See

discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/307965573

Transfer path analysis: Current practice,

trade-offs and consideration of damping

Article in Mechanical Systems and Signal Processing February 2017

DOI: 10.1016/j.ymssp.2016.09.013

CITATIONS READS

0 25

3 authors, including:

Akn Oktav Cetin Yilmaz

Bogazici University Bogazici University
8 PUBLICATIONS 1 CITATION 19 PUBLICATIONS 135 CITATIONS

SEE PROFILE SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Akn Oktav
letting you access and read them immediately. Retrieved on: 30 September 2016
 Mechanical Systems and Signal Processing 85 (2017) 760772

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Transfer path analysis: Current practice, trade-os and

consideration of damping
cross

Akn Oktav , etin Ylmaz, Gnay Anla
Department of Mechanical Engineering, Boazii University, Bebek 34342, stanbul, Turkey

A R T I C L E I N F O ABSTRACT

Keywords: Current practice of experimental transfer path analysis is discussed in the context of trade-os
Transfer path analysis between accuracy and time cost. An overview of methods, which propose solutions for structure
Damping borne noise, is given, where assumptions, drawbacks and advantages of methods are stated
Normal frequency response function theoretically. Applicability of methods is also investigated, where an engine induced structure
Leakage
borne noise of an automobile is taken as a reference problem. Depending on this particular
problem, sources of measurement errors, processing operations that aect results and physical
obstacles faced in the application are analysed. While an operational measurement is common in
all stated methods, when it comes to removal of source, or the need for an external excitation,
discrepancies are present. Depending on the chosen method, promised outcomes like indepen-
dent characterisation of the source, or getting information about mounts also dier. Although
many aspects of the problem are reported in the literature, damping and its eects are not
considered. Damping eect is embedded in the measured complex frequency response functions,
and it is needed to be analysed in the post processing step. Eects of damping, reasons and
methods to analyse them are discussed in detail. In this regard, a new procedure, which
increases the accuracy of results, is also proposed.

1. Introduction

Transfer path analysis (TPA) describes a noise or vibration response at a selected target, as a superposition of vectorial
contributions from a dened set of force inputs that excite the structure, through a chosen set of connections [1,2]. To express the
problem in this way gives an insight and a capability in the attenuation of response at the source and/or on the structure, using
various techniques of active, or passive vibration control [3]. In the widespread practice, TPA is a two-step experimental procedure:
rst step is the identication of forces and the second one is to relate identied forces to target receivers through assumed paths. The
rst step, identication of forces, is mostly the main problem that aects results and developed solutions, accordingly. Operational
forces that excite a structure are generated by a single coherent source, or multiple partially-correlated sources [46]. In the context
of vehicle acoustics, typical examples for single and multiple coherent sources are engine and road inputs, respectively, where the
former is addressed in this work.
The method chosen to identify operational forces shapes the procedure of transfer path analysis. To measure operational forces
directly is generally hard, since force transducers are required to be located between the structure and source, where vibration
isolators or mounts are present. These non-linear isolators respond dierently in the presence of transducers, since local stiness
changes signicantly [7]. Direct force measurement often gives unreliable results [8]; instead, indirect force estimation techniques
are widely used. Mount stiness and matrix inversion methods are the tools in identifying operational forces [2,8]. These indirect

Corresponding author.
E-mail address: akin.oktav@boun.edu.tr (A. Oktav).

http://dx.doi.org/10.1016/j.ymssp.2016.09.013
Received 24 June 2015; Received in revised form 29 July 2016; Accepted 10 September 2016

0888-3270/ 2016 Elsevier Ltd. All rights reserved.

A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

methods have their own advantages and drawbacks. Mount stiness method requires precise mount data, which are seldom
available, while matrix inversion method requires accelerance matrix data, which can be obtained by a troublesome and time
consuming experimental campaign. Even when mount data are available, they are sensitive to excitation amplitude due to the non-
linear nature of the material. Matrix inversion method requires removal of source to achieve both frequency response function (FRF)
and noise transfer function (NTF) measurements. In these measurements, a roving hammer with an appropriate tip for low
frequency applications is used as an exciter. Measured complex FRFs have real and imaginary parts, where the latter is associated
with damping and has smaller signal to noise ratio due to its respectively small magnitude. The real part that represents the mass
and stiness eect is relatively more reliable, since it is known that the exponential window used during impact tests specically
increases apparent damping.
The matter in question with the matrix inversion method is the decoupling of source during frequency response function
measurements. Removal of the source changes the dynamic behaviour of the passive part, which is expected to be dierent, while the
system is operating [9]. Additionally, the system may not be linearised at chosen locations for articial excitation, or linearity
assumption may be held more reasonably at another location, other than the chosen one. Brandl et al. [10] performed a sensitivity
analysis to quantify measurement errors caused by deviations at excitation locations and directions. Frequency response functions,
which are used to populate the accelerance matrix, dier substantially according to the chosen location and the direction of applied
force. Inversion of accelerance matrix further increases deviations, which results in poor outcomes. Even if these problems are
resolved, measurement of frequency response functions is prone to some other deciencies in representing actual system
characteristics. Ozgen et al. [11] investigated variance and bias type errors in measured data of an aluminium beam. They reveal
that leakage in frequency response function measurements is responsible for unexpected results. Leakage in measurements is mainly
the result of short time record, and can be reduced using window functions [12]. All common data acquisition systems oer various
window functions to compensate for leakage. Unfortunately, once entered as raw data, total elimination of leakage is not possible;
moreover, windowing functions introduce articial damping eects to processed data [13]. Schoukens et al. [14] oered a Taylor
series based method to reduce leakage in SISO systems; in follow-up studies, leakage phenomenon is discussed further, and
extension of the method to MIMO systems is given [15,16].
Above stated experimental diculties and potential measurement errors of indirect force estimation techniques induce
researchers to come up with more accurate and feasible methods. Operational transfer path analysis (OTPA), a one-step method,
is presented as one of the promising alternatives. The method, based on transmissibility concept [1719], does not require the
removal of active part, e.g. engine. Since the method requires only operational data, measurements are more easily obtained. Due to
the direct and indirect application of transmissibility, two classes of OTPA are proposed. The former uses a transmissibility matrix
instead of an accelerance matrix [20,21], whereas the latter indirectly estimates an accelerance matrix by using measured
transmissibilities [9]. Gajdatsy et al. assessed the applicability of OTPA methods in the sense of their limitations and associated
errors. Eects of neglected paths, cross-coupling in between passive side accelerations and orthogonality problems in estimated
transmissibilities are pointed out [22]. OTPA methods are further studied to increase the accuracy of results [23]. A Tikhonov
regularisation-based OTPA is proposed to solve ill-posed problems [24]. Such problems are widely examined in a recent review
paper [25]. Based on a parametric load modelling technique, a new TPA procedure [26] is also proposed, which needs extra
measurements for the identication of operational forces. This method oers to combine the widely accepted accuracy of traditional
TPA approaches with time eciency of operational TPA methods. A new approach is proposed for load identication based on mixed
and penalised cost function optimisation [27]. The proposed technique has better accuracy in force reconstruction compared to
classical methods, however, computation time is increased.
Trade-os between accuracy and time cost motivate researchers to develop several approaches. In pseudo force method [28,29],
estimation of operational forces is based on transfer function measurements in combination with operational response
measurements. However, the characterisation of source is not completely independent of the passive part of an assembled
structure, and to compare sets of pseudo forces, which are determined at dierent locations of the same source or on another source,
may lead to errors, although they might be equivalent. Power based methods are also proposed to characterise the structure borne
noise [30,31]. Dynamic coupling between active and passive parts of the structure prevents source characterisation based upon the
transmitted power. Although it is possible to assume the passive part to be dynamically stable in certain cases, to generalise this
assumption is not reasonable [32]. A blocking force approach, which isolates forces from the environment, is presented [33,34]. The
need of specied test rigs for measurements is the main disadvantage of this approach. Global transfer direct transfer (GTDT) is one
of the two-step path analysis methods, which does not require the knowledge of operational forces [35]. In the rst step, direct
transfer functions (DTFs) are calculated from measured GTDT functions, whereas in the second step, operational signal
reconstruction is made by means of calculated DTFs. In an analytical study, the path blocking GTDT method and traditional TPA
procedure are discussed, and their prediction capabilities are assessed [36]. Recently, an experimental work on a simple mechanical
set up is rst time presented [37], where GTDT method is adopted. Although, in conclusion it is said that the method can be applied
to a more complex structure like an automobile, a generic application will not be so straightforward. First, to model engine mounts
as linear springs is not realistic; second, without removing the active part, it is often impossible to access excitation locations.
Advances in testing and measurement technologies provide new solution approaches for TPA. Using PU probes, both sound
pressure and particle velocity can be simultaneously acquired [38]. Particle velocity based methods are eciently used for airborne
TPA studies [39,40]. Since the lower limit for measurements is around 200 Hz, it is not possible to use the method for structure
borne TPA, at least for the present. Unlike for airborne sources, there is no widely applicable method yet available for independent
characterisation of structure borne sound sources. As mentioned, blocked force method that characterises the source independently
is not practical, since it requires a tailored test rig. An in situ measurement method that identies the blocked force, without

