. Pass-by measurements carried out during the NOEMIE project [3] in
Europe are included in the picture. For confidentiality reasons asked by the acousticians
who provided the measurement data, the train type is not displayed. Each symbol
corresponds to a type of train and the legend provides the year of delivery.
Fig. 1. LPAeq,Tp in dB(A) measured at 25 m from the track and 3.5 m above railhead of HST in
commercial service according to the train speed. Each symbol corresponds to a HST that can be
characterised at different speeds. The year corresponds to the first year of commercial service of
the HST type.
Railway Noise Generated by High-Speed Trains 459
Table 1. LPAeq,Tp of 56 different TGV measured at 25 m on a TSI compliant track. “Ref” is the
LPAeq,Tp measured at 25 m in dB(A) during the train homologation.
The mean value of the pass-by sound pressure levels of commercial TGV is
approximately equal to the homologation value of this train type. Minimum and
maximum values are close to the mean one (-3.0 dB(A) / +2.0 dB(A)) considering that
the measurement uncertainty and the propagation effect also have an impact. As the
track quality is the same one for all the pass-bys, this homogeneity can be explained by
the quality of the rolling stock maintenance.
The results presented in Fig. 1 show that the noise of HST all around the world is
quite homogeneous. Over a period of 50 years, a large change occurred due to the
removal of cast-iron brake blocks. After this change, the pass-by noise levels have
decreased slowly. A reduction of around 2-3 dB has occurred over the last 15-20 years
(with some exceptions). On one side, no innovation has occurred to reduce drastically
the environmental noise, on the other side the reference pass-by noise of the
homologation is preserved by the maintenance of rolling stock and tracks.
The pass-by noise of HST is a combination of rolling noise, aerodynamic noise and
equipment noise. Each source can be characterised by using dedicated models and
measurement techniques. An array of microphones is a well-suited tool to characterise
all the noise sources of a given HST during its pass-by.
In case of HST, the movement of the source can not be neglected and the array
processing must be adapted to face the consequences. A method has been proposed in
[5] to adapt the beamforming to remove the Doppler effect (amplitude and frequency
modulation induced by the movement). With the speed increase, the turbulent boundary
layer around the train induces a convection of the acoustic wave. An artificial source
located on the trailer of a TGV running at 300 km/h has been located with an array of
460 F. Poisson
microphones in [6] to validate the array processing. It shows that the shift induced by
the convection can be neglected.
Aerodynamic noise sources can be also characterised in a wind tunnel with scale
models or full size subsystems. Compared to the pass-by measurement, the higher
thickness of the air flow in the wind tunnel induces a convection that must be taken into
account to localise the noise sources at the right place. A geometrical model of the
convection can be introduced in the array processing or the microphone antenna can be
installed in the air flow, close to the model [7, 8].
Some examples of noise source characterisation using an array of microphones are
presented in the following sections.
3 Rolling Noise
The rolling noise is the most important source from the railway system. Vibration is
caused by running over the wheel and rail roughness, which is radiated as noise by the
sleeper, the rail and the wheel. A linear regression on the sound pressure level
according to the train speed shows that the rolling noise is increasing according to
30·log(train speed) [9].
An array of microphones can be useful in characterising the rolling noise during the
train pass-by. The source model used in the classical beamforming method is a
monopole. As the radiation of the wheel and the rail has a complex directivity [10], the
array processing gives a rough representation of the noise source as shown in Fig. 2. A
single source is localised around the contact point of the rail and the wheel. New
approaches are proposed in [11, 12, 13] to adapt the array processing in order to
separate the wheel and the rail contributions. At this time, signal processing based on
near field acoustic measurement combined with rail vibration measurement are most
promising to extract the rail and the wheel contributions during the train pass-by [14].
Fig. 2. Rolling noise in the 4 kHz third-octave band located on a TGV using an array of
microphones. The source is located at the contact point even if the noise is mostly radiated by the
wheel in this frequency range (dynamic range: 10 dB) [15]
The rolling noise can be predicted by using for example the TWINS software [16]
and measured input parameters (rail roughness, wheel roughness and track decay rates).
Railway Noise Generated by High-Speed Trains 461
As for freight trains, the braking system has an influence on pass-by noise of HST. At
270 km/h, the difference of LPAeq,Tp between a TGV PSE (1981) equipped with cast iron
brake blocks and a retrofitted TGV PSE equipped with K blocks was around 10 dB(A).
Now, all the HST are equipped with disc brakes or K blocks which preserve the running
surface of the wheel.
Nevertheless, a measurement campaign carried out with a TGV equipped with disc
brakes shows that the wheel roughness has still a great influence at 330 km/h. The
global pass-by noise level of a TGV Duplex has been measured before and after the
train runs on a track with a too high ballast layer between the sleepers which causes
flying stones [15]. During the pass-by, the stones fly and are crushed between the rail
and the wheel. Wheel and rail roughness are affected. Table 2 shows the impact of
wheel roughness on the pass-by noise.
Table 2. Difference of LPAeq,Tp in dB(A) measured at 25 m for a TGV Duplex running with
corrugated wheels compared to the same train equipped with recently reprofiled wheels [15]
Numerical values of Table 2 give a maximum estimation of the noise reduction that
can be provided by the maintenance of the wheel roughness. These results are not
representative of daily conditions of wheel and track roughness.
The track decay rates also have an influence on the rolling noise. They can be
improved by an appropriate choice of rail pads [17], and/or by adding rail dampers [18].
As experienced in [19], the noise reduction provided by rail dampers on a French
high-speed line depends on the wheel roughness and varies from 1.5 to 2.3 dB(A) at
350 km/h.
The wheel damping can be also increased by added wheel dampers. Two bogies of
TGV Atlantique have been equipped with wheel dampers in [19]. The mean noise
reduction achieved is 2.3 dB(A) for the microphone at 7.5 m and a train speed up to
300 km/h. At a greater distance (25 m), the results were less relevant as extra bogies,
not equipped with wheel dampers for the test, contributed to the measured noise.
The combined effect of rail and wheel dampers is presented in Table 3. As the
rolling noise remains an important noise source at 300 km/h for the French high speed
railway system, the combination of rail and wheel dampers provides a significant noise
reduction of 3 dB(A).
462 F. Poisson
Table 3. Noise reduction of the LPAeq,Tp at 7.5 m in dB(A) of a TGV Atlantique. Reference is the
LPAeq,Tp at 7.5 m in dB(A) of the same TGV Atlantique running on a test site without any
dampers. Rail dampers are installed on an adjacent test site of the same track [19]
All around the world, HST are running on slab and ballasted tracks. In most of the
cases, the slab track leads to an increase of pass-by noise around 3 dB(A) at 25 m
compared to the ballasted track.
In case of the French East high speed line [20], this increase has been found to be due to
a combination of a difference of rail roughness, a difference in track decay rates and the
reflections from the slab. The difference of rail roughness is mainly due to a difference
of maintenance process in this particular case which leads to a noise increase lower than
1 dB. The influence of the difference in track decay rates is also around 1 dB. It can be
minimised by an acoustical optimisation of the slab track. The rail fastening system (rail
pad, sleeper shoe, etc) can be designed to improve the track decay rates at the conception
phase. The reflection from the slab itself has been studied in [20] with a propagation model
based on the linearized Euler equations. The noise increase due to the reflection from the
slab decreases when the source height is higher and the distance to the track becomes
larger. This propagation effect due to the slab track is around 1 dB(A) at 7.5 m and
becomes negligible at 25 m. Nevertheless, the reflection onto the slab track in near field
can be reduced by using damping material onto the slab itself (see Fig. 3).
All these results show that the global pass-by noise of high speed trains at
commercial speeds less than 320 km/h can be reduced by decreasing the rolling noise.
Solutions to reduce the rolling noise like rail grinding, wheel reprofiling, optimised rail
pads, rail and wheel dampers are not specific but remain efficient for HST. As an
example, ICE trains in Germany [22] are running with wheel dampers. Track equipped
with rail dampers are more often not high speed lines.
4 Aerodynamic Noise
With the increase of commercial speed, the contribution of the aerodynamic noise
becomes significant [23]. The noise is generated by several train parts: the “nose” of the
first car, the inter-coach spacing, the bogies, the pantograph, the pantograph recess and
the coach surfaces. Phenomena are quite similar with the noise generated by flaps and
landing gear from aircraft during landing. It is assumed that the slope of the increase of
the aerodynamic noise according to the train speed is around twice the slope of the
increase of the rolling noise. As shown in [24], the estimates of these slopes differ
widely from one study to another. The conclusion is that the transition speed where the
contribution of the aerodynamic noise to the global pass-by noise becomes higher than
the contribution of the rolling noise is about 370 km/h.
Two types of phenomena occur:
- noise generated by flow over structural elements (vortex shedding, cavity
noise),
- noise generated by turbulent flow (turbulent boundary layer, boundary layer
separation, unsteady wake).
Direct source measurement is more difficult for the aerodynamic noise than for the
rolling noise. It can be characterized with an array of microphones during the train
pass-by as shown in the example of Fig. 4. Limitations due to the monopole source
model used in the beamforming occur as the aeroacoustic sources are multipole-type
[24].
Aerodynamic noise can be also characterised in a wind tunnel with scale models (see
Fig. 5) or full size subsystems (see Fig. 6). The wind tunnel must be an anechoic
environment to perform acoustic measurements. In case of a scale model of HST, the
shear flow between the bottom of the vehicle and the ground is affected by the fixed
ground. Some wind tunnels are equipped with a moving ground plane but it increases
the background noise. Fixed ground can be used if the distance from the bottom of the
scale model to the ground is doubled [25].
Nevertheless, the transposition to full scale and the source quantification remain
difficult.
464 F. Poisson
Fig. 4. Noise source distribution of FASTECH 360S running at 340 km/h characterised with a
spiral array of microphones [26]
Fig. 5. Scale AGV mock up in front of a random microphone array in S2A wind tunnel [27]
Fig. 6. High speed train pantograph in front of a spiral microphone array in RTRI wind tunnel
[28]
Direct source measurement is possible on HST by using for example Neise probes.
The main difficulty is to locate the sensor close to the source without creating a new
aeroacoustic source. Signal processing of the output of the microphone must be also
adapted to reject noise from the air flow. Long time integration and methods based on
coherence between several probes are efficient to identify the source characteristics
[29].
Computational Fluid Dynamics (CFD) is used to simulate the air flow around the
train and the Computational AeroAcoustic (CAA) to calculate the aerodynamic noise
[30]. First, the turbulent area is characterised by Navier-Stokes equations and k-Ɛ
Railway Noise Generated by High-Speed Trains 465
turbulence modelling. The output is used as input to the Lighthill equations. This
approach is limited to simple geometries as the model of the Lighthill analogy depends
on the problem geometry. Direct Numerical Simulation (DNS) and Large Eddy
Simulation (LES) are alternatives to simulate the unsteady aerodynamic source term.
DNS deals with all the scales of phenomena but is limited to simple geometries. LES
associated with Lighthill analogy or resolution of Euler’s equations is efficient to deal
with industrial cases, in case of small components. The Reynolds-Averaged
Navier-Stokes (RANS) method is less costly to model turbulence. The flow is split into
a time-averaged part and a fluctuation part solved with the k-Ɛ approach. The increase
of computer performances makes some approaches more practicable as for example the
Lattice Boltzman Method (LBM) [31, 32, 33, 34] combined with the Ffowcs-Williams
Hawkings integral [35] to propagate the noise in the environment. The aerodynamic
noise of a complete HST is then computed in a few days taking into account all the
details of the bogie, the inter-coach spacing, etc [36]. An example of LBM output
(turbulent flow around an AGV) is presented Fig. 7.
Fig. 7. Turbulent flow calculated with LBM for a detailed model of the AGV [33]
The comparison between experimental results from array processing and simulation
results is not easy. A new approach has been proposed in [33] to compare wind tunnel
measurement and LBM simulations. As the LBM provides time signals with phase
information [37], a “virtual array” of microphones is defined in the mesh and the
beamforming is applied. An example of two colour maps is presented in Fig. 8.
Fig. 8. (left) Beamforming results in wind tunnel on a AGV scale model (right) Virtual
beamforming based on LBM results (dynamic range: 10 dB) [33]
In case of this first example, the resolution of the virtual method (right picture of
Fig. 8) is poor. The array and the processing can be optimised as the constraints are
466 F. Poisson
completely different between the wind tunnel measurement and the LBM calculation.
In wind tunnel, the duration of the time signals from microphones is long and the
number of microphones limited. With the LBM, the time signal duration is very short
and the number of “virtual microphones” is unlimited. This approach is very promising
to deal with the comparison of experimental and numerical results.
Fig. 9. Acoustic power (Lw) of the main aeroacoustic sources of a TGV Duplex estimated by
LBM (320 km/h) [36]
In case of a TGV for which the front shape is not optimised and the first bogie is not
equipped with fairings, these parts are the main sources. Acoustic powers of the
inter-coach spacing, the bogies (others than the first one) and the pantograph (including
the pantograph recess) are in the same order of magnitude. In case of a HST equipped
with noise mitigation measures, this balance can be different. For example, the main
source for a Shinkansen E2-1000 running behind a noise barrier is the pantograph at
360 km/h (see Fig. 18).
The acoustic power spectra of the aeroacoustic sources can be also compared. An
example provided by LBM for a TGV Duplex is presented in Fig. 10. The
corresponding Lw are presented in Fig. 9.
Railway Noise Generated by High-Speed Trains 467
Intercar
Fisrt car
Bogie 3
Bogie 1
Panto
Lw(dB(A))
Bogie 2
Frequency (Hz)
Fig. 10. Acoustic spectra (Lw) of the main aeroacoustic sources of TGV POS estimated by LBM
(320 km/h) [36]
The shapes of the spectra are similar and quite smooth except for the inter-coach
spacing, for which the cavity noise generates peaks. For this example, aeolian tones
generated by the pantograph are hidden in the global noise generated by the pantograph
and its recess. More details on each source are given in the following paragraphs.
Fig. 11. Front shapes of HST (left) ICE [www.shortfacts.com] (middle) AVE-S130
[www.bombardier.com] (right) Shinkansen E5 (2011) [26]
468 F. Poisson
Details of the front shape like wiper, water deflector below the windscreen and steps
for the driver’s door generate aerodynamic sources detected by array measurement.
Their contribution to the aerodynamic noise of the nose of the first car can be neglected
[24].
This experiment shows that the design of the solutions has a very important impact
on its efficiency. In this case, as the wheels are not completely covered by the fairing
and the wheel and rail roughness are low, the fairing has no impact on the rolling noise.
The design phase to integrate the first bogie into the front part of the train is crucial;
scale models can be evaluated in wind tunnel [8, 33] or by CAA [27] to reduce
expensive and time consuming measurement campaigns. In practice, constraints of
integration are strong (limited clearance, running in both directions, engine or braking
cooling, de-icing, maintenance, etc) and reduce the possibility to use efficient bogie
fairing to reduce rolling and aerodynamic noise.
Railway Noise Generated by High-Speed Trains 469
The aerodynamic noise around the inter-coach spacing is a cavity noise that has been
characterised by experimental and theoretical studies for several “length in the flow
direction” / “height” (l/h) ratios:
- l/h < 1: the cavity is deep and pure acoustic resonance occur.
- 1 < l/h < 7 or 8: the cavity is open and the coupling between acoustic waves and
shear layer oscillations creates self-sustained oscillations.
- l/h > 7 or 8: the depth of the cavity is very small compared to the length and the
shear layer reattaches on the cavity floor.
The dedicated case of inter-coach spacing has been studied in [42]. As the case l/h >
7 or 8 is the less noisy, the solution to reduce the inter-coach spacing noise is to close
the cavity. The most effective solution is the circumferential diaphragms installed for
example on Fastech 360. Instead of a complete closing of the cavity which is possible
only when the relative displacement of the cars is small, mud-flaps are installed on
most of the HST. The size of mud flaps can be optimised to close the cavity at the
maximum, preserve the accessibility for maintenance and allow the large displacement
between the two coaches when running on classical lines. An example is presented in
Fig. 13 (right).
Fig. 13. (left) Circumferential diaphragms of Fastech 360 [26], (right) Mud flaps of KTX before
and after optimisation [43]
The pantograph is mainly composed of bare cylinders and then the aerodynamic noise
is generated by unsteady air flow. Some parts generate broadband noise and also an
aeolian tone at dedicated frequencies. Vortices break away from the surface and impact
the fluid. The generated force fluctuates at the same frequency as the vortex shedding
and creates a dipole noise source. The force and the aeolian tone are characterised by a
non-dimensional frequency called Strouhal number: St=fl/U0. f is the frequency of the
vortex shedding, l is a characteristic length (for example, the diameter of the bare) and
U0 the external flow velocity. This phenomenon appears for a critical Reynolds number
470 F. Poisson
Re=U0l/v above about 3.105 with v the kinematic viscosity of the fluid (1/v~6.104 for
air). At commercial speed of HST, this phenomena occurs for bare cylinders of the
pantograph.
A dedicated method has been proposed in [44] to evaluate the noise source
distribution in combination with an acoustic technique based on Howe’s vortex sound
theory. In this approach, the noise is extracted from the flow field calculation, but the
interaction between the dipoles must be taken into account as some cancellations
between dipoles occur. Then, another approach based on the Siddon’s method
developed in [45, 46] gives the equivalent dipoles sources. A comparison between the
calculation and the measurement in a wind tunnel of the sound pressure level radiated
by the panhead of a pantograph is presented in Fig. 14. The global behaviour is well
represented in the frequency domain, especially the peak frequency.
Fig. 14. Comparison of the experimental and numerical sound pressure spectrum of a panhead of
a pantograph [45]
As the pantograph becomes an important noise source when HST are running behind
a noise barrier, numerous investigations have been also conducted in wind tunnel to
characterise the noise source [23, 28, 39] and to design mitigation measures.
The first approach consists in simplifying the pantograph itself to reduce the
interaction with the air flow. For example, the diamond shape pantograph of the first
Shinkansen has been replaced by a single arm one on Series 700 Shinkansen (similar to
the pantograph of Fig. 15). In this case, the number of bare cylinders is minimized and
then the number of noise sources. Tonal noise of the main bare cylinder is shifted to the
low frequency domain by using a larger diameter. Streamlined shapes to prevent
unsteady vortex shedding of bare, panhead, hinge, base frame and insulator must be
also considered. The PS207 type pantograph of Shinkansen series E2-1000 [26]
presented in Fig. 15 is a good example of the simplification of this subsystem.
Railway Noise Generated by High-Speed Trains 471
Fig. 15. Low noise PS207 pantograph of Shinkansen series E2-1000 [26]
For the bare pantograph itself, the noise reduction can be achieved by using ribs to
break up coherent vortex shedding as it has been tested on a DSA350SEK pantograph
of German ICE-1 and ICE-2 [39].
The flow can be controlled around the pantograph by covering the surfaces with
porous materials. The experimental characterisation in wind tunnel of a pantograph
similar to the PS207 equipped or not with porous materials presented in [28] shows that
the additional noise reduction can reach 1.9 dB(A) at 360 km/h with a significant noise
reduction of 5 dB in third octave band 250 Hz.
In the case of trains that run with multiple electrical systems, they are equipped with
multiple pantographs. For example, each power car of a TGV is equipped with two
pantographs. Then, when the train is running, some pantographs are lying in the
pantograph recess. The cavity noise is in the same order of magnitude as the raised
pantograph [39]. There is no mitigation measure in place today to avoid the flow to go
inside the recess.
The turbulent boundary layer developed on the train surface is the lowest aerodynamic
source. Nevertheless, models have been developed and some array measurement
carried out to characterize this source. Its contribution to the pass-by noise can be
neglected [24, 39]. For interior noise, the turbulent boundary layer can be an important
source in some frequency bands.
4.8 Summary
As presented in the previous paragraphs, each aerodynamic source is characterised by a
dedicated theoretical approach, numerical simulation or measurements. The definition
of the corresponding equivalent noise sources remain a challenge.
472 F. Poisson
Nevertheless, some solutions have been designed to reduce the aerodynamic noise of
HST: design of the front shape, bogie fairings, single arm pantograph, closed
inter-coach spacing, etc. Data from the bibliography can not be used to make a relevant
comparison of the noise reduction provided by each solution because the integration of
each one into the train is the key point.
Estimation can be obtained by CAA starting with an initial existing train design and
modifying it. For example, LBM simulations carried out on a TGV POS provide the
noise reduction achieved on each source [36] (see Table 4).
Table 4. LBM prediction of noise reduction achieved for each source by design modifications of
a TGV POS running at 320 km/h [36]
The mitigation measures of the aerodynamic noise are mainly in place on Japanese
trains as they are running at high speed in urban areas on elevated structures equipped
with noise barriers. Then, the rolling noise is reduced by the barriers but the
aerodynamic noise sources, on the upper part of the rolling stock, remain. In Europe,
the environmental context is different. The density of population along high speed lines
is lower. Train shapes are more classical and solutions like bogie fairings,
circumferential diaphragms, etc are not implemented due to practical difficulties. In
addition, interoperability and smaller loading gauge induce some constraints to the
train design.
5 Equipment Noise
Equipment noise of HSTs is not specific to the HSTs. As an example, the difference of
the global pass-by noise (LPAeq,Tp) of a TGV Duplex measured at 25 m is 2 dB(A) when
the two power cars are powered or not. The contribution of equipment noise becomes
the main one at stand still and speeds less than 80 km/h. The main noise sources are the
following: the cooling (fan) noise, the electrical noise (converter) and the compressor
noise.
Fan noise can be predicted with academic models [47] or CAA [48]. The physical
phenomena responsible for noise are known and some solutions exist [49]. Classical
Railway Noise Generated by High-Speed Trains 473
axial fans can be optimized to reach a noise reduction of 8 dB(A) at source. The most
efficient solution is to replace the axial fan by a radial one. Then, the tonal components
are removed and the noise level is reduced by 10 dB(A) [50].
The electrical noise is tonal. Numerical models coupling the calculation of electrical
forces and mechanical response allow optimization of the design. Then, some pure
tones can be cancelled or shifted in frequency [51].
The noise of the compressor can be also reduced. The classical piston system can for
example be replaced by a screw compressor. In this case, the noise reduction achieved
at source is around 8 dB(A).
For all the equipment noise, the integration into the train plays a major role.
Encapsulation can provide an efficient noise reduction for some sources.
The control of the equipment can be also optimized, especially at standstill, to stop
all the equipment that is not necessary and to reduce the fan speed to the minimum.
Equipment noise which is predominant at standstill and low speed (< 80 km/h) is not
specific for HSTs. For all the passenger trains, the increase of comfort leads to a higher
electrical power and then much more cooling and electrical noise sources. As presented
in this section, noise reduction solutions exist but few of them are in place on
commercial trains. They will be integrated in the future if acoustic requirements
become more severe at standstill and low speed.
6 Noise Propagation
The long range noise propagation is not particular to high speed trains. Nevertheless, as
the source spectrum contains much more energy in the low frequency bands due to the
contribution of the aerodynamic noise, this energy propagates over very long distances
in some combinations of meteorological and landscape characteristics.
Noise barriers are effective mitigation measures on the propagation path [52],
especially when the noise barrier is higher than the train. In this case, noise reduction
can reach 10 dB(A). The height of the noise barrier can be reduced to decrease its cost
and make easier the integration in the environment. Then, the noise reduction provided
by a reflective, 3 m high (2.1 m above the head of the rail), noise barrier located 5 m
from the centre of the track is only 6 dB(A) at 320 km/h [53]. The noise reduction
decreases to 4.5 dB(A) at 375 km/h. It is probably due to the increase of the
contribution of aerodynamic noise with the train speed, especially around the
pantograph and its recess which are higher than the noise barrier. For these reasons, the
very low height noise barriers installed close to tracks are not well adapted to the HST.
Damping material and/or multi diffracting edges can be added to the noise barrier.
Thirtyfour configurations have been tested with scale models in [54] and full scale tests
have been carried out in [55]. The most efficient solution seems to be the Y shape (see
an example in Fig. 16) which provides an increase of the noise reduction of 4 dB(A)
compared to the vertical reflective noise barrier of the same height.
474 F. Poisson
Fig. 16. Prototype of a Y shape noise barrier along the French east high speed line [53]
The noise reduction can be also increased by adding roof mounted noise barriers as
used on Shinkansen (see Fig. 17). The additional noise reduction of the pantograph
source is between 7 dB and 8 dB. Some investigations on the shape design are
presented in [26] to guarantee that the barrier itself is not an aeroacoustic source.
Fig. 17. Roof mounted noise barrier and low noise pantograph on Shinkansen trains (left: picture
from the author, right [26])
Fig. 18. Calculated time histories of individual noise components of series E2-1000 at 275 km/h
and 360 km/h (with noise barrier) [26]
With the definition of the “mitigated HST”, a global impact assessment can be
performed with a noise mapping software.
The main difficulties in this approach are the following:
- For the direct calculation method, the definition of each source (rolling noise,
aerodynamic noise and equipment noise) in the pass-by noise calculation software
is difficult. For each one, an equivalent source based on results of simulation tools
(TWINS, CAA, etc) or measurement (array processing, scale model in wind
tunnel, etc) is defined in position, level, spectrum and directivity.
- The efficiency of the mitigation measure on each noise source of the train and of
the track must be also defined.
- The cost effectiveness of each solution of noise reduction is not fully agreed. Data
are missing.
Then, based on the experience of existing high speed railway systems, some
mitigation measures are integrated in the design phase to be compliant with the noise
limit value required by the law.
In this context, the place for innovation is limited. New mitigation measures can not
be developed in these projects. They must be developed, tested, validated in dedicated
research projects to be ready for the design phase of new high speed railway systems.
In the research field, mitigation measures presented in the previous sections are still
under investigation in parallel with some other topics for which some examples are
given in the following.
For the rolling noise, the most promising approach remains the increase of the track
decay rates and the decrease of the combined roughness. Using new materials, the rail
pad stiffness can be adjusted at different values in several frequency bands. According
to TWINS simulation, a rolling noise reduction of 2 dB can be achieved. The influence
of the other track components such as the rail section, the sleeper mass, the ballast
stiffness, the booted sleeper and the wheel design is less than 1 dB. The decrease of the
combined roughness is linked to the maintenance process of the rail and the wheel. Rail
476 F. Poisson
grinding and wheel reprofiling can still be optimised from an acoustical point of view.
The key issues are also the process and the cost. A monitoring of rail and/or wheel
roughness which initiate maintenance operations is necessary to guarantee the
acoustical performances of the railway system along the time. Germany has developed
such an approach for the maintenance of rail roughness on classical tracks. In parallel,
investigation about the roughness growth must be continued.
For the aerodynamic noise, several innovations are proposed:
- The enclosure of the pantograph recess. As for airplane landing gears, the noise
source due to the pantograph recess can be reduced by a complete enclosure of the
cavity when a pantograph is not used.
- The multipart bogie fairings. To facilitate the integration of bogie fairing, it can be
split into several parts like jet deflectors preserving the noise reduction [SNCF
patent in progress].
- The air flow around the train can be actively controlled to reduce the noise [59].
Some experiments carried out in laboratories show several approaches: the active
control of boundary layer noise [60], the oscillating blowing [61] to control the
near-wall turbulence and the jet vortex generators [62] to control turbulent
separated flow.
For the equipment noise, the natural airflow around the train can be used for the
cooling of some devices which reduces the number of fans. An experiment has been
carried out on the Fastech 360.
An improvement of noise barrier efficiency has been developed since several years
[63]. Active noise control devices (loudspeakers and microphones) are integrated into
the wall to generate sound waves in opposite phase and cancel the incident wave. The
increase of performance is around 4 dB(A), i.e. the same as the increase provided by a
“Y” shape multidiffracting edge.
This section shows that the railway community must face two important issues in
parallel. On one side, the main challenge is to integrate existing mitigation measures in
the design phase of new high speed railway systems, the second one is to conduct
research to design new mitigations measures for all the sources, rolling noise,
aerodynamic noise and equipment noise. Cost benefit analysis is necessary to facilitate
these challenges.
In the recent past and today some research teams work more globally on new high
speed railway systems. The “Green Train” [64] is studied in Scandinavia by a
consortium around Trafikverket, KTH, Chalmers and Bombardier. This electrical
multiple unit can run up to 320 km/h. The Deutsches Zentrum fur Luft and Raumfahrt
(DLR) is working on the ‘high speed train of 2035” [65]. Environmental noise is treated
in these projects but the proposed mitigation measures are the classical ones.
8 Conclusion
High speed railway systems have been running for almost 50 years around the world.
Depending on the operating conditions, the environmental constraints and the political
Railway Noise Generated by High-Speed Trains 477
choices, each railway system has its own design. Nevertheless, physical phenomena
remain and the catalogue of noise reduction solutions is the same one. As presented in
this paper, each solution provides a noise reduction of at most a few dB(A). Then,
pass-by noise simulation is a key point to design the most efficient combination of
mitigation measures to reach a significant and robust noise reduction. Even if research
on physical topics like the definition of equivalent noise sources is still necessary, the
bottleneck of the mitigation of railway noise is the cost benefit analysis. The cost is
difficult to estimate based on prototype solutions and the benefit is locally dependent.
Nevertheless, an effort must be put on this topic to help all the stakeholders in the
design of “silent high speed railway systems”.
Acknowledgment. This paper has been prepared with measurement results provided
by several acousticians of the railway domain all around the world. Most of absolute
values are suppressed and private references are not mentioned to preserve the
confidentiality. Then, the author thanks the acousticians of ADIF, Bombardier
Transport, State Key Laboratory of Traction Power of Southwest Jiaotong University
of Chengdu (China), Deutsche Bahn, Réseau Ferré de France, Railway Technical
Research Institute of Japan, Tecnalia and SNCF colleagues for providing data and their
contribution to this article. The author would like to also thank the reviewers for their
helpful comments.
References
[1] High-speed rolling stock technical specification for interoperability (February 21, 2008)
[2] ISO 3095:2005, Railway applications – Acoustics – Measurement of noise emitted by
railbound vehicles (2005)
[3] Fodiman, P., Staiger, M.: Improvement of the noise technical specifications for
interoperability: The input of the NOEMIE project. In: International Workshop on Railway
Noise (2004)
[4] SNCF Report: Indicateur événementiel, DOC008259 / D004674 (2007)
[5] Barsikow, B., King, W.F.: On removing the Doppler frequency shift from array
measurements of railway noise. Journal of Sound and Vibration 120(1), 190–196 (1988)
[6] Poisson, F.: Localisation et caractérisation de sources de sources acoustiques en
mouvement rapide. PhD Thesis, Université du Maine, France (1996)
[7] Soderman, P.T., Allen, C.S.: Microphone measurements in and out of airstream. In:
Mueller, T.J. (ed.) Aeroacoustic Measurements. Springer (2002)
[8] Yamazaki, N., Takaishi, T.: Wind tunnel test on reduction of aeroacoustic noise from car
gaps and bogie sections. Quarterly Report of RTRI 48(4), 229–235 (2007)
[9] Poisson, F., Gautier, P.E., Létourneaux, F.: Noise sources for high speed trains: a review of
results in the TGV case. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson,
C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation
for Rail Transportation Systems. NNFM, vol. 99, pp. 71–77. Springer, Heidelberg (2008)
[10] Zhang, X.: The directivity of railway noise at different speeds. Journal of Sound and
Vibration 10, 1016 (2010)
[11] Le Courtois, F.: Caractérisation des sources acoustiques sur le matériel ferroviaire par
méthode d’antennerie. Ph D, Université du Maine, France (2012)
478 F. Poisson
[12] Kitagawa, T., Thompson, D.J.: The horizontal directivity of noise radiated by a rail and
implications for the use of microphone arrays. Journal of Sound and Vibration 329,
202–220 (2010)
[13] Faure, B., Chiello, O., Pallas, M.A.: Characterization of the acoustic field radiated by a rail
using a microphone array. In: International Congress of Sound and Vibration (2010)
[14] Letourneaux, F., Mellet, C., Coste, O.: MISO: A measurement method to separate noise
emission of railway vehicles and track. World Congress of Railway Research (2003)
[15] Gautier, P.E., Poisson, F., Létourneaux, F.: High speed trains external noise: recent results
in the TGV case. In: International Congress of Acoustic (2007)
[16] Thompson, D.J., Janssens, M.H.A.: TWINS – Track-wheel interaction noise software.
Theoretical manual, version 2.4, TNO report TPD-HAG-RPT-930214 (revised) (1997)
[17] Vincent, N., Bouvet, P., Thompson, D., Gautier, P.E.: Theoretical optimization of track
components to reduce rolling noise. Journal of Sound and Vibration 193(1), 161–171
(1996)
[18] Thompson, D.J., Jones, C.J.C., Waters, T.P., Farrington, D.: A tuned damping device for
reducing noise from railway track. Applied Acoustics 68, 43–57 (2007)
[19] Létourneaux, F., Cordier, J.F., Poisson, F., Douarche, N.: High speed railway noise:
assessment of mitigation measures. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E.,
Hanson, C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 99, pp. 56–62. Springer,
Heidelberg (2008)
[20] Poisson, F., Margiocchi, F., Bongini, E., Dragna, D.: Global pass-by noise of high-speed
train running onto slab track and ballasted track. World Congress of Railway Research
(2013)
[21] Müller, G., Möser, M.: Taschenbuch der Technischen Akustik. Springer (2003)
[22] Holzl, G.: A quiet railway by noise-optimised wheels. Zeitschrift fur Eisenbahnwesen und
Verkehrstechnikund Die Eisenbahntechnik Glasers Annalen 188(1), 20–23 (1994) (in
German)
[23] German-French cooperation DeuFraKo Annex K, final report: Noise sources from high
speed guided transport (1994)
[24] Thompson, D.: Railway noise and vibration. Elsevier (2009)
[25] Kitagawa, T., Nagakura, K.: Aerodynamic noise generated by Shinkansen cars. Journal of
Sound and Vibrations 231, 913–924 (2000)
[26] Kurita, T.: Development of external-noise reduction technologies for Shinkansen
high-speed trains. Journal of Environment and Engineering 6(4), 805–819 (2011)
[27] Paradot, N., Masson, E., Poisson, F., Grégoire, R., Guilloteau, E., Touil, H., Sagaut, P.:
Aeroacoustics methods for high speed train noise prediction. World Congress of Railway
Research (2008)
[28] Sueki, T., Ikeda, M., Takaishi, T.: Aerodynamic noise reduction using porous materials and
their application to high speed pantographs. Quarterly Report of RTRI 50(1) (2009)
[29] Fremion, N., Vincent, N., Jacob, M., Robert, G., Louisot, A., Guerrand, S.: Aerodynamic
noise radiated by the intercoach spacing and the bogie of a high speed train. Journal of
Sound and Vibration 231, 577–593 (2000)
[30] Talotte, C.: Aerodynamic noise: a critical survey. Journal of Sound and Vibration 231(3),
549–562 (2000)
[31] Marié, S., Ricot, D., Sagaut, P.: Accuracy of lattice Boltzmann method for aeroacoustic
simulations. In: 13th AIAA/CEAS Aeroacoustics Conference (2007)
Railway Noise Generated by High-Speed Trains 479
[32] Dong, Y.H., Sagaut, P., Marie, S.: Inertial consistent subgrid model for large-eddy
simulation based on the lattice Boltzmann method. Journal Physics of Fluids 20, 035104
(2008)
[33] Paradot, N., Poisson, F., Grégoire, R.: Méthodes numériques pour la prédiction du bruit
aérodynamique des trains à grande vitesse. Revue Générale des Chemins de Fer 173 (June
2008)
[34] Gaylard, A.: A comparison of a conventional RANS and a lattice gas dynamics simulation
- A case study in high speed rail aerodynamics. In: IMech Symposium (1998)
[35] Casalino, D.: An advanced time approach for acoustic analogy prediction. Journal of
Sound and Vibration 261, 583–612 (2003)
[36] Masson, E., Paradot, N., Allain, E.: The numerical prediction of the aerodynamic noise of
the TGV POS high-speed train power car. In: Maeda, T., Gautier, P.-E., Hanson, C.E.,
Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise
and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 437–444.
Springer, Heidelberg (2012)
[37] Dong, Y.D., Sagaut, P.: A study of time correlations in lattice Boltzmann-based large-eddy
simulation of isotropic turbulence. Journal Physics of Fluids 20, 035105 (2008)
[38] Gu, X., Li, Y., Liu, L.: Researches on the application of noise source identification
technique of high-speed trains in China. In: UIC High Speed Congress (2012)
[39] German-French cooperation DeuFraKo Annex K2, final report: Noise sources from high
speed guided transport (1999)
[40] Holzl, G., Fodiman, P., Schmitz, K.P., Pallas, M.A., Barsikow, B.: Deufrako K2: localized
sound sources on the high speed vehicles ICE, TGV-A and TR-07, Internoise (1994)
[41] Guccia, L., Gautier, P.E.: Aeroacoustic research applied to TGV. World Congress of
Railway Research (1996)
[42] Fremion, N., Vincent, N., Jacob, M., Robert, G., Louisot, A., Guerrand, S.: Aerodynamic
noise radiated by the intercoach spacing and the bogie of a high-speed train. Journal of
Sound and Vibration 231, 577–593 (2000)
[43] KRRI: Test and assessment of Korean high speed train, presentation of the Korea Rail
Road Research Institut to SNCF
[44] Takaishi, T., Ikeda, M., Kato, C.: Method of evaluation dipole sound sources in a finite
computational domain. JASA 116(3), 1427–1435 (2004)
[45] Ikeda, M., Mitsumoji, T.: Evaluation method of low frequency aeroacoustic noise source
structure generated by Shinkansen pantograph. Quarterly Report of RTRI 49(3), 184–190
(2008)
[46] Ikeda, M., Mitsumoji, T.: Numerical estimation of aerodynamic interference between
panhead and articulated frame. Quarterly Report of RTRI 50(4), 227–232 (2009)
[47] Gerard, A., Berry, A., Masson, P.: Control of tonal noise from subsonic axial fans. Part 1:
reconstruction of aeroacoustic sources from far-field sound pressure. Journal of Sound and
Vibration 288(4-5), 1049–1075 (2005)
[48] Roger, M.: Préambule, fondements théoriques et état de l’art. Congrès Bruit
aérodynamique des ventilateurs, Lyon (2006)
[49] Silence: Improved design of a Diesel engine cooling system, Deliverable ED17, EU-FP6
SILENCE project (2007)
[50] Poisson, F., Gautier, P.E.: The railway noise reductions achieved in the Silence project. In:
Acoustics 2008, Paris (2008)
[51] Silence: Global reference frame of the vibro-acoustic and of the dynamic behaviour of the
traction motor and of the wheel, Deliverable E.D.18, EU-FP6 SILENCE project (2006)
[52] Hemsworth, B.: Noise barriers for fast passenger trains. Internoise (1977)
480 F. Poisson
[53] Belingard, P., Poisson, F., Bellaj, S.: Experimental study of noise barriers for high-speed
trains. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T.,
Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 118, pp. 495–503. Springer, Heidelberg (2012)
[54] Nagakura, K., Kitagawa, T.: Study on effective shapes of sound barriers for Shinkansen.
RTRI Report 16(12) (2002)
[55] Ibarcq, P.: Final report of European projet EUROECRAN (1999)
[56] Poisson, F., Gautier, P.E., Bongini, E., Létourneaux, F.: Noise radiated by a TGV Duplex
running at 360 km/h: a summary of results and reduction potential. World Congress of
Railway Research (2011)
[57] Gambard, N., Sapena, J., Planeau, V.: A methodology for exterior noise prediction of
railway rolling stock. Euronoise (2009)
[58] Frid, A., Orrenius, U., Kohrs, T., Leth, S.: BRAINS - the concepts behind a quick and
efficient tool for prediction of exterior and interior railway vehicle noise. In: Acoustics
2012 (2012)
[59] Joslin, R.D., Thomas, R.H., Choudhari, M.M.: Synergism of flow and noise control
technologies. Prog. Aerosp. Sci. 41, 363–417 (2005)
[60] Pamiès, M., Garnier, E., Merlen, A., Sagaut, P.: Response of a spatially developing
turbulent boundary layer to active control strategies in the framework of opposition
control. Phys. Fluids 19, 108102 (2007)
[61] Tardu, S.F.: Active control of near-wall turbulence by local oscillating blowing. J. Fluid
Mech. 439, 217–253 (2001)
[62] Selby, G., Lin, J., Howard, F.: Control of low-speed turbulent separated flow using jet
vortex generators. Exp. Fluids 12, 394–400 (1992)
[63] Nakashima, T., Ise, S.: Active noise barrier for far field noise reduction. International
Congress of Acoustics (2004)
[64] Green train shows Swedish technology. International Railway Journal, 51–54 (September
2008)
[65] In focus Germany, Developing the high speed train of 2035, Railway Gazette International
(March 2013)
Component-Based Model for Aerodynamic
Noise of High-Speed Trains
1 Introduction
Recent studies have shown that the aerodynamic noise from high-speed trains be-
comes significant for train speeds between 300 and 320 km/h and dominant when the
train speed exceeds around 350 km/h [1]. Taking into account that high-speed trains
exceeding 300 km/h are being introduced in several countries, aerodynamic noise
becomes an important source of noise, which must be properly assessed for noise
reduction in order to fulfill environmental noise policies.
Predictions of the aerodynamic noise generation can be made, in principle, by means
of CFD (Computational Fluid Dynamics) and CAA (Computational AeroAcoustics)
software. However, normally only a small spatial domain and simple geometric con-
figurations can be covered and very high time consumption and cost are needed.
Much simpler component-based models have been successfully developed and ap-
plied to aerodynamic noise prediction from aircraft landing gear by Guo [2] and by
Smith and Chow [3]. In the present work, starting from Guo’s model, the constants
have been adapted using experimental results on noise radiated by a full scale panto-
graph in a wind tunnel [4]. Because Guo’s model does not include the prediction of
certain spectral features such as narrow peaks that seem to govern the noise spectrum,
an extended model is proposed, on the basis of Guo’s model and the model developed
for train pantographs by Behr et al. [5], showing good agreement with the tests.
For the bogie case, as no experimental data was available beforehand, an experiment
has been designed and carried out using a 1/10 scale bogie mock-up in a low noise
open-jet anechoic wind tunnel. The Sound Pressure Level (SPL) spectrum is validated
against existing pass-by test data showing a good agreement, encouraging the use of
these experimental data for model calibration and/or validation. Adaptation of the com-
ponent-based prediction model to the bogie case is the subject of further work.
2 Component-Based Model
The noise generated by vortex shedding caused by air flow over a bluff body will have
a typical “haystack” spectrum shape with the frequency of the peak coinciding with the
vortex shedding frequency. The frequency, amplitude and the width of that peak will
depend on the geometry of the bluff body and the flow speed. A high-speed train
pantograph or bogie can be simplified as a group of bluff bodies and the overall noise
from them can be approximated as the incoherent addition of the noise produced by all
those components independently.
The effect of the flow interaction between components, end effects, round edges,
upstream turbulence and surface roughness are not yet included in the model described
here but they could be added in further model developments.
Guo [2] expressed the overall SPL radiated by a landing gear as the incoherent sum of
the noise generated by different components classified by their size: big (wheels),
medium-sized (main struts) and small components (screws and other details). The noise
produced by each of these groups of components has specific spectral features covering
different parts of the noise spectrum (low-, mid- and high-frequencies). Because each
group of components gathers elements with different main dimensions the “haystack”
like spectrum for each of them will be broader than the spectrum for an individual
cylinder.
The far field mean square sound pressure given by Guo for the landing gear case has
been adapted by Thompson et al. [6] for application to the high-speed train case as
follows:
(ρ c2 )2 M 6
〈 p2〉 = 2 0 0 ∑ β 0 i S i Fi ( St )
R (1 − M cos θ ) 4 components
(2.1)
where ρ0 is the density of air, c0 is the speed of sound, R is the distance between the
noise source at the train and the receiver, M = U/c0 is the Mach number with U the train
speed, and the term (1–Mcosθ)4 is the convective amplification factor where θ is the
angle between the flow direction and the line from the source to the observer position.
The terms for each component i are the amplitude factor β0i, the surface area Si and the
normalized spectrum Fi(St), the amplitude and shape of which depends on the Strouhal
number St for the ith component and the values of some empirical constants defined by
Guo [2]. The Strouhal number is given by St=fd/U, with f the frequency and d a
characteristic dimension, e.g. the diameter of a cylinder.
Component-Based Model for Aerodynamic Noise of High-Speed Trains 483
This model is next extended, following Behr et al. [5], to account for the appearance of
distinct peaks in the pantograph noise spectrum. Whereas Guo’s model is based on
groups of components divided according to their size, here each component of the
pantograph is modelled separately. They have been identified as cylinders with dif-
ferent lengths L and cross-section geometry: circular (C), square (S), rectangular (R)
and elliptical (E). The mean square sound pressure radiated by each component at the
point x is related to the Strouhal number, correlation length (lc) and fluctuating lift
coefficient (CL) by [7]:
ρ 02U 6 St 2 C L2 Ll C
p( x)2 = (2.2)
16 c 02 R 2 (1 − M cos θ ) 4
As Fi is normalized so its integral over frequency is unity, equating equations (2.1) and
(2.2) yields:
St i2 C Lrms
2
, i L i l Ci
β 0i = (2.3)
16 S i
The peak width and roll-off slopes of the normalized spectrum were calibrated using
experimental data on noise radiated by flow over cylinders [8]. The following function
is now used for the normalized spectrum of each component:
a1 St n1 A
Fi ( St ) = (2.4)
S 2 n2
+ St n 2 A + B
where a1, n1, n2 are empirical constants that control the spectrum amplitude, the roll-on
and the roll-off slopes respectively and the values of the constants A and B were cal-
culated so that dF(St0)/dSt = 0, in order to get the maximum of the function at St0, and
the integral of F(St) over the frequency to be unity.
For the broadband noise, a different normalized spectrum function and amplitude
factor β0i are determined using empirical data from the literature [8].
To assess the influence of the tilt angle respect to the vertical axis (ψ) on the noise
radiated by the main and control struts of the pantograph, the Independence Principle
[9] is used. According to this approximation, the relevant inflow speed is the compo-
nent normal to the cylinder, UN = U0cosψ [9], so the amplitude of the noise radiated will
be reduced by a factor (cosψ)6. The vortex shedding frequency further decreases by a
factor cosψ.
In order to calibrate and validate the above models for the pantograph case, existing
measured data on the noise radiated by a full scale pantograph for different flow speeds
was available [4]. An omnidirectional microphone was used placed perpendicular to
484 E.L. Iglesias, D.J. Thompson, and M.G. Smith
the flow direction at a distance of 5 m from the pantograph centre so convective am-
plification can be neglected.
It was found that the aerodynamic noise generated by the whole pantograph followed a
speed exponent of 5.7, which is slightly lower than the theoretical value of 6 for a
circular cylinder used in Section 2. Using this value, the amplitude factor β0i for each
component had to be modified. In order to match the original model spectra at a speed
of 330 km/h, values of β0 = 7.5×10-9 for medium-sized components and β0 = 1.3×10-6
for small components were chosen. Because no noise data was available for isolated
components, the normalized spectra could not be calibrated so Guo’s values of the
constants St0 , σ, ν, q, A and B were retained.
