NSF MW 2015 Technical Papers

2015
Paper
Paper Title Author Details Company Page
No
Comparisons of Ultrasonic and Richard Steven, Josh CEESI Inc. 4
Differential Pressure Meter Kinney and Charlie
1
Responses to Wet Natural Gas Britton
Flow
Ultrasonic Meters in Wet Gas Dennis van Putten, Henk DNV GL 35
Applications Riezebos and René
Bahlmann
2
Jan Peters and Rick de Shell
Leeuw
Joe Shen Chevron
Impact of Using ISO/TR 11583 Emmelyn Graham, NEL 58
for a Venturi Tube in 3-Phase Michael Reader-Harris,
Wet-Gas Conditions Neil Ramsay, Tariq
3
Boussouara, Claire
Forsyth, Les Wales and
Craig Rooney
Evaluating and Improving Wet Alistair Collins, Mark Solartron ISA 74
4 Gas Corrections for Horizontal Tudge and Carol Wade
Venturi Meters
Setting the STANDARD - Terry Cousins and Richard CEESI Inc. 95
5 Integrating Meter Diagnostics Steven
Into Flow Metering Standards
BS7965 2013 - An Overview of Andrew Wrath SICK 121
6 Updates to the Previous (2009)
Edition
Design of a Subsea Fiscal Oil Dag Flølo Statoil ASA 139
7
Export Metering System
A New Methodology for Cost- Astrid Marie Skålvik, Christian Michelsen Research 163
benefit-risk Analysis of Oil Ranveig Nygaard Bjørk, AS
8 Metering Station Lay-outs and Camilla Sætre
Kjell-Eivind Frøysa Christian Michelsen Research
AS/Høyskolen i Bergen
Operational Experiences with Jos G.M. van der Grinten NMi EuroLoop 182
9 the EuroLoop Liquid Bart van der Stap Flowways
Hydrocarbon Flow Facility Dick van Driel Krohne
Remote Metering Monitoring Jean-Paul Couput TOTAL 192
and Smart Metering Room for
10
Cost Effective Operation of
Multiphase Meters
2015
Paper
No
Practical Experiences Obtained Jankees Hogendoorn, Krohne 200
with the Magnetic Resonance Mark van der Zande,
Multiphase Flowmeter Marco Zoeteweij, Rutger
11 Tromp, Lucas Cerioni and
André Boer
Rick de Leeuw Shell Global Solutions
International B.V.
Operation and Maintenance of Neil Sleight and Lorraine Shell 215
Multiphase Flow Meters Coelho
12
(MPFMs) on BC-10 in Casey Brister OneSubsea
Deepwater Brazil
Case Study Results on Natural Mohammed Al-Torairi Saudi Aramco 229
Gas Custody Transfer Chandramohan MC Emerson Process
13 Measurement with Coriolis Management
Meters in Saudi Arabia Marc Buttler Emerson Process
Management - Micro Motion
Gas Mass Flow Metering An Eric Sanford Vortek Instruments 240
14 Alternative Approach Koichi Igarashi Azbil North America
Kim Lewis DP Diagnostics LLC
In-situ Effects on Ultrasonic Gas Phillip Chan GDF Suez E&P Norge AS 270
Flowmeters Øyvind Storli, Sveinung FMC Technologies
Myhr and Atle
15 Abrahamsen
Reidar Sakariassen Metropartner
Kjell-Eivind Frøysa and
Camilla Sætre Christian Michelsen Research
Installation Effects and Flow Camilla Sætre and Anders Christian Michelsen Research 294
Instabilities in Gas Metering Hallanger AS
16 Station with Multipath Kjell-Eivind Frøysa Christian Michelsen Research
Ultrasonic Flow Meters AS/Bergen University College
Philip Chan GDF Suez E&P Norge AS
Realizing a Class 1 Ultrasonic Sebastian Stoof, Andreas SICK AG 308
Gas Meter without HP Ehrlich and Volker
Calibration –An Innovative Herrmann
17
Approach as Contribution to an
Economic Upstream Metering
Solution
Uncertainty in Flow Gas Felipe Matuzenetz Gasmar 321
18 Measurement Systems with Marinho
Ultrasonic Meters
Maximising Economic Recovery Craig Marshall and Alun NEL 346
19 - A Review of Well Test Thomas
Procedures in the North Sea
2015
Paper
No
Tie Backs and Partner Allocation Kjartan Bryne Berg Lundin Norway AS 372
Håvard Ausen , Steinar FMC Technologies Inc.
20 Gregersen, Asbjørn
Bakken, Knut Vannes and
Skule E. Smørgrav
Gross Error Detection: Allan Wilson, Robert Accord 385
Maximising the Use of Data Sibbald and Phillip
21 with UBA on Global Producer III Stockton
(Part2) Neil Corbett Mærsk Oil and Gas
Could Allocation be Rocket Phillip Stockton and Euain Accord 420

Science? Using the Kalman Filter Drysdale
22
to Optimise Well Allocation
Accuracy
Multiphase Flow Quantification Susithra Lakshmanan Oil &Gas Measurement 453
Using Computational Fluid Limited/University of
Dynamics and Magnetic Cambridge
Resonance Imaging Daniel Holland University of Canterbury
23
Andy Sederman University of Cambridge,
Mikhail Gurevich IMS Group
Wes Maru Oil &Gas Measurement
Limited
The Implementation of John Clarke Chevron, 478
Multiphase Meters in High Kelda Dinsdale, Lars Roxar, Emerson Process
Sulphur Environment on TCO’s Anders Ruden, Ståle Management
24
Tengiz Oil Field, Kazakhstan Gjervik, Frode Hugo Aase,
Sturle Haaland and Stig
M. Sigdestad
Experiences with the Eirik Åbro, Jan Eskeland, Statoil ASA 488
Multiphase Meter System Used Geir Sanden, Eivind Lyng
25
for Allocation of Visund South Soldal, Kåre Kleppe and
tie-in Gullfaks C Asbjørn Erdal
DP Meter Diagnostics – Multiple Jennifer Rabone Swinton Technology 505
Poster Field Results with new Andy Smith GDF Suez
4 Turbulence Diagnostic Kim Lewis DP Diagnostics
Techniques
Comparisons of Ultrasonic and Differential Pressure Meter
Responses to Wet Natural Gas Flow
Richard Steven, Josh Kinney & Charlie Britton
Colorado Engineering Experiment Station Incorporated
1. Introduction
Wet natural gas flow metering is important to natural gas producers. Whereas there are
multiphase wet gas meter designs available, due to economic constraints the majority of wet
natural gas flows are still metered by single phase gas flow meter technologies. However,
there is limited independent, neutral, third party published information regarding the direct
comparison of different gas meter design performance in wet gas flow service.
Gas meter manufacturers have to varying extents researched their respective meter’s wet gas
performance, sometimes made limited modifications, and their sales teams promote the pros
and play down the cons to their best advantage. Due to limited knowledge in this specialised
and complex subject, and the lack of published literature directly comparing different meter
types, many operators can find themselves largely reliant on the advice of these salesmen.
However, these salesmen naturally have a professional duty and vested financial interest to
promote their technology over competing technologies regardless of the true performance
comparisons. Furthermore, human nature can also cause the “Marlow’s hammer1” effect,
where salesmen come to genuinely believe their product is best for that particular application
regardless of what independent data may show. Operators would therefore benefit from more
3rd party comparisons of different gas meter technologies in wet gas service.
Two popular gas flow meter designs are the ultrasonic meter (USM) and the Differential
Pressure (DP) meter. The DP meter is one of the most widely used flow meter designs for
both gas and wet gas flow applications. The USM share of the gas market is growing, and
some USM manufacturers are now marketing the USM as a replacement for DP meters in
wet gas flow applications. However, there is limited public literature that directly compares
DP & USM meter wet gas flow performance. In this paper the wet gas flow performance of
ultrasonic, orifice DP & Venturi DP meters are discussed, using 3rd party published
information, and data taken from 8” meters being tested in series at the CEESI Wet Gas Test
Facility. The performance review includes the respective meters liquid induced gas flow rate
prediction biases and the respective liquid loading monitoring systems.
2. Wet Gas Flow Terminology
A wet gas flow is defined by ISO [1, 2] & ASME [3] as any two-phase (liquid and gas) flow
where the Lockhart-Martinelli parameter (XLM) is less or equal to 0.3, i.e. XLM ≤0.3. Note
that this definition covers any combination of gaseous and liquid components.
The Lockhart-Martinelli parameter (equation 1) indicates the relative amount of total liquid
with the gas flow. Note that m g & ml are the gas and liquid mass flow rates respectively
(where ml is the sum of the liquid component flow), and ρg & ρl are the average gas and
liquid densities respectively.
m ρg
X LM = l --- (1)
mg ρl
1
Marlow’s Hammer: “If the only tool you have is a hammer every job looks like a nail”.
1
The term ‘liquid loading’ is widely used as a qualitative term to describe the amount of liquid
with a gas flow.
The gas to liquid density ratio ( DR = ρ g ρ l ) is a non-dimensional expression of pressure.
The gas densiometric Froude number ( Frg ), shown as equations 2, is a non-dimensional
expression of the gas & liquid flow rates, where g is the gravitational constant, D is the meter
inlet diameter and A is the meter inlet cross sectional area.
mg 1 ---- (2)
Frg =
A gD ρ g (ρ l − ρ g )
Single liquid component wet gas flows have one liquid density. Multiphase wet gas flows
have two liquid densities (water and liquid hydrocarbon, ‘LHC’). In this case the liquid
density used to calculate the gas to liquid density ratio and the gas densiometric Froude
number is the average liquid density.
“Water cut” is the ratio of the water to total liquid (i.e. the sum of water and LHC) volume
flow rates when the fluid is at standard conditions. In this paper “water to liquid mass ratio”
(or “WLRm”) is defined as the ratio of the water to total liquid mass flow rates. The use of
mass flow removes the requirement to define the flow conditions. The WLRm is shown as
equation 3, where mw is the water mass flow rate and mlhc is the LHC mass flow rate.
mw
WLRm = --- (3)
mw + mlhc
The average density of a two component liquid mixture is the total combined liquid mass per
unit liquid volume. It is commonly assumed that two liquid components will be
homogenously mixed. This homogenous liquid phase ( ρl ,hom ) is calculated by equation 4.
Note that ρ w & ρ lh are the water and LHC densities respectively. For multiphase wet gas
flows it is this liquid mixture density that is used to calculate the wet gas flow parameters.
ρ w ρlh
ρl , hom =
( )
ρlh WLRm + ρ w (1 − WLRm )
--- (4)
Equation 5 shows the generic DP meter gas mass flow equation, where E is the velocity of
approach (i.e. a geometric constant), At is the minimum cross sectional area, C d is the
discharge coefficient, ε is the expansibility factor and ∆Pg is the differential pressure (DP).
Wet gas flow conditions tend to cause a DP meter to have a positive bias in the gas flow rate
prediction. This is often called an “over-reading” and denoted as “OR”. The DP created by a
wet gas ( ∆Ptp ) is different to when that gas flows alone ( ∆ Pg ). The result is an erroneous, or
“apparent”, gas mass flow rate prediction, mg apparent (see Equation 6). The over-reading is
expressed either as a ratio (equation 7) or percentage (equation 7a) comparison of the
apparent to actual gas mass flow rate.
m g = EAt C d ε 2 ρ g ∆Pg --- (5) mg apparent ≈ EAt Cd ε 2ρ g ∆Ptp --- (6)
mg Apparent ε tp Cd ,tp ∆Ptp ∆Ptp

OR = = ≅ --- (7)
mg εC d ∆Pg ∆Pg
2
 mg   ∆Ptp 
OR % =  Apparent − 1 *100% ≅  − 1 *100% --- (7a)
 mg   ∆Pg 
   
3. Flow Patterns
The liquid dispersion in a wet gas flow (i.e. the ‘flow pattern’) depends on flow conditions.
Flow patterns significantly affect any gas meters wet gas response. The flow pattern is
dictated by liquid properties and the gas dynamic pressure. Low gas dynamic pressure (i.e.
low gas density and / or low gas velocity) means the liquids weight dominates and the liquid
runs on the base of a horizontal pipe (see Figure 1). High gas dynamic pressure (i.e. high gas
density and / or high gas velocity) means the gas dynamic pressure dominates and the liquid
flows as an annular mist flow (see Figure 3). Moderate gas dynamic pressure produces a
transitional flow pattern between stratified and annular mist flow (see Figure 2). Many
horizontal wet natural gas production flows are in the stratified / annular mist transition zone.
Fig 1. Gas/HCL stratified. Fig 2. Gas/HCL transition. Fig 3.Gas/HCL annular.

The flow pattern influences a gas ultrasonic meter as it affects the local gas velocity and the
fluid phases across each individual path. The flow pattern influences a DP meter as it
influences the DP created through the DP meter body.
4. DP & Ultrasonic Meter Wet Gas Response
It is in the interest of both operators and gas meter manufacturers that relatively low cost
simple gas meters can cope with wet gas flow applications. Therefore, there tends to be a
“can do!” attitude to the problem. However, over time this attitude can begin to give a false
perception that wet gas flow metering is an adverse flow condition that various gas meter
technologies can comfortably deal with. In reality wet gas flow is an extremely adverse flow
condition that all gas meters struggle with.
Wet gas flow causes all gas meter technologies severe problems. What constitutes a ‘good’
gas meter response to wet gas flow is wholly subjective. Compared to dry gas flow meter
specifications all gas meter wet gas capabilities are poor. This reality is seldom discussed in
the industry. There tends to be more focus on any positive aspect of a gas meters performance
with wet gas flow while the negative aspects tend to be down played. DP and ultrasonic
meters responses to wet gas flows should be considered in context. The question is not really
which single phase gas meter has the best response to wet gas flow, but rather, which gas
meter design manages to deliver the most useful amount of information when used in wet gas
flow applications.
4a. Differential Pressure (DP) Meter Wet Gas Response
The response of DP meters to wet gas flow has been actively researched for nearly sixty
years. The response of the orifice meter and Venturi meter to wet gas flow is now so well
understood that ISO has published wet gas correction factors for these meters. The existence
3
Fig 4. The Liquid Loading Effect. Fig 5. Gas to Liquid Density Ratio Effect.
Fig 6. Gas Densiometric Froude Number Effect. Fig 7. The Beta Effect.
of an ISO correlation states that the wet gas response of that meter is so well understood, and
is known to be so reproducible, that there can be confidence it can be accurately predicted.
(The lack of a wet gas flow meter correlation for any meter infers the opposite is true.) It is
now known that:
• an increasing liquid loading / Lockhart Martinelli parameter produces an increasing
over-reading (see Figure 4, & Schuster & Murdock [4, 5]). Figure 4 (i.e. a XLM vs.
%OR plot) is now commonly called a ‘Murdock plot’
• the scale of the wet gas over-reading of various generic DP meter designs is related to
the gas to liquid density ratio (see Figure 5, & Chisholm [6, 7 & 8]).
the gas densiometric Froude number (see Figure 6, De Leeuw [9] & Hall et al [10]).
the DP meter beta value (see Figure 7, Stewart et al [11, 12] & Hall et al [10])
the liquid properties (see Figure 8, Steven et al [13] & Reader-Harris et al [14, 15])
the meter / flow orientation (see Britton et al [17] & Graham et al [18])
Between 1997 & 2011 de Leeuw [9], and Hall & Steven et al [10, 16] linked the DP meters
over-reading to the flow pattern. For a given DP meter design, pipe size, beta value & flow
orientation, varying the Lockhart Martinelli parameter, gas to liquid density, gas densiometric
Froude number and /or water to liquid mass ratio means changing the flow pattern, and hence
the DP meter’s over-reading.
4
Fig 8. Influence of Liquid Properties on Venturi Meter Wet Gas Flow Over-Reading.
The response of orifice and Venturi DP meters to wet gas flow is now so well researched and
understood that ISO has technical reports with orifice and Venturi meter correlations. The DP
meter’s wet gas flow response has been found via massed long term multi-company R&D to
be not just repeatable, but highly reproducible.
Orifice Meters
ISO TR 11583 [1] and ISO TR 12748 [2] both give the same orifice meter wet gas
correlation, for 2” to 4” meters, horizontally installed, with gas and liquid hydrocarbon. ISO
TR 12748 [2] includes a modification to this orifice meter correlation that accounts for the
effect of the water to liquid mass ratio. These orifice meter wet gas correlation were
developed and checked over several years by cross industry cooperation (including multiple
operators, meter manufacturers, Joint Industry Projects and test facilities). Figures 9 thru 11
show photographs of various 2” to 4” orifice meter wet gas flow tests carried out by industry
in the last decade. These orifice meter correlations are valid for all paddle plate, single & dual
chamber orifice meter designs that are ISO 5167 compliant. These ISO wet gas orifice meter
correlations are not orifice meter (or DP transmitter) manufacturer type dependent. They
were produced by industry wide collaboration, i.e. by industry for industry, and are freely
available to all. Orifice meter technology has now reached the stage where any 2” to 4”
orifice meter, supplied “off the shelf” by any reputable supplier, has a known wet gas flow
performance that is accurately predicted by a freely available ISO published wet gas
correlation.
Figure 13 reproduces a massed orifice meter wet gas data set (see Steven et al [16]) with and
without the ISO correction factor applied. For a known liquid flow rate the ISO correlation
corrected the data to within 2% uncertainty (to 95% confidence). All data used (from multiple
operators, test facilities and orifice meter manufacturers) was traceable.
Table 1 shows the wide wet gas flow condition ranges for which the ISO orifice meter wet
gas correlation is applicable. The massed data sets from Murdock, Chisholm and others were
not used as they were not traceable. However, that data was also noted to agree with the
traceable data and the ISO correlations.
5
Fig 9. 4” Dual Chamber Orifice Meter Fig 10. 2” Paddle Plate Orifice Meter
Multiphase Wet Gas Flow Tests at CEESI. Multiphase Wet Gas Flow Tests at CEESI.
Fig 11. 4” Paddle Plate Orifice Meter Fig 12. 8” Dual Chamber Orifice Meter
Two-Phase Wet Gas Flow Tests at TUVNEL. Multiphase Wet Gas Flow Tests at CEESI.
Fig 13. Massed 2” to 4” Orifice Meter Wet Gas Data With & Without the ISO Correlaton.
6
Parameter Range
Pressure 6.7 to 78.9 bara
Gas to liquid density ratio 0.0066 < DR < 0.111
Frg range 0.22 < Frg < 7.25
XLM 0 ≤ XLM < 0.55
Inside full bore diameter 1.94” ≤ D ≤ 4.026”
Beta 0.341 ≤ β ≤ 0.683
Gas / Liquid phase Gas / LHC/ Water
Table 1. ISO TR12748 Multiphase Wet Gas Flow Orifice Meter Correlation Flow Range.
Industry has more understanding of orifice meter wet gas flow response than that stated in
these ISO documents. In 1993 Ting [19] showed that at trace liquid loadings an orifice meter
may very slightly under-read the gas flow rate. However, this under-reading is so slight it is
within the uncertainty of most gas meter wet gas correction factors (i.e. 2%) and is therefore
usually ignored.
In 2014 BP (Steven et al [20]) showed 4” wet gas orifice meter data (recorded at CEESI)
where the liquid components included water, hydrocarbon liquid with heavier components
(that would form wax below approximately 970F) and MEG injection. Although the ISO
orifice meter wet gas correlation was extrapolated to different fluid properties it was shown to
operate within the stated uncertainty.
In 2014 CEESI (Steven et al [20]) showed massed CEESI 8”, 0.689β orifice meter wet gas
flow data. In 2015 new data has been added from this meter. Figure 12 shows this meter
installation. Table 2 shows the data set range.
Parameter Range
Pressure 14 ≤ Pressure (bar a) ≤ 77
Gas to liquid density ratio 0.011 < DR < 0.083
XLM 0 ≤ Xlm < 0.275
Inside full bore diameter 0.2027 m ( 7.981 inch)
Beta 0.689 ≤ β ≤ 0.752
Gas / Liquid phase Natural gas / Exxsol D80 / Water
WLRm 0 ≤ WLRm ≤ 1
Table 2. 8” Orifice Meter Wet Gas Test Data Set
As the ISO TR 12748 orifice meter wet gas flow correlation is specifically for 2” ≤ Ф ≤ 4”
the correlation is being extrapolated when used with an 8” orifice meter. The results of
applying the ISO correlation are shown in Figure 14. Extrapolating the 2011 orifice meter
wet gas correlation for use with the 8” orifice meter wet gas flow data produces results
slightly out with the correlations stated uncertainty of 2% at 95% confidence. The ‘corrected’
data show a slight positive bias. With no available alternative it is still beneficial to
extrapolate the correlation, although the uncertainty has increased. When applying the ISO
≤ 4” orifice meter wet gas correlation to an 8” orifice meter the uncertainty was found to be
-2% to +3% to 95% confidence.
Industry knows a lot about an orifice meters reaction to wet gas flow. The orifice meter does
not cause damming problems and the DP signal is pseudo steady (see Steven et al [16]). The
generic orifice meter wet gas flow response is not just repeatable, but reproducible, and
therefore very predictable. For ≤ 8” nominal meter size ISO offers a reliable orifice meter wet
7
Fig 14. All CEESI 8” Orifice Meter Wet Gas Data.
gas correlation which for a known liquid flow rate corrects the gas flow rate prediction to 3%
uncertainty. With the exception of the Venturi meter industry has no equivalent detailed
knowledge of any other gas meter’s reaction to wet gas flow.
Venturi Meters
ISO TR 11583 [1] gives a Venturi meter wet gas correlation, for ≥ 2” horizontally installed
meters with gas and LHC or gas and water. This ISO Venturi meter wet gas correlation was
released after many years of Venturi meter wet gas flow research. Although largely
uncoordinated, this research spanned more than two decades and was conducted by multiple
operators, meter manufacturers, Joint Industry Projects, academics and test facilities. Figures
15 thru 18 show photographs of Venturi meter wet gas flow tests carried out at CEESI.
This ISO wet gas Venturi meter correlation is not Venturi meter (or DP transmitter)
manufacturer type dependent and is valid for all ≥ 2” ISO 5167 compliant Venturi meters. It
was produced by a UK government funded R&D project at TUVNEL. Originally considered
controversial largely due to its release as an ISO correlation before it had been checked with
independent data by 3rd parties, subsequent industry checks are now validating the general
applicability of this correlation.
Figure 19 shows one such validation check, i.e. data from nominal 2”, 4”, 6” & 8” 0.6 beta
ISO compliant Venturi meters tested at the CEESI multiphase wet gas data (i.e. the meters
shown in Figures 15 through 18) with and without the ISO TR 11583 correction factor
applied. The 8” & 4” meters are resident downstream of commercial equipment tests at the
CEESI 8” & 4” multiphase wet gas test facilities respectively. The 2” & 6” meters were
tested as downstream additions to various commercial jobs, and these data sets are therefore
more limited. The ISO TR 11583 correlation is technically applicable only to gas with liquid
hydrocarbon or gas with water. However, it is a simple procedure to interpolate between the
two cases to apply the correlation to multiphase wet gas flows, i.e. gas with liquid
hydrocarbon and water flows. For a known liquid flow rate, within the specified range of the
ISO correlation, all four Venturi meter gas flow rate predictions predicted the gas flow to
< 3% at 95% confidence. This is very close to the correlation uncertainty claim by ISO of 3%
uncertainty for XLM ≤ 0.15, and 2.5% uncertainty for 0.15 < XLM ≤ 0.3. This independent
CEESI data showed that a 3% uncertainty for XLM ≤ 0.15 was achieved, but the uncertainty at
8
Fig 15. 8” Venturi Meter Fig 16. 6” Venturi Meter
Multiphase Wet Gas Flow Tests. Multiphase Wet Gas Flow Tests.
Fig 17. 4” Venturi Meter Fig 18. 2” Venturi Meter

Two-Phase Wet Gas Flow Tests. Multiphase Wet Gas Flow Tests.
the 0.15 < XLM ≤ 0.3 range was still 3%, not 2.5%. At the 0.15 < XLM ≤ 0.3 range the ISO
correlation appears to very slightly over correct for the over-reading. However, an ISO
Venturi meter correlation applicable over a wide range of Venturi meter sizes, that corrects a
Venturi meter wet gas over-reading for a known liquid flow rate to 3% uncertainty is very
useful to industry. Table 3 shows the range of the data sets shown in Figure 19.
Parameter CEESI Test Range ISO TR 11583 Stated Limits

Pressure 14.8 to 77 bara N/A
Gas to liquid DR 0.016 < DR < 0.085 DR > 0.02
Frg range 0.25 < Frg < 7.13 Frg > 3β2.5
XLM 0 ≤ XLM < 0.28 XLM < 0.3
Inlet Diameter 1.939” ≤ D ≤ 7.981” D ≥ 2”
Beta 0.600 0.4 ≤ β ≤ 0.75
Gas / Liquid phase Gas /HCL/ Water Gas / HCL or Gas / Water
Table 3. CEESI 2” to 8” Venturi Meter Wet Gas Test Data Shown in Figure 19 & the ISO TR
11583 Venturi Meter Wet Gas Flow Correlation Flow Ranges.
The 2”, 0.6 beta Venturi meter data shown in Figure 19 (and included in Table 3) was only
part of the total data set recorded from that meter. In Figure 19 / Table 3 the minimum
density ratio (DR) included for all meters is 0.016. The ISO TR 11583 correlation is for DR >
0.02. It was considered reasonable to extrapolate the correlation to allow for this slightly
9
Fig 19. CEESI 2” to 8” Venturi Meter Multiphase Wet Gas Data
With & Without the ISO Correlaton.
lower value of DR. The same was true of the meter diameter limit. Although the correlation is
for diameters > 50mm, the 2” sch 80 meter diameter is 1.938” / 49.2mm. It was considered
reasonable to extrapolate the correlation to allow for the very slightly lower meter size.
However, this meter’s wet gas data set also included data recorded at even lower density
ratios, approximately 0.01. This is half the stated 0.02 minimum and hence it was considered
unreasonable to include this data in the massed data set presented in Figure 19. With a slight
diameter extrapolation and a liquid type interpolation already being applied a significant
additional density ratio extrapolation would not be a fare review.
Figure 20 shows all the 2” meter data. The data is split between the ISO correlation valid
range of DR > 0.02, and the ISO correlation invalid range of DR < 0.02. The ISO correlation
predicted the gas flow rate of the DR > 0.02 data to 3% uncertainty. The extrapolated ISO
correlation predicted the gas flow rate of the DR < 0.02 data to 5.1% uncertainty at 95%
confidence. (The 34 test points at DR < 0.02 showed a correlation uncertainty of 5% at 94%
Fig 20. CEESI 2” Venturi Meter Multiphase Wet Gas Data, Full Data Set,
Inclusive of DR < 0.02.
10
confidence.) This result shows that modest density ratio extrapolation of the correlation does
not cause gross errors in this correlation’s output.
Venturi meter technology has reached the stage where any ≥ 2” ISO compliant Venturi meter,
supplied “off the shelf” by any reputable supplier, has a known wet gas flow performance
that is accurately predicted by a freely available ISO published wet gas correlation.
Industry has more understanding of Venturi meter wet gas flow response than that stated in
these ISO documents. In 2014 Graham et al [18] extended the work of Britton et al [17] on
the influence on wet gas flow over-reading of Venturi meter flow orientation. A Venturi
meter wet gas correlation for vertical up flow was offered by Graham et al. There are other
published vertical up flow Venturi meter correlations, e.g. Xu et al [21]. There are also
various ongoing academic and commercial programmes to model wet gas flow through a
Venturi meter, e.g. van Werven M. et al [22] & Lupeau A et al [23]. Wet gas flow through
Venturi meters has been, and still is, a highly researched important topic.
Industry has learned a lot about Venturi meters reaction to wet gas flow. The Venturi meter
wet gas flow response is not just repeatable, but reproducible, and therefore very predictable.
For ≥ 2” nominal meter size ISO offers a reliable horizontally installed Venturi meter wet gas
correlation. With the exception of the orifice DP meter, industry has no equivalent detailed
knowledge of any other gas meter’s reaction to wet gas flow.
4b Ultrasonic Meters Meter Wet Gas Response
The response of ultrasonic meters to wet gas flow has been sporadically researched by
competing USM manufacturers over the last twenty years. One issue with understanding
USM wet gas flow response is that unlike the orifice & Venturi meter there is no ‘standard’
design. Different USM manufacturers have different designs, some clamp-on, some integral
meters, different number of paths, different locations of paths, different transducer designs
and different software. This means any USM wet gas test result is only applicable to that
particular meter design. This significantly hinders industries development of general USM
wet gas flow performance understanding.
Fig 21 Ultraflow Daniel Ultrasonic Meter Wet Gas Data

In 1996 a Joint Industry Project (JIP) named ‘Ultraflow Wet Gas Development Project”
conducted R&D into the wet gas response of the first generation Daniel senior sonic USM.
11
The results were summarized by Wilson [24]. The test series included NEL laboratory
nitrogen / water tests and natural gas / condensate tests at UK gas terminals. The laboratory
tests had stratified flows and the gas terminal flows were annular mist (or ‘mist’) flows. This
consortium chose to use percentage LVF as the liquid loading parameter. With the stated
laboratory and gas terminal fluids and stated flow conditions this converts to the stratified
flow laboratory tests having a maximum Lockhart Martinelli parameter of approximately
0.04, and the mist flow terminal tests having a maximum Lockhart Martinelli parameter of
approximately 0.14. That is, the tests covered up to about half the modern wet gas flow range
of XLM ≤ 0.3.
The projects results are reproduced here in Figure 21. As with DP meters the different flow
patterns through the 6” horizontally installed British Gas 4 path USM design produced
different wet gas over-readings. The consortium fitted linear lines (i.e. correlations) to the
two separate flow pattern data sets. There is some scatter in the data around the fits. There is
not enough overall data, and not a high enough liquid loading to make any comprehensive
conclusions. There is a noticeable relatively large scatter in over-reading at very low liquid
loadings. There is a distinct change in response between stratified and mist flow. There is no
data for the transitional flow pattern between stratified and mist flow, and in industry that is
the most common horizontal flow pattern. In practice, even if the operator knows the gas is
wet the operator does not easily know what flow pattern exists. A comprehensive wet gas
correlation should account for not knowing the flow pattern. It is not ideal for the operator to
have to know the flow pattern before choosing an appropriate correlation. Nevertheless, this
initial research was interesting and useful, but unfortunately the USM manufacturers then
stopped wet gas flow R&D and focused on other projects.
A 1998-2002 CEESI wet gas JIP investigated the response of various horizontally installed
gas meters to wet gas flow. An early three bounce path Instromet USM design was tested (as
shown in Figure 22). This USM design is significantly different to the meter in the Ultraflow
tests. These results were released by the JIP in PowerPoint form in 2007. During these tests
these natural gas with decane wet gas flow conditions were varied considerably to produce
various flow patterns. Figure 23 shows the JIP data (where the y-axis ‘GMFR Corr’ is the
over-reading ratio, OR). The meter tended to over-read the gas flow as the liquid loading
increased up until a Lockhart Martinelli parameter of approximately 0.07. The over-reading
had a lot of scatter. Beyond this liquid loading the meter response was random.
At XLM < 0.07 the radial position of
installation had a significant effect on the
over-reading. That is, how the USM paths
are orientated relative to the wet gas flow
pattern has a significant impact on how a
USM reacts to wet gas flow. This is
analogous with AGA 9’s recommendation
that a USM for dry gas service needs to be
calibrated with the flow conditioner fixed
at one radial position so that no flow
disturbance changes due to flow
conditioner rotation can produce a flow
Fig 22. 3 Path USM During CEESI JIP. rate prediction bias.
In 2011 Cameron presented a wet gas flow data set from a horizontally installed Cameron
design 8”, 8 path USM at a Brazil Wet Gas Flow Measurement Workshop. Again, this data
set is for a significantly different USM design than the two previous discussed wet gas data
12
Instromet Ultrasonic Meter --- Wet Gas Data
Entire Data Set
1.80
1.60
1.40
GMFR Corr
1.20
1.00
Rotated -- Low Press.

0.80
Rotated -- Med. Press.
Rotated -- High Press.
Vertical -- Low Press.
0.60 Vertical -- Med. Press.
Vertical -- High Press.
0.40
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
Lockhart Martinelli Number
Fig 23. CEESI JIP 3 Path USM Wet Gas Flow Data
sets. This data was released in PowerPoint form only. Cameron chose to mainly show the
data in terms of LVF%. It was stated that the data had Lockhart Martinelli parameters up to
0.3. Two Cameron graphs are reproduced here as Figure 24. The tests were conducted at 20
& 60 Bar. Due to limitations of the test facility used the maximum gas velocity was relatively
low at 12 m/s. It is assumed by these authors that the fluids were Nitrogen and a light
hydrocarbon (Exxsol D80).
Fig 24. Cameron 8”, 8 path USM Design Wet Gas Flow Data.
Figure 24’s left hand graph shows 60 Bar data. There is a general over-reading but obvious
significant scatter due to gas velocity / flow rate. There is no obvious gas velocity trend.
Further analysis is hampered by the limited maximum gas flow rate of the data. Figure 24’s
right hand graph shows a possible pressure effect on the over-reading. However, this is an
isolated data set. There are only two pressures, and no repeat tests. There is not enough data
here to make any comprehensive statement on pressure effects. Figure 24 indicates that this
8” USM meter, while having good dry gas performance, does not have a particularly
predictable wet gas bias. This is a similar issue to that seen in the Ultraflow data in Figure 21.
13
Fig 25. Clamp On Ultrasonic Meter Under Fig 26. Integral Three Bounce Path Ultrasonic
8” Wet Gas Flow Tests at CEESI Under 8” Wet Gas Flow Tests at CEESI
In 2013 a 3rd party asked CEESI to wet gas flow test six meters in series. These were:
• an 8” schedule 40 (7.981”) clamp-on bounce path ultrasonic meter (see Fig 25),
• an 8” schedule 80 (7.625”) 3 bounce path ultrasonic meter (see Fig 26),
• an 8” schedule 80 chordal four path (Westinghouse) ultrasonic meter (see Fig 30),
• an 8” schedule 80, 0.6 beta Venturi meter (see Fig 15),
• a 6” schedule 80, 0.6 beta Venturi meter (see Fig 16), and
• an 8” schedule 40, 0.689 beta orifice meter (see Fig 12).
The data analysis of the three DP meters was included in section 4a. Table 4 shows the wet
gas test range.
Parameter CEESI Test Range
Pressure 17to 70 bara
Gas to liquid DR 0.016 < DR < 0.075
XLM 0 ≤ XLM < 0.16
Inlet Diameter 7.625” ≤ D ≤ 7.981”
Gas / Liquid phase Gas /HCL/ Water
Table 4. CEESI 8” Multiple Flow Meter Wet Gas Flow Test Range.
The clamp-on ultrasonic meter (see Figure 25) had two sets of transducers each producing a
bounce path. One path was a vertical path (with transducers installed at 12 o’clock) and the
other was a horizontal path (with transducers installed at 3 o’clock). The vertical path was
guaranteed to encounter the liquid regardless of the flow pattern. The horizontal path was
likely to encounter some liquid for most wet gas flow conditions. The transducers supplied
were not rated for the lowest pressures of the test and only data at a density ratio of 0.036 was
recorded. The local 20D spool where the clamp-on meter was installed (at mid-stream on the
spool) was schedule 40. There was an expansion / reduction 0.178” step between this spool
and the neighbouring schedule 80 pipes. The possibility of some liquid hold up in this
schedule 40 spool was discussed but considered unlikely as the gas velocities ranged between
5 & 15 m/s.
The dry gas response of the 8” clamp-on meter was good, it matched the facility reference
meters to 1%. However, this clamp-on meter was severely affected by the presence of wet
gas flow. Figure 27 shows the results. The significant gas flow rate prediction errors
produced in the limited data set seem to be random. There was no discernible relationship
14
between the over-reading and the Lockhart Martinelli parameter or gas densiometric Froude
number.
Fig 27. Data from 8” Clamp-On Bounce Path Ultrasonic Meter
Fig 28. Sketch of the Three Bounce Path Ultrasonic Meter Tested at CEESI.
Fig 29. Data from 8”, Three Bounce Path Ultrasonic Meter.
15
The 8”, three bounce path ultrasonic meter tested had two double bounce paths (paths 1 & 3)
and one single bounce path (path 2). A sketch of these paths is given in Figure 28. The left
hand sketch shows the relative position of the paths, although the transducers are spaced
axially along the meter body. The right hand side of Figure 28 attempts to show the axial
nature of paths 1 & 3. The meter shown in Figure 26 is installed (as per design) such that the
single bounce path is positioned close to (but not at) the vertical 12 to 6 o’clock plane.
The bounce path ultrasonic meter design is (like all ultrasonic & DP meters) inherently a
single phase meter. The bounce path concept has pros and cons when compared against other
common single phase gas ultrasonic meters. In the baseline single phase performance checks
this meter had a good performance, matching the facility reference gas flow rate to < 1%.
However, when used in the highly adverse and specialized case of wet gas flow each bounce
path’s coverage of the pipe area almost guarantees that each path will encounter the liquid
regardless of the flow pattern. With every path adversely affected by the presence of liquid
wet gas presents a challenge to this meter. The test results show that this is the case.
The bounce path ultrasonic meter was severely affected by the presence of wet gas flow.
Figure 29 shows the results. For eight of the eighty six wet gas data points (approximately
9% of the data) this meter failed to produce a gas flow rate prediction2. The failures began to
appear at XLM ≥ 0.05. Of the seventy eight test points where the meter gave a gas flow rate
prediction there was significant scatter. The significant gas flow rate prediction errors
produced seem to be random. There is a rough relationship between increasing liquid loading
and increasing ‘over-reading’ (and unfortunately meter failure), but no discernible
relationship between the over-reading and the gas to liquid density ratio or gas densiometric
Froude number.
Fig 30. 8” Four Chordal Path Ultrasonic Fig 31. Sketch of a Four Chordal Path
Meter at the CEESI Wet Gas Test Facility Ultrasonic Meter (not to scale)
Figure 30 shows the 8” chordal four path (Westinghouse design) ultrasonic meter installed in
the CEESI wet gas test facility. Figure 31 is a sketch of the path orientation. Each path is at a
set height in the horizontally installed meter body. Hence, for wet gas flows, the lower the
path (i.e. the higher the path #) the more likely it will encounter higher liquid concentrations.
As expected the 8” chordal four path ultrasonic meter was found to have a good dry gas
baseline performance, matching the facility reference gas flow rate to < 1%. This meter was
adversely affected by the presence of wet gas flow. Figure 32 shows the meter performance.
2
For the 9% of wet gas flow conditions where the bounce path ultrasonic meter did not produce a useable
output the meter did not ‘fail’ as such, but produced huge over-readings in the order of > 6e6%. This obviously
erroneous result is effectively a ‘no result’.
16
Fig 32. Data from 8”, 4 Chordal Path Ultrasonic Meter.
Fig 33. Data from 8”, 4 Chordal Path Ultrasonic Meter.
Fig 34. Comparison of Clamp-On, Bounce Path & Chordal Ultrasonic Meter Wet Gas Results
17
There is a general relationship between increasing liquid loading and increasing gas flow rate
error (or ‘over-reading’), however there is considerable scatter in the percentage Over-
Reading vs. Lockhart Martinelli Parameter relationship. Figures 32 & 33 do not show any
clear evidence of gas to liquid density ratio or gas densiometric Froude number effects.
Figure 34 shows the performance of the three different 8” ultrasonic meter designs when they
were tested together at CEESI. The 8” chordal four path ultrasonic meter data has less scatter
than the other ultrasonic meter designs data and it also tends to have lower over-readings.
This may be due to the upper paths of the chordal design tending to encounter less liquid than
the lower paths, or the paths in the other ultrasonic meter designs, and therefore this meter
design has the benefit of some partially operational paths. (For all but the most extreme
pressures and flow rates gravity dictates that even with annular mist flow in horizontal
installations there is a liquid density gradient, i.e. more liquid concentrated at the base of the
pipe / meter body.) The results shown in Figure 32 & 33 are similar to the results released
independently by Cameron for their chordal meter (see Fig 24).
Fig 35. Massed Wet Gas Data from 8”, Chordal Four Path Ultrasonic Meter.
The smaller scatter and lower wet gas over-reading of the chordal 4 path ultrasonic meter is
relative to the performance of the other ultrasonic meters. To review the performance in a
wider context it is necessary to compare the result to other wet gas results from the same
meter (to check for repeatability) and to compare the response to the DP meter. Figure 35
compares the 8” chordal path meter’s performance in these ultrasonic meter comparison tests
to its previous performance when it was resident in the CEESI wet gas flow facility
downstream of various 3rd party equipment tests. The new data is within the wet gas flow
condition range of the previous massed data set, however, clearly there is some scatter in the
chordal path ultrasonic meter wet gas response.
There is a 4” chordal 4 path ultrasonic meter resident in the CEESI 4” multiphase wet gas
flow test facility. This meter is the same design as the 8” chordal 4 path ultrasonic meter
being discussed here. These 4” & 8” meter have been tested over similar wet gas flow
conditions. Figure 36 compares all the 8” chordal 4 path ultrasonic meter data shown in
Figure 35 to a wet gas data set from the 4” meter of the same design. Clearly, at higher
Lockhart Martinelli parameters (i.e. higher liquid loading) there is a significant difference in
over-reading between the 4” & 8” meters. This infers it may not be possible to create a
correction factor for one size of ultrasonic meter and expect it to work on another size.
18
Fig 36. Massed Wet Gas Data from 8”, Chordal Four Path Ultrasonic Meter.
Fig 37. 4” Chordal 4 Path Data Set, Separated Density Ratio.
Fig 38. 4” Chordal 4 Path Data Set, Set DR, Separated Frg.
19
With the exception of the higher over-reading at higher Lockhart Martinelli parameters, the
wet gas performance of the 4” chordal 4 path ultrasonic meter was similar to the 8” meter.
The 4” chordal 4 path ultrasonic meter showed a general relationship between increasing
liquid loading and increasing over-reading. This is shown in Figure 37. However, again, there
was considerable scatter in the percentage Over-Reading vs. Lockhart Martinelli Parameter
relationship. Figures 37 & 38 suggest this scatter cannot be accounted for due to any gas to
liquid density ratio or gas densiometric Froude number effects.
No standards board (or ultrasonic meter manufacturer) has yet published any comprehensive
ultrasonic meter wet gas correlation. The DP meter is the only gas meter design type known
to have a reproducible wet gas response and therefore a corresponding ISO correlation.
5. DP Meter & Ultrasonic Meter Wet Gas Response Comparisons
There is commercial pressure to compare Ultrasonic & DP meter wet gas flow performance.
A sample comparison of the 8” chordal four path ultrasonic & Venturi meters that where
installed in series in the CEESI multiphase wet gas facility (see Figures 30 & 15 respectively)
is shown in Figure 39. Both meters have significant wet gas over-readings. The ultrasonic
meter over-reading tends to be smaller than that of the Venturi meter but this is a relative
term. Both have substantial gas flow rate errors. In most applications such an error will need
to be corrected, and hence it is not the comparison of the uncorrected over-reading that is of
importance, but the availability and performance of respective wet gas correlations. The ‘off
the shelf’ ISO compliant Venturi meter has the ISO TR 11583 Venturi meter correlation
available. For a known liquid flow rate the result of applying the Venturi meter correlation is
included in Figure 39. The gas flow rate prediction error after applying the ISO correlation is
within the ISO stated correlation uncertainty. There are no standards board or manufacturer
produced published USM wet gas correlations. Therefore unlike the Venturi meter there is no
ultrasonic meter wet gas correlation with which to correct this ultrasonic meter’s over-
reading.
A sample comparison of the 8” chordal four path ultrasonic & orifice meters that where
installed in series in the CEESI multiphase wet gas facility (see Figures 30 & 12 respectively)
is shown in Figure 40. Both meters have significant wet gas over-readings. The ultrasonic
meter over-reading tends to be slightly smaller than that of the orifice meter but both have
substantial gas flow rate errors. Again, in most applications such an error will need to be
corrected, and hence it is not the comparison of the uncorrected over-reading that is of
importance, but the availability and performance of respective wet gas correlations. The ‘off
the shelf’ ISO compliant orifice meter has the ISO TR 12748 orifice meter correlation
available. This correlation is technically only for orifice meters ≤ 4” in diameter, but in
industry this correlation can and will be extrapolated to larger orifice meters. For a known
liquid flow rate the result of applying this orifice meter correlation is included in Figure 40.
The gas flow rate prediction error after applying the ISO correlation was found to be -2% to
+3% to 95% confidence. There are no standards board or manufacturer produced published
USM wet gas correlations. Therefore unlike the orifice meter there is no ultrasonic meter wet
gas correlation with which to correct this ultrasonic meter’s over-reading.
When a gas meter is used for low liquid loading wet gas flows the operator may choose to
ignore that the gas flow is not dry and hope that the over-reading is negligible. In such a
scenario the gas meter design with the lowest over-reading would be desirable. Figures 39 &
40 have circled the wet gas liquid loading region of interest in such a scenario. The chordal 4
path ultrasonic meter generally has a lower over-reading than the Venturi meter but the
20
Fig 39. 8” Venturi & Chordal 4 Path Ultrasonic Meter Wet Gas Data Comparisons.
Fig 40. 8” Orifice & Chordal 4 Path Ultrasonic Meter Wet Gas Data Comparisons.
significantly higher level of ultrasonic meter scatter negates any significant advantage in this
respect. The ultrasonic meter can’t be guaranteed to have a lower over-reading in this liquid
loading region. Figure 41 highlights the wet gas over-reading comparison of the 8” orifice &
chordal 4 path ultrasonic meter at XLM ≤ 0.05. The difference in over-reading between the
orifice and ultrasonic meter is more marginal than between the Venturi and ultrasonic meter.
Again, the ultrasonic meter can’t be guaranteed to have a lower over-reading in this liquid
loading region.
These comparisons of the wet gas flow performance of 8” orifice, Venturi and chordal 4 path
ultrasonic meters have also been made with 4” meters. The results and conclusions made on
the 8” data are substantiated by the 4” data. Figure 42 shows sample CEESI 4” data.
21
Fig 41. 8” Orifice & Chordal 4 Path Ultrasonic Meter Performance at Low Liquid Loading.
Fig 42. 4” Orifice & Chordal 4 Path Ultrasonic Meter Wet Gas Data Comparisons.
Figure 42 compares the performance of 4” orifice and chordal 4 path ultrasonic meters.
Again, there is no advantage between using the two type of meters at low liquid loading (with
no correction applied). Again, the ISO complaint 4” orifice meter had the ISO correlation
available, whereas there is no available ultrasonic meter correlation.
In some applications the operator may not know if the gas is wet. Even if wet gas flow is
confirmed, all gas meters with a wet gas correlation require the liquid flow rate (or some
liquid loading parameter) supplied from an external source. The common methods (e.g. test
separator history or tracer dilution techniques) are spot checks. In these cases the operator is
blind to changing liquid loadings and blind to the associated gas flow rate prediction bias
22
between these periodic checks. Therefore, a wet gas monitoring system internal to the meter
is desirable.
It has been suggested by some that the ultrasonic meter has an advantage over DP meters in
wet gas flow service on the grounds that the ultrasonic meter has a comprehensive diagnostic
system that could be used to show changes in liquid loading. It is inherently inferred in these
statements that the DP meter has no diagnostic capabilities, and no method of monitoring
liquid loading. However, this is not correct. The Venturi meter has a very well established
widely used liquid loading monitoring system, and the generic DP meter now also has a
comprehensively proven diagnostic system. Hence, the Venturi, orifice and any generic DP
meter does have the capability to monitor changes on liquid loading.
6. Liquid Loading Monitoring
6a. DP Meter Liquid Loading Monitoring
In 1997 de Leeuw [9] showed that the ratio of a Venturi meter’s permanent pressure loss
(PPL, DPPPL) to the traditional DP (DPt), often called the Pressure Loss Ratio, or ‘PLR’, was
a fixed value for a given Venturi meter geometry with single phase flow. This is confirmed
for orifice & Venturi meters by ISO 5167 Part 2 and Part 4. De Leeuw then showed that the
PLR of a Venturi meter was sensitive to wet gas liquid loading, and hence monitoring the
PLR gives a very simple, effective and reliable Venturi meter liquid loading monitor.
Fig 43. Modified De Leeuw Sketch Showing DP Readings.
Fig 44. De Leeuw 4”, 0.4β Venturi Meter Wet Gas PLR vs. XLM Wet Gas Data.
Figure 43 shows a modified de Leeuw [9] sketch of the Venturi meter with the traditional &
PPL DPs required to read the PLR. Figure 43 also shows a later general DP meter diagnostic
development of also reading the recovered DP (DPr), as described by Steven et al [20, 27] &
Rabone et al [25] & [26]. Figure 44 shows de Leeuw sample wet gas data showing the PLR
vs. XLM relationship.
23
This de Leeuw technique has spawned multiple wet gas meter products. The Expro
‘SmartVent’, the Solartron Dualstream I, the ABB & Krohne wet gas Venturi meters etc. use
this technique to meter the gas and liquid flow rates without liquid loading information being
required from an external source. At present there are no equivalent ultrasonic meter
technology based wet gas meter products. Furthermore, ISO TR 11583 [1] describes this PLR
vs. XLM wet gas metering technique for Venturi meters. ISO TR 11583 even includes some
approximate predictions for predicting the gas & liquid flow rates via combining the
information from the standard meter and the PLR measurement for some Venturi meter
geometries. (Further discussion of this ISO prediction is outside the scope of this paper.)
This concept is applicable to all generic DP meters. The orifice meters photographed in
Figures 9 & 12 have a downstream tap for this liquid monitoring capability, as do all the
Venturi meters in Figures 15 thru 18. Any DP meter operator can monitor for liquid loading
changes via PLR changes. An operator can do this manually, or there are software packages
available, e.g., the patented comprehensive DP meter diagnostic system ‘Prognosis’ (see
Steven et al [20, 27] & Rabone et al [25]) has software that among other capabilities includes
this liquid loading monitoring capability. Prognosis also includes further dynamic pressure
field monitoring techniques that can further diagnose wet gas as the source of the diagnostic
alarm (see Rabone et al [26]). Steven [27] showed that while an orifice meters PLR parameter
can be used to monitor liquid loading, for an orifice meter in particular, the ratio of the
recovered to PPL DPs (or ‘RPR’) is more sensitive to liquid loading than the PLR. The
Prognosis software includes this RPR check (see Rabone [26]) as well as DP stability
monitoring to produce a ‘wet gas flow’ monitoring system.
6b. Chordal 4 Path Ultrasonic Meter Liquid Loading Monitoring
Most ultrasonic meters have a generic comprehensive diagnostic system consisting of the
following diagnostic checks: Speed of Sound, Path Velocity Ratios (usually shown
individually as well as split into Profile Factor & Symmetry information), Path Performance,
Path Turbulence, Path Signal to Noise Ratio and Path Gain. Different manufacturers have
their own methods of displaying these diagnostics. The diagnostic result, i.e. “diagnostic
pattern”, produced by cross referencing these seven diagnostics is said to indicate wet gas
flow. Different manufacturer displays will of course produce different patterns.
There is significantly less literature regarding the generic ultrasonic meter diagnostic
system’s reaction to wet gas flow than there is for a generic DP meters wet gas monitoring
system. This is in part because industry has released significantly less wet gas flow R&D on
ultrasonic meters than DP meters. Another hindrance to understanding ultrasonic meter wet
gas flow response is the variation of ultrasonic meter and diagnostic display designs. Test
data from one ultrasonic meter design is not automatically transferable to other designs.
Furthermore, physically similar ultrasonic meter designs (such as two competing chordal 4
path designs) may have similar diagnostic responses to wet gas flow, but due to difference’s
in the diagnostic display the response may not be obviously similar.
Lansing [28] gives a short discussion on one horizontally installed 4”, chordal 4 path
ultrasonic meter’s diagnostic reaction to wet gas flow. Lansing says “…there are four basic
diagnostic parameter that can be used to identify when liquids are present. They are
Performance, Path Ratios, Path Speed of Sound and Turbulence values.” Figures 45 thru 56
reproduces the Lansing graphs. No pressure or fluid type information was given. The liquid
loading is shown as GVF%. The relationship of the GVF% to Lockhart Martinelli Parameter
is shown by equation 8.
24
 ρl 
 
 ρg 
GVF % =   *100% (8)
X + ρ l 
 LM ρ 
 g 
Fig 45. USM Performance vs. GVF%. Fig 46. USM Speed of Sound vs. GVF%
Fig 47. USM Turbulence vs. GVF%. Fig 48. Path Velocity Ratios vs GVF%.
The horizontal flow pattern can place different amounts of liquid across each path thereby
making each paths reaction to the wet gas flow potentially different. Figs 45 thru 48 have an
unusual non-linear scale, i.e. unit lengths along the x-axis do not represent unit steps in
GVF%. It is important to realise that Figs 45 thru 48 shows post test analysis. The standard
ultrasonic meter diagnostic suites do not plot the diagnostic parameters vs. GVF% (or
alternative liquid loading). In practice the operator will have to decipher the meaning of
the generic ultrasonic meter diagnostic suite output from the standard USM diagnostic output
display only.
Figure 45 shows that the lowest path (Path 4), which will experience the highest
concentration of liquid with most flow patterns, begins to show performance abnormalities at
a GVF of about 99.5%. Path 3 shows a similar problem at GVF < 98%. Figure 46 shows
Path 4 beginning to show Speed of Sound (SOS) abnormalities at a GVF of about 99.5%.
Path 3 shows a similar problem at GVF < 98%. Figure 47 shows Path 4 beginning to show
turbulence abnormalities at a GVF < 99.9%. Path 3 shows a similar problem at GVF < 98%.
Figure 48 shows Path 4 beginning to ‘show’ velocity abnormalities at GVF < 99.5%.
In field operation the ultrasonic diagnostic display shows this result in terms of Path Velocity
Ratios, as reproduced here as Figures 49 to 52. It is difficult for an operator to judge from
these path velocity ratio plots if the diagnostics are suggesting the flow is dry or wet (or has
25
another problem). A better analysis is to use this information to plot Profile Factor (PF) vs.
Symmetry. This is reproduced as Figures 53 thru 56. These plots identify an abnormality in
the expected dry gas response somewhere between 99.9% < GVF < 99.5%. Unfortunately,
with no gas to liquid density ratio information supplied it is not possible to convert this
example to Lockhart Martinelli parameter terms.
In summary, this reproduced example (Lansing [28]) suggests that the ultrasonic diagnostic
system starts to show an unspecified problem with Path 4 failing between 99.9% < GVF <
99.5%. Wet gas flow is one of several possible problems that could cause Path 4 to fail.
When Path 3 also starts to show similar problems at GVF < 98% pattern recognition (by an
experienced ultrasonic meter operator closely monitoring the diagnostics) can then potentially
indicate wet gas flow as the likely problem.
The following example shows a direct comparison of the diagnostic / wet gas monitoring
systems of the respective 8” chordal 4 path ultrasonic, orifice & Venturi meters. A sample
mid wet gas flow condition has been selected for when these meters were tested in series.
Fig 49. Path Velocity Ratios at 100% GVF Fig 50. Path Velocity Ratios at 99.9% GVF
Fig 51. Path Velocity Ratios at 99.5% GVF Fig 52. Path Velocity Ratios at 99.0% GVF
Fig 53. PF vs. Symmetry 100% GVF Fig 54. PF vs. Symmetry 99.9% GVF
26
Fig 55. PF vs. Symmetry 99.5% GVF Fig 56. PF vs. Symmetry 99.0% GVF
6c. CEESI 8” DP Meter & USM Wet Gas Monitoring Example
Natural gas & Exxsol D80 (a light LHC) at a nominal 47 Bar(a), 350C flowed at 15.7 kg/s.
The gas to liquid density ratio was 0.045 and the gas densiometric Froude number was 2.2.
The GVF% & Lockhart Martinelli parameter range is shown in Table 5.
GVF % 100 99.87 99.64 99.33 98.89 98.53
Xlm 0 0.006 0.017 0.032 0.053 0.070
Table 5. GVF% vs. Lockhart Martinelli Parameter.
Fig 57. 8” Venturi Meter PLR vs. XLM Fig 58. 8” Orifice Meter RPR vs. XLM
Gas / LHC, DR 0.045, Frg 2.2. Gas / LHC, DR 0.045, Frg 2.2.
Figure 57 shows the wet gas response of the 8”, 0.6β Venturi meter’s PLR vs. XLM. The
Venturi meter PLR is very sensitive to wet gas. Trending a Venturi meters PLR is a simple
yet reliable and well established method for monitoring even small changes in wet gas liquid
loading. Figure 58 shows the wet gas response of the 8”, 0.689β orifice meter’s RPR vs. XLM.
The orifice meter RPR is moderately sensitive to wet gas. Although less sensitive to liquid
loading than the Venturi meters PLR, trending an orifice meters RPR is a simple and reliable
method for monitoring for moderate liquid loading changes. As the DP ratio reading
uncertainty is typically < 1% the Venturi meter can ‘see’ wet gas at XLM < 0.002, while this
orifice meter has a more modest identification threshold of XLM < 0.013.
Figure 59 shows the ultrasonic meter individual path performance vs. liquid loading. Path 4
begins to show an abnormality from normal dry gas flow performance at a liquid loading
3 Only a few DP meter malfunctions cause the PLR & RPR values to drift in these particular directions. None of
these other malfunctions have the associated DP fluctuations caused by wet gas flow. Hence, these DP meters
monitoring systems can identify wet gas flow as the specific problem (see Rabone [26]).
27
Fig 59. 8”Performance % vs. XLM Fig 60. 8” SOS % vs. XLM
Fig 61. 8”USM Turbulence % vs. XLM Fig 62. 8” Path Velocity Ratios vs. XLM
Fig 63. 8”USM Path Velocity Ratio Fig 64. 8” USM Path Velocity Ratio
Gas / LHC, DR 0.045, Frg 2.2., Dry Gas Gas / LHC, DR 0.045, Frg 2.2 XLM 0.017
Fig 65. 8”USM Path Velocity Ratio Fig 66. 8” USM Path Velocity Ratio
Gas / LHC, DR 0.045, Frg 2.2., XLM 0.032 Gas / LHC, DR 0.045, Frg 2.2 XLM 0.07
28
of about XLM ≥ 0.015. Path 3 begins to show this abnormality individual Path Velocity Ratio
at about XLM ≥ 0.03. Figure 60 shows the individual path SOS vs. liquid loading. There is no
significant SOS warning until Path 4 begins to fail at XLM > 0.02. No other path showed any
significant SOS problem across the liquid loading range tested. Figure 61 shows the
individual path turbulence vs. liquid loading. No clear indication of a problem is evident until
Path 4 fails at XLM > 0.032. Across the liquid loading range tested no other paths turbulence
diagnostic check noticed the presence of wet gas. Figure 62 shows the individual Path
Velocity Ratio vs. liquid loading. Again, in this method of presentation it is difficult to see
any liquid loading effect. A marginally better way is to look at the individual path velocity
ratio plots (see Figures 63 thru 66), but much better still is the PF vs. Symmetry plot (see
Figure 67). It can be difficult for an operator to be sure if wet gas flow is the problem
when looking at the Path Velocity Ratio plots alone (i.e. Figures 63 thru 66), although the
failure of path 4 is a potential wet gas flow
indicator. Figure 67 shows a much clearer
trend, although it is possible such a result
may be caused by other malfunctions
causing path 4 to fail. As Path 4 fails at
XLM ≥ 0.032 the PF & Symmetry
calculations stop. The lack of data from
path 4 at XLM > 0.032 indicates failure of
the path. A failed path is in
Fig 67. 8” USM PF vs. Symmetry
Gas / LHC, DR 0.045, Frg 2.2.
itself a diagnostic result. A path fails for a reason, and that reason is obviously causing
adverse effects on the metering system. A path 4 failure does not in itself indicate wet gas,
but flooding due to the presence of wet gas flow is one possibility for such a failure.
The Venturi meter diagnostic system can show a problem exists and suggests wet gas is the
probable cause by a liquid loading of XLM > 0.002. The orifice meter diagnostic system can
show a problem exists and suggests wet gas is the probable cause by a liquid loading of
XLM > 0.01. The chordal 4 path ultrasonic meter tested can show an unspecified problem
exists by a liquid loading of XLM > 0.01. This ultrasonic meter diagnostic system began to
show a specific wet gas pattern by XLM > 0.03.
7. Conclusions
Wet gas flow is an extremely adverse flow condition that significantly degrades the
capabilities of all gas meter technologies. No gas meter can come close to meeting its dry gas
performance specifications if more than trace liquid is present. An ISO wet gas meter
correlation is not just a useful method of correcting a gas meter’s liquid induced over-
reading. For a correlation to be published by ISO there must be a massed data set from
multiple meters and test facilities publicly available. This data set has to include data from
multiple independent 3rd parties to indicate that the meters wet gas performance is not just
repeatable, but reproducible, and therefore predictable. Hence, the existence of an ISO wet
gas correlation for any given meter indicates that industry has a comprehensive understanding
about that meter technologies wet gas flow performance. This is the case with Venturi and
orifice meters. Conversely, the lack of a comprehensive correlation for any given gas meter
technology suggests that industry as yet does not fully understand that technologies wet gas
flow performance. This is the present state of ultrasonic meter wet gas flow research.
29
The DP meter has a simple, effective and established liquid loading monitoring system. This
DP meter liquid loading monitoring system is well known, extensively tested and developed,
described in ISO publications, and has been widely applied for several years. The Venturi
meter liquid loading monitoring system is very effective, showing the presence of very small
liquid loadings. The orifice meter liquid loading monitoring system is moderately effective,
showing the presence of small to moderate liquid loadings. The ultrasonic meter has a single
phase diagnostic system that is potentially capable of identifying some unspecified problem
exists at low liquid loadings, and identifying wet gas as the likely problem at moderate liquid
loadings.
It is not by accident that virtually every wet gas meter on the market (most of which are
developed and owned by parent companies that also own DP and ultrasonic single phase gas
meter products) use DP meter technology and none utilize ultrasonic meter technology. The
ultrasonic meter measures gas flow by multiple discrete velocity measurements across paths.
This can be highly effective in single phase metering applications. However, with the highly
adverse condition of wet gas flow, the liquid’s presence causes these discrete measurements
to fail due to wave scatter, energy absorption, ‘bridging’, path flooding etc. Wet gas causes
ultrasonic meters to lose a portion of the signals needed to measure the flow. That is, in an
application where more information is beneficial to meter the more complex flow ultrasonic
meters tend to produce less information than they can with single phase flow. In contrast, the
DP meters interact with and manipulate the wet gas flow, causing the flow pattern to change
in a reproducible way through the meter body. This causes the DP signals, still clearly
readable, to have a reproducible relationship with the wet gas flow. This is what allows DP
meters to have published ISO correlations.
8. References
1. ISO TC30 TR 11583, “Measurement of Wet Gas Flow by Means of Pressure Differential
Devices Inserted in Circular Cross –Sectional Conduits”.
2. ISO TC193 TR 12748, “Natural Gas – Wet Gas Flow Measurement in Natural Gas
Operations”.
3. ASME MFC Report 19G, “Wet Gas Metering”.
4. Schuster R.A., “Effect of Entrained Liquids On A Gas Measurement”, Pipe Line Industry
Journal, February 1959.
5. Murdock. J.W., “Two-Phase Flow Measurements with Orifices" Journal of Basic
Engineering, Vol.84, pp 419-433, December 1962.
6. Chisholm D. and Leishman J.M., "Metering of Wet Steam", Journal of Chemical &
Process Engineering, pp103-106, July 1969.
7. Chisholm D., "Flow of Incompressible Two-Phase Mixtures through Sharp-Edged
Orifices", Journal of Mechanical Engineering Science, Vol. 9, No.1, 1967.
8. Chisholm D., "Research Note: Two-Phase Flow Through Sharp-Edged Orifices", Journal
of Mechanical Engineering Science, Vol. 19, No. 3, 1977.
9. De Leeuw R., "Liquid Correction of Venturi Meter Readings in Wet Gas Flow," North Sea
Workshop, 1997.
10. Hall A., Stobie G., Steven R., “Further Evaluation of the Performance of Horizontally
Installed Orifice Plate and Cone Differential Pressure Meters with Wet Gas Flows”,
International South East Asia Hydrocarbon Flow Measurement Workshop, Malaysia, March
2008.
11. Stewart. D and Steven R. et al., “Wet Gas Metering with V-Cone Meters”, North Sea
Flow Measurement Workshop 2002, St Andrews, UK.
30
12. Stewart, D.G., “Application of Differential Pressure Meters to Wet Gas Flow”, 2nd
International South East Asia Hydrocarbon Flow Measurement Workshop, March 2003.
13. Steven, R., “Liquid Property and Diameter Effects on Venturi Meters Used with Wet Gas
Flows”, International Fluid Flow Measurement Symposium, Mexico, May 2006.
14. Reader-Harris, M., “Ventuti-Tube Performance in Wet Gas Using Different Test Fluids”,
North Sea Flow Measurement Workshop 2006, St Andrews, UK.
15. NEL Report 2005/206 “Venturi-Tube Performance in Wet Gas Using Different Test
Fluids”, September 2006.
16. Steven R., Stobie G., Hall A. & Priddy R., “Horizontally Installed Orifice Plate Meter
Response to Wet Gas Flows” North Sea Flow Measurement Workshop Oct. 2011, Tonsberg,
Norway.
17. Britton C. et al “A Review of the Parameters Influencing Venturi Meters with Wet Gas
Flows”, North Sea Flow Measurement Workshop 2008, St Andrews, UK. 18. Graham E. &
Reader-Harris M., “Performance of a Vertically Installed Venturi Tube in Wet Gas
Conditions”, North Sea Flow Measurement Workshop 2014, St Andrews, UK. 19. Ting V.C.
et al. “Effect of Liquid Entrainment on the Accuracy of Orifice Meters for Gas Flow
Measurement”, Int. Gas Research Conference, 1995.20. Steven R. & Babajide Adejuyigbe et
al “Expanded Knowledge on Orifice Meter Response to Wet Gas Flows”, North Sea Flow
Measurement Workshop 2014, St Andrews, UK.
21. Xu, L et al “Wet Gas Flow Modelling For a Vertically Mounted Venturi Meter”,
Measurement Science Technology, Volume 23, Article 045301, 2012.
22. van Werven M., van Maanen H.R.E., Ooms G., Azzopardi B.J.. “Modelling wet gas
annular/dispersed flow through a Venturi”. AIChE J 2003;49(6):1383-91.
23. Lupeau A., Platet B., Gajan P., Strzelecki A., Escande J., Couput J.P.. “Influence of the
presence of an upstream annular liquid film on the wet gas flow measured by a Venturi in a
downward vertical configuration”. In: Flow Measurement and Instrumentation. 2006;1-11.
24. Wilson M.B., “The Development and testing of an Ultrasonic Flow Meter for Wet Gas
Applications”, Seminar on the Measurement of Wet Gas”, October 1996, NEL, Glasgow,
UK.
25. Rabone J. et al, “DP Meter Diagnostic Systems – Operator Experience”, North Sea Flow
Measurement Workshop 2012, St Andrews, UK.
26. Rabone J. et al, “Advanced DP Meter Diagnostics – Developing Dynamic Pressure Field
Monitoring (& Other Developments)”, North Sea Flow Measurement Workshop 2014, St
Andrews, UK.
27. Steven R., “The Response of Orifice Meters & Their Diagnostic System to Wet Natural
Gas Flow”, Canadian School of Hydrocarbon Measurement, Calgary, March 2014.
28. Lansing J., “Understanding Gas Ultrasonic Meter Diagnostics – Advanced”, Appalachian
Gas Measurement Short Course, Pittsburgh, Pennsylvania, August 2014
31
Ultrasonic meters in wet gas applications
Dennis van Putten, Henk Riezebos, René Bahlmann (DNV GL)
Jan Peters, Rick de Leeuw (Shell)
Joe Shen (Chevron)
DNV GL Oil & Gas

Energieweg 17, 9743 AN Groningen
dennis.vanputten@dnvgl.com
1 INTRODUCTION
Oil and gas operators today are facing a number of significant measurement challenges in their efforts
to optimize production and generate more from their reservoirs, particularly in wet gas fields.
Traditional flow measurement technologies in wet gas fields for well reservoir management (WRM)
purposes are Venturi meters or orifice plate meters that involve differential pressure (dp)
measurements. The advantage of these technologies is that the flow data can be generated under
almost any wet gas process condition and that correction algorithms are available , refs [7] and [8].
There are some disadvantages in terms of limited rangeability, cost and pressure drop.
Many oil and gas operators are aware of the disadvantages and are investigating other options. In the
past decade ultrasonic technologies, both inline as well as clamp-on, are being successfully applied in
dry gas applications. The main advantages of the ultrasonic technology are the large rangeability (high
turn-down ratio), (nearly) undisturbed flow patterns with absence of pressure drop, no obstructions or
moving parts and powerful diagnostic features. Last decade also the clamp-on ultrasonic technology
has made significant steps ahead in gas measurement applications.
Ultrasonic inline and clamp-on technologies have been applied in wet gas circumstances. Although
promising first results are available, most ultrasonic technologies applied under wet gas conditions are
currently mostly “trial and error” based and mainly focus on the performance in field applications. A
systematic approach towards a correction for liquid fractions has been lacking so far.
Typical sales allocation industry requirements express the need to have allocated gas volumes
determined with a maximum uncertainty of 2%. In practice it can be proven that the operational
uncertainties of the wet gas flow meters are in the range of 2%-5% this can be considered as industry
acceptable. However, a serious issue is to determine possible biases due to wet gas circumstances.
Due to the liquid introduction the flow meters tend to over-read in terms of gas flow which can reach
values of 20-30%. To systematically explore the application of ultrasonic flow meters in wet gas sales
allocation applications, DNV GL has executed a Joint Industry Project (JIP) with E&P companies and
ultrasonic manufacturers. The E&P companies that joined the project were NAM, Shell and Chevron.
Both inline ultrasonic manufacturers: Daniel, Elster-Instromet, Krohne and Sick; and clamp-on
ultrasonic manufacturers: Expro, Flexim and Siemens participated in this JIP.
A main goal of this JIP was to develop a correction algorithm to account for the expected over-reading
of the ultrasonic meters due to the introduction of liquid. To develop the algorithm for the ultrasonic
meters, a good understanding of the multiphase flow physics is needed to predict the behaviour of the
liquid. The fundamental study on multiphase flow has led to a physical model for the liquid hold-up
and supported the development of the test setup and the test matrix. Test data of the participating
ultrasonic meters has been generated during an extensive test campaign and was used to determine
the parameters in the over-reading model. This paper will present the results of the ultrasonic meters
tests and the obtained generic correction algorithm.
2 MULTIPHASE FLOW FUNDAMENTALS FOR WET GAS

The analysis of the multiphase flow equations is used to find analytic solutions of the liquid hold-up
within certain limits, e.g. in stratified flow. Moreover, the analysis will demonstrate which parameters
are important, from a fluid dynamical point of view, and therefore determine the setup of the test
matrix. It will be shown that the number of independent parameters to be evaluated during the
performance test can be reduced by means of a dimensional analysis.
The analysis of the multiphase flow equations is elaborated in Appendix A for a general two-phase flow.
This analysis includes all encountered flow regimes of two-phase flow, where it is assumed that the
liquid phase is well mixed and behaves as an emulsion, see Appendix B for the mixture rules. The next
section considers the limiting case of the general derivation of Appendix A for wet gas.
2.1 Application to steady two phase gas-continuous flow

In the wet gas JIP tests the multiphase flow can be considered as gas-continuous and stationary. No
significant pressure gradients are expected due to the full bore configuration of the ultrasonic meters.
This information will allow for further reduction of the dimensionless numbers due to the absence of
Sl j and Eu j , see Appendix A. In the case of wet gas, the gas and liquid flow equations are scaled by
the same parameter, i.e. the  g usg2 / D . In the remainder of the derivation of the dimensionless
equations the bar on the dimensionless variables will be omitted for clarity.
Applying dimensional analysis and this scaling results in the gas and liquid equation in dimensionless
form

   g ug ug   1
Re g
1
   g  2 g g ,
(1)
F̂rg
2
(2)
  l ul ul   LM    l 
X 1
X 2
LM 2
l g .
Rel F̂rg DR
Also the jump condition can be written in dimensionless form by scaling with  g u sg2 resulting in
 X 2LM 1  (3)
      ng  1  ng ,
 Re l
Re g 
g
We g
 l
where the Reynolds numbers, Re g / l , and the Froude number of the gas, F̂rg , were already found for
the general two-phase equation in Appendix A. In equations (2) and (3), additional dimensionless
groups are identified for wet gas
l u sl (4)
X LM  ,
 g u sg
g (5)
DR  ,
l
 g u sg2 D (6)
We g  ,

which are termed the Lockhart-Martinelli parameter, the density ratio and the gas Weber number,
respectively. It is noted that often the densimetric Froude number is used which is defined as
DR (7)
Frg  F̂rg .
1 - DR
The total set of dimensionless numbers is
Re g , Re l , Frg , X LM , We g , DR . (8)
The scaling of the Froude number with the density ratio does not lead to density ratio independence,
which is also a fundamental requirement of the Buckingham PI theorem.
3 TEST SETUP
The test facility used in this JIP is the MultiPhase Flow Laboratory of DNV GL in Groningen. A general
description of the facility including an explanation of the flow rate reconstruction and uncertainty
model can be found in ref [14]. The facility performs well for wet gas flow conditions and the
uncertainties of the reference gas and liquid flow rates are well within 1%.
The study is limited to the wet gas regime, i.e. Gas Volume Fraction GVF  95% and X LM  0.3 .
The tests in the JIP are performed at pressures between 12 and 32 bara in the temperature range
between 15 and 35°C. The fluids used are natural gas, Exxsol D120 (API 40, 4.8cSt) and salt water
with a salinity of 4%wt. In Figure ‎3-1 an example of a two phase flow regime map is depicted. For wet
gas testing the flow regimes that are typically encountered are stratified (wavy) flow and dispersed
flow.
Figure ‎3-1 : Flow regime map for atmospheric air/water mixtures in horizontal
configuration as a function of the phase volumetric flux (left) and corresponding flow
patterns (right), figures from ref [2].
3.1 Test section design

All ultrasonic meters were installed in a single 6” test line in horizontal orientation, see Figure ‎3-2. To
establish a level playing field it was mandatory to prove equal conditions for each flow meter. This was
realized by taking sufficient upstream length from distorting elements, i.e. pipe elbows and
temperature sensors, to provide fully developed multiphase flow and the same flow regime to each
meter. This was validated by installing two optically accessible pipe sections upstream and
downstream of the test section. The observed flow patterns with the optical pipe sections we used to
validate the theoretical hold-up model. Possible risk of interference of ultrasonic signals between
different flow meters was mitigated by installing the ultrasonic meters in order of alternating
frequencies and taking sufficient distance between them.
Krohne
Expro
Elster
Daniel
Siemens
Flexim
SICK
Flow direction
Figure ‎3-2 : Overview of JIP test setup at DNV GL test facility
3.2 Test matrix

The conventional methodology to define a test matrix is to specify dimension-full quantities like
pressure, temperature and phase flow rates. Since the characteristic behaviour of a multiphase flow is
not directly related to these quantities, a different approach is applied. In the previous section, it was
explained how the multiphase flow equations can be written in dimensionless form and which
dimensionless parameters determine the topology of the multiphase flow. For a gas dominated flow
the expected dominant dimensionless quantities are:
Re g , Rel , Frg , X LM , We g , DR . (9)
The core of the test matrix is constructed in terms of Frg , X LM and DR with values that span the
operational envelop of the test facility. The applied values for the test matrix are:
Frg  [0.7 1.2 1.7 2.2] (10)
X LM  [0.01 0.02 0.04 0.08 0.15 0.3]

DR  [0.01 0.013 0.016 0.02 0.026 0.032]
The Frg variation is chosen to span both the dispersed and stratified regime. The X LM variation is
chosen to cover the wet gas envelop and focus is laid on the low liquid content test points. The Frg
and X LM test points form the basis for the test matrix and the DR is chosen in such a way that the
gas-oil density ratio at a certain pressure corresponds to the gas-water density ratio at a pressure one
increment higher. Therefore, at a certain test point the difference between the properties of the oil
and water can be directed to the viscosity effect, i.e. Re l , or the surface tension effect, i.e. We g . A
distinction between these two dimensionless numbers can be established by changing the temperature
of the gas and oil flow. The viscosity of the oil is a strong function of temperature, whereas the
surface tension varies only slightly. Also mixtures of oil and water are executed leading to high values
of the viscosity near the oil-water inversion point. In total approximately 200 test points were run.
Two output signals were required from the flow meters: Actual volumetric flow rate and speed of
sound. A stable test point with a duration of 10 minutes was logged, from which averages were taken
for the correction algorithm analysis. During the test, the diagnostics of the ultrasonic meter were
logged and based on this data the manufacturer provided a valid or invalid tag to each test point.
4 TEST RESULTS
Before commencing with the wet gas test matrix, a dry gas baseline test was performed to eliminate
systematic deviations at single phase conditions. In the wet gas data only the valid test points will be
presented in this paper and this data will be used for the development of the correction algorithm.
The over-reading is based on volumetric flow rates and is defined as:
QgMUT (11)
OR  ,
Qgref
where QgMUT is the measured gas volumetric flow rate by the ultrasonic meter and Qgref is the actual
gas volumetric flow rate given by the reference system of the test facility, both at line conditions. In
Figure ‎4-1 the over-reading of all ultrasonic meters is plotted as a function of Frg and X LM . In
Figure ‎4-2 the projection of the results as a function of Frg and X LM is plotted separately. The global
behaviour of the ultrasonic meters is indicated by the green lines. Note that due to the projection the
dependence in the other dimension might be misinterpreted as scatter. Also these results are for all
liquid mixtures and all DR .
In Figure ‎4-2 a strong over-reading dependence is observed in terms of X LM , with a maximum of 25-
30%. For low Frg numbers, i.e. stratified flow, the over-reading becomes independent on Frg .
Increasing the Frg number leads to atomization of the liquid which results in a lower liquid hold-up
and consequently in a lower over-reading.

Figure ‎4-1 : Over-reading of all ultrasonic meters as a function of Frg and X LM .
Figure ‎4-2 : Projection of over-reading of ultrasonic meters as a function X LM (left) and

Frg (right). The green solid lines indicate the global behaviour and the green dashed line
indicates the transition point.
5 DEVELOPMENT OF CORRECTION ALGORITHM
The over-reading of the ultrasonic meters is assumed to be directly proportional to the liquid hold-up,
or in terms of the gas void fraction, g ,
QgMUT 1 (12)
OR   .
Qgref g
Since an ultrasonic flow meter measures the velocity, the reconstruction of the volumetric flow rate is
not evident in a wet gas environment. For the development of the correction algorithm, we assume
that the resulting gas velocity profile is flat and that the ultrasonic meter does not correct internally
for the presence of the liquid phase. Also, it is assumed that the velocity measurement itself is not
influenced by the multiphase flow, i.e. the liquid will cause the signal to be attenuated but it will not
have an influence on the speed of sound and therefore produce a valid velocity measurement. The
latter can be proven to be true for sufficiently high frequencies, see refs [9] and [16].
5.1 General construction of the correction algorithm

The procedure of constructing the correction algorithm is based on the principle of separation of
variables. Application of this method results in a set of functions that depend solely on a single
dimensionless number. For the gas void fraction one can write
 g   (X LM ) (Frg )(We g )(DR) (Re g ) (Re l ) . (13)
The function expression in (13) should obey different physical limits. The first set of conditions the
function should satisfy are based on Frg and X LM , since these parameters are expected to be of
main influence
lim  g  1 (14)
X LM  0
lim  g  0
X LM  
lim  g  stratified flow

Frg  0
lim  g  GVF
Frg 
The limits of X LM and stratified limit of Frg are evident, whereas the limit of Frg to infinity needs
further explanation. For large Frg the liquid is assumed to be dispersed and the velocity of the liquid
approaches the gas phase velocity and therefore the gas void fraction will tend to the gas volume
fraction. It should be noted that a liquid film will be present on the wall and this results in a lower
velocity of the liquid phase.
5.1.1 Stratified flow
A model has been developed for the gas void fraction in the stratified regime and is based on the
fundamental equations as described in section 2, see refs [12] and [18]. The model indicates that the
correction algorithm in the stratified flow regime should depend mainly on X LM with some minor
dependence on the ratio of the Reynolds numbers and the density ratio. The model by Van Maanen
presented in ref [11] provides a simplification of stratified flow equations and gives for the gas void
fraction
1 (15)
~m (X LM )  .
1  X LM
The model developed in this study requires the solution of a set of non-linear equations and depends
also on the Reynolds number ratio and the density ratio. The function is simply expressed as
 m (X LM , Re l / Re g , DR) . (16)
It is clear that the Van Maanen expression satisfies the limits in equation (14). The same can be
proven for equation (16). For illustration purposes both expressions are plotted in Figure ‎5-1.
0.95
0.9
 g [-]
0.85
Rel/Reg = 10-2
0.8
Rel/Reg = 10-3
0.75
Rel/Reg = 10-4
0.7
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Xlm [-]
Figure ‎5-1 Gas void fraction from Van Maanen (equation (15)) : thick solid line, and from
equation (16) for different DR: 0.015 (solid line), 0.025(dashed line) and 0.035 (dotted line)
and different Rel/Reg.
Using the data from the JIP testing, it is observed that a systematic error remains in terms of X LM
after correction with the theoretical models in equations (15) and (16), see Figure ‎5-2. The reason
for this error can be found in the theoretical construction of both stratified flow models. The
expression by Van Maanen does not take into account viscous effects, while for low X LM and thus low
liquid hold-up the contact surface of the liquid is large compared to the bulk liquid flow. Viscous effects
will increase the hold-up and therefore produce a larger over-reading. The expression developed by
DNV GL does include viscous stresses, but require as input the hydraulic diameter of the liquid flow.
This diameter is not well defined for low liquid hold-up. Increasing X LM results in a better agreement
with the experimental data.
Figure ‎5-2 : Systematic error in OR after correction by  m for stratified flow (low Frg ),
green line indicates trend
An additional correction is required to improve the predictability of the model. A polynomial expression
capable of correcting this function is given by
 a   XbLM  b2 X LM
1 (17)
0  b1  1 .
b2  0
For both expressions (equations (15) and (16)) the parameters can be fitted to minimize the error,
resulting in
~a   X0LM
.76
 1.44X LM (18)
.
a  X 0.75
LM  1.35X LM
The final expressions for both  -functions are constructed as
  m  a . (19)
5.1.2 Dispersed flow
The limit towards high Frg -numbers is covered in the  (Frg ) function. For low values of Frg , the
expression for stratified should remain unaltered and therefore
 (Frg )  1 (20)
for values of Frg lower than a critical value, denoted by Frg* . The transition of the stratified regime
towards the disperse regime is covered by an exponential function since it is expected that the onset
of the transition is relatively quick whereas the asymptotic behaviour of establishment of the dispersed
flow regime will be slowly attained. The following expression is proposed
 1 Frg  Frg* (21)

 (Frg )  
(1  c1 ) exp[  c 2 (Frg  Frg )]  c1 Frg  Frg*
*
where the constant c1 is determined by the limit
Frg   Frg  

lim  g  lim  (XLM ) (Frg )  GVF ,  (22)
and so
GVF (23)
lim  (Frg )  c1  .
Frg   (X LM )
The c2 constant in equation (21) determines the rate at which the flow regimes transit from stratified
to disperse. This parameter is fitted to the JIP data, resulting in: c2  0.4 . It is clear that due to the
matching of the boundary conditions,  also becomes a function of X LM and DR .
5.1.3 Transition point

The critical Froude number, Frg* , is determined by considering the important factors of the transition
process between stratified and disperse. Typically, Frg*  1.2 for gas-oil flows and Frg*  1.5 for gas-
water flows, see ref [7]. An obvious choice would be to linearly interpolate these values with the
WLR . However, this is not a generic approach. It is known that interface instabilities cause the
break-up of the interface into droplets. Two dominant dimensionless numbers play a role in this
process: Re g and We g . The Re g determines the shear force exerted on the interface by the gas
stream, the We g determines the forces exerted by the surface tension to keep the interface flat. One
can again construct physical limits which the equation for the Frg* should satisfy
lim Frg   (inviscid gas :  g  0) (24)
Re g  
lim Frg  0 (surface tension :   0)

Weg  
lim Frg   (surface tension :   )

Weg  0
Equations (24) indicate that an inviscid flow, g  0 , cannot exert a force on the interface and
therefore the flow will remain stratified, i.e. Frg*   . The same accounts for an infinite surface
tension,  , leading to We g  0 . When the surface tension tends to zero, i.e. We g   , the
interface will instantaneously break-up in small droplets. Combining both dimensionless numbers to
the Ohnesorge number
We g g (25)
Oh g   .
Re g  g D
The Oh g number is often used in jet break-up and atomization problems; see e.g. refs [1] and [15].
The limits defines in equations (24) convert to limits for the Oh g number and a function that satisfies
these limits is given by
 
Frg Oh g  c3Oh -gc4 . (26)
By means of data fitting the coefficients are determined: c3  2.3  10 5 and c4  1.1 . For other
break-up processes it is known that c4 is of the order of unity, see ref [1].
One could apply a simplification of equation (26) by using the fact that Frg*  1.2 for gas-oil flows
and Fr  1.5
*
g for gas-water flows and using a WLR dependent interpolation
Frg  1.2  0.3WLR . (27)
5.2 Generic correction algorithm

The generic correction algorithm is now build-up from equations (19), (21) and (26)
 g   (X LM , Rel / Re g , DR) (Frg , X LM , DR, Oh g ) . (28)
The generic correction algorithm involves all physical processes that were described in the previous
sections. The algorithm is complex and requires the solution of a system of equations and detailed
knowledge of the physical properties of the fluids.
Based on several simplifications proposed in this section one can construct an algebraic expression
which is capable of correcting for the over-reading to a large extent while maintaining the main
physical background of equation (28). For the dependence in terms of X LM , equation (15) combined
with (18) is used. Then, using equation (21) with the simplification proposed in (27) leads to
~  ~(X LM ) Frg  Frg* (29)

g   ~
 (X LM )  GVFexp[ 0.4(Frg  Frg )]  GVF Frg  Frg
* *
where
1 (30)
~(X LM )   X 0LM
.76
 1.44X LM  1  X 0LM
.71
 0.73X LM
1  X LM .
Frg  1.2  0.3WLR
The latter part of the ~ -function is an approximation by means of a data fit to the original expression
which was based on the physical description of the hold-up.
5.2.1 Application of the generic correction algorithm

Using the simplified correction algorithm instead of the generic correction algorithm based on the
physical model results in relative modest differences. The corrected results using the generic model in
equation (28) are depicted in Figure ‎5-3 and Figure ‎5-4. The error is defined as
 OR  OR g  1 . (31)
The reconstruction of the data with the current correction algorithm results in a data set without
additional systematic behaviour in other dimensions of the parameters space. The analysis included
both the dimensional and the dimensionless parameters. Other observed variations are considered to
be uncorrelated and can be considered as scatter. Also small systematic errors between different
manufacturers are observed.
Figure ‎5-3 : Corrected over-reading of all ultrasonic meters as a function Frg and X LM
Figure ‎5-4 : Projection of corrected over-reading of the ultrasonic meters as a function

X LM (left) and Frg (right)
Most of the ultrasonic meters performed similarly as can be observed from Figure ‎5-4, which will be
denoted by the core group of ultrasonic meters. Some meters demonstrate different behaviour, which
can be assigned to differences in path configuration or the ability to already correct partly for the
presence of liquid. Applying the over-reading correction to the core group of ultrasonic meters leads to
low uncertainty on the gas volumetric flow rate. It is observed from Figure ‎5-3 and Figure ‎5-4 that the
dependence on Frg and X LM has been eliminated with exception of the small region near X LM  0 .
The uncertainty of the corrected over-reading is determined on the basis of a 95% confidence interval.
For the calculation only the valid test points are used, i.e. green labelled data. It is important to take
into account these results together with the number of valid test points, since a conservative choice in
approval of the test data may result in low uncertainty values and vice versa. For the core group of
ultrasonic meters approximately 90% of all the data points were labelled green and the uncertainty is
within 4%. Most of the red labelled data points are found at high Frg and high X LM . It is obvious that
the ultrasonic meters have problems at high X LM , since this indicates high liquid volume fractions.
The high Frg numbers can be explained by the transition to annular flow in which the ultrasonic signal
is scattered and no valid data is obtained.
5.2.2 Extension of the correction algorithm

The applicability of the correction algorithm outside the operational envelop covered in the JIP test can
be verified by using the data from the literature study. It should be noted that all the data has been
supplied by the manufacturers (either via papers and conference proceedings or directly) so the data
is not fully objective. Moreover, no validation of the test points has been performed in the same line
as the traffic light system. The data of Cameron in ref [3], Flexim in ref [5], SICK in ref [10] and
Daniel in ref [19] have been added to Figure ‎5-5. A surface plot is added to visualize the shape of the
correction function. The corrected results are given in Figure ‎5-6.
Figure ‎5-5 : Over-reading of ultrasonic meters as a function Frg and X LM (black dots)
including literature data (circles); green from ref [3], red from ref [5], black from ref [10]
and blue from ref [19]. Surface plot is to indicate the shape of the correction algorithm.
Figure ‎5-6 : Projection of corrected over-reading of ultrasonic meters as a function X LM
(left) and Frg (right) including literature data (circles); green from ref [3], red from ref [5],
black from ref [10] and blue from ref [19]
As discussed before, it is difficult to use literature data when all information about the performed test
is not available. Some of the literature data demonstrate large scatter, especially at low Frg -numbers
which is surprising since at those conditions stratified flow is expected. The interesting part of
Figure ‎5-6 is the high Frg -number limit in which the correction algorithm seems to predict the over-
reading behaviour very well.
6 CONCLUSIONS
DNV GL has successfully executed the JIP on ultrasonic meters in wet gas applications. The aim of the
project was to gain knowledge to support the wider use of ultrasonic meters in wet gas and to develop
a correction algorithm to account for the expected over-reading of the ultrasonic meters due to the
introduction of liquid.
To achieve these goals a test program was executed. In general, the ultrasonic meters are capable of
handling the wet gas flow regimes and the resulting over-reading due to the additional liquid phase
has a systematic character. It has been observed that the liquid hold-up in the test section reaches
levels of approximately 30% of the cross sectional area leading to similar values for the over-reading
of the gas flow rate. Due to the cusp shaped interface, the liquid layer near the wall can reach levels
near the centreline of the pipe. Flow meters with ultrasonic paths under the centreline will experience
problems with their signal when traversing the wet gas envelope. Therefore, the consistent over-
reading results are obtained when considering velocity measurements in the domain above and
including the centreline of the pipe. Results show that a generic correction algorithm is feasible as long
as the ultrasonic path configurations are sufficiently similar.
A core group of ultrasonic meters has been identified which demonstrated corresponding over-reading
data. These meters perform well in term of the number of valid test points, typically above 90%. Also,
these meters showed good repeatability in wet gas and are capable of recovering the correct dry gas
flow rate after a wet gas test point.
To develop the correction algorithm a test matrix was constructed in terms of dimensionless numbers.
The analysis of the fundamental equations of multiphase flow indicated which dimensionless numbers
are dominant and formed the basis for the test matrix.
The generic correction algorithm is based on a physical model of the two-phase flow and depends on
the dimensionless numbers. The over-reading of the core group of ultrasonic meters is systematic and
after application of the generic correction algorithm the resulting uncertainty of the gas flow rate is
below 4% for all of the conditions tested. Data from literature is used to verify the correction
algorithm for different diameters and higher pressures and showed reasonably good results. The
ultrasonic meters that do not belong to the core group of ultrasonic meters demonstrate different
behaviour which can be assigned to different path configurations or the ability to already compensate
partly for the presence of liquid.
7 REFERENCES
[1] E. Badens , O. Boutin, and G. Charbit, Laminar jet dispersion and jet atomization in
pressurized carbon dioxide, J. of Supercritical Fluids, 36 81, (2005).
[2] C.E. Brennen, Fundamentals of Multiphase Flow (Cambridge University Press, New York,
2005).
[3] G. Brown. Wet Gas Testing of an 8-path Ultrasonic meter. Wet Gas Flow Measurement
Workshop (2011).
[4] E. Buckingham, On physically similar systems; illustrations of the use of dimensional

equations. Phys. Rev. 4:345–376 (1914).
[5] B. Funck. Report on Testing Carried Out at the CEESI Wet Gas Test Facility in Colorado.
Internal Report (2010).
[6] M. Ishii, Thermo-fluid dynamic theory of two-phase flow (Springer Verlag, Berlin, 2011).
[7] ISO TR 12748, Wet Gas Flow Measurement in Natural Gas Operations, Technical report Draft
(2014).
[8] ISO TR 11583, Measurement of wet gas flow by means of pressure differential devices
inserted in circular cross-section conduits, Technical Report (2012).
[9] L.E. Kinsler, A.R. Frey, A.B. Coppens, and J.V. Sanders, Fundamentals of Acoustics (John
Wiley & Sons, New York, 2000)
[10] J. Lansing, T. Dietz, and R. Steven. Wet Gas Test Comparison Results of Orifice Metering
Relative to Gas Ultrasonic Metering NSFMW 1.2 (2010).
[11] H. de Leeuw, R. Steven, and H. van Maanen, Venturi Meters and Wet Gas Flow, NSFMW
(2011)
[12] R. Malekzadeh, Severe Slugging in gas-liquid two-phase pipe flow, PhD Thesis, TU Delft
(2012).
[13] A. Prosperetti, and G. Trygvasson, Computational Methods of Multiphase Flow (Cambridge

University Press, New York, 2007).
[14] D.S. van Putten, Flow Rate Reconstruction and Uncertainty Model of the MultiPhase Flow
Facility Groningen, GCS-2013-2111C (2014).
[15] M.A. Rahman, Scaling of Effervescent Atomization and Industrial Two-Phase Flow. PhD
Thesis, University of Alberta (2011).
[16] S. Temkin, and R.A. Dobbins, Attenuation and Dispersion of Sound by Particulate-Relaxation
Processes, J. Acoust. Soc. Am. 40:317-324 (1966).
[17] Y. Wang, and M.A. Oberlack. Thermodynamic Model of Multiphase Immiscible Flows with
Moving Interfaces and Contact Line, unpublished, 2011.
[18] S. Wongwises, W. Khankaew, and W. Vetchsupakhun. Prediction of Liquid Holdup in

Horizontal Stratified Two-Phase Flow. J. Sc. Tech. 3:2 (1998).
[19] K.J. Zanker, and G. Brown. The Performance of a Multi-Path Ultrasonic Meter with Wet Gas.
NSFMW 6.2 (2000).
APPENDIX A FUNDAMENTAL MULTIPHASE FLOW EQUATIONS
The derivation of the fundamental equations of multiphase flow starts with the local instantaneous
formulation. This formulation states that the single phase flow equations are valid in the pure phase
sub-regions of the domain under consideration, with exception of the interface, see refs [6] and [13].
The evaluation of the balance equations across the interface results in the so-called jump conditions. A
schematic drawing of a multiphase mixture is given in Figure A-1.
Figure A-1 Control volume of a two phase flow with upper gas phase sub-region and lower
liquid phase sub-region separated by the gas-liquid interface.
In this study the multiphase flow mixture of gas, oil and water will be considered as a two-phase gas-
liquid mixture, assuming that the water-oil mixture is an emulsion, which is valid for relatively high
volumetric flow rates of the liquid phases. The liquid properties are determined by appropriate mixing
rules, see Appendix B.
A general expression for the phase j (gas or liquid) is given by
 j f j
t
 
    j f j u j     j   j j , j  g, l
(32)
with jump condition at the interface
 
j  g ,l
j  n j  ng (33)
where  j , uj and nj are the mass density, velocity and unit outward normal of phase j, respectively.
The values of fj , j and j are given in Table A-1 for mass and momentum conservation, where j
is the stress tensor (containing the pressure and shear stress term), g is the gravitational acceleration,
 is the surface tension and     ng is the surface curvature of the interface. In equations (32)
and (33) it is assumed that no mass transfer occurs between the phases and that the surface tension
is constant.
Table A-1 Values of Parameters for Conservation Equations
fj j j
Mass conservation 1 0 0
Momentum conservation
uj j g
The stress tensor is constructed as  j   p j I   j , where p j is the static pressure, I is the identity
matrix and  j is the shear stress tensor. The final equations for mass and momentum for phase j
become
 j
t
 
    juj  0 ,
(34)
 j u j
t
 
    j u j u j   p j     j   j g
(35)
and the accompanying jump condition from the momentum balance across the interface
p g   
 pl ng   l   g  ng   ng (36)
where nl   n g at the interface. Consider the example of a static spherical bubble in liquid (i.e. no
shear forces), equation (36) demonstrates that the pressure inside the bubble is higher than the
hydrostatic pressure and proportional to the size (i.e. curvature) of the bubble and the surface tension.
A.1 Dimensional analysis of multiphase flow

A dimensional analysis on the fluid dynamical equations and jump conditions for a two phase system
is performed. The dimension-full parameters are scaled by reference values
u j  u j 0u j , t  t0t ,   D1 , p j  p0 p j , g  gg ,  j   j 0  j ,    0 ,  j   j 0  j (37)
where u j 0 denotes the reference velocity of phase j (a convenient choice is the superficial velocity of
phase j denoted by usj ), D is the characteristic length scale (typically the diameter of the pipe) and
 j0 is the dynamic viscosity. The Buckingham PI theorem [4] states that the number of
dimensionless parameters is equal to the number of variables minus the fundamental units (m, kg and
s). In the case of two phase flow, equation (35) will yield 11 variables, so 8 dimensionless numbers
will be obtained.
The momentum equation in dimensionless form can be constructed and dividing by  j 0usj2 / D leads
to
1  j u j
Sl j t
 
    j u j u j   Eu j  p j 
1
Re j
1
  j  2  j g .
(38)
F̂r j
In equation (38) the dimensionless groups are identified
D (39)
Sl j  ,
u j 0t 0
p0
Eu j  ,
 j 0u 2j 0
 j 0u j 0 D
Re j  ,
 j0
u j0
F̂r j  ,
gD
where Sl j is the Strouhal number, Eu j is the Euler number, Re j is the Reynolds number and F̂r j is
the Froude number. The hat on the Froude number is to prevent confusion with the densimetric
Froude number which is denoted as Fr j .
APPENDIX B MIXTURE RULES FOR LIQUID
To obtain the dimensionless numbers for the liquid, appropriate mixing rules need to be defined. It is
assumed that the liquid are well mixed and have the same velocity. The liquid density is determine by
means of the well-known WLR scaling
l  o (1  WLR )   w WLR . (40)
The behaviour of the liquid viscosity is very non-linear and can be described by a simplified version of
the Brinkman equation [6]
l  o (1  WLR )  p , WLR  WLR * (41)
l   w (WLR)  p , WLR  WLR *
where typically p  1.75 and WLR * is called the inversion point, which is found by matching the
two expressions in equation (41)
1 (42)
WLR *  1/ p
 
1   o 
 w 
Determining the mixture surface tension is not evident. Every interface within the three phase mixture
will have its own surface tension. In this study the surface tension is needed to determine the We g -
number for the transition between stratified and dispersed flow. It is observed that when considering
the transition the oil and water phase show a volume averaged character, i.e. near the transition point
a part of the oil flow becomes a dispersion while then remaining water-oil mixture is stratified.
Therefore a volume flow averaging is used for the surface tension
 l   o (1  WLR )   w WLR . (43)

nd
33 International North Sea Flow Measurement Workshop
20-23 October 2015
Impact of Using ISO/TR 11583 for a Venturi Tube in

3-Phase Wet-Gas Conditions
E. Graham, M. Reader-Harris, N. Ramsay, T. Boussouara, C. Forsyth, L. Wales, C.
Rooney
NEL, East Kilbride, UK
1 INTRODUCTION
Venturi tubes are one of the most common types of device used for wet-gas flow
measurement as they are a simple, robust and cost-effective flow meter. The presence
of the liquid in the gas phase causes an increase in the measured differential pressure
and results in the Venturi tube over-predicting the actual amount of gas passing
through the meter. This over-reading is usually ‘corrected’ using available correlations
derived from experimental data to determine the actual gas mass flowrate. This over-
reading trend is observed in all differential-pressure meters. The flowrate of the
liquid, which can be a combination of water and hydrocarbons, is normally
determined by an external means such as from test separator data, tracer experiments
or sampling etc. Information on the liquid flowrate and density is necessary to use the
correlations.
The majority of commercial wet-gas and multiphase meters incorporate a Venturi tube
along with other measurement technology to determine the flow rate of the individual
phases.
The equations used for correcting the Venturi tube over-reading included in the ISO
Technical Report on wet-gas flow measurement using differential pressure devices
(ISO/TR 11583:2012) were developed for 2-phase flows of gas and water or gas and
hydrocarbon liquid. There is a parameter, H, in the equations which accounts for the
effect of the liquid properties on the over-reading. Values for H have been determined
for water and liquid hydrocarbon but not for a mixture with varying water cut.
There has been some criticism voiced from industry over the robustness of the
equations for real field applications partly due to not including the effect of different
water cuts which are common in the field. Until recently there has been limited
openly available 3-phase wet-gas flow data which can be used for deriving equations.
ISO/TR 11583 is a step forward in the standardisation of using Venturi tubes for
measuring wet-gas flows but another stride is necessary to align with industry’s
metrology needs for 3-phase wet-gas measurement using Venturi tubes.
The upgraded high-pressure 3-phase wet-gas facility at NEL was used to collect 3-
phase wet-gas flow data over a range of water cuts from a 4-inch, β=0.6, Venturi tube
installed in a horizontal orientation. The data was used to assess the errors from
applying the equations to determine the gas flowrate from ISO/TR 11583:2012. The
equations within the ISO Technical Report using the pressure-loss measurement
1
nd
20-23 October 2015
across a Venturi tube to determine an estimate of the amount of liquid in wet-gas

flows have been assessed to determine their robustness for 3-phase wet-gas flow.
This paper summarises this research and supports the need to revise the ISO technical
report to cover Venturi tubes for 3-phase wet-gas flow measurement.
It is anticipated that a revision to ISO/TR 11583 for Venturi tubes will follow a more
extensive research programme to develop and validate more robust equations to
correct the over-reading in 3-phase wet-gas conditions.
2 DEFINITIONS OF WET-GAS FLOW
For the research presented in this paper wet-gas flow is defined as the flow of gas and
liquids with a Lockhart-Martinelli parameter, X, in the range 0 < X ≤ 0.3.
mliq ρgas
The Lockhart-Martinelli parameter, X= (1)
mgas ρliq
where mliq and mgas are the mass flowrates of the liquid and gas phase respectively and
ρliq and ρgas are the densities of the liquid and gas phase respectively. In this work the
density of the gas phase is that at the upstream pressure tapping, ρ1,gas.
The gas densiometric Froude number, Frgas, is a dimensionless number directly

proportional to the gas velocity. It is defined as the square root of the ratio of the gas
inertia if it flowed alone to the gravitational force on the liquid phase.
vgas ρ1,gas
Gas densiometric Froude number, Frgas = (2)
gD ρliq − ρ1,gas
where vgas is the superficial gas velocity, g is the acceleration due to gravity and D is
the pipe internal diameter.
mgas
The superficial gas velocity is given by vgas = (3)
ρ1,gas A
where A is the pipe area.
The gas-to-liquid density ratio, DR, is defined as
ρ1,gas
DR = (4)
ρliq
2
nd
20-23 October 2015
The corrected gas mass flowrate, mgas, is given by
EAd Cε wet 2 ρ1, gas ∆p wet

m gas = (5)φ
where E is the velocity of approach factor defined below, Ad is the Venturi-tube throat
area, C is the discharge coefficient in the actual (wet-gas) conditions, εwet is the gas
expansibility in wet-gas conditions, ∆pwet is the actual (wet-gas) differential pressure
and φ is the wet-gas over-reading or correction. εwet was determined from ISO 5167-4
[1] using the actual value of pressure ratio.
1
The velocity of approach factor, E, is defined as E= (6)
1− β 4
where β is the diameter ratio of the Venturi tube (diameter at throat / diameter of
pipe).
3 BRIEF HISTORY OF WET-GAS CORRELATIONS FOR

HORIZONTAL INSTALLATIONS
3.1 Over-reading correlations where X is known
Correlations for the use of orifice plates in wet-gas conditions have existed since the
1960s; the most commonly used correlations are those of Murdock and Chisholm.
These correlations are still used and commonly referred to. These equations have
been applied to other types of differential-pressure meter including Venturi tubes.
Research by Murdock [2] in 1962 on orifice plates in wet-gas conditions stated that
the wet-gas over-reading was dependent on the Lockhart-Martinelli parameter.
3.1.1 Murdock Correlation

Murdock’s correlation gave the over-reading as φ = 1 + 1.26 X (7)
3.1.2 Chisholm Correlation

Chisholm’s research on orifice plates found that the wet-gas over-reading was
dependent on the Lockhart-Martinelli parameter and the gas-to-liquid density ratio [3,
4]. Many of the available correlations for correcting the wet-gas over-reading are
based on the Chisholm model.
Chisholm’s correlation gave the over-reading as φ = 1 + CCh X + X 2 (8)
3
nd
20-23 October 2015
where CCh accounts for the density ratio and is given by the following equation:
n n
 ρliq   ρ1,gas 
CCh =   +  (9)
 ρ1,gas   ρliq 
   
where n = 0.25.
3.1.3 de Leeuw Correlation

The most commonly used correlation for Venturi tubes is that of de Leeuw published
in 1997 [5]. He used data collected from a 4-inch, 0.4 diameter-ratio Venturi tube and
fitted the data using a modification of the Chisholm model. This research found that
the wet-gas over-reading was dependent on the Lockhart-Martinelli parameter, the
gas-to-liquid density ratio and the gas Froude number. De Leeuw used Equations (8)
and (9) but showed that n was a function of the gas Froude number:
n = 0.41 for 0.5 ≤ Frgas < 1.5 (10)
(
n = 0.606 1 − e
−0.746 Frgas
) for Frgas ≥ 1.5 (11)
The de Leeuw correlation or modifications of the Murdock and Chisholm correlations

are used extensively throughout industry to correct for the differential-pressure over-
reading from Venturi tubes and to determine the actual gas flowrate.
However, it is known that the extrapolation of an empirical correlation derived from a

set of data with a limited range of a particular parameter has risks and that this can
increase the measurement errors. This risk can be accounted for by increasing the
uncertainty of the measurements derived from the correlation
3.1.4 Other Research

Since the publication of de Leeuw’s correlation it has been shown by Stewart et al. [6]
that there is a diameter-ratio effect on the wet-gas over-reading. Reader-Harris et al.
[7, 8] and Steven et al. [9, 10] have shown that the liquid properties can have an effect
on the response of differential-pressure meters in wet-gas conditions.
Other correlations have been published and an in-depth review is provided by ASME
[11].
3.1.5 NEL Correlation

In 2009 NEL published a correlation based on a modification of the Chisholm model
and incorporated a wet-gas discharge coefficient term [12, 13]. This correlation found
that the wet-gas over-reading was dependent on the Lockhart-Martinelli parameter,
the gas-to-liquid density ratio, the gas Froude number, diameter-ratio (β) and a
parameter to account for the liquid type (H).
4
nd
20-23 October 2015
This new correlation can be used to determine a value for n in the wet-gas over-
reading based on the Chisholm model. In addition, the discharge coefficient in wet-
gas conditions, which has been found to differ from the value in dry-gas conditions,
can be used with the over-reading to determine the gas mass flowrate in wet-gas
conditions.
The wet-gas discharge coefficient can be derived using this equation:
-0.05Frgas,th  X 
C = 1-0.0463e min 1,  (12)
 0.016 
where the throat Froude number (Frgas,th) is calculated as:
Frgas
Frgas,th = (13)
β 2.5
The value of n was determined to be:
−0.8 Frgas / H
n = max(0.583 − 0.18β 2 − 0.578e , 0.392 − 0.18β 2 ) (14)
where H is a parameter to account for the effect of the liquid properties on the over-
reading. H = 1 for liquid hydrocarbon, H = 1.35 for water at ambient temperature and
H = 0.79 for liquid water in wet-steam flow (hence at elevated temperatures).
The correlation can be used to determine the gas mass flowrate for the following
Venturi tube parameters and wet-gas conditions:
0.4 ≤ β ≤ 0.75
0 < X ≤ 0.3
3 < Frgas,th
0.02 < ρ1,gas/ρliq
D ≥ 50 mm
 3% for X ≤ 0.15
with an uncertainty of 
2.5% for 0.15 < X ≤ 0.3
5
nd
20-23 October 2015
3.2 Correlation Developed For Determining the Wetness
The pressure loss across a Venturi tube is a function of the wetness of the gas. In dry
gas the pressure loss is generally in the range of 5 to 30% of the differential pressure
for a divergent angle of 15o and in the range of 5 to 15% for a divergent angle of 7°. In
wet-gas conditions the pressure loss can be much greater, and under certain
circumstances the ratio of the pressure loss to the differential pressure can be used to
determine X and hence determine the gas mass flowrate without a separate
measurement of the liquid flowrate. Figure 1 shows a schematic of the different
pressure measurements across a Venturi.
Figure 1: Schematic of Venturi tube illustrating total pressure loss (∆∆ϖ) and the
differential pressure (∆ ∆ϖ/∆
∆p) to calculate the pressure loss ratio (∆ ∆p).
(Note diagram is not drawn to correct scale or dimensions)
For a limited range of X it is possible to use the pressure loss ratio to determine the
Lockhart-Martinelli parameter. The formulae in this paper are valid for a Venturi tube
with divergent total angle in the range 7° to 8°.
The pressure loss, ∆ϖ, from the upstream pressure tapping to a tapping a distance
Ldown downstream of the downstream end of the Venturi tube divergent section is
measured. Ldown should be such that
Ldown
max(5,20 β − 7) ≤ ≤9 (15)
D
Then evaluate (this is an iterative procedure)
∆ϖ
Y= − 0.0896 − 0.48β 9 (16)
∆p
and
  ρ1,gas  
Ymax = 0.61exp  −11  − 0.045Frgas / H  (17)
  ρliquid  
   
If Y/Ymax ≥ 0.65 it is not possible to use the pressure loss ratio to determine X.
If Y/Ymax < 0.65 X is evaluated from
6
nd
20-23 October 2015
Y −0.28 Frgas / H
= 1 − exp(−35 X 0.75e ) (18)
Ymax
Limits of use are:
Frgas,th > 4
Frgas
≤ 5.5
H
ρ1,gas
and ≤ 0.09
ρliquid
The uncertainty in C/φ using the additional equations to determine the liquid content
is
 Y
 4% for < 0.6
 Ymax

6% for 0.6 ≤ Y < 0.65
 Ymax
where Y is the increase in the pressure loss ratio due to wetness and Ymax is the
maximum value of Y.
These equations are included in ISO/TR 11583:2012; their derivation is explained in

much more detail in references [13, 14].
3.3 ISO Technical Report for Wet-Gas Flow Measurement
An ISO Technical Report on wet-gas flow measurement using differential pressure

devices (ISO/TR 11583:2012) has been produced for horizontal meter installations
and includes the NEL correlation [14]. The NEL-derived equations for the pressure-
loss ratio to determine X are also included in the ISO Technical report.
4 3-PHASE HORIZONTAL VENTURI TUBE WET-GAS TESTS AT NEL
A 4-inch, β=0.6, Venturi tube with convergent angle 21o and divergent angle 7.5o
(which had previously been tested in 2-phase wet-gas conditions at NEL) was
installed in a horizontal orientation and tested in NEL’s new 3-phase wet-gas flow
measurement facility. Figure 2 shows a photograph of the set-up. The Venturi was
installed approximately 40D downstream of two concentric reducers (8-inch to 6-inch
and 6-inch to 4-inch). The conditions tested are shown in Table 1 over a range of
Lockhart-Martinelli parameters (X) up to 0.3. The hydrocarbon liquid used was a
kerosene substitute called Crownsol D75, which at the nominal facility operating
temperature of 20 oC has a density of 801 kg/m3 and dynamic viscosity of 1.93 cP.
Pure water was used as the third phase, and the gas used was nitrogen.
7
nd
20-23 October 2015
During the commissioning of the 3-phase facility it was noticed that the density output
of the reference liquid Coriolis flow meters was affected by pressure. This is
discussed separately in the Appendix and was corrected in this data set.
Flow
Figure 2: Photograph illustrating the Venturi and instrumentation setup
Table 1 Test Conditions
Density Ratio Gas Froude No.