761
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

decoupling the source, is proposed by Moorhouse et al. Blocked forces, or independent characterisation of a source is vital in
prediction of the vibration transmission on a dierent receiver [41]. In the case of inverse force estimation techniques that are used
as the rst step of traditional TPA, identied forces can only be used for a specic source-receiver assembly. In situ measurement of
blocked force technique is tested to characterise structure borne road noise of an automobile [42]. Results of the proposed technique
are compared to ones achieved from traditional TPA. Elliott et al. report that, compared results are acceptably similar; moreover,
their method has an important advantage in terms of reduced test time. However, in situ measurement technique gives no
information about mounts, which are obviously important in any source-path-receiver type problem. Mounts play a substantial role
in the renement of structure borne noise problems in automobiles and other machinery. Yet another possible problem for an engine
TPA type application is that measured passive side accelerations can contain extra inputs, which come from either road, or other
auxiliary components. In a recent paper, Sottek et al. [43] oer another in situ measurement approach, which uses active side
acceleration data and matrix inversion method together, to compute so called eective mount transfer functions. Unlike the former,
this approach gives information about mount data, which can be used in any renement study. Nevertheless, an important
advantage of former method is lost, i.e. force data are dependent to specic case studied. To eliminate the disadvantage, Kelvin
Voigt type parametrisation for mounts is oered. For many applications, especially for hydro and/or complex shaped mounts,
assumptions of this modelling are not suitable [44,45]. What is more, if the active part is not removed, as suggested by the method
[43], it is hardly possible to access mounts for hammer impacts. In a recent review paper, a general framework for TPA is presented
[46]. Eleven methods are derived and a classication is proposed.
As above discussion shows, current practice and new approaches have their own advantages and drawbacks, which force
researchers to consider trade-os, especially between accuracy and time cost, while making decisions. In this work, current practice
is briey discussed, and important trade-os, which tip the balance, are highlighted. Applicability of dierent methods to an engine
induced structure borne noise problem of an automobile is studied. Sources of error are examined, and ways to increase the accuracy
of results are presented. Using matrix inversion method, a new procedure is proposed that focuses on damping.

2. An overview of current practice

In this section, an overview of current practice is given, where assumptions, advantages and drawbacks of methods are outlined.
Methods are classied into two groups; rst one (Section 2.1) requires the knowledge of identied operational force vector, and
second one (Section 2.2) uses transmissibility relationships to compute a target response. Although TPA is applicable to all kinds of
assembled structures, such as aeroplanes, ships, domestic appliance and machinery, it is mostly adopted in automotive industry.
Hence, the capability of methods, in solving a particular, but a common vibro acoustics problem, is examined. The structure borne
aspect of engine induced low frequency noise of an automobile is addressed.
Note that, unless indicated otherwise, each function throughout the text is a function of frequency. For the sake of clarity, term,
which denotes the frequency dependence, is omitted. Capital and lowercase mathematical expressions in bold denote matrices and
vectors, respectively.

2.1. Force methods

Consider an assembled structure as sketched in Fig. 1, which comprises active (source) and passive (receiver) substructures,
linked by mount(s). According to TPA approach, a response (y ) at a selected target (i) can be computed as a linear superposition of
separate vectorial contributions (Nij) from a dened set of force inputs (f ) entering to the structure over a known set of connections
(j), as stated in Eq. (1). Here, Nij matrix is populated with noise transfer functions (a.k.a. propagation FRFs, or vibro-acoustic
transfer functions) that are estimated between a target location (i) and an excitation location (j). To solve Eq. (1), matrix of noise
transfer functions (Nij) and vector of operational forces (f j) have to be estimated:
yi = Nij f j (1)

2.1.1. Estimation of noise transfer functions

Estimation of noise transfer functions (NTFs) can be achieved directly or reciprocally on the account of linearity assumption, if it
holds in the frequency bandwidth of interest. As it will be seen in Section 3, for some of the methods, like in situ blocked force and
OPAX, to use reciprocal measurements is a necessity, rather than a preference, since they recommend not removing the source.
NTFs can be measured directly, by exciting the body side of mounts, where operational forces enter to the structure, and by acquiring
the response at the target receiver. As the name suggests, in reciprocal measurement, response data are acquired at mount locations,
while the excitation is performed at the target location, using a low frequency volume source. The point is that results obtained using
these two techniques do not often correlate well. In such a situation, one cannot intuitively nd out which one is more reliable.
Coster et al. [47] compared the two techniques, and investigated the accuracy of them through a sensitivity analysis. Comparisons
show that there are considerable dierences in between the two, which can aect the results at low frequencies, as well. Note that,
the source is removed in the direct measurement, while it is present in the reciprocal one. Mainly, this incompatible change in
boundary conditions alters the dynamic properties of the system.