Fig. 1(a) shows a comparison between tests and the modified model for a flow speed
of 330 km/h. The model predicts the general trend of the spectrum shape. Nevertheless
the noise spectrum from the experimental data seems to be dominated by peaks asso-
ciated with specific sound sources.
Fig. 1(b) plots the measured and predicted OASPL (Overall A-weighted SPL) for
seven different flow speeds. The differences between tests and predictions lie in the
range 2 to 4 dB for all the flow speeds considered.
a) b)
Fig. 1. a) Predicted and measured 1/3 octave noise spectrum for a flow speed of 330 km/h. b)
OASPL from model and tests for seven different flow speeds.
Table 1 shows the constants used in the normalized spectrum function given by
equation (2.4) for the peak (P) and broadband (B) noise, for each of the component
geometries.
Table 1. Values of the constants of the normalized spectrum for the peak (P = peak) and broadband
noise (B = broadband) for each of the component geometries (C = circular, S = square, R = rectangular,
E = elliptical)
a1 n1 n2 a1 n1 n2
Geom. Geom.
P B P B P B P B P B P B
C 2.21 0.11 20 3 20 3 R 6.05 0.78 7 2 14 2
S 1.82 0.02 18 2 18 2 E 4.30 1.12 8 3 12 3
Fig. 2(a) compares the results of this revised model and the wind tunnel tests for a
flow speed of 330 km/h. A much better agreement is found than in the previous figure,
the noise spectrum being governed by the noise produced by individual components at
the pantograph head (e.g. contact strip, lift limiters, etc). The noise produced by the
yawed struts of the pantograph main body is not contributing significantly to the overall
noise, for the given tilt angle of 70°.
Fig. 2(b) compares the predicted and measured OASPL. Due to the use of a speed
exponent of 6 in the present case, the noise is under-predicted by between 4.5 and 2 dB
for speeds below 250 km/h; however for higher speeds the difference is less than
1.5 dB.
a) b)
Fig. 2. a) Predicted and measured 1/3 octave noise spectrum for a flow speed of 330 km/h. b)
OASPL from model and tests for seven different flow speeds.
For the bogie case, no existing data was available for comparison with the prediction
model. For this reason, experiments were designed and carried out in the ISVR open jet
anechoic wind tunnel. A 1/10 scale bogie mock-up was constructed which can be
configured as either a powered bogie, with motor and gearbox, or as trailer bogie with
486 E.L. Iglesias, D.J. Thompson, and M.G. Smith
brake discs on the axles. The noise radiated by this was measured for different flow
speeds (25, 40, 50 m/s) using seven far-field microphones placed at a distance of 2 m at
different radiation angles. Due to the scale factor and speed range available the Reyn-
olds numbers achieved during the experiments are considerably lower than those ap-
plying to a high-speed train bogie at full scale [14]. Nevertheless, as will be seen, it
provides a useful dataset.
Fig. 3 shows a rear and front view of the experimental set-up. The bogie mock-up
was attached to a stiff plywood panel that simulates the floor of the train. This rigid
panel was surrounded by an acoustic baffle constructed of an aluminium sheet covered
in dense acoustic foam on the flow-facing side of the baffle. The centre of the bogie
model was aligned with the centreline of the nozzle in such a way that the flow can be
considered symmetric on both sides of the bogie model.
In order to simulate the inflow conditions to the bogie when this is installed in a train
two ramps were used. The upstream ramp was designed to avoid flow separation and to
provide smooth flow to the upstream wheelset components. To avoid edge effects the
upstream ramp is wider than the nozzle but this means that no flow impinges on the
sides of the bogie so this component of noise is missing in the experiments. Any effect
of the incoming boundary layer was neglected.
The data obtained from the wind tunnel tests were compared with measured data
from a high-speed train pass-by [15] for different train speeds (200, 250, 300 and
350 km/h). Because the pass-by data was acquired for the whole train, only the noise in
the low frequency part of the spectrum (below 500 Hz approximately) is likely to be
dominated by aerodynamic noise.
a) b)
Fig. 3. Anechoic wind tunnel tests experimental set-up. a) Rear view. b) Front view.
The Sound Power Level (PWL) for the motor and trailer bogie mock-up was cal-
culated using the data from the seven microphones available. This was then scaled for
the scale factor and modified to correspond to the train speed. Next, the SPL was
calculated for a train pass-by including 4 motor bogies and 9 trailer bogies. The effect
of convective amplification, Doppler frequency shift and retarded time were accounted
for. The measured data were available for a microphone at 25 m from the train in terms
of the TEL (Transit Exposure Level) as specified in ISO 3095. The influence on the
incident flow speed of the bogie position along the train was not accounted for,
choosing the train speed as incident flow speed.
Component-Based Model for Aerodynamic Noise of High-Speed Trains 487
The results are shown in Fig. 4. A reasonably good agreement is found both for
amplitude and shape of the spectra. However, the dips found at 320 and 350 km/h
between 500 Hz and 1 kHz do not appear in the wind tunnel tests. The speed exponent
is slightly lower for the pass-by measurements probably due to the influence of rolling
noise sources.
In further work the component-based approach will be extended and applied to the
bogie case.
Fig. 4. Comparison between a typical noise spectrum from a TGV-Duplex measured at 25 m and the
noise spectrum calculated at the same distance using the wind tunnel data for different flow speeds:
200 km/h, 250 km/h, 300 km/h and 350 km/h
5 Conclusions
The component-based model developed by Guo to predict aircraft landing gear noise
has been applied to the high-speed train pantograph case. A good agreement with
experimental data was obtained in terms of overall level, but discrete peaks that dom-
inate the noise spectrum cannot be represented with this model.
In reviewing the underlying physics of the noise generation a revised compo-
nent-based model has been proposed where each component is modelled independently
and identified with cylinders with different cross-section. The amplitude of the noise
spectrum from each component is related to the fluctuating lift, correlation length and
Strouhal number while the spectrum shape is calibrated empirically. A good agreement
with the wind tunnel tests is found. This, together with the very short calculation time,
means that the model can be useful for high-speed train pantograph aerodynamic noise
prediction. Further wind tunnel tests are planned to increase the available database on
noise from cylinders.
For the bogie, experiments on the noise radiated by a 1/10 scale bogie mock-up were
carried out. Comparisons with high-speed train pass-by measurements have shown the
potential application of the results. Even though it was not possible to identify the
contribution of each of the bogie component independently, the experiments were
488 E.L. Iglesias, D.J. Thompson, and M.G. Smith
useful to clarify the underlying physics behind the aerodynamic noise from bogies.
Future wind tunnel tests are planned to measure the noise radiated by simple geome-
tries that can approximate bogie components and to include the noise from the side of
the bogie. It is also intended to apply a modified form of the component-based model
for the bogie case.
Acknowledgements. This work has been carried out within the EU FP7 project
Acoutrain, coordinated by UNIFE. The assistance of the partners, and DB
Systemtechnik in particular, is gratefully acknowledged.
References
[1] Mellet, C., et al.: High speed train noise emission: Latest investigation of the aerodynam-
ic/rolling noise contribution. Journal of Sound and Vibration 293(3), 535–546 (2006)
[2] Guo, Y.: A component-based model for aircraft landing gear noise prediction. Journal of
Sound and Vibration 312(4), 801–820 (2008)
[3] Smith, M.G., Chow, L.C.: Prediction method for aerodynamic noise from aircraft landing
gear. In: Proc. 4th AIAA/CEAS Conf., Toulouse, France (1998)
[4] Lölgen, T.: Wind tunnel noise measurements on full-scale pantograph models. In: Proc.
Joint ASA/EAA Meeting, Berlin, Germany (1999)
[5] Behr, W., et al.: Low noise pantograph ASP - recent developments. In: Proceedings of
Internoise 2000, Nice, France (2000)
[6] Thompson, D.J., Smith, M.G., Coudret, F.: Application of a component-based approach to
modelling the aerodynamic noise from high-speed trains. In: Maeda, T., Gautier, P.-E.,
Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos,
P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118,
pp. 427–435. Springer, Heidelberg (2012)
[7] Blake, W.K.: Mechanics of flow-induced sound and vibration, vol. 1. Academic Press,
New York (1986)
[8] King, W.F., Pfizenmaier, E.: An experimental study of sound generated by flows around
cylinders of different cross-section. Journal of Sound and Vibration 328(3), 318–337 (2009)
[9] Zdravkovich, M.M.: Flow around circular cylinders: applications, vol. 2. Oxford Univer-
sity Press on Demand, UK (2003)
[10] Schewe, G.: On the force fluctuations acting on a circular cylinder in cross-flow from
subcritical up to transcritical Reynolds numbers. Journal of Fluid Mechanics 133(1),
265–285 (1983)
[11] Vickery, B.J.: Fluctuating lift and drag on a long cylinder of square cross-section in a
smooth and in a turbulent stream. Journal of Fluid Mechanics 25(3), 481–494 (1966)
[12] Knisely, C.W.: Strouhal numbers of rectangular cylinders at incidence: a review and new
data. Journal of Fluids and Structures 4(4), 371–393 (1990)
[13] Benini, E., et al.: Unsteady aerodynamics of stationary elliptic cylinders in subcritical flow.
AIAA Journal 8(10), 1814–1821 (1970)
[14] Lauterbach, A., et al.: Microphone array wind tunnel measurements of Reynolds number
effects in high-speed train aeroacoustics. International Journal of Aeroacoustics 11(3),
411–446 (2012)
[15] Thompson, D.: Railway noise and vibration: Mechanisms, modelling and means of control.
Elsevier, Oxford (2008)
Analysis of Aerodynamic and Aeroacoustic Behaviour
of a Simplified High-Speed Train Bogie
1 Introduction
For high-speed trains, aerodynamic noise becomes significant when their speeds
exceed 300 km/h and can become predominant at higher speeds or with the reduction
of the rolling noise [1, 2]. The aeroacoustic behaviour of high-speed trains needs
further study, especially numerical investigations which can reveal more information
on the flow physics. Predicting the aerodynamic noise in an industrial context is still
very difficult due to the large computational resources required for unsteady
numerical simulations [3]. The aim of this paper is to study the flow behaviour and
the aerodynamic noise generation and radiation mechanisms from a simplified bogie
using computational fluid dynamics (CFD). The delayed detached-eddy simulation
(DDES) turbulence model [4] is used together with the Ffowcs Williams-Hawkings
(FW-H) acoustic analogy for radiated sound prediction [5]. DDES avoids grid-
induced separation and preserves the Reynolds-averaged Navier-Stokes mode
throughout the boundary layer [4].
As an initial step, the ground effect and bogie cavity are neglected here. The work
commences with the flow behaviour and aeroacoustic characteristics around an
isolated wheelset before progressing to tandem wheelsets and then a simplified bogie
model. A 1:10 scale case with simplified geometry is modelled with inlet velocity of
30 m/s for comparison with wind tunnel tests. The configuration of the simplified
bogie is displayed in Fig. 1. The isolated wheelset and tandem wheel-sets have the
same geometry as the simplified bogie but without the frame.
2 Numerical Setup
For the simplified bogie case making use of symmetry along the axle mid-span, the
domain has dimensions of 17.7D, 10D and 6.3D (D is the wheel diameter) along the
streamwise, vertical and spanwise directions, respectively. The boundary conditions
are applied as follows: the upstream inlet flow is represented as a steady uniform flow
with a low turbulence intensity; the top, bottom, axle mid-plane and side boundaries
are specified as symmetry boundaries which are equivalent to zero-shear slip walls; A
pressure outlet with zero gauge pressure is imposed at the downstream exit boundary
and all solid surfaces are defined as stationary no-slip walls.
The influence of mesh resolution has been investigated based on the flow
simulated around a circular cylinder representing the axle. The influence of spatial
resolution has been compared by using different grid points in x-y plane (named cases
of ‘Coarse’, ‘Baseline’ and ‘Fine’) and z direction. The differences in terms of results
between cases ‘Baseline’ and ‘Fine’ are smaller than those between cases ‘Coarse’
and ‘Baseline’. There is little difference in the results between the ‘Baseline’ case and
the case with fine resolution in the z direction. Using a fine time step also made little
difference to the results. Therefore, the fully structured mesh based on the ‘Baseline’
grids is generated for each case. Inside the boundary layer, the wall-normal grids are
stretched with a growth ratio of 1.1 and the first grid point is set 10-5 m from the wall
to yield the maximum value of y+ (the dimensionless first-cell spacing) less than 1,
which is adequate for the low-Reynolds number turbulence model being used. The
final mesh used for the simplified bogie case has 14.5 million grid points. The
physical time step size is 10-5 s to ensure a sufficiently small Courant-Friedrichs-Lewy
(CFL) value of less than 2 for most of the computational domain, indicative of
adequate temporal resolution for the simulation to be converged at each timestep.
CFD calculations have been performed using the open source software OpenFOAM-
2.2.1.
Analysis of Aerodynamic and Aeroacoustic Behaviour 491
3 Aerodynamic Results
For isolated wheelset case, Fig. 2 visualizes the flow structures represented by the iso-
surfaces of the normalized second invariant of velocity gradient Q (at level of 50) and
coloured by the velocity magnitude. It can be seen that the flow past the isolated
wheelset is characterized by alternating vortex shedding with different sizes and
orientations. The incoming inflow separates and reattaches on the flat wheel side
surface where a crescent-shaped separation bubble appears and horseshoe-shaped
eddies are formed and convected downstream.
Fig. 2. Iso-surface of the instantaneous normalized Q criterion (left: side view; right: top view)
Fig. 3 shows the wake structure for the tandem wheelsets. The vortices are shed
from the upstream wheelset, impinge on the downstream wheelset, deform largely and
are merged into eddies formed behind the rear wheelset, making the wake of the
downstream wheelset highly turbulent.
The instantaneous non-dimensional spanwise vorticity fields (ωZ) in the front and
rear axle wake area of the tandem wheelsets are displayed in Fig. 4. It can be noted
here that the vortices shed alternately from the upstream axle impinge on the
downstream axle and all vortices are mixed up behind the rear axle, leading to the
highly random behaviour of the downstream axle wake.
492 J.Y. Zhu, Z.W. Hu, and D.J. Thompson
Fig. 4. Contours of instantaneous spanwise vorticity fields in vertical plane through centre of
axle
The wake structure for the simplified bogie is visualized in Fig. 5. Different with
the tandem wheelsets case, the streamwise ‘rib’ vortices from the upstream axle
between the wheels are distributed obliquely along the streamline direction since the
turbulent flow develops more quickly close to the mid-span axle region as there is
much less blockage far away from the wheel-frame area.
4 Aeroacoustic Results
Based on the near-field unsteady flow data obtained from CFD, the FW-H method
can predict far-field sound using equivalent acoustic sources such as monopoles,
dipoles, and quadrupoles [5, 6]. The power spectral density (PSD) of the sound
pressure is computed from predicted sound pressure by segmental average (50 %
overlap) using a Hanning window applied to each segment with a frequency
resolution of 6 Hz. The noise directivity is obtained based on overall sound pressure
level (OASPL) calculated for the frequency range below 5 kHz. The receivers are
distributed uniformly on a circular frame with radius 2.5 m at an interval of 5º to
measure the noise directivity from the wheelset centreline along the vertical y-z plane
and horizontal x-z plane as represented in Fig. 1.
Analysis of Aerodynamic and Aeroacoustic Behaviour 493
Fig. 6. Noise spectra from half wheelset components Fig. 7. Noise spectra from half wheelset
Fig. 6 shows the spectra of the noise radiated from the wheel and the axle
separately (as part of the half wheelset) at the side microphone located at (2.5m,
0.35m) in the vertical y-z plane and Fig. 7 shows the result from the entire half
wheelset. Tonal noises are found with dominant frequencies at 311 Hz (Strouhal
number St of 0.18 based on the axle diameter) and 622 Hz (St of 0.36) corresponding
to the axle periodic vortex shedding and the interaction between the vortex shedding
from the axle and the unsteady flow around the wheel in the drag direction. The sound
radiation generated from the wheel in the presence of the axle is mainly associated
with the oscillating drag forces, whereas the noise generation from the axle mainly
corresponds to the oscillating lift forces. As is well known, the aerodynamic forces
acting in the vertical direction fluctuate with a larger amplitude at half the frequency
of those along the streamwise direction.
Fig. 8. Comparisons of spectra of radiated noise Fig. 9. Noise directivity from isolated wheelset
from simulation and experiment (vertical y-z plane)
494 J.Y. Zhu, Z.W. Hu, and D.J. Thompson
Fig. 8 compares the measured spectrum of the radiated noise at the top receiver
(α = 90º) in the vertical y-z plane with a measurement taken in an anechoic wind
tunnel under similar conditions from simulation. Good agreement can be found for the
dominant frequency of tonal noise and the shape of the spectrum.
For the isolated wheelset case, Fig. 9 shows the noise directivity in the vertical y-z
plane normal to the flow direction (in this plane, the angle α = 0º or 180º corresponds
to the side of the train). This reveals that a typical dipole sound is generated by the
flow separation from the wheelset top/bottom surfaces and radiates predominantly in
the vertical direction.
Figs. 10 and 11 display the noise directivities from the front and rear wheelset in
the vertical y-z plane. It can be seen that the noise level from the front or rear wheelset
of the tandem wheelsets and simplified bogie are very close except at the horizontal
plane through the axle centreline (α = 0º or 180º) where the presence of the frame
causes differences between them (3 dB for front wheelset and 0.8 dB for rear
wheelset). Compared with the front wheelset, the noise radiated from the rear
wheelset is up to 9 dB lower except near α = 0º or 180º. This is because the turbulent
flow convected from the front wheelset is dissipated and the trailing wheelset is
subject to a lower mean incident flow velocity. Thus, the sound generated from
periodic shedding at the rear wheelset may be lost and replaced by a broadened
spectrum with a lower level. Nevertheless, in the horizontal x-z plane, the flow
separation is stronger from the downstream wheelset along the lateral side, leading to
about 3 dB higher noise level than from the upstream wheelset for tandem wheelsets
and slightly larger for simplified bogie.
Fig. 10. Noise directivity from front wheelset Fig. 11. Noise directivity from rear wheelset
(vertical y-z plane) (vertical y-z plane)
The directivities of radiated noise from the front and rear wheelset in the horizontal
x-z plane along the flow direction are illustrated in Figs. 12 and 13. Since the bogie
frame changes the flow behaviour around the wheelset, the noise generated from the
front wheelset of the bogie is up to 4 dB higher than in the tandem wheelsets case and
it is about 1 dB larger for the rear wheelset. Note that Figs. 10 and 12 show a vertical
Analysis of Aerodynamic and Aeroacoustic Behaviour 495
dipole pattern of directivity for the sound radiation of the upstream wheelset, whereas
Figs. 11 and 13 indicate a directivity with more similar magnitude in each direction
for the downstream wheelset. This is due to the laminar separation occurs at the front
wheelset and thereby the periodic shedding is generated at the wheelset top/bottom
surface, whereas the rear wheelset is submerged in the upstream wheelset wake and
situated in a turbulent condition, and thus the irregular flow feature makes the noise
radiation more broadband and multi-directional. Additionally, it also shows that the
amplitude of the noise radiated from the rear wheelset is much smaller compared to
the front wheelset.
Fig. 12. Noise directivity from front wheelset Fig. 13. Noise directivity from rear wheelset
(horizontal x-z plane) (horizontal x-z plane)
Fig. 14. Noise directivity from front bogie Fig. 15. Noise directivity from rear bogie
(vertical y-z plane) (vertical y-z plane)
The directivities of radiated noise from the front and rear bogie in the vertical y-z
plane are displayed in Figs. 14 and 15. Here ‘front bogie’ represents the upstream
wheelset with the front half frame and ‘rear bogie’ means the downstream wheelset
496 J.Y. Zhu, Z.W. Hu, and D.J. Thompson
and the rear half frame. This demonstrates that the frames are minor sources
compared to the wheelsets. The noise radiation of the rear frame is 2.5-4.5 dB larger
than from the front frame because of the stronger vortex shedding and flow separation
at the frame ends. Also, it can be seen that the noise radiated from the rear bogie is
weaker (up to 7.3 dB) than from the front bogie except at α=0º or 180º where the
noise is 0.7 dB higher from the rear bogie.
In order to interpret these results for a full scale geometry, it may be noted that the
tonal noise peaks presented here will occur at 1/10 of these frequencies and the noise
levels will increase in proportion to the surface area (factor 100). At a running speed
of 300 km/h (83 m/s), the noise levels will increase proportionally to the flow speed
to the power of 6; and in the presence of turbulent flow, the tonal components are
likely to be less significant and the broad-band component is likely to increase in
importance. The geometry of a real train will also lead to more complex flow
structures which will affect the noise radiated.
5 Conclusions
It is found that the flow past an isolated wheelset has a complex three-dimensional
wake. The primary behaviour of the flow past the tandem wheelsets and the
simplified bogie is that the vortices shed from the upstream bodies are convected
downstream and impinge on the downstream ones, leading to highly turbulent wake
of the downstream bodies. In the isolated wheelset case, tonal noises are generated
with dominant frequencies corresponding to the lift and drag dipoles. Furthermore, a
vertical directivity pattern of noise generation is predicted for the isolated wheelset
and the front wheelset of tandem wheelsets and simplified bogie. The rear wheelset
has an omni-directional directivity and its sound radiation is generally weaker
compared to the front wheelset.
References
[1] Thompson, D.J.: Railway noise and vibration: mechanisms, modelling and means of
control. Elsevier, Oxford (2008)
[2] Talotte, C.: Aerodynamic noise: a critical survey. Journal of Sound and Vibration 231(3),
549–562 (2000)
[3] Spalart, P.R., Shur, M.L., Strelets, M.K., Travin, A.K.: Initial noise predictions for
rudimentary landing gear. Journal of Sound and Vibration 330, 4180–4195 (2011)
[4] Spalart, P.R., Deck, S., Shur, M.L., Squires, K.D., Strelets, M.K., Travin, A.K.: A new
version of detached-eddy simulation, resistant to ambiguous grid densities. Theoretical and
Comp Fluid Dynamics 20, 181–195 (2006)
[5] Ffowcs-Williams, J.E., Hawkings, D.L.: Sound radiation from turbulence and surfaces in
arbitrary motion. Philosophical Transactions of the Royal Society of London 342, 264–321
(1969)
[6] Farassat, F.: Derivation of Formulations 1 and 1A of Farassat. NASA/TM-214853 (2007)
Derivation of Sound Emission Source Terms for High
Speed Trains Running at Speeds in Excess of 300 km/h
1 Introduction
2 Background
For conventional railways, depending on speed and train specification, the dominant
sound sources are rolling sound (the interaction between the wheels of the train and the
rail), power, traction and auxiliary systems. For trains running at high speeds (typically
defined as > 250 km/h), aerodynamic sound may significantly contribute to the overall
passby sound level. However, recent papers show that rolling noise is still important up
to 350 km/h. Aerodynamic sound is caused by the flow of air over the train body as it
travels at high speed. The most important sources of aerodynamic sound on high speed
trains vary from one train to another, but usually include [1]:
• the bogies, particularly the leading bogie; and
• the pantograph, its recess in the roof, and any other roof-mounted equipment
such as insulators.
Other important sources can be the nose of the train, gaps between coaches,
ventilation grills, projections (door handles, steps, etc.) and cavities (that can have
resonant responses). Sources located towards the top of the train are particularly
important when wayside noise barriers are present.
The level of aerodynamic sound increases more rapidly with the train speed V than
rolling sound. Rolling sound is typically assumed to have a speed dependence of
30log10V for LpAeq,Tp and LpAF,max (20log10V for SEL) [1]. Aerodynamic sound is
typically assumed to follow 60log10V [1, 2, 3], although [4] suggests a speed
dependence of 70log10V.
Given the importance of aerodynamic sound at high operational speeds, existing
train passby sound prediction methods have been modified, and new methods
developed to take aerodynamic sound into account. Examples are the German Schall 03
[5], Dutch RMR [6] and Nordic 2000 [7, 8]. A common noise assessment method in
Europe (CNOSSOS-EU) [2] has been proposed, but not yet implemented, by the
European Commission for strategic noise mapping under the Environmental Noise
Directive 2002/49/EC.
All these methods have a common concept – sound from a train passby is assumed to
emanate from a set of discrete sources situated at different heights above rail head. The
sound source powers are normally derived from national databases of passby
measurements of operational rolling stock.
The source height is an important factor, especially when considering noise barriers.
Typically assumed source heights are 0.0 m and 0.5 m for rolling sound, 0.5 – 4.0 m for
traction and auxiliaries and 0.0 – 5.0 m for aerodynamic sound.
In the UK, a calculation method [9] was developed to predict the noise impacts from
the Channel Tunnel Rail Link (HS1). The model was validated in France and UK with
an extensive series of noise measurements taken on the LGV. The method predates
the UK Calculation of Railway Noise [10] (CRN) although much of the same data was
used to develop CRN. The HS1 method is being used for the assessment of HS2
because: it predicts maximum sound levels (LpAFmax) as well as LpAeq,T levels; the
method was used to successfully design and deliver HS1; measurements have shown
that it provided a conservative estimate of actual in-service sound levels; and its
Derivation of Sound Emission Source Terms for High Speed Trains Running at Speeds 499
forecasts for maximum levels fit well with measurements made recently on LGV-Est at
distances out to 1 km [11].
The HS1 method assumes all sound originates from a source height of 0.5 m above
rail head. For trains running at very high speeds (> 300 km/h) a multiple source version
of the method is required.
2.1 TSI
HS2 is being designed under the Interoperability Directive, and hence sound emissions
from rolling stock would need to satisfy the limits specified in the rolling stock
technical specification for interoperability (TSI) of the trans-European high-speed rail
system [12].
The high-speed TSI subdivides exterior sound emitted from rolling stock into
stationary sound, starting sound and pass-by sound. Stationary sound is highly
influenced by auxiliaries, such as cooling systems, air conditioning and compressors.
Starting sound is a combination of contributions from traction components such as
cooling fans, and auxiliaries. Pass-by sound includes both rolling and aerodynamic
sound.
The limit for starting sound for Class 1 electric trainsets is 85 dBLpAFmax, defined at a
distance of 7.5 m from the centreline of a reference track, 1.2 m above railhead. The
limits for pass-by sound are 87 dBLpAeq,Tp at 250 km/h, 91 dBLpAeq,Tp at 300 km/h and
92 dBLpAeq,Tp at 320 km/h, all defined at a distance of 25 m from the centreline of a
reference track, 3.5 m above railhead. A margin of 1 dB(A) is accepted on these values.
The noise TSI reference test track includes specific requirements on rail roughness and
decay rate of the rail. Depending on the level of maintenance of wheel and track, this
means that noise TSI test track may not represent emission levels in-service, where
wheel and track can be noisier.
The HS2 method builds upon the HS1 method by introducing a multi-source concept
similar to other noise prediction methods [5 – 8].
Following a review of the different prediction methods, the following five sources
have been included in the HS2 method:
1. rolling sound, at a height of 0.0 m above rail head, which includes sound
emitted by the wheels and the track;
2. body aerodynamic sound, at a height of 0.5 m above rail head, which includes
sound generated by flow in the lower regions of the train;
3. starting sound, at a height of 2.0 m above rail head, which includes sound
generated by power, traction and auxiliary systems;
4. pantograph recess sound, at a height of 4.0 m above rail head; and
5. raised pantograph, at a height of 5.0 m above rail head.
500 T. Marshall, B.A. Fenech, and R. Greer
Because HS2 trains have not yet been procured, the source terms for these five
sources had to be derived, based upon TSI, published literature and measurements. The
following sections present the derivation process for a current scenario (base case),
where sound levels are the maximum permitted by statuary guidance, and a more
socially and economically viable scenario (assumed mitigated case), where both the
rolling stock and infrastructure incorporate proven “noise reduction at source”
technologies.
3.1 Assumptions
The rolling stock for HS2 is assumed to consist of 200 m long trainsets. The trainsets
would be distributed power (EMUs), and none of the vehicles would have cast iron
tread brakes. Traditional bogie architecture is assumed (articulated bogie architectures
could be considered as a form of noise mitigation).
The speed dependence for aerodynamic sound was assumed to follow 70log10V for
LpAeq,Tp and LpAF,max (60log10V for SEL) to allow for a conservative extrapolation of
maximum sound levels for speeds in excess of 320 km/h. The SEL relationships for all
five sources are:
• RSEL + 20log10V for rolling sound [1, 2, 4, 9, 10];
• BSEL + 60log10V for body aerodynamic sound [4];
• SSEL − 10log10V for starting sound [e.g. 10]; and
• PSEL + 60log10V for pantograph and pantograph recess sound [4],
where RSEL, BSEL, SSEL and PSEL are constants and V is the train speed.
Each base case 200 m trainset is assumed to feature one raised pantograph and one
pantograph recess (with a lowered pantograph).
sound than a TSI reference track. The constant RSEL was increased by +3 dB(A) to
allow for this in the assessment method.
A separate set of equations was developed by the same process described for LpAF,max
(and hence LpAeq,Tp). These equations use 30log10V speed dependence for rolling noise
and 70log10V speed dependence for aerodynamic sound.
There is limited published information on the absolute level of sound radiated by high
speed train pantographs. Lölgen [14] presents data for an old crossed-arm type
pantograph DSA350SEK, and a low-noise prototype actively controlled single arm
pantograph (ASP) with either one or two contact strips. Levels normalized to 320 km/h
measured 25 m from the line suggest a maximum passby noise level of around
90 dB(A) for the DSA350SEK, and around 75 to 80 dB(A) for the ASP with two pan
heads and optimised insulators. A reduction of around 3 dB(A) was measured in
changing from two panheads to one. Nagakura et al [3] show that the 700 series trains,
with their low noise pantographs, exhibit pantograph aerodynamic noise emissions that
are around 5 dB(A) lower than the earliest bullet (Shinkansen) trains, and have a
maximum noise level around 70 to 75 dB(A) at a distance believed to be 25 m from the
line at 300 km/h. Ikeda et al [15] present results that show that pantographs designed for
the E5 and 700N stock, in service since the end of 2011, are around 4 dB(A) quieter
than the equipment on the 700 series trains. Gautier et al. [16] present simulated passby
sound levels for a pantograph recess and raised pantograph on a TGV exhibiting
maximum pass-by levels of 87 and 85 dB(A) respectively at 320 km/h.
The published data therefore shows that the maximum aerodynamic noise level from
a pantograph pass-by is around 85 dB(A) at 25 m from the line for a current European
HS pantograph at 320 km/h and that this level can be reduced to 75 dB(A) or potentially
less with more aerodynamic pantographs. This ~10 dB reduction in pantograph noise at
high speed is also cited by other work, e.g. [17].
Passby sound measurements of train running behind noise barriers can provide
additional information on the relative contribution of pantograph aerodynamic noise.
The validation of the HS1 prediction method [9] with its 0.5 m source height showed
that the in-situ barrier insertion loss recorded for a range of noise barriers (including
bunds) up to 4 m above rail was not affected by contribution from the pantograph
aerodynamic noise at speeds up to 300 km/h. Belingard et al. [18] present results
showing that the barrier insertion loss provided by a 2.1 m high reflective barrier for a
TGV-POS was reduced by 1 dB as train speed increased from 320 to 375 km/h.
These outcomes were recreated in a series of multiple source barrier insertion loss
calculations assuming the rolling and body aerodynamic source terms derived in
Sec.3.2 together with each pantograph source level described in this section. These
support a maximum pass-by aerodynamic sound source level of 83 dB(A) for a current
European HS train running at 320 km/h measured 25 m from the track.
502 T. Marshall, B.A. Fenech, and R. Greer
The TSI limit for starting sound was converted to its equivalent at a distance of 25 m
from the track centreline, 3.5 m above railhead assuming a 7 dB(A) level difference
between the two positions. To determine the SEL source term, a distributed power train
with a configuration of [M–T–M–T–M–M–T–M–T–M] was assumed, where M
denotes a motor vehicle with two starting sources (one at each bogie), and T denotes a
trailer vehicle with no starting sources. A time domain sound model was used to
determine the SEL of a distribution of starting sound sources, assuming the TSI LpAF,max
limit is met.
The mitigated case was developed as a number of corrections to constants RSEL, BSEL,
SSEL and PSEL to represent currently proven noise at source mitigation technologies.
These corrections are presented in Table 1.
Table 1. Source values for the model and initial corrections assumed for mitigated case
4 Discussion
Fig. 1 shows the predicted passby sound level for an HS2 train assuming base case (just
TSI compliant train on in-service track) and the initial mitigated case, as a function of
train speed. Also shown are the current TSI limits, and data measured for the Deufrako
and NOEMIE projects [1] . The base case models the noisiest trains well – these consist
of older 1st generation European HS trains. The mitigated case, which represents what
can be achieved using already proven mitigation intervention in use around the world,
is 2-3 dB lower than the 1st generation European HS train data measured between 300 –
360 km/h, and 3 dB lower than base case at 360 km/h.
The sound modeling has demonstrated that controlling noise at source is much more
effective at minimising noise levels at communities than other forms of mitigation such
as noise barriers and bunds. For example, a mitigated train running behind a 3 m
Derivation of Sound Emission Source Terms for High Speed Trains Running at Speeds 503
110.0
90.0
LpAeq,Tp, dB
85.0
80.0
75.0
70.0
65.0
60.0
100 200 300 400
train speed, km/h
Fig. 1. LpAeq,Tp vs speed for total passby sound at 25 m from the track predicted using the HS2
base case and assumed mitigated case. The red square markers show the current TSI limits
(including the +1 dB allowance). The black markers show measured sound levels for TGV-A,
TGV-Duplex and Thalys, reproduced from [1].
References
[1] Thompson, D.: Railway noise and vibration. Mechanisms, modelling and means of control.
Elsevier (2009)
[2] European Commission. Joint Research Centre Reference Reports. Common noise
assessment methods in Europe (CNOSSOS-EU) (2012)
[3] Nagakura, K., Zenda, Y.: Prediction model of wayside noise level of Shinkansen. Railway
Technical Research Institute Japan (2003)
[4] US Department of Transportation: High-speed ground transportation noise and vibration
impact assessment (2012)
[5] SCHALL 03 2006, Richtlinie zur Berechnung der Schallimmissionen von Eisenbahnen
und Straßenbahnen, Entwurf, Stand (December 21, 2006)
[6] Ministerie van Volkshuisvesting: Ruimtelijke Ordening en Milieubeheer. Reken- en
meetvoorschriften railverkeerslawaaai 1996 (Calculation and measurement requirements
for railway traffic 1996) (2001)
504 T. Marshall, B.A. Fenech, and R. Greer
[7] Zhang, X.: Prediction of high-speed train noise on Swedish tracks. SP Technical Research
Institute of Sweden, SP report 2010, 75 (2010)
[8] Brekke, A., et al.: The Norwegian high speed rail study. In: Proc. Joint Baltic-Nordic
Acoustics Meeting (2013)
[9] Hood, R.A., et al.: Calculation of railway noise. Proc. of the Institute of Acoustics 13(8)
(1991)
[10] Department for Transport. Calculation of Railway Noise (1995)
[11] Department for Transport. HS2 Phase One environmental statement volume 5: sound,
noise and vibration: Appendix SV-001-000: Annex D2 (2013)
[12] 2008/232/CE: Commission Decision of 21 February 2008 concerning a technical
specification for interoperability relating to the ‘rolling stock’ sub-system of the
trans-European high-speed rail system (2008)
[13] Gautier, P.-E., et al.: High speed trains external noise: Recent results in the TGV case. In:
Proc. 19th International Congress on Acoustics (2007)
[14] Lölgen, T.: Wind tunnel noise measurements on full-scale pantograph models. J. Acoust.
Soc. Am. 105(2), 1136–1136 (1999)
[15] Ikeda, M., Mitsumoji, T., Sueki, T., Takaishi, T.: Aerodynamic noise reduction of a
pantograph by shape-smoothing of panhead and its support and by the surface covering
with porous material. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson,
J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation
for Rail Transportation Systems. NNFM, vol. 118, pp. 419–426. Springer, Heidelberg
(2012)
[16] Gautier, P.E., et al.: High speed trains external noise: A review of measurements and
source models for the TGV case up to 360 km/h. In: 8th World Congress on Railway
Research, Seoul (2008)
[17] Asplan Viak, A.S.: A methodology for environmental assessment – Norwegian high speed
railway project Phase 2 (2011)
[18] Belingard, P., Poisson, F., Bellaj, S.: Experimental study of noise barriers for high-speed
trains. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T.,
Schulte-Werning, B., Thompson, D., de Vos, P., et al. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 495–503. Springer,
Heidelberg (2012)
[19] Tsukanishi, M., Toyooka, M., Mori, T., Yano, H.: Development to decrease noise in place
along railway-tracks of Sanyo Shinkansen. In: Maeda, T., Gautier, P.-E., Hanson, C.E.,
Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise
and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 383–388.
Springer, Heidelberg (2012)
[20] Frid, A., Åbom, M., Bouvet, P.: Noise reduction for cooling systems on railway vehicles.
Silence Final Event (2008)
[21] Oertli, J.: The STAIRRS project, work package 1: a cost effectiveness analysis of railway
noise reduction on a European scale. Journal of Sound and Vibration 267 (2003)
[22] European Railway Agency: Preliminary draft 1.0 of the Technical Specifications of
Interoperability relating to Noise (NOI TSI). ERA/CON/2013-01/INT
Mastering Micro-Pressure Wave Effects at the
Katzenbergtunnel –
Design of Measures, Prediction of Efficiency
and Full-Scale Test Verification
DB Systemtechnik GmbH,
Voelckerstrasse 5, 80939 Munich, Germany
michael.hieke@deutschebahn.com
Summary. This paper deals with the first and comprehensive application of a
new prediction and assessment method on the new Katzenbergtunnel. The tunnel
follows a new safety concept with two single track tubes and is strongly affected
by the micro-pressure wave (MPW) phenomenon. Thus, appropriate MPW
countermeasures were designed in the early construction phase. At this stage, the
effects could only be predicted by model scale tests and simulations. During the
final homologation tests, the values were measured in full scale and compared to
the predicted ones. There was a good correlation so the tunnel could be opened
for operation in December 2012 without any further modifications.
In this paper the implemented countermeasures are presented. Also, some
comparisons between predicted and measured values are shown. A detailed
description of the assessment method which was adopted in a guideline of DB
Netz AG is given in a separate paper.
1 Introduction
In winter 2005/2006, when first test runs had been carried out on the new German
high-speed line Nuremberg-Ingolstadt (300 km/h line speed), DB had to face an
unexpected issue of strong micro-pressure waves [1] emitted from the portals of two
long tunnels. The emissions were in an order of magnitude which was not acceptable.
There was an urgent need for suitable countermeasures to retrofit these tunnels. The
issue of micro-pressure wave on new Nuremberg-Ingolstadt line and the successfully
taken countermeasures were presented at the 9th IWRN [2].
Following this incident, DB Systemtechnik developed a methodology and a set of
simulation tools to predict micro-pressure waves for other future German tunnels.
Furthermore, an authority approved regulation framework including reference values
for acceptable micro-pressure wave emissions had been developed by DB
Systemtechnik [3, 4].
The Katzenbergtunnel was the first tunnel that the prediction tools were applied to.
A prediction of micro-pressure wave emissions was done and the predicted values were
checked by full scale measurements during the homologation tests.
2 The Katzenbergtunnel
The Katzenbergtunnel on the new high-speed line Karlsruhe-Basel (250 km/h line
speed) was the first German tunnel following the new safety rules requiring a set of two
independent single track tubes. The tunnel has a length of 9385 m and a free cross
sectional area of 60.5 m2 per tube. There are air shafts from each tube to the upper
environment roughly at the tunnel center. Every 500 m there are evacuation chambers
between the tubes but these are closed by default. A slab track with a smooth surface is
installed in the tunnel to allow rescue road vehicles to drive into the tunnel for
evacuation purposes.
Both portal regions were planned to be built by cut-and-cover which means that the
portal regions were completely filled afterwards until only the plain portals remain
visible. This construction technique allowed to model appropriate underground portal
hoods which could fit into the free space of the open construction. On the one hand this
helps to reduce costs because the free space had to be filled anyway. On the other hand
additional chambers and ducts were necessary to bring the displaced air from the hoods
to the free environment.
The definition of the hood size, the shape and position of the ducts and chambers
were part of a deeper Computational Fluid Dynamics (CFD) investigation to reduce
micro-pressure wave effects to an acceptable value.
All tunnel portals have been equipped with portal hoods because the tunnel is equipped
with tracks for two-way working. Due to the cut-and-cover construction technique, all
hoods could be placed below ground and a set of ducts and chambers was implemented
to allow compressed air from the tunnel to leave. There are different constructions at
the north and at the south portal due to project constraints. At the north portal, the ducts
and chambers were placed on top of the hoods. At the south portal, they were placed
sideways in the middle region between the hoods.
The north portal hoods have a length of 50 m and a free cross sectional area of 90 m2
each. In the top wall of each hood there were 4 times 3 slots which lead into 4 chambers
which are connected to the free environment. The hood design is shown in Fig. 1.
Mastering Micro-Pressure Wave Effects at the Katzenbergtunnel 507
Fig. 1. North portal hoods – top: aerial view (picture by DB Projektbau), bottom: schematic drawing
with slots and chambers
The south portal hoods have a length of 50 m and a free cross sectional area of
90 m2. They are shown in Fig. 2. The 4 times 3 slots in each hood were mounted in the
side wall. Due to a large distance between the hoods, the chambers were placed
between the hoods with opening to the environment at the front. A large so called lean
bow was placed at the front to prevent the train’s head pressure wave entering the
chambers directly. In the large empty space in the cut-and-cover region behind the
hoods two reflection chambers were implemented which lengthen the entry
compression wave even more.
The slots in all hoods were designed a bit larger than necessary and equipped with
metal frames. A set of slides mounted on the frames was used to close some slots partly
to optimize the entry pressure wave even further. An example of the slides in the south
portal hood is shown in Fig. 3.
508 M. Hieke, C. Gerbig, and T. Tielkes
Fig. 2. South portal hoods – top: front view (picture by DB Systemtechnik), bottom: schematic
drawing with slots, chambers and additional reflection chambers between hood and tunnel
Fig. 3. Slides for variable slot adjustment in the south portal hood (picture by DB Projektbau)
Mastering Micro-Pressure Wave Effects at the Katzenbergtunnel 509
A first rough assessment showed the substantial need for countermeasures which, in a
second step, were adapted to the advanced planning of the tunnel. It was found that the
maximum entry pressure gradient needed to be reduced by around 70 %. The basic
design of the countermeasures was investigated firstly by moving model rig tests. With
the knowledge of the measured entry compression wave signal, a prediction of the
wave steepening and the emission of the micro-pressure wave was done. The prediction
was repeated using CFD simulations at a later stage when more advanced tools were
available at DB Systemtechnik. At this stage the detailed opening configuration of the
slides was also optimized. Finally, during the test runs for homologation, a set of
full-scale measurements was done to prove compliance with the prediction values to
the regulatory authorities.
A first set of experimental tests on the hood and chamber principle were carried out in
the moving model rig (TRAIR) at Derby (UK) under the ownership of DeltaRail. The
tests comprised both portals with different opening scenarios and different train speeds.
The train model in these investigations was a simplified ICE3 which is currently the
fastest German high-speed train. The train and the hoods were scaled by a factor of
1/28.8 to reproduce the correct blockage of the Katzenbergtunnel.
As more advanced simulation techniques have become available, the generation of
the entry compression wave was simulated by transient 3-dimensional CFD described
in [5] and the final opening configuration was found. Full-scale measurements have
been done during the homologation runs to measure the entry compression wave
directly.
In Fig. 4 the simulated and measured entry compression waves are shown.
CFD
TRAIR
pressure gradient
Fullscale
pressure
time time
Fig. 4. Predicted and measured entry compression waves at the south portal (left: pressure signals,
right: pressure gradients)
510 M. Hieke, C. Gerbig, and T. Tielkes
A direct comparison of pressures and pressure gradients shows a quite similar shape
of the waves, but the extreme values differ more. Comparing the pressures between
moving model and full scale tests, a good correlation could be found except for the rear
part of the wave. There, the influence of the reflection chambers became visible. The
wave propagation in the model scale occurred at the same speed of sound as in full scale
so the time for a reflection was much shorter in model scale so the pressure decrease in
the rear part happened earlier.
The simulated wave doesn’t represent the short pressure drop in the rear part
correctly. It is possible that the wave propagation and reflection in the reflection
chambers are not modeled correctly in the CFD. For these kind of chambers further
refinement is needed in future studies.
A comparison of the gradients shows a very good correlation between CFD and full
scale tests in the front part of the wave and bigger differences at the rear, mainly due to
the effects discussed above. The difference between the maximum gradients in the
relevant front part in simulation and full scale measurement is below 2 % which is a
very good correlation.
The different gradients in the model scale tests can be explained by a non-optimized
opening configuration, as the measurements were done at the very beginning of the
investigations and due to the choice of a different train type. In the moving model tests
an ICE3 was used while the CFD and the full scale tests were done with an ICE1.
With the knowledge of the entry compression wave, a prediction of the wave
steepening was done and compared to full scale measurement. The tunnel was equipped
with pressure sensors close to the south portal, one at 314 m inside the tunnel, one at a
distance of 3696 m inside the tunnel and 1092 m in front of the air shaft in the tunnel
center and one at a distance of 5693 m inside the tunnel.
P1 meas
P2 meas
pressure gradient
P3 meas
pressure
P1 sim
P2 sim
P3 sim
time time
Fig. 5. Comparison of simulated (lines) and measured values (points) at different positions in the
tunnel (P1@ 314 m, P2 @ 3696 m, P3 @ 5693 m)
A 1-dimensional calculation tool was used for the simulations as described in [5].
The empirical friction parameters in the model were set to standard values in a first
attempt. An adjustment to measured values was made afterwards. It was observed that
Mastering Micro-Pressure Wave Effects at the Katzenbergtunnel 511
the friction parameters were slightly increased. It was not clearly identified if the reason
for the increased friction was the tunnel interior with tubing segments and lots of gaps
and holes or if it was due to the decreased hydraulic diameter in the Katzenbergtunnel
compared to the Euerwangtunnel where the original friction parameters were derived
from.
In Fig. 5 a comparison between the measured and predicted wave steepening is
shown. It must be noted that the starting wave for this steepening simulation was the
measured wave at the entry portal and not the CFD result to show the accuracy of the
steepening model. Regarding the maximum gradients the difference between
prediction and measurement is below 3 %.
The MPW emissions at a distance of 25 m from the exit portal were calculated
according to the approach described in [5] from the steepened compression wave at the
exit portal. A comparison to measured values is shown in Fig. 6.
time
Fig. 6. Comparison of measured and simulated micro-pressure waves
As the measurements were done using microphones with a frequency range (± 2 dB)
from 3.15 to 20000 Hz, the predicted values were also filtered by a 2nd order
Butterworth high-pass filter with 3.15 Hz. Although the signals do not seem to be
equal, the sound pressure levels of interest show a good correlation as described in the
next section.
Table 1. Comparison of predicted and measured MPW emissions at the north portal
The predicted values of LCpeak are higher than measured and represent a conservative
treatment. The correlation is different but still within 1.5 dB(C) for the LCE values. One
explanation for this is that the duration of the measured MPW is longer than the
predicted one, see Fig. 6. If the signal length would be doubled, an increase of up to
3 dB(C) in LCE could be possible. Another explanation is the rolling noise which was
not simulated in the prediction but appeared in the measurements directly after the
micro-pressure wave.
