Embed Size (px)
Citation preview
The 27th international
North Sea Flow Measurement Workshop• The world venue for state of the art technology• Covering the latest development and experience within this field• Relevant for personnel working with flow measurement of oil and gas• Exhibition and excellent networking opportunities
20-23 October 2009, Tønsberg, Norway
Photo : Bjarne Riesto, Norway
NORWEGIAN SOCIETY FOR OIL AND GAS MEASUREMENT
1530 Refreshments
1600 (13) oiMl R 137-1, the first ultrasonic meter to be tested to accuracy class 0.5 Skule E. Smørgrav, FMC Kongsberg Metering AS, Norway
1630 (14) ultrasonic Meter condition based Monitoring - a fully automated maitanance solution Georg Kneisly, Transwestern Pipeline, USA, John Lansing, SICK, USA and Toralf Diez, SICK Germany
1700 End of day 2
1930 Dinner
Chair: Douglas Griffin,Department of Energy & Climate Change, UK
0900 (15) Three Columns gas Chromatograph analysis using Correlation between Component's Molecular Weight and its Response Factor Anwar Sutan, Charles Johnson, Emerson, UK
0930 (16) Validation of the CFD method for determining the measurement error in Flare gas ultrasonic meter installations Dr. Jeff Gibson, TUV NEL, UK
1000 (17) Nitrogen subtraction on reported Co2 emission using ultrasonic Flare gas Meter Kjell-Eivind Frøysa, CMR, Norway, Henning Ekerhovd, StatoilHydro ASA, Norway, Atle A. Johannessen, Fluenta, Norway
1030 Refreshments
1100 (18) Comparison of Multipath ultrasonic Meter Calibration Data from Two liquid Hydrocarbon Facilities and one Water Facility G J Brown, T Cousins and D R Augenstein, Cameron Measurement Systems, UK
1130 (19) High viscosity hydrocarbon flow measurement, a challenge for ultrasonic flow meters? Jankees Hogendoorn, Herman Hofstede, André Boer and Jan Drenthen, Krohne, The Netherlands
1200 (20) a Multipath ultrasonic Meter with Reducing Nozzle G J Brown, T Cousins, D R Augenstein and H Estrada, Cameron Measurement Systems, UK
1230 Lunch
Chair: Richard Paton, TUV NEL, UK
1330 (21) Cost Benefit analyses in the Design of allocation Systems Phillip Stockton, Smith Rea Energy Limited, UK
1400 (22) Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Verónica Mejía Gallardo and Diego Nelson Moncada Benavides, Centro de Tecnología Avanzada CIATEQ, Mexico
1430 (23) Realistic Pipe Prover Volume uncertainty Paul Martin and Mark Hay, Smith Rea Energy Limited, UK
1500 Refreshments
1530 (24) inferential Chemometric allocation Phillip Stockton, Smith Rea Energy Limited, UK
1600 (25) Metering atlas - a portal to create transparency in production and fiscal measurement data Lex Scheers and Richard Coomber, Shell Global Solutions International, Oi-Mee Voon, Brunei Shell Petroleum
1630 Closing remarks
1640 End of technical program
1930 Cocktails
2000 Banquet with entertainment
Breakfast from 0700
0830 Transport to Torp airport Sandefjord
TuEsday 20 oCTobER 2009 1200 Lunch and Registration
1300 Welcome Program chair: Dag Hendrik Flølo, StatoilHydro ASA, Norway
1310 Key Note New Challenges in Oil & Gas Measurement Douglas Griffin, Department of Energy & Climate Change, UK
Program chair: Dag Hendrik Flølo, StatoilHydro ASA, Norway
1330 (1) Significantly increased Capabilities of DP Meter Diagnostic Methodologies Dr Richard Steven, DP Diagnostics, USA
1400 (2) Cone DP Meter Calibration issues Casey Hodges, Charlie Britton and Bill Johansen, CEESI, USA Richard Steven, CEESI, USA
1430 Refreshments
Chair: Svein Neumann, ConocoPhillips, Norway
1500 (3) Recent field experiences using multiphase meters for fiscal allocation Eirik Åbro and Kaare Kleppe, StatoilHydro ASA, Norway
1530 (4) a Novel Subsea on-line Multiphase Fluid Sampling and analysis System Marie Bueie Holstad, Christian Michelsen Research AS, Norway Erik Magnus Bruvik, University of Bergen, Norway Bjørn Tore Hjertaker, University of Bergen Jarle Spilde and Kjell-Eivind Frøysa, Christian Michelsens Research AS, Norway
1600 (5) X-ray based densitometer for multiphase flow measurement Stein Arild Tjugum, Roxar, Norway, Romulus Mihalca, PANalytical, The Netherlands
1630 (6) Successful implementation and use of multiphase meters Gordon Stobie, ConocoPhillips, Arnstein Wee, MPM, Norway
1700 End of day 1
1930 Dinner
Chair: Kåre Kleppe, StatoilHydro ASA, Norway 0830 (7) an improved Model for Venture Tube over reading Wet gas
Michael Reader-Harris and Emmelyn Graham, TUV NEL
0900 (8) Measurement of Water in a Wet gas Arnstein Wee, MPM, Norway, Lex Scheers, Shell Global Solutions International, The Netherlands
0930 (9) Finding the optimum Wet gas Metering Solution, offshore Egypt Ingar Tyssen, Roxar, Norway, Alexis Houdusse, Roxar, France, Mohamed Baydoun, Petroleum Engineering, WDDM/ Raspetco
1000 Refreshments
1030 Manufacturers and Vendor Sessions 1300 Lunch
Chair: Per Lunde, University of Bergen and CMR
1400 (10) Reducing installation Effects on ultrasonic Flow Meters Jan G Drenthen, Jeroen van Klooster and Martin Kurth, Krohne New Technologies
1430 (11) Calibration errors of ultrasonic Meters in the Bernoulli laboratory due to Stratification Aernout van den Heuvel, Gasunie, The Netherlands, Frans Doorman, Shell, Piet van den Herik, NMi, the Netherlands, Arjan Stehouwer, Elster-Instromet, Belgium, Robert Kruithof, Gastransport Services, the Netherlands
1500 (12) Field experience of ultrasonic Flow Meter use in Co2 Rich applications Keith Harper, Sandridge Energy, USA, John Lansing, SICK, USA and Toralf Diez, SICK, Germany
dP mETERs
mulTIPhasE mEasuREmEnT
lIquId ulTRasonIC mETERs
GEnERal mEasuREmEnT ToPICs
Gas ulTRasonIC mETERs
WEdnEsday 21 oCTobER 2009
ThuRsday 22 oCTobER 2009
fRIday 23 oCTobER 2009
WET Gas mEasuREmEnT
EnvIRonmEnTal mEasuREmEnTs
New Challenges in Oil & Gas Measurement
Douglas Griffin
Head, Petroleum Measurement & Allocation
UK Department of Energy & Climate Change, Aberdeen, Scotland.
Introduction
The topics covered in this short paper by no means represent an exhaustive list.
However, some of the major oil and gas measurement issues arising from the
development of the remaining reserves in the UK sector of the North Sea are
presented from DECC’s perspective.
The purpose of this paper is to stimulate discussion in these areas at this year’s
workshop, as well as to clarify DECC’s position in one or two key areas.
Traceable Liquid Density Measurement
Approximately 5 years ago, a fairly major shortcoming in the existing practices
surrounding the calibration of liquid densitometers came to light. In a paper presented
at the 2007 North Sea Flow Measurement Workshop, the problem was summarised as
follows:
“However, the calibration of most industrial densitometers is undertaken
using fluids whose physical characteristics are significantly different to the
actual working or operational fluids. Additionally, the range of pressures and
temperatures at which the instruments are normally calibrated is limited to
near ambient conditions but many densitometers, particularly those used in
offshore applications, operate under high pressure, high temperature
conditions. This can be a significant source of error in density measurement.
Ideally, calibration should be undertaken at metering pressures and
temperatures using fluids whose volumetric properties are known accurately
across the full temperature and pressure range required for the calibration.”
[1]
A Joint Industry Project (JIP) completed its work earlier this year and a report was
circulated to the project members for their review. The consultation period has now
concluded.
The report contains a number of recommendations whereby the integrity and
traceability of the overall calibration process will be greatly improved. Unfortunately,
the implementation of these recommendations will require significant modifications
to the existing calibration facilities. This will take time, as well as a significant
investment on the part of the operators of these facilities. DECC, as UK Regulator,
cannot reasonably ask oil companies to implement the new calibration procedures
while no appropriate calibration facility exists. At the same time, it is only through
some sort of regulatory stimulus that the required modifications to the calibration
facilities will be made.
To this end, a statement will be published on the DECC website, outlining the
recommendations of the JIP and the new calibration procedures that it expects to see
followed after an 18-month ‘period of grace’. It is understood that this period will
give the operators of the calibration facilities sufficient time in which to make the
necessary modifications.
Calibration Uncertainty
One of the basic principles of uncertainty analysis is that parameters contributing less
than a tenth of the overall uncertainty may be conveniently ignored. From this, the
following basic metrological principle naturally arises:
When calibrating a measuring device, the uncertainty of the calibration
standard should be a factor of 10 less than that of the device itself.
Thus, when calibrating a temperature transmitter to ±0.5°C, the uncertainty of the
calibration standard should be ±0.05°C or less. Any inherent bias in the certifying
authority’s temperature standard is small enough to be safely ignored.
This basic principle should of course be familiar to any flow metering engineer. In
practice, however, flow metering engineers are occasionally forced to turn a blind eye
to this principle in view of the lack of any practical alternative. The impact of such a
decision must nevertheless not be overlooked.
Let us now consider two such examples from practice.
• Gas Ultrasonic Meters
In North Sea applications of 10-15 years ago, custody-transfer measurement of
gas flowrate made use of orifice plate technology, and virtually nothing else.
The design uncertainty of such systems is typically ±1.0%. Since orifice
plates designed and installed to ISO 5167 do not need to be flow calibrated,
there was never any need for a calibration facility with an uncertainty of 0.1%.
Today, of course, the situation is very different. The use of ultrasonic meters
for custody-transfer gas flow measurement has understandably become
widespread. One of the features of these devices, as well as their facility to
provide detailed diagnostic information, is their relatively low measurement
uncertainty. Multi-path meters have uncertainties in the range of 0.2%.
However, they require flow calibration. From the above, it is clear that for
true flow calibration to take place, calibration standards with flow
uncertainties in the region of 0.02% are required. Unfortunately, such
facilities do not presently exist – at least, not at the required flowrates.
Thus, in the absence of any practical alternatives, the basic principle referred
to above is ‘bypassed’. Ultrasonic meters are routinely ‘calibrated’ against
flow standards whose uncertainty is not significantly lower than the meters
themselves. While the adoption of such practices is wholly understandable, as
said above the consequences of ‘bypassing’ established metrological
principles must not be overlooked. In contrast to situation with the
temperature transmitter, an inherent bias in the certifying authority’s standard
has the potential here to be very significant. Operators are of course aware of
this and this explains the tendency to returning ultrasonic meters to the same
laboratory throughout their life in service.
There is clearly a continued place for the practice of ‘flow-calibrating’
ultrasonic meters. Even a condition-based monitoring scheme requires an
initial flow calibration. However, great care should be taken when
interpreting the results; in particular, the uncertainty and repeatability of the
calibration facility must always be considered when the results of the
calibration are used in mismeasurement calculations.
• Multiphase Meters
Equally, great care must be taken when evaluating the performance of
multiphase meters by placing them in series with test separators and
comparing the single-phase flowrates obtained. It is by no means clear that
multiphase meters’ uncertainties are any higher than the test separators. It is
most definitely not the case that test separators’ uncertainties are an order of
magnitude lower. Nevertheless, many commercial contracts require
multiphase meters to be adjusted to agree with test separator measurements if
a ‘trigger’ level (typically ±5% per phase) is exceeded.
Here too DECC recognises that comparison between multiphase meters and
test separators may be the best practical option, and can provide valuable
information, perhaps allowing drift in multiphase meter performance to be
detected. But the fact that basic metrological principles have been by-passed
should never be overlooked, and any commercial contract requiring
adjustment of one measurement to agree with the other should be treated with
great caution.
Cost v Benefit in Calibration Activities – Condition-Based Monitoring
The above considerations have led DECC to reconsider its position on calibration
activities in general.
North Sea operators face a continual challenge to keep operating costs low. While the
costs associated with the routine calibration of each of the elements of a fiscal
metering system are undoubtedly clear, the associated benefit is not always
understood. This is not a desirable situation. Where the benefit is clear, the
likelihood of such activities being overlooked is reduced.
Calibration takes place in order to detect any systematic shift that may already have
occurred in the device under test. In any fiscal measurement system, such a shift will
have financial consequences. These consequences may be very significant, they may
be trivial, or they may be somewhere in between. When weighing up the benefits of a
calibration program, one must consider the likely position on this ‘spectrum’ of any
shifts that may occur, and the likelihood of their occurrence. This is standard risk-
analysis strategy – the key parameter is the product of the consequences of an event
occurring and the probability that it will actually happen. In our field we commonly
use the term ‘exposure’ for the ‘financial risk’ arising from mismeasurement.
How are these key parameters – probability of a shift, and its likely scale – to be
determined in practice? Ultimately this must be based on the service history of the
device in question, although to some extent the performance of similar devices
elsewhere should be borne in mind. Ultimately, however, this is an area where hard-
and-fast rules are difficult to construct – it comes down to the judgement of
measurement specialists.
One the exposure has been determined, this may be compared with the cost of
removal and recalibration. On the basis of this comparison, the appropriate interval
between successive calibrations may then be agreed following discussion with the
Regulator. One of the attractions of this approach is that the benefits derived from the
calibration are made explicit. Calibrations should be seen to be in Operators’ own
interest, rather than being done simply because they are required by the Regulator or
by the pipeline authority.
DECC considers that such a risk-based approach to routine calibration programmes
should have an important part to play in Operators’s maintenance strategies in the
remaining years of North Sea production.
Multiphase and Wet Gas Measurement
It should be no surprise to find these topics included in a paper entitled ‘New
Challenges in Oil and Gas Measurement’. Multiphase, and its subset, wet gas
measurement, have dominated the programmes of North Sea Flow Measurement
Workshops for the last decade. Their use in fiscal applications is now widespread.
This is partly a reflection of the improved performance and reliability of multiphase
flowmeters, but it primarily reflects the smaller size of oil and gas fields being
developed in the North Sea, and the need to make fiscal measurement prior to the
fluids being separated into their constituent phases.
The charts in Appendix 1 have been presented by DECC and its predecessor
organisations at a number of forums in Aberdeen over the past decade, but they are
perhaps worth presenting again at the North Sea Flow Measurement Workshop, as
they illustrate why it is necessary to develop satellite fields sooner rather than later.
One of the key challenges facing the North Sea flow measurement industry is the need
to expand the range of the correlations used in wet gas applications. There is
widespread knowledge of correlations such as that first published by de Leeuw in
1997 [2]. However, like all such correlations this was never intended to be used over
the full range of conditions routinely encountered in the North Sea. It has been
widely used well beyond its intended pressure range, on fluids that are not
representative of those that were used in its derivation. Once again, the justification
for doing so has been the lack of any reasonable alternative. This does not mean that
we should lose sight of the fact that mismeasurement will result. To some extent, the
‘acceptability’ of such an approach – which is apparent rather than real - has reduced
the drive to expand the areas of the pressure/temperature/GOR matrix covered by the
existing correlations. To help address this to some extent, the UK Government is
funding ongoing research in this area. [3]
The need to derive additional confidence in the verification of multiphase meters is as
keenly felt as ever. As highlighted earlier in this paper, there are shortcomings
associated with the current standard strategy of comparison between multiphase or
wet gas meter and test separator. Additional verification methods such as data
reconciliation techniques or the use of ‘virtual’ models have the capability to build
confidence in these performance of these meters.
Sampling of Liquid Petroleum
Sampling of liquid petroleum has lost none of its importance with the increasing age
of the North Sea. On the contrary, with increasing levels of water production,
determining the true water content of contributing fields in shared transportation
systems is more challenging than ever before. This is especially true since the advent
of HT/HP condensate fields and the associated difficulties of determining the water
content of condensate samples.
The operator of at least one major North Sea pipeline has correlated pipeline
imbalance with platform shutdowns. Since large quantities of water are often
produced following field shutdowns, this finding probably confirms the suspicion of
many that the principal cause of pipeline imbalance is failure to detect quantities of
water entering the pipeline. The consequences of such a failure are often particularly
inequitable, since the apparent ‘loss’ typical of liquid pipeline systems is attributed to
fields on the basis of their throughput. Newer fields, which are by their nature ‘drier’,
are often relatively high producers. They are therefore allocated a correspondingly
high share of the pipeline ‘loss’, despite the fact that the source of the loss probably
lies with older, ‘wetter’ fields.
The operator of at least one major pipeline in the UK sector of the North Sea has
surveyed relevant liquid sampling systems to determine the level of compliance with
ISO 3171. DECC does not believe that a sampling system should stand or fall on its
performance relative to this standard; there have been instances where apparently
well-designed sampling systems have failed to meet the ‘homogeneity’ criteria of
Appendix A by several orders of magnitude(!). To determine the degree of
homogeneity of the fluid at the sampling point, it is preferable to carry out a CFD
study that takes account of the specific characteristics of the sampling system - for
example, whether the flow is horizontal or vertical at the sampling point.
Of course, the sampling system itself is only part of the problem. Failure to return
samples in their correct condition to the onshore laboratory for analysis results in the
use of water content values determined in offshore laboratories, often in less-than-
ideal conditions and often by personnel who are less well-qualified than their onshore
counterparts. While it is unreasonable to expect 100% return rates, it does appear that
sample collection and return is not always given the priority it deserves. The
penalties for failing to return samples for analysis are often minimal, and current
procedures may actively benefit operators who fail to return samples with high water
contents. DECC strongly encourages pipeline operators to look at the level of
incentivisation that exists in current transportation agreements, as this is an issue that
may reasonably be expected to increase in significance as the North Sea province
continues to mature.
Commercial Contracts and Measurement Uncertainty
As we have seen , today the North Sea is a mature province, the development of small
sub-sea satellites over existing infrastructure is very much the norm.
The ‘traditional’ method of doing this is to allocate production to the satellite field(s)
directly, perhaps by well test, more often (these days) by continuous measurement by
multiphase meter or, more rarely, via dedicated separator(s). The ‘host’ field is
allocated the out-turn from the platform, less the quantities allocated to the satellite
field(s), adjusted via a process model to take account for changes in phase behaviour
as the fluids pass through the plant.
The uncertainty associated with any ‘by difference’ measurement depends on the
uncertainty of the directly-measured quantities, and the relative proportions of the
sum of these to the ‘by difference’ amount. Ideally, the ‘by difference’ quantity
should be a small proportion of the total. Given that it is the ‘host’ field that is
normally allocated on a ‘by difference’ basis, and given that the host field will
necessarily have been in production for longer than any of the satellites, this ideal
scenario is not always the case. The quantities produced by the host field may instead
be a small, and rapidly diminishing, fraction of the total.
This problem will inevitably be exacerbated whenever a further satellite is developed
across the host facility; the uncertainty in the amount allocated to the host field will
inevitably increase.
In the UK sector of the North Sea, where a host field has been in production since
1993 there is a high probability that it will pay tax at a higher rate than its newer
satellites, so there is a direct fiscal dimension to this allocation process. DECC and its
predecessors have generally been fairly relaxed when it comes to giving consent to
such developments; while financial exposure is increased, the higher uncertainty in
the allocation to Petroleum-Revenue-Tax-paying fields has generally been seen as a
price worth paying in order to help ensure the development of the remaining reserves
in the UK sector of the North Sea.
It has to be said that commercial operators have not always been so philosophical.
There have been a number of occasions on which the operators of the host facility
have sought financial compensation, in advance, from the partners of prospective
satellite fields. On more than one occasion the subsequent commercial wrangling has
come very close to preventing the development of a satellite field from taking place.
Such a practice can only arise from a misunderstanding of the nature of measurement
uncertainty and its impact on financial exposure, and it is to be strongly discouraged.
Another possible scenario is where the Operator of the ‘host’ facility holds equity in
both ‘host’ and ‘satellite’ field. In such cases the exposure of this Operator may be
relatively small, since any under-allocation to one field will be compensated for by an
over-allocation to the other. However, where other equity-holders have interests in
only one of the fields, the financial exposures, and resultant willingness to invest in
robust measurement and allocation systems, may not be aligned between all the
interested parties. This too has the potential to cause commercial tensions.
DECC hopes that these and other scenarios will be addressed in the coming years,
perhaps by means of a joint Operators’ forum (perhaps under the auspices of UK Oil
& Gas); the issues involved have the potential to jeopardise the successful recovery of
some of the remaining reserves in the UK Sector of the North Sea.
REFERENCES
[1] GLEN,N & GRIFFIN,D Traceable Calibration of Liquid Densitometers. North
Sea Flow Measurement Workshop, Gardermoen 2007
[2] DE LEEUW, R Liquid Correction of Venturi Meter Readings in Wet Gas Flow,
North Sea Flow Measurement Workshop, Kristiansand 1997
[3] Project funded by the Department of Business and Innovation’s Engineering &
Flow Programme (http://www.nmo.bis.gov.uk/content.aspx?SC_ID=489).
APPENDIX
The red and orange areas charts are producing oil and condensate fields respectively.
Around each field a tie-back ‘bubble’ of c.50km has been drawn. Projections for field
life have been applied, allowing us to identify the diminishing extent of the areas of in
which satellite tie-backs will have the potential to be developed.
1
Significantly Improved Capabilities of DP Meter Diagnostic Methodologies
Richard Steven, DP Diagnostics PO Box 121, Windsor, Colorado, 80550, Colorado
Tel: 1-970-686-2189, e-mail: [email protected] 1. Introduction Differential Pressure (DP) flow meters are a popular generic flow meter type. DP meters are simple, sturdy, reliable and inexpensive devices. Their principles of operation are easily understood. However, traditionally there has been no DP meter self diagnostic capabilities. In 2008 a generic DP meter self diagnostic methodology [1] was proposed. In this paper these DP meter diagnostic principles are reconfirmed and a simpler methodology is also explained. These two methods will be shown to operate in conjunction increasing the overall sensitivity of a DP meters diagnostic capability. These diagnostic methods work for all generic DP meter designs. However, in this paper they are proven with extensive experimental test results from orifice plate and cone DP meters. Finally, it is recognized that it can be beneficial to have real time diagnostics where the diagnostic results are shown to the operator in a very simple and easily understood format. DP Diagnostics proposes such a method. 2. The generic DP meter classical and self-diagnostic operating principles
Fig 1. Orifice plate meter with instrumentation sketch and pressure fluctuation graph Figure 1 shows an orifice meter with instrumentation sketch and the (simplified) pressure fluctuation through the meter body. Traditional DP meters read the inlet pressure (P1), the downstream temperature (T) and the differential pressure (∆Pt) between the inlet pressure tap (1) and a pressure tap positioned at a point of low pressure (t). Note that the orifice meter in Figure 1 has a third pressure tap (d) downstream of the plate. This addition to the traditional DP meter design allows the measurement of two extra DP’s. That is, the differential pressure between the downstream (d) and the low (t) pressure taps (or “recovered” DP, ∆Pr) and the differential pressure between the inlet (1) and the downstream (d) pressure taps (i.e. the permanent pressure loss, ∆PPPL, sometimes called the “PPL” or “total head loss”). The sum of the recovered DP and the PPL equals the traditional differential pressure (equation 1). Hence, in order to obtain three DP’s, only two DP transmitters are required.
PPLrt PPP Δ+Δ=Δ --- (1)
2
Traditional Flow Equation: tdtt PYCEAm Δ= ρ2.
, uncertainty ± x% --- (2)
Expansion Flow Equation: rrtr PKEAm Δ= ρ2.
, uncertainty ± y% --- (3)
PPL Flow Equation: PPLPPLppl PAKm Δ= ρ2.
, uncertainty ±z% --- (4) The traditional generic DP meter flow rate equation is shown here as equation 2. Traditionally, this is the only DP meter flow rate calculation. However, with the additional downstream pressure tap three flow equations can be produced. That is, the recovered DP can be used to find the flow rate with an “expansion” flow equation (see equation 3) and the PPL can be used to find the flow rate with a “PPL” flow equation (see
equation 4). Note tm.
, rm.
and PPLm.
represents the traditional, expansion and PPL mass
flow rate equation predictions of the actual mass flow rate (.
m ) respectively. The symbol ρ represents the fluid density. Symbols E , A and tA represent the velocity of approach (a constant for a set meter geometry), the inlet cross sectional area and the minimum (or “throat”) cross sectional area through the meter respectively. Y is an expansion factor accounting for gas density fluctuation through the meter. (For liquids Y =1.) The terms
dC , rK and PPLK represent the discharge coefficient, the expansion coefficient and the PPL coefficient respectively. The downstream tap is ideally at the recovery point but it could be elsewhere, e.g. closer to the primary element for more compact meter designs. The flow coefficients are found by calibration (thereby calibrating out any tap position issues) and they can be set to constant values with set uncertainty ratings, or, may each be fitted to the Reynolds number, usually at a lower uncertainty rating. The Reynolds number is expressed as equation 5. Note that μ is the fluid viscosity and D is the inlet diameter. In this case, as the Reynolds number (Re) is flow rate dependent, each of the three flow rate predictions must be independently obtained by an iterative method within the flow computer. Equations 2 through 4 are derivation in detail by Steven [1].
Dm
πμ
.4Re = --- (5)
Every generic DP meter body is in effect three flow meters. As there are three flow rate equations predicting the same flow through the same meter body there is the potential to compare the flow rate predictions and hence have a diagnostic system. Naturally, all three flow rate equations have individual uncertainty ratings (say x%, y% & z% as shown in equations 2 through 4). Therefore, even if a DP meter is operating correctly, no two flow predictions would match precisely. However, a correctly operating meter should have no difference between any two flow equations greater than the sum of the two uncertainties. The calibration therefore produces three more values, i.e. the maximum allowable difference between any two flow rate equations, i.e. %φ , %ξ & %υ , as shown in equation set 6a to 6c. This allows a self diagnosing system. If the percentage difference between any two flow rate equations is less than that equation pairs summed uncertainties (found from the meters calibration), then no potential problem is found and the traditional flow rate prediction can be trusted. If however, the percentage difference between any
3
Traditional & PPL Meters % allowable difference ( %φ ): %%% zx +=φ -- (6a)
Traditional & Expansion Meters % allowable difference ( %ξ ): %%% yx +=ξ -- (6b)
Expansion & PPL Meters % allowable difference ( %υ ): %%% zy +=υ -- (6c) two flow rate equations is greater than that equation pairs summed uncertainties then this indicates a metering problem and the flow rate predictions should not be trusted. The three flow rate percentage differences, i.e. %ψ , %λ & %χ , are calculated by equations 7a to 7c:
Traditional to PPL Meter Comparison : %100*%...
⎭⎬⎫
⎩⎨⎧
⎟⎠⎞
⎜⎝⎛ −= tmmm tPPLψ -- (7a)
Traditional to Expansion Meter Comparison: %100*%...
⎭⎬⎫
⎩⎨⎧
⎟⎠⎞
⎜⎝⎛ −= tmmm trλ -- (7b)
PPL to Expansion Meter Comparison: %100*%...
⎭⎬⎫
⎩⎨⎧
⎟⎠⎞
⎜⎝⎛ −= PPLmmm PPLrχ -- (7c)
This diagnostic methodology is that which was discussed in 2008 [1]. It uses the three individual DP’s to independently predict the flow rate and then compares these results. In effect, the individual DP’s are therefore being directly compared. However, it is possible to take a different diagnostic approach. The Pressure Loss Ratio (or “PLR”) is the ratio of the PPL to the traditional DP. The PLR is constant for all DP meters operating with single phase homogenous flow, as indicated by ISO 5167 [2]. We can rewrite Equation 1:
1=ΔΔ
+ΔΔ
t
PPL
t
r
PP
PP
--- (1a) where t
PPL
PPΔΔ
is the PLR.
From equation 1a, if PLR is a constant set value then both the Pressure Recovery Ratio or “PRR”, (i.e. the ratio of the recovered DP to traditional DP) and the Recovered DP to PPL Ratio, or “RPR” must then also be constant set values. That is, all DP ratios available from the three DP’s are constant values for any given DP meter geometry and can be found by the same calibration that finds the three flow coefficients. Thus we have:
PPL to Traditional DP ratio (PLR): ( )caltPPL PP ΔΔ , uncertainty ± a% Recovered to Traditional DP ratio (PRR): ( )caltr PP ΔΔ , uncertainty ± b% Recovered to PPL DP ratio (RPR): ( )calPPLr PP ΔΔ , uncertainty ± c% Here then is another method of using the three DP’s to check a DP meters health. Actual DP ratios found in service can be compared to the calibrated values. Let us denote the difference between the actual PLR and the calibrated value as α , the difference between the actual PRR and the calibrated value as γ , and the difference between the actual RPR and the calibrated value as η . These values are found by equations 8a to 8c.
4
[ ]{ } %100*/% ncalibrationcalibratioactual PLRPLRPLR −=α --- (8a)
[ ]{ } %100*/% ncalibrationcalibratioactual PRRPRRPRR −=γ --- (8b)
[ ]{ } %100*/% ncalibrationcalibratioactual RPRRPRRPR −=η --- (8c) The standard calibration of a DP meter with a downstream pressure tap can produce six meter parameters with nine associated uncertainties. These six parameters are the discharge coefficient, expansion flow coefficient, PPL coefficient, PLR, PRR and RPR. The nine uncertainties are the six parameter uncertainties (±x%, ±y%, ±z%, ±a%, ±b% & ±c%) and the three flow rate inter-comparison uncertainties (±ψ %, ±λ , ± χ %). These fifteen DP meter parameters found by a standard calibration define the meters correct operating mode. Any deviation from this mode beyond the acceptable uncertainty limits is an indicator that there is a meter malfunction and the traditional meter output is therefore not trustworthy. Table 1 shows the six possible situations that should signal an alarm. Note that each of the six diagnostic checks has normalized data, i.e. each meter diagnostic parameter output is divided by the allowable difference for that parameter.
DP Pair No Alarm ALARM No Alarm ALARM tPΔ & pplPΔ 1%% ≤φψ 1%% >φψ 1%% ≤aα 1%% >aα
tPΔ & rPΔ 1%% ≤ξλ 1%% >ξλ 1%% ≤bγ 1%% >bγ
rPΔ & pplPΔ 1%% ≤υχ 1%% >υχ 1%% ≤cη 1%% >cη Table 1. The DP meter possible diagnostic results. For practical real time use, a graphical representation of the meters health continually updated on a control room screen could be simple and effective. However, any graphical representation of diagnostic results must be accessible and understandable at a glance by any meter operator. Therefore, it is proposed that three points are plotted on a normalized graph (as shown in Fig 2). This graphs abscissa is the normalized flow rate difference and the ordinate is the normalized DP ratio difference. These normalized values have no units. On this graph a normalized diagnostic box (or “NDB”) can be superimposed with corner co-ordinates: (1, 1), (1, 1− ), ( 1− , 1− ) & ( 1− ,1). On such a graph three meter diagnostic points can be plotted, i.e. ( φψ , aα ), ( ξλ , bγ ) & ( υχ , cη ). That is, the three DP’s have been split into three DP pairs and for each pair both the difference in the flow rate predictions and the difference in the actual to calibrated DP ratio are being compared to the calibrations maximum allowable differences. If all points are within the NDB the meter operator sees no metering problem and the traditional meters flow rate prediction should be trusted. However, if one or more of the three points falls outside the NDB the meter operator has a visual indication that the meter is not operating correctly and that the meters traditional (or any) flow rate prediction can not be trusted. The further from the NDB the points are, the more potential for significant meter error there is. Note that in this random theoretical example shown in Figure 2 all points are within the NDB indicating the meter is operating within the limits of normality, i.e. no metering problem is noted.