762
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Fig. 1. An assembled structure, where a and p denote the active (source) and passive (receiver) parts, respectively; m is a mount that links two substructures; j and k
are passive and active side interfaces, respectively; i is a response point of interest; l is an extra measurement location.

2.1.2. Estimation of forces

Advances in instrumentation technology give researchers an opportunity to measure operational forces (f j) directly, using thin
multi-axis sensors. These sensors are located between mounts and the structure, mostly using a tailored apparatus. To render a
measurement possible, original mount bolts and required torque values have to be changed. As a consequence, prestress values on
the engine mount and local stiness on the body side deviate, which result in poor outcomes. Hence, instead of a direct
measurement, operational forces are often estimated using the following techniques.
Mount stiness method requires the knowledge of complex dynamic stiness of mounts, which can be measured using multi-
axial test rigs. Mount data exhibit dierent characteristics, depending on static pre-load, frequency, temperature, and up to some
extent, on vibration amplitude [48]. Non-linear characteristics of data and variances among test specimens make this method
impractical, when the linearity assumption inherent in TPA methods is considered. Estimation of operational forces using mount
stiness method is expressed by
x k x j
f j = Km ,
2 (2)

where Km is the complex dynamic stiness matrix of mount m; xk and xj are active and passive side acceleration vectors, respectively,
which have to be measured, when the system is operating. Even when mount data are reliable, identied forces may contain other
information that is hard to decouple: measured acceleration vectors are also disturbed by road noise, and other auxiliary
components. Nevertheless, this approach is useful, especially in optimisation studies of mounts, where Km matrix is optimised
subjective to a target response.
Matrix inversion method identies operational forces using the passive side acceleration vector (xj), and expressed by

f j = H +mjx j (3)

where (H+mj) denotes the pseudo inverse of the accelerance matrix. Accelerance matrix (Hmj) is populated with FRFs, which
characterise modal properties of mount interfaces, measured when the source is decoupled. However, removal of source may change
dynamic characteristics of the assembled structure, which result in deceptive information in measurements. Hammer tests that are
achieved to measure complex FRFs, introduce leakage and unsubstantial damping eects, beside probable impact location and
direction errors. What is more, measured FRFs contain data of same resonances, which are mathematically hard to decouple. For
this reason, inversion operation of accelerance matrix may give wrong results, where condition number of inverted matrix gains
considerably high values, especially at resonance frequencies.
In a decoupled passive substructure (p), hammer excitation at a mount location (j=1) causes responses at other mount locations
( j = 2, 3, ), as well. Resulting FRFs are o-diagonal terms of Hmj matrix. One can assume that more FRF data mean more noise
and error in the measurements. From this point of view, a diagonal version of matrix inversion method can also be used, which
counts only on diagonal terms of Hmj matrix. This method can be formulated as

f j = diag (H +mj) x j. (4)

Pseudo force method [28,29] is a variant of matrix inversion technique, where equivalent forces are identied instead of actual
operating ones. Estimated pseudo forces are independent of the receiver structure, which means that they are theoretically valid,
even the source is mounted on a dierent structure, other than used. Excitation locations can be chosen arbitrarily (not necessarily
on mount interfaces), unlike the matrix inversion method. Non-uniqueness of pseudo forces is one of the important drawbacks.
Moreover, as proposed by the method, to operate the source in a freely suspended state is hard for some applications, like the engine
of an automobile.
Operational force vectors in Eqs. (3) and (4) are dependent on the dynamic properties of uncoupled receiver substructure. Notice
that, this dependence is due to the accelerance matrix (Hmj), which is constructed by measured FRFs, when the source is removed. If
blocked forces can be estimated instead of them, then it is possible to characterise a vibration source independently. Special test rigs
[34] render measurement of blocked forces possible, although this cannot be achieved ideally. In that case, Eq. (1) can be used for
dierent assemblies, on which the same source is mounted. Instead of a direct measurement, blocked forces can be estimated using
an in situ method, i.e.

+mjx j
fbl = H (5)

763
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

+
where H mj denotes the pseudo inverse of accelerance matrix in the coupled state [41]. The matter in question with this approach is
accessibility to excitation locations. As suggested by Moorhouse et al., it is also possible to identify the blocked force vector using the
operational acceleration of an extra measurement point (l) on the receiver substructure, and by relating it to the passive side of
mount (j). To dene additional measurement points (l = 1, 2, , n ) is recommended, which can be used for over determination of the
matrix (Hjl ) to be inverted. To overcome the accessibility problem, reciprocity principle can be used, and Hjl is measured instead of
Hlj , as stated in Eq. (6), provided that the linearity assumption holds:
+jlx l
fbl = H (6)

An alternative TPA approach, which proposes to identify operational forces by using parametric load models, is coined as OPAX
[23,26]. The method tends to preserve the traditional TPA approach due to its widely accepted outcomes, while working out a
solution for the time consuming FRF measurement step. The model essentially uses operational data, which have to be already
measured for all TPA approaches. Additionally, instead of a full FRF measurement step, a limited number of FRFs are measured to
be used in the construction of a parametric load model for the identication of operational forces, as oered in Eq. (7). This scalable
approach gives a chance for making trade-os between accuracy and time cost, such that one can make decisions about the number
of FRF measurements. In Refs. [23,26], it is recommended to increase number of FRF measurements and processed orders for a
more accurate model. Note that, in comparison with traditional TPA, although time cost is considerably reduced, computational cost
is signicantly increased:
f j = f (parameters, x k , x j) (7)