Though being designed simultaneously to the drafting of guidelines, the
Katzenbergtunnel complies with the limits of the new micro-pressure wave regulation
too, as documented by the measurements. The limit value for LCpeak at a distance of
25 m from the portal was 115 dB(C) and for LCE it was 70 dB(C) at the nearest
neighborhood as described in [3].
5 Conclusions
The Katzenbergtunnel was the first German high-speed tunnel that went into operation
with two single track tubes. As the tunnel was affected by the MPW phenomenon and
equipped with tracks for two-way working, all portals were equipped with subterranean
portal hoods to reduce MPW effects. A set of ducts and chambers was also used to
release the compressed air during the train entry. The efficiency of all measures was
predicted and confirmed during homologation tests. For future German tunnels the
same tools and assessment will be used, so the planning reliability can be achieved at an
early design stage.
References
[1] Ozawa, S., Maeda, T., Matsumura, T., Uchida, K., Kajiyama, H., Tanemoto, K.:
Countermeasures to reduce micro-pressure waves radiating from exits of Shinkansen
tunnels. In: Proc. 7th Int. Symp. on Aerodynamics and Ventilation of Vehicle Tunnels,
Brighton, pp. 253–266. BHR Group (1991)
[2] Tielkes, T., Kaltenbach, H.-J., Hieke, M., Deeg, P., Eisenlauer, M.: Measures to counteract
micro-pressure waves radiating from tunnel exits of DB’s new Nuremberg-Ingolstadt
high-speed line. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C.,
Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 99, pp. 40–47. Springer, Heidelberg (2008)
Mastering Micro-Pressure Wave Effects at the Katzenbergtunnel 513
[3] Gerbig, C., Hieke, M.: Micro-pressure wave emissions from German high-speed railway
tunnels - An approved method for prediction and acoustic assessment. In: Nielsen, J.C.O.,
Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D., Tielkes, T., Towers,
D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems.
NNFM, vol. 126, pp. 571–578. Springer, Heidelberg (2015)
[4] Netz, D.B., Modul, A.G.: 853.1002A01 in Richtlinie 853 “Eisenbahntunnel planen, bauen
und instand halten”, available at DB Kommunikationstechnik GmbH, Karlsruhe,
http://dzd-bestellservice@deutschebahn.com
[5] Hieke, M.: Gerbig, Ch., Tielkes, T., Degen, K.G.: Assessment of micro-pressure wave
emissions from high-speed railway tunnels. In: Proceedings WCRR 2011 (2011)
Aerodynamic Noise Reduction of a Pantograph
Panhead by Applying a Flow Control Method
1 Introduction
In Japan, Shinkansen trains run through dense residential areas. Therefore, reduction of
wayside noise caused by Shinkansen trains is necessary to preserve the surrounding
environments along railway lines. Since the sound power of aerodynamic noise grows
in proportion to the 6th to 8th power of the train speed, aerodynamic noise is the
dominant source in the high-speed range. Among aerodynamic noise sources of a
Shinkansen train, the pantograph, especially the panhead (see Fig.1), is one of the
dominant aerodynamic noise sources.
The mechanism of aerodynamic noise generation from the panhead is known to be
the following. When the panhead moves through the air, the flow around it separates
from the panhead surface and Karman vortices shed behind the panhead cause pressure
fluctuations. The pressure fluctuations propagate to the far-field and cause Aeolian
tones. Therefore, if the flow around the panhead can be controlled so that the Karman
vortices are weakened, aerodynamic noise from the panhead will be reduced.
In this study, two flow control methods are applied to the panhead. First, a plasma
actuator is applied to the panhead surface. Second, based on the flow control
mechanism obtained from the study using the plasma actuator, a more practical flow
control method is proposed using air suction.
Horn Panhead
45 mm
Panhead
support Horn
Knuckle
Articulated frames
70 mm
(a) Pantograph (b) Panhead
Fig. 1. Example of low noise Shinkansen pantograph and its panhead
The plasma actuator is a relatively new flow control device, which attracts increasing
attention of aerodynamic researchers. Fig. 2 shows a typical configuration of plasma
actuator. The plasma actuator generates plasma on its surface and produces a body
force to the surrounding air. The body force draws surrounding air and induces
tangential flow. Electric power is supplied to the actuator from a special power supply
equipment, which generates several kV voltage with a frequency of several kHz.
Although this power supply equipment is a special one, other elements such as a
dielectric and electrodes are easily available. For example, in this wind tunnel test,
0.05 mm thick KaptonⓇ (polyimide) tape is used as the dielectric and 0.035 mm thick
copper foil tape is used as the electrodes. In this way, the plasma actuator has many
advantages in that it is very thin, has no complex mechanical structure, and is easy to
set up. The plasma actuator can control only low speed (several m/s or so) flow,
because its output power is not so strong. But, the plasma actuator can perform
innovative flow control effect. Therefore, in this study, the plasma actuator was applied
to the panhead and low speed wind tunnel test was carried out to obtain a new flow
control mechanism which is applicable to the real Shinkansen trains.
The plasma actuator was applied to the half scaled panhead model as shown in Fig. 3.
The plasma actuator was installed at a location directly downstream of the upstream
corners, which are the flow separation points of the panhead. Wind tunnel tests
were carried out at RTRI’s small scale low-noise wind tunnel (the nozzle size is
760 mm × 600 mm, the maximum flow speed is 150 km/h), and the flow control effect of
the plasma actuator was verified by evaluating the flow field behind the panhead. The
flow field was observed by using PIV (Particle Image Velocimetry) technique because it
can obtain flow velocity distribution in the two dimensional plane at once. In the PIV
measurement, the fluid is seeded with oil particles. The particles are irradiated with a
laser sheet and image pairs of the particles are taken by a high-performance CCD camera.
From the time interval and displacement of the particles between the two snapshots, the
flow velocity vector field in the snapshot plane can be obtained. The experimental set up
and the coordinate system of the wind tunnel test are shown in Fig. 4.
Aerodynamic Noise Reduction of a Pantograph Panhead 517
AC
Dielectric
Flow Flow
PA
PA
(a) Without PA (b) With PA
Fig. 3. Applying the plasma actuators to the panhead surface (model scale is 1/2)
PIV Laser
Panhead Model
600mm
500mm
x
y Flow
z
PIV Camera
The plasma actuator was driven by an electric power supply specially designed for it
(KI-Tech., PSI-PG1040F). The voltage and frequency of the AC electric power were
8 kV and 4 kHz respectively. In this experiment, the flow velocity induced by the
plasma actuator was about 1 m/s. The spanwise length of the plasma actuator was
100 mm because of the capacity of the electric power supply equipment.
518 T. Mitsumoji et al.
Since the velocity that can be induced by the plasma actuator is relatively low, it is
desirable to set the free-stream velocity and model scale to be small. Therefore, the
free-stream velocity was set at 4.6 m/s, which was the minimum flow velocity of the
wind tunnel, and the model scale was set at 1/2 so that the spanwise length of plasma
actuator could be sufficiently longer than the cross-sectional scale of the panhead.
Although the free-stream velocity is quite low and the model scale is small compared to
the Shinkansen trains, it is considered that the flow control mechanism obtained by the
plasma actuator is applicable to the real Shinkansen trains.
Figs. 5, 6, and 7 show comparisons of the flow fields measured by PIV technique.
Fig. 5 shows the distribution of the x-directional mean velocity. According to Fig. 5, the
wake region in the case of “With PA” (Fig. 5(b)) is longer than that in the case of
“Without PA” (Fig. 5(a)). Therefore, it can be estimated that the plasma actuator
prevents flow separation from the panhead surface and stabilize shear layer. Fig. 6
shows the distribution of the y-directional RMS velocity. According to Fig. 6, the RMS
velocity in the case of “With PA” (Fig. 6(b)) is smaller than that in the case of “Without
PA” (Fig. 6(b)). These differences and stabilized shear layer are shown more clearly in
Fig.7. Therefore, it can be estimated that the plasma actuator weakens the Karman
vortices and reduces unsteady velocity fluctuations. From these results, the mechanism
of the flow control by using the plasma actuator can be explained as follows. The
plasma actuator has two flow control effects. One is the tangential jet flow itself, which
interferes with Karman vortices. The other is the effect of attracted flow. The tangential
jet flow velocity induced by the plasma actuator (about 1 m/s) is much smaller than the
free-stream velocity of 4.6 m/s. Therefore, the dominant flow control effect is obtained
not by the induced flow itself but by the flow-attracting effect. In other words, the
plasma actuator attracts the surrounding air, prevents the flow separation from
the panhead, and weakens the Karman vortices. If these effects are obtained in the
high-speed range, the plasma actuator can reduce aerodynamic noise. The velocity that
can be induced by the plasma actuator, however, is not sufficient. Therefore, based on
the flow control mechanism obtained from the study using the plasma actuator, a more
practical flow control method is proposed using air suction.
Wake Wake
7
50 50
Flow Flow PA
) )
m m
( 0
m ( 0
m m/s
y y
PA
-50 -50
0 50
x (mm)
100 150 0 50
x (mm)
100 150
-2
(a) Without PA (b) With PA
Fig. 5. X-directional mean velocity distribution (wind speed is 4.6 m/s, X-direction: flow
direction)
Aerodynamic Noise Reduction of a Pantograph Panhead 519
2
50 50
Flow Flow PA
) )
m m
( 0
m ( 0
m
y y m/s
PA
-50 -50
75 75
60 Without PA 60
Without PA
45 With PA 45
With PA
30 30
15 15
y (mm)
y (mm)
0 0
-15 -15
-30 -30
-45 -45
-60 -60
-75 -75
-2 -1 0 1 2 3 4 5 6 0 1 2 3 4 5
X-directional mean flow velocity (m/s) Y-directional RMS flow velocity (m/s)
From the experimental result of the plasma actuator, we focus on the air-attracting
effect of the plasma actuator, and propose an alternative but more practical flow control
method using air suction. Fig. 8 shows a 1/1 scaled panhead model with suction holes
near the separation points on both the upper and lower sides. Air suction through the
suction holes is carried out by an external exhaust fan. Each suction hole has a circular
opening with 2 mm in diameter. Suction holes are aligned in a straight line in the
spanwise direction of the panhead model. The center distance between neighboring
holes is 5 mm. In this wind tunnel test, we applied two types of spanwise suction holes
alignment as shown in Fig. 8(c) and Fig. 8(d). In the case of Fig. 8(c), suction holes are
aligned in one line -- we call this alignment “One line”. In the case of Fig. 8(d), suction
holes are aligned in two line --- we call this alignment “Two lines”. The total number of
suction holes is 103 and 206, respectively. Using these panhead models, we verified the
aerodynamic noise reduction effect.
Wind tunnel tests were carried out at the same wind tunnel as mentioned in the
previous section. The wind speed was set at 25 m/s (90 km/h). The free-stream velocity
is smaller than the Shinkansen train speed (300 km/h), but it is considered that the flow
520 T. Mitsumoji et al.
control mechanism obtained by the air suction is applicable to high speed range. Fig. 9
shows the experimental set-up. Aerodynamic noise emitted from the panhead model
was measured by a non-directional michrophone, which was located at 2 m away from
the panhead model.
In this experiment, we define the strength of the air suction by the following two
parameters; one is the suction flow speed US and the other is the suction flow rate Q.
The flow rate Q was measured by a flow meter installed in the middle of the air duct
which connected the panhead model to the exhaust fan. The suction flow speed US was
calculated from the flow rate Q and the total opening area of the holes.
Flow
Fig. 8. Panhead model with suction holes: (a) cross section; (b)-(d) suction holes alignments in
the spanwise direction which are viewed from upper side
Flow 2m
Microphone Panhead model
Fig. 10 shows a comparison of 1/3 octave band noise level according to the suction
strength. According to Fig. 10, if enough suction strength is obtained, this flow control
method can reduce the narrow band peak noise level around 63 Hz and its harmonic
frequency 125 Hz. The Strouhal number of this peak is 0.11 and this value is close to
the Strouhal number of the square rod. Therefore, this narrow band peak noise is
generated by the Karman vortices and called the Aeolian tone. In the same way as the
plasma actuator, the flow control mechanism is conjectured that the suction flow
through the suction holes attracts the air around the panhead and prevents the flow
separation from the panhead surface. As a result, the Karman vortices behind the
panhead are weakened.
Aerodynamic Noise Reduction of a Pantograph Panhead 521
70 Without suction
Without suction
Q = 0,Us=0)
(Q=0, Us = 0
60
Noise Level dB(A)
50 With suction
Q=2.4e-3 (One line)
m3/s,
m/s m3/s
Q = 2.4E-2
Us=37
Us = 37 m/s
40
With suction
Q=2.8e-4 (One line)
m3/s,
30 Q = 2.8E-2
Us=43
3
m/s m /s
Us = 43 m/s
20
With suction
Q=3.0e-4 (One line)
m3/s,
10 Us=46 m/s m3/s
Q = 3.0E-2
Us = 46 m/s
31.5 63 125 250 500 1k 2k 4k 8k 16k O.A.
Center Frequency Hz
70 Without suction
Without suction
Q = 0,Us=0)
(Q=0, Us = 0
60
Noise Level dB(A)
50 With suction
Q=2.4e-4 (Two lines)
m3/s,
m/s m3/s
Q = 2.4E-2
Us=18
Us = 18 m/s
40
With suction
Q=2.9e-4 (Two lines)
m3/s,
30 Q = 2.9E-2
Us=23
3
m/s m /s
Us = 23 m/s
20
With suction
Q=3.3e-4 (Two lines)
m3/s,
Q = 3.3E-2 3
10 Us=25 m/s m /s
Us = 25 m/s
31.5 63 125 250 500 1k 2k 4k 8k 16k O.A.
Center Frequency Hz
Focused on the narrow band peak noise level around 63 Hz, the peak noise level can
be reduced drastically if the suction strength is greater than
- Q = 3.0 × 10-2 m3/s, US = 46 m/s for the “One line” case (Fig. 10(a)),
- Q = 3.3 × 10-2 m3/s, US = 25 m/s for the “Two lines” case (Fig. 10(b)).
From these results, noise reduction effect is found to mainly depend on the flow rate
Q rather than the suction flow speed US. In these cases, the sufficient flow rate for the
flow control can be estimated about 3 × 10-2 m3/s.
Although the narrow band peak noise level is reduced, broad band noise level above
250 Hz is increased. Increment of these broad band noise levels is caused by the suction
flow itself. Therefore, the broadband noise should be reduced by improving suction
conditions such as the suction hole shape or the total opening area.
522 T. Mitsumoji et al.
4 Conclusions
To reduce aerodynamic noise of the pantograph head for high-speed trains, we applied
a flow control method to the panhead. First, we applied a plasma actuator to the
panhead surface. As a result, it is clarified that the plasma actuator installed near the
separation point can prevent the flow separation from the panhead surface, and can
reduce flow velocity fluctuations behind the panhead. This flow control effect is
considered to be obtained by a flow attracting effect of the plasma actuator. Second,
based on the flow control mechanism derived from the experiments applying the
plasma actuator, we proposed a more practical flow control method using air suction as
an alternative. Experimental results show that air suction near the separation point can
reduce narrow band peak aerodynamic noise from the panhead. Further work is needed
to reduce broadband suction effects and to apply this flow control method to the
Shinkansen train.
References
[1] Ikeda, M., Mitsumoji, T.: Numerical estimation of aerodynamic interference between
panhead and articulated frame. Quarterly Report of RTRI 50(4), 227–232 (2009)
[2] Sueki, T., Takaishi, T., Ikeda, M., Arai, N.: Application of porous material to reduce
aerodynamic sound from bluff bodies. Fluid Dynamics Research 42, 015004 (2010)
[3] Sueki, T., Ikeda, M., Takaishi, T.: Aerodynamic noise reduction using porous materials and
their application to high-speed pantographs. Quarterly Report of RTRI 50(1), 26–31 (2009)
[4] Ikeda, M., Mitsumoji, T., Sueki, T., Takaishi, T.: Aerodynamic noise reduction of a
pantograph by shape-smoothing of panhead and its support and by the surface covering with
porous material. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T.,
Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 118, pp. 419–426. Springer, Heidelberg (2012)
[5] Igarashi, T., Naito, H., Fukagata, K.: Direct numerical simulation of flow around a circular
cylinder controlled using plasma actuators. In: 9th European Fluid Mechanics Conference
(EFMC9), Rome, September 9-13, Paper 0050, 1 p. (2012)
[6] Inoue, K., Fukagata, K.: Control of a flow around a circular cylinder using plasma actuators.
In: Proc. 7th Int. Conf. on Flow Dynamics, Sendai, November 1-3, pp. 234–235 (2010)
Reduction of Train Induced Ground Vibration
by Vehicle Design
1 Introduction
The reduction of noise and vibration emissions induced by railway traffic is important
for the railway to reach its full potential as a sustainable, effective and comfortable
mode of transportation. The three-year project RIVAS started in 2010 and focused its
research on ground vibration abatement by addressing vehicle design, track design, soil
conditions and maintenance aspects. The current paper is a result of the work within
RIVAS aiming for better understanding of the influence of the vehicle design and
wheel-out-of roundness on vibration, taking various levels of track irregularity and
ground conditions into account. The research leading to these results has received
funding from the European Union Seventh Framework Programme under grant
agreement no 265754.
governs the vibration excitation. The track design and the ground conditions determine
the vibration response and the propagation of vibration to nearby buildings. Extensive
research on the influence of ground conditions has been performed at the Institute of
Sound and Vibration Research (ISVR) in the UK and at the Katholieke Universiteit
Leuven (KUL) in Belgium e.g. [1, 2]. Most previous investigations on ground vibration
induced by trains have had the track and ground influence as primary targets. Hence,
the role of the vehicle has been less established even though there are exceptions. The
RENVIB II project, funded by the UIC and a predecessor to the RIVAS project,
presented a parameter study on the influence of vehicle design and wheel and track
irregularities using a quarter-car vehicle model [3]. The most prominent vehicle
influence was observed from the unsprung mass for frequencies above 25 Hz and from
wheel irregularities which resulted in a broadband contribution of up to 10 dB to the
vibration velocity level in the ground. Simulations with a multi-degree-of-freedom
vehicle model have been carried out at Chalmers University of Technology
(CHARMEC) in Sweden. These showed that vehicle geometry clearly influences the
frequency content of the vibration at a given speed due to the excitation by various axle
passing frequencies while the influence of bogie mass and primary suspension stiffness
was found to be small. The model was validated by field measurements that also
showed the importance of the parametric excitation due to the discrete sleeper supports
of a ballasted track [4]. As the properties of track and ground may vary considerably
throughout the rail network, it may not be possible to determine one universal vehicle
design that is optimal at all sites. The current paper aims at investigating an improved
vehicle design, considering variations in the dynamic ground properties as well as
accounting for different levels of wheel OOR and vertical track irregularity
(longitudinal level).
Fig. 1. Vehicle model used in TRAFFIC consisting of rigid masses, springs and viscous damping
elements. The two investigated parameters: primary suspension stiffness, k1, and unsprung mass,
mw, are marked.
High and low levels of unsprung mass and primary suspension stiffness are
combined to provide four different vehicle designs. The values of the high and low
levels are 4 kN/mm and 7 kN/mm for the stiffness and 1425 kg and 2850 kg for the
mass. The stiffness values are reasonable variations to the nominal stiffness of a Y25
bogie. The high level of the unsprung mass is unrealistic for a freight wagon but more
typical for a locomotive. Three different classes of OOR are included to represent
wheel tread conditions ranging from good to bad. The impact on ground vibration from
the variations in vehicle design and wheel tread condition is predicted with TRAFFIC
using ground models representing three different sites in Europe and for three different
spectra of vertical track irregularity. All calculations are performed for train speeds 70,
100 and 120 km/h, and the vibration response is predicted at 8, 16, 32 and 64 m from
the track centre line. The ground models referred to as Horstwalde, Ledsgård and
Lincent are all reference sites used within the RIVAS project. Horstwalde is a site in
Germany close to Berlin with a thick and relatively stiff layer of sandy soil. Ledsgård is
located in the southern part of Sweden and has several layers of very soft ground that
have caused severe problems with train induced ground vibration. The site of Lincent in
Belgium has a combination of stiff and soft layers. The details of the models are found
in [6].
The ground is excited by the wheel and track irregularities that are included in the
calculations as the effective (energy) sum of their wavelength spectra, i.e. a combined
wheel and track irregularity spectrum. The track irregularity is modelled either
according to the ISO 3095:2005 reference spectrum (extrapolated to wavelengths of
about 30 m) [7], or by the high and low levels of the ORE (ERRI) spectrum [8]. The
ISO 3095 spectrum is intended for application in noise type tests, while the ORE levels
are used for evaluating vehicle dynamics and running stability. Hence neither of these
two was initially intended for evaluating ground vibration. The three classes of OOR
are taken from an extensive measurement campaign of out-of-round wheels carried out
by Johansson [9]. Class 1 corresponds to a freight wagon wheel in very good condition,
class 2 represents a bad wheel from a passenger vehicle and class 3 corresponds to a
526 A. Mirza, A. Frid, and J.C.O. Nielsen
freight wheel in poor condition with high tread irregularities particularly for order 1
(eccentricity). The wheel radius is 0.9 m. The combined spectra of the wheel and track
irregularities are presented in Fig. 2. The class 1 OOR has a negligible influence on the
combined spectra (less than 0.5 dB in the most affected 1/3 octave band).
60 60
OOR class 1
Wheel OOR level [dB] ref 1e-6m
50
10 30
0
20 ISO 3095 + OOR class 1
-10 ORE low + OOR class 1
ORE high + OOR class 1
-20 10 2 1 0 -1
1 2 3 4 5 6 7 8 9 10 11 12 13 10 10 10 10
order [-] 1/3 oct. wavelength bands [m]
60 60
Irregularity level [dB] ref 1e-6m
Irregularity level [dB] ref 1e-6m
50 (c) 50 (d)
40 40
30 30
Fig. 2. Wheel and track irregularity spectra. (a) Three classes of wheel OOR based on
measurements reported in [9], left bar: class 1, middle bar: class 2 right bar: class 3. Three track
irregularity spectra combined with (b) class 1 OOR, (c) class 2 OOR, (d) class 3 OOR.
4 Results
The results from the simulations are shown in the four figures below. All trends in Figs.
3, 4 and 6 are consistent for the three speeds included in this study and hence only the
results for 100 km/h are shown here. In Fig. 5 however, showing the ground influence,
the speed is 70 km/h which turned out to be a “worst case” for the Ledsgård site. Fig. 3
shows the increase in vibration level caused by the three classes of OOR assuming a
track irregularity according to ISO 3095 at the site of Lincent. The graphs show the 1/3
octave band spectra at 8 m while the bars show the increase caused by the OOR both in
1/3 octave bands and total level at several distances from the track. A class 1 OOR,
Reduction of Train Induced Ground Vibration by Vehicle Design 527
60 15
ISO 3095 class1
40
Vibration velocity level [dB] ref 1e-9
30 10
20
10
0 5
-10
-20
-30 0
2.5 5 10 20 40 80 8m 16 m 32 m 64 m
1/3 octave bands [Hz] Total vibration level
Fig. 3. Left y-axis: vibration spectra at 8 m from the track at the site of Lincent. Wheel–rail
excitation by ISO 3095 track irregularity and the three classes of OOR described in Fig. 2. Right
y-axis: bar diagram showing the increase in vibration level caused by the OOR in 1/3 octave
bands at 8 m and for the total level at 8, 16, 32 and 64 m from the track. Left bar: class 1, middle
bar: class 2, right bar: class 3.
Fig. 4 shows the influence of a class 3 OOR on ground vibration for different spectra
of track irregularity. The increase caused by the OOR is significant and largest for the
ORE low track irregularity spectrum at high frequencies. This is a consequence of the
relatively low level of the ORE low spectrum at short wavelengths (see Fig. 2). Due to
the high response of the ground at these frequencies, the impact of the OOR is a 10 dB
increase of the total level at 8 m. Fig. 5 shows the influence of OOR on the three
different ground models. The increase in vibration is similar for the three models in
most frequency bands. However, the increase in total level is higher at Ledsgård. This
is a consequence of the combination of train speed and the wavelength of the wheel
eccentricity leading to a high excitation at 6.3 Hz where the ground shows a high
response.
528 A. Mirza, A. Frid, and J.C.O. Nielsen
60
ISO 3095 ISO 3095
40
Vibration velocity level [dB] ref 1e-9
30
10
20
10
0
5
-10
-20
-30 0
2.5 5 10 20 40 80 8m 16 m 32 m 64 m
1/3 octave bands [Hz] Total vibration level
Fig. 4. Left y-axis: vibration spectra at 8 m from the track at the site of Lincent. Wheel–rail excitation
by the ISO 3095, ORE low and ORE high track irregularity spectra in combination with OOR class 3
as described in Fig. 2. Right y-axis: bar diagram showing the increase in vibration level caused by
OOR in 1/3 octave bands at 8 m and for the total level at 8, 16, 32 and 64 m from the track. Left bar:
class 1, middle bar: class 2, right bar: class 3.
60
Lincent Lincent
40
Vibration velocity level [dB] ref 1e-9
30
10
20
10
0
5
-10
-20
-30 0
2.5 5 10 20 40 80 8m 16 m 32 m 64 m
1/3 octave bands [Hz] Total vibration level
Fig. 5. Left y-axis: vibration spectra at 8 m from the track at the sites of Lincent, Ledsgård and
Horstwalde. Train speed 70 km/h. Wheel–rail excitation by the ISO 3095 spectrum and OOR class 3
according to Fig. 2. Right y-axis: bar diagram showing the increase in vibration level caused by the
OOR in 1/3 octave bands at 8 m and for the total level at 8, 16, 32 and 64 m from the track. Left bar:
class 1, middle bar: class 2, right bar: class 3.
Reduction of Train Induced Ground Vibration by Vehicle Design 529
Fig. 6 shows the influence of both the class 3 OOR and an altered vehicle design.
The interaction effect of the OOR and design changes is seen as the relatively small
differences between the bars (e.g. the two bars for low mass with either soft or stiff
spring) within a single 1/3 octave band or the bars representing the total level at a given
distance from the track. These variations are minor compared to the effects of OOR and
unsprung mass separately. The influence of the primary suspension stiffness occurs
only in the 5 to 10 Hz range where it shifts the vehicle resonance. This effect is however
masked by the influence of OOR. The increase at 8 m from doubling the unsprung mass
is of the same order of magnitude as the one caused by the class 3 OOR.
20
10
5
0
-10
-20
-30 0
2.5 5 10 20 40 80 8m 16 m 32 m 64 m
1/3 octave bands [Hz] Total vibration level
Fig. 6. Left y-axis: vibration spectra at 8 m from the track at the site of Lincent. Wheel–rail excitation
by the ISO 3095 track and OOR class 3 according to Fig 2. Four different vehicle designs according to
Table 1. Right y-axis: bar diagram showing the increase in vibration level caused by OOR in 1/3 octave
bands at 8 m and for the total level at 8, 16, 32 and 64 m from the track. Left bars: low mass, right bars:
high mass.
The current simulation approach has enabled the evaluation of vehicle design and OOR
on ground vibration for a range of track and ground conditions. Single order OOR can
induce large contributions to the vibration level in a narrow frequency range even when
the track irregularity level is high. The impact on the total level is significantly
influenced by the frequency response of the ground. Fig. 5 indicates that a vehicle with
class 3 OOR running on the three sites of Lincent, Ledsgård and Horstwalde will
produce different levels of ground vibration, although the summed wheel/rail
irregularity spectrum is equal at all sites. The reduction achieved by a lower unsprung
mass is broadband and can be clearly seen in the overall vibration level. This reduction
is of the same order of magnitude as the potential reduction from removing severe
OOR. The two levels of unsprung mass used in this study represent a freight bogie and
a locomotive bogie, and, hence, the potential reduction within a single vehicle category
530 A. Mirza, A. Frid, and J.C.O. Nielsen
is expected to be smaller than illustrated here. The interaction effect of OOR and
vehicle design is small, and the influence of the primary suspension is limited to a
narrow frequency range. The full effect of the suspension stiffness might, however, not
be revealed here since the parametric excitation was excluded. The influence from
OOR and unsprung mass on ground vibration seen in this simulation study is in
agreement with the analysis of measurement data performed by Huber et. al within the
RIVAS project [10].
Acknowledgements. The authors would like to acknowledge the support from Prof
Geert Lombaert and Dr Stijn Francois at the KUL with preparations of the simulations
in TRAFFIC.
References
[1] Sheng, X., Jones, C.J.C., Thompson, D.J.: A comparison of a theoretical model for
quasi-statically and dynamically induced environmental vibration from trains with
measurements. Journal of Sound and Vibration 267(3), 621–635 (2003)
[2] Lombaert, G., Degrande, G.: Ground-borne vibration due to static and dynamic axle loads
of InterCity and high-speed trains. Journal of Sound and Vibration 319, 1036–1066 (2009)
[3] Temple, B.P., Block, J.R.: RENVIB II Phase 2 Task 6 - Influences on the generation of
railway ground-borne vibration. AEAT (March 1999)
[4] Mirza, A., Frid, A., Nielsen, J.C.O.: Ground vibration induced by railway traffic - a
pre-study on the influence of vehicle parameters. Research report 2010:08. Chalmers
University, Department of Applied Mechanics (2010)
[5] Nielsen, J.C.O., et al.: Train induced ground vibration – optimised rolling stock mitigation
measures and their parameters. RIVAS Deliverable D5.4 (2013)
[6] Coulier, P., et al.: Scope of the parametric study on mitigation measures on the
transmission path. RIVAS Deliverable D4.1 (2011)
[7] Railway applications - Acoustics - Measurement of noise emitted by railbound vehicles
(ISO 3095:2005)
[8] Utrecht, ORE, Question B176: Bogies with steered or steering wheelsets, Report No 1
Specifications and preliminary studies. Specification for a bogie with improved curving
characteristics, vol. 2
[9] Johansson, A.: Out-of-round railway wheels – measurements of out-of-roundness,
transverse profile and surface hardness. Research report 2004:1. Chalmers University,
Department of Applied Mechanics (2004)
[10] Huber, P., Nelain, B., Müller, R.: RIVAS - Mitigation measures on vehicles (WP5);
Experimental analysis of SBB ground vibration measurements and vehicle data. In:
Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D.,
Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 126, pp. 531–538. Springer, Heidelberg (2015)
RIVAS – Mitigation Measures on Vehicles (WP5);
Experimental Analysis of SBB Ground Vibration
Measurements and Vehicle Data
1
PROSE AG,
Zürcherstrasse 41, 8400 Winterthur, Switzerland
philipp.huber@prose.ch
2
VibraTec, Ecully, France
3
SBB AG, Bern, Switzerland
In 2011, the EU started a research and development project within the scope of the 7th
Research Framework Program referred to as RIVAS. The goal of the three-year
research project, which is financially supported by the EU (5 million euro of
subsidies), is the development of innovative measures to significantly reduce the
negative impacts of ground-borne vibration from railway traffic on the environment.
At the same time it should maintain the competitiveness of European railways. The
main focus of the project is on the freight traffic, open lines and existing tracks. Not
only measures made on the infrastructure and on the propagation path (work packages
3 and 4) are investigated but also measures on the rolling stock (work packages 2 and
5). The RIVAS project does not only test how to avoid symptoms on rolling stock
(turning or replacing wheels) but is also searching for measures for existing and new
vehicles.
After a state-of-the-art report regarding measures on rolling stock, the essential
rolling stock parameters and their influence on ground vibration were investigated in
detail. On the one hand the influence of rolling stock parameters was studied on the
basis of numerical simulations, and on the other hand based on comprehensive ground
vibration measurements in Switzerland, to make statistically reliable statements on a
wide range of vehicles. This summary is concentrated on the results of the ground
vibration measurements in Switzerland.
Ground vibration measurements on open track (distance 2 m and 8 m to the track axis)
were made at three wheel load checkpoints (WLC) of the SBB for several days. Several
thousand regular trains were recorded.
Together with the detailed WLC data provided for each vehicle (type, vehicle
number, speed, wheel load), the measurements resulted in an extensive database for the
statistical vibration emission analysis of the rolling stock used on the SBB railway
network. All in all, approx. 1,000 bogies of locomotives and train sets, approximately
5,000 bogies of passenger coaches and approximately 10,000 freight wagon bogies
were statistically analysed for vibrations.
The goal of this analysis was the quantification of vibration differences within and
between the vehicle categories and to make a correlation with the rolling stock
parameters. On the basis of these findings, the characteristics of low-vibration rolling
stock can be determined.
The evaluations of the vibration measurements show that there are significant
differences. Not only the average values per vehicle category differ, but also the
variance differs (see Fig. 1 for the measuring point in Thun, stiff soil).
In Thun, vibration produced by Intercity trains as well as local trains and freight
traffic (Lötschberg axis) with different freight locomotives / freight wagons were
recorded. The minimum value, maximum value, median value and 95% value shown in
Fig. 1 clearly indicate the great difference between the vibration emissions of different
vehicle categories, and the different variance of the values for each vehicle category.
While a high median value indicates the high vibrations generally caused by a vehicle
type, high variance is a sign for different wheel conditions.
RIVAS – Mitigation Measures on Vehicles (WP5) 533
0.7
95%-value Median
0.6 Max
Min
0.5
vrms, fast [mm/s]
Median 95%-value
0.4
0.3
0.2
0.1
0.0
"piggy back"
Freight vehicle,
Freight vehicle,
Freight vehicle,
New freight loco
EMU
Passenger
Passenger
Passenger
Passenger
Re460 loco
Re420 loco
Re620 loco
axle d=1.8m
coach 1
coach 2
coach 3
coach 4
axle d=2m
RoLa
2-axle
Fig. 1. Statistical values for ground vibrations v rms,fast in Thun at a distance of 8 m from the
track, vehicle speed 60-70 km/h
The freight locomotives standing out on Fig. 1 are discussed in detail in the
following chapter. Passenger trains usually are faster than freight trains. In such
situations, the difference between the vibration values for passenger trains and freight
trains are less obvious, because the vibration is increasing with speed.
The measured freight trains in Switzerland show that freight locomotives rather than
freight wagons are responsible for maximum vibrations. Therefore, possible measures
should primarily be taken on freight locomotives of the older generation (Re420,
Re620) as well as the newer generation to efficiently and effectively reduce ground
vibrations.
The evaluations of the vibration measurements show that older freight locomotives
with cast iron brake blocks and tyred spoke wheels such as in Re420 and Re620 cause
the strongest vibrations. These are caused by spalling in the tread and by
out-of-roundness of the wheel, which are accentuated by capacity bottlenecks for wheel
maintenance in combination with higher unsprung mass.
The properties of the newer generation freight locomotives (compared to the SBB
Re460 Intercity locomotive which was originally also used for freight traffic), that are
decisive for vibrations are discussed on the basis of the example of a new generation
freight locomotive that is very common for the standard Swiss railway network. The
main difference between Re460 and the new generation freight locomotive is that the
latter has a nose-suspension drive (drive mass directly coupled to the wheelset mass)
and larger wheels with diameter 1250 mm (Re460 has diameter 1100 mm wheels) as
well as disc brakes (Re460 has weak unilaterally acting sintered block brakes combined
534 P. Huber, B. Nélain, and R. Müller
with permanent magnetic track brakes). Moreover, the Re460 has passively controlled
radial wheelset steering. Tests were also carried out for the new generation freight
locomotive regarding the introduction of radial steering. However, due to financial
reasons, increased maintenance, and problems with approvals throughout Europe,
radial steering was not introduced. Moreover, for a maximum speed of 140 km/h, the
fully suspended drive (as used for BR 146 of the German Railway and Re460) was
ruled out for financial reasons.
A direct comparison between Re460 and the new generation freight locomotive with
regard to vibrations could be made at the WLC measuring point in Thun.
156 Re460 bogies and 72 new generation freight locomotive bogies were
statistically evaluated in the frequency domain within the speed range from 60 to
70 km/h (Fig. 2).
1000
100
v rms [10^-3 mm/s]
Re 460 median
10 New freight loco median
Re 460 95%-value
New freight loco 95%-value
1
0.1
8 16 31.5 63 125 250
1/3 octave frequency [Hz]
Fig. 2. Third-octave band spectra of SBB Re460 and new generation freight locomotive in Thun at
a distance of 8 m from the track, vehicle speed 60-70 km/h
The comparison of the median curves shows significant differences between the new
generation freight locomotive and Re460 in particular in the 50-80 Hz third-octave
bands. Apart from that, the curves are similar despite the different bogie design. The
main cause for these differences regarding the median values are possibly due to the
different drive bearings and/or the huge differences in unsprung mass that are more
than a factor two. The difference in 95 % values is much more significant. The
difference between Re460 and the new generation freight locomotive in the 63 Hz
third-octave band is almost a factor of ten. The only possible cause for the significant
variance of vibrations within one vehicle type lies in the different condition of the
wheels, because any other vibration-relevant properties of the vehicle such as unsprung
mass, primary suspension stiffness etc. hardly change. In other words, the condition of
the wheels is the dominant parameter for vibrations. There are various influences on the
condition of the wheels. At least the following five aspects are decisive for this case
when comparing Re460 and new generation freight locomotives:
RIVAS – Mitigation Measures on Vehicles (WP5) 535
The diagrams of Fig. 3 show almost harmonic curves and/or polygons with 14 and
28 corners. The 14 corners exactly match the high increase in the 63 Hz third-octave
band in Fig. 2 (excitation frequency at 65 km/h). This typical curve and almost equally
high vibrations were also measured for the locomotives of the other manufacturer. The
causes for this polygonization were investigated within the scope of the RIVAS WP5,
Deliverable D 5.4 [13].
Different types of freight wagons were measured at the WLC measuring point in Thun.
The most common bogie for freight wagons is the Y25 bogie with an axle distance of
1.8 m. There are also other bogie types with the same axle distance as well as bogies
with an axle distance of 2.0 m including smaller wheels (Y33). The RoLa (rolling road,
536 P. Huber, B. Nélain, and R. Müller
“piggy-back”) has a bogie with four axles (axle distance. 0.70/0.75 m) with very small
wheels (diameter 360 mm) and disc brakes. Fig. 4 shows the third-octave band spectra
for the bogie categories with an axle distance of 1.8 m (e.g. Y25) and RoLa for the
speed range between 60 and 70 km/h.
100
v rms [10^-3 mm/s]
0.1
8 16 31.5 63 125 250
1/3 octave frequency [Hz]
Fig. 4. Third-octave band spectras of RoLa and freight wagon with an axle separation of 1.8 m in
Thun at a distance of 8 m from the track, vehicle speed 60-70 km/h
At sleeper passage frequency (31.5 Hz), the RoLa generates ground vibrations that
are less by a factor of up to three. Possible causes are: Low unsprung mass, low static
axle load, the small wheels of the RoLa leading to an out-of-roundness which has
accordingly higher frequencies, or small axle distance and/or 4 axles per bogie. The
variance within the two vehicle categories is significant, also for the RoLa where the
small wheel diameter results in high wheel-rail contact stress (noise monitoring
measurements carried out by Federal Office of Transport resulted in very high noise
variances for the RoLa; variances for other freight wagons are lower).
The scattering of wheel out-of-roundness in freight wagons is directly related to the
scattering of ground vibration. Since freight wagons have no slip-slide protection, there
are many wheel flats that produce high vibration. Measures taken on freight wagons
that reduce wheel flats are therefore important to mitigate ground vibrations. Moreover,
an improved and more homogenous wheel quality will result in a better condition of the
wheels.
The evaluations and statistical analyses of the ground vibration of vehicles on the SBB
network show that both the condition of the wheels and the unsprung mass have a
dominating influence. For the unsprung mass, this was confirmed with numerical
simulations in RIVAS WP5, Deliverable D 5.4 [13]. The condition of the wheels can be
RIVAS – Mitigation Measures on Vehicles (WP5) 537
positively influenced by a design that reduces the wear of tracks and wheels (e.g. radial
steering of the wheelsets, low unsprung mass) as well as by condition-based and
prompt wheel maintenance.
The measures taken on the rolling stock to reduce vibrations can be classified in
three categories: Maintenance / prevention, improvement of existing vehicles and
improvement of new vehicles. The following measures taken on the rolling stock to
reduce vibrations might be very effective (further measures are provided in [5, 6, 8,
10]):
- Automatic monitoring systems for wheel quality integrated in the network,
allowing for condition-based and prompt maintenance.
- Improved interaction of the brake systems, slip-slide protection and wheel
material qualities (e.g. special pearlitic-silicon and manganese carbon steels) that
avoid wheel flats occurrence and spalling.
- The reduction of unsprung mass, especially for locomotives, will not only reduce
vibrations but also the dynamic load on the wheels and tracks (e.g. more
expensive hollow-shaft drive instead of nose-suspension drives, where the
gearbox mass and most of traction motor mass are unsprung, if also reasonable
for lifecycle costs).
- Radial steering of the wheelsets in the bogies (passive or active) to reduce the
wear and polygonization of wheels and rails in narrow curves (proved by Re460
in service on the alpine lines for years, see [14].
6 Conclusion
Freight locomotives generate particularly high ground vibrations, mainly related to
high unsprung masses and a high degree of out-of-roundness.
Possible measures for the rolling stock that effectively reduce vibrations should be
developed and tested at a broader scale. Measures exist that do not require high
investment and can be implemented within short time with high efficiency. Other
measures exist that may lead to the desired result only in the long run and will require
further discussions. Additional drivers for the implementation of measures besides
vibration mitigation are, on the one hand, the reduced maintenance efforts for
infrastructure and rolling stock due to low dynamic wheel-rail forces, and, on the other
hand, the increased comfort demands of the passengers and the improved safety
standards.
New rules and standards, provisions of regulatory authorities and requirements
regarding track access charges could be an incentive to use low-vibration rolling stock
in the future.
References
[1] Müller, R.: Mitigation measures for open lines against vibration and ground-borne noise: a
Swiss overview. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C.,
Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 99, pp. 264–270. Springer, Heidelberg (2008)
538 P. Huber, B. Nélain, and R. Müller
Summary. This paper studies the efficiency of stiff wave barriers for the
mitigation of railway induced vibrations. Coupled finite element–boundary
element models developed at KU Leuven and ISVR are employed; these models
have been cross–validated within the EU FP7 project RIVAS (Railway Induced
Vibration Abatement Solutions). A first mitigation measure consists of a block
of stiffened soil embedded in a halfspace that acts as a wave impeding barrier.
The existence of a critical frequency from which this mitigation measure starts
to be effective, as well as a critical angle delimiting the area where the vibration
levels are reduced, is demonstrated. Next, a sheet piling wall is considered,
accounting for the orthotropic behaviour of this wall. Calculations show that the
reduction of vibration levels is entirely due to the relatively high axial and
bending stiffness in the vertical direction (along the profiles), while the bending
stiffness for bending waves traveling in the longitudinal direction
(perpendicular to the profiles) is too low to affect the transmission of vibrations.
Field tests are being carried out in Spain and Sweden to confirm the conclusions
of these numerical computations.
1 Introduction
Railway induced vibrations are an important source of annoyance in the built
environment that can cause the malfunction of sensitive equipment and nuisance to
people. Various measures can be taken to mitigate vibrations, either at the source
(railway track), in the propagation path between source and receiver, or at the receiver
(surrounding buildings) [1].
An open trench in the soil is a typical example of a mitigation measure in the
transmission path. Such trenches aim to reflect the impinging waves and are very
effective for trench depths of greater than about one Rayleigh wavelength [2]. The
construction of open trenches is, however, limited to shallow depths for stability; the
use of either soft or stiff in–fill materials allows an increase of depth. If a soft in–fill
material is used, the behaviour of a filled trench resembles that of an open trench. The
use of a stiff in–fill material, however, fundamentally alters the physical mechanism by
which vibration levels are reduced.
This paper focuses on the application of stiff wave barriers to mitigate railway
induced vibrations. Two such mitigation measures are presented: subgrade stiffening
next to the track and a sheet pile wall. The study is part of the EU FP7 project RIVAS
(Railway Induced Vibration Abatement Solutions) [3].
2 Methodology
For the prediction of railway induced ground vibration, project partners KU Leuven
and ISVR make use of different two–and–a–half–dimensional (2.5D) coupled finite
element–boundary element (FE–BE) models [4, 5]. In these models, the geometry of
the track–soil system is assumed to be invariant in the longitudinal direction. First, an
independent analysis was conducted by both project partners for a benchmark case of a
12 m deep barrier in a homogeneous halfspace, characterized by a shear wave velocity
Cs = 250 m/s, a dilatational wave velocity Cp = 1470 m/s, a density ρ = 1945 kg/m³,
and material damping ratios βs = βp = 0.025 in both deviatoric and volumetric
deformation. The wall barrier has a thickness t = 0.396 m, a modulus of elasticity E =
6.99 x 109 Pa, a Poisson's ratio ν = 0.30, and a density ρ = 286.6 kg/m³. The track is
disregarded. At KU Leuven, the wall is modeled with 2–noded shell elements with a
mesh size of 0.309 m. The soil at both sides of the sheet pile wall is modeled with a
conforming BE mesh, assuring 8 elements per shear wave length up to 100 Hz. In the
model of ISVR, the wall is modeled with 3–noded elastic solid elements. The node to
node distance was set to 0.1 m in the regions within 1 m from the top and bottom of the
wall and 0.25 m everywhere else.
−8
x 10
4
F
Mobility [m/Ns]
1
z y
x 0
0 20 40 60 80 100
Frequency [Hz]
(a) (b)
Fig. 1. (a) Benchmark study for a stiff wave barrier in a homogeneous halfspace. (b) Vertical
transfer mobility at a distance of 8 m (○), 16 m (×), 32 m (□) and 64 m (◊) from a vertical point
source. KU Leuven results (solid lines) and ISVR results (symbols).
Transfer mobilities for a vertical point source are shown in Fig. 1. The agreement
between both models is excellent. The largest discrepancies are found at the highest
frequencies. In the next sections, results of a parametric study for stiff wave barriers are
presented.
Stiff Wave Barriers for the Mitigation of Railway Induced Vibrations 541
3 Subgrade Stiffening
The first vibration mitigation measure consists of subgrade stiffening next to the track,
where a block of soil is stiffened (e.g. by means of jet grouting). Throughout this
section, a homogeneous halfspace characterized by a shear wave velocity Cs = 200 m/s ,
a dilatational wave velocity Cp = 400 m/s , a density ρ = 2000 kg/m 3 , and material
damping ratios βs = β p = 0.025 in both deviatoric and volumetric deformation is
considered. The vibration reduction efficiency of a block of stiffened soil with a
width w = 2 m and a depth h = 2 m will be investigated. This stiff wave barrier has a
shear wave velocity Cs = 550 m/s and a dilatational wave velocity Cp = 950 m/s ; the
same density and material damping ratios as in the halfspace are used. To facilitate
physical interpretation, an incident wavefield is generated by the application of a unit
vertical harmonic point load at the surface of the halfspace; the presence of the track is
thus neglected.