5
Fig 2. A normalized diagnostic calibration box with normalized diagnostic result.
As both techniques use the same inputs, i.e. the three DP’s, it may be asked whether it is necessary to use both techniques together. If the DP relationships are as expected both techniques indicate no meter error. If they are not as expected both techniques should indicate incorrect meter operation. However, from experience (as we will see), it has been found that the two techniques can have slightly different sensitivities to problems. The DP ratio technique is more sensitive to metering abnormalities. Therefore, if both techniques show no problem then there is no alarm. If both techniques show a problem there is a “general alarm”. However, for relatively small problems the different sensitivities of the two methods can cause one technique to indicate a problem while the other indicates no problem. This scenario gives an “amber alarm”. The amber alarm indicates that there may be a metering problem. The amber alarm arises from the fact that the DP ratio technique can find real meter problems below the flow rate comparison techniques sensitivity limit. However, there are rare cases where the DP ratio technique is too sensitive to real but very small problems that do not cause the flow rate prediction to be beyond the meters stated uncertainty. However, the flow rate comparison technique is never sensitive enough to trigger such a false alarm. Therefore, the flow rate comparison technique can counter any over sensitivity of the DP ratio technique by offering the operator objectivity. The amber alarm states that there is a possibility of a small metering problem, but if it exists the metering error is correspondingly small. Therefore, it is beneficial to use both techniques simultaneously (especially as the computational power required is relatively small). We shall now look at orifice and cone meter correct and incorrect operation data to show the usefulness of these methodologies.
Fig 2a. Normalized diagnostic box with alarm zones.
6
3. Orifice plate & cone DP meter experimental data analysis Orifice plate and cone DP meters are both popular. However, industry tends to utilize these meters in different ways. As orifice meters can be seriously affected by installation effects there are standards (e.g. [2]) that dictate where they should be installed in relation to other pipe components. Most orifice meter installers adhere to these standards and as a well made plate has a repeatable performance the standards discharge coefficient statement is used without a meter calibration being required. However, the cone DP meter (which has no standards and therefore requires calibration) is well received by industry due to the discharge coefficient being largely immune to installation effects. Therefore, orifice meters are usually installed in precise adherence to the standards and cone DP meters are usually installed in any awkward pipe work location. This means we have to treat the DP diagnostic research of the two meters differently. In this paper all orifice meter data, from correct and incorrect operation, are from plates installed according to ISO 5167 requirements, just as they are commonly installed in the field (except for deliberate tests for installation effects). Therefore, the orifice meter calibration values for all required diagnostic parameters outside of those given by ISO can be set from a single standard installation test. However, this is not so for cone DP meters. The cone DP meter research required that the meter was calibrated with long straight pipe lengths to find the meters diagnostic parameters. This procedure is all that would be done for massed produced cone DP meters. However, unlike the orifice meter, the cone DP meter can be used in awkward pipe installations. Therefore, this research has to prove that, just like the discharge coefficient, the other cone meter parameters required by the diagnostics are also acceptably immune to installation effects. Only then could the diagnostic method be shown to work successfully in all cone meter applications. Therefore, orifice meter diagnostic method tests only had to introduce problems when the orifice meter was installed according to ISO 5167. However, the cone meter diagnostic method tests had to introduce problems when the cone meter was installed in the typically extreme adverse installation conditions where it is commonly used. 4. Correctly operating orifice plate meter data Three 4”, 0.5 beta ratio flange tap orifice meter data sets have been analyzed. The first is a dry natural gas flow test on an orifice fitting installed plate. These tests were part of a CEESI wet gas meter Joint Industry Project (or “JIP”). In these tests only the traditional DP and PPL were read. The recovered DP was derived by equation 1. The other two data sets are from separate air flow, flange installed, orifice meter tests carried out at CEESI in 2008 and 2009. The 2008 tests used Daniel plates. The 2009 tests use Yokogawa plates. These air tests both directly read all three DP’s. Table 1 shows the data range of the three baseline (i.e. correctly operating) orifice meter tests. Figure 3 shows the test at the CEESI wet natural gas loop in 2000. Figure 4 shows the test at the air facility at CEESI in 2009. Note that the downstream tapping was at 6D downstream of all the plates as suggested by ISO 5167. Figure 5 shows the discharge coefficient, expansion coefficient and PPL coefficient from all three tests together. For simplicity of explaining the diagnostic concept constant values were assigned by data
7
Fig 3. Orifice fitting with natural gas flow. Fig 4. Flange installed plate with air flow.
Test 2000 Natural Gas 2008 Air 2009 Air Orifice Type & Fit Daniel Orifice Fitting Daniel Plate / Flange Yokogawa Plate /Flange No. of data points 112 44 124
Diameter 4.026” 4.026” 4.026” Beta Ratio 0.4965 (single plate) 0.4967 (multiple plates) 0.4967 (multiple plates)
Pressure Range 13.1 < P (bar) < 87.0 15.0 < P (bar) < 30.0 14.9 < P (bar) < 30.1 DPt Range 10”WC< DPt <400”WC 15”WC< DPt < 385”WC 15”WC< DPt < 376”WC DPr Range 10”WC <DPr < 106”WC 10”WC < DPr < 100”WC 10”WC <DPr < 100”WC
DPppl Range 10”WC <PPL < 293”WC 11”WC<PPL< 285”WC 11”WC<PPL< 277”WC Reynolds No. Range 350 e3 < Re < 8.1e6 300e3 < Re < 2.1e6 317e3 < Re < 2.2e6 Table 1. The three orifice plate meter baseline data sets. fitting. (It should be noted that more than 95% of the combined discharge coefficient results fitted the Reader Harris –Gallagher, or “RHG”, equation to within this equations stated uncertainty bands of ±0.5%.) All three flow coefficient constant values are given in Figure 5 with a stated uncertainty at 95% confidence. Figure 6 shows the PLR, PRR & RPR from all three tests together. Constant values were assigned by analysis of the combined data and are shown in Figure 6 with a stated uncertainty at 95% confidence. Note that the sum of the PLR and PRR is not quite unity as theoretically required due to data uncertainty. Figures 5 & 6 indicate that all six parameters exist as set values at relatively low uncertainty and they are repeatable and reproducible. Fig 7 shows the full results of calibrating this DP meter type with a downstream pressure tap. The boxed information shows the traditional DP meter calibration result, i.e. the discharge coefficient and its uncertainty to 95% confidence. The broken line box indicates a rare additional traditional result when a downstream pressure tap is included. Note even in this rare case when a downstream tap is included, only the PLR is found. Traditionally no other parameter is considered and the downstream tap only exists to help predict the PPL across the component for overall PPL predictions on the piping system. Fig 7 shows that all fifteen DP meter parameters discussed in the theory exist in reality. From adding an extra pressure tap and DP transmitter, a standard DP meter calibration run can find each of the DP meters fifteen parameters. Each parameter tells the meter user something unique and of interest about the nature of that meters response to the flow. That is, a DP Diagnostics meter calibration produces several times the standard calibration information from effectively the same effort and expense.
8
Fig 5. Combined 4”, 0.5 beta ratio orifice plate meter flow coefficient results.
Fig 6. Combined 4”, 0.5 beta ratio orifice plate meter DP ratio results.
Fig 7. The results of a full DP meter calibration.
In reality most orifice meters are not calibrated. ISO 5167 provides the RHG equation to find the discharge coefficient (Cd) and the associated uncertainty (±x%). It also offers a couple of PLR equations although no uncertainty (±a%) for these equations are given. The ISO PLR equation considered as more precise is equation 9:
9
( ){ }( ){ } 224
224
11
11
ββ
ββ
dd
dd
CC
CCPLR
+−−
−−−= -- (9) PLRPRR −=1 -- (10)
PLRPRRRPR = -- (11)
Therefore, as ISO offers a PLR equation, there are associated predictions for PRR (equation 10) and RPR (equation 11), although of course ISO does not state as much. Furthermore, it can be shown that:
PLRCK d
r −=
1ε
--- (12) PLR
CEK dppl
εβ 2
= --- (13)
In the field, Equation 9 would use the RHG discharge coefficient result. However, this is Reynolds number dependent and hence individual flow point dependent. As here we simply want to check the approximate applicability of this equation we can use our data sets averaged discharge coefficient to predict the DP ratios. (All data fitted the RHG equation to ±0.5% and the constant discharge coefficient to ±0.65% so they are very similar.) Furthermore, note that equations 12 & 13 require the expansibility (ε ) of each point to be known. However, this is a second order effect, and for our purposes here we can approximate the value to unity (i.e. assume incompressible flow). We can now use these approximations to examine the approximate effectiveness of using the ISO PLR prediction with the RHG equation to predict the DP ratios and the expansion and PPL flow coefficients. The results are shown in Table 2. Data Fit
Values Data Fit
% Uncertainties ISO Prediction
Values % Difference in ISO & Data Fits
Discharge Coefficient 0.602 ±0.65% N/A* N/A* Expansion Coefficient 1.165 ±1.1% 1.167 +0.14%
PPL Coefficient 0.178 ±1.8% 0.181 +1.95% PLR 0.732 ±1.6% 0.734 +0.23% PRR 0.262 ±1.2% 0.266 +1.64% RPR 0.360 ±1.8% 0.363 +0.82%
N/A* - Here we are using the data fit value, in the field the RHG equation will be more accurate at ±0.5%. Table 2. Comparison of ISO “predictions” to experimental results. Even with these generalizing assumptions the ISO predictions are very similar to the experimental data results. The PPL flow coefficient and PRR are out with the data fitting uncertainty bands, but in both cases just marginally so. Furthermore, it should be remembered that in the field with the use of the RHG equation and the expansibility equation the discharge coefficient and expansibility inputs will be more accurate. This will further reduce the uncertainty of these “ISO prediction” results. Therefore, an orifice meter user could calibrate his meter to find the full parameter set described here, or for some increase in uncertainty ISO based “predictions” could be used. However, note that this research chose to investigate a 4”, 0.5 beta ratio plate. The accuracy of applying ISO based predictions to other orifice meter diameter and beta ratio sizes instead of calibrating the meter is as yet unknown. We now have enough orifice meter information to apply the normalized diagnostic box (NDB) when the meter is in service, and hence we have DP meter diagnostics. When
10
using these diagnostics it should be remembered that the primary output of the meter is the traditional flow rate prediction with its uncertainty rating. All other calculations are solely to check the validity of this output. False alarms regarding the meters health are highly undesirable. Therefore, as the uncertainty ratings of the diagnostic parameters are at 95% confidence, we need to increased these uncertainties to avoid false alarms. Also note that when the third DP is not being directly measured, a small increase in diagnostic uncertainty values is prudent. (Note that these diagnostic uncertainty setting increases have nothing to do with the uncertainty rating of the primary output. The discharge coefficient can have one uncertainty rating for the output value and a separate larger uncertainty rating assigned for the diagnostic use of the parameter.) The diagnostic parameter uncertainties are set at the users discretion. Liberal uncertainty values are less likely to produce a false alarm, but, this is obviously at the expense of diagnostic sensitivity. The larger the uncertainties, the less sensitive the meter is to small but real problems. The greatest possible diagnostic sensitivity and the greatest exposure to false alarms are both achieved with the smallest possible uncertainties, i.e. the calibrated values at 95% confidence (i.e. see Fig 7 for this 4”, 0.5 beta ratio orifice meter example). Figure 8 shows a NDB with the baseline data sets. Here, the diagnostic parameter uncertainties were increased above the calibration results (see Figure 7) in order to minimize the chance of false alarms. The values chosen by engineering judgment are shown in Figure 9. The flow coefficient uncertainties were raised to the next whole number, except the PPL coefficient which was so close to ±2% it was raised to ±3% instead. The flow rate comparisons effectively compare a factor directly related to the square root of the DP ratios. Therefore, DP ratios are more sensitive to changes in the DP relationships than the flow meter equations. There is therefore a danger that they can be overly sensitive (as will be shown) and hence a full 1% was added to each DP ratio uncertainty.
Fig 8. Results from the massed data sets from the three 4”, 0.5 beta ratio orifice meters. Even though the uncertainty value for the discharge coefficient is listed here as ±1% for diagnostic methods (to help stop false alarms) this does not affect the meters flow rate uncertainty which remains in this case at ±0.65%. Again note that the uncertainty setting can be at or above the calibrated values at the discretion of the meter user.
Figure 8 appears to have a mass of data. However, note that as each flow point produces three DP pairs, every single flow point tested has three diagnostic points on the graph. In
11
Fig 9. The results of a full DP meter calibration result.
actual application only three points representing three DP pairs would be superimposed on the graph making the diagnostics result very clear. Even with 280 flow points producing 840 diagnostic results in Figure 8, it is clear that no point for these correctly operating conditions is outside the NDB meaning the diagnostics are declaring the meter to be serviceable. This result is in itself trivial as the uncertainties of the diagnostic parameters were set to this very data. However, the non-trivial results are from 4”, 0.5 beta ratio orifice meters deliberately tested when malfunctioning for a variety of reasons. 4a. Incorrectly operating orifice plate meter data There are many common orifice meter problems including incorrectly installed, damaged or contaminated plates and meters not installed in accordance with the ISO standards. These scenarios are now discussed. All orifice meter NDB points shown in the examples use parameter uncertainties shown in Figure 9. Like all DP meters the orifice meter can suffer from DP transmitter saturation or drift. However, such a worked example will be left to the discussion of another generic DP meter, i.e. the cone DP meter in section 4. 4.a.1. Reversed orifice plate installation Orifice plates are often installed erroneously in the reverse (or “backwards”) direction to the flow. Figure 10 shows the repeatable traditional error (equation 2) with backwards plates. Table 3 shows the data ranges. Figure 11 shows the data sets plotted with a NDB. The backwards plate produces a -15% error. Whereas there are no traditional diagnostics to indicate the problem the NDB data plot indicates the problem as the data falls outside the NDB. In this case as the problem is a precise geometry issue the precise pattern on the NDB indicates to the user the problem is most likely the plate is installed backwards.
Year 2008, Daniel Plate 2009 Yokogawa Plate Pressures 15 & 30 Bar 15 Bar
Traditional, DPt 13”WC < DPt < 335”WC 14”WC < DPt < 327”WC Expansion, DPr 5”WC < DPr < 100”WC 5”WC < DPr < 98”WC
PPL, DPppl 9”WC < DPppl < 235”WC 10”WC < DPppl < 229”WC Reynolds Number Range 346e3 < Re < 2.31e6 367e3 < Re < 1.66e6
Table 3. Backwards plate test data range.
12
Fig 10. Reproducible significant errors when plate is installed backwards.
Fig 11. Graph indicating metering error with NDB and all reversed plate results.
4.a.2. Damaged orifice plates – buckled (or “warped”) plates Adverse flow conditions can damage orifice plates. A buckled plate can give significant flow measurement errors. Traditionally there is no diagnostic methodology to indicate this problem. In 2008 DP Diagnostics heavily damaged a 4”, 0.5 beta plate to show the diagnostic capability of the downstream tap. In 2009 this test was re-run to prove repeatability of the new diagnostic system. Then a more moderately buckled 4”, 0.5 beta ratio plate was tested. The buckled plates are shown in Figures 12 & 13. Table 4 shows the test data ranges. Figure 14 shows the flow rate prediction (equation 2) error due to the buckling. The heavily buckle produces a -30% error. The moderate buckle plate produces
Test 2008, Severe Buckle 2009, Severe Buckle 2009, Moderate Buckle Pressures 15 & 30 Bar 15 & 30 Bar 15 & 30Bar
Traditional, DPt 11”WC <DPt< 291”WC 12”WC <DPt< 285”WC 14”WC <DPt< 352”WC Expansion, DPr 5”WC <DPr< 100”WC 5”WC <DPr< 100”WC 5”WC <DPr< 99”WC
PPL, DPppl 7”WC <DPppl< 192”WC 7”WC<DPppl< 185”WC 10”WC<DPppl<254”WC Reynolds No. Range 389e3 < Re < 2.64e6 394e3 < Re < 2.6e6 331e3 < Re < 2.2e6 Table 4. Buckled plate test data range.
13
Fig 12. Severely buckled orifice plate. Fig 13. Moderately buckled orifice plate.
Fig 14. Flow rate prediction errors due to buckled orifice plates.
Fig 15. Metering error with NDB Fig 16. Metering error with NDB & heavily buckled plate results. & moderately buckled plate results. a -7% error. The pressure had no effect on the results and the results were very repeatable. Figure 15 shows the heavily buckled plate data sets plotted with a NDB.
14
Figure 16 shows the moderately buckled plate data set plotted with a NDB. This indicates a significant problem. Note that for the moderately buckled plate the traditional and PPL DP pair does not trigger the alarm. However, all three DP pairs are always available and the other two DP pairs clearly trigger the alarm. This example highlights the extreme usefulness of the traditionally least used DP, i.e. recovered DP. (In fact, the particular DP pair to trigger any alarm is wholly dependent on the meter design the type of problem.) 4.a.3. Damaged orifice plate – worn leading orifice edge Orifice sharp edges can be worn leading to flow measurement errors. Traditionally there is no diagnostic methodology to indicate this problem. In 2008 DP Diagnostics heavily filed down a 4”, 0.5 beta orifice edge to show the diagnostic capability of the downstream tap. In 2009 smaller damage was tested with 0.01” and 0.02” chamfers being put on 4”, 0.5 beta orifice edges. The filed and 0.01” chamfer are shown in Figures 17 & 18. Table 5 shows the test data ranges. Figure 19 shows the flow rate prediction (equation 2) error due to the orifice edge wear. The approximate errors are -8% for the heavily filed orifice edge, -5% for the 0.02” chamfered edge and -2.5% for the 0.01” chamfered edge. Figure 20 shows the “worn” plate data sets plotted with a NDB. (Pressure had no effect on the results so both pressures tested are shown as one data set, i.e. three DP pairs, per plate.) In Figure 20 the data plotted on a NDB shows the heavily filed edge plate to have significant problems. In Figure 21 the larger chamfer also has a clear diagnostic indication that there are significant problems. Therefore these meter flow rate outputs should not be trusted. It is interesting to again note that not all three DP pairs always trip an alarm. Here the traditional DP and PPL pairings again do not always see the problem and again it is the rarest used of the DP’s, i.e. the recovery DP, used with either of the other two DP’s that is correctly tripping the alarm. Finally, note that as would be expected, the smallest edge damage (i.e. the 0.01” chamfer) is the most difficult to notice. With a traditional flow rate error of only -2.5% the traditional and recovery DP ratio just picks up the problem. The other two DP pairs are not sensitive enough to this particular problem to trigger an alarm. This appears to be the limit of the diagnostic systems ability. A smaller amount of damage (say causing a metering error < 2%) may not be seen.
4.026", 0.5 Beta Ratio Orifice Plate Meter
-0.5%
+0.5%
-10-8-6-4-202468
10
0 500000 1000000 1500000 2000000 2500000Pipe Reynolds Number
% F
low
Rat
e Er
rors
Filed Plate 15 BarFiled Plate 30 Bar0.02" Chamfer, 15 Bar0.02" Chamfer, 30 Bar0.01" Chamfer, 15 Bar0.01" Chamfer, 30 Bar
Fig 19. Flow rate prediction errors due to wear on orifice plate edges.
15
Fig 17. Filed orifice edge. Fig 18. Chamfered (0.01”) orifice edge.
Test 2008, Filed 2009, 0.02” Chamfer 2009, 0.01” Chamfer Pressures 15 & 30 Bar 15 & 30 Bar 15 & 30Bar
Traditional, DPt 14”WC <DPt< 356”WC 14”WC <DPt< 359”WC 15”WC <DPt< 368”WC Expansion, DPr 4”WC <DPppl< 100”WC 4”WC <DPr< 99”WC 4”WC <DPr< 99”WC
PPL, DPppl 10”WC <DPr< 256”WC 10”WC<DPppl< 256”WC 11”WC<DPppl<270”WC Reynolds No. Range 325e3 < Re < 2.23e6 352e3 < Re < 2.15e6 332e3 < Re < 2.12e6
Table 5. Worn orifice plate edge test data range.
Fig 20. Metering error with NDB Fig 21. Metering error with NDB and & heavily filed orifice edge results. chamfered orifice edge plate results. 4.a.4. Contaminated orifice plates Adverse flow conditions can deposit contaminates on orifice plates leading to flow measurement errors. Traditionally there are no diagnostics to indicate this problem. There are two types of contamination. There is fluid contamination (e.g. oil from upstream components) which is transient in nature, and difficult to test, and the more stable and easier to test case of solid deposits left on plates. Therefore, two 4”, 0.5 beta plates were given mild and severe contamination respectively. The mildly contaminated plate was lightly spray painted on the upstream side to produce light ripples and some paint drops at the sharp edge. The heavily contaminated plate was heavily painted on the upstream side and then large salt granules embedded in the painted to produce an extremely rough surface. Due to time and financial constraints no downstream side plate contamination was investigated. The contaminated plates are shown in Figures 22 & 23 respectively.
16
Fig 22. Lightly contamination. Fig 23. Heavy contamination.
Test 2008, Mild Contamination 2009, Heavy Contamination Pressures 15 & 30 Bar 15 & 30 Bar
Traditional, DPt 15”WC <DPt< 376”WC 17”WC <DPt< 368”WC Expansion, DPr 4”WC <DPr< 100”WC 4”WC <DPr< 99”WC
PPL, DPppl 11”WC <DPppl< 276”WC 12”WC<DPppl< 265”WC Reynolds No. Range 318e3 < Re < 2.18e6 346e3 < Re < 2.15e6
Table 6. Contaminated plate test data range.
Fig 24. Flow rate prediction errors due to contamination on orifice plates. Table 6 shows the test data ranges. Figure 24 shows the flow rate prediction (equation 2) errors of -4% and -1.5% for the heavily and lightly contaminated plates respectively. These results are similar to fluid contamination test results by Johansen [3] and Pritchard [4]. Figure 25 shows the two data sets plotted with a NDB. Again, as pressure had no effect, both pressures tested for each plate are simply shown as one data set, i.e. three DP pairs, per plate. Figure 25 shows the heavily contaminated plate to have problems so the meters flow rate output should not be trusted. Again note that the traditional DP and PPL pairings do not see the heavy contamination problem and again it is the recovery DP, used with either of
17
Fig 25. Graph indicating metering error with NDB and all contaminated plate results.
the other two DP’s that is correctly tripping the alarm. The light contamination only induces an error of -1.5% which less than 1% beyond the meters stated uncertainty. Figure 25 shows that the diagnostic system can not see such a small error. This is beyond the sensitivity of the diagnostic system. (It is however noteworthy that here we are using the higher set uncertainties of Table 9 to avoid false alarm trips. If we used the minimum possible uncertainties found for this meter in Table 7 the PRR comparison would correctly trip the alarm. This is an example of the choices the user must make between high diagnostic sensitivity and the danger of false alarms.) The practical meter error limit of the diagnostic system appears to be in the region of ±2%. 4.a.5. Orifice plate installation out side of ISO 5167 part 2 requirements ISO 5167 states orifice meter installation requirements. If an orifice meter is installed too close to pipe components the flow disturbances can cause flow measurement errors. DP Diagnostics tested a 4”, 0.5 beta ratio orifice meter with a half moon orifice plate (HMOP), as shown in Figure 26, installed at 2D, 12D and then 22D upstream. The HMOP blocked the upper half of the pipe area modeling a half open gate valve. Table 7 shows the test data ranges. Figure 27 shows the 22D installation. Figure 28 shows the
Fig 26. Half Moon Orifice Plate. Fig 27. Meter installed with upstream HMOP.
18
Test 2D upstream 12D upstream 22D upstream Pressures 15 Bar 15 Bar 15 Bar
Traditional, DPt 16”WC <DPt< 378”WC 18”WC <DPt< 395”WC 32”WC <DPt< 379”WC Expansion, DPr 4”WC <DPppl< 98”WC 4”WC <DPr< 94”WC 8”WC <DPppl< 99”WC
PPL, DPppl 11”WC <DPr< 281”WC 14”WC<DPppl< 301”WC 23”WC <DPr< 280”WC Reynolds No. Range 323e3 < Re < 1.52e6 373e3 < Re < 1.66e6 455e3 < Re < 1.52e6
Table 7. HMOP upstream of orifice plate test data range. flow rate prediction (equation 2) error due to the HMOP. There are no HMOP induced errors for the 22D and 12D installations. ISO state 12D is the minimum distance for a valve when fully open. Therefore, no error at 12D was considered a surprising result. There is a -6% error induced by the 2D HMOP installation. Figure 29 shows the data with a NDB for the problem installation of HMOP at 2D upstream. The diagnostics show a significant error exists and that the meter output should therefore not be trusted. Figure 30 shows the data with a NDB for the installation of a HMOP at 12D and 22D upstream. The 22D HMOP installation diagnostics correctly indicates that the meter is serviceable. However, the installation of a HMOP at 12D upstream, for which the meter gave the correct flow rate does still trigger an alarm. That is the RPR falsely indicates a metering problem. At 12D the flow profile has recovered enough from the disturbance for the traditional meter to operate correctly. However, the flow profile has not recovered enough to always give a RPR value similar enough to the
Fig 28. Flow rate prediction errors on orifice meters due to upstream HMOP’s.
Fig 29. HMOP at 2D data with NDB. Fig 30. HMOP at 12D & 22D with NDB.
19
calibrated value. That is, the RPR is too sensitive to the flow disturbance here, and it is therefore suggesting there is a metering problem when in fact there is not. This is the reason why general and amber alarms are proposed. It should be realized that a HMOP 12D upstream of an orifice meter is an extremely poor installation which is very rare in reality. DP Diagnostics has primarily developed the diagnostics for use with correctly installed orifice plate meters in which this is not. With that said, if this meter was calibrated in-situ the effect of the disturbance could be calibrated out of the diagnostic parameters. It is also of interest to note that it was found after analysis, that if the uncertainty rating of the RPR had been chosen as 4% instead of 2.8% the alarm warning would not have triggered and all other diagnostic results for correctly and incorrectly operating orifice meters would also have remained correct. However, this would be fitting the uncertainties with the benefit of hindsight which does not give a realistic review of the ability of the diagnostics, so the original research results are kept here. The issue here is that this is a new diagnostic methodology, and the best parameter uncertainty settings which produce the best balance between an over sensitive system and a system that is too insensitive will need to be found by experience. 4.b. A review of the orifice meter diagnostic system test results A general alarm almost certainly indicates a significant metering problem. An amber alarm very probably indicates either a significant or small metering problem. In reality for any generic DP meter there is virtually no chance of an amber alarm being triggered by a flow rate comparison while the associated DP ratio comparison does not trigger the alarm. This is due to the DP ratios being more sensitive to problems than the flow rate comparisons technique. Therefore, an amber alarm is going to be triggered by the DP ratio technique only. Note that the user can make engineering judgments on these alarms. There are three points of interest while judging if an amber alarm indicates a real problem and if so, is the error significant enough to merit intervention? 1. How far outside the NDB are the points? The further outside the NDB the more chance of a significant real problem. In practical terms the amber alarm can be limited to the case where the DP ratio comparison alarm values are between 1 and 1.5. Beyond the 1.5 value all experimental results show a significant real metering error. 2. Is there more than one point outside the NDB? If so the chances of a real significant problem are increased. 3. If a DP ratio point is outside the NDB, where is it located on the abscissa? The further from the ordinate, the closer the flow rate comparison technique is to signaling a general alarm and the more likely then that the amber alarm is indicating a real and significant metering problem. 5. Correctly operating ∆P cone meter data The diagnostic methods described in Section 2 are available for all generic DP meter designs. In this section the cone DP meter is discussed. A sketch of a cone DP meter design and the pressure fluctuations through the meter is given in Figure 31. Note the extreme similarity to the orifice meter sketch in Figure 1.
20
Fig 31. Cone meter with instrumentation sketch and pressure fluctuation graph DP Diagnostics built and tested a 4”, 0.63 beta ratio cone meter (with a downstream tap). The centre line of the cone support is 2D from the inlet flange face and 5.25D from the outlet flange face (as the meter is 7.25D long). The three DP’s were read directly. The meter was fully calibrated (at 14 & 41 Bar) with straight lengths upstream and downstream (see Figure 33). The results are shown in Figure 32. Table 8 shows these baseline test data ranges. The discharge coefficient was found to ±1/2% as expected. Again, the other parameters were found to be of a relatively low uncertainty.
Test Baseline Adverse Flow Conditions Pressure Range 17.2 < P (bar) < 41.1 17.2 bar
DPt Range 21”WC< DPt <301”WC 14”WC< DPt < 304”WC DPr Range 10”WC <DPr < 133”WC 6”WC < DPr < 136”WC
DPppl Range 12”WC <PPL < 169”WC 7”WC<PPL< 168”WC Reynolds No. Range 888 e3 < Re < 3.75e6 734e3 < Re < 2.9e6
Table 8. Correctly operating ∆P cone meter baseline & adverse flow condition tests.
Fig 32. The results of the standard straight run DP cone meter calibration.