2.2. Transmissibility methods

Yet another possibility is to accommodate MIMO transmissibility approach to TPA. Although transmissibility is a response-to-
response relationship, and dierent from force-to-response functions, both in terms of outcomes and their physical meanings, some
proposals are available, which enable to get TPA like results, without removing the source. In what follows, dierent transmissibility
based TPA approaches are formulated with a consideration on their applicability, using the thus far notation. One of them is OTPA
method, which claims that there is no need to remove the source to predict a target response. Rearranging Eqs. (1) and (3) yields
yi = NijH +mjx j = Tij x j, (8)

which denes a unique transmissibility matrix Tij, provided that m=j, Hmj is square and invertible. In that case, Nij and Hmj matrices
are replaced by Tij, which stands for transmissibility matrix, where entries can readily be estimated from an operational
measurement, using one-sided auto and cross spectral density functions, i.e.
Tij = G ijG+jj. (9)

This formulation corresponds to an H1 estimator, while H2, or Hs estimators can also be used for construction of the transmissibility
matrix, as shown by Leclre et al. [49].
Another transmissibility based approach is GTDT method, which prescribes that force-to-response functions can be estimated
from response-to-response functions, i.e. measured transmissibilities. In that case,
^ y +T
yi = T ^ y.
ij j ii i (10)
^ and T
Eq. (10) states that response at i is the summation of blocked transmissibilities (T ^ ), provided that i j . Somehow, if all other
ij ii
^
paths can be blocked, the response at i would be equal to T , when the system is excited only at j. Following the same approach, if the
ij

system is excited only at i, T^ would be the response at i, provided that all other responses of the system are blocked. To block a
ii
system is physically hard, if not impossible, and often may result in uncontrollable changes in the dynamic behaviour of the system.
Instead, GTDT method oers to estimate blocked transmissibilities from measured ones (See e.g. [36,37]), using operational
accelerations at j. However, the procedure requires a hammer test step, which cast doubt on any particular application. In other
words, although the applicability of method is proven by another study [50], it may not be feasible for some applications, e.g. the
engine induced structure borne noise of an automobile, where access to interface is almost impossible without removing the source.

3. Applicability of methods

In this part, applicability of above stated methods are investigated on the engine induced low frequency noise problem of an
automobile. This reference problem is chosen, since it is a common industrial problem, which needs fast and accurate solutions. In
the following subsections; rst, sources of errors, feasibility problems and resulting trade-os are stated; then, the reference problem
is described; nally, results are given for the chosen methods.

3.1. Sources of errors, feasibility problems and resulting trade-os

Sources of errors force and motivate researchers to develop new techniques. Eventually, some of the trade-os are caused by

764
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

these errors, which are itemised as the following:

(1) Non-linear characteristics of mounts: Non-linearity of mounts is important in three ways; (i) a wide spread TPA technique,
mount stiness method, uses mount data; (ii) in renement studies, mounts are important targets; (iii) variability among
measurements mostly introduced by mounts [51].
(2) Removal of the source: Methods like matrix inversion require the removal of source, which changes boundary conditions [47].
Removal of the source may also impact damping (see Section 4.2 for a discussion).
(3) Hammer excitation: Most of the TPA methods require hammer excitation to measure FRFs. Direction and location of applied
force change the estimated FRFs [10]. Among bias and variance type errors, leakage is the most important one [11]. Force and
exponential windows used as countermeasures introduce articial damping eects to the estimated FRFs [13]. Note that, FRFs
obtained from an impact hammer test match the theoretical FRFs, only when the record length is innite [52].
(4) Phase errors: When shakers are used as force inputs, phase errors may occur depending on time delays. Information about time
delays is needed to correct the corresponding phase errors [53]. It is also necessary to investigate the leakage induced phase
errors [54].
(5) Inversion of an ill-conditioned matrix: Most of the TPA approaches require matrix inversion steps. Beside orthogonality, noise
and articial damping eects gain considerably high values, during inversion process. To check the condition number is eective
in guring out orthogonality problems, but not enough for other eects.
(6) Eect of neglected paths: In TPA methods, paths are user dened. The point is that it is vital to reveal the eects of neglected
paths [22], where it is not possible in some approaches such as OTPA.
(7) Cross-coupling of inputs: Operational accelerations measured on the body side of a mount do not depend only on the force
acting at this location, other forces acting on other mounts also made contributions. This is called cross-coupling of inputs, which
cannot be neglected without being computed, when input forces are correlated.
(8) Non-linearity of the system: The foremost assumption of TPA methods is the linearity, although it is not valid in real life
structures, especially for complicated ones.
(9) Contaminated information in acceleration signals: Beside measurement noise, data coming from road and auxiliary
components contaminate operational accelerations measured.

On the other hand, measurements are also restricted by some physical obstacles, which can be classied as feasibility problems
that are itemised below.

(1) Access to excitation locations: Today, engines of automobiles are installed such that without partially, or in some cases
completely removing the engine, even to locate sensors for an operational measurement is hard, if not impossible. Some
proposed methods cannot be applied, since excitation locations are inaccessible.
(2) Space problem for a shaker: Almost all aforementioned problems associated with external excitation can be considerably
eliminated, provided that a shaker is used, instead of a hammer. Unfortunately, the irregular shapes of the engine compartments
of automobiles do not provide enough space for a shaker set up, especially to achieve excitations in x- and y-directions.

Sources of errors and feasibility problems cause the following trade-os:

(1) If the source is removed, boundary conditions change, and time cost increases. On the other hand, if the source is not removed,
cross-coupling of inputs aects results. Response-to-response functions are used instead of force-to-response functions, e.g. for
an OTPA work. To access excitation locations may not be possible, e.g. for GTDT and in situ blocked force methods.
Computational cost increases substantially, and NTFs are needed to be measured reciprocally.
(2) In cases, where a hammer is used for excitation, leakage can considerably aect results, provided that force and exponential
windows are not used. However, if these windowing functions are used, articial damping eects are introduced to the measured
responses.

3.2. Reference problem

An automobile equipped with a four-cylindered gasoline engine is tested. Engine and exhaust of the automobile are considered as
vibration sources, where each of them assembled to unibody by 3 mounts. Thus, referring to Fig. 1, active part (a) consists of engine
and exhaust; passive part (p) is the unibody of automobile, and number of links (m) between active and passive parts is 6. The
objective is to predict responses at two microphone locations in the cabin. To measure operational accelerations, triaxial sensors are
placed at active (k=18) and passive (j=18) sides of mounts (m=6), which correspond to 36 measurement locations. A uniaxial
accelerometer is placed on the engine block to get a reference signal for phase information. To provide over determination in matrix
computations, 3 extra measurement (l=3) locations are also chosen on unibody. In the cabin, two microphone (i=2) locations are
determined; driver's left ear level (mic.1), and the midpoint at rear seat, which represents the ear level of passengers (mic.2) [55].
Then, an operational measurement, which is common for all above stated methods, is performed in a run up condition, where raw
data are acquired in between 900 and 5700 rpm, over 42 channels. An industrial standard analyser is used for data acquisition,
whereas all calculations and post processing are performed using MATLAB . Since the engine of automobile has four cylinders, the

765
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Fig. 2. Comparison of predicted and measured SPL curves, where mount stiness method is used for prediction.

second harmonic of crankshaft rotational frequency is considered as the dominant contributor of low frequency noise [56,57], and
all raw data are processed accordingly, to be used in calculations. Note that, since second order is considered, frequency range of
interest is taken as 30190 Hz at x-axis, for all SPL comparison graphs below.