WC (DR) Pressure (Frgas)
/% /barg
0 0.025 16 1.5, 3
25 0.024 17 1.5, 3
50 0.025 18 2, 3
70 0.025 20 1.5, 3.5
100 0.025 21 1.5, 3
0 0.046 31 1.5, 3.5
25 0.047 33 2.5, 3.5
50 0.047 35 1.5, 2.5, 3.5
70 0.047 37 1.5, 2.5, 3.5
100 0.046 38 1.5, 3.5
0 0.089 61 2.5, 5.5
25 0.084 61 2.5, 4, 5.5
50 0.080 61 2.5, 5.5
70 0.075 61 2.5, 5.5
100 0.072 61 2.5, 5.5
A total of 63 two-phase data points (0% WC and 100% WC) and 119 three-phase data
points (25% WC, 50% WC and 70% WC) were collected.
8
nd
20-23 October 2015
5 RESULTS AND ANALYSIS
5.1 Over-readings where X is known

The parameter H, which accounts for the effect of the liquid properties on the over-
reading when using equations (12) and (14) has been determined as H = 1 for liquid
hydrocarbon (0% WC) and H = 1.35 for water (100% WC) at ambient temperature.
Values have not previously been determined for liquid flows with water cuts between
0% and 100%. Linear interpolation of values for H between 1 and 1.35 has been used
in this study to determine values for water cuts between 0% and 100%, see Table 2.
Table 2 Values of H for Different Liquid Flow Water Cuts (WC)
WC (%) H (-)
0 1
25 1.09
50 1.18
70 1.25
100 1.35
Figure 3 shows the errors in gas mass flow rate uncorrected for the presence of liquid
and corrected using equations (12) and (14), which are included in ISO/TR 11583.
For the 3-phase data “variable H” refers to a linearly interpolated value based on the
water cut.
80
ISO/TR 11583 Uncert limits
70 2-phase data corrected - ISO\TR 11583
Uncorrected 2-phase data

60
3-phase data corrected - ISO\TR 11583
Rel Error in Gas Mass Flow Rate (%)
(variable H)
Uncorrected 3-phase data
50
40
30
20
10
-10
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Lockhart-Martinelli parameter , X (-)
Figure 3 Errors in gas mass flow rate uncorrected for the presence of liquid and
corrected using ISO/TR 11583. “variable H” refers to a linearly interpolated
value based on the water cut.
9
nd
20-23 October 2015
Figure 3 shows the interpolating the value of H based on the water cut is a generally
robust approach to use in applying ISO/TR 11583 to 3-phase flows of water, oil and
gas. For the 2-phase data (0% WC and 100% WC) 95.2% of the data points were
within the uncertainty limits of ISO/TR 11583 and for the 3-phase data 84% of the
data points were within the uncertainty limits of ISO/TR 11583.
Figure 4 compares the data corrected using the well known de Leeuw correlation and
ISO/TR 11583. It should be noted that the de Leeuw correlation was developed using
a Venturi tube with diameter ratio 0.4, whereas these data were collected using a
Venturi tube with diameter ratio 0.6. The magnitude of the over-reading is dependent
on the diameter ratio.[6] For comparison, 66.7% of the 2-phase data points (0% WC
and 100% WC) and 54.6% of the 3-phase data points were within the ISO/TR 11583
uncertainty limits when applying the de Leeuw correlation.
10
All data corrected - de Leeuw

6
Rel Error in Predicted Gas Mass Flow Rate (%)
All data corrected - ISO\TR 11583 variable H

4 ISO/TR 11583 Uncert limits
-2
-4
-6
-8
-10
-12
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Figure 4 Gas mass flow rate errors using the de Leeuw and ISO\TR 11583
correlations for the 2-phase and 3-phase data.
Figure 5 compares all the data (2-phase and 3-phase) corrected using ISO/TR 11583
with H=1.35 and those corrected using ISO/TR 11583 with a linearly interpolated
value for H. For this data set using H=1.35 produces a better overall fit and a higher
percentage of points within the uncertainty limits quoted in ISO/TR 11583 with
97.3% of data using H=1.35 within the uncertainty limits compared with 87.9% when
using a linearly interpolated value for H. Using a value of H=1 rather than H=1.35 for
data with 0% WC, as recommended in ISO/TR 11583, did ensure all the data were
within the ISO/TR 11583 uncertainty limits as shown in Figure 6.
10
nd
20-23 October 2015
4
Rel Error in Predicted Gas Mass Flow Rate using ISO\TR 11583 (%) ISO/TR 11583 Uncert limits
H=1.35 All data
3
Variable H All data
-1
-2
-3
-4
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Figure 5 Gas mass flow rate errors using ISO/TR 11583 correlations with H =
1.35 and H determined by linear interpolation for all the data.
4
Rel Error in Predicted Gas Mass Flow using ISO TR
1
11583 (%)
-1
-2
ISO/TR 11583 Uncert limits
-3 0% WC (H=1)
0% WC (H=1.35)
-4
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Figure 6 Gas mass flow rate errors using ISO/TR 11583 correlations with H =
1.35 and H = 1 for data with 0% water cut, i.e. oil and gas flows.
11
nd
20-23 October 2015
5.2 Using the Pressure Loss Ratio (PLR) to Determine X
Figures 7 and 8 show the relative errors in the gas mass flow rate using the pressure
loss ratio (PLR) method as outlined in ISO/TR 11583 and Section 3.3. The PLR
method can be used in limited conditions (very small values of X) to determine the
wetness defined as X and hence use the equations to correct for the Venturi over-
reading.
20
Rel Error in Predicted Gas Mass Flow using ISO/TR 11583 (%)
ISO/TR 11583 uncert limits
Corrected Gas mass flow rate (variable H)

15
Uncorrected gas mass flow rate
10
-5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Y/Ymax (-)
Figure 7 Relative gas mass flow rate error as a function of Y/Ymax (predicted gas
mass flow rate determined using PLR equations from ISO/TR 11583 and linearly
interpolated value for H based on water cut).
The graphs provide evidence that when the criteria for using the PLR method are met
as defined in ISO/TR 11583, then the method is robust enough to use over a range of
water cuts using a value of H based on the water cut. Using the method can
significantly reduce the errors in the gas mass flow rate.
12
nd
20-23 October 2015
20
Rel Error in Predicted Gas Mass Flow using ISO/TR 11583 (%) ISO/TR 11583 lower uncert limits
Corrected Gas mass flow rate (variable H)

15
Uncorrected gas mass flow rate
10
-5
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Figure 8 Relative gas mass flow rate error as a function of X (provided Y/Ymax <
0.65) (predicted gas mass flow rate determined using PLR equations from
ISO/TR 11583 and linearly interpolated value for H based on water cut).
6 CONCLUSIONS
It has been shown from a limited data set that the current wet-gas over-reading
correlations in ISO/TR 11583 derived for Venturi meters can be used with reasonable
accuracy for 3-phase wet-gas flows with water cuts from 0% to 100% provided the
parameter H is determined by linear interpolation based on the water cut. Using this
method over 85% of the 3-phase data were corrected so that the predicted gas mass
flow rate errors using ISO/TR1183 were within the quoted uncertainty limits of either
2.5% or 3%.
The ISO/TR 11583 equations using the pressure loss ratio to determine X work well
when using a linearly interpolated value for H based on the water cut. Using this
method on this data set reduced the errors in the gas mass flow rate to within the
limits of ISO/TR 11583. This method can reduce the errors in the gas mass flow rate
by a factor of four.
This small data set provides evidence of the robustness of the equations for correcting
Venturi over-readings in ISO/TR 11385 when they are used for 3-phase wet-gas flows
with water cuts from 0% to 100%. Significantly more data will be required before
there can be a revision of ISO/TR 11583 to extend its application to wet-gas flows
with water cuts from 0% to 100% with equations more suitable for more realistic field
conditions.
13
nd
20-23 October 2015
7 ACKNOWLEDGMENTS
The work described in this paper was carried out as part of the National Measurement
and Regulation Office’s Flow Programme, under the sponsorship of the United
Kingdom Department for Business Innovation and Skills.
We would like to thank Chris Mills, Bob Belshaw and Muir Potter for support with
using Coriolis liquid reference meters, Brendan Robson for data acquisition software
modifications, Derek MacLeod for financial management, and Norman Glen and Phil
Mark for assistance with density measurements from Coriolis flow meters.
This paper is published by permission of the Managing Director, NEL.
8 REFERENCES
[1] INTERNATIONAL ORGANIZATION FOR STANDARDIZATION.

Measurement of fluid flow by means of pressure differential devices inserted in
circular-cross section conduits running full – Part 4: Venturi tubes. ISO 5167-4:2003.
Geneva: International Organization for Standardization, 2003
[2] MURDOCK, J. W. “Two-Phase Flow Measurements with Orifices”, Journal

of Basic Engineering, Vol 84, pp 419-433, 1962.
[3] CHISHOLM, D. “Flow of Incompressible Two-Phase Mixtures through

Sharp-Edged Orifices”, Journal of Mechanical Engineering Science, Vol 9, No. 1,
1967.
[4] CHISHOLM, D. “Research Note: Two-Phase Flow through Sharp-Edged

Orifices”, Journal of Mechanical Engineering Science, Vol 19, No. 3, 1977.
[5] De LEEUW, H. “Liquid Correction of Venturi Meter Readings in Wet-Gas

Flow”, in Proc. of 15th North Sea Flow Measurement Workshop, Norway, paper 21,
October 1997.
[6] STEWART, D. G. “Application of Differential Pressure Meters to Wet Gas

Flow”, in Proc. 2nd Int. South East Asia Hydrocarbon Flow Measurement Workshop,
March 2003.
[7] READER-HARRIS, M. J., HODGES, D., GIBSON, J. “Venturi-Tube

Performance in Wet Gas Using Different Test Fluids”, TUV NEL Report 2005/206,
September 2005.
[8] READER-HARRIS, M. J. “Venturi-Tube Performance in Wet Gas Using

Different Test Fluids”, in Proc. 24th North Sea Flow Measurement Workshop, October
2006.
14
nd
20-23 October 2015
[9] STEVEN, R., KINNEY, J., and BRITTON, C. “Liquid property and diameter
effects on Venturi meters used with wet gas flows”, in Proc. 6th International
Symposium on Fluid Flow Measurement, Querétaro, Mexico, May 2006.
[10] STEVEN, R. “Horizontally Installed Differential Pressure Meter Wet Gas

Flow Performance Review”, in Proc. 24th North Sea Flow Measurement Workshop,
October 2006.
[11] AMERICAN SOCIETY OF MECHANICAL ENGINEERS, MFC, Report

19G, “Wet gas flowmetering guideline”, 2008.
[12] READER-HARRIS, M., GRAHAM, E., “An improved model for Venturi tube
over-reading in wet gas”, North Sea Flow Measurement Workshop, 2009.
[13] READER-HARRIS, M., GRAHAM, E., “Venturi tubes in wet gas – improved
models for the over-reading and the pressure-loss ratio method”, South East Asia
Hydrocarbon Flow Measurement Workshop, 2010.
[14] INTERNATIONAL ORGANIZATION FOR STANDARDIZATION.

Measurement of wet gas flow by means of pressure differential devices inserted in
circular cross-section conduits. ISO/TR 11583:2012. Geneva: International
Organization for Standardization, 2012.
APPENDIX
ERRORS IN THE DENSITY OUTPUT OF A CORIOLIS FLOW METER AT

DIFFERENT PRESSURES
During commissioning of the new 3-phase wet-gas flow measurement facility at NEL
it was discovered that the density output of the reference liquid Coriolis flow meters
was providing erroneous reading at higher pressures (facility operates from 10 barg to
63 barg). The Coriolis meters were calibrated at atmospheric pressure by the
manufacturer with an uncertainty in the output density of 0.1%. One of the Coriolis
flow meters was custom calibrated for the density output in NEL’s Industrial
Densitometer Calibration Facility (reference fluid density uncertainty of 0.02%, k=2)
over a range of temperature from 10oC to 40oC and pressures from 1 bara to 71 bara.
Figure A1 shows the relative errors in the uncorrected and corrected density output
from the Coriolis flow meter as a function of pressure. There is a noticeable
temperature effect but this is much smaller than the pressure effect on the density
output. The density output was corrected by fitting coefficients to an equation form
that is commonly used in industrial offshore densitometers. More work on the
pressure and temperature effects on Coriolis flow meter will be conducted at NEL to
reduce the uncertainty in the reference measurements used in NEL’s facility.
15
nd
20-23 October 2015
0.1
0.0
Corrected
-0.1
(%)
-0.2
-0.3 Iso-octane
100 (ρ Coriolis - ρ ref)/ρ ref
10°C
10°C Uncorrected
-0.4 25°C
25°C
40°C
-0.5
40°C
Sebacate
20°C
-0.6
20°C
Toluene
-0.7 20°C
20°C
-0.8
0 10 20 30 40 50 60 70 80
Pressure (bar(a))
Figure A1: The relative errors in the uncorrected and corrected density output
from a 1.5-inch Coriolis flow meter as a function of pressure.
16
EVALUATING AND IMPROVING WET GAS CORRECTIONS
FOR HORIZONTAL VENTURI METERS
Alistair Collins, Dr. Mark Tudge, Carol Wade (Solartron ISA)
“[we are] dwarfs perched on the shoulders of giants...we see more and farther than our predecessors, not because we have keener
vision or greater height, but because we are lifted up and borne aloft on their gigantic stature.” [Metalogicon, John of Salisbury]
ABSTRACT
Solartron ISA have collated an extensive wet gas calibration data set for horizontal Venturi flow meters,
including over 5,000 two-phase and three-phase data points from a range of meter sizes (3” to 10”) and
beta ratio (0.55 to 0.70). This paper utilises the data set to provide an independent evaluation of public
domain wet gas corrections for horizontal Venturi meters, including those published by Murdock, Chisholm,
de Leeuw and in ISO TR 11583.
Furthermore, this analysis has been expanded to suggest improvements to a number of the correlations to
reduce the associated wet gas correction error.
INTRODUCTION
Wet Gas Correction Principles
Over the last three decades, many papers have been published at this workshop demonstrating that
Venturi flow meters are robust and reliable wet gas flow metering solutions. By applying a “simple” wet
gas correction term to the indicated gas mass flow rate, , (as given in ISO 5167-4 [1] and equations (1)
and (5) below), the corrected mass flow rate, , can be found:
(1)
where:
(2)
(3)
(4)
and:
(5)
where is the wet gas correction term.
Wet Gas Calibration Database

Solartron ISA have manufactured wet gas flow meters since the 1990’s, and have tested a significant
number of Venturi meters at many of the commercially available wet gas flow laboratories, including CEESI,
K-Lab, NEL, Porsgrunn, SINTEF and SwRI, primarily for the development and verification of the Dualstream
1
range of wet gas flow meters. This dataset has been collated along with the related project and
metrological data to form a useful tool for evaluating wet gas flow over a wide range of variables.
For the purposes of this paper, the database has then been trimmed to include only flow meters “typically
to ISO 5167-4 for wet gas”. This, for instance, excludes Solartron ISA’s Dualstream 2 Advanced flow meter
calibrations due to their upstream mixer section. It also means that the Venturi meters have tappings at
the top of the meter (i.e. no triple-T or piezometer ring arrangements) and are used without flow
conditioners (see also reference [2]). The cones of each Venturi have a convergent angle of 10.5° (an
included angle of 21.0°) and divergent angle of 7.5°, within the dimensional tolerances of ISO 5167-4, and
the outlet cone is not truncated.
The data for this paper therefore consists of 5,285 test points taken as part of 27 calibration tests on 22
different flow meters. The Venturi pipe size varies from nominally 3” to 10” (66.66 mm to 212.15 mm
internal diameter), with most having a diameter ratio (which is the typical value for
Dualstream 1 Advanced and Dualstream Elite meters), but with two at each of 0.62 and 0.70.
As already mentioned, these tests have been conducted at a range of different loops, and therefore also
over a range of different gases (air, nitrogen, natural gas and methane) and liquids (hydrocarbon liquids -
including condensate, Exxsol D80, Kerosene and Drakesol 205 - and fresh and saline water), an extensive
range of pressures (11 to 235 bar absolute) and moderate range of temperature (13 to 51°C), providing a
wide range of gas densities (11 to 158 kg/m3).
Within this data, 356 points are dry gas (no liquid injection), 3,281 are two-phase (made up of 1,825
hydrocarbon liquid only and 1,456 water only), and 1,648 are three-phase (gas, hydrocarbon liquid and
water).
Wet gas can be quantified by the Lockhart-Martinelli parameter, , which is given for this paper as:
(6)
(A discussion on the origin and forms of the Lockhart-Martinelli parameter is given in [3] Appendix A-1).
Nominally, a flow may be said to be wet gas where . The database for this paper has 59
points where , representing 1.1% of the dataset; however, rather than exclude them from the
analysis, these points shall be used to indicate the performance of the wet gas corrections outside their
typical bounds. Similarly, as there are 4,296 points where , this may imply a bias in the analysis;
therefore the performance of each correlation shall also be shown over the typical range of wet gas
( ) and a narrower bound ( ) for wet gas with less liquids.
Analysis Introduction
This paper summarises a number of wet gas corrections, and examines the possibility of improving the
uncertainty associated with correcting the mass flow rate.
In the first section, the form of the wet gas corrections will be described, and the whole wet gas calibration
dataset (even when outside the specified operational bounds of the correction) will be used to evaluate the
effectiveness of each correction.
In the second section, the dataset is randomised and broken into two parts with 80% of the test points
(4,228 points) used to “improve” the parameters associated with each correlation, and the other 20%
(1,057 test points) to evaluate the effectiveness of these changes.
The paper will be drawn together with discussion and conclusions.
2
The analysis of the correlations will use twice the standard relative error:
(7)
This is similar to twice the relative standard deviation (i.e. roughly 95% coverage) where the data is not
necessarily normally distributed, and therefore does not require an average value, i.e. it takes into account
any bias in the data.
It is also worth noting that each wet gas flow laboratory has an associated uncertainty in its reference flow
measurements. This is typically in the range 0.5% to 0.75% for gas mass flow rate, for instance. This paper
does not attempt to distinguish between different uncertainties for each test point, but instead notes that
the errors in gas mass flow rate (and therefore applied to the wet gas correction term) are notably smaller
than the uncertainties given for each wet gas correction.
WET GAS CORRECTIONS AND PERFORMANCE

Wet Gas Correction Design and Boundary Conditions
The design of a wet gas correction method can be limited by physically determined boundary conditions.
Where no liquid is flowing through the Venturi, the wet gas correction should give a value of 1.0, and
therefore the gas mass flow rate will be the single phase indicated gas mass flow rate as given by
equation (1).
The “dense condition” is where the pressure is so great that the gas and liquid densities have the same
value, and therefore the fluid can be considered a homogeneous mix. As such it can be shown that:
(8)
It can additionally be shown that the smallest value that a wet gas correction can take for a given liquid
content (see [4] Appendix A.2 The over-reading at minimum energy) is equal to:
(9)
Although it is not an absolute requirement for a wet gas correction correlation, it can easily be shown that
the homogeneous form given in equation (12) below fulfils all of these conditions; it is also readily
apparent, for example, that the Murdock equation does not. By not having a form that automatically tends
to these boundary conditions, a wet gas correlation should show suitable performance in these regimes, or
else risk additional flow measurement uncertainty.
No Correction
Where no correction is made for wet gas (i.e. the wet gas correction can be considered an error term in the
mass flow rate) this gives a “good” option only for a very low liquid flow rates. For this paper’s dataset, to
account for a error in the gas mass flow rate of 2.0% requires the Lockhart-Martinelli parameter to be
less than or equal to 0.0035; for typical values in this data set this implies a gas mass fraction of
approximately 98.6% or greater. If the dry gas points are ignored, this represents only 2.8% of the wet gas
test points from the dataset, thus indicating the limitations of utilizing no wet gas correction in wet gas
conditions.
3
Homogeneous
Homogeneous flow assumes that the liquid and gas phases flow at the same speed through the pipeline,
and that they are suitably distributed such that the average characteristics of the flow are applicable.
Therefore the homogeneous fluid density, , is found using:
(10)
where is the gas mass fraction. The discharge coefficient and expansibility are determined in terms of the
total homogeneous flow conditions, and this results in the total flow (gas and liquid) being determined as:
(11)
By substituting for the homogeneous density, this can be rearranged to show that the homogeneous wet
gas correction, , has the form:
(12)
where:
(13)
As Figure 1 and Figure 2 below indicate, the homogeneous wet gas correction has generally the right form
of correction for the Venturi data, particularly for lower liquid levels. However, as the Lockhart-Martinelli
parameter increases, so does the scatter in the error; this is also clearly seen in Table 2.
Murdock
J.W. Murdock was the Associate Technical Director for Applied Physics at the Naval Boiler and Turbine
Laboratory in Philadelphia in 1962 when he published “Two-Phase Measurement with Orifices”[5]. This
seminal paper took data from different two-phase measurements sources (including combinations of
steam-water, air-water, natural gas-water, natural gas-salt water, and natural gas-distillate) and a range of
orifice plate sizes and beta, to which he was able to fit a simple straight line model.
(14)
For the orifice dataset of 90 test points, Murdock used a least-squares fit to give with ±1.5%
uncertainty at 2σ (i.e. roughly 95% confidence). Murdock also used an alternative form of the Lockhart-
Martinelli parameter, where:
(15)
The extension of this form of wet gas correction to other flow meters (including [6]) necessitated alteration
of the M parameter to suitable values for each meter, and as such was suggested for Venturi
meters.
As shown in Figure 3, the line fits the orifice data well; as reported in Murdock’s paper, the data
fits to 1.5% to 2 standard deviations. However, neither the nor the lines are very
representative of the Venturi database data (see Table 2). The single-sided bias (due to most points being
under-corrected by this method) can also be seen in Figure 4.
4
Also, the spread of the Venturi points and the curve at low Lockhart-Martinelli number seen in the Venturi
dataset indicates that any straight line on this graph could only nominally fit the data. The additional
parameters and curved form seen in other wet gas correction models are therefore necessary to improve
the fit of the wet gas correction.
Chisholm
D. Chisholm was the Head of the Applied Heat Transfer Division of the National Engineering Laboratory
(NEL) in Glasgow in the 1960’s. Chisholm’s paper [7] examines eight investigations into two-phase data for
orifice plates, and uses the same form as equation (12), replacing the parameter with the generic:
(16)
with .
Even for the sharp-edged orifice plate datasets Chisholm uses, he notes that “agreement is unsatisfactory”
for two of the eight investigations, and “this requires further study”.
As can be seen in Figure 5 and Figure 6 below, Chisholm’s correction is generally a poorer fit than that of
the Homogeneous wet gas correction for the Venturi dataset.
Other wet gas corrections (including those by Lin[8], James[9], Smith and Leang[10]) have been considered,
and their performance is shown in Table 2. However, as these corrections were primarily for orifice plates,
in the same manner as Chisholm, they generally provide an inadequate correction for the Venturi data.
de Leeuw
Rick de Leeuw is a Flow Measurement and Allocation Specialist at Shell Global Solutions in the Netherlands.
de Leeuw’s North Sea Workshop paper from 1997 extended Chisholm’s idea of fitting the parameter of
the homogeneous flow correction (as per equation (16)) to data from Venturi flow meters, taken on a 4”
Venturi at SINTEF Multiphase Flow Laboratory in Norway.
de Leeuw noted a gas velocity effect that previous correlations had not taken into account. As such, the
data was fitted such that:
(17)
where the Froude gas number, , is given by:
(18)
For the data available to de Leeuw when developing this correlation the uncertainty was given to be 2% to
two standard deviations, where .
Figure 7, Figure 8 and Table 2 shows that the de Leeuw correlation is a better fit for this paper’s Venturi
data than any of the previous corrections, with a error of 3.829% (equivalent to an uncertainty of
3.673% to two standard deviations) for the bounded data. Figure 8 also shows that although the
error for this correlation has less bias than the previous corrections, there appears to be a positive bias at
larger liquid levels.
5
Steven
Richard Steven (formerly of NEL and Strathclyde University, now of CEESI and DP Diagnostics) wrote his
PhD[11] on Wet Gas Metering, including a new form of wet gas correction. This he further revised in a later
paper[12] to parameterise the A to D factors in terms of gas-to-liquid density ratio rather than pressure:
(19)
where:
(20)
The performance of this form of wet gas equation for the Venturi dataset can be seen (in Figure 9, Figure
10 and Table 2) to generally improve upon the Homogeneous and Orifice (Murdock, Chisholm, James, Lin,
Smith and Leang) wet gas corrections, but generally has greater error than the de Leeuw correction.
ISO TR 11583:2012
Emmelyn Graham and Michael Reader-Harris (both of TUV NEL) wrote a paper detailing a further wet gas
correction[13] before it later became part of a technical report for the International Organization for
Standardization[14]. In a similar manner to Chisholm and de Leeuw’s extensions of the Homogeneous
correction, the parameter from equation (16) is re-characterized to be:
(21)
H is a parameter depending on the type of liquid and its surface tensions (taken as 1.0 for hydrocarbon
liquid, 1.35 for water at ambient temperature, and 0.79 for liquid water in steam flow. Whilst this
parameter may be beneficial, its reliance on data that may not be readily known or measured in situ makes
this correction appear more arbitrary. However, as none of the Venturi data for this paper is for steam
flow, for the purposes of this paper only the following simplified equation for was used:
(22)
This standard also provides a correlation for the “discharge coefficient”:
(23)
This equation is elaborated upon in the paper[13], explaining that the initial value is “a remarkably close
approximation to the fitted value for the single-phase discharge coefficient, Cdry, which should apply both to
dry-gas flow and to homogeneous flow as Frgas,th tends to .” This is also true for the Venturi dataset
examined here, with an average dry gas discharge coefficient of 1.001, but implies that there may still be a
residual error at dry gas. For the purpose of the analysis given in this paper, is considered as an
6
additional part of the wet gas correction term. The performance of this correction is demonstrated in Table
2, Figure 11 and Figure 12, where this term corrects the initial negative bias seen in the error, further
reducing the wet gas correction errors. However, it is also noted that the paper[13] utilizes a slightly
different form for which improves the correlation (from 2.502% to 2.383% for ).
As an additional trial, the performance of the wet gas correction was evaluated for different values of .
The resulting data is given in Table 1, demonstrating that the correction is relatively insensitive to this
parameter, and therefore an assumption of water to liquid ratio for the term only (which
results in ) provides an almost negligible change in the uncertainty given in this dataset.
However, the correct liquid density is still required for the rest of the correction term.
2δ relative wet gas correction uncertainty for ISO TR 11583:2012

All Data Points X <= 0.3 X <= 0.1
H calculated using actual WLR 2.584% 2.502% 2.338%
H calculated using WLR = 0.5 2.861% 2.744% 2.310%
Table 1 - Effect of changing H (via WLR) in ISO TR 11583:2012 Venturi wet gas correction
He and Bai
Denghui He and Bofeng Bai (from the State Key Laboratory of Multiphase Flow in Power Engineering, Xi’an
Jianotong University, China) have recently published a new wet gas correction based on a two-phase mass
flow coefficient [15]. The correction values are a correlation for a term based on extensive testing of two
Venturi flow meters (2” and 8”, Sch.80, ) over both two-phase and three-phase testing at CEESI:
(24)
Figure 13, Figure 14 and Table 2 shows that the performance of He and Bai’s correction over the Venturi
dataset for this paper is generally comparable to that of de Leeuw’s correction. However, as can be seen
from Figure 15, data with greater errors are generally concentrated at gas-to-liquid density ratios far in
excess of that seen by the correlation in the He and Bai paper, which had a maximum of 0.0441 for the 2”
and 0.081 for the 8” meter.
Manufacturer-Specific Wet Gas Correction

It is noted that various manufacturers of Venturi meters (including Solartron ISA) use “undisclosed wet gas
flow Venturi meter data sets and associated confidential correlations” ([16] section 6.4.3). The benefit of
this to the operator can be seen in Table 2 where a typical Solartron ISA wet gas correction correlated to
the data for each flow meter project has significantly improved performance over all other corrections
detailed in this paper.
Further graphical representation or equations for the Solartron ISA methods are not given in this as its
performance is solely provided as an example of a manufacturer-specific correction.
7
Summary of Wet Gas Correction data
Table 2 provides a summary of the performance data for the entire Venturi dataset utilizing the various wet
gas corrections, as also shown in Figure 1 to Figure 14. This indicates that the conventional corrections are
not as good a fit for Venturi data, whilst the Venturi-specific corrections do perform better. It is notable
that the ISO TR 11583:2012 correction significantly outperforms the other standard corrections mainly due
to the removal of the bias at low liquid levels seen in other corrections.
Note however that these results include using wet gas corrections beyond their specified operational
bounds; restricting the data to the limitations given in their respective publications may therefore result in
improved uncertainties.
2δ relative wet gas correction uncertainty

All Data Points X ≤ 0.3 X ≤ 0.1
CONVENTIONAL / ORIFICE WET GAS CORRECTIONS
Homogeneous 6.036% 5.583% 3.570%
Murdock (M = 1.26) 10.061% 9.990% 8.720%
Murdock (M = 1.50) 7.774% 7.739% 7.305%
Chisholm 10.464% 10.362% 8.772%
James 14.676% 14.215% 10.309%
Lin 17.481% 16.602% 11.323%
Smith and Leang 19.158% 18.825% 16.598%
VENTURI SPECIFIC CORRECTIONS
de Leeuw 4.011% 3.829% 3.311%
Steven 4.719% 4.458% 4.208%
ISO TR 11583:2012 2.584% 2.502% 2.338%
He and Bai 4.186% 4.010% 3.198%

EXAMPLE OF MANUFACTURER-SPECIFIC WET GAS CORRECTION
Solartron ISA 1.579% 1.497% 1.217%
Table 2 – Performance of the Wet Gas Corrections
It is also worth noting that the wet gas corrections listed here are not necessarily available in all industrial
flow computers, or specified in the standard documents relevant to the wet gas industry (including [3], [14]
and [16]).
8
Homogeneous
Figure 1 - Performance of the Homogeneous WGC Figure 2 - Error in the Homogenous WGC
Murdock
Figure 3 - Murdock Wet Gas Correction, showing Figure 4 - Error in the Murdock (M = 1.50) WGC
original and paper data
Chisholm
Figure 5 – Performance of Chisholm wet gas correction Figure 6 - Error in the Chisholm wet gas correction
9
de Leeuw
Figure 7 - Performance of de Leeuw wet gas correction Figure 8 - Error in the de Leeuw wet gas correction
Steven
Figure 9 - Performance of Steven wet gas correction Figure 10 - Error in Steven wet gas correction
ISO TR 11583:2012
Figure 11 - Performance of ISO TR 11583:2012 Venturi Figure 12 – Error in ISO TR 11583:2012 Venturi wet gas
wet gas correction correction
10
He and Bai
Figure 13 - Performance of He and Bai wet gas Figure 14 – Error in He and Bai wet gas correction
correction
Figure 15 – Error in He and Bai wet gas correction plotted against Density Ratio
11
IMPROVED CORRECTIONS
Whilst the performance of the wet gas correction correlations in the previous section and Table 2
may be satisfactory for many applications, an additional benefit of the large Venturi dataset
available for this paper is to consider whether the flow measurement uncertainty can be reduced by
altering the value of parameters for the various correlations. As such, a randomized set containing
80% of the data (4,228 points) was used to update a number of the corrections by using Microsoft
Excel’s Solver routine (typically quadratic estimate, forward derivatives with conjugate search) to
optimise the selected parameters; the performance of the remaining 20% (1,057 test points) is then
shown separately to evaluate the performance of the improved correction.
For each wet gas correction, the parameters have been correlated for three different ranges of
points: all of the 80% data (i.e. 4,228 points), the 80% data where X < 0.3 (4,175 points), and the 80%
data where X < 0.1 (3,435 points). The tables given in the sections below list the parameter values,
the original correction performance (with a blue background), the performance for each optimised
region (white background), and the performance of the parameters correlated for the general wet
gas region (X < 0.3, shown with a green background) for the all data and X < 0.1 ranges.
“Improved” Murdock
The Murdock wet gas correction performance can be improved by altering the gradient ( in
equation (14)) to best fit the 80% data. However, as the correction is only a straight line, the
improvement seen to the 20% data between and is limited apart from for the
lower liquid level (low Lockhart Martinelli numbers) data, where the fitting procedure causes
significant changes to to attempt to correlate the initial curve in the Venturi data, as can be seen
in Figure 16 and Table 3.
2δ
80% 20%
M ALL DATA
Original 1.2600 10.006% 10.217%
Venturi Suggestion 1.5000 7.738% 7.877%
All Data Optimised 1.8128 6.397% 6.236%
X < 0.3 Optimised 1.8870 6.479% 6.221%
X < 0.3
Original 1.2600 9.936% 10.151%
X < 0.3 Optimised 1.8870 5.922% 6.001%
X < 0.1
Original 1.2600 8.659% 8.964%
X < 0.1 Optimised 2.3311 4.337% 4.425%
X < 0.3 Optimised 1.8870 5.336% 5.506%
Table 3 - Improved Murdock Correction Performance
As the benefits of fitting this dataset appear to be relatively small (i.e. still leaving significant
uncertainty) it is assumed that an improved “general” Murdock correction for all Venturi meters
should not to be implemented. A Murdock correction fitted for a specific Venturi meter may offer
12
some benefit, but other corrections described in this paper would better generally better suit a given
flow measurement application.
Figure 16 - Improved Murdock Correction
“Improved” Homogeneous or Chisholm

As both the homogenous and Chisholm corrections are identical except for the value of the
parameter, improvements to these two corrections can be considered together in a similar manner
to the parameter for the Murdock correction.
2δ
80% 20%
n ALL DATA
Chisholm 0.2500 10.439% 10.563%
Homogeneous 0.5000 6.005% 6.161%
All Data Optimised 0.4601 5.451% 5.519%
X < 0.3 Optimised 0.4692 5.479% 5.563%
X < 0.3
Chisholm 0.2500 10.322% 10.520%
Homogeneous 0.5000 5.493% 5.927%
X < 0.3 Optimised 0.4692 5.165% 5.431%
X < 0.1
Chisholm 0.2500 8.714% 9.001%
Homogeneous 0.5000 3.565% 3.590%
X < 0.1 Optimised 0.5245 3.402% 3.493%
X < 0.3 Optimised 0.4692 4.102% 4.110%
Table 4 - Improved Homogeneous or Chisholm Correction Performance
Table 4 demonstrates that fitting the parameter to the Venturi data provides only a small benefit
over the homogenous value, indicating the limitations of the form of this correction to match the
13
performance of the flow Venturi meter under wet gas conditions. Therefore a more complex
correction method may be required to provide better flow metering performance.
“Improved” de Leeuw
As the de Leeuw correction involves more parameters, for the purpose of this paper they are
characterised as:
(25)
Additionally we can relate these parameters when such that:
(26)
thus leaving only two free parameters to optimise. Optimising the value was also
considered and tested, but was found to offer very little additional benefit.
The parameters optimised for (i.e. the typical range for wet gas) show only a small
reduction in uncertainty from the original de Leeuw equation.
It is also worth noting the improvement shown between the improved homogeneous correction
(Figure 17) and the improved de Leeuw correction (Figure 18 – noting the difference in vertical axis
scale), where particularly at larger Lockhart-Martinelli numbers the scatter in the data is much
reduced. However, there is still a generally negative bias to the test points where .
2δ
80% 20%
AdL BdL CdL ALL DATA
Original 0.6060 -0.7460 0.4100 4.044% 3.875%
All Data Optimised 0.5731 -0.7918 0.3983 3.842% 3.706%
X < 0.3 Optimised 0.5736 -0.8245 0.4071 3.862% 3.729%
X < 0.3
Original 0.6060 -0.7460 0.4100 3.827% 3.837%
X < 0.3 Optimised 0.5736 -0.8245 0.4071 3.693% 3.703%
X < 0.1
Original 0.6060 -0.7460 0.4100 3.320% 3.275%
X < 0.1 Optimised 0.5800 -1.1127 0.4707 2.904% 2.970%
X < 0.3 Optimised 0.5736 -0.8245 0.4071 3.442% 3.384%
Table 5 - Improved de Leeuw Correction Parameters and Performance
Overall, there appears to be little benefit to “improving” the de Leeuw equation by optimising the
parameters to the larger Venturi dataset. It is therefore notable how well the original de Leeuw
equation works given the limited range of Venturi meters used in its development.
14
“Improved” Steven
For the purposes of this paper, the parameters in Steven’s wet gas correction still use equation (19),
but where equation (20) is replaced by:
(27)
where Table 6 - Parameters for improved Steven correction provides the optimised values for these
parameters.
ORIGINAL ALL DATA Optimised X < 0.3 Optimised X < 0.1 Optimised
A B C A B C A B C A B C
A 2454.5100 -389.5680 18.1460 2454.3343 -392.0140 18.1495 2454.5903 -390.9185 18.3412 2454.6565 -389.7183 24.6626
B 61.6950 -8.3490 0.2230 60.2484 -9.8381 0.3459 59.6912 -9.3426 0.3112 59.5260 -7.7311 0.2109
C 1722.9170 -272.9200 11.7520 1723.1712 -269.4366 11.2957 1722.8127 -270.9531 11.6307 1722.7832 -271.6276 17.1882
D 57.3870 -7.6790 0.1950 57.3220 -9.3315 0.3244 56.7776 -8.8500 0.2909 57.1583 -7.3915 0.1993
Table 6 - Parameters for improved Steven correction
The use of 12 parameters for this correlation provides additional risk of “over-fitting” the data to an
already complex model.
2δ
80% 20%
ALL DATA
Original 4.719% 4.713%
All Data Optimised 3.418% 3.119%
X < 0.3 Optimised 3.451% 3.175%
X < 0.3
Original 4.422% 4.591%
X < 0.3 Optimised 3.066% 3.015%
X < 0.1
Original 4.182% 4.302%
X < 0.1 Optimised 2.073% 2.082%
X < 0.3 Optimised 2.510% 2.453%
Table 7 - Improved Steven correction performance
Table 7 shows that correlating the data significantly improves the fit of this correction. However,
although Figure 19 indicates that much of the bias at lower Lockhart-Martinelli numbers is removed,
there is the appearance of a downward curve where that may signify that the underlying
model does not fit the Venturi data so well at higher liquid levels.
15
“Improved” ISO TR 11583:2012
For the purposes of this paper, the parameters in the ISO TR 11583:2012 paper Venturi wet gas
correction are characterised as:
(28)
(29)
By fully detailing the parameters for this correlation, the associated additional degrees of freedom
are made clear. As for the Steven correction, this increases the risk of over fitting the data, and a
simpler correction may be more parsimonious; however, the performance of this correlation may
justify the number of parameters.
Also, although and have the same value in the original correction, this is not assumed in
the “improved” parameters.
AISO BISO CISO DISO EISO FISO KISO LISO MISO NISO
Original 0.5830 -0.1800 -0.5780 -0.8000 0.3920 -0.1800 1.0000 -0.0463 -0.0500 0.0160
All Data Optimised 0.5211 0.0230 -0.4135 -0.6031 0.3421 -0.1797 1.0022 -0.0535 -0.0516 0.0116
X < 0.3 Optimised 0.5135 0.0475 -0.3999 -0.5901 0.3504 -0.1777 1.0023 -0.0524 -0.0499 0.0115
X < 0.1 Optimised 0.5479 0.0147 -0.3272 -0.5882 0.3223 -0.1863 1.0029 -0.0454 -0.0558 0.0053
Table 8 - Parameters for improved ISO TR 11583:2012 Venturi correction
2δ
80% 20%
ALL DATA
Original 2.608% 2.487%
All Data Optimised 2.407% 2.370%
X < 0.3 Optimised 2.410% 2.357%
X < 0.3
Original 2.509% 2.474%
X < 0.3 Optimised 2.379% 2.346%
X < 0.1
Original 2.352% 2.280%
X < 0.1 Optimised 2.198% 2.178%
X < 0.3 Optimised 2.313% 2.242%
Table 9 - Improved ISO TR 11583:2012 Venturi correction parameters and performance
As the ISO TR 11583:2012 already gave the best performance for the original wet gas corrections, it
is not surprising that the optimisation of the parameters generally gives only a small flow
measurement performance benefit. The scatter in the wet gas correction error seen in Figure 20
also shows the lack of bias due to the additional correlation parameters for low Lockhart-Martinelli
numbers.
16
“Improved” He and Bai
For the purpose of this paper, the parameters in He and Bai’s wet gas correction are characterised
as:
(30)
where , and for the correction as detailed in

equation (24).
2δ
80% 20%
AH&B BH&B CH&B ALL DATA
Original 0.5681 -0.1444 -0.1494 4.183% 4.196%
All Data Optimised 0.5434 -0.0662 -0.3237 3.193% 3.047%
X < 0.3 Optimised 0.5530 -0.0674 -0.3418 3.206% 3.038%
X < 0.3
Original 0.5681 -0.1444 -0.1494 3.984% 4.110%
X < 0.3 Optimised 0.5530 -0.0674 -0.3418 3.069% 2.997%
X < 0.1
Original 0.5681 -0.1444 -0.1494 3.214% 3.133%
X < 0.1 Optimised 0.5189 -0.0614 -0.3566 2.691% 2.680%
X < 0.3 Optimised 0.5530 -0.0674 -0.3418 2.892% 2.830%
Table 10 - Improved He and Bai Correction Parameters and Performance
For such a relatively simple correction, the performance benefit found by optimising the parameters
is significant. However, Figure 21 demonstrates that even the improved He & Bai correction has the
same initial bias at low Lockhart-Martinelli numbers that appears to be corrected by the
term (equation (23)) of the ISO TR 11583:2012 Venturi correction.
17
Figure 17 - Improved Homogenous or Chisholm correction
Figure 18 - Improved de Leeuw correction
Figure 19 - Improved Steven correction
18
Figure 20 - Improved ISO TR 11583:2012
Figure 21 - Improved He & Bai correction
“Improved” Correction Discussion

The purpose of this section was to investigate whether the wet gas corrections could be improved by
simple adjustment of their parameters. Therefore Table 11 provides a summary of the performance
improvements for some of the wet gas corrections, alongside a count of the number of parameters
that were optimised.
2δ relative wet gas correction uncertainty Optimised
for X < 0.3 Parameters
Original "Improved" – 20% data
Murdock 7.739% (M=1.5) 6.001% 1
Homogeneous 5.583% 5.431% 1
de Leeuw 3.829% 3.703% 2
Steven 4.458% 3.015% 12
ISO TR 11583:2012 2.474% 2.346% 10
He and Bai 4.010% 2.997% 3
Table 11 - Summary of "Improved" Wet Gas Correction performance
19
The Murdock correction sees significant improvement, but it still has the greatest uncertainty of the
corrections considered in this section.
The Homogeneous/Chisholm correction is almost unimproved, but due to its curved form, it is a
better one parameter wet gas correction for Venturi meters than Murdock.
de Leeuw’s correction has only one additional free parameter when compared to
Murdock/Homogeneous/Chisholm, and it provides a significant improvement. However, as it is
almost unimproved by the correlation process and the effects of the parameter changes are small, it
would appear pertinent to continue to use the original parameter values.
The “improved” Steven and He & Bai have both similar performances and similar performance
improvements, despite a significant difference in the number of parameters to be optimised.
Further investigation and potential improvements of He and Bai’s correlation, particularly for higher
density ratios is recommended.
The ISO TR 11583:2012 correction continues to provide the best overall performance. It uses a
significant number of parameters, although its wet gas correction error performance may justify its
use. Again, as the effects of the parameter changes are small, it may be pertinent to continue to use
the original parameter values.
Although beyond the scope of this paper, it would also appear that the characteristics of the
transition between dry and wet gas flow for Venturi meters would also benefit from further
investigation in a comparable manner to from equation (23). Similarly, the effectiveness of
water-to-liquid (or similar) parameters in wet gas corrections, such as the use of in equation (21),
should be further considered.
CONCLUSIONS
- The performance of a number of wet gas correction correlations has been evaluated for a large
Venturi dataset (see Table 2).
- The uncertainty of the wet gas corrections are generally increased for this dataset than those
stated in the relevant papers due to the far more extensive range of meters in this dataset.
- “Improved” corrections have been developed (using 80% of the dataset) and evaluated (using the
remaining 20% of the data). Only two (Steven and He & Bai) showed significant performance
benefits by optimising the parameters.
- Areas for further research into improving wet gas corrections should be considered, including
correlating terms at low Lockhart-Martinelli number and investigating the effects of water-to-liquid
ratio.
20
BIBLIOGRAPHY
[1]. ISO 5167-4 Measurement of fluid flow by means of pressure differential devices inserted in
circular cross-section conduits running full - Part 4: Venturi tubes. 2003.
[2]. Wet Gas Flow Measurement in the Real World. One day seminar on Practical Developments in
Gas Flow Metering. Stobie, G., National Engineering Laboratory, East Kilbride, 1998.
[3]. ASME MFC-19G-2008 Wet Gas Flowmetering Guideline. 2008.
[4]. Venturi Meters and Wet Gas Flow. de Leeuw, Rick, Steven, Richard and van Maanen, Hans.,
North Sea Flow Measurement Workshop, 2011.
[5]. Two-Phase Flow Measurement with Orifices. Murdock, J.W. s.l. : Journal of Fluids Engineering,
1962, Vol. 84.4, pp. 419-432.
[6]. High Accuracy Wet Gas Metering. Jamieson, A.W and Dickinson, P.F. 1993. North Sea Flow
Measurement Workshop.
[7]. Research Note: Two-Phase Flow through Sharp-Edged Orifices. Chishom, D., Journal Mechanical
Engineering Science, 1977, Vol. 19 No.3.
[8]. Two-Phase Flow Measurement with Sharp-Edged Orifices. Lin, Z.H. s.l. : Int. J. Multiphase Flow,
1982, Vols. 8, No.6, pp. 683-693.
[9]. Metering of Steam-Water Two-Phase Flow by Sharp-Edged Orifices. James, Russell., Proc. Instn.
Mech. Engrs, 1965, Vols. 180, Pt.1, No.23.
[10]. Evaluations of Correlations for Two-Phase Flowmeters Three Current-One New. Smith, R. V. and
Leang, J. T., Journal of Engineering for Gas Turbines and Power, 1975, Vol. 97, pp. 589-593.
[11]. Wet Gas Metering. Steven, R.N., PhD Thesis, University of Strathclyde, 2001.
[12]. Wet gas metering with a horizontally mounted Venturi meter. Steven, R.N., Flow Measurement
and Instrumentation, 2002, Vol. 12, pp. 361-372.
[13]. Venturi Tubes in Wet Gas - Improved Models for the Over-Reading and the Pressure-Loss Ratio
Method. Graham, Emmelyn and Reader-Harris, Michael., SE Asia Hydrocarbon Flow Measurement
Workshop, 2010.
[14]. PD ISO/TR 11583:2012 Measurement of wet gas flow by means of pressure differential devices
inserted in circular cross-section conduits.
[15]. A new correlation for wet gas flow rate measurement with Venturi meter based on two-phase
mass flow coefficient. He, Denghui and Bai, Bofeng., Measurement, 2014, Vol. 58, pp. 61-67.
[16]. ISO/TR 12748 Measurement of wet natural gas flow in circular cross-section conduits. 2015.
[17]. Liquid Correction of Venturi Meter Readings in Wet Gas Flow. de Leeuw, Rick., North Sea Flow
Measurement Workshop, 1997.
[18]. Evolution of Wet Gas Venturi Metering and Wet Gas Correction Algorithms. Collins, Alistair and
Clark, Steve. 1 February 2013, Vol. 46, pp. 15-20.
21
Setting the STANDARD
- Integrating Meter Diagnostics Into Flow Metering Standards
Terry Cousins
Richard Steven
Colorado Engineering Experiment Station Inc.
54043 WCR 37, Nunn, Colorado 80648, USA
1. Introduction
When industry requires to meter a flow, it requires that flow to be metered correctly. A
trusted flow meter which is subsequently found to have given an erroneous flow rate
prediction can cause financial penalties, legal arguments, and / or the process to become
inefficient. Verification of a flow meter’s performance has always been a major aspect of the
flow meter engineer’s art.
Traditional diagnostic methods to verify a flow meter’s serviceability are still widely used
today, such as mass balance checks, and the associated liquid prover / check meter
techniques. However, these methods have long been known to have severe limitations. The
latest generation of flow meters tend to have diagnostic systems that carry out internal checks
on the meter’s health. Although imperfect these diagnostic systems are a great advance over
the traditional check methods. Nevertheless, industry has been very slow on the uptake of
such an obviously beneficial development.
In this paper the authors discuss the history of flow meter verification and the issues that
hinder the permeation of flow meter diagnostic usage throughout industry. It is argued that
the flow meter diagnostic systems have now reached a maturity where they could and should
be discussed in considerably more detail in the flow meter standards. Once included in flow
meter standards, diagnostic techniques for meter verification will become included in
contractual obligations, and applying such techniques which is presently the exception,
should gradually become the general rule.
2. Verification vs. Diagnostics
Operators talk of “verifying” that a flow meter is correctly metering a flow. “Verification” is
described in dictionaries as:
 To prove, show, find out, or state that (something) is true or correct.

Diagnostics are the methodology utilized to achieve this verification. “Diagnosis” is
described in dictionaries as either:
 The act of identifying a problem by examining something, or,

 A statement or conclusion that describes the reason for a problem.
Clearly, the two definitions of diagnostics are quite different propositions to the flow meter
engineer. It is an order of magnitude easier to show that something is wrong, i.e. highlight an
unspecified malfunction, compared to also stating precisely what is wrong.
3. A History of Flow Meter Verification (Diagnostics) Methods
Where there is an industrial requirement to meter a flow there is an inherent requirement to
meter it correctly. This obvious observation leads directly to a very practical and difficult
problem; how does the operator know a flow meter output is trustworthy!?
1
Most flow meter designs require calibration at a test facility prior to being shipped to site and
installed1. However, although a good performance at a calibration laboratory is a prerequisite
to a meter’s acceptance at site it does not guarantee the meter will have the calibration
performance in operation. Many problems can adversely affect a meters performance. If the
rigors of industrial use can create various scenarios that adversely affect a flow meter how
does the operator check the integrity of the meter output? This conundrum has blighted flow
metering since the earliest days.
Early methods of checking a flow meter’s integrity are still used today. These include using
due diligence, i.e. due care in selecting, calibrating, installing and operating a flow meter,
possibly coupled with routine scheduled maintenance.
3.1 Mass Balance Checks
In pipe networks with multiple pipes and flow meters it is possible to check the mass balance
across the system. Mass balance checks inherently require knowledge of the process which is
a recurring theme in flow meter verification / diagnostics up to the modern age. There were,
and still are, severe limitations to the mass balance method. Perceived mass balance issues
can be caused not just by meter malfunctions but by incomplete understanding of a process
(e.g. flow being diverted down other pipe lines, line pack, or leakage). Also, multiple flow
rate uncertainties combine to make a mass balance check across a large pipe network
relatively unprecise. Furthermore, even if a mass balance check identifies a problem, it
doesn’t identify if the problem is loss of fluid from the pipe, or a flow meter error, or which
flow meter is in error. In short, these verification / diagnostic methods are external to the
meters themselves and even if a problem is suspected these methods can do little to pin point
a specific problem. At best, mass balance is too coarse a technique to identify many flow
meter errors.
3.1.1 Check Metering – Two Meters in Series
Fig 1. Gas Ultrasonic Check Meter Upstream of Primary Gas Turbine Meter.
An early attempt to improve on such flow meter integrity checks consisted of using check
meters, i.e. comparing the outputs of two meters in series in close proximity. This is a mass
balance check, but specific to the two meters. However, there are limitations to this approach.
If the two meters are of the same design principle they may encounter a common mode
problem. That is, they may have similar flow rate errors induced by the same underlying
issue thereby masking a flow meter problem instead of highlighting it. A partial solution is to
1
An exception to this is the orifice meter. Orifice meters are very reproducible. Industries massed orifice meter
data has been fitted & ratified by standards boards to create a discharge coefficient prediction in lieu of a
calibration. The standards discharge coefficient prediction is therefore just a data fit to massed calibration data.
Hence, orifice meters are effectively “pre-calibrated” and not exempt from this discussion.
2
use dissimilar meters in series (i.e. meters that operate on different physical principles). It is
less likely that meters operating on different physical principles will encounter a common
mode problem. Figure 1 shows an example of dissimilar flow meters in series; a non-
intrusive ultrasonic meter is installed upstream of the turbine meter. The gas ultrasonic meter
is a check meter to the primary gas turbine meter.
Dissimilar meters in series is still not an ideal check. It still requires two separate meters,
some problems can still affect dissimilar meters in similar ways, and any combination of
meters in series produces a check only as good as the combined uncertainty of the two meters
flow rate outputs. For example, if both meters have a 1% uncertainty, the meters can only be
checked to within their root mean sum uncertainty of √2%. That is, this system has a
relatively low sensitivity to any flow prediction bias. This check can only suggest the meter is
operational to an uncertainty higher than the meter’s uncertainty rating. Unfortunately, this is
a recurring theme with flow meter diagnostics, even for the latest internal diagnostic systems.
Furthermore, check metering only gives the operator a green or red light on meter
serviceability. If there is a problem it should show a discrepancy between the meters, but as a
stand-alone check it gives no information on which meter is in error or why. Nevertheless,
check metering can be described as a basic verification / diagnostic check.
Check metering is a diagnostic method that can identify if there is a metering problem. If the
two meters disagree on the flow rate outside their combined uncertainties, then it is verified
that there is a problem. The converse is not true. If the two meters agree on the flow rate
within their combined uncertainties then it is only probable that the meters are operating
correctly. However, there is still a possibility that both meters are in error together due to
some common mode problem. This issue highlights a fundamental truth about instrument
diagnostics. Diagnostic suites seldom if ever offer absolute proof a meter is fully serviceable,
they simply significantly increase the probability that the meter is fully serviceable. A
consequence of this fact is although diagnostics can be extremely useful, they cannot
completely replace the requirement for due diligence and good metering practice.
3.1.2 Check Metering with Liquid Provers
Liquid flow meter “proving” is a form of
check metering. The great advantage a
liquid prover has over a regular check
meter is that its uncertainty is so small that
the root sum square of the prover and flow
meter being checked is in practical terms
the uncertainty of the meter being checked.
The disadvantages are that it is a spot
check, effectively a periodic re-calibration
of a meter in service, and fundamentally
the process still just produces the basic
diagnostic of showing if the pair of meters
disagree for some unspecified reason.
Furthermore, due to the intermittent nature
of proving, if a discrepancy is found the
Fig 2. A Liquid Prover
operator does not know when in the time period between ‘proves’ the meter malfunction
developed. Proving is also limited to liquid flow meters, it isn’t practically possible to prove
3
gas meters. Hence, whereas liquid provers most certainly have their place in industry they are
not the panacea or benchmark of flow meter diagnostics. As such, industry has been slowly
turning to internal (“integrated”) flow meter diagnostic systems (particularly for gas flow
measurement) to verify a flow meter’s performance.
3.2 Integrated (or ‘Internal’) Diagnostics
Whereas industry still appreciates reductions in flow meter uncertainties for meter designs
operating in ideal conditions, much flow meter research has been leaning towards developing
and improving internal meter diagnostics. Flow meter verification is becoming as important
an aspect of flow metering as the uncertainty of a meter working flawlessly.
In the last twenty years flow meter designs have had great advances in internal diagnostics.
There is a slow acceptance of the inherent truths that a flow meter’s stated uncertainty is only
truly valid when there is a guarantee the meter is operating correctly, and the most assured
and precise way to continually guarantee meter serviceability is to use a diagnostic system
internal to the meter system. Strictly speaking, a flow meter’s uncertainty rating is only as
good as its diagnostic system’s ability to verify the meter is fully serviceable. As no flow
meter’s diagnostic system is perfect, this obvious truth leads to a less spoken truth, the
uncertainty rating of any flow meter is more notional than reality, it is still a faith based
measurement. It is a ‘best estimate’ based on the engineer’s judgement when accounting for
the meters calibration performance, installation & process conditions, and the capability of its
diagnostic system. This is the true state of the art, and the underlying driver for improvements
to diagnostic suites. The better the internal diagnostic system and the better the knowledge of
the process, the less uncertainty in the flow meters output.
A definition of “external meter diagnostics” is a diagnostic system that uses information
external to the meter readings, such as mass balance, the associated method of check meters,
or other process information. A definition of “internal meter diagnostics” is a diagnostic
system that uses information obtained solely from readings that are generated from sensors
internal to the metering system. An internal diagnostic system compares the resulting
readings and parameters to known laws of physics 2 and / or known correct performance
criteria (sometimes derived from a calibration baseline). Internal and external methods are
not mutually exclusive and operators can and do combine both methods.
Internal diagnostics have been developed for various flow meters such as the ultrasonic, DP
and Coriolis meters. However, these diagnostics suites tend to be relatively complicated
systems. These diagnostic systems (or ‘suites’) are a great advance in flow metering
verification, are well understood by some meter experts, but as yet they are not well
understood or utilised by the majority of flow meter operating staff.
4. Challenges with Adopting Modern Flow Meter Integral Diagnostics
In order to understand the diagnostic suite of an ultrasonic, DP or Coriolis meter it is
necessary to have a good grounding in the fundamental principles governing the operation of
the basic meter designs. Unfortunately, flow meter designs are based on various combinations
of fluid mechanics, mechanical, electronics and mathematical principles, and this can be quite
2
Examples: 1) Ultrasonic meters can prove path serviceability by inter-comparing the speed of sound (SOS)
measurements across each path because a homogenous fluid at a given thermodynamic condition has a constant
speed of sound regardless of the local path flow velocity. 2) DP meters can prove DP reading integrity by
equating the sum of the recovered & permanent pressure loss DPs with the traditional DP as this relationship is a
consequence of the 1st law of thermodynamics.
4
daunting to the average operator, who is not a meter specialist and has many other duties to
attend to other then checking the flow meters.
Many operators of flow meters see them as ‘black box’ devices and are content to use them
without understanding the details of how the operate. Standard meters predict the flow rate
via a flow computer output the operator can read, believe, and use. They do not need a
detailed understanding of the internal metering system. This level of understanding hinders
the adoption of the present diagnostic suites.
Many operations require (by legal contract) that flow meters are used for fiscal, custody
transfer, and allocation flow metering, and hence the operating staff are obliged to have a
basic working knowledge of them. But, the same is not true of the meter’s associated integral
diagnostic suites.
Legal contracts tend to state that flow metering will be conducted according to some
standards document (e.g. ISO, AGA, API documents etc.). If the stated standards document
does not promote the use of the relevant flow meter diagnostics, then use of diagnostics to
verify the flow meter will not be required by the legal contracts. As much of industry
considers the use of these diagnostics complicated, and they are not being forced by legal
requirements to adopt the use of such diagnostics / meter verification, then it is natural that
many do not pursue their use. However, such an obviously useful and beneficial advance in
technology cannot and will not fail to be adopted in the long term. Obstacles hindering the
general adoption of flow meter diagnostic suites include:
 the lack of comment in the standards documents,
o and the associated lack of legal requirement,
 the incomplete nature of diagnostic suites,
 confusing operator / machine interfaces,
 operator lack of understanding coupled with technician resistance
4.1 Meter Diagnostics and Standards Documents
There is passing comments to the ultrasonic meter diagnostic suite in AGA 9 [1] & ISO
17089-1 [2]. However, there is no comprehensive description, nor any concise statements on
the benefits, or necessity, of applying these diagnostics as part of a comprehensive meter
verification process. The DP meter and Coriolis meter standards say significantly less
regarding their respective various diagnostics. Therefore, in this 21st Century world, the
standards, and hence the legal contracts are promoting use of 20 th Century technology.
The reasons for this are varied. One reason is the fact that standards traditionally lag the state
of the art (on the grounds that they are meant to discuss mature accepted technology). As yet
there has been no strong drive to update relevant flow meter standards to include the now
matured diagnostic methodologies. Another reason is different manufacturers use different
diagnostic suites (which in reality are very similar) with different user interfaces, which
makes the diagnostics more difficult for operators to understand.
Although different manufacturers do have different diagnostics and interfaces, for a given
meter type (such as ultrasonic or Coriolis meters), or a given meter sub-system (such as DP
transmitters), the diagnostics and user interface tend to be a variation on a common theme.
This perception of difference is in part caused by the meter manufacturers aim to be seen as
different and unique. They tend to want their diagnostics to be seen as different and better
than their competitor for commercial reasons. However, in many cases they are close enough
that a generic description in a standard document is possible. All that is really hindering such
5
a development in the relevant standards is the political will. Such an addition to the standards
would benefit the meter operators, who in the long run would then adopt the very useful tool
of internal diagnostics.
4.2 The Incomplete Nature of Flow Meter Diagnostic Suites
“Part of the problem is when we bring in a new technology we expect it to be perfect in a way
that we don't expect the world that we're familiar with to be perfect.” - Esther Dyson.
Great advances in the diagnostic suites of various flow meter designs have made in the last
twenty years. Further technical developments will undoubtedly be beneficial to further
persuading industry to adopt flow meter diagnostics as a verification tool. However, although
flow meter diagnostic suites are very useful at telling an operator if the meter has a problem
they are not “all seeing”. No developer of a diagnostic suite can guarantee (or will ever
guarantee) that it will see any and all problem/s, from any source, before a flow meter bias in
excess of the meters stated uncertainty is exceeded.
For any given generic flow meter its internal diagnostic system does not have the same
sensitivity to different problems. Different problems can induce different flow rate prediction
biases before the diagnostic system can identify that a problem exists3. Furthermore, no
diagnostic system of any flow meter type can guarantee it will see all potential problems.
The fact that diagnostic systems are inherently imperfect has been cited as one reason they
are not adopted. But this is a weak argument. Such an argument is practically saying, “… if it
cannot tell me everything I’d rather know nothing.” Ignorance is bliss. Such an argument
cannot stand for long when discussing commercially sensitive metering applications.
4.3 Desired Improvements in Meter Diagnostic Suites and Their Operator / Machine Interface
"It takes a lot of hard work to make something simple, to truly understand the underlying
challenges and come up with elegant solutions." – Steve Jobs
A significant problem with industry adopting internal flow meter diagnostics has been that
the advance in flow meter diagnostic suites has not generally been matched with advances in
the machine / operator interface, nor operator training 4. It is unreasonable to assume that the
average operator has advanced knowledge of a flow meter operation principle and its
associated diagnostics suite. Hence, just as many flow meter operators wish to simply receive
a flow rate prediction from his “black box” flow meter, he would also appreciate a simple
diagnostic output. Complex diagnostics screens may contain very valuable information, but
it’s of little practical use if the average operator does not understand what it means. Presently,
diagnostic suites make the flow meter far ‘smarter’ but also far more difficult to understand
and use.
Let us consider an “ease of use vs. smartness” graph (as used by the late Steve Jobs of Apple
to discuss smart phones). Figure 3 shows the industry perception of where standard traditional
flow meters are. Not so smart but easy to use. It also shows the industry perception of where
present flow meters with diagnostics are. Quite smart, but difficult to use. The long term goal
3
Can a diagnostic suite guarantee the flow rate prediction to within “x%”? Unfortunately, the answer (as
opposed to a salesman’s answer) is always no. There are philosophical arguments that suggest it is a panacea
that is impossible to reach. Improving diagnostics can approach this ideal asymptotically but will never actually
reach that ability.
4
It does go both ways. Whereas the meter manufacturer could simplify the diagnostic output to make it easier
for the end user to understand, there is some onus on the end users to put in some effort to improve their
understanding by actively training on how to read the diagnostic output.
6
Fig 3. Perceived Meter Diagnostics Usefulness vs. Ease of Use
of industry must be to make flow meters with diagnostics very smart and easy to use. This
will only be achieved by both meter manufacturers improving their diagnostic suite,
improving the operator interface (to make it more user friendly), and by training of the
operating staff.
The authors have noted that some meter manufacturers presently think that their diagnostic
interface is just fine, and the onus is on the end users to learn the meaning of their diagnostic
display. After many years of practice and familiarity, some are so used to their diagnostic
display that its meaning is obvious to them. They cannot see a need for a simpler
representation of the diagnostics. They attempt to educate the users on what the complex
display means. However, this is effectively asking the non-specialist end user to become a
specialist. The client is being told to put the effort in to work with their system, rather than
they put the effort in to work with the client’s needs. In reality, the essential long term
requirement of industry is to significantly simplify the diagnostic output display. This of
course is easier said than done. It is difficult to simplify a complex system, and this is why it
is still a major issue. Regardless of difficulty, such is the obvious benefit to industry, and
such is the marketability of user friendly diagnostics, that it is inevitable that this will be
achieved to a certain degree in the long run. The question then, is when, not if these
developments will take place.
Figure 3 indicates that the authors consider it necessary that there is not only an improvement
in the ease of use of diagnostics but an increase in the diagnostic capability. Presently, many
flow meter diagnostic suites can indicate that there is a problem, but few make any serious
effort to tell the operator what that problem is. This is partially as the meter manufacturers are
seldom pushed to do so, and partially as the state of the art of diagnostic suites makes this a
very difficult challenge. Nevertheless, various diagnostic suites allow the operator the ability
to eliminate certain common problems while short listing other common problems. Hence,
the present state of the art potentially allows some diagnostic suites to make at least limited
predictions on the likely causes of a problem. Few however do.
4.4 Psychological Resistance
4.4.1 Technician Resistance – Fear of Change (Resulting in Redundancy)
Technicians providing flow meter maintenance are a mandatory requirement for any
reputable hydrocarbon company. As such, senior technicians in hydrocarbon production and
transmission companies can become the companies flow metering specialists. Any new
metering development is often passed by these company experts for their opinion. However,
7
these metering technicians can perceive diagnostics as a potential threat to their and their
teams livelihood. It is not always in their personal interest to highlight the great benefits, they
can perceive a flow computer with diagnostics as a “technician in a box”. That is, technology
coming to make them redundant. It is only natural that there is resistance.
Many operators have too few flow meter technicians for too large a number of meters, spread
over too far an area. Typically most of these meters have no diagnostics, or if they have
diagnostics they are not monitored as they are not fully understood. As such the technician
can be blind to many problems and is rather inefficient as he carries out routine scheduled
maintenance on meter after meter that did not need maintenance 5, while other meters with
unseen problems are untouched. The use of diagnostics would allow this technician to carry
out the philosophy “… if it ain’t broke don’t fix it”, while continually going to the meters that
do need work. That is, diagnostics allow the technician to carry out a Condition Based
Maintenance (“CBM”) scheme. CBM schemes are far more efficient, making the technicians
efforts far more valuable to his employer. CBM also reduces unnecessary staff exposure to
high pressure hydrocarbon systems. If there is proof that a meter does not need maintenance
then there is no need to break containment. Yet, by human nature technicians tend to be
nervous of the introduction of diagnostics, as it is change. Perhaps they are correct, a portion
of them may in the long run be no longer required. But this is the nature of progress, in the
end a good idea cannot be stopped, only delayed.
4.4.2. What You Don’t Know Won’t Hurt You / Plausible Deniability
The vast majority of meter operator staff / technicians naturally want to do a good job, be part
of a good team, and want the meters to operate correctly. They take pride and satisfaction
from their work. But, naturally they also want a quite life and to defend their job function.
Their management too want a quite life. Nobody likes problems. Management like to hear
that all is well. Successful mass balance check across plants, agreement between buyer and
seller metering stations, no metering alarms etc., all lead to a happy management and
harmony among the team. Ignorance is bliss. But, if you have comprehensive flow meter
diagnostic suites, they just might find a problem, a problem you otherwise would have been
ignorant of. To know of a problem, is to give the technician the conundrum of what to do
about it?
The product being flow metered is not the property of the meter operator staff / technicians.
Not at least in a financial sense. The meter operator staff typically get paid regardless of their
employers profit levels. If the operator is losing money by giving product away due to an
under-reading meter, or accidentally over charging due to an over-reading meter, it doesn’t
directly financially affect the metering staff. If a flow meter diagnostic system shows a
problem they otherwise wouldn’t have noticed, the diagnostic system has created a dilemma
they would otherwise not have had to deal with. Now, thanks to the diagnostic suite, if they
admit such a problem they get the satisfaction of doing their job properly and honourably, but
potentially give themselves the pressure of fixing it (while perhaps being unfairly blamed by
a demanding management). If they do not admit such a problem they know they are not doing
their job properly and honourably, but they avoid the pressure of fixing it (and perhaps being
unfairly blamed). However, if they can avoid using flow meter diagnostics in the first place
then “ignorance is bliss”, and they may never face this dilemma. This is a rather cynical
5
Routine scheduled maintenance can cause as many problems as it solves. For example, an orifice meter may
operating correctly before scheduled maintenance and after the maintenance the technician leaves it with a
problem, e.g. a plate re-installed backwards, a not completely re-sealed equalization valve on the 5 way
manifold etc.
8
observation, but it’s also human nature, and therefore undoubtedly a real issue. So, here
again, some technicians responsible for flow meters have natural reasons for not wanting
diagnostic systems. Still, many technicians want flow meter diagnostics regardless of these
problems. Many promote the use of diagnostics to senior management, but find some of their
senior management obstructive.
4.4.3 Resistance from Some Senior Management
“New ideas pass through three periods: 1) It can't be done. 2) It probably can be done, but
it's not worth doing. 3) I knew it was a good idea all along! ” – Arthur C. Clarke
Most operator management who are in charge of, or advise on, flow meter issues served their
time over several decades as instrument technicians or engineers. They are well versed, and
opinionated, on how flow metering should be conducted. However, along with the valuable
experience age brings, it also brings a greater resistance to change, and a tendency to not
follow or understand new technology as well as the younger generation who are brought up
with it. With flow meter diagnostics being one of the latest, most significant and complex
changes in flow meter technology for many years, it is only natural that many (although not
all) senior engineers with the authority to drive flow meter diagnostic use have not been
proactive in doing so.
There is an exception to this. Ultrasonic meters have been extensively marketed for the last
two decades. The ultrasonic meter diagnostic suite was one of several claimed benefits. Many
middle managers of the time ‘hanged their hat’ on the new ultrasonic meter technology, and
these managers are now senior managers. Nevertheless, it is curious that even though they
still promote these claimed benefits, few actively drive the use of ultrasonic meter diagnostics
in practice. For other meter designs, such as Coriolis and DP meters, the proponents of
diagnostics amongst senior instrument / flow meter engineers is significantly less, probably
as fewer have any personal ‘buy in’ to those technologies, and not all fully understand them.
Nevertheless, for hydrocarbon industry flow metering practice to evolve to include the
benefits of diagnostics it is necessary that the senior managers of the major operators support
diagnostic suite use with generic meters. This will be achieved in the long run, by some
senior managers realising their benefit, but also by the natural process described by Max
Planck:
“A new scientific truth does not triumph by convincing it’s opponents and making them see
the light, but rather it’s opponents eventually die, and a new generation grows up that is
familiar with it.”
The hydrocarbon industry has been long used to operating without flow meter diagnostics.
The contracts (and standards) do not tend to demand the use of diagnostics. Adding meter
diagnostics may add capital cost, i.e. in the case where the diagnostics are supplied external
to the basic meter, such as proving or check meters, or a stand-alone optional diagnostics
system procured separately to the basic meter. Diagnostics will add operating costs in the
form of staff training and time. Hence, some managers can resist due to the perceived
increase to the project budget.
In an effort to reduce in-house engineering staff costs many operators now out-source large
engineering projects (such as refineries, platforms, metering stations etc.). This makes the
budget issue more acute. By senior management policy operators do not get involved in
detailed engineering design decisions of these large projects. That is left to the bidding
engineering house contractors. However, such is the competitive nature between these
9
engineering houses, that for fear of losing a contract on price, they will not add any
equipment beyond the basic requirements of the “cookie cutter” specifications specifically
requested by the clients. Hence, the operators tend to trust the engineering houses judgement,
but the engineering house doesn’t use any judgement regarding extra equipment (such as
meter diagnostic considerations) for fear of over bidding and losing the job. Therefore,
metering diagnostics equipment like provers, check metering (including ultrasonic meter Z
configurations), or diagnostic systems not automatically supplied with a flow meter are often
not considered. The operator and the engineering house may both individually think the
addition sensible, but due to the political set up of the relationship neither party has the
authority to add the diagnostics. They both claim it’s the other party’s responsibility.
Inclusion in the standards and therefore contracts would induce operators to specify the
relevant flow meter diagnostic systems in the system specification and therefore help
alleviate this issue.
4.5 The Problems of Selling Insurance Policies
Fundamentally a flow meter diagnostic system is an insurance policy. A diagnostic suite is
not required for the flow meter to be fully functional. Industry has long been using flow
meters with no diagnostics. If there is no flow metering problem, the meter operates without
diagnostics just as well as it does with diagnostics. Hence, you do not need diagnostics to
meter a flow rate correctly. You just require diagnostics to give a significantly greater level of
certainty to the flow rate prediction uncertainty, and to tell you of the occasion when a meter
malfunctions.
The decision to buy into a flow meter diagnostic methodology is inherently a decision to buy
flow meter malfunction insurance. To the neutral 3rd party observer such insurance is
obviously beneficial. It can (and will) reduce the uncertainty in the operators product (and
therefore money) flow. That is, it will reduce exposure to potential large financial losses if a
meter under-reads, or perhaps law suites when the operator has over-charged due to an over-
reading. Nevertheless, selling the concept of insurance to a conservative industry that often
operates on inertia and habit can be difficult.
The application of flow meter diagnostics may take a paradigm shift in the long established
way that some operators think about their needs. Mis-measurement of fiscal, custody transfer
and allocation flows has always periodically occurred, long before comprehensive flow meter
diagnostics were available. Hence, industry has long accepted the inefficiency of finding a
flow metering bias, not knowing when the bias developed, and therefore compromising on a
mis-measurement / reallocation agreement between the affected parties. All parties are
exposed to the uncertainty of a resulting profit or loss. So entrained and routine is this process
in some organisations, that they have become blinded to the fact that this is now becoming an
obsolete methodology of conducting their affairs, a legal artefact from an early period in
technology when there was no alternative.
While inspecting a large high pressure high flow custody transfer meter, one of the authors
enquired about use of diagnostics. The response was: “Why would we need to bother with
that? We check the meter every few months. If there is a problem I just write a mis-
measurement report, they re-allocate funds, and everything is fine.” The facts that:
 they may not know the amount of bias induced by the problem found,
 they don’t usually know when the problem appeared between checks,
 they may not notice a problem without diagnostics even with standard checks,
 and hence they don’t know the financial cost of such a mis-measurement,
10
seemed lost on them. As was the fact that with diagnostics they would know immediately
when a problem appeared, and could have fixed it then, before exposure to measurement
uncertainty was an issue. Again, exposure to such mis-measurement usually does not directly
affect the individual meter technician’s personal finances. They get paid regardless, as long
as they follow the antiquated standard procedures. If the industry is to be brought into the 21st
Century, if the industry is to adopt flow meter diagnostics due the obvious advantages they
offer, a paradigm shift is required.
5. Examples of the Present State of the Art of Flow Meter Diagnostic Technologies
Flow meter diagnostics can only be considered as mature enough for general use and
inclusion in standards if they are generally accepted as potentially useful in most metering
applications. Various diagnostic methodologies for ultrasonic and DP meters at least have
been about for many years and can certainly be considered mature.
In this section the authors show a few sample examples from the positive feedback from
certain parts of the industry. (It is difficult to find any negative feedback from any source
from a technical perspective.) It is assumed that the reader has a basic understanding of flow
meter diagnostics. Detailed descriptions of ultrasonic and DP meter diagnostics can be found
throughout the industry literature. Sample documents for further reading are Lansing [3 & 4]
for ultrasonic meter diagnostics, Skelton [5] and Rabone [6] for DP meter diagnostics, and
Wehr [7, 8] for DP transmitter diagnostics. This is a very small sample. There are many more
detailed meter diagnostic documents by multiple authors and organizations that can be found
by rudimentary literature search.
Flow prediction biases tend to be induced by two distinct root causes. One is a meter system
malfunction problem (e.g. a malfunctioning sub-system such as an ultrasonic transducer or
DP transmitter, or wrong geometry or calibration information supplied to the flow computer).
The other is a flow condition induced problem on a correctly operating metering system (e.g.
disturbed flow, contamination, two-phase flow etc.) However, in some cases the flow
condition issue can in turn induce a meter system malfunction (e.g. erosion, contamination
build up, partially blocked flow conditioner, blocked pressure ports etc.) The state of the art
of flow meter diagnostics can typically identify when the meter has a problem. For a given
flow meter diagnostic system it sometimes can identify (or ‘short list’) what that problem is,
and sometimes it cannot identify the possible source of the problem. This issue is one of the
R&D aims of meter manufacturers. Nevertheless, a generic alarm stating the meter has a
problem and is therefore probably misreading is still very valuable. The examples below are a
mixture of these three general cases.
Each generic flow meter type has its own unique diagnostic suite based on the physical
principles of that meter type. It is therefore not possible, or relevant, to try and compare
different meters diagnostic systems. For example, an ultrasonic meter diagnostic suite can see
small changes in velocity profile, while a Venturi DP meter diagnostic suite cannot.
However, ISO 5167-5 indicates that a cone meter flow rate prediction is rather resistant to
this problem, whereas ISO 17089-1 indicates ultrasonic meter flow rate prediction is very
sensitive to the problem. So it can hardly be claimed that an ultrasonic meter’s diagnostic
capability to see small flow disturbances that effect it significantly is any real advantage over
the cone meter’s inability to see this issue that does not significantly affect it. In general,
different flow meter types diagnostic systems cannot be directly compared.
Most flow meter’s diagnostic suite can have the diagnostic checks broken into two categories,
i.e. ‘absolute’ and ‘relative’ diagnostic checks. An absolute diagnostic check consists of
11
comparing a known baseline to a meter result. This known baseline may be set by a physical
law, or by a fixed calibration result. A relative diagnostic check consists of trending a
parameter, i.e. checking the performance of a specific parameter over time, a ‘then’ and now’
comparison. Examples of absolute diagnostic checks are the ultrasonic meter Speed of Sound
(SOS) measurement compared to an external equation of state SOS prediction check, and the
DP meter DP reading integrity check. Both compare the results to known physical laws.
Examples of relative diagnostics is the ultrasonic meter trending of gain, signal to noise ratio,
path performance etc., and the similar diagnostics of the trending of a DP meter’s DP reading
standard deviations. Absolute diagnostic checks tend to be more powerful than relative
diagnostic checks, as they are less subjective, but trending diagnostics are still very useful.
Most flow meter diagnostic suites combine both. In these examples both types are shown.
5.1. A Proving Example – Wax Deposits on a Turbine Meter
Proving on site is a diagnostic that has been available for around 50 years. It shows change in
meter performance, i.e. if there is a shift due either an installation effect, or a change in flow
conditions with time then a proving will demonstrate this not just only qualitatively but also
quantitatively. Furthermore, proving is effectively a periodic re-calibration of the meter
within a resolution that is acceptable at present for custody transfer metering levels. A good
proving system should be able to determine the new calibration to within 0.06-0.1%.
A prover system further acts as a very competent diagnostic for meters such as turbine
meters, where it is known that such a meter should repeat during proving to within a given
value, usually set in practical terms to a spread of 0.05% in 5 runs for a fixed set of flow
conditions. If a turbine meter does not meet this criterion, then it is known that either the
meter has mechanical issues, the flow conditions are bad or have changed, or the prover has a
problem. This combined with a control chart is an excellent method of providing a
“quantitative diagnostic”.
Cleaned Cleaned
& Line
Heated
Fig 4. Control Chart for a Turbine Affected by Wax Build-up

Figure 4, shows an example of a turbine meter measuring where the meter has a buildup of
wax. In this example, the problem is inherently a flow condition problem (i.e. it contains
wax) but this problem then causes the meter to mechanically have a problem (as the wax
deposits on the meter). The proving shows the meter calibration change as a quantitative
value, while giving the operator a warning that the meter is experiencing a significant issue.
The prover is a good diagnostic system for identifying something is wrong. However, like
12
most diagnostics it does not state what the source of the problem is. It takes process
knowledge, and / or manual intervention, to identify the problem (in this case wax deposits).
Although the prover is a good diagnostic system for liquid flow metering it does have
drawbacks. Prover systems are only available for liquid applications. Provers are expensive,
requiring good operational maintenance and skilled staff to operate. It is not a practical
proposition to run provers continuously, and therefore proving is a “spot check” where the
operator is still blind in-between the periodic proves. When a prove shows a problem has
induced a meter performance shift since the last prove (with a subsequent mis-measurement),
the operator does not know when that problem appeared and at what rate it developed. The
mis-measurement has to be solved retrospectively, and usually with some guess work and
approximations, never a satisfactory situation for a high value product sales meter.
A significant modern issue is the lack of prover compatibility with the modern Coriolis and
ultrasonic liquid flow meters. By the physical principles of how they operate both these meter
types have inferior repeatability compared to turbine meters. Poor repeatability between
proves is a good diagnostic for liquid turbine meters. However, poor repeatability is not a
good diagnostic for liquid Coriolis and ultrasonic meters. The repeatability of these meters
can vary widely and unlike turbine meters repeatability is not a true diagnostic of
performance. In practical terms this results in two issues:
 The prover requires a large number of proves, making the mechanical parts, such as
the four way valve, work harder per prove with consequent maintenance issues.
 Keeping the repeat proves at constant flow conditions is difficult to achieve on site.
A further issue is several turbine meters can be replaced by one larger ultrasonic meter.
Compared to the prover size required to test one smaller turbine meter at a time, proving a
single larger meter requires a larger prover with the associated larger cost and footprint 6.
Proving therefore becomes a practical issue, and in finding some way to resolve this, internal
flow meter diagnostics have come to the forefront. Thus operators hope that the internal
meter diagnostics can identify any metering issue as it arises, so as they can decide whether
maintenance or recalibration is required.
5b. An Ultrasonic Meter Integral Diagnostic Example – Gas Trapped in Transducer Pocket
A liquid test facility calibrated an ultrasonic meter. The ultrasonic meter’s diagnostic output
was recorded during the calibration. The meter was calibrated at a high flow rate first, and
then the flow was reduced. A distinct step in meter factor was found between the high flow
rate and lower flow rates. When the flow rate was subsequently increased again to the high
flow rate, the meter factor remained similar to that found for the lower flow rates. That is,
there was a shift in the high flow rate meter factor between the two high flow calibration
results. Figure 5 shows these results. Note the step calibration change between the original
high flow rate and lower flow rate results. What caused the initial higher meter factor during
the initial high flow rate test?
6
There are always pros and cons to any engineering system. The increase in prover size required to prove large
ultrasonic meters is a disadvantage of using a single large ultrasonic meter instead of a bank of smaller turbine
meters. Ultrasonic meter marketers have responded by claiming ultrasonic meters do not need proved like
turbine or PD meters as they have no moving parts. However, such claims are tantamount to claiming an
ultrasonic meter will never malfunction for any reason. This is obviously not true, ultrasonic meters are
adversely affected by pipe fouling, flow disturbances, flow conditions out with the calibration range etc., and
hence they too benefit from periodic proving.
13
Fig 1. Change in an Ultrasonic Meter Calibration Meter Factor
The ultrasonic meter diagnostic suite has various diagnostic checks. Different diagnostic
checks are more sensitive to specific problems than others. That is, the diagnostic check
within the over-all suite that is most sensitive to any specific problem depends on the specific
problem. In this case study, let us consider the Signal to Noise Ratio (SNR), gain (the drive
voltage to give an acceptable signal strength), and the individual path velocities. Diagnostic
results of interest are:
 at the time of meter factor change there was a significant change in the SNR, Gain
and path velocity data,
 the top path in particular exhibited a significant difference to the other paths - a
warning from the outset that the meter had a possible issue,
 the top path SNR was low, showing a poor signal level compared to the noise,
 the top path transducer gain was very high, showing that the meter was trying to push
up the signal level to compensate,
 the velocity of the top path was different to its equivalent path.
The meter diagnostics were clearly showing that the top path was experiencing an issue at the
start of the calibration. The diagnostic warnings disappeared at lower flows, suggesting the
problem was only temporary. Process knowledge (external to the meter diagnostics) allows
us to surmise that air could be trapped in the liquid flow. Therefore, a likely reason for such a
diagnostic response is that air, which will gather at the top of the pipe, may have become
trapped in an upper paths transducer port. The flow rate changes are likely to have flushed the
gas free, and hence the problem soon disappeared.
There can be a tendency amongst diagnostic proponents (and ultrasonic meter salesmen) to
claim in such situations that the ultrasonic meter self-diagnosed that there was air tapped in
the transducer port. However, such claims exaggerate the ability of flow meter diagnostics. In
truth, the ultrasonic meter only self-diagnosed that the top chord had a temporary problem. It
was the skill of the operator, coupled with both the process knowledge and ultrasonic meter
14
diagnostics suite output that allowed the operator to come to the conclusion of what the
specific problem was. Nevertheless, the ultrasonic meter diagnostic warning that something
was wrong (and that it was a problem specifically with the top path) was the catalyst for the
problem being found a solved.
This is another example of the problem being inherently a flow condition problem (i.e. gas
entrained in the liquid) but this problem then causing the meter to mechanically have a
problem (i.e. gas trapped in the transducer port causing the transducer a problem due to the
significant difference in gas and liquid acoustic impedance).
5c. An Ultrasonic Meter Integral Diagnostic Example – Flow Profile Issues
An operator was periodically proving a 4 path liquid ultrasonic meter. After several years of
successful quarterly proving the prove indicated a significant shift in meter performance. The
meter factor shift was approximately 0.8%. The operator immediately surmised that there was
a fault with the flow meter.
Figure 6 shows a sketch of the meter installation, flow passes through two bends
approximately 15 diameters upstream if the meter. To mitigate the bends a 19 tube, tube
bundle was mounted upstream, 10 diameters between the tube bundle inlet and meter inlet,
similar to the installation recommended by API 5.8. In this installation there was no access to
log the meters internal diagnostics, it could only be done by taking a laptop to the meter, and
downloading a snapshot. That is, the operator did not monitor the flow meter’s internal
diagnostics.
Fig 6. USM Installation

The auditors first course of action was to review the prover data. The prover was calibrated
with turbine meters by an external US proving company with a generally good record. The
water draw data appeared in order. The turbine master meter calibrations and history were
acceptable. The temperature and pressure calibration certificates were up to date. The records
showed that they had checked the critical valves. The prove appeared correct, not totally
confirmed, but a very high probability. Suspicion turned to the ultrasonic meter.
Due to the unspecified problem identified by the prover, and the assumption that the
ultrasonic meter must have malfunctioned, the ultrasonic meter’s internal diagnostics were
reviewed. Fortunately, the ultrasonic meter had had a diagnostic baseline recorded during its
commissioning. The present results could therefore be compared to this diagnostic baseline.
The majority of the diagnostic checks had not shifted from this baseline significantly. The
SNR, gain & SOS checks had no significant change. The ‘turbulence’ diagnostic showed a
marginal shift. However, the diagnostic check sensitive to this problem was the flow profile.
Figures 7a & 7b shows the difference between the commissioned profile and the profile
during the investigation. At commissioning the profile appears symmetrical (as it should), but
during the investigation the profile was asymmetric. Therefore, the ultrasonic meter’s
diagnostic result showed that there had been a shift in the meters velocity profile. That is, the
15
Fig 7a Commissioned Profile Fig 7b Profile after Tube Bundle Lost
ultrasonic meter diagnostics specifically identified that the meter is receiving a disturbed flow
profile. This typically means the ultrasonic meter mis-measures the flow, although it is not
possible to derive from the meter diagnostics the induced flow rate bias.
By combining the ultrasonic meter diagnostic output with knowledge of the process, the first
prediction of the trained flow meter auditor was that the tube bundle was partially blocked.
That would create a disturbed flow and account for the meter’s diagnostic response. The
meter operators were not trained in understanding the ultrasonic meter diagnostics and had to
trust the auditor. In winter conditions the line was dismantled to inspect the tube bundle. The
auditor was only partially correct. Yes, the ultrasonic meter diagnostics were correct in
indicating there was a flow disturbance, but it was not due to a partially blocked tube bundle,
it was due to the tube bundle being absent. The tube bundle had become detached and had
passed through the meter and been caught by a downstream bend. So what does this simple
example teach us?
a. Proving on a quarterly basis only catches a major event at the time of the prove, and that
the error can be in place for an indeterminate time between the proves.
b. If the flow meter diagnostics had been continually monitored (by software which can send
an alarm to operator staff trained in the diagnostic system) a large and significant event such
as a tube bundle disappearing downstream, would certainly have been immediately noticed.
c. The diagnostic alarm could not confirm the precise cause of the problem, but significant ly
it would have correctly identified that a problem may exist and the meter needs to be
immediately re-proved.
d. After such a re-prove (i.e. recalibration), it would be possible to back calculate from the
before and after meter performance the flow rate prediction bias induced by the adverse
event.
e. For flow meter diagnostics to be of any use the staff operating the meter need to understand
their meaning. In this case the operating staff did not understand the ultrasonic meter
diagnostics and hence it is probable even if diagnostic results were recorded the staff would
not have realized there was a problem.
The main reasons cited by the operator for not utilizing the ultrasonic meter diagnostic suite
was that they believed the prover would be good enough to see any problems, nobody on
staff understood the ultrasonic diagnostics, and their use was not demanded in the contract or
local regulatory requirements. Both the contract and regulatory requirements followed the
recommendations of standards.
16
5d. An Orifice Meter Integral Diagnostic Example – Contamination
Orifice meters have been used widely throughout industry for a century. As with liquid flow
meter proving, it has long been known there should be site checks for gas orifice meters,
particularly for custody transfer applications. An early form of orifice meter diagnosis was
the creation of the orifice plate fitting which allowed the plate to be periodically removed and
inspected. Instrumentation could be periodically re-calibrated. These processes gave a good
indication of the meter performance. However, as with liquid provers, such techniques are
“spot checks” that leave the operator blind to problems that occur between such checks.
Again, what was needed were internal meter diagnostics that could continually monitor the
orifice meter. Such techniques now exist. DP transmitters (a vital sub-system of DP meters)
now have internal diagnostics that monitor their health. More recently a DP meter diagnostic
system based on pressure field monitoring has been developed. This case study uses the
orifice meter pressure field monitoring diagnostics (and was previously presented at the
North Sea Flow Measurement workshop by Rabone [6] in 2012)
Fig 8. Orifice Meter Run Fig 9. Plate Inspection - Contamination.
Fig 10. Orifice Meter Diagnostic Display Screenshot.

A 6” orifice meter was audited. The DP meter pressure field monitoring diagnostic system
‘Prognosis’ was installed as the initial part of the audit. Figure 8 shows the meter run prior to
the installation of the diagnostic system. Figure 10 shows a screenshot from the diagnostic
display. The diagnostic result immediately showed that the meter had a problem (i.e. note the
two points outside the box in Figure 10). Like the ultrasonic meter diagnostic suite, this DP
meter diagnostic suite cannot always tell the operator what the specific problem is, but from
the diagnostic output pattern coupled with operator process knowledge the problem can be
17
identified. The particular diagnostic pattern was indicative of a few potential problems. This
short list included possible contamination of the meter run. It was known from process
knowledge that contamination was a possibility. The meter was subsequently inspected.
Figure 9 shows a photograph of the plate. The plate was contaminated. A subsequent
borescope investigation of the meter run showed meter run contamination.
Contamination causes all flow meters to mis-measure the flow. The operator was unaware
that the meter flow rate output was in error prior to the installation of the diagnostic system
and the subsequent audit. If the operator had used such a diagnostic system from the outset,
and been trained to understand the diagnostic display, they would not have been blind to the
developing problem and would have known to intervene long before the problem was
accidentally found by a scheduled audit.
This is another example of the problem being inherently a flow condition problem (i.e.
contaminate in the flow) but this problem then causing the meter to mechanically have a
problem (i.e. contamination build up in the meter).
5e. An Orifice Meter Integral Diagnostic Example – DP Reading Error
Three 12” orifice meters are used to meter gas being supplied to a power station (see Rabone
[9]). The operator installed the pressure field monitoring diagnostic system “Prognosis”. On
commission of the meters (with the diagnostic system) two of the three orifice meter
diagnostic systems gave no diagnostic alarm. That is, for these two meters the seven DP
meter diagnostic checks found no issues. The results are represented in the diagnostic display
as seven co-ordinates (giving four points) on a graph. If the points are inside a box there is no
alarm. The meters with no problem showed a diagnostic result like that shown in Figure 11.
However, one of the three meters showed the diagnostic result reproduced in Figure 12. One
of the diagnostic checks is indicating a meter problem.
Fig 11. Normal Operation Result. Fig 12. Abnormal Operation Diagnostic Result.
This DP meter diagnostic pattern shown in Figure 12 specifically indicates that one of the DP
readings is erroneous. The operator checked the DP transmitters and stated that they looked
okay. The DP transmitter internal diagnostics showed no problem. However, the diagnostics
DP integrity check is an absolute diagnostic check, based on the 1 st law of thermodynamics.
Hence, a DP reading was wrong, regardless of the initial maintenance check suggesting
otherwise. The operator understood this, and on second review found a minor scaling issue in
the flow computer that was causing an incorrect conversion of the DP signal. The primary DP
reading used to measure the flow rate was being over-predicted by +1.6%. The associated
flow rate prediction bias was +0.8%.
This case study is an example of where the problem was not induced by flow conditions but
by the metering equipment. In this case the diagnostics were very close to identifying the
precise problem and no process knowledge was required.
18
This example shows that internal meter diagnostic suites are sometimes more useful than due
diligence, spot checks, or check metering by external systems. For example, if this orifice
meter had an ultrasonic meter in series as a check meter the problem may still not have been
seen. API 14.3 estimates the typical uncertainty of an orifice meter at about 0.7%. ISO
17089-1 states the typical uncertainty of a custody transfer ultrasonic meter as 0.7%. So the
root mean sum of these meters uncertainties is approximately 1%. That is, these meters in
series could only indicate a metering problem when the discrepancy between them is > 1%.
The discrepancy here was ‘only’ +0.8%, below the threshold of many check metering
systems, but high enough to cause significant financial mis-measurement.
Even after maintenance the initial response of the competent maintenance crew carrying out
standard instrument checks was to say the DP readings were okay. It took their trust in the
meter internal diagnostic system to make them look again, closer. Only due to their trust in
the meter’s internal diagnostic system did they save themselves from mis-measuring the flow.
5f A Venturi Meter Diagnostic Example – Showing the Calibration Report to be Wrong
An 8”, 0.4 beta Venturi meter (Fig 13), was calibrated with natural gas flow. The calibration
included setting the baseline of the pressure field monitoring diagnostic system. After
multiple witnesses had confirmed the calibration was correct and complete, the standard Cd
vs. Re calibration report was written and released. No comments or issues were raised by any
party on the integrity of the calibration report.
Fig 13. 8”, 0.4 Beta Ratio Venturi meter Under Calibration
At a later diagnostic system FAT, the diagnostic system software was checked by entering
the meter geometry, calibration diagnostic parameters, and sample data from the calibration
report. As the diagnostic system had been calibrated to this very data the diagnostic system
should by default show no problem. However, the diagnostic system signaled a malfunction.
Figure 14 shows the resulting diagnostic plot.
There is a short list of possible malfunctions that cause such a Venturi meter pressure field
monitoring diagnostic pattern. The first on the list was incorrect geometry, either too small an
inlet diameter or too large a throat diameter. The geometry in the calibration report was
therefore compared to the actual meter geometry from the manufacturer. The true 8”, 0.4 beta
Venturi meter throat diameter was found to be 73.02mm. The meter had been calibrated with
the correct geometry. However, the calibration report erroneous stated the throat diameter of
73.62mm. The diagnostic system had subsequently been given the wrong geometry read from
the report. The flow computer for this meter in service would most likely have been given the
wrong geometry from the same report. The effect on the meter flow rate prediction would
have been to over-read the flow. The Venturi meter internal diagnostic system saw the
problem before mis-measurement occurred. Conversely, standard practice of due diligence by
19
Fig 14. Diagnostic Result During Factory Acceptance Test of Diagnostic System
multiple experienced competent witnesses checking the report failed to see the problem.
Without diagnostics the problem may have permanently would have gone unnoticed. Further
details of this example are discussed by Rabone [6].
This case study is an example of where the problem was not induced by flow conditions but
by the metering equipment (and more precisely human error). In this case the diagnostics
were very close to identifying the precise problem and no process knowledge was required.
This example shows that internal meter diagnostic suites are sometimes more useful than due
diligence. No matter how diligent operators are they are human and mistakes can creep
through. Automated checks from diagnostic systems do not make such errors.
5g A Coriolis Meter Diagnostic Example – Gas in the Liquid Flow
Coriolis meters have an established diagnostic suite. The Coriolis meter manufacturers talk of
in-situ verification by using the Coriolis meter internal meter checks (i.e. diagnostic suit).
For liquid flow particularly, the Coriolis meter’s density output can be monitored for
fluctuations and compared to the expected liquid density. As the liquid density is usually
know by the operator to a high degree of certainty, and the Coriolis meter liquid density
measurement has a low uncertainty, any significant difference between the values suggests a
problem exists7. (This is another example of how many diagnostics rely on operator process
knowledge.)
The measured stiffness of Coriolis meter vibrating tubes in operation can be compared to the
original factory baseline value. Changes in stiffness can represent corrosion, erosion, or
deformation of the tubes.
Another main Coriolis meter diagnostic check is that of monitoring the drive gain. Drive
Gain is a measure of the amount of power the transmitter requires to keep the tubes
resonating. Drive gain is stated as a percentage of the total power available to the transmitter.
For given fluid properties, it takes a set amount of power to keep the tubes resonating. If the
fluid consistency is constant, and the flow is constant, the drive gain should be constant.
Changes in fluid density and / or two-phase flow cause shifts in drive gain. Therefore, drive
gain can be used to monitor fluid consistency. Shifts in drive gain can indicate qualitative
shifts in fluid properties. However, fluid properties are not the only factor that influences
drive gain. Drive gain is also affected by the physical condition of the vibrating tubes. If these
tubes are damaged or changed (e.g. by erosion, corrosion, overpressure, hydraulic shock,
7
Liquid Coriolis meters give low uncertainty flow density measurement. However, as moderate to low pressure
gas flow has such a small density compared to liquid densities, the gas Coriolis meter density measurement has
a much higher uncertainty. This, coupled with the fact that the independent gas density measurement uncertainty
is also higher than an independently measured liquid density uncertainty, means that this diagnostics is very
useful with liquid flow but less useful with gas flow.
20
contamination deposit etc.) the drive gain required will shift. Most damage or changes to the
vibrating tubes is non-symmetrical in nature, and this can make resonance very difficult to
attain, resulting in an unstable (or possible saturated) drive gain.
A Coriolis meter drive gain value is application dependent. Stable flow conditions produce a
stable drive gain. If the drive gain has significant fluctuations, say 10% or more, the
measurement might be still be acceptable, but the meter uncertainty may higher than the
specified uncertainty rating. A higher drive gain indicates more energy required to overcome
higher damping of the vibrating tubes. Gas entrained in liquid flow is one adverse operating
condition (amongst others) that induces higher tube damping.
Gas Passing
Fig 15. Coriolis Diagnostic Showing Gas in the Liquid Flow

Figure 15 shows sample trending data from a liquid Coriolis meter. The primary meter
outputs of mass flow rate and density measurement are shown along with the drive gain, all
plotted against time. The initial conditions are steady. The initial conditions give a density
measurement that can be compared to process knowledge, i.e. an external density
measurement, and shown to be correct, giving confidence in the meters operation and mass
flow prediction. The drive gain is stable (i.e. has a small standard deviation) and low, as
expected. Then trending shows all three outputs to significantly change together. It is
suspected that this particular combination of diagnostic results indicates the liquid flow
becoming a two-phase flow. The mass flow rate and density predictions begin to drop as the
drive gain magnitude and standard deviation significantly increases. Then these effects
magnify. This is suspected to be a two-phase flow, i.e. gas passing. The density measurement
and mass flow rate prediction drop as would be expected with gas in a liquid flow, although
these meter cannot be expected to be the average actual flow values! Then, the meter outputs
return to their original values, indicative of the gas having passed, and the flow being single
phase liquid flow again.
The transient nature of the response (shown by trending) indicates that the issue is likely a
flow condition issue as mechanical problems do not tend to fix themselves. Although
‘absolute’ diagnostic checks are generally more powerful than relative / trending diagnostics
(as they are less subjective), this example shows that relative / trending diagnostics have an
21
important contribution to make to diagnostic suites. As the trending allows mechanical
problems to be considered unlikely the result strongly suggests two-phase flow.
The precise effect of two-phase flow on any Coriolis meter is design, fluid property and flow
condition dependent. Drive gain is a good qualitative indicator that something has changed.
However, Coriolis meter drive gain is influenced in complex interactive ways by various
considerations and therefore it cannot offer any detail quantitative information.
This case study is an example of where the problem was induced by flow conditions and not
the metering equipment. The flow conditions did not damage the meter in any way.
6. Existing Authoritative Flow Meter Documentation - Comments on Flow Meter Diagnostics
There are limited comments in the authoritative literature promoting the use of flow meter
diagnostics. However, it is interesting to note that although these documents support the use
of meter diagnostics they tend to stop some way short of demanding their use, while they
demand adherence to other good meter practice. As we will see below (in quotes from sample
text) the organization and committees that wrote these authoritative documents clearly
understand that the use of internal flow meter diagnostic systems are as important as adhering
to good standard meter practice if mis-measurement is to be avoided. However, these
documents (which at least address diagnostics – many authoritative documents don’t) only
promote diagnostic use with words like ‘should’ and ‘can’, while other good practice is
described with language like ‘shall’ and ‘must’. In the following examples the bold
underlined italics are added by these authors to highlight the optional aspect of the
suggestions.
DECC “Guidance Notes for Petroleum Measurement, Issue 8”, released in 2012
For DP meters in section 6.7.7 DECC states:
“The use of diagnostic systems based on the use of an additional measurement of the fully-
recovered pressure is gradually becoming well established. Experience has shown that this
technique enables the Operator to detect significant deviations from normal operating
conditions as they arise. It may therefore form the basis of a condition-based maintenance
strategy, as described in Chapter 4 of these Guidelines; DECC has already agreed to the
adoption of such a strategy at a major UK terminal. Operators of new developments are
strongly encouraged to consider the adoption of such a strategy. The provision of an extra
pressure tapping costs relatively little at the design and manufacturing stages, but may
permit significant operational savings to be made during the life of the field. ”
For ultrasonic meters (called “USFMs”) in section 6.8.12 DECC states:

“…condition-based maintenance (CBM) of gas USFMs may be the most appropriate
strategy in many instances.”
Therefore diagnostic use is optional, while other good practice is not. DECC explain that by
2012 they had agreed to a condition based maintenance strategy on one orifice meter station
and one ultrasonic meter station based on these meters respective internal meter diagnostics.
The Alberta Energy Regulator’s Directive 17, “Measurement Requirements for Oil & Gas
Operations”, released 2015
In section 2.5.2 (on gas meter internal inspection) Canada’s Directive 17 states an exception
(#9) to the demanded routine scheduled maintenance / re-calibration period for a gas meter:
22
“Internal metering diagnostics may be used to determine if the structural integrity of the
primary measurement element is within acceptable operating parameters and checked at the
same required intervals as an internal inspection. Then internal inspection is not required
until an alarm or error is generated by the device or as recommended by the manufacturer.”
In section 2.6 (on liquid meters) Directive 17 repeats this statement. In section 12.3 (on
production measurement) Directive 17 states:
“Verification can be achieved using…. internal diagnostics of the measurement device if

present to check the structural integrity of the primary measurement element”
Again, diagnostic use is optional while other good practice is not.

ISO 17089-1 “Measurement of Fluid Flow in Closed Conduits – Ultrasonic Meters for Gas –
Part 1: Meters for Custody Transfer and Allocation Measurement”, 1 st Edition.
This ISO standard discusses the use of the meter’s diagnostics in considerably more detail
than most flow meter standards. However, again it is curious to note how much the authors
clearly understand the importance of the internal meter diagnostics and yet hold back from
demanding their use. The language surrounding the diagnostic system is promotional,
whereas the language discussing operation of the primary meter system itself is authoritative.
The text repeatedly alludes to the fact that if the operator is to achieve the minimum flow rate
prediction uncertainty technically achievable he needs to use the meter’s internal diagnostics.
But the text does not take that leap to state this fact outright. Sample text from ISO 17089:
In the scope:
“This part of ISO 17089 specifies requirements and recommendations for ultrasonic gas
flowmeters (USMs)…”
That is, in the scope, the scene is set where aspects of use of the ultrasonic meter will be
requirements of the standard, i.e. instructions to the operator, and other aspects of use of the
ultrasonic meter are only recommendations, i.e. suggestions as a good idea. The operators
are free to ignore those recommendations and still claim the meter installation is compliant
with this standard. It is notable that virtually all comments on the universally accepted
ultrasonic meter diagnostics system are all firmly in the recommendation camp, where they
are free to be ignored. For example:
In the Introduction:
“USMs can deliver extended diagnostic information through which it may be possible to
demonstrate the functionality of an USM.” And,
“Due to the extended diagnostic capabilities, this part of ISO 17089 advocates the addition
and use of automated diagnostics instead of labour-intensive quality checks.”
In Section 7.4.1:
“… the theoretical speed of sound (TSOS) can be compared with the measured value
[MSOS]. The SOS is an excellent tool...”
In Section 7.4.1.1:
“If both MSOS and TSOS are available, they may be compared …”
23
In Section 7.4.1.1:
“Although the SOS is one of the most important parameters to be used in verification, there
are many more parameters which may be monitored in order to ensure optimum
performance…”
Again, diagnostic use is optional while other good practice is not. This standards document is
one of the best available standards for promoting and discussing the benefits internal flow
meter diagnostics. However, even here, virtually nowhere does this ISO text demand the use
of these diagnostics. Some the wording is deliberately weak. Even if use of the meter
diagnostics are not to be demanded, wordage like “can” and “may” is significantly weaker
than, “should” and “it is highly recommended”.
Why is this? After years of research, development and field experience are the subject
specialists who write flow meter standards really still unsure if such internal meter
diagnostics can demonstrate flow meter functionality!? That is what such text infers.
However, these authors think not. There is now abundant evidence over many years, from
multiple manufacturers, laboratory and field results from multiple independent parties
(including users), to show that flow meter internal diagnostics have come of age. The authors
of the standards are the same individuals that independently publish results for their
respective companies showing flow meter internal diagnostics to be invaluable.
Perhaps this is a legacy issue, with diagnostic suites tending to be newer attributes to the
original meter designs. The standards committees (even those working on current draft
standards) are notoriously conservative and still baulk at demanding adherence to even the
most established meter diagnostics. No flow meter standards committee working on any flow
meter technology as yet has stepped up and made the not so bold step of stating what is now
surely self-evident. That is, flow meter diagnostics have come of age and now should be an
integral part of flow meter best practice. There is no real technical argument against using
modern flow meter internal diagnostics. So surely it is time standards said as much?
Flow meter diagnostics can now be made integral to the meters such that future generations
will automatically see them as integrated vital parts of the core meter system. Today, there is
still a lingering historic perception that somehow these diagnostic systems are superfluous,
not essential to guarantee the correct operation of the meter, but it is fast becoming a false
perception. Technology has moved on, and such an obviously beneficial advance cannot, and
will not be ignored indefinitely. Sooner rather then later, these diagnostic systems will
become engrained in the operators mind as essential to professional flow meter operation
conduct, and they will be automatically used. The sooner the flow meter standards board
grasp this reality and make authoritarian demands about the use of diagnostics the quicker
this change in mind set and practice will occur, to the benefit of all industry.
7. Comments
Under ideal (calibration) flow conditions many modern flow meter types have low
uncertainty, and their internal diagnostic systems are capable of seeing many common
problems before significant flow rate prediction errors occur. As long as the operator applies
the meter’s internal diagnostics flow metering has never been so assured. Time and time
again internal flow meter diagnostics have repeatedly proven themselves invaluable. They are
vital to gain maximum assurance of low uncertainty flow metering. However, the majority of
industry still does not understand or use these internal meter diagnostics. Industry still tends
to run the majority of flow meters “blind”.
24
Various common flow meter designs now have good diagnostic capabilities that are very
capable at identifying when the meter has malfunctioned. They often cannot state what the
problem is, or the size of the flow prediction bias. However, the diagnostic output of many
meters is now good enough that when the pattern of the diagnostic suite output is cross
referenced with process knowledge a competent operator can often deduce the specific flow
condition or meter problem. (Without the diagnostics a competent operator is often
effectively blind, even with good process knowledge, and often cannot even deduce the meter
is in error.)
The authors have not discussed in the paper some further advantages of the use of diagnostics
that are worthy of mention. The active use and collection of diagnostic data allows an auditor
to be able to review the historical performance of metering systems, giving a greater degree
of confidence in the operation. Further, the active use of diagnostics helps streamline the
maintenance and efficient operation of metering, which generally contributes towards the
correct operation, and sustainability of a good uncertainty of measurement.
These diagnostic advances represent a significant improvement in flow meter technology

over the last two decades. However, many operator staff do not understand the physical
principles of these various flow meter internal diagnostic suites, and they do not
understanding the need, or the implications of using these diagnostics. There is a lack of
expertise in interpreting these diagnostics. Even when some do understand the physical
principles, it can still be difficult to understand the various complex ways in which the
different manufacturers present their diagnostic results. These factors all contribute to the
poor uptake of diagnostic use. All these issues are consequences of the core issue: the lack of
standardization.
Without standardization there is no political pressure for operators to even use flow meter
diagnostics, never mind for these issues to be resolved. With standardization, use of internal
meter diagnostics will become included in contracts and fiscal regulators will then enforce
their use. This in turn will drive the permeation of flow meter internal diagnostics into the
main stream of industrial flow metering to the long term benefit of industry and society.
Flow meter internal diagnostics are one of the most guaranteed ways of knowing when a
meter has malfunctioned. (Some traditional external diagnostic methods have significant
limitations.) Internal diagnostic systems of various meters (and sub-systems of meters) are
now mature, and their inclusion in the relevant standards is due. There are two types of
comment in industry standards, requirements and recommendations. To comply with a
standard an operator must follow the requirements, but is not obliged to follow the
recommendations. Therefore, although it is preferable for diagnostics to be recommended in a
standard rather than for them to be not discussed at all, for full impact it is preferable that
internal flow meter diagnostics should be stated as a requirement in standards. Then their
subsequent inclusion in contracts will slowly dismantle the present remaining political
obstacles.
Inclusion in standards, and therefore contracts, would inevitably result in a significant
increase in use. This in turn would offer meter manufacturers commercial incentive to put
more resources into further developing flow meter diagnostics. It would also produce more
scrutiny of the various meter manufacturer’s diagnostic claims and in turn control their
claims. This increase in use would focus the industry into developing better, clearer, simpler
more intuitive user friendly ways of displaying diagnostic outputs. Such a development
would inherently make flow meter diagnostics more accessible and understandable to more
25
meter operators. Standardization of flow meter internal diagnostics and their displays would
also facilitate an increase in formal meter operator training which in turn would further
increase their use.
The authors can only think of a couple of minor potential disadvantages. The first is that as a
diagnostic system is inherently imperfect, of course there will be occasions where it could
give a false alarm (i.e. “cry wolf”). This is inevitable. However, these cases would be
relatively few, especially if the operator is initially pragmatic about choosing the sensitivity
settings of the diagnostic output until he has experience with the system. The second is the
perceived cost of change. Diagnostic systems can add cost to a metering system in some
cases via capital cost, and in most cases due to the required training and duties of the
operators to monitor these diagnostics. However, these capital and training costs are typically
a modest investment, in the hydrocarbon industry they are small relative to the potential
savings by avoiding mis-measurement. Furthermore, nobody would expect an operator to
continually monitor a diagnostic system. Modern diagnostic systems run in the background
unmonitored until the operator chooses to periodically look, or they are alerted of a problem
by the diagnostic system. Therefore, these few claimed disadvantages are more false
perception than reality.
Such is the benefit that modern flow meter diagnostics bring to industry that it is obviously
only a matter of time (although a rather long time) before they are universally adopted. In the
long run operators will come to wonder how they managed without them. The sooner their
use permeates throughout industry the better for industry. A catalyst to accelerating their use
is their inclusion in the industry standards as required practice. This development would help
the conservative hydrocarbon industry depart from its habit of using 20 th Century metering
methods and bring its metering methods into the 21st Century.
References
1. AGA Report No. 9, “Measurement of Gas by Multipath Ultrasonic Meters”, 2 nd Ed. April
2007.
2. ISO 17089 – 1, “Measurement of Fluid Flow in Closed Conduits – Ultrasonic Meters for
gas, Part 1 – Measurement for Custody Transfer & Allocation Measurement ” 1st Ed. 2010.
3. Lansing J. “Basics of Gas Ultrasonic Meter Diagnostics”, International School of
Hydrocarbon Measurement, OK, USA, Paper / Class # 1400, May 2013.
4. Lansing J. “Advanced Gas Ultrasonic Meter Diagnostics”, International School of
Hydrocarbon Measurement, OK, USA, Paper / Class # 1410, May 2013.
5. Skelton M. et al, “Diagnostics for Large High Volume Flow Orifice Plate Meters”, North
Sea Flow Measurement Workshop, October 2010,UK.
6. Rabone J. et al, “DP Meter Diagnostic Systems – Operator Experience”, North Sea Flow
Measurement Workshop, October 2012, UK.
7. Wehr D. “Detection of Plugged Impulse Lines Using Statistical Process Monitoring
Technology”, Emerson Process Management, Rosemount Inc., White Paper, website:
http://www2.emersonprocess.com/en-us/brands/rosemount/pressure/pressure-
transmitters/3051s-advanced-diagnostics/Pages/index.aspx
8. Wehr D. “Wet Gas Diagnostics with Intelligent Differential Pressure Transmitter”,
Emerson Process Management, Rosemount Inc., White Paper, website:
http://www2.emersonprocess.com/en-us/brands/rosemount/pressure/pressure-
transmitters/3051s-advanced-diagnostics/Pages/index.aspx
9. Rabone J. et al “DP Meter Diagnostics – Multiple Field Results with new Turbulence
Diagnostic Techniques”,North Sea Flow Measurement Workshop, October 2015, Norway.
26
BS7965 2013 – “An overview of updates to the previous ( 2009 ) edition“
Andrew Wrath, BS7965 Working Group
Introduction
In November 2013, the current revision of BS7965 was published, “Guide to the selection,
installation, operation and calibration of diagonal path transit time ultrasonic flowmeters for
industrial gas applications”. This paper seeks to look at some of the changes made since the previous
edition (2009) was issued and the reasons for making the changes.
The BS7965 working group sits as part of the British Standards Institute CPI/030/05 committee
covering measurement based on velocity and mass methods ( primarily Ultrasonic and Coriolis
meters). The guidelines contained within BS7965 were initially published in 2000 and then revised
in 2009.
Since the publication of BS7965:2009, ISO 17089-1:2009 ( Measurement of fluid flow in closed
conduits — Ultrasonic meters for gas — Part 1: Meters for custody transfer and allocation
measurement ) and ISO 17089-2:2012 ( Measurement of fluid flow in closed conduits - Ultrasonic
meters for gas - Part 2: Meters for industrial applications ) have been published. Part 1 of ISO170989
covers class 1 and 2 meters, part 2 covers class 3 and class 4 meters.
The revision of BS7965 covered changes needed to provide

 Consistency with ISO17089 where these improved the document;
 To capture new meter sizes;
 To reflect performance offered by modern meters; and,
 To integrate the latest operator experience.
One of the main drivers in revising the document was the more frequent use of smaller meters, e.g.
2” & 3” meters, as these sizes had not been considered in the previous versions. As previously BSI
wanted to keep all of the guidance in a single document.
At this point we would like to acknowledge members of the working group, the wider CPI/030/05
committee and all those who commented on draft versions of the documents. The working group
covered a cross section of manufacturers, auditors and end users. The aim of the group was to
produce a document that was practical, pragmatic & performance based.
Members of BS7965:2013 Working Group
David Beecroft Accord Energy Solutions Ltd.
Colin Lightbody Petrofac Ltd
Stephen Peterson Shell UK Ltd
Eddie Spearman CNR International
Mike Thackray Elster Instromet Ltd
Andrew Wrath SICK UK Ltd

Scope of BS7965
BS7965 covers all industrial gas applications from class 1 to class 4. It covers full bore, reduced bore
and insertion type meters. It does not cover clamp-on/externally mounted, domestic or stack/chimney
(combustion exhaust) meters.
“This British Standard provides recommendations and guidance on the selection, installation,
operation and calibration of transit time ultrasonic flowmeters (USMs) for industrial gas
applications (including flare gas).
It is applicable only to those devices in which the entire gas stream flows through the body of the
USM and the transducers (wetted or invasive type), or a wetted interface, are in contact with the
fluid. This includes full-bore/reduced-bore and insertion type meters employing direct or reflective
paths. It is also applicable to “dirty gases”, i.e. those contaminated with solids and/or liquids in
sufficiently small quantities, depending upon the application.
It is not applicable to clamp-on/externally mounted, domestic or stack/chimney (combustion exhaust)

meters.
As an aid to selection, this standard incorporates an uncertainty classification for USMs. This takes
into account the components of uncertainty relating to the meter plus those additional uncertainties
associated with installation and operation effects. There are four classifications offered. These range
from high accuracy applications (Class 1) to process flow indication and control applications
(Class 4).”
Cartridge Type Meters
This is a new design of meter introduced to the market since the previous 2009 revision. They consist
of an adaptor ( the mechanical body ) which holds the measurement section ( the meter ). The meters
are designed with sizes of 2” - 6” and maximum pressures of 20 Bar, predominantly for onshore use.
There are design differences to traditional multipath meters which have led to the inclusion of
specific clauses relating to the upstream installation requirement, typically no straight lengths being
required and the inclusion of a thermowell being part of the adaptor rather than being installed in the
downstream pipework.
Figure 1 : Typical Cartridge Meter

Variations in Path Configuration
There are many different designs of path configurations in existing meters. These are usually based
on a direct path ( chordal ), reflective path configuration or a combination of both. BS7965:2013
aims to show the configurations that were being implemented by meter manufacturers at the time of
publication. Selection of any design can be influenced by process conditions ( i.e. low pressure or
high levels of attenuating gas components such as CO2 ) or by the installation conditions ( i.e.
perturbations caused by upstream pipework ). The drawings show typical configurations however as
BS7965:2013 is focused on performance, although a meter does not follow one of configurations
shown, it may still conform with the requirements laid down within the document and therefore be
compliant.
Figure 2 : Typical Path Layouts
Meter Size Specification
The major change, which was one of the primary drivers for updating the document, was the
reclassification of meter sizes. During previous versions of the documents the meters available on the
market ranged in size from 4” – 48”. Due to falling offshore gas production rates there has been an
increase in use of the smaller meters, the 4” meter that has always been available and with the
introduction of 3” and 2” meters. It was felt that performance figures for the previous small meters
were not applicable for 3” & 2” meters. With this in mind the sizes were reclassified as follows;
BS7965:2009
Large meters are defined as those meters with nominal bores equal to or greater than 254 mm
(10 inches), whilst small meters are defined as those meters with a nominal bore below 254 mm
(10 inches).
BS7965:2013
Large meters are defined as meters with nominal bores equal to or greater than 300 mm (12 inches).
Medium meters are defined as meters with a nominal bore below 300 mm (12 inches) but greater
than 100 mm (4 inches). Small meters are defined to have a nominal bore equal to or less than 100
mm (4 inches).
The performance requirements for each size meter are defined within Table 2 of BS7965:2013
Figure 3 : BS7965:2013 Class performance criteria
Performance figures should be treated as a minimum requirement i.e. if a meter complies with the
criteria then it is in compliance with the requirements of BS7965:2013. There may be contractual
requirements where the operator may choose to define more stringent requirements. i.e. to have a
small meter operating under the criteria defined for a medium meter.
Repeatability
When looking at repeatability it should be noted that the way that this is defined within standards, by
manufacturers and calculated by test sites is not consistent. BS7965:2013 defines repeatability as;
“Repeatability is defined as the closeness of agreement between successive flow rate measurements
with the same flow meter when obtained under the same conditions (same fluid, same flow meter,
same operator, same test facility and a short interval of time) and without disconnecting or
dismounting the flow meter.”
Within the Performance Specification charts in BS7965:2013, and also ISO17089, the repeatability
is shown as a tolerance band with positive and negative limits. For example for a class 1 meter where
flow is greater than qt, then the requirement is +/-0.2%. This requires that when a meter is subject to
calibration, to comply with the standards, any individual calibration point should fall within this
band. However manufacturers quote a statistical value and calibration facilities also calculate a
repeatability based on statistical values. Depending upon the number of calibration points and spread
of points it may be possible that the manufacturers stated repeatability is met with a point falling
outside of the acceptance band. When calibrating a meter the acceptance criteria should be agreed in
advance and agreed by all parties. If the acceptance is based on a statistical number then the method
of calculation should also be agreed.
Performance Specification Summaries
Performance specifications were modified for all four classes of meters. For Class 1,2 and 3 meters
this involved revising the small meter error limits to medium meter error limits and adding a third set
of error limits to cover the new small meter category. Zero flow reading values were also modified.
For class 1,2, and 3 meters the existing Qmin and Qt values apply. For class 4 meters the
performance limits are only shown as Qmin and Qmax ( of Qt can also be assumed to equal Qmin ).
Class 4 meters tend to be associated with flare gas applications and on review of manufacturers’
literature the conclusion drawn was that these devices are usually only quoted with a minimum
velocity ( Qmin ) and not with a Qt parameter.
Figures 4 to 7 on the following pages illustrate the revised performance specifications.