(a) (b)
10
(c) −10
Fig. 2. Real part of the vertical displacement uˆ z (x, ω ) due to a unit harmonic vertical excitation
at the origin of the coordinate system at 45 Hz (a) on a homogeneous halfspace and (b) when a
block of stiffened soil is included. The vertical insertion loss is shown in (c).
uˆ zref ( x, ω )
IL z ( x, ω ) = 20log10 [dB] (3.1)
uˆ z ( x, ω )
542 P. Coulier et al.
reference case (without barrier) and in the case with the barrier, respectively. Positive
values of the insertion loss indicate a reduction of the vertical free-field vibration. The
insertion loss IL z ( x, ω ) is shown in Fig. 2c. A delimited area where significant
reduction of vibration levels is achieved is observed. The insertion loss reaches values
of 10 dB and more in this region. The reduction is not only obtained at the surface of
the halfspace, but also at depth, although some localized areas can be identified with
increased vibration levels with respect to the reference case. Furthermore, lines of
constructive and destructive interference between direct and reflected Rayleigh waves
can be observed at the opposite side of the track (i.e. where no soil stiffening is
applied). The physical interpretation for these observations will be formulated in the
subsequent subsection.
To explain the behaviour of a stiff barrier, the cylindrical wavefield should be decomposed
into a superposition of plane waves. Each of the plane waves is characterized by a
dimensionless longitudinal wavenumber k y , defined as k y = k yCs / ω = Cs / C y , with Cs
the shear wave velocity of the halfspace and C y the phase velocity of the waves. The
efficiency of the stiff wave barrier can hence also be quantified in the
frequency–wavenumber domain through the vertical insertion loss IL~ z ( x, k y , z, ω ) , where a
tilde above a variable denotes its representation in the frequency–wavenumber domain.
~
Fig. 3 shows the insertion loss IL z ( x = 8m, k y , z = 0m, ω ) in a range 0 ≤ k y ≤ k R , where the
dispersion curve k y = k R = Cs / CR corresponds to a Rayleigh wave propagating in the
y–direction. A clearly delimited area in the (ω, k ) –domain
y corresponding to a
significant insertion loss can be distinguished in this figure.
Fig. 3 can be interpreted by considering the interaction of the Rayleigh wave in the
soil and bending waves in the block of stiffened soil [6]. Superimposed on Fig. 3 is the
dispersion curve k y = k b of a free bending wave in an infinitely long Timoshenko
beam with the same properties as the block of stiffened soil. The region where a
( )
substantial insertion loss is obtained in the ω , k y –domain is clearly bounded by the
Rayleigh wave dispersion curve k y = k R and the free bending wave dispersion
curve k y = k b . At low frequencies, the wavenumber k b is larger than the wavenumber
k R , and the Rayleigh wave propagates unhindered through the block of stiffened
soil. Above a critical frequency, the wavefield contains propagating plane waves
with a wavenumber k y larger than k b (i.e. with a wavelength λ y smaller
Stiff Wave Barriers for the Mitigation of Railway Induced Vibrations 543
than λb ). The transmission of these plane waves is impeded by the block of stiffened
soil, as the admittance of a beam of infinite length is then dominated by its bending
−4
stiffness and decreases proportionally to k y at a given radial frequency ω . This
explains why the zone of significant insertion loss in Fig. 3 is delimited by the free
bending wave dispersion curve.
1.5 10
5
1
k̄y [-]
0.5
−5
0 −10
0 20 40 60 80 100
Frequency [Hz]
~
z ( x = 8m, k y , z = 0m, ω )
Fig. 3. Vertical insertion loss IL in the case of a block of stiffened soil
embedded in a homogeneous halfspace. Superimposed are the dispersion curve of a Rayleigh
wave in the y–direction (solid black line) and the free bending wave dispersion curve in an
infinitely long beam (dashed black line)
The critical radial frequency ωc from which the block of stiffened soil can act as a
wave impeding barrier is determined by the intersection of the Rayleigh wave and the
free bending wave dispersion curves [6]:
ρA E μκ
ωc = CR2 (3.2)
EI ( E − ρC )( μκ − ρC )
2
R
2
R
which equals 2π × 12 Hz in the actual case. E is the Young's modulus, μ the shear
modulus, ρ the density, A the cross-sectional area, I the moment of inertia, and κ the
shear coefficient of the block barrier. As a square cross section is considered, the area
moment of inertia is the same for bending around the x-x and the z-z axis, respectively;
this is in general not the case. Numerical simulations indicate that bending around the
x-axis is dominant, provided that the block of stiffened soil has a minimum width to
ensure that it behaves as a beam [6]. The discussion above indicates that the mitigation
measure can only be effective for frequencies above ωc . Equation (3.2) indicates that
the critical frequency strongly depends on the stiffness contrast between the
surrounding soil and the block of stiffened soil, indicating that this mitigation measure
544 P. Coulier et al.
will be more effective at sites with a soft soil. Equation (3.2) is of great practical
importance, as it provides a simple design guideline to assess the expected vibration
reduction efficiency of soil stiffening next to the track, without the need of an extensive
FE–BE calculation.
The propagating plane waves are propagating in the direction given
( )
by θ = sin −1 k y / k R . This angle is situated between θ = 0 (plane wave propagation
perpendicular to the barrier) and θ = π / 2 (plane wave propagation parallel with the
barrier). As a result, a reduction of vibration levels will only be obtained in an area
( )
delimited by a critical angle θc (ω ) = sin −1 k b / k R , defined as:
4 E ( μκ ) A
2
E + μκ + ( E − μκ )
2
+
ρ Iω 2
sin θ c (ω ) = CR ρ . (3.3)
2 E μκ
The physical mechanism outlined above explains the pattern of insertion loss that is
observed in Fig. 2c. As the considered frequency of 45 Hz is above the critical
frequency of 12 Hz, a significant reduction of vibration levels is achieved, but only in a
limited area due to the existence of a critical angle.
Within the frame of the EU FP7 project RIVAS [3], a field test is being carried out in
Spain to confirm the findings of the numerical simulations. A jet grouted wall with a
width of 1 m, a depth of 7.5 m and a length of 60 m will be constructed at a site with
relatively soft soil layers.
At Furet, Sweden, a sheet pile wall has been installed next to a track to reduce train
induced vibrations in adjacent buildings [7]. The sheet pile wall consists of VL 603-K
profiles (Fig. 4). The depth of the sheet piles is 12 m with every fourth pile extended to
18 m. The soil profile at Furet consists of very soft clayey silt up to a depth of
approximately 12 m. Here, the effectiveness of a 12 m deep sheet pile wall installed in a
homogeneous halfspace is investigated, to facilitate the physical interpretation. The soil
characteristics as given in Section 2 are used in this example. An equivalent orthotropic
plate model of the sheet pile wall was adopted in the 2.5D calculations to account for
the fact that the bending stiffness in the vertical direction (along the profiles) is much
larger than the bending stiffness in the longitudinal direction (perpendicular to the
profiles). In the frequency range of interest (0-100 Hz), the bending wavelength in the
sheet pile wall is much larger than the repetition distance dR = 1.2 m of the sheet pile
wall (Fig. 4). Therefore, the profiling of the plate can be disregarded. The thickness,
density, moduli of elasticity, and Poisson's ratios of the equivalent orthotropic plate are
chosen such that the plate has approximately the same mass, bending stiffness in two
directions as well as vertical axial stiffness as the VL 603-K profile.
Stiff Wave Barriers for the Mitigation of Railway Induced Vibrations 545
Fig. 4. Cross section of the sheet pile wall (VL 603-K profile)
Fig. 5a shows the vertical insertion loss at 25 Hz. Fig. 5b shows the vertical
insertion loss at a distance of 32 m in the frequency-wavenumber domain. Very little
vibration reduction is seen below 20 Hz. The frequency above which a reduction in
vibration levels can be expected, is in the first place determined by the depth of the
wall with respect to the Rayleigh wavelength [2]. At low frequencies, the penetration
depth of the Rayleigh waves is very large and a significant amount of the vibrational
energy passes underneath the sheet pile wall. Increasing the depth improves the
insertion loss at low frequencies [7].
(a) (b)
Fig. 5. (a) Vertical insertion loss at 25 Hz and (b) vertical insertion loss
~
IL z ( x = 32m, k y , z = 0m, ω ) in the case a 12 m deep sheet pile wall is embedded in a homogeneous
halfspace. Superimposed is the dispersion curve of a Rayleigh wave in the y–direction.
axis, with second moment of area Izz, is too low to affect the transmission of vibration in
the frequency of interest. At 25 Hz, the sheet pile wall reduces the vibration levels at the
surface with insertion loss values around 3 dB. The insertion loss is fairly homogeneous
at the surface behind the sheet pile wall. The reduction in vibration levels, however, is
restricted to the top five meters of soil. In the present case, the efficiency of the sheet
pile wall is relatively low compared to other mitigation measures on the transmission
path. A larger reduction in vibration levels is predicted for sites with a soft soil, like that
of the test site in Furet [7].
5 Conclusions
Numerical simulations show that a stiff buried barrier like a block of stiffened soil or a
sheet pile wall can act as an effective wave impeding barrier. The mitigation measure
will be more effective at sites with a relatively soft soil. Increasing the barrier’s depth is
beneficial to obtain a reduction at low frequencies. Field tests are being carried out in
Spain and Sweden to confirm these results.
Acknowledgments. The results presented in this paper have been obtained within the
frame of the EU FP7 project RIVAS (Railway Induced Vibration Abatement
Solutions) under grant agreement No. 265754 [3]. The first author is a doctoral fellow
of the Research Foundation Flanders. The financial support is gratefully
acknowledged.
References
[1] Andersen, L., Nielsen, S.R.K.: Reduction of ground vibration by means of barriers or soil
improvement along a railway track. Soil Dynamics and Earthquake Engineering 25,
701–716 (2005)
[2] Woods, R.D.: Screening of surface waves in soils, Journal of the Soil Mechanics and
Foundation Division. Proceedings of the ASCE 94(SM4), 951–979 (1968)
[3] http://www.rivas-project.eu
[4] François, S., Schevenels, M., Lombaert, G., Galvín, P., Degrande, G.: A 2.5D coupled
FE–BE methodology for the dynamic interaction between longitudinally invariant structures
and a layered halfspace. Computer Methods in Applied Mechanics and
Engineering 199(23-24), 1536–1548 (2010)
[5] Nilsson, C.-M., Jones, C.J.C.: Theory manual for WANDS 2.1. ISVR Technical
Memorandum No. 975. University of Southampton (2007)
[6] Coulier, P., François, S., Degrande, G., Lombaert, G.: Subgrade stiffening next to the track
as a wave impeding barrier for railway induced vibrations. Soil Dynamics and Earthquake
Engineering 48, 119–131 (2013)
[7] Dijckmans, A., Ekblad, A., Smekal, A., Degrande, G., Lombaert, G.: A sheet piling wall as a
wave barrier for train induced vibrations. In: Proceedings of COMPDYN 2013, Kos Island,
Greece (2013)
Ground-Borne Vibration Mitigation Measures
for Turnouts: State-of-the-Art and Field Tests
1 Introduction
Within the frame of the EU FP7 project ’Railway Induced Vibration Abatement
Solutions (RIVAS)’, measures to reduce ground-borne noise and vibration for
tracks at grade were studied. Work package (WP) 3.3 of RIVAS focused on railway
infrastructure-based vibration reduction technologies for curves and turnouts (see
Fig. 1). For curves, three different types of soft USP have been tested, while for
turnouts a stiff USP [1] has been investigated. This paper summarizes the results of the
turnout tests as well as a state-of-the-art study on ground-borne vibration issues
describing vibration generating mechanisms and possibilities for cost-effective
mitigation measures [1].
Fig. 1. Components of a turnout with through and diverging routes. From [2].
A brief literature survey, mainly based on papers [2, 3, 4, 5] from the EC funded
INNOTRACK project, describes the main excitation mechanisms and shows the
influence of mitigation measures on dynamic wheel-rail contact forces. Dynamic
interaction between vehicle and track is more complex in turnouts than on tangent or
curved tracks. For example, wheel-rail impact loads with large magnitudes and
significant contributions from high-frequency vehicle-track interaction are generated
when the nominal wheel-rail contact conditions are disturbed at various locations in the
turnout, such as at wheel transfers between wing rail and crossing nose [2] .
Mitigation measures described in the literature mainly aim at reducing the vertical
wheel-rail contact force on the crossing. Possibilities to improve the design of the
crossing panel (crossing nose and wing rail) of a turnout have been identified. A
smooth wheel trajectory, determined by the combination of wheel and rail geometry, is
important to achieve low energy impacts on the crossing nose. Track (rail) geometric
irregularities are located at the switch and crossing panels, where the wheel transfers
Ground-Borne Vibration Mitigation Measures for Turnouts 549
from one rail to another. Optimization of these transition zones to reduce contact forces
is achieved by smoothing the wheel trajectory at the transitions, for example by
reducing the slope and depth of the transition at the crossing panel. Track
characterization measurements have indicated that the amplitude of the impact related
to the variation in track geometry is more important than the impact due to the variation
in track stiffness [3].
In the INNOTRACK project, the so-called ‘MaKüDe’ design developed by DB
Systemtechnik for improved rail profile geometry showed good performance (see
Fig. 2), in particular for normally worn wheel profiles, in both facing and trailing
moves [4]. In connection with reduced support stiffness (e.g. by use of more resilient
rail pads, see Fig. 3), this crossing panel design could lead to a significant reduction of
impact loads and potentially also a reduction of ground-borne vibration.
Fig. 2. Influence of transition geometry and wheel profile on maximum normal contact force at
the crossing (left bars), and corresponding equivalent stress (right bars). From [4].
200
150
100
50
0
230 200 160 120 25
Speed [km/h]
Fig. 3. Influence of vehicle speed and track stiffness (resilient rail pads) on maximum normal
contact force at the crossing. From [4].
Fig. 4. Amplification of vibration at about 8 m distance from a fixed crossing nose compared to
tangent track. The numbers after “EW” indicate curve radius of the turnout (e.g. EW 300 = radius
300 m). From [5].
Ground-Borne Vibration Mitigation Measures for Turnouts 551
40%
35%
Short
30% Medium
Long
25%
20%
15%
10%
5%
0%
4 6 8 10 15 >15
Radius [m]
Fig. 5. Statistical distribution of radius evaluated at deepest point of the wheel trajectory based on
CTM measurements. Influence of turnout type distinguished by curve radius of the diverging
route (Short: 500 m 1:12-1:14, Medium. 760 m 1:14, Long: 1200 m 1:18.5). From [6].
The CTM (Continuous Track Monitoring) coach developed by the Deutsche Bahn
(DB) is instrumented with axle box accelerometers. This coach can be used to evaluate
the performance and quality of turnouts by measuring the wheel trajectory at crossing
transitions (see Fig. 5). Different types of turnout, distinguished mainly by their curve
radius and the angle between the through and diverging routes (crossing angle), were
included in the investigation reported in [6]. For these turnouts, the estimated radius of
the wheel trajectory at its deepest point was evaluated. This radius gives an indication
of the vertical wheel acceleration and the vertical wheel-rail contact force.
According to Fig. 5, turnouts with small curve radius display a smaller radius of the
wheel trajectory indicating a sharper reversal of vertical wheel displacement at
transition from wing rail to crossing nose, and therefore higher acceleration and
wheel-rail impact load. The results in Fig. 5 could serve as an explanation for the results
in Fig. 4, since a smaller curve radius leads to higher impact loads and hence more
vibration.
of traffic is encouraging. However, the reason for the high insertion loss in the
frequency range from 20 to 40 Hz, which is significantly lower than what can be
expected for a nominal tangent track, is unknown (see Fig. 6). Possible explanations
include different maintenance status of the two compared turnouts, different excitation
mechanisms for turnouts compared to tangent track, different soil conditions, etc.
Fig. 6. Insertion loss by UBM (cstat= 0.06 N/mm3) for normal track and turnouts, see [7]
Fig. 7. RMS max vibration measurements at Rubigen with and without USP
Several vehicle and track parameters clearly influence the ground-borne vibration
induced at a turnout. Some mitigation measures have the potential of reducing the
vibration in a cost-efficient way, such as soft USP, resilient rail pads, and well
maintained rail profiles. A difficult challenge is to design the rail profiles of the
crossing panel such that it provides a smooth transition for different vehicle types and
speeds and for different wheel profiles (nominal and worn). The best approach to
reduce the turnout amplification is a system approach where the different parameters
are designed to interact in a harmonious and robust way.
In the remaining work of WP3.3, vehicle dynamics and wheel-rail contact forces in
turnouts will be simulated, aiming for an optimization of the crossing geometry that is
robust for a range of nominal and worn wheel profiles. In late 2012, SNCF carried out
a test campaign on a turnout to quantify the contribution of each part of the turnout
(joints, switch end, crossing), and therefore of each excitation mechanism (impact
load at different joints and crossing, parametric excitation), to the generated
ground-borne vibration. For a selection of turnouts, the influence of rail profiles on
vibration excitation will be studied by Empa by analyzing axle box acceleration
measurements repeated over time. These measurements will show the influence of
crossing panel geometry (maintenance status) on the acceleration of the wheelsets.
The benefit of the new mitigation measure (SBB ‘Turnout 2015’), which has recently
been installed at Wichtrach between Bern and Thun, will be evaluated by vibration
measurements.
The research leading to these results has received funding from the European Union
Seventh Framework Programme under grant agreement no 265754.
554 R. Müller et al.
References
[1] Müller, R., et al.: Description of the vibration generation mechanism of turnouts and the
development of cost effective mitigation measures. RIVAS Deliverable D3.6, Berne,
Switzerland (March 2013)
[2] Kassa, E.: Dynamic train-turnout interaction, Mathematical modelling, numerical
simulation and field testing. Ph.D. thesis, Department of Applied Mechanics, Chalmers
University of Technology, Gothenburg, Sweden (2007)
[3] Pålsson, B.: Towards optimization of railway turnouts. Licentiate thesis, Department of
Applied Mechanics, Chalmers University of Technology, Gothenburg, Sweden (2011)
[4] INNOTRACK Deliverable 3.1.5: Recommendation of, and scientific basis for, optimization
of switches & crossings - part 1, 30 pp + 2 annexes (2009)
[5] Müller, R.: Mitigation measures for open lines against vibration and ground-borne noise: a
Swiss overview. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C.,
Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for
Rail Transportation Systems. NNFM, vol. 99, pp. 264–270. Springer, Heidelberg (2008)
[6] Lölgen, T., et al.: Classification of track conditions with respect to track irregularity
database, RIVAS Deliverable D2.3, Munich, Germany (March 2013)
[7] Müller, R.: Dämmwirkung von Unterschottermatten. Rothrist Bereich III Weichen. SBB
Internal Report Mai 2000, Berne, Switzerland (2000)
[8] Said, S., Auersch, L., Knothe, E.: Mitigation measures for switches in Rubigen and Le
Landeron, Switzerland. RIVAS Report 2012, Berlin, Germany (2012)
Reducing Railway Induced Ground-Borne Vibration
by Using Trenches and Buried Soft Barriers
1 Introduction
The present study is part of the EU funded project RIVAS [7]. Here, results are
presented for a parametric study of the effectiveness of open and soft-filled trenches for
different soil conditions. In Section 2 the methodology used in this study is briefly
introduced. In Sections 3 and 4, parametric studies of open trenches and buried soft
barriers are presented respectively. Conclusions are given in Section 5.
2 Methodology
The geometry of a railway line with a trench adjacent to it is two-dimensional but the
loading due to a train introduces a dependence on the third (longitudinal) dimension.
So-called 2.5D numerical simulations have been carried out by project partners ISVR
and KU Leuven using two different models [8, 9]. In these models, the profile of the
track, ground and trench are modelled in 2D using a combination of finite elements and
boundary elements (e.g. Fig. 1). By assuming homogeneity of the geometry and
material properties in the track direction, the response along this third dimension is
formulated in terms of the wavenumber. A Fourier transform is used to recover the
three-dimensional response. This method is computationally more efficient than a full
3D approach.
-9
x 10
2.5
32m
2
Mobility [m/s/N]
1.5
0.5
0
0 50 100
Frequency [Hz]
Fig. 1. Example sketch of track, soil layers and Fig. 2. Vertical insertion loss at 32 m from a
trench used in 2.5D model vertical point force. KU Leuven results (solid
lines) and ISVR results (circles)
Initially, independent analyses were conducted by both project partners for three
benchmark cases with typical soil conditions: (i) a homogeneous half space, (ii) a soft
upper layer over a stiffer soil (e.g. Fig. 1), and (iii) a stiffer upper layer over a softer
sub-layer. In each case, transfer mobility and insertion loss were considered for a single
trench type. Fig. 2 shows results of one of the benchmark cases. In this case, a point
force is applied directly on the soil at the centre of a homogeneous half-space with a
rectangular open trench between the force and receiver locations. The track structure
was not included in this case. The open trench is 0.5 m wide, 11.75 m deep and 5.6 m
away from the force. The homogeneous half-space has a shear velocity Cs = 250 m/s,
Reducing Railway Induced Ground-Borne Vibration 557
a compressional wave velocity Cp = 1464 m/s, a Young’s modulus E = 3.61 x 108 Pa, a
Poisson’s ratio ν = 0.485 and a density ρ = 1945 kg/m³. Both partners used boundary
elements to simulate this case, although different approaches were used. KU Leuven
used Green’s functions for a half-space, including layers if necessary, whereas ISVR
used a full space Green’s function but needed to mesh the ground surface to sufficient
distance. Calculations were performed in the frequency range from 1 to 100 Hz, with
every 1 Hz for KUL and every 5 Hz for ISVR. The vertical transfer mobility at 32 m
from the point force is shown in Fig.2. At all frequencies, the results agree reasonably
well.
Subsequent to the benchmarking, ISVR undertook a parametric study aimed at
investigating the effectiveness of open trenches and buried soft barriers (trench filled
with soft material).
3 Open Trench
Calculations are presented here for an open trench. A two-layer ground was considered.
The upper layer has a shear velocity Cs = 150 m/s, a compressional wave velocity
Cp = 298 m/s, a Young’s modulus E = 1.07 x 108 Pa, a Poisson’s ratio ν = 0.33 and a
density ρ = 1800 kg/m³. The underlying half-space has a shear velocity Cs = 600 m/s, a
compressional wave velocity Cp = 1191 m/s, a Young’s modulus E = 1.92 x 109 Pa, a
Poisson’s ratio ν = 0.33 and a density ρ = 2000 kg/m³. The depth of the trench is kept
fixed at 6 m (unless otherwise stated). A line force is applied to the top of the upper
layer. The line source is simulated by a series of point loads at various locations
representing the axle locations of a four-coach Bombardier Regina EMU train, which
is used as a reference in RIVAS. The parameters varied are the width of the trench, w,
the distance from the force point, d, and the depth of the upper layer of soil, h1.
Fig. 3 shows the insertion losses at 32 m away from the force for five different widths
of open trench: 0.25, 0.75, 1.25, 2.25 and 3.25 m. In each case the nearest edge of the
trench is 3 m away from the force point. These results clearly indicate that the width of
the open trench does not have much influence on the insertion loss.
Next, the location of the trench is varied. Four locations are studied: 1, 3, 5 and 7 m
away from the force point. Fig. 4 shows the insertion losses at the same positions as
before. The location of the trench also clearly has little influence on the insertion loss.
Although there are some differences, these do not appear to be systematic. Therefore, a
fixed location, 3 m away from the force, was used in the remainder of the study to
reduce the number of variables and associated computational cost.
558 J. Jiang et al.
32 m 32 m
40 40
w=0.25m d=1m
30 w=0.75m d=3m
Insertion loss [dB]
10 10
0 0
1 10 100 1 10 100
Frequency [Hz] Frequency [Hz]
Fig. 3. Insertion loss for an open trench of Fig. 4. Insertion loss for an open trench at
different widths obtained at 32 m away from different distances from the force point
the force. Trench is 3 m away from the force, obtained at 32 m away from the force. Trench is
ground with h1 = 3 m. 6 m deep, ground with h1 = 3 m.
To investigate the influence of the ground condition, five different cases are
considered, with upper layer depths of 0, 3, 6, 9 and ∞ m. Fig. 5 shows the insertion
losses obtained at 32 m away from the force using an open trench which is 6 m deep,
0.25 m wide and 3 m away from the force.
As seen in Fig. 5, the 6 m deep open trench has no effect below 8 Hz for any of the
grounds considered. For the stiff homogeneous half-space there is no benefit below
30 Hz. In this case the criterion that the depth should be at least 0.6 times the Rayleigh
wavelength [1, 2] is only met above 60 Hz; in this frequency region an attenuation of
over 10 dB is found. For the soft homogeneous half-space the criterion relating to the
Rayleigh wavelength is met at about 15 Hz and the attenuation is greater than 10 dB for
frequencies above this, increasing to around 30 dB by 30 Hz.
For the layered ground with h1 = 9 m, the trench (6m deep) does not cut the upper
layer completely. The insertion loss resembles that of the half-space of soft material
although with a reduced performance at higher frequencies. For the layered ground
with h1 = 6 m, the trench does cut the upper layer completely. Here, the insertion loss
rises sharply between 10 and 20 Hz, which is the region where surface waves become
localised in the upper layer. Consequently, between 10 and 30 Hz the benefit of the
trench is greater than for the half-space of soft material.
For a layered ground with h1 = 3 m, the 6 m deep trench cuts the upper layer and
penetrates the underlying stiffer soil by a further 3 m. Here, the benefit of the trench
increases sharply between 20 and 30 Hz which, again, is the range where waves start to
propagate within the upper layer.
To investigate this further, a case is considered for the ground with layer depth
h1 = 3 m where the trench depth is reduced to 3 m, the same as the depth of the upper
layer. Results are shown in Fig. 6. For this case, for frequencies up to 40 Hz, a similar
insertion loss is achieved as with the 6 m deep trench. At higher frequencies the deeper
trench has some additional benefit. However, it seems that for this soil there is little
benefit in taking the trench any deeper than the upper surface layer into the stiffer soil.
Reducing Railway Induced Ground-Borne Vibration 559
32 m 32 m
40 40
h1=0m h1=3m, h=6m
h1=3m
h1=3m, h=3m
h1=6m
20 h1=9m 20
h1=inf
10 10
0 0
1 10 100 1 10 100
Frequency [Hz] Frequency [Hz]
Fig. 5. Effect of ground condition on the Fig. 6. Effect of trench depth for a 3 m deep
insertion loss of an open trench obtained at soil layer of an open trench obtained at 32 m
32 m away from the force. Trench is 3 m away away from the force. Trench is 3 m away from
from the force with 0.25 m width. the force with 0.25 m width.
Some results are shown in Fig. 7. This shows the insertion loss at 32 m away from
the force for a 0.25 m wide, 6 m deep trench located 3 m away from the force point, and
the upper layer depth h1=3m here. The performance of two fill materials (A1 and B1)
and the open trench are compared. From the figure it can be seen that an open trench
gives the best attenuation. Both materials have a much lower performance, with an
attenuation limited to about 5 to 10 dB. Material B1 performs better than material A1,
which is stiffer (see Table 1).
560 J. Jiang et al.
To illustrate the results further, Fig. 8 shows the results for each material in a single
graph with the lines corresponding to the different ground types. The material A1 is not
effective below 40 Hz for any ground, while material B1 is effective from 10 Hz for the
soft homogeneous soil and the soil with 6 m or 9 m deep upper layer.
32 m, h1=3 m
40 20
B1 (a) A1
10
30
Open trench 0 h1=0m
1 10 100 h1=3m
20
20 h1=6m
(b) B1 h1=9m
10
10
h1=inf
0 0
1 10 100
10 100 Frequency [Hz]
Frequency [Hz]
Fig. 7. Insertion loss of trenches filled with Fig. 8. Insertion loss of trenches filled with
different materials obtained at 32 m away different materials obtained at 32 m away from
from the force. Trench is 3 m away from the force. Trench is 3 m away from the force
the force point with 0.25 m width and 6 m point with 0.25 m width and 6 m depth.
depth. h1=3 m.
Results are shown in Fig. 9 for the different variants of material group A as identified
in Table 1. From this it appears that varying the Young’s modulus (Fig. 9a) has more
effect than varying the shear wave speed (Fig. 9b), with a lower Young’s modulus
being more beneficial.
An ideal material therefore has a low Young’s modulus, although in practice the
value that is achievable is limited by the ability of a material to sustain the confining
pressure in the ground.
A3 30
Insertion loss [dB]
w=0.5 m
0
w=0.75 m
10 100 20
(b) Differnt shear wave velocity
20
A1 10
10 A4
A5 0
0
1 10 100 10 100
Frequency [Hz] Frequency [Hz]
Fig. 9. Insertion loss of trenches filled with Fig. 10. Effect of trench width on insertion
material group A. Trench is 3 m away from the loss of using martial B1 to fill a trench 6 m
force point with 0.25 m width and 6 m depth; deep 3 m away from the force. h1=3 m.
response is obtained at 32 m away from the
force. h1=3 m.
Reducing Railway Induced Ground-Borne Vibration 561
Fig. 10 shows the results of changing the width of the trench for the case of material
B1 and h1=3 m. It indicates that a wider trench gives a higher insertion loss at high
frequencies when such soft materials are used, although the width does not affect the
frequency at which attenuation commences.
To show the effect on train vibration, Fig. 11 shows an example of a predicted RMS
velocity spectrum at 32 m during a train pass-by, obtained using the ISVR vibration
model (TGV) [10]. By applying the insertion losses from the previous sections as
corrections to the predicted vibration spectra, estimates are given of the response for the
open trench and trenches filled with materials A1 and B1. The vibration without
mitigation can be seen to rise sharply at about 10 Hz for this ground due to the layering.
While the hypothetical open trench can reduce the level of vibration effectively, the
more practical solution (B1, in Fig. 11) has a modest but still significant effect. The
improvement achieved commences at a frequency just above the first rise in vibration
associated with the upper layer.
-4 A1
10
-5
10
-6
10
10 100
One-third octave band centre frequency [Hz]
Fig. 11. One-third octave band RMS spectra of vertical free-field velocity for the ground with
h1=6 m at 32 m from track. The trenches are 6 m deep and 3 m away from the force.
5 Conclusion
Mitigation applied in the transmission path as considered here is one of the few
measures with the potential to reduce vibration at lower frequencies. Overall, under all
ground conditions investigated in this paper, the notional rectangular open trench
performs better than the other constructions. The width of an open trench has little
influence on the performance, whereas increasing the width of a filled trench reduces
the stiffness of the in-fill, improving the performance of the trench. Likewise, fill
materials with lower Young’s modulus give higher insertion losses. The distance of the
trench from the track is of secondary importance. When the soft soil layer is 6 m deep,
a 6 m deep filled trench could provide benefit all frequencies above about 10 Hz. At
these low frequencies, trenches have the potential to significantly mitigate vibration
impacts. Field tests are proposed to confirm these conclusions.
562 J. Jiang et al.
Acknowledgments. The research leading to these results has received funding from
the European Union Seventh Framework Programme under grant agreement number
265754, project RIVAS (Railway Induced Vibration Abatement Solutions) [7].
References
[1] Richart, F.E., Hall, J.R., Woods, R.D.: Vibrations of soils and foundations. Prentice Hall
(1970)
[2] Beskos, D.E., Daskupta, B., Vardoulakis, I.G.: Vibration isolation using open or in filled
trenches. Journal of Computational Mechanics 1(1), 43–63 (1986)
[3] May, T., Bolt, B.A.: The effectiveness of trenches in reducing seismic motion. Earthquake
Engineering and Structural Dynamics 10(2), 195–210 (1982)
[4] Hung, H.H., Yang, Y.B., Chang, D.W.: Wave barriers for the reduction of train-induced
vibrations in soils. Journal of Geotechnical and Geo-Environmental Eng. 130(12),
1283–1291 (2004)
[5] Garcia-Bennett, A., Jones, C.J.C., Thompson, D.J.: A numerical investigation of railway
ground vibration mitigation using a trench in a layered soil. In: Maeda, T., Gautier, P.-E.,
Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos,
P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 118,
pp. 315–322. Springer, Heidelberg (2012)
[6] Jones, C.J.C., Thompson, D.J., Andreu-Medina, J.I.: Initial theoretical study of reducing
surface-propagating vibration from trains using earthworks close to the track. In:
EURODYN, Leuven, Belgium (2011)
[7] http://www.rivas-project.eu
[8] Nilsson, C.-M., Jones, C.J.C.: Theory manual for WANDS 2.1, ISVR Technical
Memorandum No. 975. University of Southampton (2007)
[9] Lombaert, G., Degrande, G., Clouteau, D.: Numerical modelling of free field traffic
induced vibrations. Soil Dynamics and Earthquake Engineering 19(7), 473–488 (2000)
[10] Sheng, X., Jones, C.J.C., Thompson, D.J.: TGV - A computer program for train-induced
ground vibration. ISVR Technical Memorandum No. 878, University of Southampton
(December 2001)
Pantograph Area Noise and Vibration Transmission
Characteristics and Interior Noise Reduction Method
of High-Speed Trains
J.Q. Guo1,2, J.M. Ge1, Z.J. Sun2, S.Q. Liu2, Y.J. Zhao2, and J.S. Lin2
1
TongJi University,
No.1239 Siping Road, Shanghai, P.R. China,
hfutguo@163.com
2
National Engineering Research Center for High-speed EMU,
CSR Qingdao Sifang Co., Ltd.,
No.88 Jinhongdong Road, Chengyang District, Qingdao, P.R. China
Summary. For high-speed trains, the interior noise of the pantograph area is
usually higher than other areas. A lot of achievements have been made to reduce
the aerodynamic noise of the pantograph and its facilities. Yet, the exterior noise
and vibration of the pantograph area is still much higher while the train operates at
300 km/h or more. So research on the vehicle structure is necessary.
Aiming at reducing the interior noise, operating tests were carried out to get
the noise and vibration of the pantograph area and the force on the pantograph
mounting. Spectrum characteristics and statistical laws all prove that the interior
noise is closely related to the ceiling vibration.
In order to reduce the vibration, an absorber structure is proposed for the
pantograph mounting. A specially designed real-car-simulating test bench is used
to evaluate the application performance of the absorber. The test results show that
the noise and vibration are significantly reduced after adopting the absorber.
The research of this article proposes a new thinking and method for interior
noise reduction of the pantograph area from the aspect of vibration attenuation.
1 Introduction
Pantograph noise is one of the main noise sources of high-speed trains. At the 350 km/h
speed level, pantograph noise becomes one of the biggest noise source since
aerodynamic noise becomes as important in some situations as wheel-rail noise [1,2].
Therefore, pantograph noise control becomes an inevitable and important issue for
high-speed train development. Numerous researchers have done a lot of work in this
field.
Kurita and Wakabayashi [3] have done some comparative test research about two
pantograph structures on FASTECH360 high-speed test trains, and have evaluated the
application performance of low-noise pantograph and pantograph noise insulating
boards at the speed of 360 km/h. Yang and Zheng [4] have researched the influences of
the current-guide cover on aerodynamic noise of pantograph by numerical simulation
Taking the reduction of the interior noise level as the research purpose, the exterior
noise on the vehicle surface was tested rather than the pass-by noise at 25 m away.
B&K surface microphones were used for exterior noise measurement while the interior
noise was tested by a microphone array.
The interior standard point (1200 mm above the floor) noise of the pantograph area
and the end of the compartment without pantograph was shown in Fig. 1. The noise of
the pantograph area was 5 dB (A) to 8 dB (A) higher at different train speed.
Moreover, as the train speed increase, the difference value becomes larger. Fig. 2
shows the influence in terms of the spectrum. It can be observed that the exterior noise
(roof) has a broadband peak value between 200 Hz and 3150 Hz. The noise level of the
pantograph area was about 10 dB (A) higher in the whole band. During transmission,
the noise above 1000 Hz is attenuated rapidly, so that the noise of these two areas
above 1000 Hz tends to be equal and the noise below 1000 Hz is much more important
for interior noise control.
According to the principle of adding dB values, if one sound is 10 dB higher than
the other, the lower sound can be ignored. So the exterior noise of the pantograph area
is dominated by the aerodynamic noise of pantograph and its subsidiary facilities. For
vehicle interior noise, the difference of 5-8 dB (A) illustrates that the pantograph
noise is at least 3 dB (A) larger than the other noise.
Pantograph Area Noise and Vibration Transmission Characteristics 565
To reduce the influence of the pantograph noise on vehicle interior noise, a heavy
noise insulation enclosure on the ceiling is usually used, so that the noise of
pantograph area can nearly keep the same level with the other areas.
Fig. 1. Interior noise (1200 mm above the Fig. 2. One-third octave spectra of exterior
floor) level at different train speed and interior noise at 350 km/h
In the frequency domain, the roof vibration and ceiling vibration are significant
different, as shown in Fig. 4. To compare with the noise spectra, the vibration levels
are also represented by their one-third octave spectra. The following conclusions are
obtained from the figure:
566 J.Q. Guo et al.
The higher the frequency, the bigger the roof vibration level becomes while the
ceiling vibration level observably attenuates above 1000 Hz. This variation trend
is similar with noise spectrum (see Fig. 2);
Compared with the end of the compartment without pantograph, the ceiling
vibration level of the pantograph area is much higher below 1000 Hz, especially
below 630 Hz. The 10 dB difference in Fig. 3 is mainly caused by the vibration
in these frequency bands.
As the pantograph system is the only difference between the two areas, we can
firmly conclude that the influence of the pantograph on the ceiling vibration is
restricted to the frequency below 630 Hz.
The force acting on the pantograph mountings can be measured by sensors between
the pantograph and the vehicle when the train is operating on line. The forces at
350 km/h are shown in Fig. 5.
Obviously, the Z axis (vertical) force is much bigger than that of the X axis
(transverse) and the Y axis (longitudinal). Therefore, only the Z axis force is
considered in the following experiment shown in section 5.
Fig. 5. The forces acting on pantograph Fig. 6. Schematic diagram of pantograph noise
mountings and vibration transmission approach
Theoretically, the noise and vibration sources of the pantograph area fall into four
categories, as shown in Fig. 6:
All the four sources above can result in train body vibration and then radiate noise.
There are two possible theoretical models of train body vibration: plane vibration due
to the mass law and bending vibration. The noise generated by the former model will
be transmitted in the form of plane waves. This kind of noise is usually called
airborne noise. The noise generated by the latter model will be transmitted in the form
of bending waves. This kind of noise is usually called structure-borne noise.
4 Solutions
There are two aspects of vibration attenuation structure design for pantograph area:
one is vibration absorber for pantograph mounting; the other is vibration damping
structures for the vehicle section. The last measures often cause weight gain.
Proceeding from the former aspect, this article adopts vibration absorber for
pantograph mounting, so that the vibration can be reduced before it is transferred to
the vehicle.
Safety, aerodynamic performance and maintainability must be considered during
the design process of the vibration absorber. Optimized over a period of time, an
elliptical-cone-shaped vibration absorber was finally designed [9], as shown in Fig. 8.
The interior and outer enclosures are made of metal material with 70° cone angle.
This structure ensures the safety of the vehicle even if the damping material is
damaged. The gap between the enclosures is filled with damping material. The
performance of the absorber mainly depends on the damping material property.
Insulator
Absorber
Carbody
In order to make experimental results much closer to the real-car performance, the
evaluation experiment of the absorber was carried out on a real-car-simulating test
bench, as shown in Fig. 9. The bench is 5 m in length taking into account the need of
pantograph mounting and the vibration model of the roof. The roof of the bench is made
of real-car aluminum extruded sections while the walls around are concrete. The roof
568 J.Q. Guo et al.
vibration and the interior noise before and after adopting the absorber were tested when
the mountings on the roof were excited. Then the difference was calculated to represent
the application performance of the absorber.
To eliminate the influence of differences in excitation force, the test data was
normalized based on the excitation force. Consequently the unit force vibration level
and unit sound pressure level were obtained.
The vibration level for a unit force in each one-third octave can be calculated as
follows:
aRMS ,i
VLi = 20 * log (5.1)
a0 f i
The sound pressure level for a unit force in can be calculated as follows:
pe,i
SPLi = 20*log (5.2)
p0 fi
Fig. 9. Application performance of the Fig. 10. The relationship of interior standard
absorber point (1.2m) noise and exciting force
6 Conclusions
In this paper, the noise and vibration sources of the pantograph area are divided into
four categories and a qualitative analysis is made of their way of influence on interior
noise. But the quantitative contribution analysis of each source needs more work.
Nonetheless, the data tested on CRH380A high-speed trains during on-line operation
has illustrated that the pantograph noise has significant influence on interior noise.
Moreover, structure bending vibration is shown to be a most important means of noise
transmission. Based on the above analysis, a special absorber for the pantograph
mounting is designed.
570 J.Q. Guo et al.
References
[1] Thompson, D.: Aerodynamic noise, Chapter 8 in Railway noise and vibration, pp. 307–310.
Elsevier Ltd., The Netherlands (2009)
[2] Mellet, C., Létourneaux, F., Poisson, F., Talotte, C.: High speed train noise emission: Latest
investigation of the aerodynamic/rolling noise contribution. Journal of Sound and
Vibration 293, 535–546 (2006)
[3] Takeshi, K., Masaaki, H., Haruo, Y.: Reduction of pantograph noise of high-speed trains.
Journal of Mechanical Systems for Transportation and Logistics 3(1), 63–74 (2010)
[4] Yang, F., Zheng, B., He, P.: Numerical simulation on aerodynamic noise of power
collection equipment for high-speed trains. Computer Aided Engineering 19(1), 44–47
(2010)
[5] Takaishi, T., Yamazaki, N., Sueki, T., Uda, T.: Recent studies on aerodynamic noise
reduction at RTRI. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson,
J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation
for Rail Transportation Systems. NNFM, vol. 118, pp. 411–418. Springer, Heidelberg
(2012)
[6] Chen, Z., Guo, J., Wang, D., Ge, J.: Study on vibration of high-speed train near pantograph.
In: Proceeding of the 10th National Conference on Vibration Theory and Application,
Nanjing, China, vol. 1, pp. 268–270 (2011)
[7] Liao, X., Liang, J., Sun, Z., Ge, J.: Pantograph’s impact on the noise of high-speed train.
Technical Acoustics 30(4), 171–174 (2011)
[8] ISO 1683:2008. Acoustics – Preferred reference values for acoustical and vibratory levels
[9] Lin, J., Sun, Z., Song, X., Guo, J.: The vibration attenuation device and its method of
pantograph system. China invention patent. CN103129400A (June 05, 2013)
Micro-Pressure Wave Emissions from German
High-Speed Railway Tunnels – An Approved Method for
Prediction and Acoustic Assessment
DB Systemtechnik GmbH,
Voelckerstrasse 5, 80939 Munich, Germany
christian.gerbig@deutschebahn.com
1 Introduction
At present, several new high-speed lines are in the planning stage or under construction
in Germany. For safety reasons, the tunnels will be twin tube single-tracked with a free
cross-sectional area of about 60 m² per tube. Due to the small cross-section in
conjunction with slab track and high train speed the emission of micro-pressure waves
(MPW) is an important issue, which must be considered already during the planning
process.
In the homologation procedure for high-speed lines it is stated that the impact on the
environment has to be predicted and any harmful environmental effects are to be
avoided. Up to now, no generally accepted methods suitable for an assessment of the
effects of micro-pressure waves from tunnels were available, neither on a national level
in Germany nor on a European level [1]. Hence, in the last years DB Systemtechnik
developed a prediction method for noise levels from micro-pressure waves and
guidance values for different situations, which are considered as upper limits for noise
levels. This combination of prediction and limit values is essential for the proper choice
of efficient measures to reduce the micro-pressure waves from tunnels. It guarantees
the required protection on the one hand while avoiding over-engineered solutions on
the other hand.
Train
Generation of Propagation of
compression wave compression wave
Radiation of micro-
pressure wave
Without countermeasures the micro-pressure wave can become clearly audible even
at distances far away from the tunnel portal. The perceptibility depends both on the
spectrum and on the amplitude of the pressure pulse. The micro-pressure wave is
particularly audible when the duration is short and the amplitude is large.
The spectra of MPW are characterized by pronounced contributions at low
frequencies (e.g. < 100 Hz). Typical third octave band spectra from MPW measured
during test runs at distances 50 m, 100 m and 485 m from the portal are shown in Fig. 2.
dB
110.0
100.0
90.0
80.0
70.0
60.0
50.0
40.0
30.0
4 8 16 31.5 63 125 250 500 1 2
Frequency [Hz]
Hz kHz
Fig. 2. Third octave band spectra of a typical micro-pressure wave. Measuring points located at
50 m, 100 m and 485 m from the exit portal [2].
In order to efficiently plan countermeasures for new tunnels on high-speed lines as well
as to predict residual micro-pressure waves even after the implementation of
countermeasures, a detailed and well established prediction process was developed by
DB Systemtechnik. By using a four-stage model for (i) generation of the compression
wave, (ii) the steepening process during the propagation inside the tunnel, (iii) the
emission of the MPW wave at the opposite tunnel portal and (iv) the transmission of
the MPW from the portal, all relevant acoustical sound levels can be calculated at the
relevant receiver points [3].
574 C. Gerbig and M. Hieke
•L
L = 13.5 log ( s / 25 m) dB for 10 m ≤ s ≤ 50 m (3.1)
•L
L = [4 + 16.5 log ( s / 50 m)] dB for 50 m ≤ s ≤ 200 m (3.2)
•L
L = [14 + 20 log ( s / 200 m)] dB for 200 m < s (3.3)
In Fig.3 the trend of the propagation loss corresponding to Eq. 3.1 to 3.3 is shown
with some measured values.
20
10
(reference distance s = 25 m)
measurements
directly on track
propagation loss in dB
-10
measurement point measurements in a residential
16 m from center of the track area, influence from
-20 background noise is possible
measurements
-30 measurements in a directly on track
residential area
-40
0 250 500 750 1000 1250
distance s between portal and immission point [ m ]
Fig. 3. Propagation loss of MPW corresponding Eq. 3.1 to 3.3 and some measured values [2]
Reference conditions for the prediction have been defined. The meteorological data
is set to be p = 1013 kPa, air humidity 70 %, air temperature 10 °C outside and 12 °C
inside the tunnel and air velocity 0 m/s inside the tunnel. The maximum regular train
speed of the line is to be used. Two reference trains have been defined: a high-speed
train (v > 230 km/h) with shape of ICE1 (see Fig. 4, left side) and a conventional loco
hauled train with shape of BR 101 (see Fig. 4, right side). For double-tracked tunnels it
is assumed that no train encounter occurs when the train passes through the tunnel.
Micro-Pressure Wave Emissions from German High-Speed Railway Tunnels 575
Fig. 4. Reference trains – left: high-speed train with shape of ICE1; right: conventional train with
shape of BR 101 [4]
During the years 2007 to 2009, DB Systemtechnik has conducted a study how to assess
and to limit the immission of micro-pressure waves. The purposes were to exclude
negative health effects for persons staying temporarily in close vicinity to the tunnel
entrances (e.g. railway workers), to avoid a strong audible boom close to the tunnel
portal as well as annoyance of residents and effects on the structures of buildings
including secondary acoustical effects due to vibrations excited by low frequencies in
the spectrum of micro-pressure waves. The following boundary conditions were
defined:
Based on the above stated requirements, a set of quantities and guidance levels was
elaborated. An overview is given in Fig. 5.
Two representative receiver points were chosen. The first one is located at a fixed
distance of 25 m from the portal and at a height of 1.5 m above rail head (representative
for the area, where people might temporarily stay but outside the near field of the sound
emission). The second point is chosen to be at the front of the nearest building 3.5 m
above the ground.