3.3. Results

Since excitation locations are inaccessible, GTDT and in situ blocked force methods are not applicable (n/a) for the reference
problem. Blocked force method and pseudo force methods require special test rigs and equipment. These methods are not so feasible
for this particular problem, although they can be applied. Results obtained using OTPA method are not given here, but successful
applications are reported in the literature. For OTPA method, post processing steps have a vital importance in the sense of accuracy
(see [58,59] for a detailed framework). Likewise, OPAX results are not reported here, since outcomes change depending on the
number and locations of dened extra measurement points (the number is 2 for this work). Additionally, in the post processing step,
parameters used in identifying forces are obviously user dened. Hence, using same data it is possible to get dierent outcomes in
post processing steps. This situation can be assessed as another trade-o. The time gained in reduced measurement eort can be
better used in additional post processing of data. In what follows, results of calculations carried out for mount stiness, matrix
inversion and diagonal matrix inversion methods are given, where they are compared with actual measurements.
Employing mount stiness method, operational forces are evaluated using Eq. (2). To measure NTFs of Eq. (1) directly, the
engine of the automobile is removed, since to access mount locations is impossible for the case studied. Available frequency
dependent complex dynamic stiness matrix data are put into calculation. Prediction and measurement results for the target
microphone locations are given in Fig. 2.
Matrix inversion method also requires removal of source to achieve both FRF and NTF measurements. In these measurements, a
roving hammer with an appropriate tip for low frequency applications is used, and average of 10 impacts are put into calculation in
Eq. (3), where Hv estimator is adopted due to its nearly perfect (above 0.95) ordinary coherence curves. Nevertheless, ordinary
coherence values do not guarantee that measurements are free of errors, such as leakage and articial damping eects (see Section
1). Although an electromagnetic shaker would provide more accurate FRF estimates, as recommended in e.g. Ref. [11], the test
structure is not appropriate for this type exciter due to lack of enough space. After inverting the accelerance matrix as expressed in
Eq. (3), condition number of matrix is also monitored, which reads values well below 100, in the corresponding frequency range of
interest. Same data are also employed in the diagonal version of matrix inversion method to investigate, whether if less data mean
less measurement error, or not. Results of matrix inversion and diagonal matrix inversion methods are given in Figs. 3 and 4,
respectively. Observing these results, one can conclude that measurement errors are not less important, than the eect of cross talk
of mounts, i.e. o-diagonal terms of Hmj matrix.

4. Consideration of damping

In this section, improvement of the accuracy of FRF based TPA methods is investigated in what concern damping in measured
responses, where a roving hammer is used as an excitation source. As discussed in the preceding sections, hammer tests are prone to
some measurement errors, such as leakage, even applied perfectly. Hence, force and exponential windows are used in hammer tests,
where the former and latter are applied to input and outputs, respectively. Inversion of matrix Hmj further increases these errors,
which results in poor outcomes. Hence, it is reasonable to analyse damping in the estimated FRFs, before inverting the accelerance
matrix.

4.1. Identication of structural and viscous damping

In time domain, the governing equations of motion of a multi degree of freedom (mdof), nite dimensional, linear structure can

766
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Fig. 3. Comparison of predicted and measured SPL curves, where matrix inversion method is used for prediction.

be written as

Mx (t ) + Cx (t ) + (K + j D) x (t ) = f (t ), (11)

where M , C , K and D matrices indicate the mass, viscous damping, stiness, and structural damping of the system, respectively;
j = 1 , f and x are force and displacement vectors, respectively. Assuming harmonic excitation, f (t ) = fe jt , and x (t ) = xe jt , then
Eq. (11) becomes

( K M ) x + jCx + jDx = f,
2
(12)

in the frequency domain. Rearranging, we get

K M 2 + j (C + D) x = f,
(13)

where the statement in curly brackets is the complex dynamic stiness matrix. As stated in Eq. (14), it is nothing but the inverse of
complex receptance type FRF matrix (HC ), which is estimated using hammer, or shaker test. That is

1
(H )C
(
= K M 2 + j C + D . ) (14)

Lee and Kim [60] showed that it is possible to separate damping from the measured FRF as shown in Eqs. (15) and (16):

C 1 1
(H )R = K M 2 H N( ) , (15)
C 1
(H )I = C + D, (16)

where subscripts R and I denote real and imaginary parts, respectively. Here, the real part that represents the mass and stiness
characteristics is also called as normal frequency response function (NFRF), and will be denoted as HN . Since, all structures have
damping up to some extent, NFRF cannot be measured directly; instead, it is calculated using the measured complex FRFs.
Structural and viscous damping can be calculated as

Fig. 4. Comparison of predicted and measured SPL curves, where diagonal matrix inversion method is used for prediction.

767
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

C
^
I ^I

+
(
H I (1) ) +

D 1

= ,
C +
^
2n n ^
I s I sn 2n (


H C
I s )
( )
sn n (17)

here ^I is the identity matrix and s are the spectral line indices. Alternatively, using Eqs. (13) and (15), we have
1
(H ) N x + j (C + D) x = f, (18)

pre-multiplying both sides with HN gives

x + j HN (C + D) x = HN f, (19)

and dening the following transformation matrix, after Chen et al. [61]
G = HN (C + D), (20)

we have
x + jGx = HN f. (21)

From experimental modal analysis, we know that

x = H Cf , (22)

which can be written as

(
x = H CR + j H CI f . ) (23)

Comparing Eqs. (21) and (23), we get

(
H CR + j H CI + jG H CR + j H CI = HN . ) (24)

Rearranging Eq. (24) yields

(
H CR GH CI + j GH CR + H CI = HN , ) (25)

where right hand side of Eq. (25) has only real part, which implies that the imaginary part of the expression on the left hand side
must be zero for all frequencies; hence,
GH CR + H CI = 0. (26)

Substituting Eq. (20) into Eq. (26), we get

1
(
HN C + D = H CI H CR) ( ) . (27)

Finally, structural and viscous damping can be calculated by

C
D
HN 1 HN + H CI (1)

(
H R (1) )
+

=
C
,
+
2n n
HN s HN sn 2n H CI (s )
sn n
(
H C
R (s ) ) sn n (28)

where resulting damping matrices are both symmetric and positive denite.