Figure 4: BS 7965 Figure 5 performance specification for Class 1 meters


Temperature Measurement , Thermowell Position
BS7965:2009 stated, “For unidirectional flow, the thermowell should be installed downstream of the
USM and the distance from the downstream flange face to the thermowell should be at least 5D”.
This contradicted manufacturers’ instructions, contradicted other standards and also allowed for the
thermowell to be installed at an infinite distance downstream of the meter !
The BS7965:2013 revision addresses these issues, “For uni-directional flow, the thermowell should be
installed downstream of the USM and the distance from the downstream flange face to the thermowell
should be a minimum of 2D and a maximum of 5D. If the meter has a feature to have inbuilt
thermowell(s) these may also be used. For bi-directional flow installations, requiring optimum
accuracy, the thermowell should be located either: at least 10D from either USM flange face; or
between 5D and 10D from either USM flange face, in which case, the meter should be
flow-calibrated in both directions with the thermowells installed and two correction factors
derived.”
The reference to in built thermowell(s) is applicable to cartridge type meters which are commonly
installed with either 1 or 2 measurement points.
Temperature Measurement, Thermowell Insertion Depth
In the previous document the insertion depth was not referenced but it has been customary practice to
install temperature measurement within the middle third of the pipe. Research carried out and
presented at previous North Sea Flow Measurement Workshops [1] has shown that measurement
can be closer to the pipe wall or as an alternative surface mount technology may be used. Surface
mount technology may be preferential to the use of thermowells on the smallest pipes e.g. 2” or 3”
where the use of a thermowell may cause a major obstruction within the pipe or expanded pipe
sections to be used
BS7965:2013
“The recommended insertion depth for the thermowell and the pockets is between D/10 and D/6.67.
Where the insertion depth >D/3.33 then the meter run might not need to be insulated. Where
insertion depth <D/3.33 (or where surface mount technology is used) then the meter run should be
insulated from the meter inlet until at least 1D downstream of the temperature measurement point”.
Previous papers [1] concluded,
“Gas temperatures measured on profile probes were <0.2ºC of mean gas temperature
 For gas velocities > 1m/s

 Demonstrating centre third thermowell measurement not necessary”
Zero Flow Test
BS7965:2013 contains updated acceptance criteria for zero flow tests on meters primarily associated
with Class 1 and Class 2 meters.
“This subclause describes the zero flow test that can be carried out as part of the field verification of
the meter performance. The system would have to be taken off line for about 30 min but this could
prove to be beneficial, particularly for Class 1 and Class 2 meters. The zero flow test is carried out
on a product, at operating pressure and zero flow conditions and isolated at the closest valves to the
meter; once flow is confirmed to be zero the individual path velocities are measured. During the
Factory Acceptance Test under controlled conditions, a limit of <±0.006 m/s for
each acoustic path should be achievable when averaged over a period of not less than 100 s. In the
field under true, verified zero flow conditions, a limit of <±0.012 m/s for each acoustic path should
be achievable when averaged over a period of not less than 100 s. If in any doubt, the manufacturer
should be consulted.”
A zero flow test will be carried out in one of three situations,
1. Factory acceptance tests;

2. Flow Calibrations;
3. In service tests.
The factory zero flow test is carried out in a controlled situation, it is primarily a verification of the
geometry of the meter body and the parameters entered into the meter signal processing unit. The
factory test is carried out on gas with a known composition, which it typically pure Nitrogen or Air
with a stable pressure and temperature. A typical factory test arrangement is shown in figure 8.With
this high level of stability it will be possible to meet a limit of <±0.006 m/s for each acoustic path.
Typical factory zero test values are shown below in figures 9a and 9b.
Figure 8 : Factory Zero verification

Figure 9a : Typical Factory Zero Flow Results _ 10” Meter
Figure 9b : Typical Factory Zero Flow Results _ 6” Meter
For class 1 and 2 meters it is also usual practice to carry out a zero flow test as part of a high pressure
calibration. In this case data is taken either before the calibration commences or as flow is stopped at
the end of the calibration. There will, however, be an optimum time to take the data. As flow ceases
there will be a period before stabilisation occurs, once the gas is stable data can be taken. An
example is shown in Figure 10. After a period of time stability will be lost mainly due to temperature
effects and it will not possible to take useful data. Experience shows that at a calibration facility it
will be possible to meet a limit of <±0.012 m/s for each acoustic path. Under some circumstances it
is possible to meet the lower limit of <±0.006 m/s for each acoustic path but this is not a
requirement.
Figure 10 : Typical Zero Flow values at a test facility
A zero flow test can also be a useful indication of meter performance once the meter has been
installed and is operational on site. The test situation will not be as stable as in the factory or at the
high pressure calibration and a zero flow test can become a valve leakage test if valves are not fully
sealing. Temperature effects will also be greater than the previous examples especially if the meter is
exposed to direct sunlight or there is a high differential temperature between the gas temperature and
the ambient temperature. Assuming that metering stream isolation valves are not leaking then it
should be possible to meet a limit of <±0.012 m/s for each acoustic path however it will probably not
be possible to meet a limit of <±0.006 m/s for each acoustic path.
Example of on site zero flow test with an 18” meter ; Table 1
Data from an 18” chordal meter, exposed to direct sun with a differential temperature of 15 degC (
gas temperature lower than ambient temperature ) shows that once flow is stopped the velocities for
the 4 chords are ;
The meter would pass the limit of <±0.012 m/s for a period of 112 seconds and then 448 seconds
once flow ceases.
But the meter would pass the limit of <±0.006 m/s for a period of only 11 seconds and then 32
seconds once flow ceases.
On this basis, this confirms that the limit of <±0.012 m/s is practical for in situ testing.
Path 1 Path 2 Path 3 Path 4
m/s m/s m/s m/s >0.012 >0.06
Flowing conditions 4.628 5.051 5.054 4.409 m/s m/s
T=0s -0.005 -0.012 0.01 0.01
T=21s 0.004 -0.001 -0.003 0.005
T=32s 0.002 -0.001 -0.007 0.004
T=112s 0 -0.002 -0.002 0.015
T=128s 0.006 -0.003 0.002 0.009
T=249s -0.003 0.006 0.001 0.001
T=291s 0.001 0.007 0 0.001
T=455s 0.007 0.004 -0.012 0.004
T=576s 0.004 0.014 0.06 0.09
Table 1 ; 18” meter zero test on site
For class 4 meters, typically flare gas meters, the zero flow test may take place using specialised test
equipment such as a zero flow box. The function of a zero flow test box is to give a stable, controlled
environment where transducer performance can be verified.
Zero Flow Test : Flare Gas Meters
For flare gas meters ( typically class 4 meters ) the transducers are usually removed and are subject
to a test in a zero flow test box; see Figure 11. This gives a set path length and allows the air pressure
and temperature to be measured. The measured speed of sound can be compared against a theoretical
value as well as the gas velocity ( zero ) to be checked.
Figure 11 : Zero Flow Test Box

Shift between calibrations
BS7965: 2009 gave a maximum allowable shift between calibrations on ±0.3% of FWME.
Experience showed that this figure was not always practical for small meters so the requirements
were modified as follows ;
“For meters >4 in, a typical tolerance of ±0.3% of FWME (flow weighted mean error) should be
allowed between subsequent calibrations. For meters ≤4 inches, the allowable shift is ±0.5% of
FWME.
Examples for shifts in performance are given in figures 12,13 and 14, these show that the defined
values for allowable changes in FWME can be met for the three sizes of meters.
Large Meter
2005 Calibration FWME : + 0.0187 %

2004 Calibration FWME : + 0.0985 %
Change in FWME : - 0.0797
Figure 12 : Shifts at Calibration for a Large Meter
For the large, 20” ; figure 12, we can see that the change in performance results in a change in
FWME of less than ±0.3%. Although we show only one set of results the meter has been to
subjected to multiple recalibrations and in all cases the change in performance has complied with
requirements laid down in BS7965:2013
Medium Meter
Figure 13 : Shifts at Calibration for a Medium Meter
For the medium, 10”, size meter there are 3 sets of results; figure 13. The worst case change occurs
from the November 2013 to July 2014 calibration, a change in FWME of 0.24 to 0.28 ( a change in
FWME of 0.04 ) It can be seen that the change is less than ±0.3% of FWME so the results of the
calibrations would be acceptable under BS7965:2013
Small Meter
Figure 14 : Shifts at Calibration for a Small Meter

For the smallest size meter, 2”, there are results for two meters; Figure 14. When we look at repeat
calibrations we see a range of change of between 0.08% of FWME to 0.43% of FWME with an
average change of 0.28% of FWME ). It can be seen that some of the changes would be greater than
±0.3% of FWME however the change is always less than ±0.5% of FWME so the results of all of the
calibrations would be acceptable under BS7965:2013
For all sizes of meters there may be occasions where the change in FWME is greater than that
allowed. There may be a number of reasons for this not limited to, but including the following;
1. Meter contamination, either through pipeline particulates, liquid or in the process of

removing the meter from service. To this end it is recommended that a diagnostics file is
taken from the meter prior to removing the meter from service. If a change in performance is
due to contamination it should be possible, using logged data in the meter, flow computer or
supervisory system to determine when this occurred,
2. Change in test site conditions, this can raise the following questions,
a. Is the meter being calibrated at the same site ?
b. Is the site set up the same, if a flow conditioning plate is being used is it the same
type, at the same location and of in same orientation as for the previous calibration ?
3. Parameter check, if values have changed since the previous calibration this may result in a
change in performance. This can usually be checked either through checksum comparisons or
by comparing parameter list automatically.
Path failure simulation and exchange of components
“Where a Class 1 or Class 2 meter remains in service in the event of path failure, the effect of the
failure should be determined during meter calibration by simulating the failure of one or more paths.
The test should be carried out at or around the mid-point of the expected operating range of the
meter. During the test, the flow rate should be varied by 20% of the flow rate to ensure that the meter
responds appropriately.
The manufacturer should demonstrate the capability of the meter to replace or relocate transducers,
electronic parts and software without a significant change in meter performance. This should be
demonstrated for: the electronics; transducers of different path types.
When components are exchanged, the resulting shift in the FWME of the meter should not be more
than 0.2%”.
This new section within BS7965:2013 seeks to address the effect of replacing a component in the
field and to put a limit on the change in performance after such a replacement. Component failure
may require the operator to change out either the electronics unit or a pair of transducers.
Results are shown for a 3” meter tested as part of a high pressure calibration carried out on natural
gas at 50 Bar; Figure 15. Results show the initial calibration, three points repeated with one set of
transducers exchanged ( for a spare set ) and the meter with the original transducers installed.
For all scenarios the resulting shift in the FWME of the meter is less than 0.2%, the meter would be
compliant with the requirements of BS7965:2013 and this also shows that the procedure in the
document is valid even for small meters.
USM ........
1.50 Error Curve
1.25
1.00
0.75
0.50
0.25
DN80 Cal
0.00 March 2015
0 50 100 150 200 250 300 350 400 450
-0.25
+ve Tol
-0.50
-0.75
-ve Tol
-1.00
-1.25
-1.50
Flow (m3/hr)
Figure 15 : Change in Performance with transducer change out for a Small Meter
FWME with original transducers ( as found ) 0.15
FWME with 1 pair of transducers replaced 0.05
FWME with original transducers replaced 0.12
CBM ( Condition Based Maintenance / Monitoring )
UK offshore operators have, historically, calibrated meters using a time based maintenance
philosophy. This has been followed independent of the value of product flowing through a meter and
the level of diagnostics available in the metering system. The adoption of a risk based philosophy,
using the diagnostics available either directly from the meter or the metering system addresses the
following concerns [2];
 Safety concerns around the removal and handling of large pieces of equipment
 Planning and scheduling to ensure minimal down time without any gas deferral
 ETS legislation governing the venting and release of Hydrocarbon gases
 Availability of test slots, particularly for meter sizes above 10”
 The calibration facilities claiming uncertainties of approximately ± 0.23%, not the
“order of magnitude” smaller uncertainty (± 0.03%) that fundamental metrological
principles require for a true calibration.
 Potential for damage to the meters during removal, transportation or reinstallation .
Gas USFM meter diagnostics may be classified depending on the type of information that they
provide:
 Functional (information on the physical operation of the meter)

 Process (information on the fluid properties, flow profile, etc.)
 System Performance (information on the overall measurement system)
This new section, covering CBM, was implemented based on information first included in DECC
“Guidance Notes for Petroleum Measurement Issue 8”. The section within the DECC document was
formulated following a seminar hosted by DECC in Aberdeen in June 2011. Attendees were the main
manufacturers of fiscal meters currently being used in the UK and meter operators. Latest advice
from the OGA [3] recommends that operators of meters should adopt a combined “risk-based” and
“conditioned base” rather than a simple “time based” approach to maintenance where possible.
The guidance was based on common values ( listed below ) that are generated by all recognised
meters and does not include bespoke manufacturers solutions, the guidelines cover data such as ;
 Speed of Sound
 Automatic Gain Control
 Signal to Noise Ratios
 Performance
 Temperature
The approach to CBM is qualative rather than quantative. There is a recognition that there is an
opportunity to add to this as ISO17089 part 1 is updated.
Summary
The BS7965:2013 working group believe that the document, which is based upon extensive end user
experience, is practical, pragmatic and is performance focused. As a group our hope is that the
industry will adopt the practices contained within the latest revision and that the work carried out
will be a valuable input into other documents, especially the work being carried out to revise part 1
of ISO17089.
References
1. Experimental Research into the Measurement of Temperature in Natural Gas Transmission

Metering Systems – Sarah Kimpton and Bob Ingram ( Oil and Gas Focus Group ; April 2015)
2. On-line Condition Based Maintenance of Gas Ultrasonic Flow Meters – Stephen Peterson
and Jan Peters ( CEESI/VSL European Flow Measurement Workshop ; March 2014 )
3. OGA Policy Statement – Maintenance Strategies on Fiscal Measurement Systems ( August

2015 )
NSFMW 2015
Design of a subsea fiscal oil export metering system
Dag Flølo
Security Classification: Open - Status: Final Page 1 of 24

Table of contents
1 Introduction ................................................................................................................................................... 3
1.1 Why a subsea fiscal oil export metering station? ............................................................................................ 3
1.2 Requirements to design - Authority requirements........................................................................................... 3
1.3 Recognized standards .................................................................................................................................... 4
2 Basic design of the metering system ......................................................................................................... 4
2.1 Introduction to basic design ............................................................................................................................ 4
2.2 Components that already exists for subsea application.................................................................................. 5
2.3 Component to be developed and qualified - Subsea ultrasonic flow meters .................................................. 5
2.4 Selection of proving method for a subsea design ........................................................................................... 5
2.5 Sampling system............................................................................................................................................. 6
2.6 Basic design: 2 metering runs – 1 line meter in each –1master meter in a bypass ........................................ 8
2.7 Uncertainty analysis for the basic design........................................................................................................ 9
3 Design considerations ............................................................................................................................... 15
3.1 Failure Mode and Effects Analysis................................................................................................................ 15
3.2 Consideration of measurement uncertainty .................................................................................................. 16
4 An alternative metering system................................................................................................................. 17
4.1 Arguments for a metering system with 3 meters in series ............................................................................ 17
4.2 Metering system with 3 meters in series - Design ...................................................................................... 18
4.3 Metering system with 3 meters in series - General arrangement ............................................................... 19
4.4 Metering system with 3 meters in series - Uncertainty analysis ................................................................. 19
5 Selection of the preferred system ............................................................................................................. 20
5.1 Methodology for selection of preferred system ............................................................................................. 20
5.2 Assumptions put into the evaluation ............................................................................................................. 21
5.3 Evaluation of alternative metering design ..................................................................................................... 22
5.4 Preferred design for a subsea fiscal oil export metering system .................................................................. 23
6 Summary...................................................................................................................................................... 24

1 Introduction
1.1 Why a subsea fiscal oil export metering station?
Tankers used as floating storage and offloading units has an expected life time of 20-30 years. This life time is
often shorter than the expected lifetime of an oil field. During the life time, these tankers use a substantial
amount of fuel, emit large amounts of CO2 and have significant operational costs. At the end of the life time
extensive repairs and recertification or replacement is required.
A subsea storage will have a longer life time than a floating tanker and also have lower operational costs.
Hence there are substantial business drivers to develop a subsea storage concept which can replace tankers
used as floating storage units. To export oil from a subsea storage, a fiscal oil export metering system will be
required under the Norwegian statutory regime. This is thus a strong driver to start developing a subsea fiscal
oil export metering system.
A subsea fiscal oil export metering system must:

1. Measure flow, pressure, temperature, density and water fraction
2. Provide a way of collecting representative liquid samples
3. Provide traceable measurements within the statutory uncertainty requirements
4. Be retractable / exchangeable via module replacement
5. Provide redundancy and have condition monitoring capabilities
6. Be robust to the expected variation of fluid parameters
Statoil is currently developing a subsea fiscal oil export metering concept. The paper will present this concept.
1.2 Requirements to design - Authority requirements
In Norway, the authority requirements to fiscal measurement can be found in “The Measurement Regulations: “
REGULATIONS RELATING TO MEASUREMENT OF PETROLEUM FOR FISCAL PURPOSES AND FOR
CALCULATION OF CO2-TAX
Requirements that are significant for station design are included in this section.
Section 13 – Parallel metering runs

“On sales metering stations the number of parallel meter runs shall be such that the maximum flow of
hydrocarbons can be measured with one meter run out of service,”
Section 13 – Shut of valves

“Shutoff valves shall be of the block and bleed type. All valves of significance to the integrity of the metering
station shall be accessible for inspection to secure against leakage.”
Section 13 – Design in accordance with recognized standards

“The measuring system shall be planned according to the requirements in this regulation and according to
recognised standards for such measuring systems.”

Recognized standards are identified in the document: “Standards relating to measurement of petroleum for
fiscal purposes and for calculation of CO2-tax” available at the Norwegian Petroleum Directorate” net site.
Section 14 – Permanent equipment for calibration shall be available

“If other types of flow meters [than turbine meters] are used for liquid metering, permanent equipment for
calibration of the metering device shall be available.”
1.3 Recognized standards
API MPMS 5.1

API MPMS 5.1:2008 “General Considerations for Measurement by Meters” is regarded by the measurement
regulations to be a recognized and significant standard for design of fiscal oil export metering systems.
A few essential sections are included here.

API MPMS 5.1 – Meter proving
“5.1.9.4.1 Each meter run should be connected to a permanent prover or connections should be provided for a
portable prover or master meter to obtain and demonstrate the use of meter factors that represent current
operations. The proving methods shall be acceptable to all parties involved.”
API MPMS 4.5 – Master meter provers

“3.3 Indirect master meter proving method.
This proving method requires that the line meter and a master meter be in series. The line meter is proved by
comparison to the master meter whose meter factor was determined by a previous direct proving on a different
flow stream and/or conditions. This method has a significant higher uncertainty than the other methods
because a displacement prover is not in series with the master meter and the line meter.”
“Master meter proving is used when proving by the direct method cannot be accomplished because of meter
characteristics, logistics, time, space, safety, and cost considerations.”
ISO 17089 – Reference meter method for ultrasonic meters in series

This standard is valid for gas metering applications. For gas metering applications it has become common to
put two ultrasonic meters in series. This standard describes a method for monitoring of the quality of the flow
meters:
“Annex C (Informative) The flow reference meter method for ultrasonic meters is series
C1. General
With two ultrasonic meters in series, a systematic approach, the flow reference meter method, may be
employed to monitor the quality of the meters (with the exception of common-mode errors).”
2 Basic design of the metering system
2.1 Introduction to basic design
First step in the development is to put up a feasible basic design. The intention with the basic design is to arrive
at a design that meets statutory requirements and industrial standards. The basic design will then form the
basis for a Failure Mode and Effect Analysis and an analysis of the total risk for loss of profit. The following

sections summarize these assessments. Also it is important to identify components that are already existing
and qualified, and which components will have to be developed.
2.2 Components that already exists for subsea application
A number of subsea components already exist and can be regarded as qualified. The following components
are here regarded as sufficiently qualified:
 Horizontal and vertical pipe connectors
 Tubing connectors
 Relevant signal and power connectors
 Block valves
 Double block valves
 Temperature and Pressure transmitters
2.3 Component to be developed and qualified - Subsea ultrasonic flow meters
Due to long term stability, large turn down range, low pressure loss, linearity and low uncertainty - ultrasonic
flow meter has become the most used flow meter in new topside metering station designs.
Several manufacturers provide ultrasonic flow meters for topside application. Without putting more
consideration into it, it is taken as granted that a subsea version can be designed for any of the topside
ultrasonic flow meters. On a preliminary basis it is also taken for granted that the uncertainty requirements can
be met by using an ultrasonic flow meter. The uncertainty is further addressed in a later section.
2.4 Selection of proving method for a subsea design
Without further consideration it is here taken for granted that due to the need for frequent maintenance it will
not be feasible to develop a conventional large volume prover or a conventional small volume piston prover for
subsea application. In such proving systems there are too many moving parts that are susceptible for wear and
tear and hence will need some kind of maintenance.
The only remaining proving system will then be an indirect master meter proving system. The master meter,
and the line meter, will have to be adjusted and calibrated on a similar fluid at a topside calibration facility
before installation in the subsea facility.
Note: It is explicitly stated in API MPMS that the master meter must be in series with the line meter. It is not
explicitly stated that the master meter need to be in a by-pass to the line meter. However, in a typical drawing
in the standard, it is indicated that the master is in a bypass to the line meter.
Conventional master meter solution contains a master meter in a by-pass. However, the development of
ultrasonic flow meters has opened up for the possibility to install the master meter in line with the line meter. In
gas metering systems an inline master meter in series with the line meter has become quite common. Also in
oil metering systems this is now becoming more common.

2.5 Sampling system
It is considered that for a subsea fiscal metering system, a sampling system in accordance with the Norwegian
statutory requirements for fiscal measurement will be required. It is further considered that a subsea sampling
system, with transport of the sample in the umbilical to a nearby topside facility and sub-sampling by a
conventional topside sampling will be feasible. This will be elaborated in the following section.
The following extracts are intended to give an overview over the most fundamental statutory requirements.
 “Sampling shall be carried out in a manner which ensures that representative amounts are sampled.”
 “Sampling shall be automatic and flow proportional. In addition it shall be possible to carry out manual
sampling.”
 “The sampling probe shall be placed at a location where a representative sample will be
obtained. A mixing device may be required.”
 “Manual spot sampling shall be available also with automatic sampling/fast loop out of service.”
The following high level considerations are made regarding the design.
 The sample can be extracted from the main pipe by using a conventional sample probe
 The sample probe should be placed in a vertical section upstream the flow meters
 The sample stream can be transported to the top side facility via tubing in the umbilical
 To overcome the pressure loss in the sampling line, and to elevate the sample above sea level up to
the platform, a sampling pump will be required in the subsea system
 To be able to perform maintenance on sample probe, pump and valves - the sampling system should
be included in a retrievable module or system.
 At the top side facility, a conventional system for fiscal sampling will have to be installed for sub-
sampling of the sample stream from the umbilical.
 For redundancy, the system should contain two independent sampling systems
 A conventional flow metering control system should be located at the topside facility to control both the
metering system and the sampling system.
 The time delay for transport of the sample from the subsea must be accounted for by the metering
control system to obtain representative flow proportional sampling
The system should be designed with redundancy within the sampling system. However – If one of the sampling
systems fails, retraction and replacement of the sampling module will have to be scheduled. Detail design will
decide if the sampling system is part of the metering module or in a separate module.
2.5.1 Detailed design of the sampling system
2.5.1.1 Design considerations – location of the sampling probe
If the sampling probe is placed in a horizontal pipe, a static or dynamic mixing device will be required to ensure
proper mixing. In a horizontal pipe, velocities above 6 – 7 m/s will be required to ensure proper mixing and
equal distribution of water in oil over the pipe cross section. The system may be driven by difference in density
between water and oil – and not by pumps. In this case static mixing is not desired, as a static mixer will
introduce a pressure loss, and consequently also lower flow rates. A dynamic, pumped mixing system is not
desirable because of the complexity it would add to a subsea system.

However, it is evident from ISO 3171 sub - clause 5.2 that installation of a sample probe in a vertical pipe
section is preferable to get a homogeneous mixture of water in oil at the sampling point. It is also indicated in
Figure 10.5 til 10.8 in NFOGM Handbook of Water Fraction Metering, that in a vertical pipe, equal distribution
of water over the cross section can be expected for velocities above 1 m/s even for water fractions up to 5 %.
As the velocities will be well above 1 m/second in normal operation, sufficient mixing of water in oil can be
expected in the vertical pipe sections. To avoid removal of liquid from the pipe after it has been measured, the
sampling probe should be placed at the vertical pipe section at the entry of the metering system. Exact location
in the vertical section will have to be optimized based on evaluation of effects of the upstream pipe
configuration. Due consideration will have to be paid to the centrifuge effect of Pipe bend.
2.5.1.2 Design considerations - Flow velocity in the sampling tube
To maintain a representative mix of water and oil, and to avoid settlement of free water, the flow velocity in the
umbilical need to be kept in the turbulent flow range. It is assumed that a flow velocity higher than 1 m/s will be
required to avoid accumulation of free water, or solids, in the sampling line inside the umbilical.
A theoretical upper limit for flow velocity is assumed to be in order of magnitude 5 m/s. However, due to the
pressure loss it is assumed that a velocity of approximately 1 m/s will be achievable. Velocity down to 0,5 m/s
may be evaluated.
Tubing in the umbilical is typically 12 mm super duplex tubing. The maximum operational pressure in the tubing
is typically 690 Bar. It is regarded fully feasible to pump the sampling stream at a sufficient flow rate through
tubing in the umbilical.
The system should be design for a sample flow rate of minimum 1 m/s through the tubing in the umbilical.
2.5.1.3 Design considerations - Sample transport time
Another design consideration to be made is how to handle the time delay from the subsea metering station to
the top side facility.
An offshore loading operation will take about 15 – 20 hours. Further it is assumed that the subsea facility will be
placed about 1000 m from a top – side facility. Assuming a flow velocity of 1 m/s, the sample will use 1000
seconds to be transported from the subsea facility to the top-side facility. 1000 seconds is equal to about 17
minutes. Hence the automatic sampling system at the topside system will have to be able to handle a time
delay of 17 minutes.
The following design will handle a time delay of 17 minutes:

First, topside sub - sampling of the content coming from the sampling line will have to be delayed 17 minutes to
be sure that the fluid sample is representative for the batch. A flowmeter should be placed topside to allow
more exact online calculation of the delay, based on flow rate and volume of the sampling line.
Second – Fluid sampling shall be flow proportional. Consequently the sampling rate should be proportional to
the flow rate measured 17 minutes ago, and not to the flow rate at the moment.

Third – sub - sampling of the flow from the sampling line will have to continue for 17 minutes after the flow
through the subsea fiscal oil metering station has been stopped. Hence the pumps will have to be running, and
the fluid reservoir for the main flow in the main pipe should be open for 17 minutes after the offloading has
stopped.
It should thus be evident that the sampling can be both representative and flow proportional by just taking into
account the transport time for the sample from the subsea facility to the top – side facility. The sampling system
can be controlled by a conventional metering control system placed at the top – side facility.
2.6 Basic design:

2 metering runs – 1 line meter in each –1master meter in a bypass
The following design is regarded to meet significant statutory design requirements:
QS = Sampling System, FT = Ultrasonic Flow Meter, PT = Pressure Transmitter, TT = Temperature

Transmitter.
In this figure, the fluid is flowing in from top left side, passing the sampling systems and then passing through 1
of the 2 metering runs in parallel. The second metering run is a spare metering run. The flow from each of the
two metering runs can be routed via the master meter run in the middle for calibration and adjustment of the
line (duty) meter.
Rather than having many pipe connectors within the metering system and thereby increasing the complexity,
the system design shown here has only 2 pipe connectors. These are shown as a little rectangle where the line
is passing the stippled box. The systems within the stippled box are retractable.

As the metering system is placed at the seabed, calibration will require retraction of the system, and
replacement with a calibrated spare system. Also valve leakage, or failure to prove valve tightness, will require
the system to be retracted and replaced. The complete system will have to be replaced if any of the
components in the system need maintenance.
2.7 Uncertainty analysis for the basic design
2.7.1 Introduction to the uncertainty analysis
An uncertainty analysis is required both by the measurement regulations and by authority guidelines for Plan
for Installation and Operation of facilities for transport and utilisation of Petroleum.
Norwegian Society for Oil and Gas measurement is currently developing a tool for performing uncertainty
analysis of fiscal oil metering stations. This tool is utilised for the uncertainty analysis in this document.
As the actual metering systems has not been built and test, the uncertainty analysis is an a priori estimate of
uncertainty. This a priori estimate is based on a number of assumptions which are identified below. The
accuracy of the assumptions below is not of critical importance for the result.
Operational pressure 30 bara

Operational temperature 45 degC
Viscosity 1 cP
Velocity 7 m/s
Pipe diameter 0,5 m
Density 800 kg/m3
Reynolds number 2,8 E6
Further – it is assumed that all uncertainties for the individual instruments in the metering system are equal to
the uncertainty requirements in “The measurement regulations”. It is also assumed that internal corrosion of the
pipe and wax deposits is avoided by design.
The fluid will be sampled through the umbilical and analysed at a topside facility in accordance with
conventional methods. Hence density and water cut will also be determined within the uncertainty requirements
in The measurement regulations. The uncertainty in determination of water cut (0,05%) is excluded from this
evaluation. It will be showed that this uncertainty contribution will not have a significant effect on the result.
2.7.2 Flow profile and fluid effects on ultrasonic meters
The ultrasonic meter will have an uncertainty component which may be termed “Flow profile and fluid effects on
the master meter” this uncertainty component is meant to cover:
 The effect of the difference in pipe configuration between the laboratory and field installation.
This effect is regarded to be neglible as the meter will be calibrated using the same upstream pipe
configuration.
 The effect of difference in pipe roughness between the laboratory and field installation.
This effect is regarded to be neglible as the pipe will be made by a non-corrosive material

 The reproducibility of the master meter between laboratory conditions and operational conditions
where viscosity, density, Reynolds number and flow profile will vary
Note that uncertainty contribution arising from the linearity of the meters is covered by a separate uncertainty
contribution.
For the last bullet point it is recognized that the meters will be operated at Reynolds numbers far into the
turbulent range. It is also recognized that the flow profile is quite flat at such Reynolds numbers. The following
calibration curve is obtained by using original weighting factors and no velocity profile correction for an
ultrasonic meter. The curve demonstrates that the calibration curve becomes flat at high Reynolds numbers.
The curve also demonstrates that the Reynolds number has to be taken into consideration when evaluating
uncertainty for ultrasonic meters.
The figure is taken from the following paper: “Qualification of Fiscal Liquid Ultrasonic Meter for Operation on
Extended Viscosity Range” By Øyvind Nesse and Tore Bratten - presented at the North Sea Flow
Measurement Workshop in 2013.
2.7.2.1 Uncertainty estimate for “Flow profile and fluid effects” based on experience from a
recent project
Experience from a recent project provides some insight into the possible influence from fluid variation on the
measurement result. 12” ultrasonic meters were used in this project. The three meters were calibrated on
gasoline 0,5 cP, jet fuel 1,5 cP and domestic fuel 4 cP. The calibration results demonstrates that the meter
factor varied approximately 0,15 % from the average meter factor for these three fluids. Investigations are
currently being conducted to understand and reduce the fluid effect for these particular meters. However, these
tests give an indication about the level of uncertainty that may arise from flow profile and fluid effects.

The calibration results are presented in the following figure. The meters were neither calibrated nor adjusted for
the calibration fluids before the calibration documented in these figures.
It should be noted that the calibration curve clearly and consistently is influenced by the viscosity of the
product. When a systematic effect has been determined during calibration, it is reasonable to assume that
operational routines will ensure some monitoring of the viscosity of the product. The calibration curve for the
viscosity which is closest to the viscosity of the measured product can then be selected in operation.
As demonstrated by this calibration curves, the change in meter factor due to viscosity was up to 0,15 % in this
particular case. If we assume that the calibration curve for the viscosity closest to the operational viscosity is
selected, it is reasonable to assume that the maximum uncertainty due to variation in viscosity is approximately
half this change (0,075 %).
2.7.2.2 Uncertainty estimate for “Flow profile and fluid effects” based on experience from the
Snorre platform
The following figure contains operational experience for a meter at Snorre A, which were operated on a similar
fluid as studied in this analysis. The fluid had a viscosity of approximately 2,5 cP. The experience is
documented in the paper: “Operational experience with liquid ultrasonic meters” A paper for Presentation at the
29’th North Sea Flow Measurement Workshop in Tønsberg 2011 By Dag Flølo et al.
The charts below shows all K-factors achieved at the yearly on-site calibration performed by using real product.

Operational experience with this meter demonstrates that the variation in meter factor due to both linearity, flow
profile and fluid effects is approximately 0,1 % @ 95 % confidence level in this particular case. Similar results
were obtained for a similar meter at Snorre B. The operational experience also demonstrates that there is no
sign of significant aging effects. Hence it is reasonable to regard the influence from aging to be insignificant.
2.7.2.3 Uncertainty estimate for “Flow profile and fluid effects” based on manufacturers data for
a newly developed meter.
The presentation by Pico Brand from Krohne at NFOGM annual technical workshop 2015 “Are liquid ultrasonic
flowmeters independent of fluid properties?” contains data for a newly developed 7 beam meter that were
tested on a variety of products. As can be seen in the figure below, the meter demonstrated an exceptional
linearity over a large range of Reynolds number.

Judging from these data it will be reasonable to assume that the influence of flow profile and fluid effects on
meter factor will be well within 0,1 % for this meter. The capabilities will have to be confirmed by testing.
2.7.3 Conclusion about the uncertainty component “Flow profile and fluid effects”
Based on the experience presented in this document and taking into account the additional premises that
significant systematic errors will be revealed and rectified under governance of the operational procedures it
seems reasonable to assume that the uncertainty component “Flow profile and fluid effects” will be in the order
of magnitude 0,15 % @ 95 % confidence level.
2.7.4 Uncertainty budget for the basic design
The assumptions above were then used as input variables in the uncertainty tool currently being developed by
the Norwegian Society for Oil and Gas measurement. As can be seen from the configuration picture below, the
tool matches perfectly the basic design:

Based on the assumptions in previous sections in this document, the following chart shows the uncertainty
contributions and relative expanded uncertainty at 95 % confidence level for the measured standard volume of
oil.

The total uncertainty in Standard Volume is estimated to be approximately 0,19 % @ 95% confidence level. If
we take into account an additional uncertainty of 0,05 % @ 95% confidence level from determination of water
in oil the resulting uncertainty will be 0,2 % @ 95% confidence level for the basic design.
Hence, the design will meet the uncertainty requirement of 0,3 % of standard volume as required by The
Norwegian measurement regulations.
3 Design considerations
3.1 Failure Mode and Effects Analysis
3.1.1 Failure mode – Need for calibration and adjustment
For the purpose of this evaluation it is assumed that calibration will only be performed if the condition
monitoring indicates that calibration and adjustment of one or several components will be required. It is further
assumed that the stability of the components is such that replacement for the sole purpose of calibration will
not be required during the lifetime of the system. However it is assumed that the system will be retrieved during
the lifetime. And that the system will then be replaced by a calibrated spare system.
3.1.2 Failure mode – Corrosion of the internal surface of the pipe
It is recognised that change of the internal pipe surface may affect the performance of a flow meter. A master
meter will be exposed to much of the same conditions as the line meter. Hence the master meter will also be
exposed to the same failure mode as the line meter. Consequently corrosion will have to be avoided by
constructing the metering systems with materials than cannot corrode under the operating conditions.
Without putting more consideration into the selection of pipe material or coating strategy, it is here concluded
that the metering system must be constructed in a way that prevents corrosion of the internal surfaces.
3.1.3 Failure mode – Wax deposits on the internal surfaces
It is recognised that deposits on the internal surface will affect the performance of a flow meter. A master meter
will be exposed to much the same conditions as the line meter. Hence the master meter will also be exposed to
the same failure mode as the line meter. Consequently wax deposits will have to be avoided by heating and/or
chemicals.
Wax can be avoided by keeping fluid temperature above the wax appearance temperature. Wax can be
removed by heating the fluid to above the wax dissolution temperature. Without putting more consideration in
the wax strategy here, it is here concluded that if there is a wax potential, the metering system must be
constructed in a way that prevents wax deposits on the internal surface of the pipe.

3.1.4 Failure mode – Flow meter - Failure or deterioration
A remaining failure mode to be evaluated is failure or deterioration of the meter electronics or transducer. It is
reasonable to consider that such failure can happen. It is also reasonable to assume that it is equally likely that
such failure can happen to the master meter, as it is that the failure can happen to the line meter.
By putting the master meter in a by-pass, it is regarded implicit that the master meter will only be used for
calibration of the line meter, and that the duration of periods where the meters are run in series will be limited to
the time needed for calibration within the repeatability requirements.
It is also considered that the best way to reveal such failure would have been long term comparison of the flow
meter with the master meter. One operational method that has been developed for monitoring of flow meters is
described in ISO 17089 “Measurement of fluid flow in closed conduits – Ultrasonic meters for gas.
ISO 17089 contain a reference meter method for ultrasonic meters in series. The method is based on the
establishment and maintenance on the difference between hourly volume totals for the two meters at metering
conditions.
By putting two flow meters in series a large historic database over the difference between the two flow meters
can be established and maintained. By putting the master meter in line with the line meter, the reference meter
method can be fully employed. By having the master meter in a temporarily used bypass, the reference meter
method cannot be fully employed.
3.1.5 Failure mode – Valve leakage or failure to confirm that a valve is not leaking
If a master meter is placed in a bypass to the flow line, it is clear that the outlet valve from line meter run will
have to be tight. A leakage through the metering run outlet valve will cause a systematic mismeasurement. As
a consequence, a system which can be used to demonstrate that the valve is tight will be required. Leakage
through one of the two seals within the valve will remove the possibility to confirm that the valve is tight.
Consequently if one of the seals is leaking, or if the leakage detection system fails, the system will have to be
retracted, replaced and overhauled.
From experience, the frequency for this failure is quite high. The consequence is also quite dire, as it is
required to keep the valves tight and hard to quantify a leakage. This failure mode will be quite significant for
systems having this kind of valves in the system.
3.2 Consideration of measurement uncertainty
When the master meter is placed in a by-pass it is implicit that it will be used for calibration and adjustment of
the line meter, and that it will not be used for continuous measurement. Consequently the measurement
uncertainty of the master meter will be transferred to the line meter at proving, with the additional uncertainty
contributions from repeatability and the difference between flow rate at calibration and flow rate at
measurement. The resulting measurement uncertainty will be higher than the measurement uncertainty for the
master meter alone.
If the master meter is placed in line with the line meter, the master meter can be used for measurement, in
addition to its use as a master meter. In addition it can also be used use in the reference meter method. From

Guide to the expression of Uncertainty in Measurement, it is known that the measurement uncertainty will be
reduced if the measurand is the average of 2 or more measurements. For uncorrelated uncertainty
contributions, the resulting measurement uncertainty will be equal to the measurement uncertainty for each
meter divided by the square root of the number of meters.
If we first assume that the uncertainty of one meter is 0,15 % at 95% confidence level, the resulting uncertainty
for measurements based on the average of meters in series will then be:
For measurement based on 1 meter: 0,15 % * 1 / (Square root of 1) = 0,15 * 1,00 = 0,15 %
For measurement based on 2 meters: 0,15 % * 1 / (Square root of 2) = 0,15 * 0,71 = 0,11 %
For measurement based on 3 meters: 0,15 % * 1 / (Square root of 3) = 0,15 * 0,58 = 0,09 %
Consequently a metering station where the master meter is put in line with the line meter, and where the
measurement result is the average of the flow meters in series, will have a significantly lower uncertainty than a
conventional solution with the master meter in a by-pass to the flow line.
4 An alternative metering system
4.1 Arguments for a metering system with 3 meters in series
One main reason for putting a master meter in a bypass is to avoid that the master meter is exposed to the
same deteriorating conditions as the line meter. It is here recognised that an ultrasonic flow meter does not
contain any moving parts; hence this failure mode is completely removed by the flow meter design.
It is also considered that corrosion can be avoided by using non-corrosive materials and that wax deposits can
be avoided by heating the fluid or by applying heating to the metering section at no flow conditions. It is also
worth noting that the basic design, with a master meter in a bypass, will not be more robust to these two failure
modes than other designs. If you have a problem with the internal pipe surface in the metering run, you will be
exposed to the same failure mode in the master meter run. So if you get a problem with the flow metering run,
you are likely to get the same problem with the master meter run. Or you may be unable to tell if you have the
same problem with the master meter run, or if it is the master meter run that have the problems and not the
flow metering run.
A remaining failure mode is failure or deterioration of the meter hardware. It is assumed that such failure is best
revealed by long term comparison of the line meter with the master meter. By having the master meter in a by-
pass the possibility for close long term monitoring is greatly reduced. By putting the master meter in series with
the line meter a large historic database over the difference between the two flow meters can be established
and maintained. Hence the best possible database for monitoring quality is established.
Furthermore, it is recognized that a design with master meter in a bypass will have two failure modes related to
valve leakage. If one of the two seals in each valve is passing, or if the valve leakage detection system has
failed, the metering system will have to be retracted, replaced and overhauled. In a design with the master
meter in series with the line meter, this failure mode is completely removed.
Hence, a design with the master meter in line with the line meter is recommend over a design with the master
meter in a by-pass. Mainly because a design with the master meter in series with the line meter will:
 Have lower uncertainty
 Avoid the failure modes related to valve leakage

 Have better quality monitoring capabilities
In addition, cost, size, complexity and weigh will be much lower than a system with the master meter in a by-
pass to the main line.
In addition to the arguments for putting the master meter in-line with the line meter, there are three arguments
for adding a third flow meter into the single flow metering run.
The first reason is that the measurement uncertainty will be somewhat reduced (20 %) as the measurement
uncertainty will be equal to the measurement uncertainty for each meter divided by the square root of the
number of meters. 1 / (Square root of 2) = 0,71 and 1 / (Square root of 3) = 0,58.
The second reason to add a third meter into the design is that it will make it more ease to find out which meter
has failed or deteriorated, if so happens.
The third reason is that even if one of the three meters fail the metering station will be fully operational.
Replacement of the defective meter can be postponed and maybe entirely avoided during the lifetime of the
metering station.
4.2 Metering system with 3 meters in series - Design
The basic system is in compliance with the significant requirements the measurement regulation. Mainly as it
has parallel metering runs and permanent equipment for calibration of the flow meter. However, the Failure
Mode and Effect Analysis and the uncertainty evaluations have revealed that this is not be the best design.
Based on the Failure Mode and Effect Analysis and the uncertainty consideration above the following design
have been identified as a better design of the metering system.
QS = Sampling System, FT = Ultrasonic Flow Meter, PT = Pressure Transmitter, TT = Temperature

Transmitter.
In this figure, the fluid is flowing in from the left side, passing the sampling system and then passing through 1
metering run containing a duty meter, a master meter and a spare meter installed in series.
This system has the following main features when compared with basic design:
1. The measurement uncertainty is reduced
2. The failure mode related to leakage through high integrity valves is completely removed
3. Size, weight, complexity and costs are much lower

4. The facilities for monitoring of the flow meters are improved
5. The fault finding capabilities for the flow meters are improved
4.3 Metering system with 3 meters in series - General arrangement
Based on the considerations about station design, and design of the sampling system, the following figure
indicates one possible arrangement that will meet significant design considerations for the fiscal metering
system. The system will be retrievable.
4.4 Metering system with 3 meters in series - Uncertainty analysis
From Guide to the expression of Uncertainty in Measurement it is known that the measurement uncertainty will
be reduced if the measurand is the average of 3 measurements. For uncorrelated uncertainty contributions the
resulting measurement uncertainty will be equal to the measurement uncertainty for each meter divided by the
square root of the number of meters.
If we accept the uncertainty analysis in a previous section, the uncertainty of one meter will be approximately
0,15 % @ 95% confidence level. The resulting uncertainty for a measurement based on the average of 3
meters in series will then be approximately: 0,15 % * 1 / (Square root of 3) = 0,15 * 0,58 = 0,09 % - if the
uncertainty contributions are uncorrelated.
The following uncertainty budget chart illustrates the uncertainty that may be expected for a metering system
consisting of 1 metering run with 1 line meter, 1 master meter, 1 spare meter which are all used for
measurement and monitoring. The tool cannot be configured for this particular uncertainty analysis. However,
by combining the uncertainty contributions for the 3 flow meters into the master meter we can simulate this
configuration. In the uncertainty budget below the uncertainty for the three meters are combined into the
uncertainty component for the Master meter and sat equal to 0,09 %.

The total uncertainty in Standard Volume, for a metering system with 3 meters in series is now estimated to be
approximately 0,13 % @ 95% confidence level.
If we take into account the additional uncertainty of 0,05 % @ 95% confidence level from determination of
water in oil the estimated uncertainty will be 0,14 % @ 95% confidence level. It seems clear that this design will
meet the uncertainty requirement of 0,3 % of standard volume as required in the measurement regulations.
If the uncertainty contributions from the three ultrasonic meters are fully uncorrelated, the uncertainty of the
average from the three meters will be reduced by 1/ (square root of 3) to 0,09 % @ 95% confidence level. The
resulting uncertainty will then be 0,14 % @ 95% confidence level.
If the uncertainty contributions from the three ultrasonic meters are fully correlated, the uncertainty of the
average from the three meters will remain equal to the uncertainty of one ultrasonic meter, estimated to 0,15%
@ 95% confidence level. The resulting uncertainty will then be 0,19 % @ 95% confidence level.
5 Selection of the preferred system
5.1 Methodology for selection of preferred system
A principle for performance of cost benefit analysis is described in NORSOK I-106 Fiscal metering systems for
hydrocarbon liquid and gas (Edition 1, November 2014) ANNEX C System selection criteria (informative).
Measurement uncertainty represents a risk for loss of income. A risk for loss of income represents an equal risk
for loss of profit. The risk for loss by measurement uncertainty (NOK) can be calculated as 0,2 * Measurement

uncertainty @ 2 standard deviations. In this calculation the measurement uncertainty (m3) has to be converted
to a monetary unit (NOK) by taking into account the product value (NOK/m3).
Risk is regarded to be Probability times consequence. A cost is a consequence having probability 1.

Consequently all direct and indirect costs related to the measurement system will also represent risks for loss
of profit. Probability and costs can also be estimated for all the failure modes.
The methodology for comparing alternative systems will then be to add all risks for each metering system
design and compare the total measurement risk for these systems. The purpose should be to arrive at the
system design that minimizes the total measurement risk. The total Risk for loss of profit by the metering
system.
A total measurement risk evaluation has to take into account:

 Risk for loss of income, and profit, by measurement uncertainty
 Risk for maintenance costs as revealed by the Failure Modes and Effects Analysis
 Costs for spare components and systems
 Direct and indirect costs over the whole lifetime of the metering system
5.2 Assumptions put into the evaluation
A few key assumptions are included here to facilitate an overall understanding of the summary:
1. Net present value of the total oil export 2,00E+11 NOK
2. Indirect project costs = 2 * purchase order costs for the metering system
3. The retractable / replaceable unit is shown within the stippled box
4. A spare unit is available for the systems within the stippled boxes
5. A fixed development cost is included for the sampling system and subsea ultrasonic meter
6. Failure mode related to corrosion is avoided by ensuring that the internal surface is non – corrosive
7. Failure mode related to wax deposits is avoided by design or by the wax strategy
8. Failure frequency for valve tightness is set to 1 / (valve * lifetime (25 years))
9. Failure frequency for sampling system is set to 1 / (system * lifetime (25 years))
10. Flow meters can be calibrated at full flow rate at an accredited laboratory
11. Due du complexity and extent of the basic system with master meter in a bypass,
the total costs for replacing this system is sat to 2,5*10E7 NOK.
12. As the preferred system with 3 meters in series is a much simpler system,
the total costs for replacing this system is sat to 1*10E7 NOK.
For the basic design, with a master meter in a bypass, the failure frequency for each flow meter is set to 0,5 /
(meter * lifetime (25 years)). As there are 3 flow meters – and each failure will cause a replacement to take
place the total failure frequency will be 3*0,5=1,5.
For the preferred design, with 3 meters in series, failure frequency for each flow meter is set to 0,5 / (meter *
lifetime (25 years)). However, with 3 meters in place there is redundancy in place for the 2 meters that are
required for the station to be fully functional. The resulting failure frequency for the station is therefore sat equal
to the combined failure frequency for 2 meters: 0,5 * 0,5 = 0,25.

5.3 Evaluation of alternative metering design
In the original evaluation several systems were considered. Only the results for the highlighted systems are
presented here:
2 metering runs, 1 line meter in each + 1 master meter in a bypass
1 metering run , 1 line meter in line with 1 spare meter + 1 master meter in a bypass
2 metering runs, 1 line meter in line with 1 master meter in each run
1 metering run – 1 line meter in line with 1 master meter
1 metering run – 1 line meter in line with 1 master meter and 1 spare meter
Ship survey
The figure below shows the result of the analysis.

5.4 Preferred design for a subsea fiscal oil export metering system
It is very clear from the analysis that the complexity of a conventional system, with parallel metering runs and a
master meter, has huge direct and indirect cost consequences. It is also clear that a conventional system does
not have the benefits of redundancy, lowest possible measurement uncertainty and best possible monitoring
capabilities as a system with three meters in series will have.
Hence, a conventional system can be improved by simplifications. The measurement uncertainty will be
reduced by putting the master meter in line with the line meter and by letting the measurement result be the
average of the master meter and the line meter. By putting the master meter in line with the line meter, high
integrity metering valves are not needed. As high integrity metering valves are removed from the design the
failure mode related to valve tightness also vanish. This will significantly reduce the direct and indirect costs.
It is also worth noting that a master meter run will be exposed to the same failure modes as the flow metering
run. So if you get a problem with the flow metering run, you are likely to get the same problem with the master
meter run. Or - you may be unable to tell if you have the same problem with the master meter run, or if it is the
master meter run that have the problems and not the flow metering run.
By putting the master meter in line with the line meter some redundancy is built into one metering run. However
there will then be only one master meter and one line meter. There will not be any redundancy if one of the two
meters fails. Such redundancy can be achieved by adding a third flow meter into 1 single metering run.
This third flow meter will contribute to reduction of the measurement uncertainty because the measurement will
be the average of three meters. The third flow meter will also improve the fault finding capabilities. The fault
finding capabilities are improved as it will be easy to see if one of three meters has drifted relative the two other
meters. This third flow meter can also serve as redundancy both for the line meter and for the master meter. If
one meter fails, the metering system will still be fully operational. Replacement can be planned to a suitable
time. At the end of the lifetime it may even be acceptable to operate the metering system with one meter out of
operation.
The solution consisting of 1 metering run consisting of 1 line meter, 1 master meter and 1 spare meter is
therefore preferred because it has:
1. Installed spare functionality

2. The best quality monitoring capabilities
3. The best fault identification capabilities
4. The lowest measurement uncertainty of all the evaluated systems
5. Much lower costs than conventional systems
6. Much lower total risk for loss of profit than conventional systems
The design rests on the assumption that it will be possible to calibrate the system at full flow rates, If not,
parallel metering runs will most likely be required. The cost will be significantly impacted by parallel runs.
The preferred system will deviate from the measurement regulation by not having a parallel metering run and
by not having a conventional solution with a master meter in a by-pass to the flow line. These deviations will
have to be formalized with the authorities. The design also has to be accepted by the partners.
Under the Norwegian tax system the company will have to pay 78 % of the profit in tax to the Norwegian state.
Consequently, in relative terms, the company and the state will be equally exposed to the total measurement

risk for the alternative metering systems. The evaluation of total measurement risk can then form the basis for
an application for deviation from statutory requirements. What is good for company profit will also be good for
the tax income to the state.
6 Summary
Based on the work presented in this paper it can be concluded that it will be feasible to design a subsea fiscal
oil export metering station within the uncertainty requirements in the measurement regulations.
A basic design consisting of 2 metering runs, with 1 ultrasonic flow meter in each run, and an ultrasonic master
meter in a bypass will meet the most significant requirements in “The measurement regulations”.
It is also regarded feasible to route a sample stream via the umbilical to a nearby topside facility for sampling
with a conventional sampling system.
However it has been demonstrated that a metering system consisting of 1 metering run with 1 line meter, 1
master meter and 1 spare meter in series is preferable because it has:
1. Installed spare functionality

2. Better quality monitoring capabilities
3. Better fault identification capabilities
4. Lower measurement uncertainty
5. Much lower costs than the basic system
6. Much lower total risk for loss of profit than the basic system
The uncertainty requirement in The Norwegian measurement regulations is 0,3 % of standard volume.
The uncertainty for the preferred system is estimated to be less than 0,2 volume % @ 95 % confidence level.

33st International North Sea Flow Measurement Workshop
20. – 23. October 2015
A new methodology for cost-benefit-risk analysis of oil

metering station lay-outs
Astrid Marie Skålvik1,
Ranveig Nygaard Bjørk1,
Kjell-Eivind Frøysa12 and
Camilla Sætre1
1
Christian Michelsen Research AS, Bergen, Norway
2
Høyskolen i Bergen, Bergen, Norway
ABSTRACT
Custody transfer oil metering stations are traditionally equipped with spare meter runs
and proving device with on-site calibration possibilities for the proving device. Such a
layout is expensive (CAPEX and OPEX). The gain is that metering uncertainty is low
to secure national and company income.
Currently, there is major focus on cost-reduction in the oil industry. This has initiated
increased focus on metering station costs, and increased need for cost-benefit analysis
for proposed metering station layout. Such analysis traditionally address balance
between investment, operational costs and uncertainty.
Simplified metering stations may have larger measurement uncertainty than more
complex stations. In addition, if a flow meter or other essential components fail, the
metering station uncertainty may increase significantly in the period before repair or
replacement. For metering stations with simpler layout it may also be more time
consuming to take repairing actions, due to lack of access. All this increases the risk of
loss of income from the exported oil.
The methodology proposed in this paper combines situations when flow meters are
malfunctioning and when they are working. Response times for repair are included.
Probabilities of the different states (functioning, malfunction, etc.) are derived using
steady state Markov models. Total risk due to normal and increased uncertainty over a
metering station life time can then be calculated.
Several metering station layouts, from complex to simple, are analysed using the new
proposed method, where the risk of loss due to normal and increased uncertainty is
combined with CAPEX and OPEX to identify optimal metering station layout with
respect to risk of loss of income for a given field.
The new method enables the derivation of the overall risk associated with the
malfunction of one or several flow meters in a metering station. Enhanced cost-benefit
analysis with this additional risk are presented, for a series of metering station layouts.
1
20. – 23. October 2015
1 INTRODUCTION
The aim of this paper is present a new methodology to show how the risk associated
with different solutions for the fiscal measurement of oil flow can be calculated. Risk
is here the risk of loss of income from the deviation between the actual amount and the
measured amount of oil or gas, or the risk of loss during production shut-down.
Furthermore, a cost-benefit analysis combining the risk and cost (CAPEX + OPEX) of
each metering configuration is presented. Quantifying risk and comparing it to fixed
costs can be an informative contribution in the process of choosing between different
metering station solutions.
In order to establish a quantitative estimate of the probability, consequence and thus

risk associated with the different states or conditions that the metering station may be
in, the following input parameters must be provided to the model:
 Flow rate of metering station

 Oil price
 Measurement uncertainties associated with the different metering
configurations
 Mean time to repair (MTTR) a meter
 The expected time between planned production stops
 The mean time to failure (MTTF) of a meter
 Operating hours per year of a meter
 Number of years the analysis should cover
 Expected life cycle costs for the different metering configurations
These parameters do not have any definitive, “correct” value, and may vary from
project to project. In this paper, we have studied the risk associated with different
metering configurations based on an example set of these input parameters.
The first part of this paper explains the method used for the risk analysis. Then we
calculate the probabilities of the metering stations being in different states depending
on whether the individual meters function or fail. We continue by estimating the
consequences of being in these different states. Then we calculate the risks associated
with each of the states using the formula: 𝑅𝑖𝑠𝑘 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 ∙ 𝐶𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒. In the
end, we compare the overall risk with the CAPEX and OPEX costs associated with the
different metering configurations.
2 DESCRIPTION OF DIFFERENT METERING STATION

CONFIGURATIONS
The risk and cost-benefit study is carried out for six different fiscal oil metering station
configurations, ranging from a conventional solution to more simplified solutions.
The following nomenclature is used:

 DM: Duty Meter
 PD: Prover Device
 CP: Compact Prover
 MM: Master Meter
2
20. – 23. October 2015
2.1 Configuration 1: Conventional system with prover device
Figure 1 shows a traditional solution with inline calibration using a large volume prover
(typically used every 4th day) and a backup run for the DM. If both DM fail, the
production is shut down. If one DM fails, there is enough capacity in the other DM, and
production can continue without increased uncertainty in the flow metering. The failed
DM can then be replaced or repaired without waiting for the next planned production
stop, as it is placed in a parallel run.
Figure 1: Configuration 1, conventional solution with a prover device for proving.
2.2 Configuration 2: Conventional system with master meter
Figure 2 shows a modified configuration of the conventional metering station

configuration, with parallel runs (one duty run and one backup run). This configuration
has inline calibration in a bypass loop using a master meter with a yearly connection of
a compact prover. This configuration is expected to have a slightly higher uncertainty
than configuration 1, as the inline calibration is performed with a master meter instead
of a prover device.
Figure 2: Configuration 2, use of a master meter for proving instead of a prover

device.
2.3 Configuration 3: Simplified solution with master meter bypass
Configuration 3 consists of one duty meter in a single run, with inline calibration using
a master meter in a bypass loop (typically used every 4th day) with possibility of a yearly
connection of a compact prover. In normal operation configuration 3 has comparable
uncertainty to configurations 1 and 2 with only one DM operational. The risk is higher
for configuration 3 as there is no bypass loop around the DM. Hence, if the DM fails
3
20. – 23. October 2015
one has to wait for a planned production stop for repair. There will be an increased
uncertainty of the metering station in this period, as the MM will be used as a DM.
For this metering station configuration, the MM can be changed or repaired independent
of production pause.
Figure 3: Configuration 3, with DM in a single run and a bypass loop with a MM.
2.4 Configuration 4: Simplified solution with master meter inline
Configuration 4 consists of one single run with inline calibration using a master meter,
with a compact prover bypass loop for yearly calibration. The risk for this configuration
is higher compared to the previous configurations, since the master meter cannot be
disconnected and maintained during operation.
There is no bypass loop around the DM nor the MM. Hence, if the DM or MM fail, one
has to wait for a planned production stop for repair. There will be an increased
uncertainty in this period.
The differences between configuration 3 and configuration 4 are:

 The MM in configuration 3 is protected from daily wear and tear, whereas the
MM in configuration 4 is subject to daily wear and tear, as well as scale build-
up etc.
 In configuration 4 it is only possible to do a comparison between the MM and
the DM and compare the deviation between the measured values. This is not
considered to be a normal calibration.
Figure 4: Configuration 4, with DM and MM in the same single run.
4
20. – 23. October 2015
2.5 Configuration 5: Simplified solution with MM inline and no CP
Configuration 5 is a simplified configuration with one single run with inline calibration
using a MM, and no compact prover. The risk for this configuration is comparable to
configuration 4, since the master meter cannot be disconnected and maintained during
operation.
The difference from configuration 4 is that configuration 5 has no onsite calibration

possibility for the master meter, and that offsite calibration may only be performed
during planned production stops. This increases the uncertainties of the meters
compared to configuration 4.
Figure 5: Configuration 5, only a DM and a MM, no bypass loop for a compact

prover.
2.6 Configuration 6: Simplest configuration with one duty meter and no

on-site proving possibility
Configuration 6 is the simplest configuration possible, with a single run with only one
duty meter, no master meter and no compact prover. Calibration of the duty meter has
to be performed offline. This metering station thus has a higher uncertainty compared
to the previous configurations, and the probability of a production shut down is higher
as there is only one meter. As expected, this configuration represents the highest risk.
Figure 6: Configuration 6, a single run with only one duty meter, no master meter
and no compact prover.
3 METHOD FOR RISK ANALYSIS
Based upon the six different configurations outlined in section 2, we develop a steady
state Markov model for the problem. A metering station may be in different states i
depending on the functioning of the different individual meters. For each of the
metering station configurations studied in this paper, we have calculated the probability
𝑃𝑖 of being in each of these states 𝑖. Then we have established the consequence 𝑄𝑖 of
each state. The consequence 𝑄𝑖 has been limited to either the potential loss of revenue
by mis-allocation due to measurement uncertainties, or the potential loss associated
with production shut-down.
5
20. – 23. October 2015
Then the overall risk 𝑅𝑖 for each state i is calculated using: 𝑅𝑖𝑠𝑘 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 ∙
𝐶𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒:
𝑅𝑖 = 𝑃𝑖 ∙ 𝑄𝑖
The risk associated with a system that has N possible states, is then expressed as the
statistical expected loss, using the following equation [1, p. 7]
𝑁 𝑁
𝑅𝑡𝑜𝑡𝑎𝑙 = ∑ 𝑅𝑖 = ∑ 𝑃𝑖 ∙ 𝑄𝑖
𝑖=1 𝑖=1
In this analysis, we assume that depending on their complexity, the metering stations
could be in the following different states:
 Normal operation
 Failure of duty meter (DM)
 Failure of master meter (MM) or prover device (PD)
 Failure of all meters
4 PROBABILITY OF DIFFERENT STATES
In order to set up the risk-budget, it is necessary to estimate the probabilities for all the
different combinations that are possible for each metering station.
4.1 Input parameters to the probability calculations
Input parameters for calculating the probability distribution of the different scenarios
are described in the following. Note that the example values that are assigned to each
of the parameters here are only examples. When the method is used to evaluate a
specific project, these parameters must be thoroughly determined. The goal of this
paper is not to estimate these parameters in detail, but to demonstrate the use of a new
framework for risk-calculation.
T - Operating duration
T is the operating duration in hours, for which the probability is calculated. The
operating duration depends on how the meter is placed in the metering station.
 Main run: For a meter in the main run, it is assumed that the meter is in operation
100 % of the time, and 𝑇𝑚𝑎𝑖𝑛 𝑟𝑢𝑛 is typically set to 8760 hours times the number
of year for the analysis.
 Bypass loop: If the meter is situated in a bypass loop, the operational duration
is much lower than for a meter situated in the main run. For a meter that is
situated in a bypass loop, 𝑇𝑏𝑦𝑝𝑎𝑠𝑠 depends on how frequently the proving is
performed, and for how long periods of time. If, for example, the proving is
performed every 4th day, with a duration of 1 hour each time, then 𝑇𝑏𝑦𝑝𝑎𝑠𝑠 is
only a small fraction of 𝑇𝑚𝑎𝑖𝑛 𝑟𝑢𝑛 .
In reality, the meter components may degrade with time, even if the meter is not
in operation. Thus, it is necessary to include this passive degradation in the
6
20. – 23. October 2015
calculation of probability of failure. This may be done by assuming that the

passive degradation is proportional to the actual time that passes, and that for
example every four days that passes when the meter is passive, result in
comparable degradation as 1 day in operation. This can be simplified and
implemented by dividing the operation time of the metering station by 4, so that
𝑇
𝑇𝑏𝑦𝑝𝑎𝑠𝑠 = 𝑚𝑎𝑖𝑛 𝑟𝑢𝑛
. Note that this is just an assumption that is used in the
4
example calculations shown.
𝜆 - Failure rate
The failure rate 𝜆 is inversely proportional to the mean time to failure (MTTF ) for the
meter in operation, expressed in hours. It is important to notice that we here assume
that the meter has not exceeded the expected service lifetime, and that it is used during
the normal operating period where the failure rate can be assumed to be constant. It is
possible that the MTTF may increase directly after installation due to possible
adjustment problems1, or at the end of the expected service lifetime due to wear and
tear. For simplicity, the MTTF used in this analysis is assumed to be constant and cover
failures in other critical components (temperature transmitters, pressure transmitters,
and densitometer, as well as valves etc.) associated with the meter in question. The
MTTF is in the examples shown here assumed to be in the order of 4 years of operation,
but this is just an example value and must be determined for each specific meter.
𝜇 – Repair rate
𝜇 is a parameter that is inversely proportional to the mean time to repair (MTTR), the
expected time to repair the meter, in hours. MTTR depends on how the meter is placed
in the metering station.
 Main run – wait before repairing : If the meter is placed in the main run, with
no possibility of bypass, it is assumed that one has to wait until the next planned
production stop in order to repair the meter. If it is assumed as an example that
there will be a planned production stop every 5 years, the mean time to repair
could be estimated to be half of this period, and 𝜇𝑊 is set to
1
in the example calculations in this paper.
2.5 𝑦𝑒𝑎𝑟𝑠∙365 𝑑𝑎𝑦𝑠∙24 ℎ𝑜𝑢𝑟𝑠
 Bypass loop – direct repair possible: If it is possible to disconnect the meter
and repair it, without stopping the production, it can be assumed as an example
here that the meter will be repaired or replaced during 48 h. 𝜇𝐷 is therefore set
1
to 48 ℎ𝑜𝑢𝑟𝑠 in the example calculations in this paper.
4.2 Probability of a dependent system in different states - Steady state

probabilities from a Markov model
If a system only consists of components where the probability for one component being
in a failure or function state is independent of the state of the other components, it is
straightforward to find the probabilities of the overall state of the system by multiplying
together the probabilities of the individual components being in the different states.
1
As fiscal oil flow meters are expected to be sufficiently calibrated onshore before
installation this may not be applicable in this case.
7
20. – 23. October 2015
If, on the other hand, the state of an individual component is dependent on the state of
one or several of the other components, as is the case for the metering station
configurations studied in this report, then the correlation between the probabilities must
be taken into account. An example is the dependence of the DM and the MM in
configurations 4 and 5. If only one of the meters is in a failure state, it is assumed that
the meter will not be changed/repaired before the next planned production shut down.
However, if the meters are in a failure state at the same time, it is assumed that the
production will be shut down directly and that the meters will be repaired/replaced
within hours.
[1, p. Appendix D.4] outlines a method for taking this interdependence between the
individual meters of the metering station into account. This method is based on Markov
models, assuming that in the long run, the rate of arrivals into a specific state will equal
the rate of departures from that state. Figure 7 shows a diagram with the different states
a system consisting of one DM and one MM can be in, and the transition rates between
the different states. Here λ is the failure rate and 𝜇𝑊 is the repair rate waiting for a
planned production stop before repairing, and 𝜇𝐷 is the repair rate when both meters
are in a failed state at the same time and will be repaired directly.
Figure 7: Markov model for a DM and MM in series (configuration 4 and 5).
Table 4-1: State / probability budget for a DM and MM in series (configuration 4 and
5).
Departures Arrivals
State 0 𝑃0 ∙ 𝜇𝐷 (𝑃1 + 𝑃2 ) ∙ 𝜆
State 1 𝑃1 ( 𝜆 + 𝜇𝑊 ) 𝑃3 ∙ 𝜆
State 2 𝑃2 ( 𝜆 + 𝜇𝑊 ) 𝑃3 ∙ 𝜆
State 3 2𝑃3 ∙ 𝜆 (𝑃1 + 𝑃2 ) ∙ 𝜇𝑊 + 𝑃0 ∙ 𝜇𝐷
Table 4-1 shows a budget of probability times failure and repair rates for the expected
departures from and arrivals to each state. This forms a set of linear equations, and
together with the fact that the sum of all four probabilities must equal 1, it is possible
to solve the set of equations and find the expected long term probabilities.
This method does not take into account any exponentially distributed probabilities of
failure, as it is only the long term, steady state probability. In the special case where a
meter is assumed to be repaired or replaced during every planned production shut down,
8
20. – 23. October 2015
and it can be assumed that the probability of failure increases exponentially with time,
this may result in a slight overestimation of the probability of shut down in the first few
years after such a replacement or repair2.
The probabilities for the different states for the other configurations are calculated using
the same method.
5 CONSEQUENCES ASSOCIATED WITH DIFFERENT

SITUATIONS
5.1 Consequence associated with metering uncertainty
The potential loss associated with metering station uncertainty can be expressed in
terms of the expanded uncertainty using the following equation based on Stockton [2]:
𝑈 ∗ ∙ 𝑁𝑃𝑉
𝑄 = 𝑝𝑜𝑡𝑒𝑛𝑡𝑖𝑎𝑙 𝑙𝑜𝑠𝑠 = ≈ 0.2 ∙ 𝑈 ∗ ∙ 𝑁𝑃𝑉 (5.1)
√8𝜋
Here 𝑈 ∗ is the relative expanded uncertainty (2 standard deviations, 95 % confidence

level) associated with the measurement of the flow, and NPV is the net present value of
1
the oil. The factor comes from the integration from -∞ to 0 of the assumed normal
√8𝜋
or Gaussian uncertainty distribution.
The relative uncertainty 𝑈 ∗ used in Equation (5.1) depends on the configuration of the
metering station.
Table 5-1 gives an overview of uncertainties for the meters in the different metering
station configurations, which are used in the analysis in this report. Note however, that
these are estimated uncertainties based on experience with similar systems. A more
thorough, project-specific analysis would be needed in order to establish these
uncertainties for a specific case. For cases with high water fraction and/or non-
homogenous flow, as well as for metering stations with gas break out due to pressure
loss, the uncertainties used here may be underestimated.
2
It is possible to include this time-dependent nature of the probabilities of failure by stating
that the arrivals to and departures from a state i equals the derivative of this state’s probability,
𝑃𝑖 ′. This would result in a system with linearly interdependent differential equations that may
be solved using the Laplace Transformation of the system or by finding the eigenvalues of the
system matrix. This is a more exhausting operation, and is subject to further work.
9
20. – 23. October 2015
Table 5-1: Overview of uncertainties for the meters in the different metering station
configurations denoted V1 to V6 (relative expanded uncertainty 𝑼∗ at 95 %
confidence level). These are estimated example uncertainties based on experience
with similar system, and must be updated specifically for each case.
Metering 𝑼∗ (95
Meter Comments
station % c.l.)
V1 DM 0.25 Duty meter proved with prover device in bypass
V2 DM 0.30 Duty meter proved with master meter in bypass
DM 0.30 Duty meter proved with master meter in bypass
V3
MM 0.50 Master meter with bypass calibration possibility
DM 0.40 Duty meter proved with master meter inline
V4
MM 1.00 Master meter inline
Duty meter prover with inline master meter,
DM 0.60
V5 without CP
MM 1.50 Master meter inline, without CP
V6 DM 1.00 Duty meter without proving or calibration
All
DM 1.00 Duty meter without proving or calibration
stations
5.2 Consequence associated with failure of all meters
There is a probability that several meters will fail in the metering station. If it is no
longer possible to measure the flow, the failure of the meters may result in a shutdown
of the metering station, during the time period of which the meters are repaired. The
cost or consequence of a shutdown is in this study for simplicity taken as the lost cash
flow due to lost production during the time to repair. Another way of calculating the
cost associated with a shut-down is to calculate the present value loss, which is the cost
of delaying the production.
𝑄𝑎𝑙𝑙 𝑚𝑒𝑡𝑒𝑟𝑠 𝑓𝑎𝑖𝑙 = 𝑁𝑃𝑉 (5.2)
However, in some special cases, it may be possible that the production should not be
shut down, but that the flow should be estimated from other metering points or from
performance curves, history etc. In such cases, it is possible to define the uncertainty
∗
𝑈𝑁𝑀 that may for example be in the range of 50 % and above. In this special case, the
consequence of a “no measurement” situation is the following:
∗
𝑄𝑎𝑙𝑙 𝑚𝑒𝑡𝑒𝑟𝑠 𝑓𝑎𝑖𝑙 = 0.2 ∙ 𝑈𝑁𝑀 ∙ 𝑁𝑃𝑉
(5.3)
10
20. – 23. October 2015
6 RISK ASSOCIATED WITH DIFFERENT SITUATIONS
As explained in chapter 2, the risk is calculated as the product of the probability of a

state and the consequence or potential loss associated with this state. For configuration
5, which was taken as an example in chapter 4, the risks are the following:
𝑅0 = 𝑃0 ∙ 𝑄0 = 𝑃0 ∙ 𝑁𝑃𝑉
∗
𝑅1 = 𝑅2 = 𝑃1 ∙ 𝑄1 = 𝑃2 ∙ 𝑄2 = 𝑃1 ∙ 0.2 ∙ 𝑈DM inline MM,no CP ∙ 𝑁𝑃𝑉
(6.1)
∗
𝑅3 = 𝑃3 ∙ 𝑄3 = 𝑃3 ∙ 0.2 ∙ 𝑈MM inline,no CP ∙ 𝑁𝑃𝑉
All parameters are defined in section 4.2 and 5.
The total risk associated with configuration 5, related to the functioning of the meters
and their uncertainty, is thus:
𝑅𝑇𝑜𝑡𝑎𝑙,𝑐𝑜𝑛𝑓𝑖𝑔.5 = 𝑅0 + 𝑅1 + 𝑅2 + 𝑅3 (6.2)
6.1 Discussion of results of risk analysis
Figure 8 shows a comparison between the risks associated with the different metering
stations, for an example set of input parameters.
As expected, the risk increases from metering station configuration 1 to configuration

6.
11
20. – 23. October 2015
Figure 8: Example comparison between risks associated with different metering

configurations. The green parts represent risk associated with normal operation, the
yellow/brown parts the risk associated with failure of one DM, the blue parts the risk
associated with MM or PD failure, and the red parts the risk associated with shut down
of production.
The following conclusions can be drawn from the results shown in Figure 8:
 The risk associated with normal operation increases from configuration 1 to

configuration 6:
o The DMs in configuration 1 are proved regularly with a prover device,
which again is calibrated annually against a compact prover.
o The DMs in configuration 2 are proved regularly with a master meter,
which again is calibrated annually with a compact prover. The use of a
master meter instead of a prover device results in a small increase in the
normal operation uncertainty and consequently in the associated risk.
o The DM in configuration 3 is proved and calibrated in the same manner
as the DMs in configuration 2, but as there is only one DM in
configuration 3, if this fails the metering station is no longer in normal
12
20. – 23. October 2015
operation. Therefore the probability that the metering station is in a

normal operating state, and thus the risk associated with this state, is
lower for configuration 3 than for configuration 2.
o The DM in configuration 4 is proved against a MM that is inline, and
more subject to wear and tear than for a MM in a bypass loop. This
results in a higher uncertainty during normal operation, but the
probability of being in the normal state is lower for configuration 4 than
configuration 3. The combination of probability and consequence results
in a slightly higher risk for configuration 4 compared to configuration 3.
o The DM in configuration 5 is only proved against a MM that is not
regularly calibrated. This results in a higher uncertainty during normal
operation, and thus a higher risk during normal operation compared to
configuration 4.
o The DM in configuration 6 is not regularly proved. The higher
uncertainty associated with this metering configuration results in a
higher risk during normal operation.
 The risk associated with a failure of one DM is negligible in configuration 1

and configuration 2, as there are two DMs here, and if one of them fails, the
other should be able to measure the flow without increased uncertainty. For
configurations 3 to 5, there is a potential loss associated with a DM failure, as
the flow then has to be measured by the MM, with increased uncertainty. For
configuration 3, with a MM in a bypass loop the increase in uncertainty is less
than for configuration 4 where the MM is placed in line, and the uncertainty is
highest for configuration 5 where the MM is inline and has no possibility of
regular calibration.
 The risk associated with a failure of the MM or PD: Even if the potential loss
associated with a failure of the MM or PD is identical for configurations 1 to 5,
the probability of this situation is considered to be minor for configurations 1 to
3, and higher for configurations 4 and 5. This is because when the MM or PD
are in a bypass loop, they can be changed /repaired directly, without having to
wait for the next planned production shut down. For configurations 4 and 5, the
MM is in line with the DM in the main run. The period from the failure happens
to the meter is replaced/repaired is thus longer. The probability of the situation
with a failed meter in any chosen period is higher, and the risk associated with
this is thus higher too. For configuration 6, this situation is not applicable.
 The risk associated with a production shutdown is much smaller for

configurations 1 to 5 than for configuration 6. This is due to the fact that
configurations 1 to 5 have to be shut down only if two meters fail at the same
time, the two DMs for configurations 1 and 2, and the DM and the MM for
configurations 3 to 5. For configuration 6, on the other hand, it is sufficient that
the only meter present fails, and the probability for a shutdown is thus higher.
The probability that the two meters are in a failed state simultaneously is lower
for configurations 1 and 2 than for configurations 3-5. This is because the DM
in configurations 3-5 is in-line, and the probability that it is a failed state is here
higher, since we assume in this example that it will not be repaired before the
next planned production stop.
13
20. – 23. October 2015
7 COST-BENEFIT ANALYSIS
A cost-benefit analysis of a metering configuration takes into account all the costs
associated with a metering solution, including installation and operation costs, the
metering accuracy, functionality and reliability. Based on the combined risk and cost
of each metering solution, it is then possible to do a qualified choice of which metering
solution should be chosen for an application from a cost-benefit perspective.
7.1 Background information
According to Chan [3], the net present value for a metering option is usually analysed
with the traditional cost-benefit analysis using the following equation3:
𝐸 ∙ 𝑁𝑃𝑉 ∙ 𝑖𝑚𝑝𝑟𝑜𝑣𝑒𝑚𝑒𝑛𝑡 𝑜𝑓 𝑚𝑒𝑡𝑒𝑟𝑖𝑛𝑔 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦[%] ∙ 𝑅

100% (7.1)
= 𝑚𝑎𝑥. 𝐶𝐴𝑃𝐸𝑋 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑚𝑒𝑡𝑒𝑟𝑖𝑛𝑔 𝑐𝑜𝑛𝑠𝑒𝑝𝑡
Here 𝐸 is the ownership factor (equals 1 if there is only one owner for the field), 𝑁𝑃𝑉 4
is the net present value and 𝑅 is the risk factor (𝑅 = 0.2 is used in this analysis, ref.
paragraph 5.1). The improvement of metering uncertainty using an alternative metering
configuration is in this way compared with the maximum cost increase that should be
associated with this alternative.
NORSOK I-106, Annex C [4], uses a similar5 approach to compare two metering
concepts A and B, using the following equation to decide if concept B may be
acceptable6:
(𝐶𝐴 − 𝐶𝐵 ) > (𝑈𝐵 − 𝑈𝐴 ) ∙ 𝑅 ∙ 𝑁𝑃𝑉 (7.2)
Here 𝐶𝐴 and 𝐶𝐵 denote the total life cycle costs for each concept, and 𝑈𝐴 and 𝑈𝐵 denote
the uncertainties associated with each concept. This equation can be rearranged into the
following in order to compare the total life cycle costs and risks associated with each
metering concept:
(𝐶𝐴 + 𝑈𝐴 ∙ 𝑅 ∙ 𝑁𝑃𝑉) > (𝐶𝐵 + 𝑈𝐵 ∙ 𝑅 ∙ 𝑁𝑃𝑉) (7.3)
3
In [3], another equation is proposed to perform an allocation cost-benefit analysis between
more than two owners. This is not the case in this paper.
4
[3] uses the term “NVP”.
5
The main difference between the two approaches is that the NORSOK I-106 appendix C
considers the total life cycle cost of the meters, whereas the GDF approach considers the
CAPEX of the metering concepts.
6
NORSOK I-106 proposes another equation when there are more than one owner. The risk of
allocation with several owners is not the subject of this paper.
14
20. – 23. October 2015
In addition NORSOK I-106 [4] states that the cost benefit analysis should also take into
account the expected regularity of the possible concepts, and that a metering concept
may be acceptable if:
𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛 𝑝𝑟𝑜𝑓𝑖𝑡 𝑏𝑦 𝑟𝑒𝑑𝑢𝑐𝑒𝑑 𝑟𝑒𝑔𝑢𝑙𝑎𝑟𝑖𝑡𝑦

< 𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛 𝑝𝑟𝑜𝑓𝑖𝑡 𝑏𝑦 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑑 𝑐𝑜𝑠𝑡
7.2 Cost-benefit analysis with quantified regularity and different metering

station uncertainties
The methods presented in the previous section are useful when it comes to comparing
two metering station concepts where the main difference is the risk associated with their
metering uncertainty. However, it was found in chapter 5 and 0 that a metering station
may have different metering uncertainties depending on which state it is in, and the risk
associated with a potential shut down may be quantitatively included in the total risk
associated with a metering concept.
The following formula is then established to calculate the total cost and risk associated
with a metering concept:
𝑇𝑜𝑡𝑎𝑙 𝑐𝑜𝑠𝑡 𝑎𝑛𝑑 𝑟𝑖𝑠𝑘 𝑎𝑠𝑠𝑜𝑐𝑖𝑎𝑡𝑒𝑑 𝑤𝑖𝑡ℎ 𝑎 𝑚𝑒𝑡𝑒𝑟𝑖𝑛𝑔 𝑐𝑜𝑛𝑐𝑒𝑝𝑡 = 𝐶 + 𝑅
𝐶 = 𝑡𝑜𝑡𝑎𝑙 𝑙𝑖𝑓𝑒𝑡𝑖𝑚𝑒 𝑐𝑜𝑠𝑡 𝑎𝑠𝑠𝑜𝑐𝑖𝑎𝑡𝑒𝑑 𝑤𝑖𝑡ℎ 𝑎 𝑚𝑒𝑡𝑒𝑟𝑖𝑛𝑔 𝑐𝑜𝑛𝑐𝑒𝑝𝑡
𝑅 = 𝑅𝑛𝑜𝑟𝑚𝑎𝑙 𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑜𝑛 + 𝑅𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑑 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦 + 𝑅𝑠ℎ𝑢𝑡𝑑𝑜𝑤𝑛
In this simplified approach, the total lifetime cost, 𝑪, includes the fixed costs that are
not subject to risk or probability of failure:
 procurement and installation cost of the meter, including engineering resources
and planning costs, as well as weight and dimension footprint at the platform
(CAPEX)
 operational costs associated with planned maintenance, proving and calibration
(OPEX)
The risk term, 𝑹, includes the risks associated with the following (as found in chapter
5 and 0):
 risked exposure to lost revenue due to mis-allocation during
 normal operation
 periods with increased uncertainty as a result of failure of one or several
meters
 production loss or delay during shut down, in the case where all meters that can
be used for measuring production have failed
The total expected cost and risk calculated this way is then compared between the
different metering solutions.
15
20. – 23. October 2015
7.3 Total lifetime costs associated with each metering configuration
The total lifetime cost, 𝐶𝑗 , includes the fixed costs that are not subject to risk or
probability of failure. This cost may vary depending on the specific project due to
differences in for instance oil properties, maximum expected flow rates, platform
facilities and available resources. Furthermore, this cost may depend on the planning
time of the metering solution, as the economical rates for material and resources may
increase or decrease with time
In order to have a basis for a quantitative cost benefit analysis for this generic study,
the total life cycle costs of configuration 1 (conventional solution) and configuration 5
(simplified solution) have been taken from Flølo [5]. The other configurations have
then been assigned an expected total life cycle cost compared with configuration 1 and
5.
It is important to stress the fact that the main goal of this study is to establish a
framework and a method for comparing different metering station configurations,
and not to perform detailed calculations of the total lifetime cost associated with
each configuration. The lifetime cost depends on the expected lifetime of the metering
station, and it is possible that the numbers presented here are under- or overestimating
a 10 year lifetime cost, which is used as an example. The following numbers are
therefore only illustrative and the only reason that the numbers are quantified here is to
illustrate the cost benefit analysis.
Configuration 1, conventional solution: According to the example in [5], a

conventional offshore metering station may weigh approximately 100 tons, have a size
of 15m x 5 m x 5m, a package cost of 5 million euros and a life cycle cost of 20 million
euros, which roughly correspond to 160 million NOK.
Configuration 2: According to an article from Faure Herman [6], the use of a master
meter instead of a conventional proving device may reduce the overall cost of the
metering skid with up to 40 %. The total life cycle cost associated with configuration 2
will therefore be set to 160 million NOK * 60 % = 96 million NOK.
Configuration 3: Since this configuration has only one DM and therefore only three
meters, whereas configuration 2 has four meters, it is roughly estimated that the cost is
proportional to the number of meters, and the cost of configuration 3 is therefore set to
96 million NOK * 75 % = 72 million NOK.
Configuration 4: It may be assumed that the use of the MM in line with the DM instead
of in a parallel run, may save space and weight and therefore result in a 20 % reduction
in total life cycle cost compared with configuration 3. The cost of configuration 4 is
therefore set to 72 million NOK * 80 % = 57,6 million NOK.
Configuration 5, simplified solution: According to the example in [5], a simplified

offshore metering station may weigh approximately 10 tons, have a size of 15m x 1 m
x 1 m, a package cost of 1 million euros and a life cycle cost of 4 million euros, which
roughly corresponds to 32 million NOK.
16
20. – 23. October 2015
Configuration 6: As this configuration has only one meter, compared with

configuration 5 that has two meters, it may be expected that the total lifetime cost may
be roughly divided by 2. The cost of configuration 6 is therefore set to 32 million NOK
/ 2 = 16 million NOK.
8 DISCUSSION OF EXAMPLE RESULTS OF COST/BENEFIT

ANALYSIS
Figure 9 and Figure 10 show graphically the results of cost and risk for each metering
configuration, for two example cases. The examples show that higher risk is often
associated with lower cost, as expected.
The optimal configuration(s) will depend on the input parameters of the specific case.
The lifetime cost for each configuration, the production rate and the oil price are of
particular importance in order to get a good cost benefit estimation. Other parameters
as mean time to fail and to repair, metering station uncertainties and operating duration
are equally important parameters that must be estimated specifically for each case.
The goal of this study is to show how a new framework can be used to include risk
calculations related to increased uncertainty or shut-down in a cost-benefit budget, not
to find a universally optimal metering station configuration.
It is important to note that it is not necessarily the configuration with the smallest sum
of life cycle costs and risk that is the optimal choice, as this will also be dependent on
acceptable risk contribution.
Figure 9: Total life cycle cost and risk associated with each metering station
configuration, in million NOK over a lifetime of 10 years. Example case with a
production of 30 200 barrels/day.
17
20. – 23. October 2015
Figure 10: Total life cycle cost and risk associated with each metering station
configuration, in million NOK over a lifetime of 10 years. Example case with a
production of 50 000 barrels/day.
9 CONCLUSIONS
In this study, a method for quantifying the risk associated with different metering
station configurations has been developed. The focus has been on oil metering stations,
but the principles and methods that are used are transferable to other kinds of metering
stations, i.e. multiphase or gas metering stations. Furthermore, an example cost-benefit
analysis combining the risk and cost of each metering configuration has been carried
out.
The risk modelling was based on that the metering stations can be in the following
states:
 Normal operation
 Failure of duty meter resulting in the need to measure flow using only master
meter or prover device, resulting in increased uncertainty
 Failure of master meter or prover device, resulting in increased uncertainty
 Failure of all meters, resulting in no measurements from the metering station,
hence the uncertainty will be defined by any back up measurements, or
production shut down.
In order to calculate the risk, the probabilities for each state were estimated and
multiplied with the consequences of each state. The consequences were lost revenue
due to mis-allocation resulting from measurement uncertainties, as well as potential
loss associated with production shut down.
The overall risk for all the metering station configurations were compared for an
example case, and as expected, it was found that the conventional configuration had a
low overall risk, and that the risk increased as the metering configuration was
simplified. It was found that this increase in risk both originates from the lower
reliability of the simpler metering configurations, as well as the higher potential loss
due to increased metering uncertainties.
18
20. – 23. October 2015
The aim of this study was to develop a framework and a method for comparing and
displaying different metering station configurations in terms of risk and cost. The
optimal metering station configuration will depend on the various input parameters such
as life cycle costs, flow rate, years between planned production stops, the failure rate
of the meter, oil price and the expected mean time to repair a meter.
ACKNOWLEDGEMENT
The authors would like to thank the Norwegian Petroleum Directorate, and especially
Steinar Vervik, for bringing the idea of this paper forward to CMR and financing the
development of the model.
REFERENCES
[1] Terje Aven, Pålitelighets og risikoanalyse, Universitetsforlaget, 1994.

[2] Philip Stockton, Smith Rea Energy Limited, UK, Cost Benefit Analysis in the
Design of Allocation Systems, North Sea Flow Measurement Workshop, 2009.
[3] Philip Chan, GDF Suez E&P Norge AS, Bruk av usikkerhetsanalyser og
nåverdiberegninger i konseptfasen, NFOGM nytt nr 2, 2014.
[4] NORSOK, “Annex C, System selection criteria (informative),” NORSOK, 2014.
[5] Dag Flølo, Statoil, Cost benefit analysis for measurement at pipeline entry,
NFOGM Hydrocarbon Management Workshop – Field Allocation 2014, 2014.
[6] Faure Herman, Advantages of Master Metering Mathod of Proving Custody
Transfer Flows, June / July 2012.
19
33rd International North Sea Flow Measurement Workshop
20 - 23 October 2015
Technical paper
Operational experiences with the

EuroLoop Liquid Hydrocarbon Flow Facility
Jos G.M. van der Grinten, NMi EuroLoop

Bart van der Stap, Flowways
Dick van Driel, Krohne
Abstract
The operational experiences obtained during the calibration of several types of liquid flow
meter clearly demonstrate EuroLoop’s ability to test the viscosity dependent performance
of these instruments. The temperature as observed during a series of successive runs
recorded at a constant flow rate, demonstrates a stability that is a factor 50 better than
originally specified.
Extension of a previous intercomparison between the EuroLoop and Trapil facilities using
an 8” 7 path ultrasonic transfer flow meter and EuroLoop’s small prover [2] was extended
using the big prover. All 19 results match a normalized deviation . For 16 results
.
Sample analysis performed at two independent laboratories as part of EuroLoop’s
metrological maintenance, reveals a significant viscosity difference for the lowest and
highest viscosity liquids. The density of the samples agrees within their mutual
uncertainties, which is important for mass flow measurements.
1. Introduction
For most types of liquid flow meters for custody transfer purposes like ultrasonic flow
meters, turbine meters and coriolis flow meters, the performance is depending on the
Reynolds number dependent velocity profile or on the liquid viscosity directly. Also the
factor of Dp devices like orifices, nozzles and Venturis, is depending on the Reynolds
number and therefore on the viscosity as well.
The EuroLoop Liquid Hydrocarbon Flow Facility operates on three different liquid
viscosities – 1, 10 and 100 mm²/s (centi-Stokes, cSt) – covering a large range of Reynolds
numbers (approximately 1·103 – 4·106). The rationale behind the construction of EuroLoop
is that calibration of a flow meter using different liquid viscosities gives the user of the
meter insight on the viscosity dependent performance of the meter. For the manufacturer
this knowledge enables him to compensate for the different flow profiles or different
viscosities in the meter electronics. In addition EuroLoop can operate at a very constant
liquid temperature which will keep the viscosity constant as well. This is extremely
important for the calibration of turbine meters, where the calibration needs to be
performed with a constant viscosity over the entire range of flow rates.
1
20 - 23 October 2015
Technical paper
NMi EuroLoop operates the facility and is responsible for the metrological performance.
The design of the facility was described earlier [2]. Accreditation is in the process of being
acquired [1] and the first round robin tests have been performed [2]. All calibration
certificates are issued by NMi EuroLoop. Maintenance is done by Krohne, the designer and
constructor of the facility. The actual specifications of the facility are shown in the table
below.
After years of construction and testing the current paper gives an anthology of operational
experiences obtained during calibrations using the EuroLoop Liquid Hydrocarbon Flow
Facility and its metrological maintenance.
Table 1: Specifications of the EuroLoop Liquid Hydrocarbon Flow Facility

Small circuits Large circuits
Flow 10 – 1 200 m3/h 30 – 5 000 m3/h
Dimensions of meters 4” – 12” (100 – 300 mm) 12” – 30” (300 – 750 mm)
Flanges and pressure classes ANSI 150, PN 10 ANSI 150, PN 10
Kinematic viscosity 1, 10, 100 mm2/s (cSt) 1, 10, 100 mm2/s (cSt)
Temperature stability Better than 0.1 °C Better than 0.1 °C
Maximum back pressure 12 bar(g) 12 bar(g)
Overall uncertainty volume by piston / master 0.02% / 0.06% 0.02% / 0.06%
Overall uncertainty mass by piston / master 0.04% / 0.07% 0.04% / 0.07%
2. Description of facility
Each circuit, schematically displayed in Fig. 1, is split between a small and a large diameter
loop. The temperature increase, resulting from the work exerted on the liquid by the
pumps, is reduced by a cooling system using water. A back pressure of 6 – 9 bar(g) avoids
cavitation in the liquid. If necessary, e.g. during the calibration of Venturi tubes, the back
pressure can increased to 12 bar(g) to avoid cavitation in the throat of the Venturi. In that
case operation has to be performed with great caution as safety valves open at 12.5 bar.
The primary references for calibration consist of a large piston prover and a small piston
prover, which are both operated on all three liquids. The secondary references are master
meters, which is a combination of an upstream monitoring full bore and downstream
reduced bore ultrasonic meter.
The temperature in each circuit is monitored by fast ultrasonic sensors that measure the
average temperature in a cross section of the pipe without disrupting the flow pattern. The
ultrasonic temperature sensor is combined with an upstream pressure transmitter and a
downstream Pt100. If the liquid properties change the readings of the ultrasonic
temperature measurement and the Pt100 will start to deviate. In this way the operator is
triggered to have the liquids analyzed again. In addition the operator has a diagnostic tool
for checking the liquid homogeneity.
2
20 - 23 October 2015
Technical paper
Fig. 1:. Schematic drawing of the EuroLoop Liquid Hydrocarbon Flow Facility. One circuit is shown, the
others are identical. The piston provers are shared by all three circuits.
Each calibration starts with the specification of the calibration set-up. For the small circuits
24 meter straight length is available, for the large circuits 28 meter is available. In each
configuration sufficient length is planned behind flow disturbances, like reducers or
expanders. Before the start of the calibration all air and vapours are eliminated from the
circuits and leak tests are performed. The liquid is homogenized, which is tested by the
ultrasonic temperature measurements in the circuits. The last step in the preparation is an
integrity check of the system. For each calibration there is a choice to use only the master
meters as reference, or to use both master meters and piston prover. The latter option
guaranties a substantially lower uncertainty, but requires more time.
The calibration itself starts with temperature equalization by circulating the liquid at a
flowrate above 0.5 . The calibration flow rates and repeats can be performed according
to API MPMS, OIML R117 or client wishes.
3
20 - 23 October 2015
Technical paper
3. Traceability
The traceability of the EuroLoop Liquid Hydrocarbon Flow Facility is schematically shown
in Fig. 2. The piston provers are directly traceable to length, the master meters are traceable
to the piston provers and the meter under test is traceable to either the master meters or
the piston provers. A more elaborate description on the traceability of EuroLoop was
discussed in an earlier paper [2].
Piston
m
Prover
Master Meter
Meter under Test
Fig. 2: Traceability chain of the EuroLoop Liquid Hydrocarbon Flow Facility.
4. Metrological maintenance
Part of the metrological maintenance is the regular determination of the density and the
kinematic viscosity of all three liquids. The density and kinematic viscosity are obtained
from analyzing samples from the test fluids. For this purpose we use two labs. A test
laboratory is located close to EuroLoop and performs routine tests according to the
appropriate ASTM standards [3][4]. Test results are available with 24 hours. The lab has an
ISO 9001 certificate but does not hold an ISO 17025 accreditation. The calibration
laboratory holds and ISO 17025 accreditation for viscosity, however the accredited density
is outside the range we need at EuroLoop. This lab is further away and delivery times are
between one and three weeks. Both laboratories obtain their traceability from reference
liquids that are externally purchased from an ISO 17025 accredited laboratory.
EuroLoop needs results fast and also needs an ISO 17025 accredited lab. For that reason a
small proficiency test (PT) was organized between both labs to verify if they produce
equivalent results.. The liquids were circulated in the test sections in order to homogenize
them. Then samples were taken and for each liquid two bottles were filled directly after
each other. The samples were labelled and send to the labs. Lab A analyzed the sample at
10, 20 and 30°C. Due to time constraints Lab B could perform the analysis at 20 and 30°C
only. Results were returned by calibration certificates and are graphically displayed in Fig.
3. The upper graphs show the results of the analysis. In order to distinguish the results of
both labs the results of Lab A were plotted at the nominal temperature minus 0.5°C and for
Lab B at the nominal temperature plus 0.5°C. The date of the analysis is shown in the legend
of the graph. On the scale used, all results look identical.
4
20 - 23 October 2015
Technical paper
In order to compare results between laboratories the so-called normalized deviation is

used, which is defined as:
(1)
This is the ratio of the difference between two results and the uncertainty of the difference.
If the difference of two measurement results is smaller than the uncertainty of the
difference, i.e. , results are in agreement. When the uncertainties have a coverage
factor of , the confidence level of this assessment is at least 95.4%.
Viscosity Density
1000 880
1 mm²/s Lab A 1 mm²/s Lab A
31-jul-2015 860 31-jul-2015
1 mm²/s Lab B 840 1 mm²/s Lab B
100 6-aug-2015 6-aug-2015
ν [mm²/s]
820
ρ [kg/m³]
10 mm²/s Lab A 10 mm²/s Lab A

31-jul-2015 800 31-jul-2015
10 mm²/s Lab B
10 mm²/s Lab B 780
10 6-aug-2015
6-aug-2015
760 100 mm²/s Lab A
100 mm²/s Lab A
740 31-jul-2015
31-jul-2015
100 mm²/s Lab B
1 100 mm²/s Lab B 720
6-aug-2015
5 15 25 35 6-aug-2015 5 15 25 35
t [°C] t [°C]
En of Viscosity En of Density
12.00 1.00
0.90
10.00 0.80
8.00 0.70
0.60
En [-]
En [-]
6.00 1 mm²/s 0.50 1 mm²/s

10 mm²/s 0.40 10 mm²/s
4.00 0.30
100 mm²/s 100 mm²/s
2.00 0.20
0.10
0.00 0.00
5 15 25 35 5 15 25 35
t [°C] t [°C]
Fig. 3: Proficiency test results for kinematic viscosity [mm²/s] (left hand side) and density (right hand
side) of EuroLoop’s three hydrocarbon liquids. The upper graphs show the results of the analysis, where
the viscosities are plotted on a logarithmic scale. The lower graphs depict the normalized differences.
In the lower part of Fig. 3 the values are plotted. For the density all values are less
than 1, which means the labs agree. For the viscosities the results for the 10 mm²/s liquid
agree very well. The results for the other liquids differ significantly. The laboratories were
asked for clarification but have not responded so far.
5
20 - 23 October 2015
Technical paper
The agreement of the density results is important for the comparison of a mass flow meter
with a volume reference. For the uncertainty of the density the root-sum-square value of
both labs are taken, i.e. 0.025%.
In the viscosity comparisons the 10 cSt results are extremely well in agreement and for this
reason the differences observed for the other liquids are remarkable. For the 1 cSt liquid
, which means a difference in the measured viscosities of 7%, too much for
comparing facilities that operate on different liquid viscosities. The viscosity values
obtained at the accredited lab will be used for reference. EuroLoop will use a 1%
uncertainty for all liquid viscosities and corresponding Reynolds numbers.
Further steps will be taken which will include a critical evaluation of our own sampling
procedures, a repetition of the comparison to see if differences are consistent over time,
looking for laboratories that have a relevant ISO 17025 accreditation and supplier audits
that will focus on the details of the measurement processes for density and viscosity.
5. Calibration experiences
Now the facility is operational we are acquiring experience with the different types of flow
meters. The examples shown comprise two turbine meters in series, two different coriolis
meters and an ultrasonic meter. All results are anonimized unless the owner has given us
explicit consent to mention details of the meters.
5.1 Turbine meters

The two 16” turbine meters were calibrated in series. Both meters were configured with a
pipe bundle 20D upstream of the meter. The schematic setup is displayed in Fig. 4. The
meters are red, the inlet pipe with the flow conditioners is yellow and the flow conditioners
are white. Below the spools the length of each section is indicated. Above the spool the pipe
number are written.
A B
Fig. 4: Schematic configuration of the calibration setup of two 16” turbine meters.
6
20 - 23 October 2015
Technical paper
The meters were calibrated with the 10 cSt liquid, however due to elevated temperature of
25.9 °C the effective kinematic viscosity was 7.88 mm²/s. The as-found results of the
calibrations are shown in Fig. 5. The deviation of both meters is in the range of -0.15% and
0.00%. Both meters show a relatively flat curve in the lower operating range. Between 60%
and 100% of one meter raises while the other curve drops. The meters have been sent
back to the manufacturer for furter examination.
One of the observations made during the calibrations is that the temperature variation
during a series of succesive repeats at one specific flow rate is very low. The double
standard deviation equals =0.002°C, which corresponds to a viscosity stability of
better than 4·10 mm²/s. Both values are less than the calibration uncertainty of the
-4
quantities. The temperature stability appears to be a factor 50 better than the design
specification in Table 1, which is beyond expectation. Despite this impressive result the
uncertainty analysis will be based on a temperature repeatability (2s) of 0.05°C.
0.1 0.1
MuT A, 10 mm²/s MuT A, 10 mm²/s

MuT B, 10 mm²/s MuT B, 10 mm²/s
Deviation [%]
Deviation [%]
0.0 0.0
-0.1 -0.1
-0.2 -0.2
0 1000 2000 3000 4000 10 000 100 000 1 000 000
Volume Flow Rate [m³/h] Reynolds number [-]
Fig. 5: As-found calibration results of two turbine meters A (in blue) and B (in red). The deviation is
plotted as a function of the flow rate [m³/h] (left) and the Reynolds number [-] (right).
5.2 Coriolis meters

An example of the calibration of a coriolis meter is shown in Fig. 6. The coriolis meter was
calibrated using the piston prover as a reference. On the left hand side the deviation is
plotted versus the mass flow rate, on the right hand side the deviation is plotted versus the
Reynolds number. The error bars in both graphs indicate the expanded overall uncertainty
( ). All results range between -1.5% and +0.5%. In the left graph of Fig. 6 the viscosity
dependency of the meter is clearly visible. The difference between the deviations at a
specific flow rate exceeds the uncertainties and is therefore significant. The higher the
viscosity the more the curve drops to minus at low flow rates. The right hand graph of Fig. 6
shows that the curves on the different product connect well in the overlapping Reynolds
range. This knowledge allows the manufacturer to compensate for the Reynolds
dependency.
7
20 - 23 October 2015
Technical paper
For a different coriolis meter Fig. 7 shows an example what will happen if such
compensation is in place. Here calibration results are plotted using a liquid with a nominal
viscosity of 100 mm²/s. The results are cumulated from runs performed on two dates.
Where in Fig. 6 the results of a regular coriolis meter for the highest viscosity liquid range
from -1.5% to 0.0%, the deviations for the viscosity compensated meter are between -0.2%
to 0.05%. This is a reduction in the dispersion of the observed deviations with a factor six.
Before putting out the flag, some remarks are to be made here. During the measurements
the coriolis meter showed substantial variations of the viscosity indication while the actual
viscosity was constant. Unfortunately, there was no opportunity to test the meter with
other liquid viscosities. It would be interesting to test the meter with both lower and higher
viscosities. And lastly the question is whether the viscosity compensation mechanism is
influenced by other parameters or conditions.
1.0 1.0
0.5 0.5
0.0
Deviation [%]
0.0
Deviation [%]
-0.5 -0.5
(1.161 ± 0.002) mm²/s (1.161 ± 0.002) mm²/s

-1.0 -1.0
(11.13 ± 0.37) mm²/s (11.13 ± 0.37) mm²/s
(91.6 ± 4.5) mm²/s (91.6 ± 4.5) mm²/s
-1.5 -1.5
-2.0 -2.0
0 200 400 600 800 1 000 1 000 10 000 100 000 1 000 000 10 000 000
Mass flow rate [kg/h] Reynolds number [-]
Fig. 6: Results of the calibration of a coriolis meter with three different viscosity liquids. The deviation
[%] is plotted versus the mass flow rate [kg/h] (left graph) and versus the Reynolds number [-]. Error
bars represent the overall expanded uncertainty.
0.15 0.15
0.1 0.1
0.05 0.05
Devaition [%]
Deviation [%]
0 0
-0.05 -0.05
-0.1 -0.1
-0.15 -0.15
100 mm²/s, 29 Apr 2015
-0.2 100 mm²/s, 29 Apr 2015
100 mm²/s, 24 Apr 2015 -0.2
100 mm²/s, 24 Apr 2015
-0.25
-0.25
0 500 1000 1500 2000 2500 3000
1000 10000 100000
Mass flow rate [kg/h] Reynolds number [-]
Fig. 7: Calibration results of a coriolis meter with viscosity compensation using a liquid with a nominal
viscosity of 100 mm²/s. The deviation [%] is plotted versus the mass flow rate [kg/h] (left graph) and
versus the Reynolds number [-]. Error bars represent the overall expanded uncertainty.
8
20 - 23 October 2015
Technical paper
5.3 Ultrasonic meters

For ultrasonic meters Reynolds-based profile corrections are already common practise.
This makes the meter suitable for intercomparison exercises. In a previous paper [2] an
intercomparison was described between the Trapil and EuroLoop facilities, in which only
EurpLoop’s small piston prover was used. At EuroLoop the calibrations were repeated with
the big piston prover using a single meter (i.e. 8” Altosonic 5 NG meter s/n A14050085 –
19YA20901_2002) that was used in the previous experiment. The calibration results were
compared with the previously obtained results at Trapil [2].
The left picture in Fig. 8 gives an overview of all calibration results obtained both at Trapil
and at EuroLoop utilizing the big piston prover. All calibration results are found in a
bandwidth of 0.2%, which is the result of the implementation of a profile correction
algorithm. Error bars indicate the overall expanded uncertainties, black for Trapil’s results
and blue for EuroLoop’s results. Uncertainties range between 0.03% and 0.11%, which is
caused by differences in repeatability.
The agreement between Trapil and EuroLoop was evaluated using defined in Eq. 1.
The Trapil results were obtained at Reynolds numbers that differ from the Reynolds
numbers of the EuroLoop results. In order to compare the results a linear interpolation was
used to translate the observed deviations at Trapil and their uncertainties to the Reynolds
numbers of the EuroLoop observations. In formula:
(2)
in which is the Reynolds number of the observation at EuroLoop, and are the
adjacent Reynolds numbers corresponding to the deviations and in Trapil, and
are the expanded uncertainties corresponding to and .
0.15 2.0
0.10 1.8
En 1 mm²/s
Normalized deviation En [-]
1.6
0.05 En 10 mm²/s
1.4
Deviation [%]
0.00 En 100 mm²/s

1.2
-0.05 1.0
-0.10 0.8
Gasoline 0.80 mm²/s
Heating oil 5.6 mm²/s 0.6
-0.15 Heavy crude 64 mm²/s 0.4
Euroloop 1 mm²/s LP
-0.20 0.2
-0.25 0.0
1000 10000 100000 1000000 10000000 1000 10000 100000 1000000 10000000
Reynolds number [-] Reynolds Number [-]
Fig. 8: Meter curves obtained at EuroLoop and Trapil for meter A (left). The deviation of the meter [%]
is plotted versus the Reynolds number [-]. Error bars indicate the total uncertainty (k=2) of the
measurement results. The grey and the blue bars are attached to Trapil and EuroLoop results
respectively. On the right the normalized deviation [-] is plotted versus the Reynolds number [-].
9
20 - 23 October 2015
Technical paper
The results are depicted in the right graph of Fig. 8. All 19 values are less than 1,
16 values are less than 0.5. These results mean that all Trapil and EuroLoop results
agree with at least 95.4% confidence.
6. Conclusions
Now EuroLoop is fully operational the characteristics of the facilities can be compared with
the specifications the facility was designed to. From the operational experiences described
in the previous chapters to following conclusions can be drawn.
1. EuroLoop is very suitable to test the viscosity dependence of flow meters.
Instruments with viscosity compensation installed can be calibrated and verified at
EuroLoop leading to improved meter performance at lab conditions.
2. The temperature stability observed during a series of successive calibration runs at
one specific flow rate appears to be much more constant than originally expected.
Instead of 0.1°C double standard deviations of 0.002°C were observed. Despite this
impressive result a temperature repeatability (2s) of 0.05°C will be used in the
uncertainty analysis.
3. After earlier intercomparisons using the small piston prover [2], even better
comparison results were obtained between Trapil and EuroLoop using the big piston
prover. All observations have a normalized difference smaller than 1, which means
that the results agree with more than 95.4% confidence. From the 19 observations,
16 had a normalized difference of smaller than 0.5.
4. Point of attention for the metrological maintenance is the analysis of liquid samples.
Although traceability is in order, there is a significant difference between the two
test and calibration laboratories for the lowest and highest viscosity liquids. The
density of the samples agrees within their mutual uncertainties. Further steps will be
necessary to resolve the differences.
7. References
[1] Jos van der Grinten, Bart van der Stap and André Boer (2015): Preparing the new EuroLoop
liquid flow facility for accreditation, Lecture presented at the 3rd European Flow Measurement
Workshop, 17-19 March 2015, Noordwijk, The Netherlands
[2] Jos van der Grinten, Bart van der Stap and Pico Brand (2015): Round robin testing for the new
EuroLoop liquid flow facility, International Symposium for Fluid Flow Measurement, 14-17
April 2015, Arlington, Virginia, USA
[3] ASTM D 7042-14 (2014): Standard Test Method for Dynamic Viscosity and Density of Liquids
by Stabinger Viscometer (and the Calculation of Kinematic Viscosity), ASTM International, PA,
USA
[4] ASTM D 4052-11 (2011): Standard Test Method for Density, Relative Density, and API Gravity
of Liquids by Digital Density Meter, ASTM International, PA, USA
10
20-23 October 2015
Technical Paper
REMOTE METERING MONITORING AND SMART METERING ROOM FOR