Cone DP meters are popular due to their discharge coefficients proven immunity to flow disturbances. Hence, after calibration it was necessary to test the meter with a myriad of adverse flow conditions in order to prove that all the diagnostic parameters are acceptably immune to flow disturbances and therefore the ∆P Cone Meter has a practical diagnostic capability. The adverse flow conditions tested were a double out of plane bend (DOPB) at 0D (Figure 34), 2D & 5D upstream, a DOPB at 0D upstream with a half moon orifice plate (HMOP) at 2D downstream (Figure 35), a DOPB at 0D upstream and a triple out of plane bend (TOPB) 0D downstream (Figure 36), a HMOP 6.7D & then 8.7D upstream (Figures 37 & 38), a HMOP 2D downstream (Figure 39) and a 540 swirl
21
Fig 33. Baseline Fig 34. DOPB, 0D up
Fig 35. DOPB 0D up & HMOP 2D down Fig 36. DOPB 0D up & TOPB down
Fig 37. HMOP 6.7D up Fig 38. HMOP 8.7D up generator with a 3” to 4” expansion 9D upstream (Figure 40). Very few real applications would create worse flow conditions at the meter inlet. Table 8 shows the test data ranges. Figure 41 shows the disturbance effects on the discharge coefficient. Two installations cause it to vary beyond the baseline ±½% uncertainty. They are the HMOP 6.7D and
swirl generator with expander upstream installations. Both installations are extreme. The HMOP at 6.7D models a gate valve at 5D upstream. This is below the minimum
22
Fig 39. HMOP 2D down Fig 40. 3” Swirl Generator + Expansion 9D up
Cd = 0.803,+/-1% to 95% confidence+0.5%
-0.5%
+1%
-1%
0.76
0.77
0.78
0.79
0.8
0.81
0.82
0 500000 1000000 1500000 2000000 2500000 3000000 3500000 4000000Reynolds Number
Dis
char
ge C
oeffi
cien
t
BaselineDouble Out of Plane Bend 0D upstreamDouble Out of Plane Bend 2D upstreamDouble Out of Plane Bend 5D upstreamDouble Out of Plane Bend 0D upstream, Half Moon Plate 2D dow nstreamDouble Out of Plane Bend 0D upstream, Triple Out of Plane Bend 0D dow nstreamHalf Moon Plate 6.7D upstreamHalf Moon Plate 8.7D upstreamHalf Moon Plate 2D dow nstreamSw irl Generator w ith 3" to 4" Expansion 9D upstream
Fig 41. 4”, 0.63 beta ratio, ∆P cone meter disturbed flow discharge coefficient results. recommended upstream distance (of 6D) for this meter. The HMOP at 6.7D increases the discharge coefficient by approximately 0.8%. Extending the upstream distance to 8.7D (i.e. a gate valve at 7D) drops the discharge coefficient to with the baseline uncertainty. The extreme swirl with expansion 9D upstream dropped the discharge coefficient below the baseline uncertainty at lower Reynolds numbers. Even so all discharge coefficient data from all the disturbance tests are spread around the baseline calibrated value to ±1% at 95% confidence. (Nevertheless, DP Diagnostics suggests valves are installed no closer then 7D upstream of cone meters and inlet swirl conditions are limited to moderate swirl, say <300, especially if there is an expansion close by upstream.) Figures 42 & 43 show the disturbance effect on all the flow coefficient and DP ratios. All parameters are more affected than the discharge coefficient, but, crucially they are still relatively immune to disturbances. Commercial ∆P cone meters would be calibrated with straight pipe lengths only, so the baseline flow coefficient and DP ratio results must be held and only the uncertainty limits changed (i.e. increased) by standard agreed amounts to account for the possible real world use in extreme installations. It is proposed here that the standard uncertainties assigned to the calibration found diagnostic parameter values are those found in these extreme tests. Therefore the 15 parameters defining this ∆P cone
23
Fig 42. Flow coefficient disturbance test results for the 4”, 0.63 beta ratio ∆P cone meter.
Fig 43. DP ratio disturbance test results for the 4”, 0.63 beta ratio ∆P cone meter. meter are as shown in Figure 44. With the six diagnostic parameters and the nine associated uncertainties set, all baseline data sets (minus the 6.7D HMOP which is below the minimum distance allowed) can now be plotted on a NDB (see Figure 45). By design all points for the correctly operating meter are within the NDB. For a far more detailed discussion on these test results see Steven [5].
24
Fig 44. The ∆P cone meter diagnostic parameters and practical uncertainties.
Fig 45. All correctly operating 4”, 0.63 beta ratio ∆P cone meter data with a NDB.
5a. Incorrectly operating ∆P cone meter data Common ∆P cone meter problems include a partially blocked minimum area (or “throat”), damaged cones and DP transmitter issues. These scenarios are now discussed. All NDB’s used in the examples use parameter uncertainties shown in Figure 44. 5a1. Incorrectly operating ∆P cone meter – a partially blocked throat DP meters can trap objects in the flow. In this example a field blockage is simulated by a nut trapped at the cone. To make the test realistic the DOPB 0D upstream & HMOP 2D downstream installation was used (see Figure 35). Figures 46 & 47 show the trapped nut.
Fig 46. Trapped nut looking downstream Fig 47. Trapped nut looking upstream
25
Fig 48. Trapped nut induced flow rate error Fig 49. Trapped nut data with NDB. Table 8 shows the adverse flow condition data range used. Figure 48 shows the resulting traditional (i.e. equation 2) flow rate error. Traditionally there is no diagnostics available to alert the user to the problem. However, Figure 49 shows the data plotted with a NDB. The diagnostics clearly show a significant meter problem. 5b. Incorrectly operating cone DP meter – a saturated DP transmitter DP meters are reliant on DP transmitters. Modern transmitters have DP turn downs of up to 100:1 (depending on transmitter type, flow conditions and the allowable uncertainty). This allows flow rate turndowns of up to 10:1 without stacking transmitters. Most DP transmitters are extremely reliable. However, they are not infallible. There are several common DP transmitter problems. Incorrect DP readings result in incorrect flow rate predictions. The diagnostic system can indicate metering problems caused by many DP transmitter problems. We will discuss one common problem here as way of an example. DP transmitters have maximum ranges (or “Upper Range Limits”, URL’s). When the DP exceeds the URL the transmitter is said to be “saturated”. The following is a theoretical example of when the above calibrated 4”, 0.63 beta ratio ∆P cone meter has a saturated DP transmitter. In this example, the calibration is accepted as the meters performance. Let us consider a flow through this meter of 5.35 kg/s, with a pressure of 17.3 Bara, and a gas density of 20.76 kg/m3. The Reynolds number is 3.77e6 and the isentropic exponent is 1.3. Say the system uses a 400”WC URL transmitter for the traditional DP and 250”WC URL transmitter for the PPL. Say this particular system does not measure the recovered DP directly but calculates it by equation 1. Equation 2 shows the actual traditional DP to be 432”WC. This exceeds the transmitters URL by 32”WC. The transmitter is saturated and falsely reads a DP of 400”WC. As the meters actual PLR is 0.559 the PPL is 241”WC. This value is within the 250”WC DP transmitters range. (Note that the diagnostic method works regardless of whether this is so or not. If both transmitters saturate the stated PLR is calculated falsely as 0.625, i.e. an 11.8% shift from the calibration value, thereby exceeding the allowable 4.5% difference limit.) The actual recovered DP is therefore 191”WC but due to the saturated transmitter the system calculates it incorrect as 159”WC. The resulting traditional meter flow prediction is 5.16 kg/s, i.e. a -3.6% error. Traditional meters have no diagnostics to indicate this error. However, when the data is plotted with the NDB a clear error is indicated (see Figure 50).
26
Note that all correctly operating DP meters put the diagnostic data inside the NDB. However, if the meter is not operating correctly, but the DP’s being produced are being read correctly, the diagnostic data tends to be outside the NDB and in the first and third quadrants of the diagnostic plot. This is regardless of whether the problem is an incorrect meter installation, a physically damaged meter, a disturbed inlet flow, a contaminated meter or a throat blockage etc. There are fluid dynamic reasons why all real DP’s read must produce such a result. However, when the problem is with DP instrumentation, and not the meter body itself, the DP’s no longer represent the physical reality of what is happening in the meter. Therefore, the problem can produce DP’s that have no association with the real DP’s and hence, the results no longer have to obey the laws of physics. In this case the diagnostic data can end up anywhere on the diagnostic plot. Note therefore, that unlike any other test shown in this paper the saturated DP result has data in the second and fourth quadrant of the plot (see Figure 50). This is a sign that not only is the system not working but the reason is with the DP transmitters and not the meter body.
Fig 50. 4”, 0.63 beta cone DP meter with a saturated DP transmitter.
5c. Incorrectly operating ∆P cone meter data – a damaged cone assembly If a non-gusseted cone DP meter is dropped or heavily bumped during transportation or installation, or struck in service by a pressure spike or slug, the shock loading on the cone assembly can cause plastic deformation. Furthermore, a very substantial over speed beyond the design limits can also cause plastic deformation of the cone assembly.
Fig 51. Sketch of a 4”, 0.75 beta ratio cone DP meter.
The design of cone meters has the cone and meter body aligned. In reality manufacturing tolerances mean that the cone usually has a small deviation from the meter centre line.
27
Manufacturers attempt to make this deviation small, but it is unreasonable not to expect some deviation. DP Diagnostics built a wafer style 4”, 0.75 beta ratio ∆P Cone Meter (see Figure 51). The cone to body centerline deviation (i.e. xo) measured was 0.2380. This is as small a deviation as is practical for mass production techniques. Note that for this meter the downstream tap was located on the downstream pipe work. The CEESI straight pipe calibration results for the flow coefficients and DP ratios are shown in Figure 52 with the calibration uncertainties replaced with the larger uncertainties of Fig 44 in order to cover any meter installation. The discharge coefficient was actually calibrated to 0.5% uncertainty as shown in Figure 53.
Deflection x1 x2 y1 y2 Base (0.24o) 0.650” 0.645” 0.636” 0.659”
Very Mild (1.2o) 0.650” 0.645” 0.599” 0.697” Mild (2.1o) 0.650” 0.645” 0.562” 0.733”
Table 9. Cone deflection tests.
Fig 52. As built 4”, 0.75 beta ratio cone DP meter diagnostic parameter calibration.
Fig 53. 4”, 0.75 beta ratio, cone meter discharge coefficient variation with deflection.\ The cone assembly was then deliberately damaged to show the effects. The deflection was kept to that which was difficult to see with the naked eye. This is because this is unlikely to be found by any visual inspection between calibration and installation. The
28
Parameter Ranges Pressure Range 13.8 < P (bar) < 31.0
DPt Range 14”WC< DPt <276”WC DPr Range 8”WC <DPr < 156”WC
DPppl Range 6”WC <PPL < 121”WC Reynolds No. Range 946e3 < Re < 6.07e6
Table 10. Nominal flow conditions for the three cone angles tested. first deflection increased the misalignment to 1.20, the second to 2.10. The cone centering effect is shown in Table 9. Table 10 shows the test data ranges. Figure 53 shows the effect of the damage. The increase to 1.20 deflection caused a 0.4% over reading. The increase to 2.10 deflection caused a 1.2% over reading. There is no traditional method of diagnosing if a cone DP meter calibration is still valid. Figure 54 shows all this meters data on a diagnostic plot using the Figure 52 parameter uncertainties. The 0.4% shift of the 1.20 deflection is below the sensitivity of the diagnostic system. However, even with the elevated uncertainties imposed to account for all adverse flow conditions, the 2.10 deflection triggers the alarm through the PLR thereby alerting the user to the fact the meter is not behaving as it was when calibrated. Note more significant damage would be far more easily seen by the diagnostic system.
Fig 54. Normalized diagnostic box for as manufactured and after damage performance.
Conclusions The patent pending diagnostic methods are simple but very effective and of great practical use. The two different methods of comparing DP pairs both work well for cases where there are substantial problems. As the DP ratio technique is seen to be more sensitive than the direct flow rate prediction comparison technique it can typically see smaller metering problems (sometimes with flow prediction errors less than 2%) but very occasionally give false alarms. Thus, with the diagnostic method giving qualitative and not quantitative information, it can be beneficial to apply both methods simultaneously to gain some objectivity on the diagnostic results. If both techniques state a problem it is likely to be a significant problem and a general alarm is given to the meter user. If the DP ratio technique states a problem but the flow rate prediction comparison does not, a warning is given to the user. In this incidence the use of both techniques together allows the meter user to better judge the level of the likely error.
29
The proposed method of plotting the diagnostic results on a graph with a NDB aids the meter user in this task. Furthermore, although the diagnostics are qualitative and not quantitative, even in this stage of early development certain problems are known to produce a particular signature on the NDB plot thereby indicating what the problem is. Examples are the particular tell tale pattern of points if an orifice plate is installed the wrong way round, or points in the second or fourth quadrant suggest a DP reading problem rather than a meter body problem. As more experience is gained more understanding of the NDB results will be obtained. References 1. Steven, R. “Diagnostic Methodologies for Generic Differential Pressure Flow Meters”, North Sea Flow Measurement Workshop October 2008, St Andrews, Scotland, UK. 2. International Standard Organisation, “Measurement of Fluid Flow by Means of Pressure Differential Devices, Inserted in circular cross section conduits running full”, no. 5167. 3. Johansen, W.R. “Effects of Thin Films of Liquid Coating Orifice Plate Surfaces on Orifice Flowmeter Performance”, Colorado Engineering Experiment Station Inc. report for Gas Research Institution, No. GRI-96/0376. 4. Pritchard M. et al “An Assessment of the Impact of Contamination on Orifice Plate Meter Accuracy”, North Sea Flow Measurement Workshop, St Andrews, Scotland, 2004. 5. Steven, R. “Diagnostic Capabilities of ∆P Cone Meter”, International Symposium of Fluid Flow Measurement 2009, Anchorage, Alaska, USA.
1
Cone DP Meter Calibration Issues
Authors: Casey Hodges, Charles Britton, William Johansen & Richard Steven CEESI, 54043 Weld County Road 37, Nunn, CO 80648
Email: [email protected] , Telephone: US 970-897-2711 1. Introduction Cone DP flow meters are becoming increasingly popular in the oil and gas industry. A cone DP meter is a member of the generic Differential Pressure (DP) meter family and operates according to the same physical principles as other DP meter types. ISO 5167 [1] states the performance of orifice plate, nozzle, Venturi nozzle and Venturi DP meters across set geometry designs, over particular ranges of flow conditions. ISO 5167 covers these meters as they have a long history of research where the massed data sets are publicly available for scrutiny. However, ISO 5167 does not cover cone DP meters as the patent protection has only recently lapsed and no independent research has yet shown that cone DP meters of set geometries have repeatable and reproducible performances over given flow condition ranges. This paper reviews cone DP meter data from CEESI independent research, a CEESI wet gas Joint Industry Project and multiple third party1 tests. The cone DP meters discussed are produced by multiple manufacturers. Performance comparisons are made between nominally identical cone DP meters. The relative merits of calibrating cone DP meters with low Reynolds number water flows or high Reynolds numbers gas flows will be discussed. The pros and cons of cone DP meter periodic re-calibration is also discussed. The effect of damage changing the cone alignment will be considered. Finally, the prospect of cone DP meters being eligible for inclusion in ISO 5167 is discussed. 2. The Cone DP Meter Geometry and Principles of Operation Figures 1 shows a schematic sketch of a cone DP meter. There is an inlet (denoted with suffix “1”) of known area ( 1A ) where the inlet pressure ( 1P ) is read. Inlet pressure is usually read with one pressure port at the same radial position as the cones support bar but this can change between meters. A cone DP meter primary element consists of a support bar (which can be a variety of sizes relative to the cone) downstream of the high pressure port, holding a cone. The cones apex is attached to this support bar and pointing into the flow (at a half angle, θ = 260). A second cone of shorter length extends from the base of the first upstream cone, hence with the apex pointing downstream (at a half angle of 67.50, i.e. a base to cone angle, ψ = 22.50). The low pressure port2 (denoted with suffix “t”) extends through the cones and up through the support bar. The centre line of the cones should be aligned with the centre line of the pipe / meter body. The minimum cross sectional flow area (or “throat”) of a cone DP meter (At) is therefore the annular ring between the circumference of the communal base of the cones (or “beta edge”) and the 1 All third party data has been blinded to protect the privacy of individual cone DP meter owners. 2 The most common design of cone DP meter reads the low pressure at the back face of the cone, i.e. in the cones wake. However, a minority of cone DP meters have the low pressure port on the meter wall, in the vicinity of directly downstream of the cone. This paper only discusses the common design.
2
Fig 1. Sketch of a standard cone DP meter with flow profile, pressure and DP transmitters meter body wall. The distance between the centre lines of the upstream pressure port and the cone support is usually 2⅛” in order to facilitate close coupling with a DP transmitter. However, this distance can be changed as required for specific applications.
When metering a mass flow rate (.
m ) or volume flow rate (.
Q ) with a cone DP meter, the high pressure is read from the inlet pressure port (P1) and the differential pressure (or “DP”) is read as the difference between the inlet pressure and the low pressure (Pt) at the centre of the back of the cone. (i.e. tPPP −=Δ 1 ). That is, the low pressure is the pressure in the cones “wake”. A wake is a highly turbulent area of flow, deficient in momentum, directly behind bodies immersed in a fluid flow. The cone (i.e. the generic) DP meter mass and volume flow rate equations are shown here as equations 1 & 2 respectively.
PCEAm dt Δ= ρε 2.
--- (1) ρε PCEAQ dt
Δ=
2.
--- (2)
Note that ρ represents the fluid density. E is the “velocity of approach” and tA is the “throat area”. These are both constants for a set DP meter geometry. The velocity of approach is solely a function of the beta ratio as shown in equation 3. The cone DP meter beta ratio is calculated by equation 4 (where D is the meter inlet diameter and cd is the base cone diameter). The throat area tA is calculated by equation 5.
411β−
=E -- (3),
2
1
1 ⎟⎠⎞
⎜⎝⎛−==
Dd
AA ctβ -- (4), ( )22
4 ct dDA −=π
-- (5)
The expansibility (ε ) accounts for gas density fluctuation through the meter. In the US this parameter is called the expansion factor and denoted by the letter “Y”. The cone DP meter expansibility was developed by Stewart et al [2] and is shown as equation 6. Note that κ is the fluids isentropic exponent. For liquids expansibility is unity.
3
( )( )⎭⎬⎫
⎩⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛ Δ+−=
1
4 *696.0649.01PP
κβε --- (6)
The final component to the mass and volume flow rate equations is the discharge coefficient, dC . The discharge coefficient of cone (and all generic) DP meters is defined by equation 7. The numerator of equation 7 is the actual mass flow rate. The denominator is the DP meter theoretical flow rate calculation where no real world effects are accounted for. Therefore, the discharge coefficient of cone (and all generic) DP meters is defined as the ratio of the actual to the theoretically calculated mass flow rates.
tt
actual
ltheoretica
actuald PEA
m
m
mCΔ
==ρε 2
.
.
.
--- (7)
The discharge coefficient corrects the theoretical flow rate prediction for all inaccurate assumptions made by the theoretical equation development. That is, the discharge coefficient accounts for all unpredictable factors. Calibration of cone (and all generic) DP meters means testing the meter to find the discharge coefficient. As the actual flow rate through any flow meter is never truly known, it is standard practice to use a calibration facilities low uncertainty reference flow meter output as the “actual” flow rate. A consequence of this is that a cone flow meter uncertainty rating can therefore never be less than the reference meter uncertainty rating. A simple calibration method is to take an average or mid point discharge coefficient across the calibration Reynolds number (Re) range. However, it is sometimes necessary to data fit the discharge coefficient to the Reynolds number to get a lower flow rate uncertainty. The Reynolds number is shown in equation 8, and the data fit to the discharge coefficient in equation 9.
DQ
DmDU
forcesviscousforcesinertia
πμρ
πμμρ
..44Re ====
−
--- (8), ( )Re1fCd = --- (9)
Note that −
U is the average velocity, μ is the fluid viscosity and 1f is a unique data fit per DP meter, of discharge coefficient to Reynolds number. If a data fit is used rather than a constant discharge coefficient the mass or volume flows are calculated by substituting equation 8 into equation 9, and equation 9 into equations 1 or 2 respectively and iterating. Note, that the cone DP meter does not have a special, unique or different flow rate equation that fundamentally differs from any other DP meter flow equation. The basic theory of operation is the same as with all DP meters. However, it should be noted that compared to other DP meters the cone DP meter does have some practical advantages, e.g. a significantly higher resistance to upstream disturbances (see Peters et al [3]). 3. Cone DP Meter vs. Venturi Meter Calibration Requirements Two popular rival DP meter designs are the cone DP meter and the Venturi meter. The Venturi can sometimes be perceived to have an advantage as unlike the cone it is
4
included in ISO 5167 [1]. Hence, for precise Venturi meter geometries, under a given range of flow conditions, no calibration is required to find the Venturi meters discharge coefficient. However, the cone DP meter must always be calibrated to find its discharge coefficient. Therefore, within the Venturi meter ISO scope the Venturi has an advantage, as not requiring calibration can significantly reduces the cost of the meter. However, it should be noted that ISO 5167 is only valid over a stated range of Venturi meter geometries and flow conditions. In fact, in many (if not most) natural gas production meter applications, the flow conditions are out with the limits of the ISO Venturi meter standard. Extrapolating the ISO discharge coefficient prediction to other conditions is not valid. In this situation the Venturi meter reverts to the status of all “non-standard3” DP meters (e.g. the cone DP meter) in as much as the discharge coefficient must be found by calibration at the range of flow conditions for which the meter will be used. ISO 5167 includes a discussion on the high precision machined convergent section Venturi meter. This is the common type used to meter natural gas production flows. The limits of this meters ISO performance declaration are:
50 mm (2”) ≤ D ≤ 250 mm (10”) 0.4 ≤ β ≤ 0.75
2e5 ≤ Inlet Reynolds Number (D) ≤ 1e6
(Annex B of ISO 5167 gives an informative discussion only on the possible affects of higher Reynolds numbers with significantly increased discharge coefficient uncertainties suggested.) It is stated that under performance declaration the discharge coefficient is a constant, i.e. 995.0=dC to an uncertainty of ±1%. However, ISO 5167 also states: “Research into the use of Venturi tubes in high-pressure gas [ ≥ 1 MPa ( ≥ 10 bar)] is being carried out at present. In many cases for Venturi tubes with machined convergent sections discharge coefficients which lie outside the range predicted by this part of ISO 5167 by 2% or more have been found. For optimum accuracy Venturi tubes for use in gas should be calibrated over the required flowrate range.” Furthermore, ISO also explain that a simultaneous use of the extreme values of D, β, Re(D) shall be avoided as otherwise the Venturi meter flow rate uncertainty is likely to increase. They therefore state that for installations outside theses diameter, beta ratio, pressure and Reynolds number limits, it remains necessary to calibrate the meter in its actual conditions of service. Whereas Venturi meters are popular in the natural gas production industry most applications have pressures greater than 10 bar (abs) and Reynolds greater than 1e6 and many applications have pipe diameters greater than 10”. Therefore, in many (if not most) natural gas production applications the ISO Venturi meter standard is inapplicable. Furthermore, Venturi meter performance outside the ISO limits has been researched. It is known that the ISO discharge coefficient is not always applicable and there can be reproducibility problems between nominally identical meters (e.g. Geach et al [4]). 3 “Non-standard” is defined here as any meter that does not have a standards body approved discharge coefficient prediction. This term is not to be associated with whether or not a meters manufacturing process is standardized.
5
Hence, the blind application of the ISO stated discharge coefficient could lead to flow measurement error. Therefore, Venturi meters with flow conditions outside the ISO scope should be calibrated across the full Reynolds number range of the meters application. Cone DP meters must be calibrated across the full Reynolds number range of the meters application. Failure to calibrate the cone DP meter correctly can lead to high flow rate uncertainties. The scale of these cone DP meter uncertainties will be discussed in this paper. Therefore, both Venturi and cone DP meters should be individually calibrated, and hence, in most natural gas production applications the cone DP meter does not have a calibration cost disadvantage against a Venturi meter. The cone DP meter is an increasingly popular meter, in part due to its high resistance to upstream disturbances. The cone DP meter is no longer patent protected. It is therefore a candidate for future inclusion in standards documents. However, an independent evaluation of cone DP meter repeatability and reproducibility is required as an initial step to discussing this meters possible future inclusion in standard documents. The authors now offer an evaluation of cone DP meter performance. 4. Cone DP Meter Performance and Calibration Issues Although the cone DP meter has no standard it is known from experience that the cone DP meter discharge coefficient is approximately 0.8. However, it is also known from experience that this can vary between individual meters by ±8% (and occasionally more). This can be due to several reasons such as the relative size of the support bars changing between meters, slightly different cone designs (due to manufacturing considerations for different diameter meters) and deliberately liberal manufacturing tolerances to ease manufacturing complexity. The manufacturers can allow this as they typically state that each cone DP meter should be calibrated across the full Reynolds number range of the application. It is stated by the manufacturers that if each cone DP meter is properly calibrated the meter has “up to 0.5% uncertainty”. It is a performance fact of DP meters that, as long as the expansibility is accounted for, it does not matter what fluid4 is used to calibrate the meter. Therefore, a natural gas flow DP meter can be calibrated in an air flow calibration facility or a gas DP meter can be calibrated with a water flow calibration facility etc.. The single, but critical, stipulation is that the Reynolds number range of the application is met. If this stipulation is met, a DP meter calibration carried out with one fluid is applicable to when the meter is in use with any other fluid. Water flow meter calibration facilities are simpler and less expensive to operate than gas flow meter calibration facilities. Therefore, calibrating a DP meter with a water flow can be attractive to both manufacturer and DP meter users. However, there can be a significant potential problem with this approach. This problem hinges on the fact that the Reynolds number range of the application must be met. Equation 8 shows that the Reynolds number is a function of the fluid density, average fluid velocity, the inlet diameter and the fluid viscosity. For a given meter the inlet diameter is of course set. However, if we consider a set velocity value we see that the Reynolds number is a 4 In this statement, “fluid” means a Newtonian fluid. This statement is not true of non-Newtonian fluids.
6
function of the fluid density and viscosity. Liquids are considerably denser than gases (even at extremely high pressures) but gas is typically a couple of orders of magnitude less viscous than liquids. Hence, for any DP meter with a flow with a set average velocity, a liquid flow has a Reynolds number an order of magnitude less than a gas flow. The effect this has on DP meter calibration is that it is unlikely that a water calibration facility can reach the upper Reynolds number values required for many (if not most) gas flow metering applications. Therefore, if water flow calibration data is used to find a DP meters discharge coefficient for a gas flow application, it is likely that only the lower end of the gas applications Reynolds number range will be reached even at the maximum capabilities of the water flow calibration facility. Where calibration data from liquid and gas facilities have matching Reynolds numbers the discharge coefficient will be the same. However, at higher gas flow application Reynolds numbers, where water flow test data must be extrapolated, the discharge coefficient is being estimated only. As DP meters often have a discharge coefficient vs. Reynolds number relationship that is not constant this extrapolation can and does lead to substantial metering errors. Cone DP meters operate according to generic DP meter rules. If these rules are met then the cone DP meter is likely to be a reliable flow meter capable of giving flow rates to an uncertainty of 0.5% across a 10:1 turndown5. As with all common DP meters there are no moving parts and therefore a properly calibrated and installed cone DP meter should be reliable and need little to no maintenance. If no damage, wear, trapped foreign objects or contamination issues occur then the calibration result, and hence the performance, should remain constant (as long as the instrumentation is properly maintained). However, failure to calibrate the cone DP meter correctly can lead to an increase in flow rate measurement uncertainty or a significant bias on the flow rate measurement. The following discussion shows examples of potential calibration issues with cone DP meters. 4a. The Necessity for Calibration Across the Applications Full Reynolds Number Range If a DP meter is calibrated across a relatively low Reynolds number range (e.g. with water flow calibration facility) it is often not possible to see any discharge coefficient relationship with the Reynolds number over a larger turndown. As a result a low Reynolds number range / water calibration can give the illusion that the meters discharge coefficient is constant, and / or can suggest the performance at higher Reynolds numbers is different to what it actually is. Often calibration across a larger turndown (usually by means of gas flow tests) shows extrapolation of lower Reynolds number data to be incorrect. Hence, a cone DP meters uncertainty rating is only applicable within the Reynolds number range of its calibration. Extrapolating a low Reynolds number calibration for use with high Reynolds number flows invalidates the uncertainty rating and may lead to significant bias in the flow measurement. Examples are now given. 5 There is a debate about the flow rate turndown capability of DP meters. A traditional limit is a very modest 3:1. This corresponds to an approximate DP transmitter turndown of 9:1 (see equation 1). However, DP transmitters have improved considerably since this traditional DP meter limit was set many years ago. Modern DP transmitters typically give reliable DP turndowns between 50:1 and 100:1 which corresponds to DP meter flow rate turndowns between approximately 7:1 and 10:1. This can be further extended by stacking DP transmitters of different ranges. Naturally, the uncertainty of any DP transmitter increases at the lower end of its range. A DP meters turndown rating depends on the calibrated Reynolds number range as well as acceptably accurate DP readings.
7
Fig 2. 2”, 0.75 beta ratio cone DP meter water and gas calibration results.
A 2”, 0.75 beta ratio cone DP meter was supplied to a gas flow application with a water calibration. The discharge coefficient was stated to be 0.7754 based on a water flow calibration which had a maximum Reynolds number of 114,606. During use at considerably higher Reynolds numbers a potential performance problem was noted. By plotting subsequent gas flow calibration data at Reynolds numbers up to 4e6 it was found that extrapolation of the water calibration data was causing a 4.5% under-reading of the gas flow rate. Data fitting all the data across the full Reynolds number range gave a meter uncertainty of 0.5% as required. Figure 2 shows these results.
Fig 3. 4”, 0.45 beta ratio meter X, water flow calibrated and gas flow calibrated.
CEESI tested several cone DP meters as part of a wet gas flow research Joint Industry Project (JIP). The initial research was to confirm the meters gas flow performances as stated by the manufacturer. A cone DP meter manufacturer supplied a 4”, schedule 80, 0.45 beta ratio meter (say meter X) with a stated discharge coefficient set to 0.833 at 0.5% uncertainty. CEESI applied this manufacturer supplied discharge coefficient and discovered that the meter was consistently under predicting the gas flow rate. As the first
8
meter was removed for inspection a second meter (say meter Y) of the same specification (built from the same drawing as meter X) was therefore dispatched to CEESI by the manufacturer. This had a manufacturer stated discharge coefficient of 0.807 and 0.5% uncertainty. It was subsequently discovered that both meters had been calibrated by the manufacturer with water flows only with a maximum Reynolds number of 340,000. The JIP had tested the meter at CEESI with gas flows at Reynolds numbers up to 4e6. All meter X data (i.e. the water and natural gas flow data) is shown plotted in Figure 3. As the Reynolds number increased beyond the limit of the water calibration it was seen that the discharge coefficient was not constant (as implied by the manufacturers supply of a constant value) but rather it increased with Reynolds number. Therefore, when the JIP extrapolated the water calibration found discharge coefficient to the highest Reynolds numbers tested the flow was under-predicted by 2.46%. Data fitting all the data, i.e. the water and the various pressure6 gas flow data sets gave a flow rate uncertainty of ±0.6%7 for the 10:1 turndown. (It is very likely that the meter would have calibrated to ±0.5% if a gas calibration facility had been used.)
Fig 4. 4”, 0.45 beta ratio meter Y, water flow calibrated only.
The replacement cone DP meter (Meter Y) was produced to the same specification as the original meter (Meter X). However, this time the manufacturer had calibrated Meter Y across a range of Reynolds numbers within their water calibration facilities range and not simply the highest Reynolds number obtainable. However, this data could be interpreted as the discharge coefficient having a constant value or having a relationship with Reynolds number. Figure 4 shows the manufacturer water flow data and the CEESI gas flow data. The results on the whole combined data set of taking a constant discharge
6 Note that varying the pressure / gas density made no difference to the calibration. As long as the Reynolds number value is met any DP meter is calibrated correctly regardless of the line pressure. 7 Note that the CEESI wet gas loop is not designed as a gas flow calibration system and therefore does not give the same standard of gas reference metering as a meter calibration system. For example, one CEESI gas flow meter calibration system has a reference (critical flow Venturi) meter with an uncertainty < ±0.35%. The CEESI wet gas test facility running dry has a reference (turbine) meter with an uncertainty < ±0.7%. All uncertainties discussed in this paper refer to the difference between test and reference meters.