4.2. Eects of the identied damping on the reference problem

Structural and viscous damping calculations are performed for the reference problem that described in Section 3.2, using Eqs.
(17) and (28). 18 paths described are used to calculate damping in 30190 Hz bandwidth with a 0.5 Hz resolution, i.e. n=18 and
s=321. Results are plotted in Fig. 5, where diagonal elements of structural and viscous damping matrices are shown.
Identied damping characteristics of all paths are examined in detail. It is observed that among diagonal paths, the 15th and
17th exhibit strong damping characteristics. The essential observation related to damping characteristics of these paths is that they
take negative damping values, which means that they create energy, rather than dissipating. This situation indicates a measurement
error, provided that the calculation procedure of damping characteristics is correct.
A similar observation is revealed in Ref. [11], where it is suggested to employ a MIMO testing, instead of an impact hammer test
to resolve such measurement errors. Unfortunately, the case study of present work is not appropriate for a MIMO test, as explained
in Section 3.1. Exponential windowing that introduces articial damping is necessarily used to compensate leakage, through

768
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Fig. 5. Plots of 18 diagonal elements of structural and viscous damping matrices calculated by Eq. (28). Paths 15 and 17 exhibit strong damping characteristics.

hammer test [62]. To quantify the eects of damping on the problem studied, a transfer path analysis is performed through normal
frequency response function matrix, HN (See Eqs. (15) a nd (27). Considering the outcomes of Section 3.3, and to minimise variance
type errors, Eq. (4), which counts only on diagonal elements of Hmj is employed for the analysis. Alternatively, instead of diag (Hmj),
N
diag (H mj), which is populated using normal frequency response functions is put into calculation, i.e.,
+
( )N
f j = diag H mj x j. (29)

The aim is to reveal the eect of damping in the foregoing analysis. Then, identied force vector is used in Eq. (1), to calculate
SPLs at y1 and y2 (i.e. mic.1 and mic.2). As before, predicted and measured SPLs at the targets are compared, where results are
plotted in Fig. 6. As expected, predicted SPL curves take higher values than measured ones, nearly in the whole bandwidth of
interest, although a few exceptions are present.
The foregoing discussion suggests that the damping eect, which is embedded in the measured complex frequency response
functions, is needed to be analysed in the post processing step. The need arises due to errors in measured FRFs, which cannot be
avoided depending on trade-os and physical obstacles faced in application, where expressed in detail, through Section 3.1. It is
apparent that Paths 15 and 17 exhibit large measurement errors, and among them articial damping eects introduced by the
exponential windowing function are dominant. To further inspect the situation, an analysis is conducted as described in the
following:

(1) All diagonal paths are put into calculation in force identication step as suggested in Eq. (4).
(2) Among them, Paths 15 and 17 are selected for a further post processing operation depending on the identied structural and
viscous damping outcomes presented in Fig. 5. Damping information embedded in the measured frequency response functions
of these paths is decontaminated.
mj)), i.e.
(3) Finally, sound pressure level calculations are performed through the manipulated diagonal matrix (diag (H
H1,1 0 0

0

H14,14
diag (H mj) =
N
H15,15

H16,16
N
H17,17
0 H18,18 (30)

The described procedure is successfully applied to the reference problem. The procedure proposed increases the accuracy of
results, as observed in Fig. 7.
Note that, although the proposed procedure provides better accuracy, it introduces new trade-os, and has its own drawbacks.
First of all, to ignore the damping eect in some specic paths seems unrealistic. Another drawback is the time cost of the detailed
post processing step.
On the other hand, the damping eect in the passive side acceleration vector (xj) is not ignored. This vector is measured under
operational conditions, and put into calculations as it is. Hence, only the damping in determined paths is ignored with respect to
measurement errors, which cannot be decoupled.
Removal of the source may impact damping, as well. To gure out this eect, FRFs have to be measured with and without the

769
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Fig. 6. Comparison of predicted and measured SPL curves, where diagonal normal frequency response function matrix method is used for prediction.

source. Unfortunately, the studied case is not suitable to perform measurements with the source, as explained in Section 3.1.
Finally, along with the proposed procedure, all TPA methods that can, or cannot be applied for engine induced low frequency
noise of an automobile are tabulated in Table 1. Notice that as excitation locations are inaccessible, GTDT and in situ blocked force
methods are not applicable (n/a) for the reference problem (See Section 3.3).

5. Conclusion

In transfer path analysis, practitioners often complain of the measurement time, which is increased by the following issues: (i)
removal of the source; (ii) to locate sensors and cables; (iii) to measure frequency response functions (FRFs) and noise transfer
functions (NTFs) by applying a known force (shaker), or a normalised impact force (hammer). Today, engines of automobiles are
installed such that without partially, or in some cases completely removing the engine, even to locate sensors for an operational
measurement is hard, if not impossible. Once the source is removed, and sensors are located, to apply an external force is not an
inconvenience in terms of time cost.
During the population of FRF matrix, unavoidable measurement errors are introduced to the results. Articial damping,
measurement noise, nite record length, phase, variance and bias errors certainly aect the outcomes of transfer path analysis.
Although many of these measurement errors are reported in the literature, damping and its eects are not considered in detail.
Present study suggests considering eects of damping in any TPA framework.
Dynamic stiness matrix is the inverse of experimental FRF matrix. During inversion process, measurement errors of FRFs are
highly amplied in the dynamic stiness matrix, since it is dominated by the weakest modes. Whether measured directly or
reciprocally, all FRFs contain the damping information, and in any method that inverts the FRF matrix, same problem is valid.
Current practice of TPA methods is studied on a reference problem, where the engine induced structure borne noise of a vehicle
is considered. Applicability of methods is investigated, and results are presented. It is observed that some of the methods are not
applicable for the case studied. Normal FRF concept is introduced and eects of structural and viscous damping on the results are
discussed. Based on the well-known matrix inversion method, a new procedure is proposed. In the context of trade-os outlined
through the paper, the accuracy of results is focused. Following the post processing operations described, promising outcomes are
achieved.
In many studies, predictions of proposed methods are compared with traditional transfer path analysis (TPA) results. In fact
traditional TPA has its own assumptions and drawbacks, which motivate new research. In the present study, unlike many reported
results in the literature, comparisons are made between calculated and measured sound pressure levels. Validity is greatest for the

Fig. 7. Comparison of predicted and measured SPL curves, where the described procedure is used for prediction.