576 C. Gerbig and M. Hieke
Nearest Tunnel
building All requirements have portal region
to be fulfilled and give the
basis for the design of
3,5 m countermeasures
above ground
25 m
Threshold value for the C-weighted sound Threshold value for the C-weighted
exposure level dependent on type of use sound peak level
LpC,peak < 115 dB(C)
Assessment of the A-weighted sound
exposure level according to the German traffic
noise protection directive
Fig. 5. Schematic design of locations and levels for the assessment quantities
Concerning the protection against hearing damage, the European directive 2003/10/EC
[6] gives clear guidelines based on the C-weighted sound peak level LpC,peak. Although
the C-weighting of levels is not very common in railway noise assessment, it takes into
account, that most of the micro-pressure wave energy is found at low frequencies (<
100 Hz), even extending into in the infrasound region ( < 20 Hz).
The directive defines a lower actuation level at LpC,peak =135 dB(C), which should not
be exceeded. Based on this definition, the following guidance level was recommended:
• At a distance of 25 m from the tunnel portal a C-weighted sound peak level LpC,peak =
115 dB(C) must not be exceeded.
With this limit value at 25 m distance, the lower actuation level LpC,peak = 135 dB(C)
will not be exceeded at any point even very close to the tunnel portal. With this limitation,
also secondary acoustical effects caused by the excitation of vibrations in buildings are
avoided. Potential damage to nearby buildings is excluded as well.
These guidance levels ensure that residents are protected against significant
disturbance due to MPW, whether it is direct or secondary acoustical effects or
vibrations.
Independent of the acoustic assessment of MPW on the basis of C-weighted levels,
the A-weighted sound exposure level has to be assessed according to the German traffic
noise regulations. The sum level of A-weighted MPW-immissions and noise rating
levels corresponding to the German federal traffic noise protection directive
(16. BImSchV) [5] should be determined. The result is to be compared to the relevant
limit value. In the case that the A-weighted MPW-immissions are 20 dB or more below
the limit value according to 16. BImSchV, the influence of the MPW is insignificant
and the contribution is negligible.
There are different possibilities for reducing MPW-emissions: measures on train (e.g.
stretch “train nose”), infrastructure (e.g. portal hoods) and operating (e.g. reducing train
speed). It has been shown in planned constructional arrangements for reduction of
MPW-emissions in Germany during the last three years that hoods at the tunnel
entrance are the most efficient reduction measure and have the best cost-benefit ratio.
Different hoods were already built and these tunnels will be brought to service in the
coming years. The first German tunnel with hoods – the Katzenbergtunnel – was
brought to service in December 2012 [7]. In Fig. 6 the realized hood at the southern
portal is shown.
6 Conclusions
The presented prediction and assessment procedure is the basis for all current German
tunnel projects either in the planning phase or under construction. It ensures dependable
planning of measures for reducing MPW, so that planning reliability can be achieved at
an early design stage.
578 C. Gerbig and M. Hieke
The coordination process with the German national railway authority (EBA) and the
German Federal Environment Agency (UBA) was finalized in 2012. The
authority-approved corresponding regulations in the standards of DB Netz were put
into effect on Feb, 1st 2013 [4].
References
[1] Gerbig, C., Degen, K.G.: Acoustic assessment of micro-pressure wave emissions from
high-speed railway tunnels. In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B.,
Nelson, J.T., Schulte-Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 118, pp. 389–396. Springer,
Heidelberg (2012)
[2] Gerbig, C., Hieke, M.: Mikrodruckwellen-Emissionen an Tunnelportalen – Prognose,
Akustische Bewertung und Minderungsmaßnahmen. In: Lutzenberger, S., et al. (eds.)
Fachtagung 2012 - Bahnakustik – Infrastruktur, Fahrzeuge, Betrieb, pp. 71–80.
Bahn-Fachverlag (2012)
[3] Hieke, M., et al.: Assessment of micro-pressure wave emissions from high-speed railway
tunnels. In: Proceedings WCRR 2011 (2011)
[4] DB Netz AG: Modul 853.1002A01 in Richtlinie 853 Eisenbahntunnel planen, bauen und
instand halten, available at DB Kommunikationstechnik GmbH, Karlsruhe,
dzd-bestellservice@deutschebahn.com[5] German federal traffic noise protection directive:
Sechzehnte Verordnung zur Durchführung des Bundes-Immissionsschutzgesetzes
(Verkehrslärmschutzverordnung – 16. BImSchV) (1990)
[5] Directive 2003/10/EC of the European Parliament and of the Council, minimum health and
safety requirements regarding the exposure of workers to the risks arising from physical
agents (noise) (2003)
[6] Hieke, M., Gerbig, C., Tielkes, T.: Mastering micro pressure wave effects at the
Katzenbergtunnel – Design of measures, prediction of efficiency and full scale test
verification. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M., Nelson, J.T.,
Thompson, D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and Vibration Mitigation
for Rail Transportation Systems. NNFM, vol. 126, pp. 505–513. Springer, Heidelberg
(2015)
Three Noise Mitigation Measures
for Steel Railway Bridges
H. Venghaus
ACCON GmbH,
Gewerbering 5, D-86926 Greifenberg, Germany
helmut.venghaus@accon.de
1 Introduction
Steel bridges, as a part of railway lines, can cause considerable noise and associated
annoyance, especially when they are located close to residential areas. Noise radiated
by the structure of these bridges increases the radiated noise by up to 15 dB(A)
compared to a normal track. Nevertheless, steel bridges are still being constructed in
urban areas due to their low construction height compared to other bridge designs.
Various investigations have been undertaken to achieve noise reduction of steel
bridges without requiring them to be replaced or reconstructed. The common measure
is to add damping to the bridge, either with mechanical dampers or sheeting the main
radiating surfaces by constraint layers.
In recent years, some investigations were carried out based on decoupling the source
(wheel-rail-contact) from the structure of the bridge by inserting high resilient
baseplates underneath either the rails or sleepers. Depending on the resulting resonance
frequency considerable noise reductions were achieved.
Most studies related to mitigation of steel bridge noise concentrate on high level
numerical calculations like FEA and/or experimental measurements. The present paper
is trying to establish a more simplified estimation procedure, based on measurements at
the existing structure of the bridge and an easy to use mathematical process to
determine which type of mitigation method will achieve best results.
2 Mathematical Basics
The majority of noise from a bridge´s structure is radiated from the wide linked areas
such as girder plates and the base plate of the bridge. These sources are discussed below.
sleeper
stabilizing structure
Fig. 1. Sketch of a top hat rail bridge with solid web girders
The sound pressure field radiated by a vibrating surface (see Fig.2) surrounded by an
infinite baffle can be calculated as (Wallace [2])
( , )
( , Θ, Φ) = (2.1)
with
Fig. 2. Schematic model of sound radiation of a simply supported panel with a point force
excitation [2]
Three Noise Mitigation Measures for Steel Railway Bridges 581
( )
= ∗ 1+ (2.2)
"
In this paper, Equation 2.2 will be pursued to highlight and discuss the main
contributors influencing the radiated noise of a bridge for frequencies up to 200 Hz.
This is the main frequency range of radiated noise by the structure of steel bridges.
Changes of the radiation index σ are not discussed, as they will mainly shift resonances
to higher frequency ranges and not reduce the vibration energy and in consequence, in
most cases, the resulting overall noise reduction is poor.
= ∗ = (3.1)
It is obvious, that an increased mass will lead to lower acceleration a of the structure
by a constant excitation force F.
As the surface density in equation (2.2) is squared in the denominator, increasing
this mass could be seen as an effective measure for noise reduction of a bridge. With
increasing surface density, there is a shift of the coincidence frequency fg. If the
coincidence frequency fg, is lowered this will result in higher noise, caused by an
increase in the radiation index, σ of the plate. In Fig. 3 the increased radiation index, s,
is shown while Fig. 4 shows the resulting noise reduction (normalized to the excitation
force over frequency).
582 H. Venghaus
2
h=1.5cm
-4 h=1.5cm
h=1.6cm
0.1 -6 h=1.9cm
h=2.5cm
-8
-10
0.01 -12
10 16 25 40 63 100 160 250 400 10 100
Frequency [Hz] Frequency [Hz]
Fig. 3. Increase of the radiation index σ due Fig. 4. Reduction of the radiated noise due to the
to the increase of the surface density m” increase of the surface density m”
From Fig. 4 can be taken, that the increase of the surface density m” of a plate will
have noticeable effects only if the mass is increased in such a way, that the role of the
increasing radiation index can be neglected. From this it can be concluded, that the
increase of the surface density m” as a noise mitigation measure is only viable for a
replacement of the complete bridge. This procedure was undertaken by the Deutsche
Bahn (DB) by installing a new bridge with a base plate of 10 cm thickness. The
achieved noise reduction was about ΔL ≈ 10 dB(A) [4] compared to the replaced bridge
noise.
On ballast-less steel bridges, wooden sleepers are used very often, and these in
themselves can be seen as an initial measure of decoupling the structure from the
excitation forces. Highly resilient base plates (e.g. Cologne Egg) and resilient
under sleeper pads are being developed to provide further decoupling from the forces.
Fig. 5 depicts a highly resilient base plate, DELKOR ALT1, at the bridge across
the river Limmat in Switzerland, used to decouple the rail from sleeper. In Fig. 6 a
highly resilient base plate (normally used as under ballast mats) is shown (before the
sleeper is replaced); in these cases the rail and sleeper are decoupled from the bridge
structure.
Threee Noise Mitigation Measures for Steel Railway Bridges 583
Fig. 5. High resilient base platte DELKOR ALT1 Fig. 6. Highly resilient pads before sleeeper
is laid on
To calculate the efficiency in noise reduction of a highly resilient base plate at llow
frequencies, a first approacch is to simplify the resilient base plate as a simple sprring
with low stiffness and high damping.
d By this a noise reduction of the steel bridge cann be
calculated (Fig. 7). The ach hieved noise reduction – calculated and measured [5] - is
displayed. In Fig. 8 the calcu
ulated noise reduction of the under sleeper pad is presentted.
For these calculations the measured
m insertion loss of the under sleeper pad was taaken
into account. It can be seen that despite the simplicity of this model, the measured ddata
is predicted reasonably well for frequencies up to 125 Hz. Above this frequency raange
the noise radiation of wheeels and rails are dominating the radiated noise, which is not
included in these calculationns.
15 15
calculated
12 12 measured
noise reduction of the bridge [dB]
Noise reduction of the bridge [dB]
9 9
6 6
3 3
calculated
0 measured 0
-3 -3
10 100
0 10 100 1000
Frequency [Hz]] Frequenz [Hz]
Fig. 7. Noise reduction by use of a high Fig. 8. Noise reduction by an inserted unnder
resilient baseplate sleeper pad
sources become more impo ortant and therefore this model may be considered sufficient
for many applications due to o noise radiated by the structure of the bridge.
It is important to note thhat both mitigation measures will lead to an increase in the
radiated noise of the rail. Thhis is caused by the increased mobility of the rail due to the
reduced coupling of the raill with its supports. Rail dampers may be considered in thhese
applications to provide addiitional noise reduction. This can lead to an overall reducttion
of up to ΔL ≈ 10 dB(A) [6]..
12
l
loss factor bridge
welded - riveted bridge
diff. noise radition [dB]
welded
w bridge 9
riveted
r bridge
loss factor η
0.1
6
0.01 0
0 50 100 150 200 0 25 50 75 100 125 150
Frequency [Hz]] Frequency [Hz]
Fig. 9. Loss factors of weldeed and riveted Fig. 10. Calculated difference in radiated sound
bridges power (welded – riveted bridge)
The changes of the loss factors, η, are presented in Fig. 12 and Fig. 13 for two
different bridges with and without dampers. The same damper type was used for both
bridges. It can be seen, that both the original loss factor of the bridge and the
effectiveness of the dampers depend on the type of bridge.
loss factor η
factor factor
0.1 0.1
0.01 0.01
0 50 100 150 200 0 50 100 150 200
Frequency [Hz] Frequency [Hz]
Fig. 12. Loss factors of a riveted bridge and Fig. 13. Loss factors of a welded bridge and
with mounted dampers with mounted dampers
The effective loss factor of the damper is calculated by subtracting the loss factors of
the bridge with dampers from that of the bridge without dampers at each frequency.
There is good general agreement between the measured and predicted noise
reductions of the two bridges types (Fig. 14 and Fig. 15). Although the damping effect
at most frequencies is somewhat lower than predicted. The riveted bridge bridges are
less noisy than welded bridges due to their higher internal structural losses. To obtain a
noticeable noise reduction using dampers, the increase of the loss factor must be higher
than the loss factor of the structure of the bridge itself.
5 riveted bridge 5
calculated
4 4
Noise reduction of the bridge [dB]
measured
3 3
2 2
welded bridge
1 1 calculated
measured
0 0
-1 -1
0 25 50 75 100 125 150 0 25 50 75 100 125 150
Frequenz [Hz] Frequenz [Hz]
Fig. 14. Calculated and measured noise Fig. 15. Calculated and measured noise
reduction of a riveted steel bridge reduction of a welded steel bridge
586 H. Venghaus
4 Conclusions
An easy-to-use equation for calculating the radiated sound power of a vibrating surface
of finite length given by Cremer and Heckl leads to three potential ways of noise
mitigation measures for ballast-less steel bridges:
By calculating the mitigation measures
the applicability of this equation at low frequencies was demonstrated comparing the
results to measured noise reductions. As the results showed a good agreement it can be
said that this equation gives a useful estimation for which type of mitigation measure
might be recommended for a particular steel bridge under investigation. Detailed
information about the loss factors of the steel bridge has to be acquired to increase the
quality of the results. This can be done by measurements at existing bridges or by FEA
calculations for new bridges.
References
[1] Cremer, L., Heckl, M.: Körperschall. Springer, Heidelberg (1967)
[2] Wallace, C.E.: Radiation resistance of a rectangular panel. Journal of the Acoust. Soc. of
Am. 51, 946–952 (1972)
[3] Kuik, S.S., et al.: Tuned vibration absorbers for control of noise radiated by a panel. In:
Proceedings of ACOUSTICS 2009, Adelaide, Australia, November 23-25 (2009)
[4] DB Systemtechnik GmbH, internal information not published
[5] Köstli, K.P., Jones, C.J.C., Thompson, D.J.: Experimental and theoretical analysis of
railway bridge noise reduction using resilient rail fasteners in Burgdorf, Switzerland. In:
Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson, J.,
Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 208–214. Springer, Heidelberg (2008)
[6] Poisson, F.: Performances des solutions de reduction du bruit des ponts metalliques, Unité
Physique du Système Ferroviaire, Direction de l’Innovation et de la Recherche, 45 Rue de
Londres, 75379 Paris cedex 08 (July 2006)
[7] Deutsche Bahn: Innovative Maßnahmen zum Lärm- und Erschütterungsschutz am Fahrweg
Schlussbericht Berlin (June 15, 2012)
The Mechanisms of Curve Squeal
1 Introduction
2 Conventional Theory
Curve squeal is a high level, strongly tonal noise occurring in sharp curves. The noise
level is usually 10~20 dB higher than the rolling noise. When measured near the
wheel, the noise level can reach 130 dBA. The dominant single frequency of squeal
often occurs in the frequency range 1000 to 5000 Hz, the range at which human
hearing is most sensitive. Squeal occurs more frequently when the curve radius is
smaller than 100 times the bogie wheelbase, or equivalently the wheel angle of attack
exceeds 10 mrad [2].
Current theory points to lateral creepage of the wheel as the prime cause of squeal,
particularly for the leading inner wheel of a bogie [1]. The effect on squeal of
longitudinal creepage and flange contact are generally discounted as excitation
mechanisms. Remington [3] concluded from laboratory experiments that flange contact
does not produce squeal. Reported observations indicate that squeal is usually
generated by the leading inner wheel [1].
Due to the rigid design of a railway bogie, one axle of the bogie (usually the leading
one) cannot align itself radially to the curve centre; as a result the wheels have to slide
laterally while rolling in the longitudinal direction. When the lateral creepage exceeds
the critical value around 10 mrad, gross wheel lateral sliding occurs. This becomes
unstable if the friction force reduces at high slip values, according to the classic
stick-slip induced instability mechanism.
The schematic diagram as shown in Fig.1 (a) is usually used to illustrate the
‘stick-slip’ or negative damping mechanism for squeal. A solid mass sitting on a
moving belt experiences a transient motion, and stops eventually if the interface friction
coefficient remains constant or does not drop with increasing sliding speed. With the
introduction of a dynamic friction coefficient decreasing with increasing sliding speed,
as the so called 'negative damping', it may become unstable. This is the well-known
‘stick-slip’ or negative damping mechanism.
Fig. 1. A comparison of the single degree-of-freedom model, which is usually used to illustrate
the negative damping mechanism (a), and a two degree-of-freedom model for the mode coupling
mechanism (b)
The negative damping mechanism has been the primary mechanism of curve squeal
since the first prediction model was proposed by Rudd in 1970s [1-4]. However, in the
very closely related automotive brake squeal community, the mode coupling instability
mechanism is gaining increasing support while the negative damping mechanism of
brake squeal has not received much attention in recent years [5].
The Mechanisms of Curve Squeal 589
The simple model in Fig. 1(b) is used by Hoffmann [6] to illustrate the mechanism of
the mode coupling instability. The mass sitting on a moving belt moves in both vertical
and lateral directions. In addition to the structural coupling, friction also plays a
cross-coupling role and it alters both the vertical and lateral oscillation frequencies.
When the friction coefficient increases to a critical threshold, the vertical and lateral
oscillations become out-of-phase and the friction force, which depends on the vertical
displacement, is therefore out-of-phase with the lateral motion. This means the friction
force no longer dissipates energy, but instead feeds energy into the vibrating system
and the system becomes unstable.
If this mechanism is applied to curve squeal it suggests that the likelihood of squeal
increases with increasing wheel/rail interface friction coefficient and the presence of
closely spaced lateral and vertical wheel vibration modes. This implies that squeal can
be controlled by reducing the friction coefficient or by modifying wheel lateral and
vertical modes so that they are sufficiently far apart in the frequency domain, or out of
the sensitive frequency region.
A key difference between the negative damping and the mode coupling instability
mechanism is that the former only considers the dynamics in the tangential direction
and requires a negative damping friction characteristic to generate instability; whereas
the latter takes into account both the tangential and normal dynamics, and relaxes the
requirement of a negative damping friction characteristic.
The mode coupling mechanism and the role of the vertical dynamics in squeal
generation have been recognized in some recent models [7-9], but there are conflicting
opinions on whether a negative damping friction characteristic is critical.
4 Discussion
The conventional theory of curve squeal gives explanations of some observations, but
it appears to be deficient in explaining others, which are discussed in the following
sections.
Fig. 2 shows a typical example of measured rail accelerations when a squealing wheel
passes [10]. The rail vibrations were measured simultaneously from a tri-axial
accelerometer mounted onto the web of the outer rail. The rail accelerations were very
high (around 500 g) in both vertical and lateral directions when the squealing wheel
passed. Both vibrations exhibit amplitude modulation.
When zoomed in, it can be seen that the lateral and vertical vibrations changed from
out-of-phase to in-phase over approximately 100 ms. The corresponding spectra of rail
vibrations show slightly different peaks (see Fig.3).
590 J. Jiang, D.C. Anderson, and R. Dwight
Vertical Lateral
500
-500
-1000
-1500
45 45.1 45.2 45.3 45.4 45.5 45.6 45.7 45.8 45.9
500
Acceleration (g)
-500
-1000
-1500
45.5 45.51 45.52 45.53 45.54 45.55 45.56 45.57 45.58 45.59
400
200
0
-200
-400
45.55 45.551 45.552 45.553 45.554 45.555 45.556 45.557 45.558 45.559
Time (s)
Fig. 2. Measured rail vertical and lateral accelerations when a squealing outer wheel passes [10]
Vertical Lateral
100
50
Acceleration (g)
2 3 4
10 10 10
100
50
The rail is expected to vibrate at the same frequency as the wheel due to the
wheel/rail interaction so the coupled rail vibrations reflect the coupled wheel vibrations
in either direction. The different rail vibration frequencies in two directions may
indicate that two wheel modes around 2150 Hz, possibly one vertical and one lateral,
were involved with squeal generation.
The Mechanisms of Curve Squeal 591
The coupled rail vibration or beating oscillation, which reflects the coupled wheel
vibration, can be readily explained by the mode coupling mechanism, but are not
available from the negative damping mechanism. Closely spaced radial and lateral
wheel modes are common from a railway wheel [1]. According to Hoffmann [11], one
pair of the closely spaced wheel modes may lead to beating oscillations even when the
friction is below the critical value.
Fig. 4. Friction curve measured from a test rig with and without water (reproduced from [12])
N.B. There is no squealing with water, whereas there is squealing without water.
Visual inspection indicated that the outer rail gauge corner/face was nearly dry at
Site 1 when the unexpectedly high percentage of squeal was detected at the outer rail;
whereas the very low percentage (3 %) at Site 4 was associated with good lubrication of
the outer rail. This highlighted that gauge face lubrication may play a more significant
role in squeal generation than previously recognised and triggered further field trials as
reported in [16].
The results [16] confirm that the lubrication condition of the outer rail has a
significant impact on curve squeal. The overall number and severity of the curve squeal
was greatly reduced when the outer rail was lubricated and the proportion of events
arising from the outer rail changed from 90 % to 30 %.
These findings contradict the general view that flange contact helps to reduce the
likelihood of squeal due to lateral slip. Due to the contact angle between the wheel
flange and rail gauge or corner, the creep of the outer wheel along the rail is not limited
to the lateral direction, as it is for the inner wheel. The wheel lateral and vertical
vibration are coupled naturally, which may result in squeal under the right conditions,
such as sufficient friction coefficient and closely spaced wheel modes as suggested by
the mode coupling mechanism.
Why squeal occurs at one frequency but not others is not predicted precisely by an
existing model. However, a future model based on the modal coupling mechanism
appears promising.
70 50%
45%
Number of squeal events
60 36.64% 40%
50 30.53% 35%
Percentage
40 30%
25%
30 20%
12.98%
20 15%
5.34% 5.34% 6.11% 10%
10 3.05%
5%
0 0%
<1 1-2 2-3 3-4 4-5 5-6 >6
Frequency (x1000Hz)
5 Conclusions
Field observations such as the generation of squeal from the outer wheel (including
wheel flange contact), squeal occurring at various wheel natural frequencies, coupled
rail vibrations when squealing wheels pass, and the obvious influence of trackform on
squeal occurrence and severity, are better explained by the mode coupling mechanism
than the traditional negative damping mechanism.
References
[1] Thompson, D.: Railway noise and vibration: Mechanisms, modelling and means of control,
p. 168. Elsevier (2009)
[2] Rudd, M.: Wheel/rail noise—part 2: wheel squeal. Journal of Sound and Vibration 46,
381–394 (1976)
[3] Remington, P.J.: Wheel/rail squeal and impact noise: what do we know? What don’t we
know? Where do we go from here? Journal of Sound and Vibration 116(2), 339–353
(1985)
594 J. Jiang, D.C. Anderson, and R. Dwight
[4] Vincent, N., et al.: Curve squeal of urban rolling stock—Part 1: State of the art and field
measurements. J. Sound and Vibration 293, 691–700 (2006)
[5] Kinkaid, N.M., et al.: Review: Automotive disc brake squeal. Journal of Sound and
Vibration 267, 105–166 (2003)
[6] Hoffmann, N., et al.: A minimal model for studying properties of the mode-coupling type
instability in friction induced oscillations. Mechanics Research Communications 29,
197–205 (2002)
[7] de Beer, F.G., et al.: Squeal noise of rail-bound vehicles influenced by lateral contact
position. Journal of Sound and Vibration 267(3), 497–507 (2003)
[8] Chiello, O., et al.: Curve squeal of urban rolling stock—Part 3: Theoretical model. Journal
of Sound and Vibration 293, 710–727 (2006)
[9] Glocker, C., et al.: Curve squealing of trains: Measurement, modelling and simulation.
Journal of Sound and Vibration 324, 365–386 (2009)
[10] Jiang, J., Dwight, R.: CRC Rail Innovation Project R1-105 Report A2, Consolidation of
curving noise theory through field testing field verification of curving noise mechanisms
(March 2011)
[11] Hoffmann, N., Gaul, L.: Non-conservative beating in sliding friction affected systems:
Transient amplification of vibrational energy and a technique to determine optimal initial
conditions. Mech. Syst. Signal Process 18, 611–623 (2004)
[12] Anderson, D., et al.: Mitigation of wheel squeal and flanging noise on the Australian rail
network. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth,
B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Trans
Systems. NNFM, vol. 99, pp. 399–405. Springer, Heidelberg (2008)
[13] Koch, J.-R., et al.: Curve squeal of urban rolling stock—Part 2: Parametric study on a 1/4
scale test rig. J. Sound and Vibration 293, 701–709 (2006)
[14] Jiang, J., Dwight, R.: CRC Rail Noise Project R1-128 Report of Activity 1.2,
Consolidation of curving noise theory through field tests (June 2012)
[15] Jiang, J., Dwight, R., Anderson, D.: Field verification of curving noise mechanisms. In:
Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-Werning,
B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 118, pp. 349–356. Springer, Heidelberg (2012)
[16] Curley, D., Anderson, D.C., Jiang, J., Hanson, D.: Field trials of gauge face lubrication and
top-of-rail friction modification for curve noise mitigation. In: Nielsen, J.C.O., Anderson,
D., Gautier, P.-E., Iida, M., Nelson, J.T., Thompson, D., Tielkes, T., Towers, D.A., de Vos,
P. (eds.) Noise and Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 126,
pp. 449–456. Springer, Heidelberg (2015)
[17] Jiang, J., Ying, I., Hanson, D., Anderson, D.: An investigation of the influence of track
dynamics on curve noise. In: Nielsen, J.C.O., Anderson, D., Gautier, P.-E., Iida, M.,
Nelson, J.T., Thompson, D., Tielkes, T., Towers, D.A., de Vos, P. (eds.) Noise and
Vibration Mitigation for Rail Transportation Systems. NNFM, vol. 126, pp. 441–448.
Springer, Heidelberg (2015)
Proposals for Improved Measurement Methods
for Curve Squeal and Braking Noise
Summary. Improved methods for measuring curve squeal and braking noise
have been proposed within the ACOUTRAIN project, intended as input to future
standards. In this paper an outline of these methods is given and the underlying
considerations are explained. The EU TSI regulation limiting noise emission of
new rail vehicles does not yet specify limits for curve squeal or braking noise,
which are both relevant noise sources but not easy to characterise reliably due to
their variability. No ISO or EN standard exists for curve squeal. Braking noise is
included in the current EN ISO 3095 noise type testing standard, but the
procedure requires improvement in relation to measured quantities. For curve
squeal, two methods are proposed, an occurrence test using on board
measurements along a route with various curves, and a trackside test for curves or
switches. The existing UIC trackside test is taken as basis using the percentage
exceedance indicator L10.
For braking noise, improvements to the existing trackside measurement
standard are proposed, taking the variety of potential noise sources and their
variable position and operating condition into account. The measurement
quantity, positioning and braking operation are the main topics. Without
proposing limit values, some considerations are put forward that may be taken
into account when applying the methods for compliance testing for curve squeal
and braking noise.
1 Introduction
Improved methods for measuring curve squeal and braking noise of railway vehicles
have been proposed within the European project ACOUTRAIN [1], which is
developing methods for virtual type testing, replacing running tests by simulation or
equivalence procedures. These methods which are intended as input to future standards
are outlined in this paper and the underlying considerations are explained.
The European TSI regulation limiting noise emission of new rail vehicles does not
yet specify limits for curve squeal or braking noise, which are both relevant noise
sources but not easy to characterise reliably due to their variability. No ISO or EN
standard exists for curve squeal. Braking noise is included in the current EN ISO 3095
noise type testing standard, but the procedure requires additional measured quantities to
address the statistical nature of these sources.
2 Curve Squeal
2.1 Characteristics
Curve squeal is known as the tonal noise that occurs during curving either at curves or
in points. It depends on many parameters both on vehicle and track [2][3], but also
environmental parameters such as moisture, wheel and rail temperature, and friction
coefficient due to lubricants and dust. Track parameters include the curve radius, track
cant and gauge, railhead profile and track dynamics. Vehicle parameters include speed,
wheel base, surface profile, wheel and bogie dynamics and wheel damping. Usually the
wheel is the main sound radiator emitting frequencies from around 250 Hz upwards,
depending on the wheel type, and corresponding to particular wheel vibration modes.
Curve squeal is an instability phenomenon which is rather complex to predict due to the
many influence parameters and mechanisms involved. The most predominant type of
curve squeal is due to lateral excitation caused by slip-stick in the contact patch. Also
lower frequency curving noise can occur due to juddering of the wheel-rail contact, or
higher frequency noise can occur due to flange contact, sometimes sounding like
hissing. Both of these phenomena tend to produce lower noise levels than lateral
friction excitation.
The variability of curve squeal makes it difficult to assess as it can differ from one
curve to another and for different environmental conditions. A measurement method
was previously developed by TNO for UIC in 2004 [4] which takes this behaviour into
account. Its main purpose was to be able to assess the effect of noise control measures
for reduction of curve squeal using statistical indicators quantifying both the level and
the duration of tonal noise.
The new method proposed for curve squeal is intended to assess rolling stock and
consists of two parts, an occurrence test using on board measurements while passing a
series of curves, and a trackside measurement at a curve where squeal occurs. The on
board test can be combined with dynamic tests required by the TSI specified in UIC
Proposals for Improved Measurement Methods for Curve Squeal and Braking Noise 597
leaflet 518 [5] which already includes a range of curves, although it will require some
sharper curves (if necessary switches).
2.3.2 Processing
The following quantities are determined for each curve and measurement position:
- LpAF(t) and LpAF(t,f) over the curve section and over the reference section;
- LpAFmax, LpAeq,T, L10 for the whole trajectory of the microphone along the curve;
- ΔLpAeq, ΔL10, ΔL10(f) for the difference between the whole curve trajectory and the
reference section.
598 M.G. Dittrich and H.W. Jansen
- The histogram and cumulative histogram of the overall A-weighted level and third
octave bands is determined from the LpAF(t,f) for each curve in 1 dB intervals (for
presentation example see Fig 1). The overall L10, L50 and L80 A-weighted levels are
determined from the cumulative histogram.
LN is the percentage exceedance level, the (sound pressure) level that is exceeded for
N% of the measurement time. LN(f) is the same for the one third octave band with
centre frequency f. ΔLN(f) is the difference between LN(f) on the curve and on the
reference track section.
The processing for each microphone position begins when it enters the start of
the curve and stops when it passes the end of the curve. The start and stop of the
processing time may be identified from the lateral acceleration signal, GPS or trackside
markers.
Occurrence of curve squeal is confirmed if for any on board microphone, for any of
the curves and for any third octave band centre frequency fi, the difference between the
curve and straight track in any frequency band is more than 5 dB,
The benefit of using the L10 percentage exceedance indicator is that it takes both
duration and level of a noise event into account. The L10 neglects short bursts of noise
and gives comparable results to the equivalent noise level for continuous constant
noise.
The above criterion could be further refined in terms of frequency range or
distinction between flange noise and curve squeal.
A m plitud e [P a ] 1 20
% o f tim e
0 10
-1 0
0 5 10 70 75 80 85 90
Time [s] LpA [dB(A)]
LpAFmax=89 dB(A) L10=86 dB(A); Dominant L10=82 dB(A) in 5000Hz band
C um m . % of tim e
100 100
L p A (t) dB (A )
Overall
90 dominant band
50
80
70 0
0 5 10 40 60 80 100
Time in [s] Δ t =0.125[s] LpA [dB(A)]
A-weighted third octave spectrogram LpAeq,T= 83 LpAFmax= 89 L10= 86 L50= 82 L80= 80 dB(A)
Frequency [H z]
8000 80
80 LpAeq,T
L p A d B (A )
60 LpAFmax
60
40 L10
40
31.5 20 L
0 5 10 63 125 250 500 1k 2k 4k 8k 16k 50
Time [s] Frequency [Hz] L80
Fig. 1. Measurement results for intercity coaches passing a set of points at approx. 30 km/h, with
varying curve squeal measured at the trackside. Top left: sound pressure time signal; middle left:
A-weighted level history; bottom left: third octave A-weighted spectrogram; top right: histogram
of A-weighted level; middle right: cumulative histogram of overall level and level in dominant
frequency band, with L10, L50 and L80 lines indicated; bottom right: A-weighted spectra for LAeq,
LpAFmax, L10, L50 and L80.
The histogram and cumulative histogram of the overall A-weighted level (see Fig. 1)
and of the unweighted third octave bands are determined from the Lmax(t,f) for each
pass-by in dB intervals. The overall L10, L50 and L80 levels are determined from the
cumulative histogram. Each level is arithmetically averaged over the pass-bys. Both the
running test and the trackside test have been evaluated by Talgo and reported in the
Acoutrain project [8].
3 Braking Noise
3.1 Characteristics
Braking noise potentially covers a wide range of conditions and noise sources, for
example braking at speed, braking to standstill, tonal noise from squeal, broadband
noise from friction, noise from pneumatic components and from regenerative braking.
Methods for braking noise to standstill are included in the EN ISO 3095: 2013 standard
600 M.G. Dittrich and H.W. Jansen
[5]. Some of the highest braking noise levels are often due to brake squeal occurring
while braking to standstill, causing disturbance around stations, stopping locations and
railway yards. Similarly to curve squeal, brake squeal is strongly dependent on several
parameters including brake type, wheel design, wear, friction conditions, brake forces
and speed, and can be stochastic in nature. In particular, the position of the noise source
on a braking train can be hard to identify in relation to the trackside measurement point.
This leads to a requirement for more measurement positions along the track or
alternatively, larger measurement distance. A similar difficulty occurs for measuring
acceleration noise.
The current test method for braking noise in EN ISO 3095:2013 [5] specifies braking
from 30 km/h to standstill. The measurement positions are located on either side of the
train at 7.5 m distance from the centre of the track and 1.2 m height above the rail
surface, at the front measurement cross section. This is defined as being opposite the
centre of the first car of the unit when it stops. Due to the difficulty in predicting the
precise braking distance, a tolerance of ±10 m is allowed. The current measured
quantity is LpAF(t), from which LpAFmax is determined by averaging at each microphone
position and taking the maximum of these levels. So as to take into account the
variability of brake squeal in time, position and frequency content, it is proposed to add
the L10(f) and overall A-weighted L10 indicator in analogy to the curve squeal pass-by
method, but only applied to the braking duration. A braking noise example using the
same processing applied for curve squeal is shown in Fig. 2. In contrast to the example
in Fig. 1, the squeal level is higher and more continuous. In this case all the indicators
demonstrate the presence of strong tonal noise.
As the spread in brake squeal noise can be larger than other braking noise sources, it
is recommended to take the coverage of microphone positions into account. Spread in
results may be reduced by allowing an alternative or provisional 25 m microphone
position, or adding one or more extra positions at 7.5 m distance, ensuring a level
difference between the positions of no greater than 2 or 3 dB. For microphones at a
distance of d from the track centerline, assuming monopole point sources, the
maximum separation distance a along the track is 2d to ensure differences between
measured levels of less than 3 dB. To obtain a difference of less than 2 dB, microphone
separation a should be no greater than d.
So for full coverage of braking noise sources on a vehicle up to 45 m long, 3
microphones placed 15 m apart would be sufficient for a maximum spread in the result
of 3 dB. At 25 m from the track, a single microphone is sufficient to cover this same
length, under the same assumptions (monopole point sources).
For a better definition of the braking operation, the measured train deceleration
might be specified in a given range. Fairly continuous brake application is required to
ensure an uninterrupted braking noise level, but without full or sudden braking. The
chance of detecting the maximum braking noise level at a single microphone position
could be increased by slightly varying the starting point of brake application. Another
way of achieving this is to allow braking in both directions, which in some situations
may also be time saving.
Proposals for Improved Measurement Methods for Curve Squeal and Braking Noise 601
5 20
A m plitu de [P a ]
% of tim e
0 10
-5 0
0 5 10 15 40 45 50 55 60 65 70 75 80 85 90 95100
Time [s] LpA [dB(A)]
LpAFmax=95 dB(A) L10=91 dB(A); Dominant L10=87 dB(A) in 4000Hz band
C um m . % of tim e
100
Overall
L p A (t) dB (A )
100
dominant band
50
80
0
0 5 10 15 40 60 80 100
Time in [s] Δ t =0.125[s] LpA [dB(A)]
A-weighted third octave spectrogram LpAeq,T= 87 LpAFmax= 95 L10= 91 L50= 85 L80= 72 dB(A)
80
F requency [H z]
8000
80 LpAeq,T
L p A dB (A )
60 LpAFmax
60
40 L10
40
31.5 20 L
0 5 10 15 32 63 125250500 1k 2k 4k 8k 16k 50
Frequency [Hz] L80
Time [s]
Fig. 2. Trackside measurement results for a freight train braking from 30 km/h down to standstill,
with continuous loud squeal. Figure legend same as in Fig 1.
Any limit should be set appropriately for the distance of 25 m and for 7.5 m. Given the
potential uncertainty in measurement results, a margin of at least 5 dB should be taken
into account when considering comparison with a limit.
5 Conclusions
A new method for assessing curve squeal for railway vehicles has been proposed,
consisting of an on board occurrence test and a trackside test. A test route including a
series of curves is required, which can be combined with dynamic vehicle running tests.
602 M.G. Dittrich and H.W. Jansen
For occurrence testing, a comparison is made between a straight reference track and a
curve, with the L10 percentage exceedance used as indicator. The method has been
evaluated in field tests in the Acoutrain project by Talgo [8].
For braking noise, improvements to the existing trackside test are proposed, mainly
in the distribution of microphones to be used, the use of the L10 indicator and smooth
application of the braking regime.
The L10 indicator is more suitable than LpAFmax which quantifies instantaneous levels,
or the equivalent sound pressure level LpAeq.tp, which in some cases conceals the
presence of squeal.
After further evaluation these methods may be used for input to measurement
standards. As virtual testing (simulation for type testing) of curve and brake squeal are
not yet considered feasible, the test methods described here are currently considered
most appropriate.
References
[1] Dittrich, M.G., Jansen, H.W.: Braking noise and curving noise: corresponding indicators
and measurement procedures (interim), Acoutrain report no. ACT-T1_2-D-TNO-001-04,
TNO (April 11, 2013)
[2] de Beer, F.G., Janssens, M.H.A., Kooijman, P.P.: Squeal noise of rail-bound vehicles
influenced by lateral contact position. Journal of Sound and Vibration 267, 497–507 (2003)
[3] Thompson, D.: Railway noise and vibration. Elsevier (2009)
[4] Jansen, H.W., Janssens, M.H.A.: Squeal noise measurement protocol, TNO report
DGT-RPT-040028, Delft (July 9, 2004)
[5] UIC Leaflet 518: Testing and approval of railway vehicles from the point of view of their
dynamic behaviour - Safety – Track fatigue - Running behavior, UIC (October 2009)
[6] EN ISO 3095:2013: Acoustics - Railway applications - Measurement of noise emitted by
railbound vehicles
[7] Jansen, H.W., Dittrich, M.G., Sikma, E.: Brake noise measurements on mixed Freight trains
with composite brake blocks. In: Proceedings of Acoustics 2008, Paris (2008)
[8] Martín López, B., Gómez, J.C.: Curve squeal noise measurements, Acoutrain report no.
ACT–WP1-D–TALGO-003-03, Talgo (October 2013)
Curve Squeal in the Presence of Two
Wheel/Rail Contact Points
Summary. A real case of a city tramcar generating very high curve squeal noise
levels has motivated research in this area considering the simultaneous presence
of two contact points. In this paper, available measurements are summarised in a
qualitative manner in order to highlight the most important frequencies involved
and a theoretical model in the frequency domain is developed with the aim of
predicting curve squeal tones. Good matching is found between numerically
predicted and measured unstable frequencies and a peculiar shift toward higher
frequencies is found both in measurements and predictions.
1 Introduction
Curve squeal is one of the most serious noise problems connected with the operation of
trains, tramways and metros in urban environments. The phenomenon has been
analysed by many authors, by means of both experimental tests and theoretical models,
and it is agreed that it is produced by the self-excited vibration of the wheel, as a result
of the contact phenomena taking place at wheel-rail interface during curve negotiation.
Although some research has been carried out in order to point out the influence of
the different contact conditions at the inner and outer wheel of a leading axle on the
occurrence of squeal noise, no specific investigation for urban tramways with grooved
rails is available. This paper describes research into squeal noise measurements and
modelling for a case of a city tramway with independent resilient wheels on very sharp
curves [1]. Very high squeal noise levels were recorded and both the flange back of the
inner wheel and the flange of the outer wheel were believed to be in contact with the
head and grooved-head due to the highly worn condition of the rail.
Since Rudd’s paper [2], there has been a general understanding that flange contact
does not represent an important source of noise in curves and in [3] it was even shown
that squeal noise was not occurring with the contact point located on the flange while it
appeared as soon as the wheel was slightly moved to have the contact point on the tread.
However, it is known that squeal noise can be also generated by flange contact (see
e.g. [4]) involving predominantly radial modes of the wheel. Indeed, measurements
described in [1] represent exactly this situation and give an important motivation for
investigating the effect that two contact points have on squeal noise.
To the authors’ knowledge there are no models available in the literature to address
curve squeal in the presence of two-point contact; therefore Section 3 is devoted to
describing how the model developed by Huang [5] has been extended and updated to
consider the simultaneous presence of two contact points. It is known that the curve
squeal phenomenon is far from being deterministic and parameters like temperature,
humidity and even dirt and particles on the rail can have a strong effect on its
occurrence. For this reason, in the model presented here, some input parameters are
defined as uncertain variables. As a result, the model is used to give the frequency
values of possible unstable eigenvalues of the system while the range of uncertain
parameters is randomly spanned. The definition of the nominal kinematic condition at
the contact (position of the contact points and creepages) has been determined by
means of a vehicle dynamics numerical model described in [6]. The model developed is
of general purpose and can be used for any wheel-rail combination; as a first
application results have been verified against measurements of [1] showing a good
reliability in predicting the frequencies involved in curve squeal.
2 Measurements
-3
10 a
Mobility, m/s/N
-4
2090 Hz
10 3370 Hz 3740 Hz
-5
10
-6
10
-7
535 Hz 1270 Hz 2480 Hz
10-3
10 b
Mobility, m/s/N
-4
10
-5
10
-6
10
-7 1420 Hz 2230 Hz
10
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency, Hz
Fig. 1. Wheel mobilities at nominal contact point. a: axial direction. b: radial direction. − − −:
measurements; ⎯⎯: FE model.
To summarise the wheel behaviour in terms of squeal noise the inner wheel of the
leading axle is taken as example; in this case the flange back is in contact with the
grooved head. Fig. 2 shows the spectrogram of the wheel vibration velocity in radial
and axial directions during the passage through the curve. Although the darker
corridors in the figure are relatively wide it is still possible to identify the following
as the frequencies mainly involved in squeal: 530 Hz, 1330 Hz, 1550 Hz, 2100 Hz,
2550 Hz, 3600 Hz and 3730 Hz. The contribution at 3100 Hz is likely to be the
second harmonic of 1550 Hz which appears due to the non-linearity of the
phenomenon.
According to curve squeal theory each tone recorded during the vehicle passage is
expected to be related to a wheel mode. In this case this is not always straightforward.
In fact, it is peculiar to observe that the modes at 1270 Hz and 1420 Hz seem to generate
squeal at higher frequencies (1330 Hz and 1550 Hz) and the mode at 2480 Hz has a
similar behaviour.
3 Theoretical Model
The model consists of a linear description of the wheel and rail dynamics while the
non-linear representation of friction coefficients as a function of creepages
presented by Kraft [7] is adopted to calculate the contact forces. A stability analysis
of the open loop transfer function of the overall system yields the possible unstable
frequencies.
606 G. Squicciarini et al.
25 a
Time, s
20
15
25 b
Time, s
20
15
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency, Hz
Fig. 2. Spectrogram of inner wheel vibration velocity during tram pass-by. a: axial; b: radial.
Darker regions correspond to higher levels.
Fig. 3a shows the general structure of the model. The wheel is represented by means
of its modal basis to give the mobilities at the nominal contact point. By knowing the
steady state position of the contact points along the curve, direct and transfer mobilities
W
at these locations are calculated by means of rigid translations and rotations. Yij ,kl in
the figure represents a single element of the wheel mobility matrix, subscripts i and j
represent the degrees of freedom (d.o.f.) at the contact while k and l represent the
contact points (see Fig. 3b). The d.o.f. considered in the model are longitudinal, lateral,
normal to contact plane and spin (rotation around the normal); it is believed that the
other two rotations can be neglected.
For the track, different analytical models are used depending on the d.o.f.
considered. For the vertical direction a Timoshenko beam on a double elastic layer is
adopted while a multi-beam model described in [8] allows lateral vibrations to be
coupled with torsion. Simple formulae are implemented to estimate longitudinal and
spin mobilities. Again, track mobilities are transferred to the contact points by means of
R
rigid rotations and translations. Yij ,kl in the figure represents one element of the rail
mobility matrix.
The contact stiffness is considered in the vertical direction for the wheel-rail contact
C
mobility matrix ( Yij ,kl ) and is calculated according to Hertz theory.
Complex dynamic sliding velocities Vs and complex dynamic contact forces F at
the contact points are finally related through mobility matrices as follows:
⎧⎪V s ⎫⎪ ⎡ Y Y12 ⎤ ⎧ F1 ⎫
V s = ⎨ 1s ⎬ = ⎢ 11 ⎨ ⎬ = YF
Y22 ⎥⎦ ⎩F2 ⎭
(3.1)
⎪⎩V2 ⎪⎭ ⎣Y12
where each element of the matrix Y is calculated by adding wheel, rail and contact
mobilities and the subscripts in Eq. (3.1) indicate the two contact points.
Curve Squeal in the Presence of Two Wheel/Rail Contact Points 607
a b
Fig. 3. a: general structure of the model. b: reference frames at contact points 1 and 2 (0 is
nominal contact point).
After algebraic manipulation taking into account that the vertical sliding velocity is
zero, the following quantities can be defined:
dTn = {Y13,1n Y23,1n Y63,1n Y13, 2 n Y23, 2 n Y63, 2 n } n = 1,2 (3.2)
Y33,11Y33, 22 ⎛ T Y33,12 T ⎞
bTn = ⎜ − an + a 3 − n ⎟ n = 1,2 (3.4)
Y33,11Y33, 22 − Y332 ,12 ⎜ Y33, nn ⎟
⎝ ⎠
and Eq. (3.1) rewritten as
(
V s = C + d1b1T + d 2bT2 F = EF ) (3.5)
The matrix C in Eq. (3.5) is formed from the matrix Y by removing the rows and
columns related to the vertical direction; vectors d and b are defined in Eqs. (3.2)-(3.4).
E is one of the key elements of the self-excited loop represented in Fig. 3a.
To describe the relationship between friction coefficients and creepages the
FASTSIM algorithm is adopted (see [9]). To introduce the falling region, a heuristic
correction formula (see [5, 7]) of the form:
(
τ k = 1 − λk e −κ k γ tot
) (3.6)
is assumed; λk and κk are empirically determined for each contact point k and γtot
combines the contribution of all creepages.