9
Fig 5. 4”, 0.45 beta ratio meter Y, calibrated with all data.
coefficient from the water data and fitting a linear line through the water data is shown. The water flow data based constant discharge coefficient again under-read the gas flow rate at the upper end of the applications Reynolds number by up to -1.71%. The linear line fitted to the water data only (which simulates only having the water data available for meter calibration) improved the situation. This linear fit fitted all the data to an uncertainty of ±0.93% across a 10:1 turndown. However, Figure 5 shows that by data fitting all the water and gas flow data (i.e. by calibrating the meter correctly) the resulting uncertainty was ±0.5%, across a 10:1 turndown. That is, when calibrated correctly, the meter met the manufacturer’s uncertainty claim.
3.810", 0.6021 Beta Ratio Cone DP MeterMeter Z
Water Data Fitted to All DataCd = 0.8232
+/- 0.85% to 95% confidence(in reality mostly negative bias)
All Data FittedCd = 0.8275 + (-7E-10*Re)+/-0.5% to 95% confidence
0.8
0.81
0.82
0.83
0.84
0.85
0 1000000 2000000 3000000 4000000 5000000 6000000Pipe Reynlods Number
Dis
char
ge C
oeffi
cien
t
CEESI JIP Natural GasCEESI Air LabWater Lab
Fig 6. 4”, Schedule 80, 0.60 Beta Ratio Meter, Water and Gas Flow Calibrated. Figure 6 shows the calibration results of a 4”, 0.6 beta ratio cone DP meter (Meter Z). The water calibration point is the maximum Reynolds number achieved at a water flow facility. The resulting discharge coefficient (Cd=0.8232) was supplied. The natural gas flow test data at various pressures is shown (as one data group) along with a later air calibration at CEESI. Fitting all three data sets allowed a linear fit to predict the discharge
10
coefficient to the manufacturers claimed uncertainty of ±0.5% across a 10:1 turndown. If the water based low Reynolds number calibration had been accepted and used by extrapolation, the resulting uncertainty would have been ±0.85%, across a 10:1 turndown. However, note from Figure 6, it could be argued that this is not really an increase in uncertainty but the introduction of a small bias as the water calibration data has a lower discharge coefficient than more than 95% of the higher Reynolds number gas flows. Note that Figures 2 thru 5 show a discharge coefficient to Reynolds number relationship with a positive gradient. This is not always the case as shown by Meter Z in Figure 6. It is not good measurement practice to extrapolate cone DP meter (or any meter) calibrations to higher Reynolds numbers. Extrapolating low Reynolds number data sets to predict the discharge coefficient at high Reynolds numbers can give at best an increase in uncertainty, or a small bias, or worse, give gross errors. It is good practice to calibrate the meter across the full Reynolds number range of the application. Only then is a flow meter uncertainty statement meaningful. If a cone DP meter is calibrated across the full Reynolds number range of the application the meter usually gives ±0.5% at 95% confidence across a turndown of 10:1. If it is not calibrated across the full Reynolds number range the uncertainty in flow measurement is simply unknown outside the calibrated range. 4b. The Necessity to Calibrate Each Individual Cone DP Meter Many flow meter applications are in systems where multiple identical meters are required. If multiple meters are ordered, which are on paper said to be identical, there is a temptation to calibrate one or two meters only and apply that calibration to all meters of that specification. The rational of this proposed approach is based on the assumption that because the meters are said to be identical their performance under the same flow conditions should also be identical. Therefore, this common argument is wholly based on the assumption that because the meters are identical on paper they are also identical in reality. However, in reality there are manufacturing tolerances. No two flow meters are truly identical. With the current typical cone DP meter manufacturing tolerances, although meters are identical on paper they can be subtly different in practice. As the manufacturers state each meter should be individually calibrated this is not in itself an issue. However, as it would be advantageous to not have to calibrate multiple meters of the same specification it is interesting to know what shifts in discharge coefficients are caused by the subtle differences between the nominally identical cone DP meters. Unfortunately there is little in the literature that describes the level of geometric variation between nominally identical cone DP meters before they begin to have significantly different characteristics. Hence, at the time of writing, industry has no guarantee two cone DP meters built from the same drawing are in fact identical or have the same performance characteristics. The two 4”, schedule 80, 0.45 beta ratio meters discussed in section 4a where identical on paper. They were built by the same manufacturer, at the same fabrication shop, from the same drawing. However, it should be noted that the first meter (Meter X) had an ID of 3.812” and a beta ratio of 0.4512, whereas, the second meter (Meter Y) had an ID bore of 3.823” and a beta ratio of 0.4500. When the two meter calibrations were compared (see Figure 7) it was found that there was approximately 4% difference between the meters
11
Fig 7. Comparison of two nominally identical cone DP meter calibrations.
Fig 8. Applying data fits from one cone DP Meter to another of the same specification.
discharge coefficients. If one meter was calibrated only, and the result applied to the other meter (on the assumption it will behave the same), then the un-calibrated meter would have a significant bias in its flow rate predictions. Figure 8 shows this scenario. However, it is noteworthy that each meter individually calibrated across the Reynolds number range to give a low uncertainty across a relatively large turndown. There is plenty of evidence that this is a typical result. Figure 9 shows the comparison of three 4”, schedule 80, 0.75 beta ratio cone DP meters, two from one manufacturer and one from another. On paper they should have an identical performance. In reality all three meters have similar but not identical performances. Again, individually the meters calibrate to give an uncertainty of < ±0.6%8 across relatively large turndowns. (Note that meter 2 required a non-linear fit to meet the ±0.5% uncertainty specification. Such fits, 8 Meter 1 only has data from a wet gas facility running dry. This facilities reference meter is not of gas meter calibration standard, hence causing higher uncertainty. It is therefore likely all three meters could be calibrated to ±0.5% uncertainty.
12
Nominal 4", Sch 80, 0.75 Beta Ratio Cone DP Meters
Cd = 0.8083 - (2E-09*Re)+/- 0.6% to 95% confidence Cd = 0.7953 + (4E-10*Re)
+/- 0.3 to 95% confidence
0.760.770.780.790.8
0.810.820.830.840.850.86
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000 8000000
Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
tManufacturer 1, Meter 1, Natural Gas 3.815", 0.7512 Beta Ratio Manufacturer 2, Meter 2, Air & Natural Gas 3.826", 0.7500 Beta RatioManufacturer 2, Meter 3, Air, 3.826", 0.7500 Beta Ratio
Cd = 0.79 - (3343948/(Re 1̂.5))+/- 0.5% to 95% confidence
Fig 9. Comparison of nominally 4”, 0.75 beta ratio cone DP meter calibration data sets.
4", sch 80, 0.75 Beta Ratio Cone DP Meter Data
+0.5%
-0.5%
-3
-2
-1
0
1
2
3
0 2000000 4000000 6000000 8000000 10000000 12000000
Pipe Reynolds Number
% E
rror
Meter 3, Calibrated to Meter 3 DataMeter 1, Calibrated to Meter 3 DataMeter 2, Calibrated to Meter 3 Data
Fig 10. Results of applying one meter calibration to other same specification meters. when required, are standard practice across industry.) If we assume that the meters have the same performance then we assume one calibration describes all three meter performances. Again, this assumption leads to significant flow metering errors. Figure 10 shows the induced metering error of applying meter 3’s calibration on meters 1 and 2. The resulting flow metering error can therefore be several percent. Note that these errors are a bias, not an increase in flow rate uncertainty. Finally, note that Figure 9 shows meter 2 with a distinct rise in discharge coefficient at the lower Reynolds number range before it levels off. If calibrated at low Reynolds number only and the result extrapolated to high Reynolds number a significant bias could exist. A third example is shown in Figures 11 and 12. Four 8”, schedule 40, 0.7 beta ratio cone DP meters (say meters A, B, C & D) were built by one manufacturer and calibrated at CEESI across the full Reynolds number range of the application. Individually the meters were all found to have a ±0.5% performance across the full turndown. (The individual
13
Nominal 8", 0.7 Beta Ratio Cone DP Meter
0.780.790.8
0.810.820.830.840.850.860.870.88
0 400000 800000 1200000 1600000 2000000Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t
Meter A, 7.974", 0.6943 Beta RatioMeter B, 7.974", 0.6943 Beta RatioMeter C, 7.993", 0.6985 Beta RatioMeter D, 7.977", 0.6975 Beta Ratio
Fig 11. Four nominally 8”, 0.70 beta ratio cone DP meter calibration data sets.
Nominal 8", 0.7 Beta Ratio, Cone DP MetersMeter A Calibration Applied to all Nominally Similar Meters
+0.5%
-0.5%
-2-101234567
0 400000 800000 1200000 1600000 2000000Pipe Reynolds Number
% E
rror
Meter A, 7.974", 0.6943 Beta RatioMeter B, 7.974", 0.6943 Beta RatioMeter C, 7.993", 0.6985 Beta RatioMeter D, 7.977", 0.6975 Beta Ratio
Fig 12. Results of applying meter A’s calibration to meters of the same specification. linear data fits are presented in section 4d.) The difference in performance of meters of the same specification is again clearly noticeable. Figure 11 shows that the variation in discharge coefficient value is approximately 5%. Figure 12 shows the result of accepting meter A as the calibration data for all four meters. This introduces significant biases for Meters B, C & D of up to 4%. It is possible that two cone DP meters built from the same specification could have identical performances. Figure 13 shows two 14”, schedule 40, 0.46 Beta Ratio cone DP meter calibrations. Here it was found that not only did the two meters (say meters E & F) both individually calibrate with a linear fit to a ±0.5% performance across the full turndown, but that they both had the same performance. Figure 14 shows the result of using each meters calibration on the other meter. In this case the meters uncertainty rating stayed at ±0.5%. It is therefore clear that multiple cone DP meters built to the same specification may or may not have the same performance. This situation will continue until such time as the cone DP meter manufacturers can guarantee that any small
14
Nominal 14", sch 40, 0.46 Beta Ratio Cone DP Meters
Cd = 0.831 + (2E-09*Re)+/- 0.5% to 95% confidence
0.8
0.81
0.82
0.83
0.84
0.85
0.86
0 500000 1000000 1500000 2000000 2500000Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t
Meter E, 13.126", 0.4589 Beta RatioMeter F, 13.170", 0.4585 Beta Ratio
Fig 13. Calibrations of meter E & F, i.e. 14”, 0.46 beta ratio cone DP meters.
Nominal 14", Sch 40, 0.46 Beta Ratio Cone DP Meter
+0.5%
-0.5%
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0 500000 1000000 1500000 2000000 2500000Pipe Reynolds Number
% E
rror
Meter E, 13.126", 0.4589 Beta RatioMeter F, 13.170", 0.4585 Beta Ratio
Fig 14. Applying meter E’s calibration to meter B’s performance and vice versa.
variations in meters built to one specification are small enough as to not affect the meters performance. However, as yet there is not enough information available to produce guidelines on the required manufacturing tolerances to ensure cone DP meters of the same specification have reproducible calibrations. Therefore, until such times as this information becomes publicly available with substantial third party data backing the reproducibility claims, each cone DP meter should be individually calibrated, across the full Reynolds number of the application, if a low flow measurement uncertainty is to be achieved. It therefore appears to be premature to be considering developing a cone DP meter standard. 4c. A Discussion on the Requirement for Periodic Re-Calibration If a DP meter is of a set geometry then its performance (i.e. the relationship between the Reynolds number and the discharge coefficient) will be set. This is the basis for the ISO 5167 standards. It is the fact that cone DP meters often have differences in their
15
manufacture that causes the discharge coefficient to shift between meters that are built from the same specification. If two cone DP meters were truly identical their performance would be identical and only one cone DP meter out of a batch of identical meters would require calibration for the performance of all the meters to be found. It therefore follows that if a particular cone DP meter is calibrated, then as long as the geometric shape of the meter has not changed (e.g. from contamination, erosion, corrosion, plastic deformation of the cone assembly due to adverse flow conditions, or no foreign object is trapped at the cone) then its calibration result will remain constant. That is, periodic re-calibration of any DP meter is only required if confirmation is required that the geometry has not changed. (However, note that all instrumentation must be regularly re-calibrated.) Proof of this can be seen if we discuss meters A, B, C & D in section 4b. These data sets shown in Figures 11 are actually multiple data sets for each meter. These meters are recalibrated every year to assure the owner that the previous calibration is still valid. Each meters separate data sets are shown in Figures 15 through 18 respectively. Meters A, B & D had the same calibration result each year. Note that meter D has two calibration data sets on the same day. This is due to the procedure of the meter being calibrated as delivered to the facility and then cleaned of any contamination. As no significant difference is seen this indicates no significant contamination problem existed. Meter C has four data sets. One pair is a repeat set from 2004. This again is calibrations before and after cleaning. Here the meter was delivered as it left service, calibrated and then cleaned and re-calibrated. A small shift is seen indicating that the contamination level of this meter was enough to have caused a small shift in performance (see Figure 17, 2004 set 1). On cleaning, the meter returned to its standard performance (see Figure 17, 2004 set 2). The ±0.7% data fit in Figure 17 is for all the data, including the contaminated cone DP meter data set. If this data set is removed the meter can be shown to calibrate to ±0.5% like the other meters. This is a very common result. For example, meters E & F (in section 4b) also show repeat calibrations after periods of service. Figures 13 and 14 actually show multiple calibration data for each meter. No significant difference exists between periodic calibrations.
7.974", 0.6943 Beta Ratio Cone DP MeterMeter A
Cd = 0.8425 + (1E-08 * Re)+/- 0.5% to 95% confidence
0.82
0.83
0.84
0.85
0.86
0.87
0.88
0 200000 400000 600000 800000 1000000 1200000 1400000 1600000 1800000Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t
4/8/20045/21/20056/8/2007
Fig 15. A comparison of repeat calibration results for meter A.
16
7.974", 0.6943 Beta Ratio Cone DP MeterMeter B
Cd = 0.8505 + (1E-08*Re)+/- 0.5% to 95% confidence
0.83
0.84
0.85
0.86
0.87
0.88
0 400000 800000 1200000 1600000 2000000Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t
8/24/200411/3/200511/7/2006
Fig 16. A comparison of repeat calibration results for meter B.
7.993", 0.6985 Beta Ratio Cone DP MeterMeter C
Cd = 0.8257 + (4E-09*Re)+/- 0.7% to 95% confidence
0.8
0.81
0.82
0.83
0.84
0.85
0 400000 800000 1200000 1600000 2000000
Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t
7/19/2004 set 27/19/2004 set 110/19/20057/26/2006
Fig 17. A comparison of repeat calibration results for meter C.
7.977", 0.6975 Beta Ratio Cone DP MeterMeter D
Cd = 0.8412 + (3E-09*Re)+/- 0.5% to 95% confidence
0.8
0.82
0.84
0.86
0.88
0.9
0 400000 800000 1200000 1600000
Pipe Reynolds Number
Dis
char
ge C
oeffi
cien
t 7/19/20047/20/2004 set 17/20/2004 set 2
Fig 18. A comparison of repeat calibration results for meter D.
17
Hence, if no physical damage, contamination or throat blockage occurs to a cone DP meter, the calibration of the meters remains valid indefinitely. However, if physical damage, contamination or throat blockages are potential events in the meters application and the meter user requires a low flow measurement uncertainty, then periodic recalibrations of the meter can give the user assurance that the stated calibration flow measurement uncertainty is being achieved. 4d Issues with Estimating Discharge Coefficients with no Calibration ISO 5167 [1] states that the Venturi meter has a constant discharge coefficient of 0.995, if the meter is within a set geometry range and the flow conditions are within set condition ranges. Although these geometry and flow condition ranges are rather limiting, within these ranges it has been shown that the Venturi meter discharge coefficient is repeatable and therefore calibration is not required. As the cone DP meter is a competitor to the Venturi meter and a candidate for future ISO standards work, it is of interest to plot all available cone DP meter data together to investigate the variation of the discharge coefficient between meters. The meter size restriction here is, for the sake of comparison, held as the same as for the Venturi (i.e. 50 mm ≤ D ≤ 250 mm) but the beta ratio range has been slightly altered to the common cone DP meter range in industry (i.e. 0.45 ≤ β ≤ 0.85). The Reynolds number and pressure limitations of ISO 5167 Part 4 were ignored as they are largely inappropriate to the oil & gas industry. Figure 19 shows all thirty available, non-manufacturer, third party owned cone DP meter data sets plotted on one discharge coefficient vs. Reynolds number graph. It can be seen in Figure 19 that each individual meter gave data that could be (and was) fitted to give meter flow rate uncertainties of ±0.5% across their respective turndowns. However, there is a wide spread of discharge coefficients between the meters. With the mid discharge coefficient of the spread being approximately a value of 0.8 the spread in discharge coefficient is approximately ±8.5%. There may be some relationship between discharge coefficient, meter diameter, and beta ratio but as yet the authors know of no research in the public domain regarding this. With that said it is evident from Figure 19 that several meters have the same specification and yet have significantly different performances. This is likely due to some relationship between manufacturing tolerance parameters (e.g. angle of cone alignment with pipe centerline) and the discharge coefficient. Currently, it does not seem possible to predict a cone DP meters discharge coefficient to low uncertainty (i.e. the typically desired ±0.5%) and hence to achieve a low flow rate uncertainty each cone DP meter must be individually calibrated across the full Reynolds number range for which it will be used. 4e The Effect of Misaligned Cone Assemblies on a Cone DP Meter Performance Many cone DP meters are manufactured to a simple technique where the cone and meter body centre lines are not assured to be aligned. It is not unusual for there to be an off set of up to several degrees. Manufacturers state that each meter should be individually calibrated and hence any effect caused by this misalignment can be calibrated out of the metering system. This is true. However, not all cone DP meters are calibrated, at least not across the full Reynolds number range of the application. Furthermore, there is no information in the public domain regarding what variation in discharge coefficient is
18
Fig 19. Cone DP meter blinded data sets, Reynolds Number vs. discharge coefficient.
19
caused by two otherwise identical meters having different cone misalignments. It is widely thought likely that the significant differences in discharge coefficient that can exist between cone DP meters of the same specification (e.g. see Figure 19) is mainly due to this cone misalignment effect. That is, small variations in cone alignments are expected to have a significant effect on the cone DP meters performance. Deliberately generous tolerances for cone DP meter manufacture (to reduce cost and shorten delivery time) are not the only mechanism that can cause a cone to be misaligned. Mild damage can also occur. The cone is a cantilever9 with a hollow support. This support is designed to be easily strong enough to hold the cone element under normal conditions. However, if a cone DP meter is dropped, generally mishandled in transit, experiences abnormal flow conditions etc., a small amount of plastic deformation of the support can occur. If the cone alignment shifts from the calibration position it is unlikely that the calibration is still valid.
Fig 20. Sketch of the actual 4”, schedule 80, 0.75 beta ratio cone DP meter built.
Fig 21. Sketch of the damaged 4”, schedule 80, 0.75 beta ratio cone DP meter
9 Some cone DP meters have gussets supporting the cantilever cone design. Gussets increase the cone assemblies strength, stiffness and therefore natural frequency, thereby reduces the likelihood of damage by shock loading or fatigue. Most small cone DP meters (i.e. <6” diameter) have non-gusseted cantilever cone designs due to the difficulty of inserting gussets in small meters. Most large cone DP meters are gusted cantilever cone designs. Gussets are not known to affect meter performance and they do help prevent cone assembly damage considerably.
20
Fig 22. Correctly aligned cone. Fig 23. Approx 10 misaligned cone. It is important to know what affect cone misalignment has on a cone DP meters performance. This information can aid understanding on why meters built to the same specification have different discharge coefficients. It would also aid understanding on what level of metering error can be expected if a cone alignment is shifted due to damage after calibration. Therefore, a 4”, 0.75 beta ratio cone DP meter with a well centred cone was calibrated and then the cone assembly was deliberately bent out of alignment with the meter body. The damage was slight both to simulate a typical variation of cone alignment between manufactured meters and to simulate typical small damage levels that could go unnoticed by meter operators. Figure 20 shows the meter as built. The reality of the manufacturing process produces some asymmetry. The cone is misaligned from the meter body centre line by approximately 0.24°. This is a well manufactured meter and this is as close to no deflection as is reasonable for a standard mass produced meter. Figure 22 shows a photograph looking upstream into the meter body (with the cone support at 9 o’clock). No significant deflection is noticeable with the naked eye. Figure 21 shows a sketch of the same meter after the cone assembly was bent upwards to simulate damage due to over stressing of the support. The deflection is very modest at approximately 0.95° (making a cone angle of 1.19° with the meter body centre line). Figure 23 shows the cone DP meter after the cone deflection. Note how small the deflection is to the naked eye and how unlikely it would be that this would be noticed by a standard visual inspection. In fact, this change in cone orientation is small enough to be within the typical manufacturing tolerance of a cone DP meter. That is, two meters built from the same drawing could also have this difference alignment due to manufacturing tolerances alone. The undamaged meter was calibrated as shown in Figure 24. A linear data fit gave the required ±0.5% uncertainty. Borders at ±0.35% are shown to give perspective on the effect of damage. Figure 24 also shows the cone deflection caused the discharge coefficient to reduce by approximately 0.4%. Thus even small changes in cone alignment seem to significantly affect calibrations. This modest change in cone alignment can be obtained by rough handling of cone DP meters in transit or adverse field conditions. It is therefore advisable to treat a cone DP meter as an instrument that requires care and not simply as a pipe spool. Measurement of the cone alignment during calibration allows it to be periodically rechecked. Any significant shift indicates recalibration is advisable.
21
Fig 24. Calibration data for before and after cone deflection. It is suspected that larger cone / smaller beta ratio cone DP meters will be more susceptible to shifting discharge coefficients as the cone alignment shifts because there is less distance between the beta edge and the meter wall. Hence, there is a more severe change in the meter throat’s annular ring shape for a given cone deflection angle. Unfortunately, the heavier the cone is relative to a set support bar size the more susceptible the meter is to damage. Gussets greatly help reduce this problem. The strength of a cone DP meter without gussets is dependent on the outside diameter, schedule and material specifications of the support bar. Naturally, for a given material, the larger the outside diameter and the higher the schedule (i.e. the smaller the inside bore) the higher the support bars second moment of area and the higher the applied force required before the support bars yield stress is exceeded causing plastic deformation. Therefore, it is advisable for cone DP meters without gussets to be built with as large a support and schedule as possible before the support starts interfering with the cones operation. There is currently little data available on how large a support can be before it starts to interfere with the operation of the cone DP meter. 5. The Potential for Cone DP Meter Sizing Errors DP meter uncertainty is related to DP transmitter uncertainty. The DP transmitters output has its lowest uncertainty at its upper range limit (or “URL”). It is therefore common practice for a cone DP meter to have the beta ratio sized for a given pipe size, schedule and flow condition range such that the maximum expected DP flow conditions give the URL of the chosen DP transmitter. However, this procedure inherently assumes the precise actual maximum DP the meter will see in service can be predicted from discharge coefficient and application flow rate estimates. In reality, if the actual flow conditions produced a greater DP than the transmitters URL the transmitter is said to be “saturated”. A saturated DP transmitter registers the URL and not the actual higher DP. A DP meter can not meter the flow rate correctly with a saturated DP transmitter. Here then, is a very significant potential pitfall to an engineer sizing a DP meter. The estimated DP from a DP meter is found by rearranging equation 1 to give equation 1a:
22
2.
21
⎟⎟⎟
⎠
⎞
⎜⎜⎜
⎝
⎛=Δ
dt CEAmPερ --- (1a)
In order to predict the maximum DP the meter geometry, meter performance and flow conditions must be known. Although in reality, there are manufacturing geometric tolerances, and variations in pressure and temperature from the predicted conditions can change the precise meter geometry, these are all second order effects, and therefore, we can practically consider the meter geometry known. The required geometry terms are the inlet diameter and the beta ratio (i.e. the cone size). From these inputs the velocity of approach, the throat area and geometric terms for the expansibility can be predicted. At the design stage the meter performance is not known and therefore an estimated constant discharge coefficient is used. The required flow conditions are the flow rate, inlet pressure and the fluids density and isentropic exponent. However, for hydrocarbon production applications the flow conditions are not always precisely known. The reservoir engineers give estimated flow condition ranges which typically have a few percent uncertainty. Therefore, when designing a cone DP meter to give a maximum DP in service that matches the chosen DP transmitters precise URL two vital pieces of information, i.e. the discharge coefficient and the flow condition range, are being estimated only. Whether the actual maximum DP in service is below, on or exceeds the chosen DP transmitters URL is dependent on the discharge coefficient and the flow condition range estimations. If the actual DP produced when the meter is in service exceeds the DP transmitters URL the meter is not capable of metering the full flow condition range, i.e. it will fail to meter the largest flow rates correctly. Figure 19 shows that the average discharge coefficient is approximately 0.8 (±8.5%). The scatter is significantly less for set meter diameters and beta ratios. Therefore manufacturers may be able to estimate more precise discharge coefficients for some meter specifications. However, manufacturers still state each cone DP meter requires calibration in order to meet a ±0.5% flow rate uncertainty. Therefore, it is not very likely that any pre-calibration discharge coefficient prediction will be any closer than ±2% to the calibrated value. Furthermore, the authors know from experience that the reservoir engineer flow condition estimates can have uncertainties of several percent. Cone DP meters with beta ratios set to the fourth decimal place based on these initial discharge coefficient and flow condition estimations are common place. However, in the extremely likely event that these estimations are not precisely correct there is no point choosing a beta ratio to four decimal places. In fact it can give users a false sense of precision and security. Due care should be taken to account for these estimations when choosing a beta ratio as otherwise there is a significant chance that the actual application will saturate the DP transmitter at the applications actual high end flow, therefore making the meter not fit for purpose. An example will highlight this issue: Worked Example: A natural gas production company requires a 6”, schedule 80, DP cone meter. The reservoir engineers estimate the pressure will be 28.5 Bara, the temperature 300K, the
23
molecular weight is 20.15, therefore the density will be 24.2 kg/m3, and the flow rate is 30 MMSCFD. Say the cone DP meter manufacturer estimates a discharge coefficient of 0.80 and as the URL of the transmitter stipulated by the user is say 250”WC (i.e. 62.2 kPa), the manufacturer sets the beta ratio to 0.5844 so as the maximum expected flow conditions produce a DP equal to the URL, i.e. 250.00”WC. However, after the meter is built and calibrated, say it is found that the discharge coefficient is actually 0.788 (i.e. -1.5% difference). After installation, the actual applications pressure is found to be 28.1 Bara (a -1.4% difference), the temperature is 305K (a +1.7% difference), the molecular weight is 20.25 (a +0.5% difference), therefore the actual density is 23.5kg/ m3 (a -2.9% difference), and the actual flow rate is 30.3 MMSCFD (a +1% difference). The actual DP produced by the supplied meter is therefore 274.2”WC (i.e. 68.2 kPa). That is, the actual DP is 24.2”WC (6 kPa) greater than the URL of the stipulated DP transmitter. The DP transmitter is saturated. The DP transmitter therefore reads the URL, and typically there is no system alarm stating there is a problem. If this erroneous DP reading is therefore accepted as true, in this example the DP reading of 250”WC predicts a flow rate of 28.97 MMSCFD where as the real flow is 30.3 MMSCFD. That is, the meter would under-read the flow rate by -4.4% or 1.33 MMSCFD. Note that this example is not extreme. The predicted to actual parameter variations quoted and the resulting flow rate errors are all modest compared to what could happen in a real application. Furthermore, this kind of metering problem can be difficult to identify as subsequent re-calibrations of the meter with appropriate DP ranges shows no problem. Unfortunately, this is not a trivial theoretical discussion. This design method is common for many generic DP meters. In this example, to make this meter serviceable (if the problem is ever noticed) a DP transmitter with a larger URL would be required to replace the existing undersized DP transmitter. However, this problem should never be allowed to occur. It can be avoided at the design stage by taking due account of the uncertainties associated with discharge coefficient and flow condition estimations. Therefore, when sizing a cone DP meter (or any DP meter) it is best to set the desired maximum DP at a value lower than the stipulated DP transmitters URL. The actual maximum DP designed for should be down to engineering judgment per application. For example, in this case let us say the engineer played safe and sized the meter to a maximum DP of 220”WC. At the estimated discharge coefficient (0.80) and estimated flow conditions a beta ratio of 0.6006 gives a maximum 220”WC result. However, note that with the discharge coefficient not known to within ±2% at best before calibration there is no real meaning to the last two beta ratio decimal points. Engineers should round up the beta ratio value to the second decimal place. Here, that is a beta ratio of 0.60, giving an estimated maximum DP of 221.1”WC, well below the URL of 250”WC. Now, if this meter was used in the actual field conditions described above the maximum DP would be 242.5”WC, which is still below the applications set URL limit. That is, the saturating of the stipulated DP transmitter would be avoided from the initial design stage and the meter is now fit for purpose. 6. Standardizing the Cone DP Meter Design and ISO Standards It would be beneficial for industry to have a standard cone DP meter design. Currently, there can be different size cone supports, different orientated upstream pressure tappings
24
relative to the support bar, slightly different cone designs (due to fabrication issues) and different inlet lengths between the inlet of the meter spool and the cone location. Different diameter meters even of the same schedule and beta ratio are not typically geometrically similar by design. It is also common practice to size the beta ratio according to estimated flow conditions up to the fourth decimal place. Therefore, the available public data from cone DP meter calibrations is often not for the same meter design and flow conditions twice. In order for the performance of the cone DP meter to be ready for inclusion in a standard there needs to be massed data sets on the same geometry cone DP meter showing the performance of that design to be reproducible. The phrase “same geometry” includes tight tolerances on the vital dimensions of the meter, and this clearly includes the angle of deviation of the cone to meter centerlines. With the discharge coefficient not being currently predictable to a low uncertainty without calibration the beta ratio needs only to be set to the second decimal place. Furthermore, as there is a possibility that the cone DP meter discharge coefficient is related to the beta ratio (as it is for an orifice plate meter) it may be beneficial for manufacturers to concentrate the beta ratio range to set values. For example, 0.45 to 0.80 beta ratios with 0.05 increments. This, along with geometric similarity, would in the long run allow a better chance of a detailed comparison of the data and give standards organizations a better chance of making a definite statement regards the meters performance. However, for the foreseeable future each cone DP meter must be individually calibrated across the full Reynolds number range of its application. 7. Conclusions Cone DP meters operate according to the generic DP meter principles. Cone DP meters must be calibrated across an applications full Reynolds number range for the calibration to be valid for that application. It should be noted that most hydrocarbon production applications have flow conditions such that ISO 5167 states a Venturi meter must also, for optimum accuracy, be calibrated across the Venturi meters flow range. Therefore, the cone DP meter and Venturi meter are on par with calibration costs for most hydrocarbon production applications. The requirement to calibrate a cone DP meter across an applications full Reynolds number range means it is often inappropriate to calibrate the meter in a water flow facility with a limited Reynolds number range and then extrapolate the results to significantly higher Reynolds numbers. Such practice can lead to gross errors in flow measurement. Analysis of the available data indicates that the current manufacturing tolerance on cone DP meters is not of a level where meters built to the same specification have the same performance. Therefore each individual cone DP meter must be individually calibrated but periodic re-calibration is only required if the operator suspects contamination or damage. However, currently it can be difficult to judge if this is the case while the meter is in service. Therefore, if a meters stated uncertainty rating is of vital importance to the meter operator it can be prudent to periodically re-calibrate the meter. Owners of cone DP meters should treat the cone DP meter as they would treat any precision instrument – i.e. with care. Initial research indicates that a cone deflection from
25
the meter body centerline of less than 1° was enough to shift the discharge coefficient by approximately 0.4%. Once the cone is deflected, the uncertainty rating is no longer valid and the meter should therefore be re-calibrated. It is concluded that the more firmly the cone is held in place the better. Care must be taken when sizing a cone DP meter. It is not appropriate to size the beta ratio to four decimal places when the meter performance and application flow range are only estimations. It is prudent to size the beta ratio such that the predicted maximum DP is lower than the DP transmitters URL. This produces a safety factor for when the actual meter performance and application flow range create larger DP’s than expected. It would be premature to attempt to include the cone DP meter in a standard. However, there is abundant evidence that a properly calibrated cone DP meter has a flow rate uncertainty of ±0.5% across a 10:1 flow rate turndown. Acknowledgments CEESI thanks Citizens Thermal for releasing multiple cone DP meter data sets for analysis and presentation. CEESI also thanks Cameron and DP Diagnostics for supplying various data sets. Thanks is also given to all companies that had a small numbers of cone DP meter data sets included as part of the massed blinded data set. References 1. International Standards Organisation, “Measurement of Fluid Flow by Means of Pressure Differential Devices Inserted in Circular Cross-Section Conduits Running Full,” ISO 5167, Parts 1-4. 2. 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. 3. Peters R., Steven R., George D., Bowles E., Nored M., “Tests of the V-Cone Flow Meter at Southwest Research Institute and Utah State University in Accordance with the New API Chapter 5.7 Test Protocol”, North Sea Flow Measurement Workshop 2004, St Andrews, Scotland, UK, Paper number 2.1. 4. Geach D. and Jamieson A. “Wet Gas Measurement in the Southern North Sea”, North Sea Flow Measurement Workshop, St Andrews, UK, Oct 2005.