COST EFFECTIVE OPERATION OF MULTIPHASE METERS
Jean–Paul COUPUT
TOTAL
1 INTRODUCTION
Since 15 years, multiphase and wet gas metering techniques have been largely
deployed by industry for well testing and well metering especially for subsea and
unmanned development. The main reasons for that were to avoid expensive test
separator solutions associated with test lines requirements mainly for CAPEX
reduction.
TOTAL for instance is operating more than 150 multiphase flow meters (MPFM)
worldwide from several suppliers with a majority located subsea (one meter per
well) with minimum access due to high water depths.
There have been a lot of feedbacks challenging accuracy & reliability of

multiphase metering solutions which may be subject to drift, fluid property
changes and also sensors failure.
Maintaining data customer confidence in flow measurements and required

accuracy is essential both for
- Fiscal measurements
- And technical application like reservoir monitoring and well production
optimization
For development using MPFM, decisions are dependent on information’s provided

by such meters and well allocation for reservoir monitoring as well as forecast
and shortfall calculations are based on well flow data provided by MPFM.
This is more critical in applications cases in which either MPFM are directly used to
calculate overall marginal field productions or sum of well measurements
provided by MPFM allow to calculate field production for fiscal allocation purposes.
Conventional way of operating MPFM by using both maintenance and operation

practices adapted for topside applications as well as relying on suppliers spot or
periodic assistance has been found not efficient and costly. Mobilization of
manufacturer’s specialists with offshore visits is also quite expensive and not
compatible with the current climate of cost reduction.
In order to improve data quality for technical and fiscal applications but also to
reduce OPEX related to maintenance and validation, real-time Remote Metering
Monitoring ( RMM ) as well as collaborative work in Smart Metering Room
facilities have been developed and implemented within TOTAL affiliate and
headquarters to follow and trouble shoot multiphase meters & wet gas meters as
well
1
20-23 October 2015
Technical Paper
2 MULTIPHASE METERING TECHNICAL & COST CHALLENGES [1]
MFFM & WGFM which are bringing clear advantages on traditional approaches
based on separation are largely applied in Oil & Gas productions by Operators...
Most of technology are based on known physical principles and are using static
sensors which per definition should be quite reliable.
Nevertheless, maintaining accuracy of oil, water & gas phase flow rates
determinations and confidence in information provided has been a challenge.
Most of technology is deriving phase fractions using on fluid property contrast
measurements like density, gamma attenuation, and permittivity: this requires a
reasonably good knowledge of individual phase fluid property and generates a
rather high sensitivity to fluid property changes.
Partial failures have also been noticed due to problems on some sensors related
to effect of severe process conditions able to generate conductive films with short
circuits, deposits...
Taking samples and going to the laboratory or sending samples to specific labs for
PVT analysis as recommended by some suppliers is rather difficult, costly and not
suitable at all in subsea & unmanned situations.
In subsea environment where normal practice is to have one MPFM per well,
removing and changing fully retrievable systems is also very costly. To reduce
development costs of subsea fields, the tendency is to use non retrievable or
partially retrievable systems: it is obvious in that case, meter cannot be retrieved
and new operating approaches have to be implemented.
Flow calibration using a reference flow system like a test separator may be
required in some circumstances to comply with regulations or agreements when
MPFM measurement is used for fiscal allocation to determine ownership.
In addition to technical challenges, MPFM are associated with
- Complexity requiring specialists and experts
- Significant OPEX for configuration verification and reprogramming.
OPEX are due both to supplier mobilisation & trips (from 2 to 4 times a year) and
also to in house time spent by metering specialists, operation people and data
customers like Reservoir or Well performance to troubleshoot MPFM used at well
level.
2
20-23 October 2015
Technical Paper
This may result on a significant time spent annually per multiphase metering
systems. Time may exceed one week or more per MPFM if no specific follow up is
deployed.
3 SMART METERING SOLUTIONS FOR MPFM OPERATION [2]
To improve both data quality and also to reduce operating costs related to
multiphase metering systems , so called “smart metering “ solutions have been
developed and implemented in Operating assets & in TOTAL headquarters in
France .
Basically, such approaches may include
- Remote metering monitoring
- Remote operation
- Collaborative work in Smart Rooms
- Dedicated Smart Metering room
4 REMOTE METERING MONITORING ( RMM ) and OPERATION
Basically RMM consists of monitoring continuously flow data as well as raw data
and all relevant sensor data to follow and trouble shoot metering.
Data is acquired through real time field data made accessible in EIS environment
through historical data base like PI systems.
Data is transferred from site and made available to Support entity of Operational
Assets or to TOTAL headquarters.
3
20-23 October 2015
Technical Paper
In order to make approach clear and standard, data and information are
structured and described through data asset modelsin PI-AF .
Following screens give examples of typical information category used for MPFM.
In house routine have been developed and implemented in PI – AF modules to run

verification calculations or to check consistency.
Knowhow and in house expertise of users is implemented in RMM through choice

of detection criteria or customized calculations.
Two main levels of RMM can be used : Standard Remote Metering Monitoring
and Advanced Monitoring.
Standard RMM
Standard or routine RMM is applied to carry out continous monitoring and to

check operating performance of meters for instance MPFM operating points
should remain within enveloppes .
Static period are also analyzed remotely to verify fluid property like gas density
or permittivity depending on MPFM technology .
4
20-23 October 2015
Technical Paper
8000 0
GVF 10% ‐50 ‐45 ‐40 ‐35 ‐30 ‐25 ‐20 ‐15 ‐10 ‐5 0
GVF 30% ‐2
7000
Operating Point ‐ ORQ552 ‐ P1012
‐4
6000
GVF 50%
5000 mbar
‐6
Liquid flow rate (actual Am3/day)
5000
‐8
High Energy
4000
GVF 70%
‐10
3000 ‐12
‐14
2000
Period 1
Period 2 ‐16
Period 3
1000
Period 4
Period 5 ‐18
50 mbar
0 0.1
Triangle Data ‐20
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 Low Energy
Gas flow rate (actual Am3/day) Operating Point ‐ 23/06 20:40 to 23/06 21:00
Thanks to continuous monitoring , normal production events like production

shutdowns are used to make some simple calibration of permittivity or gas
density without stopping productions
By facilitating failure detection on some electrical sensors , monitoring influence

of chemical injection and identitying scale deposition on some meters , RMM
allows to improve operation of MPFM and bring some added value to Operations
for flow assurance and production optimisation.
Advanced RMM
In addition to routine tools and simple calculations, RMM may take advantage of
advanced tools and software to calculate fluid property to validate data using &
data reconciliation software (DVR).
Data validation & reconciliation approach is using information redundancy and

input data uncertainty to
- Detect sensor & meter failure
- Calculate uncertainty
- To calculate back up data
5
20-23 October 2015
Technical Paper
DVR brings a back up for multiphase flow meters by using other sensors like
choke valve opening, pressure and temperature combined with flow models.
This is illustrated in the following recording in which there is a failure on water

measurements given by a multiphase meter.
Thanks to redundancy, DVR calculated a back up value for Water Liquid Ratio.
5 SMART METERING ROOM
A Smart metering room putting together specialists and experts through

communication and visualization tools is in operation within TOTAL technical
center located i n Pau, France.
The Smart Metering Room (SMR) is equipped with cutting-edge data analysis
tools and modern technical devices and interactive system that enable several
specialists to work on the same files and to interface with participants’ computers.
TOTAML smart bmetering room is currently use to monitor, validate and
troubleshoot most of subsea multiphase meters ( > 130 units ) installed
worldwide as well as topside flow meters.
After MPFM flow data is captured and analyzed, specialists, experts and data
users are sharing information locally or remotely to make diagnostics and solve
metering issues.
6
20-23 October 2015
Technical Paper
For MPFM used for well metering & testing , Smart Metering Room allows to share
data between HQ specialists , Site Operation and Well performance people as well
as with Suppliers in order to trouble shoot systems in a quicker way .
This has also resulted in a significant cost reduction for MPFM operation : two
times less visits for instance , no shutdown for calibration , less people on site .
6 BENEFITS
Optimal data quality
● Real time monitoring
● Imbalances reduction and better well allocation
● More confidence in MPFM data
● Ccorrect data to improve short- and long- term production forecasts and
to adjust well performance models
Reactivity
 Remote monitoring and answers to affiliate queries in real time;
 Spot, periodic, or continuous assistance (from daily service to monthly
reporting).
Organization and training
● MPFM Skills development in Operations “ RMM and SMR are highly
valuable for us to acquire and develop specific competencies and
expertise”
● Experts available for the users
● Constant dialog between the specialists in headquarters and the
Operations
● Standardized monitoring practices;
Safety and security
 Less site intervention
 Reduction of risks
 Less people on board
Economics & cost reductions
● Wells are no longer shut down for meter calibration
● Use of unplanned production shutdowns to check the systems
● Less costly interventions from specialists
● Suppliers visit divided by 2
● Intervention cost reduction through active preparation
7 CONCLUSIONS
Smart Metering solutions like remote metering monitoring or smart metering

room have been successfully applied and implemented by TOTAL on several field
locations in order to improve multiphase metering systems operations.
Thanks to them, the full added value of installed multiphase meters is used and
confidence in data has increased.
Implementation of RMM has reduced cost of MPFM operation with better data
quality for reservoir and well monitoring.
.
For full efficiency, we had to incorporate the business and metering metier know-
how and to combine it with digital technology .This is particularly true for Remote
Metering Monitoring and collaborative work carried out in Smart Metering room?
A lot of improvement may still be made through standardisation of required data
and supplier cooperation to give access to raw data. For MPFM , black boxes
should be prohibited for the benefits of everyone.
7
20-23 October 2015
Technical Paper
It is also important to continue to develop associated monitoring tools and

technology through RD initiatives.
RMM & Smart Metering Rooms are part of the digital solutions promoted within
TOTAL operations to improve quality of MPFM metering data but also others
systems like fiscal ones
8 REFERENCES
[1] J.P. Couput , TOTAL ; Subsea multiphase measurements , 29th North Sea
Flow Measurement workshop , 2011
[2] J.P Couput, TOTAL, SPE Smart Metering Solutions: How to Combine Digital
Technology with Business Expertise to Improve Metering Performance, 2015
Abbreviations
RMM: Remote Metering Monitoring

SMR: Smart Metering Room
DVR: Data Validation and Reconciliation
MPFM: Multiphase Flow Meter
WGFM: Wet Gas Flow Meter
Acknowledgement
The Author would like to thank Renaud Caulier & Violaine Sontot for their
contribution to Remote Metering Monitoring deployment .
8
Practical Experiences Obtained with the
Magnetic Resonance Multiphase Flowmeter
Jankees Hogendoorn*, Mark van der Zande*, Marco Zoeteweij*,
Rutger Tromp*, Lucas Cerioni*, André Boer*, Rick de Leeuw+
*Krohne, +Shell Global Solutions International B.V.
1 INTRODUCTION
Magnetic Resonance (MR) is a measurement technology which is well known from its
application in medical imaging. Next to this application in Healthcare, MR can also be
applied in the upstream oil and gas industry to measure multiphase flow of oil/water/gas
mixtures upstream of a separator. To this end, an industrial magnetic resonance multiphase
flowmeter has been developed.
In previous papers [1],[2] the MR measurement principle is introduced, and explained how
the flowrate of oil, water and gas are measured. Additionally, experimental results are shown
of extensive tests performed at various independent multiphase flow loops. The main
conclusion of that study is that MR is a powerful technique for measuring multiphase flow.
Good accuracy can be achieved over a wide range of condition such as GVF ranging from 0-
95% and Water Liquid Ratio (WLR) ranging from 0-100%, with a sweet spot of the
flowmeter at high WLR of 85% or higher.
With the experience gained in research and development, and the lessons learned from the
tests performed, further improvements have been made to the flowmeter and a first series of
MR multiphase flowmeters has been produced (see Fig. 1). One of the flowmeters has been
installed in a field trail by a Dutch exploration and production company at a production
location in the Netherlands. The main focus of this field trial is to validate practical aspects of
the multiphase flowmeter such as installation procedures and long term stability of the
flowmeter in changing environmental conditions. Furthermore, the concept of in-line fluid
characterization (without the need of taking samples) is evaluated. The practical experiences
gained from the field trial are described in this paper.
Figure 1: The magnetic resonance multiphase flowmeter (M-PHASE 5000) as developed by

Krohne and Shell.
2 DESCRIPTION FIELD TRIAL APPLICATION
After the successful tests on various multiphase test loops, the magnetic resonance multiphase
flowmeter has been installed in a field trial by a Dutch exploration and production company
on a production location in the Netherlands.
The field comprises 16 different wells that are connected to a main production line.
Additionally, each well is connected to a test manifold. In this way, the production of each
individual well can be tested by means of a test separator. During the field trial 9 wells have
been tested. In the test separator the liquid and gas are separated, and each flow is measured
separately. The oil-water mixture is measured using a Coriolis flowmeter and the WLR is
determined based on density. Prior to the start of the field test, the base densities have been
sampled. The gas flow is measured using a vortex meter. Both the Coriolis and the vortex
meter have been calibrated before the start of the field test.
The magnetic resonance multiphase flowmeter has been installed upstream of the test
separator (see figs. 2 and 3).
M-PHASE 5000
Figure 2: The magnetic resonance multiphase flowmeter (M-PHASE 5000) installed in a field
test by a Dutch exploration and production company on a production location in the
Netherlands.
To test
separator
M-PHASE 5000
3.
1.
Multiphase
2. flow from well
Figure 3: Schematic representation of the configuration at the field test. The flowmeter is
positioned in a bypass, which can be controlled via valves indicated by 1, 2 and 3. When the
flowmeter is in normal operation, valve 1 is closed and valve 2 and 3 are open; in this way
the flowmeter is in-line with the test separator. In case the in-line fluid characterization is
performed, valve 1 is open and valve 2 and 3 are closed to block the fluids in the meter.
At the start of the field test, it was clear that the flow conditions of the wells are outside the
optimal application range of the flowmeter; for most wells the GVF is high (>95%) as
significant gas lift is applied. Nevertheless, it was decided to perform the field test to gain
valuable practical experience regarding the installation procedure, the in-line fluid
characterization procedure and the robustness against environmental conditions. The
Netherlands has a moderate maritime climate where on average the relative humidity is
between 75% and 90% and the average precipitation (rain, snow, hail) per month is 70mm
distributed over 15 days per month. The average ambient temperature ranges between 0 oC up
to 23 oC, however during the field trial the temperature ranged between -4 oC and 34 oC.
3 INSTALLATION OF THE METER AND START UP
The meter was installed in October 2014. Before being allowed to install the meter all official
documents had to be in place like material certificates, CE declaration, PED approval
documents Lloyd’s, approval for applying the meter in a hazardous area, Site Acceptance Test
documents, etc..
Since the meter didn’t have an explosion safety approval at the time of installation, an active
nitrogen purge with flow control unit has been used to cover the explosion safety related
aspect of the meter. Meanwhile, the MR meter has an explosion safety approval.
Consequently, the nitrogen purge is not necessary anymore for new installations of the
flowmeter. The installation of the meter went smoothly without difficulties. The same holds
for starting up the meter.
4 IN-LINE FLUID CHARACTERIZATION
Advantage of In-line Fluid Characterization

One of the goals of this field trial is to test how the in-line fluid characterization procedure
works out in the field. The outcome of this part of the test is very important, as the operating
concept of the meter in the field becomes very straight forward: block-in the flowmeter for a
certain moment, run the ‘auto-fluid-characterization’ procedure, open the valves and start the
flow measurement.
This means that no samples are necessary anymore and testing procedures in the lab at similar
pressures and temperatures are superfluous. Furthermore, issues like differences between live
oil and dead oil properties are gone, no additional equipment is necessary, the hassle of
sample handling is over, and so on, and so on.
Summarizing, we may conclude that in-line fluid characterization has many advantages
compared to off-line characterization by means of samples.
Description of the In-Line Fluid Characterization

For a proper multiphase flow measurement, the MR meter requires the relaxation times (T1
and T2) and the Hydrogen Index (HI) both for oil and water [1]. These parameters can be
measured by the MR meter itself. The only requirement is a static situation i.e. no flow.
The characterization procedure is as follows:

 The multiphase mixture is guided through the meter.
 Once the representative oil and water are present in the meter, valve 1 (figure 3) is
opened and valves 2 and 3 are closed.
o Note that the ratio of oil, water and gas doesn’t have to correspond to the multiphase
flow composition. The only criterion is to have representative oil and water present in
the meter.
o Furthermore, the presence of gas is not necessary since T1, T2 and Hydrogen Index
(HI) of gas is not critical in the gas measurement and can be calculated from the
pressure and temperature.
 When the flow has been blocked-in and the multiphase mixture is static, the automated
fluid characterization procedure can be carried out.
 At the end of the characterization procedure, the relevant fluid parameters are stored
the meter.
Fluid Characterization Duration

At the very beginning of the field test, the in-line characterization procedure took 10 to 24
hours! This was partially caused by the fact that too much data was collected. Right from the
start, it was concluded that such a long characterization time is not practical and therefore not
acceptable.
By optimizing the routine, the characterization period has been reduced to about 45 minutes.
It turned out that this time interval works out well in the field. Meanwhile, the
characterization time is being reduced further. The current estimation is that it will be reduced
with another factor of two.
Live Oil and Dead Oil

As mentioned before, differences in properties between dead oil and live oil are observed. In a
dead oil sample the pressure is sufficiently low (typically ambient pressure) such that all
dissolved gas and volatile fractions have left the oil phase. In a live oil sample (at a higher
process pressure) part of the gas phase is dissolved in the oil phase. This dissolved gas
fraction does affect the oil properties. One of the properties that is affected is the oil viscosity.
Typically, the dissolved gas leads to a lower viscosity. Consequently the T1 and T2 times are
higher. Very little quantitative information is available on the difference in relaxation times
between live oil and dead oil.
The differences between live and dead oil are shown in figures 4 and 5. These figures show
the logmean value of the T2 time distribution as function of temperature for the live oil as
obtained by the in-line fluid characterization procedure in the field (purple triangles).
0,6
0,5 dead oil sample: off-line at ambient pressure

T2 Relaxation time [s]
live oil sample: in-line at 5 [bara]

0,4
0,3
0,2
0,1
Well 1
0
0 10 20 30 40 50 60
Temperature [degC]
Figure 4: The logmean value of the T2 distribution as function of temperature for the
oil of well 1, obtained by means of in-line fluid characterization at a process pressure
of 5 bara (purple triangles) and by means of off-line characterization performed in the
laboratory at ambient pressure (red squares).
The dead oil results obtained off-line at ambient pressure in the laboratory are represented by
the red squares. In order to protect confidential information of the Dutch exploration and
production company the well numbers which are mentioned do not correspond with the well
numbers from the field. Three T2 measurements have been performed during the in-line
characterization. During the measurements the fluid temperature has changed somewhat due
to heat exchange with the environment. The dead oil samples have been measured in the
laboratory over a temperature range of about 10 to 50 °C.
The T2 time of the live oil sample is slightly higher than the T2 time of the dead oil sample in
the same temperature range. It is expected that the differences between dead and live oil are
more pronounced for lighter crude oils which contain a larger fraction of short, volatile,
hydrocarbon molecules. At ambient pressure, a relative larger part of the volatile components
will be released from the oil. Consequently, the viscosity and T2 time will be more heavily
affected. This is confirmed by measurement results as shown in figure 5. The oil as shown in
this figure has a somewhat higher T2 value than the oil in figure 4. A higher T2 time means a
lower viscosity because of a relative larger part of shorter hydrocarbon molecules. The
difference between the live oil (purple triangles) and the dead oil sample (red squares) is more
pronounced.
0,9
0,8
dead oil sample: off-line at ambient pressure
0,7
live oil sample: in-line at 5 [bara]

0,6
0,5
0,4
0,3
0,2
0,1
Well 2
0
0 10 20 30 40 50 60
Temperature [degC]
Figure 5: The logmean value of the T2 distribution as function of temperature for a lighter oil
obtained from well 2. The purple triangles are the results obtained by means of in-line fluid
characterization at a process pressure of 5 bara. The red squares are the results from off-line
characterization in the laboratory at ambient pressure.
Accuracy of In-Line Oil Characterization (Live Oil)

An important question is how accurate the in-line and laboratory fluid characterization is. But
an even more important question is how an error in the T2 determination propagates in the
flow measurement results. Therefore, the sensitivity to T2 has been quantified.
First of all, an estimation is made for the error in the in-line fluid characterization result. Both
figures 4 and 5 show a clear temperature dependency of T2. When this temperature
dependency trend is used as a basis and the standard deviation of the variation in the in-line
characterization result is quantified, a standard deviation of 6% is obtained in T2 time for the
in-line results as shown in figure 4. A standard deviation of 8% is obtained for the results as
shown in figure 5. The standard deviation can be taken as a measure for the error in the T2
measurement. The standard deviation in T1 measurements is comparable to the standard
deviation in T2 measurements. Table 1 shows the effect of 8% standard deviation in T1 and T2
measurement of oil on the gas flowrate, the liquid flowrate and the WLR by means of a
sensitivity analysis. This well has a WLR of 90%. Qgas[%] is defined as the percentage
change in gas flowrate as result of the standard deviation (measure for the error) in relaxation
time. The same definition holds for the liquid flowrate and WLR.
Parameter . . ∆ [%] ∆ [%] ∆ [%]

T1 8% 0.0% 0.7% 0.6%
T2 8% 0.6% 0.3% 0.0%
HI 5% 0.1% 0.9% 0.8%
Table 1: Sensitivity analysis to clarify the relation between the inaccuracy in T1 ,T2 and HI for
oil and the resulting inaccuracy in the measured flowrates of gas and liquid and WLR. This
result is obtained on a well with a WLR of 90%.
Due to the high GVF (> 95%) in this study, the total amount of liquid present in the meter is
small. This small amount of liquid is enough for T1 and T2 determination. However, it is not
so obvious to accurately determine the Hydrogen Index (HI). For this situation the alternative
solution is to use the HI for water as given in literature [3]. The uncertainty on HI by using
this relationship is estimated to be 2%. A similar approach can be applied for oil. The
uncertainty in the HI of oil by following this approach is 5%. The effect on the individual
flowrates and WLR is quantified in tables 1 and 4.
Accuracy of Off-Line Oil Characterization at Ambient Pressure (Dead Oil)

In case an in-line fluid characterization is not possible (for whatever reason), the oil sample
should be analyzed off-line at ambient pressure. For this situation the differences in T2 time
between the in-line and off-line measurements are more pronounced and vary from 11% for
the measurement as shown in figure 4 to 50% for figure 5. The effect of this 50% shift in T1
and T2 time for the off-line characterization is quantified in table 2.
Parameter ∆ [%] ∆ [%] ∆ [%]

T1 50% 0.2% 4.2% 4.0%
T2 50% 3.5% 2.1% 0.3%
Table 2: The effect of a difference of 50% in T1 and in T2 between live and dead oil on the
measured flowrate of gas and liquid and WLR.
We expect that the differences are even more pronounced for higher process pressures, since a
relative larger part of gas will be dissolved in the oil. Although the difference in T1 and T2
between dead and live oil can be quite significant, the effect on the measured flowrates is
much more limited. It can be concluded that off-line fluid characterization is suitable,
however a higher accuracy in measured flowrates can be achieved in case in-line fluid
characterization is applied. It should furthermore be noted that the error in the off-line
characterization can be reduced by introducing a model that accounts for the effect of the
dissolved gas fraction in the oil.
Frequency of In-Line Fluid Characterization
Another item to be addressed is the frequency at which the fluid characterization has to be
carried out. Is this for example once a month, once a year or only one time during the meter
start-up after installation?
Of course, the answer to this question is related to the stability of the oil and water
composition. As long as the composition of each phase hasn’t changed, a fluid
characterization is not required. In order to get a feel for the stability of oil and water
composition, two samples from the same well have been analyzed which have been retracted
with an intermediate period of two years. The result of this experiment is shown in figure 6.
This graph shows the T2 time as a function of temperature for both samples. The red triangles
show the measurement results of the sample collected in February 2013. The red squares are
the results of the sample collected in February 2015 from the same well.
Well1, Feb 2013 - Feb 2015

0,30
Lab measurement on sample of 19-feb-15
0,25 Lab measurement on sample of 26-feb-13
T2 relaxation time [s]
0,20
0,15
0,10
0,05
0,00
10 15 20 25 30 35
Temperature [degC]
Figure 6: T2 relaxation time as function of temperature for oil samples of the same well
collected with an intermediate period of two years. These results are obtained by means of an
off-line characterization and give an indication for the stability of the composition of both the
oil and water phase in time.
The result as shown in figure 6 shows that the T2 relaxation time for oil over time is pretty
stable. The curve obtained on the 2013 sample is less smooth than the curve obtained from the
2015 sample. It is not clear yet what the reason is.
The standard deviation on the differences between the two T2 curves is 11%. In accordance
with the same sensitivity analysis as discussed before, the effect on the flowmeter reading can
be quantified. The result is shown in table 3. This experiment seems to show that the fluid
characterization procedure for this particular well can be carried out with a low frequency
(e.g. once every 6 months), just to check the validity of the fluid parameters.
It should be noted that this well is producing for many years and seems very stable.
Experience has to be collected for each well individually.
Parameter . . ∆ [%] ∆ [%] ∆ [%]
T1 11% 0.0% 0.9% 0.9%
T2 11% 0.8% 0.5% 0.0%
Table 3: Sensitivity analysis to clarify the relation between the difference in T1 , T2 of oil for
two samples taken at a 2 year interval and the resulting deviation in the measured flowrates
of gas and liquid and WLR.
It is recommended to run a fluid characterization more frequently during the first month after
the installation of a multiphase flowmeter to quantify the stability of the oil composition and
determine the fluid characterization frequency accordingly.
Differences between Wells from the Same Reservoir

Another interesting topic is the assessment of the differences between two wells from the
same reservoir. A result of this comparison is shown in figure 7. This graph shows the T2
relaxation times obtained in the laboratory as function of temperature for two different wells
producing from the same reservoir. The vertical axis is in a log scale to show the constant off-
set between the two curves over the entire temperature range.
Two different wells, one reservoir

1
Well2: sample from 19Feb2015
well1: sample from 18Dec2014
T2 relaxation time [s]
0,1
0 10 20 30 40 50 60
Temperature [degC]
Figure 7: A comparison between the T2 relaxation times obtained in the laboratory as a

function of temperature for two different wells producing from the same reservoir.
This graph clearly shows that there are differences between wells from the same reservoir. It
is hard to make a general statement on basis of this single set of data. For this specific
situation, the error could be estimated on basis of the sensitivity analysis, in case the T1 and T2
relaxation time of one well is used for the other one. The error would end up in approximately
4% for the gas and liquid flowrate as well as the WLR. This is too much to neglect the
difference in properties. So, in this case both wells should be characterized. Therefore, the
software of the flowmeter is designed such, that the flowmeter can store (and retrieve)
multiple sets of fluid characteristics, for example one set of fluid characteristics for each well
that is connected to the flowmeter.
In-Line Water Characterization
Next to the oil characterization, the water has to be characterized. The in-line characterization
of water is being carried out at the same time as the oil characterization. These are not two
separate measurements, but carried out in the same characterization run. The same holds for
the off-line characterization in the laboratory.
An example of an in-line and off-line water characterization result of well 1 is shown in figure
8.
2
1,8 water sample: off-line at ambient pressure
1,6 water sample: in-line at 5 [bara]
1,4
1,2
1
0,8
0,6
0,4
0,2 Well 1
0
0 10 20 30 40 50 60
Temperature [degC]
Figure 8: T2 relaxation time as a function of temperature for an in-line (green triangles) and
off-line water characterization run (blue squares) for well 1.
If we take the temperature dependency into account and determine the standard deviation on
the relaxation times of the in-line results for the different wells, values in the range from 0.5
to 3% are obtained. Analogue to the oil results, the effect of this standard deviation can be
translated to the estimated error in the individual flowrate and WLR determination. An
overview is given in table 4.
Parameter . . ∆ [%] ∆ [%] ∆ [%]

T1 3% 0.0% 1.8% 0.1%
T2 3% 0.0% 0.0% 0.0%
HI 2% 0.0% 1.6% 0.2%
Table 4: Sensitivity analysis to clarify the relation between inaccuracy in T1 ,T2 and HI for
water and the resulting inaccuracy in the measured flowrates of gas and liquid and WLR.
This table shows that the determination of T1 of water is important for the liquid flow
determination. The other measurement values are very weakly dependent on the accuracy of
the relaxation time. Something similar holds for HI. The HI value is important for the liquid
flowrate and much less for the gas flowrate and WLR.
Another phenomenon that can be observed from figure 8 is the difference between the in-line
and off-line characterization. Most likely, this difference is caused by oxygen diffusion into
the water phase. The effect of oxygen on HI is negligible, but it is known from literature
[4][5][6] that the paramagnetic behavior of oxygen significantly reduces the T1 and T2
relaxation times (up to 20% at 15°C).
The sampling procedure during the field test has been carried out in the open air without any
protective gas like e.g. nitrogen. The result as shown in figure 8 illustrates the importance to
prevent oxygen being dissolved in the water during sampling. Another phenomenon that
could lead to a pronounced difference in T1 and T2 for water between the in-line and off-line
conditions is the presence of volatile components in the sample. An investigation to quantify
this effect is ongoing.
It is important to note, that in case in-line fluid characterization is performed, the effect of
oxygen and volatile components on the measured properties of water does not need to be
taken into account, as the fluid characterization takes place under the actual line conditions.
5 OTHER LESSONS LEARNED
Process Temperature Variations

As mentioned earlier, measurements have been carried out at 9 different wells. The
measurements at the different wells have demonstrated that the process temperature at the
measurement location can vary significantly. For this field trial, the largest part of the
temperature differences is caused by heat transfer to the environment. Due to the lower
ambient temperature in combination with long distance between the well heads and the
flowmeter (several hundreds of meters) and the differences in flowrates, the temperature
change can be as large as 20 °C.
An illustration of such a process temperature change at the measurement location is given in
figure 9. The purple line indicates the process temperature at the location of the multiphase
flowmeter. The blue line is the process pressure. A very distinct temperature drop is observed
during a well change at 19:00h. After the well change, we still observe a variation in the
temperature. This is caused by the unsteady flow condition in the second well.
Figure 9: Process temperature and process pressure as function of time during a well change.
The temperature between two wells can vary significantly. After the well change the
temperature of the second is varying due to the nature of the unsteady flow in combination
with heat transfer to the environment (ambient temperature is about 0°C).
The field test has shown that the multiphase flowmeter is able to cope without difficulties
with these significant changes in process temperature over a short time.
Flowrate and Composition Variations

A very pronounced phenomenon that has been observed during the field tests is related to the
unsteady nature of multiphase flow; the liquid flowrate can vary significantly over time.
During the field test, time scales have been observed from tens of seconds up to hours. In
addition, the fluctuations in water and oil flow can run out of phase and lead to large
variations in the WLR. A clear example of such a situation is shown in figure 10. The upper
graph shows the total liquid flow as function of time as given by the test separator. The
second graph from the top shows the oil and the third graph from the top shows the water
flowrate. By comparing the oil and water flowrates, it can be observed that there is a very
strong variation in the WLR since the oil and water flowrate are running out of phase. The
WLR is shown in the bottom graph. The vertical axis for the latter figure runs from 0 to
100%. This means that the WLR varies from less than 10% up to about 90%!
Figure 10: Illustration of the unsteady behavior of some wells. This holds for the flowrate as
well as the WLR. The upper graph shows the total liquid flowrate. The second and third graph
the oil and water flowrate respectively. It can be observed that the oil and water flowrates are
running out of phase. This leads to a significant variation in WLR (fourth graph). Due to
confidentiality reasons, the numbers on the vertical scale have been removed. The intercept of
the horizontal axis with the vertical axis corresponds to a flow of 0 m3/h for the upper three
graphs. The vertical scale for the lowest graph runs from 0 to 100%.
The strong WLR variation could be a result of the relative long distance between the well
head and the measurement location. Due to periods of low flowrate in combination with the
oil-water density differences, it is likely that gravimetric separation takes place. The resulting
stratification of the individual phase in combination with periodic slug flow could explain the
strong variation in WLR. As these fluctuating conditions occurred during the field tests, this
had to been taken into account. Therefore, during the field test a modification has been made
to the flowmeter to be able to measure the flowrates under these fluctuating conditions.
A comparison between the separator measurement and the meter reading is shown in figure
11. The upper graph shows the total liquid flow of the separator. The lower graph shows the
meter reading. The blue, red and green line is the liquid flowrate after applying 5 seconds, 1
and 5 minutes averaging respectively. It can be seen that applying averaging inside the
flowmeter, results in deviations between the measured flowrate and the actual flowrate.
Therefore, during the field test the measurement strategy of the flowmeter has been adapted,
and single shot data is used for flowrate determination, instead of using averaged data.
Figure 11: Comparison of the separator and the multiphase meter. The upper graph shows
the total liquid flow of the separator. The lower graph shows the meter reading. The blue, red
and green line is the liquid flowrate after applying 5 seconds, 1 and 5 minutes averaging
respectively.
The separator shows more pronounced liquid flowrate variations than the multiphase
flowmeter. This is most likely caused by the combination of the separator tank volume and
the active level control operation. When the liquid level is too low, the flow to the liquid flow
line is reduced or even stopped. And vice versa, when the liquid level is too high, the liquid
flowrate is increased. This leads to larger variations in the liquid flowrate.
Another observation that has been made during the field test is the enormous dynamic range
in flow velocity and volume flow that occur in practice. It is the task of the flowmeter to cover
the entire range in flow conditions.
A test that demonstrates the enormous variation in dynamic range in flow velocities is shown
in figure 12. In this graph the liquid hold-up as measured by the flowmeter is plotted versus
the measured liquid flow velocity. The liquid flow velocity has been scaled by the maximum
velocity and plotted along a log scale such that the reader is able to quantify the low flowrates
as well. The liquid flow velocity is varying over a range of 1:33. When the dimensionless
volume flowrates are derived from this graph (by multiplying the hold-up with the
dimensionless velocity) one can see that dynamic range is even significantly larger (larger
than 1:1000).
Figure 12: Liquid hold-up versus dimensionless flow velocity as measured by the flowmeter,
over a period of 18 hours. This figure illustrates the enormous dynamic range in flow
conditions. The velocity varies over a range of 1:33. The dimensionless volume flow variation
(hold-up times the dimensionless velocity V/Vmax) is larger than 1:1000.
The entire range of flow velocities must be covered by the multiphase flowmeter. Significant
errors in the end result are obtained if the low flow velocities are not measured correctly. Note
that one of the benefits of the magnetic resonance measurement principle is that it allows
measuring the large range in liquid velocity and liquid holdup that are encountered in the field
test.
An important lesson learned from the field test is that we have experienced that non-averaged
flow measurement data should be used in the calculation of the individual phase flowrates. By
doing so, the large variations in flowrates and composition are correctly taken into account
and prevent the introduction of non-linear effects which lead to additional errors.
6 SUMMARY AND CONCLUSIONS
The M-PHASE 5000, the magnetic resonance multiphase flowmeter, has been installed and
tested on an oil field, in which the flowmeter was connected to multiple wells via a test
header. The installation and start-up of the flowmeter went smoothly. The flowmeter
remained operational during the full 7 months of the field trial. The changing environmental
conditions and process conditions during the test period had no effect on the performance of
the flowmeter, which is in line with the results obtained earlier during environmental and
robustness tests in specialized laboratories.
For proper operation of the flowmeter, the correct magnetic resonance fluid properties of the
multiphase fluids need to be determined. The in-line fluid characterization procedure has been
tested and optimized (the characterization time has been reduced to well below one hour). It
has been shown that the in-line fluid characterization procedure is very convenient in practice;
the fluid properties can be determined inside the flowmeter, without the need of taking
samples. The accuracy of the in-line procedure is better than for the off-line fluid
characterization which is based on samples in the lab. This is related to the fact that for in-line
fluid characterization, the fluids are characterized under the actual line conditions, and there is
no need for conversion from lab conditions to line conditions.
The flowrates encountered in the field test showed large fluctuations. The timescales of the
fluctuations ranged from tens of seconds up to hours. As a result of these fluctuations, for one
of the tested wells the measured dynamic range for fluid velocity was 1:33 and for the
volumetric flowrate the dynamic range was over 1:1000. As the oil flowrate and water
flowrate varied out of phase, the WLR ranged for a single well between 10% and 90%. In
case averaging is applied to the raw measurement data, valuable information gets lost.
Therefore, during the field test the measurement strategy of the flowmeter has been adapted to
handle the raw measurement signal, without applying averaging. In this way the flowmeter is
able to manage the fluctuating flowrates, as well as the very large dynamic range encountered
in the field test.
7 OUTLOOK
Although in the current field test the field conditions were outside of the application range of
the flowmeter, a quantitative analysis is being performed. This work is still ongoing.
Furthermore, at present a second more extensive field test is being prepared at a different
location, in the Middle East. In this field test the conditions will be inside the application
range of the flowmeter, and the flowmeter will be tested for a large range of viscosities, WLR
and GVF.
Next to this, the first series of commercial flowmeters has been produced and has been
delivered to a customer.
8 ACKNOWLEDGEMENTS
The authors of this paper like to express their appreciation to the entire development team of
the magnetic resonance multiphase flowmeter at Krohne. Also the support of Piet Moeleker
and Danny Kromjongh from Shell Global Solutions International B.V. and the field staff from
the Dutch E&P company during the preparation, installation and operation of the flowmeter in
the field test is greatly acknowledged.
9 REFERENCES
[1] Jankees Hogendoorn, et al., Magnetic Resonance Technology, A New Concept for
Multiphase Flow Measurement, 31st International North Sea Flow Measurement
Workshop, Tønsberg, Norway, 22.-25. Oct., 2013.
[2] Jankees Hogendoorn, et al., Magnetic Resonance Multiphase Flowmeter: Gas Flow
Measurement Principle and Wide Range Testing Results, 32nd International North Sea
Flow Measurement Workshop, St. Andrews, Scotland, 21.-24. Oct., 2014.
[3] K.J. Dunn, D.J. Bergman and G.A. Latorraca, Nuclear Magnetic Resonance
Petrophysical and Logging Applications, Pergamon, 2002.
[4] J.H. Simpson and H.Y. Carr, Diffusion and Nuclear Spin Relaxation in Water, Phys.
Rev., Vol. 111, No.5, 1958.
[5] F. Franks (ed.), Water, a Comprehensive Treatise, The Physics and Physical
Chemistry of Water, Vol.1, p. 235, Plenum Press, 1972.
[6] B.B. Benson and D. Kraus, The concentration and isotopic fractionation of oxygen
dissolved in freshwater and seawater in equilibrium with the atmosphere, Limnol.
Oceanogr., 29(3), American Society of Limnology and Oceanography, 1984.
Operation and Maintenance of Multiphase Flow Meters (MPFMs) on BC-10 in
Deepwater Brazil
Neil Sleight, Shell Brazil; Lorraine Coelho, Shell Brazil; Casey Brister, OneSubsea;
Abstract
The Shell operated BC-10 field located in deepwater, 1650-1920m, offshore Brazil produces heavy oil in
the range of API 16-24. The fields are produced using an entirely subsea development concept tied back
to the host FPSO located approximately 8-10km from the producing fields. Given the high cost of subsea
wells and the associated pipelines and artificial lift pumps, surveillance and optimization of each well
becomes increasingly important in the operation of an efficient production system. In the case of BC-10,
the entire development concept was based around having accurate subsea metering as this allowed
significant design changes including:
 CAPEX reduction through the elimination of test separators and the use of fewer but larger
production risers,
 The elimination of well testing deferment caused by testing by difference,
 A robust flow assurance strategy as wells would not need to flow alone over long distances (up to
16km)
In addition to this, the ability to measure continuously and accurately has been a huge driver for
production optimization.
What makes the operation of MPFMs successful is continuous monitoring and tuning as project
assumptions are replaced by field data.
In this paper real world examples of operating and tuning MPFMs will be given including:
 Continuous monitoring of topsides measurements vs subsea measurements

 Setup of MPFM after first oil prior to samples being available
 Adjustments for changes in reservoir fluid properties
 Tuning when water salinity changes
 Setup for high watercut wells
 Validation of gas rates using GVF models and MPFM measurements
The goal is to share the challenges encountered operating subsea MPFMs over 5 years of production
together with solutions for maintaining accuracy in an environment where well testing was not feasible.
Whilst by no means exhaustive, it is hoped that some of the learnings here will be applicable to other
fields which have similar concerns.
2
Introduction
An MPFM is multiphase flow meter, capable of measuring at least 3 discrete phases: oil, gas and water.
In operation the advantageous of using MPFMs include:
1. Real time production data for reservoir and flow assurance management
2. Allocation of fiscal oil and gas back to individual wells without the need for well tests
3. Reduction in CAPEX as test separators and test flowlines are not required
4. Removal of deferment associated with by difference well testing
5. Field/well optimization on a daily basis
6. Real time rate data for leak detection models
Multiple technologies for multiphase metering are available, however the basic principle is to determine
the phase fractions present and then apply a venturi equation to convert the measurement data into phase
flow rates. The models required to do this must factor in the phase behaviour and phases densities
together with their impact on the primary measurements. The phase measurement typically is performed
by deploying a radioactive source together with a detector system which can differentiate the presence of
gas and liquid. The proportions of the liquid can then be determined either through conductivities or
through further radiation attenuation. The meters used in the BC-10 project are all One Subsea Vx type
52 meters and from this point on all references to MPFMs are references to this specific meter.
The BC-10 Field

BC-10 is an all subsea development located in deepwater Brazil. Partners in the development are Shell
(50%), ONGC (27%) and Qatar Petroleum International (23%). Production comes from 4 fields located
in water depths ranging from 1640-1920m, these are moderately heavy oil relatively low gas content –
see table 1. The subsea architecture which enabled the development of these separate reservoirs consists
of multiple drill centres coupled to production manifolds. Manifolds are routed to caisson ESPs
(Electrical Submersible Pumps) which provide the necessary energy to bring the fluids to surface. To
minimize equipment costs, each field has only two production flowlines routed to the host, one for
production and the second for hot oil displacement and/or production. In the case of Ostra, a third riser
for gas separated subsea is also present – figure 1. Such a design reduced the number of risers required -
an important consideration as on the host, a turret moored FPSO, turret routes are bulky and expensive. In
line with this philosophy, there are only 3 production trains on the host, one for each field with Abalone
being produced together with Ostra. In total the use of MPFMs allowed the removal of 3 additional risers
and the test separator, not to mention additional lower capacity ESPs which could deal with single well
flow rates. Further benefits come from integration of water flood management with continuous producer
well monitoring and its ability to observe if a particular injector – producer pair are in direct
communication. In Brazil the regulatory regime requires that individual wells be tested at least every 45-
90 days, which in a system with reduced CAPEX, would lead to additional deferment as testing by
difference would be required.
3
Field Phase Density (API) GOR (SCF/bbl) Oil Viscosity at MPFMs

[Sm3/Sm3] flowing installed /
wellhead Wells
conditions (cP)
Ostra 1 24 278 [49] 7-8 7 /7
B-West 1 16 200 [35] 90-100 0 / 2*
Abalone 1 42-45 3200-4000 1-2 1/1
O-North 2 16 350-325 [62-58] 50-70 7/7
*Future well tie-ins require MPFMs to be retrofitted.
Table 1. Oil properties for the various operating fields in the BC-10 development
Figure 1. Diagram showing the field layout: Gas lines – red, Oil/Water/Gas lines – Green and Water Injection -
Blue
4
History of Petroleum Legislation in Brazil

On August 6th 1997, the Brazilian government issued the Law 9.478, known as “Lei do Petróleo”, that
liberalised the petroleum industry in Brazil and created the “Agência Nacional do Petróleo (ANP –
Petroleum National Agency)” as the state department responsible to control, regulate and inspect the oil
and gas industry. In order to ensure oil production was being correctly measured, which has a direct
impact on royalties payment, ANP has issued on June 19th 2000, the “Portaria Conjunta N°1”, which was
the first Brazilian legislation on Oil and Gas Metering. Multiphase metering was not included as an
approved method for measurement, but the legislation indicates that other technologies could be used
following prior approval from ANP.
The BC-10 Allocation philosophy was based on well allocation using the MPFM results as it is
challenging to perform a good well test due to the oil properties and field hardware design. In this sense,
ANP approved the use of the MPFM for well allocation in Ostra and Abalone Fields, part of the BC10
Phase 1.
On 1st of June 2013, ANP issued a reviewed legislation for metering, “Resolução Conjunta ANP/Inmetro
n°1”, which included for the first time the multiphase metering. The legislation stated that the
“measurement systems that uses MPFMs should comply with the technical metrological requirements
established by Inmetro”. However, Inmetro has indicated that the regulation requires further metrological
and technological knowledge that is currently not available in the agency. In this case, MPFM
applications continues to be approved as deviations by ANP, which is the case of the approval for the use
of MPFM for well allocation in O-North, authorized in September 2013.
Principles of Operation
The Vx type MPFM works using a radioactive Barium
133 (Ba-133) source and a full spectrum gamma
detector. The Ba-133 source is located in the throat of a
metallic venturi meter and is installed behind a ceramic
window which allows the radiation to pass directly
though the fluid inside the venturi throat and into the
ceramic window protecting the gamma detector
opposite it. The venturi is instrumented with dual
differential pressure gauges (throat to inlet) and dual
line pressure and temperature sensors which measure
the conditions at the inlet. Critical to obtaining a
repeatable fluid inlet pattern, a blind tee is installed
upstream and conditions the fluid such that the feed to
the venturi always conforms to the assumptions
contained in the slip model. Figure 2 shows the general
assembly.
The raw gamma counts are grouped into three energy

peaks (32keV-Low, 81keV-High and 356keV) based on Figure 2. Overview of meter construction showing
the output of the gamma detector. the key sensors and blind tee
5
From these raw inputs, the flow computer calculates

the attenuation of the radiation coming from the
source. This requires an “empty pipe reference”, a
measurement point when the venturi throat was
known to be clean and air filled – the three gamma
peaks are recorded together with the date and time.
Such a profile is show in figure 3. A nuclear decay
model is then applied and a new empty pipe reference
created for the current moment in time. From this the
attenuation is calculated using the ratio of measured
counts to the calculated current empty pipe counts. Figure 3. Detector count of the Ba-133 source for an
empty pipe
Each peak, the Low (32keV), High (81keV) and 356
keV is thus converted into an attenuation.
Atomically different substances absorb varying
radiation levels differently according to inherent
properties and this is modeled by the attenuation
equation below:
𝐼 = 𝐼0 𝑒 −𝜇𝑑
Where: I – the gamma intensity, I0 – the initial gamma

intensity, µ - the attenuation coefficient for the
medium and d – length travelled through the medium.
Note that this equation needs a different mass
attenuation coefficient for each energy level and for
each phase. Mass attenuation is obtained from
attenuation thus:
Figure 4. Example solution triangle showing the 3
𝜇 pure phases and the operating point
𝜌
Higher radiation energy levels are far too energetic to react with anything less massive than the nucleus
of an atom. This makes the attenuation of high energy gamma rays dependent on the number of nuclei
between the source and the detector which of course is proportional to density, adjusted for composition
and pressure and temperature.
Lower energy gamma rays, specifically much lower than 300 keV, begin to interact more and more with
the electron clouds of the medium. There are two of these effects that dominate in the electron cloud
interaction, Compton Scattering and Photoelectric Absorption. Whichever is dominant will depend on
the energy level of the gamma rays and the specific medium. This secondary dependence on the medium
allows us to differentiate between mediums of similar densities such as oil and water.
The denser the substance, the more atoms there are present to interact with the radiation and hence the
higher the attenuation coefficient. Ions also increase the attenuation of a substance. Formation water
always contains some level of salt. Gas, with a density lower than either water or oil, is the easiest phase
to detect. Mass attenuation of a substance is a largely fixed property, assuming of that the composition
does not change (H2S, CO2 as these dissolve in oil and flash off changing the oil properties and salinity).
6
The specific mass attenuation of each phase can be measured using a pure sample of formation brine and
a separate sample of oil. This is carried out using a test cell consisting of a source and detector of the
same type as will be used in the field. The meter measures linear attenuation, the amount of peak
attenuation for a given throat size and as such it is necessary to convert mass attenuation into linear
attenuation by multiplying by the phase density. Here is where the configuration file in the flow computer
comes in; it provides a pressure and temperature dependent relationship for the gas, oil and water
densities allowing the line pressure and temperature measured to be used to calculate accurately the
density of each phase. With this the meter can now solve the two energy level attenuations to derive the
gas volume fraction and water – liquid ratio inside the venturi. This can be graphically represented in a
solution triangle, an example of which is shown in figure 4. By visual inspection it can be seen than each
pure phase has a unique linear attenuation of the high and low energy (LE and HE respectively) gamma
peaks. The gas fraction is the area between the operating point and the pure oil and pure water points,
divided by the total area of the triangle from gas to oil to water. In the same manner the ratio of oil to
liquids may be found and also converted to a volumetric fraction. By difference what is not gas or oil is
assumed to be water. Due to phenomena known as slip, the gas in the venturi travels much faster than the
liquid phases and consequently the gas fraction measured in the venturi needs to be corrected to account
for this effect. In the Vx meter this process is handled by a propriety slip law equation. At this point the
meter computer has the venturi throat phase fractions, this allows the density to be determined of the
multiphase mixture and then the standard venturi equation used to solve for the total volumetric flow rate
(where: q- volumetric flow rate, d – throat diameter, D – inlet diameter, Cd – discharge coefficient, ∆P –
throat to inlet differential pressure, ρ fluid density):
𝜋𝑑 2 2∆𝑃
𝑞 = 𝐶𝑑 √[ ]
4 𝑑4
𝜌(1 − 4 )
𝐷
The meter computer also adjusts the discharge coefficient based on the fluid fractions measured in the
meter which is used to calculate a mixture viscosity and Reynolds number used to determine the correct
coefficient. This is important as the basis for the venturi equation is zero viscosity (implying that all
differential pressure change is the result of acceleration, not due to friction on the venturi wall) and in
heavy oil or emulsion prone systems, this assumption is not correct and would otherwise compromise
meter accuracy.
From this point, the actual volumetric flows are known and it is required to convert them to standard
conditions. This is calculated using volume formation factors, as are commonly used in reservoir
engineering, where the actual volume is converted by the following equation:
𝑄(𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑) = 𝑄(𝑎𝑐𝑡𝑢𝑎𝑙) . 𝐵𝑥
Where Q is the flow rate in standard and actual (also called line) conditions and Bx is the formation
factor also called the shrinkage factor. As an example, a typical oil shrinkage factor could be in the range
0.7-0.95 depending on the temperature, pressure and GOR. The shrinkage factor is computed from the
line pressure and temperature using the information contained in the configuration file which has been set
up based on a PVT report for the specific crude being produced.
In addition to the measured phases, there are certain quantities which cannot be measured by the meter.
7
One example is the solution gas, since the meter typically operates at pressures well above standard
conditions, there will be gas which is dissolved in the oil and water phases which will flash off at stock
tank conditions but not in the meter. At high pressures the solution gas can easily be a factor higher than
the measured “free gas” and is accounted for using the PVT behaviour contained in the configuration file,
which calculates a solution GOR and applies this to the standard oil rate to determine the solution gas
rate. To this the gas dissolved in the water phase is also calculated and added to the free gas (at standard
conditions) to produce the final gas rate.
Metering Assurance
Having MPFMs installed and producing flow rate data, the question arises as to how to validate it. In
traditional oil and gas production facilities to measure the production of a well it would be removed from
the other production and routed to a test separator, which would have separation and metering of the
individual phases together with the option of taking samples to determine gas and oil density together
with BS&W to allow the readings to be corrected for incomplete separation. This option may be available
in some facilities which have MPFMs and allows periodic verification of the results. However MPFMs
add the most value in situations where such well tests are either difficult or impossible, for example in
complex subsea production systems which have long tie backs and where flow instabilities in the pipeline
make out flow unstable and hence testing by difference challenging. Gas condensate wells provide such a
challenge, as at low rates, pipeline liquid hold up can take days or even weeks to establish steady
conditions, leading to an underestimate of well liquid rates if testing by difference. For oil wells in deep
water, hydrate prevention by heat retention may make it risky to flow wells at low rates into a dedicated
test line for fear of forming hydrates or wax in the pipeline. One solution to this problem is the use of
reconciliation factors (RF) comparing the topsides metering with the sum of the MPFMs aligned to it as
below:
𝑆𝑢𝑟𝑓𝑎𝑐𝑒 𝑑𝑎𝑖𝑙𝑦 𝑡𝑜𝑡𝑎𝑙(𝑝ℎ𝑎𝑠𝑒)

𝑅𝐹(𝑝ℎ𝑎𝑠𝑒) =
∑𝑛𝑢𝑚𝑏𝑒𝑟
1
𝑜𝑓 𝑎𝑙𝑖𝑔𝑛𝑒𝑑 𝑚𝑒𝑡𝑒𝑟𝑠
𝑀𝑃𝐹𝑀 𝑑𝑎𝑖𝑙𝑦 𝑡𝑜𝑡𝑎𝑙(𝑝ℎ𝑎𝑠𝑒)
With this approach and careful surveillance a reconciliation factor of 1.0 indicates that the MPFMs are
reading correctly. When wells are brought offline, the RF should not change once the system returns to
steady state, otherwise this suggests that that meter was contributing more or less than is being measured.
Other methods such as applying a well bore model to the production and matching the measured pressure
data by adjusting the rates, GOR and BS&W until it matches can further confirm if the meter is likely
correct. In most instances an adjustment of GOR is not acceptable for non-gas lifted wells where the
GOR is fixed by operating the reservoir above bubble point.
A method which was successfully used on BC-10 was turret BS&W sampling, this was helpful in early
field life where the watercut of the combined fluid was very low <1% and as such was required to prove
the wells were dry. In-well water soluble tracers provide additional support to determine that no
formation water is being produced.
Finally step tests can be employed to try and detect that performance is correct over several well bean
ups. However this method is least reliable as it involves disturbing the steady state and if other wells are
of similar BS&Ws it may be hard to determine anything accurately.
8
The effects of Salinity on the Vx

The Vx was developed with the conscience
choice of Barium as its gamma ray source.
This was done for many reasons, not the least
of which was the ability to distinguish
between the three phases of Water, Oil, and
Gas, but also the relative insensitivity to the
salinity of the flow. The measurements
robustness to salinity makes it possible to
reliably use the Vx in any conditions with a
change of Salinity of around 5-6%.
In order to differentiate the fresh and briny

solutions, figure 5 displays the attenuation of
100% of each phase. If we assume a typical
WLR of no more than 50 to 60% as in a gas
well then the salinity change from 0 to Figure 5. Variation in water liquid ratio for various salinity
10% affects the WLR by less than 2%. changes.
Concurrently the oil and water flow rate
are equally minimally affected by the same
amount.
This strength can also be a weakness where salinity change is a concern for flow assurance and first water
production requires a sensitivity greater than 125ppm. It should be noted that it is possible to detect
salinity changes with the Barium system through the use of the third energy level, but this method is
limited to WLR’s > 30% and GVF’s over 70%.
9
Phase 2 Start-up Tuning

BC-10 Phase 2 (O-North) was started in Start up of Wells ON-9 and ON-6 using initial oil
September 2013 and it was agreed with ANP that attentuation data
each meter would be individually verified against 50
45
the topsides separator meters before the next well
MPFM measured BS&W (%)

40
could be opened, hence at any one time there was 35
never more than one well with an unproved 30
MPFM. As there was not sufficient O-North oil for 25 3.5% BSW
20 ON-6
a test cell to measure the mass attenuation of the 15 ON-9
oil the meters were started up with estimated 10
parameters for oil and an assumed formation water 5
0
salinity. Due to the design of the BC-10 system, 0 25 50 75 100 125
there is a minimum flow rate which can be Well hours since first start-up (hours)
produced without encountering slugging or Figure 6. Start up of two new wells with the same
running into viscosity and hydrate issues. This configuration files as the completition brine unloads
rate, around 15000 BPD, was well above the rates and the well becomes dry.
expected from a single well and as such the system
was started up by circulating gas spiked dead oil from the Ostra field. This allowed for the initial well
unloads and clean up but prevented the taking of a pure oil sample of O-North. Once four wells had been
brought online and all their completion brine had cleaned up, all their watercuts tended to a flat line in the
range 3-4%, the figure 6 shows this for 2 wells where the initial clean up can be clearly observed
followed by a steady state. Topsides samples were taken and this confirmed that there was <0.5%
BS&W meaning the MPFMs were not correct with respect to watercut and water rate. After examining
the meter solution triangle it was noted that the pure oil operating point was most likely incorrect, since
this had been estimated due to the lack of a suitable sample. Adjusting it for both LE and HE attenuation
resulted in a reduction of the BS&W from 3 to 4% to -0.25 to 0.5%. The solution triangles below
illustrate the changes – figure 7 showing the initial setup which had produced the 3.5% watercut and
figure 8 which shows the final tuned values matched for BS&W and GVF. As the position on the gas-oil
line also affects the GVF and hence free gas rate, the new oil point had to be adjusted such that the GVF
was in line with the PVT expectations. This required several iterations to achieve as only small changes
were necessary.
Initial Tuning: LE and HE Energy Levels Revised Tuning: LE and HE Energy Levels
0 0
-2 Linear Attenuation -2 Linear Attenuation
measured at Act measured at Act
-4 conditions -4 conditions
Linear Attenuation HE (1/m)
Linear Attenuation HE (1/m)
-6 Oil Water Gas -6 Oil Water Gas
-8 -8
-10 Oil -10 Oil
-12 -12
-14 Water -14 Water
-16 Oil -16 Oil
-18 Gas -18 Gas

Water Water
-20 -20
-60 -55 -50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 Average point -60 -55 -50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 Average point
Linear Attenuation LE (1/m) Linear Attenuation LE (1/m)
Figure 7. O-North initial MPFM configuration which Figure 8. O-North final MPFM configuration after
resulted in a meter BS&W of 3.5%. Surface samples minor tuning to reduce watercut to match the surface.
indicated no measurable water.
10
Phase 2 Comparison with Lab Data

Once a pure sample of O-North oil uncontaminated by the gas spiked Ostra oil was collected it was sent
for lab analysis to determine the density and the mass attenuation. However the lab results were
significantly different from the tuned values, which had given good agreement with the topsides
measurements of zero water and GOR. The table 2 below shows what happened after these lab results
were uploaded to the config file. Note that the comparison included the change in oil mass attenuation but
also the use of the original PVT which had a target GOR at reservoir initial conditions of 62 Sm3/Sm3
[350SCF/bbl] which had been proven to be higher than surface metering suggested.
Using the new lab values, the GOR dropped by about 10% despite using a higher solution GOR in the
configuration file – had this not been the case, the lower GVF of around 5% would have had an even
more dramatic impact on the overall GOR reported by the meter. After this testing period, the original file
was put back into service and metering reconciliation factors reverted to normal.
Test Period 18th July 2014 9th December 2014

Parameter Start up tuning with Lab data for oil Start up tuning Lab data for oil
GOR of attenuation, GOR 62 with GOR of attenuation,
58Sm3/Sm3 Sm3/Sm3 58Sm3/Sm3 GOR 62
File: 005 File: 006 File: 005 Sm3/Sm3
File: 006
Oil Rate (Sm3/h) 21.2 22.5 18.1 19.0
Water Rate (Sm3/h) 2.4 2.0 8.0 7.6
Gas Rate (Sm3/h) 1231 1167 1030 983
Free Gas Rate 439 247 382 262
(Sm3/h)
GVF (%) 9.2 4.8 7.9 5.0
GVF using PVT data 8.5 8.3
at line temperature
and pressure (%)
BS&W 10.2% 8.2% 30.7% 28.6
GOR (Sm3/Sm3) 58.1 51.9 56.9 51.7
[SCF/bbl]
GOR (Sm3/Sm3) 62.4 / 57.9 [350]/ [325]
[SCF/bbl] :
PVT Report / Surface
Rates
Table 2. Two test periods where the oil attenuation used was changed to that obtained from the lab analysis for
mass attenuation. Between tests several months elapsed as can been seen from the different watercuts.
The lab data config file reduced the GVF from 9.5% to 5%, indicating that the oil point had moved up,
towards the gas point. This was suspicious as it did not agree with the GVF which would be expected
based on PVT behaviour alone, which was 8.5% at line conditions using a GOR of 350SCF/bbl. This in
turn affected the GOR, since the free gas is a direct function of the measured GVF, reducing it below the
GOR measured at the surface and in the original PVT report range. Alone these results were enough to
give confidence that the lab data did not accurately represent the field operating conditions and the
“revised tuning” parameters were retained in all meters. One possible explanation, and one which was
11
not examined at the time, was that the slip law which converts the measured GVF into the reported GVF,
could be introducing some error into the calculation – for example if the viscosity model was wrong then
the slip between the phases might be effected. A second explanation could be errors in the oil density
measurement in the original PVT report; however no attempt was made to tune the density since there
was no subsequent PVT analysis to base changes on.
Ostra Gas Rate Offset

During the operation of Ostra, the
Ostra MPFM and Topsides Gas Before and After Water Density
reconciliation factor on gas had been gradually Update
rising until it was above the recommended 25000
limit (1.1). Investigation was focused on the

20000
water attenuation initially but in consultation
Gas Rate (Sm3/h)

with experts it was noticed that the water 15000
MPFM Total
density, which for Ostra should be around 10000
Sm3/h
Topsides Total
1110kg/m3, was in fact 990 kg/m3 at operating Sm3/h
5000 Update Period
conditions.
This may have been a remnant of the 0
0 2 4 6 8 10
commissioning phase, where condensed water Time (days)
was expected, but it was unsuitable for higher
watercuts and had reached the point where it Figure 9. Trend of surface gas and MPFM gas before,
was seriously affecting the gas reading. Once during and after the water density update.
this was realised, the configuration files were
updated to match the water density expected Ostra MPFM Oil and Water Before and After Water Density
Update
and the gas reading improved dramatically. 350
The entire update process of the 7 meters took 300
Oil or Water Rate (Sm3/h)
almost 2 days, owing in part to the technician

250
carrying out multiple tasks, but also due to the
200
low communications speed on the subsea MPFM Total
150 Water Sm3/h
communications architecture. Figure 9 shows MPFM Total Oil
100 Sm3/h
the metering of subsea gas and topsides gas for Update Period
50
10 days, before, during and after the update.
0
The impact to the oil and water rates was not 0 2 4 6 8 10
significant, figure 10 shows them before and Time (days)
after and only the total water shows a notable
Figue 10. Trend of the MPFM oil and water rates
reduction of 2-3%.
during the update process.
Phase 1 High Watercut Well

As the average BS&W of the Ostra field continued to rise as the field aged, some of the MPFMs started
to show suspicious results, especially a well (C7) which exhibited a BS&W of around 95-100%, with
some occasions that showing it exceeding even 100%. Initially scale or a weak nuclear source was
suspected, but after the correction for the incorrect salinity was made, as covered in the previous section,
the offset became far worse – suggesting there was something related to the attenuation. As a cross check
for the possibility of scale, the meter was filled with methanol and the nuclear counts compared to what
would be expected for the scale free signal, this was unfortunately inconclusive. Figures 11-13 show 3
different well start-ups on C7 from a 1 year period with different configuration files, the changes are
summarised in table 3.
12
Figure 11. Figure 12. Figure 13.

The start up data from well C7 showing the gross flow rate and BS&W. Note that the time when
methanol injection is stopped is also marked since this introduces an error on the meter since methanol
appears like a light hydrocarbon plus water and hence gives a false reading. All well starts have a meter
pressure higher than the bubble point of the oil so no free gas is present.
Figure 11 12 13
Change Base file, water density Water density corrected (using Mass attenuation for
incorrect (using fresh water formation water density and water adjusted to give
density but actual salinity for actual salinity for attenuation) feasible BS&W
attenuation)
Table 3. Meter configurations for the corresponding figures above.
Adjusting the water density, which was around

995 kg/m3 when using the fresh water setup
(figure 11) and 1110 kg/m3 when set up for the
salinity measured at the surface (figure 12), made
a huge impact. This is to be expected if the
solution triangle is considered. At these higher
watercuts, the meter is operating extremely close
to the pure water point, so moving this even
slightly will have a dramatic impact - as can be
seen by the BS&W exceeding 100% (figures 11
and 12), which requires a negative oil flow rate,
results from the meter having an operating point
beyond the water point in the solution triangle –
see figure 14. Tuning to get good operation at
this set of conditions is not easy, but is necessary
as the lab attenuations did not give physically
meaningful results and other likely explanations
Figure 14. Solution triangles for the figures 11-13
had been exhausted. Once completed, the new
together with the operating point.
tuning parameters were applied to all of the Ostra
field MPFMs, since they all produce the same oil
and same water salinity, what works for one meter must work for the rest if the assumption that the meter
is not at fault. This also allows the reconciliation factor to be checked to confirm that the change was
correct overall or not.
13
Ostra Gas Oil Ratio

During an investigation into GOR present on some Ostra wells, specifically ones which had very high
watercuts, the GOR was noted to be unexpectedly high in the range 80-100 Sm3/Sm3 compared to the 49
Sm3/Sm3 of the initial PVT report. Due to gas injection into the reservoir via the gas disposal well,
certain wells did already have a higher GOR, but they also exhibited a higher GVF. These other wells did
not. The answer lay inside the configuration file, which had a water gas ratio (WGR) of around 2.2
Sm3/Sm3 at operating conditions, corresponding to the gas saturation of fresh water. The true saturation
of gas in the Ostra formation water was 0.8Sm3/Sm3 due to the salt making dissolution of gas more
difficult. At the high water cuts on the well, for example at 95%, and at high pressure so with no free gas,
the total gas rate would be:
Oil Rate x Solution GOR + Water x WGR = 5 x 48.8 + 95 x 2.2 = 453 Sm3 / 100Sm3 liquid
(GOR = 453 / 5 = 90.6 Sm3/Sm3)
Compared to the correct WGR of 0.8Sm3/Sm3 giving:
5 x 48.8 + 95 x 0.8 = 320 Sm3 / 100Sm3 liquid

(GOR =320 / 5 = 64 Sm3/Sm3)
Thus the discrepancy was due only to the added gas from the water by the meters’ computer. This
highlights the importance of understanding the contents of the configuration files, as there can be some
confusion around a number such as this – how can the GOR exceed that of the initial reservoir, and for
modelling purposes, should a GOR of the pure oil or the combined fluid be used? Normally, the gas held
in the water would not be considered for metering purposes, as it represents a very small stream when at
surface processing conditions and indeed may not even be directly measured – gas leaving the water
treatment plant for example may be mixed with nitrogen complicating metering (if it is metered at all). So
at a high level a choice must be made as to whether pursue as high a fidelity model as possible, and hence
include these effects, or to make the model simple to allow easier detection of anomalies. It also raised
the question of whether the entire aquifer is saturated in gas, and if so what was the mechanism and will
this change as new water in-flows.
Data Verification Tools

With the MPFM running and producing data, this can be validated against the expected behaviour. One
measure which has proven to be useful is GVF, since this is a physical property of the oil and should only
vary with pressure, temperature, watercut and is a good reference point for the meter operation. For
example, a meter operating at a pressure above the bubble point should not read any GVF, since there
should not be any gas present, conversely, when the meter is operated inside the 2 phase region, gas must
be present and any absence indicates that something is not correct. A more sophisticated approach is to
compare the expected GVF with that measured. This type of surveillance is similar to checking GOR, but
has the added benefit that it is less dependent on the PVT information in the configuration file. For the
Ostra and O-North meters, online calculations were created which allow the theoretical PVT GVF to be
calculated and compared to the output of the meter. Two well start ups are shown in figures 15 and 16,
indicating that after the initial gas cap is blown off the well, the meter GVF and theoretical GVF match
well. Start-up is a good time to check, as the meter will pass through a wider range of pressures and
temperatures as the flow rate is increased – allowing a wider operating range to be validated. Some wells
in the field are not suitable for this treatment, as due to gas injection they have a variable GOR and hence
14
the GVF cannot be calculated without already assuming that the meter gas and oil are correct. This would
also apply in the case of a gas lifted well, however BC-10 do not use gas lift so this technique is helpful.
Figure 15. Start up of Ostra well C-1 showing the Figure 16. Start up of O-North well ON-10
metered GVF compared to the GVF calculated showing the metered GVF compared to the GVF
using PVT software. calculated using PVT software.
Conclusions
Several examples of troubleshooting subsea MPFMs have been shared here. The central learning is that
MPFMs, whilst still a flow meter, require a different approach which is much more cross disciplinary
than a more conventional meter. They require a deep understanding of the physical behaviour of the oil,
gas and water being produced as the conditions are radically different from those in a surface facility.
This information, traditionally obtained and owned by the reservoir engineers and flow assurance
engineers must be shared and understood in order to correctly setup the meter and to understand possible
reasons when results may appear to be drifting as the field ages and conditions change. Challenges exist
in the future for BC-10 MPFMs, not least the addition of 3 new fields which will bring new PVT data and
the continued use of water flood, which will impact the salinity of the produced water on a per well basis,
making recalibration necessary more frequently. The BC-10 project has proven the value of continuous
well monitoring, for production optimisation in addition to allocation using subsea MPFMs.
Acknowledgements
The author wishes to thank the partners of the BC-10 Project; Shell Petroleo Brasil, Qatar Petroleum
International and ONGC of India for allowing the publication of this paper.
NSFMW 2015 Tønsberg, Norway, October 20 to 23, 2015
Case Study Results on Natural Gas Custody Transfer

Measurement with Coriolis Meters in Saudi Arabia
Mohammed Al-Torairi: Saudi Aramco

mohammed.torairi@aramco.com
WC-1337, Al-Midra Tower
Dhahran 31311, Saudi Arabia
Chandramohan MC: Emerson Process Management

Chandramohan.MC@emerson.com
P.O. Box 17033Jebel Ali Free ZoneFZE JAFZA South zone II
Dubai 17033, United Arab Emirates
Marc Buttler: Emerson Process Management – Micro Motion

marc.buttler@emerson.com
7070 Winchester Circle
Boulder, CO, USA 80301
Introduction
Saudi Aramco has conducted a long-term trial of two Coriolis flow meters in an industry gate
natural gas custody transfer application. The long-term measurement results of the Coriolis
duty meter are compared to both a second reference Coriolis meter and an orifice meter.
The trial was conducted employing elements of the of American Gas Association Report No. 11
(AGA 11) Measurement of Natural Gas by Coriolis Meter - Second Edition that was published in
February of 2013 (a.k.a, American Petroleum Institute Manual of Petroleum Measurement
Standards Chapter 14.9 (API MPMS 14.9)). The Second Edition of AGA 11 includes many
improvements and additions to the First Edition. One of the new additions describes how
manufacturers can demonstrate that a meter designed for natural gas service has the flexibility
to be calibrated with fluids other than natural gas (e.g., water). Another significant addition to
AGA 11 describes how it is possible to perform a secondary verification check on a Coriolis
meter in service without interrupting gas flow measurement.
Description of the Trial Application and Installation
Two Micro Motion CMF200 Coriolis meters were mounted together on the skid to allow gas to
flow through both meters in series under normal operating conditions. Valves were installed
and piping arranged as shown in Figure 1 to allow for either meter to be bypassed at any time.
Additionally, two block valves were located on either side of each meter so that either meter
could be blocked in on both the upstream and downstream side. High quality block valves were
selected to ensure that no leakage through any of the valves would occur to influence the test
results. The flow indication from the two meters was taken to a control room where an Emerson
FB 107 flow computer calculated and recorded the flow totals from the meters on a daily basis.
1
Fig.1. Diagram of the Trial Skid Piping Arrangement
With this arrangement it was possible to check the zero of either meter regularly without the
need to interrupt the flow through the skid. It was also possible to remove either meter from
service temporarily to allow the meters to be transported and tested in a calibration laboratory
setting at another location without interrupting the functioning of the other meter during the
course of the trial period.
The site of the test was a glass factory located in Dammam, Saudi Arabia. The skid shown in
Figure 2 was delivered and installed in series with an existing orifice meter run at the point
where natural gas is sold from a pipeline into the factory.
Fig.2. Trial skid with Two Micro Motion CMF200 Meters
2
The skid was designed to meet the following requirements:
• Gas flow rate: 700 - 1000 Nm3/hr (0.6 - 0.85 MMSCFD)

• Mass flow rate at operating conditions: 560 - 800 kg/hr
• Operating pressure: 250 - 300 psig
• Gas temperature: 60 - 100 °C (140 - 212 °F)
• Pressure loss across each meter: 0.35 psi
• Maximum velocity passing through the meters: 18 m/s (57 fps or 0.04 Mach)
The trial period began May 25, 2011 and was successfully concluded on July 14, 2012 after a
total duration of approximately 14 months. Toward end of trial period, the flow rates were
allowed to increase to as high as 1180 Nm3/hr (1.0 MMSCFD) with no appreciable adverse
change observed in the results.
Determining Energy and Standard Volume Units from Coriolis Mass Flow Measurements
The standard practices for trading in natural gas are based on units of standard volume and
energy. Standard volume is the volume of a fixed quantity of gas when it is at mutually agreed
upon reference conditions of fixed pressure and temperature. It is necessary to work in units of
standard volume instead of actual volume because the actual volume of a fixed amount of gas
will change dramatically as actual pressure and temperature are changing.
To determine flow in energy units, a gas chromatograph or other method is needed to establish
the heating value of the natural gas, which will vary with changing composition. The heating
value can either be represented as energy units per unit of standard volume, such as BTU per
standard cubic foot, or as energy units per unit of mass, such as MJ per kg.
The simplest and most accurate path to energy units is to multiply the mass by the heating
value, expressed in energy units per unit of mass to arrive at units of energy (or energy flow).
𝐸𝑛𝑒𝑟𝑔𝑦 (𝑀𝐽) = 𝑀𝑎𝑠𝑠 (𝑘𝑔) × 𝐻𝑐 (𝑀𝐽⁄𝑘𝑔)
OR
𝐸𝑛𝑒𝑟𝑔𝑦 𝐹𝑙𝑜𝑤 (𝑀𝐽⁄𝑑𝑎𝑦) = 𝑀𝑎𝑠𝑠 𝐹𝑙𝑜𝑤 (𝑘𝑔/𝑑𝑎𝑦) × 𝐻𝑐 (𝑀𝐽⁄𝑘𝑔)
Where:
Mass = Mass of gas (Coriolis flow meter measurement)
𝐻𝑐 = Heating Value (as determined by gas analysis)
3
If, instead, heating values are defined in units of energy per unit of standard volume, it is easy to
make the conversion from mass to standard volume before multiplying by the heating value.
𝑀𝑎𝑠𝑠 (𝑙𝑏𝑚)
𝑆𝑇𝐷 𝑉𝑂𝐿 (𝑆𝐶𝐹) = 3 (AGA 11 Equation D.2) 1
𝜌𝑏 (𝑙𝑏𝑚⁄𝑓𝑡 )
Where:
𝑆𝑇𝐷 𝑉𝑂𝐿 = Volume at reference temperature (Tb) and reference pressure (Pb)
Mass = Mass of gas (Coriolis flow meter measurement)
ρb = Base density of gas at reference temperature (Tb) and reference pressure (Pb)
𝑃𝑏 ×𝑀𝑟 (𝑔𝑎𝑠)
𝜌𝑏 = (AGA 8 Equation 7) 2
𝑍𝑏 ×𝑅×𝑇𝑏
Pb = Pressure at base conditions
Mr = Gas Molar Weight
Zb = Base compressibility at reference temperature (Tb) and reference pressure (Pb)
R = Universal gas constant
Tb = Temperature at base conditions
The relationship between energy units and mass total or mass flow is also, therefore,
straightforward as shown in the following equations.
𝑀𝑎𝑠𝑠 (𝑙𝑏𝑚)
𝐸𝑛𝑒𝑟𝑔𝑦 (𝐵𝑇𝑈) = 𝑆𝑇𝐷 𝑉𝑂𝐿 (𝑆𝐶𝐹) × 𝐻𝑐 (𝐵𝑇𝑈⁄𝑆𝐶𝐹 ) = 3 × 𝐻𝑐 (𝐵𝑇𝑈⁄𝑆𝐶𝐹 )
OR
𝑀𝑎𝑠𝑠 𝐹𝑙𝑜𝑤(𝑙𝑏𝑚/𝑑𝑎𝑦)
𝐸𝑛𝑒𝑟𝑔𝑦 𝐹𝑙𝑜𝑤 (𝐵𝑇𝑈/𝑑𝑎𝑦) = 𝑆𝑇𝐷 𝑉𝑂𝐿 (𝑆𝐶𝐹𝐷) × 𝐻𝑐 (𝐵𝑇𝑈 ⁄𝑆𝐶𝐹 ) = 3 × 𝐻𝑐 (𝐵𝑇𝑈 ⁄𝑆𝐶𝐹 )
Note that base density is a fixed constant that will not change unless the composition of the gas
changes. Actual density will change, just as actual volume will change, when the actual
pressure and temperature are changing. In contrast, base density will not change due to
changes in actual pressure and temperature.
4
When the flow is measured directly by mass with a Coriolis meter it is still necessary to know
the composition in order to convert to either energy or standard volume, but it is never
necessary to know the actual pressure or temperature to convert units to base conditions
because the mass of a fixed quantity or amount of gas will never change, even when the actual
pressure and temperature are changing. For this reason, the trial skid did not include any
pressure or temperature transmitters.
The initial gas molar weight in the trial installation was 16.42. The base density, ρb, was 0.78
kg/m3 (0.044 lb/ft3) at 14.7 psia and 15.5 °C. Therefore, the daily volume flow rate could be
recorded in units of MSCFD in the FB107 by simply dividing the daily mass flow reading in
kg/day by the base density, ρb (0.78 kg/m3), and then multiplying by the conversion factor
0.035315 MSCF/m3. As the trial progressed, the gas composition was sampled and updated in
the flow computer on a weekly basis to calculate base density and standard volume (using the
detailed method from AGA 8).
Measurement Traceability
Both the test (duty) and the reference Coriolis meters were installed and placed into service with
the original factory water-based flow calibrations left intact.
AGA 11 states in the beginning of Section 7 Gas Flow Calibration Requirements that it may
be valid to use an alternative calibration fluid, such as water, to calibrate Coriolis meters for gas
measurement, so long as transferability of the calibration from the alternative fluid to gas has
been demonstrated by the meter manufacturer through tests conducted by an independent flow
calibration laboratory. Transferability of the calibration from an alternative fluid will include an
added uncertainty relative to gas measurement that must be quantified by the manufacturer and
verified by an independent flow calibration laboratory or laboratories. Coriolis meter designs
that have not yet demonstrated calibration fluid flexibility are required to be flow calibrated on
gas as prescribed in Section 7.1 of AGA 11.
Emerson has verified transferability of water calibration to gas flow measurement for the Micro
Motion® ELITE® CMF series of flow meters through testing at multiple independent flow
calibration laboratories. No linearization or adjustment was applied during this series of tests
after the original factory calibration on water was performed other than compensation for the
known effect of pressure on the meter that was applied. The maximum difference observed
during testing between the original water calibration and the tests on natural gas and nitrogen
was ±0.5% 3. One of the Coriolis meters that were tested to verify the transferability of water
calibrations is shown in Figure 3 as it was installed in an independent gas calibration laboratory
during the testing.
5
Fig.3. Coriolis Meter Installed for Independent Gas Laboratory Testing 4
The manufacturer’s published gas measurement accuracy specification for the two Micro Motion
meters used in the trial is ±0.35% of rate. This accuracy specification for gas measurement
applies to the meters as they leave the factory calibration lab, after the standard flow calibration
procedure that uses water as the fluid.
The measurement of the test (duty) Coriolis meter was confirmed on a daily basis during the
trial by comparing the reading with that of the reference Coriolis meter. The two meters were
required to agree by ±0.7% or better at all times or the test (duty) meter reading would not be
used for custody transfer and the sale point would revert back to the orifice meter, which was
left active.
The traceability of each of the Coriolis meters was also independently confirmed well into the
trial period. The ability to remove each meter during operation was an important element of the
design of the skid because the trial plan included provisions for testing each of the meters, one
in a liquid calibration laboratory and one in a gas calibration laboratory, but only after both
meters had already been in service at the trial site for some time.
After approximately 8 months in service, the test (duty) meter was removed from the skid and
sent to the Emerson calibration laboratory in Abu Dhabi for a scheduled calibration test on
water. The laboratory is CMC Certified by VSL. The original test plan called for agreement
between the test (duty) meter and the water calibration standard of ±0.1% or better to pass.
The meter passed this test and was returned to the skid and placed back into service the
following month. During this time period, the reference meter acted as the duty meter.
6
As soon as the test (duty) meter had been returned into service, the plan then called for the
reference Coriolis meter to be removed and sent to a third-party gas calibration laboratory to be
checked against reference standards on natural gas flow in a controlled environment. The
Southwest Research Institute (SwRI) natural gas laboratory in San Antonio, TX, USA was
selected for this test and the plan called for agreement between the meter and the reference
standards on natural gas of ±0.7% or better to pass. The data in Figure 4 from the test at SwRI
shows that the meter agreed with the reference standards to within ±0.25%. The meter was
returned to the skid and placed back into service as the reference meter on April 27, 2012.
+/- 0.7% minimum

performance
requirement as per
AGA 11
Fig.4. Reference Coriolis Meter Test Results at the Independent Gas Laboratory (SwRI)
Ongoing In-Situ Meter Verification
One of the most sought-after benefits of Coriolis meters is that the calibration remains very
constant over time if nothing occurs to damage the meter structurally. This is why diagnostic
tools that accurately monitor the structural health of a Coriolis meter are useful as a tool for
secondary verification of the meter calibration that can be relied on, even after lengthy periods
of time have passed since the most recent calibration against a primary or secondary flow
reference standard.
Micro Motion ELITE CMF meters can be equipped with an optional diagnostic feature called
Smart Meter Verification (SMV) that uses a sophisticated analysis the Coriolis meter flow tube
vibration response characteristics to assess and trend the structural consistency of the meter
flow tubes. If the structure of the meter is found by the SMV tool to be consistent over time, this
result indicates that the meter’s flow calibration has remained unchanged.
7
A Coriolis meter may be installed directly into service for natural gas custody transfer after the
factory calibration on water, or it may be sent to an independent gas calibration laboratory for
calibration. In either case, the value of the investment in equipment and calibration can be
prolonged and preserved by using an automated secondary verification diagnostic tool like the
Micro Motion SMV feature.
During the course of the trial, the SMV diagnostic test was used on a weekly basis as further
confirmation of the integrity of the measurements. The SMV feature tested both the flow meter
electronics (transmitter) and the sensor. In particular, the SMV test measured and recorded the
stiffness of the flow tubes, which is the secondary attribute that is used to verify the structural
integrity and calibration stability of the meter. The full history of the structural integrity flow tube
stiffness tests for both of the Coriolis meters used in the trial is shown in Figure 5.
Fig.5. SMV Meter Structural Integrity Test Results History of the Two Coriolis Meters
The SMV test results showed that the both the meters were undamaged and that both of their
original calibrations were intact. Therefore, it was no surprise when the test meter passed the
liquid calibration test later in Abu Dhabi and the reference meter passed the gas calibration test
at SwRI.
In addition to checking the calibration and health of the meters using the SMV test, the zero of
each meter was checked on a weekly basis. To perform this check, one of the meters was shut
in by closing both the upstream and downstream valves on either side of that meter and then
confirming that the meter indicated zero flow properly. Then the flow through that meter was
resumed while the same check was performed on the other meter. Both meters consistently
passed this check.
Previous experiences in this application had identified a risk of coating by a black powder
contamination. The Zero Check and SMV diagnostic test were used to protect against any
undetected build-up of the black powder coating in the meter. No significant coating was
detected while the meters were in service and this was verified by inspection with a borescope
when the test meter arrived for calibration in water flow Lab in Abu Dhabi.
8
Summary Observations of the Trial
Confidence in the measurements made by both of the Coriolis meters was high. The weekly
SMV and zero checks confirmed that the meters had not suffered any damage and that the
calibrations of both meters were remaining stable over time as expected.
The planned removal of each of the Coriolis meters so that they could be sent to a laboratory
offsite and tested against reference standards strengthened the confidence in the
measurements even more. By first verifying that the test (duty) meter was still within ±0.1% of
the Emerson lab when tested on water after 8 months in natural gas service, and then verifying
that the reference meter was within ±0.25% when tested on natural gas at SwRI after 9 months
in service, it was concluded that the meter calibrations had remained stable during the entire
course of the trial.
Confidence in the calibration stability of the meters was also bolstered by the consistent
agreement between the two meters to within ±0.7% during the entire trial period of 14 months.
Figure 6 shows this agreement between the two Coriolis meters in terms of daily captured flow
rate averages. Note that the black line of the reference Coriolis meter is almost always
completely obscured by the grey line of the test (duty) Coriolis meter. The only time the black
line of the reference Coriolis meter is fully visible is during the one month period when the test
meter was sent to Abu Dhabi for testing in the Emerson lab. During that time, the grey line was
absent since the meter was not present in the skid.
Fig.6. Daily Average Flow Rates of the Two Coriolis Meters and the Orifice Meter
9
It is easier to confirm the agreement between the two Coriolis meters in the plot of the deviation
between the two meters shown in Figure 7. The green line, plotted as Deviation 2, represents
the % difference between the two Coriolis meters and shows how well they agreed to within
±0.7% during the length of the trial period. A gap exists in the plot of Deviation 2 because
during the time period where the gap appears is the time when either one or the other of the
Coriolis meters was away at one of the offsite labs.
Fig.7. Deviations between the Two Coriolis Meters and the Orifice Meter
It can be seen in both Figures 6 and 7 that the orifice meter reading agreed reasonably well with
the Coriolis meters up until roughly November or December of 2011. The red line, plotted as
Deviation 1, indicates the difference between the test (duty) Coriolis meter and the orifice meter.
It is plain to see that the orifice meter began to read higher and higher relative to both of the
Coriolis meters from November 2011 until the end of the trial data collection in July of 2012.
The orifice meter was over registering relative to the Coriolis meters and it eventually reached a
point toward the end of the trial where it indicated higher than the Coriolis meters by roughly
10% of rate. The root cause of the orifice drift relative to the Coriolis meters has not yet been
identified.
10
Conclusions
The trial has captured data demonstrating the long term performance of the test (duty) Coriolis
meter. The results demonstrated the successful application of both calibration fluid flexibility
with water and secondary meter verification diagnostics as described in the second edition of
AGA 11.
The case study concluded that the Coriolis meters that were used are well suited for custody
transfer measurement of natural gas. Furthermore, the Coriolis meters provided substantial
benefits:
• improved system accuracy as a result of the long-term calibration stability

• quicker start-up straight from the factory due to the calibration fluid flexibility
• extended confidence in the measurements due to the secondary verification diagnostics
• confirmed water calibration for ongoing calibration tests with the results being
transferable to gas flow measurement
1
AGA Report No. 11, API MPMS Chapter 14.9, Measurement of Natural Gas by Coriolis Meter,
American Gas Association, 400 N. Capitol Street, N.W., 4th Floor, Washington, DC 20001
2
AGA Report No. 8, Compressibility Factor of Natural Gas and Related Hydrocarbon Gases,
American Gas Association, 400 N. Capitol Street, N.W., 4th Floor, Washington, DC 20001
3
Test Report Number NMi-12200340-02, Project Number 12200340, NMi Certin B.V., Hugo de
Grootplein 1, 3314 EG Dordrecht, The Netherlands
4
Wyatt, T., Stappert, K., Large Coriolis Meters and the Applicability of Water Calibrations
for Gas Service
11
GAS MASS FLOW METERING
AN ALTERNATIVE APPROACH
Eric Sanford, Vortek Instruments
Koichi Igarashi, Azbil North America
Kim Lewis, DP Diagnostics
1 INTRODUCTION
Metering a fluid’s mass flow rate in a pipe with methods that do not require an external
fluid density prediction is an attractive option in many flow meter applications. Such
meter designs tend to be described as ‘mass flow meters’. The development of a simple,
robust and compact gas mass flow meter concept is described here.
Gas flow must ultimately be metered by mass flow. A steady gas flow throughout a
pipeline has a constant gas mass flow rate but a varying volume flow rate. A gas mass
flow rate reading is an absolute statement of the gas quantity flowing in the system.
However, a gas volume flow rate reading is only true at the meter location. Flow meters
that read a volume flow rate prediction require an independent gas density prediction at
the meter location in order to give the required mass flow rate. The Achilles heel of most
gas meter designs is that they are wholly dependent on the independent gas density
prediction being available (and trustworthy). Therefore, for some meter applications it
can be preferable to meter mass flow using physical principles that that do not require the
independent measurement of fluid density.
Although sonic nozzles and thermal mass meters are good mass flow meters for select
niche applications, the Coriolis meter is widely considered to be the only practical low
uncertainty general use industrial mass flow meter available. It is widely assumed that
there is no viable alternative, i.e. no competing general purpose mass flow meter
technology. However, an alternative gas mass flow meter design has been known for at
least 50 years. In 1956 Boden [1] patented the concept of cross referencing the outputs of
a density sensitive and density insensitive flow meters. This combination allows the
prediction of the fluid density, volume flow rate & mass flow rate without any fluid
density information being required from an external source. The two meters could be
placed in series or a hybrid meter design that blends the two separate technologies into
one meter body could be considered. There have been multiple improvements and
independent “re-inventions” of this concept, and yet the concept remains obscure and an
academic curiosity. There appears to be three main reasons for this:
• It was many years after Boden’s initial invention before computers were available
to make such a system practically and economically viable,
• Two meters in series can be perceived as a heavy & expensive “contraption”
meaning that a practical design needs to be a compact hybrid design,
• There is a practical design complication that becomes evident when developing a
compact hybrid design.
In this paper the Boden concept is described along with the practical design problem
when developing such a mass meter. A new hybrid design consisting of a DP meter & a
1
vortex meter will be introduced that has been developed to circumvent this design
problem. Data from test meters and the first commercial meter will be shown. The DP
meter is fully equipped with the modern DP meter diagnostic system ‘Prognosis’.
2 THE HISTORY OF THE BODEN MASS METER CONCEPT
In 1956 Boden [1], working in the aeronautical industry, stated that cross referencing
density sensitive and density insensitive meters in series produces a density prediction
along with a volume and mass flow rate predictions. Boden placed a turbine (density
insensitive) meter in the throat of a Venturi (density sensitive) meter to produce a mass
meter (see Figure 1).
Fig 1. Boden et al, Turbine + Venturi Meter Combination 1956

In 1967 Pfrehm [2], an Esso engineer, modified & improved this design. Considering the
adverse effects of a turbine meter in a Venturi meter throat too excessive on the Venturi
meter performance Pfrehm suggested that the DP meter would operate better if the low
pressure port was located away from the turbine meter. Pfrehm, suggested a heavily
modified Venturi tube. The Pfrehm design is shown in Figure 2. However, the Pfrehm
design still produces a highly unorthodox Venturi meter with questionable performance.
Fig 2. Pfrehm, Turbine + Modified Venturi Meter Combination 1967
2
Fig 3. Lisi’s Mass Meter Arrangement Fig 4. Mottram’s Mass Meter Arrangement
In 1974 Lisi [3], another Esso engineer, considered a different combination of density
sensitive & density insensitive flow meters. Lisi suggested that as a vortex meter has no
moving parts it may be a better choice than a turbine meter to put in series with a DP
meter. Lisi considered a vortex (density insensitive) meter and an orifice (density
sensitive) meter in series. Lisi then dispensed of the independent orifice (DP) meter by
reading the DP created across the vortex meter bluff body, i.e. using the vortex bluff body
as a DP meter primary element (see Fig 3).
In the early 1970’s Klaus Zanker of Kent Industries in the UK also investigated this
generic concept. At that time Kent Industries judged the concept not commercially viable
due to these paired flow meters having distinctly different performances. This R&D
project would also have been hindered by early flow computer capabilities and the then
lesser commercial demand for gas mass metering.
In 1986 Mottram [4] patented another density insensitive and density sensitive flow meter
combination. Mottram placed a vortex (density insensitive) meter in an extended throat of
a nozzle (density sensitive) DP meter (see Fig 4). Vortex meters operate at peak
performance at moderate to high flow velocities. Hence, Mottram placed the vortex meter
in the DP meter throat to increase the fluid velocity at the vortex meter and achieve
enhanced vortex meter performance. Mottram’s list of prior art did not mention any of
these earlier developments.
By 2005 no such device had been marketed. Knowledge of this concept was not even
known to many who would have reasonably considered themselves ‘skilled in the art’ of
flow meter design. As such, in 2005 Dimarco S. et al [5] of Rosemount Inc (owned by
Emerson Process) submitted a patent application on this very same generic idea.
Rosemount did not state Boden [1], Pfrehm [2], Lisi [3] or Mottram [4] as prior art. Like
Mottram, Rosemount Inc / Emerson Process were unaware that there were reinventing a
concept that had been reinvented and modified several times over several decades.
In 2006 Steven [6], while being just as uninformed as Mottram & Dimarco et al also
submitted a patent on this generic idea. This patent application listed various different
potential designs where various meters were combined to produce hybrid mass meter
designs. These suggested designs inadvertently including the prior art of Pfrehm & Lisi
designs. However, other designs were also suggested. These included placing an
3
ultrasonic volume flow meter in the throat of a Venturi or nozzle meter, or combining
vortex & cone DP meters by replacing the cone element support with a vortex meter bluff
body.
If knowledge of the prior art had been more widespread, then Mottram [4], Dimarco [5],
& Steven [6] would have known the generic concept was prior art, the various patents
had expired and the generic methodology was now obscure but free public knowledge.
However, there are obvious under lying questions. There is an industrial requirement for
mass flow metering, and this concept has been continually reinvented as a good idea. So
why has this concept repeatedly failed to be developed into a product? Why is this
concept not better known amongst flow meter engineers who were skilled in the art?
One issue is that when Boden, Pfrehm & Lisi were researching the concept the computer
power was not easily available to cross reference the two meter outputs. It was a good
theoretical idea but difficult to implement in practice. Once, the computer power was
readily available as of the 1980’s practical problems appeared when attempting to
develop a commercial product.
The primary idea by Boden [1] is to put separate meters in series. This was also discussed
by Lisi [3], Damarco [5] & Steven [6]. Few operators wish to purchase two separate flow
meters for a single flow metering application. It is more expense, more maintenance,
more footprint etc. A single hybrid meter is far more attractive. However, whereas
combining density sensitive & density insensitive flow meters to produce a single hybrid
meter design is theoretically sound, in practice it can suffer from some practical
limitations. The devil is in the detail. The Boden [1], Pfrehm [2], Lisi [3], Mottram [4] &
some of Steven’s [6] hybrid designs suffer from the same practical flaw. The two meters
are equally important for the concept to operate successfully. However, these designs
tend to inherently chose one meter as the primary meter. The other meter’s performance
is compromised by attempts to fit it into the other primary or ‘main’ meter. For example:
• Boden [1]: The primary meter is the turbine (density insensitive) meter. It is
positioned in the Venturi meter throat where flow velocity is highest and the turbine
meter works best. The turbine meter will disturb the flow in the Venturi meter throat.
The turbine meter size limits the choice of the Venturi meter throat. That is, the
Venturi meter cannot be independently sized to the most appropriate beta (throat to
inlet diameter ratio) value for any particular application. Without free range to fit an
appropriate DP meter beta value, it is not possible to set the Venturi meter flow range
as the same as the turbine meter. The Venturi meter performance is compromised.
• Pfrehm [2]: The primary meter is the turbine (density insensitive) meter. It is still
positioned in the throat of the Venturi meter. Pfrehm did state that the Boden [1]
design compromised the Venturi meter performance. The Pfrehm design attempted
to somewhat improve the Venturi meter performance. However, the improvement
was marginal. The Venturi meter still can not be independently sized to its best beta
value. The Venturi meter performance is still compromised.
• Lisi [3]: The primary meter is the vortex (density insensitive) meter. The ‘DP meter’
is the vortex meter bluff body with a DP read across it. A vortex meter bluff body
4
makes a poor DP meter primary element. It produces low DPs at even moderate to
high flow rates meaning a limited useable flow range, and a range that is mis-
matched with the vortex meter range. In this hybrid design the DP meter is
compromised.
• Mottram [4]: The primary meter is the vortex (density insensitive) meter. It is
positioned in the throat of the nozzle meter. The nozzle meter may be affected by the
presence of the vortex meter in its throat and the throat cannot be sized
independently of the vortex meter. In this hybrid design the nozzle meter
performance is compromised.
• Steven [6], Venturi / Nozzle with ultrasonic meter design. When placing an
ultrasonic meter in the throat of a Venturi meter both meters can be compromised.
The minimum practical size of ultrasonic meters limits the choice of the Venturi
meter throat. That is, the Venturi meter cannot be independently sized to the most
appropriate beta value for any particular application. Therefore, the Venturi meter
performance is compromised. Furthermore, the local gas velocity in a mid-beta
Venturi meter throat is much greater than in the pipe line. For example a 0.5 beta
Venturi meter with a 20 m/s inlet velocity has an 80 m/s throat velocity. Such high
velocities compromise the ultrasonic meter due to excessive noise. Also, ultrasonic
meters do not operate at peak performance in very low pressure gas applications.
Modern day computers are of course easily capable of cross referencing these meters
outputs. If a hybrid design can be produced such that the meters operate together across
the same flow range both to reasonable accuracy then the concept should work
successfully. The ‘trick’ to a successful hybrid design is finding a combination of density
insensitive and density sensitive flow meters that can be combined into a hybrid design
without significantly affecting either meters performance, while allowing each meter to
be independently ‘sized’ to operate well across the same flow range. DP Diagnostics has
now developed such a meter with Vortek Instruments. This consists of a vortex (density
insensitive) meter in combination with a cone DP (density sensitive) meter.
In order to describe how this meter operates it is first necessary to review vortex meter
and DP meter operating principles.
3 VORTEX SHEDDING FLOW METER THEORY
A vortex meter operates by exposing a bluff body to the fluid stream. Vortices shed from
the bluff body in a cyclic fashion (see Figure 5). This series of downstream vortices is
called a “von Karman vortex street” after the aerodynamicist Theodore von Karman. The
vortex shedding frequency has a nominally linear relationship with the average fluid
velocity. Hence, reading the vortex shedding frequency allows the average flow velocity
to be found.
Vincenc Strouhal studied the phenomenon of vortex shedding. The Strouhal number,
defined by equation 1, is a constant over a large turn down (at least for non-insertion
vortex meters). For even larger turn downs the Strouhal number may be sensitive to
velocity (or Reynolds number). When the Strouhal number ( St ) is found by flow
5
Fig. 5 Vincenc Strouhal (left), Theodore von Karman (right), and the principle of cyclic
vortex shedding from a bluff body.
calibration, then the average velocity ( U 1 ) can be found by reading the vortex shedding
frequency ( f ), and knowing the bluff body width ( d ), (see equations 1a & 2).
fd fd f St
St = ---- (1) U1 = = ---- (1a) where C= ---- (2)
U1 St C d
.
The volume flow rate, Q is therefore calculated by:
. f f C St
Q = A1U 1 = A1 = ---- (3) where K v = = ---- (4)
C Kv A1 Ad
where “A1” is the cross sectional area of the meter inlet and “ K v ” is the vortex meter
“K-factor” which is usually found by calibration. Therefore, the generic vortex meter
volume flow rate equation is equation 3. As the vortex meter K-factor ( K v ) is either set
as constant or data fitted to the average gas velocity the vortex meter volume flow rate
prediction is independent of the fluid density.
If the vortex meter operator chooses to plot K factor against velocity (U1) the resulting
calibration fit (function “ f 1 ” as shown in Equation 5), means that an iteration on the
average velocity is required to solve for volume flow rate, i.e. Equation 6 requires an
iterative solution.
K v = f1 (U 1 ) --- (5)
. f f
Q = A1U 1 = = --- (6)
Kv f 1 (U 1 )
The mass flow rate is found from Equation 7.

m = ρQ --- (7)
Therefore, in order to predict the mass flow rate, the stand-alone vortex meter requires
the fluid density from an external source, e.g. use of a densitometer or a pressure, volume
and temperature (or “PVT”) equation of state calculation based on a known fluid
properties and pressure & temperature readings.
For a more in-depth discussion on vortex shedding flow meter technologies see Storer
[7].
6
4 DIFFERENTIAL PRESSURE (DP) FLOW METER THEORY
The first widely publicized DP meter was the Venturi meter developed by Clemens
Herschel in 1885 based on the work of Giovanni Venturi in 1797. Giovanni Venturi
combined the laws of conservation of mass (as discovered by Lavoisier) and conservation
of energy in fluid flows (as described by Bernoulli) to produce the DP meter concept.
Since Herschel’s first ‘Venturi’ design different shapes of flow obstruction element (i.e.
the “primary element”) have been developed. However, they are all generic DP meters
and they all operate according to the same physical principles.
Fig 6. Left to Right, Antoine Lavoisier, Daniel Bernoulli, Giovanni Battista Venturi, &
Clemens Herschel.
There is a common misconception in the modern industry (often promoted by the
marketing of competing flow meter technologies) that DP meters are an aging
technology. It is all too often claimed, in a rather derogatorily tone, that the DP meter is
“just the application of Benoulli’s principle”. However, such claims are disingenuous.
Flow meters are not fashion accessories, they are an industrial requirement, and their aim
is to meter the flow as simply, reliability and inexpensively as possible. Therefore, if this
can be achieved by “just applying the Bernoulli principle”, then why not just apply the
Bernoulli principle!? However, the common belief that DP meters operate by just
applying the Benoulli principle is technically wrong. Bernoulli’s principle alone cannot
be used to find a fluids flow rate. The Bernoulli principle (i.e. the conservation of the
flows energy) must be cross referenced with Lavoisier’s principle of the conservation of
mass for the flow rate to be found. This is how a DP meter actually operates. Neither of
the physical laws of the conservation of mass or energy alone can be used in isolation to
find the flow rate. It is the concept of combining these laws that creates the DP meter.
Therefore, DP meters operate by cross-referencing two of the most fundamental laws of
physics. This makes DP meters extremely simple, yet robust and reliable.
Figure 7 shows a sketch of a generic DP meter, with an inlet (subscript “1”) and a
minimum cross sectional area, or ‘throat’ (denoted by the subscript “t”). The inlet
pressure P1 is usually read (with the system temperature). This information is
traditionally combined with independent knowledge of the gas composition to allow a
‘PVT’ equation of state calculation of gas density. The differential pressure (DPt) is
directly read. The conservation of mass flow (m) between the inlet cross sectional area
(A1) and the throat cross sectional area (At) for an incompressible flow is expressed by
Equation 7a. Note that ρ denotes the fluid density, Q denotes the volume flow rate, and
U1 & Ut denote the average fluid velocity at the inlet and throat sections respectively.
7
Fig 7. Sketch of Generic DP Flow Meter
Conservation of mass expression: m = ρQ = ρA1U 1 = ρAt U t --- (7a)
At
Re-arranging the conservation of mass expression: U1 = Ut --- (8)
A1
For incompressible flow the density is constant and hence Equation 7a can be expressed
as Equation 8. The conservation of mass flow dictates that the DP meter geometry (i.e.
the throat to inlet area ratio) sets the throat to inlet fluid velocity ratio. On its own, this is
all the law of conservation of mass can tell us. It is not possible to find the flow rate
through the DP meter using the law of conservation of mass alone.
P1 U 12 Pt U t2
Conservation of energy expression: + = + --- (9)
ρ 2 ρ 2
2∆P
Re-arranging the conservation of energy expression: U1 = U t2 − --- (10)
ρ
A horizontal incompressible no loss flows conservation of the energy between the inlet
cross sectional area (A1) and the throat cross sectional area (At) is expressed by
Equation 9. This is Bernoulli’s theorem. For incompressible flow the density is constant
and hence Equation 9 can be expressed as Equation 10. The conservation of energy
dictates that the relationship between the average inlet and throat fluid velocities is
dictated by the differential pressure and fluid density. On its own, this is all Bernoulli’s
theorem can tell us. That is, for any given inlet velocity, the throat velocity is dictated by
the differential pressure and the fluid density. It is not possible to find the flow rate
through the DP meter using the law of conservation of energy, i.e. the “Bernoulli
theorem”, alone.
The relationship between the inlet and throat velocities must satisfy both the laws of
conservation of mass and energy. Therefore, equating the throat velocity expression from
the conservation of mass & energy expressions gives Equation 11. Re-arranging gives
Equation 12. Substituting equation 12 into equation 7 produces the mass flow expression
equation 13. Note that beta (β), a geometry value for a given DP meter, is defined by
equation 14. The “Velocity of Approach” (denoted as ‘E’) is defined as equation 15.
8
At 2∆P 2 ∆P
U1 = U t = U t2 − --- (11) Ut = --- (12)
A1 ρ   A  2 
ρ 1 −  1  
  At  
2∆P
m = ρQ = ρAt U t = ρAt = EAt 2 ρ∆Pt --- (13)
  A  2 
ρ 1 −  1  
  At  
At 1
β= --- (14) E= --- (15)
A1 1− β 4
Equation 13 is the DP meter idealized mass flow rate equation. The actual gas DP meter’s
flow rate equation also includes an expansion term denoted by “ε” to account for density
changes through the meter, and a discharge coefficient denoted by “Cd” to account for
other differences between the ideal and actual operation of the meter. The actual generic
DP meter mass flow rate equation is shown as equation 16. It follows that the DP meter
expression for actual volume flow rate is equation 16a. Note, that to predict the mass or
volume flow rate with a DP meter requires the density be known from an external source.
As with the vortex meter this is usually via a densitometer or a “PVT” equation of state
calculation.
m = ρQ = EAt εC d 2 ρ∆Pt --- (16)
2∆Pt
Q = EAt εCd --- (16a)
ρ
5 BODEN MASS FLOW METERING

Let us consider a vortex meter and a DP meter in series. A vortex meter finds the volume
flow rate by equation 6, where the vortex meter K-factor is usually considered to be a
constant value. A DP meter (such as a cone meter) finds the volume flow rate by equation
16a. Equating the vortex meter and DP meter volume flow rate equations (i.e. Equations
6 & 16a) gives Equation 17. Re-arranging this expression produces an expression for the
fluid density, i.e. Equation 18.
f 2∆Pt
Q= = EAt εCd --- (17)
Kv ρ
2
 K EA εC 
ρ = (2∆P ) v t d  --- (18)
 f 
The expansion factor is a function of the inlet pressure, the differential pressure, the DP
meter’s beta and the isentropic exponent (see Equation 19). The beta value is a known
constant geometric value, and the inlet pressure and differential pressure are measured.
9
The isentropic exponent is a gas property. However, for any given type of gas it is a
known value that is effectively constant over a large thermodynamic range. Furthermore,
the flow rate prediction sensitivity to isentropic expansion value uncertainty is extremely
low. Uncertainty in a gases isentropic exponent is a third order issue.
ε = f (P, ∆Pt , β , κ ) --- (19)
Therefore consider Equations 18 & 19. The velocity of approach (E), beta value (β) &
throat area (At) are known geometric constants. The inlet pressure (P), the DP (∆P) & the
vortex shedding frequency (f) are read. Therefore, with a known isentropic exponent, if
the vortex meter K-factor and the DP meter discharge coefficient values are constants, the
density of the fluid is directly found via Equation 18. Then, as the volume flow rate is
known from the vortex meter Equation 6, the mass flow rate is found via Equation 7.
In cases where the lowest possible uncertainty is required, the vortex meter K-factor may
be expressed as a function of gas velocity (see Equation 5). This function is formed by
data fitting the calibration results. In such a case the volume flow rate is still predicted by
the stand alone vortex meter, by iterating Equation 5 on the gas velocity, to give the
volume flow rate prediction (Qvortex). The DP meter discharge coefficient may need to be
expressed as a function of the Reynolds number (see Equation 20). This function is
formed by data fitting the calibration results.
C d = f (Re ) --- (20)
Such a scenario is represented Equation 17a. The associated fluid density calculation is
shown as equation 17b. However, this expression has two unknowns, i.e. the fluid density
and Reynolds number. Equation 21 shows the Reynolds number calculation. Equation 21
is substituted into equation 17b. There are still two unknowns, i.e. the fluid density and
viscosity (µ). It is necessary to supply the fluid viscosity to the fluid density calculation.
This is the price of requiring reduced uncertainty.
2∆Pt
Qvortex = EAt ε { f (Re )} --- (17a)
ρ
 EA ε { f (Re )} 
2
ρ = (2∆P ) t  --- (17b)

 Q vortex 
.
4ρ Q
Re = --- (21)
πµD
In summary, when using a constant DP meter discharge coefficient and an approximate
isentropic exponent value this methodology can produce a fluid density prediction, and a
volume and mass flow rate prediction. For optimized performance (i.e. minimum output
uncertainty) the discharge coefficient can be fitted to the Reynolds number, meaning that
the operator has to supply the fluid viscosity. For gas flows this is often not a serious
hindrance. Gas viscosities are reasonably well known at a given temperature, and the
calculation has a low sensitivity to viscosity uncertainty.
10
5a INITIAL DP DIAGNOSTICS AND VORTEK INSTRUMENTS RESEARCH
The initial DP Diagnostics and VorTek Instruments research tested Lisis’s idea of using a
stand alone vortex meter and reading the DP across the bluff body as a ‘DP meter’. The
problems found by Kent Industries in the 1970’s were confirmed. Across the vortex
meter’s flow range the DPs produced were low. There was a mis-match in this designs
‘DP meter’ & vortex meter practical flow ranges. As the concept requires both meters
operate well across the same flow range this ‘DP meter’ design is not a practical
replacement of a stand alone DP meter.
This is the Achilles heel of all previous hybrid designs. The DP meter has been
compromised by the requirements of the accompanying density insensitive meter limiting
the choice of the DP meter’s beta (and hence practical flow range). Furthermore, the DP
meters chosen in these existing designs were all to an extent compromised by the flow
disturbance caused by the presence of the density insensitive meter.
The vortex / cone DP meter combination offered a solution to these problems. The cone
meter is so resistant to upstream disturbances that it should not be significantly adversely
affected by the presence of a vortex meter close coupled upstream. A series of tests of
separate vortex and cone DP meters in series were conducted. These 4” meter tests
confirmed that both the vortex & cone DP meters continued to operate normally down to
the minimum spacing tested, i.e. the vortex meter two pipe diameters (2D) upstream of
the apex (i.e. nose section) of the cone. (A detailed discussion on this early research
project is given by Storer et al [8].)
Fig 8. First 4” Prototype Mass & Volume Flow Meter and Densitometer
A 4” spool containing a vortex meter 3D upstream of the cone meter was produced, as
shown in Figure 8. Figure 8 shows the vortex meter systems flow computer casing
connected above the vortex shedding bluff body at the inlet to the 4”, schedule 80, meter
body. Close to the mid-point of the meter body there are two pressure tappings. These are
the standard cone DP meter inlet and low pressure cone pressure taps. Figure 9 shows a
downstream view. The cone meter had a 0.75β. This beta was chosen to show that even a
relatively small cone was resistant to upstream disturbance caused by the vortex meter.
Figure 8 also shows a wall pressure tapping downstream of the cone. This is to allow the
cone DP meter to incorporate the generic DP meter diagnostic system ‘Prognosis’. With
this extra length to accommodate the downstream pressure tapping this 4” meter was 11D
11
long at 42 inches (3.5 ft) long. If DP meter diagnostics were not required this 4” mass
flow meter would have been approximately 8D at 31 inches (2 ½ ft) long.
The CEESI air tests had a turndown of 14:1 (i.e. 6.4e6 ≤ Re ≤ 4.3e5). Three nominal
pressures were set at 14 Bar (16 kg/m3), 27 Bar (32 kg/m3) & 41 Bar (48 kg/m3). The
subsequent calibration of the independent meters are shown in Figures 9 & 10. The cone
DP meter expansibility calculation was taken from Stewart [9]. Figure 9 shows the vortex
meter K-factor set to a constant value with an uncertainty of ±0.5%. Figure 10 shows the
DP meter discharge coefficient fit. A constant discharge coefficient fitted the data to an
uncertainty of ±0.5%. A linear fit to Reynolds number fitted the data to ±0.35%.
With a constant K-factor the vortex meter predicted the volume flow rate directly (see
Figure 11). The mass flow rate and density were predicted by combining this volume
meter output with the cone DP meter with the liner discharge coefficient fit. The results
are shown in Figures 12 & 13 respectively. The volume flow rate prediction had < 0.5%
uncertainty. The mass flow rate prediction had < 0.5% uncertainty. The gas density
prediction had < 1% uncertainty.
4" Prototype Volume, Mass & Density Meter

1260
Kf = 1220.9
1240
+/- 0.5%
K-Factor
1220
1200
1180
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000
Reynolds Number, Re
Fig. 9. 4” Prototype Meters Vortex K-Factor, Fitted as a Constant Value.
4" Prototype Volume, Mass and Density Meter

0.82
+0.35%
Discharge Coefficient
0.81
Cd = 0.8069+ (7E-10*Re)
-0.35%
0.8 +/- 0.35%
or
Cd = 0.8087 +/- 0.5%
0.79
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000
Reynolds Number, Re
Fig. 10. 4” Prototype meter DP Discharge Coefficient, as a Function of Reynolds No.
12
4", Prototype Volume, Mass & Density Meter
1
% Volume Flow Rate
+0.5%
0.5
Uncertainty
-0.5
-0.5%
-1
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000
Reynolds Number, Re
Fig 11. 4” Prototype Meter Volume Flow Rate Prediction.
4", Prototype Volume, Mass & Density Meter

2.5
2
Mass Flow Rate %
Constant Kf & Cd = f(Re)

1.5
Uncertaity
1
+0.5%
0.5
0
-0.5
-0.5%
-1
-1.5
-2
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000
Reynolds Number, Re
Fig 12. 4” Prototype Meters Mass Flow Rate Prediction.
4" Prototype Volume, M ass & Density M eter

2.5
Density % Uncertainty
2 Constant Kf & Cd = f (Re)

1.5 +1%
1
0.5
0
-0.5
-1
-1.5 -1%
-2
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000
Reynolds Number, Re
Fig 13. 4” Prototype Meter Density Prediction.

Note that the system is required to know the gas flow’s isentropic exponent and the gas
viscosity. In this example approximations of a gas isentropic exponent of 1.4 and the gas
13
viscosity estimate of 1.84e-5 Pa-s were held constant across all the flow conditions to
give these results.
5b DESIGNIING A HYBRID VORTEX / CONE DP METER
Having confirmed that a vortex meter in close proximity to a downstream cone DP meter
was a viable mass meter system it was now necessary to produce a hybrid system in order
to make the system more compact, lighter and attractive to industry. This task proved to
be more difficult than first envisaged. It took four design iterations to achieve this goal.
The devil was in the detail.
5b.1 How Not To Design a Hybrid Vortex / Cone DP Meter
The standard cone DP meter has a cone support that extends vertically down from the
wall to the centre line, where it is attached to the apex of the cone. The initial hybrid
design replaced this circular cone support with a vortex meter bluff body. This bluff body
/ cone support bar extended down to the centre line only. The vortex shedding sensing
device was positioned at the standard position downstream of the bluff body. The cone
element was positioned at its normal location downstream of its support bar. This design
is drawn in Figure 14.
Fig. 14. Initial Hybrid Vortex / Cone DP Meter Design

Air flow testing this design at CEESI showed that whereas the cone meter operated
normally the vortex meter was unserviceable with no vortex shedding read. Vortek
Instruments postulated that the problem was that the bluff body did not extend across the
entire diameter of the meter body. Insertion vortex meters, like integral vortex meters,
usually have bluff bodies that extend the across the full pipe diameter. Shorter bluff body
lengths have been known to have vortex shedding problems. This hybrid design was not
viable.
The second hybrid design extended the bluff body / cone support to the full meter body
diameter. This second iteration design is drawn in Figure 15. This design now had a
standard vortex meter bluff body design. As the bluff body was the length of the meter
body diameter it was attached at both ends to the wall. This design gave the bluff body
more chance of producing the required vortex shedding while it significantly increased
the stiffness of the bluff body / cone assembly. This reduces the likelihood of any long
term fatigue failure, while also increasing the overall strength of the assembly.
14
Fig. 15. Second Hybrid Vortex / Cone DP Meter Design
Air flow testing this design at CEESI showed that again the cone meter operated correctly
but the vortex meter was still unserviceable. The vortex shedding was intermittent and
not reliable. Vortek Instruments postulated that the problem was that the proximity of the
vortex shedding sensor to the downstream cone element. The vortex shedding sensor
overlapped the nose of the cone. As the reducing cross sectional area / flow acceleration
is a known flow disturbance mitigater, it was suspected that the cone was dissipating the
von Karman vortex street. This hybrid design was therefore not viable. The cone meter
had to be withdrawn further downstream to allow the vortex shedding sensor to be
upstream of the cone element. With successful earlier tests of a stand-alone vortex meter
2D upstream of a cone meter it was realized that the distance did not have to be
excessive.
For the third design iteration the question became how far downstream must the cone
element be from the vortex meter sensor in order for the vortex meter to be unaffected by
the presence of the cone element? It was decided to build a test meter with a replaceable
cone element such that the distance between the vortex meter bluff body and the apex of
the cone was variable. The bluff body and cone element were connected by a hollow
circular bar. This ran along the meter body centre line and screwed into the bluff body
and cone element. The distance between the bluff body and cone element could be varied
by changing the bar length between tests. This would avoid the necessity to build
different test meters with different bluff body / cone element distances.
Such a replaceable cone meter design was used in the early 2000’s by McCrometer for
R&D purposes, Cameron considered such a design in the mid 2000’s, and Dynamic Flow
purports to have ‘invented’ this concept more recently. However, this particular design
was singularly used here for R&D convenience, it was not considered as part of any
commercial finished design. This third iteration design is drawn in Figure 16.
Air flow testing this design at CEESI, with a 1.5D spacing between the bluff body and
cone apex, showed that this time the vortex meter operated correctly but that the cone
meter had highly variable and untrustworthy DP readings. Subsequent inspection found
galled threads on the connecting bar. This was a lesson on the practical problems of
replaceable cone element designs. Replaceable cone elements usually means a thread is
required to connect the cone element to the meter. This is a weak point on the design. It is
essential for the successful operation of a cone meter that there is no leak in the conduit
15
Fig. 16. Third Hybrid Vortex / Cone DP Meter Design
running from the back face of the cone through the cone and up the support bar to the low
pressure port. The galled thread was leaking. The low pressure port therefore read a
varying and incorrect pressure, and hence the associated DP was erroneous. The thread
was galled between manufacture and the first test. It was considered that reasonable due
care had been taken while transporting and installing this meter. Hence, there would be a
considerable concern about the practicality of such a replaceable cone element design in
typical industrial applications. The R&D project reverted to building fixed cone meter
designs for the fourth iteration of tests.
5b.2 A Successful Hybrid Vortex / Cone DP Mass Meter Design
For the fourth design iteration a 4” 0.563β cone element meters were built with a 1D
bluff body to cone apex spacing. For simplicity this prototype meter was a flangeless
‘wafer style’ meter (as shown in Figures 17 & 18). The inlet pressure tap and the
downstream diagnostic pressure tap were on the upstream and downstream pipes
respectively. The cone extended into the downstream pipe. This 4th iteration 1D spacing
(‘short cone design’) meter was tested with air flow at CEESI.
As a proof of concept test the test matrix was limited to two pressures. Figure 18 shows
the meter installation. Note that unlike the drawing in Figure 17, the actual meter
produced (Figure 18) had the cone low pressure port located at 1800 to the vortex
shedding sensor and the vortex meter head. In practice it was found that this produced a
less congested design without compromising performance.
Figure 19 shows the cone meter discharge coefficient vs. Reynolds number relationship
fitted by a linear equation. This discharge coefficient was slightly lower than would be
expected by a standard cone meter. However, the discharge coefficient across the tested
range is not significantly outside of the accepted range for standard cone meters, i.e. 0.8
±8% (i.e. 0.736 ≤ Cd ≤ 0.864). Furthermore, this meter is not a standard cone meter, with
an unconventional support bar. Hence, the calibration result is understandable, and the
discharge coefficient was fitted to a linear line at 0.3% uncertainty. As expected the
pressure has no significant effect on the discharge coefficient.
Figure 20 shows the calibration result of the vortex meter. A constant K-factor fitted the
reference meter to 0.75% uncertainty. The mass flow rate prediction of the system is
shown in Figure 21. The mass flow rate is predicted to within 1% and 95%. The volume
16
flow rate prediction of the vortex meter is shown in Figure 22. The volume flow rate
prediction is within 1% and 95% confidence. The density prediction of the system is
shown in Figure 23. The density prediction is within 1.5% and 95% confidence.
Fig. 17. Fourth Hybrid Vortex / Cone DP Meter Design
Fig 18. The Prototype Mass Meter Installed at CEESI.
Fig 19. Discharge Coefficient Calibration of the ‘Short Cone Design’ Cone Meter.
17
Figure 20. K-Factor Calibration of the ‘Short Cone Design’ Vortex Meter.
Fig 21. Mass Flow Rate Prediction by Combining Traditional Cone & Vortex Meters.
Fig 22. Vortex Meter Volume Flow Rate Prediction.