At a contact point k the linearized relation between contact forces and creepages can
be expressed, independently of the other contact point, as:
Fk = K k Vk + H k F3,k (3.7)
In Eq. (3.8) N0,k is the normal force at point k, μi,k is the friction coefficient in
direction i at contact point k, γi,k is the creepage in direction i at contact point k and f3,k is
the dynamic force in the direction normal to the contact plane at point k.
Combining Eqs. (3.7) and (3.4) with (3.5) the equation for the self-excited loop of
Fig. 3a results as
⎧⎪b T ⎫⎪ ~
F = KEF + H ⎨ 1T ⎬F = KEF + HF (3.9)
⎪⎩b 2 ⎪⎭
In order to find the possible unstable frequencies, the eigenvalues of the open loop
transfer function are computed and those having an imaginary part equal to zero and a
real part greater than unity are classified as possible sources of instability.
4 Results
The theoretical model described in Section 3 has been verified against measurements
summarised in Section 2. In order to obtain reliable values for the vehicle dynamic
parameters a multi-body model of the tram has been developed by means of the
software described in [6]. Steady state creepages and contact point positions have been
obtained. Where necessary, variability is added to these parameters. The most
important uncertain and deterministic input parameters are summarised in Table 1.
Table 1. Input parameters at contact points 1 and 2 for inner and outer wheels. U indicates a
uniform distribution between the stated limits.
Values
Name Unit Description
Inner Outer
N0,1 kN Normal force 1 U(17,25) U(14.5,21.7)
N0,2 kN Normal force 2 U(0,6.5) U(0,12)
V0 m/s Speed U(1.9,3.6) U(1.9,3.6)
κ1 - Eq. (3.6) 1 U(0.04,0.06) U(0.04,0.06)
κ2 - Eq. (3.6) 2 U(0.12,0.17) U(0.12,0.17)
λ1 - Eq. (3.6) 1 0.8 0.8
λ2 - Eq. (3.6) 2 0.8 0.8
θ1 ° Contact angle 1 U(2.4,3.4) U(-3.4,-2.4)
θ2 ° Contact angle 2 U(-73,-63) U(-76,-66)
γx1 - Creepage x 1 U(0.0038,0.0057) U(0.0067,0.0101)
γy1 - Creepage y 1 U(-0.0651,-0.0434) U(-0.0651,-0.0434)
γsz1 1/m Spin-creep z 1 U(0.12,0.18) U(0.12,0.18)
γx2 - Creepage x 2 U(-0.0310,-0.0207) U(-0.0193,-0.0128)
γy2 - Creepage y 2 U(-0.1587,-0.1058) U(-0.2000,-0.1334)
γsz2 1/m Spin-creep z 2 U(2.2560,3.3840) U(2.2960,3.4440)
P1 mm Contact 1 [0 U(-9.5,0.5) -0.22] [0 U(-9,1) 0.2]
P2 mm Contact 2 [50.1 40.2 10.4] [50.6 -20 8]
Curve Squeal in the Presence of Two Wheel/Rail Contact Points 609
In order to produce a single set of inputs and to run a single case, uncertain
parameters are randomly extracted from a uniform distribution and, for each case, the
eigenvalues of the open loop transfer function are calculated as a function of frequency.
Fig. 4 summarises the results for the inner and outer leading wheels, with black dots
depicting the unstable frequencies for each run; the wheel mobility at the flange contact
point is also shown for reference. A total of 450 cases have been calculated giving an
idea of the variability of the phenomenon.
The model predicts unstable frequencies that are very close to measured ones
(Fig. 2). In particular the most important frequencies 1330 Hz, 1550 Hz and 2550 Hz
are the most likely to be unstable in the model as well. Frequencies such as 2100 Hz and
3700 Hz that, although present in the measurements, have less importance in terms of
levels and persistency, are also found in the predictions.
Interestingly, in the predictions it is often found that unstable eigenvalues have
higher frequencies than the corresponding natural mode. For example, the mode at
1420 Hz appears to generate squeal at around 1550 Hz. This has been verified by
adding fictitious damping to the mode at 1420 Hz; as a consequence the unstable
frequencies at around 1550 Hz disappeared. Further research is in progress to
understand better the details of the phenomenon. In this sense it has been verified that
neither the high damping values related to the resilient wheel nor the presence of the
rail in the model are responsible. Moreover it has also been determined experimentally
that the positions of the peaks in the wheel FRF are hardly influenced by the vehicle
load; therefore this also cannot explain the shift in frequency.
Finally, by modifying damping ratios in the wheel modal model, those values
capable of decreasing the chances of instability can be sought. In this case, multiplying
the damping ratios of the modes at 2480 Hz, 1420 Hz and 1270 Hz by factors of 15,
5 and 6, respectively, would greatly decrease the occurrence of squeal. By repeating the
same analysis that generated Fig. 4 it has been found that, with this amount of added
damping, these modes are likely to be involved in squeal in less than 10 % of the
450 cases considered.
5 Conclusions
Despite the relatively low interest in the railway noise literature, curve squeal can also
occur due to flange contact; moreover, when a grooved rail or a check rail is fitted onto
the track, the flange-back contact can also have a significant role.
A real case of a tramcar negotiating a sharp curve has given the motivation for
developing a model capable of considering two-point contact. By performing stability
analysis of the open loop transfer function, possible unstable frequencies have been
predicted. A degree of uncertainty has been given to the most important input parameters.
Qualitative comparison with measurements showed that the most important tones
recorded during the tram pass-by can be found in the predictions. Noticeably, the same
frequency shift between wheel natural frequency and unstable frequency value as
measured is found in the predictions. Further investigations are in progress to
understand the physical reason for this.
610 G. Squicciarini et al.
In order to verify whether all the predicted unstable frequencies would really appear
and what the noise levels would be, a time domain solution of the same loop can be
developed.
a 10
-3
400
Mobility, m/s/N
300
Case no.
-5
200 10
100
-7
0 10-3
b 10
400
Mobility, m/s/N
300
Case no.
-5
200 10
100
-7
0 10
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency, Hz
Fig. 4. Black dots: curve squeal occurrences. Lines: mobility at flange contact point. − − −:
normal; ⎯⎯: tangent. a: inner wheel. b: outer wheel.
References
[1] Corradi, R., et al.: Experimental investigation on squeal noise in tramway sharp curves. In:
Proc. EURODYN 2011, Leuven, Belgium (2011)
[2] Rudd, M.J.: Wheel/rail noise—Part II: Wheel squeal. Journal of Sound and Vibration 46(3),
381–394 (1976)
[3] Chiello, O., et al.: Curve squeal of urban rolling stock—Part 3: Theoretical model. Journal
of Sound and Vibration 293(3-5), 710–727 (2006)
[4] Monk-Steel, A.D., et al.: An investigation into the influence of longitudinal creepage on
railway squeal noise due to lateral creepage. Journal of Sound and Vibration 293(3-5),
766–776 (2006)
[5] Huang, Z.: Theoretical modelling of railway curve squeal. PhD Thesis, University of
Southampton, Institute of Sound and Vibration Research (2007)
[6] Cheli, F., Corradi, R., Facchinetti, A.: A numerical model to analyse the dynamic behaviour
of modern tramcars. In: Proc. Mini Conf. on Vehicle System Dynamics, Identification and
Anomalies, pp. 183–190 (2002)
[7] Kraft, K.: Der Einfluß der Fahrgeschwindigkeit auf den Haftwert zwischen Rad und
Schiene. AET 22 Addendum to ETR, 58–78 (1967)
[8] Wu, T.X., Thompson, D.J.: Analysis of lateral vibration behaviour of railway track at high
frequencies using a continuously supported multiple beam model. J. Acoust. Soc.
Am. 106(3), 1369–1376 (1999)
[9] Kalker, J.J.: A fast algorithm for the simplified theory of rolling contact. Vehicle System
Dynamics 11(1), 1–13 (1982)
A Review of Measurement Data on the Performance
of a Resilient Track Form as a Mitigation Measure
for Ground-Borne Noise
1 Introduction
Measurements have been made at several different sites that show the change in
ground vibration obtained by replacing the existing track form with baseplates that
have a very low vertical stiffness. In each case, the particular assembly retro-fitted
into the track was the Pandrol Vanguard baseplate, which has a vertical stiffness of
about 5 kN/mm. While a significant reduction in ground vibration was found at each
site, the magnitude of this reduction and the frequency range in which it was achieved
differ between the sites.
The reasons for this variability from one site to another are not well-understood at
present. This presents difficulty in making predictions for the effectiveness of very
low stiffness baseplates at a given site, in advance of their installation. Such a
prediction is normally a requirement for a commercial project in which vibration
mitigation is a primary consideration.
The present work is intended to address the need for an improved understanding of
the behaviour of very low stiffness baseplates and also a means to predict their
effectiveness at a given site.
Insertion Loss
f1
f2 Frequency
Fig. 1. Form of the insertion loss spectrum given by a typical ground vibration model used in
commercial projects, for a reduction in track stiffness: , negative insertion loss; ,
positive insertion loss
When Pandrol Vanguard baseplates are used to replace the existing track form on
plain slab track, the new loaded track resonance frequency, f1, is typically about
25 Hz and that for the existing track, f2, normally lies in the range from 50 Hz to
100 Hz. This corresponds to an existing track stiffness of up to about 100 kN/mm.
When these existing track forms are to be replaced by Vanguard baseplates, the
model predicts an increase in vibration, i.e. negative insertion loss or insertion gain,
up to a frequency of about 35 Hz (1.5 times f1), and a reduction in vibration, i.e.
positive insertion loss, for frequencies greater than about 35 Hz. The maximum
insertion gain expected from this type of model is typically 10 dB to 20 dB, in the
25 Hz one-third octave frequency band, with a maximum insertion loss of a similar
level at frequency f2.
In addition to the excitation of the rail due to the combined wheel-rail roughness,
there are other kinds of excitation that may be important at the lower end of the
frequency range of interest here.
i. Quasi-static excitation due to the moving wheel loads.
ii. Parametric excitation due to the discrete rail supports.
iii. Unevenness in the track foundation.
iv. Variability in the stiffness of the track supports along the length of the track.
The above are not normally accounted for in the models used to predict the
effectiveness of a given track form as a mitigation measure for ground-borne noise in
commercial projects. Excitation types i. and ii. can be accounted for in more complex
models. For example, it was found in [6] that for at-grade track a more resilient track
form can bring about a significant reduction in ground vibration due to excitation type
614 S.J. Cox and D. Herron
i., but the effects of this are near-field in nature and are therefore unlikely to be
significant at typical receiver positions. It is not clear if this also applies to the case of
ground vibration above underground railways.
Excitation types iii. and iv. were identified in the analytical study of Hunt [7] as
ones that might explain the difference between the results of the ground vibration
models used in commercial projects and vibration measurements made at some sites
before and after fitting a more resilient track form. The effects of these excitation
types are more difficult to include in computer models and require measurement data
that is specific to the site (type iii) or the track forms (type iv) under consideration. At
present, there is a paucity of such measurement data.
Ground vibration measurements have been made above railway tunnels at several
sites and also beside an at-grade track at one site. These measurements show the
effect of replacing the existing track form with Pandrol Vanguard baseplates. Four
sets of measurements have been chosen for presentation here. These are described in
Table 1 below.
These four cases were selected from an assessment of the quality of the
measurements and the requirement that the ‘before’ and ‘after’ data sets provide a
basis for a direct comparison of the vibration performance of the two track forms.
This means that the ‘before’ and ‘after’ measurements were made on the same section
of track, with only a short period of time between them, with no rail change or rail
grinding between them.
Note that the timber sleepers at site B were replaced by concrete sleepers at the same
time that the baseplates were changed. While this introduces an additional variable to
the comparison, it provides an opportunity to evaluate the possibility that variability in
the track support stiffness along the length of the track is an important excitation
A Review of Measurement Data on the Performance of a Resilient Track Form 615
mechanism with regard to ground vibration (excitation type iv in Section 2.3). This is
because the timber sleepers had been in service for several years and many of the
sleepers contained large cracks and there were voids under some sleepers.
Measurements for the vertical point accelerance on the rail head at positions directly
above a number of sleepers showed that this varied considerably before the track
structure was changed, but that this variability was reduced enormously by fitting
concrete sleepers and Pandrol Vanguard baseplates.
The insertion loss for vertical ground vibration achieved by replacing the existing
track form with Pandrol Vanguard baseplates (and also for replacing the sleepers at
Site B) is shown for the four selected sites in Fig. 2 below.
f1 1.5 x f2 f2
10 20 30 40 50 60 70 80 90 100
Fig. 2. Insertion loss for vertical ground vibration shown versus one-third octave band centre
frequency
Comparing the basic form of the curves in Fig. 2 with that in Fig. 1 shows that the
insertion loss curves found for all four sites are quite consistent with expectations
based on the BOEF model in the upper part of the frequency range shown. In all cases
there is a broad peak in the insertion loss spectrum, centred on the expected loaded
track resonance frequency of the existing track form.
There is a significant range in the peak value of insertion loss for the four sites,
from about 11 dB for Site B to 22 dB for Site C. The relatively low peak value for
Site B may be a result of the known variation in the stiffness of the existing track
form from one support position to another at this site, such that the loaded track
616 S.J. Cox and D. Herron
resonance frequency f2 is ill-defined at this site. The range in peak insertion loss is
about 7 dB between the other three sites, with no obvious dependence on the
properties of the existing track forms or the characteristics of the site.
In the frequency range up to and including the 31.5 Hz frequency band the
insertion loss spectra are less consistent with each other and with the exception of Site
D, they are also less consistent with expectations based on the BOEF model than they
are at higher frequencies. In this lower frequency range, the insertion loss at the four
sites can be summarised as follows:
Since it does not seem possible to attribute the differences shown in the insertion
loss achieved by fitting Pandrol Vanguard baseplates at the four sites to known
characteristics/conditions at these sites it seems likely that it is characteristics/
conditions at the sites that are not well understood that are responsible for these
differences. Unevenness in the track foundation (mechanism iii. from above) and
variations in the track support stiffness (iv.) seem to fit this description. Since these
are characteristics/conditions of the track structure, rather than the ground, it is also of
interest to consider vibration measurements made on the track at these sites.
With the exception of Site C, suitable measurement data for the vertical vibration of
the track slab/sleepers is available for the test sites for which the ground vibration was
considered in Section 3 above. The insertion loss for the vertical vibration of the track
slab/sleepers at these sites is shown in Fig. 3 below.
10 20 30 40 60 70 80 90 100
Fig. 3. Insertion loss for vertical vibration of the track slab/sleepers shown versus one-third
octave band centre frequency
For site A, the insertion loss spectrum for track slab/sleeper vibration in Fig. 3 is
very similar to that for ground vibration in Fig. 2. This would be expected, because
the ground vibration measurements were made at a position close to the track slab.
However, there is a small difference between them in the frequency bands up to and
including 25 Hz. The insertion gain shown for the slab vibration is significantly
smaller than that for the ground vibration in this range.
The change in slab/sleeper vibration on fitting the Vanguard baseplates is also
more favourable than the change in ground vibration at Sites B and D. This is the case
618 S.J. Cox and D. Herron
in much of the frequency range shown for Site B, but only in the lower part of the
frequency range for Site D, where significant insertion loss is shown for the slab
vibration but not for the ground vibration.
Comparing Figs. 2 and 3 therefore indicates that the effect(s) by which reducing
the track stiffness bring about a reduction vibration in the lower part of the frequency
range shown, up to and including the 31.5 Hz frequency band, are consistently greater
at the track slab/sleeper than they are at the ground surface. This is consistent with
the finding in [6] for the particular case of the quasi-static excitation mechanism for
at-grade track.
5 Conclusion
The measurement data presented in this work for all four sites shows that the Pandrol
Vanguard baseplates are an effective mitigation measure for ground-borne noise.
Their performance in the lower part of the frequency range of interest does vary
significantly from site to site and is in some cases quite different to that expected from
the models used to predict their performance in advance of their installation in
commercial projects.
The performance of the Vanguard baseplates is significantly better than expected
in this lower part of the frequency range for two of the four sites. The reasons for this
are not clear. There does seem to be a case for performing additional measurements
during future projects. The aim of these measurements would be to quantify
characteristics of the site that may relate to excitation mechanisms that cannot be
easily accounted for using the models.
References
[1] Jones, C.J.C.: Chapter 12 of Railway noise and vibration: Mechanisms, modelling and
means of control. Thompson, D.J. Elsevier, Oxford (2009)
[2] Jones, C.J.C.: Chapter 13 of Railway noise and vibration: Mechanisms, modelling and
means of control, Thompson, D.J. Elsevier, Oxford (2009)
[3] Brennan, M.J., Ferguson, N.S.: Vibration control. In: Fahy, F.J., Walker, J.G. (eds.)
Advanced Applications in Acoustics, Noise and Vibration. Spon Press, London (2004)
[4] Hussein, M.F.M., Hunt, H.E.M.: A software application for calculating vibration due to
moving trains in underground railway tunnels. In: Proceedings of NOVEM 2009: Noise
and Vibration: Emerging Methods, Oxford, UK (2009)
[5] Forrest, J.A., Hunt, H.E.M.: Ground vibration generated by trains in underground tunnels.
Journal of Sound and Vibration 294, 678–705 (2006)
[6] Triepaischajonsak, N., Thompson, D.J., Jones, C.J.C., Ryue, J.: Track-based control
measures for ground vibration – the influence of quasi-static loads and dynamic excitation.
In: Maeda, T., Gautier, P.-E., Hanson, C.E., Hemsworth, B., Nelson, J.T., Schulte-
Werning, B., Thompson, D., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail
Transportation Systems. NNFM, vol. 118, pp. 249–257. Springer, Heidelberg (2012)
[7] Hunt, H.E.: Types of rail roughness and the selection of vibration isolation measures. In:
Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson,
J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 341–347. Springer, Heidelberg (2008)
Challenges in the Design and Fabrication
of Elastomeric Springs for Floating Slab Tracks
ATS Consulting,
215 N Marengo Avenue, Pasadena, CA 91101, USA
srajaram@ATSConsulting.com
Summary. Floating slab tracks (FST) are mass-spring systems that are used to
reduce groundborne vibration generated by trains. The effectiveness of FSTs can
be estimated using the single degree-of-freedom (SDOF) model. More detailed
study of the modes and shapes of the FST are sometimes performed using Finite
Element Method (FEM) models. This paper also explores the suitability of a
natural rubber (NR) based elastomeric spring for FST designs that require
relatively high deflection and good performance at low service temperatures
(-40 °C). The paper also evaluates the utility of simple SDOF models as a
prediction tool, and the potential factors that would result in higher damping at
the resonance frequencies (f0).
1 Introduction
2 Background
The typical properties that are specified for elastomeric springs are static and dynamic
moduli, hardness, resilience, hysteresis, damping, creep, fatigue, compression set and
temperature sensitivity. A variety of resilient polymers including elastomers and foams
are used as springs for vibration isolation. The most notable advantages of NR are low
cost, ease of fabrication and its moderate to good ratings for a wide range of
performance parameters associated with vibration isolators. The NR-based Malaysian
Rubber formulation has been almost exclusively used for FSTs in North America over
the last three decades. As noted earlier, there is insufficient information on its durability
when subjected to greater than 10 % deflection and its cold temperature properties. The
cold temperature performance of NR compounds can be improved by blending it with
synthetic rubbers and adjusting the cure system, but usually it is not feasible to invest in
an R&D effort in the middle of an infrastructure project.
The basic concept of floating slab systems can be explained with the SDOF system
shown in Fig. 1. The parameters in the model are mass (m), stiffness (k), viscous
damping (c), exciting force (Fin) and transmitted force (FTransmitted). Key derived values
of an SDOF system are resonance frequency (f0), critical damping, damping percent
and transmissibility. The transmissibility is the difference in decibels between the
output and input forces to a system. Fig. 2 shows the force transmissibility for an SDOF
system for damping ranging from 5 % to 100 % of critical damping. The maximum
amplification occurs at f0 and the system starts to attenuate the transmitted force at
frequencies greater than 1.4 times f0. Amplification is controlled by damping. For the
considered damping range, the amount of amplification is 20 dB for 5 % critical
damping and drops to just over 1 dB when damping is 100 % critical. At frequencies
well above f0, the FST is more effective with reduced damping.
Field measurements of the performance of 15 different floating slab systems have been
identified through the review of literature and reports produced for previous rail transit
projects [4-13]. Table 1 summarizes the information on each FST. It is not always clear
from the literature and reports what the design f0 was for the FST. In these cases, the f0
has been estimated based on the reported insertion gain. We assumed that the resonance
is at the same frequency as the maximum insertion gain. Fig. 3 shows the measured
insertion gain for each FST after normalization to an f0 of 7 Hz. It also shows the
insertion gain estimates for a 7 Hz FST with 20 % damping based on the SDOF model.
The figure illustrates that the estimates from the SDOF model agree well with the
measured insertion gain.
4 Test Results
Elastic Deflection
Linear Regime
The primary question was whether the relatively high deflection (14 %) for Malaysian
Rubber formulation during service could adversely affect longevity of the pads by
reducing fatigue life and accelerating creep and ageing. Fatigue testing of coupons was
performed based on standardized DeMattia tests. The material failed after 100K cycles.
The conclusion was that the loading cycles were unrealistic and the undue heat buildup
from the rapid flexing accelerated polymer degradation and caused premature failure.
Therefore, a custom test that simulated the actual loading conditions was developed and
included two steps. In Step 1, a full-size pad was instrumented with thermocouple and a
cyclic load that was equivalent to the maximum service load was applied to the pad.
The optimal cycling rate for the test was determined. In Step 2, the optimal conditions
were applied to a virgin full-size pad and the results showed that higher deflection
624 S. Rajaram and H.J. Saurenman
would not adversely affect the fatigue life of the material. Analysis also showed that the
pads under higher deflection can be expected to resist ozone degradation and fatigue
cracking better than pads that are under relatively lower deflection [1].
A major challenge with the Malaysian Rubber formulation is that the springs are
custom-built for every project and the tolerances need to be very clearly specified to
maintain quality. For example, the prototype pads for the CCLRT project had static
stiffness in two regimes: (1) linear and (2) non-linear regime (see Fig. 5). At the
anticipated maximum service load of 60 kN, the apparent total deflection of prototype
pads was 23 % although the deflection in the linear regime was only 14 %.
Investigation revealed that the top and bottom surfaces of the tested support pads were
softer than specified. The softening was attributed to heat flow issues during
fabrication. The problem was addressed, and the production pads were free of surface
softening and the resulting non-linear regime (see Fig. 5) eliminated.
5 Modeling
The dimensions of the typical Cedar Street FST slab are 9100 mm longitudinally,
2440 mm laterally and the nominal depth is 510 mm. The weight of each slab is
approximately 27200 kg and the total volume of each slab is approximately 11.3 m3.
Each slab is supported by 12 main support pads (‘vertical support pads’) spaced at
1500 mm in two rows. The rows are parallel to the rails and are located directly under
the rails. The under-slab cavity has a nominal uncompressed height of 150 mm. The
vertical support pads are circular and the dimensions in uncompressed condition are
343 mm x 178 mm. There are 12 side pads per slab and one end pad. The rails are
embedded in the floating slabs with a railboot system.
The FST was modeled using Abaqus 6.11. The FEM modeling was performed by
Duschlbauer and Roulo [14]. The model comprises four 9.1 m slabs, which total
approximately 36.5 m of FST. The floating slabs were meshed with quadratic
(20-node) brick elements. All elastic support pads were represented using translational
springs. The foundation under the floating track was not modeled. Instead, all springs
from the slabs to the troughs were connected ‘to ground’. The total stiffness of the air
trapped under the slab (air spring stiffness) was divided by the number of main support
pads and was then added to the dynamic vertical stiffness of each main support pad.
The rails were modeled using 125 mm long quadratic beam elements. The railboot
system was modeled with a series of translational springs also spaced at 125 mm.
Two axles (134 kg/axle) and four wheels (200 kg/wheel) at a nominal offset of
1.75 m were introduced. The wheels were modeled as lumped masses attached to the
ends of the axles. The translational DOFs of the rail and the axle ends were coupled.
The influence of the gage ties were assumed to be secondary because they are fully
embedded in concrete and represent a negligible volume fraction. All analyses were
Challenges in the Design and Fabrication of Elastomeric Springs for Floating Slab Tracks 625
conducted both for room temperature stiffness 20 °C and for stiffness at cold
temperatures –30 °C.
The response of the system due to non-deterministic excitation in the vertical
direction at the four wheels was calculated in a random response analysis. A ‘white
noise’ force spectrum of unit value was applied over the frequency range of 1 Hz to
225 Hz at the four wheel/rail contact points in the vertical direction. This excitation was
assumed to be stationary. The random response analysis was preceded by a mode
extraction step in which all modes up to 350 Hz were extracted. The analysis was based
on 20 % critical damping for frequencies up to 40 Hz. Between 40 and 350 Hz the
damping was linearly reduced with increasing frequency from 20 % critical at 40 Hz to
5% critical at 350 Hz. Within the framework of the FEM model, damping was
implemented as modal damping. The rail stresses were estimated for the maximum
static loading scenario W5 ‘Centre Truck’ of the AW4 weight (677.5 kg). This loading
scenario assumes a maximum of 271 kg (or 40 % of the AW4 weight) to load the rails at
axle spacing between 1.7 to 1.9 m. The model has been loaded with four vertical loads
of 67.75 kN which were applied to the rail at a spacing of 1.625 m.
6 Conclusions
This paper identifies the challenges in designing FSTs with NR-based springs for
demanding service conditions. The NR formulation used in North America is suitable
for FST designs with up to 15 % deflection and that require good spring performance
when the ambient temperature reaches -40 °C. Also, the testing and analysis showed
that fatigue and ageing of the material would not be affected by the relatively high
deflection designed for the CCLRT project.
626 S. Rajaram and H.J. Saurenman
This study also demonstrates that at the resonance frequency, the damping of FSTs
is significantly higher than the material damping of the elastomeric pads. The potential
factors contributing to this high damping and the approach required to refine the FEM
model were discussed. Also, the studies have shown that SDOF models can provide
reasonable estimates of the FST insertion gain.
References
[1] Rajaram, S., et al.: Railway floating slab track design: Review of specification for
elastomeric springs. In: InterNoise 2012, New York, USA (2012)
[2] Lindley, P.B.: Engineering design with NR, NR Technical Bulletin, 4th edn. Malaysian
Rubber Producers Research Association, Brickendonbury (1978)
[3] Morton, M.: Rubber technology, 3rd edn. Kluwer Academic Publications, Dordrecht
(1999)
[4] Jaquet, T., Garburg, R.: Measurements and investigations at the floating-track-bed system
in the north-south tunnel in Berlin. In: Schulte-Werning, B., Thompson, D., Gautier, P.-E.,
Hanson, C., Hemsworth, B., Nelson, J., Maeda, T., de Vos, P. (eds.) Noise and Vibration
Mitigation for Rail Transportation Systems. NNFM, vol. 99, pp. 150–157. Springer,
Heidelberg (2008)
[5] Saurenman, H.J.: Develop noise and vibration mitigation strategies for SVRT project,
Final Report, ATS Consulting (January 2005)
[6] Wolfe, S.L.: Floating slab design and performance for embedded in-street track - a case
study at San Francisco municipal railway. In: APTA, Rapid Transit Conference, San
Diego, USA (1995)
[7] Nelson, J.T., Jessop, A.: Preliminary engineering vibration control for the north link
preferred alternative at the University of Washington, WIA, for Puget ST Consultants,
Report #: CIN 2100-0604-1679 (April 2006)
[8] Wolfe, S.L.: Ground vibration measurements of train operations on segment 2A of the Los
Angeles red line. Draft Report, WIA, for PB/DMJM (November 1996)
[9] Lawrence, S.T.: Toronto’s double tie trackbed system. In: APTA, Rapid Transit
Conference, Chicago, USA (1978)
[10] Duschlbauer, D.: unpublished data
[11] Wilson, G., Wolfe, S.L.: Vibration and noise performance characteristics of resilient track
support systems used in MARTA subways, Report, WIA, for STEDEF, Inc. (August 1980)
[12] Nelson, J.T., et al.: Metrorail operational sound level measurements: Ground-borne
vibration and noise levels for the Washington metropolitan area transit authority,
Preliminary Report, WIA, for De Leuw. Cather & Company (December 1979)
[13] Hui, C.K., Ng, C.F.: The effects of floating slab bending resonances on the vibration
isolation of rail viaduct. Applied Acoustics 70(6), 830–844 (2009)
[14] Saurenman, H.J., et al.: CCLRT Civil East Floating Slab Track: FE modeling of vibration
isolation efficiency, Draft Report, for AECOM (June 2012)
[15] Shimozawa, K., Tohtake, T.: An air spring model with non-linear damping for vertical
motion. QR of RTRI 49(4) (2008)
[16] Li, Z., et al.: Study on damping of air spring with additional chamber. In: Consumer
Electronics, ICCN 2011, Xianning, China (2011)
[17] Zhang, G., Li, J.: Dynamical model and characteristics analysis of air spring system of
low-speed Maglev train. In: 4th International Conference on Advanced Computer Theory
and Engineering 2011, Dubai, UAE (2011)
Vibration Mitigation by Innovative Low Stiffness Rail
Fastening Systems for Ballasted Track
Summary. The collaborative work presented in this paper is part of the European
project RIVAS, dedicated to the mitigation of ground-borne vibrations from rail
traffic. In this paper, focus is made on innovative low stiffness fastening systems
designed for ballasted track. These kinds of systems reduce ground-borne
vibration by the filtering effect provided by their low stiffness in comparison to
the global track stiffness. Numerical simulations performed as a first step
confirmed this expected behaviour: ground vibration is strongly attenuated above
a cut-off frequency at which resonance occurs. Following these results, two
fastening systems designed by Pandrol (member of the RIVAS consortium) were
chosen for testing, first in laboratory and then in a commercial track. The
laboratory tests mainly consisted in the evaluation of the systems stiffness in
realistic conditions (single sleeper in a ballast box), using an appropriate
excitation. Combined with acceleration measurements, those tests confirmed the
main results expected from simulations but also raised specific issues for the
testing of low stiffness fastening systems. In order to be tested in a commercial
track in France, some preliminary work was required on these systems; it is also
presented in this paper.
1 Introduction
The RIVAS project (Railway Induced Vibration Abatement Solutions) aims for the
reduction of the environmental impact of ground-borne vibration from rail traffic. In
this three-year European project started in 2011, the problematic of ground-borne
vibration is assessed in many aspects, from the generation mechanisms to the human
perception. In the work package 3 dedicated to mitigation measures for ballasted track,
the effect of soft fastening systems and of under sleeper pads was investigated [1]. This
paper focuses on the soft fastening systems. In order to assess the efficiency of soft
fastening systems for ground-borne vibration reduction, a large parametric study was
first performed, taking into account the influence of various parameters. In a second
step, two fastening systems designed by Pandrol (member of the RIVAS consortium)
were tested in realistic laboratory conditions. The results obtained from the numerical
simulations and from the laboratory tests are presented respectively in Sections 2 and 3.
In Section 4, some information is given concerning the implications arising from the
installation of very soft fastening systems in a commercial ballasted track.
The simulations were carried out with the software TRAFFIC developed by KU
Leuven, based on a coupled 2.5D FEM-BEM approach for the track-soil system [2].
The track and the soil are assumed to be invariant in the longitudinal direction. Periodic
support by sleepers and rail pads is replaced by an equivalent continuous support. The
continuous model of the rails, the rail pads and the sleeper contains four degrees of
freedom in the plane of the cross-section as indicated in Fig. 1.
(a) (a)
(b) (b)
(c)
(d) (e)
The rails are modelled as Euler beams (a), the rail pads as continuous spring–damper
layers (b) and the sleepers as a uniformly distributed mass, rigid in the plane of the
cross-section (c). The sleepers are supported by a ballast layer located on a
homogeneous soil. The ballast is modelled as an elastic continuum with 2.5D finite
volume elements (d). The soil is modelled by means of boundary elements at the
interface between the ballast and the soil (e).
• The spatial variation of the track stiffness is not taken into account (no parametric
excitation).
• The excitation mechanism, the unevenness, is formed by the irregularities of the
wheel and rail rolling surfaces. The unevenness profile considered in this study is a
standard profile. It is described by a PSD ΓU(ky) in the spatial domain, transposed in
the frequency domain through the train speed v.
• The vehicle and track dynamic behaviours are described by their respective
compliances CkV(ω) and CT(ω), so that the PSD of the kth wheel/rail interaction force
can be written as given by Eq. (2.1).
ΓU (−ω / k )
Γk (ω ) =
[
v C (ω ) + CkV (ω )
T
]
2
(2.1)
The track compliance CT(ω) is obtained from the track model of Fig. 1 and the
vehicle compliance CkV(ω) is obtained from a bogie model with four degrees of
freedom (vertical displacement of both wheelsets and the body, and rotation of the body
around the centre of gravity). Traction bogies and coach bogies are differentiated.
The output of the model is the vibration velocity of the soil surface in the free-field
due to a train pass-by at speed v, assimilated to its PSD function ΓS(ω). It is computed
from the PSD of all wheel/rail interaction forces, using the soil mobility h(Mk,M,ω)
(transfer function between the kth wheel/rail contact point on the rail and the point M at
the soil surface in the free field). Under the hypothesis of uncorrelated loads and a
stationary train, the PSD of the free-field velocity is given by Eq. (2.2):
ΓS (ω ) = ∑ Γk (ω ) h( M k , M , ω )
2
(2.2)
k
The rail pad stiffness varied from 25 MN/m to 150 MN/m (reference case).
630 B. Faure et al.
2.4 Results
In terms of track dynamics, very soft resilient layers behave in the track as simple
vibration filters characterized by a resonance frequency at which amplification occurs
and after which vibrations are attenuated. The frequency affected in our case is the
resonance of the unsprung mass between the stiffness of the track and the vehicle, which
occurs at around 40 Hz. As a consequence, very soft fastening systems lead to an increase
of ground vibration up to 10 dB at 40 Hz. At higher frequencies, ground vibration is
attenuated up to 15 dB. An example of the IL computed for 25 MN/m rail pads versus
150 MN/m rail pads is given in Fig. 2. The softer the rail pad, the lower the resonance
frequency and the more important the filtering effects (resonance and attenuation). These
results depend of course on some parameters such as the soil stiffness and layering, the
ballast height and the sleeper mass. Nevertheless, simulations showed only small effects
of these parameters on the mitigation efficiency [1].
Fig. 2. IL for a rail pad stiffness of 25 MN/m relative to a rail pad stiffness of 150 MN/m
The results from this numerical study indicate that very soft fastening systems can
induce a large mitigation effect for ground borne vibration, i.e. for frequency above
40 Hz (below 40 Hz, no effect or even a negative impact is noticed). On the other hand,
it is expected that very soft fastening systems will induce a reduction of parametric
excitation (homogenization of track stiffness) which is not taken into account in these
numerical simulations. Therefore, better performances could be observed in track,
especially at the vehicle on track resonance frequency.
Following the promising numerical results, laboratory tests were performed to assess
the efficiency of different systems in more realistic conditions [4]. Two original designs
were tested as well as rail pads of different stiffness on classical fastening systems.
The two original designs are manufactured by Pandrol, member of the RIVAS
consortium.
Vibration Mitigation by Innovative Low Stiffness Rail Fastening Systems 631
The DFC Valiant is a fastening system with two resilient layers: a rubber pad (a) is
placed between the sleeper and a cast-iron baseplate (b) on which the rail is mounted
through a Fastclip-like system including a classic rail pad (c). A wide range of stiffness
can be achieved with the DFC Valiant by using different rail-pad/baseplate-pad
combinations. The dynamic stiffness for the softest DFC Valiant was measured at
20 MN/m (following EN-13146-4). The Vanguard is a fastening system where the rail
is supported under the head and in the web with rubber wedges (d) and cast-iron side
brackets (e), leaving the foot of the rail suspended. The dynamic stiffness of the
Vanguard was measured at 6 MN/m. These systems and their components are
represented in Fig. 3.
d
c
Fig. 3. Low stiffness fastening systems tested within the RIVAS project
Rail Rail
Sleeper Sleeper
Ballast
Ballast
Fig. 4. Experimental setup. Case small amplitude (left). Case high amplitude (right)
632 B. Faure et al.
For the small amplitude set-up, a static preload was applied to the rails by a
hydraulic jack while a dynamic excitation of small amplitude was also applied on the
rail by a modal shaker. The static load was 60 kN for DFC and Fastclip and 30 kN for
Vanguard. The dynamic excitation amplitude was 50 N from 8 Hz to 250 Hz for all
systems.
For the high amplitude set-up, a sinusoidal excitation was applied to the rail by the
hydraulic jack at 5 Hz, with an amplitude of +/- 10 % relatively to a mean value (60 kN
for DFC and Fastclip, 30 kN for Vanguard). A steel I-beam was used between the jack
and the rail as the jack was unable to reach the rail.
3.3 Results
The two systems presented in Section 3.1 are planned to be tested in a commercial track
(tests have been performed since), requiring two preliminary actions.
Vibration Mitigation by Innovative Low Stiffness Rail Fastening Systems 633
As the Vanguard and the DFC are designed for slab tracks, they do not fit to any
existing sleeper. Thus, a new sleeper was designed by Sateba, member of the RIVAS
consortium, in accordance to Pandrol and SNCF requirements.
Fig. 5. Insertion Loss calculated using the stiffness measured with the “high amplitude” set-up.
DFC (dashed) and Vanguard (dotted) in comparison to Fastclip with 9 mm rail pads.
This sleeper, called M260-DFC, is adapted to both Vanguard and DFC. The test
section will be built with 150 m of these sleepers on which the different systems will be
mounted one after the other, including different assemblies for the DFC (different
combinations of pads). At the end of the tests, the softest DFC assembly will remain on
track for 2 years; it will then be retrofitted in its stiffest version (equivalent to the
French regular fastening). This required a complete certification process for both the
sleeper and the DFC (soft and stiff). The Vanguard, which will be installed on track
only for a few days, required just some standard tests to guarantee the safety of the
commercial operations.
As the test section will be inserted in a regular track, there will be a large gap in the
track stiffness between the reference and the test section resulting in high transient
excitation that could be damaging for the track. In order to prevent this risk, transition
zones of intermediate stiffness have to be inserted in-between as shown in Fig. 6.
Finally, simulations were performed to predict the dynamic behaviour of the rolling
stock along the transition. The acceleration of the leading bogie of a Freight wagon
running at 90 km/h was calculated with a SNCF validated software. When regular track
defects were added in the model, almost no influence of the stiffness gaps was found.
5 Conclusion
Acknowledgments. The research leading to these results has received funding from the
European Union Seventh Framework Programme under grant agreement no 265754.
References
[1] THE RIVAS PROJECT: Results of the parameter studies and prioritization for prototype
construction for ballasted track. Deliverable 3.2. UIC (2012)
[2] François, S., et al.: 2.5D coupled FE-BE methodology for the dynamic interaction between
longitudinally invariant structures and a layered halfspace. Computer Methods in Applied
Mechanics and Engineering 199(23-24), 1536–1548 (2010)
[3] Lombaert, G., Degrande, G.: Ground-borne vibration due to static and dynamic axle loads of
InterCity and high speed trains. Journal of Sound and Vibration 319(3-5), 1036–1066
(2009)
[4] THE RIVAS PROJECT: Optimization of the fastening system and the sleeper: numerical
and test bench results. Deliverable 3.4. UIC (2012)
[5] Wu, T.X., Thompson, D.J.: On the rolling noise generation du to wheel/track parametric
excitation. Journal of Sound and Vibration 293(3-5), 566–574 (2006)
Control of Railway Induced Ground Vibrations:
Influence of Excitation Mechanisms on the Efficiency
of Resilient Track Layers
Summary. This paper focuses on the insertion of resilient layers in the track as
mitigation measures for ground-borne noise and vibration. Soft rail pads,
under-sleeper pads or ballast mats are considered as mitigation measures.
Common train-track interaction models using unevenness excitation predict
reduction of ground vibrations around and above the natural frequency of the
rolling stock unsprung mass on the track stiffness, in agreement with
experimental observations. In the lower frequency range, i.e. below this natural
frequency, computations show an increase of vibration that is not always
confirmed by experimental results recorded on ballasted tracks.
The excitation mechanisms play a role in the measured efficiency. In this
paper the insertion losses computed with an unevenness and a parametric
excitation model are compared. The results show that, depending on the train
speed, the low frequency range insertion loss, related to resilient layer insertion
in the track, is not always negative for parametric excitation.
1 Introduction
Ground-borne vibration and noise is a growing issue in railways, as the traffic load
capacity and the train frequency intensify. The development of mitigation measures
requires validation steps based on measurement and simulations. Resilient components
- such as rail pads, under-sleeper pads or ballast mats – are current mitigation measures
in the track. The experimental results usually show a reduction of ground vibration
above the vehicle unsprung mass natural frequency on the track stiffness. Below this
frequency, reduction or increase can be observed depending on the situation [2].
Excluding the quasi-static excitation related to the moving load, which vanishes
rapidly in the ground away from track, the ground vibrations related to train pass-by
can be considered as a sum of two main components: the unevenness excitation and the
parametric excitation.
The unevenness excitation is related to geometrical defects of rolling surfaces such
as wheel-out-of–roundness, flats or squats for the wheel and unevenness, squats, or
weld defects for the rail [3]. The parametric excitation is related to the variation of track
Insertion loss is computed using a dedicated numerical approach for each excitation
type. The unevenness excitation model is treated using 2.5D models in the
frequency-wavenumber domain, while the parametric excitation model is treated in the
space-time domain.
A second equation (2.2) is used to calculate the soil vibration v(ω) using H(ω) the
vertical transfer mobility from rail to soil.
The unevenness excitation is given by the equation from ORE document [9]:
UR=A*(1+n/n0)3, with A = 308x1.15x10-6 m2/cycles/m, and n0 = 0.0489 (medium rail
quality).
Control of Railway Induced Ground Vibrations 637
Fig. 2. Modeling strategy for parametric excitation; K(t = x/v) represents the varying track
stiffness seen from the wheel vs distance or time
Fig. 3. Computed track stiffness using a FEM with random foundation stiffness along track
Three track cases are simulated. For the reference case (case 1), the vertical stiffness
per sleeper is considered as Gaussian with a mean value of 210 MN/m and a standard
deviation of 20 MN/m. A correlation length of 1 m is accounted for in the random
sample generation [10]. The vertical stiffness of cases 2 and 3, represents a track with
the same stiffness per sleeper as for case 1, but with rail pads of 150 MN/m, and
90 MN/m respectively (Fig. 3).
Control of Railway Induced Ground Vibrations 639
-7
10
case1
case2
case3
Compliance [m/N]
-8
10
-9
10
-10
10
20 40 60 80 100 120 140
Frequency (Hz)
Fig.5 presents the computed insertion loss (IL) relative to case 1, for case 2 and case 3.
A speed of 300 km/h is considered. The IL values depend on the excitation type.
With unevenness excitation, IL is close to zero in the low frequency range, up to
about 40 Hz. At low frequencies, the vehicle receptance Aw(ω) is predominant over the
track receptance Ar(ω). Then Eq. (2.1) becomes
which does not depend on track properties. Consequently, ground vibration amplitudes
are the same for all three cases (see Fig. 4).
Around 50 Hz for case 2 and 63 Hz for case 3, a minimum negative value of IL is
observed. This minimum corresponds to the unsprung mass resonance on the track
stiffness, where an amplification of contact force is observed in Fig. 6.
Above these frequencies, the track receptance becomes higher than the vehicle one.
The contact force from Eq. (2.1) does not depend on the vehicle anymore.
Consequently, the contact force related to case 1 exhibiting a low track receptance will
be higher than the contact force related to cases 2 and 3 with higher track receptance.
640 B. Nélain et al.
20 20
10 10
0 0
-10 -10
-20 -20
8 16 31.5 63 125 250 8 16 31.5 63 125 250
One third octave band frequency [Hz] One third octave band frequency [Hz]
Fig. 5. Computed insertion loss with parametric and unevenness excitation for a train speed of
300 km/h
Unevenness Unevenness
100 100
case 1
GBV @8m dB [re 5e-8 m/s]
90 case 2 90
80 80
70 70
case 1
60 60 case 3
50 50
8 16 31.5 63 125 250 8 16 31.5 63 125 250
Parametric Parametric
100 100
GBV @8m dB [re 5e-8 m/s]
case 1 case 1
80 80
case 2 case 3
60 60
40 40
20 20
0 0
8 16 31.5 63 125 250 8 16 31.5 63 125 250
Frequency (Hz) Frequency (Hz)
Fig. 6. Computed ground vibration at 8 m from the track center with parametric and unevenness
excitation for a train speed of 300 km/h
For parametric excitation, the behavior is completely different since the contact
force is proportional to the track stiffness variation. The standard deviation for case 1 is
about 11 % of the mean stiffness, while it is 5 % for case 2 and 2 % for case 3. This
explains why the insertion loss is not zero at low frequencies, in contrast with
unevenness excitation. At the unsprung mass resonance, the IL values are positive.
Even if amplifications are present at these frequencies for case 2 and 3, the amplitude of
the excitation is reduced compared to case 1, explaining the positive IL results.
Control of Railway Induced Ground Vibrations 641
The peak observed for all cases in the octave band 125 Hz for parametric excitation
in Fig. 6 is due to the sleeper passing frequency amplification, located near 140 Hz for
this train speed. Positive IL values are computed due to the reduced excitation for cases
2 and 3 compared to case 1, due to lower stiffness variation.
The IL for parametric excitation also depends on train speed. For a train speed of
90 km/h, giving a sleeper passing frequency close to the unsprung mass resonance
frequency of case 3, negative IL values are computed for case 3 at 50 Hz, as shown in
Fig. 7.
For unevenness excitation, the IL does not depend on the speed. The excitation is the
same for a given speed, regardless of the track stiffness.
case 2
30 case 3 60
case 2
case 3
20 GBV @8m [dB]
40
10
20
0
-10 0
-20 -20
8 16 31.5 63 125 250 8 16 31.5 63 125 250
One third octave band frequency [Hz] One third octave band frequency [Hz]
Fig. 7. IL(left) and ground vibration (right) for parametric excitation – 90 km/h
The present results highlight the influence of excitation type, on expected efficiency of
resilient layer in a ballast track.
For parametric excitation, negative insertion losses are not always observed in the
low frequency range, below and around the unsprung mass resonance on the modified
track stiffness. Moreover, unlike for unevenness excitation, the insertion loss is
speed-dependant; it is affected by the relative position of the sleeper passing frequency
and of the resonance frequency of the unsprung mass on the modified track.
These results can account for measurement of positive IL at low frequencies, when
implementing resilient layers in ballast track.
Further investigations are now required to assess the relative importance of
unevenness and parametric excitations versus wavelength on ballast track. Current
wheel unevenness and out-of-roundness values should also be compared to track
excitation.
Acknowledgment. The results presented in this paper have been obtained within the
frame of the EU FP7 project RIVAS (Railway Induced Vibration Abatement Solutions)
under grant agreement No. 265754.