Recent field experiences using multiphase meters for fiscal allocation Eirik Åbro, StatoilHydro ASA Kåre Kleppe, StatoilHydro ASA Leif Jarle Vikshåland, StatoilHydro ASA
Introduction
StatoilHydro has about 150 multiphase and wet gas meters in operations, with additional 70-80 multiphase and wet gas meters to be operative in near future. Most of the meters in operation today are used for production optimisation and production management. During the last years, several oil/wet gas fields operated by StatoilHydro have been developed by use of multiphase and wet gas meters for fiscal allocation purposes. The fields are developed as subsea productions system (SPS) where unprocessed multiphase flows are transported to process platforms through pipelines. In this paper the allocation metering installed on Kristin and Åsgard B in connection with the tie-in of the Tyrihans Field is presented. The ownership allocation between the Halten West Unit (Kristin) and Tyrihans Unit is based on multiphase metering of the Tyrihans flow line production, i.e two parallel topside multiphase meters installed onboard Kristin platform. Subsea multiphase meters are installed for the 11 subsea producing wells, 8 oil producers initially in 2009, one gas producer in 2015 and converting of two gas injectors to gas producers in 2023. The Tyrihans field started to produce in July 2009 and data from the multiphase meters and the test separator meters have been compared frequently in order to verify the topside multiphase meters. The contributions of the overall fiscal metering system uncertainties of the allocation metering for the Tyrihans production are identified, such as test separator metering, PVT compositions, gas lift measurements and multiphase meters. Field experiences with results from the multiphase meters used for fiscal allocation of the Tyrihans productions from the first months in operation are presented.
Tyrihans field development
The Tyrihans field is located in PL 073 (block 6407/1) and PL091 (block 6406/3) in the Halten Nordland area. Figure 1 illustrates a prospect map including the location of Tyrihans.
Figure 1. Tyrihans location map at Haltenbanken
The development concept of Tyrihans comprises five separate four slots templates. One production template located on Tyrihans Nord, three on Tyrihans Sør and one injection template in the saddle point between the two accumulations. The hydrocarbon production from Tyrihans is tied in to the Kristin platform. The Tyrihans Sør well stream produces through a 14 km 16” flow line to Tyrihans Nord and commingled production of Tyrihans Sør and Nord through a 29 km 18” flow line to Kristin.
Reservoir pressure support is achieved by utilising both gas- and water injection. Gas injection is provided from Åsgard B. Export gas from Åsgard B is injected into Tyrihans Sør through a 10” flow line. The water injector is a stand-alone subsea raw seawater injector on a separate template placed between Tyrihans Sør and Nord.
Table 1. Ownership interests of Tyrihans field
Tyrihans Unit (PL073/PL091)
StatoilHydro ASA 58.8351 % ExxonMobil 11.7549 % Total 23.180340% ENI 6.22966 %
*) Pre unit split is used in PL073B (Based on a PL073 (80%) and PL091 (20%) split).
Figure 2. Tyrihans development sketch
Measuring concept and testing philosophy
The topside multiphase meters are used for fiscal allocation purposes. With the test separator onboard Kristin, the topside multiphase meters are verified periodically against the test separator. The first year of production, however, the availability of the test separator is high. When shut-in of new wells, the well production shall be routed to the test separator.
Figure 3. Tyrihans metering concept
Each of the subsea and topside meter have a measurement uncertainty specification. Two different multiphase meter technologies have been chosen for subsea and topside. The topside multiphase meters, however, are equal and both the meters can be routed to the test separator, as indicated in Figure 3.
The subsea multiphase meters, which are used for production optimisation, are also back-up for the topside multiphase meters. While both topside meters are in normal operation, and verified against test separator, the subsea meters are continuously compared to the top side meters. The total hydrocarbon mass rate measured by the topside meters is compared to the sum of the hydrocarbon mass rate measured by the individual subsea multiphase meters and the mass rate from the gas lift. Gas lift in the riser base is accounted for. With continuous gas lift in the wells, updated PVT input to the subsea multiphase meter is required and the amount lift gas must be accounted for. However, for gas lift in the wells during start-ups, it is not possible/practical to update the PVT data in the subsea multiphase meters.
By comparing continuous HC mass and water rates, measured by topside meters and subsea meters, included gas lift in riser base, any deviation between the topside measurements and subsea measurements is revealed. More extensively testing by using the test separator is required in order to map the source of deviation.
Kristin
Test separator
Inlet separator
MPM
Tyrihans Nord
A-2 H
A-3 H A-4 H
A-1 H
Spare slotGas injectorOil producer
16” production flowline
18” production flowline
14 km
29 km
Future gas producer
Riser
16”
12”
MPMChoke
ChokeMPM
12”
Tyrihans metering concept
Revision 14.09.05
B-2 H
B-3 H B-4 H
B-1 H
MultiPhase Meter
MPM
MPM
Tyrihans Sør
C-2 H
C-3 H C-4 H
C-1 H
MPM
D-2 H
D-3 H D-4 H
D-1 H
Tyrihans Sør
10” gas injection lineÅsgard B
DP
DP Cell
MPMMPM
MPM
DP MPM
Tyrihans Sør
In addition there will be installed DP cells over the gas lift chokes on the oil producers and over the choke on the SRSWI for metering purposes.
Kristin
Test separator
Inlet separator
MPM
Tyrihans Nord
A-2 H
A-3 H A-4 H
A-1 H
Spare slotGas injectorOil producer
16” production flowline
18” production flowline
14 km
29 km
Future gas producer
Riser
16”
12”
MPMChoke
ChokeMPM
12”
Tyrihans metering concept
Revision 14.09.05
B-2 H
B-3 H B-4 H
B-1 H
MultiPhase Meter
MPM
MPM
Tyrihans Sør
C-2 H
C-3 H C-4 H
C-1 H
MPM
D-2 H
D-3 H D-4 H
D-1 H
Tyrihans Sør
10” gas injection lineÅsgard B
DP
DP Cell
MPMMPM
MPM
DP MPM
Tyrihans Sør
In addition there will be installed DP cells over the gas lift chokes on the oil producers and over the choke on the SRSWI for metering purposes.
Gas Lift Riser BaseKristin
With known fluid compositions of Tyrihans Nord and Tyrihans Sør, model-based PVT-input to the topside multiphase meters at Kristin is implemented. Export gas from Åsgård B used as gas lift and gas injection, and export gas from Kristin is used for gas lift in the Tyrihans riser will influence the overall PVT data. Gas lifts in the wells and riser base are planned primarily for start up purposes. Gas lift in the wells may also be used for flow stabilisation when or where this is considered necessary from an operational point of view. The system is designed for continuous use of lift gas, and as a consequence, updated PVT input to the subsea multiphase meters and topside meters are required. Both Åsgard B and Kristin gas lift and injection for Tyrihans are measured and included in the overall measurement philosophy.
Tyrihans 14” riser is connected to a 16” flow line manifold. From the 16” manifold the total Tyrihans stream is split into 2 lines upstream the chokes. On each line there is installed a multiphase meter that is connected to a valve arrangement so that each flow line can be directed either to Kristin test manifold or Kristin production manifold. Each multiphase meter can be directed to Kristin test separator in order to be individually calibrated.
Figure 4. Simplified sketch for Tyrihans inlet arrangement at Kristin
Measurement uncertainties are given for each multiphase meter by the vendors. The specified measurement uncertainties given by the vendors are not based on the specific field application. Additional uncertainties due to PVT related uncertainty and flow dynamics, such as slugging must be accounted for.
Measurement Uncertainty
Oil metering
6” 5-path ultrasonic
2” 2-path ultrasonic
+/- 0.32% mass1
+/- 1.45% vol2
Gas metering
10” V-cone
4” V-cone
+/- 0.7% mass or vol
Water metering
3” 2-path ultrasonic
1” 2-path ultrasonic
+/- 1.2% mass or std vol3
+/- 7.2% mass4
Table 2. Metering of test separator at Kristin.
Table 2 shows the Kristin test separator metering. For Tyrihans the situation is high flow and high temperature. High temperature will give a slightly higher contribution to the total uncertainty than low temperature. The uncertainty for the test separator oil metering with high flow and high temperature is therefore assumed to be within +/- 0.4 %.
Measurement for gas lift and gas injections are defined by the NPD requirement and in accordance to NORSOK I-104.
Sensitivity of PVT Data input to the topside multiphase meters
For the topside multiphase meters, changes in the PVT data (composition) will change due to the relative production from the Tyrihans Sør and Tyrihans Nord reservoirs. In addition gas lift from Kristin and Åsgård B to riser base and the wells, respectively, contribute to the changes in the fluid composition.
It has been performed a study to map the sensitivity of changes in the PVT data in the topside multiphase meters. The sensitivity study is based on that the PVT data input to the multiphase meters is a mixture of Tyrihans Nord and Tyrihans Sør compositions. The assumption is that the PVT mixture is 20% Tyrihans Nord and 80% Tyrihans Sør during plateau production.
Maximum deviations due to PVT setup will then be when production is only from one reservoir, either Tyrihans Sør or Tyrihans Nord. When the PVT input to the multiphase meter is the 20/80 mixture of Tyrihans Nord and Tyrihans Sør: 1 high flow and low temperature. 2 low flow and high temperature. 3 high flow and high temperature. 4 low flow and low temperature.
The relative deviations of oil volume rates are 10% and 11% when producing only from the Tyrihans Nord and Tyrihans Sør reservoirs, respectively.
The relative deviations of gas volume rates are -0.5% when producing only from the Tyrihans Nord or Tyrihans Sør reservoirs.
The relative deviations of water volume rates are -10 % and 11% when producing only from the Tyrihans Nord and Tyrihans Sør reservoirs, respectively.
The relative deviations of HC mass are 1.2% and -0.8% when producing only from the Tyrihans Nord and Tyrihans Sør reservoirs, respectively.
Considerable errors in the split between the liquid volume phases are calculated by changes in the PVT in the topside multiphase meters. Therefore, online PVT calculations, based on the relative productions from the Tyrihans Nord and Tyrihans Sør reservoirs, are required. The calculated updated PVT data can be downloaded to the multiphase meters.
Experiences from the start-up of the Tyrihans field
As shown in Figure 5, the calculated accuracies of the topside multiphase meter for hydrocarbon mass rate measurement are about 4-6%, depending on the production profile. Similarly, the calculated accuracies of the measured oil volume rate are about 4-8%.
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
8,0
2009 2011 2013 2015 2017 2019Year
Rea
ltive
Unc
erta
inty
[%]
HC Mass RateOil Volume Rate
Figure 5. Calculated measurement uncertainty of HC mass rate and oil volume rate for one topside multiphase meter.
To minimize the uncertainties in the allocations, the Tyrihans production allocation is based on HC mass rate, which is independent on different pressures and temperatures subsea, topside multiphase meters and test separator.
The start-up of the first wells was in July 2009. During August 2009 extensive testing was performed on the production and for verifications of the two topside multiphase meters (MPFM1 and MPFM2). The tests have been performed using two different combinations of wells.
For each of the tests, one (f.ex MPFM1) of the two topside multiphase meters was isolated, thus all production was routed through the other multiphase meter (MPFM2) and further to the test separator.
Then the sequence was repeated by isolate the other multiphase meter (MPFM2), thus verification of the multiphase meter (MPFM1) was performed. Figure 6 shows a trend plot of oil, gas and water at actual and standard conditions measured by topside MPFM1.
Figure 6. Trend plot of MPFM1 during a test period. Both actual and standard conditions are presented.
When comparing the measurements of multiphase meters and the test separator, the combined uncertainty of the multiphase meter and the test separator must be accounted for. Here, the uncertainty of the test separator is insignificant compared to the overall calculated measurement uncertainties of the multiphase meters, thus it is assumed that any deviation between test separator and multiphase meters can be derived by the multiphase meter measurement uncertainties.
To minimize systematic errors, k-factors are used for standard mass rate for each phase. During, the first months of production it is important to establish k-factors based on several tests, covering several months. For the tests presented here, a new set of k-factors is used for each test date and for each multiphase meter. The k-factors are based on previous calibrations of the multiphase meters. For the k-factors for MPFM1, the k-factors are 0.952 - 0.970 and 1.114 – 1.148 for oil mass and gas mass, respectively. Similarly, for MPFM2 the k-factors are 1.001 – 1.056 and 0.987 – 1.044 for oil mass and gas mass, respectively. The k-factors for water are 1.000 for both meters.
The comparisons for each multiphase meter are done for oil and gas volume flow rates at standard conditions and hydrocarbon mass flow rates. Relative deviations are presented for standard oil volume rates, standard gas volume rates and hydrocarbon mass rates. Due to the low water production, water cut is used to compare the water measurements.
The test periods last over several hours, and each test point represents one hour test under stable condition. By dividing each test into several test points, the repeatability of the tests is also demonstrated. Each test point therefore represents an averaged value over one hour.
MPFM1
200
210
220
230
240
250
260
270
280
290
300
200 220 240 260 280 300HC mass [ton/h] test separator
HC m
ass
[ton/
h] M
ultip
hase
met
er
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
MPFM2
200
210
220
230
240
250
260
270
280
290
300
200 220 240 260 280 300HC mass [ton/h] test separator
HC
mas
s [to
n/h]
Mul
tipha
se m
eter
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
Figure 7. Comparison of hydrocarbon mass rate and with +/- 5% dashed lines.
Figure 7 shows the hydrocarbon mass flow rates measured by both multiphase meters compared to the test separator measurements. Except for the tests done 2. Aug, the repeatability of the hydrocarbon mass rate is within +/-1%, as shown in Figure 8.
-3,0
-2,5
-2,0
-1,5
-1,0
-0,5
0,0
0,5
1,0
1,5
2,0
18.07.2009 25.07.2009 01.08.2009 08.08.2009 15.08.2009 22.08.2009 29.08.2009 05.09.2009Date
Rel
dev
iatio
n H
C m
ass
[%]
MPFM1MPFM2
Figure 8. Relative deviation of hydrocarbon mass rate versus time for both topside meters. Figure 9 and Figure 10 show the comparison of the measured oil volume flow rates at standard conditions and the relative deviations of the oil volume rates. The plots show that the deviations are within measurement uncertainty specification. However, a negative offset of about 1% of the oil volume rates is observed, thus an updated k-factor may be needed. The repeatability of the oil volume rates appears to be similar to the hydrocarbon mass rates, i.e. +/-1%.
MPFM1
150
160
170
180
190
200
210
220
230
240
250
260
270
280
290
300
150 170 190 210 230 250 270 290Oil volume flow rate [Sm3/h] test separator
Oil
volu
me
flow
rate
[Sm
3/h]
Mul
tipha
se m
eter
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
MPFM2
150
160
170
180
190
200
210
220
230
240
250
150 170 190 210 230 250Oil volume flow rate [Sm3/h] test separator
Oil
volu
me
flow
rate
[Sm
3/h]
Mul
tipha
se m
eter
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
Figure 9. Comparison of oil volume flow rate at standard conditions and with +/- 5% dashed lines.
-4,0
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
18.07.2009 25.07.2009 01.08.2009 08.08.2009 15.08.2009 22.08.2009 29.08.2009 05.09.2009
Date
Rel d
evia
tion
oil s
td. v
ol [
%]
MPFM1MPFM2
Figure 10. Relative deviation of oil volume flow rate at standard condition versus time for both topside meters. In Figure 11 and Figure 12 show the comparison of the measured gas volume flow rates at standard conditions and the relative deviations of the gas volume rates. The plots show that the deviations are within +/-3%. However, the repeatability of the gas volume rates appears to be about 4%, and poorer than hydro carbon mass rate, as expected.
MPFM1
60000
65000
70000
75000
80000
85000
90000
95000
60000 65000 70000 75000 80000 85000 90000 95000Gas volume flow rate [Sm3/h] test separator
Gas
vol
ume
flow
rate
[Sm
3/h]
Mul
tipha
se m
eter
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
MPFM2
60000
65000
70000
75000
80000
85000
90000
60000 65000 70000 75000 80000 85000 90000Gas volume flow rate [Sm3/h] test separator
Gas
vol
ume
flow
rate
[Sm
3/h]
Mul
tipha
se m
eter
Well A4 C2 B1Well A4 B1 B3Ideal+/-5%
Figure 11. Comparison of gas volume flow rate at standard conditions and with +/- 5% dashed lines.
-4,0
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
18.07.2009 25.07.2009 01.08.2009 08.08.2009 15.08.2009 22.08.2009 29.08.2009 05.09.2009
Date
Rel d
evia
tion
gas
std.
vol
[%
]
MPFM1MPFM2
Figure 12. Relative deviation of gas volume flow rate at standard condition versus time for both topside meters. A 24 hour test was carried out 2. Aug. As seen in Figure 6, it is observed a significant spread of the gas volume rate measurements. If the gas volume rates are averaged over the whole 24 hrs period, the relative deviation of the gas volume rate is reduced to -1.1%.
-1,0
-0,5
0,0
0,5
1,0
1,5
18.07.2009 25.07.2009 01.08.2009 08.08.2009 15.08.2009 22.08.2009 29.08.2009 05.09.2009
Date
Abs
devi
atio
n W
C [%
]
MPFM1MPFM2
Figure 13. Absolute deviation of water cut versus time for both topside meters.
Conclusions
By evaluating the deviations between the multiphase meters and the test separator, the k-factors during this period have to be taken into account for a complete mapping of the uncertainties. The ranges of the k-factors in addition to the observed deviation, show that by using multiphase meters topside at the Kristin platform for Tyrihans allocation purposes are within our expectations. The repeatabilities of hydrocarbon mass rates are within +/-1% for both topside meters for about 90% of the test points. The comparisons of the single phases show deviations for oil and gas volume rates at standard conditions higher than +/-1%. For oil volume rates the deviations are in the range -3% -0%, indicating an offset of about -1%. For gas volume rates the deviations are about +/-3%. A test period of 1 hour appears to be too low, and reduced deviation of the gas volume rate can be achieved by increasing the testing time. The results show that the tests of the topside multiphase meters for Tyrihans are within the measurement specifications. The multiphase meter’s sensitivity of the PVT data is high, which can result in significant measurement errors. Carefully follow-ups are needed to maintain these results with high focus on PVT data needed as input to the meters.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
Fluid characterisation in a subsea on-line multiphase fluid sampling and analysis system
E.M. Bruvik2, M.B. Holstad*,1,3, J. Spilde1,3, B.T. Hjertaker2,3, S.H. Stavland1, K-E. Frøysa1,3, S.K. Meyer2 1 Christian Michelsen Research AS, Fantoftvegen 38, P.O. Box 6031, NO-5892 Bergen, Norway
2 University of Bergen, Department of Physics and Technology, Allégaten 55, NO-5007 Bergen, Norway
3 The Michelsen Centre for Industrial Measurement Science and Technology
* Corresponding author: [email protected], tel.: +47 55 57 42 93
ABSTRACT. The trend in subsea petroleum production systems in offshore field developments points towards integrated production and processing facilities at the seabed along with more extensive use of multiphase transportation technology. The SOFA (Subsea On-line multiphase Fluid sampling and Analysis) system designed by Christian Michelsen Research in cooperation with the University of Bergen is an autonomous metering station for permanent installation subsea. The SOFA carries out fluid analysis subsea, and the current method of transfer of fluid samples to surface by a remote operated vehicle is avoided.
A laboratory prototype for capturing multiphase fluid samples in a dedicated measurement chamber/sample container has been built and tested. This is equipped with ultrasound sensors, a dual modality densitometer (DMD) gamma-ray system, pressure and temperature sensors, which together with conductivity, permittivity or similar measurements will be used as a multi-modality system for fluid characterisation.
This paper presents experimental results for fluid sampling and characterisation based on multimodality. The focus is on detection and monitoring of water salinity, water density and oil density. In addition, perspectives for operational use are discussed.
1. INTRODUCTION The use of subsea production systems in offshore oil and gas field developments has increased dramatically during recent years, both in Norway and abroad. The trend is towards integrated production and processing facilities at the seabed along with more extensive use of multiphase transportation technology. This causes an increasing demand for accurate multiphase flow measurements and measurement of fluid properties. Fluid properties are
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
needed for more accurate multiphase flow meter (MPFM) readings, process optimisation, well management and production allocation. Fluid characterisation should be carried out near the subsea wellhead or the subsea manifold in order to detect changes in composition or fluid properties as early as possible. Christian Michelsen Research (CMR) started the development of a new concept called “Subsea Online Fluid sampler and Analyser” (SOFA) in 2003 (Baker et al. 2007, Spilde et al. 2008), with the aim of enabling subsea multiphase fluid sampling, single-phase fluid characterisation, and possibly also measurement of fluid fractions and multiphase flow rates.
1.1. The SOFA concept The principle of the SOFA concept is to build an autonomous metering station for permanent installation subsea so that transfer of fluid samples to the surface by remote operated vehicle (ROV) is avoided. The system could be installed as an integrated part of a subsea tree or manifold, and is expected to become a particularly valuable and cost efficient tool for optimisation of subsea separation processes and well management. Figure 1 shows a schematic which compares the SOFA concept to a traditional multiphase flow meter.
Figure 1. Schematic of a traditional multiphase flow meter (left) the SOFA concept (right). Here q is the volumetric flow rate of the components, Q is the mass flow rate, WLR is the volumetric water liquid ratio, and GVF is the volumetric gas volume fraction.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
The SOFA concept comprises a number of separate modules and technologies, and CMR and the University of Bergen (UoB) have built a first experimental prototype of the SOFA system, see figure 2. The concept poses two major challenges; the representative sampling of the multiphase flow into the measurement chamber, and secondly an accurate measurement of the fluid properties. A prototype design for capturing multiphase fluid samples in a dedicated measurement chamber/sample container has been built and tested. The sampling system will be briefly discussed in section 2.1. The focus for the remainder of this article is the measurement methods applied after the multiphase sample has been taken. Measurements on single separated phases allow for more accurate results and additional measurement results depending on the number of parameters influencing the various measurement technologies.
water
oil
gas
inlet
outlet
bypassflow str.
T−bend
ven− turi
Figure 2. An artist’s view of a SOFA installation at the sea bed (A), and (B); a schematic of the laboratory prototype for flow rig testing. See figure 3 for detailed image and schematics of the fluid analysis chamber.
1.2. Fluid analysis technologies The fluid sample is taken by means of a pitot tube inserted into the main pipe flow as shown in figure 2B. During sampling both the inlet and outlet pitot tubes are open so that the chamber is flushed with the process flow, and after approximately one minute the flow through the chamber is redirected outside the chamber through the bypass line. The fluid sample will then be left to separate into single phases, and single phase fluid analysis will be attempted using technologies based on gamma-rays (transmission and scattering) and ultrasound (level and speed of sound). The configuration of the measurement chamber is
A) B)
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
shown in figure 3. In addition electromagnetic characterisation will be used, but is not currently implemented in the chamber design. Also, the chamber is equipped with temperature and pressure transducers. The volume of the chamber is approximately 5 litres (~10x10x50cm3 oriented at 45 degrees).
B)
OIL
WATER
GAS
D2
D3
D4
D1
v3S
C)
OIL
WATER
GAS
Figure 3. A) Prototype fluid analysis system with ultrasound and gamma-ray instrumentation. In addition the chamber is also equipped with pressure and temperature sensors. The gamma source (labelled 2 in the image) irradiates the radiation detectors (all detectors labelled 3 in the image) B) Gamma-ray measurement setup (cross-section of fluid analysis chamber in the plane of gamma-source and detectors). The source irradiates detectors 1-3 directly for transmission measurements through the respective phases, while detector 4 only measures scattered radiation from the water-phase. C) There are four pairs of ultrasound transducers (labelled 1 in the image) operating both as transmitters and receivers; four transducers on top of the chamber and four at the bottom.
1.2.1. The acoustic measurement system comprises four pairs of ultrasonic (US) transducers operating in pulse-echo and transmission modes (Stavland 2005). Since a level measurement on a single interface usually is possible with two transducer pairs, both level and speed of sound in the two phases can be found. The transducers are in direct contact with the fluids and will have to deal with the high pressures and temperatures (HPHT) of the process. The chamber is currently instrumented with CMR’s HPHT NISEP-transducers (5 MHz, max. 150 oC and 700 bar).
1.2.2. The Nucleonic measurement system will allow for transmission measurements of gamma-rays at 60 keV emitted from a Am-241 radio-isotope source (Berntsen 2005). The source is inserted slightly into the chamber in a titanium housing in order to irradiate all the transmission detectors. For detection of the gamma-rays, semiconductor CdZnTe-detectors are used. In addition to transmission measurements, detector 4 is shielded from directly
A)
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
transmitted photons and only measures photons scattered from the water phase at the bottom of the chamber. This provides for Dual Modality Densitometry (DMD) as described by Holstad and Johansen (2005). All gamma-ray measurements presented in this paper are based on a counting time of 2 minutes.
1.2.3. Electromagnetic measurement system. In addition to acoustic and nucleonic measurement technologies, electromagnetic (EM) measurement technologies are considered used in the SOFA. Several electromagnetic technologies are being considered and some initial tests show that some of the technologies have potential for use in the SOFA chamber.
2. EXPERIMENTAL RESULTS AND DISCUSSION
2.1. Sampling and water cut measurements Although fluid characterisation is and has been the main focus for the SOFA project, representative sampling of the flow would give valuable additional information. This information is, however, available also from commercially available multiphase flow meters. If the sampling in the SOFA system is representative, the volume and mass flow rates can be found by the use of a venturi meter as shown in Figure 1. CFD modelling was used in the first design of the sampling probe, and such simulations will also be used in the further developments of the system. Experiments have confirmed that the current sampling geometry does not give a volumetric representative sample for the gas fraction, but it ensures collection of sufficient amounts of all phases. And notably; the water cut of the sample is representative (see figure 4) in all flow regimes, although the gas volume fraction (GVF) is not (Bruvik and Hjertaker 2009). Further development of the measurement geometry will be needed in order to get accurate flow rates, and the initial experiments have indicated how the sampling geometry can be further refined and optimised.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
0 17 33 50 67 83 1000
10
20
30
40
50
60
70
80
90
100
WLRref
(%)
WLR
(%
)
experimentalWLR = WLR
ref
0 17 33 50 67 83 100
−4
−3
−2
−1
0
1
2
3
4
WLRref
(%)
Err
or W
LR (
% a
bs)
experimental
Figure 4. Water cut measurements as measured with the US instrumentation in the SOFA chamber. For water cuts larger than a few % the error is within the +/-2 % absolute accuracy of the Coriolis reference meter in the flow rig. Repeated water cut measurements are done in different flow regimes in the vertical process tube while sampling. The challenges regarding representative sampling may be solved in several ways. For example, density measurements of the main flow can be used for GVF measurements so that only the water cut is needed from the SOFA. Also, if the SOFA is used as an analysis unit only, volume fractions are not needed from the SOFA. This will be the case if the unit is utilised to give process parameter values needed as input to other instruments, such as multiphase meters relying on accurate salinity, density and permittivity values. The main objective, characterisation of the fluids, is therefore possible as long as a significant amount of each phase is present in the sample.
2.2. Oil density The oil density can be measured by gamma-ray transmission measurements using detector 2 as shown in figure 3B. The levels of the interfaces will be known from the US level measurement, and the attenuation in water will be found from transmission measurement to detector 3. The only unknown parameter then, in the path from the source to the detector 2, is the density of the oil (assuming the density of the gas is small compared to that of the liquids).
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
0 5 10 15 20 25
−2
−1
0
1
2
thickness of oil layer available for density measurement (cm)
Oil
dens
ity e
rror
(%
)
Figure 5. Relative error in oil density measurement by gamma transmission (D2 in figure 3B) measurement for all relevant measurement points in figure 4. The oil density (diesel oil density, ρ = 840 kg/m3) was found accurate to 1 % by gamma-ray transmission measurements (figure 5) if the thickness of the oil layer between source and transmission detector was larger than 10 cm (WLR < 60 %). Since the speed of sound (u) in the oil is found by the US measurements, the adiabatic compressibility (κS) can also be found; κS=1/(ρ•u2).
2.3. Water salinity and density by gamma-ray-DMD The density of the water depends on the salinity and possible process additives such as Mono-Etylene-Glycol (MEG) (to prevent hydrate formation in multiphase transport systems). Water salinity might be an indicator of well production conditions, and salinity and density are important inputs to MPFMs. Demonstrated here are results for Chloride (Cl) salts of Sodium (Na), Calsium (Ca) and Strontium (Sr). The results plotted below are from gamma-ray transmission and scatter measurements (i.e. DMD) in the fluid analysis chamber as shown in figure 3.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
0 2 4 6 8 100.6
0.7
0.8
0.9
1
Tran
sm. /
H2O
Tra
nsm
.