18
Fig 23. Density by Combining Traditional Cone & Vortex Meters.
6 A 2” Hybrid Vortex / Cone DP Mass Meter Steam Flow Test
In 2015 Vortek Instruments tested this prototype mass meter at the Spirax steam flow test
facility in Cheltenham, UK. Steam flow test facilities are rare and the few that exist are
generally private facilities not available for hire by third parties. A 2”, 0.5β meter was
built and tested with single phase saturated steam flow (see Figures 24 and 25). The dry
gas saturated steam test data ranged between 4 & 10 Bar.
Fig. 24. 2” Vortex / Cone DP Mass Meter

The Spirax facility is a steam flow facility for general saturated steam flow component
testing. However it was not primarily designed as a flow meter calibration facility. The
steam flow rate reference was a vortex meter. The nominal volume flow uncertainty was
1%. Due to data logging issues between the facility control panel and the meter under test
each test point’s average reference flow rate was manually estimated and recorded from
observing the facility control system output over time. It was assumed, but not
guaranteed, that the steam maintained 100% quality. It is therefore conservatively
estimated that the reference volume flowrate uncertainty was 2.5%, the reference steam
density uncertainty was estimated at 1%, and therefore the corresponding mass flow rate
19
uncertainty was estimated at 3%. It is not possible to give any instrument a lower
uncertainty rating than the reference system on which it was calibrated. Hence, in this
steam flow test, although steam flow is a rare and interesting flow meter test medium, the
flow meter can only be shown to predict the mass flow to 3% uncertainty. This is a
limitation of the specialized steam flow facility, and not necessarily the best performance
of the meter under test.
Figure 25. 2” Vortex / Cone DP Mass Meter at the Spirax Steam Test Facility.
Figure 26 shows the cone DP meter’s discharge coefficient plotted against Reynolds
number. A constant discharge coefficient fitted the data to 1.4%. Figure 27 shows the
vortex meter K-factor plotted against average gas velocity. A constant K-factor fitted the
data to 0.9%. (These values would be reduced if the meter was calibrated on a single
phase gas flow calibration facility.) The relatively high uncertainty of the paired meter
calibrations will have an adverse effect on the density and mass prediction.
Figure 28 shows the percentage difference between the vortex meter’s predicted and
reference volume flow rate. The error bars represent the volume flow rate reference
uncertainty. The meter under test predicts the volume flow rate to within 1% of the
reference. Figure 29 shows the percentage difference between the meter under test’s
predicted and reference mass flow rate. The error bars represent the mass flow rate
reference uncertainty. The meter under test predicts the mass flow rate to 3% of the
reference meter. Figure 30 shows the percentage difference between the meter under
test’s predicted and reference fluid density. The error bars represent the reference density
uncertainty. The system predicts the fluid density to within 3% of the reference. Due to
the nature of the steam test facility and the available reference uncertainties during this
test, these results are as good as could be expected. This mass meter design has predicted
the volume and mass flow rates to the same uncertainty as the available reference data.
Vortek Instruments, DP Diagnostics (and Swinton Technology) also tested the DP meter
diagnostic system ‘Prognosis’ during these steam tests. The extra downstream pressure
port required by the DP meter diagnostic system is evident in Figure 25. Furthermore, the
metering system was also tested with wet saturated steam. However, these research topics
are out with the scope of this paper and would require a separate dedicated paper to
discuss.
20
Figure 26. Cone Meter Cd vs. Reynolds Number Calibration Data
Figure 27. Vortex Meter K-factor vs. Gas Velocity Calibration Data
Figure 28. Volume Flow Rate Prediction vs. Reference Meter Data
21
Fig 29. Mass Flow Rate Prediction vs. Reference Meter
Fig 30. Density Prediction vs. Reference Value

7. 3” Hybrid Vortex / Cone DP Mass Meter Tested with Water & Air Flows
A 3” 0.68β mass meter was manufactured. (This meter was sized for an industrial steam
flow application field test.) A drawing of the meter is shown in Figure 31. The vortex
shedding sensor is again positioned 1800 to the pressure ports to avoid sensor mounting
congestion. This meter was first tested with water flow at Vortek Instruments. No
photograph of this test was taken, although Figure 31 also shows a photograph of a later
2” meter of the same design being water flow tested in the same facility.
The Vortek Instrument water flow facility utilizes a weigh tank and has a mass flow rate
uncertainty of 0.1%, a water density uncertainty of 0.05%, and a volume flow rate
uncertainty 0.12%. Figures 32 & 33 show the cone DP & vortex meter water flow
calibration data respectively. Figure 34 shows the resulting density prediction and volume
& mass flow rate predictions. This mass meter technique is applicable to liquid or gas
flow applications, and here when the metering system has been calibrated on a very low
uncertainty water flow facility the meter outputs also have a corresponding low
uncertainty.
22
Fig. 31. 3” 0.68β Meter Drawing & a 2”, 0.5β at Vortek Water Facility
This 3” meter was also extensively air flow tested at CEESI. The air test matrix had three
pressures nominally of 56 Bara (812 psia), 27.6 Bara (400 psia) & 10.4 Bar (150 psi),
which relate to three nominal gas densities 66.7, 33.0 & 12.3 kg/m3. The Reynolds
number range was 3.4e5 < Re < 6.2e6, while the gas velocity range was 5 <U1(m/s)< 30.
The cone DP meter discharge coefficient data and the data fit are shown in Figure 37.
The vortex meter K-factor data and the data fit to average gas velocity are shown in
Figure 37. The K-factor is seen to be very steady at high to moderate velocities, although
it begins to become non-linear at < 7 m/s. This result was in part due to an over-sight
during the testing. The vortex meter was inadvertently operated without the standard
signal filtering system activated. This filter helps signal analysis by filtering out signal
noise. Such noise filtration is usually most beneficial at lower velocities. This lack of the
standard practice noise filtering accounts for at least some of the perceived K-factor non-
linearity at lower velocities. A data fit was therefore preferred over a constant K-factor.
Figure 38 shows the gas mass flow rate prediction uncertainty. (The uncertainty bars
represent the reference mass flow rate uncertainty of 0.35%). The mass flow rate is
predicted to 1% at 95% confidence. Figure 39 shows the volume flow rate prediction
uncertainty. (The uncertainty bars represent the reference volume flow rate uncertainty of
0.3%). The volume flow rate is predicted to 1% at 95% confidence. Figure 40 shows the
gas density prediction uncertainty. The gas density is predicted to < 2% at 95%
confidence.
23
Fig 32. 3” Vortex / Cone DP Meter DP Meter Water Calibration Result
Fig 33. 3” Vortex / Cone DP Meter Vortex Meter Water Calibration Result
Fig 34. 3” Vortex / Cone DP Meter Water Density, Volume Flow & Mass Flow
Prediction Result
24
Fig 35. 3” 0.68β Hybrid Vortex / Cone DP Meter Under Air Flow Testing at CEESI.
Fig 36. 3” Cone DP Meter Discharge Coefficint vs. Reynolds Number Air Data.
Fig 37. 3” Vortex Meter K-Factor vs. Gas Velocity Air Data.
25
Fig 38. 3” Mass Meter Mass Flow Rate Prediction Uncertainty with Air Flow.
Fig 39. 3” Mass Meter Volume Flow Rate Prediction Uncertainty with Air Flow.
Fig 40. 3” Mass Meter Gas Density Prediction Uncertainty with Air Flow.
26
7.1 3” Hybrid Vortex / Cone DP Mass Meter Combined Gas & Liquid Data
Figure 41 shows cone DP meter discharge coefficient derived from the combined data
from the gas and water labs. The two different labs give very similar discharge
coefficients across the same Reynolds number range. Where the water and air Reynolds
number ranges over-lap the discharge coefficients agree.
Note that due to the low Reynolds number range of the water data a constant discharge
coefficient fitted this range well (see Figure 32). However, if we were to have
extrapolated the water calibration result to the much higher Reynolds number values of
the gas flow Figure 41 shows that a significant discharge coefficient bias would have
been introduced. Only by calibrating across an applications full Reynolds number range
can the operator be assured that the discharge coefficient data fit is applicable for that
Reynolds number range. When combining the water and air calibration data one linear
discharge coefficient vs. Reynolds number data fit predicts the discharge coefficient
across both data sets to 0.5%.
Fig 41. Air & Water Cone Meter Discharge Coefficient Calibration Data Set & Data Fit
Fig 42. Gas & Water Lab Vortex Meter K-Factor Coefficient Data Set & Data Fit
27
Figure 42 shows the vortex meter K-factor derived from the combined data from the gas
and water labs. If the low air flow velocity result is ignored (as it has been shifted due to
the lack of noise filtering) the two different labs give very similar K-factors across the
same Reynolds number range. One linear K-factor vs. velocity data fit predicts the K-
factor across both data sets to 1%.
Fig 43. Combined Gas & Water Lab VorCone Mass Meter
Mass Flow, Volume Flow & Density Predictions
In practice operators nearly always know enough about their process that they will know
the fluid type they are metering. As such, it is not a realistic example to show this mass
meters ability to measure the gas and liquid flow ranges using the combined water & air
calibration data. Furthermore, prior knowledge of whether the application has a gas or
liquid flow dictates if the flow calculation applies expansibility on the cone DP meter
calculation, and the fluid viscosity input. However, in order to show this mass meter’s
range ability, i.e. wide applicability, Figure 43 shows the meter’s performance when
applying the combined calibration factors shown in Figures 41 & 42. In these
calculations, the gas data had the expansibility correction applied and the air and water
data had the appropriate fluid viscosity applied to the discharge coefficient calculation.
The range in flow conditions in this combined data set is extreme, with a density range
between water and low pressure air of 82:1, and volume and mass flow ranges (i.e. turn
downs) of 20:1 and 59:1 respectively.
Figure 43 shows the predicted mass flow, volume flow, and fluid density from the
combined water and gas calibration data fits. Whereas as a liquid meter or a gas meter the
system predicted the mass and volume flows to 1%, and the density to 2% at 95%
confidence, applying the meter across the combined air and water flow ranges produces
only a marginal increase in uncertainty. Figure 43 shows the performance across both
fluids and their respective ranges was the mass flow to 1.3%, volume flow to 1%, and the
density to 2% at 95% confidence. (The few points outside these limits are the low
velocity gas flow where the vortex meter had no noise filtering activated.)
28
8 CONCLUSIONS
Traditional gas flow meter designs such as ultrasonic, turbine, vortex and DP meters
require that the gas density, and possibly viscosity and isentropic exponent, be supplied
from an external source in order to predict the mass flow rate (or the equivalent volume
flow rate at standard conditions). The gas viscosity and isentropic exponent can usually
be estimated to reasonable uncertainty. Furthermore, the meter output tends to have a
relatively low sensitivity to these two parameters. However, the requirement for fluid
density can be more problematic. The uncertainty in a traditional meter’s mass flow rate
prediction is heavily influenced by the density prediction uncertainty. Density uncertainty
is a significant (and sometimes primary) contributor to traditional mass flow rate
prediction uncertainty. As such, direct mass measurement is sometimes beneficial, either
to eliminate the requirement for density measurement or to act as a check against the
independent density measurement.
While Coriolis meters are large by volume (footprint) and weight, have high permanent
pressure loss and are relatively expensive, for clean high value liquid flows their
exceptional measurement performance has established them as the benchmark of mass
flow and density metering. However, Coriolis meter performance declines somewhat
with gas flow. Although still a good meter the Coriolis meter does not measure density of
low pressure gas (the output is called an indication and not a measurement), and the gas
mass flow rate uncertainty increases to be equivalent to standard gas meters. This
reduction in the main pro, i.e. performance, coupled with the continued existence of the
cons, i.e. size and weight issues, high permanent pressure loss and expense, has meant
that Coriolis meters are less favoured for many gas applications, especially moderate to
low pressure flows where the value of the product is less. There is therefore still a niche
market need for a simple, relatively inexpensive gas mass flow meter.
The simple Boden mass flow meter concept has been known for sixty years. The concept
historically failed to become a mainstream meter design due to various reasons, including
early computer limitations and practical problems in combining the two technologies. As
such the concept remained relatively obscure. However, modern computers easily cope
with the demands of this technology and the practical problems of combining the flow
meter technologies are surmountable.
DP Diagnostics and Vortek Instruments have developed a practical Boden type mass flow
meter by combining a vortex meter and a cone DP meter. Multiple meter tests have
shown this hybrid meter design to be a viable practical industrial gas mass meter. This
particular design allows the vortex and DP meter sub-systems to be independently sized.
Critically, this allows both meters to be sized for use across the same flow ranges thereby
solving one of the most challenging problems that hampered the R&D of this general
concept for many years. The research and development project has solved the mechanical
and fluid mechanics problems arising from creating this hybrid metering system.
The mass meter system has now been tested in three different test facilities with air, water
and steam. The results show that when tested at a reputable flow calibration test facility
across the applications full Reynolds number range, the technology can meter even low
pressure gas flows to 1% by mass and 1% by volume while measuring the fluid density to
29
< 2%. The system can be used as a stand-alone meter, or in conjunction with a density
measurement system. If a density measurement is independently available both the cone
DP meter and the vortex meter can use the independent density prediction to individually
measure the mass flow rate. The combination of the two meters supplies a density
measurement check independent of the traditional density measurement.
9 REFERENCES
1. Boden., “Mass Flow Meter” US Patent No. 2772567A, Granted 1956.
2. Pfrehm. “Modified Turbine Mass Flow Meter”, US Patent No. 3430489, Filed 1967.
3. Lisi, E.L., “Mass Flow Meter”, United States Patent No. 3,785,204, Filed April 6th,
1972, Awarded January 15th, 1974.
4. Mottram, R.C., “Mass Flow Meter”, United Kingdom Patent Application No. GB
2161941A, Filed 19th July 1984.
5. DiMarco S et al “Simplified Fluid Property Measurement”, US Patent No. 7258024
B2, Filed March 2005.
6. Steven R., “Flow metering”, US Patent No. 2010/0224009 A1, Published Sept. 2010.
7. Storer, J., “Flow Measurement Utilizing Vortex Shedding Meters”, International
School of Hydrocarbon Measurement, Norman, Oklahoma, May 22nd -24th, 1990.
8. Storer, J. “Advances in Vortex Shedding Flow Metering”, International Symposium of
Fluid Flow Measurement, Anchorage, Alaska, August 2009.
9. Stewart D., Reader Harris M., Peters R., “Derivation of an Expansibility Factor for the
V-Cone Meter”, Flow Measurement International Conference, Peebles, Scotland, UK,
June 2001.
30
20nd-23rd October 2015
In-situ effects on ultrasonic gas flowmeters

Philip Chan, GDF Suez E&P Norge AS
Øyvind Storli, FMC Technologies
Sveinung Myhr, FMC Technologies
Reidar Sakariassen, Metropartner
Atle Abrahamsen, FMC Technologies
Kjell-Eivind Frøysa, Christian Michelsen Research
Camilla Sætre, Christian Michelsen Research
Abstract
This paper describes the course of events which led to the discovery of a significant mis-measurement on
Gjøa platform’s gas export from the operator's point of view. The mis-measurement has persisted since Gjøa's
start-up, almost five years ago. The paper closes with lessons learnt from a two year long investigation and
tries to clear up some common misconceptions on ultrasonic gas flowmetering.
The gas export on Gjøa is a cross-border measurement between Norway and UK. There has therefore been
a large emphasis on the accuracy, calibration and diagnostic capabilities of Gjøa’s gas export since the
beginning, for operators and regulators from both countries alike. The design of the gas export system
comprise of two 12’’ ultrasonic meter runs with 100% capacity to facilitate removal of flowmeters for calibrations
without stop of production. The diagnostic capabilities of the export USM are fully utilised by logging of hourly
and daily diagnostic data.
An accidental switchover from duty to stand-by meter run led to the discovery that metering run one was
overmeasuring metering run two by ca. +2%, appearing as a sudden change in flow rate. Furthermore, sudden
but stable flow profile changes were observed on meter run two which temporary led to 2-3% more flow.
Process venturi meters in vicinity of the export meters could not confirm the flowrate changes measured by
the gas export station during the switchover. Tracable calibrations onshore show no evidence that the
flowmeters have drifted or are erroneous. Three flowmeters are swapped among two metering runs, all
flowmeters have therefore been installed in any of the metering runs. The mis-measurement is therefore not
related to any of the individual flowmeter but the meter run in operation, which increases the suspicion on flow
profile effects as the root cause.
In conclusion, operators have to increase their awareness over profile parameters and its influence on
flowmeter performance. Non-ideal in-situ flow profiles might have serious effects on the measured flow, even
when the USM has passed the ISO 17089 type test. The type test should not be seen as a guarantee that the
flowmeter is able to handle all perturbations, but rather as an uncertainty budget when the profile is unknown.
When the flow profile is non-ideal, a more detailed assessment of the profile and its implications is required. A
non-ideal profile can cause a systematic bias and lead to significant mis-measurements over time.
Transferability of traceable onshore calibrations must be questioned when the profile measured on-site is
different from that obtained during calibration.
1 Introduction
Gjøa is a semi-submersible floater, located in the North Sea ca. 47km from the Norwegian shore. Construction
started in 2007 with Statoil as the operator during the construction phase, while Engie took over as operator
for the production phase in end of 2010. Five subsea templates from Gjøa field and three from Vega field are
tied-back to the Gjøa platform.
Fig 1-1: Gjøa illustration (Statoil 2015)
Fig 1-2: Position of Gjøa semi (Norsk Oljemuseum 2015)

The produced oil is exported to the Mongstad refinery via the Troll pipeline, while the gas is exported via the
Gjøa gas pipeline (operated by Gassco) and the Far North Liquids Associated Gas System
(FLAGS, operated by SEGAL consortium) for further processing at the St.Fergus gas plant in the UK. Gjøa
represents the majority of gas transported in the FLAGS, as its capacity is ca. 33 MSm3/d (Shell 2015a),
while Gjøa's export capacity is approx. 18 MSm3/d (Norwegian Petroleum Directorate 2015). Gjøa's fiscal
gas export measurement is therefore critical due to:
 Gjøa's substantial contribution to quantities transported through FLAGS

 Cross border export from Norway to UK
 Allocation of various platforms connected to FLAGS
The importance of Gjøa's gas export measurements for both nations has been present from day one and has
therefore been designed to comply to metering regulations from both UK and Norway.
Fig 1-3: FLAGS Overview (Shell 2015b)

2 System description
2.1 Gjøa's fiscal gas export system
2.1.1 Arrangement and layout
Following equipment are of relevance for this investigation:
 Fiscal ultrasonic meter in meter run 1 (in spring 2013: SN# 2259)
 Fiscal ultrasonic meter in meter run 2 (in spring 2013: SN# 2257)
 Spare fiscal ultrasonic meter in logistics base onshore (in spring 2013: SN# 2258)
 Non-fiscal process venturi 27FE1131 (compressor venturi)
 Non-fiscal process venturi 27FE1312 (PPS venturi)
Fig. 2.1.1-1: Gjøa's fiscal metering station with process flowmeters up- and downstream
The fiscal USMs are arranged in a parallel configuration designed for 2 x 100% flow. Under normal operation,
one run is on duty while the other run is on stand-by. The non-fiscal process flowmeters are package
flowmeters of the compressor system and the Pipeline Protection System (PPS). These flowmeters are not
relevant for the fiscal export, but are used as benchmark during the investigations.
The fiscal metering station features following elements (in order of flow direction):
 Equally sized, symmetric header with blind tees at both ends
 Double block & bleed isolation valve of "split gate" type
 CPA flow conditioner flanged on the isolation valve
 15D upstream straight lenght
 Ultrasonic flowmeter
 9D downstream straight length
 Dual gas chromatographs
 Dual pressure and temperature transmitters and single densitometer
Fig 2.1.1-2: General arrangement fiscal metering station (GDF SUEZ E&P NORGE 2010)
2.1.2 Fiscal Ultrasonic Flowmeter (USM)

All three flowmeters are transit time USMs with six sound paths on four planes. Sound paths are crossed on
the first two planes, while the two lower paths are single sound paths without crossing. Hence, path one and
two are crossed on the 1st plane while path three and four are crossed on the 2nd plane. Path five and six
are single paths, located in plane three and four respectively.
Fig 2.1.2-1: Cross section of an ultrasonic gas flowmeter used on Gjøa semi (FMC Technologies 2008)
The ultrasonic technology and the setup of the transducer paths is well known for all models of all
manufacturers. However, the interpretation of the transit time measurements and their conversion into a
flowrate is naturally proprietary information and well protected. It is nevertheless of utmost importance for the
operator to know the basics of the proprietary algorithms in order to find the reason for an eventual
mis-measurement. Transit time measurements from all six sound paths are used for evaluation of profile
flatness, asymmetry, cross-flow and swirl.
Profile flatness
The ratio between the flow velocities in the center of the pipe and the top/bottom of the pipe is defined as the
profile flatness. This ratio is expressed as a percentage of the core velocity. Hence, values close to 100%
means that the flow velocities on the top and bottom of the pipe are the same as the velocities in the center of
the pipe. A low profile flatness means that the top/bottom velocities are lower than the core velocities and
indicate a more "pointy" axial velocity profile.
Symmetry
The ratio between the flow velocities on the top two levels and the bottom two levels is defined as symmetry.
This ratio is expressed as a percentage. Positive values mean that the flow on the top is faster than the flow
on the bottom.
Swirl and Cross-flow

The ratio between the axial and tangential flow velocities is defined as swirl and cross-flow. Hence, 1% swirl
means that the tangential velocities amount to 1% of the axial velocities. The profile is interpreted as swirling,
if the tangential flow directions on the first and second level are the same. Hence, assuming that the gas flow
turns around the entire pipe. The profile will be intepreted as cross-flow if the tangential flow directions are
opposite for the first and second level. Hence, assuming that there is at least one swirl center in the top section
of the pipe.
Fig. 2.1.2-2: Interpretation of path velocities (FMC Technologies 2008)

2.1.3 Accredited onshore calibration of fiscal USMs

All gas export USMs are subject to a six monthly re-calibration where the duty flowmeter is replaced by a
calibrated USM and sent to an accredited calibration laboratory onshore. All flowmeters performed well during
all calibrations and showed less than 0,3% drift from their previous calibration.
Fig 2.1.3-1: Profile flatness during a typical onshore calibration
2.1.4 Condition monitoring

All fiscal flowmeters are under constant supervision. The diagnostic capabilities of the ultrasonic flow meters
are fully utilised on Gjøa. Following parameters are logged hourly and monitored on a monthly basis:
-Deviation of measured Speed of Sound (SOS) against calculated SOS (AGA10)

-Profile flatness
-Swirl
-Symmetry
-Signal to noise ratio
Cross-flow is usually not used for condition monitoring as the cross-flow experienced on Gjøa is not stable
enough to be monitored. Footprint as described in ISO 17089 is available, but usually not monitored as even
small changes in flow profiles lead to significant changes in the footprint.
3 Order of events
Several incidents in spring 2013 raised doubts on the technical integrity of the fiscal gas export station with
USM SN#2259 and SN#2257 installed in meter run one and two respectively.
 December 2012 Meter run two taken into operation with SN#2257
Installation of SN#2257 in run two as new duty flowmeter, with SN#2259 switched to stand-by in run one.
SN#2258 was removed from run two for scheduled re-calibration.
 February 2013 Accidental swap from meter run two to meter run one
The control system accidentally swapped from metering run two to metering run one. A sudden step change
of ca. +2% in flow could be observed, while all diagnostic parameters were normal without any alarms.
 April 2013 Detection of profile peaks on metering run two

Sudden and temporary changes of the flow profile were observed on metering run two with SN#2257. The
profile flatness increased from ca. 85% to ca. 95%, while swirl and asymmetry changed in value and direction.
The increased profile flatness led to an increase of average flow velocity and resulted to an increase of flow
by ca. 3%. These profile peaks were extremely repetitive and stable for ca. 10s – 3h, before returning back
to its usual flow profile. SN#2257 showed no errors or configuration changes since it was installed in
December 2012. All diagnostic parameters were normal.
 June 2013 Flowtest with metering run one

An in-situ flowtest was performed to check whether metering run one is exposed to the same profile peaks
as in metering run two. When the export was swapped to run one, a +2% step change could be reproduced
like earlier. However, neither the downstream PPS venturi or the upstream compressor venturi could confirm
the change of flow. Metering run one showed no signs of profile peaks as expected.
 September 2013 Parallel flowtest with both meter runs

A parallel flow test was initiated to measure the flowprofile of both flow meters at the same time. The export
was switched from single run to parallel export, then run two was closed for single export with run one. Single
run export with run one was measuring more than single export with run two as observed earlier.
Furthermore, it could be observed that parallel export with both runs measured almost no swirl and
assymmetries. The measured flow during parallel export was between single export with run two and run
one.
 October 2013 New parallel flowtest with replacement of SN#2257

A new parallel flow test was initiated in October 2013 as SN#2257 (the USM which measured profile peaks)
shall be re-calibrated and replaced with SN#2258. The purpose was to see whether the profile peaks and
step change phenomenon can be reproduced with a different flowmeter in run two. This made it also possible
to check SN#2257 with an onshore calibration. SN#2257 was replaced with SN#2258 and the same test
done in september 2013 was repeated. Both step change and profile peak phenomenons could be
reproduced, with a different meter in run two. All observations during the test were predicted, and the test
proved that both phenomenons are independent of the flowmeter in use and applies to all export USMs. This
raised suspicion on other flowmeters on Gjøa, as the same brand and model of flowmeters were also utilised
in other parts of the process system. All USMs were switched to parallel operation in order to limit the amount
of swirl and a potential mis-measurement.
Further investigations and Computational Fluid Dynamic (CFD) simulations were carried out by Christian
Michelsen Research (CMR) and Metropartner. The investigations were concluded in January 2015, details
of the executed CFD simulation are covered in a separate NSFMW paper (Sætre, et al., 2015).
4 Observations for profile peaks

The properties of the observed profile peaks in metering run two can be summed up as follows:
4.1 Stable and predictable profile changes

Changes in the measured flow profiles are very stable and predictable. Profile flatness increases, swirl and
asymmetry change value and direction. All other diagnostic parameters are stable. The observed changes in
flow profile have been observed on every investigated profile peak. The increase of the profile flatness leads
to an increase of measured flowrate of ca. 3%, as an increase of the flow velocity in the vicinity of the pipe wall
leads to an increase of average flow velocity. Flow peaks have not been detected so far during any calibration.
Run two Run two operation

normal operation during peaks
Profile flatness 83% 96%
Swirl -1,8% +1,8%
Symmetry +5,2% -2,9%
Tab. 4.1-1: Typical profile parameters in periods with profile peaks
4.2 Unpredictable occurrence

Occurance of profile peaks cannot be predicted. The appearance of profile peaks vary between once every
48h and several times within 12h. Profile peaks could be observed since SN#2257 was installed in
December 2012, but not before. It was not possible to relate the profile peaks to opening, closing & leakage
of valves or operation of any other process equipment. Profile peaks have been observed with any flow
velocity between 0% - 87% of design flowrate.
4.3 Variable in duration

The duration of profile peaks lasts typically 10-25min, but durations from 10s up to 3h have also been
observed.
Fig. 4.3-1: Changes in profile flatness (red), symmetry (yellow) and swirl (green) in periods with peaks
5 Observations for step change

Gas export was switched from the duty run to the stand-by run in three occasions. A significant increase of
flow could be observed every time when the export was switched from run two to one. Below figure shows
the flow during single and dual run export with the PPS and compressor venturi as benchmark.
690
680
670
Mass flow [tons/h]
660
Meter run 1 SN#2259
Meter run 2 SN#2257
PPS venturi
650
compressor venturi
640
630
620
0:00 12:00 0:00 12:00 0:00 12:00 0:00 12:00
Fig. 5-1: Single export test in June 2013
Deviation against PPS venturi
Dec 2012 Feb 2013 June 2013
Metering run 2, SN#2257 +0,39 % +0,16 % +0,15 %
Metering run 1, SN#2259 +1,88 % +2,08 % +2,04 %
Compressor venturi +3,16 % +3,44 % +3,22 %
Tab. 5-1: Deviation against PPS venturi
Flow deviations from the PPS venturi were systematic, stable and reproducible for all incidents. In general,
measured values from the PPS venturi agrees well with metering run two, while undermeasuring in
comparison with metering run one and the compressor venturi. It was possible to reproduce the sudden
change of flow each time when the metering run was changed.
6 Discussion
6.1 Malfunction and external influences
It is impossible that the step changes and profile peaks are a result of malfunctions or external influences, as
both phenomenons are reproducible in occurance and predictable in outcome. Malfunctions and external
influences like transducer failure, influences from weather, compressor load or variation in power supply
were checked and quickly ruled out as the reason for step changes and profile peaks.
6.2 Benchmark flow meters

The compressor venturi and PPS venturi up- and downstream were used as flow benchmark against the fiscal
flowmeters in this investigation. A possible explanation for these observations could be the inability of the
process venturis to measure the small and sudden flow changes. The export flowmeters are frequently
calibrated at an accredited flow laboratory, while the Venturi dP transmitters are not calibrated and the
condition of the venturi body is unknown. Further doubts can be raised, as USMs have a higher accuracy and
are based on a different physical principle. It is possible that the venturis are physically not able to measure
the subtle change of flow which has been registered by the high accuracy USMs. This possibility has been
investigated and is explored in this chapter. The base equation for massflow measurement for a venturi is
defined in ISO 5167-4 as:
C π
(Equ. 6.2-1) ṁ= ε 4 d2 √2∆p∙ρ
√1-β4
with:
C flow coefficient
𝜀 expansion coefficient
d throat diameter
𝛽 beta coefficient
∆p differential pressure
ρ line density
A rough uncertainty calculation indicated an uncertainty of ca. 2% for the Venturi meters. The high
uncertainty results mainly from the very high Reynolds number, leading to a high uncertainty of the discharge
coefficient, and an inaccurate fabrication of the venturi body.
The measured venturi flow corresponds well with the USMs, except during switch of runs and profile peaks.
This increases the suspicion that the USMs are exposed to non-idealities in the flow profile, while Venturi
meters are not affected as they are based on mass/energy conservation along a single streamline. The offset
of the venturis against the fiscal export could be explained with inaccurate fabrication of the venturi, which
would offset the flow metering with a systematic error. The uncertainty assessment showed that the venturis
should have been able to detect a sudden 2-3% flow change during step changes and profile peaks, as their
uncertainty is estimated to roughly 2%. This, despite the different functional principle of an USM and dP flow
meter.
690
680
670
Mass flow [tons/h]
660
Meter run 1 SN#2259
Meter run 2 SN#2257
PPS venturi
650
compressor venturi
640
630
620
0:00 12:00 0:00 12:00 0:00 12:00 0:00 12:00
Fig. 6.2.5-1: Single export test 20th June 2013 with venturi uncertainties
6.3 Layout
Errors in the design of the metering skid or disturbances upstream of it can lead to significant disruptions of
the flow profile and the performance of USMs.
Fig. 6.3-1: Isometrics of the export metering skid (GDF SUEZ E&P NORGE 2010)
15D and 9D have been implemented as up- and downstream lenghts for the USMs respectively, this is more
than required according to NORSOK I-104. The flow conditioner is flanged directly on the isolation gate valve,
which again is flanged directly on the inlet header. Upstream straight lenghts are therefore not available for
the flow conditioners. One could argue whether the isolation gate valve and parts of the blind tee from the
header arrangement could be seen as upstream straight lenght as the gate valve is full bore.
The gas flows through a u-turn after which it follows ca. 38D of straight lenght, before it enters the metering
skid header after a right bend. Although the inlet arrangement indicate that some swirls could be produced,
especially after the last bend in front of the header, there are no evidences that this setup could generate
severe disturbances in the flow profile.
Fig. 6.3-2: Piping configuration upstream of the eksport metering skid (GDF SUEZ E&P NORGE 2010)
6.4 Tracable flow calibration

Gjøa's export USMs are subject to tracable re-calibrations after six months flowing time. Any defects on the
flowmeters should be visible on the calibration results.
The export flowmeters are calibrated in accordance with ISO 17089 and the obtained flow weighted mean
error (FWME) is compared against its previous calibration. A shift of 0,3% from its last calibration is
contractually set as threshold value for claims on re-allocations.
Fig. 6.4-1: Drift of export flowmeters vs. time in operation
Although very close, none of the calibrations have exceeded the threshold value of 0,3% shift from the last
calibration. A relation between flowing time and shift of the flowmeters could not be found. Preliminary results
from the investigations in 2013 indicated that operation of both export runs mitigated the
mis-measurement. As a result, Gjøa started exporting with both runs since October 2013, causing extended
flowing time for each flow meter, as only three flowmeters rotate among two metering runs. Plotting the shift
against the calibrated flowrate for all calibrations gives a clear picture on the performance of each flowmeter
at each flowrate.
Fig. 6.4-2: Meter factor shift vs. flowrates for all calibrations
Although some of the meter factors have shifted more the 0,3% from the last calibration, none of the
calibrations triggered any re-allocations as the FWA meter factor was below the threshold value and gas export
was operated with meterfactors which have not shifted. Clusters of calibration points can be detected with no
shift at all. FWA shift was below +/- 0,1% if a 4th order polynom is utilised to account for the outliers. Comparing
the flowmeters against the flow laboratory yields an even better performance.
Fig. 6.4.-3: Deviation from flow laboratory vs. flowrates for all calibrations
Maximum allowed deviation according to ISO17089 is +/- 0,7% and +/- 1,4% for flowrates > QTransition and
< QTransition, respectively. The frequent recalibrations do not indicate that any of the flowmeters have been
mis-measuring or shifted significantly. A correlation between time in operation and shift of the flowmeter
could not be found.
6.5 Flow profile
6.5.1 Profile during periods with peaks

Although the profile measured during periods with peaks is significant different from the normal operation in
run two, profile abnormalities are not large enough to explain the observations offshore.
Run two normal Run two operation Run one normal

operation during peaks operation
Profile flatness 83 % 96 % 93 %
Swirl -1,8 % +1,8 % +0,9 %
Symmetry +5,2 % -2,9 % -1,0 %
PPS deviation +0,24 % +3 % +1,9 %
Tab. 6.5.1-1: Typical profile parameters and flow during normal operation and periods with peaks
Note that the profile parameters during single export with run two becomes similar to the profile of run one,
positive swirl and negative symmetry, leading to a positive deviation from the PPS venturi. The increase of
profile flatness during periods with peaks explains the higher flowrate of ca. + 3% due to a higher average
flow velocity. All diagnostic parameters such as signal to noise, SOS, path velocities etc. made sense and
indicated no errors or conflicts in the flowmeter. As mentioned previously, the change in flow was not
registered by the process venturis.
Fig. 6.5.1-1: Change of profile flatness during periods with peaks compared with SOS and signal to noise ratio
6.5.2 Historic flow profile meter run one and two

The diagnostic capabilities of the export USMs are fully utilised on Gjøa and extensive information on the
flow profiles are available since startup. Deviation between theoretical and measured speed of sound is
usually below 1 m/s while signal/noise ratios show no degradation. The investigation focusses therefore
mainly on profile flatness and swirls.
Fig. 6.5.2-1: Historic profile flatness for both meter runs
Fig. 6.5.2-2: Historic profile swirl for both meter runs

Meter run one has always measured with a profile flatness > 90% and a positive swirl, while run two's profile
is more pointy, < 80% flatness with a negative swirl. A different flow profile for different metering runs does not
raise any immediate concerns, as both pipes have different flow paths. The specific flow profile for run one
and run two has been confirmed by all flowmeters, as three flowmeters rotate between two metering runs. The
flow profile anomalies are also not significant so that they should raise any immediate concerns. At an axial
velocity of 20 m/s, swirls in the range of +/- 2% corresponds to a tangential velocity of 0,4 m/s only.
It is also not contradicting that run one gives a higher flow than run two as its profile is flatter. A flatter flow
profile means that the top and bottom flow velocities are similar to the velocities in the core of the pipe, in
contrary to a pointy profile where the top and the bottom is lagging behind. An integration across the entire
pipe diameter results then into a higher average flow velocity for the profile which is flatter, when the core
velocity is similar.
The measured flow profile during change of export runs is no different than during export with a single run.
This increases the suspicion that the step change has always been present and further investigations
confirmed that run one has always measured more than run two, since start-up.
Fig. 6.5.2-3: Historic deviation against PPS venturi
Note that both runs were operated for a short period in 2011, resulting in a deviation from the PPS between
that of run one and two. Although the behaviour of both metering runs is strange, there is still no firm proof
that the meters are mis-measuring, as in theory, this effect could be caused by a higher friction or
obstructions in run two.
6.6 In-situ flow tests

All incidents and flowtests where meter runs were switched during export showed, without exceptions, a step
change of measured flow while all other process parameters were constant. An examination of the theoretical
flow was made to assess the possibility whether pipe friction/restriction could have led to the observations
made offshore.
Meter run 1 SN#2259 Meter run 2 SN#2257 PPS venturi

compressor venturi dP across metering station
670 2,9
660 2,8
650 2,7
dP across metering station [bar]

Mass flow [tons/h]
640 2,6
630 2,5
620 2,4
610 2,3
600 2,2
590 2,1
0:00 6:00 12:00 18:00 0:00 6:00 12:00 18:00 0:00 6:00 12:00 18:00 0:00
Fig. 6.6-1: Single export flow test
Flowtests where export was switched from duty to stand-by run did not give any clues on the root cause.
Pressure transmitters directly upstream or downstream of the export station are not available, the measured
dP includes therefore up- and downstream piping. It is not possible to calculate the theoretical flow explicitely
as the roughness of the pipe is unknown and the dP measurement is not isolated for the metering station.
However, based on the general flow equation, some qualitative assessements can be done nevertheless.
When the pipe length and the gas properties are equal, then the general flow equation (Finch, et al., 1988)
can be simplified to:
2 2 5 0,5
(𝑃 −𝑃 )𝑑
(Equ. 6.6-1) 𝑄̇ = 𝐾1 ( 1 𝑓2 )
Where:
𝑄̇ Flowrate [Sm3/d]
𝑃1 Pressure upstream [kPA]
𝑃2 Pressure downstream [kPA]
𝑑 Pipe diameter [mm]
𝑓 Darcy-Weisbach friction factor [-]
1
𝐾1 Constant for lenght, physical and base parameters [ ]
𝑘𝑃𝐴 𝑘𝑚
As the diameter and physical properties for run one and run two are identical, their difference in flow must arise
from a difference in pipe friction or differential pressure. Pressure measurements further up- and downstream
the meter station suggests that run one creates a lower dP than run two. If meter run one should indeed
measure more than run two, then the friction factor of run one must be significant lower than run two. Not only
in order to measure more than run two, but the effect of run one's lower dP must be overcome in addition. It is
strange that two identical pipes should have such a different friction factor. However, there is no proof that this
is not the case.
The parallel flow test in September 2013 marked a significant turning point in the investigations and revealed
that the export USMs are indeed mis-measuring.
PPS venturi compressor venturi Total station flow upstream pressure downstream pressure
700 150
149
690
148
680
147
line pressure [barg]

ton/h
670 146
145
660
144
650
143
640 142
07:30 07:58 08:27 08:56 09:25 09:54 10:22 10:51 11:20
Fig. 6.6-2: Parallel export test
Export from run two is switched to export with both runs and then to export with run one only. The test data
showed that run one measures most, followed by parallel export with both runs, while metering run two
measures least. Fluid dynamics dictate that this is impossible.
The qualitative assessment of the change from run one to run two suggested that the friction factor of run one
must be significant lower than run two, for the observations on the step change to become true. The friction
factor for parallel export must be between that of run one and two as the same pipes are used. All things being
equal, parallel export should deliver the largest flow due to the larger available pipe cross section, and not run
one. The upstream pressure actually drops significantly during parallel export, compensating the gains from
an increased diameter. This is expected when the pipe diameter is increased without increasing the
compressor output. The gas flows out more easily and the system balances at a new, reduced upstream
pressure. In order to establish the same conditions as during single run export, compressor load would have
to be increased to create a higher upstream pressure. That again would, naturally, lead to a higher flow during
parallel export than single run export.
Furthermore, flow peaks with ca. +3% deviation from the PPS venturi were observed when switching over from
run two to parallel export. This flow peak is reproducible and occurs every time when switching from run two
to parallel export. However, a similar peak cannot be observed when switching from run one to parallel export.
Neither the venturis or the dP across the metering station could confirm that this surge of flow has ever
happened. This increases the suspicion that the USMs are influenced by the different in-situ flow profiles, while
the Venturi meters are not affected.
The parallel test done in September 2013 was the first time when it was proved that the export flowmeters are
mis-measuring and suggests that different friction factors or flow restrictions for run one and two cannot be the
cause for the observations made offshore.
It was also observed that the characteristic profile flatness and swirls for run one and two have vanished
during parallel export.
Single run export Parallel export Onshore

calibration
Run two
Run one Run two (during Run one Run two -
peaks)
Profile flatness [%] 93 83 96 90 89 89 %
Swirl [%] +0,9 -1,8 +1,8 + 0,1 - 0,4 0,1 %
Profile symmetry [%] -1,0 +5,2 -2,9 + 0,1 + 1,3 1,6 %
PPS deviation [%] + 1,9 + 0,24 +3 + 0,7 -
Tab. 6.6-1: Typical profile parameters
Swirls seem to cancel each other out when exporting with both runs at the same time, which resulted in similar
flow profiles as observed during onshore calibrations. Although logically on the first hand, it is not selfevident
that the swirls of both runs will cancel each other out. Run two measures actually still a small amount of swirl
and assymmetry.
When considering table 6.6-1, a pattern between flatness and swirl can be observed. There is a correlation
between a positive swirl, increased flatness and increased flow. Although it is logical that an increased flatness
coincide with a higher flow, there is no logic that a positive (hence, clockwise) swirl should generate more flow
than a negative swirl. The swirl direction should not influence the measured flow, only the amount of swirl.
A further parallel test was done in October 2013, this time flowmeter SN2257 was replaced with SN2258 to
test whether phenomenons like the step change and profile peaks could be repeated with a different flow meter
in meter run two. SN2258 in run two behaved exactly as predicted and showed step change and profile peaks
like SN2257. Both meter runs have been put into operation after this test, as it is expected that the flowmeters
will measure correctly when experiencing a similar flowprofile as during the calibration onshore.
It is now evident that the export meters are mis-measuring and that the error is related to all flowmeters
regardless where they are installed. The investigations was handed over to CMR at this point, as
computational fluid dynamic simulations and a detailed knowledge of the proprietary flow algorithm was
required in order to find the cause for the observations on-site.
7 Conclusion
A mis-measurement on Gjøa's gas export was detected in early 2013. The reason for the mis-measurement
was an incorrect swirl compensation of the flowmeters caused by a wrong path angle configuration for the
bottom two paths. As a consequence, tangential velocity components were not compensated for but instead
increased, causing the meter to overmeasure with positive swirl and undermeasure with negative swirl. The
gas export has been mis-measured since start-up, since the flowmeter configuration was part of the factory
configuration. Although this incident demonstrates that even very small errors can accumulate to a significant
mis-measurement over time, the message of this paper is another.
The reason for the profile peaks is still not known, but subsequent CFD simulations carried out by CMR showed
that the blind tee upstream might be the cause for the instable profile peaks in run two. This is explored in
detail in another NSFMW paper (Sætre, et al., 2015). Lessons learnt from this incident may include the
prevention of blind tees, improved location for the flow conditioner, installation of two meters in series with
redundant technology and a detailed review of the flowmeter performance during the ISO 17089 type test.
Gjøa's export flowmeters are frequently traceable calibrated to ensure that they have not drifted more than
0,3% between calibrations. In addition, traceable flow laboratories are often thought to be the basis to compare
against to ensure that the flowmeters are measuring "correctly". USM flow calibrations prove the measured
flow for exactly one flow profile only, the flow profile during calibration. Even when calibrating the meter with
up-/downstream pipespools and flow conditioner may not solve the problem, as the profile entering the spool
is different between calibration and on-site. The transferability of the traceable calibration is dependent on the
results of the ISO 17089 type test, as the profile in-situ will be different than during calibration. Calibrations so
far did not show any evidence on any drifting of any flowmeter.
The ISO 17089 type test gives a hint on how the meter performs when exposed to a non-ideal profile. A
flowmeter passes the ISO type test if it can prove that its FWME has not changed more than 0,3% due to a
variation of upstream perturbations. Operators need to resist the urge to think of the ISO 17089 type test as a
guarantee that the flowmeter performs under all upstream configurations, but rather consider it as an
uncertainty budget when the expected profile is unknown. However, when the flowmeter is installed, it is
subject to one single flowprofile only, as piping configurations do not change and profile non-idealities tend to
be stable due to the highly turbulent flow. Therefore, any non-compensated profile anomaly will result in a
systematic mis-measurement.
An analogy can be drawn to venturis and liquid USMs, which are calibrated by associating different meter
factors to different Reynolds numbers, as their performance are strongly dependent on it. This is not relevant
for gas USMs, as gas flow is usually highly turbulent due to the low viscosity and the high pipe velocity. Instead,
the flow profile is of uttermost importance for the operation of gas ultrasonics. One can generally expect an
USM to perform similarily when exposed to a similar flowprofile. So why are gas USMs not calibrated against
their flow profile by associating a different meter factor to each flow profile?
One reason could be the flow dynamics in turbulent gas flows. Their impact on gas ultrasonics is multi-
dimensional, in contrary to venturi meters and liquid USMs, and a gas flow profile is described by many factors
such as flatness, swirl, cross-flow, symmetry and their different combinations. It is therefore not a trivial task
to assign correction factors to every possible flow profile which could occur.
Operators need a better understanding on how their USMs cope with non-ideal flow profiles. How does non-
ideal flow profiles influence their performance? What are the limits under which the USM can be operated
without mis-measurements?
The performance of USMs, when exposed to a non-ideal profile, is still not fully understood and requires
further investigations. A critical analysis of the flow profile and its impact on the measurement has to be
done, once non-idealities are experienced during operations. The transferability of calibrations needs to be
questioned, once the flow profile in-situ is different from the profile measured during calibration.
1 Bibliography
Finch, J.Christopher and Ko, David W. 1988.
Fluid Flow Formulas. 1988.
FMC Technologies 2008.

User Manual.
GDF SUEZ E&P NORGE 2010.

Internal project documentation.
International Organization for Standardization 2003.

ISO5167-4: Measurement of fluid flow by means of pressure differential devices inserted in circular cross-
section conduits running full - Part4: Venturi tubes.
International Organization for Standardization 2010.

ISO17089-1: Measurement of fluid flow in closed conduits - Ultrasonic meters for gas - part 1: Meters for
custody transfer and allocation measurement.
Norsk Oljemuseum 2015.

Petroleumskartet. [Online] [Cited: 29 09 2015.]
http://www.petroleumskartet.no/.
Norwegian Petroleum Directorate 2015.

Factpages. [Online] [Cited: 29 09 2015.]
http://factpages.npd.no/FactPages/default.aspx?nav1=field&nav2=PageView.
Shell 2015a.
FLAGS asset fact sheet. [Online] [Cited: 29 09 2015.]
http://www.shell.co.uk/business-customers/upstream-oil-and-gas-infrastructure/segal-system.html.
Shell 2015b.
SEGAL Overview Schematic. [Online] [Cited: 29 09 2015.]
http://www.shell.co.uk/business-customers/upstream-oil-and-gas-infrastructure/segal-system.html.
Statoil 2015.
Gjoa illustration. [Online] [Cited: 29 09 2015.]
http://www.statoil.com/no/OurOperations/ExplorationProd/partneroperatedfields/Gjoea/Pages/default.aspx.
Sætre, Camilla, et al. 2015.

Installation Effects and Flow in-stabilities in a Gas Metering Station with Multipath Ultrasonic Flow Meters".
2015.
20. – 23. October 2015
Installation effects and flow instabilities in gas metering

station with multipath ultrasonic flow meters
Camilla Sætre1,
Kjell-Eivind Frøysa1, 3,
Anders Hallanger1 and
Philip Chan2
1
Christian Michelsen Research AS, Bergen, Norway
2
GDF Suez E&P Norge AS, Stavanger, Norway
3
Bergen University College, Norway
ABSTRACT
This paper will address the analysis carried out in order to understand and correct flow
meter effects on the Gjøa gas export metering station.
The metering station consists of two parallel metering tubes, each equipped with a
multipath ultrasonic flow meter with an upstream flow conditioner. The header
upstream of the two metering tubes has T-bends at the end sections. Through
Computational Fluid Dynamics (CFD) it will be shown that the T-bends in the upstream
header can cause severe distortions of the flow profiles, and that the positioning of the
flow conditioners can be essential for preventing flow instabilities. It will be
demonstrated that both the pipe geometry upstream the inlet header and the geometry
of the inlet header itself may affect the flow instability through the flow meters. The
results will to a large extent explain the effects found in practice.
Furthermore, it will be demonstrated how errors in the meter configuration file can
affect the meter performance, even in cases where flow calibration has been carried
out successfully.
1 INTRODUCTION
This work was based on the mis-measurements for the Gjøa gas export, which were
discovered by Engie due to difference in flow rates when switching from one flow line
of the metering station to the other. A thorough description of the observations is given
in the paper of P. Chan et al. [1]. The difference in the measurements of the ultrasonic
meters (USM) for the two parallel lines was initially assumed caused by installation
effects. The metering response to possible installation effects were investigated based
on the use of computational fluid dynamics (CFD) model.
The work also included the evaluation of the effect of wrong input to configuration files
of the USMs. Additionally, it was evaluated whether the mis-measurements caused by
the wrong input could be corrected within fiscal requirements based on the measured
path velocities from the USMs.
The CFD model and analysis of the USM response to the modelled flow profiles are
presented in Chapter 2. The analysis of the effect of wrong input to the configuration
files of the USMs are presented in Chapter 3.
1
20. – 23. October 2015
2 CFD MODEL AND RESPONSE OF INSTALLATION EFFECTS ON USM
In this chapter the installation effects on the ultrasonic measurement system are studied.
The CFD model and analysis of the USM response to the modelled flow profiles are
presented.
2.1 Measurement principle
The measurement principle for the gas export at Gjøa is flow measurements from
ultrasonic flow metering (USM). For Gjøa the USMs are of the type MPU 1200
ultrasonic meters from FMC, consisting of six ultrasonic transmitter pairs. These will
define six acoustic paths, each with an inclination angle relative to the flow and a
defined distance from the centre of the pipe.
The acoustic paths numbered 1 and 2 are at the same plane with equal distance from
the centre of the pipe. Likewise for the acoustic paths numbered 3 and 4. The acoustic
paths numbered 5 and 6 are on the other hand at different distances from the centre of
the pipe. Figure 2.1 illustrates how the inclination angles of the acoustic paths are
aligned when the pipe is seen from above.
Figure 2.1 The MPU 1200 acoustic paths seen from above
The measurement principle of the USM is based on the measured six flow velocities,
or path velocities, v1, v2, v3, v4, v5, and v6. To these flow velocities, there are applied
six weight factors, w1, w2, w3, w4, w5, and w6. The average axial flow velocity in the
pipe is calculated from the combination of these. The volume flow rate is calculated
from the average axial flow velocity and the cross sectional area of the metering pipe.
Additional measurements are temperature and pressure.
For the MPU 1200 flow meter, the measured flow velocities are also used to calculate
four quality parameters for the flow profile. These parameters are
- Profile flatness: a quantitative description of the distribution of axial flow in the
pipe. High flatness number means that the axial flow velocity in the centre of
the pipe is not significantly larger than the velocity off-centre. Low flatness
number indicates that the flow velocity is largest in the centre of the pipe.
- Profile symmetry: a quantitative description how symmetric the flow velocity
is with respect to the centre of the pipe. If the profile symmetry is close to zero,
the flow is symmetric in the pipe cross section.
2
20. – 23. October 2015
- Swirl: a quantitative description of the tangential flow velocity relative to the

axial flow velocity in the pipe. For swirl, the rotational flow will form one
rotation cell seen in the cross section of the pipe.
- Cross flow: similar to swirl, but describes a rotational flow forming two cells
seen in the cross section of the pipe.
2.2 Observations
The mis-measurements for the Gjøa gas export were discovered by Engie due to
difference in flow rates when switching from one flow line of the metering station to
the other. The meter shift could not be explained by operational issues, and the shift
was not seen on comparison with upstream and downstream Venturi meters. The
observations are presented in the paper of P. Chan et al. [1], with the conclusion that a
non-ideal profile can cause a bias on the measurements and lead to significant errors of
the measurements over time.
In order to clarify the reasons for the observed differences between the two flow meters,
Engie initiated a study involving CFD simulations and a subsequent ultrasonic metering
analysis.
During single operation of run 2, shorter time periods of increased flow rate were
observed. These events are referred to as peaks. In addition to increased flow rate, the
observations also showed flow profile changes in swirl, symmetry, and flatness values.
These effects were also investigated in the CFD simulations.
2.3 CFD
The CFD code MUSIC is used in the simulations. MUSIC is an in-house finite volume
code developed by CMR, and has been used in simulations of pipe flow [2].
In the simulations, equations for momentum (Navier-Stokes) and pressure are solved
together with the equations for the industry standard k-ω turbulence model. With an
average flow velocity of 20 m/s the Reynolds number is 4·107. Due to these high
Reynolds numbers, the turbulent boundary layers at the pipe walls are very thin. The
wall friction is therefore modelled with wall functions.
In the simulations, the gas is considered to be incompressible.
2.4 USM simulation
The flow profiles simulated by the CFD code are interpreted by the CMR developed
program USMSIM, for virtual ultrasonic flow metering in CFD calculated flow profiles,
used to simulate the readings from the MPU 1200 flow meter. Here the average flow
velocities are calculated using the actual acoustic path location and integration
weighting factors and the algorithms used in the MPU 1200 software. In this way, the
effect of swirl on the flow meter volume flow rate output, is found.
2.5 Simulations and analysis
The initial objectives for the CFD analysis were:
3
20. – 23. October 2015
- Investigate the effect of swirl on the volume flow reading compared to fully
developed flow pattern for the gas export metering station on Gjøa. Calculate
the corresponding profile flatness and profile symmetry.
- Simulation of specified flow profiles with simplification of piping configuration
upstream the position of the flow meter in order to replicate these flow profiles.
This part of the study was intended for an off-site calibration with replication of
the observed flow profiles at Gjøa.
- Simulation of the flow pattern at the location of the USM for the given piping
configuration with the existing flow conditioner at Gjøa gas export metering
station. Simulation of the effect on the volume flow reading from these flow
patterns compared to fully developed flow pattern.
- Simulation as listed in the previous point, with an optimised location of the flow
conditioner. Investigate whether a re-location of the flow conditioner can lead
to a fully developed flow profile.
The initial objectives were adjusted according to the findings of incorrect input to the
configuration files of the USMs. The simulations were performed based on a
configuration with correctly implemented inclination angles of the ultrasonic
transducers, and thereby the results would apply to the corrected and future operation
of the Gjøa metering station for export gas.
2.5.1 Simulation geometries
The simulations were based on the following geometries:

- An arbitrary two-bends-out-of-plane geometry with regular bends for the
investigation of swirl effects in general for USM measurements
- Similar to the existing installation geometry, but with regular bends instead of
T-bends at the header.
- Similar to the existing installation geometry of the export gas metering station,
o without flow conditioner
o with flow conditioner positioned as is
o with flow conditioner in a new position closer to the flow meter
The different simulation geometries are illustrated in Figure 2.2 and Figure 2.3. The
distance between the flow conditioner and the flow meter is 15 D (4.465 m).
Flow direction
Flow direction
Figure 2.2 Single operation simulation geometry, simplified metering station with
regular bends. The arrows indicate the direction of the flow. The upper part of the figures
show the upstream piping. Left: single export run 1. Right: single export run 2.
4
20. – 23. October 2015
Figure 2.3 Simulation geometry of the metering station with T-bends at the header,
parallel operation.
2.5.2 Simulation parameters and reference base case
The simulation parameters were based on the typical values of the USM gas export
metering station:
- Velocity 9.5 and 19 m/s
- Pressure 149 bar
- Temperature 51.5 °C
- Density 139.3 kg/m3
- Viscosity 1.86·10-5 kg/ms
- Pipe spool diameter 288.53 mm
- Reynolds numbers 2·107 and 4·107
The simulations referred to as Base case, are for fully developed flow in straight pipe.
The velocity profiles derived from the MUSIC CFD simulations are symmetric axial
velocities, and transversal velocities approximately equal to zero. The USMSIM
analyses of MPU 1200 calculates for the Base case a flatness as function of Reynolds
number, symmetry of 0.001 %, swirl of 0 %, and cross-flow of 0 %. Initially the
simulations were performed assuming smooth walls (roughness of 0.01 mm). For
Reynolds number in the range of 2-4·107, the flatness of developed flow in this case
will be approximately 95 %. This is considered to be a high value, and the simulations
were repeated with higher roughness values. The roughness value is equivalent to the
grain size of the sand corns on a sand paper coating the inner walls of the pipe. In the
calibration tests, the reported profile flatness from the MPU 1200 was typically 90 %
and constant over the meter velocity range. Test simulations in a very long straight pipe
giving developed flow profiles were carried out with different wall roughness. The flow
profiles were then run through USMSIM. Based on the evaluation of flatness for
different types of roughness values, the simulations were performed with an assumed
coefficient of roughness set to 0.3 mm.
5
20. – 23. October 2015
2.5.3 Simulation results: piping configuration with regular bends
Flow simulations were carried out for simple piping configuration in order to
investigate effects of swirl on the USM flow meter. These simulations were compared
to the fully developed flow pattern, referred to as base case. The effect of swirl on the
volume flow depends on the orientation of the swirl relative to the meter. Hence, a
specific swirl value will not have a unique effect on the volume flow reading. From the
simulations, an absolute swirl value of 2 % would have an effect spanning over 0.3 %
on the volume flow reading, depending on the orientation of the swirl relative to the
meter.
Simulations were performed with regular 90° bends instead of T-bends in the header of
the flow metering station. The upstream piping configuration were kept similar as is,
with a clock-wise snail house orientation.
Figure 2.4 shows the simulated flow velocities in the horizontal cross section of the 90°
bend and first section of run 2, single operation. The initial velocity of the simulations
were 19 m/s. For a configuration with regular bends, there is no sign of recirculation
zones in run 2, neither in run 1 during single operation.
Flow Direction
Figure 2.4 Flow velocities in the entrance of run 2 single operation, regular 90°
bends. The flow pattern is shown in a horizontal plane through the meter pipe axis. It is
shown as vectors at each grid point. Blue means low flow, yellow and red means higher
flow.
2.5.4 Simulation results: piping configuration with T-bends in the header
Simulation of the metering station as is, with T-bends at the header, but without the
flow conditioner, showed the following:
- Parallel export: Recirculation zones with negative flow velocities in part of the
pipe cross-section downstream of the T-bends of the header. Especially
prominent for run 2. The recirculation zone for run 2 extends over 1 m
downstream of the manifold. The flow conditioner is at 1.8 m from the
manifold. For run 1, the simulated swirl is 4 %, and for run 2 the simulated swirl
is -5 % at the flow meter position. Note that these simulations were performed
without flow conditioner. Simulated swirl are significantly larger than observed,
as expected, since the flow conditioners should smooth the flow and reduce
swirl.
6
20. – 23. October 2015
- Single export through run 1: Recirculation zones after the T-bends of the header
are more prominent than for parallel export. At the position of the flow
conditioner, 1.8 m from manifold, the flow profile symmetry parameter
indicates an asymmetric flow profile.
- Single export through run 2: Recirculation zones after the T-bends, as for single
export run 1. Disturbed flow with significant asymmetric flow is present also at
the position of the flow conditioner. Note again that these simulation results are
performed without flow conditioner.
The flow recirculation, which was found in the simulations of the metering station with
T-bends, was not found in the simplified geometry with ordinary 90° bends (see Figure
2.4). The flow recirculation region and the flow profile further downstream (1.8 m from
the manifold) appear to be quite different in run 1 compared to run 2.
The main difference between run 1 and run 2 from the results of the simulations of flow
without flow conditioner: The flow profiles at the position 1.8 m from the manifold are
in a way inverse of each other, where the flow at a position in the cross section of the
pipe in one of the run is low, it is high in the other run, and vice versa.
2.5.5 Simulation results: piping configuration as is with flow conditioner
Simulations were performed of the metering station as is with flow conditioner present
1.8 m from the manifold. The three-dimensional flow profile, including flow effects
like swirl, was found, and the equivalent measurements of the flow meter were found
by use of the USMSIM program. The distance between the flow conditioner and meter
is 15 D.
The flow conditioner (FC) is a CPA (Canadian Pipeline Accessories) plate with 25
holes. The pressure build-up on the upstream side of the plate will act to redistribute
skew axial flow profiles. The flow will emerge from the holes on the downstream side
as jets from each hole. The high turbulence levels will quickly mix the jets and give a
developed profile. It is, however, possible that the asymmetries of the flow can
propagate through the flow conditioner and give a reduced symmetry on the
downstream side.
The simulations showed recirculation regions, with flow in reverse direction in part of
the pipe cross section, at the beginnings of the meter pipes as in the simulations of the
meter station without the flow conditioners. This is believed to be a result of the T-
bends between the manifold and the meter pipes. The recirculation is shown in Figure
2.5, for single export in run 1, and in Figure 2.6, for single export in run 2. The flow
pattern is shown in a horizontal plane through the meter pipe axis. It is shown as vectors
at each grid point of the CFD simulations. Blue colour means low flow, whereas green
and yellow mean higher flow, up to a maximum at yellow of 58 m/s. The recirculation
zone with reverse flow is marked within the red line in the plot. The simulation results
indicate that the nature of the recirculation region and the flow profile downstream that
region is quite different in run 1 compared to run 2 when single export.
The simulation results with the flow conditioner as positioned in the metering station,
show that the swirl and cross flow are close to zero for all orientations of the USM
relative to the flow. The deviation from reference simulations of fully developed flow,
7
20. – 23. October 2015
is estimated to be between 0 and 0.3 % for single export through run 1, and between -
0.1 and 0.4 % for single export through run 2.
The simulated profile symmetry, flatness, and deviation varies with the orientation of
the flow profile relative to the USM.
0.5 m from manifold 1.0 m from manifold
Flow Direction
T-Bend 0.5 m from 1.0 m from Flow

Outlet manifold manifold conditioner
(manifold)
Figure 2.5 Simulation results for run 1 single export. Upper part: axial flow profile
0.5 m, 1.0 m and 1.8 m downstream the manifold. Lower part: Flow velocities in run 1
between the manifold and the CPA-plate (flow conditioner). The flow pattern is shown in
a horizontal plane through the meter pipe axis. It is shown as vectors at each grid point.
Blue means low flow, green and yellow means higher flow. The recirculation zone (with
reverse flow) is marked in red.
8
20. – 23. October 2015
0.5 m from manifold 1.0 m from manifold
Flow Direction
T-Bend 0.5 m from 1.0 m from Flow

Outlet manifold manifold conditioner
(manifold)
Figure 2.6 Simulation results for run 2 single export. Upper part: axial flow profile
0.5 m, 1.0 m and 1.8 m downstream the manifold. Lower part: Flow velocities in run 2
between the manifold and the CPA-plate (flow conditioner). The flow pattern is shown in
a horizontal plane through the meter pipe axis. It is shown as vectors at each grid point.
Blue means low flow, green and yellow means higher flow. The recirculation zone (with
reverse flow) is marked in red.
The simulated flow pattern at the location of the USM for the given piping
configuration with the existing flow conditioner shows a low value of the swirl and
cross flow for both lines, single operation. The other flow parameters vary with the
orientation of the flow profile relative to the USM. This variation seems to be more
prominent in run 2, single export. The deviation from reference volume flow reading is
in the range of 0.3 to 0.4 % for both lines based on simulation of the metering station
as is. Figure 2.7 shows the simulation result of the axial velocity profile at the position
of the meter for single export for run 1 (left) and run 2 (right).
Figure 2.7 Axial velocity profile at position of meter, simulations of metering station
as is, with flow conditioner. Left: Single export, run 1 open, Right: Single export, run 2
open.
9
20. – 23. October 2015
The simulations showed the occurrence of a recirculation zone between the T-bend
header and the flow conditioner. There are major differences between the flow profiles
upstream of the CPA-plate in run 1 compared to run 2. It might be that this causes the
observed peaks in run 2, as this region probably varies with the variations in the flow.
The CFD simulations may have overestimated the effect of the flow conditioner plate
since both swirl and cross flow appear to be almost zero at the meter position, for both
lines. However, the flow profiles upstream of the flow conditioner more likely indicate
that there is a large flow profile difference between run 1 and run 2.
2.5.6 Simulations of a possible optimum location of the flow conditioners
The simulations of the metering station as is were repeated for other locations of the
flow conditioner in order to search for a more optimal location of the flow conditioner.
The results presented here are with the flow conditioner positioned 8.5 D upstream of
the USM. The initial distance between the flow conditioner and meter is 15 D.
The flow velocities between the manifold and the CPA-plates (at new position) are
shown in Figure 2.8 and Figure 2.9. The recirculation at the entrance to the meter pipes
persists as in the simulation with the CPA-plates at the present location. However, the
flow will tend to smooth out and become more uniform upstream of the flow
conditioner plate with the modified flow conditioner position. This may possibly lead
to less occurrences of peaks, which are observed in run 2.
Figure 2.8 Flow velocities in run 1 single operation between the manifold and the
modified position of the CPA-plate. The vectors are shown in a horizontal plane through
the meter pipe axis. Similar plot as Figure 2.5, but with flow conditioner 3.8 m after the
manifold.
Figure 2.9 Flow velocities in run 2 single operation between the manifold and the
modified position of the CPA-plate. The vectors are shown in a horizontal plane through
the meter pipe axis. Similar plot as Figure 2.6, but with flow conditioner 3.8 m after the
manifold.
10
20. – 23. October 2015
Figure 2.10 Axial velocity profile at position of meter, simulations of metering station
as is, with flow conditioner in new position 3.8 m from header. Left: Single export, run 1
open, Right: Single export, run 2 open.
The simulated flow pattern at the location of the USM for the given piping
configuration with the existing flow conditioner shows a low value of the swirl and
cross flow for both lines, single operation. The other flow parameters vary with the
orientation of the flow profile relative to the USM. This variation is lower than the
results when the flow conditioner was placed 1.8 m from inlet (as is). This may indicate
that a positioning of the flow conditioner closer to the flow meter gives a less disturbed
flow pattern than the installation as is. Figure 2.10 shows the simulation result of the
axial velocity profile at the position of the meter for single export for run 1 (left) and
run 2 (right), when the flow conditioner is placed 3.8 m from header.
The recirculation zones between the T-bend header and the new positioned flow
conditioner, seem from simulations to be more smooth also for run 2, and the flow is
more uniform upstream of the plates with the modified position. Although the
simulations at hand did not show any peaks, since the simulations run were stationary,
it is found probable that a new location of the flow conditioner further away from the
T-bend header will provide more stable flow conditions.
3 CONFIGURATION ERRORS, EVALUATION OF EFFECTS
In this chapter, the effect of wrong input for the configuration files of the flow meters
is discussed, and the calculated results expected for a correctly implemented flow meter
for Gjøa gas export are presented.
3.1 Configuration files – error in parameter values
The flow meters installed at the Gjøa gas export station, measure six flow velocities, or
path velocities. These six path velocities are combined with weight factors specified for
the meter in order to calculate the average axial flow velocity.
For the meters at Gjøa gas export metering station, the configuration of the inclination
angles of the sound paths were correct for the double set transducers of acoustic paths
1, 3, 2 and 4. For the two lower sound paths, however, the inclination angles in the
configuration file were implemented with opposite sign than the actual design of the
meter.
11
20. – 23. October 2015
For the flow profile parameters, flatness and symmetry will depend on the inclination
angles and weight factors defined in the configuration of the meters. Whereas for the
calculated swirl value, cross flow value and velocity of sound, there are close to no
dependency of the specified inclination angles and weight factors for the two lower
acoustic paths.
The consequence of the wrong sign on the inclination angles for acoustic paths 5 and 6
is that the correction of transversal flow components, that is swirl and cross flow, was
performed with the wrong sign.
3.2 Analysis of effect of error
In the analysis of the effect of error in the configuration setup of the meters, an
evaluation of the metering results was calculated as if the correct angles had been used
in the software.
Figure 3.1 shows an example of single export in run 2, with the flow velocity measured
by the meter with wrong inclination angles for the two lower acoustic paths in orange,
and how the flow velocity would have been if correct angles were implemented, shown
in blue. For the example presented here, there is a peak in the time series of the observed
flow velocity. Note that with the incorrect angles as implemented, this jump in flow
velocity was in the range ~0.3 to 0.4 m/s. Now, if correct angles were implemented the
jump in flow velocity would have been in the range ~0.1 to 0.2 m/s. Thus, the effects
of the peaks in run 2 are not as severe with correctly implemented inclination angles.
Figure 3.1 Example of the implications of the wrong inclination angles, single export
run 2. Y-axis displays the flow velocity in m/s, and x-axis is time. In
orange: The metering result as implemented. Blue: The metering result if
correct angles had been used.
Figure 3.2 shows the deviation between the measured total flow velocity as
implemented relative to how it should have been with correct angles for single export
in run 2. For this example, two peaks were apparent in the observed flow velocity. In
general, for this example, the flow velocity was underestimated with approximately -
0.6 % during regular flow, and overestimated by approximately 0.4 % when peaks
12
20. – 23. October 2015
appeared in the flow. Note that these calculations are performed based on the measured
flow velocity for each acoustic path and their correctly corresponding weight factors.
The influence of calibration factor on the measurement results when the calibrations
were performed with errors in the configuration files is not included here.
Figure 3.2 Deviation between measured total flow velocity with angles as
implemented relative to correct angles. Example with single export in run
2.
A detailed analysis of the flow parameters derived from the USM measurements,
indicated that the flatness was more stable and at a higher value than for the
measurements with wrong inclination angles for the two lower acoustic paths. There
were no signs of change in the flatness value during peaks. The symmetry was also
more stable and with a lower value than for the result as implemented. The peaks are
shown as dips for the symmetry value. However, the amplitude of the dip relative to
the stable level is significantly reduced when correct angles are used in the calculations.
Calculated flow parameter values for swirl and cross flow are not affected by the
switched sign in inclination angle for the two lower paths.
3.5 Configuration errors – what about flow calibration?
The flow meters were calibrated with the incorrect inclination angles. How does this
affect the measurement results of the separate axial flow velocities for each of the six
acoustic paths?
If there are swirl and cross flow present during calibration, although minor, the effect
would have been enhanced by the opposite sign of the inclination angles. Investigation
showed that with a swirl value of 0.10 % and cross flow of -0.12 %, the result would
be a 0.26 % shift in the calibration K-factor. If the swirl values were as low as 0.02 %
and cross flow -0.04 %, the shift would have been 0.09 %.
Hence, the effect of shift from the calibrations are also included in the correction of the
flow measurements in order to meet the fiscal requirements.
13
20. – 23. October 2015
4 CONCLUSIONS
This paper describes the evaluation of installation effects and flow instabilities in the
Gjøa gas metering station, which consists of two parallel runs with multipath ultrasonic
flow meters. The analysis was carried out based on Computational Fluid Dynamics
(CFD) with additional calculation of the metering results of the flow meters in use. The
study also covered the evaluation of errors in the meter configuration file, how it
affected the meter performance and factors derived from calibration.
The results of the CFD analysis of the metering station geometry, with the upstream
conditions equivalent to the Gjøa gas export metering station, showed that the geometry
of the header and pipe geometry upstream the header can affect the flow stability
through a flow meter. The positioning of the flow conditioner can therefore be of high
importance.
The limitation of space on the metering site might require the upstream piping to be not
ideal for the metering principle at hand. However, with an evaluation of the expected
effects on the flow geometry at the positions of the meters, the functionality of a six
beam ultrasonic meter may be optimized with an evaluation of the upstream header and
positioning of the flow conditioner.
The measurement errors caused by errors in the flow meter configuration file, can be
corrected for based on the initial ultrasonic measurements for each acoustic beam. This,
naturally, will rely on the frequency of which the original ultrasonic measurement data
are stored. The evaluation based on the initial axial flow velocities measured for the six
acoustic beams, showed that the ultrasonic flow meter provides acceptable fiscal
measurements of the flow for the Gjøa gas export.
5 ACKNOWLEDGEMENT
The authors would like to gratitude FMC Technologies, and especially Atle
Abrahamsen and Sveinung Myhr, for valuable discussions and for the permission to
use the integration weights and the algorithms implemented in the MPU 1200 meter in
this project. Acknowledgement is also given to Reidar Sakariassen, Metropartner, for
valuable discussions and input throughout the work.
6 References
[1] P. Chan, Ø. Storli, S. Myhr, R. Sakariassen, A. Abrahamsen, K.-E. Frøysa, C. Sætre, "In-
situ effects on ultrasonic gas flowmeters," in 33rd International North Sea Flow
Measurement Workshop, 2015.
[2] A. Hallanger, K.-E. Frøysa , P. Lunde , "CFD simulation and installation effects for
ultrasonic flow meters in pipes with bends," Int. J. of Applied Mechanics and Engineering,
vol. 7, no. 1, pp. 33-64, 2002.
14
20 – 23 October 2015, Norway
Realizing a class 1 Ultrasonic Gas Meter without HP calibration – An innovative

approach as contribution to an economic upstream metering solution
Sebastian Stoof, SICK AG, Germany

Dr. Andreas Ehrlich, SICK AG, Germany
Dr. Volker Herrmann, SICK AG, Germany
1 Introduction
Diversification in gas production worldwide with an increasing production of unconventional
natural gas, such as shale gas, leads to an increasing amount of gas wells. For the purpose of
production control and well monitoring, the gas production from these wells is typically
measured by a gas flow meter close to the wellhead or at the gathering station while the gas
may contain liquids and contaminants due to limited gas treatment equipment on a wellhead
skid. In addition, there is an urgent need for high rangeability, since the output of the well
changes over time and is generally not predictable. Since the amount of wells significantly
increase with shale gas and coal seam gas production, there is also a need for high availability
and low maintenance equipment.
Gas metering points in this production environment are mainly covered by differential
pressure meters driven by two major aspects – their robustness against liquids and their
economic pricing. New technology gas meters like ultrasonic meters provide several
advantages over differential pressure technologies such as a greater rangeability, strongly
reduced pressure drop and extended diagnostics. However, ultrasonic meters usually mean a
higher capital expenditure to the gas companies compared to differential pressure technology.
This applies specifically for small diameter gas production pipelines. Even though the lower
operational expenditures of ultrasonic meters may compensate for the higher initial invest,
ultrasonic meters for gas production applications have not yet become accepted on a larger
scale.
In order to overcome this issue, SICK presents an innovative approach for an ultrasonic gas
production meter that provides high reliability, high rangeability and class 1 uncertainty in dry
conditions (acc. ISO 17089-1) without need for an individual high pressure natural gas
calibration. High pressure calibration with the related logistics usually means a 30-50% price
adder for small size meters compared to the meter itself. The main factors that must be kept
under control by the meter manufacturer are minimized manufacturing tolerances in order to
ensure systematic low-pressure to high-pressure test transferability.
The paper describes the technical challenges that have been overcome in order to ensure a
proper meter performance in high pressure natural gas without high pressure calibration.
Both, low pressure and high pressure calibration lab results are presented as well as wet gas
test results with water and oil components. Additional measures make the meter highly robust
even against high LVF and provide a reliable liquid presence detection possibility by using
the meter diagnostics. This results in an accurate (in dry-conditions) and economic upstream
ultrasonic meter with large turn-down that indicates the presence of liquids (with reduced
accuracy of readings) on a real time basis.
2 Initial considerations
Dynamic (wet) calibration has become a standard for Class 1 gas flow meters in order to
ensure a high level of measurement accuracy in the field. ISO17089-1 recommends to choose
calibration conditions such as pressure, temperature and flow rate similar to the designated
operating conditions of the meter and to include upstream and downstream piping with flow
conditioners whenever possible. [1]
The type of gas, pressure and temperature clearly have an effect on the flow profile inside
ultrasonic meters and thus on the calibration result. Those effects are largely understood while
influences from geometric meter tolerances, upstream piping characteristics and effects from
flange connections have been less investigated. Usually these effects are neglected during
flow calibration of ultrasonic meter packages. That’s why there is a clear trend to calibrate
meter packages in order to reduce the influencing factors that may cause a higher uncertainty
of measurement in the final installation.
Ultrasonic meters that meet class 1 performance requirements, are mainly multipath ultrasonic
meters with 3 or more paths for custody transfer applications. While meters for gas production
applications have to be robust, provide highly available readings with good reliability and
must be economic in CapEx and OpEx. Additionally, the gas stream in gas production
applications may contain contaminants and liquids where 4-path meters have not shown an
advantage in accuracy over dual-path meters. [2] Based on these considerations, a decision
has been taken for a more economic dual-path concept that should meet Class 1 performance
requirements. For this, it is necessary to understand the effects of changing operating
conditions on the meter performance and to identify the main influencing factors to keep them
down to an uncritical level of variation.
The main challenges for the realization of the project were:
• Create a dual-path meter with high accuracy and reproducibility

• Reduce flow profile variations from manufacturing tolerances and installation effects
• Develop a factory test procedures to guarantee for class 1 performance
3 Meter design
FLOWSIC600 has shown its accuracy and long-term stability in custody transfer and gas
storage applications. Thus, sensor technology, electronics and manufacturing standards have
been taken from FLOWSIC600 to ensure a thorough technical basis for the dual-path meter.
It has been shown that upstream piping conditions like steps and wall roughness do not have a
significant effect on the accuracy of multipath custody ultrasonic flow meters [3, 4].
However, it can be assumed that their effect on dual path meters is more significant. This has
been confirmed by the initial flow tests with a welded design for the initial upstream meter
design.
Typical variations that increase the uncertainty of a dual-path ultrasonic meter are:
• Misalignments of flanges due to tolerances in flange connections

• Gaps between connection flanges
• Influences of weldings on pipe roundness
• Manufacturing tolerances of upstream piping
The influence of these variations on the flow profile and reproducibility of the dual-path
meter are decreasing with increasing distance to the measurement section. Therefor the
upstream meter section has been increased to a length of 10*nominal diameters. This
increases the distance to flow disturbances from installation effects and allows the meter
manufacturer to control the upstream pipe conditions close to the measurement section. The
meter design as shown in figure 1 has also proven to be very beneficial in wet gas conditions.
[5]
Figure 1: FLOWSIC600 DRU

Another measure to increase insensitivity of a dual-path meter to the specifics of an actual
installation is to make the meter more robust against variations in the flow profile even in
undisturbed conditions e.g. optimize the linearity of the meter with respect to Reynolds
number. For symmetrical profiles in turbulent flows, changes are mainly caused by a change
of the Reynolds number. Gätke [6] has defined a specific point in the flow profile, which
shows minimal dependency on profile variations. This point is located in 0.4*radius distance
to the pipe wall. (Figure 2) FLOWSIC600 DRU ultrasonic paths are positioned at this
0.4*radius position what makes the meter less affected by flow profile variations.
Figure 2: Calibration factor at turbulent flow [6]

A flow conditioner upstream of the meter is required to get a stable and symmetrical flow
profile and ensure measurement performance also with typical perturbations like elbows in
the upstream piping. The type of flow conditioner has been chosen under consideration of a
proper flow straightening effect while the FC should ideally not limit the turn-down ratio of
the ultrasonic meter. In several tests the CPA Type 55E has been found to provide a good
compromise for this purpose. Typical meter installation setup is shown in figure 3.
Figure 3: Meter installation setup

The first idea for realizing the meter design with integrated inlet section was a welding
assembly made of inlet flange, pipe and measurement section. However, even after substantial
improvements of the welding procedure the residual variations in inner diameter and axis
offset were too high to create reproducible meter characteristics.
The ambient air flow test result of 8 meters with welded inlet section is shown in figure 4. The
meter-to-meter scatter is too high resulting in the fact that most of the meters did not meet the
accuracy requirements of ISO 17089-1 in ambient air. It is assumed, that a high variance of
the baselines in low pressure will cause a larger deviation of the derived high pressure
baselines.
Figure 4: As-found results of 8 meters with welded meter design in ambient air flow test
The test result of the welded meters has shown that a main criterion for reduction of the
scattering between different meters besides the transit-time uncertainty is a low level of
geometric tolerances and minimal meter-to-meter variations by optimization of the
manufacturing and testing procedures..
The meter body design was revised for a second approach. Each meter body was now
manufactured from a steel bar in one machining step with utmost precision. Geometric
dimensions of each meter body are measured precisely afterwards to prove compliance of
tolerance criteria. Figure 5 shows the low level of variations at a sample of 25 machined meter
bodies.
Figure 5: Tolerances in meter body geometry of the machined meter design

The combination of the integrated inlet piping with a low spread in geometric variations and
the ideal positioning of the ultrasonic measurement paths ensure highly reproducible ambient
air flow test results with high reproducibility and a very good linearity of the curves.
Figure 6: Reproducibility results of one sample meter in ambient air, 3 runs over 5 days
In figure 6 it is shown that the machined meters have a reproducibility of <0.35 % over the
full flow range which exceeds the reproducibility requirements of ISO 17089-1:2011. The
machined meter body design made out of a single bar has been approved for series
production.
4 Test results
4.1 Ambient air tests

With the design aspects and manufacturing tolerances from chapter 3, a set of 25 meters have
been tested in a 7-point flow test at SICKs ambient air flow lab. Figure 7 shows the as-found
results of the test. It can be seen that the special design of the meters and the narrow
tolerances result in meeting the Class 1 accuracy requirements with a dual-path meter.
Figure 7: As-found results from 7-point ambient air flow test of 25 meters over flow rate
The meter-to-meter variations are less than ±0.5 % for flow rates above 100 m³/h. The meters
show a very high linearity, in particular for higher flow rates. It can also be seen in figure 7
that the meter-to-meter variations are increasing with low flow rates. This can be explained by
two effects:
First, Reynolds number is decreasing with decreasing flow velocity and cause a less stable
flow profile by the lower inertial forces in the fluid. For example, the Reynolds number at
Qmin is approx. 4500 which represent a “transition” flow which is not fully turbulent. The
effect of Reynolds number on the uncertainty of measurement can be seen in figure 8 which
shows the as-found error of the 25 meters over Re number.
Figure 8: As-found results from 7-point ambient air flow test of 25 meters over Re number
Second, the total uncertainty of measurement of an ultrasonic flow meter is mainly defined by
the error of the transit-time difference determination. While the uncertainty budget of all other
influencing factors are independent from the flow rate, the uncertainty of transit time
difference measurement is increasing at lower flow rates. (Figure 9)
Figure 9: Uncertainty budget of dual-path USM with k=3

The results in figure 7 are encouraging, since the meter-to-meter variations and the linearity of
the meters are good and improve with higher Reynolds numbers. Higher Reynolds numbers
can be expected in the target operation conditions.
The meter electronics corrects for both, geometry changes of the meter body caused by
temperature and pressure variations and for changes in the flow profile caused by a change of
the Reynolds number. These corrections are made based on mathematic models and typically
contain an uncertainty. In order to calculate the estimated total uncertainty caused by
operation of the meter in operating conditions with high pressure natural gas instead of
ambient air, the uncertainty budgets of the test labs and the uncertainty of the internal
corrections have to be considered. All single uncertainties are normally distributed and
considered with 95% confidence level (k=2):
, , , , 0,32 % 2
with single uncertainties considered with:
, 0,2% (Test lab uncertainty of ambient air test stand)
, 0,2% (Typical harmonized lab uncertainty of high pressure test labs)
, 0,1% (Uncertainty of geometric correction function [7])
, 0,1% (Uncertainty of Reynolds number correction)
The ambient air test results give the required margin for the additional uncertainty of 0.32 %
when the increasing Reynolds number in high pressure is considered. Thus, it can be
expected that the meter will stay within the error limits even in high pressure natural gas.
4.2 High pressure natural gas tests
In order to answer that question, 10 out of the 25 meters have been tested at traceable
European high pressure natural gas test benches. Tests have been performed at two different
test benches at 4barg and 50 barg. 4 barg have been chosen to verify the flowmeter
characteristics at slightly elevated pressures from ambient while the 50 barg tests represent a
typical operation pressure of the meter. Figure 10 shows the as-found results of the tests in
high pressure natural gas.
Figure 10: As-Found results of 10 meters in high pressure natural gas

The results confirm the high linearity of the meter characteristics and are within the expected
additional uncertainty of ≤ 0.34 %. All test results are well within the Class 1 performance
requirements of ISO 17089-1. The meter to meter variations are within the expected range.
Significant differences between the 4bar tests and the 50 bar tests could not be found.
If the test results are plotted over Reynolds number (figure 11), it can be found that the spread
of As-found errors is further declining with increasing Reynolds numbers. No systematic bias
can be found. It can be assumed that the spread of error will converge to a level of approx. ±
0.5 % which may represent the physical limitations in uncertainty of a dual path meter.
Figure 11: Test results from ambient air and high pressure natural gas over Re number
4.2 Quality assurance
The test results of chapter 4.1 and 4.2 show that a dual path ultrasonic meter can provide high
accuracy in operating conditions even if it is not calibrated in similar conditions. The key
aspects that must be ensured for that by the meter manufacturer are:
• High linearity of the meter characteristic over the whole flow range
• Minimum meter-to-meter variations in geometric parameters
• Highly repeatable and reproducible ultrasonic measurements also in low flow
conditions
State-of-the art transducer technology and advanced meter electronics with compensation for
geometric shifts caused by temperature and pressure and for changes in the flow profile
caused by changes of the Reynolds number in the gas are clearly a basic precondition.
However, it has been found that even minor deviations in manufacturing tolerances or the
manufacturing process can significantly influence the meter performance. Due to that, SICK
has implemented a quality assurance concept that not only monitors the tolerances in each
manufacturing step, but finally evaluates each meter characteristic in a 5-point ambient air
flow test. In addition, random samples from the series meter production are validated on high
pressure natural gas test benches in order to detect drifts in series production.
5 Ensuring production environment robustness

Another vital aspect for a meter for gathering stations and gas processing plants is a high
robustness against liquids and pollutants that may occur in the gas stream.
Several concepts for ultrasonic meters exist to provide a higher tolerance in polluted gas
streams. SICK has tested and realized numerous meters in upstream applications also with
different concepts over the past 5 years. [5] Based on this experience and the valuable and
trustful cooperation with our customers SICK has found a beneficial combination of meter
design aspects and diagnostic functions for services with potentially wet gas.
The ultrasonic transducers in the meter have been improved for wet gas service. The sensors
are made of titanium, hermetically sealed and protected with a special metallic encapsulation
against moisture. They are mounted in retracted position with enlarged sensor pockets (flush-
mounted) so that contaminants cannot harm or damage the sensors and liquids can easily
drain from the sensor pockets. (Figure 12) Even in the unlikely event of damage to the sensor
head by solids in the gas stream, constructive measures prevent any process leakage. The
pressure bearing feedthrough (red marking in figure 12) from the sensor to the environment is
located in the back part of the sensor, so that even if the sensor head would get damaged, no
safety risk or leakage can occur.
5.1 Liquid indication
It has been shown that ultrasonic meters are well
capable for use even in gas production
applications with wet gas and that the diagnostic
values provide valuable information that may
indicate the presence of liquids. [2] Since
ultrasonic meters as well as orifice meters tend to
over-register with the presence of liquids,
signalizing this presence can be a valuable
diagnostic feature. Thus, the diagnostic data
from wet gas tests with the current meter design
have been analyzed in detail and an idea for a
liquid detection algorithm has been developed.
Figure 12: Sensor positioning inside the meter
The simplified detection algorithm is shown in figure 13. The algorithm uses standard
diagnostic parameters such as Speed of Sound and Turbulence in order to identify a potential
presence of wet gas with LVF of typically >0.5%. This is typically the amount of LVF where
significant over-registering of a meter starts. In a next step, the algorithm has been validated
during a Joint-Industry-Project (JIP) at DNV-GL in the Netherlands in 2014 where SICK
participated with the FLOWSIC600 DRU meter concept. It could be proven that the algorithm
detects liquid volume fractions of 0.5% or more with a confidence level of >95% in
laboratory conditions. This liquid detection algorithm has now been implemented into the
FLOWSIC600 DRU in order to gain field experience and real application data with the new
diagnostic feature that can provide valuable information about potential liquids in the gas
stream. It must be pointed out, that the liquid indication diagnosis is not used to compensate
for over-reading of the meter caused by liquids in the gas.
Figure 13: Simplified algorithm of Liquid Indication Diagnosis by SICK

5.2 Wet gas test results
Wet gas tests at different test facilities as well as field experience in numerous wellhead
applications contributed to increase SICKs knowledge of wet gas measurement over the last
10 years. The latest part in that was the Joint Industry Project “US meters in wet gas
applications” by DNV-GL [8], where the meter has been tested in two-phase flows with the
liquid phase being water, oil or a mixture of water and oil. Variations in Froude number,
Lockhart-Martinelli-Parameter and Density Ratio have been tested in more than 200 test
points. The results are in line with the results from former tests with the meter concept.
Figure 14 shows a ramp-up/ramp-down test with increasing Lockhart-Martinelli-Parameter
(XLM) indicating the “wetness” of the gas. FLOWSIC600 DRU continuously measured over
the full test where at the maximum of XLM=0.9, an equivalent of 13.2% LVF has been
reached. It can be seen also, that the over-registering of the meter in wet gas can reach up to
150% which shows the big importance of identification of those flow conditions. It can be
seen at the beginning and end of the ramp that the meter readings are within an acceptable
allocation accuracy range for XLM < 0.1.
Figure 14: Overreading of FLOWSIC600 DRU in wet gas flow ramp-up/ramp-

down test
During the blind test, the manufacturers were asked to label the measurement result based on
the diagnostic informations from the meter with a traffic light system. SICK decided to set a
“yellow” status for each test point were the diagnostic data indicated, that the meter readings
may be inaccurate. Figure 15 shows the results over XLM and Froude number FrG and it can be
seen that most of the green labeled test points remain close to the actual gas reference flow
rate and show a low uncertainty of typically <2.5%. For test points with higher liquid loads
that caused the meter to over-register, this was signalized by the diagnostic data in most of the
cases. It is also true that in some test points the meter over-registered while the meter
diagnostics did not indicate this and vice versa. SICK did not make pre-tests at the test facility
so that the data of the JIP become higly beneficial since we were able to optimize the
performance of a dual-path meter in wet gas based on this.
Figure 15: Wet gas test results over XLM and FrG during test
After slight modifications in the evaluation method, we re-calculated the test points and
improved the result by eliminating the outliers. Figure 16 shows the result after the
optimization in the evaluation method. Obviously, it is easy to adapt meter diagnostics in
post-processing when the desired result is known. However we are convinced that the meter
performance and meter diagnostics will improve with the optimized diagnosis also in the
field. Results from various ongoing field installations will be presented soon.
Figure 16: Wet gas test results over XLM and FrG after improvements
Even though some test conditions (e.g. with mist flow) have caused the lower ultrasonic path
to fail because liquids were blocking it, the path recovered rapidly when the test conditions
improved. Before and after the wet gas tests, the meters baseline has been determined. No
change in accuracy or stability was observed.
Finally, the results from the JIP Project have verified the meters capability to
• withstand wet gas flow conditions in mixtures with water and oil.
• provide highly accurate readings in dry gas before and after the test.
• provide readings with acceptable accuracy for allocation metering up to 0.5% LVF of
the gas stream.
• provide continuous readings even with heavy liquid loads
• reliably detect the presence of liquids that affect the meters readings
6 Conclusion
The use of ultrasonic meters in gas production applications, at gathering stations or in gas
processing plants is not limited by the ultrasonic technology itself.
It has been shown that the combination of ultrasonic technology, a special meter design, low
manufacturing tolerances and minimal meter-to-meter variations allow transferring ambient
air flow test results to high pressure natural gas. Understanding the key influencing factors on
the meter characteristic and keeping them in an uncritical level of variation is vital in order to
meet Class 1 performance requirements.
Another vital requirement to meters for upstream application is to be robust against liquids
and contaminants and to provide reliable readings under all circumstances. It has been shown
by wet gas test results from former tests and the Joint-Industry-Project of DNV-GL that
several measures in sensor positioning and transducer design allow to provide continuous
readings in wet gas conditions up to a XLM of 0.9! Moreover, it could be shown that for XLM <
0.1, which can be typically expected after the first stage of separation, the meter provides
reliable readings with an accuracy which is usually acceptable for allocation purposes.
Based on wet gas tests and field experience from upstream applications, a new diagnostic
feature has been presented that indicates the presence of liquids in the gas stream which may
cause the meter to over-register significantly. In the lab tests, LVF of >0.5% could be reliably
detected.The reliability of the new diagnostic feature of liquid detection and possible side
effects on it will be investigated in various ongoing field applications.
It can be expected that the installed base of ultrasonic meters in gas production applications
will increase in the future. At least since the low oil prices may force operators to put their
focus more on the operational cost of wellhead equipment. Here, the virtually maintenance-
free ultrasonic meters provide thorough advantages also by their high turn-down ratio.
Precondition for that is that ultrasonic meters show no significant disadvantage compared to
traditional meters in reliability, availability and cost.
References
[1] Internation Standard ISO 17089-1:2010. “Measurement of fluid flow in closed conduits –
Ultrasonic meters for gas. Part 1: Meters for custody transfer and allocation”. First
Edition 2010-11-15.
[2] Jakschik A., Wenzel J., Warmenhoven T., Horst T. “Accuracy and long-term stability of
ultrasonic gas meters at varying operating pressures and different liquid loadings – field
experience.”. North Sea Flow Measurement Workshop, 2012, Norway.
[3] Karnik U., Geerligs J. “Effect of steps and roughness on Multi-path ultrasonic meters”.
International Symposium on Fluid Flow Measurement, 2011, USA.
[4] Horst T., Jakschik A., Dietz T., Riezebos H., Herrmann V. “Considerations on the
influence of deposits or changes in wall roughness on the validity of the calibration and
long term accuracy of ultrasonic gas flow meters”. North Sea Flow Measurement
Workshop, 2011, UK.
[5] Lansing J., Dietz T., Steven R. “Wet Gas Test Comparison Results of Orifice Metering
Relative to Gas Ultrasonic Metering”. AGA Operations Conference, 2012, USA.
[6] Gätke, Johann. “Akustische Strömungs- und Durchflussmessung”. Rostock: Wiley-VCH,

1991.
[7] Herrmann V., Ehrlich A., Dietz T. “Multipath Ultrasonic Gas Flow Meter – How can
Design Iimprovements reduce total Measurement Uncertainty?”. North Sea Flow
Measurement Workshop, 2004, Norway.
[8] van Putten, D., Riezebos, H. “First Results of the JIP on Ultrasonic Meters in Wet Gas
Applications”. European Flow Measurement Workshop, 2015, Netherlands
18
UNCERTAINTY IN FLOW GAS MEASUREMENT SYSTEMS

WITH ULTRASONIC METERS
Felipe Matuzenetz Marinho1
1 ABSTRACT
Although there are, in many countries, uncertainty ranges acceptable by law and by contracts
signed in fiscal metering of natural gas, in many cases is not defined the methodology for calculation of
this uncertainty. These data, in addition to verify the certainty of the measurement, can be used in
management decisions in equipment and process applications.
This paper applied the methodology shown in GUM (JCGM, 2008), mapping the uncertainties
involved in the metrological process. In the analysis were verified several parameters that are normally
discarded such as the differences between the calibration and process conditions, errors due the heat
transfer between the pipe and the thermowell, variations in atmospheric pressure, and chromatographic
uncertainty. In total, fifty four sources of uncertainty were verified. In addition to these data, were
considered the uncertainties that affect the amount billed, producing a monetary representation of
uncertainty values.
As a case study, was used a custody transfer station in Maranhao / Brazil, with ultrasonic meters
in a flow rate of 2,200,000 Sm³/day, at 32 bar.
The results showed the relevance of factors that generate uncertainty in the final measurement,
generating reliable data to guide management decisions, showing, for example, if is economically feasible
to purchase more expensive and precise equipment, comparing them with the uncertainty values obtained
with these equipment.
2 OBJETIVE
This paper aims to demonstrate the natural gas measurement uncertainty calculations applied at
Companhia Distribuidora de Gás Natural do Maranhão - Gasmar.
The uncertainty calculation in accordance with this document aims to:
• Ensure the measurement reliability of consumed volumes;
• Ensure contractual and regulatory compliance with respect to the criterion of uncertainty
maximum;
• Add value to the product and service provided in volume measurement;
• Ensure traceability and transparency in uncertainty calculation;
• Provide auditable material to give basis for an arbitration in volumes disputes;
• Ensure that invoiced volumes are fair and not unduly benefit any of the parties;
• Give basis for management decisions.
3 INTRODUCTION
The Joint Resolution ANP/INMETRO No. 1, from June 10, 2013, Item 6.4.7 - b, sets the
maximum uncertainty of flow measurement for custody transfer in Brazil at 1.5%. Several other standards
in other countries establish maximum uncertainty parameters. However, the methodology for calculation
of this uncertainty is not defined. This document defines a methodology for calculation of uncertainty and
error in the fiscal measurement based on ISO GUM (JCGM, 2008).
The methodology consists of the individual analysis of each source of uncertainty, combining
uncertainties that interfere in each variable separately and calculating the contribution of each variable on
__________________________________________________________________________________________________________
1
Chemical Engineer, Operation and Maintenance Manager - Companhia Maranhense de Gas – Gasmar
North Sea Flow Measurement Workshop 2015
Uncertainty in flow gas measurement systems with ultrasonic meters
___________________________________________________________________________________
the final value. After the calculation of the uncertainty of the measured volume, the uncertainty is then
analyzed over the invoiced amount corresponding to the volume, which includes other sources of
uncertainty, such as tariff calculation.
The methodology was tested and applied in a custody transfer measurement system in Maranhao
/ Brazil using a 10” measurement system, pressure approximately at 32 bar, 40 °C gas temperature,
average flow of 2,200,000 Sm³/day and ultrasonic meter with four pairs of transducers. The straight
segment, as well as the entire system, was designed in accordance with AGA Report No. 9.
In addition to the confirmation of the legal and contractual compliance and measurement
transparency and reliability, these data give basis for management decisions. It can be seen, for example,
if the application of an in line chromatograph at the point of delivery reduces the financial risk of this
application in order to enable the investment cost, just by comparing the difference in billing uncertainty
with and without the chromatograph with the cost of purchasing and installation of the equipment.
4 CALCULATION METHODOLOGY
4.1 Concept and volume uncertainty method

The method demonstrated below is defined in ISO GUM (JCM 2008), applied to the fiscal
measurement of natural gas, in order to quantify the uncertainty arising from this measurement. There
may be variations depending on the measurement system design (equipment, dimensions, etc.) and
processes adopted.
4.1.1 Calculation Method

The calculation of the uncertainty is performed from the partial uncertainties of all variables
that act on the system. To analyze which variables impact the system the definition of the
mathematical model is required. Typical mathematical models in the gas measurement are detailed
in Item 4.2.
After defining the mathematical model, the variables that comprise it and the uncertainties of
each variable are analyzed.
4.1.1.1 Partial Uncertainties

Initially, the individual analysis of each variable that impacts the final result is performed.
A given variable can result in several sources of uncertainty. For example, the variable
temperature may have uncertainty due to repeatability, response time, drift, resolution, etc.
All the variable uncertainties must be identified. The quantification of individual
uncertainties is detailed in Item 4.3.
In order all uncertainties are on the same level of confidence, every uncertainty must be
divided by their respective coverage factor.
4.1.1.2 Combined uncertainty of the variables

After the definition and quantification of the partial uncertainties, the combined uncertainty
is calculated for each variable, algebraically equal to geometric sum of the partial uncertainties
(Equation 1).
__________________________________________________________________________________________________________
Page 2/25
___________________________________________________________________________________
𝑛𝑗
𝑘𝑗
𝑢𝑐 𝑗 = . √∑ 𝑢𝑗 2𝑖 (1)
2
𝑖=1
Where:
𝑢𝑐 𝑗 : Combined uncertainty of variable j (same unit of variable j);
𝑘𝑗 : Coverage factor of uncertainty 𝑢𝑐 𝑗 calculated according to Item 4.1.1.2.1 for a
confidence level of 95,45% (dimensionless);
𝑛𝑗 : Number of uncertainties identified from variable j (dimensionless);
𝑢𝑗 𝑖 : Partial uncertainty i of variable j (same unit of variable j).
4.1.1.2.1 Effective degrees of freedom

Some quantified uncertainties may have been obtained by empirical measurement, in
this case, the number of measurements can affect the results obtained. It is applied then the t-
Student method to correct the measured value to a theoretical condition in which endless
measurements would be performed. This is the case, for example, of the repeatability, which
is obtained by repeated measurement of the same value, checking the variations that occur.
The more measurements are made, the more reliable the result.
Algebraically, this method is reflected in the coverage factor shown in Equation 1,
which is obtained from Table 1 or by the formula shown in
Figure 1 (MS Excel - BR) using the effective degrees of freedom, calculated according
to Equation 2.
𝑢𝑐 4
𝜈𝑒𝑓𝑓 = (2)
𝑢𝑖 4
∑𝑁
𝑖=1 𝜈
𝑖
Where:
𝜈𝑒𝑓𝑓 : Effective degrees of freedom (dimensionless);
𝜈𝑖 : Uncertainty degrees of freedom u (dimensionless)i;
𝑢𝑐 : Combined uncertainty of the variable analyzed (same unit of the variable analyzed);
𝑢𝑖 : Uncertainty i of the variable analyzed (same unit of the variable analyzed);
Figure 1. Calculation of the coverage factor in Excel
__________________________________________________________________________________________________________
Page 3/25
___________________________________________________________________________________
Table 1. Coverage Factor in function of Effective Degree of Freedom and Level of Confidence
Effective Degrees Level of Confidence
of Freedom 68.27% 90% 95% 95.45% 99% 99.73%
1 1.84 6.31 12.71 13.97 63.66 235.8
2 1.32 2.92 4.3 4.53 9.92 19.21
3 1.2 2.35 3.18 3.31 5.84 9.22
4 1.14 2.13 2.78 2.87 4.6 6.62
5 1.11 2.02 2.57 2.65 4.03 5.51
6 1.09 1.94 2.45 2.52 3.71 4.9
7 1.08 1.89 2.36 2.43 3.5 4.53
8 1.07 1.86 2.31 2.37 3.36 4.28
9 1.06 1.83 2.26 2.32 3.25 4.09
10 1.05 1.81 2.23 2.28 3.17 3.96
11 1.05 1.8 2.2 2.25 3.11 3.85
12 1.04 1.78 2.18 2.23 3.05 3.76
13 1.04 1.77 2.16 2.21 3.01 3.69
14 1.04 1.76 2.14 2.2 2.98 3.64
15 1.03 1.75 2.13 2.18 2.95 3.59
16 1.03 1.75 2.12 2.17 2.92 3.54
17 1.03 1.74 2.11 2.16 2.90 3.51
18 1.03 1.73 2.1 2.15 2.88 3.48
19 1.03 1.73 2.09 2.14 2.86 3.45
20 1.03 1.72 2.09 2.13 2.85 3.42
25 1.02 1.71 2.06 2.11 2.79 3.33
30 1.02 1.7 2.04 2.09 2.75 3.27
35 1.01 1.7 2.03 2.07 2.72 3.23
40 1.01 1.68 2.02 2.06 2.7 3.2
45 1.01 1.68 2.01 2.06 2.69 3.18
50 1.01 1.68 2.01 2.05 2.68 3.16
100 1.005 1.66 1.984 2.025 2.626 3.077
∞ 1 1.645 1.96 2 2.576 3
4.1.2 Sensitivity Coefficient

For the calculation of the global uncertainty, combined to the same analysis variable (in this
case the volume converted to reference conditions), it is necessary to calculate the Sensitivity
Coefficient, which represents the impact the uncertainty of a variable causes in the uncertainty of
the result. When the variables are independent, the sensitivity coefficient is calculated according to
Equation 3. When the sensitivity coefficient of a variable is multiplied by the uncertainty of this
variable, it results in the Contribution of this uncertainty. The geometric sum of the contributions of
each variable generates the overall uncertainty (Taylor series considering that the system is linear),
which multiplied by a coverage factor results in the Expanded Uncertainty (Equation 4).
__________________________________________________________________________________________________________
Page 4/25
___________________________________________________________________________________
𝜕𝑓
𝑐𝑗 = (3)
𝜕𝑥𝑗
𝑁
2
𝑈 = 𝑘. √∑ (𝑢𝑐𝑗 . 𝑐𝑗 ) (4)
𝑗=1
Where:
𝑓: Mathematical model to obtain the results from the variables analyzed;
𝑥𝑗 : Variable j;
𝑐𝑗 : Sensitivity Coefficient of variable j;
𝑈: Expanded overall uncertainty;
𝑘: Coverage Factor considering infinite degrees of freedom;
𝑁: Number of variables of the mathematical model.
Follow below the sensitivity coefficient of each variable of the mathematical model defined
in Item 4.2.
𝜕𝑉𝑅 𝑉𝑅 𝜕𝑉𝑅 𝑉𝑅 𝜕𝑉𝑅 𝑉𝑅 𝜕𝑉𝑅 𝑉𝑅

= ‖ = ‖ = ‖ =−
𝜕𝑁 𝑁 𝜕𝑃𝐹 𝑃𝐹 + 𝑃𝐴𝑇𝑀 𝜕𝑃𝐴𝑇𝑀 𝑃𝐹 + 𝑃𝐴𝑇𝑀 𝜕𝑇𝐹 𝑇𝐹
𝜕𝑉𝑅 𝑉𝑅 𝜕𝑉𝑅 𝑉𝑅 𝜕𝑉𝑅 𝑉𝑅

= ‖ =− ‖ = (5)
𝜕𝑍𝑅 𝑍𝑅 𝜕𝑍𝐹 𝑍𝐹 𝜕𝑃𝐶𝑆𝐹 𝑃𝐶𝑆𝐹
4.2 Definition of the mathematical model

The volume conversion for the base conditions is performed from the Clapeyron equation and it
is shown in Equation 6, considering as negligible the systematic error of the flowmeter (pulse
compensation on the meter).
(𝑃𝐹 + 𝑃𝑎𝑡𝑚 ) 𝑇𝑅 𝑍𝑅
𝑉𝑅 = 𝑉𝐹 ∙ ∙ ∙ (6)
𝑃𝑅 𝑇𝐹 𝑍𝐹
Where:
𝑉𝑅 : Volume at base conditions (m³);
𝑉𝐹 : Volume at flow conditions (m³);
𝑃𝐹 : Pressure at flow conditions (Pa);
𝑃𝑎𝑡𝑚 : Local atmospheric pressure (Pa);
𝑃𝑅 : Pressure at base conditions (Pa);
𝑇𝑅 : Temperature at base conditions (K);
𝑇𝐹 : Temperature at flow conditions (K);
𝑍𝑅 : Compressibility at base conditions (dimensionless);
𝑍𝐹 : Compressibility at flow conditions (dimensionless).
To offset the calorific value of the gas, a calorific value factor is included as per Equation 7.
__________________________________________________________________________________________________________
Page 5/25
___________________________________________________________________________________
(𝑃𝐹 + 𝑃𝑎𝑡𝑚 ) 𝑇𝑅 𝑍𝑅 𝑃𝐶𝑆𝐹
𝑉𝑅 = 𝑉𝐹 ∙ ∙ ∙ ∙ (7)
𝑃𝑅 𝑇𝐹 𝑍𝐹 𝑃𝐶𝑆𝑅
Where:
𝑃𝐶𝑆𝑅 : Reference Superior Calorific Value (J/m³);
𝑃𝐶𝑆𝐹 : Superior Calorific Value of the Gas at flow conditions.
Assuming a constant factor K, the volume in flow conditions is detailed depending on the amount
of pulses, according to Equation 8, which also details the calorific value calculated as the method
shown in ISO 6976.
̃ °𝑗
∑𝑛𝑗=1 𝑥𝑗 ∙ 𝐻
𝑁 (𝑃𝐹 + 𝑃𝑎𝑡𝑚 ) 𝑇𝑅 𝑍𝑅 𝑍𝑅
𝑉𝑅 = ∙ ∙ ∙ ∙ (8)
𝐾 𝑃𝑅 𝑇𝐹 𝑍𝐹 𝑃𝐶𝑆𝑅
Where:
𝑁 : Accumulated amount of pulses (pulses);
𝐾 : Pulses correspondence factor per cubic meter (pulse/m³);
𝑥𝑗 : Mole fraction of the jth component (dimensionless);
𝐻̃ °𝑗 : Ideal calorific value in volumetric base at the base conditions (MJ/m³);
𝑛 : Quantity of natural gas constituent components (dimensionless).
Considering the expansion of the flowmeter body due to pressure and thermal expansion, and
detailing the calculation of the atmospheric pressure based on local altitude, Equation 9 is formed as
follow.
−𝑀∙𝑔
𝑎∙ℎ 𝑎∙𝑅
[𝑃𝐹 + 𝑃0 ∙ (1 + 𝑇 ) ]
0 ̃ °𝑗
𝑇𝑅 𝑍𝑅 ∑𝑛𝑗=1 𝑥𝑗 ∙ 𝐻
𝑉𝑅 = 𝑉𝑟𝑎𝑤 ∙ 𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 ∙ 𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇 ∙ ∙ ∙ ∙ (9)
𝑃𝑅 𝑇𝐹 𝑍𝐹 𝑃𝐶𝑆𝑅
Where:
𝑉𝑟𝑎𝑤 : Raw volume measured by flowmeter (m³);
𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 : Correction factor of the flowmeter expansion due to pressure (dimensionless);
𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇 : Correction factor of the flowmeter expansion due to thermal expansion
(dimensionless);
𝑃0 : Atmospheric pressure at reference location (Pa);
𝑎 : Rate of temperature variation in relation to altitude (K/m);
ℎ : Altitude in relation to atmospheric base point (m);
𝑇0 : Temperature at atmospheric base point (K);
𝑀 : Air mole mass (kg/m³);
𝑔 : Local gravity (m/s²);
𝑅 : Universal gas constant (J/mol.K).
Details of the correction factors of the flowmeter expansion due to pressure and temperature is
shown in Equation 10.
__________________________________________________________________________________________________________
Page 6/25
___________________________________________________________________________________
2 2
[𝐷𝑜𝑢𝑡 ∙ (1 + 𝜐)] + [𝐷𝑖𝑛 ∙ (1 − 2 ∙ 𝜐)]
𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 ∙ 𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇 = 1 + [3 ∙ ( 2 2) ) ∙ (𝑃𝐹 + 𝑃𝑎𝑡𝑚 − 𝑃𝑅 )] (10)
𝐸 ∙ (𝐷𝑜𝑢𝑡 − 𝐷𝑖𝑛
Where:
𝐷𝑜𝑢𝑡 : Outer diameter of the piping or flowmeter (m);
𝐷𝑖𝑛 : Inner diameter of piping or flowmeter (m);
𝜐 : Poisson Coefficient (dimensionless);
𝐸 : Young Module (MPaa).
To simplify the analysis and calculation process of the uncertainty of the parameters analyzed, it
is used the basic model defined in Equation 7, details shown in Equations 8, 9 and 10 are calculated
separately.
4.3 Uncertainty Sources

To check the uncertainties in the reference volume, the individual variables are analyzed, which
are shown below.
4.3.1 Uncertainties related to the quantity of accumulated pulses (N)
4.3.1.1 Reading Resolutions

The transmission of flow by pulses has the range uncertainty of a pulse due to the
resolution. The higher the frequency of pulses per cubic meter is, the lower the impact of this
uncertainty on the volume.
4.3.1.2 Loss of Pulse

Accumulated pulses differences between the generated by the flow meter and the received
by flow computer accumulator may eventually occur. These differences may happen for various
reasons, such as the interferences in the inductive environment, contact with corrosion, and
incompatible impedance. The estimation of loss of pulse is performed based on the analysis of
the history between the flow meter accumulator and the flow computer accumulator.
4.3.1.3 Uncertainty Inherited From the Calibration Laboratory

The analyzed measurement system has a calibration curve correction in the flow meter.
The values used in the curve correction are subject to the calibration laboratory uncertainties and
should be taken into consideration in the analysis. It is obtained with the multiplication of the
uncertainty informed in the calibration certificate of the flowmeter (comprising all the
uncertainties of the laboratory, such as the ones inherited from patterns, repeatability, etc.) by the
calibrated flow range and the K factor.
4.3.1.4 Drift of Flowmeter

Every measurement instrument has a tendency to increase the uncertainty with the use of
the equipment. This can be observed in the As Found calibration report. In order to estimate how
much the uncertainty increases during the inter calibration period, a linear regression is made
considering the last adjustment period and the geometric sum of the repeatability and the
systematic error. The resulting equation is applied to estimate the increase of uncertainty for the
analyzed period considering the last calibration performed in the flowmeter.
__________________________________________________________________________________________________________
Page 7/25
___________________________________________________________________________________
4.3.1.5 Expansion by Pressure
The uncertainty related to the increase in distance between the ultrasonic transducers due
to the expansion by pressure. This is obtained with the individual analysis of the uncertainties
that compose the equation of expansion by pressure, as seen in Equation 11.
2 2
[𝐷𝑜𝑢𝑡 ∙ (1 + 𝜐)] + [𝐷𝑖𝑛 ∙ (1 − 2 ∙ 𝜐)]
𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 = 1 + 3 ∙ 2 2) ∙ (𝑃𝐹 + 𝑃𝑎𝑡𝑚 − 𝑃𝑅 ) (11)
𝐸 ∙ (𝐷𝑜𝑢𝑡 − 𝐷𝑖𝑛
In this case study, the uncertainties that came from the measurements of the flowmeter’s
inner and outer diameter, the Poisson coefficient, and the Young Module, were not taken into
account, considering only the uncertainties that came from the pressure and atmospheric pressure
variations, which the flowmeter is submitted. The uncertainty of the pressure effect expansion is
obtained with Equation 12, where k stands for the coverage factor obtained by comparing the
values fixed in the GUM (JCGM, 2008) to the effective degrees of freedom (calculated with
Equation 13).
𝑠𝑃 2
𝑢𝑁−𝑝𝑟𝑒−𝑝𝑓 = 𝑘 ∙ √( ) + 𝑢𝑃𝐹 2 (12)
𝑛
√ 𝑃
4
2
𝑠
(√( 𝑃 ) + 𝑢𝑃𝐹 2 )
√𝑛𝑃
𝜈𝑒𝑓𝑓𝑝𝑢𝑙𝑠𝑜𝑠−𝑝𝑟𝑒−𝑝𝑓 = (13)
𝑠 4
( 𝑃 ) 𝑢𝑃𝐹 4
√𝑛𝑃 +
𝑛𝑃 − 1 𝜈𝑒𝑓𝑓−𝑝𝑟𝑒𝑠𝑠ã𝑜
Where:
𝑢𝑁−𝑝𝑟𝑒−𝑝𝑓 : Uncertainty of the gas pressure for the uncertainty calculation due to the
expansion by pressure effect, including the values variation (MPa);
𝑢𝑃𝐹 : Combined uncertainty of the gas pressure measurement (MPa);
𝑛𝑃 : Quantity of pressure measurements collected in the period (dimensionless);
𝑠𝑃 : Standard deviation of the pressure measurements collected in the period (MPa);
𝜈𝑒𝑓𝑓−𝑝𝑟𝑒𝑠𝑠ã𝑜 : Effective degrees of freedom of the combined uncertainty in the pressure
measurement (dimensionless).
The atmospheric pressure uncertainty is obtained as shown in Item 4.3.7. The sensitivity
coefficient is equal to the partial derivative of the pressure effect correction in relation to the
measured pressure and in relation to the atmospheric pressure, as shown in Equation 14.
2 2
𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 [𝐷𝑜𝑢𝑡 ∙ (1 + 𝜐)] + [𝐷𝑖𝑛 ∙ (1 − 2 ∙ 𝜐)]
= = 3∙ 2 2) ∙𝐾 (14)
𝜕𝑃𝐹 𝜕𝑃𝑎𝑡𝑚 𝐸 ∙ (𝐷𝑜𝑢𝑡 − 𝐷𝑖𝑛
Where:
𝐾 : Correspondence Factor of pulses per cubic meter (pulse/m³).
The uncertainty in the pulses emitted by the flowmeter due to the pressure effect expansion
is calculated as shown in Equation 15.
__________________________________________________________________________________________________________
Page 8/25
___________________________________________________________________________________
2 2
√(𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 ∙ 𝑢𝑁−𝑝𝑟𝑒−𝑝𝑓 ) + (𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑃 ∙ 𝑢𝑃𝑎𝑡𝑚 ∙ 𝑘𝑃𝑎𝑡𝑚 )
𝜕𝑃𝐹 𝜕𝑃𝑎𝑡𝑚
𝑢𝑁−𝑝𝑟𝑒 = (15)
2
Where:
𝑢𝑁−𝑝𝑟𝑒 : Uncertainty of the number of pulses due to the pressure effect expansion (pulses);
𝑢𝑃𝑎𝑡𝑚 : Uncertainty of the atmospheric pressure (MPa).
4.3.1.6 Expansion by Temperature Effect