642 B. Nélain et al.
References
[1] http://www.rivas-project.eu
[2] Terno, H.-J.: State of the art review of mitigation measures on track, RIVAS Deliverable
D3.1 (2011)
[3] Nielsen, J.C.O.: Classification of track conditions with respect to vibration emission,
RIVAS Deliverable D2.1 (2012)
[4] Faure, B., Bongini, E.: Results of the parameter studies and prioritization for prototype
construction for ballasted track, RIVAS Deliverable D3.2 (2012)
[5] Wu, T., Thompson, D.J.: On the parametric excitation of the wheel/track system. Journal of
Sound and Vibration 278(4-5), 725–747 (2004)
[6] Faure, B., Bongini, E.: Optimisation of the fastening system and the sleeper: numerical and
test bench result, RIVAS Deliverable D3.4 (2012)
[7] Hunt, H.E.M.: Types of rail roughness and the selection of vibration isolation measures. In:
Schulte-Werning, B., Thompson, D., Gautier, P.-E., Hanson, C., Hemsworth, B., Nelson,
J., Maeda, T., de Vos, P. (eds.) Noise and Vibration Mitigation for Rail Transportation
Systems. NNFM, vol. 99, pp. 341–347. Springer, Heidelberg (2007)
[8] Lombaert, G., Degrande, G., Kogut, J., François, S.: The experimental validation of a
numerical model for the prediction of railway induced vibrations. Journal of Sound and
Vibration 297(3-5), 512–535 (2006)
[9] ORE. Question C116: Wechselwirkung zwischen Fahrzeugen und gleis, Bericht Nr. 1:
Spektrale Dichte der Unregelmässigkeiten in der Gleislage. Technical report, Office for
Research and Experiments of the International Union of Railways, Utrecht, NL (1971)
[10] Schevenels, M., Lombaert, G., Degrande, G., Clouteau, D.: The wave propagation in a
beam on a random elastic foundation. Probabilistic Engineering Mechanics 22(2), 150–158
(2007)
Measurement of Long Wavelength Irregularities on Rails
S.L. Grassie
RailMeasurement Ltd., The Mount, High Street, Toft, Cambridge, UK CB23 2RL
stuart.grassie@railmeasurement.com
1 Introduction
Dynamic loads and noise are excited by irregularities on the running surfaces of both
wheels and rails. There are a few references on wheel irregularities and rather more on
rail irregularities, one of the earliest of which is in the classical series of papers on
wheel/rail noise by Remington and his colleagues in 1976 [1]. There exist significant
shortcomings in measurements of rail roughness, not least in the fact that the main
acoustic standards (EN ISO3095 [2] and EN15610 [3]) consider roughness
measurements in wavelength ranges that are scarcely sufficient even for wheel/rail
rolling noise (3.15-630 mm for EN ISO3095 and 3.15-250 mm for EN15610).
Excellent work of a similar form has been done by Trafikverket in Sweden [4] and
ISVR in Southampton [5] to plug the gaps in measurements of rail roughness by
extending long wave measurements from track recording cars (TRC) to a shorter
wavelength and short wave measurements using the CAT (Corrugation Analysis
Trolley) [6, 7] to a longer wavelength. These analyses show a persuasive overlap of the
two sets of data, despite the fact that TRC measurements are essentially of track
geometry (which is corrected by maintenance activities such as tamping) and CAT
measurements are of irregularities on the running surfaces of the rails themselves
(which are corrected by reprofiling trains).
The range of irregularities that are present on different types of railway system and
the effects of reprofiling on rail roughness are discussed in Ref. [7]. The analysis in that
reference, in which all measurements were obtained using the CAT equipment, was
restricted to 1000 mm wavelength. This is significantly longer than considered in EN
ISO3095 and EN15610 but is nevertheless inadequate for those who are concerned
with ground-borne noise and vibration. This is particularly the case in locations where
low frequency vibration has an impact on manufacturing processes and laboratories.
The principal purpose of this paper is to present novel measurements of long
wavelength irregularities on rails (up to 5 m), obtained by an alternative technique to
that pioneered by Trafikverket and ISVR. This makes use of so-called RCA equipment
(Rail Corrugation Analysis) manufactured by RailMeasurement Ltd [8]. This
equipment has been designed for use in routine rail maintenance and to help develop a
railway’s reprofiling programme. It is typically mounted on reprofiling trains or hi-rail
vehicles. On reprofiling trains (Fig. 1(a)) the operating speed is usually 0.5-15 km/h.
The RCA mounted on a hi-rail in Fig. 1(b) is used to develop a railway’s reprofiling
programme, and was designed for speeds of up to 55 km/h. Both the CAT and the RCA
are based on inertial principles since this offers the greatest promise of being able to
measure typical rail irregularities from a moving reference frame [9].
(a) (b)
A typical standard for which the RCA can demonstrate compliance is the European
Standard EN13231-3 [10], which is significantly more demanding than its 2012
revision. This standard contains limits on residual longitudinal irregularities in
wavelength ranges of 10-30 mm, 30-100 mm, 100-300 mm and 300-1000 mm. The
reprofiling standard for the Australian standard-gauge system, from which
measurements are shown here, contains similarly demanding requirements to those in
Ref. [10] for wavelengths of up to 1000 mm combined with an extremely demanding
limit on allowable irregularities in the 1000-1500 mm wavelength range, in which
corrugation has previously been problematic on their system [11]. It is demonstrated in
Ref. [12] that the roughness limits in Ref. [2], which is an acoustic standard, and
Ref. [10], which is a reprofiling standard, are remarkably similar despite their having
been derived by different groups with different objectives.
Although this equipment is used for routine rail maintenance rather than for
“scientific” purposes, it is demonstrated here that the measurements are of sufficiently
high standard to be used for these purposes. Measurements from an RCA are also
compared to those from the CAT equipment. A comparison of this type is made as part
of the validation procedure when the RCA equipment is supplied. The CAT equipment
is widely used to obtain measurements of acoustic roughness and was shown in a
Measurement of Long Wavelength Irregularities on Rails 645
2 Measurements of Irregularities
The RCA shown in Fig. 1(a) is attached to a rail grinding train. Part of the validation
testing for this equipment was to undertake successive measurements with the CAT and
with the RCA of several hundred metres of both rails on a mixed passenger and freight
line. The test section had recently been ground, with grinding scratches apparent on the
rails but no residual grinding debris and relatively low overall roughness. There were
welds at an interval of about 12 m as a consequence of the short rails that were
historically used on this railway.
Fig. 2. “Raw displacement” from RCA runs at similar speed in same direction over 50 m of track
The “raw displacement” for both rails is shown in Fig. 2 from successive RCA runs
in the same direction at a speed of about 8 km/h over 50 m of the site. The two curves
are so similar that it is difficult to separate them at this scale, or indeed at even
considerably higher magnification. Repeatability of the equipment is extremely good:
RMS amplitudes measured on the two runs (noted above the graphs) differ by less than
5 % from one another and there is little difference in the waveforms. The so-called
“percentage exceedence” measured on the two runs differs by 0.1 %. This is typical of
correlation over the site. Only 50 m of data are shown here so that the details are
clearer.
One-third octave spectra have been calculated from CAT and RCA measurements of
both rails over the section 31.30-31.89 km of the overall site (Figs. 3 and 4). This was
undertaken for the wavelength range 6.3-5000 mm using an extension of the digital
filtering algorithms presented in Ref. [3]. The spectra are shown for two sets of
measurements with both instruments, with the TSI limit (which is slightly lower than
646 S.L. Grassie
Fig. 3. One-third octave spectra for the left rail, from CAT and RCA
Fig. 4. One-third octave spectra for the right rail, from CAT and RCA
the roughness limit for “smooth” rail in Ref. [2]) superposed. For the RCA these are the
measurements shown in Fig. 4. For the CAT, the measurements on the left rail were
taken in opposite directions, as these were the only measurements available. For the
right rail, the CAT measurements shown in Fig. 4 were taken in the same direction.
Both sets of RCA measurements and both sets of CAT measurements differ little
from one another throughout the wavelength range 6.3-5000 mm. There is also little
difference between RCA and CAT measurements in the wavelength range 10-700 mm.
At wavelengths shorter than 10mm and longer than 700 mm the RCA gives
consistently lower measurements of roughness than the CAT. The differences arise for
Measurement of Long Wavelength Irregularities on Rails 647
different reasons. In the short wavelength range, the measuring wheel on the RCA
(Fig. 1) tends to smooth out short wave roughness: it essentially provides contact patch
filtering. The reason for the difference at long wavelengths is not so clear. It is also not
at all clear from these results alone whether the CAT or the RCA provides more
accurate measurements nor is it possible to provide an independent method of
demonstrating the accuracy of these measuring instruments for such long wavelengths.
An alternative method of assessing the validity of the measurements is objectively to
quantify repeatability of the different instruments. This has been done by calculating the
RMS amplitude of irregularities in different wavelength ranges for successive 50 m
sections of the site 31.30-31.85 km. For the RCA there are 22 sections in total for both
rails whereas for the CAT there are only 11 sections for the right hand rail. Measurements
on the left hand rail were made in opposite directions and are therefore not a true guide to
repeatability. The wavelength ranges considered are those in the European reprofiling
standard [10] (10-30 mm, 30-100 mm, 100-300 mm, 300-1000 mm) plus the longer
wavelength range 1000-3000 mm. If sA and sB are the RMS amplitudes of records A and B
of one instrument (either CAT or RCA) for one wavelength range and one block of 50m
length, the percentage difference in RMS values is simply,
3 Conclusions
References
[1] Galaitsis, A.G., Bender, E.K.: Wheel/rail noise – Part V: Measurement of wheel and rail
roughness. Journal of Sound and Vibration 46, 437–451 (1976)
[2] Railway applications – acoustics – measurement of noise emitted by railbound vehicles,
European Standard EN ISO 3095:2005, CEN, rue de Stassart 36, B-1050 Brussels,
Belgium (August 2005)
[3] Railway applications – noise emission – rail roughness measurement related to rolling
noise generation, European Standard EN 15610:2009, CEN, rue de Stassart 36, B-1050
Brussels, Belgium (May 2009)
[4] Berggren, E.G., Li, M.X.D., Spännar, J.: A new approach to the analysis and presentation
of vertical track geometry quality and rail roughness. Wear 265, 1488–1496 (2008)
[5] Triepaischajonsak, N., Thompson, D.J., Jones, C.J.C., Ryue, J., Priest, J.A.: Ground
vibration from trains: experimental parameters and validation of a numerical model.
Journal of Rail and Rapid Transit, Procs of ImechE 225F, 140–153 (2011)
[6] Grassie, S.L., Saxon, M.J., Smith, J.D.: Measurement of longitudinal rail irregularities and
criteria for acceptable grinding. Journal of Sound and Vibration 227, 949–964 (1999)
[7] Grassie, S.L.: Rail irregularities, corrugation and acoustic roughness: characteristics,
significance and effects of reprofiling on different types of railway system. Journal of Rail
and Rapid Transit, Procs of I Mech. E 226F, 542–557 (2012)
[8] http://www.railmeasurement.com/rail-corrugation-and-roughnes
s-measurement/rca/ (accessed at June 12, 2013)
[9] Grassie, S.L.: Measurement of railhead profiles: a comparison of different techniques.
Wear 191, 245–251 (1996)
[10] Railway applications – track – acceptance of works – Part 3: Acceptance of rail grinding,
milling and planing work in track, European Standard EN 13231-3:2006, CEN, rue de
Stassart 36, B-1050 Brussels, Belgium (May 2006)
[11] Grassie, S.L.: Corrugation on Australian National: cause, measurement and rectification.
In: 4th Intnl Heavy Haul Railways Conference, Brisbane, Instn of Engrs Australia (1989)
[12] Grassie, S.L.: Controlling irregularities in rail to reduce noise. International Railway
Journal, 38–42 (April 2010)
[13] Railway applications – noise emission – Road test of draft standard for rail roughness
measurement prEN15610:2006, CEN, rue de Stassart 36, B-1050 Brussels, Belgium
(March 2008)
[14] Jones, C.J.C., Létourneaux, F., Fodiman, P.: Testing a new rail roughness measurement
standard. In: Proceedings of Euronoise / Acoustics 2008, Paris, June 29-July 4 (2008)
Statistical Description of Wheel Roughness
1 Introduction
The role played by the combined wheel and rail roughness in generating rolling noise is
of fundamental importance [1]. Along with the mobilities of the rail, of the wheel and
of the contact, the roughness can be used in estimating the wheel-rail contact force. The
motion of a wheel rolling along a track generates a force with an amplitude that is
proportional to the roughness amplitude and at a frequency given by the ratio between
the speed of the wheel centre and the roughness wavelength. A method to measure and
analyse rail roughness is described in the standard EN15610:2009 [2]; however no
equivalent standard exists for the measurement of wheel roughness.
The EU project ACOUTRAIN is exploring the possibility of introducing some
elements of virtual testing in the acoustic certification process of new rolling stock. In
doing so, it is important to estimate the effect of uncertainty and variability. For
example, in rolling noise prediction models, variability in measured wheel roughness
will result in variability in the predictions of rolling noise. A measurement procedure
has been proposed aimed at reducing this variability [3]. Construction of a database of
measured wheel roughness is also desirable. Such a database can be used in the first
instance to estimate the statistical distribution of roughness. Moreover, it can be used to
identify the important factors influencing wheel roughness, e.g. braking system and
powered or unpowered wheels. Finally, since it is not always possible to measure all the
wheels of a train, the database could be used to estimate the number of wheels required
to be measured to obtain a representative sample for a whole train.
The paper will first describe wheel roughness data measured in the last two decades
among the ACOUTRAIN project partners in terms of statistical properties. The
question of how many wheels should be measured from a particular train is addressed
in Section 3. Finally, Section 4 is devoted to describing the results of a round robin test
performed within the ACOUTRAIN project to assess the effect on wheel roughness
measurements of adopting different equipment used independently by different teams.
Wheel roughness data are currently available from seven campaigns between 1999 and
2012 (see Table 1) and are continuously being updated. In total, data from 310 wheels
have been input into the database. The database is implemented with MATLAB
objects, and the main wheel properties are described in Table 2.
These campaigns took place over a time of 15 years and they were performed by
different people with different equipment and with different post-processing software,
thus adding a certain amount of additional variability, as will be discussed in Section 4.
As a consequence, the wavelength range is not always consistent among the campaigns;
some of them in fact do not report the very long and/or the very short wavelengths.
However, only the line corresponding to the nominal contact point (70 mm from the
flange back) is considered here.
The quantity observed is the wheel roughness level expressed in decibels as:
2
Lr=10 log10 ⎛⎜ rrms rref ⎞⎟ ; rref = 1μm (2.1)
⎝ ⎠
where rrms is the root mean square value of the measured roughness in a particular
one-third octave band. The statistical analysis is based entirely on processed data in
one-third octaves after application of suitable pre-processing such as removal of pits
and spikes [2, 3].
T-Tests [4] are used to compare different sets of data to see if the factors
characterising the set are sufficient to indicate a significant difference in the population.
The brake type is known to have a significant effect on wheel roughness; statistical tests
have confirmed this but results are not reported here. The t-test has also been performed
to compare trailer wheels with powered wheels or to compare different campaigns.
Once a set is defined, for each wavelength, a Lilliefors normality test [5] at 5 %
significance level is used to understand the likelihood of the data being normally
distributed.
To summarize the variability of measured wheel roughness data, the measurements
of each set will be presented in terms of the arithmetic mean of the levels, the energy
average (as the noise level is expected to depend on the energy average contribution of
each wheel) and of width of the distribution defined by the 15.8th and the 84.2th
percentile, corresponding, for a normal distribution, to one positive and one negative
standard deviation relative to the mean.
20 20
15 (a) 15 (b)
10 10
Lr, dB re 1 μ m
Lr, dB re 1 μ m
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
80 20 5 1.25 0.315 80 20 5 1.25 0.315
Wavelength, cm Wavelength, cm
Fig. 1. Roughness levels for disc brake wheels from campaign A. (a): powered; (b): trailer.
: [6]; ⎯⎯: energy mean; ⋅⋅⋅⋅⋅⋅⋅: sample mean; − − −:± one standard deviation corridor.
20 20
15 (a) 15 (b)
10 10
Lr, dB re 1 μ m
Lr, dB re 1 μ m
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
80 20 5 1.25 0.315 80 20 5 1.25 0.315
Wavelength, cm Wavelength, cm
Fig. 2. Roughness levels disc brake wheels. (a): all campaigns; (b): campaign A. : [6];
⎯⎯: energy mean; ⋅⋅⋅⋅⋅⋅⋅: sample mean; − − −:± one standard deviation corridor.
Fig. 3a presents average wheel roughness levels and variability for all 84 wheels with
cast-iron brake blocks. In this case not all the datasets are available over the same
wavelength range; therefore the variability appears not to be consistent, being quite low
at short wavelengths. Again, introducing more data in the analysis results in a wider
range of variability. The null hypothesis of the data being normally distributed is to be
rejected for seventeen wavelength bands out of thirty. It can be observed that the shape
of the roughness curve as a function of wavelength shows a peak at around 5 cm. In
three cases a similar behaviour can be found by considering the campaigns one at time.
However, both the normality test (null hypothesis is often to be rejected) and the t-test
between two sets defined by the campaigns suggest that it is not straightforward to
consider all the available wheels with cast-iron brake blocks as members of one single
population. Again, other factors, such as mileage or running condition, might be too
important to be discarded.
Fig. 3b presents average roughness levels for wheels with cast-iron brake blocks of
campaign G. These measurements were on a single vehicle type and consequently show
a much smaller variability.
20 20
15 (a) 15 (b)
10 10
Lr, dB re 1 μ m
Lr, dB re 1 μ m
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
80 20 5 1.25 0.315 80 20 5 1.25 0.315
Wavelength, cm Wavelength, cm
Fig. 3. Roughness levels wheels with cast-iron brakes block. (a): all campaigns; (b): campaign G.
: [6]; ⎯⎯: energy mean; ⋅⋅⋅⋅⋅⋅⋅: sample mean; − − −:± one standard deviation corridor.
656 G. Squicciarini et al.
20 20
15 (a) 15 (b)
10 10
Lr, dB re 1 μ m
Lr, dB re 1 μ m
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
80 20 5 1.25 0.315 80 20 5 1.25 0.315
Wavelength, cm Wavelength, cm
Fig. 4. Roughness level, wheels with composite brakes. (a): all campaigns; (b): campaign F.
: [6]; ⎯⎯: energy mean; ⋅⋅⋅⋅⋅⋅⋅: sample mean; − − −:± one standard deviation corridor.
n = ( zα/2σ E )2 (3.1)
where zα/2 is the ‘standard normal deviate’, or ‘standard normal random variable’ [4],
corresponding to an upper tail of α/2 (for example: z0.025=1.96 ).
Statistical Description of Wheel Roughness 657
Where the variance of the population is not known, the Student T distribution should
be used instead of the normal one and an estimator for the variance should be adopted.
However, as the Student T distribution itself is dependent on the sample size, n, a
closed form cannot be found (see e.g. [4]). Instead, the information acquired from the
previous sections can be used to estimate the value of the variance of each population
(disc brakes, cast-iron brake blocks etc.) and such an estimator can be used to replace
the actual value of σ. This can be either the upper bound of the standard deviation
confidence interval or the sample standard deviation itself.
A calculation for n should be done for each wavelength and the greatest value will
give the actual sample size. Eq. (3.1) should be applied only at those wavelengths
where data appear to be normally distributed.
To give an example, for a hypothetical campaign on wheels with disc brakes (see
Fig. 2a), by setting E = 4 dB and using the upper bound of the standard deviation
confidence interval (6.5 dB) the number of wheels to be measured should be 10, in
accordance with [7]. Adopting the lower bound of the standard deviation (4.9 dB)
would result in 6 wheels to be measured.
-5 -5
-10 -10
-15 -15
80 20 5 1.25 0.315 80 20 5 1.25 0.315
Wavelength, cm Wavelength, cm
Fig. 5. Wheel roughness from the ACOUTRAIN project test. (a): cast-iron brakes; (b):
composite brakes. ⎯⎯: [6]; ○: P1; □: P2; +: P3; ×: P4; ◊: P5.
Measurements made by the five teams (P1 to P5) are shown in Fig. 5a for the vehicle
with cast-iron brakes and in Fig. 5b for the vehicle with composite brakes. Lines P1, P2
and P5 agree, apart from two one-third octave bands, within 2.5 dB in the entire range
for both the brake types. Measurements labelled as P3 and P4, to differing extents,
represent anomalous values, indicating either a malfunction of the equipment and/or
incorrect data post-processing.
658 G. Squicciarini et al.
5 Conclusions
With the aim of estimating the effect of uncertainty and variability in preparing input
data for rolling noise models, this paper describes, in terms of the average and standard
deviation, wheel roughness data for seven different campaigns collected in a database.
By addressing their statistical distribution, considering one campaign at a time, it was
shown to be likely that the roughness level is normally distributed. If all the available
campaigns are analysed at once, however, the likelihood of data being normally
distributed decreases. However, the likelihood is expected to increase again by
introducing into the database more measurements, although, as a side effect, the
estimated deviation from the average would increase.
A simple method to estimate the minimum number of wheels necessary to be
measured to represent, with enough confidence, the whole train has been addressed.
The method has been shown to give results in agreement with previous research
projects [7].
Finally, a round robin test performed within the ACOUTRAIN project, has shown
that, in the particular case considered, the amount of measurement variability due to
different equipment used by different teams following the same protocol can be
estimated to be within 2.5 dB, provided that no major equipment malfunctions or
post-processing errors occur.
References
[1] Thompson, D.J.: Railway noise and vibration: Mechanisms, modelling and means of
control, 1st edn. Elsevier Science (2009)
[2] European Standard EN 15610:2009
[3] Thompson, D.J.: Proposed analysis method for wheel roughness. Deliverable 2.4 for
ACOUTRAIN project (2013)
[4] Montgomery, D.C.: Design and analysis of experiments, 8th edn. John Wiley & Sons (2012)
[5] Lilliefors, H.W.: On the Kolmogorov-Smirnov test for normality with mean and variance
unknown. Journal of the American Statistical Association 62(318), 399–402 (1967)
[6] European Standard EN ISO 3095:2013
[7] Fodiman, P.: NOEMIE Project Final report (Project 2002/EU/1663), AEIF, see Appendix D
(July 2005)
Rail Corrugation Growth on Curves – Measurements,
Modelling and Mitigation
1 Introduction
“Curves are a critical area of any railway system and it is best to avoid them
completely” [1].
Considering the demand for expensive maintenance of railway curves [2],
infrastructure managers worldwide would probably approve of the above statement by
Grassie and Elkins. The large magnitude wheel–rail contact forces (and associated
sliding of the wheel–rail contacts) created when railway bogies negotiate curves may
introduce both rolling contact fatigue (e.g. head checks and shelling of the high rail)
and wear (e.g. on the gauge face and crown of the high and low rails, respectively).
Complaints regarding high noise levels are regularly received by Stockholm Public
Transport (SL) from passengers and from people living close to their track network [3].
To manage the corrugation problem, SL is forced to run regular and expensive rail
grinding programmes. This has no potential for preventing the corrugation from
reappearing and obviously does not offer a satisfying long-term solution to the
problem. In his PhD dissertation [4], the first author studied the development of
corrugation on small radius curves by mathematical modelling, numerical simulations,
field measurements and laboratory investigations. A 120 m radius curve on the metro
of SL experiencing severe corrugation growth was used as a reference. A summary of
the thesis is outlined in the current paper.
2 Measurements
In the studied curve corrugation develops exclusively on the low rail. The track consists
of continuously welded BV50 rails (50 kg/m and steel grade 350 HT) with inclination
1:40 on ballast subgrade. The rails are mounted, via rail pads, to monobloc concrete
sleepers separated by a distance of 600 mm. The track gauge is 1435 mm and
super-elevation is 90 mm. To reduce the gauge face wear on the high rail, in-house
designed passive lubricators are applied at the entrance of the curve. The traffic consists
of a single fleet of trains with an estimated axle load of less than 12.5 tonnes (loaded
condition). The wheel profile is S1002. Approximately 80 000 trains per year pass the
curve in each direction.
The development of corrugation on the low rail has been monitored during a
grinding interval of one year. Roughness was measured with a corrugation analysis
trolley (CAT) [5] in the following approximate number of months after rail grinding:
0-, 0+, 2, 3, 5, 10, 12 (0- and 0+ denote directly before and after rail grinding,
respectively).
Roughness level spectra in 1/3 octave bands from the northbound track are presented
in Fig. 1. The measurement of the fully grown corrugation from the previous grinding
interval shows short-pitch corrugation in the wavelength interval 4 - 14 cm with peaks
at approximately 5 cm and 8 cm. Only a moderate additional growth of roughness is
observed after the measurement 291 days after grinding.
Fig. 1. Roughness level spectra in 1/3 octave bands based on measurements on the low rail of
the northbound track. Dates (year, month, day) and number of days after grinding when the
measurements were performed are listed in the legend [6].
Rail Corrugation Growth on Curves – Measurements, Modelling and Mitigation 661
Fig. 2. The surface irregularity of a 60 cm long section cut from the low rail of the northbound
track as measured with a coordinate measuring machine (bottom). On the top, micro-structures
are presented for longitudinal and lateral cuts of the rail crown [6].
2.2 In Laboratory
Fig. 2 shows results from a laboratory investigation based on a piece of rail cut from the
low rail of the curve. The topology measured with a coordinate measuring machine
clearly shows the dominant corrugation wavelength of approximately 5 cm. By
studying the microstructure of the rail surface layer, material flow was found in the
lateral direction oriented towards the field side. No visible differences were found when
comparing the microstructures of corrugation troughs and peaks.
3 Modelling
(a) (b)
Fig. 3. Curving diagrams illustrating bogie 21 in a C20 trainset under steady-state curving.
Similar curve geometry (radius 120 m) and running conditions (vehicle speed 32 km/h) as in the
studied curve of SL. Resultant creep forces and their longitudinal and lateral components in the
track plane are outlined at the locations of the wheels and rails, respectively. All displayed forces
are acting on the wheels. (a) Friction coefficient 0.6, (b) friction coefficient 0.3 [4].
The curving diagrams in Fig. 3 indicate the lateral plastic flow observed in Fig. 2 to
be caused by the large magnitude lateral creep force developed at the low rail contact of
the leading wheelset.
A time-domain model for the prediction of long-term growth of rail roughness on small
radius curves has been developed [7]. The dynamic interaction between a full three-unit
C20 trainset and a curved track is simulated in a broad frequency range. Non-Hertzian
and non-steady effects in the wheel–rail contact are accounted for using Kalker’s exact
theory [8].
During negotiation of the curve, the amount of wear generated by the low rail
contact of the leading wheelset exceeds that of the trailing wheelset [6]. Further, the
predicted generation of wear shows peak magnitudes at wavelengths in close
agreement with observations in the field. The dynamic behaviour of the vehicle–track
system at the corresponding frequencies defines the so-called wavelength-fixing
mechanisms. Their identification is essential to reach a comprehensive understanding
of the causes of corrugation growth on the curve.
Fig. 4 illustrates the wavelength-fixing mechanism associated with the dominant
corrugation wavelength of 5 cm as primarily determined by the first antisymmetric
bending eigenmode of the leading wheelset. A stick-slip oscillation occurs in the
contact area (see the distribution of stick and slip presented for three time instances).
The wavelength-fixing mechanism is primarily influenced by the structural flexibility
of the wheelset. The wavelength of the generated wear decreases with increasing
bending stiffness of the wheel axle (which would increase the eigenfrequencies of the
bending modes) [6].
Rail Corrugation Growth on Curves – Measurements, Modelling and Mitigation 663
Model validation is performed versus field measurements [6]. Friction coefficient 0.6
(dry conditions) and vehicle speed 30 km/h are assumed for all train passages. The
parameters in an Archard wear model [9], the non-dimensional wear coefficient kw and
the hardness of the rail material H are taken as 1 · 10-4 and 3.4 GPa, respectively. In an
iteration scheme, the wear depth caused by one wheel passage is multiplied with a
magnification factor of 10 000. In Fig. 5(a), roughness level spectra in 1/3 octave bands
based on calculated results, using the roughness level for a newly ground rail as input, and
measurements on the fully grown corrugation on the northbound track are compared.
Both the predictions and measurements show corrugation growth at wavelengths 5 cm
and 8 cm. For these wavelengths the assumption of uniform speed for all passing trains
makes the predicted peaks too distinct. The increasing roughness level at around 17 cm is
caused by the strong (non-realistic) excitation of the P2-resonance created when the train
enters the updated rail irregularity and should be disregarded.
For corrugation wavelengths of 5 cm and 8 cm, the predicted roughness magnitudes
after 30 000 and 40 000 wheel passes are overlapping and hence the corrugation has
stopped to grow. This is due to a gradual decrease in phase between the calculated wear
and the present rail irregularity for an increasing number of wheel passages [7].
Additionally, this phase difference causes the predicted development of roughness to
occur with a simultaneous translation of the corrugation profile. In Fig. 5(b), the
predicted roughness growth for a single wavelength initial irregularity on the low rail of
the curve is shown. After a substantial initial growth of roughness, the rail irregularity
moves backwards with almost constant amplitude. Field observations verify that
corrugation eventually develops a constant amplitude, see Fig. 1 and [3]. However, the
verification of the suggested longitudinal translation of the corrugation profile is an
important part of future work.
664 P.T. Torstensson and J.C.O. Nielsen
(a) (b)
Fig. 5. (a) Comparison between measured and predicted roughness levels in 1/3 octave bands on
the low rail of the studied curve. Curves with markers show predicted results for 10 000, 20 000,
30 000 and 40 000 wheel passages. The solid line shows measurement 111130 on the northbound
track, cf. Fig. 1. Initial roughness of magnitude according to measurement on newly ground rail.
Friction coefficient 0.6 and vehicle speed 30 km/h [6]. (b) Development of rail roughness after
100, 200, …, 1200 wheel passages on an initial single wavelength irregularity of wavelength
3.8 cm. Results shown for one single wavelength. Curve radius 120 m, vehicle speed 25 km/h
and friction coefficient 0.6 [7].
4 Mitigation
(a) (b)
Fig. 6. (a) Roughness level spectra in 1/3 octave bands measured on the low rail of the studied
curve approximately every second month during one year. The solid curve shows the full grown
corrugation of the previous grinding interval and curves with markers show results during the
studied grinding interval [6]. (b) Magnitude (1/24 octave bands) of transfer function between
calculated accumulated wear depth and initial rail irregularity after one wheel passage.
Steady-state equilibrium curving of bogie 21 in a C20 trainset negotiating curves of different
radii. : v = 30 km/h, R = 100 m, : v = 32.5 km/h, R = 125 m, : v = 35 km/h, R = 150 m,
: v = 40 km/h, R = 200 m. Results marked with circles are calculated for the corresponding
vehicle speeds and curve radii but for a bogie having a soft primary suspension (results are partly
overlapping) [4].
rail irregularity is used. For steady-state curving, the magnitude of the transfer function
H is observed to decrease with increasing curve radius, see Fig. 6(b). In particular, for
curve radii above 150 m the wavelength-fixing mechanism associated with the 5 cm
corrugation wavelength (corresponding to about 180 Hz) is no longer observed. The
same effect is obtained if the bogie steering is improved. For longitudinal primary
suspension stiffness reduced by multiplication factor 0.01, the wavelength-fixing
mechanism does not appear in the entire interval of studied curve radii.
5 Concluding Remarks
range (up to at least 200 Hz) and captures non-linear phenomena such as the limiting
amplitude of corrugation growth. Calculations show the dominant wavelength-fixing
mechanism observed in field to be primarily determined by the first antisymmetric
bending eigenmode of the leading wheelset.
The application of a top-of-rail friction modifier was found to effectively control the
corrugation growth on the curve, and numerical predictions show the positive effect of
improved steering of bogies.
References
[1] Grassie, S.L., Elkins, J.A.: Traction and curving behaviour of a railway bogie. Vehicle
System Dynamics 44(1), 883–891 (2006)
[2] Lewis, R., Olofsson, U.: Wheel-rail interface handbook, p. 842. Woodhead Publishing
Limited, Cambridge (2009)
[3] Torstensson, P.T., Nielsen, J.C.O.: Monitoring of rail corrugation growth due to irregular
wear on a railway metro curve. Wear 267(1-4), 556–561 (2009)
[4] Torstensson, P.T.: Rail corrugation growth on curves. Ph. D. Dissertation, Department of
Applied Mechanics, Chalmers University of Technology, Göteborg, Sweden (2012)
[5] Grassie, S.L., et al.: Measurements of longitudinal rail irregularities and criteria for
acceptable grinding. Journal of Sound and Vibration 227(5), 949–964 (1999)
[6] Torstensson, P.T., Schilke, M.: Rail corrugation growth on small radius curves –
Measurements and validation of a numerical prediction model. Wear 303(1-2), 381–396
(2013)
[7] Torstensson, P.T., et al.: Simulation of rail roughness growth on small radius curves using a
non-Hertzian and non-steady wheel–rail contact model. Wear (2013),
doi:10.1016/j.wear.2013.11.032
[8] Kalker, J.J.: Three-dimensional elastic bodies in rolling contact, p. 314. Kluwer Academic
Publishers, Dordrecht (1990)
[9] Archard, J.F.: Contact and rubbing of flat surfaces. Journal of Applied Physics 24, 981–988
(1953)
[10] Grassie, S.L., Kalousek, J.: Rail corrugation: characteristics, causes and treatments.
Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid
Transit 207(16), 57–68 (1993)
Effects of Track Stiffness and Tuned Rail Damper
on Rail Roughness Growth and Rail Vibration Levels
on Metro System
Summary. This paper describes track vibration levels and rail roughness growth
measurements carried out on a number of tracks with different values of track
stiffness and rail damping at a number of metro systems in China. At each test
site, measurements of rail surface roughness were repeated several times over a
period of time. The results provide useful data on the progress of rail roughness
growth, and of change in track vibration levels in the medium term. Rail
roughness increased noticeably on the stiffer track, whereas it decreased on the
track with tuned dampers attached to the rail. Comparing with the standard
baseplate track, the vibration levels measured on the rails in both vertical and
lateral directions are all lower on the track with lower stiffness support and tuned
vibration dampers.
1 Introduction
Fast growth of city population as the result of urbanisation has increased the demand
for reliable public transportation systems. Metro systems are constructed in many cities
in China and the total length is growing rapidly. One of the important issues associated
with this development is the environmental protection for residential, commercial,
educational, and hospital developments, i.e., control of noise and vibration levels
caused by trains running on tracks in order to satisfy environmental requirements.
There are two main noise and vibration problems associated with running trains on
tracks. The first is transmission of ground borne vibration into structures around the
track. The second is airborne noise from the track and vehicle.
Track support systems designed to control ground vibration from railways do so by
offering low dynamic track stiffness. Generally speaking, a low stiffness track will
have a low frequency above which effective ground vibration reduction is achieved,
resulting in better overall performance. However, lower track stiffness may cause
higher rail vibration and rolling noise [1].
Rolling noise is important at low speeds of most metro systems. Rolling noise from
trains is radiated by the train wheels and the rails, driven by dynamic forces arising at
the contact between the wheel and the rail. One of the most important sources of this
dynamic excitation is the roughness [2, 3] and the rolling impact contact force which is
referred to the dynamic wheel-rail contact patch and its time history at a fix point on the
rail can be approached as a half sine wave, particularly on the curved tracks.
Within the vehicle-track system, three components radiate rolling noise: These are
the slab (which contributes most at low frequencies), the rail (which is important over a
broad range of frequencies, between say 200 Hz and 2 kHz), and the wheel (which is
most significant at the higher frequencies). For the metro systems at low speeds, the
overall noise at the low and middle ranges of frequencies is the major contributor to the
environmental noise problem. To control both ground-borne vibration and airborne
noise, it is essential to understand the influence of track stiffness on track dynamic
behavior, and the tuned vibration damper applied to the rail to control vibration and rail
roughness growth on metro track.
Rail vibration and rail roughness growth measurements were carried out on different
tracks with different values of track stiffness and rail damping in Chengdu and Nanjing
metros in China. These include six trackforms, including standard baseplate, Cologne
Egg baseplate, two layers resilient baseplate, “floating rail” fastener, tuned damper
fastener and floating slab track with steel springs. Comparisons of the rail vibration and
roughness growth on different tracks have also demonstrated the performance of direct
fixation fasteners with integral tuned vibration dampers.
2 Rail Corrugation
π EI ⎧ 1 π rg 2 2(1 + v) ⎫ (2.1)
f = ⎨1 − ( ) (1 + )⎬
2l 2 mr ⎩ 2 l K ⎭
Effects of Track Stiffness and Tuned Rail Damper on Rail Roughness Growth 669
where E is the modulus of elasticity of rail steel, I is the moment of inertia of the rail,
mr is the rail mass per unit length, l is the fastener spacing, rg is the radius of gyration,
ν is Poisson’s ratio, and K(≈ 0.34) is the shear constant of the cross section.
Fig. 1 shows typical rail corrugation on metro track. The wavelength and its amplitude
are both important parameters that define the rail corrugation level. Therefore the
corrugation excitation frequency related to the running train can be calculated if the train
speed is known, i.e.,
fc = s / λ (2.2)
Rail roughness and vibration measurements were carried out at two metro systems
and six different trackforms. The train speed range is 45-65 km/h. The details are listed
in Table 1, where the dynamic stiffness of rail fastener is also listed in the table.
670 A. Wang et al.
To compare the effect of rail roughness on airborne noise, the A-weighted roughness
levels for the main frequency range between 160 Hz and 500 Hz are presented for
equivalent wavelengths of 20-80 mm at the train speed of 45-65 km/h on the metro
systems.
The A-weighted roughness levels for the measurements at four sites on Chengdu
Metro and Nanjing Metro are listed in Table 2 and shown in Fig. 3 and Fig. 4
respectively.
20 20
Roughness dB(A) (ref 1E-6m)
10 10
5 5
U - Up Track
D - Down Track
0 0
PV Egg Std TDF
Fig. 3. Rail roughness on Chengdu Metro Fig. 4. Rail roughness on Nanjing Metro
Effects of Track Stiffness and Tuned Rail Damper on Rail Roughness Growth 671
On the whole, the rail roughness on the Nanjing Metro is lower than that on Chengdu
Metro. As shown in Fig. 3, on the Chengdu Metro system the roughness levels on the
stiffer track were generally higher than those on the soft track. Roughness on the
floating slab track is higher than that on others. Tracks with TDF demonstrated a good
performance with the lowest rail roughness.
20
-10
Up-Jul.12 Down-Jul.12
-20 Up-Mar.13 Down-Mar.13
-30
630 315 160 80 40 20 10
Wavelength (mm)
Fig. 5. Tuned damper fastener (TDF) Fig. 6. Roughness with Std baseplate on Chengdu Metro
Tuned rail vibration dampers can be combined with the rail fastener, such as the rail
suspension fastener. The tuned mass elements were bonded to the elastomeric support
block to form a “mass-spring” tuned damper. The frequencies were tuned to the same
range as the rail pinned-pinned frequencies for the track.
-10 -10
Up-Jul.12 Down-Jul.12
-20 -20
Up-Mar.13 Down-Mar.13
-30 -30
630 315 160 80 40 20 10 630 315 160 80 40 20 10
Wavelength (mm) Wavelength (mm)
Fig. 7. Roughness growth with FST on Fig. 8. Roughness growth with tuned damper
Chengdu Metro on Nanjing Metro
Data collections were made for three different trackforms at two sites on Chengdu
Metro Line 1. At site one the existing slab track was installed with standard rail
baseplates that were replaced with tuned damper fasteners on August 2012. Site two is
the steel spring FST on the same line with the same curvature. Vibration accelerations
were recorded for more than 30 train passes during the rush hours for each trackform
and averaged to obtain the mean level.
Rail vibration velocity levels are shown in Fig. 9 for the vertical direction and Fig. 10
for the lateral direction. The rail vibration on the track with tuned damper fastener is
lower than that with standard baseplate and significant lower than that with the FST in
both directions.
Taking the track with standard baseplate as a reference track, the total rail vibration
velocity level on FST is 3.4 dB(A) higher in the vertical direction and 7.4 dB(A) in the
lateral direction. However, on the track with tuned damper fasteners, it is slightly
lower, 1.5 dB(A) in the vertical direction and 0.8 dB(A) in the lateral direction. This
shows that the lower stiffness of tuned damper fastener is able to not only significantly
reduce the ground vibration but also to reduce the rail vibration in a broad range of
frequencies in both vertical and lateral directions.
110
Vibration velocity dB(A) (ref 5E-8m/s)
110
Vibration velocity dB(A) (ref 5E-8m/s)
Std_Vertical Std_Lateral
100 100
TDF_Vertical TDF_Lateral
90 FST_Vertical 90
FST_Lateral
80 80
70 70
60 60
50 50
40 80 160 315 630 1250 2500 40 80 160 315 630 1250 2500
1/3 Octave Central Frequency (Hz) 1/3 Octave Central Frequency(Hz)
Fig. 9. Rail vertical A-weighted velocity Fig. 10. Rail lateral A-weighted velocity
674 A. Wang et al.
5 Conclusions
Rail roughness and vibration measurements on different tracks with different values of
track stiffness and rail damping in a number of metro systems in China were carried
out. The results provide useful data on the evolution of rail roughness growth rate and
on changes in track vibration levels in the medium term.
The rail roughness levels on the stiff track were generally higher than those levels on
soft track. Tracks with tuned damper fasteners (TDF) demonstrated good performance
with the lowest rail roughness growth rate. Compared to standard baseplate track over a
period of eight months, the roughness growth rate on the track with tuned damper
fasteners was about 5 dB(A) lower; whilst the roughness growth rate on the floating
slab track was about 2-3 dB(A) higher.
Compared to the standard baseplate track, the vibration velocity levels measured on
the rail were lower on the track with lower support stiffness and tuned damper attached
by about 1.5 dB(A) in the vertical direction and 0.8 dB(A) in the lateral direction. On
the FST, the total rail vibration velocity level was 3.4 dB(A) higher in the vertical
direction and 7.4 dB(A) in the lateral direction compared with the standard baseplate
track.
References
[1] Vincent, N., Bouvet, P., Thompson, D.J., Gautier, P.E.: Theoretical optimization of track
components to reduce rolling noise. Journal of Sound and Vibration 193(1), 161–171 (1996)
[2] Wang, A., Cox, S.J.: Noise characteristics of high speed track with railpads of different
stiffness. In: Proceedings of EuroNoise 1998, Munich, Germany, pp. 295–300 (1998)
[3] Wang, A., Cox, S.J.: Effect of railpad stiffness on rail roughness growth and wayside noise
levels on high speed track. In: The 6th World Congress on Railway Research, Edinburgh,
UK (2003)
[4] Grassie, S.L.: Corrugation: variation on an enigma. Railway Gazette International 146(7),
531–533 (1990)
[5] Grassie, S.L., Kalousek, J.: Rail corrugation: characteristics, causes and treatments. Journal
of Rail Rapid Transit, Proceeding of Institute of Mechanical Engineering 207F, 57–68
(1993)
[6] Ilias, H.: The influence of railpad stiffness on wheelset/track interaction and corrugation
growth. Journal of Sound and Vibration 227, 935–948 (1999)
[7] Grassie, S.L., Cox, S.J.: The dynamic response of railway track with flexible sleepers to
high frequency vertical excitation. Proceedings of Institute of Mechanical
Engineering 198D(7), 117–124 (1984)
[8] Grassie, S.L., Edwards, J.W., Jones, C.J.C., Verheijen, E.: Benchmark testing of equipment
to measure rail roughness. In: Proceeding of the 8th International Workshop on Railway
Noise, Buxton, UK, pp. 273–284 (2004)
[9] Wu, T.: Effects on short pitch rail corrugation growth of a rail vibration absorber/damper.
Wear 271, 339–348 (2011)
Prediction of Acoustical Wall Pressure Levels
of Rolling Stock Vehicles
ALSTOM Transport,
Rue Albert Dhalenne 48, 93482 Saint Ouen, France
andrea.bistagnino@transport.alstom.com
Summary. This paper presents a prediction of the acoustic wall pressure levels
on a rolling stock vehicle from wheel-track interaction. This study has taken into
account two different approaches: beam-tracing and boundary element
calculations. The two types of calculations are benchmarked against results
measured on an existing train.
1 Introduction
Passenger and crew comfort is an important requirement in essentially all rolling stock
projects. Noise is one of the most important parameters to define the train’s comfort
and its prediction is a common task of all rolling stock manufacturers.
Different strategies for airborne interior noise predictions can be employed, for
example Statistical Energy Analysis [1] or beam-tracing techniques [2]. A key input for
these calculations is the acoustic wall pressure around the train [3]. One typical way to
obtain the acoustic wall pressure loads around the train is to create an experimental
database for different type of trains. This approach presents some drawbacks; mainly
due to train availabilities for tests and complicated test setups, the cost of building such
a database is very high. Moreover, the architectural particularities of single projects
cannot be taken into account, resulting in a reduced reliability on the results that can be
obtained.
To overcome these limitations, it is possible to replace the expensive field tests with
calculations. In this study we compare the results obtained with two computation
techniques (beam-tracing and Fast Multipole Method BEM) with experimental values
measured on an existing train.
Beam-tracing techniques have been applied for years to predict noise levels, thanks to
their good precision and relatively simple setup. Recent developments have included
the possibility to include diffraction effects in the computations, which (albeit
computationally quite expensive) expands the range of possible applications, including
wall pressure prediction.
Boundary Element Method (BEM) is a very well-known numerical method and can
be used to predict the diffracted sound levels around a train. However, the
computational resources needed to solve a conventional BEM model for a full vehicle
could limit its use just for low frequencies. One option to extend the computation to
high frequencies is to use the FMM BEM [4, 5]. With this approach, both the required
memory and the CPU time are greatly reduced.
In this study, the results and resources needed to compute the train wall pressure are
compared for the two approaches and, in this particular application to rolling stock, the
emissions of a set of sources placed close to the train’s wheels were analyzed. The
sound pressure level from these sources was predicted in a number of positions in the
bogie area, in the floor area away from the bogie and in the sidewalls.
Graphical representations of the calculations are shown in Fig. 1.
Fig. 1. Views of the three-dimensional models of the two computations, with sources and
receivers. On the left, a visualization of the beam-tracing calculation with train’s model: one
source (on the wheel), one receiver (on the far right) and one of the many paths connecting the
two (ground not shown). On the right, the FMM BEM model with two sources on the wheels and
several reception points in the bogie and on the sidewalls.
3 Experimental Setup
The main objective of this study was the comparison of the two calculation techniques
described in Section 2 with experimental results; the test setup was dedicated to
wheel/rail interaction noise, as it is the most prominent noise source in most rolling
stock applications.
The tests were conducted on an Alstom-built regional train, in free-field conditions.
With the train at standstill, loudspeakers were placed on the ground on both sides of the
wheels. They were set up to emit a given random noise; their sound power level was
characterized by intensity measurement. The sound pressure level in several positions
around the train was then measured, including the train bogie area, underframe and
sidewalls. Pictures of the setup are shown in Fig. 2.