770
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

Table 1
Summary of TPA methods that can be used for engine induced low frequency noise of an automobile, where the symbols, , and n/a means yes, no and not
applicable, respectively.

Method Applicability Source removal Independent source Time cost Post process

Blocked force Hard Very high Moderate

Diagonal matrix inversion Moderate High Moderate
GTDT n/a Moderate Moderate
In situ blocked force n/a High Moderate
Matrix inversion Moderate High Moderate
Mount stiness Easy Moderate Low
OPAX Moderate Moderate High
OTPA Easy Low High
Pseudo force Hard High High
Proposed procedure Moderate High High

procedure proposed.

References

[1] K. Wyckaert, H. Van der Auweraer, Operational analysis, transfer paths analysis, modal analysis: tools to understand road noise problems in cars, in:
Proceedings of SAE Noise and Vibration Conference, 1995, pp. 139143.
[2] J. Plunt, Finding and xing vehicle NVH problems with transfer path analysis, Sound Vib. 39 (11) (2005) 1217.
[3] X. Wang (Ed.), Body Structure Noise and Vibration Renement, Wood Head Publishing, 2009.
[4] D. Otte, Development and evaluation of singular value analysis methodologies for studying multivariate noise and vibration problems (Ph.D. thesis), K.U.
Leuven, 1994.
[5] K. Wyckaert, W. Hendricx, Transmission path analysis in view of active cancellation of road induced noise in automotive vehicles, in: Inter-Noise and Noise-Con
Congress and Conference Proceedings, no. 5, Institute of Noise Control Engineering, 1994, pp. 653656.
[6] B.K. Yoo, K.-J. Chang, Road noise reduction using a source decomposition and noise path analysis, SAE, 2005-01-2502.
[7] P.E.F. Padilha, J.R. de Frana Arruda, Comparison of estimation techniques for vibro-acoustic transfer path analysis, Shock Vib. 13 (4) (2006) 459467.
[8] H. Van der Auweraer, P. Mas, S. Dom, A. Vecchio, K. Janssens, P. Van de Ponseele, Transfer path analysis in the critical path of vehicle renement: the role of
fast, hybrid and operational path analysis, SAE Technical Paper, 2007-01-2352.
[9] G. De Sitter, C. Devriendt, P. Guillaume, E. Pruyt, Operational transfer path analysis, Mech. Syst. Signal Process. 24 (2) (2010) 416431.
[10] S. Brandl, H.-H. Priebsch, F. Brandl, W. Biermayer, R. Hldrich, A. Sontacchi, Challenges and new solutions for transfer path analysis, ISMA, Leuven,
September 2008.
[11] G.O. Ozgen, J.H. Kim, Error analysis and feasibility study of dynamic stiness matrix-based damping matrix identication, J. Sound Vib. 320 (1) (2009) 6083.
[12] Hewlett Packard Application Note 243-3, The Fundamentals of Modal Testing, Technical Report, HP Application, 1986.
[13] W. Fladung, Windows used for impact testing, in: Proceedings, SPIEThe International Society for Optical Engineering, 1997, pp. 16621666.
[14] J. Schoukens, Y. Rolain, R. Pintelon, Leakage reduction in frequency response function measurements, IEEE Trans. Instrum. Meas. 55 (6) (2006) 22822287.
[15] R. Pintelon, J. Schoukens, G. Vandersteen, K. Barb, Estimation of nonparametric noise and FRF models for multivariable systemspart 1: theory, Mech. Syst.
Signal Process. 24 (3) (2010) 573595.
[16] R. Pintelon, J. Schoukens, G. Vandersteen, K. Barb, Estimation of nonparametric noise and FRF models for multivariable systems, applicationspart 2:
extensions, Mech. Syst. Signal Process. 24 (3) (2010) 596616.
[17] A. Ribeiro, N. Maia, J. Silva, Experimental evaluation of the transmissibility matrix, in: Proceedings of IMAC, Kissimmee (FL), USA, February 1999.
[18] A. Ribeiro, J. Silva, N. Maia, On the generalisation of the transmissibility concept, Mech. Syst. Signal Process. 14 (1) (2000) 2935.
[19] N. Maia, J. Silva, A. Ribeiro, The transmissibility concept in multi-degree-of-freedom systems, Mech. Syst. Signal Process. 15 (1) (2001) 129137.
[20] K. Noumura, J. Yoshida, Method of transfer path analysis for vehicle interior sound with no excitation experiment, in: Proceedings of FISITA, 2006, p.
F2006D183.
[21] K. Noumura, J. Yoshida, Method of transfer path analysis for interior vehicle sound by actual measurement data, JSAE Paper 55 (20065041) (2006) 712.
[22] P. Gajdatsy, K. Janssens, W. Desmet, H. Van Der Auweraer, Application of the transmissibility concept in transfer path analysis, Mech. Syst. Signal Process. 24
(7) (2010) 19631976.
[23] P. Gajdatsy, L. Gielen, K. Janssens, H. Van der Auweraer, P. Mas, W. Desmet, A novel TPA method using parametric load models: validation on experimental
and industrial cases, in: SAE 2009 Noise and Vibration Conference and Exhibition Technical Papers, 2009.
[24] W. Cheng, Y. Lu, Z. Zhang, Tikhonov regularization-based operational transfer path analysis, Mech. Syst. Signal Process. 75 (2016) 494514.
[25] J. Sanchez, H. Benaroya, Review of force reconstruction techniques, J. Sound Vib. 333 (14) (2014) 29993018.
[26] K. Janssens, P. Gajdatsy, L. Gielen, P. Mas, L. Britte, W. Desmet, H. Van der Auweraer, OPAX: a new transfer path analysis method based on parametric load
models, Mech. Syst. Signal Process. 25 (4) (2011) 13211338.
[27] A. Rezayat, V. Nassiri, B. De Pauw, J. Ertveldt, S. Vanlanduit, P. Guillaume, Identication of dynamic forces using group-sparsity in frequency domain, Mech.
Syst. Signal Process. 70 (2016) 756768.
[28] M. Janssens, J. Verheij, A pseudo-forces methodology to be used in characterization of structure-borne sound sources, Appl. Acoust. 61 (3) (2000) 285308.
[29] M. Janssens, J. Verheij, T. Loyau, Experimental example of the pseudo-forces method used in characterisation of a structure-borne sound source, Appl. Acoust.
63 (1) (2002) 934.
[30] A. Moorhouse, On the characteristic power of structure-borne sound sources, J. Sound Vib. 248 (3) (2001) 441459.
[31] X. Han, Y. Guo, Y. Zhao, Z. Lin, The application of power-based transfer path analysis to passenger car structure-borne noise, Proc. Inst. Mech. Eng. Part D: J.
Autom. Eng. 222 (11) (2008) 20112023.
[32] R. Fulford, B. Gibbs, Structure-borne sound power and source characterisation in multi-point-connected systems, part 1: case studies for assumed force
distributions, J. Sound Vib. 204 (4) (1997) 659677.
[33] J. Park, P. Gu, M.-R. Lee, A. Ni, A new experimental methodology to estimate tire/wheel blocked force for road NVH application, SAE Trans. 114 (6) (2005)
25962602.
[34] A. Gaudin, L. Gagliardini, Recent Improvements in Road Noise Control, Technical Report, SAE Technical Paper, 2007-01-2358, 2007.
[35] O. Guasch, F. Magrans, The global transfer direct transfer method applied to a nite simply supported elastic beam, J. Sound Vib. 276 (1) (2004) 335359.
[36] O. Guasch, Direct transfer functions and path blocking in a discrete mechanical system, J. Sound Vib. 321 (3) (2009) 854874.