The distance between the ultrasonic transducers can also be changed due to the thermal
expansion of the flowmeter body. The uncertainty from this variation is obtained similarly to
what was demonstrated for the pressure expansion, based on Equation 16. The uncertainty is
calculated in accordance with Equation 17 (disregarding the uncertainty of the linear expansion
coefficient), with the coverage coefficient proportional to the effective degrees of freedom
calculated in the Equation 18.
𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇 = 1 + 3 ∙ 𝛼 ∙ (𝑇𝐹 − 𝑇𝑅 ) (16)
𝑠 2
𝑘 ∙ √( 𝑇 ) + 𝑢 𝑇𝐹 2
√𝑛 𝑇
𝑢𝑁−𝑡𝑒𝑚−𝑡𝑓 = (17)
2
4
2
𝑠𝑇
(√( ) + 𝑢 𝑇𝐹 2 )
√𝑛 𝑇
𝜈𝑒𝑓𝑓𝑝𝑢𝑙𝑠𝑜𝑠−𝑡𝑒𝑚 = (18)
𝑠 4
( 𝑇 ) 𝑢 𝑇𝐹 4
√𝑛 𝑇 +
𝑛𝑇 − 1 𝜈𝑒𝑓𝑓−𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑎
Where:
𝛼 : Coefficient of thermal expansion of body material of flowmeter.
𝑠𝑇 : Sampling standard deviation of the temperature variation during the period (K);
𝑛𝑇 : Number of temperature samples (dimensionless);
𝑘 : Coverage factor (dimensionless);
𝑢 𝑇𝐹 : Combined uncertainty of the temperature measurement (K);
𝜈𝑒𝑓𝑓−𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑎 : Effective degrees of freedom of the partial uncertainties that compose the
uncertainty of temperature measurement (dimensionless);
𝑢𝑁−𝑡𝑒𝑚−𝑡𝑓 : Gas temperature uncertainty for the uncertainty calculation due to the
expansion by temperature effect (K).
The sensitivity coefficient is equal to the partial derivative of correction by temperature

effect in relation to the gas temperature, resulting in what is demonstrated in Equation 19, which
also demonstrates the uncertainty calculation in the pulses due to the flowmeter thermal
expansion.
__________________________________________________________________________________________________________
Page 9/25
___________________________________________________________________________________
𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇 𝜕𝐸𝑥𝑝𝐶𝑜𝑟𝑟𝑇
( = 3 ∙ 𝛼 ∙ 𝐾) → 𝑢𝑁−𝑡𝑒𝑚 = 𝑢𝑁−𝑡𝑒𝑚−𝑡𝑓 ∙ (19)
𝜕𝑇𝐹 𝜕𝑇𝐹
Where:
𝐾 : Correspondence Factor of pulses per cubic meter (pulse/m³);
𝑢𝑁−𝑡𝑒𝑚 : Uncertainty of the number of pulses due to the expansion by temperature effect
(pulses).
4.3.2 Uncertainties Related to the Gas Pressure’s Measurement (PF)
4.3.2.1 Repeatability
Uncertainty related to the capacity of the instrument to show the same measurement in the
same point repeatedly. It is estimated during the calibration of the pressure transmitter and is
numerically equal to the greatest calculated value of the division between the standard deviation
and the square root of the number of measurements, for each measurement point during the
calibration, and multiplied by a coverage factor that equals two, as per demonstrated in Equation
20.
2
𝑛
∑𝑛𝑗=1 𝑥𝑗
√∑𝑖=1 (𝑥𝑖 − 𝑛 )
𝑠 𝑛−1
𝑢𝑃𝐹−𝑟𝑒𝑝 = = (20)
√𝑛 √𝑛
Where:
𝑢𝑃𝐹−𝑟𝑒𝑝 : Uncertainty related to the repeatability of the instrument (Pa);
𝑠 : Standard Deviation of Measurements (Pa);
𝑛 : Number of Measurements (dimensionless);
𝑥𝑖 , 𝑥𝑗 : Pressure Measurement (Pa).
4.3.2.2 Hysteresis
The hysteresis is related to the capacity of the instrument to show the same measurement
at the same point in ascending and descending measurements. It is estimated during the
calibration of the pressure transmitter and is numerically equal to the greatest calculated value
found in the module of the average difference between the ascending and descending calibration
measurements, as per Equation 21.
𝑛 𝑛
2 2
𝑥𝑎 𝑥𝑑
𝑢𝑃𝐹−ℎ𝑖𝑠 = |∑ −∑ | (21)
𝑛 𝑛
𝑎=1 𝑑=1
Where:
𝑢𝑃𝐹−ℎ𝑖𝑠 : Uncertainty due to hysteresis of the pressure meter instrument (Pa);
𝑛 : Number of measurements (dimensionless);
𝑥𝑎 : Ascending pressure measurement (Pa);
𝑥𝑑 : Descending pressure measurement (Pa).
__________________________________________________________________________________________________________
Page 10/25
___________________________________________________________________________________
4.3.2.3 Standard Instrument Resolution
The resolution of the standard instrument used in the calibration is also a source of
uncertainty and is obtained in the certificate of calibration of the standard instrument.
4.3.2.4 Uncertainty inherited from calibration standard instrument

Like the resolution, the uncertainty inherited from the standard instrument is also
considered and obtained from the certificate of calibration of the standard instrument multiplying
by the calibrated amplitude of the standard instrument.
4.3.2.5 Instrument Resolution

The instrument resolution represents the truncation of the measured value. In the analyzed
system, the signal transmission of the instrument is made via the 4-20mA communication and
the resolution is limited by the D/A converter, which is obtained from the transmitter manual,
multiplying the value by the signal transmission amplitude.
4.3.2.6 Uncertainty of the 4-20mA current meter

The uncertainty related to the 4-20mA current meter instrument used in the calibration. It
is obtained in the certificate of calibration of the current meter. It is algebraically obtained
multiplying the uncertainty in the certificate of calibration of the current meter by the amplitude
of the signal transmission (in pressure unit) and divided by the amplitude signal transmission in
current (with the 4-20mA, the amplitude would be 16mA), as shown in Equation 22.
𝑢𝑐𝑜𝑟𝑟𝑒𝑛𝑡𝑒−𝑚𝑐𝑐 ∙ 𝑎𝑝
𝑢𝑃𝐹−𝑚𝑐𝑐 = (22)
𝑎𝑐
Where:
𝑢𝑃𝐹−𝑚𝑐𝑐 : Uncertainty inherited from 4-20mA signal meter used in calibration (Pa);
𝑢𝑐𝑜𝑟𝑟𝑒𝑛𝑡𝑒−𝑚𝑐𝑐 : Uncertainty of 4-20mA signal meter used in calibration (mA);
𝑎𝑝 : Transmitter measurement amplitude (Pa);
𝑎𝑐 : Signal transmitter amplitude (for transmission of 4-20mA ac=16mA) (mA).
4.3.2.7 Signal Generator Uncertainty in Loop Test

In the analyzed system, the instruments are removed from installation site for calibration
and the communication mesh is verified separately. The values obtained are compensated in the
calibration values. A signal generator is used for the test of the communication mesh. The signal
generator uncertainty multiplied by the signal transmission amplitude (in pressure unit) divided
by the signal transmission amplitude in current units is the value considered as uncertainty due
to the use of this equipment (similar to what was demonstrated in Equation 18).
4.3.2.8 Pressure Variations

The flow computer has a calculation cycle that performs the data reading of the instruments
and calculates the volume at base conditions. If there are pressure variations in time shorter than
the flow computer sampling, these variations will not be recorded. The shorter the sampling time
is, the lower the uncertainty due to the pressure in the sampling intervals. The uncertainty value
is the standard deviation of the variations in time of the flow computer cycle divided by the
square root of the quantity of analyzed variations.
__________________________________________________________________________________________________________
Page 11/25
___________________________________________________________________________________
4.3.2.9 Environment Temperature Effect
The electronic components of the transmitters go through minor changes due to the
temperature. These changes can be adjusted in the calibration, but some differences may happen
when the temperature in the calibration is different from the temperature in the process. The
closest the process temperature is to the temperature in the calibration, the lower this uncertainty
will be. Each manufacturer has a calculation method to achieve this uncertainty, which can be
found in the manual of the instrument. For the Rosemount transmitter 3051, the uncertainty due
to the effect of the environment temperature is calculated according to Equation 23.
(25 ∙ 𝐿𝑆𝐹 + 125 ∙ 𝑎) ∙ 10−5

𝑢𝑃𝐹−𝑒𝑡𝑎 = ∙ |𝑇𝑎𝑚𝑏−𝑝𝑟𝑜 − 𝑇𝑎𝑚𝑏−𝑐𝑎𝑙 | (23)
28
Where:
𝑢𝑃𝐹−𝑒𝑡𝑎 : Uncertainty due to
the environment temperature variation (Pa);
𝐿𝑆𝐹 : Upper range limit of transmitter (bar);
𝑎 : Calibration amplitude of transmitter (bar);
𝑇𝑎𝑚𝑏−𝑝𝑟𝑜 : Average environment temperature at the installation site (°C);
𝑇𝑎𝑚𝑏−𝑐𝑎𝑙 : Average environment temperature during the calibration (°C).
4.3.2.10 Pressure Meter Drift

Similar to what was demonstrated in Item 4.3.1.4, the pressure meter is subject to the drift
effect, and this uncertainty is calculated by the linear regression considering the last adjustment
period and the geometric sum of the repeatability, the hysteresis, and the systematic error. The
resulting equation is used to estimate the increase in uncertainty for the analyzed period
considering the last calibration performed in the instrument.
4.3.2.11 Assembling position of Transmitter

The assembling position (horizontal, vertical, or inclined) can interfere the pressure
measurement. This uncertainty can be eliminated in the calibration adjustment. In the analyzed
measurement system, the calibration is performed in the same vertical inclination of the
installation position. However, once the instrument is moved during the calibration, there might
be a slight difference between the calibration measurement and the installed instrument
measurement. The manufacturer informs this uncertainty estimative.
4.3.2.12 Electric Variation

Voltage variations of the instruments might propagate in signal transmission and result in
interferences in measurement. The uncertainty value depends on the instrument voltage and is
informed by the manufacturer.
4.3.3 Uncertainties related to flow temperature (TF)
4.3.3.1 Repeatability
Such as the calculation for the pressure transmitter (Equation 20), the repeatability is also
taken into consideration for the temperature measurement.
__________________________________________________________________________________________________________
Page 12/25
___________________________________________________________________________________
4.3.3.2 Uncertainty inherited from the standard instrument
The uncertainty of the standard instrument used in the calibration of temperature instrument
constitutes the variable uncertainty calculation and must be in absolute values of temperature.
4.3.3.3 Resolution of Calibration Standard Instrument

The standard temperature instrument used in the calibration should be considered
(instrument of measurement or indication of temperature bath).
4.3.3.4 Axial Homogeneity of the temperature bath

There might be differences in the measurement of the calibration bath depending of the
vertical insertion position of the sensor. These differences create errors in the calibration process.
The estimate of the calibration bath homogeneity is achieved with tests in various conditions,
and is informed by the manual of manufacturer equipment.
4.3.3.5 4.3.3.5 Radial Homogeneity of temperature bath

Similar to the axial homogeneity, there is also an uncertainty due to the insertion horizontal
position of the sensor. This value can also be found in the manual of the equipment.
4.3.3.6 Uncertainty of the 4-20mA meter

Similar to the calculation of the pressure instrument, the 4-20mA signal meter equipment
is considered in the temperature uncertainty.
4.3.3.7 Precision of the Signal Generator in Loop Test

Similar to the calculation of pressure meter instrument, when the Loop Test is used to
compensate the error in the communication mesh, the uncertainty of the signal generator is
considered.
4.3.3.8 Thermal Transfer of the Thermowell

When a thermowell is used in the temperature measurement, there might be thermal
transfers between the thermowell and the piping, considering that normally the device is welded,
threaded, or flanged in the piping, and the piping is subject to environmental conditions. These
transfers are greater when the gas temperature and the pipe temperature are not in thermal
equilibrium, or when there is consumption intermittence.
A study has been conducted to verify the impact of the environmental conditions in the
measurement of temperature in the analyzed system. Several computer simulations of the fluid
dynamics (CFD system) were performed, using the exact dimensions of the meter system with
the materials installed, and using the hourly measurements of temperature in various parts of the
system and environment temperature for 30 days. The data were statistically analyzed providing
indications as shown in Figure 2. Due to the complexity of the study, this will not be detailed in
this paper.
__________________________________________________________________________________________________________
Page 13/25
___________________________________________________________________________________
Figure 2. CFD Simulation of the interference estimation of the environmental conditions in the
temperature measurement with a thermowell
4.3.3.9 Response Time of the Temperature Sensor

The temperature sensor has a period that is necessary to identify the temperature variation
and to stabilize the signal in the real value. The sensors are tested by the manufacturer under
determined conditions to verify the necessary time for the sensor to reach 50% of the step
disturbance value. The uncertainty estimation regarding the response time is achieved by
verifying the greatest variation registered by the meter in a period corresponding to the thermal
response time of the instrument. This value corresponds to half of the disturbance theoretical
maximum value registered by the meter (considered as a Type B uncertainty). Figure 3 shows
the method used.
T (°C)
Perturbação tipo degrau
Resposta do sensor à perturbação
utemp-res tRe Tempo necessário para 50% da variação
TRe TRe Indicação de 50% da variação
utemp-res Incerteza no tempo de resposta
t (s)
tRe
Figure 3. Response Time Estimation of the Meter
4.3.3.10 Digital-to-Analog Conversion

Inside the temperature transmitter, there is a digital-to-analog converter that comes with a
rounding proportional to the quantity of bits of the converter. This rounding contributes to the
temperature’s uncertainty and is obtained directly from the transmitter’s datasheet, multiplying
the value by the amplitude set up in the transmitter.
__________________________________________________________________________________________________________
Page 14/25
___________________________________________________________________________________
4.3.3.11 Drift of temperature meter
Similar to what has been demonstrated in Item 4.3.2.10, the temperature meter is subject
to the drift effect, and this uncertainty is calculated by the linear regression considering the last
adjustment period, the geometric sum of the repeatability, and the systematic error. The resulting
equation is used to estimate the increase in uncertainty for the analyzed period considering the
last calibration made in the instrument.
4.3.3.12 Effect of ambient temperature on the temperature transmitter

Similar to Item 4.3.2.9, the temperature transmitter also shows differences in measurement
proportional to the ambient temperature’s difference in the calibration and at flow conditions.
The calculation method depends on the meter’s model used and can be found in the
manufacturer’s manual. The Rosemount 3144P transmitter has an uncertainty of 0.0015°C for
each 1°C of difference between the ambient temperature in the calibration and the ambient
temperature in the installation place (by using the average ambient temperature)
4.3.3.13 Ambient Temperature Effect on Digital Analog Converter

The difference between the ambient temperature in the calibration and in the installation
place also produces an uncertainty in the D/A converter. For the Rosemount 3144P transmitter
this uncertainty can be found in the manufacturer’s manual as being numerically equal to 0.001%
of the amplitude set up in the transmitter for each 1°C of difference between the ambient
temperature in the calibration and in the installation place.
4.3.3.14 Electrical Variation

Similar to the pressure’s meter, the electrical variation in the voltage of the temperature’s
transmitter produces an uncertainty in the measurement that can be obtained in the
manufacturer’s manual.
4.3.4 Compressibility at Flow Conditions
4.3.4.1 Calculation Methodology

The method used to calculate the compressibility is an important factor to be taken into
account. Each calculation method holds an uncertainty related to the constants used and to the
method itself, which is normally empirical. The compressibility calculated according to the AGA
Report nº 8 Detailed Method (2003), shows the uncertainties exposed in Figure 4 according to
pressure and temperature.
__________________________________________________________________________________________________________
Page 15/25
___________________________________________________________________________________
Figure 4. Ranges of uncertainty of compressibility calculation methodology by AGA 8
4.3.4.2 Uncertainty Inherited from the Measurement of Pressure and Temperature

In the mathematical model shown in Item 4.2, the compressibility was inserted as an
independent variable to simplify the analysis. In order to compensate this fact, the insertion of
the impact that the pressure and temperature uncertainties cause in the compressibility’s
uncertainty is necessary. The compressibility’s calculation method is explained in Equation 24.
The uncertainty inherited from the gas pressure measurements, the atmospheric pressure, and the
gas temperature is calculated according to Equation 25 (the parameters and constants are
calculated according to AGA 8 Detailed Method).
18 58
𝐷∙𝐵
𝑍 = 1 + 3 − 𝐷 ∙ ∑ 𝐶𝑛 ∗ ∙ 𝑇 −𝑢𝑛 + ∑ 𝐶𝑛 ∗ ∙ 𝑇 −𝑢𝑛 ∙ (𝑏𝑛 − 𝑐𝑛 ∙ 𝑘𝑛 ∙ 𝐷𝑘𝑛 ) ∙ 𝐷𝑏𝑛 ∙ 𝑒𝑥𝑝(−𝑐𝑛 ∙ 𝐷𝑘𝑛 ) (24)
𝐾
𝑛=13 𝑛=13
Where:
𝑍 : Compressibility
𝐵 : Second virial coefficient;
𝐾 : Size parameter;
𝐷 : Gas reduced density;
𝐶𝑛 ∗ : Coefficient as a result of the composition;
𝑇 : Gas temperature;
𝑢𝑛 , 𝑏𝑛 , 𝑐𝑛 , 𝑘𝑛 : Constants contained in AGA 8.
__________________________________________________________________________________________________________
Page 16/25
___________________________________________________________________________________
2 2 2
√( 𝜕𝑍 ∙ 𝑢𝑃𝐹 ∙ 𝑘𝑃𝐹 ) + ( 𝜕𝑍 ∙ 𝑢𝑃𝑎𝑡𝑚 ∙ 𝑘𝑃𝑎𝑡𝑚 ) + ( 𝜕𝑍 ∙ 𝑢𝑇𝐹 ∙ 𝑘 𝑇𝐹 )
𝜕𝑃𝐹 𝜕𝑃𝑎𝑡𝑚 𝜕𝑇𝐹
𝑢𝑍𝐹−𝑝𝑡 = (25)
2
Where:
𝜕𝑍
𝜕𝑇𝐹
𝜕𝐵 ∗ −𝑢𝑛 ∗ 3.𝑘𝑛 . 𝑑𝑘𝑛 ). 𝑇 −𝑢𝑛 . (𝑐 . 𝑘 . (𝐾 3 . 𝑑)𝑘𝑛 . 𝑢 − 𝑏 . 𝑢 )]
𝑇𝐹 . 𝑑. − 𝑑. 𝐵 + 𝐾 3 . 𝑑. ∑18
𝑛=13 𝐶𝑛 . 𝑇𝐹 . (𝑢𝑛 + 1) − 𝛼 + ∑58 3 𝑏𝑛
𝑛=13[𝐶𝑛 . (𝐾 . 𝑑) . 𝑒𝑥𝑝(−𝑐𝑛 . 𝐾 𝐹 𝑛 𝑛 𝑛 𝑛 𝑛
𝜕𝑇𝐹
=
𝑇 𝑇 𝑇
𝑇𝐹 + 𝐹 . 𝐵. 𝑑 − 𝐹 . 𝛽 + 𝐹 . 𝛼
𝑍𝐹 𝑍𝐹 𝑍𝐹
𝜕𝑍 𝜕𝑍 𝐵. 𝑑 − 𝛽 + 𝛼
= =
𝜕𝑃𝐹 𝜕𝑃𝑎𝑡𝑚 𝑃 + 𝑃 . 𝐵. 𝑑 − 𝑃 . 𝛽 + 𝑃 . 𝛼
𝑍 𝑍 𝑍 𝐹 𝐹 𝐹
58
𝑒𝑥𝑝(−𝑐𝑛 . 𝐾 3.𝑘𝑛 . 𝑑 𝑘𝑛 ). 𝐶𝑛 ∗ . (𝐾 3 . 𝑑)𝑏𝑛
𝛼= ∑ [ . (𝑏𝑛 2 − 𝑏𝑛 . 𝑐𝑛 . (𝐾 3 . 𝑑)𝑘𝑛 . 𝑘𝑛 − 𝑐𝑛 . 𝑘𝑛 . (𝐾 3 . 𝑑)𝑘𝑛 . (𝑏𝑛 + 𝑘𝑛 )
𝑇𝐹 𝑢𝑛
𝑛=13
+ 𝑐𝑛 2 . 𝑘𝑛 2 . (𝐾 3 . 𝑑 2 )𝑘𝑛 . 𝐾 3 )]
18
𝛽 = 𝐾 . 𝑑. ∑ (𝐶𝑛 ∗ . 𝑇𝐹 −𝑢𝑛 )
3
𝑛=13
18 𝑁 𝑁
𝜕𝐵 3
𝑢
= − ∑ 𝑢𝑛 . 𝑎𝑛 . 𝑇𝐹 −(𝑢𝑛+1) . ∑ ∑ 𝑥𝑖 . 𝑥𝑗 . 𝐸𝑖𝑗𝑛 . (𝐾𝑖 . 𝐾𝑗 )2 . 𝐵𝑛𝑖𝑗
∗
𝜕𝑇𝐹
𝑛=1 𝑖=1 𝑗=1
𝑘𝑃𝐹 , 𝑘𝑃𝑎𝑡𝑚 , 𝑘𝑇𝐹 : Coverage factor, calculated with Distribution of Student;

𝑑 : Molar density;
𝐵 : Second virial coefficient;
𝐾 : Size parameter;
𝐶𝑛 ∗ : Coefficient as a result of the composition;
𝑢𝑛 , 𝑏𝑛 , 𝑐𝑛 , 𝑘𝑛 , 𝑎𝑛 : Constants;
𝑃 : Pressure (equal to Patm or PF, depending on sensibility coefficient calculated);
𝑥𝑖 , 𝑥𝑗 : Molar fraction of components i and j;
𝐸𝑖𝑗 : Binary energy parameter of second virial coefficient;
𝐾𝑖 , 𝐾𝑗 : Parameter of component size;
∗
𝐵𝑛𝑖𝑗 : Coefficient of binary characterization.
__________________________________________________________________________________________________________
Page 17/25
___________________________________________________________________________________
4.3.4.3 Uncertainty Inherited from the Gas Composition’s Measurement
The gas composition, normally verified with the chromatography, holds a mixed
uncertainty of various factors, such as repeatability, uncertainty inherited from the standard gas
used in the calibration, mix normalization, method of analyses, etc. Regardless of the method for
the compressibility calculation, the gas composition will be necessary in the calculation, hence,
the uncertainty in the measurement of the fractions of the gas components will be passed to the
compressibility.
To obtain the composition’s uncertainty in the compressibility, the method of extremes
with triangular distribution was adopted. By using the certificate of chromatograph’s calibration,
the uncertainty of the mole fraction of each component was verified, and by considering the
fractions in extreme points of uncertainty, both minimum and maximum compressibility values
were verified in the points of maximum uncertainty with constant pressure and temperature
(equal to flow pressure, atmospheric pressure, and average flow temperatures). The difference
between these data is regarded as a type B uncertainty.
4.3.4.4 Composition Variations

Even with the online equipment, the chromatographic analysis occurs by batch. Any
change in the gas composition in the interval between the analyses will not be taken into account.
When a change in the composition is verified between one analysis and another, it is not possible
to deduce in which moment between the analyses this event took place. Considering the
chromatography used for the compressibility calculation is the last valid one, there is an
uncertainty concerning the possibility of a change in composition variation since the last analysis.
Algebraically, this uncertainty is achieved with the compressibility’s calculation at constant
pressure and temperature (utilizing the average gas pressure, atmospheric pressure, and gas
temperature) analyzing the greatest variation in the period inter samplings. This value represents
the uncertainty with rectangular distribution.
4.3.5 Compressibility at reference conditions
4.3.5.1 Calculation Methodology

Like in the compressibility calculation at flow condition, the uncertainty related to the
calculation method should be taken into account.
4.3.5.2 Uncertainty Inherited from the Gas Composition’s Measurement

Similar to what has been demonstrated in Item 4.3.4.3, when it comes to the compressibility
at flow conditions, the uncertainty in compressibility at base conditions inherited from the
chromatography is reached with the analysis of the uncertainty’s extremes, however, considering
the constant pressure and temperature equal to the base conditions.
4.3.5.3 Composition variations

Just like in the compressibility at flow conditions, the variations in the composition in time
shorter than the one in the chromatographic sampling bring about an uncertainty in the
compressibility at reference condition. In order to reach this uncertainty, the maximum variations
of compressibility at constant pressure and temperature equal to the base conditions must be
calculated.
4.3.6 Superior Calorific value of the Gas
__________________________________________________________________________________________________________
Page 18/25
___________________________________________________________________________________
4.3.6.1 Uncertainty Inherited from the Chromatograph
The gas calorific value is contained in the volume conversion’s equation for the
compensation of the quantity of energy that the gas can provide in the combustion. The gas
calorific value is calculated based on the average individual calorific value of each fraction that
constitutes the gas, and as such, the chromatograph’s uncertainty used to obtain the molar
fractions will impact upon the calorific value uncertainty, which is calculated in Equation 26.
𝑁 𝑁 2
𝑢𝑃𝐶𝑆−𝑐𝑟𝑜 = √∑ [𝑢𝑥𝑗 ∙ (𝐻°𝑗 − ∑ 𝐻°𝑖 ∙ 𝑥𝑖 )] (26)

𝑗=1 𝑖=1
Where:
𝑢𝑃𝐶𝑆−𝑐𝑟𝑜 : Uncertainty of calorific value due to molar fractions uncertainty (MJ/m³);
𝑢𝑥𝑗 : Uncertainty of j (adm) component molar fraction;
𝐻°𝑖 , 𝐻°𝑗 : Ideal calorific power of i and j components in volumetric base, obtained in ISO
6976 (MJ/m³);
𝑥𝑖 : Molar fraction of i (MJ/m³) component;
𝑁 : Quantity of components in the gas (adm).
4.3.6.2 Calculation Method

The calorific value’s calculation method used in the case study is demonstrated in the ISO
6976 and holds an uncertainty inherent to the method of 0.015% of the calculated calorific value.
4.3.6.3 Base Values Uncertainty

The calculation method used in the case study is based on the pondered average of the
individual calorific value of the components. The values used as base were measured empirically
and hold an uncertainty inherited from these measurements. The base values demonstrated in
ISO 6976 cause an uncertainty of 0.05% in the calculated calorific value.
4.3.6.4 Daily Average

When the daily arithmetical average of the calorific value, which is applied for the
conversion of the daily volume, is accomplished instead of an online application, this causes an
uncertainty that is algebraically equal to the calorific value’s standard deviation of the analyses
throughout the day, divided by the square root of the quantity of analyses. Evidently, the greater
the variation of calorific value throughout the day is, the greater uncertainty this will be.
4.3.6.5 Composition Variation

Equal to what has been demonstrated for the compressibility, the chromatograph’s
sampling period may hide possible variations in the gas composition. In conservative way, in
order to estimate the uncertainty due to this sampling period, the greatest variation in the
composition between one analysis and another is verified, and then considering this variation as
an uncertainty of rectangular distribution.
4.3.6.6 Rounding
In some cases, the rounding of the calorific value occurs before the application of the
__________________________________________________________________________________________________________
Page 19/25
___________________________________________________________________________________
calculation. This rounding causes an uncertainty similar to the uncertainty due to the resolution.
4.3.7 Atmospheric Pressure
4.3.7.1 Atmospheric Variations

In the case of manometric transducers, the atmospheric pressure used in the calculation is
constant. Nonetheless, variations effectively occur mainly because the variations of the density
due to the humidity, composition, and temperature in the atmospheric layer, and from the
atmospheric tides that can change the air layer’s height. In order to estimate these variations,
periodic measurements of the local atmospheric pressure are made. The sampling standard
deviation divided by the square root of the quantity of measurements provides the uncertainty
value. It is important that the sampling collection take place in a representative period (at least a
year), so that the data be collected in various weather conditions. The application of the
Chauvenet method for elimination of spurious errors can be necessary depending on the data
source involved..
4.3.7.2 Barometric Calculation

In the case study, altitude measurements were made, and with this value the barometric
equation was applied, as shown in Equation 27. The uncertainties inherent to the method were
disregarded, applying only the uncertainties acquired from the variables used. In the case study,
the uncertainties inherent to the measured values of reduced gravity at the reference point and
the air’s molar mass were also disregarded. The calculation of the atmospheric pressure
uncertainty inherited from the variables that form the barometric equation is demonstrated in
Equation 28. The values were reached with the instrumentation in atmospheric balloons, such as
the ones provided by the INPE/CEPETEC’s metrological station.
−𝑀∙𝑔
𝑎∙ℎ 𝑎∙𝑅
𝑃𝑎𝑡𝑚 = 𝑃0 ∙ (1 + ) (27)
𝑇0
Where:
𝑃𝑎𝑡𝑚 : Local atmospheric pressure (mbar);
𝑃0 : Atmospheric pressure in the reference place (mbar);
𝑎 : Temperature variation rate in relation to altitude (K/m);
ℎ : Geopotential altitude in relation to the reference point (m);
𝑇0 : Temperature at the reference point (K)
𝑀 : Air molar mass (Kg/m³);
𝑔 : Local gravity (m/s²);
𝑅 : Constant universal of gases (J/mol.K).
__________________________________________________________________________________________________________
Page 20/25
___________________________________________________________________________________
2 2 2 2
√(𝜕𝑃𝑎𝑡𝑚 ∙ 𝑢𝑃0 ∙ 𝑘𝑃0 ) + (𝜕𝑃𝑎𝑡𝑚 ∙ 𝑢𝑎 ∙ 𝑘𝑎 ) + (𝜕𝑃𝑎𝑡𝑚 ∙ 𝑢ℎ ∙ 𝑘ℎ ) + (𝜕𝑃𝑎𝑡𝑚 ∙ 𝑢 𝑇0 ∙ 𝑘 𝑇0 )
𝜕𝑃0 𝜕𝑎 𝜕ℎ 𝜕𝑇0
𝑢𝑎𝑡𝑚−𝑐𝑏𝑎 = (28)
2
Where:
−𝑀∙𝑔
−1
𝜕𝑃𝑎𝑡𝑚 𝑃𝑎𝑡𝑚 𝜕𝑃𝑎𝑡𝑚 𝑃0 ∙ 𝑀 ∙ 𝑔 ∙ ℎ 𝑎∙ℎ 𝑎∙𝑅
= ‖ =− ∙ (1 + )
𝜕𝑃0 𝑃0 𝜕𝑎 𝑇0 ∙ 𝑎 ∙ 𝑅 𝑇0
−𝑀∙𝑔 −𝑀∙𝑔
−1 −1
𝜕𝑃𝑎𝑡𝑚 𝑀 ∙ 𝑔 ∙ 𝑃0 ∙ ℎ 𝑎∙ℎ 𝑎∙𝑅 𝜕𝑃𝑎𝑡𝑚 −𝑀 ∙ 𝑔 ∙ 𝑃0 𝑎∙ℎ 𝑎∙𝑅
= 2 ∙ (1 + ) ‖ = ∙ (1 + )
𝜕𝑇0 𝑅 ∙ 𝑇0 𝑇0 𝜕ℎ 𝑅. 𝑇0 𝑇0
𝑢𝑃0 : Uncertainty of atmospheric pressure estimated in the reference place (mbar);

𝑢𝑎 : Uncertainty of estimate of temperature variation rate in relation to altitude (K/m);
𝑢ℎ : Uncertainty of altitude measurement in relation to the reference point (m);
𝑢 𝑇0 : Uncertainty of temperature estimate in the reference point (K);
𝑘𝑃0 , 𝑘𝑎 , 𝑘ℎ , 𝑘𝑇0 : Coverage factor, calculated through the Student (adm) distribution.
4.3.8 Volume at reference conditions
4.3.8.1 Rounding of volume

The uncertainty from volume rounding should also be considered.
4.4 Methodology for the calculation of billing uncertainty
For the application of uncertainty values in management decisions, it is necessary to verify how
much this uncertainty impacts on the billing. This is done by applying the same methodology
demonstrated in Item 4.1. The mathematical model for the billing will depend upon the contractual
conditions. In this case study, the mathematical model for the billing in a specific type of contract is
shown in Equation 29 (the model may vary according to the contract).
𝑉𝑅 ∙ 𝑡𝑅 ∙ 𝑓 ∙ 0,0094
𝑅𝐵 = (29)
1 − (𝑃𝐼𝑆 + 𝐶𝑂𝐹𝐼𝑁𝑆 + 𝐼𝑆𝑆)
Where:
𝑅𝐵 : Gross revenue (R$);
𝑡𝑅 : Gas tariff in BTU (R$/BTU);
𝑓 : Conversion factor from kcal to BTU (kcal/BTU);
𝑃𝐼𝑆 + 𝐶𝑂𝐹𝐼𝑁𝑆 + 𝐼𝑆𝑆 : Gross revenue (adm).
The billing uncertainty is obtained with the calculation shown in Equation 30, disregarding the
uncertainties from the taxation.
__________________________________________________________________________________________________________
Page 21/25
___________________________________________________________________________________
𝜕𝑅𝐵 𝑢𝑡𝑅 2 𝜕𝑅𝐵 𝑈𝑉𝑅 2
𝑈𝑅𝐵 √
=2∙ ( ∙ ) +( ∙ ) (30)
𝜕𝑡𝑅 √12 𝜕𝑉𝑅 2
Where:
𝜕𝑅𝐵 𝑅𝐵 𝜕𝑅𝐵 𝑅𝐵
= ‖ =
𝜕𝑡𝑅 𝑡𝑅 𝜕𝑉𝑅 𝑉𝑅
𝑢𝑡𝑅 : Uncertainty due to rounding of tariff (similar to the resolution of a instrument) (R$);
𝑈𝑅𝐵 : Gross revenue expanded uncertainty with level of trust of 95,45% (R$).
5 PRESENTATION OF RESULTS
It is recommended that all the calculation memory is recorded evidencing the background of the
values utilized and the results presented according to the following model, with the information in the
same unit and with the same quantity of decimal places:
Y=(y E ±U) un : k=k’, level of confidence=IC
Where:
Y: Variable analyzed (Volume in the flow conditions or Gross Revenue);
y: Value of variable;
E: Systematic Error, including the signal;
U: Expanded uncertainty;
un: Unit;
k’ = Coverage factor;
IC = level of trust.
For example:
VR = (2.124.356,5 -86,2 ±261,0) m³ : k=2, level of trust=95,45%;
RB = (658.352,32 +91,51 ±154,01) R$ : k=1,96, level of trust =95%.
6 CASE STUDY
This methodology was applied in the Companhia Maranhense de Gás – Gasmar, in measurement
systems with ultrasonic meters average flow of 2.2 million m³ at reference conditions (101.325 kPa,
20°C), average manometric pressure of 32 bar, gas temperature of approximately 40°C, in 10”
measurement drains. Considering the processes applied by the company, by applying the proposed
method, an expanded uncertainty of a ±0.46% measurement and a billing uncertainty of ±1.35% were
reached. Figure 5 demonstrates the contributions of each variable in the measurement uncertainty value
composition. Through the analysis of this image it is possible to verify which variables cause more impact
to trigger a reduction in the uncertainty.
__________________________________________________________________________________________________________
Page 22/25
___________________________________________________________________________________
Patm VR N
<1% <1% 5%
PCS
31%
PF
39%
ZR
7%
ZF
10% TF
8%
Figure 5. Contribution of each variable to measurement uncertainty
7 APPLICABILITY AND BENEFITS
The analysis of the measurement uncertainty together with the financial impact assessment
provide data that can be utilized as subsidy to beacon the management decisions, such as the process mode
changes, so that the risk inherent to uncertainty be compatible with the application cost of process change.
This is the case, for example, of applying the risk based maintenance of components inherent to the
measurement (in addition to the continuity related risks), where the risk is defined by the uncertainty.
For the purpose of analysis, please see below examples of some cases where the methodology
can be used in management decisions:
i. Precision instruments generally have a higher cost because they use differentiating
technologies and systems. The acquisition of more precise and more expensive instruments
should precede a more detailed analysis concerning the impact that the variables may have
on the uncertainty, for instance, a simulation of the uncertainty using a pressure instrument
with a greater accuracy and a simulation of the uncertainty using an instrument with a lower
accuracy. The difference in invoicing between the application of both instruments should
be compared to the cost difference of the equipment investment, aiming at the best payback.
__________________________________________________________________________________________________________
Page 23/25
___________________________________________________________________________________
ii. The instruments calibration in the field provides a lower uncertainty in comparison to a
calibration in the laboratory. A higher cost is generated, though. With the analysis of the
measurement’s system uncertainty, it is possible to verify the financial risk of holding a
calibration with a higher uncertainty and compare it with the cost. The best cost relation by
risk is the ideal operation.
iii. The monthly cost of the chromatographic analyses should not override the cost concerning
the uncertainty that the chromatography causes in the measurement. The best periodicity
of chromatographic analysis can be defined from that.
iv. The periodicity of instruments calibration demands a more improved analysis than the legal
requirement’s compliance only. The measurement may hold an uncertainty that represents
a financial risk big enough to demand the adoption of a shorter calibration period. This
analysis is based on the As Found, verifying the instrument uncertainty before the
calibration, and on the As Left, verifying how much the uncertainty decreases executing
the adjustments. The uncertainty also guides the acceptance criteria in the calibrations,
analyzing which impact the uncertainty in the calibration would cause in the system as a
whole, adopting a criterion that would not cause a considerable impact on the volume.
v. In financial decisions, where the maximum possible difference between buying and selling
is controlled, based on independent measurements, providing subsidies for the
dimensioning of values that the company should keep in cash reserves to absorb the
possible differences. Moreover, it is possible to use this analysis in the decisions related to
the tariff plan, encompassing the financial risks inherent to the uncertainties regarding
measurements in buying and selling in the gas unit price.
The uncertainty of measurement generates intangible values related to the company’s image,
such as the increase in measurement’s reliability for both internal and external clients, thus enabling the
traceability for audits and arbitrations. The measurement becomes more transparent and clear, showing
the compliance to the contractual and legal requirements.
The possibility of analyses using the measurement system’s uncertainty is not limited to the
examples presented and can be used in various applications providing distinct results.
8 CONCLUSION
The guided analysis of uncertainty of gas flow measurement and billing can provide inputs
for management decisions, improving process efficiency, however, the use of those values for this purpose
must have great accuracy. The methodology presented in this paper outlines many variables that are
normally disregarded without any study performed to estimate its impact providing the required accuracy.
Due to the volume of data generated, the results obtained in the case study have not been
explained, since the values can differ depending on the systems and processes applied. The impact analysis
of each uncertainty should be made for each measurement system, so that the relevant uncertainties can
be defined.
__________________________________________________________________________________________________________
Page 24/25
___________________________________________________________________________________
9 REFERENCES
AGÊNCIA NACIONAL DO PETRÓLEO, GÁS NATURAL E BIOCOMBUSTÍVEIS;

INSTITUTO NACIONAL DE METROLOGIA, QUALIDADE E TECNOLOGIA. Resolução Conjunta
ANP/Inmetro nº.1. 10 de junho de 2013.
AMERICAN GAS ASSOCIATION. AGA Report No. 9: Measurement of Gas by Multipath
Ultrasonic Meters. Second Edition. 2007
AMERICAN GAS ASSOCIATION. AGA Report No 10: Speed of Sound in Natural Gas and
Other Related Hydrocarbon Gases. 2003.
DANIEL MEASUREMENT AND CONTROL. Daniel Ultrasonic Gas Flow Meters with Mark
III Eletronics. Revision 5. 2013.
EMERSON PROCESS MANAGEMENT. Folha de dados do transmissor de pressão Rosemount
3051. 2011.
EMERSON PROCESS MANAGEMENT. Folha de dados do transmissor de temperatura
Rosemount 3144P. 2011.
INSTITUTO NACIONAL DE PESQUISAS ESPACIAIS, CENTRO DE PREVISÃO DE
TEMPO E ESTUDOS CLIMÁTICOS. http://www.cptec.inpe.br. Acessado em 01/04/2015.
INTERNATIONAL ORGANIZATION FOR STANDARDIZATION. International Standard
ISO 6976: Natural gas – Calculation of calorific values, density, relative density and Wobbe index from
composition. Second Edition. 1999.
JOINT COMMITEE FOR GUIDES IN METROLOGY. Evaluation of measurement data –
Guide to the expression of uncertainty. First edition. 2008.
STARLING, K. E., SAVIDGE, J. L.. AGA Report No. 8: Compressibility Factors of Natural
Gas and Other Related Hydrocarbon Gases. Second Edition. 2003.
__________________________________________________________________________________________________________
Page 25/25
20th – 23rd October 2015
Maximising Economic Recovery –

A Review of Well Test Procedures in the North Sea
Craig Marshall, NEL

Alun Thomas, NEL
1 INTRODUCTION
1.1 General
In terms of Maximising Economic Recovery (MER) in the UKCS, the measurement of well
production rates is essential to best optimise the hydrocarbon production strategy from within
the well itself. This is achieved during a process called a well test where a snapshot of
production is monitored by measurement equipment and instrumentation. The data produced
is then used to optimise the wells production rates.
However, how accurate that measurement has to be to provide sufficient control has not been
established and there is little information in the public domain that shows what the current
typical operation measurement uncertainty is.
NEL have reviewed current industry practice in regard to well testing on the UKCS and have
received assistance from operators in the region. The information provided was used to
generate typical uncertainties found in well testing using test separator systems. These results
have then been compared with results from models assessing the impact of flow measurement
errors on recovery factor and hence economic recovery of hydrocarbons.
In addition, alternative methods available to complete well testing and how each of these can
impact on MER have been described.
1.2 Oil and Gas Authority
The requirement for MER is clear; since the first licenses were issued for production in the
North Sea, the industry has spent more than £500 billion in exploration, development and
production activities [1]. There are over 450,000 people employed in the oil and gas industry
in the UK and HM Treasury has received more than £310 billion in production taxes alone;
not to mention the subsequent revenues generated from income tax, exports and
manufacturing.
There are many impressive facts relating to the industrial and economic benefits of the
offshore oil and gas industry and to continue to reap these benefits would be in the best
interest of the UK. However, not only to reap the rewards, but to maximise those rewards
should be a plausible and achievable goal.
In February 2014, Sir Ian Wood published the Wood Review [1], an independently led
review of the UKCS’s oil and gas recovery potential in the coming decades. The aim of this
review was to assess the current state of operations, production and exploration in the UK and
the methods of licensing and regulatory control in order to make recommendations for MER.
In essence, the Review calls for a stronger and more authoritative regulator to work with both
Industry and HM Treasury in a new tripartite strategy for MER within the UKCS.
1
The Oil and Gas Authority (OGA) was formed on 1st April 2015 with the responsibilities
being transferred from the Department of Energy and Climate Change (DECC) by the Energy
Bill Bill, which is currently (as of 1st October 2015) at the Committee stage in the House of
Lords. The progress of the bill, and the bill itself, may be reviewed at:
http://services.parliament.uk/bills/2015-16/energy.html [2].
OGA’s responsibilities include the regulation of fiscal oil & gas measurement & allocation
[3]. This work is carried out by the OGA’s Petroleum Measurement & Allocation Team
(PMAT), which is part of OGA’s Exploration & Production directorate. The role of OGA is
to work with government and industry to ensure that the UK obtains the maximum economic
return from its oil & gas resources.
A more in depth overview of OGA and their role within the offshore oil and gas industry can
be found at:
https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/458593/A5_O
GA_Overview.pdf [4]
1.3 Well Testing
1.3.1 What is well testing?
Well testing is extremely valuable in the oil and gas industry as it allows operators to assess
the production performance of their wells to glean important information about its structure
and characteristics. As a whole it is an expensive operation involving significant resources
and logistics, and differs from most techniques as it requires the reservoir to be in a dynamic
state, as opposed to a static state, in order to activate the responses needed for mathematical
modelling.
A basic well test system consists of a subsurface string, incorporating downhole tools such as
gauges, check valves, flow switching valves, isolation valves and packer assemblies, together
with a surface or deck system for separating, sampling and metering the fluids flowing from
the well.
Well tests incorporate many aspects of operations from drilling and process plant production
and are performed in order to estimate reservoir properties. They are used to obtain dynamic
data from a reservoir during different stages of that reservoir’s life. During the exploration
phase, the results of well tests provide the key dynamic data which will directly affect
decision making regarding further development.
Well testing objectives are diverse and can be used to confirm the existence of hydrocarbon
fluids in the drilled wells, to obtain downhole samples and to characterise the reservoir. The
duration of a typical well test is usually short, of the order of tens of or hundreds of hours.
The main well test deliverables that can influence MER and will be discussed further in this
report are:
 Reservoir parameter characterisation

 Reservoir model selection
 Production flowrate determination
2
These three deliverables link closely to MER through reservoir optimisation and the ability to
maximise the recovery factor for the well.
1.3.2 Reservoir characterisation
Typically, reservoir characterisation is achieved by finding a model that matches the

empirical data which can provide the well characteristics such as flow capacity (i.e.
permeability-thickness product), skin factor, and the structural and/or hydrodynamic
boundaries.
The interpretation and ultimate utility of the well test data, which necessarily encompasses its
uncertainty, is linked to knowing the particular reservoir’s storage capacity i.e. porosity. One
important use of well test data is to determine (a) if a static model e.g. a geological one,
behaves in the same way as the real reservoir and (b) to enhance the predictability of that
model.
Well testing can also be used anytime during the life of a reservoir to diagnose strange
behaviour e.g. an unexpected gas-oil ratio, or an unexplained reduction in productivity.
Differing types of well testing procedure are designed to serve particular purposes, but in all
cases there will be some sort of controlled constant productions (or injections) while
recording the pressure data. For those cases where a well flows at a particular production rate,
it is termed a draw-down test, whilst when the well is closed or shut-in, it is termed a build-up
test as shown in Figure 1.
Figure 1: The concept of draw-down and build-up well tests [5]
The ultimate goal of either is to describe a reservoir such that it can reproduce the same
output for a given input signal. Therefore, because well testing is effectively an inverse
problem - one which needs the data to match the model - its interpretation largely depends
upon the quality of input and output data. Hence, the focus of the study in investigating the
role of measurement uncertainty upon MER.
3
1.3.3 Production flowrate determination
Owing to the adverse conditions upstream of the well head, there is great difficulty in
monitoring component flowrates with great accuracy. Instead, the produced fluids are
isolated from other producing wells and sent to a test separator. The test separator separates
out the individual components of the flow into either liquids or gases – as in a two phase
model – or oil, water and gas – a three phase separator (more commonly used in the North
Sea). The separated components are then measured individually by single phase flow
measurement technologies as shown in Figure 2.
Figure 2: Three phase test separator flow measurements [6]
Using these measurements over the length of the well test it is possible to acquire a snapshot
of the production rates for that point in time. These values are then used as the well rates until
they are updated by the next series of well test data collected.
Coupling this flowrate data with reservoir parameters allows reservoir engineers to model
specific wells in order to optimise its production profile i.e. get as much hydrocarbon
produced versus as little water as possible.
It is intuitive that completing reservoir optimisation successfully depends on accurate data

being supplied. However, how accurate the data needs to be is unknown and one of the
questions the study aims to answer.
From Figure 2, it is easy to imagine process situations where measurement performance can
be worsened. In addition, other practical issues may arise that result in measurement errors.
Essentially, any deviation from non-ideal conditions will cause a detrimental effect on the
flow measurement which leads directly to poorer data for reservoir optimisation and hence
recovery factor.
There are many influences that can cause flow measurement error at the outlet of test
separators. Some common concerns are [7]:
 Presence of a second component (carry-under and -over)
4
 Installation effects
 Fouling on measurement device
 Calibration expired
 Poor maintenance/inspection
 Secondary instrumentation error
 Pulsating flows
 Representativeness (do the flow rates during the well test represent the actual typical
conditions)
With so many potential measurement issues present, some being persistent, it is beneficial to
know what the overall cost of the measurement problems are. This way an appropriate cost
benefit analysis can be completed and the risk in regard to MER can be properly assessed.
This ties in with current OGA measurement uncertainty strategies and the recommendations
from the Wood Review.
3 REGULATION OF OFFSHORE OIL & GAS PRODUCTION
3.1 Current Guidelines
Currently, the OGA guidelines [3] do not specify a requirement for the performance of test
separator measurements unless they are used for, or to verify a multiphase meter used for,
allocation or fiscal measurement. However, the guidelines do recognise that it is normal and
good oilfield practice to periodically test production wells for the purpose of reservoir
management.
Typically, a well test frequency is in the region of 30 – 90 days and is dependent on a number
of variables. When required by OGA, the frequency is decided in agreement with the
regulator. This is normally based on a risk-based approach concerned with financial exposure
of any produced hydrocarbons with a normal frequency close to 30 days.
If the well test frequency is not met for wells where OGA are interested, then dispensations
must be obtained with appropriate reasons included.
Proof of measurement uncertainties for test separator systems may also be required to
validate measured values for OGA. In fields where OGA have no interest the driver for
obtaining the proof through measurement audits and calculations of uncertainties lies solely
with the operator. Anecdotal evidence from OGA suggests that in these installations, the
measurement uncertainties are typically larger than recommended by good oilfield practice
owing to the influences suggested in section 1.3.3.
In fields where OGA has an interest, proof of measurement uncertainty should be provided
either through operator’s obtaining audits from independent companies or through OGA’s
own inspection team. However, the Wood Review noted that DECC (before OGA were
formed) were severely understaffed and current OGA staff levels still do not permit the
inspection of every test separator measurement system. This results in a dependence on
operators obtaining their own verifications.
5
4 CURRENT WELL TEST PROCEDURES
4.1 General
In order to assess the current state of well test procedures in the North Sea a questionnaire
was issued to operators to gauge typical well test frequency, success and methods.
Information was also sought about the equipment used to ascertain how fit for purpose it is in
regards to helping to maximise economic recovery [8].
A very low response rate was achieved predominantly due to the decrease in oil price over
the consultation period. Therefore the results obtained are a snapshot of industry and
unfortunately do not represent a significant sample to be applied to all. From the 32 operators
contacted, responses were received from 6 and for 7 installations. Overall this has led to data
from over 100 wells included in the uncertainty analysis.
4.2 Well Testing Methods
From the responses received it was clear that there are obstacles to accomplish successful
well testing. It is important to point out that some of the responses were negative in terms of
how companies approach well testing i.e. none or very little were completed. This was
typically due to the physical set up of wells either being in satellite fields or tie-backs. This
meant that wells could not be tested individually without shutting down other production
wells. Clearly this would be in conflict with company financial policies especially given the
time constraints to test each well and the number of wells.
Another issue was concerned with the amount of wells for an individual installation. In one
case, over 100 wells were in use to varying degrees and as such it was almost impossible to
accurately well test each one within a suitable time frame.
Of the submitted questionnaires where well tests were completed it was found that the vast
majority reported that a 4-8 week frequency was typical and this timescale was achieved in
close to 100% of wells. This timescale ties in well with expected values as prior to receiving
these answers 4 weeks was believed to be industry norm.
The typical well test length was in the region of 1-3 days with variations from 1 hour up to 7
days. This accounts for set-up, attaining steady state conditions and holding for the well test
duration. It is not uncommon for well tests to last up and over 70 days but these are usually
only in rare circumstances. Again, the values attained were similar to what was expected
from experience.
Considering an average 60 day well test campaign period and average duration to be around 2
days, it is easy to understand how installations with more than 30 wells begin to place
pressure on test separator systems. It is important to note this does not include change over
time or more importantly, maintenance and calibration time for the equipment.
In terms of operations it was reported that it is the operator themselves who have written the
procedures, perform and analyse the data coming from well tests. Service companies were
not used in this sample.
6
4.3 Uncertainty
In terms of the measurement uncertainty for each phase the data provided was used in generic
uncertainty budgets to give estimates for typical values. The requested data included
information about the single phase measurement devices and stream configurations for the
test separator systems:
 Flow meter used

 Make/Model
 Meter size
 Upstream straight pipe length
 Downstream straight pipe length
 Temperature measurement
 Pressure measurement
 Calibration method
 Calibration frequency
Data was also provided for each well that the test separator is used to test. Component flow,
pressure, temperature, physical properties and contamination levels were stated or estimated
for use within the uncertainty model.
The generic uncertainty budgets consisted of the following sources of uncertainty with the
values being determined from provided data or through previous history of the metering
technologies used:
 Baseline meter uncertainty

 Calibration uncertainty
 Installation effects
 Contamination of second component
 Drift since calibration
 Temperature
 Pressure
As the supplied data was relatively low in number multiple runs were completed on the data
where certain sources of uncertainty were generated randomly based on previous history. For
instance, it is known that some fluids contain more impurities than others and hence cause
larger amounts of drift of measurement equipment through deposition. In order to account for
this, levels of drift of between 0.1% and 0.5% per annum were randomly generated and
applied to the data.
The data supplied from operators included information on the metering technologies and
dimensions, installation criteria, secondary instrumentation and calibration history. Then
flowrate, temperature, pressure and physical property data were used in the uncertainty model
of the equipment to generate estimated and representative uncertainty values for the data
given. As stated before, additional runs were incorporated given random values of the
uncertainty sources so as to give more data and to include extreme cases. This should give
more confidence that the results obtained are representative given the low population sample.
The ranges of uncertainties found were quite varied for the wells and were different for
liquids and gases. For liquids, both oil and water uncertainties were similar and were in the
7
region of ± 1-5% where the larger numbers were associated with extreme cases of deposition
and contamination of second phases. In general, the uncertainty values were between 2 and
3%.
For gases, the uncertainties were slightly higher at values between ± 2-7%. Again the larger
numbers were attributed to cases under extreme conditions such as contamination and drift.
In general, the uncertainty values were between 3 and 5%.
These values are for measurement uncertainty at the outlets of the test separator. When
discussing the uncertainty of the flows in the well (more useful for reservoir modelling) then
volume corrections, flow pulsations and other contributing factors must be included. Once
they are accounted for the uncertainty can rise to over ± 10% (estimated).
The calculated numbers are similar to values expected from experience and general industry
held beliefs. The uncertainty relating to certain sources are subjective and may vary from
case to case. Only general rules of thumb have been applied which give an indication of
expected values. In addition, other separation methods for example a two-phase system will
have other contributing uncertainty factors.
The results are based on a desk-based assessment of provided data. A site audit of each
facility would give better indication of the current state of the art for test separator
measurement. This will allow for more accurate uncertainty budgets to be developed and give
a larger sample of uncertainties to assess.
5 WELL TEST SIGNIFICANCE
5.1 General
Throughout this report the term ‘accurate measurement’ has been used to discuss the
requirement for the quality of topside measurements for reservoir optimisation. However, it
has also been noted that the absolute level of quality required to ensure successful
optimisation remains unknown.
To investigate this problem, Coventry University completed a study relating to the

uncertainties in well test measurements and their consequent impact on MER [5]. The study
was a high level assessment of generic well test metering, including multiphase measurement
used during a standard well test. The focus of the Coventry work lay in running a number of
simplified models in order to explore the importance of rate measurement for well test
interpretations; as opposed to developing in-depth models akin to those in use commercially.
Coventry’s scope encompassed downhole rate measurement as a necessary means of

comparing and contrasting such measurement with surface techniques and, overall, the study
had much to say about downhole techniques. Nevertheless, the intent of the modelling was to
establish, in broad terms, the nature and strength of the link between surface well test
measurement uncertainty and its importance to maximising future extraction. This section of
the report summarises the work completed by Coventry and elaborates on the conclusions
drawn.
Firstly, it is important to understand the process in which flowrate measurements are used in
reservoir optimisation and production. In order to successfully optimise production from a
8
particular well, the well itself has to be characterised so that its future production can be
accurately modelled with a low uncertainty. Only once production can be predicted can the
most optimum production pattern be obtained.
There are two parts to this prediction, the characterisation of the parameters within the well
itself e.g. porosity, permeability, skin factor etc and the model used to calculate the outputs
given the input parameters. Both of these parts are determined through data provided through
well tests. Traditionally, surface flow measurements have been a key component in the
analysis.
Characterising well parameters and selecting the most appropriate model to use is influenced
by data taken during well tests but the process involved is outside the scope of this study.
This report and the conclusions from Coventry’s work do not consider this process; only the
quality impact of the data on the recovery of oil and gas is considered.
5.2 Description of Work
An example reservoir was created and a series of test runs were conducted to assess the
output from the example with respect to the changing input parameters. The example
reservoir was based on a 100 ft vertical well within a fractured reservoir with a radius of
5000 ft. The following were used as the flow parameters of the reservoir:
 Storativity ratio (ω) – 0.1

 Inter-porosity flow coefficient (λ) – 2 x 106
 Permeability (K) – 500 md
 Bulk porosity (S) – 0.27
The parameters can be taken as descriptors of how fluids flow through a reservoir and there
exact definitions can be found in various sources. However, for the purpose of this study they
can be thought of as inputs to a model where the closeness of the predicted values of these
inputs to the actual values dictates the accuracy of the model as a whole.
During each test run, the example reservoir was ‘produced’ with varying levels of
measurement information recorded and utilised. This measurement data was then used to
generate reservoir models and the predicted reservoir parameters. A comparison could then
be made between the accuracy of the model and correct parameters in the example reservoir.
The test runs consisted of a single phase oil drawdown phase at a constant flow of 9200
STB/D with a duration of 158 hours. Then the well is then shut-in for 8 hours for a build-up
phase before being produced again.
The second stage production could be applied for any time frame and for these tests the well
was assumed to produce fluids for 20 years allowing for a direct comparison of overall
production rates i.e. how much total hydrocarbon was recoverable over the timeframe
compared with values obtained during other test runs. This then allows comparison as to
which methodology allows for maximising recovery factors and hence MER.
The test runs considered during these tests were:

1. Correct flow rate measurements taken at the surface
2. Correct flow rate measurements taken downhole
3. 10% random error in flow rate measurement taken at the surface
9
4. 10% random error in flow rate measurement taken downhole

The surface measurements are defined as measurements above the well head i.e. either test
separator measurement systems or multiphase flow meters. Downhole measurements are
defined as measurement taken in the well bore typically at the well perforations.
5.3 Results
Consider the comparative cases of well tests where the flow rate measurements are taken at
the surface and downhole respectively (test run 1 and 2). During the drawdown and build-up
phases the measured flow rates would be as shown in Figure 3.
10000
9000
8000 Downhole Rate Measurements
Oil reate, STB/day
7000
Surface Rate Measurements
6000
5000
4000
3000
Build-up
2000 Drawdown
1000
0
140 145 150 155 160 165 170
Time, hr
Figure 3: Flow rate measurement location during well test
It is clear that there are different flow rates measured using both methods particularly at the
transition between drawdown and build-up. This is due to the fact that once the well is shut-
in, there will be no flow at the surface i.e. the measured rate drops instantaneously to zero.
However, with measurement downhole, once the well is shut-in the reservoir still flows until
it reaches equilibrium where there is a pressure balance and the produced area becomes stable
again.
Surface flow rate measurements do not record this additional flow post well shut-in and
therefore do not include them in parameter predictions which can cause error as shown in
Figures 4 and 5.
1000 1000
Realistic Data
(Normalized Superposition Curves) ΔPn, ΔPn'
Model Fit
Homogenous Model Fracture and

K= 500 md 100
100 Skin λ matrix flow
ΔP & ΔP', psi
Skin= 0
(S)
ω
Dual Porosity Model
10 WBS 10 K=470 md
S=0.07
ω=0.1
λ=2e-6
1 1
1E-4 1E-3 1E-2 1E-1 1E+0 1E+1 0.001 0.01 0.1 1 10
Time, hr
Time, hr
a) b)
Figure 4: Well test diagnostics plot for test run 1 (a) and 2 (b)
10
Figures 4a) and b) show the well test diagnostic plots for test runs 1 and 2 respectively. The
shape, slopes, and plateaux of the curves found on a well test diagnostic plot are used to
estimate parameters to describe a reservoir which will be used in a reservoir model.
Figures 4a) and b) are describing the same reservoir but it is clear that Figure 4b) conforms to
a model better than Figure 4a). This results in lower modelling uncertainties and better
prediction capabilities suggesting that for MER, surface flow rate measurements are not the
best method of measurement available.
Figure 4a) denotes an area as Well Bore Storage (WBS) on the curve. This is an effect that
masks well flow rates from surface flow rate measurements through essentially a dampening
effect. Owing to the distance, pressure differential and other factors, the production profile at
the well perforations and any associated pulsations or changes in component fractions at
these points will be ‘smoothed’ out as the fluids flow to the production platform. What could
have been a high pressure region or high water cut region will be averaged out by the rest of
the fluids meaning the information will be lost. This is apparent in Figure 4a) and is one of
the key reasons, the recorded data is more scattered and does not fit the suggested model
well. Again, it is important to point out the data itself is used to help choose the correct model
to use.
WBS will be present in all wells but will affect the measured results to varying degrees
depending on the measurement location. This is another example suggesting that for MER,
downhole measurements are potentially the most promising.
1000
No-Error
10% Error
Normalized Superposition Curves) ΔPn, ΔPn'
100
10
No erro in downhole rates 10% erro in downhole rates
K=470 md K=550 md
S=0.24 S=0.07
ω=0.1 ω=0.1
λ=2e-6 λ=2e-6
1
0.001 0.01 0.1 1 10
Time, hr
Figure 5: Well test diagnostic plot for test run 2 and 4
For test runs 3 and 4 a 10% error was introduced into the flow rate measurements to assess
the impact of these errors on the recovery factor of hydrocarbons. It was found that a 10%
flow rate measurement error resulted in a 10% reservoir parameter estimation error (Figure
5). The location of the flow rate measurements did not impact this relationship. This error is
different from the error introduced through whether or not flow after shut-in is utilised in the
estimation.
However, these parameter errors were not found to be very significant on recovery factor.
Even with large errors present in the reservoir parameters, this does not impact linearly with
11
how much hydrocarbon is produced from the well by itself. As an extreme case, when the
flow rate error is ± 50% under the conditions in the example reservoir, the recovery factor
after 20 years was found to have 3% error only. Essentially, for every 1% error in flow rate
measurement there is an error in recovery factor of 0.06%.
As discussed earlier there are two parts in reservoir prediction in order to optimise production
successfully. The main impact on MER from flow rate measurement errors is not from the
reservoir parameter estimation but from the use of the data to select the most appropriate
model.
For each test run, the data generated on reservoir parameter estimations and the most
appropriate reservoir model were used to ‘produce’ the example reservoir for a period of 20
years. For test runs 1 and 3 a single medium model was selected from the data and for test
runs 2 and 4 a dual medium model was selected. Dual medium denotes a reservoir fracture
was detected whereas single medium denoted no fracture.
Figure 6 shows the effect on recovery factor after 20 years using the each model and reservoir
parameter estimations.
0.4
Dual Medium
0.35
0.3
Single Medium
Recovery Factor
0.25
0.2
0.15
0.1
0.05
0
0 5 10 15 20 25
Time, Year
Figure 6: Recovery factor after 20 years
The results show that the model uncertainty has higher impact on the final recovery. Using a
single medium model the reduction in the recovery factor is around 12% compared with the
dual medium model. Potentially this could be a huge number in terms of financial
calculations. To give an idea that how much this would be consider that the initial oil in place
for the example reservoir is around 3.78×107 barrels of oil. Incorrect flow rate measurement
data either from errors or from location factors can result in around 5% decrease in producing
the original oil in place. Or at $60 per barrel, this is equivalent to £75 million for this
example.
In this study Coventry have highlighted the potential of flow rate measurements and in
particular downhole rate measurements for improved well test data, and demonstrated the
contribution of accurate measurements on reducing uncertainty in modelling, parameter
estimation and the ultimate recovery. Simulations were performed using the models that are
frequently used in the well test interpretation routines and only include the “direct” impact of
flow rate uncertainty on the recovery factor by extending the well test models for long term
predictions. As a consequence, only natural production mechanisms are considered and other
improved recovery techniques (such as water or gas injection) are not considered. In practice
there are “indirect” uncertainties induced from inaccurate rate measurements. Such
12
uncertainties could result in making incorrect or sub-optimal decisions for reservoir

developments that can ultimately impact the secondary or enhanced oil recovery plans.
Assessment of the impact of such uncertainties in general cases is difficult, as each reservoir
requires its own considerations and different policies than the other reservoirs.
It is important to rationalise the study completed by Coventry for its applicability and
representativeness to the industry. The example reservoir modelled was simple in design in
comparison with ‘real world’ reservoirs. The results obtained are specific to the example and
are only given as indications of the potential significance of flow rate measurement error (and
hence uncertainty) to MER in the UKCS.
For the test runs including 10% error, the results consider the error introduced to the
permeability estimation only. Coventry recommended further work to determine the whole
effect of all reservoir parameters on recovery factor.
5.4 Summary
The results obtained in the work completed by Coventry show relatively small effects of
surface flow rate measurement errors on recovery factor (considering reservoir parameter
estimations only). However, production optimisation is not the only use of well testing in the
industry and there will be more significant consequences from flow rate measurement errors
in these applications.
Well flow rate measurements can be used in the allocation of produced hydrocarbons
between wells or fields, in the verification of multiphase meters and the decision making
process of field developments i.e. where to drill the next well. The significance of flow rate
measurements here can influence MER in different ways.
Where multiphase meters are used to monitor well flow rates continuously or during well
tests they will still be verified periodically against separator measurement systems. This acts
as a verification of the performance of the equipment and potentially can be used to apply
corrections (this assumes a better uncertainty in the separator measurements). There can be a
compounded effect of measurement error from separator system through to multiphase meter
through to reservoir parameter estimations.
In field developments the data from producing wells is used to assist in the positioning of
future drilling. Depending on where a new well is drilled may influence how much
hydrocarbon can be produced or the effectiveness of any injections for EOR.
Allocation is used to apportion produced fluids to particular wells in the event of comingling.
The allocated values can be determined from separate meters i.e. multiphase meters or by test
separator systems. Allocated fluids tie in with MER not through normal well test processes
but through continual monitoring of streams. The ability to quickly and in real-time
determine if water breakthrough is occurring is extremely valuable in MER.
It is clear from the completed study that the significance of flow measurement uncertainty
(and errors) cannot be linked directly to one output as the whole process is intrinsically linked
to varying degrees. Where one input is weakly linked to the output through one mechanism it
can be strongly linked through another. This coupled with significant secondary affects
makes it difficult to place absolute values on the implications of measurement accuracy.
13
Instead, some examples have been presented with their stated limitations and sample values
calculated.
In summary, in terms of MER in the UKCS the study has found flow measurement is vitally
important in ensuring its success. However, the most significant contribution is not from the
measurement error or uncertainty itself but through the use of the data to choose an
appropriate reservoir model to use for reservoir optimisation. This ties in with not only the
quality of the measurements themselves but on the location and appropriateness of the
measurements.
Coventry University would recommend the use of downhole flow rate measurements as these
provide additional information on well flows not available from surface measurements. This
information has been shown to be extremely valuable in terms of MER through the models
run.
Another recommendation from Coventry is the use of multiphase meters instead of test
separator systems for surface flow rate determination. Primarily, for the reason they are able
to record and trend data continuously resulting in an improvement in data quality and
confidence of estimated reservoir parameters.
6 ALTERNATIVE METHODS AVAILABLE
6.1 General
The need of well testing for maximising oil and gas production and economic recovery,
formulating fiscal policy and attracting investment have been evident for years. In
conjunction with conventional well testing technologies, a number of alternative methods
have been developed to aid such operations since the early eighties by research organisations,
meter manufacturers, oil and gas production companies and others. A combination of
technologies has emerged, albeit their prototypes were quite dissimilar in functions and
designs. These technologies have become commercially available and their applications and
users are rapidly expanding.
Well testing for production analysis [9] involves measuring the contribution of oil/water/gas
from individual wells. Indeed, the knowledge and data obtained from well testing are then
utilised to facilitate reservoir management, field development, operational control, flow
assurance, production and fiscal measurements.
In this study, three popular alternative methods are chosen to be reviewed. These are:
 virtual flow metering

 multiphase flow metering
 mobile well testing.
Virtual flowmeters (VFMs) are calculation based software suites that incorporate existing
measurements instruments to create estimation of flow in real-time at any point in the
process; with or without flow measurement technology being available. Multiphase
flowmeters (MPFMs) are physical measurements systems installed inline that measure the
component flows of oil, water and gas in real-time. Finally, mobile well testing (MWT) is
where portable separator systems, often combined with two or three phase measurements, are
14
installed for short periods before being removed. The following section will discuss these
three popular methods for the use in MER.
6.2 Virtual Flow Metering
Although there are differences among VFM technologies, there is no attempt to undertake a
product-by-product comparison in this review. The perceived advantages and disadvantages
of VFMs are generalised and summarised in Table 1.
TABLE 1
Advantages and Disadvantages of VFMs
Advantages Disadvantages
Provision of real-time continuous monitoring Not suitable for responding transient pressure
of reservoir characterisation, well due to rapid inflow. Detailed knowledge is
optimization & economic recovery. required to set up VFMs.
Cheaper than multiphase flow meters & Installation effects could affect the accuracy
mobile well testing facilities. of flow rate calculation.
Useful for visualisation of component phases Accuracy of component phases are required to
and flow assurance risks [10]. be predicted based on pressure and
temperature measured in and around the wells.
Output flow rate uncertainties are unavailable.
Capable of recalibration if required. Regular verification is required to be
undertaken using real time measurements to
ensure accuracy & representativeness of flow
rates predicted
Elimination of maintenance requirements, Corrosion, sand plugging with wax and
such as corrosion, sand plugging with wax, methyl hydrates would potentially affect the
methyl hydrates etc. accuracy of VFMs.
Cost effective and efficient for monitoring a Less efficient and accurate if meter fouling
number of wells simultaneously as opposed were not identified in and round the well.
to a physical well testing.
Elimination of operational risks, such as fluid Operational risks are simply transferred to
surges, leakages, formation of hydrates, etc. service providers who have to carry out
maintenance work.
Reliable and has good fault tolerance Less reliable if failed hard temperature and
pressure sensors were lack of regular
calibrations.
Useful for validation of well test, VFMs consistently either over- or under-
reconciliation for and replacement of flow predict as the choke size increased (i.e.
meters [11]. increase of flow rate). Some programs can
freeze and crash and significant time may be
spend opening, closing and switching between
analyses and post-processing the data.
Useful for tracking liquid slugs, MEG, pigs, The drop in differential pressure also increases
identification of hydrates, control of chokes, the uncertainty of the flow rate through the
monitoring for erosion, corrosion, wax or choke.
leaks.
Useful for looking ahead and for performing The risk of bias or error in the flow rate
what if scenarios and uncertainty analysis. estimations increase, if the VFMs were
operating outside the trained range.
15
VFMs are capable of reliably determining the well flow rate, if and only if the pressure,
temperature and other physical sensors were periodically calibrated and VFM flow rates
against calibrated and accurate multiphase flow meters were verified [12]. Preferably,
uncertainty performance of the various system components, e.g. downhole and subsea
pressure and temperature sensors, subsea or topside multiphase meters, a length of tieback
pipeline, a topside choke valve, separator with single-phase flow and water cut meters, are
regularly carried out for calibrating and verifying VFMs. At present, uncertainty ranges of
VFMs are rarely known and are not widely published. Critical and independent evaluation of
the performance and accuracy of VFMs for subsea well tests are virtually unavailable in the
public domain. Nonetheless, it was reported that VFMs could deviate from the multiphase
flow measurements by a range of 10% to 20%.
Even when accuracy is not considered sufficient, VFMs still offer valuable information from
their repeatability. Comparison and trending of the VFM result with other technologies, such
as MPFMs, can indicate when they require verification. In this sense they can qualitatively
check primary measurements for reproducibility and act as a redundancy.
6.3 Multiphase Flow Metering
Before 1980s, single phase measurements alone were sufficient to satisfy the needs of oil and
gas industry. The maturity of oil reserves together with smaller and deeper wells with higher
water contents saw the needs of multiphase flow meters (MPFMs). Since 1994, the
installation of MPFMs has steadily increased, with substantial growth from 1999 onwards. A
recent study suggested that there were approximately 2,700 MPFM applications covering
production optimisation, field allocation and mobile well testing.
Initially, the purpose of MPFMs was to analyse the increasing amounts of water and natural
gas that were producing in a greater part of fluid from some North Sea wells. However, oil
wells have a mixture of oil, natural gas, water and other chemical compounds. Ultimately, the
multiphase flow metering is defined as the utilisation of MPFMs, which are devices used to
measure the individual flow rates of constituent phases in a given flow of petroleum, gas and
water mixtures produced during oil production and well testing processes [13].
The initial interest in multiphase flow metering came from the offshore industry. In fact, the
North Sea oil field played a pivotal role in the development of MPFMs. The key factors
which have instigated the speedy uptake of multiphase flow metering technology are
identified as follows:
 increases in oil prices

 decreases in meter costs
 improved meter performances
 wider competition amongst operators
 deployment of more compact meters for mobile systems.
Prior to the development of MPFMs, the only way to determine the fluid mixture was to
physically separate the fluids and to individually measure each phase. The main advantage of
state-of-the-art MPFMs is the elimination of the use of three-phase separators. The flow rates
and composition of the fluid from the well can be determined by MPFMs without using a test
16
separator. Consequently, this offers a substantial economic and operating advantages over
conventional phase separating predecessors.
MPFMs provide important information and data about the fluid compositions in the well.
They also provide indications over the increase in the percentage of water coming out of a
well and the decline of oil production. MPFMs give a faster response time than their test
separator counterparts, because they log and analyse data in real time. MPFMs [14] are more
useful for allocation metering where produced fluids from different wells need to be
combined and sent to a processing station.
The main disadvantage of MPFMs, however, is their high capital and operating costs
involved. In particular, there are currently no single MPFM on the market meeting all
multiphase metering requirements. Other disadvantages include contaminations, sand
erosions, blockages and difficulty of calibration. All of these impose a higher level of
measurement uncertainties to MPFMs [15].
Table 2 generalises the advantages and disadvantages of MPFMs identified from literatures.
TABLE 2
Advantages and Disadvantages of MPFMs
Advantages Disadvantages
Elimination of test separators High capital, maintenance and software
costs
No requirements of regular intervention of High measurement accuracy requires more
test separator by qualified personnel intense radiation sources, which affect
safety aspects and mobility considerations.
Continuous well monitoring possible. This The phase fractions will only be
is not practical using a test separator representative over a cross section if the
phases are homogeneously mixed
No requirements of long period of steady Contaminations, sand erosions, blockages
operation to achieve accurate and difficulty of calibration.
measurements. Information is available to
the users in a minutes after starting the
operation
Ability to track, in real time, any changes in Intrusive
fluid composition, flow rates, pressure and
temperature.
Ability to respond quickly to changes in High measurement accuracy requires more
fluid composition and the reduced time to intense radiation sources, which affects
stabilize flow H&S and mobility considerations
No moving parts Static holes for pressure measurement may
be blocked by contaminants, sand and etc.
MPFMs vary in uncertainty estimates from manufacturer to manufacturer. The aim of this
review is not to compare the different MPFM models with each other but MPFMs
collectively with other well production monitoring techniques.
17
Device performance is commonly split into the operating ranges it can measure within its
stated specifications. Some devices can handle higher gas volume fractions or water cuts than
others and this is based on the technology used in the measurement.
Typical ranges of uncertainties for components measured by MPFMs are 2-10% for liquids
(lower gas volume fractions allow for lower uncertainties) and 4-10% for gases [16]. Again,
the quantity of gas present has a large impact on uncertainty achieved.
6.4 Mobile Well Testing
In contrast with permanent flow-line well testing installations, which facilitate continuous
production measurements, mobile well testing (MWT) is a discrete testing approach. Mobile,
also known as portable, well testing units are fully portable production facilities and are
capable of both well test and clean-up operations. Typically, for offshore testing, MWTs
consist of partial separation units with two or three phase meters on the outlets. They take a
significantly less footprint than full separation systems and can be transferred between
installations after a well testing campaign is complete. Other models are smaller again, acting
like MPFMs with some capable of non-intrusive operation.
Table 3 briefly generalises the common benefits and drawbacks of MWTs, as most MWTs
are MPFMs which have been detailed in the last Section of this review.
TABLE 3
Advantages and Disadvantages of MWT Technologies
Technology Advantages Disadvantages

Separator, Portable / mobile Process fluids may need conditioning
MPFMs for optimum separation
Accessible to harsh arctic Separation may be inefficient due to
operating conditions liquid slugging or foaming
Complete package of well testing Complication and risk due to
and clean-up solutions operations and logistical needs of
heavy and bulky equipment to the well
site
Data to desk facilitated Size, cost and time constraints due to
limited numbers of well tests to be
performed at one time.
- Flow data is dependent on well test
frequency
- Impractical operations for running
multiple test packages throughout the
field.
- Intrusive
Radioactive - Risk of H&S due to exposure to
MPFMs gamma rays.
Clamp-on Small-footprint to reduce the Working accuracy is within  10%,
Sonar flow amount of equipment or which is less accurate as opposed to
meter disruption on process operations MPFMs.
Achieve periodic sampling of -
18
fluids at line conditions

Non-intrusive -
Radioactive- Eliminate risk of H&S -
free MPFMs
The uncertainties of multiphase flow measurements by means of test separators have been
described in Section 4 of this review, while the major uncertainties of MPFMs are
generalised in Section 6.3.
6.5 Impact on Maximising Economic Recovery
These alternative methods of well testing can provide very real substitutes to test separator
systems in one way or another. Each has their advantages and disadvantages as discussed in
the preceding section. As an impact to MER, there are varying levels of influence from the
alternative methods that must be discussed.
The main constraints on test separator systems are over use and intermittent data being used
between well test periods i.e. lack of real-time data. It is these two issues where these
alternative methods can provide significant advantages over test separator systems.
VFMs cannot replace measurement systems entirely. At best they can be used to reduce the
burden on other measurement systems by extending times between well tests. This is
accomplished through real-time monitoring of production rates from the software. Once the
software has been calibrated it can determine flows that can be used to optimise production.
However, the software can drift from the ideal response and needs to verified or even
recalibrated periodically. The frequency of this is likely to be lower than what would be
required acquiring well data from a periodic well test alone. This enables better economic
recovery through continual monitoring of flow rates and a faster response time.
Multiphase flowmeters can vastly impact on MER through continual measurement and faster
response times. When installed either subsea or topside in a multi-well capacity there is very
little stabilisation time required in comparison to test separator systems. This means a larger
number of wells can be tested or longer well tests can be obtained. If operating in a single-
well capacity, then real-time and continual measurements can be given. In either regard a
multiphase meter can be a viable alternative to test separator systems if verification can be
established through another means; potentially another multiphase meter, mobile well testing
equipment or even using the test separator itself.
It is the continuous measurement criteria where multiphase meters can offer a significant
advantage to test separator systems. Real time data vastly improves the ability to control
wells. The uncertainties associated with multiphase meters are comparable with test separator
systems too. Although larger, any influence on recovery factor should be minimised through
the larger amount of data that will be provided through continuous measurement.
Incorporating MPFMs into individual wells can only improve well optimisation and impact
positively on MER. The cost of MPFMs can be weighed against the perceived advantage in
MER it provides. It is likely this will be a case by case basis.
19
7 DISCUSSION
The impact of flow measurement on MER will be significant on the UK economy and oil and
gas industry. Without accurate knowledge of the fluids produced from individual wells it will
be impossible to efficiently optimise and maximise the amount of hydrocarbons produced in
the UKCS, and hence maximise the economic contribution to the UK treasury. What type and
method of measurement is the best for MER is still not defined however the various methods
have been outlined in the preceding sections. Current industry practice shows the test
separator system as the system of choice for the large majority of installations.
From feedback from industry there appears to be significant restrictions in the capability of
test separator systems to accomplish MER from the point of view of overuse and equipment
quality. For example, pipe set ups can inhibit well testing through the requirement to shut all
producing wells down in order to test one well and there are many issues that can affect the
single phase measurement methods. It is also clear that a periodic review of a wells
production is not enough to ensure the production is kept optimized throughout its lifetime.
For instance, for a well being tested every 30 days, there can be a sudden water breakthrough
after 15 days that would not be found for another 15 days. This may have a detrimental effect
on the whole reservoir. But how much of an effect does mis-measurement have on the UK
economy?
According to the 2013 FT, the UK’s oil and gas industry made a substantial contribution to
the British economy, accounting for some 450,000 jobs and £5bn tax revenues. The UK has
one of the largest budget deficits among European countries at 5.8% of GDP [17]. As such,
the accurate field databases derived from real time well tests would certainly have an
influence on the forecast of UK’s economic prospects and reservoir life to allow the
processing, transport and export of the UK’s petroleum and investment in new key
infrastructure. It helps avoid the premature decommission of assets to the detriment of
production hubs and infrastructure critically needed for maximising economic recovery from
UK’s valuable production assets and for achieving the maximum economic extension of field
life.
Based on the OBR’s most recent long term forecasts, the oil and gas production from 2014 to
2044 inclusive is expected to be 9.1 billion barrels of oil equivalent (boe) in a central case,
7.7 billion boe in a low case and 10.7 billion boe in the high case, Figure 7 [18].
Figure 7: OBR Scenarios for Future Production, 2014 Onwards
20
As can be seen, there would be a significant different of 3 billion boe between a low
production and high production future, all of these projections would fall short of maximising
economic recovery. Nonetheless, in his final report, Sir Ian Wood indicated that the recovery
of 15 to 16.5 billion boe is a realistic ambition [1].
In order to achieve the recovery of 15 to 16.5 billion boe, real time well tests based on
MPFMs would likely be the best option, as real time field data provides vital information to
facilitate exploration success and to optimise oil and gas production, petroleum revenue tax
(Figure 8) and financial performance in terms of net present value, internal rate of return,
profitability index, saving index and so on [19].
Figure 8: Government Revenues from UK Oil and Gas Production [20]
In its 2014 budget, the UK government announced a review of the fiscal regime to ensure that
it supports MER UK [21]. The fiscal regime which specifically targets (i) the maximisation
of economic recovery from oil and gas production and (ii) the profitability and cost
effectiveness of the industry would certainty needs accurate real time field data for striking a
balance between providing sufficient incentive for companies to operate in the UK, whilst
ensuring the nation gets a fair share of the proceeds. If the UKCS field databases were
discrete and uncertain, this would undoubtedly disadvantage the government's effort to
simplify the fiscal regime because of the incomplete picture which imposes additional
uncertainties. Ultimately, it could weaken the North Sea oil and gas investment and
production (Figure 9), as well as the uptake of improved and enhanced oil recovery
techniques and technologies as a whole.
Figure 9: North Sea Oil Production [22]

21
The US Energy Information Agency (EIA) commented that UK North Sea oil production
would decrease again in 2016 to just 500,000 barrels per day. In contrast, the sharp decline in
Norwegian North Sea oil production has stopped and production rates, while down from
previous highs, has been broadly stable since 2012 because of effective government policies
intended to support the industry [23]. With the recommendations from the Wood Review
being actively pursued by the UK, the value of 500,000 barrels per day can hopefully be
increased. Reservoir optimisation through better measurement methods is one way of helping
to achieve this goal.
Though many hydrocarbon reservoirs are profitably produced, very few are depleted
efficiently and economically. To optimise reservoir productivity and performance and to
predict recovery trends, the time series mapping of depositional environmental, flow barriers,
flow rates and core data are required. In particular, well tests that provide the vital time series
data [24] for reservoir management and economic recovery can:
 provide a better description of reservoir contents via flow quantification

 reduce investment and recovery uncertainty by evaluating production history or
historical matching
 establish a basis for total management and dynamic development for optimising
reservoir operation in all phases of depletion.
The real time flow data gathered from the field and well tests can provide actual evidences as
to whether or not an irregular fluid displacement is being hampered by “pockets” of flow
barriers, e.g. rocks. The reservoir reaction team can then rapidly take appropriate action(s) to
maximise volumetric flooding efficiency, and hence optimise the oil recovery factor.
In reality, every practical approach and decision making are essentially based on a set of real
time and accurate & representative data. In the nut shell, to economically achieve best
possible hydrocarbon recovery from reservoirs, real time field and well test data for decision
making are undoubtedly indispensable.
Figure 10 reviews the variation of flow patterns that can be captured by three alternative well
testing methods during reservoir depletion.
Figure 10: Flow Rates Plots of Well Testing Data
Table 4 summarises the three alternative methods available for well testing based on their real
time capability.
22
TABLE 4
Real Time Continuity
Alternative Method Real Time Comments

Capability
Virtual Flow Metering No Pseudo real time series, flow rates are
simulated by computer software.
Multiphase Flow Yes Genuine real time series, flow rates are
Meters physically measured and are not inferred by
computer software.
Mobile Well Testing No Discrete patterns, flow rates are measured on
a "pay-as-you-go" basis.
It should be noted from previous Sections that virtual flow metering is merely computer
simulations based on certain assumptions and boundary conditions, which are measured by
temperature and pressure sensors in and around the wells. The sensors once installed in and
around subsea wells would be hardly re-calibrated and verified. In contrast, mobile well tests
are carried out on discrete bases. Field data captured by virtual flow metering and mobile
well testing are lacking real-time continuity. To economically maximise hydrocarbon
recovery, each flow parameter captured must be verified against its consistency in real time
continuity. As such, the usual “garbage in, garbage out” approach can be avoided.
MPFMs allow both real time and continual flow measurements and have been subjected to
extensive and robust verifications [25]. Technically speaking, MPFMs would probably be the
best alternative option for well tests because MPFMs are capable of capturing real-time
continuous and accurate flow data which are essential for optimising the economic recovery
of deposits from hydrocarbon reservoirs. Nevertheless, virtual flow metering and mobile well
testing are also considered to be cost effective alternative options for complementing MPFMs
measurements.
In terms of errors in measurement or measurement uncertainty it was found that there is a

very weak link to recovery factor or MER. Instead, the biggest contributor to the recovery
factor from measurement is when the well test data is used to select a reservoir model. This
means that the estimated uncertainties found in this work seem to be acceptable in terms of
measurement requirements, as long as the reservoir model is correct. The mechanism for
selecting the model was not assessed in this work but it is recommended to complete further
work in this area.
The errors introduced from periodic testing of a well can clearly be seen in Figure 10 where
the dramatic changes in production profile is monitored to varying degrees by the different
alternative methods. Clearly, a continuous measurement system offers significant advantages
to periodic checking.
This work has shown that the current uncertainties in well testing, typically through test
separator systems, are fit for purpose in terms of measurement uncertainty they deliver for
their current use for operators. However, in terms of MER UK they lack the ability to provide
data in real time to fully optimise production. In addition, the restrictions of test separators in
terms of overuse will only continue and worsen.
23
It is clear that some kind of continuous measurement system that can offer trending of the
data would be best. This will enable better response and predictability of reservoir models
which are vital in optimising the production of hydrocarbons. This could include multiphase
flow meters, virtual flowmetering or downhole measurements. Practically, multiphase flow
meters used in combination with virtual metering is probably the most viable option at
present due to availability, reliability and current technology level for downhole measurement
techniques.
Test separator systems should not be eliminated in an ideal scenario. They would still be used
for the validation of the continuous measurement techniques. In this situation the stresses of
overuse would not be present and the process could be planned and equipment maintained
more regularly.
There is clearly much work that has to be completed in the area of measurement in order to
maximise economic recovery in the UKCS. Guidance will likely be provided in the latest
edition of the OGA Measurement Guidelines when they are made available. However, in the
meantime there are important conclusions and recommendations that can be implemented to
ensure the UK is optimising the revenue from its hydrocarbon reserves.
8 CONCLUSIONS
From the research conducted within the scope of work of this project the following
conclusions can be drawn from current well testing procedures used to acquire flow rate
information:
 In order to MER within the UKCS successful reservoir optimisation will be key. This can
only be achieved through knowledge of the production rates of any well and accurate
models of the reservoir performance
 The current practice of well testing for this information using a test separator system can
result in large uncertainties and potential bias. Further work is required to fully appreciate
the extent of test separator measurement uncertainties i.e. audits and additional data
 Recovery factor is weakly affected by flow measurement errors directly. Only when flow
measurement data is used to select reservoir models does the effect become significant.
Potentially an additional 5% in recovery factor (absolute) can be achieved
 The nature of well production is dynamic, periodic assessment (even every 30 days) is not
acceptable for accurate monitoring of well flows. Changes in water cut may not be picked
up till the next well test resulting in inefficiencies in production i.e. wells not optimised
 For surface or subsea flow measurements, multiphase meters offer the most robust and
accurate continuous monitoring method. The use of multiphase meters on each well will
enable faster responses to changes conditions within a well
 Well bore storage issues affect all surface flow rate measurements to some degree
resulting in dampened measurement results that can induce inaccuracies
24
 Downhole flow rate measurements are the most valuable sources of information for MER
as they provide real-time, continuous, and undampened reservoir responses. This provides
the most accurate and useful data for reservoir engineers in production optimisation
The above conclusions state the findings from the work from the scope of work items.
However, it is important to view these in terms of how they influence MER in the UKCS in
terms of flow measurement:
 The current levels of flow measurement uncertainty found in industry are acceptable for
their impact on economic recovery as long as reservoir models are correct
 A periodic test of production rates is not acceptable to ensure MER. Continuous

measurements would be the preferred option with periodic verification of the continuous
measurement
9 RECOMMENDATIONS
The following recommendations are made to provide industry with flow measurements that
do not hinder MER in the UKCS:
 The use of multiphase meters on wells where it is economically viable to install one
(subject to well lifetime and production rates)
 Development of historical trending or history matching methods for well test data to
ensure full use of the considerable data resources available
 Investigation into the current applicability of downhole flow rate measurement

technologies and their reliability
 The current assessment of well test uncertainties be expanded to the point where it is
statistically representative. This will include sourcing additional production and
process data as well as audit reports
 More stringent reporting requirements of well production data to the regulator
10 ACKNOWLEDGMENTS
The authors would like to thank the following for their input to the study:
 National Measurement and Regulation Office for providing funding for the project
 Oil and Gas Authority for their guidance and direction
 Operators in the North Sea for providing data to complete the uncertainty analysis
 Coventry University for assessing the significance of measurement error on recovery
factor and hence MER
25
11 REFERENCES
[1] Sir Ian Wood, UKCS Maximising Recovery Review: Final report, Feb 2014
[2] http://services.parliament.uk/bills/2015-16/energy.html
[3] OGA, Guidance Notes for Petroleum Measurement – Issue 9.2, Oct 2015
[4] https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/458593/A5_O
GA_Overview.pdf
[5] Coventry University, Well Test Report, March 2015
[6] http://www.emersonprocessxperts.com/2014/01/importance-of-flow-measurement-for-
separators/
[7] Ross & Stobie, Well Testing – An evaluation of test separators and multiphase meters,
NSFMW 2010
[8] NEL Questionnaire issued to North Sea Operators, Jan 2015
[9] Basic concepts in well testing for reservoir description. Heriot Watt University -
http://www.afes.org.uk/uploads/files/1.%20Well%20Testing%20Res%20Des%20Concepts.pdf
[10] Ingvil, Uncertainty in multiphase flow estimates for a field development case. Norwegian
University of Science and Technology, MSc thesis, June 2013.
[11] E Toskey, RPSE An evaluation of flow modeling. The Letton-Hall Group, Final Report
07121-1301-Task4.Final, Sept 20, 2011
[12] Bekt R., & et al, 2011. Comparison of commercial multiphase flow simulations with
experimental and field databases. BHR Group’s Multiphase Production Technology
Conference in Cannes 2011
[13] Tekna. Handbook of Multiphase Flow Metering, Revision 2, March 2005. Norwegian society
for oil and gas measurement. Norsk Forening for OLJE Gassmaling
[14] Gaviria F & et al, 2010. First Ever Complete Evaluation of a Multiphase Flow Meter in
SAGD and Demonstration of the Performance Against Conventional Equipment. 28 th
International North Sea Flow Measurement Workshop 26th- 29th October 2010.
[15] Bjrlo J., 2013. Uncertainty in multiphase flow estimates for a field development case. Master
of Science in Engineering and ICT. Norwegian University of Science and Technology,
Department of Energy and Process Engineering.
[16] Mulitphase manufacturer data sheet
[17] Trading Economics - http://www.tradingeconomics.com/united-kingdom/indicators
[18] HM Treasury. Driving investment: a plan to reform the oil and gas fiscal regime, December
2014
[19] Manaf N.A.A., & et al, 2014. Effects of fiscal regime changes on investment climate of
Malaysia’s marginal oil fields: Proposal model. International Conference on Accounting
Studies 2014. ICAS 2014, 18-19 August 2014, Kuala Lumpur, Malaysia
[20] HM Revenue & Customs, 2014. Statistics of Government Revenues from UK oil and gas
production
[21] Gov.UK. Review of the oil and gas fiscal regime: a call for evidence –
www.gov.uk/government/consultation s/review-of-the-oil-and-gas-fiscal-regime
[22] Will The UK Rescue North Sea Oil? By Investopedia, March 12, 2015 -
http://www.investopedia.com/articles/investing/031215/will-uk-rescue-north-sea-oil.asp
[23] Am K., and Helberg S., 2014. Public-private partnership for improved hydrocarbon recovery -
Lesson from Norway's major development programs. Energy Strategy Review 3 (2014) 30-48
[24] Smith R.G & et al, 1998. The road ahead to real-time oil and gas management. Institution of
Chemical Engineers. Trans IChemE. Vol 76, Part A July 1998.
[25] NEL, Multiflow I, II and III JIPs, 2001 - 2011
26
20-23 October 2015
Tie Backs and Partner Allocation

A Model Based System for meter verification and monitoring
Kjartan Bryne Berg, Lundin Norway AS, Håvard Ausen, Steinar Gregersen, Asbjørn
Bakken, Knut Vannes, Skule E. Smørgrav, FMC Technologies Inc, Norway
Abstract/Introduction
This paper is addressing the operation and management of flow instrumentation in complex
subsea and topside infrastructures. Special attention is given to situations where measurement
inaccuracies may influence the allocation process over time and cause significant imbalance in
the monetary flow, and eventually lead to legal disputes between partners over the allocated
volumes.
This paper will offer a description of how a model based system can be used to overcome the
challenges of describing measurement accuracy, need for reallocation, when/what to re-calibrate
as well as a total overview of the flow instrumentation status– utilizing information from every
instrument and device from subsea and downhole through multiphase meters, separators all the
way through the custody transfer metering system.
A Condition based Performance Monitoring system for a complete overview of the

instrumentation system including instrument status and accuracy indication is also described.
By utilizing flow modelling techniques, fast fault detection can be achieved to minimize the need
for re-allocation. In a potential fault situation, an online model will provide backup calculations
based on the existing instrumentation and include the custody transfer meters as the validator and
total system watchdog.
In cases where re-allocation is required, the model can be used to provide model based re-
calculation.
By tracking compositions from the wells to the fiscal metering point and comparing trends a well
test advisory is given.
20-23 October 2015
1 Challenge
In many fields the flow instrumentation is used for both managing the production process and
allocation. If a biased instrument is allowed to influence the allocation process over time this can
cause significant imbalance in the monetary flow, and could eventually lead to legal disputes
between partners. Early detection allows the operator to take appropriate action and moves the
response from unplanned events to planned operations.
Quality control of the upstream flow instrumentation system is normally done by verification
using topside test facilities. Dedicated well meters may not be verified at all due to failure to
produce a representative rate to test facilities. Therefore meter verification might only be possible
during rare and costly test campaigns. Depending on the subsea architecture, such as in Figure 1,
the tests themselves could also be very uncertain. Testing wells can also defer production. In
contrast, pipeline flow meters can be verified at normal production rates. The difficulty of
verification can lead to some meters operating with biases for long periods of time.
The challenge becomes how to verify and maintain the flow metering equipment while at the
same time minimising production loss due to testing. A good metering system for the allocation
process will also be a good basis for production optimization.
Figure 1 Complex subsea field infrastructure can make meter verification very difficult and costly
2 Model based system

A real-time model based system analyzes information from all available sources such as the
process control system and production databases and present them in a unified way. Combining
multiphase and fluid modelling with statistical analysis and mathematical optimization creates the
foundation for condition based monitoring of the upstream process as well as the topside process
facilities that can provide early error detection and bridge the gap between metering, allocation
and well performance.
A model based system utilizes all available sensors such as temperature, pressure, pressure drop
over choke and Multiphase meter (MPFM) primary instrumentation and calculated rates
combined with the multiphase models in calculating the actual flow. By including all available
information in the model based analysis redundancies can be exploited to enable a more fault
tolerant system.
20-23 October 2015
This paper describes an online model based system based on WATCH from FMC Technologies
comprised of three different parts: Upstream module, Overall mass balance module and Well
Test support module. WATCH is a common software platform for Metering, Production
Performance, Flow Assurance and Asset Integrity systems. WATCH is used in more than 30
online field applications modelling more than 700 wells and have been installed encompassing
both virtual flow metering systems (VFMS), Overall Flow Metering Systems (OFMS) and flow
assurance advisory systems (FAS) using state of the art steady state and transient multiphase flow
modes. Since 2011 the technology was expanded to include “Condition and Performance
Monitoring” (CPM) of subsea equipment. Figure 2 shows a screen shot from an online VFMS
application.
Figure 2 A screenshot from a model based monitoring system containing a single well, flowline and topside separation
process
The Upstream Module includes the wells and subsea production network. The overall mass
balance module uses the output from the Upstream Module, topside instrumentation and custody
transfer metering system to calculate accurate status indications for all meters. The Well Test
support module can flag wells as candidates for well tests. It can also help with the test itself by
providing real time estimates of stabilization time and source tracking during a test, and, through
the an integrated PVT package, act as a conversion tool between topside and subsea conditions,
mass and volumetric rates.
Through these modules, early error detection, status indication for all meters, well test advice and
real time support during well testing can be achieved. Figure 3 shows how the different modules
interact. Each module can work alone, or together with any other.
20-23 October 2015
Figure 3 How the different modules can interact, each module can also work independently
2.1 Early fault finding

A model based system is able to compare sensors through mass, energy and momentum balances
and in this way uses all available instrumentation to validate meters. Analysis is then performed
using built-in statistical tools for error detection and mathematical optimization to extract
information from the system, which is presented as technical conditions.
Early indication of meter failure reduces the need for re-calculation and allows the operator to
include meters in scheduled maintenance tasks ahead of time and avoid unplanned maintenance
and thereby reducing cost. An illustration of this is given in Figure 4.
Figure 4 Detecting errors early allows minimizes the impact from errors
2.1.1 Technical Conditions

The technical condition of the meters is quantified by technical condition indexes (TCIs). The
TCIs are normalized values that are aggregated upwards in an established hierarchy to the desired
level, usually following the structure of the field layout or different use cases.
A technical condition is compounded of many parameters for example the availability of the tag,
if the value communicated from the database within set parameters; frozen, max rate of change,
min/max limits and comparing the meter with redundant or neighboring meters etc.
These are hard failures. The second contribution comes from the model based system, where the
deviation between a calculated value and measured value is compared relative to the sensor and
20-23 October 2015
model uncertainty. The statistical analysis performs Gross Error detection on the system and
reports the Gross Error as an alarm and as a TCI indication as seen in Figure 5
Figure 5 An example of how a TCI tree can be built up with according to specific users need, for Metering the application
has detected a Gross Error in the 1st Stage Separator Liquid Meter.
The TCI's are organized in a tree structure so that each discipline receives relevant information
for their use cases. Examples of different users could be:
 Allocation and metering
 Petroleum Technology
 Operations
 Maintenance
 Commercial reporting
This is illustrated in Figure 6.
Figure 6 The TCI's are aggregated information reflecting the status of all layers under it and customized for each
discipline
The TCI's are linked to a notification system which can be set up to send push notices as emails
or other services such as SMS. This allows the user to get a quick overview and can locate errors
quickly before they can create large negative impacts.
20-23 October 2015
3 Overall Mass Balance

The overall mass balance is used to validate the well rates and topside meters against the fiscal
metering stations.
WATCH includes process models for separation and merging of streams and integrates a PVT
package which enables compositional tracking and accurate multistage flashing. This means that
the mass balance is also a full compositional balance.
The mass balance is built as a high level robust model to represent the field layout with wells,
templates, flow lines, gas lift and chemical injection lines.
An example of the model layout is given in Figure 7. Both virtual information and physical
meters can be included together with PVT information. The input and output for the Overall
Mass Balance is shown in Figure 8.
Figure 7 The mass balance follows the layout of the field which allows for an intuitive approach
3.1 Sensor uncertainty

Each measurement included in the mass balance has a specified uncertainty corresponding to the
model and sensor uncertainty.
The initial uncertainty is taken from the data sheets for each meter; the result from the overall
mass balance is used to adjust the assumed uncertainty and more accurately reflect the plant
conditions.
The uncertainty specified for each meter increase the awareness of the operator on how each
meter in the allocation system is currently performing.
20-23 October 2015
Figure 8 The Overall Mass Balance validates the input measured values and uncertainties and performs error detection.
Edvard Grieg is an oil field on the Norwegian continental shelf operated by Lundin Norway AS
which will implement an Overall Mass Balance. The Overall Flow Metering System (OFMS)
will be implemented for the process topside, in an effort to qualify this technology and bridge the
gap from wells to fiscal metering station. The OFMS will also be used to perform monitoring and
early error detection for the topside metering system as well as the wells. In addition Edvard
Grieg will have traditional well instrumentation plus a dedicated MPFM for each well. All wells
can be routed individually directly to a test separator for testing.
4 Upstream validation
The upstream metering validation is normally compromised of multiphase models covering the
wells from inflow all the way to the receiving facilities topside or onshore. The model can also
include gas and water injection networks.
In WATCH the wells are modelled using inflow correlations, steady state flow models for the
wellbore and a detailed choke model. All the models are based on first principles for mass and
energy balance and can easily be adjusted to match field conditions.
The upstream metering validation can also be used to estimate rates or other direct or indirect
flow properties such as velocities, densities and phase distribution where no meters are available.
4.1 Brynhild
A WATCH FAS (Flow Advisory System) is installed on Brynhild, a field operated by Lundin
Norway AS. The FAS implements several of the principles mentioned in this paper. Brynhild has
two/three production wells producing into a common MPFM, the FAS is used to validate the
MPFM and monitor the individual well performance.
By having the online and offline capabilities of the FAS the operators on Brynhild can quickly
run simulations where they can test hypothesis and compare them with historical measured data
from the field, this provides a deeper understanding of how the wells and field performs, and
allow for more optimal operations.
20-23 October 2015
4.1.1 Slugging
Well slugging was experienced on Brynhild. The FAS was used to identify that the slugging only
occurred outside certain velocity constraints; a virtual velocity measurement was added in the
FAS and could be directly applied for well control without any installation costs or
commissioning period.
4.1.2 Water Cut increase

The MPFM on Brynhild is used as the primary tool for allocation and the FAS works as a
verification and backup system. When the MPFM suddenly showed a 17% increase in WC the
FAS was used to investigate whether the increase was due to an actual change in the production
process fluid content or a problem with the WC measurement in the MPFM. As Figure 9 shows
the total flow rate were slowly decreasing while the WC was increasing. The step increase in WC
however is only present in the MPFM and not visible in the calculated rate from the FAS. The
FAS rate is based on the model response for all pressures and temperatures. The pressure and
temperature responses do not show any indication of a step increase of WC and this is reflected
directly in the calculated WC. This indicates that the measured MPFM WC increase is not due to
a change in the well performance. The independent verification of the MPFM gives the operator
room for action, and avoids unnecessary tests or well shut downs.
Figure 9 A water cut increase is measured in the MPFM, the FAS shows no increase indicating a mismatch between the
pressure and temperature instrumentation and the MPFM. Values are normalized.
4.2 Out of range meter

Many operators' struggle with out of range meters, this can in some cases mean that the operator
needs to defer production since the actual production is not known and safe operations are
difficult.
In Figure 10 one such case is shown for a wet gas meter. The differential pressure sensor of the
venturi of the meter is saturated meaning the full production rate cannot be measured. The VFMS
calculated rate that normally would be used for backup and surveillance, is then used as the
primary measurement. During the initial ramp-up in Figure 10 the meter is not saturated and we
20-23 October 2015
can see a good match between the measured value in red and the calculated value in green. Once
the meter reaches saturation the calculated values can be used as a replacement avoiding deferred
production.
Figure 10 A saturated wet gas meter, the red line is the measured value and the green is the calculated value. The VFMS is
used as a replacement allowing the operator to continue producing at high rates.
4.3 Offline tool and re-calculations

If significant errors have been detected in the metering system, a re-calculation can be necessary.
Re-calculation is then performed on the impacted period, but since the error is known it can either
be eliminated or compensated for. The model based tool can read data from the historical
database; the error can be removed from the calculation, either by adjusting the model or
excluding parts of the input.
4.3.1 Redundancy and backup to MPFM

The upper plot in Figure 11 shows an MPFM during first oil for a well. During the initial period
the MPFM PVT was not configured, thus the flow rates were very inaccurate. In this period
WATCH was used as the primary rate measurement, allowing the operator to monitor the
production without the MPFM online. Once the MPFM PVT was configured, marked by a red
circle in Figure 11Error! Reference source not found., the WATCH system was adjusted to the
MPFM rate and the entire period could be re-calculated using the offline capabilities in WATCH
for more accurate production rates. The re-calculation is shown in red in the bottom plot of
Figure 11.
20-23 October 2015
Figure 11 The top figure shows the WATCH as an online tool where WATCH was used as the primary measurement. The
red circle shows when the MPFM came online. The bottom figure shows flow rates prior to and after calibration.
5 Well test support

The goal of the Well test module is to perform well tests only when required thus reducing testing
time, number of well tests and costs due to deferred production.
5.1 Advice
By monitoring trends and changes in the well response, it is possible for the application to offer
well test advice.
Deviations between the modelled response and the measured response indicate that a change has
happened and the well is a candidate for test. The raw data is presented in Figure 12, while it is
clear that some of the data stands out, without analysis it is difficult to see which producers
contributes to the error.
Statistical analysis of the deviation between calculated and measured values provides this insight
and correlates each well's contribution to the deviation. An example of this is given in Figure 13
where the deviation between the sum of subsea production and custody transfer system is
correlated with the well production, especially B-13 and B-12 are candidates for the next well test
campaign.
20-23 October 2015
Figure 12 Well monitoring comparing the well production with topside metering. Without analysis it is difficult to find the
contributors to each error.
Figure 13 Statistical analysis can be used to identify which wells are recommended for test
5.2 Source tracking

One challenge with running tests on a subsea field is the stabilization times which are often
conservatively estimated, leading to unnecessary test time.
By using transient models WATCH can calculate hold up and transportation time and use this to
automatically track each wells production into the test meter or separator to ensure that the
correct well fluid enters the testing facilities.
A graphical example is given in Figure 14. The y-axis shows the arriving fraction. The time axis
is divided into 4 different sections
1. Initialization of the well test
2. Pipeline flushing, the fluid currently in the pipe is being produced out.
3. Transition period, final stages of flushing
4. Data logging, this is the optimal time for testing.
A model based tool can in this way calculate a real time estimate of the actual state of the system
and stabilization time thereby optimizing test durations.
Tracking fluid residence time and stability enables the operator to minimize the period on test
without compromising data quality.
20-23 October 2015
Figure 14 Example of source tracking during a well test, 1: Verify test conditions. 2: Flow line flushing. 3: Transition
period. 4: Data logging.
5.3 Rate conversion and PVT calculations

One of the challenges when performing well testing is the calibration process. The measured rates
are usually performed at very different conditions, such as the difference between pressure and
temperature for the test separator and well head. It could be that the subsea instrumentation
accurately measures mass rates, while the separator measurements are volumetric.
Fully compositional calculations convert between flow rate at volumetric or mass rates and
fractions between any given set of conditions.
They convert actual rates and fractions at the well and compare them to the measurements made
topside; it is also possible to compare the performance at several specified conditions at the same
time. This saves the operators time and reduces the uncertainty in the calibration process.
6 Flexible installation and maintenance
The WATCH system is a software installation and has a very flexible installation package with
its own web server and web GUI. A web browser and an open connection to the server is all that
is required to access the system, together with the appropriate credentials. This makes it very well
suited for organizations where the daily operations are split between onshore and offshore and
allows easy collaboration. The Watch software can also connect to a multitude of data sources
such as the production control system or to the customer database through one of several
supported protocols. As seen in Figure 15.
20-23 October 2015
Figure 15 The model based system gathers data from all data sources available
7 Summary
The paper has described a real-time model based system that can analyze information from all
available sources and present them in a unified way in an online overview.
The overview is customized for each profession and allows the operator to get a quick status
overview of the field.
The automated error detection can notify users quickly in the event of sensor, meter or
communication failures in the system and depositions in the production system. This enables
systematic and efficient verification of meters and well production and allows operators to only
test when needed.
The model based system enables fault tolerant operations of the online system and maximizes the
value of the available information reducing costs for the operator and maintains metering quality
through field life.
Gross Error Detection:

Maximising the Use of Data with UBA on Global Producer III (Part 2)
Neil Corbett, Maersk

Robert Sibbald, Accord
Phillip Stockton, Accord
Allan Wilson, Accord
1 INTRODUCTION
Validation of input data and results is a key component of all allocation workflows
and applies to a variety of data: metered quantities, well estimates, component
fractions, plant operating conditions, and allocation results. Data validation is
addressed throughout the length of the allocation data chain by different stakeholders
and by a variety of means. Allocation computer systems are no exception: they
normally include validation packages as standard and typically alert users when data
or calculated results are, for example, missing, stuck, or lie outside an expected range.
Some of the test thresholds can seem arbitrary. This need not be the case; knowledge
of the uncertainties in an allocation system’s input data can be utilised in rigorous
statistical tests based on conservation laws and data reconciliation.
In a previous paper [1] we presented a non-linear uncertainty based allocation method

for Maersk’s GPIII FPSO serving the Dumbarton and Lochranza fields. In this paper
we demonstrate several gross error detection techniques for validating GPIII input
data and allocation results based on statistical tests adopted from linear data
reconciliation. The gross error tests are natural extensions to GPIII’s non-linear
uncertainty based allocation method.
Though the statistical tests are rigorously derived for the linear case we demonstrate
the ability to detect the location and size of gross errors in non-linear data
reconciliation through Monte Carlo simulation and investigate their performance
when applied to real field data.
Data reconciliation has been an active area of research since the mid-1960s [2] and
has been applied in the chemical, power and oil and gas industries to name a few.
However its use remains uncommon in the UKCS, and North Sea in general. Some
examples are published in [3],[4],[5]. Data reconciliation provides an optimal estimate
of process variables consistent with physical constraints, such as conservation of mass
and energy, but is reliant on the input process variables being subject to random errors
only. Any gross errors in the input variables will skew the results away from their true
optimal values.
The intent of this paper is to highlight some of the statistically rigorous gross error
detection (GED) tests which extend data reconciliation beyond simply providing
allocation results. We do this through studying a simple fictitious allocation system,
and Maersk’s GPIII Uncertainty Based Allocation system. In doing so, we hope to
show that GED techniques are not only a natural extension for allocation systems
utilising data reconciliation techniques, but that with a little extra information on
1
measurement uncertainties some of the tests can be applied to more traditional pro-
rata based allocation systems.
In Section 2, the concepts and tests used for gross error detection in the rest of the
paper are introduced. Section 3 describes the use of Monte Carlo techniques to
demonstrate a variety of GED tests when applied to a simple fictitious allocation
system. From here the focus turns to GPIII with Section 4 describing the subsea
configuration and topsides process, along with the uncertainty-based allocation
system presented in the previous paper [1]. Section 5 demonstrates the outcome of
applying GED tests to GPIII’s non-linear uncertainty-based allocation system. The
results of the gross error detection tests when applied to a selection of real GPIII
production data are presented in Section 6 and conclusions are provided in Section 7.
For the interested reader, Section 8 contains mathematical detail regarding the
uncertainty-based allocation calculations.
.
2
2 CONCEPTS
Before introducing and implementing gross error detection techniques it is worthwhile

explaining concepts used throughout this feasibility study.
2.1 Random Errors and Gross Errors
Random errors are ever present in oil and gas measurements, caused by small changes
in ambient conditions. These random errors result in imbalances in physical conserved
quantities such as mass and energy across processing facilities. Since the mid-1960s
data reconciliation techniques have been developed to deal with the random errors and
enable the best estimate of true measurement values to be obtained from actual
measurements subject to physical conservation laws, based on rigorous mathematical
and statistical techniques.
Gross errors can arise in oil and gas measurements if a meter is miscalibrated, poorly
maintained, damaged, operated outside its certified regime or fouled. In a steady-state
production environment a gross error will manifest itself as a sudden change or as a
drift over time in a measured value from the steady-state value. A great deal of effort
has been spent and continues to be spent implementing and developing standards and
tools to ensure measurement data quality. This varies from regular meter inspection,
auditing and calibration of metering packages, online diagnostic checks in flow
meters or additional diagnostic packages to data quality checks in allocation systems.
Yet all this effort is no guarantee of data quality. As was reported in [6] errors may
arise in the allocation workflow where there are many links between an initial data
source, be it meter or well production estimate and the allocation computer system.
Gross errors will skew allocation results away from the correct results, and if
undetected for a long period of time can result in misallocations worth millions of
Dollars, reputational damage, not to forget the time and effort lost to corrective work.
2.2 Confidence Levels and Gross Error Tests
For the purposes of this paper it is assumed that all measurements are normally
distributed about a true value, μ, with a standard deviation, σ. Based on the properties
of the normal distribution we could reject any new measurement more than 2σ
different from the mean with 95% confidence. Or put another way, at the 95%
Confidence Level the new measurement is not consistent with the mean.
In the tests that follow, rather than directly comparing measurements to a hypothetical
mean and standard deviation of a Gaussian distribution, calculations will be used to
derive a test statistic upon which gross errors in the measurements can be detected.
The test statistics all make use of the same idea of a confidence level, which will be
evaluated based on either a Gaussian or Chi-squared (χ2) distribution, depending on
the details of the test.
The choice of Confidence Level is arbitrary, but affects the outcome of Gross Error
tests. If the confidence level is set too low, gross errors may be detected when none
exist; the test has raised a false alarm and is said to have committed a Type 1 error. If
the confidence level is set too high, gross errors are not detected when one or more
3
exist; the test is said to have committed a Type 2 error. If we denote the Type 2 Error
rate of a test as, 𝜃, then the ‘Power’ of a statistical test is defined as,
𝑃𝑜𝑤𝑒𝑟 = 1 − 𝜃 [1]
When setting the confidence level at which to identify gross errors there is a trade-off
between the Type 1 Error rate (the probability of raising a false alarm) and Power (the
probability of correctly identifying gross errors when they exist). A confidence level
of 95% has been used throughout this paper.
Three different tests are used in this paper for identifying gross errors:
1- Allocation Factors;
2- Global Test;
3- Generalised Likelihood Ratio (GLR) Test.
Allocation Factors are routinely used in allocation systems as a data quality check.
Defined as the ratio of allocated to measured mass (after accounting for shrinkage) for
a field or well Allocation Factors are typically expected to be about 1. A margin of
uncertainty for Allocation Factors could be determined analytically or by Monte Carlo
means for any system, but in reality often seemingly arbitrary margins of 5, 10 or
20% are applied.
The Global Test and GLR Test are standard GED tests from data reconciliation
literature. Detailed definitions of the tests are given in sections 3 and 5 where they are
used. They are a natural extension to the data reconciliation approach used in the
GPIII allocation system. These are some of the simplest tests available for gross error
detection documented in [5] and references therein.
The Global test enables data to be tested for the presence of gross errors but does not
identify the erroneous measurement(s) without further effort. The GLR Test enables
the presence of gross errors to be detected and in theory the erroneous measurement to
be identified. It also has the additional benefit of enabling the size of a single gross
error to be estimated and therefore applied as a correction to the input data. It has
been shown theoretically that if there is at most a single gross error in a system then
the GLR test has the maximum power, compared to other tests not reviewed in this
paper (Constraint Test , Maximum Power Constraint Test, Measurement Test and
Maximum Power Measurement Test). Although not included in this study, the GLR
test can also be used to identify unknown leaks in a system.
GED tests have been derived for systems subject to linear constraints in the process
variables and operating in steady-state. For the GPIII system considered here the gas
mass constraints are nonlinear (actually bilinear as they contain a term which is a
product of two quantities to be reconciled). This analysis uses the gross error tests
under the assumption that the non-linearity effects will either be small or will be
identified through the simulation studies. The assumption of a steady-state operating
system is often not valid in reality and the effects of this will be seen in analysing the
4
actual GPIII data. The success, or otherwise, of the tests are examined under these
conditions.
5
3 SIMPLE LINEAR ALLOCATION SYSTEM EXAMPLE
A demonstration of the gross error tests used in this paper is provided using a simple
fictitious example system consisting of three fields, Alpha, Bravo and Charlie. The
produced hydrocarbons from the three fields are commingled and processed to
produce export gas and export oil. All streams are assumed to be metered. The
process, assumed feed and export stream masses and measurement uncertainties are
shown in Figure 1. The example is very contrived having no shrinkage and perfect
mass balance! Export Oil and Gas are allocated back to Alpha, Bravo and Charlie
pro-rata to their measured mass.
Figure 1 - The simple linear allocation system used to demonstrate GED tests
Feed Mass Relative Export Mass Relative
(tonnes) Uncertainty (tonnes) Uncertainty
Alpha Gas 600 10%

Oil 100 5% 1200 1.00% Gas
Bravo Gas
Oil
400
50
10%
5%
PROCESS
Charlie Gas 200 10% 300 1.00% Oil
Oil 150 5%
3.1 Test Definitions
For the GED tests used in this paper it is necessary to define the constraints applicable
to this process:
−𝑚𝐸𝑥𝑝𝑜𝑟𝑡,𝑂𝑖𝑙 + 𝑚𝐴𝑙𝑝ℎ𝑎,𝑂𝑖𝑙 + 𝑚𝐵𝑟𝑎𝑣𝑜,𝑂𝑖𝑙 + 𝑚𝐶ℎ𝑎𝑟𝑙𝑖𝑒,𝑂𝑖𝑙 = 0 [2]
−𝑚𝐸𝑥𝑝𝑜𝑟𝑡,𝐺𝑎𝑠 + 𝑚𝐴𝑙𝑝ℎ𝑎,𝐺𝑎𝑠 + 𝑚𝐵𝑟𝑎𝑣𝑜,𝐺𝑎𝑠 + 𝑚𝐶ℎ𝑎𝑟𝑙𝑖𝑒,𝐺𝑎𝑠 = 0 [3]
We define 𝑦 to be a vector of measured data such that

𝑚𝐸𝑥𝑝𝑜𝑟𝑡,𝑂𝑖𝑙 [4]
𝑚𝐸𝑥𝑝𝑜𝑟𝑡,𝐺𝑎𝑠
𝑚
𝑦 = 𝑚 𝐴𝑙𝑝ℎ𝑎,𝑂𝑖𝑙
𝐴𝑙𝑝ℎ𝑎,𝐺𝑎𝑠
⋮
[𝑚𝐶ℎ𝑎𝑟𝑙𝑖𝑒,𝐺𝑎𝑠 ]
and a Constraint matrix, 𝐴, which contains one row for each constraint applicable to
the process. In any row the entries in a column are +1 for any stream flowing into the
process, -1 for any stream exiting the process, and 0 for any stream which is not
involved in the constraint. The columns of 𝐴 are ordered identically to the rows in the
measured data vector, 𝑦.
Thus the Constraint matrix for the above process can be written as
−1 0 1 0 1 0 1 0 [5]
𝐴= [ ]
0 −1 0 1 0 1 0 1
6
The first row of A represents the oil mass balance and the second row the gas mass
balance across the system. The physical constraints applicable to the system can then
be written succinctly as,
𝐴𝑦 = 0 [6]
Uncertainties in each measurement are assumed to be independent of one another thus

their Covariance matrix, V, is diagonal, with each stream’s entry on the main diagonal
𝑚∙𝜀 2
calculated from its mass,𝑚, and relative uncertainty, 𝜀, as ( ) .
2
Allocation Factor
For this simple example where product is allocated pro-rata to each field’s production
the Oil Allocation Factor for oil will be
𝑚𝐸𝑥𝑝𝑜𝑟𝑡,𝑂𝑖𝑙 [7]
𝐴𝐹𝑂𝑖𝑙 =
𝑚𝐴𝑙𝑝ℎ𝑎,𝑂𝑖𝑙 + 𝑚𝐵𝑟𝑎𝑣𝑜,𝑂𝑖𝑙 + 𝑚𝐶ℎ𝑎𝑟𝑙𝑖𝑒,𝑂𝑖𝑙
The Gas Allocation Factor is calculated analogously. We will assume an Allocation

Factor in the range 0.9 < 𝐴𝐹 < 1.1 is acceptable for both oil and gas, thus 𝐴𝐹 < 0.9
or 𝐴𝐹 > 1.1 would signal the presence of a gross error.
Global Test
The Global test uses the test statistic calculated from the matrices defined above
𝛾 = 𝐴𝑦(𝐴𝑉𝐴𝑇 )−1 (𝐴𝑦)𝑇 = 𝑟 𝑇 Σ −1 𝑟 [8]
where the constraint residual vector 𝑟 = (𝐴𝑦)𝑇 represents the amount of violation of
each constraint and the Covariance matrix Σ = 𝐴𝑉𝐴𝑇 represents the uncertainty in
each constraint.
Under the null hypothesis (that the data are free of gross errors) the above statistic
follows a χ2 distribution with 2 degrees of freedom (one per constraint). A gross error
is detected if γ ≥ χ2 (α) , for the chosen confidence level, α. At a 95% confidence
level, a gross error is detected if γ ≥ 5.99.
Generalised Likelihood Ratio Test
The GLR Test is calculated from the constraint residual vector 𝑟 = (𝐴𝑦)𝑇 as used in
the Global Tests. The test establishes which is more likely; the null hypothesis that
there is no gross error in the input data, or an alternative hypothesis, that there is one
or more gross errors in the input data.
For each measurement the test statistic is calculated as:
7
𝑑𝑘2 [9]
𝑇𝑘 =
𝐶𝑘
where
𝑑𝑘 = 𝑓𝑘 𝑇 (𝑨𝑽𝐴𝑇 )−1 𝑟 = 𝑓𝑘 𝑇 Σ −1 𝑟 [10]
𝐶𝑘 = 𝑓𝑘 𝑇 (𝐴𝑉𝐴𝑇 )−1 𝑓𝑘 = 𝑓𝑘 𝑇 Σ −1 𝑓𝑘 [11]
and each vector 𝑓𝑘 is the column of the Constraint matrix corresponding to the
measurement.
Under the null hypothesis the above statistic follows a χ2 distribution with 1 degree of
freedom. As explained in [8], to reduce the probability of a false-alarm (Type 1 error)
due to multiple applications of a univariate test, a modified confidence level is used
and a gross error is detected if any of the test statistics exceed the test criterion χ2 (β)
where
𝛽 = 1 − (1 − 𝛼 )1⁄ 𝑚 [12]
𝛼 is the chosen confidence level and 𝑚 is the number of measurements. This reduced
confidence level ensures that the Type I error rate is at most 𝛼.
For the purposes of this study, gross errors were detected at a 95% confidence level.
This equates to setting 𝛼 =0.05 and with 8 measurements, 𝛽 = 0.006. A gross error is
then detected if any of the measurement test statistics, χ2 (𝛽) ≥ 7.44.
For the measurement with the highest test statistic above the threshold, the GLR test
allows the error on that measurement to be estimated as
𝑑𝑘 [13]
𝑏𝑘 =
𝐶𝑘
3.2 Monte Carlo Simulation
A Monte Carlo (MC) simulation approach has been taken to demonstrate the expected
outcomes of each of the gross error tests. Two separate simulation runs each with
100,000 trials were generated.
In the first simulation run production and export stream masses were generated from
normal distributions with mean and standard deviation given by the masses and
uncertainties in Figure 1 without gross errors. From this simulation each test’s
Average Type 1 Error rate (AVTI) can be established as,
8
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑎𝑙𝑠𝑒 𝑎𝑙𝑎𝑟𝑚𝑠 [14]

𝐴𝑉𝑇𝐼 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑖𝑎𝑙𝑠
In the second simulation run the sign and magnitude of a gross error of between 5 and
10 standard deviations were generated at random from a uniform distribution then
applied to one randomly selected stream in each trial. From this simulation each test’s
Power can be calculated as,
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑔𝑟𝑜𝑠𝑠 𝑒𝑟𝑟𝑜𝑟𝑠 𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑙𝑦 𝑖𝑑𝑒𝑛𝑡𝑖𝑓𝑖𝑒𝑑 [15]

𝑃𝑜𝑤𝑒𝑟 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑔𝑟𝑜𝑠𝑠 𝑒𝑟𝑟𝑜𝑟𝑠 𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑
In this analysis the number of gross errors simulated is the same as the number of
trials.
Allocation Factors
Figure 2 shows the probability density 1distribution of the Oil and Gas Allocation
Factors from the simulation with gross errors. (For simplicity in the rest of the paper
we will refer to probabilities or probability distributions rather than probability
density.)
Both distributions have large peaks about the value 1. These peaks are due to the
random variation in the Oil Allocation Factor for trials where the gross error was
simulated in a gas measurement and vice-versa.
The chosen acceptable range 0.9 < 𝐴𝐹 < 1.1 is delimited by the vertical red lines.
The Oil Allocation Factor has a narrower distribution than the Gas Allocation Factor
reflecting the lower uncertainties assumed on the oil streams and suggesting it is less
sensitive to gross errors of the magnitude simulated.
1
Probability density is defined as the fraction of trials with an Allocation Factor in a
bin of the histogram relative to the total number of trials, divided by the width of the
bin. Thus the total area depicted in the figures is 1.
9
Figure 2 - Distribution of Oil and Gas Allocation Factor from MC simulation

with gross errors.
6
10
Probability Density
Probability Density
4
0 0
0.50 0.75 1.00 1.25 1.50 0.50 0.75 1.00 1.25 1.50
Oil Allocation Factor Gas Allocation Factor
Figure 3 shows the same Oil and Gas Allocation Factor distributions, according to
gross error location. Oil Allocation Factors are shown only for trials where the gross
error is located in the associated oil stream, and likewise for the Gas Allocation
Factor. The distributions therefore do not contain the trials with an Oil or Gas
Allocation Factor about 1, which form the central peaks in Figure 2. The results of
simulations with positive and negative valued gross errors are shown on the same
plots; this leads to the distributions having two peaks. Allocation Factors less than 1
arise when the gross error is an over-reading, and these errors give a narrower peak
when compared to gross under-readings
Figure 3 - Distribution of Oil and Gas Allocation Factors from MC simulation

with gross errors according to location.
Export Oil Export Gas
6
9
6 4
3 2
0 0
Alpha Oil Alpha Gas
6
9
6 4
Probability Density
Probability Density
3 2
0 0
Bravo Oil Bravo Gas
6
9
6 4
3 2
0 0
Charlie Oil Charlie Gas
6
9
6 4
3 2
0 0
0.50 0.75 1.00 1.25 1.50 0.50 0.75 1.00 1.25 1.50
Oil Allocation Factor Gas Allocation Factor
Figure 3 demonstrates the Oil Allocation Factor is fairly insensitive to gross errors.
Except for Charlie’s Oil Allocation Factor, the results suggest tests based on the Oil
Allocation Factor would be insensitive to gross errors of the size simulated, and
choice of acceptable range. Figure 3 also demonstrates the differing sensitivity of the
Gas Allocation Factor according to location of the gross error. The Gas Allocation
10
Factor is most sensitive to a gross error on Alpha, followed by Bravo then Charlie.
There is no or little sensitivity to gross errors in the export streams. The order reflects
the stream flow rates and measurement uncertainties assumed in this example.
Table 1
quantifies the performance of the Oil and Gas Allocation Factor test
demonstrated in Figure 3 according to the test’s AVTI and Power. The first two
columns report the performance irrespective of the gross error location; subsequent
columns indicate the simulated gross error’s location
Table 1 – Allocation Factor test performance statistics.