0 2 4 6 8 100.6
0.7
0.8
0.9
1
1.1
Salinity (wt%)
Sca
tter /
H2O
Sca
tter
NaClx
MEG=0.1 NaCl
xMEG
=0.4 NaCl
CaCl2
SrCl2
0.6 0.7 0.8 0.9 1 1.10.5
0.6
0.7
0.8
0.9
1
1.1
Tranmission / H2O Transmission
Sca
tter
/ H2O
Sca
tter
NaCl: S=0.0, 2.5, 5.0, 10.2 %x
MEG=0.1 NaCl: S=0.0, 2.5, 5.0, 10.0 %
xMEG
=0.4 NaCl: S=0.0, 2.5, 5.0, 9.9 %
CaCl2: S=0.0, 1.0, 2.0, 4.0, 8.0 %
SrCl2: S=0.0, 0.5, 1.0, 2.0 %
IncreasingSalinity
Figure 6. Gamma-ray transmission and scatter measurements (relative to pure water). Top figures shows transmission and scattered radiation as a function of salinity, and in the lower figure they have been combined to provide a ‘map’ of the DMD-space.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
As figure 6 shows; the salinity and MEG content (given as the molar fraction of MEG to MEG and H2O) can be found in the DMD-plane for single salts and MEG/NaCl-brines. However, an arbitrary mixture of salts and/or MEG cannot be uniquely identified and additional measurements, such as permittivity measurements must be added in order to quantify more variables. See section 3 for further details. If the salinity and salt composition is found then the density of the produced water can be found from the equations and data in Krumgalz et al. (2000) (but this is deemed outside of the scope of this paper). See also Holstad and Johansen (2005) for salinity and density measurements found through utilization of the DMD principle.
2.4. Speed of sound in the aqueous phase As mentioned in section 1.2.1 the speed of sound (SOS) in the phases can usually be found in addition to the level measurement by US. In figure 7, measured values for the SOS are shown1.
0 5 10 15
1500
1520
1540
1560
1580
1600
1620
1640
1660
1680
1700
1720
spee
d of
sou
nd (
m/s
)
salinity (wt%)
NaClCaCl
2
SrCl2
xMEG
=0.1 NaCl
xMEG
=1.0 NaCl
Figure 7. Speed of sound as a function of salinity and MEG content (mol fraction). Different types of salt impact the SOS differently, and notably; MEG increases the SOS significantly. Still, nothing can be said conclusively from a SOS measurement alone – a high 1 Measurements with MEG are not conducted in the SOFA chamber, but with a dedicated ultrasonic velocity meter.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
SOS of about 1620 m/s could be due to 10 % mol MEG or 13 % salinity of NaCl. These alternatives could be distinguished by looking at the DMD-measurement result represented in figure 6.
2.5. Permittivity of water (and oil). The (relative) permittivity (ε) of liquids can be used as an indicator of liquid composition and quality.
Initial measurement results show that the uncertainty of the permittivity measurements of oil achieved with some electromagnetic technologies is higher than what is desired since crude oil permittivity is usually about 2.3 with small variations not detectable due to excessively short propagation times. However, water/MEG-mixtures can be analysed with reasonable accuracy since the permittivity is higher with larger variations (εH2O = 80, εMEG=44). Experiments have shown that a change in the MEG concentration in the water phase causes a detectable reduction in propagation time and permittivity. More EM technologies (Folgerø 1996) will be evaluated for electromagnetic characterisation of the fluids to achieve more accurate results, in particular of the oil phase. The permittivity of the oil is one of the parameter values utilized in traditional multiphase meters, and accurate permittivity measurements will help reduce the uncertainty in such meters.
2.6. Separation properties Separation of oil and water depend on a multitude of parameters such as the viscosity, surface tension, and amount of surfactants in the liquids. Since the sample taken into the chamber will separate, the time this takes can be used to give an estimate of the ‘separability’ of the oil and water mixture. Figure 8 shows how both the gamma-ray system and the acoustic system can be used for monitoring of the separation process.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
time
coun
ts /
s
D1D2D3D4
Magnitude of echo [dB]
Time of arrival of echo
EMULSION
OIL−AIR INTERFACE
OIL−WATER INTERFACE
Time
Figure 8. Oil/water separation time measured with gamma-ray instrumentation (A) and ultrasonic pulse-echo measurements (B). The ultrasonic measurements are not done in the SOFA chamber, but a separate test chamber with a different transducer and oil-water mixture for evaluation. Both methods can perform measurements during the separation of oil and water in a chamber and thereby give an estimate of a qualitative separation time; about 5 minutes for the diesel-water mixture in A) and 10 minutes for the vegetable oil-water mixture in B).
3. MULTIMODALITY The technologies presented above work together to analyse the fluids - e.g. the ultrasonic level measurements are primarily used to get a fraction measurement, but the fluid levels found are also used as input to the gamma transmission measurement to find the density of oil, and to the measurement of permittivity. Further, each of the measurement technologies (gamma-DMD, US and EM) presented are sensitive to different parameters of interest. However, individually none of them produce conclusive measurements of any basic key parameters such as produced water composition and density. For instance, a high speed of sound in the aqueous phase might be due to ‘some’ MEG or a high salinity, and, a low level of gamma ray transmission might be due to ‘some’ MEG, ‘some’ NaCl, or a ‘little’ SrCl2. The electromagnetic measurements might be considered to be sensitive to the permittivity only and hence the content MEG alone. However, in practice the permittivity measurements will be influenced by the conductivity and hence the salinity and composition of the water. Consequently all measurements must be combined in order to characterise the produced water.
Separation Separation
A) B)
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
Figure 9 and the text below gives one example of the need for multimodality when characterising fluids.
Figure 9. Example of multimodality. A solution similar to a realistic brine with MEG, is measured with DMD (at coordinates 0.57, 0.72 - and plotted over the ‘DMD-grid’ of figure 6). The DMD-measurement is not conclusive since an alternative (and not highly improbable) alternative solution also exists. The speed of sound measurement also happens to be equal (within the measurement uncertainty) for the two brines, and the correct brine can only be identified by including permittivity measurements.
4. PERSPECTIVES ON OPERATIONAL USE CMR is currently evaluating the possibilities for continuation of the SOFA project. Depending on partners and feedback from the industry, water characterisation, liquid characterisation (oil and water) or three phase characterisation will be focused. Future applications include both permanently installed and mobile, e,g, suitable for ROV operations, analysis chambers. Status evaluations and initial feedback from the industry have indicated various possible applications for the SOFA. As previously mentioned, the concept was first developed to give online measurements of parameters needed as input to flow meters. This had been identified as a technology gap due to the current need for time consuming and costly ROV-operations to collect samples for laboratory analysis. The SOFA can be installed in connection with multiphase flow meters and give continuously updated information for more accurate flow measurements. Another possible application is a diagnostic tool used together with ROVs. In some cases where permanently installed instrumentation does not give sufficient information about the process, ROVs may be used to collect samples of the fluid. A SOFA measurement chamber
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
mounted onto the ROV will then be able to give prompt information about the fluid properties, and the sample can be directly re-injected into the flow.
5. SUMMARY AND CONCLUSIONS The SOFA setup and fluid analysis scenario may in the future provide online information on the parameter values that are needed as input to traditional multiphase flow meters in addition to information for well- and reservoir management, as well as production optimisation. This strongly reduces, or possibly eliminates, the need for samples collected by ROVs. Combined with gas fraction measurements, and a Venturi flow meter, the SOFA setup can also in some applications replace multi-phase meters since, as demonstrated here, it may give accurate information on water cut in addition to water and oil density.
ACKNOWLEDGEMENTS The authors wish to thank all personnel at CMR and UoB who have been involved in the development work, in particular Anders Hallanger at CMR for executing CFD simulations. The work was funded by the Petromaks programme of the Research Council of Norway and StatoilHydro Petroleum AS.
REFERENCES Baker A., Bjelland C., Spilde J., Frøysa K.-E., Holstad M.B., Johansen G.A., Hjertaker B.T., Berntsen B. and Stavland S.H.,(2007) A novel subsea on-line multiphase fluid sampling and analysis system, Proc. 5th WCIPT (World Congress on Industrial Process Tomography), 2007, Bergen Berntsen B.S.(2005) Development of a measurement chamber for characterisation of brine using absorption and scattering of gamma radiation, MSc. thesis (in Norwegian), Department of Physics and Technology, University of Bergen. Bruvik E.M. and Hjertaker B.T.,(2009) Characterisation of on-line fluid sampling using gamma-ray tomography, IWPT3 (International Workshop on Process Tomography), April 2009, Tokyo Folgerø, K. (1996) Coaxial Sensors for Broad-Band Complex Permittivity Measurements of Petroleum Fluids. Ph. D. thesis, University of Bergen, Department of Physics. Holstad M.B., Johansen G.A. (2005) Produced water characterisation by dual modality gamma-ray measurements, Meas. Sci. & Technol. 16 1007-1013. Krumgalz B.S, Pogorelskii R., Sokolov A., and Pitzer K.S. (2000) Volumetric Ion Interaction Parameters for Single-Solute Aqueous Electrolyte Solutions at Various Temperatures, J. Phys. Chem. Ref. Data, Vol. 29, No. 5.
27th Int. North Sea Flow Measurement Workshop 2009, 20-23 Oct., Tønsberg, Norway
Spilde J., Frøysa K.-E., Holstad M.B., Johansen G.A., Hjertaker B.T., Bruvik E.M. and Stavland S.H., (2008) A novel subsea on-line multiphase fluid sampling and analysis system, Presentation UTC (Underwater Technology Conference), 2008, Bergen Stavland S.H, (2005) Implementation of an ultrasound measurement system for hydrocarbon fraction monitoring and fluid characterisation in the Subsea Online Fluid Analyser, MSc thesis (in Norwegian), Department of Physics and Technology, University of Bergen.
X-ray based densitometer for multiphase flow measurement
Stein-Arild Tjugum, Roxar, Norway
Romulus Mihalca, PANalytical, The Netherlands
1 Introduction Flow measurements by use of gamma-ray attenuation has been utilised in multiphase flow
meters for many years. This represents a robust and reliable type of flow composition
measurement[1]. The use of a radioactive isotope does however introduce some challenges.
When using radioactive sources there is additional paper work to be done, the source has to be
tracked through its life-time and after end of life-time of the instrument the source has to be
disposed of in a safe way. The radiation hazard has to be handled during storage,
transportation and installation. An X-ray based system will only be switched on after it has
been installed and it does not represent any radiation hazard when switched off. Multiphase
flow meters are installed worldwide and Roxar has experienced that in some parts of the
world it is more difficult to install nucleonic gauges and there is a request for multiphase
meters with no radioactive source.
Most Roxar multiphase meters are equipped with a Cs-137 source for density measurement.
For some years Roxar has also supplied non-gamma multiphase meters where the gamma-ray
measurement is replaced with a patented algorithm that combines the information from the
other flow measurements in the meter. The non-gamma meter does however have flow-range
limitations and higher measurement uncertainty compared to the standard meter with gamma-
ray source. The X-ray based flow meter is a new option without radioactive source that does
not have these performance limitations.
A prototype X-ray based flow densitometer has been developed in collaboration between
Roxar (Norway) and PANalaytical (the Netherlands). The prototype meter has been
successfully flow loop tested at Christian Michelsen Research (CMR) in Bergen and TUV
NEL in Scotland. The flow testing includes both measurements on multiphase flow and on
wet gas flow
2 Flow composition measurement by gamma-ray and X-ray
Radioactive sources or X-ray tubes produce high energy photons that can be used in flow
composition measurements. The basic measurement principle is the same independent on the
origin of the photons. The attenuation of photons is measured along the narrow beam path in
the flow, see Figure 1. The beam is shaped by using a collimator that shields the radiation in
other directions than that of the beam. The average attenuation coefficient of the flow
components is thus the measured variable. This attenuation coefficient is dependent on the
energy of the photons, the atomic composition of the flow and the density (or pressure) of the
flow. The photon attenuation is not dependent on the origin of the photons. But it is possible
to produce higher photon energy with a gamma-ray source than with an X-ray source. Due to
the relatively low energy obtainable with an X-ray source it is required to use material with
low density to transmit the beam through the pipe wall. In the Roxar prototype X-ray meter
there are titanium plugs in the steel wall, where the beams are transmitted through.
A traditional X-ray tube produces a continuous energy spectrum, while a gamma-ray source
has discrete energies. Interpreting the attenuation of a continuous energy spectrum is more
complex compared to measuring the attenuation of a single energy beam. In the Fluor’X tube
used in this work mainly discrete characteristic energies are emitted from the tube.
The X-ray tube produces photon energies close to 60keV and the gamma-ray densitometer
used in this work is based on Cs-137 with principal energy of 662keV. Both of these energies
are in the range where Compton scattering is the dominating interaction mechanism. The
attenuation due to Compton scattering is about proportional to the density of the material and
the measurements can be used for finding the density of the flow. There is, however, also
some contribution from photoelectric effect, in particular for the X-ray beam. The
photoelectric effect is strongly dependent on the atomic composition. A result of this is a
higher salinity dependency of the low energy beam.
In the work presented here the gas volume fraction (GVF) has been calculated from the X-ray
measurements and the gamma-ray measurements. The water cut (WC), fraction of water in
the liquid, is found by capacitance measurements in the MPFM and this measurement has
been used as input in the GVF calculations. The GVF measurements from the X-ray system
and the gamma-ray system are compared. In these measurements flow regime effects and slip
(different flow velocity for gas and liquid [2]) is not compensated for. Due to these effects it
is expected that there will be deviation between the measured and the true GVF unless the
flow is homogenous and with no slip. In the Roxar multiphase meter slip is measured and
flow regime effects are also compensated for based on other flow measurements. Flow regime
effects can also be compensated for by using multi-beam gamma or X-ray measurements [3].
Figure 1. Principal sketch of gamma-ray or X-ray based flow composition measurement.
3 The Fluor’X tube The underlying principle of an x-ray tube is to accelerate electrons over a high voltage
potential before hitting a heavy metal target, the anode. During the deceleration process the
kinetic energy of the electrons is converted into x-rays and other forms of energy (mainly
heat). A complex, highly polychromatic x-ray spectrum is generated in a conventional x-ray
tube. The spectral definition consists of a continuous brehmsstrahlung contribution
superimposed by the characteristic fluorescence lines [1] of the anodic material being thus
inadequate for multiphase flow measurements.
Based on a patented design, the Fluor`X tube is capable to generate a near monochromatic
spectrum which is much more suitable for gauging applications. In the Fluor`X tube a primary
conical anode surrounds a beryllium cup holding a secondary conical target outside of the
vacuum envelope (see Figure 1). The inner side of the anode is coated with a high Z material,
in this case gold, over a surface of few cm2. Bombarded with electrons supplied by a circular
filament, the primary anode becomes a source of polychromatic radiation which penetrates the
beryllium window to excite the characteristic fluorescence lines of the secondary target
material (W in our case). Since the secondary target is shielded by the beryllium cup against
the scattered electrons, the brehmsstrahlung contribution in the out coming beam is
significantly suppressed resulting in an output spectrum with mainly characteristic lines of the
secondary target.
Figure 2. Architecture of the Fluor’X tube. The location and arrangement of the filament, Be
window, primary and secondary target are indicated.
A previously reported version of the Fluor`X tube [4] used electrons accelerated to 160 kV,
water cooling system and a combination of Au / W as primary and secondary anodes,
respectively. The spectral output of this tube is presented in figure 2. This radiation has a
strong characteristic line pattern which simplifies spectral interpretation and increase the
confidence and quality of the collected data.
Figure 3. Energy spectrum for Fluor’X tube with 160kV HV. The spectrum is scaled to 100% at
the W k-alfa1 fluorescence peak. The insert shows on logarithmic scale the remainder of the
background.
In line with this concept, a recently developed low power version of the Fluor`X tube was
used to carry out this study (see Figure 4).
Figure 4. The low power version of the Fluor`X tube. The tube length is about 20cm.
This tube is compact and smaller in dimensions than its predecessor and is capable to work at
maximum 100kV supplied by a relatively small generator. Due to low power consumption the
anode cooling is achieved only through heat conduction and free air convection.
The Fluor`X design allows easy reconfiguration of the spectral output by using a wide range
of secondary target materials. The user has the freedom to adjust the beam intensity and to
choose even a composite material for the secondary target, thereby widening the application
range of the Fluor`X tube.
4 The Prototype flow densitometer based on Fluor’X The current version of the Roxar prototype X-ray densitometer was developed in 2007/2008
and first flow-tests were done in 2008. Main challenges for the design are combining the high
voltage (100kV), explosion safety and the transmission of the low-energy X-ray beam from
the X-ray tube to the detector. In the prototype meter this was solved by placing the X-ray
tube and the HV generator inside an aluminium Ex d casing on the outside of the meter body.
Titanium plugs are used to allow the beam to be transmitted through the walls of the meter
body. The prototype is designed as a separate spool piece that can be installed alone or in
series with a multiphase meter. Drawings of the meter are shown in Figure 5. This first design
of the gauge is fully operational and can be used in explosion hazardous atmosphere.
Concerning radiation safety, the meter has been designed to comply with the German
“Vollshutz” standard according to Roentgenverordning 1987. Safety switches will shut down
the beam if the box containing the X-ray tube and HV generator is out of position. The
external dose-rate at 10 cm distance from the surface is below 1 microSv/h.
In this first version of the meter we have used standard components such as the Ex d casing
for the generator and the product is thus not very compact and does not have a Roxar design
Figure 5. Principle sketch and 3d drawing of prototype meter body. Dimension is given in mm.
5 Prototype testing X-ray tubes are often used in laboratory instruments, where the environmental conditions such
as temperature and vibration is quite different from the requirements for gauges used in oil
and gas production. The X-ray tube and generator for the prototype meter has thus been
specially designed for this application and the main design criteria are high stability of the X-
ray beam, robustness and reliability. PANalaytical has done computer simulations and
extensive testing (according to the ISO 13628-6 standard) on key components to obtain an
optimal design
Roxar has done both static tests on the complete system and flow loop testing. The flow loop
testing has been done both at CMR in Bergen and at TUV NEL in Scotland.
The main results from this testing are presented here.
5.1 Static tests
Static tests were done by positioning polypropylene phantoms in the pipe spool. The plastic
phantoms, which have about the same properties as oil concerning gamma/X-ray attenuation,
are shaped as stratified and annular flow regimes.
Table 1 shows the gas fraction measured along the centre beam. Measurements are compared
with the true gas fraction along the beam. We can see that the difference is less that 0.7 %.
Taking into account that there is some uncertainty in the dimensions of the plastic phantoms
this is about as accurate as it is possible to demonstrate.
In all gamma-ray or X-ray densitometers there is a contribution from scattered radiation in the
measured count-rate. This is because radiation is not only absorbed but also scattered by the
flow, pipe wall and other objects between the source and detector. The buildup is defined as
the ratio between the measured detector intensity and the theoretical detector intensity with no
scatter. So the number is always larger than 1. The contribution from scatter complicates the
measurement analysis and it is thus preferable to have the buildup as low as possible. The
scatter is dependent on the GVF and the flow regime and is difficult to predict. The buildup in
the FluorX meter has been measured at different flow regimes, and it is found to be smaller
than that of a Cs-137 based densitometer.
Table 1 Results from tests with polypropylene phantoms in FluorX meter.
Gas fraction [%]
Plastic phantom Count-rate av att coeff theory measured
1. empty pipe 10912 0.0000 100 100.0
2. stratified flow 5788 0.1747 50 50.4
3. stratified flow 5806 0.1738 50 50.6
4. annular flow 5181 0.1749 41 41.7
5. pipe filled 3042 0.1762 0 0.0
Figure 6. Measurement cross-section showing polypropylene phantoms. Dimensions are given in
mm.
5.2 Multiphase flow
In all flow tests the prototype meter body has been installed in a vertical section of the
pipeline after a T-blend. This is the standard way of installing the Roxar multiphase meter.
The prototype has been installed in series with a multiphase meter with a gamma-ray
densitometer, so the flow goes through the prototype before passing through the multiphase
meter. Figure 7 show the installation at the TUV NEL multiphase loop.
Figure 7. drawing of Fluor’X meter and MPFM2600 with pipes for NEL installation.
First flow testing of the prototype X-ray densitometer was done at CMR in March and April
2008. The prototype meter body was installed in series with an MPFM1900 meter. In Figure 8
time series of countrates and GVF measurements are plotted for the X-ray meter and the
gamma-ray meter. The change of countrate is bigger for the X-ray meter and the measurement
sensitivity is thus better for X-ray based system.
Figure 8. Test results Friday 15 February. Left: count rates (Fluor’X on top). A small change in
GVF is easier to detect with FluorX as can be seen from the data indicated. Right: GVF
When comparing the GVF values from the X-ray meter with gamma-ray measurements
(Figure 9) it is evident that there is good correlation. But there are some systematic
differences. This is as expected because the measurements are from two different cross-
sections along the vertical pipe. The gamma-ray and X-ray measurements do not take flow
velocity of the flow components into account. These test measurements are thus based on the
assumption that the velocities of the different flow components are the same. This is often not
the case. There will often be slip, which means that the gas is flowing with a higher velocity
than the liquid. Due to different slip in the two cross-sections there will be deviation between
the two gas fraction measurements. Other factors that could cause deviation is different flow
regime and backflow of liquid in the vertical pipe section. causing a higher GVF at the highest
flow cross-section. This is probably the reason why the measured GVF was higher at the
upper measurement cross-section (the gamma-ray densitometer) for the CMR flow data.
Figure 9. Gas fraction measured by the Fluor’X system vs. the gamma-ray densitometer.
In June/July 2009 flow loop tests on both multiphase and wet gas flow were carried out at
TUV NEL in Scotland. The X-ray meter was installed in series with the Roxar MPFM2600
with gamma-ray densitometer (Figure 7). A time series plot of GVF measurements are shown
in Figure 10. The gas fraction measured in the gamma-ray densitometer is lower because of
higher gas velocity (slip) in the upper measurement cross-section. The MPFM2600 data
shown are also based on the gamma-ray densitometer, but these data are also combined with
other measurements in the multiphase meter and the GVF is thus corrected for slip and other
flow-regime effects.
Figure 10. The NEL reference data for the measurement points are plotted in green. GVF
calculated from Fluor’X measurements, from the gamma-ray data and GVF from the
MPFM2600 is shown. Data are average values over 25s.
Figure 11 show the X-ray measurements compared with gamma-ray measurements and
compared with the NEL reference data. Also here we see the deviation between gas fraction
in the measurement cross-section of the X-ray meter and the gamma-ray meter. Further, we
see that the GVF from the X-ray measurement is underestimated at high GVF when compared
with reference data. This is due to higher slip at high GVF and can be compensated for in a
multiphase flow meter.
Figure 11. GVF measured by Fluor’X plotted against GVF measured by gamma-ray data (left)
and against NEL reference data (right). Right: Dashed lines show +/- 5%
5.3 Wet gas flow
The X-ray meter was also tested in the wet gas loop at TUV NEL. The test-section was
identical to the one used in the multiphase loop. The X-ray meter was placed under a
MPFM2600 meter in a vertical section of the pipe. The GVF was above 90% and the flow
pressure was between 30 and 55 bar. The flow components were nitrogen gas and kerosene
liquid.
Time series plots of GVF measurements are shown in Figure 12. The X-ray data show smaller
fluctuations from the NEL reference data than the gamma-ray data. When compared with
multiphase flow the wet gas flow is more or less homogenous and liquid is flowing at the
same velocity as the gas. There is thus less influence from slip and flow regime effects on the
measurements.
Figure 12. Measured GVF and NEL reference values, Measurements were recorded in NEL gas
loop. Integration time: 25s.
The measured GVF values for a number of test points are shown in Figure 13. The gamma-
ray (662keV) measurements show larger deviation. This can also be seen in Figure 14 where
the measurement error for all the wet-gas test points are plotted in a histogram. The standard
deviation of the GVF X-ray measurements are about half that of the gamma-ray
measurements: 0.16 % vs. 0.37 % absolute. The low energy used in the X-ray meter is more
suitable for the low density wet gas flow. Theoretical calculations show that at wet gas flow
with 50 bar pressure the relative uncertainty of the measured attenuation coefficient is about
double for 662keV compared to at 60keV. So the measurement results are in compliance with
the theory.
Figure 13. Test points recorded in NEL gas flowloop. Measured and reference GVF values are
plotted.
Figure 14. Histogram of measurement deviation for Fluor’X measurements and gamma-ray
measurements.
5.4 Stability of the beam intensity
For the gamma-ray densitometer the intensity of the beam is proportional to the activity of the
radioactive source. The beam intensity is thus very predictable. For the X-ray meter the beam
intensity is dependent on a number of factors, such as the high voltage, the emission current
and mechanical dimensions inside the X-ray tube. It is important that the photon intensity
does not fluctuate due to temperature variations, vibrations or other environmental factors.
The Fluor’X tube has been specially designed to minimize such fluctuations. The Fluor’X
tubes that have been tested have shown an acceptable stability. At the NEL test the beam
intensity for empty pipe was recorded every day, and the standard deviation for these
measurements for the whole test period was 0.1%. Beam stability is further improved in a
later version of the Fluor’X tube.
6 Installation on multiphase flow meter In a final design of the X-ray system it is planned to be integrated as a part of the multiphase
flow meter. It is today possible to get the Roxar topside multiphase meter in a non gamma
version and with gamma-ray densitometer. The X-ray system will then be a third option for
the meter, with this option it will be possible to have a flow meter without the limitations on
measurement accuracy and flow ranges that the non gamma meter has.
A concept design for X-ray system installed on MPFM2600 is shown in Figure 15. By placing
the HV generator in a separate Ex d casing a more practical design is achieved.
Figure 15. Industrial design of the Roxar MPFM2600 fitted with X-ray system
It could also be useful to have the X-ray system as a separate spool piece, a sketch of this is
shown in Figure 16. It could then be installed as a stand alone meter or in series with a non-
gamma meter.
Figure 16. Industrial design of stand alone spool piece with X-ray system.
7 Possibilities for further development The use of an X-ray system opens for new possibilities for flow composition measurement.
Both the intensity of the beam and the photon energy can be adjusted to fit the application.
In the Fluor’X technology there is a possibility to choose the energy of the photons by
choosing the material for the secondary anode. It is possible to use lower photon energy than
what has been used in our prototype tests, and this enables higher measurement accuracy in
low density flow, such as wet gas. There is also a possibility for using multiple energy beams,
which makes it possible to find more flow information such as water cut and water salinity.
Water salinity can also be measured by combining the measurement of scattered and
transmitted radiation (Dual modality densitometry) [2].
The beam intensity can be changed during the measurement. When increasing the beam
intensity the time-resolution of the measurements is improved and more detailed flow
information can be obtained
8 Conclusions A prototype X-ray flow composition meter based on the patented Fluor’X tube been built and
flow tested. The tests show that it is possible to replace the traditional gamma-ray
densitometer on the Roxar multiphase flow meter with the X-ray system. The X-ray system
shows higher measurement sensitivity in all flow tests, and in particular for low density flow.
The radiation safety during operation of the meter is fully acceptable and during installation,
storage and transportation the radiation hazard is removed. The X-ray based flow composition
gauge will be a valuable alternative to the gamma-ray densitometer in future Roxar
multiphase meters.
References
1. G. A. Johansen and P. Jackson, Radioisotope gauges for industrial process
measurements John Wiley & Sons, Ltd. (2004) 336pp
2. S. Corneliussen et al. Handbook of multiphase flow metering, rev 2 2005
(www.nfogm.no)
3. G. A. Johansen and S A Tjugum, Fluid composition analysis by multiple gamma-ray
beam and modality measurements North Sea Flow Measurement Workshop 2007
4. J. Marfeld, G. E. van Dorssen, M. Niestadt, Fluor`X –A near monochromatic X-ray
source, Proc. SPIE Vol. 4502, p. 117-125, Advances in Laboratory-based X-Ray
Sources and Optics II, Ali M. Khounsary; Carolyn A. MacDonald; Eds.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
1
Successful Implementation and Use of Multiphase Meters
Øystein Fosså – ConocoPhillips Gordon Stobie – ConocoPhillips
Arnstein Wee – Multi Phase Meters
1 INTRODUCTION Allocation of production using multiphase meters offers significant benefits in terms of both CAPEX and OPEX. For successful implementation, it is important to have thorough understanding of the application and identify the main issues which influence the performance of the multiphase meter. In order to find a workable solution, the need for metering functions, installation requirements and operation procedures must be addressed. In order to perform accurate measurement of oil, water and gas production the measurement principle needs to be able to cover a wide range of flow rates and combinations of oil water and gas. This is particularly important for slug flow applications where the flow may instantly change from multiphase to wetgas flow conditions and improper use or design of the meter may introduce significant measurement uncertainties. It is also important to understand the influence on the measurements related to uncertainty in PVT configuration data and how this can be dealt with in a practical and cost effective manner. An extensive operator driven development and qualification program has been performed by 10 oil companies in co-operation with MPM to develop a solution that can handle a wide range of operating conditions and be tolerant to significant uncertainties in the PVT configuration data. This paper presents the technical principles of the MPM meter, some multiphase metering challenges which needs to be overcome and the test results from a blind test of the meter at Ekofisk, a field located in the North Sea, operated by ConocoPhillips.
Figure 1 -
MPM HighPerformanceFlowmeter
Figure 1 - MPM High Performance Meters installed in the MPM flow
laboratory
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
2
1.1 The well dynamics challenge Many fields may start out as an oil field in the early years of production and develop into a gas field as the pressure is reduced. Production from multiple zones in a reservoir may also cause significant changes in the GVF over time. The watercut may also increase over time, particularly for water flooded reservoirs. As a consequence, the watercut and GVF in the multiphase flow may change significantly over time. For such fields it is quite common that a multiphase meter is required in the early years of production, however, as the field matures, a wetgas meter would be the more correct choice. A typical well trajectory for such a field is shown in Figure 2. In the example above, the GVF may change from a multiphase GVF of 60% to a wetgas GVF condition of 95%+ in several years where the switching between multiphase and wetgas occurs gradually. In other cases, such as gas lifted wells or long horizontal wells at low pressure, the GVF may continuously change from multiphase to wetgas conditions as illustrated in Figure 3. Here the GVF is continuously changing from 5-95% in a matter of seconds. This corresponds to sudden changes in flow conditions from multiphase to wetgas. These conditions have traditionally been difficult to handle for multiphase and wetgas meters. However they are typical for many field applications in the real world. 1.2 The wet gas challenge
For wetgas applications, the challenge is clearly the accurate measurement of small liquid fractions in a gas dominated production stream. Once the liquid volumes have been successfully measured, the small liquid fraction must then be split into water and
Figure 2 - Typical well trajectory over time
Figure 3 - Slug Flow Example for a GVF of 72%
0.00 %
10.00 %
20.00 %
30.00 %
40.00 %
50.00 %
60.00 %
70.00 %
80.00 %
90.00 %
100.00 %
1 1.5 2 2.5 3 3.5 4
[%]
Time [Minutes]
Slug Flow Example
GVF
WLR
Average GVF = 72%WLR = 8 %
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
3
hydrocarbons. Hence, a metering system that is capable of extremely high resolution is required for this task. On top of this, operators would often like to know the conductivity and salinity of the produced water in order to determine its source. In many real applications the makeup of a wetgas production stream could correspond to as much as 99.9 vol% gas, 0.05-0.1 vol% condensate and a water fraction of only 0.01-0.05 vol%. In such cases, variations in the properties of the dominating phase (gas) will usually correlate strongly to the measurement uncertainties as pointed out by H. van Maanen in [2]. It is therefore essential, as far as possible, that the metering system be insensitive to variations in the gas properties. This has been a particular area of focus in development of the MPM meter. 1.3 The gamma challenge
There are different types of gamma meters used in multiphase metering applications. The MPM meter uses a single high energy (Cesium ) gamma source, with an energy level of 662 keV, and an associated half-life of 30.7 years.