771
A. Oktav et al. Mechanical Systems and Signal Processing 85 (2017) 760772

[37] O. Guasch, C. Garca, J. Jov, P. Arts, Experimental validation of the direct transmissibility approach to classical transfer path analysis on a mechanical setup,
Mech. Syst. Signal Process. 37 (1) (2013) 353369.
[38] D.F. Comesaa, J. Wind, H.-E. de Bree, A scanning method for source visualization and transfer path analysis using a single probe, SAE International, 2011-01-
1664.
[39] D. F. Comesaa, K. Holland, J. Wind, H.-E. de Bree, et al., Comparison of inverse methods and particle velocity based techniques for transfer path analysis,
Acoustics 2012 Nantes.
[40] A. Inoue, Y. Tanabe, Development of particle velocity transfer path analysis, in: ASME 2012 Noise Control and Acoustics Division Conference at InterNoise
2012, American Society of Mechanical Engineers, New York City, New York, 2012, pp. 359368.
[41] A. Moorhouse, A. Elliott, T. Evans, In situ measurement of the blocked force of structure-borne sound sources, J. Sound Vib. 325 (4) (2009) 679685.
[42] A. Elliott, A. Moorhouse, T. Huntley, S. Tate, In-situ source path contribution analysis of structure borne road noise, J. Sound Vib. 332 (24) (2013) 62766295.
[43] R. Sottek, B. Philippen, An Unusual Way to Improve TPA for Strongly-Coupled Systems, Technical Report, SAE Technical Paper, 2013-01-1970, 2013.
[44] J. Vilsecker, W. Mack, A. Falkner, R. Hanfstingl, Eects of non-linear modelling of a hydromount in full vehicle simulation, Int. J. Veh. Noise Vib. 7 (2) (2011)
109128.
[45] W.-B. Shangguan, Y. Wei, S. Rakheja, X. Zhao, J.-W. Rong, Y.-J. Wang, A study on calculation method of natural frequency for rubber damped torsional
vibration absorbers, in: ASME 2014 International Mechanical Engineering Congress and Exposition, American Society of Mechanical Engineers, Montreal,
Quebec, 2014, pp. V04BT04A072.
[46] M.V. van der Seijs, D. de Klerk, D.J. Rixen, General framework for transfer path analysis: history, theory and classication of techniques, Mech. Syst. Signal
Process. 68 (2016) 217244.
[47] C. Coster, D. Nagahata, P.J.G. Van der Linden, On the accuracy reciprocal and direct vibro-acoustic transfer-function measurements on vehicles for lower and
medium frequencies, in: ISMA 2010-USD 2010 Conference, Leuven, 2022 September 2010.
[48] T.M. Lewis, The Eects of Dynamic Strain Amplitude and Static Prestrain on the Properties of Viscoelastic Materials, Technical Report, SAE Technical Paper,
911084, 1991.
[49] Q. Leclre, B. Roozen, C. Sandier, Experimental estimation of transmissibility matrices, in: Proceedings of ISMA-USD Conference, Leuven, Belgium, 1719
September 2012.
[50] N. Zafeiropoulos, A. Moorhouse, A. Mackay, U. Senapati, P. Lane, C. Crewe, A comparison of two in-situ transfer path analysis methods, in: RASD 20, 11th
International Conference, Pisa, 13 July 2013.
[51] A. Oktav, G. Anla, C. Ylmaz, Assessment of vehicle noise variability due to engine mounts by means of structural transfer path analysis, in: 7th Automotive
Technologies Congress, OTEKON'14, Bursa, 2014.
[52] S.J. Ahn, W.B. Jeong, W.S. Yoo, Improvement of impulse response spectrum and its application, J. Sound Vib. 288 (4) (2005) 12231239.
[53] V. Ganguly, T.L. Schmitz, Phase correction for frequency response function measurements, Precis. Eng. 38 (2) (2014) 409413.
[54] H. Dishan, Phase error in fast Fourier transform analysis, Mech. Syst. Signal Process. 9 (2) (1995) 113118.
[55] ISO, Acoustics measurement of noise inside motor vehicles, ISO 5128, International Organization for Standardization, Geneva, Switzerland, 1980.
[56] S.-H. Shin, J.-G. Ih, T. Hashimoto, S. Hatano, Sound quality evaluation of the booming sensation for passenger cars, Appl. Acoust. 70 (2) (2009) 309320.
[57] W.-B. Shangguan, X.-A. Liu, Z.-P. Lv, S. Rakheja, Design method of automotive powertrain mounting system based on vibration and noise limitations of vehicle
level, Mech. Syst. Signal Process. 76 (2016) 677695.
[58] J. Yoshida, K. Noumura, Noise reduction method for running TPA using signicance probability, J. Syst. Des. Dyn. 5 (8) (2011) 15721581.
[59] J. Yoshida, Y. Onishi, Y. Narita, Consideration and verication of accuracy evaluation indices for operational TPA, J. Syst. Des. Dyn. 6 (5) (2012) 616625.
[60] J.-H. Lee, J. Kim, Development and validation of a new experimental method to identify damping matrices of a dynamic system, J. Sound Vib. 246 (3) (2001)
505524.
[61] S. Chen, M. Ju, Y. Tsuei, Estimation of mass, stiness and damping matrices from frequency response functions, J. Vib. Acoust. 118 (1) (1996) 7882.
[62] W. Fladung, R. Rost, Application and correction of the exponential window for frequency response functions, Mech. Syst. Signal Process. 11 (1) (1997) 2336.

772