Allocation Factor Oil Gas Oil Gas Oil Gas Oil Gas Oil Gas
Error Location Overall Overall Export Export Alpha Oil Alpha Bravo Oil Bravo Charlie Charlie
Oil Gas Gas Gas Oil Gas
Type I Error 0.000 0.002 0.000 0.002 0.000 0.002 0.000 0.001 0.000 0.002
Power 0.055 0.233 0.000 0.029 0.051 0.964 0.000 0.717 0.397 0.132
AVTI is very low for all streams irrespective of gross error location. The Allocation
Factor test is exceedingly unlikely to generate a false alarm. However the test’s Power
reveals a different tale; its ability to identify a gross error is strongly subject to the
location, and for the most part, poor.
11
Global Test
The Global Test statistic probability distribution from the simulation with gross errors
is depicted in Figure 4 and according to gross error location in Figure 5. The red
vertical line denotes the critical test threshold, the 95% confidence level above which
a gross error is detected, γ ≥ 5.99
Figure 4 - Distribution of Global Test Statistic from MC simulation with gross
errors.
0.06
0.04
Probability Density
0.02
0.00
0 20 40 60
Global Test Statistic

errors according to location.
0.20
0.15
0.10
0.05
0.00
Alpha Oil Alpha Gas
0.20
0.15
0.10
Probability Density
0.05
0.00
Bravo Oil Bravo Gas
0.20
0.15
0.10
0.05
0.00
0.20
0.15
0.10
0.05
0.00
0 20 40 60 0 20 40 60
12
Figure 4 shows the Global Test statistic distribution has a long tail extending far
above the critical test threshold, indicating a higher power than the Allocation Factor
test
Figure 5 shows that for certain measurements the Global Test statistic is very
sensitive to a gross error of the magnitude simulated.
Table 2 quantifies the performance of the Global Test demonstrated in Figure 5

according to the test’s AVTI and Power. The first column reports the performance
irrespective of the gross error location; subsequent columns indicate the simulated
gross error’s location
Table 2 - Global Test performance statistics.

Global Test
Error Location Overall Export Export Alpha Oil Alpha Bravo Oil Bravo Charlie Charlie
Type I Error 0.048 0.051 0.045 0.045 0.045 0.047 0.056 0.040 0.057
Power 0.662 0.521 0.183 0.893 0.995 0.391 0.907 0.993 0.408
AVTI is low for all streams irrespective of gross error location, of the order of 5%, so
the possibility of a false alarm is not negligible. Although the overall Power indicates
that about 2/3 gross errors would be detected, the Power is much higher than for the
Allocation Factor test. The Global Test Power is higher when compared to the
Allocation Factor test according to the gross error location, and rises above 90% for
some locations.
GLR Test
The GLR Test statistic’s probability distribution from the simulation with gross errors
is depicted in Figure 6 and according to gross error location in Figure 7. The red
vertical line denotes the critical test threshold, the 95% confidence level above which
a gross error is detected, γ ≥ 7.44
13
Figure 6 - Distribution of GLR Test Statistic from MC simulation with gross

errors.
0.12
0.09
Probability Density
0.06
0.03
0.00
0 20 40 60
GLR Test Statistic

errors according to location.
0.4
0.3
0.2
0.1
0.0
Alpha Oil Alpha Gas
0.4
0.3
0.2
Probability Density
0.1
0.0
Bravo Oil Bravo Gas
0.4
0.3
0.2
0.1
0.0
0.4
0.3
0.2
0.1
0.0
0 20 40 60 0 20 40 60
GLR Test Statistic
As with the Global Test, Figure 6 shows the GLR Test statistic distribution has a long
tail extending far above the critical test threshold, indicating a higher power than the
Allocation Factor test
Figure 7 shows that for certain measurements the GLR Test statistic is very sensitive
to a gross error of the magnitude simulated.
14
Table 3 quantifies the performance of the GLR Test demonstrated in Figure 6

according to the test’s AVTI and Power.
Table 3 - GLR Test performance statistics.

GLR Test
Error Location Overall Export Export Alpha Oil Alpha Bravo Oil Bravo Charlie Charlie
Type I Error 0.006 0.007 0.003 0.010 0.006 0.004 0.008 0.003 0.007
Power 0.559 0.340 0.072 0.806 0.985 0.224 0.827 0.982 0.233
AVTI is extremely low for all streams irrespective of gross error location, an order of
magnitude lower than for the Global Test, suggesting the possibility of a false alarm is
exceedingly unlikely. Although the overall Power indicates that just over half of gross
errors would be detected, this is partly due to the low Power for export streams. The
test’s Power is much higher than for the Allocation Factor test and is higher when
compared to the Allocation Factor test according to the gross error location, rising
above 90% for some locations.
The GLR Test also enables the magnitude of the gross error to be established. Figure
8 shows the distribution of calculated corrections, standardised relative to the
measurement uncertainty. The orange vertical lines delineate the range of true values
with which gross errors were simulated.
Figure 8 - Distribution of corrections predicted by the GLR test from MC

simulation with gross errors according to location
0.100
0.075
0.050
0.025
0.000
Alpha Oil Alpha Gas
0.100
0.075
0.050
Probability Density
0.025
0.000
Bravo Oil Bravo Gas
0.100
0.075
0.050
0.025
0.000
0.100
0.075
0.050
0.025
0.000
-20 -10 0 10 20 -20 -10 0 10 20
GLR Test Standardised Correction (sd)
15
The accuracy with which the GLR Test calculates the value of the gross error varies
according to location. The accuracy improves with the test’s Power. A scatter plot of
calculated correction against the true correction shows the expected correlation.
Figure 9 - Calculated correction predicted by GLR Test plotted against the true
gross error.
20
10
-10
-20
Alpha Oil Alpha Gas
Calculated Standardised Correction (sd)
20
10
-10
-20
Bravo Oil Bravo Gas
20
10
-10
-20
20
10
-10
-20
-20 -10 0 10 20 -20 -10 0 10 20
True Standardised Correction (sd)
3.3 Comparison of Allocation Factor vs. Global Test
A revealing comparison of the Allocation Factor and Global tests is made in Figure
10. It demonstrates once more the Power of the Global Test in comparison to an
Allocation Factor test. The horizontal red lines denote the limits outside of which the
Allocation Factor test would identify a gross error. The vertical red line denotes the
critical test threshold for the Global Test. For certain gross error locations the
Allocation Factor changes slowly in comparison to the Global Test statistic and
therefore does not exceed the accepted range for the Allocation Factor test.
Of course, we could revise the accepted range, but in doing so we would have to
consider the measurement uncertainties. We would therefore arrive at a position of
being able to apply the Global Test, because for the linear case it only requires the
measurement data, associated measurement uncertainties and the process constraints.
It is not dependent on the results of an allocation system being derived by data
reconciliation, pro-rata or even by-difference allocation.
16
Figure 10 – Scatter plot of Oil and Gas Allocation Factor against Global Test
statistic for MC simulation according to gross error location.
Export Oil
Export Gas
1.50
1.50
1.25
1.25
1.00
1.00
0.75
0.75
0.50
0.50
Alpha Oil
Alpha Gas
1.50
1.50
1.25
1.25
1.00 1.00
Oil Allocation Factor
Gas Allocation Factor

0.75 0.75
0.50 0.50
Bravo Oil Bravo Gas
1.50 1.50
1.25 1.25
1.00 1.00
0.75 0.75
0.50 0.50
1.50 1.50
1.25 1.25
1.00 1.00
0.75 0.75
0.50 0.50
0 50 100 150 0 50 100 150
Global Test Statistic Global Test Statistic
17
4 DESCRIPTION OF GPIII SYSTEM
4.1 Process
A schematic of the sub-sea well configuration is presented in Figure 11:
Figure 11 – Dumbarton, Lochranza and Balloch Sub-Sea Configuration
18
Figure 12 shows the subsea and GPIII topsides process and associated topsides
metering:
Figure 12 – GPIII: Simplified Schematic of Subsea and Topsides Process
USM HP Flare Fuel

F F
USM Import
F
Gas Lift
Compressed Gas
USM Gas Export

F F
NGL
MV-032-A
1st Stage Sep
LP Compressor
LP Flare
USM LP Flare
F
MV-032-B
1st Stage Sep
MV-0401
LP Sep
Oil to
F Storage
P17 P18
Riser Base
Balloch Metering Skid
P1 P2 P3 P12 P16 P15 P13
Lochranza
DCC Manifold DC2 Manifold
Metering Skid
P6 P7 P4 P8 P9 P11 P14
The GPIII FPSO handles production from the Dumbarton, Lochranza and Balloch
fields. Production is metered with a combination of export product meters and subsea
multiphase flow meters (MPFM). Each of the Balloch wells and three Lochranza
wells have a dedicated MPFM. There is a single MPFM for each of the Dumbarton
drill centres (DCC & DC2) which are used to test the performance of the Dumbarton
wells and one Lochranza well. Lift gas to each Lochranza well is also individually
metered.
As shown in Figure 12, the Dumbarton, Lochranza and Balloch fluids are
commingled upstream of the 1st stage separators.
Oil separation on GPIII is achieved using two-stage separation with inlet and 2nd
stage heating. Gas from separation is sent to the Low Pressure (LP) and High Pressure
(HP) compression trains with produced water passed to the produced water handling
package. The plant recycles large quantities of NGL from the compression trains to
the separators. This presented difficulties with the modelling of the process in
simulation packages as discussed in the earlier paper.
19
The oil is stabilised, offloaded by tanker and shipped to market. Gas is not currently
being exported from GPIII but is utilised as lift and injection gas. Prior to MacCulloch
cessation of production, the export gas route was via the MacCulloch FPSO tie-in,
onto Piper B and into the Frigg system at St Fergus.
All oil and gas product streams (including fuel and flare) are measured.
4.2 Motivation for Uncertainty Based Allocation
Historically, once Lochranza commenced flowing the Dumbarton field was allocated
By-Difference. For example the Dumbarton’s allocated oil was determined by
subtracting the totalised Lochranza MPFM dry oil flow, after allowing for shrinkage,
from the commingled oil export meter.
Similarly, the total produced gas was calculated by summing fuel, flare, export (and
netting import) measured flows and subtracting the Lochranza MPFM gas flow, after
allowing for lift and process effects, to obtain Dumbarton allocated gas.
This was initially acceptable. However, as indicated in the previous paper, once
Dumbarton became the minority field it caused problems with the allocation results;
Dumbarton was sometimes allocated oil but not allocated any gas.
One of the key features in the selection of an uncertainty based allocation approach
was the identification of the fact that the two fields’ wellstream compositions are
essentially constant resulting in a stable GOR. Both the Dumbarton and Lochranza
reservoir pressures are maintained above the bubble point. This means that the
hydrocarbons in the reservoir rock will be in a single phase and hence when produced
up the well bore the composition of each field’s hydrocarbon fluids entering the GPIII
process should be essentially constant. The GOR may vary from day to day depending
on operating conditions and any process dynamical instabilities but it should not vary
as widely as is observed in the allocated data. Indeed the variation in the allocated
GOR has been used as a metric to judge the quality of, and consequently question, the
allocation results.
The GORs connect the oil and gas allocated to each field and because they can be
estimated to within a tolerance or nominal uncertainty they can be incorporated as
inputs into the allocation system. Since the development of GPIII uncertainty based
allocation system Balloch has come online. Its reservoir pressure is again above the
bubble point and the uncertainty based allocation scheme has been extended to
incorporate Balloch.
Mathematical details of the Uncertainty Based Allocation scheme required in this

paper are provided in section 8; see the earlier paper [1] for more information.
20
5 GPIII SIMULATION STUDIES
A similar approach to the MC studies presented in section 3 was undertaken to

establish the suitability of the different tests for gross error detection with GPIII. The
tests rely on the same concepts used in the GPIII uncertainty based allocation system,
with an important addition: the uncertainties in the export gas, export oil, fuel, flare,
injection and import gas masses are non-zero allowing them to feature in the GED
tests.
The measurement values and relative uncertainties for the MC simulation were taken
from a typical recent day. The values shown take into account shrinkage effects. For
the purposes of the test they were reconciled to provide a perfect mass balance from
which to start the GED tests. Their values and uncertainties are shown in Table 4
Table 4 – Measurement values and uncertainties used in the GPIII MC GED

tests
Mass or GOR Relative
(tonnes or Uncertainty
tonnes/tonnes) (%)
Export Oil 4556.9 1.0%
Export Gas 0.0 1.0%
Inj Gas 80.9 2.0%
Fuel Gas 6.8 5.0%
HP Flare Gas 46.3 5.0%
LP Flare Gas 413.1 2.0%
Import Gas 0.0 1.0%
P13 Oil 0.0 0.0%
P13 Gas 0.0 0.0%
P14 Oil 138.9 21.6%
P14 Gas 12.2 128.1%
P15 Oil 252.6 29.6%
P15 Gas 22.1 179.7%
Loch GOR 0.088 56.8%
P17 Oil 1688.1 4.6%
P17 Gas 226.8 6.0%
P18 Oil 1600.1 4.6%
P18 Gas 215.0 6.0%
Balloch GOR 0.134 57.0%
Non-metered Oil 877.2 77.2%
Non-metered GOR 0.081 57.1%
The MPFM uncertainties are affected by both high water cut and uncertainty in the lift
gas provided to each well. MPFM uncertainty is initially calculated based on [9] and
are in accordance with a GVF below 90% and operating pressure above 20 barg. For
an MPFM the dry oil flow uncertainty is function of the measured liquid and WLR
and their associated uncertainties. The relative uncertainty in the oil flow is given by:
21
Liq * 1  WLR 2  e WLR 2 [16]

MOil 
1  WLR 
Inspection of the above equation reveals that the relative uncertainty in the oil
becomes very large as the WLR approaches 1.
All gas production figures quoted are nett of lift gas. The lift gas metering
arrangements require the nett gas production uncertainty at each well to reflect the
uncertainties in the Injection Gas meter, individual lift gas meters and subsea MPFM.
Propagating these errors leads to the extremely high nett gas production uncertainty
on the Lochranza P14 and P15 wells in this example. The Balloch wells do not use lift
gas and so their uncertainties are derived from the vendor’s datasheet.
The GED tests for GPIII are defined slightly differently to those presented in section
3.1 to take account of the nonlinear constraints.
5.1 Test Definitions
Allocation Factor
The allocation factor for each measurement is defined as, the ratio of allocated mass
to measured mass (after taking shrinkage into account), so for example,
𝑎𝑚𝑃15,𝑂𝑖𝑙 [17]
𝐴𝐹𝑃15,𝑂𝑖𝑙 =
𝑚𝑃15,𝑂𝑖𝑙
Where 𝑚𝑃15,𝑂𝑖𝑙 denotes the P15 MPFM unreconciled shrunk mass of oil and
𝑎𝑚𝑃15,𝑂𝑖𝑙 denotes the mass of oil allocated to P15. We will test for an Allocation
Factor in the range 0.9 < 𝐴𝐹 < 1.1 as before.
Global Test
Using the Constraint matrix, 𝐴, and Jacobian matrix, 𝐽, defined for GPIII UBA in
section 8 we have approximated the Global Test from linear data reconciliation by
calculating the test statistic,
𝛾 = 𝐴𝑦(𝐽𝑉𝐽𝑇 )−1 (𝐴𝑦)𝑇 [18]
The Constraint and Jacobian matrices used for this paper have been calculated from
the unreconciled data. We have assumed that under the null hypothesis the above
statistic will approximately follow a χ2 distribution with four degrees of freedom for
the GPIII system under consideration. A gross error is detected if γ ≥ χ2 (α) , for the
chosen confidence level, α. At a 95% confidence level, a gross error is detected if
γ ≥ 9.49.
22
Generalised Likelihood Ratio Test
As earlier, the GLR Test statistic is calculated as:
𝑑𝑘2 [19]
𝑇𝑘 =
𝐶𝑘
Where we have approximated the numerator and denominator as

𝑑𝑘 = 𝑓𝑘 𝑇 (𝐽𝑽𝐽𝑇 )−1 𝑟 [20]
𝐶𝑘 = 𝑓𝑘 𝑇 (𝐽𝑉𝐽𝑇 )−1 𝑓𝑘 [21]
and each vector 𝑓𝑘 is the column of the Jacobian corresponding to the measurement.
We have followed the approach in [8] where the Jacobian used in a study of a
nonlinear GLR test is that obtained after data reconciliation.
Assuming again that under the null hypothesis the above statistic approximately
follows a χ2 distribution with one degree of freedom a gross error is detected if any of
the test statistics exceed the test criterion χ2 (𝛽) for the reduced confidence level
where 𝛽 is given by equation 12. For 𝛼 =0.05 and with 21 measurements, 𝛽 = 0.002.
A gross error is then detected if any of the measurement test statistics, χ2 (𝛽) ≥ 9.19
For the measurement with the highest test statistic above the threshold, the error on
that measurement is again estimated according to equation 13.
5.2 Monte Carlo Simulation
A similar Monte Carlo approach to section 3.2 has been taken to analyse expected
results for GED tests on GPIII. However, due to the magnitude of some of the
measurements’ relative uncertainties causing negative simulated unreconciled data
only positive gross errors have been simulated. The results are presented in Table 5.
23
Table 5 – Performance statistics of GED tests applied to GPIII MC simulated

data.
Allocation Factor Global Test GLR Test
Error Location Type I Power Type I Power Type I Power
Error Error Error
Overall 0.215 0.273 0.060 0.544 0.006 0.420
Export Oil 0.000 0.000 0.063 0.217 0.008 0.045
Export Gas 0.000 0.000 0.062 0.159 0.012 0.009
Inj Gas 0.000 0.000 0.052 0.171 0.009 0.017
Fuel Gas 0.000 0.000 0.067 0.157 0.012 0.012
HP Flare Gas 0.000 0.000 0.059 0.179 0.010 0.018
LP Flare Gas 0.000 0.000 0.056 0.289 0.011 0.092
Import Gas 0.000 0.000 0.056 0.170 0.008 0.010
P13 MPFM Oil 0.000 0.000 0.060 0.162 0.006 0.009
P13 MPFM Gas 0.000 0.000 0.065 0.163 0.010 0.012
P14 MPFM Oil 0.000 0.000 0.068 0.184 0.003 0.011
P14 MPFM Gas 0.939 1.000 0.060 1.000 0.003 1.000
P15 MPFM Oil 0.004 0.000 0.056 0.282 0.003 0.064
P15 MPFM Gas 0.909 1.000 0.054 0.964 0.003 0.914
Loch GOR 0.481 0.681 0.067 0.829 0.008 0.663
P17 MPFM Oil 0.000 0.000 0.055 0.808 0.004 0.613
P17 MPFM Gas 0.000 0.011 0.059 0.955 0.005 0.895
P18 MPFM Oil 0.000 0.000 0.057 0.808 0.004 0.608
P18 MPFM Gas 0.000 0.009 0.059 0.952 0.004 0.886
Balloch GOR 0.721 1.000 0.060 1.000 0.001 0.999
Non-metered Oil 0.786 1.000 0.062 0.998 0.003 0.997
Non-metered GOR 0.654 1.000 0.063 0.988 0.009 0.951
Table 5 shows two sides to the use of Allocation Factors to identify errors. For export
and disposal streams where the combination of flow rate and measurement
uncertainty is low the Type I error rate is found to be zero. However the production
streams have high Type I error rates and so the false alarm rate would be high. The
Power of the Allocation Factor test is high for several streams. However since these
streams are also those where the false alarm rate is high, it seems unlikely that the
Allocation Factor test can identify gross errors in the GPIII data without further
information.
In contrast, both the Global and GLR tests have respectable Type I error rates
irrespective of where a gross error may be located. For certain product streams the
Power of the Global and GLR tests are in the region 80 to 90 % indicating good
capability to identify gross errors if they do occur.
24
6 APPLICATION TO GPIII DATA
In light of the MC simulation studies the Allocation Factor test has not been applied
to actual GPIII data. The Global and GLR tests have been applied to recent GPIII data
and a selection of the results are presented in Figure 13 to Figure 16.
The period had regular stretches of stable production interrupted by occasional upset
days. From mid-March to mid-April the Dumbarton P3, Lochranza P15, and Balloch
P17 wells were online continuously. From the 9/5 onwards production was again
stable with both Balloch wells, the Lochranza P14 and P15 wells and several
Dumbarton wells online continuously. Although there are several days during this
period when the Global Test indicates a potential gross error, only one day does not
coincide with a start-up or shutdown and it seems the most obvious explanation is that
the Global Test has raised a false alarm. The GLR Test results are compatible with the
Global test.
Figure 13 – Global Test results for March to May 2015
50
45
40
35
30
25
20
15
10
Global Test Statistic Global Test Threshold
25
Figure 14 – GLR Test results for Lochranza’s P15 well March to May 2015
10000.000
1000.000
100.000
GLR Test Statistic
10.000
1.000
0.100
0.010
0.001
01-Mar-15 15-Mar-15 29-Mar-15 12-Apr-15 26-Apr-15 10-May-15 24-May-15
P15 Metered Oil P15 Metered Gas Lochranza GOR GLR Test Threshold
Figure 15 - GLR Test results for Balloch’s P15 well March to May 2015
10000.000
1000.000
100.000
GLR Test Statistic
10.000
1.000
0.100
0.010
0.001
P17 Metered Oil P17 Metered Gas Balloch GOR GLR Test Threshold
26
Figure 16 – GLR Test for the non-metered group March to May 2015
10000.000
1000.000
100.000
GLR Test Statistic
10.000
1.000
0.100
0.010
0.001
Non-Metered Oil Non-Metered GOR GLR Test Threshold
27
7 CONCLUSIONS
The use of Allocation Factors for detecting errors in process data has been compared
to the Global and Generalised Likelihood Ratio tests adapted from linear data
reconciliation for use in the GPIII UBA system.
Monte Carlo simulation suggests the Global and GLR tests offer a greater chance of
accurately identifying gross errors compared to Allocation Factors.
A simple example based on a pro-rata allocation system showed how the Global and
GLR tests can be applied to systems which do not implement UBA.
The Global and GLR Tests have been applied to a period of actual GPIII data. Given
the assumed uncertainties in measured and estimated quantities, there are relatively
few days where the tests identify possible gross errors. The possible gross errors
mostly coincide with start-up/shutdown days and are assumed to reflect the non-
steady-state nature of these days rather than the presence of gross errors.
28
8 MATHEMATICAL ANALYSIS UBA MATRIX SOLUTION

TECHNIQUE
This section has been taken from [1]. The addition of Balloch wells changes the data
reconciliation calculations by requiring an additional gas constraint relating the
Balloch gas, oil and GOR. The principles of the UBA approach defined in the
following section are otherwise unchanged.
This method is described in [10] and [11] and is based upon the principles of data
reconciliation as described more generally in [10].
It should be noted that a rigorous data reconciliation method would reconcile the
product measurements (DOILM, EXPGM, etc.) as well as the allocated quantities
(ADOILMg, etc). The sum of allocated quantities would not, therefore, exactly equal
the recorded measurements. For allocation, it is generally required that the sum of the
allocated quantities is equal to the recorded measurement. So although the product
(fiscal) measurements are included in the equations below their uncertainties are
assumed to tend to zero, and this ensures the sum of the allocated quantities is equal
to the recorded measurement. This is justifiable because the fiscal product
measurements are generally substantially more accurate than production estimates.
It should also be noted that total produced gas, with its associated uncertainty,
represents the combined fiscal gas export, fuel, flare and injection gas streams less
import gas, and their uncertainties. The total produced gas term can be replaced by the
individual stream quantities and their associated uncertainties in the following
equations. This simply leads to matrices of higher dimension in the equations. The
entries representing fiscal export, fuel and flare gas would all be analogous to those
for total produced gas shown here. Similarly, water could be included in the data
reconciliation, with an additional constraint and inclusion of the necessary metered or
estimated stream masses and uncertainties leading to a further increase in the
dimensions of the matrices involved in the equations.
Theory
The full system of equations to be solved for the GP III system is shown below:
2 2
𝐴𝐷𝑂𝐼𝐿𝑀𝑃13 − 𝑇𝐻𝑊𝑂𝑀𝑃13 𝐴𝐷𝑂𝐼𝐿𝑀𝑃14 − 𝑇𝐻𝑊𝑂𝑀𝑃14
𝜓=( ) +( )
𝑈𝑇𝐻𝑊𝑂𝑀𝑃13 𝑈𝑇𝐻𝑊𝑂𝑀𝑃14
2 2
𝐴𝐷𝑂𝐼𝐿𝑀𝑃15 − 𝑇𝐻𝑊𝑂𝑀𝑃15 𝐴𝐷𝑂𝐼𝐿𝑀𝐷𝑢𝑚𝑏 − 𝑇𝐻𝑊𝑂𝑀𝐷𝑢𝑚𝑏
+( ) +( )
𝑈𝑇𝐻𝑊𝑂𝑀𝑃15 𝑈𝑇𝐻𝑊𝑂𝑀𝐷𝑢𝑚𝑏
2 2
𝐴𝑃𝐺𝐴𝑆𝑀𝑃13 − 𝑇𝐻𝑊𝐺𝑀𝑃13 𝐴𝐺𝑂𝑅𝑃14 − 𝑇𝐻𝑊𝐺𝑀𝑃14
+( ) +( )
𝑈𝑇𝐻𝑊𝐺𝑀𝑃13 𝑈𝑈𝑇𝐻𝑊𝐺𝑀
𝑃14
29
2 2
𝐴𝑃𝐺𝐴𝑆𝑀𝑃15 − 𝑇𝐻𝑊𝐺𝑀𝑃15 𝐴𝐺𝑂𝑅𝐿𝑜𝑐ℎ − 𝑁𝑂𝑇𝐺𝑂𝑅𝐿𝑜𝑐ℎ
+( ) +( )
𝑈𝑇𝐻𝑊𝐺𝑀𝑃15 𝑈𝑁𝑂𝑇𝐺𝑂𝑅𝐿𝑜𝑐ℎ
2
𝐴𝐺𝑂𝑅𝐷𝑢𝑚𝑏 − 𝑁𝑂𝑇𝐺𝑂𝑅𝐷𝑢𝑚𝑏
+( )
𝑈𝑁𝑂𝑇𝐺𝑂𝑅𝐷𝑢𝑚𝑏
(1)
The mass balance constraints on the oil phase and gas phase are:
Oil  0   DOILM  ADOILM P13  ADOILM P14  ADOILM P15

 ADOILM Dumb
(2)
1Gas  0  TPGASM  APGASM P13  APGASM P14  APGASM P15
  ADOILM Dumb * AGORDumb 
(3)
2Gas  0  APGASM P13  APGASM P14  APGASM P15

  ADOILM Loch * AGORLoch 
(4)
The optimum solution to the system is found by minimising the value of Ψ (psi) in
Equation (1), subject to the constraints of Equations (2), (3) and (4).
For systems with two fields, simultaneous equations can be easily written out
explicitly and solved iteratively. However, for systems with more than two fields, the
equations are more complex, and a matrix-solution method is recommended. Such a
solution is described below.
Matrix Solution Method – Inputs
The input data to the matrix solution method are provided in the form of arrays and
vectors. The integer n represents the number of variables to be reconciled.
Y (Input) vector of measured data (dimension n, 1).

X (Calculated) vector of reconciled data (dimension n,1).
V Variance-covariance matrix for Y (dimension n,n). The covariance of each
element to itself is calculated from the square of the absolute uncertainty (U) of
the measurement (Ym) divided by 2, (Um/2)2. The covariance of any element
with any other element is zero because the quantities are independent.
J Jacobian matrix (dimension number of constraint equations n). This contains
the coefficients of the derivatives of the oil and gas constraints Equations (2), (3)
and (4) – see below for derivation.
30
For example, for the 2-field Lochranza and Dumbarton application, the “measured”
data comprised the stream measurements and the theoretical oil and gas production (at
export conditions) for each Field. These were mass values, based on MPFM
measurements for Lochranza Field and on well-tested oil quantities and constant GOR
for Dumbarton Field. All theoretical production quantities were calculated within the
allocation system.
For a higher-order system, such as GP III, the principles are the same but the matrices
are extended to include the additional Field values.
The subsequent matrices are shown with only 2 Fields. Equivalent terms for
additional Fields should be inserted where indicated by “…”.
DOILM Metered Export Oil

TPGASM Produced Gas
THWOMP13 Theoretical P13 MPM oil (at export conditions)
THWGMP13 Theoretical P13 MPM gas (at export conditions)
Y= =
… …
NOTGOR Loch Notional GOR Lochranza (at export conditions)
THWOMDumb Theoretical Dumbarton oil (at export conditions)
[NOTGOR Dumb] [ Notional GOR Dumbarton (at export conditions) ]
𝑈𝐷𝑂𝐼𝐿𝑀 2
( ) 0 0 0 … 0 0 0
2
𝑈𝑇𝑃𝑅𝑂𝐷𝐺𝐴𝑆𝑀 2
0 ( ) 0 0 … 0 0 0
2
𝑈𝑇𝐻𝑊𝑂𝑀,𝑃13 2
0 0 ( ) 0 … 0 0 0
2
𝑈𝑇𝐻𝑊𝐺𝑀,𝑃13 2
𝑉= 0 0 0 ( ) … 0 0 0
2
… … … … … … 0 …
𝑈𝑁𝑂𝑇𝐺𝑂𝑅,𝐿𝑜𝑐ℎ 2
0 0 0 0 0 ( ) 0 0
2
𝑈𝑇𝐻𝑊𝑂𝑀,𝐷𝑢𝑚𝑏 2
0 0 0 0 0 0 ( ) 0
2
𝑈𝑁𝑂𝑇𝐺𝑂𝑅,𝐷𝑢𝑚𝑏 2
[ 0 0 0 0 … 0 0 ( ) ]
2
The matrix solution is an iterative method, based on the “Jacobian matrix” (J). The
Jacobian terms reflect the non-linear terms in the least-squares-type method used to
determine the minimum value of Ψ in Equation (1). The Jacobian terms represent the
coefficients of the derivatives of the oil and gas constraints (Equations (2), (3) and (4))
with respect to each reconciled quantity, Xm , e.g., ∂ФOil/∂Xm and . ∂Ф1Gas/∂Xm.
𝜕Φ𝑂𝑖𝑙 𝜕Φ𝑂𝑖𝑙 𝜕Φ𝑂𝑖𝑙 𝜕Φ𝑂𝑖𝑙 𝜕Φ𝑂𝑖𝑙

⋯ ⋯ ⋯ ⋯
𝜕𝐷𝑂𝐼𝐿𝑀 𝜕𝑇𝑃𝑅𝑂𝐷𝐺𝐴𝑆𝑀 𝜕𝐴𝐷𝑂𝐼𝐿𝑀𝑃13 𝜕𝐴𝑃𝐺𝐴𝑆𝑀𝑃13 𝜕𝐴𝐺𝑂𝑅𝐿𝑜𝑐ℎ
𝜕Φ1𝐺𝑎𝑠 𝜕Φ1𝐺𝑎𝑠 𝜕Φ1𝐺𝑎𝑠 𝜕Φ1𝐺𝑎𝑠 𝜕Φ1𝐺𝑎𝑠
𝐽= ⋯ ⋯ ⋯ ⋯ 𝜕𝐴𝐺𝑂𝑅𝐿𝑜𝑐ℎ
𝜕𝐷𝑂𝐼𝐿𝑀 𝜕𝑇𝑃𝑅𝑂𝐷𝐺𝐴𝑆𝑀 𝜕𝐴𝐷𝑂𝐼𝐿𝑀𝑃13 𝜕𝐴𝑃𝐺𝐴𝑆𝑀𝑃13
𝜕Φ2𝐺𝑎𝑠 𝜕Φ2𝐺𝑎𝑠 𝜕Φ2𝐺𝑎𝑠 𝜕Φ2𝐺𝑎𝑠 𝜕Φ2𝐺𝑎𝑠
[𝜕𝐷𝑂𝐼𝐿𝑀 ⋯ ⋯ ⋯ ⋯ 𝜕𝐴𝐺𝑂𝑅𝐿𝑜𝑐ℎ ]
𝜕𝑇𝑃𝑅𝑂𝐷𝐺𝐴𝑆𝑀 𝜕𝐴𝐷𝑂𝐼𝐿𝑀𝑃13 𝜕𝐴𝑃𝐺𝐴𝑆𝑀𝑃13
−1 0 1 0 … … 0 1 0
𝐽 = [ 0 −1 0 1 … … 0 𝐴𝐺𝑂𝑅𝐷𝑢𝑚𝑏 𝐴𝐷𝑂𝐼𝐿𝑀𝐷𝑢𝑚𝑏 ]
0 0 𝐴𝐺𝑂𝑅𝐿𝑜𝑐ℎ −1 … … 𝐴𝑃𝐺𝐴𝑆𝑀𝐿𝑜𝑐ℎ 0 0
31
The matrix solution is therefore an iterative method, because some of the coefficients
of the non-linear Jacobian terms (AGORDumb, ADOILMDumb, AGORLoch and
ADOILMLoch) are dependent on the previous solution.
For the first iteration only, the Jacobian matrix uses the theoretical estimates of the
non-metered Field GOR and Oil.
Matrix Solution Method
The reconciled measurements X which result in the minimum value of Ψ in the

system of equations described above may be described as follows and are calculated
using the method described in [10] and shown in Equation (5) below:
𝐴𝐷𝑂𝐼𝐿𝑀
𝐴𝑇𝑃𝑅𝑂𝐷𝐺𝐴𝑆𝑀
𝐴𝐷𝑂𝐼𝐿𝑀𝑃13
𝐴𝑃𝐺𝐴𝑆𝑀𝑃13
𝑋= …
…
…
…
[ 𝐴𝐺𝑂𝑅𝐷𝑢𝑚𝑏 ]
X  Y  K  f  X 0   J Y  X 0 
(5)
Where,
X is the vector containing the reconciled measurements calculated by this iteration.
Y is the vector containing the initial measurements, as defined above.
K is an intermediate matrix, defined as:

K  VJ T JVJ T 1
(6)
V is the covariance matrix for Y, as defined above.

JT is the transpose of the Jacobian matrix, J.
f(X0) is the imbalance vector, and is calculated from the product of the Jacobian
matrix (J), and the current estimated measurements (X0).
f ( X 0 )  JX
(7)
J is the Jacobian matrix, as defined above.

X0 is the vector containing the reconciled measurements from the previous iteration.
32
Matrix Solution Method – Initialisation
1. Specify elements of measurements matrix, Y.

2. Calculate elements of variance-covariance matrix, V.
3. Specify initial elements of initial Jacobian matrix, J1, using the theoretical
field quantities.
4. Calculate intermediate matrix K from Equation (6): K=V JT (J V JT)-1.
5. Initialise value of reconciled measurements vector, X0 = Y.
6. Calculate new values of reconciled measurements vector, X from Equations
(5) and (7).
Matrix Solution Method – Iteration
7. Update value of reconciled measurements vector, X0 = X from previous

iteration.
8. Update elements of Jacobian matrix, J, using the latest reconciled
measurements.
9. Update intermediate matrix K from Equation (6): K=V JT (J VJT)-1.
10. Calculate new values of reconciled measurements vector, X from Equations 4
and 6.
11. Calculate absolute change in reconciled measurements vector: ABS(X- X0).
12. If the sum of the absolute changes in reconciled measurements has changed by
more than the specified tolerance, repeat steps 7 to 12.
33
NOTATION
ADOILM Allocated dry oil mass X0 Reconciled or allocated

AGOR Allocated GOR data from previous
APGASM Allocated produced gas iteration vector
mass Y Input data vector
AVTI Average Type I Error rate
DOILM Measured dry product oil Greek
mass
e Uncertainty (absolute)  Uncertainty (relative)
EXPGM Export gas mass θ Type II Error Rate (1-
f(X0) Imbalance vector Power)
J Jacobian matrix ψ Objective function
K Intermediate matrix 𝜙 Constraint
NOTGOR Notional GOR
P Constraint projection
matrix Subscripts
THWGM Theoretical well gas mass
THWOM Theoretical well oil mass Dumb Dumbarton
TPGASM Total produced gas mass liq mass of liquid
U Absolute uncertainty Loch Lochranza
V Variance covariance matrix moil mass of oil
WLR Water Liquid Ratio P13, etc Well P13, etc
X Reconciled or allocated
data vector
9 REFERENCES
[1] Allocation in an Uncertain World. Maximising the Use of Data with

UBA on Global Producer III, N. Corbett, J. Johnston, R. Sibbald, P.
Stockton and A. Wilson, 31st North Sea Flow Measurement Workshop,
22-25 October 2013.
[2] Ripps, D. L. Adjustment of Experimental Data, Chem. Eng. Prog.
Symp. Ser 55, 1965, 61, 8.
[3] Experiences in the Use of Uncertainty Based Allocation in a North Sea
Offshore Oil Allocation System, P. Stockton and A. Spence,
Production and Upstream Flow Measurement Workshop,12-14
February 2008.
[4] Production Quantification Through Full Field Modelling – Case Study.
Kjartan Berg and Daniel Fonnes, 32nd North Sea Flow Measurement
Workshop, 21-24 October 2014.
[5] Data Reconciliation and Gross Error Detection, An Intelligent Use of
Process Data, Shankar Narasimhan and Cornelius Jordache, published
in 2000 by Gulf Publishing Company, Houston Texas, ISBN 0-88415-
255-3.
[6] Performance Improvement of Large Installation Base of Wellhead
Venturi Wet Gas Measurement in Petroleum Development Oman
34
(PDO). Abdullah Al Obaidani et al., 31st North Sea Flow Measurement

Workshop, 22-25 October 2013.
[7] Determination of Measurement Uncertainty for the Purpose of Wet Gas
Hydrocarbon Allocation, R. A. Webb, W. Letton, M. Basil, North Sea
Flow Measurement Workshop 2002.
[8] Renganathan, T. and Narasimhan, S., A Strategy for Detection of Gros
Errors in Nonlinear Processes, Ind. Eng. Chem. Res., 1999, 38, 2391.
[9] Allocation Uncertainty: Tips, Tricks and Pitfalls, Phil Stockton and
Allan Wilson, Proceedings of the 30th International North Sea Flow
Measurement Workshop, 23-26 October, 2012.
[10] "Wring more information out of Plant Data", Robert Kneile, Bailey
Controls Co, Chemical Engineering March 1995, pps 110 - 116. Box
"deriving the SDF" p115, equations (9) and (10).
[11] The Estimation of Parameters in Nonlinear, Implicit Models, H. I. Britt
and R. H. Luecke, Technometrics Vol 15, No.2.
35
20th– 23rd October 2015
Could Allocation be Rocket Science?

Using the Kalman Filter to Optimise Well Allocation Accuracy
Euain Drysdale, Accord

Phillip Stockton, Accord
1 INTRODUCTION
In the 1960s, the Kalman filter was applied to navigation for the Apollo Project,
which required estimates of the trajectories of manned spacecraft going to the Moon
and back. The Kalman filter is an optimal estimation technique that uses dynamic
mathematical models and physical measurements to obtain the most probable
estimates of underlying variables as they vary with time. It incorporates both the
measurement and also model uncertainties into its highly efficient estimation
algorithm. Today, the use of the Kalman filter is extremely widespread throughout
many science and engineering applications.
This paper describes the application of the Kalman filter to the problem of estimating
(and hence allocating) well production based on intermittent and possibly poor quality
measurement data from well tests. The technique recognises the increasing
uncertainty in the estimated well flows as the time since the last well test elapses.
It also utilises the additional measurement of the wells’ daily, aggregate, commingled
production as it exits a process. It can take advantage of process upsets in that when
wells shut in, the commingled production of the remaining wells is measured and the
drop in production is an estimate of the shut in wells. This information is smoothly
incorporated into the well estimates and propagates forward in time.
The efficacy of the Kalman filter is demonstrated using simplified examples and then
applied to typical real world, anonymised data.
The industry has called for better reservoir management techniques (Wood Review
[1]). As part of the response, adoption of the Kalman filter can provide a mechanism
to improve well production estimates.
Section 2 introduces the main concepts of the Kalman filter and illustrates its
widespread practical use in numerous applications. Section 3 describes the potential
uses of the filter in terms of oil and gas allocation, specifically focussing on its
application to improve well production estimates. Section 4 explores the suitability of
the filter using theoretical models and Section 5 applies the filter to real data. Section
6 provides some conclusions. Sections 7, 8, 9 and 10 present: Mathematical Analyses,
Notation, References and an Appendix of Figures, respectively.
1
2 INTRODUCTION TO THE KALMAN FILTER
2.1 What is the Kalman Filter?
The Kalman filter is an algorithm that uses a series of measurements observed over
time, containing statistical noise (i.e. uncertainty), and produces estimates of unknown
variables that tend to be more precise than those based on a single measurement
alone.
The algorithm works in a two-step process. In the prediction step, the Kalman filter
produces estimates of the current state variables1, along with their uncertainties. Once
the outcome of the next measurement (necessarily corrupted with some amount of
uncertainty, including random noise) is observed, these estimates are updated (in the
update step) using a weighted average, with more weight being given to estimates
with lower certainty. The algorithm is recursive in that it uses only the present input
measurements and the previously calculated state and its uncertainty matrix; no
additional past information is required.
2.2 History and Applications
The filter is named after Rudolf E. Kálmán, who wrote the original paper describing
the filter [2] in 1960. (It is referred to as a filter as it filters the noise (or uncertainties)
in the system and measurements). It was subsequently developed at NASA [3] and
applied to the problem of trajectory estimation for the Apollo program, leading to its
incorporation in the Apollo navigation computer. Apollo 8 (December 1968), the first
human spaceflight from the Earth to an orbit around the Moon, would certainly not
have been possible without the Kalman filter.
Since then, Kalman filters have been employed in a wide variety of practical
applications, which include:
 Aircraft autopilots
 Navigation
 Dynamic positioning
 Computer vision
 Economic modelling
 Weather forecasting
 Climate modelling
 Missile guidance
 Speech enhancement.
1
In this paper, these state variables are the well flow estimates or potentials.
2
3 APPLICATION TO WELL ESTIMATION AND ALLOCATION
3.1 Concept
The Kalman filter is applied to dynamic systems and uses process models, along with
noisy measurements, to provide best estimates of variables in the system.
The mathematics of steady state data reconciliation has previously been used to
develop an approach to uncertainty-based allocation (UBA) [5]. UBA reconciles
noisy measurements to perform a daily allocation. An extension of these ideas to
dynamic systems involves the use of the Kalman filter to perform the data
reconciliation and is described further in [10]. Hence, the use of the Kalman filter
appeared a natural extension of the ideas used to develop UBA.
There are potentially a number of applications of the Kalman filter in allocation

systems. Though daily hydrocarbon allocation is essentially steady state, there can be
a number of dynamic variables that feature in allocation systems. For example:
 Storage tank volumes, which fill over several days then are emptied
periodically;
 Oil and gas pipelines, in which hydrocarbons of different compositions are
commingled and through which they flow, possibly being accelerated and
decelerated, over a number of days;
 Well production estimation.
It is the final example which has been used in this study to explore the feasibility and
effectiveness of the Kalman filter for allocation purposes.
3.2 Application to Well Production Estimation for Allocation
The system envisaged in which the Kalman filter could be applied is the case where
individual well production is measured intermittently using well tests. In the interval
between well tests the well oil production can vary due to a number of factors, for
example: rising water cut, change in downhole pressures, changes in lift flow or
choke settings, etc. Ideally, a well should be retested if a significant adjustment takes
place. However, in practice this is not always possible.
So, in the interval between tests it is possible that the uncertainty in the estimate of the
well production can become significant. Various methods may be used to estimate the
interim production: well decline equations, productivity indices, bottom hole and well
head pressures, gas lift curves, etc. However, these estimates are in effect based on
simplistic models of the well production, which cannot capture the complexity of the
noisy real well production, and hence these estimates may still exhibit significant
uncertainty.
A typical allocation approach in such a system is to allocate produced oil on a day in

proportion to the estimated well production, based on well tests, after accounting for
any processing effects (e.g. shrinkage) and well uptime (hours on production):
3
𝑥𝑤,𝑑 𝑞𝑤,𝑑
𝑎𝑤,𝑑 = 𝑚𝑑 ∗ ( ) (1)
∑𝑤 𝑥𝑤,𝑑 𝑞𝑤,𝑑
Where,
aw,d Oil allocation for well, w, on day, d

aw,d Oil allocation for well, w, on day, d
md Produced oil measurement, on day, d
qw,d Daily flow potential estimate for well, for well, w, on day, d
xw,d Fractional uptime (i.e. hours on production24), for well, w, on day, d.
In this equation, the shrinkage is ignored (assumed to be 1) and the well flow estimate
is in terms of a daily (24 hour) potential production qw, i.e. what the well would
produce if it was operating steadily over a day. A well therefore still has a potential
even if it is shut in. The well potential is the key variable that is estimated throughout
this paper.
On a day when a well is tested qw,d can be equated to the test rate tw,d (adjusted to a 24
hour rate). However, it should be borne in mind that the well test measurement is
noisy and hence exhibits a degree of uncertainty. In addition, as time elapses since a
well was last tested, the uncertainty in qw,d will rise.
There is additional information provided by Equation (1), in that the sum of all wells’
production is measured by md, which though it also exhibits uncertainty, is normally a
fiscal meter and therefore more accurate than the well test measurement. Changes in
qw,d will be reflected in md, although without further information it is not possible to
determine confidently which wells’ qw,d to assign a change in md to. However, in real
systems, wells are frequently shut in and started up (due to operational issues). So, for
a specific well (w=A), xA,d reduces from 1 if it is shut in (possibly falling to zero if
shut in for the whole day), during which time the the drop in md may to some extent
be attributed to qA,d. However, it must also be borne in mind that the remaining wells
which have continued producing may also have changed their collective potential to
some extent. What is evident from these considerations though is that the total
measured flow, md, does provide some information about individual well production,
but knowledge of how md varies from day to day is required, which means the system
has a temporal dependency, i.e. it is dynamic.
If changes in md are providing updates in qw,d, it also means that previous values of
qw,d, will have an impact on the current day; again the temporal dependency and
dynamic nature of the problem arises.
In summary, an approach is required that produces better estimates of wells’ daily oil
potentials (qw,d) that incorporates the following features:
 Uses well test measurements;

 Accounts for the uncertainties in the well test measurements;
 Recognises the rising relative uncertainty in the potential as time elapses since
the last well test measurement was obtained;
 Utilises the daily product oil flow measurement;
4
 Accounts for the uncertainty in the product oil flow measurement;

 Infers information about well potentials from changes in the product oil flow,
from one day to the next;
 Infers information about well potentials from a knowledge of their uptimes;
 Carries the most recent well potential estimates into the next day’s allocation
and hence updated well potential estimates.
The answer to this problem is provided by the Kalman filter.
3.3 Specific Implementation of the Kalman Filter for Well Estimation
A number of extensions and generalised methods have been developed but the
Kalman filter proposed here is a linear discrete filter. The proposed implementation is
relatively simple, in that the full Kalman filter includes terms for control variables
which are not required in the well estimation system. A more complete description of
the Kalman filter and its applications is provided by [4].
A mathematical presentation of the Kalman filter equations, as developed for this well
test based allocation system, is provided in Section 7; the various steps are described
briefly below.
Prediction Step
The first equation in the predict phase of the Kalman filter employs the transition
matrix, which predicts how the well flow potential from the previous day propagates
to the current day. This normally involves some multiplier, to account for well
decline, etc. and could simply be 1 if the well flow is assumed constant. In effect, this
is our model of how the well potentials evolve over time. In the real system described
in Section 5, the transition matrix accounts for changes in well choke position.
The uncertainty in the previous day’s potentials (this is an output from the Update
Step of the Kalman filter run on the previous day) is used - with the addition of
uncertainty due to process noise - to obtain the uncertainty in the predicted well flow
potentials. This process noise reflects the uncertainty in the model that we have
assumed.
Update Step
This step incorporates any measurements that are made. If any wells are tested the
well test along with its measurement uncertainty are used to update an individual
well’s potential. Similarly the total metered product oil is a measurement of the sum
of the production of all wells flowing on that day. This is used to update the potential
of all wells flowing on the day.
These measurements are necessarily uncertain and hence the well flow potentials are
updated using a weighted average of all estimates and measurements, with more
weight being given to those with lower uncertainty.
5
Because the uncertainty of the measurements and the process model uncertainty
(process noise) may be difficult to determine precisely, it is common to discuss the
filter's behaviour in terms of gain. The Kalman gain is a function of the relative
uncertainty of the measurements and current estimate of the predicted well potentials,
and can be tuned to achieve particular performance. With a high gain, the filter places
more confidence on the measurements, and thus follows them more closely. With a
low gain, the filter follows the model predictions more closely, smoothing out noise
but decreasing the responsiveness.
The uncertainties of the updated well potentials are also output by the filter. This is in
the form of the covariance matrix which is a familiar feature in data reconciliation
techniques.
Recursion
The calculated well potentials and their associated uncertainties form the input to the
calculations the following day. This is the recursive nature of the Kalman filter and all
the information required to perform the calculations the next day is contained within
the estimates and uncertainties from the current day.
This feature makes the Kalman filter computationally very efficient, which was one of
its attractions in the days of limited computing power available during the Apollo
missions. Though computing power is not such an issue now, the computational
efficiency is attractive from an allocation viewpoint because the algorithm is concise
and only requires input from the previous day. This means the filter does not increase
effort required to perform allocation re-runs, etc. It is analogous to any other balance
that is carried forward from one day to the next in allocation systems, e.g. pipeline
stocks, gas substitution accounts, etc.
4 FEASIBILITY WITH SIMPLIFIED EXAMPLES
4.1 Introduction
A simplified theoretical process model is utilised in this section, in which the true
well production can be modelled. Random process noise is introduced to reflect a
more realistic process. Meter uncertainties are then applied to mimic the real world
measurements and well estimates generated based on the Kalman filter equations. The
results of the Kalman filter will be affected by the measurement noise but its
performance can be determined by comparing the results with the known true
underlying well production. In particular, the Kalman filter estimates can be
compared against more conventional well test allocation approaches and hence an
assessment of the filter’s efficacy assessed.
4.2 Description of System
There are assumed to be three wells, labelled Apollo (), Gemini (γ) and Mercury (μ)
that are normally commingled in a main process but individual wells can be tested in
a dedicated test separator. The process is illustrated schematically in Figure 1:
6
Figure 1 – Simplified Model Process Schematic
Test Separator
M
Main Process
M Oil Export
Apollo
Gemini
Mercury
In this idealised theoretical model it is possible to know the true production potential
of each well as it declines with time. This is modelled assuming a simple exponential
decline curve and the potential from one day to the next changes according to:
𝑞,𝑑 = 𝑒 −𝑏𝛼∆ 𝑞,𝑑−1 (2)
Where,
b Well Apollo’s exponential decline constant

∆ Day step interval (=1 day).
In order to introduce more realistic noise into the system the decline well flow is
randomly varied in accordance with a Gaussian random variable with an uncertainty
equal to 1% of the flow. This is to reflect other unknown process effects that affect
the flow but which cannot be modelled.
The well uptime is also modelled simplistically in that a well can be shut in or
producing at maximum rate, i.e. the fractional daily uptime can only be 0 or 1. It was
assumed that there was a 10% chance that an individual well would be shut in on any
day.
It is acknowledged that when a well is producing for less than 24 hours but not shut in
for the whole day, the fractional daily uptime introduces an extra level of uncertainty
since production hours themselves will be estimates. The uncertainty in the flow is
exacerbated because wells do not attain full production immediately on start up. This
issue is discussed further in Section 5 when the Kalman filter is applied to real data.
Hence, the true production from each well can be modelled on a day by day basis and
the combined flow calculated as the sum of daily well production. Ignoring
7
processing effects (shrinkage) this summated production represents the true measured
daily oil production from the process.
In addition each well is tested at fixed intervals, although the interval is different for
each well.
In reality, the measured total production and well test measurements would include
measurement uncertainties and these were assumed to be 1% and 20% respectively.
4.3 Simplified Model Results
The calculations were run for a 50 day period. The results from the simplified model
are presented in Figure 2, Figure 3 and Figure 4 for Apollo, Gemini and Mercury,
respectively:
Figure 2 – Simplified Model Apollo Well
8
Figure 3 – Simplified Model Gemini Well
Figure 4 – Simplified Model Mercury Well
9
The black line is the true modelled flow from each well which declines exponentially
with some daily variation due to other random process noise.
The blue diamonds are the well tests which include measurement uncertainty.
The red line is the amount that would be allocated to wells based on the latest well
tests allocated in a pro rata fashion – which is a typical approach. The green line
shows the results obtained from the Kalman filter.
These plots are intended to illustrate some of the features that were anticipated in
Section 3.2. As can be observed, the Kalman filter predicts the well flows and
associated potentials better than the simple pro rata approach. Use of the commingled
meter production to inform the individual well rates in the Kalman filter mitigates the
impact of relatively high well test uncertainties which distort the pro rata results.
With the well test based pro rata approach, a poor well test for one well will have a
deleterious impact on the other wells, as can be observed on Day 22 when the Apollo
well test significantly under-measured the flow and the Gemini and Mercury wells
saw their allocated oil increase in a step wise fashion. These particular relative
distortions persist until further well tests are conducted.
Also, the days when wells shut in reveal further information about the individual well
rates and this is most noticeably observed on Day 37 when Gemini shuts in. In the
Kalman filter this results in the other two wells’ oil flows being corrected even after
Gemini restarts. This is not the case for the pro rata approach which continues to
estimate the Apollo and Gemini rates poorly.
These plots were generated randomly but a specific example was selected to illustrate
the features discussed. A more meaningful analysis is to perform a whole series of 50
day allocation runs allowing the various input parameters to vary randomly in each
Monte Carlo trial and continuing with the random daily fluctuations also within the
trial. This analysis is presented in the next section.
4.4 Monte Carlo Analysis
In the analysis, each well’s initial flow was independently, randomly varied between
50 and 200 tonnes per day2. The well decline factor (b) was similarly randomly varied
between 0.00 and 0.03 for each well independently. Both the initial production and
well decline factor were held constant for the duration of each Monte Carlo trial of 50
days.
Similarly, the assumed process noise for the Kalman filter was varied randomly
between 0% and 10% (i.e. normally significantly over estimating the true noise value
of 1%) between trials.
2
This is an arbitrary range and the designation of tonnes per day is not of special
significance.
10
The test interval was varied randomly between 10 and 30 days, independently for
each well but the intervals were held constant for the duration of a single Monte Carlo
trial.
1,000 trials were run in the Monte Carlo simulation and the absolute difference, or
mis-allocation, between the estimated and true production from each well was
recorded each day for each method. These daily absolute differences were then
summed over the 50 days for each trial. The averages of these 50 day mis-allocations
from the 1,000 trials for each method are compared in Figure 5 for all three wells.
Figure 5 – Average 50 Day Total Well Mis-allocation

from 1,000 Monte Carlo trials
This illustrates the Kalman filter estimates to be significantly more accurate than the
simple well test based approach: the average mis-allocation is reduced by
approximately 75%.
The above analysis illustrates the potential benefits of the application of the Kalman
filter to well flow estimation.
The theoretical environment allows us to assess the efficacy of the Kalman filter in a
unique way in that we do know the true values of the variables we are estimating.
Significant effort has been taken to try and account for all foreseeable issues that may
confound the estimation, for example - accounting for the fact that the Kalman model
does not assume the correct initial well flow, the decline factor or process noise. This
has been accounted for by randomly varying these assumed values (around the true
11
values) between the Monte Carlo trials. However, this is still a relatively idealised
environment and though many variables have been accounted for and their influence
tested, the theoretical model may not necessarily include the possibilities of unknown
process noise, gross errors, human error, etc. The next section addresses this by
applying the Kalman filter to real data.
5 APPLICATION TO REAL DATA
5.1 Introduction
The chaos and noise of real world data presents a more formidable test of the Kalman
filter than any theoretical data in assessing the feasibility and robustness of the
method.
The problem with real data is that we can never know the true values of well flows we
are estimating and the performance of the filter has to be assessed by indirect means.
This has been accomplished by: some direct numerical estimations of performance, by
comparison with the simple well based pro rata approach and the use of engineering
judgement.
5.2 Description of System
A schematic of the process and well configuration is presented in Figure 6:
Figure 6 – Simplified Schematic of Process
} Compression and
Gas Export
HP Separator
LP Separator
M Oil Export
M
}
Further wells
Well A
Well B
Well C
Further wells }
12
The process handles production from five wells (labelled A to E, though only the first
three are shown in the schematic). The oil and gas products are fiscally metered.
There is a single subsea multiphase flow meter (MPFM), through which the wells can
be individually routed to test their performance.
As shown in Figure 6, the well stream fluids are commingled upstream of the 1st stage
separator. Oil separation is achieved using two-stage separation and gas from
separation is sent via compression trains to fuel and export.
Though real, the data is anonymised and several years old. The period examined
covers just over 150 days (5 months) of production. Only the oil is considered in the
following analysis.
5.3 Kalman Filter Process Model
The assumed model for the real system is similar in many respects to that described
for the theoretical model presented in Section 4. However, the exponential decline
was not applied to the real wells as they were choked back during the period
examined. In fact the choke positions were varied daily and significantly so between
well tests.
It was found by an analysis of the well test data that the well flows varied roughly
linearly with the choke position. The choke position was the dominant variable in
determining flow. The well rates appeared much less well correlated with other
variables such as bottom-hole pressure, etc.
The effect of the choke changes was therefore incorporated into the process model
and the change in potential from one day to the next was calculated according to:
𝑐ℎ𝑤,𝑑
𝑞𝑤,𝑑 = 𝑞 (3)
𝑐ℎ𝑤,𝑑−1 𝑤,𝑑−1
Where,
chw,d Choke opening position, for well, w, on day, d.
w represents any of the wells, A to E. In effect, the well potential, (i.e. the state
variable), is determined as the flow at the new choke position.
5.4 Estimated Process Noise
The process noise has to be estimated using engineering judgement to some extent,
though it is adjusted as part of the tuning process, discussed in Section 5.7. A value of
10% was assumed as the baseline uncertainty in the model. This was further
increased in proportion to the choke change (up to maximum of 50%), to reflect the
additional uncertainty introduced by the choke changes and the assumption of a linear
dependence on choke position.
13
These values were determined after some trial and error. In fact, the performance of
the filter was relatively insensitive to the process noise uncertainties unless they were
extremely low (<1%) or high (>100%).
5.5 Measurement Model
Well Tests
The well test oil represents a direct measurement of the oil potential.
𝑞𝑤,𝑑 = 𝑡𝑤,𝑑 (4)
Where,
tw,d Well test rate (24 hour basis) for well, w, on day, d.
Product Oil Measurement
The product oil is a measurement of the sum of the producing wells’ flow, after
allowing for shrinkage:
∑ 𝑞𝑤,𝑑 ∗ 𝑥𝑤,𝑑 ∗ 𝑠ℎ𝑤,𝑑 = 𝑚𝑑 (5)

𝑤
Where,
shw,d Oil shrinkage from well test to export for well, w, on day, d.
That is, the sum of each well’s potential from the update step in Equation (3),
multiplied by their fractional up time for the day (i.e. hours producing/24), multiplied
by the shrinkage from well test to export product oil conditions, should be equal to the
total measured product oil on the day.
5.6 Measurement Uncertainties
The oil product and sub-sea multiphase well test flow meter measurement
uncertainties are presented in Table 1:
Table 1 –Well MPFM and Oil Product Uncertainties
Uncertainty Uncertainty
Meter
type (±%)
Oil Export Meter relative 1%
MPFM Water Liquid Ratio (WLR) absolute 3%
MPFM Liquid Flow relative 3%
14
The relative uncertainty in the oil export meter (from Table 1) was assumed to be
1%. This was based on a nominal uncertainty of 0.5% for the fiscal meter but
degraded to reflect the presence of a low, but still significant, water content.
The quoted MPFM uncertainties are typical nominal values appropriate for the type of
meter installed. For an MPFM, the dry oil flow uncertainty is a function of the
measured liquid and WLR and their associated uncertainties and the relative
uncertainty in the oil flow is given by:
2
√(𝑒𝑡,𝑙 (1 − 𝑊𝐿𝑅)) + (𝜀𝑊𝐿𝑅 )2
(6)
𝑒𝑡,𝑜 =
(1 − 𝑊𝐿𝑅)
Where,
et,l Relative uncertainty in measured liquid flow rate

et,o Relative uncertainty in oil flow rate
WLR Water liquid ratio
WLR Absolute uncertainty in measured WLR.
The above equation was presented in [7] and derived using the approach described in
the GUM [8], termed Taylor Series Method (TSM), which is used to model the
propagation of uncertainties.
Though the meter was used to measure the flow from a number of different wells,
they were all produced from the same reservoir which was at a pressure above the
bubble point. Hence, the wellstream composition for all wells should remain
ostensibly constant. Similarly, the produced water composition should also remain
stable.
The measured oil in the MPFM also undergoes some shrinkage from the point of
measurement to export. Monte Carlo simulation (described in a Supplement to the
GUM [9]) of the process found the shrinkage to be 0.928 on average with an
uncertainty of 3%. The approach used in [6] was used to obtain the average
shrinkage and associated uncertainty.
5.7 Kalman Gain Tuning
It is unlikely that the assumed uncertainty for the process noise is entirely accurate. It
is also unlikely that the product oil and MPFM measurement uncertainties are wholly
precise. This issue is well recognised in the implementation of Kalman filters in real
world applications and such filters are termed sub-optimal.
The behaviour of the filter is determined by the ratio of the process to measurement
uncertainties in the form of the Kalman gain. Hence, the filter is less sensitive to the
accuracy of the uncertainties so long as their values relative to each other are
representative. This was accomplished in this application by adjusting the process
noise variance, in effect tuning the Kalman gain. This tuning was optimised by
15
observing the behaviour of the filter in estimating the wells’ future test rates and the
daily produced oil.
The tuning process is an attempt to optimise the sub-optimal filter to render it as close
as possible to a truly optimal filter. Sub-optimal filters, Kalman gain tuning and
practical implementation considerations have been thoroughly analysed in the
literature and are discussed for example in [4].
5.8 Results
Figure 7 is a plot of the predicted total oil from the Predict Step in the Kalman filter
(Qd│d-1) versus the actual measured rate, i.e. prior to the reconciliation of the
potentials against the oil export meter on that day.
Figure 7 – Kalman Predicted versus Measured Oil
As can be observed the Kalman totals are a close match with the metered oil. To put
this in context, compare the total predicted oil from the well test based pro rata
approach in Figure 8:
16
Figure 8 – Well Test Based Predicted Total Oil versus Measured Oil
Looking at the well test based estimates, the match is not as good and there are clear
periods when the well tests appear to be significantly under- or over-estimating
production. The diamonds (plotted against the right hand axis) indicate the number of
well tests conducted each day. The periods of drift in the estimated oil production do
appear to coincide roughly with periods when the well test frequency is low.
The advantage the Kalman filter offers is its ability to exploit the additional
information provided by each day’s measured product oil.
This difference in predictive ability was calculated from the absolute difference in the
predicted and the measured oil for the two methods over the 153 day period. The
cumulative absolute differences are plotted in Figure 9:
17
Figure 9 – Cumulative Absolute Difference Predicted versus Measured Oil
At the end of the period the Kalman filter cumulative difference is 42% lower than the
well test pro rata value.
A second metric with which to compare the methods is to determine the absolute
difference between each well’s predicted flow just prior to a well test and the
measured well test flow. This is plotted in Figure 10:
Figure 10 – Cumulative Absolute Difference Predicted versus Measured Well

Test Oil
18
At the end of the period the Kalman filter cumulative difference is 36% lower than the
well test pro rata value for the well test data.
The following figures examine the results of the Kalman filter for individual wells.
Figure 11 is a plot of the Kalman estimated potentials for Well A:
Figure 11 – Well A: Kalman Oil Potentials
A B
The orange line plots the Kalman filter estimate of the well’s potential. The blue
diamonds are well test rates which have error bars to indicate their measurement
uncertainty. The grey lines, labelled UCL (upper confidence limit) and LCL (lower
confidence limit), represent the  uncertainties in the Kalman potentials, which are
generated from the covariance matrix (Pd│d). As can be observed, the uncertainty rises
as the time elapsed since the last well test increases, which was one of the objectives
presented in Section 3.2. It also reduces when a well test is conducted as the
confidence in the potential increases when a direct measurement is made.
An attractive feature of the Kalman filter is along the evolution of the potential, it
predicts changes in potential prior to a well test taking place. This is most obviously
illustrated at the points labelled A and B. Around point A, the predicted potential is
varying but falls smoothly to agree with the well test on Day 96. Even more markedly
at point B, Well A’s predicted potential falls from around 2,000 tonnes/d to 800
tonnes/d on Day 146 (due to a choke change – see Figure 12), the new reduced rate is
then confirmed by the well test on Day 152.
19
There is considerable variability in the potential and this is largely driven by changes
in the well choke position. The changes in flow roughly correspond with the changes
in potential as illustrated in Figure 12:
Figure 12 – Well A: Potential and Choke Position
The potential is plotted in orange, identical to that in Figure 11, but the choke %
opening is also plotted in purple against the right hand axis.
Similar charts for all the wells are presented in the Section 10 Appendix of Figures,
where similar features for the other wells can be observed.
The Kalman potential for Well E, as plotted in Figure 13, is worthy of further
comment however:
20
Figure 13 – Well E: Kalman Oil Potentials
An interesting feature of Well E is that it is not tested after Day 41 until Day 153,
when there is an extremely significant drop in its potential from 1,024 tonnes/d to 123
tonnes/d as indicated by C.
Examination of its choke position in Figure 14 shows that the variation in its flow
generally reflects its choke position:
21
However, this is not borne out for the day (Day 153) of the well test when the choke
position remained the same as the previous day and the flow of all producing wells
had been steady. In fact this drop leads us to suspect the well test itself, especially
when the next test on well E, performed less than two weeks later, measured an oil
rate of 1,174 tonnes/d.
The Kalman filter provides information on the uncertainty in the estimates which,
when coupled with the measurement data, allows statistical tests to be performed on
the data. One such test is the Global Test which considers the mass imbalances
(residuals in the constraint equations) in the system and determines statistically
whether these are within the range expected given the measurement and well potential
uncertainties. The mathematics of the test are presented in Section 7.4. The Global
Test identifies the potential presence of gross errors but does not identify the source of
the errors.
The statistic is plotted in Figure 15:
22
Figure 15 – Global Test Statistic
The black line is a critical value of the statistic, which is dependent on the number of
measurements made each day. The circles indicate the calculated daily Global
statistic: blue are below the critical value and red are above it indicating the potential
presence of gross errors in the data.
The Global test statistic is above the critical value on Day 153, indicating there is
potentially a problem and perhaps confirming our suspicion of the test on Well E.
In fact, there are a number of other days where the critical test statistic is exceeded
and these days have been examined on a day by day basis and ignored if they are
deemed to corrupt the estimates. These days when gross errors occur are generally
when wells restart after being shut in, as was alluded to previously in Section 4.2).
6 CONCLUSIONS
The Kalman filter brings new dimensions to estimation techniques. It utilises not only
all the available measurement data, but also employs a process model to enable
estimates to be propagated from one day to the next – hence prior information gained
is retained. It performs this in a statistically near optimal manner by utilising the
uncertainty in the measurements and the process model.
The Kalman filter has been demonstrated to be a feasible tool to optimise the
estimation of well potentials using theoretical models. In addition, Kalman filter has
been successfully applied to a real system and produces improved estimates of oil
potentials compared with simply using well test data.
23
7 MATHEMATICAL DEVELOPMENT OF THE USE OF A KALMAN

FILTER APPROACH TO OPTIMAL WELL FLOW ESTIMATES
7.1 Introduction
The mathematical development presented in this section is based on the simplified

theoretical system discussed in Section 4. A similar development is applicable to the
real world data in Section 5, the principal difference being the elements of the phase
transition matrix which are based on change in choke position rather than well
decline.
Notation is presented in Section 8.
7.2 Predict Phase of Filter
In first phase of the filter, the state transition matrix Fd is used to estimate the updated
well flow q (on day d) from the previously estimated flow (on day d-1) in Equation
(7). This is in effect the model estimated update and, for the case of the three wells,
this is given by the exponential decline equation.
𝑄𝑑│𝑑−1 = 𝐹𝑑 𝑄𝑑−1│𝑑−1 (7)
More explicitly,
𝑞,𝑑│𝑑−1 𝑞,𝑑−1│𝑑−1
𝑒 −𝑏𝛼∆ 0 0
[𝑞𝛽,𝑑│𝑑−1 ] = [ 0 𝑒 −𝑏𝛽∆ 0 ] [𝑞𝛽,𝑑−1│𝑑−1 ] (8)
𝑞𝛾,𝑑│𝑑−1 0 0 𝑒 −𝑏𝛾∆ 𝑞𝛾,𝑑−1│𝑑−1
Similarly the predicted variance (directly related to the uncertainty) of the well flows
is calculated from:
𝑃𝑑│𝑑−1 = 𝐹𝑑 𝑃𝑑−1│𝑑−1 𝐹𝑑𝑇 +𝑁𝑑 (9)
Ft remains fixed and initially P has estimates of the uncertainty in the well flows,
which can be set somewhat arbitrarily (safer to assume a high value).
2
𝜀
( 𝛼,0⁄2) 0 0
2
𝑃𝑑=0 = 𝜀
0 ( 𝛽,0⁄2) 0 (10)
2
𝜀
0 0 ( 𝛾,0⁄2)
[ ]
24
Initially, the P matrix only has diagonal elements, but as it is updated, the off-
diagonal co-variance terms become populated. Nd is the process noise matrix.
7.3 Update Phase of Filter
In the second phase of the filter any measurements are taken into account. The
residual vector Zd (in effect the mass balances) is calculated from:
𝑍𝑑 = 𝑌𝑑 − 𝐻𝑑 𝑄𝑑│𝑑−1 (11)
The observation matrix Hd, relates the measurements to the well rates. There are two
sets of measurements, the total oil production measured daily (i.e. every update) and
individual well tests which occur at intervals (possibly irregular). On a day without
any well tests and all wells flowing:
0 0 0
𝐻𝑑 = [ ] (12)
1 1 1
The top row relates the wells to measured well tests. On this day there aren’t any, so
all zeroes are entered. The second row relates the wells to the total measured flow; on
this day all wells are flowing and are added to produce the total flow. If a well isn’t
flowing then a zero would be correspondingly entered in the second row. For
example, in (13),  well is being tested and γ is not producing:
1 0 0
𝐻𝑑 = [ ] (13)
1 1 0
The vector Yd has two elements, the top corresponding to the measured well test rate
and the bottom the total produced oil measured rate:
𝑡
𝑌𝑑 = [𝑚𝑑 ] (14)
𝑑
With this arrangement it is assumed that only one well is tested on any day. If there
are multiple tests on the same day then Hd and Yd have additional rows.
The residual matrix variance, Sd, is calculated using the variances in the well rate
estimates from the predict phase and the variance in the measurements (obtained from
the well test and produced oil measurement uncertainties):
𝑆𝑑 = 𝐻𝑑 𝑃𝑑│𝑑−1 𝐻𝑑𝑇 +𝛴𝑑 (15)
This is equivalent to the calculation of the propagation of uncertainties described in

the GUM [8] but expressed in matrix notation. Σd is given by:
25
2
𝜀
( 𝑡,𝑑⁄2)
𝛴𝑑 = 2 (16)
𝜀
( 𝑚,𝑑⁄2)
[ ]
The updated state estimation vector of well flows (i.e., the solution for day d) is
calculated in the final innovation step by the filter:
𝑄𝑑│𝑑 = 𝑄𝑑│𝑑−1 +𝐾𝑑 𝑍𝑑 (17)
Where Kd is the Kalman gain matrix and is calculated, using the results from (9) and
(15), according to:
𝐾𝑑 = 𝑃𝑑│𝑑−1 𝐻𝑑𝑇 𝑆𝑑−1 (18)
In essence, this incorporates the uncertainties in the model estimates and any
measurements in a statistically optimal way.
Finally, the updated well flow estimate variance matrix is calculated (i.e. the variance
associated with Qd│d obtained in (17)):
𝑃𝑑│𝑑 = (𝐼 − 𝐾𝑑 𝐻𝑑 )𝑃𝑑│𝑑−1 (19)
Qd│d and Pd│d in (17) and (19) now become Qd-1│d-1 and Pd-1│d-1 in equations (7) and
(9) on the next day.
7.4 Global Test for Gross Errors
The Global Test, described in [10], is used to test if there are any gross errors in the
data. It does not identify the source(s) of each error, simply their presence. The Global
Test essentially considers whether the observed mass balance deviation is within the
uncertainty that would be expected given the uncertainties in the variables and
measurements used to compute it. In the context of the Kalman filter this means using
the residuals matrix Zd from (11) and its associated variance Sd calculated in (15). The
Global Test statistic γ, is computed from:
𝛾 = 𝑍𝑑𝑇 𝑆𝑑−1 𝑍𝑑 (20)
Under the null hypothesis, that the data does not include any gross errors, this statistic
follows a χ2 distribution with ν degrees of freedom, where ν is the number of
measurements. Each day there is at least one measurement - specifically, the produced
oil, plus any well tests on that day. Hence, if the calculated value of γ is greater than
this figure, the hypothesis that the data do not contain gross errors is rejected.
26
8 NOTATION
Generally, upper case letters have been used to represent vectors and matrices, whilst
lower case has been used to denote individual variables. Symbols in the general text
(i.e. outside if a numbered equation) have been emboldened.
a Oil allocation for well x Fractional uptime (i.e.

b Well exponential decline hours on production24)
constant Y Measurement vector
ch Choke opening position Z Measurement residual
d Day vector
e Uncertainty, relative
F State transition matrix Greek
H Observation matrix
I Identity matrix γ Global Test statistic
K Kalman gain matrix ∆ Day step interval
m Produced oil measurement  Uncertainty, absolute
N Process noise variance Σ Measurement variance
matrix matrix
Pd-1│d-1 Matrix of well flow ν Degrees of freedom
estimate uncertainties on χ Chi statistic
day, d-1
Pd│d-1 Predicted matrix of well Subscripts
flow estimate uncertainties
on day, d d Day number
Pd│d Final matrix of well flow l Liquid
estimate uncertainties on m Measured product oil
day, d o Oil
Qd-1│d-1 Vector of well flow t Well test
estimates on day, d-1 w Well
Qd│d-1 Predicted vector of well WLR Water liquid ratio
flow estimates on day, d  Well  (Apollo)
Qd│d Final vector of well flow γ Well γ (Gemini)
estimates on day, d μ Well μ (Mercury)
q Daily flow potential
estimate for well Superscripts
sh Oil shrinkage from well T
test to export Matrix transpose
-1
S Residuals variance matrix Matrix inverse
t Well test flow
measurement
WLR Water liquid ratio
27
9 REFERENCES
[1] UKCS Maximising Recovery Review: Final Report, 24 February 2014,

Sir Ian Wood.
[2] Kalman, R.E. (1960). "A new approach to linear filtering and
prediction problems". Journal of Basic Engineering 82 (1): 35–45.
[3] Discovery of the Kalman Filter as a Practical Tool for Aerospace and
Industry, Leonard A. McGee and Stanley F. Smith, NASA Technical
Memorandum 86847, Ames Research Center, November 1985.
[4] Applied Optimal Estimation, Arthur Gelb, 1974, The Analytical
Sciences Corporation, ISBN 0262570483.
[5] Proceedings of the Production and Upstream Flow Measurement
Workshop (The Americas’ Workshop) 12-14 February 2008,
“Experiences in the Use of Uncertainty Based Allocation in a North
Sea Offshore Oil Allocation System”, Phillip Stockton, Alan Spence.
[6] Proceedings of the 32nd International North Sea Flow Measurement
Workshop, 21-24 October 2014, “Process Simulation Uncertainties”,
Phil Stockton and Juan Martin, Accord Energy Solutions Limited.
[7] Allocation Uncertainty: Tips, Tricks and Pitfalls, Phil Stockton and
Allan Wilson, Proceedings of the 30th International North Sea Flow
Measurement Workshop, 23-26 October, 2012.
[8] Guide to the Expression of Uncertainty in Measurement, International
Organisation for Standardisation, ISO/IEC Guide 98:1995.
[9] Joint Committee for Guides in Metrology (JCGM), “Evaluation of
Measurement Data – Supplement 1 to the ‘Guide to the Expression of
Uncertainty in Measurement’ – Propagation of Distributions Using and
Monte Carlo Method” JCGM 101: 2008, France 2008.
[10] Data Reconciliation and Gross Error Detection, An Intelligent Use of
Process Data, Shankar Narasimhan and Cornelius Jordache, published
in 2000 by Gulf Publishing Company, Houston Texas, ISBN 0-88415-
255-3.
28
10 APPENDIX OF FIGURES
Figure 16 – Well A: Kalman Oil Potentials
Figure 17 – Well A: Potential and Choke Position
29
Figure 18 – Well B: Kalman Oil Potentials
Figure 19 – Well B: Potential and Choke Position
30
Figure 20 – Well C: Kalman Oil Potentials
Figure 21 – Well C: Potential and Choke Position
31
Figure 22 – Well D: Kalman Oil Potentials
Figure 23 – Well D: Potential and Choke Position
32
Figure 25 – Well E: Potential and Choke Position
33
33rd International North Sea Flow Measurement Workshop, 20 th-23rd October 2015, Tønsberg, Norway
Multiphase flow quantification using

Computational Fluid Dynamics and Magnetic Resonance Imaging
Susithra Lakshmanan1, 2, Daniel Holland3, Andy Sederman2, Mikhail Gurevich4 and Wes Maru 1‡
1
Oil &Gas Measurement Limited, OGH, Ely, Cambridgeshire, UK
2
Department of Chemical Engineering & Biotechnology, University of Cambridge, UK
3
Department of Chemical & Process Engineering, University of Canterbury, Christchurch, NZ
4
IMS Group, Moscow, Russia
Abstract
We investigated oil-water flow primarily with a view to quantify the mixing efficiency of a
newly developed device with a Liquid Jet In Cross Flow (LJICF) configuration but also to
predict the resulting flow regime and water droplet size distribution. Such devices are utilised
in important applications such as liquid sampling in custody transfer, where the degree of
homogeneity of the mixture affects the accuracy of the representative sample withdrawn and
hence the oil transaction between producers and operators as well as taxation by
governments. Despite their importance however, liquid-liquid flows under LJICF
configurations have not featured as prominently in the literature as those of gas-liquid flows,
resulting in a significant knowledge gap that needs to be addressed.
As part of a wide R&D programme at OGH, four experimental and calibration facilities are
developed to conduct investigations in the general area of multiphase flows in a synergistic
manner. These facilities are composed of: (a) small multiphase flow loop (SMPFL) primarily
designed to fit in the University of Cambridge’s Magnetic Resonance Imaging unit; (b) large
multiphase flow loop (LMPFL), which is a scale up version of the SMPFL, (c) Flow Meter
Calibration Loop (FMCL) to calibrate the flow meters for both the SMPFL and LMPFL, and
(d) Heterogeneous High Performance Computing (HHPC) facility, where it is used to
conduct and validate numerical experiments for any flow of interest based on new models
developed in-house using the OpenFOAM® computational platform.
We also developed Magnetic Resonance (MR) techniques to be used in conjunction with the
SMPFL. To quantify the water-cut, a chemically shift selective (CHESS) MR pulse sequence
was developed. The CHESS technique saturates the unwanted signal from the oil component
and only excites the water resonance. To verify that this sequence was quantitative, various
water-oil mixture phantoms were prepared with water-cuts between 2.5% and 25%. By
completely suppressing the oil spectrum peak, the water content can be readily quantified by
integrating the signal intensity at each water-cut, where a quantitative comparison of the
integrated signal intensity as a function of the known water-cut for the samples showed a
linear relationship and demonstrate that the water cut is measured with an accuracy of
approximately ± 0.2 %. The CHESS sequence was then combined with imaging sequences to
enable visualisation of the water distribution in 2D. The developed technique is then used to
study liquid-liquid flows and to characterise a novel mixing device that is developed by our
partners to homogenise oil-water flows for custody transfer applications. The experimental
results will be used to create accurate correlations as well as to validate our CFD simulation
tools that are developed in-house on the OpenFOAM® computational platform.
‡
Corresponding author email address: wes.maru@oghl.co.uk
1
33rd International North Sea Flow Measurement Workshop, 20th-23rd October 2015, Tønsberg, Norway
1 Introduction ........................................................................................................................ 2
1.1 Automatic pipeline sampling ................................................................................................... 3
1.2 Review of liquid-liquid flow diagnostics in pipes ..................................................................... 3
1.3 Jet mixing and droplet dispersion............................................................................................ 4
2 Synergistic physical and numerical experimentation ......................................................... 6
2.1 SMPFL construction and operation ......................................................................................... 7
2.2 Magnetic resonance technique ............................................................................................... 9
2.3 Computational Fluid Dynamics Simulations .......................................................................... 10
3 Results and discussion ...................................................................................................... 11
3.1 Design of experiments .......................................................................................................... 11
3.2 Experiment using the SMPFL without MRI ........................................................................... 12
3.3 Numerical experiment using the HHPC facility ..................................................................... 14
3.4 MR Measurements of static samples .................................................................................... 15
3.5 MR Measurements during flowing conditions ....................................................................... 17
4 Conclusion ........................................................................................................................ 22
5 Acknowledgement ............................................................................................................ 22
6 References ........................................................................................................................ 23
1 Introduction
Multiphase flows are very complex but also ubiquitous both in nature and in industry.
Typically, multiphase flows occur in at least two-phases in the form of gas-liquid, gas-solid,
liquid-liquid, liquid-solid or a combination of gas-liquid-solid flows. The complexity of such
flows could be further compounded where any of the gas, liquid or solid phases each manifest
as multicomponent fluids. For example, in the oil and gas industry, multiphase flows of oil-
water or gas-oil-water with the presence of sand in either of the cases are very common.
Therefore, in the oil and gas industry, there is significant interest to understand multiphase
flows not only to harness and efficiently utilise our resource to maintain the energy needs and
economic development of current civilisation but also to develop smart solutions to maintain
the health of our environment. To that end, we undertook the development of holistic
approaches and integrated facilities that may allow us to predict, verify, quantify and control
the evolution of multiphase flows. The current investigation is part of this large scale
program but only considers liquid-liquid (or oil-water) flows, which have significant
commercial importance in fiscal, allocation and custody transfer sampling applications. For
example, in new oil-fields, the proportion of water-cut is relatively low (usually < 5%) while
the water cut increases significantly as production declines and oil-fields mature. In both
situations, it is important to quantify the water-cut relatively accurately.
The industry guide for the mixing, sampling, analysis (or calculation) and certification of a
device to quantify the water-cut is described in the ISO 3171, API 8.2 and/or IP 6.2 (or IP
476:2002) standards - hereafter called Automatic Pipeline Sampling Standards (APS
Standards). One of the key drivers to create and withdraw a representative sample so that the
water-cut measurement will be accurate is homogenisation of the oil-water mixtures.
Because, errors arising from poor mixed-ness could cost both industry and government
millions of dollars a day in operational costs and revenue as well as in lost taxes. While our
2
main objective is to investigate a newly developed LJICF mixing system by our partners
according to the APS Standards, we also took the liberty to posit the requirements for
homogeneous mixing by focusing on the mechanistic behaviour of multiphase turbulence.
Therefore, we will briefly address the practical issues of pipeline sampling from a point of jet
break up and droplet distribution under non-steady flow conditions.
1.1 Automatic pipeline sampling
The APS standards provide a general guide covering all the essential steps of automatic
pipeline sampling from the mixing process up to the sample analysis. The procedures for the
sampling system including how to quantify the uncertainty of the measurements and its
limitations are also provided. The sampling system certification proving test, which operates
in a controlled or steady flow condition utilises a water injection procedure and is
accompanied by the entire sequence of sampling as stipulated in the APS standards.
Recently, Potten and Wright (2014) further elucidated the subject by providing historical data
for a generation of their mixing systems. However, in real field operations, the flow is non-
steady and so is the water cut. In this type of dispersed phase intermittency, although the type
and frequency of the sampling equipment is believed to be crucial to gain an overall accurate
result, this theory must be tested against experiment. Therefore, a review on the mechanistic
behaviour of the jet mixing and droplet dispersion process is crucial as it will have significant
effect on the mixing efficiency of the device to be used for custody transfer application.
1.2 Review of liquid-liquid flow diagnostics in pipes
Various techniques are employed by investigators to capture the behaviour of oil-water flows
in pipes. A brief review on the most relevant techniques was provided by Lakshmanan et al
(2015), which include visual observation, impedance probes, conductivity probes, particle
imaging velocimetry (PIV), 𝛾-ray Computer Tomography (CT), x-ray CT, wire mesh sensors
and hot-film anemometers (Ismail et al. 2005; Xu 2007; Bieberle et al. 2013; Wang 2015).
Magnetic resonance (MR) imaging is also proving to be a useful tool for characterising
single- and two-phase flow in pipes (Fukushima 1999; Elkins & Alley 2007; Gladden &
Sederman 2013). MR has several advantages over the above techniques; it is completely
non-invasive, can image optically opaque systems and can measure parameters including
concentration, velocity, and diffusion. It is also possible to resolve each of these parameters
spatially in one, two or three dimensions. However, there are some limitations of MR
including a restricted sample geometry which is imposed by the need for a strong
homogeneous magnetic field, an inability to image magnetic materials and difficulties
associated with very heterogeneous materials (Fukushima 1999). MR techniques to study two
phase flow include fast imaging techniques such as FLASH (Haase et al. 1986), RARE
(Hennig et al. 1986), SPRITE (Sankey et al. 2009) and spiral EPI (Tayler et al. 2011). These
techniques can be combined with velocity encoding to resolve the flow field (Tayler et al.
2012). When imaging liquid-liquid flows, it is important to resolve the signal from each
phase independently. In this situation, chemical shift imaging (Maudsley et al. 1983) and
chemical shift selective imaging (CHESS) (Götz & Zick 2003) are used. The acquisition
time of these techniques can be reduced using advanced signal processing techniques like
3
compressed sensing (Holland & Gladden 2014). In this investigation, MR measurement is

done using the SMPFL. However, in the section to follow, we first examine the mechanistic
processes that may be responsible in achieving homogeneous mixing of liquid-liquid flows
under LJICF configuration.
1.3 Jet mixing and droplet dispersion
LJICF configuration is one of the most common and efficient way of breaking up a liquid jet
in to uniform droplets to create homogeneously dispersed flow in a wide range of operating
conditions. Common examples and advanced applications of LJICF (or JICF in general) are
smokestacks, volcanic plumes, atmospheric dispersion, oil spills, liquid fuel injection in
scramjets and turbine blade film cooling. These have been investigated by many researchers
(Clark 1964; Geery & Margetts 1969; Holdeman 1993; Mahesh 2012). Most studies of LJICF
focus to predict and gain understanding of the jet trajectory but the subsequent jet break-up
and the dispersion of droplets in the cross flow remains very complex and unexplored.
Recently, oil-water homogenisers that are used for custody transfer applications typically use
LJICF configuration, often using multiple jets (Fernando & Lenn 1990). It is assumed, the
trajectory of a single jet describes the extent to which the jet penetrates into the crossflow and
also provides an estimate on how the droplets are dispersed with the flow. If sufficient
turbulence is created, it gives rise to homogeneous mixing (Mahesh, 2012). However, there
are various ways of defining and scaling the jet trajectory, mainly depending on the maxima
of the local velocity, scalar mixing, or vorticity (Margason 1993; Smith & Mungal 1998;
Mahesh 2012). Many studies have focused on single jets. However, multiple jets ensure
better mixing and may also decrease the power requirements compared to if a single jet was
to be used (Fernando & Lenn 1990; Wang et al. 1999). Most investigations on LJICF have
been on gas-liquid flows and relatively little effort has been applied to study liquid-liquid
flows. However, there are significant differences between the behaviour of gas-liquid and
liquid-liquid flows under LJICF configuration. The liquid-liquid density ratio is two to three
orders of magnitude higher than those of gas-liquid systems. There is also a relatively narrow
band in the viscosity ratios of liquid-liquid systems than those of gas-liquid systems making
the viscous dissipation significant. As such, liquid-liquid systems are more difficult to draw
general conclusion compared to gas-liquid systems (Simmons and Azzopardi, 2001).
Furthermore, the continuous phase in liquid-liquid system is much denser and more viscous
than those of gas-liquid systems thereby resulting in slower settling velocity of droplets and
having a higher propensity of dispersion. But, care should be taken when comparing liquid-
liquid systems of varying densities and viscosities. In this case, there is a trade-off between
energy input, to create jet break up and dispersion, and droplet suspension time. Contrary to
the guidance from the APS standards, high viscosity and high density oils will provide longer
suspension time for water droplets, maintaining its mixed-ness provided the stream velocity
is not too low. Therefore, it is low density, low viscosity and low stream velocity conditions
that are the worst case scenario in pipeline mixing as pointed out by Potten and Wright
(2014). However, these authors didn’t provide evidence on the performance of their mixing
systems at very low stream velocities, as we will demonstrate later. However, to achieve the
optimal homogeneity, the LJICF configuration problem has to be posed as an exercise to
achieve a narrow droplet distribution that will remain suspended for a reasonable period but
with minimal energy input and without an opportunity to create permanent emulsion. Of
4
course, this will be valid for relatively low water cut or dilute dispersed phase situations. For
higher water cuts or dense dispersed phases, droplet collision and coalescence will
complicate the dynamics (Valle, 1998; Trallero, 1995; Simons and Azzopardi, 2001) as well
as the droplet size and its size distribution. The current jet trajectory formulations (Clark
1964; Geery & Margetts 1969; Holdeman 1993; Mahesh 2012) are empirical correlations for
gas-liquid systems and can’t be used for liquid-liquid systems under LJICF configurations. A
novel multi-jet homogeniser was design on the basis of a jet penetration model for a single jet
in a liquid-liquid system under LJICF configuration. In this model, the jet penetration was
characterised (Lakshmanan et al, 2015) by:
2
𝑥𝑛 𝑦𝑛 𝛽 𝐶𝐷 (Re𝐽∞ ) Ga𝐽 𝑦𝑛
= (𝑑 ) 𝑟 + [ − 2Re2 ] (𝑑 ) , (1)
𝑑𝐽 𝐽 𝐽𝑚 𝐽 𝐽
Where 𝑦𝑛 and 𝑥𝑛 represent the jet penetration into the cross flow and the jet trajectory length
along the stream-wise direction, respectively, CD is the drag coefficient due to a cylindrical
liquid jet in cross flow as a function of the Reynolds number ( ReJ∞ ), which is based on the
relative velocity of jet and stream, ReJ is the jet Reynolds number, GaJ = gd3J ⁄v 2 is the
Galileo number as a relative measure of the gravitational and viscous forces affecting droplet
settlement, Jm = ρJ u2J ⁄ρ∞ u2∞ is the important momentum flux ratio, with ρJ and ρ∞
representing the jet and stream densities, respectively; and r = uJ ⁄u∞ is the jet-to-stream
velocity ratio for non-stationary streams, β is a correlation parameter with a value ≥ 1. The
measurement and prediction of jet break up and droplet size distribution of both gas-liquid
and liquid-liquid systems under LJICF configuration is much more challenging (Fernando &
Lenn 1990; Wang et al. 1999; Simmons and Azzopardi, 2001; Mahesh 2012). Recently,
Bolszo et al (2014) considered a pure liquid and emulsion jet break up and droplet formation
under LJICF configuration and provided a relation of the form:
𝐵
𝐷32 𝑥
= 𝐴 (𝑑𝑛) 𝐽𝑚 𝐶 We𝐽 𝐷 Re𝐽∞ 𝐸 (2)
𝑑𝐽 𝐽
Where 𝐷32 is the Sauter mean droplet diameter, We𝐽 is the jet Weber number while
𝐴, 𝐵, 𝐶, 𝐷 and 𝐸 are empirical parameters that are dependent on the fluid properties. We
suggest that determining these empirical parameters using both physical and numerical
experimentation may provide a more relevant estimate than those provided by the APS
standards and those cited by others (Pacek et al, 1998; Potten and Wright, 2014). In fact, the
work of Pecek et al (1998) is based on a batch stirred vessel with no continuous flow, whose
break up mechanism is quite distinct from LJICF under continuous dispersion (Bolszo et al,
2014). The maximum droplet size (𝑑𝑚𝑎𝑥 ) and hence the 𝐷32 values may not be similar for
equal amount of energy inputs either. To that end, experimental results of 𝑑𝑚𝑎𝑥 for dilute
dispersed phase flows are shown to agree well (Simmons and Azzopardi, 2001) with those
predicted by Hinze (1955). However, for the case of dense dispersed phase flows, there is a
disagreement on the values of 𝑑𝑚𝑎𝑥 predicted by the widely used model of Brauner (2004).
This only necessitates for more systematic, detailed and synergistic investigation to provide
better understanding that can be utilised in a wider area of applications and industries. This
important area of investigation will be carried out in our LMPFL using a combination of high
speed imaging camera, dual modality Electrical Capacitance Tomography (ECT) and wire
mesh tomography (WMT).
5
2 Synergistic physical and numerical experimentation

As part of a wide range of research and development programme, four experimental and
calibration facilities are developed to conduct investigations in the general area of multiphase
flows in a synergistic manner. These facilities are depicted by Figure 1.
Figure-1: OGH’s integrated multiphase flow and computational facility with the aim to
understand, predict, verify, quantify and control the evolution of multiphase flows.
The first facility is a small multiphase flow loop (SMPFL) with nominal diameter of 2.5” and
with a straight flow stabilisation horizontal section of 25-diameter long, primarily designed to
fit in Cambridge University’s Magnetic Resonance Imaging unit. The second is a large
multiphase flow loop (LMPFL), which is a 4-times scaled up version of the SMPFL both
geometrically and dynamically. It has a nominal diameter of 10” and a straight flow
stabilisation horizontal section in excess of 100-diameter long. The larger loop has a capacity
of up to 350 m3/hr flow with 10% water. The third facility is a liquid Flow Meter Calibration
Loop (FMCL) with a nominal diameter of 12” and with 4” and 8” HTM Master Meter
calibration stations to calibrate flow meters of various sizes. It has a flow capacity of 15-600
m3/hr. Both the SMPFL and LMPFL fall under the calibration and verification audit of the
integrated facility. The fourth facility is a Heterogeneous High Performance Computing
(HHPC) facility to conduct and validate numerical experiments for any flow of interest and
thereby utilise the competence in the optimisation and validation of various multiphase
technologies. The system uses the latest and innovative NVIDIA GPU computational
hardware heterogeneously coupled with Sandy Bridge CPU cores delivering both high power
efficiency and a sustained aggregate performance of 32 “Teraflops”. The facility’s capacity,
accuracy and results from various representative applications will be published elsewhere.
6
The central theme of these facilities is the quality audit of the practices and procedures by a
national accreditation body such as UKAS according to ISO 17025 so that instruments and
computer codes are calibrated and certified. A well-controlled and audited experimental data
set will provide not only a knowledge pool that can be shared with similar facilities and
research centres but also to serve as a testing hub for device manufacturers worldwide with
significant benefit for end users too. It will also act as a catalyst for an open forum where a
more robust data- or evidence-based prediction, verification, quantification and control of
multiphase flows can be made to improve the efficiency and effectiveness of our industry. In
this paper, we focus on the use and capabilities of the SMPFL and the HHPC.
2.1 SMPFL construction and operation
The construction and operation of the SMPFL is described in detail elsewhere (Lakshmanan
et al, 2015). However, for completeness, we will provide the salient feature of the facility.
The loop was constructed so that any device under test (DUT) could be studied with a view to
characterise its behaviour and optimise its performance. The DUT is a replaceable unit that
permits the characterisation of multiple designs. The nozzles are used to inject a fluid and
produce homogeneous flow.
Figure 2: Schematics for the layout of the small multiphase flow loop (SMPFL) as it is fit
within Cambridge University’s MRI laboratory.
The SMPFL will be used to study the mixing behaviour and efficiency of a prototype
homogeniser for liquid sampling in custody transfer applications. The entire SMPFL is
designed to be compatible with MR instruments. As such, all piping is manufactured from
Durapipe HTA® C-PVC pipework, with the exception of the section where the Zanker plate
is positioned and water is injected in to the main oil flow. The layout of the SMPFL and
associated MR instrumentation is shown in Figure 2. The section of piping that runs through
7
the bore of the magnet is constructed from 2.5” diameter schedule 40 clear PVC tubing. The
SMPFL is approximately 6 m in length and 3.5 m in width. The 3.5 m width is required to
enable the pumps, tanks and coalescer to be positioned outside the 50 G line for the stray
field of the magnet. The safety and operability of the flow loop was analysed using HAZOP
prior to commissioning. The modules of the SMPFL are:
Coalescer and Holding Tanks Module: The coalescer separates the oil-water mixture and
delivers each pure fluid in their respective tanks, where it will be pumped and fed back to the
main line to mix again at the flow conditioning and water injection module. The oil and
water leaving the coalescer each contains 50 ppm or less of the water and oil phases,
respectively. Rustlick EDM-250 oil and de-mineralised water are used. The oil has a specific
gravity of 0.81 and viscosity of 2.7 cSt at 40°C. The experiment is carried out within ±2 °C
throughout the flow loop.
Flow Conditioning and Water Injection Module: The oil is fed to the loop line via a pump
and joins a 2.5” line, where it passes through stainless steel (SST) pipe that contains a Zanker
plate designed according to ISO 5167 to ensure flow uniformity and elimination of swirl.
Water is then injected 10 pipe diameters (10D) downstream of the Zanker plate in the same
direction as the oil flow but at an angle of 23° from the bottom of the pipe so that the desired
stream velocities could be achieved for water cut between 0.5% and 5%.
Dynamically Equilibrating Flow (Straight Pipe) Module: This is one of the key modules
where the flow is expected to achieve a fully developed multiphase flow to mimic pipeline
conditions. However, the length of the straight line pipe was restricted to 25D by the
geometry of the MR laboratory in which the experiments are to be performed. As such,
extensive baseline tests are carried out to quantify these limitations.
Semi-Automatic Control and Data Capture Module: The loop is operated using a control
panel driven by a programmable logic controller (PLC) and data is managed using in-house
developed human-machine interface (HMI). All instrument data transmitters capture at a rate
of 5Hz except for the temperature transmitters that have a data capture rate of 0.5 Hz.
Homogeniser Module or DUT:: This module consists of a prototype homogeniser device

developed by our partners based on a LJICF configuration, where the creation of a horse-shoe
turbulent vortex structure provides an exceptionally efficient mixing (Mahesh; 2012). The
DUT consists of the nozzle assemblage, the scoop assemblage and a pump as well as
pressure, temperature, and flow transmitters. It has also an oil-water mixture sampling point
to physically withdraw a sample of known volume to measure its homogeneity.
In the initial experiment without MR, a volume of ~10 mL of the mixture is poured into a
graduated flask every other minute in a 10 minute single trial window. For each trial a
volume of ~100 ml is captured to verify whether the mixing created by the nozzle is
representative as it is withdrawn by the scoop. The scoop take-off that is positioned within
0.1D of the pipe centre is designed to have a take-off area to achieve a nearly isokinetic flow
with the pipe stream as stipulated by the APS standards. Of course, the wall effect on
laminar flow velocity at 0.4D makes it almost a third to that at 0.1D. That is not the case for
turbulent flow. The suggestion for the position of the scoop to be within 0.1D, we argue, must
has to do with avoiding droplet crossing trajectories, coalescence and fast stratifications.
8
2.2 Magnetic resonance technique
MR experiments were carried out on a Bruker Biospec (Horizontal Bore) AV spectrometer

operating at a 1H frequency of 85 MHz. The spectrometer was equipped with a three-axis
shielded gradient coil capable of producing a maximum gradient strength of 10 G cm -1 in
each direction. An 88 mm diameter radiofrequency coil was used to excite and detect the
signal from the fluid. A brief description on the MR technique used in this work is given
here, a background on general MR is found elsewhere (Callaghan 1991; Nishimura 2010).
A chemical shift selective (CHESS) imaging sequence was adapted to selectively image the
water signal (Haase et al. 1985). In the CHESS method, the radiofrequency pulses are
designed to selectively excite the signal from the water whilst suppressing the signal from the
oil phase. The pulse sequence for the CHESS excitation used in this work is shown in Figure
3. A 3000 𝜇s Gaussian pulse was used to saturate the signal from the oil. The duration and
shape of this pulse is such that a Gaussian distribution of frequencies with a full width at half
maximum of approximately 6 ppm is excited. This first pulse acts to saturate both the oil and
water signals. Immediately following excitation of the oil peak, crusher gradients are applied
in all three directions at a strength of 4 G cm-1 to eliminate the signal from (i.e. fully dephase)
any magnetisation arising from the oil excitation. The cycle of oil excitation and saturation
was repeated four times with the duration of the crusher gradient varied between 3 ms and
5 ms for each cycle. A 7000 𝜇s Gaussian pulse was then used to excite the signal from the
water. This pulse will excite a frequency window approximately 2.5 ppm across that is
centred upfield of the water peak by 2 ppm. Therefore, this pulse will selectively excite the
signal from the water and hence the oil will not be detected. A third, 512 𝜇s Gaussian pulse
was used in conjunction with a gradient in the magnetic field to ensure that signal is only
detected from a slice of fluid 2 mm thick. This experiment provides a quantitative
measurement of the amount of water in the excited volume.
Figure 3: Pulse sequence diagram for the CHESS sequence used for selective imaging of the
water in oil-water mixtures. Shaded boxes indicate homospoil gradients; radiofrequency
pulses are labelled according to the tip angle and species affected by the pulse. Imaging
gradients are not shown, but when these are used they were placed either side of the third,
slice selective radiofrequency excitation.
The magnetic susceptibility of oil and water are different and this difference can result in
inhomogeneity in the magnetic field. Inhomogeneity in the magnetic field can affect
quantification of the water signal. To minimise the inhomogeneity 1.5 mM gadolium
9
chloride was added to the water. At this concentration, gadolinium does not influence the
apparent surface tension of the water phase (Tayler et al. 2011). The addition of gadolinium
also has the effect of reducing the T1 relaxation time of the water to approximately 60 ms,
which reduces the time required between measurements and enables quantitative observation
of fluid at higher velocity. Furthermore, the short relaxation time of the water will enhance
the ability of the CHESS sequence to selectively detect the water signal, in preference to the
oil which has a longer relaxation time. The CHESS method was also combined with one-
and two-dimensional imaging sequences to resolve the distribution of water spatially. The
distribution of water was imaged with a field of view of 193 mm (× 193 mm) and a spatial
resolution of 1.5 mm (× 1.5 mm). Oil-water mixtures were prepared with water cuts of 2.5%,
5%, 10% and 25% and measured using each of the pulse sequences in a static phantom.
Subsequently, both static and flowing experiments were performed using SMPFL in the
absence of mixing, or under well-mixed conditions for water cuts of up to 5%.
2.3 Computational Fluid Dynamics Simulations
Computational Fluid Dynamics (CFD) proves to be a useful design tool to understand the
mechanistic behaviour of processes and hence to optimise the efficiency of products. If
properly verified and validated, it is also highly cost effective and reliable tool. For example,
Boeing’s A L Velcci stated with regard to the design of the Boeing 787: “…wind tunnel tests
were not used directly for design, but it only helped to verify the accuracy of the
computational tools that are used directly to design the aircraft” (Bushnell, 2006). This
statement is a testament to the credible use of CFD as a design tool. However, to gain reliable
information from such models, validation is crucial. To that end, we set up CFD simulations
for the device under test (DUT) as depicted by Figure 4. To verify and also in the future to
quantitatively validate our work, the interFOAM code in the OpenFOAM® computational
framework was adapted and modified to conduct numerical experiments to simulate the
LJICF behaviour of the DUT. The modified interFOAM code is based on the volume of fluid
(VOF) approach (Hirt and Nichols, 1981), which is a modelling technique where the free
surfaces of immiscible fluids are tracked so that the phase distribution could be predicted. It
does so by solving the phase-fraction transport equation and a single momentum equation for
both the phases present.
Interface tracking techniques prove to have unique capabilities to capture liquid jet break up
accurately but at a high computational cost. Quantifying the resulting droplet size and its
distribution is however still an active area of research which we will address in the future. To
obtain a quantitative phase distribution, the phase fraction and momentum equations are
solved together with both the standard Reynolds Averaged Navier-Stokes (RANS) and the
Large Eddy Simulation (LES) turbulence models. Preliminary tests show that all the RANS
based turbulence models were not capable of capturing the mixing characteristics of a LJICF
configuration while the LES provided excellent results. However, this was possible after
significant modification and implementation of new models as discussed in Lakshmanan et al
(2015). Because, like some of the leading VOF based commercial codes (Danielson & Bansal
2012; Desamala et al. 2014), the standard interFOAM package was unable to simulate an oil-
water flow in a pipe with 50% water cut that exhibits phase inversion. One of the key
challenges in the VOF formulation for two phase flow is to preserve the sharpness of the
interface between the oil and water phases (Drew & Passman 1998). In most VOF methods,
10
this issue is handled by a sub-grid level reconstruction using linear or quadratic polynomials
(Renardy & Renardy 2002; Diwakar et al. 2009). Our OpenFOAM implementation takes a
different route by modifying the advection term for the phase fraction (or water cut) itself.
The model is solved using an LES turbulence model using a dynamically-computed eddy-
viscosity coefficient for the dynamic Smagorinsky model. Full details of the modified VOF
approach will be described elsewhere. The modified VOF model is solved on the mesh
shown in Figure 4. The model is applied only to one variant of the nozzle (N1a) while the
physical experiments considered four nozzle variants – namely, N1a, N1b, N2a and N3a.
While each “Ni”-nozzle serious represent a family of nozzles, the a and b identification
represents a variant with the type or family of nozzles (see Table 1).
Figure 4: (a) photograph of the Small Multiphase Flow Loop (SMPFL), (b) close up of the
homogeniser or device under test (DUT) section, (c) CFD mesh for the DUT showing the
nozzle (upstream – x=0m) and scoop (downstream – x~0.4m) positions.
3 Results and discussion

3.1 Design of experiments
The experimental campaign involves three routes or stages. For the SMPFL, two campaigns
were carried out where the first involves a physical experiment without MR measurements
while the second involves MR measurements both on a phantom of static samples but also
with the DUT under static and flowing conditions. When the experiment didn’t involve MR
measurement, a liquid oil-water sample was withdrawn from the mixed flow instead so that
the mixing efficiency can be evaluated. As described by Lakshmanan et al (2015), 16
11
experiments were conducted using Taguchi’s design of experiment. While various parameters
were investigated, we would like to note that only one type of nozzle (N1) is used in
experimental campaigns 1 and 3. . In the second experimental campaign MR measurement
was carried out and two more nozzles types were utilised (N2, N3). Note that the third
campaign was a numerical experiment and only N1 was investigated with further
optimisation numerical experiments being conducted currently on all types of nozzles. We
provide the typical range of operating conditions in Table 1.
Exp. Campaign 1 Exp. Campaign 2 Exp. Campaign 3

(No MR) (With MR) (Numerical)
Nozzle Type (-) N1 N2, N3 N1
Range of Nozzle-Scoop 0.63,1.0 0.66, 0.87, 1 N/a
distance ratio (or N-S) (-) (B1, B2, B3)
Water Cut (%) 1, 2, 3, 5 0, 1, 2, 4, 6 2
Stream Velocity (m/s) 0.25, 0.5, 0.75, 0.2, 0.8, 1.2 0.5, 1.5
1.25, 1.5 (Q1,Q2,Q3)
Range of percent Injected 0.1, 0.3, 0.5, 1.0 0.3, 0.4, 1.0 1.0
flow rate ratio (or QJ) (-) (QJ1,QJ2,QJ3)
Table 1: Ranges of parametric and dimensionless values used during the three experimental
campaigns. N-S distance is based on normalised values from the nozzle centre (B1, B2, B3).
Ranges of QJ is normalised based on its floor and ceiling values.
3.2 Experiment using the SMPFL without MRI
Initially, the SMPFL is used to study the mixing of a standard nozzle that was designed with
two variants. The two variants are expected to provide different degree of entrainment
atomisation. Therefore, the purpose of the initial experiment is to test the entrainment
efficiency of the nozzles so that only a selected few nozzles will be used for the MRI
experiments at a later stage of the development. The design of experiment (DoE) for this
initial stage of work used the Taguchi technique and 16 experiments were conducted with
three trials in each as reported by Lakshmanan et al (2015). Depending on the variations of
the chosen factors, the 16 experiments were conducted in 4 groups. Figure 5 shows a
summary of the accuracy of the water cut for these experiments. These measurements were
obtained by sampling the oil-water mixture that was withdrawn through the scoop, as
illustrated in Figures 2 and 4. The oil-water samples were allowed to settle for two more
days before the water fraction was measured. No permanent emulsification is observed on
these samples. The samples in the graduated cylinder were measured for total liquid height
and water height using a TOMARCH PN 31761 digital height gauge with measurement
accuracy of 0.0127 mm. The water cut was calculated by dividing the water volume by the
total volume of liquid. Measurements by three operators were taken and checked for
repeatability. Similarly, each of the three experiments captured via the HMI were checked for
repeatability.
The measured fraction of water obtained from the scoop is compared with the known fraction
of water based on the measured oil and water flow rates. A relative accuracy was established
and the results are shown in Figure 5. Except for three data points, the mixing achieved falls
within the requirements of the APS standards. The results demonstrate a complex behaviour
12
on the degree of mixing that can be achieved. Although the APS standards stipulate that for a
horizontal pipe with stream velocity in excess of 2.4 m-s-1 adequate natural mixing may be
achieved, our results indicate that this is not the case. Furthermore, the APS standards
recommend the scoop take off velocity to be isokinetic and, for a horizontal pipe, the scoop
inlet area is depicted by a figure to be placed within 0.1D below its centre, which will only
allow 1% injection for a circular inlet. Of course, the APS standards indicate wall effect to be
of consideration. But, velocity profile encompasses the wall effect and at that location it is
likely to provide best option to avoid droplet crossing trajectories so that a more
representative sample is withdrawn. In addition, our results indicate that an injection of 5% is
often required. In fact, injecting 1.8% of the total flow rate with stream velocity of 1.47 m/s
proved to provide the worst mixing. For a weak jet with 1.8% injection, the stream
momentum flux is overwhelming and hence the jet collapses without providing sufficient
mixing. This indicates a limitation on APS standards both in their specification or the manner
it could be interpreted. More importantly, how mixed-ness is quantified. However, for a
stratified flow with stream velocity less than 0.75m/s and with an injection of 8% of the total
flow rate, it is possible to achieve homogeneous mixing in a horizontal pipe that satisfies the
minimum requirements of the APS standards. Indeed the results indicate that the extent of
mixing is dependent primarily on the velocity and momentum flux ratios, and is not very
sensitive to the water cut, which was not expected. However, eq. (1) suggests that the mixing
will be a strong function of water-cut if the water cut becomes high or the stream velocity is
too low and the resulting entrainment atomisation creates a dense dispersed phase that may
result in collision and coalescence of water droplets thereby enhancing stratification.
Figure 5: Plot of the relative error of water cut as a function of the injected flow into the
mixing nozzles grouped by velocity (identified by symbols) and water cut (identified by
colour). The symbols indicate water cut (black) ~1%; (red) ~2%; (blue) ~3%; and (green)
3.5% and stream velocities (●) ~0.25 m/s; (▲) 0.35 m/s; (■) ~0.43m/s; (♦) ~0.54m/s; (▼)
~0.77m/s; (+) ~0.81 m/s; ( ) ~1.15 m/s and (×) ~1.47 m/s. The x-axis shows the injected
flow rate relative to the total oil-water flow rate through the pipe. The red dashed lines
indicate a level of better than 90% mixing as required by the APS standards. However, the
blue dotted lines indicate a level of 5% mixing, which is twice better than the minimum
requirement. The mixing also appears less sensitive to the water cut, which is unexpected.
13
The results obtained using the SMPFL demonstrate that there is a need for a better
understanding of the homogenisation process when using the LJICF arrangement. In the next
sections, we present the numerical- and physical-experiment results that enabled us to study
the homogenisation process in more detail than has hitherto been possible.
3.3 Numerical experiment using the HHPC facility
CFD simulations were carried out on the LJICF configuration for an initially stratified oil-
water flow for a range of nozzle orientations, fluid velocities and spacing of the scoop.
Figure 6 gives a summary of the results of one of these simulations. Figure 6(a) shows the
instantaneous speed (magnitude of the velocity) of the fluid in the pipe. High speeds are
observed in the vicinity of the jets of the nozzle, as expected. These high speed liquid flows
enhance the jet-break up and generate the turbulence and mixing that is used to homogenise
the distribution of water throughout the cross section of the pipe.
Figure 6: Example results from the CFD simulations showing (a) speed of the liquid in the
homogenization section (DUT), (b) maps of the local water cut at different positions along
the length of the pipe, (c) plot of the water cut at the top (black), middle (blue) and bottom
(green) of the pipe for overall water cuts of 2.0% with stream velocities of 0.5 m s-1 (dashed
lines) and 1.5 m s-1 (solid lines). (d) shows the same data as in (c), but with the vertical scale
reduced to highlight differences in the local water cut.
Figure 6(b) illustrates the improvement in the homogenization that is achieved downstream of
the nozzle for the simulation conditions shown in Figure 6(a). Upstream of the nozzle
injection, the flow is stratified with water only visible in a narrow channel along the base of
14
the pipe. As soon as the flow approaches the nozzle, the water is rapidly distributed across
the cross section of the pipe. Downstream of the nozzle the flow becomes increasingly
homogeneous. To quantify the distribution of water and compare this simulation with the
industry standard criteria, the mean water-cut averaged over time as well as over the
immediate neighbouring computational cells is shown in Figure 6(c). The local water cut is
shown at the top, bottom and in the middle of the pipe for stream velocities of 0.5 m s-1 and
1.5 m s-1. Initially the flow is fully stratified and so the local water cut taken from the middle
and top of the pipe show no water, and as expected, the local water cut at the bottom of the
pipe is 100%. The nozzle is defined as the origin of the coordinate system and at this point
the water cut at the bottom of the pipe rapidly decreases, whilst the water cuts in the middle
and at the top of the pipe increase. Figure 6(d) shows the same data as Figure 6(c), but with a
finer scale such that it is possible to resolve the differences in the water cut at the different
heights in the pipe. Downstream of the nozzle, the average water cut in the middle of the
pipe, for both the 0.5 m s-1 (dashed lines) and 1.5 m s-1 (solid lines) stream velocities is
close to the 2% water cut that is expected for a homogenously mixed flow in this case, but
significantly greater than those at the top or bottom of the pipe. However, both the top and
bottom water cut values are less than those in the middle which means that the water is
sufficiently lifted from the bottom wall, which is key. Overall the stratified water is now
fairly evenly distributed as manifested by the overlapping graphs at 2% water cut values in
Figure 6(d), satisfying the APS standards. Further simulations (not shown) indicate that the
flow rate through the nozzle has a strong effect on the homogenization process. This result is
consistent with the experiments performed on the SMPFL during experimental campaign 1.
3.4 MR Measurements of static samples
Four different commercial oils (Rustlick EDM-250, VG 5, kerosene, and SAE 30) with
viscosities between 3 cSt and 100 cSt were tested using MR. Rustlick EDM-250 and VG 5
both have a high flash point and low viscosity making them a suitable model at low
temperature for oil-water experiments and safe to use.
The spectrum of mixtures of oil and water for all four oils were similar, with the resonance of
the oil peak appearing at a chemical shift approximately 4.5 ppm lower than the water peak in
all cases. These oils are therefore essentially equivalent from the perspective of MR. At this
point it is worth noting that the imaging spectrometer used for these experiments does not
provide the homogeneity or stability of the magnetic field that are required for high resolution
spectroscopic measurements; therefore the detailed chemical structure of the oil species is not
visible in the spectra. Static oil-water mixtures were prepared with water-cuts between 2.5%
and 25%. Figure 7(a) shows the magnetic resonance spectrum following CHESS excitation
of two samples. The water peak is located at 5.5 ppm, whilst the oil peak would be located at
1 ppm. The oil peak is almost completely suppressed, meaning that the amount of water is
calculated easily by integrating the signal intensity from the remaining peak. The integral of
the signal between 5 and 6 ppm, which is attributed to the water, is calculated and plotted
against the known water cut, as shown in Figure 7(b). Each experiment was repeated between
two and five times at different positions within the same static sample to provide an
indication of the reproducibility of the experiments. Error bars, given by the 95% confidence
interval, are shown in Figure 8 for both the measured signal intensity and the water cut. The
error bars are mostly comparable in length to the size of the symbols in the plot and so are
15
difficult to resolve. A larger error is observed for the water cut of 10% which is attributed to
a poor shim for this sample. The total signal intensity associated with each measurement had
a standard deviation of about 0.1 (a.u.) for each water cut. The integrated signal intensity
shows a linear relationship with the water cut. A linear fit through these data yields a slope
of 0.385 ± 0.005%-1 with an R2 value of 0.999. These results demonstrate that MR provides
quantitative measurements of the water cut. On the basis of the calculated slope, an
uncertainty in the signal intensity of 0.1 corresponds to an error in the measured water cut of
approximately 0.2%, even for the poorest quality measurements obtained with the 10% water
cut the uncertainty of the water cut is less than 1%.
25 % y=
2.5 % 10
signal intensity (a.u.)
signal intensity (a.u.)