The measured count rate N of a gamma detector is described according to the following equation:
where
Figure 4 shows the calculated mass attenuation coefficient for some selected hydrocarbon fractions, H2S and water in the salinity range 0-20% NaCl. The mass attenuation is calculated using the XCOM database at National Institute of Standards and Technology [1].
For high energy levels, the mass attenuation coefficient is almost constant for all materials, whereas the mass attenuation coefficients vary quite significant at lower energy levels. In particularly the H2S content of the flow will have a large impact on the mass attenuation coefficient.
101
102
103
0
0.5
1
1.5
2
2.5
3M ass A ttenuation vs . Energy Level
Energy Level [kEV ]
Mas
s A
tten
uatio
n [c
m2/
g]
X
Cs 662
X
Ba 133
X
Ba 133
X
Ba 133
M ethaneEthane
Pentane
Decane
H2SFresh W ater
Salt W ater (5% )
Salt W ater (10% )Salt W ater (20% )
Figure 4 : Mass attenuation as a function of energy level Source : National Institute of Standards and Technology,
XCOM Database
N : Gamma Photon Count Rate No : Empty Pipe Calibration Value μ : Mass attenuation coefficient ρ : Density x : Pipe diameter
xeNoN γρ−= *
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
4
For high energy levels like the 662 keV received from Cs 137 source, the gamma measurement is almost a ”true” density measurement, since the mass attenuation for gas and condensate is almost constant at this energy level, and the mass attenuation coefficient for water is a function of water salinity.
For lower energies such as the energy levels obtained with Barium 133 (32, 80 and partly 356 keV), the mass attenuation coefficient is much more dependent on the composition comparred to the 662 keV energy level. As a consequence, detailed information of the composition of the hydrocarbon fluids and water is required when using a low energy gamma source. In particularly the H2S content has a large influence on the mass attenuation coefficient for the 80 and 32 keV energy levels. Using 662 kEV makes the gamma measurement much more tolerant towards changes in the PVT properties of oil, gas and water comparred to low energy gamma measurements.
As a consequence, a simple field configuration of the meter can be done avoiding the need for detailed composition data in order to calculate the mass absorption coefficient of oil, water and gas as required for lower energies.
1.4 The PVT challenge All multiphase meters using a gamma source need to be configured with the fluid properties of oil, water and gas. In most applications the fluid properties will change significantly over time. If the meter is installed on a test header with many wells from different reservoirs, it is important that the meter is tolerant with respect to uncertainty in the PVT configuration data since it is difficult, if not impossible, to maintain accurate PVT data over time for such installations. Many multiphase meters are also used on comingled well streams from subsea tie backs or used to measure the production from wells producing from multiple zones. Under such circumstances, significant variation in the PVT properties can and do occur. Sea or fresh water flooded wells will also experience changes in the water properties as the amount of injected water dilutes the water from the formation. This is particularly the case if there is a large salinity difference between the injected and reservoir water. In order to obtain accurate measurement over time, it is therefore important that a multiphase meter is able to cope with significant variation in the PVT configuration data. Alternatively, procedures have to be put in place for regularly sampling of the well streams. If frequent updates of PVT data is required, the lifecycle cost (OPEX) of obtaining PVT data can easily exceed the cost of the multiphase meter itself and may also introduce significant HSE issues, particularly for subsea, H2S rich well streams and HP/HT applications. There may also be significant time delay from a sample is taken to the results are received from the lab. In most cases the oil is fairly stable, however the water conductivity may change quickly particularly in connection with well stimulation operations. If the user wants to check the well after a stimulation job is completed, the
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
5
changes in water conductivity may ruin the oil/water split if the meter is not capable of handling variations in the water conductivity. Figure 4 shows a typical example from Ekofisk where the water conducitivty from the lab samples taken of the well showed a change from 10 S/m to 6 S/m in less than two months. Another practical problem arrises is if the multiphase meter is connected to a header. At Ekofisk a header typical contains 15-20 wells. Even taking one sample/ Well/month involves a considerable amount of work. More samples would be required if the meter is not capable of handling variations in the water properties. The logistics involved in handling test samples and implementing the result into the meter is also significant and prone to human error. Typically it may also take as much as 1-2 months from the time the sample is taken until the data is entered into the multiphase meter. During this time period the well may have changed significantly as illustrated in Figure 4. Another practical issue which may introduce errors in the PVT models is related to the method for obtaining the GOR for recombination of the oil and gas samples. The GOR when the sample is taken may vary significantly due to the changing slug flow. In such cases, should the average GOR or instantaneous GOR be used when recombining the oil and gas samples? If the instantaneous GOR is used, how should the instantaneous GOR be calculated if the gas sample and oil sample is not taken at exactly the same time? The different vendors of multiphase meters have quite different views on this matter which makes it even more complicated from a users perspective.
Figure 4 : Measured water conductivity (samples) for a typical well at Ekofisk
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
6
2 MPM 3D BROADBAND™ TECHNOLOGY The MPM HighPerformanceFlowmeter is based on patented and licensed technology (5 patents and 2 pending) using a combination of a Venturi flow meter, a gamma detector, a multi dimensional - multi frequency dielectric measurement system [6, 10, 11] and advanced flow models [3-5], which are combined to a multi modal parametrical tomographic measurement system. The Venturi is also used to create radial symmetric flow condition in the 3D Broadband™ section downstream the Venturi. These flow conditions are ideal for use of tomographic inversion techniques. The technology is marketed as 3D Broadband™ and is used to establish a three dimensional picture of what is flowing inside the pipe. The basis for the technology is often referred to as ‘process tomography’ which has many parallels to tomography used in medical applications. In the oilfield, the challenges are however different than in a hospital. Firstly, the meter is measuring fluids and gases under high temperature and pressure. Secondly, the multiphase mixture can be moving at velocities exceeding 30 meters per second inside the pipe, and the amounts of gas, water and oil are unstable and changing all the time. The 3D Broadband™ system is a high-speed electro-magnetic (EM) wave based technique for measuring the water liquid ratio (WLR), the composition and the liquid/gas distribution within the pipe. By combining this information with the measurements from the Venturi, accurate flow rates of oil, water and gas can be calculated. The MPM meter is extremely fast where all the measurement directions are measured in the entire frequency range in a tenth of a second. Averaging of measured raw data is limited, to avoid errors due to non-linearity in the flow. The result is a measurement with an unparalleled performance in multiphase and wetgas flowing regimes. With its dual mode (multiphase/wetgas with automatic switching) functionality, both multiphase and wetgas applications are addressed with the same hardware and software, bridging the measurement gap between multiphase and wetgas meters. 2.1 Full three phase wetgas measurements For ultra high GVF’s the Droplet Count® functionality is an add-on feature that contributes to significantly improving the measurement performance. The Droplet Count® was commercially released in 2009 but has been in field operation in MPM Meters since early 2008. By using Droplet Count® , the MPM meter can perform precise measurements of small quantities of liquid and is very tolerant towards
Figure 5 -
3D Broadband™ tomography based meter
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
7
uncertainties in fluid PVT properties (i.e. gas density and water properties). This is achieved by a patented methodology with unique properties compared to a gamma meter, and for which the measurement accuracy is better the higher the GVF. The the Droplet Count® functionality is further described in [12]. 2.2 Dual Measurement Mode The MPM meter is a combined multiphase and wetgas meter. The meter can be software configured to operate either as a multiphase or as a wetgas meter. This is often referred to as multiphase or wetgas mode.
MPM meter Dual Mode and Droplet Count
Operating range
99,9%
99%
90%
50%
Gas V
oid Fraction (G
VF)
Gas
–flo
w r
ate
Oil – flow rate
* * * * * * * **
* *
*
*** *
***
***
***
** *
* *
* * * * * * * **
* *
*
*** *
***
***
***
** *
* *
Multiphase mode
Wetgas mode
Droplet Count
The MPM meter automatically switches between modes in a fraction of a second.
Covers FULL operating range, and all flow cases
Min
Max
Gas - flow rate
Liqu
id -
flow
rate
The standard MPM meter is delivered either as a multiphase or a wetgas meter. The hardware parts are, however, identical and the difference between the two meter versions is the software. It has two modes: one for wetgas and one for multiphase flowing regimes. Equipped with MPM Dual Mode®, the meter can be configured to automatically switch between the two modes. The switching is done in less than 0.2 seconds. The switching pointy is selectable by the user. 2.3 Water Salinity The MPM meter can measure the conductivity of the produced water. The measured conductivity is converted into water salinity and the water density is calculated, assuming a certain composition of the salt (for instance NaCl). The measurement method is based on RF measurements and MPM’s patented 3D Broadband technology. All multiphase meters require information about the gas and fluid properties (PVT data) as configuration constants. Even though the MPM meter has a low sensitivity to changes in the water salinity, the water properties are important for accurate measurement of the oil flow rates for wells with high water contents. The MPM meter can automatically measure the water conductivity and density in water-continuous emulsions. This eliminate the need for sampling and analysis in order to
Multiphase
Multiphase
Wetgas
Figure 7 - Dual Mode Operating Range
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
8
obtain the water properties. For subsea, H2S rich and HP/HT applications, this is particularly valuable. For low watercuts, the water conductivity has little effect on the measurement uncertainty, provided the specified value is within reasonable limits of the true value. If, however, the Water Liquid Ratio (WLR) is expected to increase during the life of the field and the flow turns to water continuous, then configuring a multiphase meter with the correct water conductivity is important. With MPM’s automatic configuration the water conductivity and water density are automatically and continuously measured by the meter. This eliminates the risk of getting wrong measurement as a consequence of incorrect configuration data. It also eliminates the need to take the produced water/ liquids samples in order to update the configuration constants when the watercut is increasing, and when the salinity of the produced water is changing. This is very valuable for unmanned and remote operations, as well as for subsea installations. The Watercut for which the flow turns into water-continuous depends on the application, but normally it occurs when the watercut gets in the 30-60% range – although water- continuous has been seen at the lower end and the upper end depending on the general flow regime and fluid properties. If slugging is expected, then measuring the water conductivity could be important even for lower watercuts. The reason is that if the water comes in slugs, then the watercut during the slug can be well above the water-continuous threshold. If so, and if the water conductivity is wrongly specified, the oil and water flow rates will be heavily distorted. Another benefit of the method is that the water conductivity is measured at actual temperature conditions avoiding discrepancies in the models which convert the conductivity from one temperature to another. As an example, it is common to use the conductivity at 25 ºC as a configuration parameter since the water conductivity in most cases is measured in a laboratory at room temperature and converted to 25 ºC. The multiphase meter requires the water conductivity at actual conditions, and hence the water conductivity needs to be converted from 25 ºC to the actual line temperature which may differ significantly. This conversion model may be quite inaccurate, introducing a secondary source of error for meters which rely on the water conductivity as a configuration parameter. This is avoided when the water conductivity is measured at line conditions. The salinity measurement is based on a patented method using a dielectric measurements carried out locally at the pipewall using a differential principle with one transmitting and two receieving antennas. Electromagnetic phase measurements are performed over a broad frequency range, and each measurement frequency provides a separate independent equation. All the measurements are combined in such a way that the measured water conductivity represents a “best fit” of the measured water fraction for all the measurement frequencies assuming that the ratio between the real and imaginary part of the dielectric constant of the multiphase mixture is related to the ratio between the real
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
9
and imaginary part of the dielectric constant for pure water. The salinity measurement is further described in reference [9] and [10]. 3 TEST RESULTS The MPM meter has been subject to a very comprehensive user-driven qualification program as outlined in [8]. 3.1 Tests at Ekofisk In 2007 ConocoPhillips purchased a topside MPM Meters in order to test it in a real field application. The test was performed as a blind test where MPM had no knowledge about the test program before or during the test. Following successful testing, the MPM meter has now been installed permanently and is used for well testing and well optimization. 3.2 Test Configuration The 5” topside MPM Meter was installed in series with the test separator at EKOM as shown in Figure 8 below.
Figure 8 - Ekofisk Test Setup
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
10
The MPM meter was installed in series with the test separator for the wells on the south test header. A picture of the MPM meter is shown in Figure 9. The gas outlet on the test separator is measured by a 10” Instromet Ultrasonic Q-Sonic 3S wetgas meter. The oil leg is measured by a Krohne 6” ultrasonic UFM 3030 and the water is measured by a Krohne 4” Coriolis Optimass 7000. A total of 15 wells could be routed through the MPM meter with the following range of fluid and process characteristics:
- GVF : 88-97% - WLR : 1.5 – 48% - GOR : 10 – 43 m3/m3 - Pressure : 20 – 22 barg - Temperature : 26 – 96 °C - Oil Density : 810 – 840 kg/m3 at 15 °C and 1 bara - Gas Density : 0.72 – 0.83 kg/m3 at 15 °C and 1 bara - Water Density : 1030 – 1100 kg/m3 at 25 °C and 1 bara - Water Conductivity : 50 – 170 mS/m at 25 °C and 1 bara
As seen from the listing above, there is significant variation in the PVT properties for both oil, gas and water. The test also contained both oil and water continuous wells with stable flow and slug flow conditions. At the start of the test, the test separator reference flow measurement uncertainty were assessed to be within ± 5% for all phases.
Figure 9 - MPM Meter at Ekofisk
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
11
The measurements from the multiphase meter and test separator were comparred at process conditions without flashing the multiphase meter to test separator conditions. This was considered to be a good approach since the meter and test separator operates at virtually the same temperature and pressure such that the introduced additional uncertainty is deemed to be small. 3.3 Meter Configuration and Comissioning The MPM meter is very tolerant towards uncertainties and variation in the PVT configuration data. Prior to commissioning of the MPM meter, the variation in PVT data was inverstigated and it was decided that a common PVT setup could be used for all the wells. Using a common PVT configuration was considered to be a great benefit since there would be no need for sampling, analysing and management of multiple PVT data configuration setups for the wells. The PVT configuration was obtained by calculating look-up tables for each well based on the available composition prior to commissioning of the meter. Based on the individual tables, an average look-up table was prepared. Prior to delivery from MPM, the meter was configured with the average look-up table and the meter was therefore fit for service upon delivery to Ekofisk. The ‘average’ look-up table has been used for all the wells, and it has remained unchanged since commissioning of the meter in 2007. The MPM Meter was commissioned in October 2007 as per standard procedures. The meter was mechanically installed and then connected to 24 volt supply and a fibre cable for communication with the MPM Terminal in a local control room. The software on the MPM terminal was later installed on an exisiting terminal server onshore. The MPM service engineer performed a check of the meter and an empty pipe calibration of the gamma detector was performed. The entire commissioning, including mechanical installation of the meter were performed over several days. Once the meter was electrically and mechanical commissioned, the remaining tasks of the commissioning was done in less than four hours. No flow testing or tuning towards the test separator were performed during the commissioning of the meter and the MPM service engineer left the platform before there was flow to the meter. The meter have been untoutched by MPM and COP personel since commissioning in October 2007. 3.4 Test Procedure The test of the MPM meter was performed as a blind test. Hence, MPM did not get any information of flow rate neither during commissioning of the meter nor during the testing of the meter. Measurement data was logged continuously by COPNO and handed over to MPM on a regular basis together with start and stop times for the test periods. COPNO then received average flows and time series during the well test periods from MPM.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
12
The MPM data was flashed to test separator conditions. The test separator data was also corrected for differences in liquid volume (levels) in the separator between start and end of the test. Test separator water was measured as mass flow and converted manually to volume flow using water density for each well. No corrections were made for water in the oil leg of the separator. The test period started 14th December 2007 and ended 17th February 2008 comprising a total of 364 test hours based on 76 well tests from 13 wells. At the end of the test, when COPNO had received all the data from MPM and made comparison tables with the test separator, the data was handed over to MPM for comments. The MPM meter is still in operation, delivering measurement data with the same quality as obtained during the testperiod in 2007 and 2008, without any calibration or maintenance need since it was installed in October 2007. 3.5 Test Results Below are chart of the liquid, oil, water and gas measurements for all the well tests.
The average GVF range is 88.3 – 97.7%. 94% of the liquid flow rate measurements are within ±10% and 82% of the data are within ±5%. Well 3 appear to deviate more than the others. Excluding well 3, all tests are within +/- 10% difference and 90% of the data is within ± 5%. It is not clear why well 3 deviated more during the test. For GVF’s in the range 80-95%, the uncertainty spec of the MPM Meter is ± 5%. For GVFs above 95%, the uncertainty spec is ± 10%. Hence, the measurements are well within the specification of the meter.
Figure 10 - Liquid Measurements
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
13
83% of the data is within ± 10%. No correction is made for water in the oil leg of the separator, it is therefore expected that the test separator oil measurement will be slightly higher compared to the MPM meter. The oil content constitutes 2.7 to 7.7 % of the total multiphase flow.
The WLR is in the range 1.5% to 47.3% and the water content constitute 0.2% to 5.5% of the total multiphase flow. Some of the wells were both oil and water continuous.
Figure 11 - Oil Measurements
Figure 12 - Water Measurements
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
14
A larger deviation was seen on the gas measurement which caused some concern. The test separator was fairly new and instrumented with an ultrasonic meter so the MPM gas measurement was initially considered to be suspect. 3.6 Investigation of Gas Measurement Discrepancy MPM investigated the gas flow rate measurement in more detail to look for potential issues which could explain the observed discrepancy. Some of the wells contained severe slugging with fairly long periods with almost pure gas followed by a short liquid slug as seen in Figures 14 and 15. During the periods indicated with red circles, the dP is extremely low and the MPM meter does not measure flow due to a preset cutoff value for the dP transmitter.
Figure 13 - Gas Measurements
Figure 14 -GVF & WLK for slugging Well
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
15
When the well is producing mostly gas, the dP is under the cutoff value and it sets the flow to zero. This caused a small under reading of the gas for some of the wells. The cutoff value was reconfigured to a lower value, however there was still a large deviation between the gas measurement from the separator and MPM meter. The test results were also re-simulated by MPM with automatic switching between wetgas and multiphase mode (dual mode). When the MPM meter was commissioned in October 2007, automatic switching between wetgas and multiphase mode was not available and the meter was therefore configured to operate in multiphase mode. Figure 16 shows a re-simulation of the entire test using automatic switching between multiphase and wetgas mode.
The difference between the test separator and MPM meter was reduced with an average value of approximately 2%. In general, the measurements from the MPM meter are lower compared to the test separator measurements (only 3 tests are higher on the MPM meter).
Figure 15 -Flow Rates of a slugging well
Figure 16 -Resimulated gas measurement in Dual Mode with automatic switching
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
16
Automatic switching between measurement modes also improved the liquid measurement at high GVF, removing some of the positive bias on the liquid flow rate for these wells. Based on a deeper analyse of the measurements of the MPM meter, it was concluded that the measurements between the MPM meter and test separator was different and further investigation was required. There was no correlation between the observed deviation on the gas flow rate and parameters such as temperature, pressure, gas rate, liquid rate, flow conditions (stable or slug flow), GVF, WLR or date and time. However, comparing the measured GOR of the MPM meter and the test separator towards the 30 year historical trend of the GOR for the wells at Ekofisk, it was determined that the GOR measured by the MPM meter was more in line with the historical trend . This suggested that a closer look at the test separator gas measurement was required. Further tests and investigation of the test separator gas measurements revealed that a reason for the observed discrepancy was related to liquids in the USM transducers causing the ultrasonic signals to fall out periodically. In adition, liquid was causing cycle jump/pulse detection problems leading to wrong velocity measurements on some of the paths. The measurement from the test separator was improved by removing some of the measurement chords and rotating the meter in order to minimise the cycle jumps/pulse detection problems. The chart below shows the comparison between the test separator and MPM meter for all the wells on the test header after correcting the gas measurement on the test separator. The difference between the test separator and the MPM meter is now found to be wthin 5-7% for most of the tests. (Figure 17).
0,00
100,00
200,00
300,00
400,00
500,00
600,002/4M
-23
2/4M-23
2/4M-18
2/4M-22
2/4M-18
2/4M-19
2/4M-26
2/4M-26
2/4M-22
2/4M-28
2/4M-28
2/4M-29
2/4M-16
2/4M-29
2/4M-16
2/4M-29
2/4M-30
2/4M-24
2/4M-24
2/4M-27
2/4M-30
2/4M-20
2/4M-25
2/4M-25
2/4M-27
2/4M-27
2/4M-17
2/4M-17
2/4M-23
2/4M-17
2/4M-17
45 34 53 42 47 40 37 57 44 59 41 60 51 50 31 32 61 55 39 36 46 43 56 38 49 58 33 35 54 52 48
Gas
Am
3/h
-15
-10
-5
0
5
10
15
20
% D
iffe
ren
ce
MPM Gas Corrected Gas Corrected Deviation
Figure 17 -Test of gas measurement after correcting the test separator gas measurement
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
17
3.7 Oil/Water Continuous liquid flow Two of the wells had nominal WLR averages in the range 30-45%. During the well test, the WLR was observed through the MPM meter as varying in the range 0-100%. This has been observed many times though the use of multiphase meters. The MPM meter handles transitions between oil and water continuous flows and no difference in performance compared to other wells has been observed. Later well tests has indicated that the WLR has increased in some of the wells. Fugure 18 indicates the observed water liquid ratio (through a MPM) of a well with a nominal 60% WLR.
Oil and water flow, in the short term, may give an irregular cycle and give unstable data with respect to WLR measurement. This is due to low velocity in the pipe such that separaction occurs and also the dynamic head and liquids ‘available for lift’ in the reservoir. Hence, there may not be enough force to drag the water together with the oil. As seen from the graph to the right, the WLR varies in the range from 20 to almost 100 % during a few hours. These WLR changes often form a repeatable pattern over time. 4 OBSERVATIONS AND EVALUATIONS 4.1 Well test timing By using a multiphase meter, it is possible to test the wells in a significantly shorter time compared to the test separator. The chart in Figure 19 shows the measurement from the MPM meter when a new well is routed through the meter. After approximately 5 minutes, at 10:50, averaging 15 minutes will give the same results as the average over the whole period (within ± 1.5%). The chart shows the same period for the test separator. The normal well test is 4 hours. In addition, the well requires 1 – 1.5 hour to stabilise. For the test separator, a 3.5 hour
Figure 18 –Example of water separation in the well
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
18
average is required in order to give the same result over the whole test period (within ± 1.5%). Clearly using shorter well tests, significantly more frequent well testing can be achieved with a multiphase meter compared to the test separator, which can only increase well testing accuracy.
Figure 19 –Well stabilisation on MPM Meter
Figure 20 –Well stabilisation on Test Separator
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
19
4.2 Observation during installation, commissioning and testing The following points summarises the main observations by COPNO from the installation, commissioning testing phase:
- The installation of the meter was supervised by MPM personel . - In general there were no problems during installation, commissioning and
operation. Configuration and commissioning of the meter once the mechanicall installation was complete in less than 4 hours. The MPM meter was configured with PVT data based on an average for all wells at the factory. A single look up table for oil and gas density and viscosity and oil/gas surface tension was created.
- Communication to COP offshore was trouble free. - After a few weeks, MPM personnel checked the status of the meter from the
onshore computer. A problem with a sub-supplier software module was discovered. This was a known software bug, detected a year earlier. The error was corrected from onshore and the meter (and test) was restarted. No problems thereafter and the meter has been continuously in use with an uptime of 100%.
- During the test period, MPM checked the status of the meter every two weeks and downloaded rawdata log files from the offshore service computer.
- The flow tests were completed during normal offshore operations. The test separator/MPM meter was used during well cleanups, milling operations (to remove scale) and scale squeeze operations etc. All tests were logged and reported. It is considered that these operations have no impact on the MPM meter performance.
4.3 Design and FAT Test Points Prior to delivery, the MPM meter was configerred based on the latest well data. A standard FAT, containing approximately 20 test points, was performed in the MPM flow lab prior to delivery. Figure 21 shows the two-phase flow map for the design well data and the FAT test points as recommended in the Handbook of MultiPhase Flow Metering [7].
Figure 21 –Design & FAT Data in Flow Map
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
20
The chart in figure 21 shows the same data in a flow regime map. See Figure 22. From the charts it is seen the GVF would expected to be in the 90-95 % area with a mix of churn and annular flow conditions.
Whilst we often specify a well’s performance as a single point in time on the Two Phase Flow map, the Real World is very different in that in a short period of time many wells exhibit considerable changes. Typically this is shown in Figure 23 The red points are the average flow rate during the test whereas the scattered datapoints are all the individual measurements from 12 wells during a one hour period at a resolution of 0.3 second. The black lines are the designed minimum and maximum operating limits for the meter. From Figure 23 it is seen that in real life, the GVF varies in the entire range from 50-100% GVF. with a much larger variations in the flow conditions than can be expected from the average data. In fact, the measurements are outside the design envelope for the meter for some wells a significant part of the time.
Figure 22 –Design & FAT Data Flow Regime Map
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
21
Figure 24 below shows the span in the real flow conditions and the well characteristics. Whereas the expected flow map was limited to churn and annular flow as indicated by the brown oval ring, the real flow condtions covered both annular, churn, slug and dispersed bubble flow as shown by the larger red rectangle. The real well characteristics is also summarized in the Figure 24 as:
Figure 23 –Real Life Data – in Flow Map
Figure 24 –Real Life Data – Flowregimes
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
22
From this it may be concluded that a multiphase meter for this application needs to be designed to handle all flow conditions. If slug flow is expected, care must be taken when using “average flow rates” as it is likely that the meter will have to operate beyond its normal operational envelope. During slug flow, the dP of the liquid slug may peek to a value many times greater than the average value. Similarly, the venturi dP during the gas slug may be just a fraction of the average dP for the well, and possibly below its normal dP cutoff point. Automatic switching between multiphase and wetgas is an important feature for these flow conditions. 4.4 Future use of Test Separators The existing EKOM design with two headers is a product of the high well count used on Ekofisk (30 wells is current and 40 and 50 wells are envisaged in the future). With a single header and a high well count effective well testing can only be carried out with two headers and two MPFM’s. The design adopted however with the two MPFM able to access a Test Separator is an acknowledgement of the fact that a test separator is required by operations for many purposes. These are:
• The ability to test the MPFM against a recognised flow measurement ‘standard’ – although as has been shown here – not all test separator measurments are necessarily ‘good’. Test separator is required by operations to verify performance of multiphase meter. This is also useful in order for operations to gain confidence in the measurements from the multiphase meters.
• The need for fluid samples – multiphase sampling is currently a ‘leap into the unknown’ and samples from a separator are a known technology, effective and comparitively safe.
• The ability to kick off low pressure wells after a shutdown later in field life can only be done effectively from a test or an independent LP separator. This is a known and valid economic driver in order not to leave ‘oil in the ground’.
• The ability to clean up wells after well work. Whilst this does no good to the test seperator meters, arguments from metering engineers against this policy are rarely effective – as the replacement of flow meetrs is seen as small compared to the extended stay of a drilling rig and specialised clean up and product disposal procedures.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
23
5 SUMMARY AND CONCLUSIONS
• The MPM meter has been developed, tested and qualified in JIP’s together with 9 major oil companies.
• In 2007 an MPM Meter was installed at Ekofisk. The MPM meter has been in
continuous operation since it was installed in 2007 without any operational problems. The meter is considered to be a good tool for well testing and production optimisation.
• The MPM meter has had no operational problems during the test period. It is now
in continuous operation following the test program.
• One common PVT configuration has been used for all the 15 wells at the test header, and the measurements from the MPM meter have proved to be robust with respect to variation in the PVT configuration data with no need for sampling of the wells.
• The MPM meter operates well – over a large operating envelope brigding the gap
between multiphase and wetgas meters.
• The meter handles oil and water continuous flows, high and low GVF’s and the transitions with no discernable loss of perfromance.
• The meter handles extreme salinity changes without recalibration.
• For all process conditions observed, the MPM meter has operated equally well.
• The user interface is easy to use and the log files are a good diagnostics tool.
• The MPM meter does not seem to be affected by the different well operations
(scale squeeze, milling operations, clean-up, or water conditions) experienced.
• The existing EKOM design with a multiphase meter on each header seems to be a practical and cost effective implementation of multiphase meters for well optimisation and well testing for the Ekofisk field.
• The MPM meter is still in operation, delivering measurement data with the same
quality as obtained during the testperiod in 2007 and 2008, without any calibration or maintenance need since it was installed in October 2007.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
24
6 ACKNOWLEDGEMENTS The development of the MPM HighPerformanceFlowmeter has been supported technically and financially by the six oil companies; ENI, ConocoPhillips, Hydro, Shell, Statoil, and Total. In addition, Anadarko, Chevron and Gaz de France have participated in the test of the MPM subsea meter at South West Research Institute. Nine oil companies; ENI, Chevron, ConocoPhillips, Gaz de France, Shell, StatoilHydro, Petronas, Total and Woodside are also participating in an ongoing “In Situ Verification JIP” to develop and qualify in-situ operational procedures and functions to automatically monitor and verify the measurement integrity and configuration data of the MPM meter. Flow models implemented in the MPM meter have been developed in collaboration with Onera, Total and Gaz de France. The Ekofisk field is operated by ConocoPhilips and the partners in the field are Total, ConocoPhilips, ENI, Petoro and StatoilHydro. 7 REFERENCES [1] M.J. Berger, J. H. Hubbel et al., XCOM Photon Cross Section Database, National
Institute of Standards and Technology [2] Hans R E van Maanen, Shell Global Solutions, Measurement of the Liquid Water
Flow Rate Using Microwave Sensors in Wet-Gas Meters: Not As Simple As You Might Think, North Sea Flow Measurement Workshop 2008.
[3] J.P. Couput, G. Salque, P. Gajan, A. Strzelecki, J.L. Fabre, New Correction Method For Wet Gas Flow Metering Based on Two Phase Flow Modelling: Validation on Industrial Air/Oil/Water Tests at Low And High Pressure, North Sea Flow Measurement Workshop 2007.
[4] R. de Leeuw, Liquid Correction of Venturi meter Reading in Wet Gas Flow, North Sea Flow Measurement Workshop 1997
[5] M. van Werven, H. R. E. van Maanen Modelling Wet Gas Annular/Dispersed Flow Through a Venturi AIChE Journal , June 2003, Vol. 409, No. 6
[6] A Wee, H Berentsen, V.R. Midttveit, H. Moestue, H.O. Hide, Tomography powered multiphase and wetgas meter providing measurements used for fiscal metering, North Sea Flow Measurement Workshop 2007.