Prediction of Acoustical Wall Pressure Levels of Rolling Stock Vehicles 677
Fig. 2. Views of the experimental setup, with loudspeakers placed close to the wheels and several
microphones placed on the train’s walls
4 Calculation Details
The comparison between experimental and calculated results is done in this paper on
the basis of the sound pressure level predicted around the train, having simulated the
sources with their measured sound power level. Thus, this approach suffers from
uncertainty in the value of the sources’ sound power level, but allows for a
straightforward comparison with experimental results, and follows the process that
would be used in the real-life acoustical engineering of a rolling stock vehicle.
Four sources, one on each side of two wheels of the same bogie, were modeled as
omnidirectional point sources. This modelization does not thus take into account the
directivity of the loudspeakers used for the tests.
Four reception points have been chosen for the comparison, located in the three main
areas of interest:
• Underframe - the receiver is placed at 4.1 m from the first bogie axle, at 2.6 cm
from the underframe, along the train centerline (beam-tracing only).
• Bogie - the receiver is placed above the second bogie axle, at 5 cm from the
underframe, along the train centerline (beam-tracing only).
• Sidewall - two receivers are placed above the second bogie axle, at 5 cm from the
sidewall, at 1 and 2.7 meters from the ground.
In beam-tracing calculations, the absorption of the ground was set to standard
ballast, namely the one measured in [6]. In FMM BEM calculations the ground was set
to fully reflective conditions, to reduce the computation time. The results obtained with
FMM BEM below the train were found to be overestimated, as it was expected, and are
not shown here. On the other hand, the impact of the ground absorption on the results
on the sidewall was found to be low, so, at these points, results of both measurements
can be compared with experiments.
In both cases, all of the train’s surfaces are set to be fully reflective.
The convergence of the acoustical calculations has been checked by computing the
results with an increasing number of reflections, until stabilization of the results is
obtained for all frequency bands.
Diffraction edges have been placed on the bogie frame, on the junction between the
underframe and sidewall, and on wheels, as shown in Fig. 3.
Fig. 3. Views of the beam-tracing model, with diffraction edges marked as white lines
The kD limitation is a key parameter because by increasing the BEM domain the
maximum allowable frequency decreases. For a train application where the domain
normally is large, if the objective is to evaluate a complete wall pressure the maximum
frequency will be limited. An example to visualize this limitation is shown in Fig. 4.
The maximum frequency for the BEM domain (complete vehicle) is 800 Hz, while with
a smaller domain figure it is possible to solve up to 3000 Hz. When the ground is added
to this second, smaller, domain, the upper limiting frequency is about 2000 Hz. This
latter option was finally chosen for the calculations shown in this study, as a
compromise between the frequency range of the beam-tracing calculations and the need
for a sufficiently large model. The average length of the elements is 25 mm in order to
have at least 6 elements per wave length. An infinite, fully reflective plane has been
added 500 mm below the rail level simulating the ground; the fluid has been defined as
air.
a) b)
Fig. 4. BEM domain corresponding to a) complete train vehicle; b) part of a vehicle
BEM calculations can then be used to calculate the acoustical load over the whole
vehicle portion, as shown in Fig. 5.
One of the limitations of the BEM calculations is the computational resources
required. FMM methods reduce this impact, but still require two days of calculations on
a 4-cores machine (CPU Intel @5580) with 48 GB of RAM. Preliminary tests with a
ground with homogenous impedance characteristics show much higher computation
times, making it difficult to apply industrially with current resources.
Fig. 5. Wall pressure results calculated by FMM BEM for two 1/3 octave bands
680 A. Bistagnino, A. Vallespín, and J. Sapena
Fig. 6. Comparison of experimental and beam-tracing sound pressure levels in the bogie position
(left) and in the underframe position (right). dB(A) values between 200 Hz and 3150 Hz.
Fig. 7. Comparison of experimental, beam-tracing and FMM BEM sound pressure levels in the
sidewall at 1 meter height (left) and at 2.7 meters height (right). dB(A) values between 200 Hz
and 3150 Hz, limited to 2000 Hz for BEM results.
Prediction of Acoustical Wall Pressure Levels of Rolling Stock Vehicles 681
The average absolute difference between the calculated and measured results can
again be calculated. For the lower receiver, the average differences are 2.5 dB for the
beam-tracing results and 3.1 dB for the FMM BEM results. For the upper receiver, the
average differences are 3.1 dB for the beam-tracing results and 4.3 dB for the FMM
BMM results.
For the upper receiver at low frequencies, the beam-tracing results are below the
reference ones, while the BEM results are above them. This is believed to be due
to an insufficient diffraction effect being taken into account (notably from the
bogie’s axle) in the beam-tracing case, and due to the ground reflection in the BEM
case.
In general the two calculations give fairly consistent results and both appear to be
able to correctly predict wall pressures. A number of improvements are under test to
further improve the predictions. The main ones are:
6 Conclusions
In this study we compare beam-tracing and FMM BEM approaches in their capability
to give industry-ready results for the prediction of wall pressures for rolling stock
vehicles. From a purely acoustical standpoint, the two approaches give consistent
results, and have the potential to replace experiments. Fine tuning of these methods will
deliver results of a quality acceptable for use in interior noise predictions. From the
point of view of applicability, the beam-tracing approach is more beneficial in at least
three areas:
References
[1] Lyon, R.H.: Statistical energy analysis of dynamical systems: theory and applications. MIT
Press, Cambridge (1975)
[2] Tsingos, N., Funkhouser, T., Ngan, A., Carlbom, I.: Modeling acoustics in virtual
environments using the uniform theory of diffraction. In: Proceedings of the 28th Annual
Conference on Computer Graphics and Interactive Techniques (SIGGRAPH 2001),
pp. 545–552 (2001)
[3] Gabet, P., Drobecq, V., Noé, N., Jean, P.: Prediction of acoustic comfort of a trainset using
ray-tracing technology. In: Proceedings of SIA (2009)
[4] Cordioli, J., Müller, S., Conelly, T.: SEA prediction of vehicle interior noise using exterior
loading computed by fast multipole BEM. In: Proceedings of Internoise 2009 (2009)
[5] Müller, S., Cotoni, V., Conelly, T.: Use of fast multipole BEM to calculate exterior acoustic
loads in SEA models. In: Proceedings of Internoise 2009 (2009)
[6] Broadbent, R., Thompson, D., Jones, C.: The acoustic properties of railway ballast. In:
Proceedings of Euronoise 2009 (2009)
[7] Noé, N., Gaudaire, F., Jean, P., Vermet, M.: A general ray-tracing solution to reflection on
curved surfaces and diffraction by their bounding edges. In: Proceedings of ICTCA
Conference (2009)
[8] Vermet, M., Noé, N.: Using asymptotic methods to compute diffracted pressure by curved
surfaces. In: Proceedings of ICTCA Conference (2009)
[9] VAOne 2012.5 User’s Guide. ESI Group (2013)
Study on Effective Sound Barriers for High Speed Trains
Summary. In order to reduce noise along railway lines, sound barriers have been
widely used. However, the effects of these barriers on railway noise have not
been understood thoroughly. In this paper, the relative acoustical performances
of a number of sound barriers in typical Shinkansen railway situations are first
investigated through laboratory measurements using scale models. In the scale
model studies, the effects of various types and shapes of the sound barriers on
Shinkansen noise are investigated. The results show that the most effective
barrier is Y-profile type. This is because the Y-profile barrier has noticeable
effects on both mitigating the noise increase due to the multiple reflections
of sound between the barrier and the vehicle surface and widening the shadow
zone behind the barrier because of double diffractive effects. Secondly, the
effect of the Y-profile barriers is examined by field tests with running
Shinkansen vehicles. The results show similar trends to those in the laboratory
measurements.
1 Introduction
Sound barriers are devices to prevent sound from propagating directly from a noise
source on a train to the receiver, and have been widely installed along Shinkansen
railway lines in Japan. The insertion loss of the barrier is estimated mostly by using a
Maekawa chart for a point source [1] or a Yamashita-Koyasu chart for a line source [2].
Both charts apply to a semi-infinite barrier in free space. However, for railway noise,
the insertion loss of the barriers cannot be evaluated directly by using the charts. This is
because, especially for the noise generated from the lower part of cars, multiple
reflections of sound occur between the surface of a car and the barrier. This is also due
to the fact that Shinkansen noise consists of various noise sources, e.g. rolling noise,
aerodynamic noise (pantograph, unevenness on car surface), and so on. As the insertion
loss of each source depends on its location, it is complicated to understand the effect of
the sound barrier on total Shinkansen noise. Furthermore, new barriers with novel
configurations, e.g. Y-profile barrier and T-profile barrier, have recently been
developed and, in some cases, absorptive materials have been installed on the trackside
of the barriers. However, the attenuation effects of these barriers have not been studied
quantitatively and thoroughly.
(d) inward-
(c) inverted-L (e) inward-
(a) straight (b) thick-straight inclined-α
inclined-β
30̊
45̊ 60̊ 30̊ 45̊
(f) downward- (g) downward- (h) downward- (i) outward- (j) outward-
inclined-α inclined-β inclined-γ* inclined-α inclined-β
30̊ 45̊ 45̊
2m
(k) Y-profile-α (m) inward- (n) inverted-L
(l) Y-profile-β (o) T-profile
inclined plus plus straight*
straight*
30̊ 30̊
45̊ 45̊
(a)
(a) (b)
Fig. 4. Distribution in terms of noise level for the straight type sound barrier. (a) Lower part
noise, (b) pantograph noise.
Table 1 shows the relative noise level of each barrier compared with the results
corresponding to the straight-type barrier with no absorptive materials. The estimations
were made at the position at 25 m away from the adjacent track and 6 m below rail level
(hereafter, “standard point”). In Table 1, a positive value means that it is more effective
than the straight-type barrier. The results are summarized as follows:
(a) Effects on the lower part noise:
The most effective barrier is Y-profile type, although the T-profile, inward
inclined-profile and inverted-L type barriers are also effective. These barriers have a
noise reduction effect of more than 3 dB at the standard point. This is because these
barriers have double diffractive effects for the lower part noise.
Study on Effective Sound Barriers for High Speed Trains 687
45 ■:Y-profile
Y-type
(c) Effects of the absorptive materials:
□:Y-profile
Y-type with
with acoustic
acoustic tubes
tubes
The wider the area of the materials fitted on 40
the barrier is, the greater the noise reduction 125 250 500 1k 2k 4k AP
Octave band center frequency (Hz)
is (not shown due to limitations of space).
This is because the materials mitigate the Fig. 5. Effect of the acoustic tube arrays
multiple reflections effectively. on lower part noise
Table 1. Relative noise level in A-weighted levels at the standard point of the barrier compared
with the results corresponding to the straight-type barrier with no absorbing materials. ‘L-noise’
and ‘P-noise’ stand for ‘Lower-part-noise’ and ‘Pantograph-noise’.
Based on the scale-model studies, the most effective barrier is the Y-profile. Fig. 6
shows the relative noise levels corresponding to the Y-profile barrier compared with
those of the straight-type barrier. The positive values mean that the Y-profile barrier is
more effective. For the lower part noise, greater noise reduction can be seen at the
688 T. Kitagawa et al.
measurement points close to the viaduct above the Y-profile barrier. This is due to the
screen effect of the inward inclined part of the Y-profile barrier. It is also found that the
noise levels are reduced considerably in the area close to the viaduct below the barrier.
This is closely related to double diffractive effects of the Y-profile barrier. For the
pantograph noise, the outward part of its top makes the shadow zone behind the barrier
wider.
(a) (b)
Fig. 6. Relative noise levels in dB(A) corresponding to the Y-profile barrier compared with those
of the straight-type barrier with 3 m in height. (a) Lower part noise, (b) pantograph noise.
A test barrier with a total length of 30 m was installed along a Shinkansen railway line.
The test barrier had a Y-profile cross-section, and, during the test campaign, sound
absorptive materials and the acoustic tubes were set on the barrier in turn. Fig. 7 shows
an overview of the test barrier and its conditions. For the sound absorbing materials,
glass wool panels were used and fitted on the trackside of the barrier. For the acoustic
tubes, the depths of the tubes are in the range of 0.1-1 m. Since the finite length of the
barrier allows for some sound to diffract around the ends of the barrier, the acoustical
performance of the barrier was investigated by using three one-dimensional
microphone arrays [5]. The array measurements were carried out at three positions
located 12.5 m, 25 m and 50 m away from the centre of the adjacent track (hereafter,
“P12.5”, “P25” and “P50”) during a train pass-by.
1m
Sound-
吸音材
absorbing
Sound-
吸音材
absorbing
2m
2m
materials
2m
2m
materials
R.L. R.L. R.L. R.L.
Fig. 7. Test barrier and its conditions. (a) Overview (concrete bridge, 8 m in height), (b) Straight
type (Case 1), (c) Y-profile type (Case 2), (d) Y-profile barrier with sound-absorbing materials
(Case 3), (e) Y-profile barrier with sound-absorbing materials and acoustic tubes (Case 4).
Study on Effective Sound Barriers for High Speed Trains 689
3.2 Results
10dB
there are two remarkable peaks, LPs,
which mainly consists of the
pantograph noise. The other peaks
except front and rear cars, LQs,
correspond to the lower part noise. Fig. 8. Sound pressure level measured with a
one-dimensional microphone array at P25
Fig. 9 shows the results measured at
P25 and P50. It is found that LQ at P25
was reduced by about 2 dB in all cases. This noise reduction was much smaller than that
obtained in the laboratory measurements. This is probably because the bridge noise has
a great influence on LQ at P25 in the field test. For LQ at P50, the noise reductions of up
to about 5 dB can be seen, and these results are closer to those obtained in the
laboratory measurements. This suggests that the bridge noise was considerably
attenuated at P50. Since, in the laboratory measurements, the acoustical effect of the
Y-profile barrier for the lower part noise at P25 was almost the same as that at P50, it is
reasonable to expect that the same trends would be found in the field tests. Therefore, in
the fields test, the acoustical effect of the barrier on the lower part noise at P25 is
investigated based on the results at P50.
A-weighted sound pressure level (dB)
Case 1 Case 1
Case 2
(a) Case 2
(b) (c)
Case 3 Case 3
Case 4 Case 4
Case 1
Case 2
5dB
5dB
5dB
Case 3
Case 4
150 200 250 300 150 200 250 300 150 200 250 300
Train velocity (km/h) Train velocity (km/h) Train velocity (km/h)
Fig. 9. Measured results with microphone arrays. (a) LQ at P25, (b) LQ at P50, (c) LP at P25.
Table 2 summarizes the relative noise levels compared with the results
corresponding to the straight-type barrier (Case 1). The results obtained in the
laboratory measurements are also shown for comparison. For the LP, it is clear that the
results obtained in the field tests show good agreement with those in the laboratory
measurements. For the LQ, the results obtained in both tests have similar trends. Fig. 10
shows the result for LQ at P50 for each barrier condition. It is found that the result for
Case 2 is lower than that for Case 1 above 315 Hz. By comparing the result for Case 3
with that for Case 2, it can be seen that noise reduction occurs above 1000 Hz due to the
effect of the sound-absorptive materials. Furthermore, the installation of the acoustic
tubes leads to noise reduction below 800 Hz. These global trends show good agreement
with the results obtained in the laboratory measurements.
690 T. Kitagawa et al.
Barrier condition LP LQ
10dB
Case 2 5dB 3.5dB*
Test barrier Case 3 5dB 4dB*
125 250 500 1000 2000 4000 OA
Case 4 5dB 4.5dB*
Frequency (Hz)
1/20 Scale
Case 2 6.5dB 6.2dB* Fig. 10. Results for LQ at P50. Shinkansen
model train runs at 275km/h.
4 Conclusion
Through laboratory measurements on 1/20 scale models, the relative acoustical effects
of sound barriers in typical Shinkansen railway situations were investigated.
Furthermore, the effect of the Y-profile barriers developed through the laboratory
measurements was examined carefully by field tests in full-scale with running
Shinkansen vehicles. The conclusions are as follows:
(1) The Y-profile barrier is the most effective on Shinkansen noise. This is because the
Y-profile barrier has noticeable effects on both mitigating the noise increase due to
the multiple reflections of sound between the barrier and the vehicle surface and
widening the shadow zone behind the barrier because of its diffractive effects.
(2) The noise reduction due to the sound absorbing materials occurs above 1000 Hz,
and depends on its area on the trackside of the barrier. Also, the sound absorbing
materials mitigate the multiple reflections. For the array of the acoustic tubes on
the upper surface of the Y-profile barrier, the noise reduction occurs in the
frequency range below 800 Hz.
(3) For the effect of the Y-profile barriers in full-scale, the results obtained in the field
test show good agreement with those in the laboratory measurements.
References
[1] Koyasu, M., Yamashita, M.: Scale model experiments on noise reduction by acoustic barrier
of a straight line source. Applied Acoustics 6(3) (1973)
[2] Nagakura, K., Kitagawa, T.: Study on effective shapes of sound barriers for Shinaknsen.
RTRI Report 16(12) (2002)
[3] Murata, K., Nagakura, K., Kitagawa, T., Tanaka, S.: The insertion loss of Y-shaped Noise
barrier for Shinkansen. S, RTRI Report 20(1) (2006)
[4] Nagakura, K., Zenda, Y.: Prediction model of wayside noise level of Shinkansen. Wave
2002 (2002)
[5] Moritoh, Y., Zenda, Y., Nagakura, K.: Noise control of high-speed Shinkansen. Journal of
Sound and Vibration 193(1) (1996)
Study on Abnormal Interior Noise of High-Speed Trains
1 Introduction
Noise is one of the key issues of high-speed railways. It is becoming more serious with
the increasing of speed and operating mileage of high-speed trains, the mechanism of
noise generation is not clearly understood. Hence, the study of abnormal interior noise
is of great importance.
Previous studies on the interior noise of high-speed trains emphasize the analysis of
sound pressure level based on numerical simulations and experiments. However, tests
show that the interior noise of high-speed trains can’t be entirely characterized by its
sound pressure level [1, 2]. In addition, to sound frequency distribution characteristics,
sound variation characteristics in the time domain need to be considered. So far, few
published papers have detailed discussions on the mechanism and control measures for
the type of abnormal interior noise that occurs suddenly and randomly in high-speed
train operations.
A type of abnormal interior noise was found in the 9th coach of a high-speed train which
is a dining car. The noise in this coach is much more annoying than in other passenger
coaches. In order to determine the characteristics of this abnormal interior noise and the
root causes for its generation, field experiments on a high-speed train in operation were
carried out. The high-speed train is made up of 16 coaches and it runs at 300 km/h.
Usually, the interior noise of high-speed trains is affected by two different types of
sources. One is the vehicle external noise sources including wheel-rail noise, traction
noise, aerodynamic noise, etc. which is transmitted into the coach interior. The other
is the vibration of the coach structure itself which radiates sound. Because the
abnormal interior noise occurs in the 9th coach that has a special structure inside, the
characteristics of the interior structure vibration and sound radiation may be different
from other coaches. For the external sources, wheel-rail noise and aerodynamic noise
are dominant when a high-speed train is running at 300 km/h [3]. Because the external
structure of the 9th coach is the same as other coaches, we can assume that its
aerodynamic noise characteristics are close to those of other coaches. That is to say,
wheel-rail noise is the most important external noise which is related to the abnormal
interior noise generation. Fig. 1 shows the research ideas.
The 9th coach has an abnormal interior noise problem and noise in the 8th coach is
normal when operating at 300 km/h. First, it should be clear what the abnormal interior
noise is from the sound quality analysis, in which the characteristics of sound variation
in the time domain and the sound level distribution in the frequency domain are
investigated in detail. Second, it should be possible to find the relationship between
the interior structure of the 9th coach and the abnormal interior noise generation in the
coach by comparing the structure vibrations and sound qualities in the same locations
of the 8th and 9th coaches after the re-profiling of their wheels. Finally, a study was
conducted on how the wheel re-profiling or the wheel roughness affects the interior
noise by comparing and analyzing wheel roughness, bogie vibrations and interior
sound qualities before and after the wheel re-profiling. Through these detailed
analyses, the root cause for this abnormal interior noise should become clear.
Fig. 2 shows the measuring points on the high-speed train where “●” refer to the
acoustic measuring points and “■” denote the vibration measuring points. There are 17
microphones at a vertical height of 1.5 m above the floor used for testing the interior
Study on Abnormal Interior Noise of High-Speed Trains 693
noise. There are 14 accelerometers installed on the floor to measure the vertical
vibration of the floor and there are 12 accelerometers fixed on each axle box, bogie
frame and car body to measure the vertical vibration of the bogie.
Sound quality analysis here refers to the objective evaluation including sound loudness,
sound sharpness, sound roughness, and sound fluctuation strength analyses. Loudness
describes the sound magnitude measured in “sone”. The level of 40 dB for a 1 kHz tone
is used as the reference for loudness, i.e. 1 sone. Sharpness is an acoustic index used to
describe the proportion of the high frequency components of a sound spectrum. Its unit
is “acum”. The reference sound producing 1 acum is a narrow band noise, and is one
critical band wide at a centre frequency of 1 kHz with a level of 60 dB. Sound
roughness and sound fluctuation strength are used to describe people’s perception of
the instantaneous change of sound. The sound change at high frequencies causes the
roughness change and the sound change at low frequencies influences the fluctuation
strength. Their units are “asper” and “vacil”, respectively. For sound roughness,
1 asper is defined by a 60 dB, 1 kHz tone modulated by 100 % at a modulation
frequency of 70 Hz. For sound fluctuation strength, 1 vacil is defined by a 60 dB,
1 kHz tone modulated by 100 % at a modulation frequency of 4 Hz. In this paper,
computational models of all the above sound quality parameters proposed by Zwicker
[4] are used.
Fig. 3 shows the interior sound quality results of the 8th and the 9th coaches at a
running speed of 300 km/h before the wheel re-profiling.
From Fig. 3 we can see that sound loudness at the ends of the coach is higher than
that in the middle. The changes of the sound sharpness, the sound roughness and the
sound fluctuation strength are different at the different positions. Fig. 3(a) shows that
there is a great difference between the sound loudness of the 9th and the 8th coaches.
The difference is 15~20 sone. However, the sound loudness is very similar at the end
of the 9th and the 8th coaches (S8 and S17). The sound loudness at point S8 is much
higher than that at point S2 in the 8th coach though the two points are both 3 m from
the end of the 8th coach. This is because the interior structures and external excitation
694 J. Zhang et al.
inputs are different at the two measuring points (S2 and S8) in the 8th coach. In
addition, the loudness at the vestibule diaphragm of the 9th coach is the highest in the
whole coach, reaching 71 sone.
Fig. 3(b) shows that sound sharpness at most measuring points of the 9th coach is
higher than those in the 8th coach. This means that the proportion of the high frequency
components of the sound spectrum in the 9th coach is higher than in the 8th coach.
Fig. 3(c) and 3(d) show that there are no significant differences between the two
coaches in sound roughness and sound fluctuation strength. Consequently, the key
characteristics of the abnormal interior noise are significantly higher sound loudness
and slightly higher sound sharpness.
The system response generally depends on its inherent characteristics and external
excitation inputs. The 9th coach is a dining car of the high-speed train. Its interior
structure which reflects its inherent characteristics is different from other coaches.
Furthermore, it is found through the sound quality analysis that the loudness at the
vestibule diaphragm of the 9th coach is the highest in the whole coach. This phenomenon
Study on Abnormal Interior Noise of High-Speed Trains 695
probably correlates strongly with the external excitation inputs of the coach. Thus, this
section discusses how the inherent characteristics and external excitation inputs of the
coach cause the abnormal interior noise generation based on the analysis of the vibration
acceleration level and sound quality of the interior structure for different wheel
roughness conditions.
It is assumed that the roughness of all the wheels is not different (after the wheel
re-profiling) and the other external excitation to the coach is similar in the analysis with
regard to the effect of the interior structure on the abnormal interior noise generation.
Fig. 4 shows the interior sound loudness and structure acceleration levels of the 8th and
the 9th coaches at the running speed of 300 km/h after the wheel re-profiling.
From Fig. 4(a) we can see that sound loudness at the vestibule diaphragm of the 9th
coach is similar to that in the same position of the 8th coach after the wheel
re-profiling. But before the wheel re-profiling, it was about 17 sone higher than that in
the 8th coach, as indicated in Fig. 3(a). Fig. 4(b) shows that the floor vertical vibration
acceleration levels are also similar for the two coaches after the wheel re-profiling. So
the external vibration and noise of the two coaches should be considered to be almost
equal after the wheel re-profiling. However, the sound loudness at the most points of
the 9th coach is still 5 ~ 15 sone higher than in the 8th coach. That is to say, the 9th
coach abnormal interior noise generation is mainly attributed to its interior structure
effect. Such a special interior structure not only causes stronger vibration and sound
radiation, but also changes the characteristic of the interior diffuse sound field.
In order to investigate the difference of the interior structure vibration and its
sound radiation in the two coaches, the vibration accelerations on an interior cross
section are measured. Fig. 5 shows the measuring points and the results.
696 J. Zhang et al.
From Fig. 5 we can see the vibration acceleration levels at points on an interior
cross section which includes the roof, the side plate, the up side wall, the middle side
wall and the down side wall. The vibration acceleration levels in the 9th coach are
lower than those at the same positions of the 8th coach on the whole. It can be seen
that sound loudness in the 9th coach is higher than in the 8th coach even though the
vibration acceleration levels at points on the interior cross section in the 9th coach is
lower and the external excitation inputs are similar. As a result, the special interior
structure which changes the interior diffuse sound field is the root cause of this
abnormal interior noise generation of the 9th coach.
Left Right
Measuring Points axle bogie car axle bogie car
box frame body box frame body
Before the wheel
46.23 27.72 12.74 41.48 27.65 11.52
re-profiling
After the wheel
26.62 10.83 -3.79 28.31 15.12 -2.39
re-profiling
From Table 1, we can see that the vibration acceleration levels of the 9th coach
bogie are generally reduced by 10~20 dB after the wheel re-profiling, especially on
the left axle box for which the left wheel diameter difference was very high before the
wheel re-profiling. The higher the wheel roughness is, the higher the vibration
acceleration level is. Fig. 7 shows the interior sound loudness and sound sharpness
before and after the wheel re-profiling.
From Fig. 7(a) we can see that the sound loudness in the 9th coach is generally
reduced by 3 ~ 19 sone after the wheel re-profiling, especially at the vestibule
diaphragm. However, Fig. 7(b) shows that the sound sharpness in the 9th coach
increases a little after the wheel re-profiling. This may be because after the wheel
re-profiling, the reduction of the external excitation inputs reduces the interior noise
power in low and middle frequency range and increases high frequency noise. Since
the sound loudness in the 9th coach after the wheel re-profiling is higher than it was in
the 8th coach even before the wheel re-profiling, this abnormal interior noise problem
with the 9th coach cannot be solved through the wheel re-profiling. The problem is
probably related to the interior furnishings of the coach, which is still an open
problem so far.
698 J. Zhang et al.
5 Conclusions
In this paper, noises in two types of high-speed coaches are tested and analyzed. The
conclusions are as follows: (1) the significantly higher sound loudness and slightly
higher sound sharpness are the key characteristics of the abnormal interior noise in the
9th high-speed coach, (2) the special interior structure which changes the interior
diffuse sound field is the root cause of this abnormal interior noise, (3) wheel
re-profiling can reduce the abnormal interior noise, but cannot solve this problem, and
(4) optimizing the interior structure and improving the interior acoustic environment
is a feasible way to eliminate this abnormal interior noise.
Acknowledgements. The present work was supported by the National Science and
Technology Support Program of China (2009BAG12A01-B06), the National High
Technology Research and Development Program of China (863 Program)
(2011AA11A103-2-2), the Funds for Innovation Research Team of Ministry of
Education of China (IRT1178), and the Fundamental Research Funds for the Central
Universities and the Science (SWJTU12ZT01).
References
[1] Hardy, A.E.J.: Measurement and assessment of noise within passenger trains. Journal of
Sound and Vibration 231(3), 819–829 (2000)
[2] Parizet, E., et al.: Noise assessment in a high-speed train. Applied Acoustics 63(2002),
1109–1124 (2002)
[3] Van Beek, A., et al.: Rail sources: State of the art, Harmonoise Report
HAR12TR-020118-SNCF10. European Commission (2002)
[4] Zwicker, E., Fastl, H.: Psychoacoustics: Facts and models, 2nd edn. Springer, Berlin (1999)
Interior Noise Prediction of High-Speed Train
Based on Hybrid FE-SEA Method
Y.J. Zhao, X. Deng, S.Q. Liu, R. Shuai, Z.J. Sun, J.Q. Guo, and Y. Xu
1 Introduction
such as modal density, interior loss factor of the body structure and coupling loss factor
were obtained. The external noise and vibration excitations were acquired from running
tests. Finally, the noise at the central section of a high-speed train was predicted by
using the Hybrid FE-SEA model. Compared with test results, in the frequency range of
100 ~ 1000 Hz, the predicted and test noise spectra followed basically the same trend,
so the proposed Hybrid FE-SEA method can efficiently be applied to the railway field.
a) b) c)
Fig. 1. (a) Geometry model, (b) Equivalent multilayer FE model, (c) General laminate model in Va
One
Interior Noise Prediction of High-Speed Train Based on Hybrid FE-SEA Method 701
Taking a local model of the aluminum alloy extrusion for example, the model area
size is 1m × 0.95 m , and the thickness is 30 mm. According to the above principle of
equivalence, a same size equivalent multilayer finite element model was built by means
of ANSYS software (see Fig. 1(b)); the specific material parameters are shown in
Table 1. Then, the modes of the equivalent model and original structure are computed,
Table 2 shows the computation results for the first five modes; the comparison results
indicate that the modes frequencies are basically the same, with errors within 5 %.
The equivalent multilayer finite element model is imported into VA One software.
Due to the different thicknesses of the upper and lower layers of the aluminum alloy
extrusion, the General Laminate plate in the VA One software is adopted to build a
SEA subsystem (see Fig. 1(c)). Then, the sound insulation of the SEA equivalent
model is calculated, and Fig. 2 shows the corresponding graph of the simulation result
and test result. The results show that in the 100 Hz ~ 1000 Hz frequency range, the
sound insulation performance of the SEA equivalent model and original structure is
similar, with errors in each frequency band of not more than 2.5 dB.
Table 2. The first five modes of equivalent model and original structure
Fig. 2. Sound insulation graph of equivalent computation result and test result
Based on comprehensive comparison analysis of sound insulation for the first five
order modes, the original aluminum alloy extrusion and equivalent multilayer model
have consistent dynamic characteristics similar sound insulation performance.
Therefore, this equivalent method can be used to simplify the car body profile to build
SEA subsystems.
The Hybrid FE-SEA model of the car body was established based on geometry and FE
models by means of commercial VA One software. According to the characteristics of
the car body and the basic principle of the Hybrid FE-SEA method, the beams and
stiffeners, such as the draft sill and bolster and cross beams with low modal density
were described by a FE subsystem, and the hollow aluminum alloy extrusion of the
body frame, trim, glass windows and other thin and larger plates and panels with high
modal density compared to the beams were divided into a SEA subsystem. The
acoustic cavities of car body and the air layer between the profile and interior trim are
modeled as a SEA acoustic cavity, connected to FE and SEA subsystems. Fig. 3 (a) is
the hybrid FE-SEA model of whole car body, Fig. 3(b) and Fig. 3(c) are the interior
structure and the interior cavity of the acoustic subsystem, respectively.
a) b) c)
Fig. 3. (a) Hybrid FE-SEA model of whole car body, (b) interior trim, (c) interior SEA acoustic
cavity
Interior Noise Prediction of High-Speed Train Based on Hybrid FE-SEA Method 703
Fig. 4. Internal loss factor test Fig. 5. Internal loss factor curve test result
Parameters and measured external excitations are loaded into the prediction model.
Fig. 7 shows the hybrid model with imported excitations. Then, this model is used to
predict the interior noise of the car body. The prediction results of noise at a 1.2 m
standard point in the central section of the car body are shown in Fig. 8. A
comparison between prediction and test results can be seen in this figure; in the low
frequency range of 100~500 Hz, there is a mean deviation of 2.5 dB below the test
result, but in the range of 500 ~ 1000 Hz, the mean deviation only approaches 2 dB.
We are justified in observing that the prediction and the test results follow basically
the same trend. Therefore, the proposed Hybrid FE-SEA model can meet engineering
needs and the Hybrid FE-SEA method can efficiently be applied to the railway field.
Fig. 7. Hybrid model with imported Fig. 8. Prediction and test result comparison
excitations
5 Conclusions
In this paper, the Hybrid FE-SEA method is presented to predict the interior noise of a
high-speed train. First, the equivalent model of a General Laminate was proposed to
simplify the special structure of hollow aluminium alloy extrusion in the paper.
According to the characteristic of the car body and the basic principle of the Hybrid
FE-SEA method, the subsystems of the car body are divided to set up a hybrid
Interior Noise Prediction of High-Speed Train Based on Hybrid FE-SEA Method 705
FE-SEA model. Then, through theoretical calculation and test, subsystem parameters
such as modal density, interior loss factor of body structure and coupling loss factor
are obtained. The external excitations are acquired from running test. Finally, the
noise at the central section of car body is predicted by using the Hybrid FE-SEA
model. Compared with the test results, in the frequency range of 100 ~ 1000 Hz, the
predicted and test noise spectra follow basically the same trend, so we can effectively
use the proposed Hybrid FE-SEA method to predict the interior noise of rolling stock
to evaluate passenger ride comfort.
References
[1] Langley, R.S., Bremner, P.: A hybrid method for the vibration analysis of complex
structural-acoustic systems. J. Acoustical Society of America 105(3), 1657–1671 (1999)
[2] Langley, R.S., Cordioli, J.A.: Hybrid deterministic-statistical analysis of vibro-acoustic
systems with domain couplings on statistical components. Journal of Sound and
Vibration 321, 893–912 (2009)
[3] Shorter, P.J., Langley, R.S.: Vibro-acoustic analysis of complex systems. Journal of Sound
and Vibration 288, 669–699 (2005)
[4] Chadwyck, T., Musser, A.B.R.: Mid-frequency prediction accuracy improvement for fully
trimmed vehicle using hybrid SEA-FEA technique. In: Proceedings of the SAE BRASIL
Noise and Vibration Conference, Brazil (2008)
[5] Charpentier, A., Sreedhar, P., Fukui, K.: Using the hybrid FE-SEA method to predict
structure-borne noise transmission in a trimmed automotive vehicle. In: Proceedings of
Noise and Vibration Conference and Exhibition, Charles, Illinois (2007)
[6] Manning, J.E.: Hybrid SEA for mid-frequency. In: Proceedings of Noise and Vibration
Conference and Exhibition, Charles, Illinois (2007)
[7] Faly, F.J.: Statistical energy analysis: Noise and vibration. M. Ed. By R.G. White, J.G.
Walker. Ellis Horwood Limited (1982)
[8] Yao, D., Wang, Q.: Statistical energy analysis of theory and application. Beijing Institute
of Technology Press (1995)
[9] Cheng, G., Zhu, S., Wu, X.: A summary of statistical energy analysis method and its loss
factor determination. J. Ship Engineering 26(4), 10–15 (2004)
[10] Irie, Y., Nakamura, T.: Prediction of structure borne sound transmission using statistical
energy analysis. J. Bulletin of the Marine Engineering Society in Japan 13, 60–72 (1985)
Attractive Train Interiors: Minimizing Annoying Sound
and Vibration
1 Introduction
1.1 Background
allowing for robust and practical maintenance schemes, is a real challenge for the
industry. The present paper is devoted to this challenge and constitutes a summary of
the work reported in reference [1].
1.3 Notation
When we discuss annoying sounds we need a common notation. A large variety of
terms, often sound-mimicking, has been developed to characterize annoying sounds.
The term buzzing sound is used for low-frequency sounds like the sound generated by
electric lights, 50 Hz plus harmonics. With knocking or tapping sounds we mean a
sound consisting of a distinct, often regularly repeated, series of soft impacts, like
when we knock on a door or a table. Rattle is a more complex sound consisting of a
series of more or less random impacts followed by reverberant sound with wide
frequency content, like that generated when shaking steal beads in a tin can. Squeal is
a high frequency, typically 600 – 2000 Hz, tonal sound with long duration, like a
squealing railway wheel during braking or curving. Squeak is a short duration high
frequency tonal sound, like when the rubber sole of a shoe rubs a polished floor. A
hissing sound is a long duration high frequency broad-band sound, like when gas
flows out from a small leak. Scratch is a short duration high frequency sound, like
two sandpapers in sliding contact. Grunt, hum and moan are all low frequency
sounds, typically from say 100 Hz to say 500 Hz, like a door slowly turning on its
hinges. And so on.
Frequently, all the above mentioned annoying noises are denoted “BSR noise”, from
Buzz, Squeak and Rattle, even though the sounds addressed may have a different
character.
Ten measurement sessions on four different train types, were performed. During
these sessions the trains were searched for annoying sounds and vibrations. In addition
to the sound recordings, vibration velocity levels of tables and chairs were recorded
together with the floor velocity level. When an annoying sound or vibration issue was
detected and roughly localized, a microphone and an accelerometer were placed close
to the supposed origin of sound. Sound pressure and acceleration signals were recorded
for later analysis.
The sounds recorded were categorized in three groups with respect to the sound
character: (i) knocking or tapping sounds, (ii) rattling sounds and (iii) scratching or
rubbing sounds. Annoying sound generating mechanisms were identified together with
interior systems with high vibration levels. In reference [1] the analysis methods
applied are presented, results are detailed and the sounds recorded are categorized into
subgroups as indicated above. Annoyance was to a large extent rated subjectively but
different approaches and metrics to rate annoyance of sounds are discussed in general
terms in reference [2] and with reference to particular train application in reference [1].
5
Sound pressure/(Pa)
-5
0 10 20 30 40 50 60
Time/s
Acceleration/(m/s2)
20
-20
0 10 20 30 40 50 60
Time/s
Fig. 1. Sound pressure (top) and acceleration signals (bottom) measured on a light cover panel
making scratching sounds. The signal to noise ratio is much higher for the acceleration signal.
A visual inspection reveals that the acceleration signals is much more favorable for
further analysis than the sound pressure signal. The reason is that a large part of the
energy in the sound pressure signal originates from other sources like talking
passengers etc. This is illustrated in Fig. 1 where sound pressure and acceleration
signals from a scratching light cover are shown.
Most annoying sounds on vehicles are caused either by impact forces or by friction
forces. The first group of excitation mechanisms includes tapping and rattling sounds.
The second group includes squeal, squeak, squelch and moaning sounds, all caused by
710 U. Orrenius and U. Carlsson
20
10
Acceleration/(m/s 2) 0
-10
-20
-30
10 12 14 16 18 20 22 24 26 28 30
Time/s
Fig. 2. Accelerometer placed on cabinet door and the signal measured. Strong bursts of vibration
are excited when the door impacts the door frame.
Light covers and other ceiling panels are also prone to rattle or produce rubbing
sounds. In the Fig. 3, a rattling light cover is illustrated together with the acceleration
signal recorded. The acceleration spectrum of Fig. 4 illustrates the reverberant character
of the signal. One possible reason for rattling of this component is that dismounting and
remounting of light covers during maintenance leads to misalignments which increase the
risk for annoying sound generation.
10
4
Acceleration/(m/s 2)
-2
-4
-6
-8
-10
52 52.5 53 53.5 54 54.5 55 55.5 56
Time/s
Fig. 3. Rattling light cover on regional train and acceleration signal measured. Note the impacts
appearing with roughly 0.1 s time interval and the long reverberation time.
Attractive Train Interiors: Minimizing Annoying Sound and Vibration 711
0.12
0.1
rms: Acceleration/(m/s2)
0.08
0.06
0.04
0.02
0
0 500 1000 1500 2000 2500 3000
Frequency/Hz
Occurrences of high vibration levels were also found. Fig. 5 shows the vertical
acceleration spectrum measured on a seat-mounted table together with that of the floor
below.
0.25
0.2
rms: Acceleration/(m/s )
2
0.15
0.1
0.05
0
0 10 20 30 40 50 60 70 80 90 100
Frequency/Hz
Fig. 5. Measuring of sound and acceleration on seat mounted table (left) and vertical acceleration
spectra measured (right): Solid blue, acceleration on floor; Dashed green, acceleration on table
The seat-table system analyzed has a few resonances below 30 Hz, which are the
main contributors to the table vibration level. In this frequency range, the floor
vibrations are magnified by roughly a factor three due to the chair resonance leading to
high lateral and vertical table velocities. This observation suggests that the fundamental
resonances of the chair-mounted table system should be increased for reduced table
vibration. Also, seat loading strongly affects the vibration level. With a passenger in the
seat where the table is mounted the level is reduced. The table vibration level also
increases with train speed, see reference [1].
4 Discussion
4.1 Vibration Effects on Passenger Activities
According to Griffin and Hayward [5] reading, writing and typing difficulties are due
to relative vibration. Having this in mind it is clear that body resonances, in addition
712 U. Orrenius and U. Carlsson
doors and ceiling panels to identify potential BSR problems before the component is
mounted in the vehicle. Increased use of specific BSR focused component vibration
testing may reduce the problems in rail vehicles. However, the approach is fairly
complex as several samples of each component need to be tested and usually an
experienced BSR engineer needs to be present at the tests. It is essential to mount the
system in a similar manner as applied in the vehicle.
The approach developed within Ford Motor Company is described in the book by
Chen and Trapp [3] where step-by-step procedures to systematically provoke, rank and
evaluate potential BSR issues from the components tested are also described. Certain
advances to automatically sort out poorly performing components have been
developed; see for example reference [8] where the feasibility of an automated system
for detection of BSR events is presented that can potentially replace “subjective”
detection with a human jury.
For rail vehicles no such procedures are standardized and excitation spectra and
levels to be applied need to be adapted to the operating conditions of railway vehicles.
A starting point may be to use the standards applied for vibration testing of electronic
equipment on rail vehicles (EN 61373), but the levels need to be adapted to
systematically provoke BSR events. Another aspect for trains is that the excitation may
vary significantly depending on the propulsion system, as diesel drive systems may
cause strong interior vibrations leading to a higher low frequency content of the
excitation.
5 Conclusions
From the survey on Swedish intercity trains in operation it is found that most train
sets studied have issues with annoying sounds and vibration related to interiors. For
some vehicles the number of issues is substantial. This observation underlines the
need for systematic abatement procedures and proactive activities from the
manufacturers and operators to ensure comfortable train journeys.
As the correlation between ISO vibration rating standards and perceived reading and
writing difficulty is found to be poor, a dedicated standard for assessing reading and
writing related vibration comfort is judged to be of interest for the railway industry.
Most issues with annoying sounds from interiors are directly related to the product
assembly process and designers are advised to work with interior component design
and, more important, with methods and processes to test and assemble the key
components including component quality control.
Routines, tools and procedures in design, production and maintenance programs
should be adopted to minimize occurrences of annoying sounds, e.g. by avoiding
relative motion between components. Some mitigation activities proposed are [1]:
Acknowledgement. The funding of this work was made available by the Swedish Gröna
Tåget research program. The in-situ testing in operating trains was conducted by KTH
students Max Zapka and Claes Kastby and the authors are grateful for their valuable
support. Susanne Rymell and Jonas Strömblad, SJ AB, are kindly acknowledged for their
assistance in making trains available for measurements.
References
[1] Carlsson, U., Orrenius, U.: Attractive train interiors: Minimizing annoying sound and
vibration. KTH Railway Group and Bombardier Transportation, Stockholm (2013) ISSN
1651-7660; TRITA-AVE 2013:28
[2] Griffin, M.J.: Handbook of human vibration. Academic Press Ltd. (1996) ISBN-13:
978-0123030412
[3] Sundström, J.: Difficulties to read and write under lateral vibration exposure. Doctoral
Thesis in Railway Technology, KTH, Stockholm (2006)
[4] Chen, F., Trapp, M.: Automotive buzz, squeak and rattle: Mechanisms, analysis, evaluation
and prevention. Butterworth-Heinemann, Oxford (2012) ISBN-10: 075068496
[5] Griffin, M.J., Hayward, R.A.: Effects of horizontal whole-body vibration on reading.
Applied Ergonomics 25(3), 165–169 (1994)
[6] Khan, S., Sundström, J.: Effects of vibration on sedentary activities in passenger trains.
Journal of Low Frequency Noise, Vibration and Active Control 26(1) (2007)
[7] ISO 2631-4. Mechanical vibration and shock: evaluation of human exposure to whole-body
vibration - Part 4: Guidelines for the evaluation of the effects of vibration and rotational
motion on passenger and crew comfort in fixed guideway transport systems (2001)
[8] Cerrato-Jay, G., et al.: Automatic detection of buzz, squeak and rattle events. SAE paper
2001-01-1479 (2001)
Author Index
Han, J. 117, 345 Lombaert, G. 245, 253, 539, 555, 627, 635
Hanson, D. 85, 441, 449 Lurcock, D.E.J. 297
He, B. 345 Lutzenberger, S. 21
Heiland, D. 237
Hejzlar, L. 101 Malinský, J. 189
Herron, D. 611, 627 Marshall, T. 497
Hieke, M. 505, 571 Martens, A. 181
Hlaváček, J. 101 Mirza, A. 523
Hoogzaad, S.N. 353 Mistler, M. 237
Hu, Z.W. 489 Mitsumoji, T. 515
Huber, Ph. 531 Moehler, U. 181
Hunt, H.E.M. 321 Moorhouse, A.T. 305
Hussein, M.F.M. 321 Müller, R. 531, 547
Murata, K. 55, 683
Ikeda, M. 515
Nagakura, K. 55, 683
Nélain, B. 531, 547, 635
Jakielaszek, L.J. 393
Nelson, J.T. 313
Jansen, H.W. 213, 595
Nielsen, J.C.O. 523, 547, 659
Jaquet, T. 385
Noh, H.M. 229
Jean, P. 289
Nordborg, A. 229
Jensen, T. 205
Jerson, T. 79 Oertli, J. 1
Jiang, J. 441, 449, 539, 555, 587 Ögren, M. 79
Jin, X.S. 117, 345, 691 Orrenius, U. 707
Jones, C.J.C. 39, 651
Jones, R.R.K. 369 Peris, E. 305
Jones, S.W. 321 Phamová, L. 189
Junge, S. 205 Pieringer, A. 417, 433
Poisson, F. 457
Kaewunruen, S. 377 Pouzet, A. 141
Kawaguchi, T. 55 Rajaram, S. 619
Kengang, A. Kengni 149 Reichart, U. 21
Kern, T. 337 Renoncourt, A. 141
Kitagawa, T. 55, 683 Richardson, M.L. 393
Koh, H.I. 229 Richter, M. 189
Kolmačka, R. 101 Roovers, M.S. 93, 353
Kouroussis, G. 329 Rücker, W. 401
Kropp, W. 417, 433 Ryue, J. 555
Kuijpers, A.H.W.M. 93
Kuo, K.A. 321 Sapena, J. 675
Sato, Y. 515
Latorre Iglesias, E. 481 Saurenman, H.J. 619
Lei, T. 667 Schleinzer, G. 337
Leth, S. 31 Shuai, R. 699
Létourneaux, F. 197 Smith, M.G. 481
Li, Z. 63 Song, W. 205
Liepert, M. 181 Spehr, C. 165
Lin, J.S. 563 Squicciarini, G. 39, 47, 213, 603, 651
Liu, S.Q. 563, 699 Starnberg, M. 173
Lölgen, T. 173, 409 Stegemann, B. 31
Author Index 717