800
m1 0.3
Chisq 0.05
2 0.9
R
400 5
0 0
10 0 0 10 20 30
chemical shift (ppm) water cut (%)
(a) (b)
Figure 7. (a) MR spectra obtained from water-oil mixtures at 2.5% and 25% water. Spectra
were obtained using the CHESS sequence to suppress signal from the oil peak. The water
peak is located at approximately 5.5 ppm; the peak arising from the oil is largely suppressed
but would be located at approximately 1 ppm. (b) Plot of the integrated signal intensity as a
function of the water-cut. The solid line is a linear fit to the data passing through zero. The
slope of the line is 0.385 ± 0.005%-1
Static oil-water phantoms were prepared with water cuts of between 5% and 25% to
demonstrate two-dimensional (2D) MR imaging of water within the pipe. However, ultrafast
2D MR measurements on fast-flowing oil-water mixtures are extremely challenging. Instead
here we obtain time-averaged images of the distribution of oil and water. An example of the
two-dimensional images that will be obtained of the oil-water flow is shown in Figure 8 for
static samples of 10% and 25% oil-water mixtures. In these images, in each case, two distinct
layers can be seen; one is almost pure water while the other is almost pure oil. To aid
visualisation, the approximate outline of the pipe is sketched on Figure 8 using a dashed
white line. The cross section of the pipe that contains water in Figure 8 is clearly visible, and
can be calculated to be 3.0 ± 0.3 mm2, and 7.6 ± 0.3 mm2 for the 10% and 25% samples,
respectively. The total pipe area is 30 mm2, thus, the measured cross sectional areas are in
good agreement with the volume fraction of the samples. The spatial resolution of the images
is 1.5 mm × 1.5 mm and the acquisition time was 120 s. The acquisition time could be
reduced but for later experiments on flowing systems reducing the acquisition time would
have the effect of reducing the apparent signal-to-noise ratio owing to fluctuations arising
16
from the time varying nature of the flow. The signal intensity is proportional to the water
content, as shown in Figure 7(b).
Figure 8. 2D MR images of stratified water cut distribution for a static oil-water system
through the cross section of the pipe with (a) 10% and (b) 25% water-cut.
Therefore, these images provide a direct measure of the amount of water at each position
within the pipe. The water cut at each location is estimated to be accurate to within ±1%,
based on the apparent signal-to-noise ratio of the image. Additional changes in signal
intensity of up to 10% arise from variation in the sensitivity of the radiofrequency coil. In the
future these variations will be eliminated by using a map of the sensitivity distribution of the
coil (Holland et al. 2009). The accuracy of these time-averaged measurements could be
further improved by increasing the total acquisition time, provided that the pipe flow is
operating in a steady state, which is crucial. The CHESS sequence was then combined with
imaging sequences to enable visualisation of the water distribution through the pipe. A
stationary system in a horizontal pipe was imaged, where the denser water phase remains on
the bottom of the pipe whilst the lighter oil phase fills the top of the pipe. Stratified flows of
such type are commonly encountered in the oil and gas sector and represent a worst case
scenario from which to achieve a homogeneous mixture.
3.5 MR Measurements during flowing conditions
In order to demonstrate that the MR measurements are quantitative, the signal intensity from
the spectral acquisitions was plotted for well mixed flow of oil and water at water cuts
between 1% and 6%, as shown in Figure 9. The plot shows a linear correlation, as was seen
with the static measurements shown in Figure 7(b). A linear fit through these data gives
slopes of (1.9 ± 0.1) × 107 %-1 and (2.1 ± 0.1) × 107 %-1 for stream velocities of 0.8 m s-1 and
1.2 m s-1, respectively. The fact that this slope is essentially constant for changing stream
velocity indicates that the effects of in-flow are insignificant in this case. The inherent noise
level in the data is ~1 × 106, indicating that the water cut can be measured with an accuracy
of approximately ±0.2 % using MRI.
17
Figure 10 shows 2D MR images of a stratified oil-water flow at water cuts of 2% and 7.5%,
based on the flow rate of oil and water in the system. These images are equivalent to those
shown in Figure 8, but acquired under flowing conditions such that the oil and water phases
are stratified and do not mix. It is clear from these images that the water remains largely
confined to the base of the pipe and little water mixes into the bulk oil flow, as expected. It is
also interesting to examine the cross sectional area of the pipe filled with water in both cases.
Figure 9. Signal intensity from MR experiments performed on a well-mixed system

operating with a stream velocity of the oil phase of (○) 0.8 m s-1 and (×) 1.2 m s-1 and for
water cuts of up to 6% based on the measured volumetric flow rate of oil and water. The solid
line indicates a linear fit to the data, with the dashed lines indicating the 95% confidence
interval for this fit. The error bars for the measured data points indicate two standard
deviations in the measured signal intensity.
Figure 10. 2D MR images of the distribution of water through the cross section of the pipe
during stratified flow with (a) 2% and (b) 7.5% water, based on the measured volumetric
flow rate.
The interface between the water and the oil in Figure 10 is no longer characterised by a sharp
interface as was seen in Figure 8. This is due to some flow instability at the oil-water
interface, even if it is at low velocity. In these time-averaged images, this instability
18
manifests as a gradual change in water content from pure water to pure oil. In spite of this, it
is still possible to estimate the cross sectional area of the pipe that is filled with water. In this
case the proportion of the pipe area occupied by the water at nominal water cuts of 2% and
7.5%, is 9 ± 2% and 15 ± 1%, respectively. The uncertainty is calculated on the basis of the
sharpness of the interface between the oil and water phases. It is clear that for these
experiments, the cross sectional area of the pipe occupied by water does not scale with the
water cut measured on the basis of the flow rate of oil and water through the system.
In order to investigate these observations further, 1D profiles were acquired along the height
of the pipe, as shown in Figure 11. These plots provide an indication of the amount of water
present at each vertical position, integrated across the width of the pipe. Profiles are shown
for stream velocities of 0.2 m s-1, 0.8 m s-1, and 1.2 m s-1, with nozzles N2 and N3. For
perfect mixing, the curves would be expected to be shaped like a hemi-circle, as this is the
projection of the cross section of the pipe (Lakshmanan et al, 2015). It is clear that the
profiles do not conform to this ideal shape for any of the results shown. The discrepancy is
largely due to the inhomogeneity of the magnetic- and radio-frequency fields used but also
the inhomogeneity of the mixing and unsteadiness of the flow. Despite, the profiles only
providing a qualitative indication of mixing, some insight into the flow and mixing within the
pipe can be gained.
As the water cut increases, it is clear that the fraction of the pipe filled with water increases,
as expected. At the very low stream velocity of 0.2 m s-1 that is typical of oil fields with end
of production expectancy, high signal is obtained from the bottom of the pipe while using
nozzle N2, indicating the presence of a relatively high concentration of water along the base
of the pipe. However, the amount of water at the base of the pipe is slightly less for N3 than
nozzle N2. In both cases, at the very low velocity, full mixed-ness is not achieved.
Figure 11. Profiles of the average water distribution in the pipe 330mm downstream of the
injection nozzle N2 (a-c) and nozzle N3 (d-f). The stream velocities are (a, d) 0.2 m s-1, (b, e)
0.8 m s-1 and (c, f) 1.2 m s-1.
19
It is instructive however to observe that the minimum requirement for homogeneous mixing
could be met by using the so called C1/C2 ratio in the APS standards, which is obtained by
taking samples above and below the centre-line of the pipe at say 0.5 × R, where R is the
radius of the pipe. The profile is approximately symmetric for up to 2/3rd of the pipe radius,
therefore the C1/C2 ratio will be close to one. However, what is demonstrated here is that the
C1/C2 ratio is not a good measure of mixed-ness as it does not account for the stratification
of the flow seen here. At higher stream velocities, the distribution is more homogeneous and
for both nozzles, the distribution of water is essentially the same at stream velocities of 0.8 m
s-1 and 1.2 m s-1. The similarity in the shape of the profiles for a variety of water cuts and
stream velocities indicates that the water is likely well mixed in these cases.
To elucidate the mixing further, a quantitative water-cut profile was obtained by dividing the
profiles from Figure 11 by the profile of a static oil sample, as shown in Figure 12. This
normalisation of the signal by the signal from a static oil sample helps correct for variations
in the signal intensity arising from inhomogeneity in the radiofrequency and magnetic fields
used in these experiments. The remaining signal can then be correlated with the known total
water cut, as in Figure 9, to provide an estimate of the local water cut. At a velocity of 0.2 m
s-1 the normalisation is effective as this corresponds to laminar flow and thus the flow is
stable over time. In this case, the local water cut is measured with an estimated accuracy of
±0.5%. In the case of turbulent flow, the measurements are not as stable over time and a
systematic bias in the measurement is introduced at different locations within the pipe. Thus,
for turbulent flow the error is estimated to be approximately ±3%.
Figure 12 is organised to show the parametric dependence and the characterisation of nozzles
N2 and N3. For example, row (a,b,c) shows the mixing effectiveness as a function of stream
velocity and water cut, whilst column (a,d,g) shows the evolution of stratification at different
distances downstream from the nozzle injection position. In a perfect mixing condition, the
water cut profiles are expected to be flat indicating a constant water cut throughout the pipe.
The profiles are significantly flatter than those shown in Figure 11, however some curvature
remains. For the turbulent flow profiles, the instability of the flow means that the uncertainty
in the measurement is large and the flow may be regarded as being well mixed within the
experimental uncertainty (±3% in this case). For the lowest velocity, the flow was laminar
and the results are more accurate (±0.5%). The results confirm that stratification is significant
for the lowest flow rate and that this increases downstream of the injection point. Using these
quantitative measurements of the local water cut, it is possible to estimate the error that
would be introduced into a flow measurement that does not sample from the base of the pipe.
At a water cut of 1% the error is negligible (about 1.04% vs 1%), however, at a water cut of
6% the error is significant (true water cut about 9% vs 6%) for the worst case condition of the
data at position B3 (see Figure 12(g)).
Two dimensional images of the flow were also obtained for a few of these flow conditions.
Figure 13 illustrates two of these images, taken from the (a) worst and (b) best mixing
scenarios at a water cut of 4%. These experiments correspond to the profiles in Figures 12
(a) and (k)). A bright region is clearly visible at the bottom of the pipe in Figure 13(a),
indicating the presence of almost pure water in a thin line around the base of the pipe. By
contrast at the higher flow velocity, the image intensity is fairly uniform indicating that the
fluid is well mixed.
20
N2B1Q1 N2B1Q2 N2B1Q3
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
(j) (k) (l)

Figure 12. Water cut profiles obtained by normalising the data from Figure 11. (a-i) are for
Nozzle Type 2 (N2), whilst (j-l) are for Nozzle Type 3. (a-c) and (j-l) were taken XX
downstream of the nozzle, (d-f) were taken YY downstream of the nozzle and (g-i) were
acquired ZZ downstream of the nozzle. For all nozzles types, the columns correspond to
different stream velocities of 0.2 m s-1 (Q1), 0.8 m s-1 (Q2), and 1.2 m s-1 (Q3). An important
comparison between nozzle types N2 and N3 is depicted by rows (a,b,c) and (j,k,l), for the
same dynamic condition and “N-S” value B1. (j-l) shows a clear reduction in the quantity of
water at the base of the pipe when compared with (a-c), particularly at the most challenging
Q1 value. The range of water cut values is shown in colour blue (1%), gold (2%), yellow
(4%) and purple (6%). Not all the experiments utilise 6% water cut.
21
Figure 13. 2D MR images of the distribution of water through the cross section of the pipe
during a flowing condition for water cut of 4% with (a) nozzle N2 at stream velocity of
0.2 m s-1 (b) nozzle N3 at a stream velocity of 0.8 m s-1., both at N-S distance B1. The bright
white region at the base of (a) indicates a high local water fraction, however small it looks.
4 Conclusion
As part of a large research and development work, we have commissioned a small multi-
phase flow loop (SMPFL) with a Liquid Jet In Cross Flow (LJICF) configuration to study
liquid-liquid flows using computational fluid dynamics (CFD) and magnetic resonance
imaging (MRI). First, a numerical simulation technique was developed using the open source
computational platform OpenFOAM®, where a modified Large Eddy Simulation (LES)
method was utilised to capture the oil-water flow regimes and the resulting mixing behaviour
due to a novel homogeniser. This initial simulation allowed optimising the design of the
homogeniser that ought to comply with the APS standards. Once the homogeniser is
manufactured and the SMPFL is operational, MR measurement technique using a CHESS
sequence was used to capture the data for different nozzles and mixing operating conditions.
The static oil-water composition is quantified to within 0.2%. In addition, by combining the
CHESS sequence with imaging, one- and two-dimensional profiles of the water distribution
throughout the cross-section of the pipe were obtained. For the case of laminar flow, the
accuracy of the water cut is ±0.5% while for turbulent flow case, the accuracy is ±3%, which
is within measurement uncertainty. It is important to interrogate the large data set we have to
extract the hydrodynamic artefacts that may have likely pronounced some of the water cut
values observed. However, these results highlighted various issues related to both the general
area of LJICF for liquid-liquid systems but also some unresolved issues in the current APS
standards, not only on their interpretation but also on key assumptions. Understanding the
mechanisms of Jet break-up, droplet formation and size distribution may hold the key.
5 Acknowledgement
The authors are grateful for the financial support from Innovate UK and the Engineering and
Physical Sciences Research Council through grant KTP009424. Technical support has been
provided by Dr Mick Mantle, Magnetic Resonance Research Centre (MRRC), University of
Cambridge; engineering and operational support was provided by Oil & Gas Holdings. The
authors also acknowledge the continuing support from Mr Steve O’Donnell, Chairman of the
Innovate UK Local Management Committee (LMC) and MD, Oil & Gas Systems Ltd.
22
6 References
ANGELI P., HEWITT G.F., (2000), Drop size distributions in horizontal oil-water dispersed flows,
Chemical Engineering Science., 55 3133–3143.
API 8.2 (1995). Standard Practice for Automatic Sampling of Liquid Petroleum and Petroleum
Products, American Petroleum Institute.
BIEBERLE A., HARTING H., RABHA S., SCHUBERT M., HAMPEL U., (2013), Gamma-Ray
Computed Tomography for Imaging of Multiphase Flows. Chemie Ingenieur Technik., 85 1002–1011.
BOLSZO, C. D.; McDONELL, V. G., GOMESZ, G. A., SAMUELSEN, G. S., (2014), Injection of
water-in-oil emulsion jets into a subsonic crossflow: An experimental study, Atomisation and Sprays,
24(4):303-348.
BRAUNER, N., (2001). The prediction of dispersed flows boundaries in liquid-liquid and gas-liquid
systems, Int. J. Multiphase Flow 27:885-910.
BUSHNELL DM, (2006). Scaling: Wind Tunnel to Flight, Ann. Rev. Fluid Mechanics, (38): 111-128
CALLAGHAN P.T.,(1991), Principles of nuclear magnetic resonance microscopy, Oxford:

Clarendon Press.
CLARK B.J., (1964), Breakup of a liquid jet in a transverse flow of gas, NASA TN D., 2424.
DANIELSON T.J., BANSAL K.M., (2012), Simulation of Slug Flow in Oil and Gas Pipelines Using
a New Transient Simulator. In Offshore Technology Conference., ConocoPhillips.
DESAMALA A.B., DASAMAHAPATRA A.K., MANDAL T.K., (2014), Oil-Water Two-Phase

Flow Characteristics in Horizontal Pipeline – A Comprehensive CFD Study. International Journal of
Chemical, Nuclear, Materials and Metallurgical Engineering., 8 336–340.
DIWAKAR S. V., DAS, S.K., SUNDARAJAN T., (2009), A Quadratic Spline based Interface
(QUASI) reconstruction algorithm for accurate tracking of two-phase flows. Journal of
Computational Physics., 228 9107–9130.
DREW D.A., PASSMAN S.L., (1998), Theory of multicomponent fluids, New York: Applied
Mathematical Sciences., vol 135, Springer-Verlag.
ELKINS C.J., ALLEY M.T., (2007), Magnetic resonance velocimetry: applications of magnetic
resonance imaging in the measurement of fluid motion. Experiments in Fluids., 43 823–858.
FERNANDO L.M., LENN C.P., (1990), Jet mixing of water in crude oil pipelines for representative
sampling. Analytica Chimica Acta., 238 251–256.
FUKUSHIMA E., (1999), Nuclear magnetic resonance as a tool to study flow. Annual Review of
Fluid Mechanics., 31 95 –123.
GALINAT S., MASBERNAT O., GUIRAUD P., DALMAZZONE C., NOIK C., (2005), Drop break-
up in turbulent pipe flow downstream of a restriction. Chemical Engineering Science., 60 6511–6528.
GEERY E.L., MARGETTS M.J., (1969), Penetration of a high-velocity gas stream by a water jet.
Journal of Spacecraft and Rockets., 6 79–81.
23
GLADDEN L.F., SEDERMAN A.J.,(2013), Recent advances in Flow MRI. Journal of Magnetic
Resonance., 229 2–11.
GOTZ J., ZICK K., (2003), Local velocity and concentration of the single components in water/oil
mixtures monitored by means of MRI flow experiments in steady tube flow. Chem. Eng. and Tech.
26 59–68.
HAASE A., FRAHM J., HANICKE W., MATTHAEI D., (1985), 1H NMR chemical shift selective
(CHESS) imaging. Physics in medicine and biology., 30 341–344.
HAASE A., FRAHM J., MATTHAEI D., HANICKE W., MERBOLDT K.D., (1986), FLASH
imaging. Rapid NMR imaging using low flip-angle pulses. Journal of Magnetic Resonance., 67 258–
266.
HENNIG J., NAUERTH A., FRIEDBURGH H., (1986), RARE imaging: a fast imaging method for
clinical MR. Magnetic Resonance in Medicine., 3 823–833.
HINZE, J., (1955), Fundamentals of the hydrodynamic mechanism of splitting in dispersion

processes, AIChE J. 1:289-295
HIRT, C. W., NICHOLS, B. D., (1981). Volume of Fluid (VOF) method for the Dynamics of Free
boundaries, J. Comp. Physics, 39:201-225.
HOLDEMAN J.D., (1993), Mixing of multiple jets with a confined subsonic crossflow. Progress in
Energy and Combustion Science., 19 31–70.
HOLLAND D.J., MARASHDEH Q., MULLER C.R., WANG F., DENNIS J.S., Fan L.S., Gladden
L.F., (2009), Comparison of ECVT and MR Measurements of Voidage in a Gas-Fluidized Bed.
Industrial & Engineering Chemistry Research., 48 172–181.
HOLLAND D.J., Gladden L.F., (2014), Less is More: How Compressed Sensing is Transforming
Metrology in Chemistry. Angewandte Chemie International Edition., 53 13330–13340.
ISMAIL I., GAMIO J.C., BUKHARI S.F.A., YANG W.Q., (2005), Tomography for multi-phase flow
measurement in the oil industry. Flow Measurement and Instrumentation., 16 145–155.
ISO/IEC 17025:2005: General requirements for the competence of testing and calibration laboratories
ISO 3171 (1999). Methods of test for petroleum and its products, Petroleum liquids –automatic
pipeline sampling, International Standardisation Organisation..
KUMARA W.A.S., HALVORSEN B.M., MELAAE M.C., (2010), Particle image velocimetry for
characterizing the flow structure of oil-water flow in horizontal and slightly inclined pipes. Chemical
Engineering Science., 65 4332–4349.
LAKSHMANAN S., MARU W., HOLLAND D., SEDERMAN A., (2015), Multiphase flow
measurement using Magnetic Resonance Imaging, 7th Int. Symp. in Process Tomography, 1-3 Sept.
MAHESH K., (2012), The Interaction of Jets with Crossflow. Ann. Rev. Fluid Mechanics, (45) 379–
407.
MARGASON R.J., (1993), Fifty years of jet in cross flow research, In Computational and
Experimental Assessment of Jets in Cross Flow., AGARD–CP-534 1-33.
24
MAUDSLEY A., HILAL S.K., PERMAN W.H., SIMON H.E.,(1983), Spatially resolved high
resolution spectroscopy by “four-dimensional” NMR. Journal of Magnetic Resonance., 51 147–152.
NISHIMURA D.G., (2010), Principles of Magnetic Resonance Imaging, www.lulu.com.
NYDAL O., PINTUS S., ANDREUSSI P., (1992), Statistical characterization of slug flow in
horizontal pipes. Int. J. Multiphase Flow., 18 439–453.
OpenFOAM: Open Source CFD Toolkit - http://www.openfoam.com/
PACEK A. W., MAN C. C., NIENOW A. W (1998), On the Sauter mean diameter and size
distribution in turbulent liquid/liquid dispersion in a stirred vessel, Chemical Engineering Science,
53(11):2005-2011.
POTTEN G., WRIGHT S., (2014). Methods of Determining and Verifying Fiscal Sampling System
Uncertainity by Analysing 25 Years of Real Field Proving Data and Laboratory Tests Compared with
International Acceptance Criteria, 32nd North Sea Flow Measurement Workshop, 21-24 October.
RENARDY Y., RENARDY M., (2002), PROST: A Parabolic Reconstruction of Surface Tension for
the Volume-of-Fluid Method. Journal of Computational Physics., 183 400–421.
SANKEY M., YANG Z., GLADDEN L., JOHNS M.L., LISTER D., NEWLING B., (2009), SPRITE
MRI of bubbly flow in a horizontal pipe. Journal of Magnetic Resonance., 199 126–35.
SIMMONS MJH, AZZOPARDI BJ, (2001). Drop size distributions in dispersed liquid-liquid pipe
flow, Int. J. Multiphase Flow, 27:843-859.
SMITH S.H., MUNGAL M.G., (1998), Mixing, structure and scaling of the jet in crossflow. Journal
of Fluid Mechanics., 357 83–122.
TAYLER A.B., HOLLAND D.J., SEDERMAN A.J., GLADDEN L.F., (2012), Applications of ultra-
fast MRI to high voidage bubbly flow: measurement of bubble size distributions, interfacial area and
hydrodynamics. Chemical Engineering Science., 71 468–483.
TAYLER A.B., HOLLAND D.J., SEDERMAN A.J., GLADDEN L.F., (2011), Time resolved
velocity measurements of unsteady systems using spiral imaging. Journal of Magnetic Resonance.,
211 1–10.
TRALLERO JL, (1995). Oil-water flow patterns in horizontal pipes, PhD Thesis, Univ. of Tulsa.
UKAS: United Kingdom Accreditation Services - http://www.ukas.com/
VALLE A., (1998). Multiphase pipeline flows in hydrocarbon recovery, Multiphase Science and
Technology, 10:1-139.
WANG M., (2015), Industrial Tomography: Systems and Applications, Cambridge: Woodhead
Publishing.
WANG X., FENG Z., FORNEY LJ.,(1999), Computational simulation of turbulent mixing with mass
transfer. Computers & Structures, 70 447–465.
XU, X.X., (2007), Study on oil-water two-phase flow in horizontal pipelines. Journal of Petroleum
Science and Engineering., 59 43–58.
25
The Implementation of Multiphase Meters in a High Sulphur Environment on
TCO’s Tengiz Field, Kazakhstan
Authors:
John Clarke, Chevron,
Kelda Dinsdale, Roxar, Emerson Process Management,
Lars Anders Ruden, Roxar, Emerson Process Management
Co-authors:
Ståle Gjervik, Roxar, Emerson Process Management,
Frode Hugo Aase, Roxar, Emerson Process Management,
Sturle Haaland, Roxar, Emerson Process Management,
Stig M. Sigdestad, Roxar, Emerson Process Management
1. Overview
High hydrogen sulfide (H2S) levels within multiphase flow are a frequent point of
discussion relating to multiphase meter performance. This paper will show how the
Roxar Multiphase meter from Emerson Process Management is operating reliably and
accurately in areas of high H2S concentrations on TCO’s Tengiz field.
The paper will examine the technical challenges seen – in particular unstable mixed
density measurements and extreme fluctuations of daytime to nighttime and winter to
summer ambient temperatures.
It will examine how these challenges have been overcome and the long-term
collaboration over the field’s lifecycle between TCO and Emerson. The result - in the
words of TCO – is a “world-class multiphase metering environment.”
2. The Operator & The Field
Tengizchevroil LLP (TCO) was formed between the Republic of Kazakhstan and
Chevron Corporation in 1993 to explore and develop the super-giant Tengiz Oil Field,
discovered in 1979.
The Tengiz Field (see figure 1) is among the top 10 producing fields in the world. It
covers 565 square kilometers. Today, the Tengiz Oil Field is one of the world's
deepest developed oil fields, with the oil column measuring one mile thick. The field
has recoverable reserves estimated at between six billion and nine billion barrels.
In the reservoir, Tengiz oil is mixed with ‘sour gas’ that is hydrocarbon gas with high
concentrations of toxic hydrogen sulfide. Processing facilities separate the oil and
sour gas, and stabilize and sweeten the oil. The sour gas is processed into sales gas,
natural gas liquids, and elemental sulphur, and sold for fertilizer and other products.
1
2008 also marked the completion of a US$7.2 billion expansion project that boosted
production capacity from around 310,000 barrels per day to 540,000. The expansion -
the Sour Gas Injection / Second Generation Plant - includes the largest crude oil and
sour gas processing units in the world.
Current partners in the field are Chevron (50%), ExxonMobil (25%), KazMunayGaz
(20%) and Lukoil (5%). Today TCO is the largest company in Kazakhstan and
produces and markets crude oil, gas and sulphur.
Figure 1 – The Tengiz oil field
3. The Challenges of Sulphur
As mentioned, the oil from Tengiz contains a high amount of sulphur (up to 18%). In
2014, TCO sold over 3.8 million metric tons of sulphur to customers in many
countries, including Kazakhstan, Russia, Ukraine and China.
Such sour service conditions, however, bring with them significant challenges when it
comes to multiphase measurement. These include fluctuating conditions in the field
which typically produce H2S of up to 18%, in addition to Gas Volume Fraction (GVF)
of 65% to 80%, 0% to 5% Water Liquid Ration (WLR), and pressures of 70 to 85
bars.
Due to high H2S values and limited capacity of water treatment facilities, TCO
remained highly focused on achieving water measurement with the lowest possible
uncertainty and highest sensitivity and reliability.
2
4. Multiphase Meters – Applications in H2S Field
Emerson Process Management delivered 16 second-generation Roxar Multiphase

Meters (MPFM) 1900VI to TCO, with commissioning taking place from May 2007.
The meters have been in use in the field ever since. Emerson replaced three of these
meters with the third generation MPFM 2600, with the commissioning of these new
meters taking place in May 2014.
The ability to measure flow rates reliably and accurately under the presence of high
and fluctuating H2S is a central feature of the multiphase meters. To this end, the
Roxar Multiphase Meters have a number of features to address this.
The Measurement Principle: The Roxar multiphase meter applies fractional

measurements using electrical impedance measurements, in combination with either
non-gamma software or a single high-energy gamma for density measurements (see
figure 2)
Figure 2: Roxar MPFM 2600 main components
These measurement principles are highly robust against variations in H2S

concentration. H2S that is present in the flow as a gas will be measured as gas by the
Roxar Multiphase meter. There is no need for any precautions, special calibrations or
compensations for high H2S operations.
The Roxar electrical impendence principle measures the mixed conductivity and
permittivity to determine the phase fractions. It is not likely that the oil permittivity
and water conductivity will change significantly in the presence of H2S gas and
sulphur atoms. Roxar has never seen issues with the watercut determination or gas
measurement due to high and fluctuating levels of H2S and sulphur.
3
Low Maintenance Requirements: When dealing with well streams containing
hazardous sour gases such as H2S, it is important to consider safety and
environmental implications that cleaning and interruption of the well flow might
have. The normal maintenance schedule of the meter will be a yearly empty pipe
calibration of the gamma system and checkup of the meter’s electronics and
transmitters to see if there is no drift. It is important to mention that these
maintenance activities do not require the meter to be opened up or that the process
fluid is exposed to the atmosphere in any sense. All these maintenance actions can be
performed remotely from the service console if the infrastructure required to this is
planned for.
The advantage of annual rather than monthly maintenance is of high value for such
operations, where any exposure to the gases involves a health hazard. Less
maintenance means less people exposed to potentially dangerous areas.
Material selection: Process conditions are taken into account during the material
selection for the meter. Stainless- or duplex like steels are normally suitable for
process conditions with low H2S concentrations. For applications with high H2S
concentrations, inconel 625 material for wetted parts is often offered.
Sampling requirements: Atmospheric sampling of high H2S fluids is generally

unwanted, and should be minimized due to the HS&E aspects. Thanks to the
inherently robustness of the measurement principles in the Roxar meter, there is no
need to take samples from individual wells.
5. Addressing H2S Challenges on Tengiz
In order to put in place an effective and accurate multiphase metering strategy on

Tengiz, a number of criteria had to be addressed – both at installation and as part of
ongoing operations:
Material Selection: It is important to ensure that material selection is appropriate for

the conditions in which the meter will be operating. Due to the high H2S levels seen at
Tengiz, Emerson delivered Roxar Multiphase meters in Inconel 625. All wetted parts
were considered and evaluated to ensure appropriate material selections and that pipe
integrity was maintained and assured.
Accurate Measurement: As mentioned, the Roxar Multiphase meter applies a

combination of electrical impedance measurements and single high-energy gamma for
determining phase fractions, combined with venturi and cross correlation for velocity
measurements. There were subsequently no negative effects on the performance of
the Roxar Multiphase meter due to the presence of H2S.
Providing PVT Input to the Meters: Using fluid composition analysis, TCO created
Pressure, Volume & Temperature (PVT) tables that were provided to Emerson for
input into the Roxar multiphase meters. Due to high H2S levels, TCO used a modified
equation of state with updated properties table and binary interaction parameters
(BIP), to ensure suitability to the H2S levels experienced. These PVT tables provide
4
gas, oil and water densities for use within the algorithms to provide high quality
outputs.
Direct Measurements – Mixed Density: The Roxar Multiphase meters use a number
of direct measurements, adding to the robustness and redundancy of the meter. One of
these direct measurements is mixed density measurement. As the Roxar Multiphase
meter uses high-energy gamma, there is a negligible effect on the gas attenuation
coefficient and this need not be considered with regards to measurement quality and
the effects of H2S. As the values for mass attenuation are very similar, the Caesium-
137 high-energy gamma system is insignificantly affected by variations in these
substances.
It is worth noting that depending on the measurement techniques utilized by a

multiphase meter, some may see much higher affects, degrading measurement quality
due to the presence of H2S. In particular it is worth considering the gamma
technology used, as the mass attenuation for sulphur (and other components) can be
considerably higher than for typical hydrocarbons depending on the gamma system. It
is therefore important for the operator to carefully consider the selection criteria in
order to successfully implement multiphase meters in sour service conditions.
Using Capacitance/Conductance Technology: Roxar Multiphase meters also have

direct measurements relating to the permittivity or dielectric constant of the flow,
using capacitance / conductance technology and the Clausius-Mossotti relation. The
permittivity of H2S and hydrocarbon gas are so close that H2S in gas form will be
recognized and measured as part of the gas phase, without any negative effects on the
measurement quality.
The permittivity of H2S in liquid form is in the range of 5 to 6, giving consideration to

temperature fluctuations. This is higher than the permittivity of hydrocarbon liquid
with a permittivity of 2 to 2.5. If H2S levels are high, then this can affect the
measurement quality relating to hydrocarbon liquid. The Roxar meter counteracts this
effect and maintains measurement quality by adding a compensation factor. In this
way, the H2S in liquid form will not degrade the measurement quality, and will
simply be reported as part of the hydrocarbon liquid, without negatively affecting any
other measurement parameters.
6. Ongoing Challenges on Tengiz – Addressing Meter Damage
Inevitably, some challenges have occurred over the years, with the Emerson service
team and TCO working in close cooperation to address these. Such challenges include
very high flow velocities resulting in damage to meters, unstable mixed density
measurements, differential pressure impulse line clogging, and extreme fluctuation of
daytime to nighttime and winter to summer ambient temperatures.
Damage to Meters Due to High Velocities: In 2011, three of the second-generation

Roxar Multiphase meters (MPFM 1900VI), commissioned in 2007, experienced
deviations in the capacitive sensor reading. During a visit by a Service Engineer, it
was initially suspected that this could be due to contamination of the capacitance
electrodes. Internal inspection was therefore arranged. When this was carried out, it
5
was found that each meter was missing an electrode. It is believed that this was due to
a combination of high flow velocities, vibration and perhaps solids passing through.
The meter design back then secured each electrode with one bolt and there was a very
small lip in the connection between the electrode and the insulating PEEK material in
which the electrode was installed. This presented a degree of weakness to the
conditions seen, allowed the flow to gain traction on the electrode and over time,
causing the electrode to come loose.
Addressing the Damage & Ensuring Continued Measurement: Since this time, the
design within the MPFM 1900VI was improved by splitting each electrode into
smaller sections, with a bolt per section and the electrodes being entirely flush with
the PEEK material. This is further improved within the design of the third generation
Multiphase meter (MPFM 2600) where the electrodes are all smaller in size than the
previous design.
As TCO operated three second-generation Multiphase meters with an absent

electrode, this negatively affected the permittivity measurement and resulted in the
absence of WLR measurements. It should be noted, however, that TCO operated with
very low watercut rates and these are as such also relatively stable.
As a temporary solution, the meters were supplied with fixed WLR inputs. As the
permittivity measurement is one of a number of direct measurements, by providing a
fixed WLR input, all other measurements could continue to perform as usual. This
provided TCO with continued valuable measurement from the three damaged meters.
Without the capacitance measurement and with a fixed WLR input, it is also still
possible to identify changing WLR within the Roxar Multiphase meter.
Production of water results in higher mixed density measurement and higher

temperature readings. If the WLR increases while using a fixed WLR input, the
reported GOR will be lower than expected. If we assume stable Gas Oil Ratio (GOR),
which in the case of TCO’s production was a fair assumption, then it is possible to
quantify the change in WLR by updating this in order to restore the expected GOR.
The fixed WLR input can then be updated to reflect this change, maintaining
measurement results that are representative of the production.
This highlights once again the value of a robust measurement technology where
measurement is not lost if there are issues in just one area.
In addition, TCO also then began to look into options for the replacement of the three
damaged meters. They decided to purchase three third generation (MPFM 2600)
Roxar Multiphase meters.
This also provided the opportunity for TCO to install a larger meter size in these
locations and to better accommodate the increased volume flow seen, due to the
combination of high GVF and reducing pressure. These meters were commissioned in
2014, replacing the damaged meters and restoring full measurement capabilities and
performance.
6
7. Ambient Temperature Fluctuations
The second generation Roxar multiphase meters were also found to be sensitive to the
temperature fluctuations observed between winter and summer, day and night, and
between wells with hot production and cooler production. Here, the body temperature
of the meter influenced the meter’s temperature sensor, which may not always be the
same temperature as the fluid being measured.
Initially, the GOR was seen to fluctuate between winter and summer, night and day,
and when different wells (with similar GOR, but different surface temperatures) were
flowing through the meter. During a visit by a Roxar Service Engineer, the design of
the temperature measurement within the second-generation Roxar multiphase meters
was considered. The design has a thermowell in contact with one of the electrodes
and is not directly in the flow. It was therefore concluded that given the extreme
ambient temperature changes, combined with the positioning of the thermowell in the
multiphase meters, this was not providing optimum temperature measurement.
The problem was solved by installing temperature probes in the actual flow stream at
the inlet blind-T of the meter. The temperature probes measured the correct fluid
temperature, and eliminated the temperature fluctuations caused by the meter body
temperature not fully stabilizing, or due to ambient temperature variations.
Figure 3 shows the rapid temperature stabilization after the installation of the new
temperature sensor and the resulting stability of the GOR measurement.
Figure 3: Production information for one well, showing the change in results
achieved. The point where the external temperature probes were taken into use is
pinpointed. Prior to this one can see the temperature fluctuations as measured by the
MPFM 1900VI, and the effect of this on the reported GOR. One can then see the
stable GOR after installation of a temperature measurement directly into the flow and
used as an input to the MFPM 1900VI.
7
The design within the third generation MPFM 2600 multiphase meter is no longer the
same regarding thermowell placement. The thermowell for the new meter has contact
direct with the flow, improving the reliability of the measurements achieved and
reducing any risk of ambient temperature affects.
Temperature measurements achieved on the MPFM 2600s in use on the Tengiz Oil
Field have been meeting expectations and requirements to ensure quality meter
performance.
In addition, the initial meters delivered and commissioned in 2007 were insulated and
had low power heat tracing applied. It was found that the heat tracing was not
sufficient for the conditions experienced and therefore in the period from 2010 to
2012, the heat tracing on these meters was replaced with higher power heat tracing.
The meters were delivered from Emerson in 2014 with full higher power heat tracing
and insulation. This arrangement further protects the meters from any ambient
temperature effects.
8. Mixed Density Measurement
In 2008, some issues were seen regarding the density measurement achieved by the
gamma systems on a number of the Roxar Multiphase meters. The Roxar Service
Engineer therefore suggested performing calibrations on the relevant gamma systems,
which was completed. This was a two-point calibration with gas or air and portable
water. However the calibration was not seen to resolve the issue.
The log files were seen to show some time-outs relating to the internal
communication link between the flow computer and the gamma pulse interface unit.
Other internal communication links were not showing any time-outs and it was
therefore possible to identify the gamma pulse interface unit as the probable issue.
At the time of the TCO delivery, Emerson was in the process of updating the gamma
pulse interface unit, however this was not yet complete. TCO therefore received a
modified version, designed to work with previous software. The work was completed
within Emerson to replace the gamma pulse interface unit and this was installed in
other locations without issue.
The Gamma Pulse Interface Unit was therefore replaced on one of the meters, to
evaluate if this resolved the issue. This replacement was carried out in conjunction
with a zero calibration of the gamma system - a one point calibration with gas or air.
This was found to resolve the challenges experienced and the update was therefore
performed on the other meters, restoring full measurement performance.
9. An Ongoing Dialogue: Trust and Collaboration
A key factor in the successful rectifying of challenges has been the open dialogue and
communication between TCO and Emerson/Roxar. During an intensive period from
2007 to 2009, pre-commissioning and commissioning of the multiphase meters took
place, and many service engineers were involved.
8
From 2011 onwards, Emerson changed this approach to appoint one Roxar Senior
Service Engineer to maintain the contact, follow up and support. This has proven to
be of great value to both TCO and Emerson. The Service Engineer was able to gain a
deeper understanding of TCO’s requirements, production, and installed base of Roxar
Multiphase meters. This has allowed for a systematic approach to the challenges that
have been seen through the years, ultimately ensuring strong service support and
prompt resolution to any challenges seen. The Senior Service Engineer followed up
and optimized the multiphase meters, including calibrations, performance, trouble-
shooting and any other support functions required.
This teamwork-based approach has been developed further. Between the summer of
2014 until January 2015, Emerson had a more frequent presence at TCO. This has
been achieved through a rotation plan with the appointed Senior Service Engineer and
two other identified Service Engineers rotating attendance.
This allowed the continuation of the knowledge sharing between TCO and the
appointed Senior Service Engineer, whilst also building knowledge, trust and support
with two additional Service Engineers. This not only provides TCO with more service
support from Emerson, but builds resilience into the support available with respect to
site knowledge and TCO’s measurement requirements.
John Clark, TCO advises: “The Roxar service team has provided TCO with many
quality service reps from Europe, Asia, South America and the Middle East since we
started commissioning the meters in 2007.”
“Once TCO and Roxar established continuous support from a small group of reps, we
were better able to collaborate and solve the issues we experienced at Tengiz. This
relationship has worked well, allowing TCO to trust Emerson and Emerson to trust
us.”
He concludes: “It would be nice if MPFMs were a simple plug and play device, but
we have learned that the meters require some attention to get the full benefit of the
technology.”
10. Conclusion – Meter Performance
And what of the meter’s performance?
John Clarke, TCO has been pleased with the performance of both second and third
generation meters.
He continues: “Most of our wells produce with extremely low water cuts and have the
same GOR. The Roxar meters report accurate GORs and water cuts with very little
tuning. We provide our PVT data to Roxar, and the gamma detectors are calibrated
at commissioning. The meters work very well, and help TCO easily identify any
significant changes in the water cut or the GOR. These meters are meeting our testing
needs and helping to identify problem wells.”
9
He concludes: “Together, we have created a team that has solved the problems we
encountered, and as a result we have world class MPFM measurement in Tengiz”
This paper has demonstrated the applicability of Roxar multiphase meters in a high
H2S environment. It also shows how challenges, such as unstable mixed density
measurements and extreme temperature fluctuations were addressed and solved to
deliver a sustainable and long-term multiphase metering strategy.
10
Experiences with the multiphase meter system used for
allocation of Visund South tie-in to Gullfaks C
Eirik Åbro, Jan Eskeland, Geir Sanden, Eivind Lyng Soldal, Kåre Kleppe, Asbjørn Erdal
Statoil ASA, Norway
1. Abstract
Visund South field is a subsea tie-in to Gullfaks C in the North Sea. Visund South is developed with one 4-
slots subsea template with 4 producing wells. One 10” flow line is used for transport of Visund South
production to Gullfaks C for processing and export.
Due to different owner structures between Gullfaks C and Visund South, multiphase metering is used for
ownership allocation. Subsea multiphase meters (5’’) are installed on each individual well and one topside
multiphase meter (10’’) is installed upstream the topside choke. In addition Live PVT (software) to handle
changes in the PVT is implemented [1]. As part of the Visund South project, one test separator at Gullfaks C
was upgraded with additional instrumentation according to NPD Measurement Regulations.
In this paper, the measurement philosophy for the Visund South subsea tie-in is presented. This includes the
location of multiphase meters, PVT data handling and verification of the multiphase meters. Experiences
during the project, commissioning and operational phases are described. Test results from K-lab,
comparisons of measurements from subsea meters and topside meter, and finally, verifications of multiphase
meters against test separator are presented.
The first Visund South producer was started up in November 2012. The field experiences so far show that the
build-in redundancy has been necessary due to the redesign of the topside meter. Deviations in HC mass
rates between subsea multiphase meters, topside multiphase meter and test separator measurements are
within a range of +/-1%-5% during the first 2.5 years of production.
The multiphase metering system has met our requirements in terms of measurement uncertainties.
2. Introduction
Statoil has about 250 multiphase and wet gas meters in operations, where several subsea tie-in fields have
been developed with multiphase meters for ownership allocation. Visund South subsea field is such a field,
developed as subsea productions system (SPS) where unprocessed multiphase flows are transported to
Gullfaks C (GFC) process platform through one pipeline.
The Visund South (Pan/Pandora structures) was discovered in 2008 and the field started to produce in 2012.
Visund South was the first Fast Track project in Statoil. Fast track is a method that will cut development times
in half (from about 5 years to just 2.5 years) for uncomplicated fields on the Norwegian continental shelf. Due
to standardization half development time has created profitability in small fields where commerciality has
previously been an issue.
In this paper the ownership allocation metering for Visund South is presented. It is based on multiphase
metering of the Visund South flow line production installed at GFC. Subsea multiphase meters are installed
on each of the 4 subsea producing wells.
Figure 2-1. Gullfaks C in North Sea.
Facts about Visund South
Here are some facts for Visund South [2]:
• Discovered: 2008/2009
• Production start-up: November 2012
• Location: In the North Sea between the Visund and Gullfaks fields
• Reservoir: Two structures, Pan and Pandora, with reservoir pressure of roughly 340 bar
• Volumes: 67 mboe (1/4 oil and 3/4 gas)
• Depth: 290 metres and 2900 metres below seabed
• Estimated lifetime: More than 15 years
• Visund South is located 10 kilometres southwest of the Visund platform and approximately 10
kilometres northeast of GFC. The water depth in the area is about 290 metres. Visund South is
developed with a subsea template tied to GFC.
• The reservoirs lie at a depth of 2 800 - 2 900 metres and contain oil and gas in Middle Jurassic
sandstones in the Brent Group.
• Visund South is produced by pressure depletion.
• The wellstream is routed to GFC for processing and export.
Company name Nation Company
code share [%]
Statoil Petroleum AS NO 53.200000
Petoro AS NO 30.000000
ConocoPhillips Skandinavia AS NO 9.100000
Total E&P Norge AS NO 7.700000
Table 2-1. Licensees – Visund South current
Company name Nation Company

code share [%]
Statoil Petroleum AS NO 51.000000
Petoro AS NO 30.000000
OMV (Norge) AS NO 19.000000
Table 2-2. Licensees – Gullfaks current
3. Visund South allocation metering

The Visund South field was planned developed with three production wells in a four slot template. A 4th well
was decided drilled after start-up of the field. Each well is equipped with a subsea multiphase flow meter
installed in a retrievable flow control module.
Figure 3-1. Visund South metering concept.
Production from Visund South is commingled with the production from GFC. Visund South and GFC are
separate licenses with different ownership; hence ownership allocation of the Visund South production is
required. The commingled Visund South production is routed directly to the GFC Platform. At GFC one
multiphase flow meter (MPFM) with 100% total capacity is used as primary multiphase meter for the Visund
South allocation. The subsea multiphase flow meters are used as backup for the ownership allocation. In
addition the subsea multiphase meters are essential for well allocation and production optimization.
The production from GFC is determined as the difference between the single phase export measurements
and the allocation measurement of Visund South. Hence, any mismeasurement of Visund South will influence
the allocated HC mass to GFC.
The flow through the topside MPFM can be routed either to the production header or the test separator for
verification.
The initial production profiles showed that the individual wells covered large GVF spans, from about 40% to
99%. As a consequence it was required to select a multiphase flow meter technology which automatically
covered both multiphase and wet gas conditions and had a good measurement performance in both domains.
In addition, measurements of salinity and formation water detection were important to ensure a robust flow
assurance, which minimizes potential hydrate and scaling risks.
3.1. Ownership allocation and metering philosophy

Due to space and weight restrictions at GFC, it was not regarded feasible to install an inlet separator with
single phase instruments for the Visund South production. In addition, it was not economically feasible.
Instead, it was decided to use multiphase metering for ownership allocation of Visund South.
Subsea multiphase flow meters were introduced as backup flow meters which reduced the verification
frequency of the topside multiphase meter to a minimum.
3.2. Measurement uncertainty requirements of MPFM

In Table 3-1 the uncertainty estimation of HC mass rate for Visund South is given. For subsea MPFM, the
estimation is given for all four meters in total.
Level Method Estimated HC mass rate

uncertainty
1 Test separator 1%
2 Topside MPFM 5%
3 Subsea MPFM 7%
Table 3-1. The estimated HC mass rate uncertainty levels

The project specific measurement uncertainty requirements are given in Table 3-2. The requirements have
been given separately for the multiphase and wet gas domain and are valid for subsea and topside meters:
Multiphase Wet Gas

Parameter
(GVF < 95%) (GVF > 95%)
WLR Uncertainty Uncertainty
Hydrocarbon mass rate 0 - 100 % 4% 4%
Gas volume flow rate 0 - 100 % 5% 5%
Condensate volume flow rate 0 - 50 % 7%
50 - 75 % 10 %
75 - 85 % 15 %
Water Liquid Ratio (WLR) 0 - 100 % 2%
Water Volume Fraction (WVF) 0 - 100 % 0.05 %
Salinity (wt% abs) 0 - 100 % 1% 1%
Table 3-2. Project specified measurement uncertainty requirements (95% confidence level).
3.2.1 Test separator upgrade

As part of the Visund South project, the test separator instrumentation was upgraded. Two single phase flow
meters were installed at the oil leg; one 4’’ 5-path ultrasonic meter and one 4’’ turbine meter in series. Further
a clamp-on gamma densitometer and a 4’’ water-in-oil meter were installed. For the gas phase it was installed
one 8’’ 4-path ultrasonic meter with a densitometer installed in a bypass loop.
3.3. Subsea multiphase flow meters

Based on the production profiles 5” subsea multiphase flow meters were selected. The subsea multiphase
flow meters were installed upstream the production choke and the continuous chemical injection point. The
meters were installed vertically in the flow control module with the flow direction upwards through the meters.
Communication between the subsea meters and topside is bi-directional by optical fiber. This allows for
running diagnostic communication in addition to obtaining frequent measurements and PVT data are updated
frequently.
Figure 3-2. Visund South subsea schematics
3.4. Topside multiphase flow meter

One 10” multiphase flow meter covering full capacity of Visund South production was installed on GFC. This
solution was preferred due to space constraint at GFC, instead of two smaller topside meters in parallel [1].
Figure 3-3. Model of Visund South 10” topside MPFM

3.5. Method of Visund South MPFM verification
The subsea multiphase flow meters are redundant measurement to the topside multiphase meter. As long as
deviation between the total subsea measurement and topside measurement are within certain limits, initially
set to +/-10%, verification of the subsea multiphase meter towards the test separator may be waived. The
verification of the topside meter against test separator can be limited to cases where there are discrepancies
between the sum of subsea meters and the topside meter.
Because of long stabilization time due to the length between subsea and topside meters, verification and
comparison is only performed when the production is stable. During start-up of new wells, the production is
routed to test separator thus the actual subsea meter are compared to test separator measurements.
3.6. System implementation and supervisory metering system

All measurements from the multiphase flow meters are in actual condition, and all measurements are
transmitted to the supervisory metering system and control system.
Upgrading of the GFC supervisory metering system was required in order to facilitate:
• PVT calculation of MPFM rate measurements and densities in actual condition to standard condition
• PVT calculations of required composition for MPFM input data
• PVT calculations on the test separator measurements for comparison of MPFMs
• Software for calibration of k-factor for topside multiphase meter
Flow weighted PVT data to topside MPFM is implemented, thus the metering system calculate the overall
PVT data based on the individual measured flow rates. Updated PVT data (densities) are downloaded to the
topside MPFM regularly. This is known as LivePVT [1].
The subsea multiphase flow meters receive the calculated densities based on PVT data by automatic down-
load from the supervisory metering system.
Secured remote access to the multiphase meters and the metering system is implemented. This gives fast
diagnostics and service on the metering system from the vendor which requires less travelling to the field by
service personnel.
3.7. Topside MPFM technical challenge

During FAT testing of the topside multiphase flow meter it was detected a small leakage in one of the
microwave probes during hydrostatic testing. After the FAT the meter was scheduled for flow loop test at K-
lab.
As the time was critical with respect to start-up, it was decided to continue with the K-lab test. Subsequent risk
analysis called for a redesign of the probe seal arrangement and it was decided to manufacture a new topside
flow meter, thus the original flow meter was rejected.
Due to the probe redesign it was decided to test the new flow meter at K-lab. However, it was not possible to
have the new flow meter installed and ready for use at start-up of the field.
4. Full scale flow loop test of MPFM

It is a Statoil requirement to perform a flow test of multiphase flow meters before installation, by executing a
flow test of minimum one flow meter in a batch of identical meters.
In this case identical subsea multiphase flow meters were also used in several fast track projects running in
parallel; hence a combined flow test of the subsea flow meter was performed covering a total of 159 test
points from four different projects.
In total three K-lab flow tests were performed for Visund South:
1. Subsea multiphase flowmeter (August 2011)
2. Topside multiphase flowmeter (original – June 2012)
3. Topside multiphase flowmeter (redesigned – May 2013)
4.1. K-lab
The K-Lab test facility is located at the Kårstø gas plant in Norway. Kårstø is a nodal point for natural gas
being produced in the North Sea, treating the gas before its being exported to Europe.
The K-lab facility is especially made for testing of wet gas and multiphase flow meters, as well as pumps,
separators, etc, at realistic conditions. The test site uses gas, condensate and water (salt and fresh), and
operates at pressure up to 130 bars, and temperature up to 60 degC.
K-lab capacity Unit Range

Pressure Barg 30 - 130
Temperature degC 30 - 60
GVF % 30 - 100
WLR % 0 - 100
Liquid flow rate Am3/h 0.01 - 150
Gas flow rate Am3/h 40 - 2000
Table 4-1. K-lab capacities

Test section
Figure 4-1. K-lab schematics.
4.2. K-lab test of subsea flow meter
Qg vs Qo GVF vs WLR
140 100
Production profile K-lab Production profile K-lab
90
120
80
100 70
Qo [Am3/h]
60
WLR [%]
80
50
60
40
40 30
20
20
10
0 0
0 200 400 600 800 1000 1200 1400 1600 30 40 50 60 70 80 90 100
Qg [Am3/h] GVF [%]
Qhc vs Qhc-ref Qg vs Qg-ref

140
2500
120
MPFM gas volume rate [m3/h]
2000
MPFM HC mass rate [t/h]
100
80 1500
60
1000
40
500
20
0 0
0 20 40 60 80 100 120 140 0 500 1000 1500 2000 2500
Reference HC mass rate [t/h] Reference gas volume rate [m3/h]
Figure 4-2. Test matrix and test results of subsea MPFM.

The test points follow the production profiles as close as possible. Test results of the subsea MPFM
demonstrate acceptable performance, where 90% of the test points where within the acceptance criteria. It is
noted a bit more spread in the data, at HC mass rate about 70 t/h, which corresponds to a WLR near 50% or
oil-water continuous transition point.
The test was conducted for 60 bar and 120 bar with temperature at 55C for both pressures.
4.3. K-lab test of topside flow meter

For the topside meter, the original topside MPFM was tested at two pressures, 50 bar and 130 bar. The
temperature was 60C for both pressure. The test results of gas volume flow and HC mass rate are within
acceptance criteria, as shown in Figure 4-3.
QTotHC_Mass Qgas
250 1800
1600
200
MPFM gas volume rate[m3/h]

1400
1200
150
1000
100 800
600
50 400
50 bar
50 bar
130 bar 200
0 130 bar
0 50 100 150 200 250 0
0 200 400 600 800 1000 1200 1400 1600 1800
Reference HC mass rate [t/h]
Reference gas volume rate [m3/h]
Figure 4-3. Test results of original topside MPFM.
4.4. K-lab test of redesigned topside MPFM

For the redesigned topside meter, the test matrix was similar as for the original topside MPFM and tested at
two pressures, 50 bar and 130 bar, and one temperature, 60C. Also for this test, the results of gas volume
flow and HC mass rate are within acceptance criteria.
QTotHC QGas
250 2500
MPFM gas volume rate [m3/h]
200 2000
150 1500
100 1000
50 50 bar 500 50 bar

130 bar 130 bar
0 0
0 50 100 150 200 250 0 500 1000 1500 2000 2500
Reference HC mass rate [t/h] Reference gas volume rate [m3/h]
Figure 4-4. Test results of redesigned topside MPFM.

5. Field experiences
An extensive test program against the test separator at GFC has been performed in order to verify the
multiphase metering system. Verifications of subsea multiphase meters against test separator are done
frequently. There are different combinations of wells that are produced during three major test campaigns
performed with approximately one year, namely autumn 2012, autumn 2013 and autumn/winter 2014.
Different combinations of single well testing and testing of multiple wells are available. The data used in this
paper are taken from monthly production reports and logs for Visund South.
Due to different conditions (P,T) for the subsea MPFM and topside MPFM, it is in principle two possible
methods to perform the comparison; by volume rates in standard condition or in HC mass rates. By using HC
mass rate instead of oil and gas in standard volume flow rates, PVT issues due to differences in pressure and
temperature topside and subsea conditions are expected to be minimized, as discussed in [3]. In addition, the
ownership allocation for Visund South is based on the HC mass rates.
It is not possible to present all data available, but the intension here is to give a brief overview of the results
achieved for Visund South multiphase metering system.
5.1. Well V-4H during start-up in 2012

The start-up date for Visund South was 22nd November 2012, when the well V-4H was started.
Due to the lack of topside meter in the first months of production, the V-4H subsea meter was used as
primary input to the ownership allocation. The sequenced start-up wells from Visund South gave valuable
verification of individual subsea meters against the test separator.
The first data from the MPFM at Visund South is presented in Figure 5-1. Only well V-4H was producing, thus
the comparison of subsea multiphase meter against test separator was for one well. As shown in Figure 5-1,
there are only minor differences between the daily production of HC mass rate measured by the subsea
MPFM and the test separator. The difference in the figure is in a range between -1.6% to -1.9% during 24hrs
HC mass rates the first week of production. Due to the small amount of water produced, only the oil leg of test
separator was used to measure the liquid flow rates. The water-in-oil analysis showed water content in oil
between 0.02% and 0.07% the first days of production.
Figure 5-1. HC mass rate measured with subsea MPFM at V-4H compared against GFC test separator outlet
during the test campaign 29. Nov - 4. Dec 2012. Each column represents 24hrs production.
Figure 5-2. Hourly HC mass rate measured with subsea MPFM at V-4H compared against GFC test separator
outlet during a 24 hrs test campaign 30. Nov 2012.
As shown in Figure 5-2, hourly production HC rates presented for 30. November 2012 have a rather constant
offset of about 1 tonnes/hr between the subsea meter and the test separator measurements. Similar
deviations were observed for the other days during this test campaign. In relative terms, this means a
deviation of 1.5%. This is within the limit, but in order to further reduce the deviation, more detailed analysis is
required.
5.2. Single well – start-up of V-3H in 2013
In Figure 5-3, hourly production rates from well V-3H are presented. Only the well V-3H was producing and
routed to test separator at GFC. Due to the small about of water produced, only the oil leg of test separator
was used to measure the liquid flow rates.
This means that water mass rates are included in oil mass rates measurements for the test separator
measurements. The water mass rates are in a range of 0.6 tons/hr measured by the subsea V-3H meter,
which means a water mass fraction about 1.2 %.
Figure 5-3. HC mass rate measured with subsea MPFM at V-3H compared with GFC test separator outlet
during a 24 hrs test campaign 19-20. Dec 2013.
By subtracting the water mass rates measured by V-3H subsea meter from the test separator liquid
measurements, comparison of the HC mass rates measured by the subsea meter at V-3H and test separator
shows that the ratio between test separator measurements and the subsea meter at V-3H are less than 2%,
as indicated in Figure 5-4.
Figure 5-4. Ratio in HC mass rate measured with GFC test separator and subsea MPFM at V-3H during a 24
hrs test campaign 19-20. Dec 2013.
5.3. Multiple wells – V-3H and V-4H
Figure 5-5 shows the hourly production of well V-4H in addition to the production from well V-3H. The topside
MPFM and test separator measurements represent therefore the sum of the two wells. For the topside meter,
the HC mass rate is determined on an averaged 24 hrs production. The three independent measurements
show a very good correlation with minor spread in the data.
Figure 5-5. HC mass rate measured with GFC test separator, sum of subsea MPFMs and topside MPFM with
production from V-3H and V-4H during a test campaign 21-22. Dec 2013.
When only V-3H well is producing, the deviation is an offset about 1%. The offset is about -2% when V-3H
and V-4H wells are producing. This means a shift in the offset which may be a result of changes in the
commingled PVT composition.
The subsea multiphase meters indicate a water mass rate of 0.6 tons/hr and 6 tons/hr for the V-3H well and
V-4H well, respectively. Only the oil leg of test separator is used to measure the liquid flow rates. The water-
in-oil measurements show no water content in the oil leg.
5.4. Comparison of subsea and topside multiphase meters

On a daily basis the mass balance between HC flow rates subsea and topside is the first stage to verify that
the measurement system is working according to the +/-10% limit . If comparison of the topside meter and the
sum of subsea meters are within +/- 10%, this is a good indication that both the topside meter and subsea
meters are working well.
Total mass flow rates will also include water, thus comparison of total mass flow subsea and topside will
provide useful information about the water rate measurements, but so far the water production at Visund
South in this phase has been limited.
Figure 5-6. Daily reported HC mass measured with topside MPFM and sum of subsea MPFMs with
production from V-1H, V-2H and V-4H during May 2015.
In Figure 5-6 it is seen a relatively good correlation, despite an offset, between the sum of subsea meters (V-
1H, V-2H and V-4H) and the topside meter. The average deviation of HC mass rate is 5.6% for the whole
period (1.May – 31.May 2015). The average water volume pr day measured by the topside MPFM during May
2015 is about 720 Sm3/d, representing an overall water cut of 39% in average.
5.5. Comparison of topside multiphase meter against test separator
Different combinations of producing wells are routed to test separator, which enable comparisons between
test separator, topside multiphase meters and different combinations of subsea multiphase meters. With
different compositions from each well, the commingled production will have a varying and changing
composition. When measuring the commingled production with the topside MPFM, it is required to have
correct PVT input (oil density, gas density) for the topside MPFM to work properly. By implementing flow
weighted PVT calculations, based on individual production from each well, the commingled composition is
calculated using a flow weighed average in composition. Updated PVT input (oil density, gas density) is then
downloaded to the topside MPFM.
Figure 5-7. HC mass rate measured by topside multiphase meter for Visund South compared to test
separator at GFC and relative difference between test separator and topside multiphase meter for a 24 hrs
test campaign 12. Oct 2014.
Due to the transport time between subsea wells and topside, comparison of subsea MPFM during multi rate
testing against topside MPFM and/or test separator is rather challenging, and we have not identified
appropriate data to present. However, in Figure 13 HC mass rates of topside MPFM and the test separator is
shown during a 24hr test 12.Oct 2012. The figure shows that when all three wells (V1+V2+V4) are producing,
there is negligible deviation between the HC mass rates measured by the topside MPFM and the test
separator, even if there are some changes in production. The HC mass ratio between the test separator and
topside MPFM is seen to be within 1-2 % during the multi rate test.
As shown in Figure 5-7, the comparison in measured HC mass rates shows a good performance of the
topside MPFM against test separator, even after a change in production. Due to differences in the PVT
compositions from each individual well, changes in flow rates from one or several wells will result in changes
in the commingled composition. Therefore changes in production could result in measurement errors due
changes in flow rates from one or several wells, if not flow weighted based PVT input are utilized.
6. Conclusions
Visund South was the first fast track project, with start-up in 2012. The concept is a subsea development with
unprocessed flow to a processing unit. Ownership allocation was required between Visund South and GFC
production. The only economical feasible metering system for ownership allocation for Visund South, was
implementing a cost effective multiphase metering concept, which has shown encouraging results over time.
The results presented here are typically achievements for Visund South from start-up and today. Some
deviations are observed, but deviations are well within the deviation limits of +/-10% HC mass rate that trigger
corrective actions. The water production has so far been limited, thus more multiphase metering challenges
with higher water cuts are expected to come within the next few years.
7. Acknowledgment
The authors thank the Visund South license to allow publishing some of the measurement data of the Visund
South production.
8. References
1. Experience from Implementing a Metering Solution for Subsea and Topside Multiphase Flow
Metering, David E. Olaussen, FMC Technologies - Kåre Kleppe, Statoil ASA, North Sea Flow
Measurement Workshop 2014 in St. Andrews.
2. Experience of Visund South MPM subsea meters used for allocation metering by Vidar Rune
Midttveit. Presented at MPM User Forum 2013.
3. Recent field experiences using multiphase meters for fiscal allocation, North Sea Flow
Measurement Workshop 2009 in Tønsberg, Norway. By Eirik Åbro, Kåre Kleppe, Leif Jarle
Vikshåland.
4. Information sheet for Visund South, www.statoil.com

20-23 October 2015
DP Meter Diagnostics – Multiple Field Results with new

Turbulence Diagnostic Techniques
Jennifer Rabone, Swinton Technology (UK)
Andy Smith, GDF Suez (UK)
Kim Lewis, DP Diagnostics (US)
_______________________________________________________________________________________________________
1 INTRODUCTION
A comprehensive DP Meter diagnostic system (called ‘Prognosis’) has been

repeatedly proven with successful laboratory tests and field trials in the UK, USA,
Russia, Mexico, Malaysia and Middle East. It is now established with multiple
systems in industrial service. The present DP meter diagnostic system employs
“static pressure field monitoring”, i.e. the comparison of a live or averaged
pressure field with a baseline expected pressure field. New developments include
the development of “dynamic pressure field monitoring”. Named “turbulence
diagnostics”, dynamic pressure field monitoring is the analysis of the pressure
field stability over time. This analysis can reveal significant additional information
regarding the nature of the flow and metering problems. These turbulence
diagnostics, initially researched and developed in a laboratory setting, have now
been shown to provide significant benefits in the field. The turbulence diagnostics
are part of the ongoing development of the Prognosis software, driven by the end
user’s requirement to have the most useful and comprehensive diagnostic
information possible.
In 2014 GDF Suez installed Prognosis on three 12” 0.58 beta orifice meters at the
Saltend Natural Gas Power Station. This paper describes issues GDF Suez initially
had with these meters and the reasons for applying diagnostics. The results from
combining the existing static pressure field monitoring diagnostics with the new
‘turbulence’ diagnostics are shown both before and after the issues were
identified by the diagnostics and corrected. Separate previously undisclosed
blinded operator field data from in service DP meters with the diagnostic system
are also shown. This data includes the diagnostic response from DP meters with
real unintended field problems.
The addition of turbulence diagnostics is shown to significantly improve the DP

meter diagnostic capability; considerably reducing the number of potential
problems that could cause specific diagnostic patterns. Such information
significantly reduces the time and cost of maintenance and inspection activities.
These latest developments in DP meter diagnostic systems increase the resolution
of the diagnostic output providing further reductions in the risk of mis-
measurement and hence strengthening the basis for a condition based
maintenance strategy.
This paper covers:
 A review of the latest diagnostic suite,
 “Turbulence Diagnostics” test meter data,
 GDF Suez (Saltend) meters and application of Prognosis,
 Issues identified using static and dynamic pressure field monitoring

(“turbulence Diagnostics”) applied to GDF Suez field data,
 “Turbulence Diagnostics” applied to blinded operator field data.
1
20-23 October 2015
2 DP METER DIAGNOSTICS REVIEW
2.1 STATIC RESPONSE
Swinton Technology partnered with DP Diagnostics to produce the generic DP

meter diagnostic suite ‘Prognosis’ (see Steven [1, 2]). An overview of these
patented ‘pressure field monitoring’ diagnostics is given in the Appendix. For
further details the reader should refer to the descriptions given by Steven [1, 2],
Skelton et al [3] & Rabone et al [4].
Fig 1. Orifice Meter with Prognosis Set Up & Sample Pressure Field.
Figure 1 shows a sketch of an orifice meter with three DP readings and the
pressure field created by the meter. The seven diagnostic checks created by the
diagnostic suite are displayed to the operator in an easy to understand plot.
Figure 2 shows a screenshot of this Prognosis diagnostic display. The seven
diagnostics are plotted as four points (three points with x & y coordinates, and a
fourth with just an x coordinate) on a graph. A box encompasses the area where
the points must fall within if the meter is working correctly. If any of the points
are outside of this area (i.e. outside of the box) the diagnostics have identified an
abnormality in the meter’s operation. The pattern of points outside the box offer
extra information to the nature of a potential meter problem, short listing
potential issues while discounting other problems that do not cause such a
pattern.
Fig 2. Screenshot of the Prognosis Output from an Orifice Meter in Service.
2
20-23 October 2015
2.2 DYNAMIC RESPONSE
The present DP meter diagnostic suite does not directly monitor the time
dependent response of the DP signals and the associated diagnostic parameters.
The present diagnostics compares either a single result, or a time averaged result
from a given set of results recorded over a given period of time, to the fixed
expected (calibration or standard) baseline. Time is only considered in the
present DP meter diagnostic suite in two indirect ways:
 The averaging of multiple inputs read over a set span of time to give a
single averaged diagnostic output, or,
 The act of the operator manually playing back, in chronological order, the
archived individual diagnostic results in order to check for trending.
However, there is further valuable information imbedded in monitoring the three

DPs and the associated seven diagnostic checks relative to time. That is, there is
value in monitoring both the instantaneous (or averaged) ‘static’ pressure field
diagnostic output (as is presently done) and the time dependent ‘dynamic’
response of the pressure field diagnostic output. This additional DP meter
diagnostic approach can produce further discrimination regarding what adverse
operating conditions the DP meter may be exposed to.
Monitoring the DP meter’s pressure field ‘dynamic’ fluctuations is analogous to the

USM’s ‘turbulence’ diagnostics. These USM diagnostics monitor the standard
deviation (or ‘stability’) of that meter’s primary signals, i.e. the variation in
difference in time measurements. Likewise, DP meter ‘turbulence diagnostics’ is
the monitoring, cross referencing and analysis of the standard deviation (or
‘stability’) of that meter’s primary signals, i.e. the read DPs, and the associated
diagnostic parameters.
The DP meter turbulence diagnostic technique is different to the existing DP

meter diagnostic techniques in one very significant aspect. The existing
diagnostics (reviewed in the Appendix) could be described as “absolute
diagnostics”. Here, the meter output is compared to either a particular physical
law, or, a calibration result that is guaranteed to low uncertainties to be the
performance characteristic of the correctly operating meter system in the field.
Either the metering system output is shown to agree with physical law and / or
the guaranteed calibration results, or it does not agree. There is nothing
subjective about this type of diagnostic check. It is an absolute pass or fail
statement irrespective of any operator’s opinion. However, the DP meter
turbulence diagnostic methodology does not fall into this category.
The DP meter turbulence diagnostic methodology could be described as a

“relative diagnostic” method. Here, the system output is compared to the
historical system output. There is no output guarantee fixed by either a particular
physical law, or by any applicable laboratory calibration result. Unlike absolute
diagnostics, relative diagnostics are subjective. Primary DP signal standard
deviation (i.e. turbulence) levels differ (by small to moderate amounts) for
different DP flow meters and DP transmitters. Normal turbulence levels for one
meter system is not necessarily normal turbulence levels for another nominally
identical metering system, even in the same application. If a DP meter has DP
signal turbulence levels calibrated at a laboratory, even when using the same DP
transmitters the operator has no absolute guarantee that these levels will be
representative of the correctly operating meter in the field. Many influences can
affect the correctly operating system’s DP signal standard deviation, such as
different DP transmitters, slight differences in the pipe work etc. The DP meter
3
20-23 October 2015
turbulence diagnostic concept must have its baseline set by recording the values
across a known period of actual meter operation in service. This subjective
diagnostic method is therefore comparing the relative turbulence of ‘then and
now’, which is of course different to the absolute diagnostics comparing precise
performance characteristics to absolute known performance requirements.
The DP meter turbulence diagnostic method is an eighth DP meter diagnostic

check. Unlike the existing seven diagnostics it is subjective, but it does give the
meter operator extra useful information. Furthermore, by cross referencing this
relative diagnostic method with the existing diagnostic suite it is possible to
further distinguish between certain types of DP meter malfunction. The DP meter
turbulence diagnostic method development and operation is described in [5]
where an explanation is provided on how the particular problems of blocked
impulse lines and wet gas flow can be distinguished using the turbulence
diagnostics together with the static diagnostic response.
3 DP METER TURBULENCE DIAGNOSTICS TEST METER DATA
3.1 8” 0.689 BETA ORIFCE
A CEESI 8”, 0.69β orifice meter was in use with dry natural gas flow at 45 Bar(a)
flowing 46 MMSCFD. The traditional, recovered and PPL DPs were read at
approximately 84”WC, 38”WC & 46”WC. The standard deviations of these DPs
were approximately 0.41, 0.27 & 0.12 respectively. This is shown in Figure 3. At
1837 seconds into the data logging sequence the low pressure port (i.e. the mid-
stream pressure port) was blocked by the shutting of a valve on that impulse line.
The resulting effect is very obvious. The PPL DP continues to read the same DP at
the same low standard deviation. The other two DPs begin to have drifting DPs
(that follow the small natural line pressure fluctuation) and their respective
standard deviations significantly increase to 17.0. Figure 3 shows the reaction
expected for an orifice meter with blocked low pressure / midstream pressure
port. The maximum & minimum variation of the traditional DP around the correct
value of 84”WC is approximately 115”WC, which is a difference of +31”WC (or
+0.077 Bar), i.e. a line pressure variation of < 0.2%. The DP meter turbulence
diagnostic method is very sensitive to blocked impulse lines.
Fig 3. Orifice Meter Response to Blocked Mid-Stream (“Throat”) Impulse Line.
4
20-23 October 2015
The same CEESI 8”, 0.69β orifice meter was tested with dry and wet natural gas
flow. Figure 4 shows data at 45 Bar(a) & 74 MMSCFD. With dry gas the
traditional, recovered and PPL DPs were read at approximately 172”WC, 96”WC &
76”WC. The standard deviations of these DPs were approximately 0.8, 0.24 &
0.35 respectively. This is shown in Figure 5. Then, at 770 seconds into the data
logging sequence the liquid was injected at a rate of 60 bbls/MMSCFD (i.e. XLM of
0.12). As expected the DPs increased. With this wet gas flow the traditional,
recovered and PPL DPs were read at approximately 203”WC, 87”WC & 116”WC.
The standard deviations of these DPs were approximately 5.5, 1.26 & 0.58
respectively. This is also shown in Figure 5. Figure 4 shows the reaction as
expected for an orifice meter with dry and then wet gas flow. The DP meter
turbulence diagnostic method is sensitive to, and can potentially identify wet gas
flow.
For all the DP meter turbulence diagnostic method can potentially identify wet gas
flow, this relative diagnostic check is certainly is not as powerful as an absolute
diagnostic check. Examples of the standard diagnostics reaction to wet gas flow
are shown in Figure 5, using data from this same 8” orifice meter. As described in
[5] wet gas produces a strong diagnostic pattern that can be recognised as such,
the higher the liquid loading the further from the origin the diagnostic points
move. However, as this pattern is not unique to wet gas flow (i.e. there are a few
meter malfunctions that can cause this pattern) the addition of the DP meter
turbulence diagnostic method can help identify wet gas flow from the list of
possibilities.
Fig 4. Orifice Meter Response to Wet Gas Flow.
If the operator was to use this relative check alone it is subjective if this is wet
gas flow. Although all three DP values and their standard deviations rose, not all
three DPs had similar increases in standard deviation. Whereas the traditional
DP’s standard deviation increased five fold, and the recovered DP standard
deviation quadrupled the permanent pressure loss increased by just two and a
half times. If the operator did not include the main absolute diagnostic checks
then the operator could interpret the DP standard deviation shifts as showing a
5
20-23 October 2015
blocked mid-stream / throat impulse line. This is an example of the reactive

diagnostic check methods being subjective. Hence, we can cross reference this
relative check with absolute diagnostic checks.
Fig 5. Sample diagnostic response to wet gas flow with changing liquid loading.
3.2 VENTURI WET GAS TEST DATA
Figure 6 shows a DP Diagnostics 6”, 0.7β Venturi meter undergoing dry and wet
gas flow testing at CEESI. Flow is from right to left. Note the three pressure ports
and three DP transmitters. This CEESI test facility logged data once every six
seconds.
3 DP Transmitters
3 pressure ports
Fig 6. 6”, 0.7β Venturi Meter at CEESI Wet Gas Loop
Figure 7 shows the standard Venturi meter Prognosis response to wet gas flow. Note
that there is a difference in the relative distances of (x1,y1), (x2,y2), & (x3,y3) to the
origin between a Venturi meter and an orifice meter when they are exposed to wet
gas flow. This is clear if you compare Figures 5 & 7.
Figure 7 shows that the Venturi meter diagnostics are so sensitive to this
moderate wet gas flow that the box has shrunk to a dot on the origin. The liquid’s
presence causes the DPs to change and substantially increases all three DP
6
20-23 October 2015
Fig 7 (left). 6”, 0.7β Venturi Meter Wet Gas Prognosis Result.
standard deviations (see Figure 8). The standard deviations of the DPs read from
a Venturi meter in wet gas service reacted in the same way as they would if read
from an orifice meter in wet gas service, i.e. they increase.
This is another example of DP meter turbulence diagnostics. The existing

diagnostic result in Figure 7 is averaged data and the pattern is the pattern
produced by wet gas. In this particular example, the wet gas flow had a gas flow
rate of 3.28kg/s, a Lockhart-Martinelli parameter of 0.01, a liquid flow 0.17 kg/s,
a GVF of 99.82% and the over-reading was +5.84%. However, the pattern is not
unique to wet gas flow. There are a few other problems that could cause such a
pattern. However, these other problems do not tend to have high DP standard
deviations. Wet gas produces the additional signature of DPs with high standard
deviations. By combining the existing diagnostic suite and DP meter turbulence
diagnostics it is possible to identify wet gas as the most likely causes of the issue.
7
20-23 October 2015
Figure 8 shows the average standard deviations increasing with Lockhart-

Martinelli parameter. Table 1 shows the DPs and corresponding standard
deviations using one particular wet gas test point.
Fig 8. 6”, 0.7β Venturi Meter Data from CEESI Flow Tests.
Lockhart-Martinelli Parameter (XLM) 0 (Dry) 0.1 (Wet)

Traditional DPt (kPa) 4.752 7.086
Standard Deviation of DPt (%) 0.142 1.062
Recovery DPr (kPa) 4.405 3.912
Standard Deviation of DPr (%) 0.154 3.158
PPL DPppl (kPa) 0.353 3.189
Standard Deviation of DPppl (%) 0.951 2.448
Table 1. 6”, 0.7β Venturi Meter example data from CEESI Flow Tests.
As with the orifice meter wet gas flow example all three Venturi DP standard
deviations rose, but they did not have proportional increases. The PPL DP
standard deviation increase was significantly less than that of the other two DPs.
Hence, as a stand-alone diagnostic tool the turbulence diagnostics could be
falsely interpreted as showing a blocked mid-stream / throat impulse line. This is
an example of ‘relative’ diagnostic check methods being subjective. With
Prognosis, however, we can cross reference this relative check with absolute
diagnostic checks. This means we can be confident of the correct prognosis.
4 PROGNOSIS FIELD DATA INCLUDING TURBULENCE DIAGNOSTICS
DP Diagnostics & Swinton Technology have been monitoring turbulence

diagnostics in various applications since 2014 in order to investigate whether it
was a viable and useful practical addition to the standard Prognosis suite. In this
section three field examples of turbulence diagnostics as part of Prognosis are
discussed. The first example shows data from the first orifice meters to have
Prognosis installed and the associated turbulence diagnostics monitored. In this
example the standard Prognosis system identified a real problem, although the
turbulence diagnostics were not the relevant diagnostic checks for the particular
issue the meter was found to have. The following two examples go on to show
real world case studies where the Prognosis suite identified problems where the
turbulence diagnostics were a beneficial addition to the existing diagnostic suite.
8
20-23 October 2015
4.1 GDF SUEZ SALTEND POWER STATION
Saltend is a natural gas fuelled power station located in the East Riding of
Yorkshire (Figure 9). GDF own 75% of the power station which has the capacity
to generate 1,200 MW, of which 100 MW is allocated to supply the adjacent
Saltend Chemical Park. The valuable by-product of 240 tonnes/hour of steam is
sold to Saltend Chemical Park to use in their process. This makes Salt End one of
the most efficient power stations in the UK. The natural gas to fuel the power
station is metered by an Above Ground Installation (AGI). Within the AGI are 5
gas orifice metering streams; three 50% streams and two smaller 10% streams.
Under normal operation gas flows through two of the (12”) 50% streams.
GDF had been experiencing

orifice plate fouling,
possibly due to compressor
oil carryover. GDF were
aware that Prognosis would
monitor for such issues. In
2014 Prognosis was
installed on the three 12”
orifice meters to allow a
condition based monitoring
approach. The Prognosis
Saltend system installation was
part of a metering
computer system upgrade.
Fig 9. (Left) Geographic

location of GDF Saltend
power station.
Figure 10 shows stream 1 of the 5 stream metering skid (with the other 4
streams behind). The instrument enclosures housing the traditional meter DP
Transmitters (medium and high range) and the ‘diagnostic’ DP Transmitters (DPr
and DPppl) are indicated. Meters 2 and 3 have the same instrumentation
arrangement with all DP readings accommodated in the stream flow computers.
‘Traditional DP’
transmitter
housing
‘Diagnostic DP’
transmitter
housing
Fig 10. Saltend AGI gas metering skid with instrument enclosures.
9
20-23 October 2015
The Prognosis software was integrated with the metering supervisory computer in
order that diagnostic alarms be produced from the MSC and data was acquired
from flow computers via Modbus.
The position of the available downstream pressure tap was further downstream
than the standard 6D so the standard Prognosis correction for the additional PPL
was duly applied.
Figure 11 shows a typical static (or averaged) response observed for meter 1
together with a multiple data plot showing the ‘spread’ of the points over a
randomly chosen hour long period. This data was recorded in April 2014 at a
sample rate of once per 30 seconds. The diagnostic calculations are performed
once every second using process data averaged over the previous 20 seconds.
Diagnostic variances (see Appendix) were set to a common default level of x =
1%, y = 2.5%, z = 2.5%, a = 3%, b = 2.5%, c = 4%, θ = 1%. During this test
period the flow rates slowly varied between 9 to 18 kg/s (i.e. the DPt ranged from
60 to 211 mbar). The actual meter performance matches the expected meter
performance and no alarms were raised (i.e., no points lay outside of the box).
Fig 11. Meter 1 averaged Prognosis response single plot (left) and multiple
individual data plot (right)
However, both meters 2 and 3 initially saw problems, inclusive of ‘DP Integrity’
issues; that is the ‘DP Summation’ diagnostic check indicated a problem. The
‘inferred DPt’ (i.e. the sum of the recovered and PPL DPs) was higher than the
measured DPt by >60%. Figure 12 shows Meter 3’s repetitive static Prognosis
responses over time. This states there was a problem with the DP readings, and
suggests that the problem was with more than one DP transmitter.
Figures 13 and 14 show standard deviation vs. time plots for meters 1 and 3.
Both meters have standard deviations of traditional, recovered and PPL DPs that
shadow each other, i.e. they tend to trend the same, increasing and decreasing in
synchronization. This is what would be expected when turbulence diagnostics
monitors healthy DP readings. However, we know from the standard Prognosis
suite on Meter 3 that there is a problem with Meter 3’s DP readings. From
examining Figure 14 further it is clear that the standard deviations of Meter 3’s
recovered and PPL DPs may have the correct trending with the traditional DP
standard deviation, but they are far smaller than the traditional DP standard
deviation. However, in this case there is not enough information in the turbulence
diagnostics to make any further conclusions. Industry does not know enough
10
20-23 October 2015
about relative DP standard deviations to make any defensible predictions about

the meaning of this result.
Fig 12. Meter 3 initial response showing DP integrity error (Meter 2 similar).
Fig 13. Meter 1 DP standard deviations over an hour 00:00 to 01:00.
In this real example, it is the standard Prognosis suite that states the DP readings
are wrong. On investigation GDF found that the Meter’s 2 & 3 had incorrect
scaling applied to the recovered and PPL DPs. On correction Meter 2’s diagnostic
response showed the metering system was operating correctly. However, Meter
3’s standard Prognosis response still showed a smaller but still significant DP
integrity alarm. Figure 15 shows the averaged static Meter 3 response and a
corresponding multiple data plot over a minute period. The sampling rate had
been increased to 1 per second.
11
20-23 October 2015
Fig 14. Meter 3 DP standard deviations over an hour 00:00 to 01:00.
Fig 15. Meter 3 Prognosis response following DP scaling correction, single plot
(left) and multiple data plot (right)
The ‘DP Sum’ diagnostic result of +1.6 means that the measured DPt is 1.6%
lower than the sum of the measured DPr and DPppl. If the traditional DP was
under-read by -1.6%, then the flow rate would be under-predicted by -0.8%.
GDF investigation found a wiring issue in the instrument panel which was causing
a small bias in DP measurements being received by the flow computer (and
subsequently by Prognosis). Figures 16 and 17 show the diagnostic response
once this issue was rectified. The ‘static’ diagnostic response shows a healthy
meter and the ‘turbulence’ diagnostics show the response expected for a healthy
meter.
This example shows that GDF made a reasonable decision to install Prognosis on
these three orifice meters. Prognosis correctly identified a real world problem with
the DP measurement. This DP error on Meter 3 would have resulted in a small but
significant long term metering error that is unlikely to have been noticed without
the presence of the diagnostic system.
12
20-23 October 2015
Fig 16. Meter 3 Prognosis response over 2 hours following correction of DP

integrity issue
Fig 17. Meter 3 Prognosis “turbulence diagnostics" response over 2 hours

following correction of DP integrity issue
4.2 BLINDED 10” 0.4 BETA VENTURI FIELD DATA
A 10” 0.4 beta Venturi meter was supplied to the field with Prognosis. The
meter’s calibration had included setting the Prognosis baseline characteristics.
The operator was not initially certain if the production would be dry or wet gas
flow. Prognosis identified wet gas flow from the outset. Figure 18 shows an
averaged data plot of the diagnostic response over five days. Figure 19 shows the
massed individual data plots that produced Figure 18. The data was gathered at
half hour intervals over the course of 7 days.
13
20-23 October 2015
Fig 18. 10” 0.4 beta Venturi average wet gas Prognosis data over a 7 day period
The static response produces a pattern (Figure 18) for which wet gas flow (e.g.
see Figure 7 in section 3.2) is one of a small list of problems that can cause such
a pattern, including wrong meter geometry and a partial blockage of the Venturi
meter throat. Figure 19 shows that the diagnostic response of (x1,y1), (x2,y2), &
(x3,y3) is very unsteady, while (x4,0) indicates that the DPs are being read
correctly. The combination of the standard ‘static’ Prognosis result and the
dynamic ‘turbulence’ diagnostic result narrowed down the problem to be wet gas
flow. Due to the Prognosis prediction the operator conducted a detailed study and
confirmed that the gas flow was wet.
14
20-23 October 2015
Fig 19. 10” 0.4 beta Venturi massed wet gas Prognosis data over a 7 day period
Figure 20 shows the three DPs’ standard deviations for the same (7 day) period.
The DP standard deviations are unsteady. Different periods of time produce
significantly different levels of standard deviation amongst the DPs. Figure 20
splits the time period into four distinct sections where different turbulence
diagnostic response was observed.
15
20-23 October 2015
Section 1 Section 2 Section 3 Section 4
Fig 20. 10” 0.4 beta Venturi wet gas Prognosis data (turbulence diagnostics)
Sections 1 & 4 show a typical light liquid loading wet gas flow turbulence
diagnostic response. The corresponding static / averaged Prognosis plot produced
the typical unsteady wet gas pattern, i.e. see Figure 21. (Section 4 gave a similar
response to Section 1). Section 2 show a much heavier (and unsteady) wet gas
flow. The corresponding static / averaged Prognosis plot produced the typical
very unsteady wet gas pattern, i.e. see Figure 22. Note that Figure 22 has a
wider spread of points than Figure 21. Section 3 shows a much drier flow. The
corresponding static / averaged Prognosis plot produced the typical relatively
steady wet gas pattern, i.e. see Figure 23. The flow is still wet, but it is not as
wet. This is evident from both the static diagnostic response (i.e. the relative
closeness to the origin of the points), and the relative stability of the points (i.e.
the relatively small spread of results).
Therefore, by using turbulence diagnostics as an eighth diagnostic check, the

Operator is provided with more useful information than with the standard ‘static’
diagnostics alone.
16
20-23 October 2015
Fig 21. 10” 0.4 beta Venturi wet gas Prognosis data#1
17
20-23 October 2015
18
20-23 October 2015
4.3 BLINDED 10” 0.5 BETA ORIFICE FIELD DATA
Prognosis was applied to 4 x 10” and 2 x 12” orifice meters at a hydrogen facility.
The 4 x 10” (0.5 beta) meters were installed on a 2 stream skid measuring
hydrogen rich fuel gas with a ‘pay’ and ‘check’ orifice meter on each stream.
Fig 24. 10” 0.5 beta Orifice Meter 4 averaged field data over a ten minute period
Fig 25. 10” 0.5 beta Orifice Meter 4 individual field data over a ten minute period
Figures 24, 25 and 26 show the initial static & dynamic Prognosis response of
Meter 4 (the ‘check’ meter on metering Stream 2) over a ten minute period with
data sampled once every 10 seconds. No issues are detected. The operator has
confirmation that meter is operating correctly.
19
20-23 October 2015
Fig 26. 10” 0.5 beta Orifice Meter 4 field data (turbulence diagnostics)
Figures 27, 28 and 29 show the initial static & dynamic Prognosis response of
Meter 3 (the ‘pay’ meter on metering stream 2). The standard static diagnostic
response indicates that the meter has a problem. It also indicates that at least
part of the source of the problem is with the integrity of the DP readings. As
(x3,y3) mostly stays inside the NDB while the other points do not it can be
inferred that the most likely DP reading to be in error is the traditional DP. The
standard Prognosis software states this to the operator.
Figure 29 shows the corresponding dynamic turbulence diagnostic result. It is

immediately evident that the recovered and PPL DP standard deviations are
synchronized, as they usually are for a healthy DP metering system. However,
the traditional DP standard deviation is not synchronized with the other two DP
standard deviations. This is abnormal for a fully serviceable DP meter. The
dynamic ‘turbulence’ diagnostics were therefore independently verifying the
standard static Prognosis systems warning that the traditional DP transmitter
system requires checking (i.e. “Condition Based Maintenance” or CBM).
Fig 27. 10” 0.5 beta Orifice Meter 3 averaged field data over a ten minute period
20
20-23 October 2015
Fig 28. 10” 0.5 beta Orifice Meter 3 field data over a ten minute period
Fig 29. 10” 0.5 beta Orifice Meter 3 field data (turbulence diagnostics)
On being prompted by Prognosis, site technicians carried out “CBM”. This CBM
was directed immediately at the traditional DP transmitter as the diagnostic
system had guided them there. The traditional DP transmitter system was
discovered to have contamination in the impulse lines. The impulse lines were
subsequently cleared of contamination.
Figure 30 shows the Meter 3 standard static diagnostic response observed once
the issue was resolved. Prognosis now shows that the problem has been fixed and
the meter is now fully serviceable. (The authors were not provided with any
subsequent recorded diagnostic data aside from this one static response but were
informed after cleaning that the static response was very stable and it may be
assumed that all three DP reading standard deviations were small and
synchronized with each other).
By applying Prognosis, the Operator can monitor for this (and any other) kind of
issue recurring; with the ‘turbulence diagnostics’ providing additional useful
information as to the nature of any issue. In this case study Prognosis correctly
identified a meter malfunction due to impulse line contamination that is likely to
21
20-23 October 2015
have gone unnoticed without the presence of the diagnostic system. In this real
world case the operator discovered via static and dynamic Prognosis diagnostics
that although the traditional DP reading was untrustworthy, the recovered and
PPL DP readings were trustworthy. This allowed the operator to use archived data
to calculate the correct historical flow rates and compare it to the traditional
meters output. That is, in this particular case study the actual flow error was
derivable from Prognosis. The actual flow rate prediction error induced in this
case remains confidential to the operator.
Fig 30. 10” 0.5 beta Orifice Meter 3 field data once issue resolved
6 CONCLUSIONS
DP meter diagnostics based on pressure field monitoring is now becoming widely

accepted by the natural gas production industry. As a result of this the software
‘Prognosis’ is now used by multiple operators and has been tested by 3rd parties
globally.
The existing generic DP meter pressure field monitoring diagnostic suite is a very
capable diagnostic tool. This existing diagnostic suite (without DP meter
turbulence diagnostics) allows the system to identify a DP meter malfunction. The
particular diagnostic pattern (i.e. combination of all diagnostic results together)
which shows a meter malfunction also allows a ‘short list’ of potential problems to
be produced, whilst discounting the malfunctions that could not produce that
pattern. DP Diagnostics and Swinton Technology are continuing to develop and
strengthen this diagnostic system.
The addition of DP meter turbulence diagnostics allows further discrimination of

the diagnostic result. The present short list of possible malfunctions that could
produce that diagnostic pattern will be further reduced as DP meter turbulence
diagnostics rule out certain possibilities on that list and highlight others as still
possible.
DP meter diagnostics have advanced significantly over a relatively short period of

time. However, the possibilities offered by pressure field monitoring are so wide
22
20-23 October 2015
and diverse that it is expected that the DP meter diagnostic suite will continue to
expand and improve steadily for the foreseeable future.
7 REFERENCES
1. Steven, R. “Diagnostic Methodologies for Generic Differential Pressure Flow

Meters”, North Sea Flow Measurement Workshop, October 2008, UK.
2. Steven, R. “Significantly Improved Capabilities of DP Meter Diagnostic

Methodologies”, North Sea Flow Measurement Workshop, October 2009, Norway.
3. Skelton M. et al, “Developments in the Self-Diagnostic Capabilities of Orifice

Plate Meters”, North Sea Flow Measurement Workshop, October 2010, UK.
4. Rabone J. et al, “DP Meter Diagnostic Systems – Operator Experience”, North

Sea Flow Measurement Workshop, October 2012, UK.
5. Rabone J. et al, “Advanced DP Meter Diagnostics – Developing Dynamic

Pressure Field Monitoring (& Other Developments)”, North Sea Flow Measurement
Workshop, October 2014, UK.
8 APPENDIX
Fig 1. Orifice meter with instrumentation sketch and pressure field graph.
Figure 1 shows a sketch of a generic DP meter and its pressure field. The DP
meter has a third pressure tap downstream of the two traditional pressure ports.
This allows three DPs to be read, i.e. the traditional (ΔP t), recovered (ΔPr) and
permanent pressure loss (ΔPPPL) DPs. These DPs are relate by equation 1. The
percentage difference between the inferred traditional DP (i.e. the sum of the
recovered & PPL DPs) and the read DP is δ%, while the maximum allowed
difference is θ%.
DP Summation: Pt  Pr  PPPL , uncertainty ± % --- (1)
m trad  f t Pt  ,
.
Traditional flow calculation: uncertainty ± x% --- (2)
m exp  f r Pr  ,
.
Expansion flow calculation: uncertainty ± y% --- (3)
m PPL  f PPL PPPL  , uncertainty ± z%

.
PPL flow calculation: --- (4)
Each DP can be used to meter the flow rate, as shown in equations 2, 3 & 4. Here
. . .
mtrad , mexp & m PPL are the mass flow rate predictions of the traditional, expansion &
PPL flow rate calculations. Every DP meter is three flow meters in one body.
23
20-23 October 2015
Symbols f t , f r & f PPL represent the traditional, expansion & PPL flow rate calculations
respectively, and, x% , y % & z % represent the uncertainties of each of these flow
rate predictions respectively. Inter-comparison of these flow rate predictions
produces three diagnostic checks. The percentage difference of the PPL to
traditional flow rate calculations is denoted as  % . The allowable difference is
the root sum square of the PPL & traditional meter uncertainties,  % . The
percentage difference of the expansion to traditional flow rate calculations is
denoted as  % . The allowable difference is the root sum square of the expansion
& traditional meter uncertainties,  % . The percentage difference of the expansion
to PPL flow rate calculations is denoted as  % . The allowable difference is the
root mean square of the expansion & PPL meter uncertainties,  % .
Reading these three DPs produces three DP ratios, the ‘PLR’ (i.e. the PPL to
traditional DP ratio), the PRR (i.e. the recovered to traditional DP ratio), the RPR
(i.e. the recovered to PPL DP ratio). DP meters have predictable DP ratios.
Therefore, comparison of each read to expected DP ratio produces three
diagnostic checks. The percentage difference of the read to expected PLR is
denoted as  % . The allowable difference is the expected PLR uncertainty, a% .
The percentage difference of the read to expected PRR is denoted as  % . The
allowable difference is the expected RPR uncertainty, b% . The percentage
difference of the read to expected RPR is denoted as  % . The allowable
difference is the expected RPR uncertainty, c% . These seven diagnostic results
can be shown on the operator interface as plots on a graph (see Figure 2). That
is, we can plot the following four co-ordinates to represent the seven diagnostic
checks:
 %  % , % a% ,  %  %,  % b% ,  %  % , % c% &  %  % ,0 .
For simplicity we can refer to these points as (x1,y1), (x2,y2), (x3,y3) & (x4,0).
3 flow rate
predictions
Fig 2. Display showing NDB and diagnostic results (good meter performance)
Dividing the seven raw diagnostic outputs by their respective uncertainties is

called ‘normalisation’. A Normalised Diagnostics Box (or ‘NDB’) of corner
24
20-23 October 2015
coordinates (1, 1), (1,-1), (-1,-1) & (-1, 1) can be plotted on the same graph
(see Figure 2). This is the standard user interface with the diagnostic system
‘Prognosis’. All four diagnostic points inside the NDB indicate a serviceable DP
meter. Any point outside of the NDB indicates a possible meter system
malfunction and potential measurement bias (Figure 3).
By analysing the ‘static’ diagnostic response continually and in real time, the
Prognosis software will automatically provide a system alarm when any point is
outside of the NDB for longer than a configurable ‘alarm delay’ period and also a
‘shortlist’ of possible issues which are known to cause the observed response. In
some cases (e.g., DP instrumentation issue), the software is able to tell the end
user specifically what issue exists. In the case of an issue with the ‘traditional’
meter DP reading the diagnostic system’s flow rate prediction over-determination
provides two alternative flow rate predictions.
DPt Integrity
Alarm
Fig 3. Operator display showing NDB and diagnostic results (with DP transmitter
error)
Using the existing analysis of the ‘static’ response together with the new
‘dynamic’ response provides even more useful information in identifying a meter
system malfunction.
25

NSF MW 2015 Technical Papers

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

NSF MW 2015 Technical Papers

Загружено:

Авторское право:

Доступные форматы

2015

Could Allocation be Rocket Phillip Stockton and Euain Accord 420

m g = EAt C d ε 2 ρ g ∆Pg --- (5) mg apparent ≈ EAt Cd ε 2ρ g ∆Ptp --- (6)

mg Apparent ε tp Cd ,tp ∆Ptp ∆Ptp

Fig 1. Gas/HCL stratified. Fig 2. Gas/HCL transition. Fig 3.Gas/HCL annular.

Fig 17. 4” Venturi Meter Fig 18. 2” Venturi Meter

Parameter CEESI Test Range ISO TR 11583 Stated Limits

Fig 21 Ultraflow Daniel Ultrasonic Meter Wet Gas Data

Rotated -- Low Press.

Fig 27. Data from 8” Clamp-On Bounce Path Ultrasonic Meter

Fig 33. Data from 8”, 4 Chordal Path Ultrasonic Meter.

Fig 37. 4” Chordal 4 Path Data Set, Separated Density Ratio.

Fig 43. Modified De Leeuw Sketch Showing DP Readings.

Jan Peters, Rick de Leeuw (Shell)

Joe Shen (Chevron)

DNV GL Oil & Gas

2 MULTIPHASE FLOW FUNDAMENTALS FOR WET GAS

2.1 Application to steady two phase gas-continuous flow

The total set of dimensionless numbers is

3.1 Test section design

Figure ‎3-2 : Overview of JIP test setup at DNV GL test facility

3.2 Test matrix

Re g , Rel , Frg , X LM , We g , DR . (9)

Frg  [0.7 1.2 1.7 2.2] (10)

X LM  [0.01 0.02 0.04 0.08 0.15 0.3]

The over-reading is based on volumetric flow rates and is defined as:

and consequently in a lower over-reading.

Figure ‎4-2 : Projection of over-reading of ultrasonic meters as a function X LM (left) and

5.1 General construction of the correction algorithm

 g   (X LM ) (Frg )(We g )(DR) (Re g ) (Re l ) . (13)

lim  g  stratified flow

 1 Frg  Frg* (21)

where the constant c1 is determined by the limit

5.1.3 Transition point

lim Frg  0 (surface tension :   0)

lim Frg   (surface tension :   )

Frg  1.2  0.3WLR . (27)

5.2 Generic correction algorithm

 g   (X LM , Rel / Re g , DR) (Frg , X LM , DR, Oh g ) . (28)

~  ~(X LM ) Frg  Frg* (29)

5.2.1 Application of the generic correction algorithm

Figure ‎5-4 : Projection of corrected over-reading of the ultrasonic meters as a function

is scattered and no valid data is obtained.

5.2.2 Extension of the correction algorithm

[4] E. Buckingham, On physically similar systems; illustrations of the use of dimensional

[13] A. Prosperetti, and G. Trygvasson, Computational Methods of Multiphase Flow (Cambridge

[18] S. Wongwises, W. Khankaew, and W. Vetchsupakhun. Prediction of Liquid Holdup in

A general expression for the phase j (gas or liquid) is given by

with jump condition at the interface

Table A-1 Values of Parameters for Conservation Equations

A.1 Dimensional analysis of multiphase flow

u j  u j 0u j , t  t0t ,   D1 , p j  p0 p j , g  gg ,  j   j 0  j ,    0 ,  j   j 0  j (37)

In equation (38) the dimensionless groups are identified

l  o (1  WLR )   w WLR . (40)

l  o (1  WLR )  p , WLR  WLR * (41)

l   w (WLR)  p , WLR  WLR *

 l   o (1  WLR )   w WLR . (43)

Impact of Using ISO/TR 11583 for a Venturi Tube in

NEL, East Kilbride, UK

across a Venturi tube to determine an estimate of the amount of liquid in wet-gas

2 DEFINITIONS OF WET-GAS FLOW

The gas densiometric Froude number, Frgas, is a dimensionless number directly

where A is the pipe area.