[7] Handbook of Multiphase Flow Metering, Rev. 2, March 2005, Tekna [8] MPM Meter Qualification – MPM White Paper No. 4 [9] Water Salinity Measurement & Auto Configuration – MPM White Paper No. 2 [10] A.Wee, I. M. Skjældal, Ø. L. Bø, Multiphase metering with early detection of
changes in water salinity – Americas Workshop, 2009 [11] Ø. L. Bø, A. Wee, I.M. Skjældal, Tomography powered 3-phase flow metering in
the wet gas regine, 8th International South East Hydrocarbon Flow Measurement Workshop, March 2009
[12] MPM Droplet Count, MPM White Paper No. 7
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
1
AN IMPROVED MODEL FOR VENTURI-TUBE OVER-READING IN WET GAS
Michael Reader-Harris, TUV NEL
Emmelyn Graham, TUV NEL 1 Introduction Venturi tubes are commonly selected for the measurement of wet-gas flows. Reasons for this include their physical robustness to withstand erosion and the impact of liquid slugs at high velocities, familiarity with their use and the availability of standards for their use in dry-gas conditions. The presence of the liquid causes an increase in the measured differential pressure and results in the Venturi tube over-reading 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 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 correlations currently available for correcting the over-reading of Venturi tubes have been derived from a limited set of data and may only be suitable to cover restricted ranges of Venturi tube parameters, for example, a specific diameter ratio. Use of correlations outside the conditions used to define them can result in large errors in the calculation of the gas mass flowrate. This paper describes a new Chisholm/de Leeuw-type model for the over-reading, which covers a broader range of Venturi parameters such as diameter ratio and pipe diameter. The model also accounts for the behaviour of the over-reading with different liquids. 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.
The Lockhart-Martinelli parameter, liq gas
gas liq
mX
m
ρρ
= (1)
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.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
2
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.
Gas densiometric Froude number, gas 1,gas
gasliq 1,gas
vFr
gD
ρρ ρ
=−
(2)
where vgas is the superficial gas velocity, g is the acceleration due to gravity and D is the pipe internal diameter.
The superficial gas velocity is given by gasgas
1,gas
mv
Aρ= (3)
where A is the pipe area. The gas-to-liquid density ratio, DR, is defined as
1,gas
liqDR
ρρ
= (4)
The corrected gas mass flowrate, mgas, is given by
wet 1,gas wet gas,apparentgas
2dEA C p mm
ε ρφ φ
∆= = (5)
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, φ is the wet-gas over-reading or correction and mgas,apparent is the apparent or uncorrected gas mass flowrate. εwet was determined from ISO 5167-4 [1] using the actual value of pressure ratio.
The velocity of approach factor, E, is defined as 4
1
1E
β=
− (6)
wet-gas
gas
p
pφ
∆≈
∆ (7)
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
3
3 Brief History of Wet-Gas Correlations 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. Murdock’s correlation gave the over-reading as 1 1.26Xφ = + (8)
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 2Ch1 C X Xφ = + + (9)
where CCh accounts for the density ratio and is given by the following equation:
liq 1,gas
Ch1,gas liq
n n
Cρ ρ
ρ ρ
= +
(10)
where n = 0.25. 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 (9) and (10) but showed that n was a function of the gas Froude number: 0.41n = for gas0.5 1.5Fr≤ < (11)
( )gas0.7460.606 1
Frn e
−= − for gas 1.5Fr ≥ (12)
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.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
4
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. It is worth noting that increased errors are more likely if using correlations at pressures lower than that covered by the correlation, rather than at higher pressures. This is due to the fact that at lower pressure the density ratio (of the gas to the liquid) is lower, and hence the fluid combination is less homogeneous. 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]. There are no openly available correlations based on data that cover the range of fluid conditions and parameters that are encountered by industry. 4 Derivation of New Correlation Most of the available correlations are based on Chisholm’s model, which produces an almost straight-line fit though the wet-gas data. This type of model generally provides an acceptable fit for orifice-plate data. However, with Venturi data it is noticeable that as the Lockhart-Martinelli parameter, X, tends towards zero the over-reading does not tend to zero in a linear fashion. The gradient changes as the liquid fraction and hence the Lockhart-Martinelli parameter, X, tend to zero. This is in contrast to data from orifice plates (although sometimes with orifice plates the straight line fails to go exactly through the origin). Figure 1 clearly illustrates the problem for a 0.6 diameter-ratio Venturi tube using different fluid combinations with either nitrogen or argon as the gas phase and water or Exxsol D80 (a kerosene-substitute fluid) as the liquid phase.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
5
1
1.1
1.2
1.3
1.4
1.5
1.6
0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.25 0.275 0.3 0.325
Ven
turi
Met
er O
verr
ead
ing
, φφ φφ
Lockhart-Martinelli Parameter, X
Nitrogen-Exxsol D80
Argon-Exxsol D80
Nitrogen-Water
Equation (13)
Figure 1 Over-reading data for a 4-inch Venturi tube with ββββ = 0.6, Frgas = 1.5, ρρρρ1,gas/ρρρρliquid = 0.024
One solution to account for this effect would be to add an additional term to the model to provide a better fit to the data (e.g. as in [7]), but this has no obvious physical significance. For the data presented in Figure 1 (excluding data with X < 0.02) if a multiplicative term is acceptable the best fit of a Chisholm-type model is:
2Ch1.0532 1 C X Xφ = + + , where
0.3162 0.3162liq 1,gas
Ch1,gas liq
Cρ ρ
ρ ρ
= +
(13)
The over-reading is ‘corrected’ and provides a better fit to the data (with X > 0.02) by modifying the power index in the Chisholm constant (Cch) to a value between those used by Chisholm and de Leeuw, and by adding a multiplicative constant to the over-reading equation. It is proposed here that the reason for the non-linearity is that in wet-gas conditions the discharge coefficient, C, is not equal to its dry-gas value, as is often assumed. As soon as even a very small quantity of liquid is added to the gas the discharge coefficient reduces. It is interesting to note that, whereas in dry gas a Venturi tube can often emit an audible tone (it is sometimes referred to as a singing Venturi), in wet-gas conditions Venturi tubes never ‘sing’; even a tiny quantity of liquid seems to change the flow in such a way as to prevent the emission of an audible tone. For the Venturi tube whose data are in Figure 1 the dry-gas discharge coefficient is a function of Reynolds number with a mean value of 1.009 over all its calibrations. However, the effective discharge coefficient in wet gas when calculated for X > 0.02 is more than 5% lower than the value in dry gas. The effective discharge coefficient
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
6
was calculated by determining best-fit values of C and n in Equations (5), (9) and (10) for the data with X > 0.02. Following this observation the discharge-coefficient values in wet-gas conditions were calculated for more data sets to establish if the discharge coefficients in dry and wet gas were generally different. The discharge coefficients were calculated from Equations (5), (9) and (10) (allowing n to vary) for the data sets in Table 1.
Table 1 Wet Gas Data
Beta ratio Pipe size Gas phase Liquid phase Reference
0.4 4-inch nitrogen Exxsol D80 5
0.6 4-inch nitrogen water 6, 7
0.6 4-inch nitrogen Exxsol D80 6, 7
0.6 4-inch argon Exxsol D80 6, 7
0.75 4-inch nitrogen water 6, 7
0.75 4-inch nitrogen Exxsol D80 6, 7
0.75 4-inch argon Exxsol D80 6, 7
0.55 6-inch nitrogen Exxsol D80 11
The discharge coefficient in dry-gas conditions varies between the different Venturi tubes. So it is not surprising that there is a significant variation in the value of the discharge coefficient in wet-gas conditions. This variation makes it harder to see what parameters affect the value. Flow calibrations usually relate the discharge coefficient to the Reynolds number; however, from analysis of the data it seemed more reasonable to use the throat Reynolds number or throat Froude number. Of these the throat Froude number seems more reasonable to use on physical grounds. The throat Froude number is calculated as:
gasgas,th 2.5
FrFr
β= (14)
Figure 2 shows the discharge-coefficient data plotted as a function of the throat Froude number.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
7
0.9300
0.9400
0.9500
0.9600
0.9700
0.9800
0.9900
1.0000
1.0100
0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0
Fr gas,th
C
beta = 0.4 nitrogen/Exxsol D80
6-inch beta = 0.55 nitrogen/Exxsol D80
beta = 0.6 nitrogen/Exxsol D80
beta = 0.6 argon/Exxsol D80
beta = 0.6 nitrogen/water
beta = 0.75 nitrogen/Exxsol D80
beta = 0.75 argon/Exxsol D80beta = 0.75 nitrogen/water
Figure 2 Values of the discharge coefficient as a function of Frgas, th. Data from
Table 1 The power index, n, from the Chisholm correlation (in Equation (10)) has been calculated and plotted in Figure 3 against a suitable function of Frgas. The function exp(-0.8Frgas/H) was used to linearise the data. The choice of the liquid has a significant effect and this has been accounted for in the use of an additional term, H. H depends on the liquid type and was defined as equal to 1 for Exxsol D80. H = 1 is found to be a satisfactory value for other hydrocarbon liquids, and H = 1.35 was determined for water from fitting the data. Analysis of the available data shows that the choice of gas has a negligible effect on the relationship between n and Frgas, provided that the density ratio between the gas and the liquid is kept constant.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
8
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
exp(-0.8Fr gas/H ) (H = 1 Exxsol D80; H = 1.35 water)
nbeta = 0.4 nitrogen/Exxsol D80
6-inch beta = 0.55 nitrogen/Exxsol D80
beta = 0.6 nitrogen/Exxsol D80
beta = 0.6 argon/Exxsol D80
beta = 0.6 nitrogen/water
beta = 0.75 nitrogen/Exxsol D80
beta = 0.75 argon/Exxsol D80
beta = 0.75 nitrogen/water
Figure 3 Values of the power index, n, from the Chisholm constant, as a function
of the gas Froude number. The strong effect of the liquid could suggest that the phase boundaries for a wet-gas flow in which the liquid is water are actually different from those in which the liquid is Exxsol D80 (or a general hydrocarbon liquid). Figure 4 shows the two-phase flow pattern map produced by Shell, which is commonly referred to in many wet-gas documents. This shows the flow conditions where certain flow patterns, for example stratified flow, are likely to exist. The phase boundaries, illustrated in this figure, between the flow regimes may depend on the liquid properties. This work and previous CFD [7] suggest that the effect of the liquid on the over-reading of Venturi tubes is strongly related to the surface tension of the liquid.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
9
Figure 4 Horizontal two-phase flow pattern map (courtesy of Shell) The analysis presented so far established appropriate dependencies for the values of n and C to develop a new correlation based on the Chisholm correlation. The range of data used for development of the correlation was extended and is summarised in Table 2.
Table 2 Wet gas data
Diameter ratio Pipe size Gas phase Liquid phase Reference
0.4 4-inch nitrogen Exxsol D80 5
0.6 4-inch nitrogen water 6, 7
0.6 4-inch nitrogen Exxsol D80 6, 7
0.6 4-inch argon Exxsol D80 6, 7
0.75 4-inch nitrogen water 6, 7
0.75 4-inch nitrogen Exxsol D80 6, 7
0.75 4-inch argon Exxsol D80 6, 7
0.55 6-inch nitrogen Exxsol D80 11
0.6 4-inch nitrogen Exxsol D80 -
0.6 4-inch natural gas Exxsol D80 -*
0.6 2-inch natural gas Stoddard solvent 12
0.6 2-inch natural gas water 12
0.4 4-inch natural gas decane 13
0.7 4-inch steam very hot water 14 * data points with the r.m.s of the fluctuating component of the differential pressure greater than 0.98% of the mean differential pressure were excluded.
0.0001
0.001
0.01
0.1
1
10
0.001 0.01 0.1 1 10
Densiometric Gas Froude No. (Frg)
Den
sio
met
ric
Liq
uid
Fro
ud
e N
o. (
Fr l)
Slug
Stratified
AnnularDispersed
X=1
X=0.1
X=0.01
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
10
The values for n determined from the data from Table 2 were plotted in a similar manner to those shown in Figure 3 and fitted to determine a formula for n. The value of n was determined as the maximum value from one of two equations. These equations describe the two patterns in the data that can be seen in Figure 3 with the initial negative gradient changing to a horizontal line. n is the larger of the values returned by
0.8 /20.583 0.18 0.578 gasFr Hn eβ −= − − (15) and
20.392 0.18n β= − . (16) This is summarised in one equation as
0.8 /2 2max(0.583 0.18 0.578 ,0.392 0.18 )gasFr Hn eβ β−= − − − (17) As can be seen from Figure 1 the over-reading value changes its slope as X tends to zero. This is due to a change in discharge coefficient. Therefore an appropriate equation was fitted to the data from Table 2 to account for this change in the discharge coefficient, C. The discharge coefficient was fitted based on a threshold value of X, called Xlim. Xlim was defined as the value where the over-reading data showed a distinct change in gradient. Then C is given by
fully wet lim
dry dry fully wet limlim
( )
C X X
C XC C C X X
X
≥
= − − <
(18)
where, from the overall fit of the data, Xlim = 0.016 An effective discharge coefficient for wet gas, Cfully wet, was determined by fitting the data from Table 2. This was based on an exponential fit as a function of throat Froude number, Frgas, th
gas,th0.05fully wet 1 0.0463
FrC e
−= − (19)
where the value of 1 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 ∞. A summary equation to determine the discharge coefficient, C, accounting for the full range of X is
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
11
gas,th-0.05 = 1-0.0463e min 1,
0.016Fr X
C
(20)
Figure 5 shows the errors in the gas mass flowrate determined when using these new equations as a function of the Froude number. The data from Table 2 are used. The percentage errors were calculated as
correlation actual
actual100
m m
mσ −= (21)
where σ is the percentage error in the gas mass flowrate, mcorrelation is the gas mass flowrate determined from using the correlation and mactual is the actual measured gas mass flowrate.
-4
-3
-2
-1
0
1
2
3
4
0 1 2 3 4 5 6 7 8
% E
rro
r in
gas
mas
s fl
ow
rate
Frgas Figure 5 Errors in gas mass flowrate using Equations (17) and (20) as a function
of Frgas As stated earlier the liquid type affects the over-reading value. Using the data for water, hydrocarbons and very hot water (in a steam/water flow) in Table 2 the value of H was determined. This value is shown in relation to the liquid properties of surface tension and viscosity in Table 3. This shows that there is no obvious correlation of H with the liquid viscosity and that it is more likely that the surface tension is the dominant factor in the value of H.
Table 3 Liquid properties Liquid H (-) Surface tension (N/m) Viscosity (mPa s) Water 1.35 0.060 1
Hydrocarbon 1 0.027 2.4 Very hot water 0.79 0.022 0.1
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
12
5 Testing the correlation The correlation was mostly fitted to data recorded for research purposes in which accurate dimensions were measured to check compliance with ISO 5167-4. However it is important to test the correlation using data that were not used to derive the correlation. This illustrates the applicability of the correlation to other Venturi tubes and can provide confidence in the use of the correlation. Krohne very kindly agreed to allow their data on a large number of Venturi tubes calibrated at TUV NEL in nitrogen/Exxsol D80 to be used. The data for 0.4 ≤ β ≤ 0.75, where β is the diameter ratio, are summarised in Table 4.
Table 4 Wet gas data from Krohne
Diameter ratio Pipe size Gas phase Liquid phase Number of Venturi tubes
0.6 4-inch nitrogen Exxsol D80 6 off
0.47 4-inch nitrogen Exxsol D80 1 off
0.43 4-inch nitrogen Exxsol D80 1 off
0.4 4-inch nitrogen Exxsol D80 1 off
0.57 6-inch nitrogen Exxsol D80 6 off
0.61 6-inch nitrogen Exxsol D80 5 off
0.61 10-inch nitrogen Exxsol D80 2 off
The data for one of the 4-inch and for one of the 6-inch Venturi tubes were not used as each had a mean discharge coefficient in dry gas below 0.97. On the basis of Table B.2 of ISO 5167-4 it is expected that the discharge coefficient will be equal to 1.010 ± 3% for 108 > Red > 2 × 106, where Red is the throat Reynolds number. Overall the data from 20 Venturi tubes were used to test the correlation. Figure 6 shows the percentage error in the gas mass flowrate of the data used to derive the correlation (Table 2) and of the validation data from Krohne (Table 4) as a function of the gas Froude number.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
13
-4
-3
-2
-1
0
1
2
3
4
0 1 2 3 4 5 6 7 8
% E
rro
r in
gas
mas
s fl
ow
rate
Frgas
Derivation data (Table 2)
Validation data (Table 4)
Figure 6 Errors in gas mass flowrate using Equations (17) and (20) as a function
of Frgas It can be seen that the Krohne validation data show only very slightly increased errors in the gas mass flowrate compared with the derivation data. This provides confidence that the correlation is robust enough to be used for Venturi tubes that are within the tolerances of the standard ISO 5167. The standard deviation of the errors in gas mass flowrate is given in Table 5.
Table 5 Standard deviation of errors in gas mass flowrate Data fitted to give best-fit
equations Data tested to give errors Standard deviation of
errors in gas mass flowrate Table 2a Table 2 0.86% Table 2a Tables 2 and 4 0.980%
Tables 2 and 4b Tables 2 and 4 0.969% Table 2a Table 4 1.34%
a Equations (17) and (20) b Equations (17) and (20) with revised constants On the basis of Table 5 there is no benefit in using the Krohne data as derivation data as the standard deviation of the errors only reduces from 0.980% to 0.969%. Moreover, it is better to use the Krohne data only for validation purposes since no metrological information on the Venturi tubes was available and since it was good to have test data to which the constants in Equations (17) and (20) had not been fitted. The error in the gas mass flowrate as a function of the Lockhart-Martinelli parameter, X, is presented in Figure 7 as this may be more informative. It is interesting to note that when the errors for the Venturi tube with mean Cdry = 0.9694 (omitted because it was less than 0.97) are added to Figure 7 they are approximately 3% for small X but are less for larger X. This supports the view that Equations (17) and (20) can be used to determine the gas mass flowrate with an uncertainty of 3% for X ≤ 0.15 and 2.5% for 0.15 < X ≤ 0.3.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
14
-4
-3
-2
-1
0
1
2
3
4
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
% E
rro
r in
gas
mas
s fl
ow
rate
X
All derivation and validation data
Cdry = 0.9694
Figure 7 Errors in gas mass flowrate using Equations (17) and (20) as a function
of X In summary the new correlation can be used to determine the discharge coefficient in wet-gas conditions and a value for n to use in determining the wet-gas over-reading based on the Chisholm model. This can be used to determine the gas mass flowrate. From this research the over-reading was found to be dependent on the Lockhart-Martinelli parameter, the gas-to-liquid density ratio, the gas densiometric Froude number, the diameter ratio and the liquid properties. This new 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
with an uncertainty of 3% for 0.15
2.5% for 0.15 < 0.3
X
X
≤ ≤
It has been assumed here that there is no need for a maximum limit on Frgas,th as at higher values of Frgas,th the flow becomes more homogeneous and less likely to deviate from the behaviour accounted for in the correlation.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
15
6 Using pressure-loss measurements to determine the liquid content 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 this can be exploited to determine the wetness. 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 measure of the liquid flowrate. All the information presented in this paper is based on Venturi tubes with a divergent angle of 7.5°. The downstream tapping is placed around 6D downstream of the downstream end of the divergent of the Venturi tube. Figure 8 Schematic of Venturi tube illustrating total pressure loss (∆∆∆∆ϖϖϖϖ) and the differential pressure (∆∆∆∆p) to calculate the pressure loss ratio (∆∆∆∆ϖϖϖϖ/∆∆∆∆p). (Note diagram is not drawn to correct scale or dimensions) Table 6 shows the sets of data that were used to derive a correlation between the pressure loss ratio and X.
Table 6 Wet gas data
Diameter ratio Pipe size Gas phase Liquid phase
0.4 4-inch nitrogen Exxsol D80
0.6 4-inch nitrogen water
0.6 4-inch nitrogen Exxsol D80
0.6 4-inch argon Exxsol D80
0.75 4-inch nitrogen water
0.75 4-inch nitrogen Exxsol D80
0.75 4-inch argon Exxsol D80
0.6 4-inch nitrogen Exxsol D80
0.6 4-inch natural gas Exxsol D80
From these data the ratio of the pressure loss, ∆ϖ, to the differential pressure, ∆p, was determined in dry gas as
∆∆∆∆ϖϖϖϖ
∆∆∆∆p
Flow
6D
3.75o
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
16
9
dry
0.0896 0.48p
ϖ β∆ = +∆
(22)
The increase in pressure loss due to wetness was defined as
dry
Yp p
ϖ ϖ∆ ∆= −∆ ∆
(23)
The maximum value of the increase in pressure loss due to wetness cannot be exactly determined from the database, but a good approximation is obtained by defining Ymax, as
maxdry 0.3X
Yp p
ϖ ϖ
≈
∆ ∆ = − ∆ ∆
(24)
Ymax was first plotted as a function of ρ1,gas/ρliq. This is shown in Figure 9.
0
0.1
0.2
0.3
0.4
0.5
0.6
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Ym
ax
ρρρρ1,1,1,1,gas/ρρρρliq
Nitrogen/Exxsol D80
Argon/Exxsol D80
Nitrogen/water
Figure 9 Ymax as a function of ρρρρ1,gas/ρρρρliq for nominal values of Frgas in the range
1.5 to 4.5 Figure 9 shows that that Ymax is dependent on the density ratio as well as on the liquid type but appears to show no effect of the gas type. The argon/Exxsol D80 and nitrogen/Exxsol D80 fluid combinations have overlapping data points, whereas when the liquid is water (as in the Nitrogen/water combination) Ymax has a higher value.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
17
From examination of the data it appears that the same dependence on liquid as was found when determining the value of n (Equation (17)), is appropriate to use here. Therefore the data for Ymax were plotted against a function of Frgas/H in order to account for the liquid type (Figure 10).
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7 0.75 0.8 0.85 0.9 0.95 1
exp(-0.045Fr gas/H )
Ym
ax
Density ratio 0.024Density ratio 0.046Density ratio 0.090
Figure 10 Ymax as a function of Frgas/H
Figure 10 clearly shows a dependence on the density ratio, ρ1,gas/ρliq, and this dependence can be represented by a simple function that becomes extremely small as ρ1,gas/ρliquid tends to 1. The following form for Ymax was then assumed:
1,gas
max gasliq
exp /Y a b cFr Hρρ
= − −
(25)
On fitting the available data from Table 6 Ymax was determined as
1,gas
max gasliquid
0.61exp 11 0.045 /Y Fr Hρρ
= − −
(26)
Y/Ymax was then evaluated for each point and plotted in Figure 11.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
18
0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
X
Y/ Y
max
Figure 11 Y/Ymax as a function of X
In Figure 11 the levelling off (tending to a zero gradient) of the graph at higher values of X confirms that Y reaches its maximum value around X = 0.3. Even if X = 0.3 did not give the true maximum of Y this would not invalidate the final formula as it is based on Equation (24), not the true maximum. However, Figure 11 is inadequate for practical use. For example, a value of Y/Ymax = 0.6 could give a value of X anywhere between 0.015 and 0.07. Therefore the data were replotted for small Y/Ymax divided by ranges of Frgas/H (Figure 12). Data obtained with Frgas,th < 4 increased the uncertainty and were excluded.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
19
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Y/Y
max
X
0.5
1 - 1.1
1.4 - 1.5
1.8 - 2.2
2.3 - 2.6
2.8 - 3.1
3.3 - 3.6
4.2 - 4.6
5.6
7.1
1.5: Equation (28)
3.5: Equation (28)
5.5: Equation (28)
Frgas/H
Figure 12 Y/Ymax as a function of X, for Y/Ymax < 0.7 and Frgas,th > 4
The data in Figure 12 are fitted using the following form:
( )gas/
max1 exp
cFr HdYbX e
Y
−= − − (27)
This gives the following equation:
( )gas0.28 /0.75
max1 exp 35
Fr HYX e
Y
−= − − (28)
Provided a value of the pressure loss is known from measurements it is then possible to use Equations (22), (23) and (26) to determine values for Y and Ymax. A value of X can then be determined from Equation (28), which can be used to determine the wet-gas over-reading using Equations (17), (9) and (10). The gas mass flowrate can be determined from Equation (5) using the over-reading and the value for the wet-gas discharge coefficient, C, determined from Equation (20). The percentage error in the gas mass flowrate as a function of Y/Ymax is shown in Figure 13. The results are remarkably good, but as soon as Y/Ymax exceeds 0.7 the method becomes very inaccurate.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
20
-5
-4
-3
-2
-1
0
1
2
3
4
5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Y/Y max
% e
rro
r in
gas
mas
s fl
ow
rate
Figure 13 Errors in gas mass flowrate for Frgas,th > 4 using pressure loss and Equations (22), (26), (28), (17) and (20)
To use these equations with Venturi tubes other than those from which the equation was derived it is desirable to be cautious as the equation has not been validated against additional data, as was done for the equations for n and C. A reasonable estimate for the uncertainty in gas mass flowrate using pressure loss ratio measurements and Equations (22), (26), (28), (17) and (20) is
max
max
4% for 0.6
5% for 0.6 < < 0.7
Y
Y
Y
Y
≤
Additional limits to those for the use of Equations (17) and (20) are that the divergent angle shall be between 7° and 8°
max < 0.7
Y
Y
Frgas/H ≤ 5.5 ρ1,gas/ρliq ≤ 0.09 and Frgas,th > 4.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
21
7 Conclusion The 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. This research found that the over-reading was dependent on the Lockhart-Martinelli parameter, the gas-to-liquid density ratio, the gas densiometric Froude number, the diameter ratio and the liquid properties. 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
with an uncertainty of 3% for 0.15
2.5% for 0.15 < 0.3
X
X
≤ ≤
The ratio of pressure loss to the differential pressure can be used to determine the liquid content in the wet-gas stream using the equations derived in this paper. Under certain circumstances this can eliminate the need for a separate technique to determine the liquid content. The uncertainty in gas mass flowrate using the additional equations to determine the liquid content is
max
max
4% for 0.6
5% for 0.6 < < 0.7
Y
Y
Y
Y
≤
where Y is the increase in the pressure loss ratio due to wetness and Ymax is the maximum value of Y. 8 Acknowledgments The work described in this paper was carried out as part of the National Measurement System’s Engineering and Flow Programme, under the sponsorship of the United Kingdom Department for Business Innovation and Skills (formerly Department for Innovation, Universities and Skills). Krohne kindly agreed to allow their data to be used. Their help is gratefully acknowledged.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
22
We would like to thank CEESI for the release of their data which were used in this paper. This paper is published by permission of the Managing Director, TUV NEL. 9 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. of 2nd International 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. of 24th North Sea Flow Measurement Workshop, October 2006. [9] STEVEN, R. “Liquid Property and Diameter Effects on Venturi Meters Used With Wet Gas Flows”, International Fluid Flow Measurement Symposium, Mexico, May 2006. [10] STEVEN, R. “Horizontally Installed Differential Pressure Meter Wet Gas Flow Performance Review”, in Proc. of 24th North Sea Flow Measurement Workshop, October 2006. [11] AMERICAN SOCIETY OF MECHANICAL ENGINEERS, MFC, Report 19G, “Wet gas flowmetering guideline”, 2008.
27th International North Sea Flow Measurement Workshop 20th – 23rd October 2009
23
[12] STEVEN, R. Wet Gas Metering. PhD Thesis, Department of Mechanical Engineering, Strathclyde University, Glasgow, April 2001. [13] 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. [14] STEVEN, R., BRITTON, C. and STEWART, D. In CEESI Wet Gas JIP Data Release, December 2008. [15] HARRIS, D.M., and SHIRES, G. L. “Two-phase pressure drop in a Venturi”, in Proc. Two-phase Flow Through Orifices and Nozzles, Report of a Meeting at NEL, 29th November 1972, NEL Report No 549, Paper 4, East Kilbride, Glasgow: National Engineering Laboratory, 1973.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
1
Measurement of Water in a Wet Gas
Arnstein Wee – Multi Phase Meters Lex Scheers – Shell Global Solutions
1 INTRODUCTION Accurate measurement of liquids in a wetgas stream is a complex and challenging task. Acceptable performance in detection of extremely small liquid volumes requires a highly sensitive measurement system. Furthermore, measurement of water fractions is particularly important since it has a direct impact on scale, hydrate and corrosion management in long pipelines on the seabed. Water measurement is conversely, the most challenging one since water typically constitutes the smallest volume fraction. Water volume fractions may be as low as 0.001 % of the total volume in the pipe. In order to perform accurate measurement of water, the measurement principle must be repeatable over time and able to sense small variations in the water content. Furthermore, the measurement principle must be able to tolerate significant variations in the hydrocarbon PVT properties. Operationaly, regular measurement of PVT properties is both expensive and time consuming and low dependence on sampling is desirable for the overall success of the measurement system. The dominating configuration parameter for measurement of the phase that occupies the smallest volume fraction is the PVT properties of the dominating phase. Hence, in order to achieve reliable water measurements in wetgas, it is critical that the system be tolerant to changes in the gas PVT properties. An extensive operator driven development and subsequent qualification program has been executed by 10 oil companies in co-operation with MPM to find a workable solution to the challenge. Tomographic techniques for measurement of liquid content and methods for dealing with variations in the hydrocarbon PVT properties have been developed and tested. The measurements proofed to be very robust, even with significant changes in the PVT properties of the gas.
Picture of a 5”MPM Subsea meter tested at SwRI, San Antonio, Texas
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
2
The paper presents the technical principles, results from the test and qualification program that in particular focused on the water measurement in a wet gas environment. 1.1 Why is water measurement important In order to ensure a continuous production of hydrocarbons from remotely located subsea wells, management of water production is essential. Water in the production lines can cause scale and hydrates, which can block the pipes. In order to optimise the scale and hydrate inhibition, it is important that water production rates are accurately measured such that safe production can be achieved at the optimum production capacity. For many fields it is also important to know the salt content of produced water in order to prevent corrosion in the pipelines and production equipment. In addition, knowledge about salt content is also important from a reservoir management perspective. Discrimination between salt and fresh water can be used to identify the origin of produced water, either condensed water vapour or produced formation water. 1.2 Measurement of small liquid fractions
For wetgas applications, the challenge is clearly the accurate measurement of tiny liquid fractions in a gas dominated production stream. Even once the liquid volumes have been successfully measured, this tiny liquid fraction must then be split into water and hydrocarbons. Hence, the requirement for a metering system that is capable of extremely high resolution. In many real applications the makeup of a wetgas production stream could correspond to as much as 99.9 vol% gas, 0.05-0.1 vol% condensate and a water volume fraction of only 0.01-0.05 vol%. In such cases, variations in the properties of the dominating phase (gas) will usually correlate strongly to the measurement uncertainties as pointed out by H. van Maanen in [2]. It is therefore essential, as far as possible, that the metering system be insensitive to variations in the gas properties. This has been a particular area of focus in development of the MPM meter. Typical configuration parameters for commercially available wetgas flow meters are density, permittivity (dielectric constant), mass absorption coefficients and viscosity data for the three phases. If the determination of phase fractions is entirely based on the measurement of the average dielectric properties, or the average density of the phases, the result is often liquid measurement errors of several hundred percent. As an example, a density based fraction measurement and a typical wetgas case with an operating pressure of 150 bar, the measured mixture density may be 112.7 kg/m3. Assuming an input configuration gas density of 110 kg/m3 and condensate density of 650 kg/m3, the calculated GVF becomes 99.5 vol% i.e. 0.5% of the volume in the pipe is liquid. If on the other hand, the input configuration gas density was wrong by 5% such
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
3
that the true gas density was 104.5 kg/m3 instead of 110 kg/m3, then the calculated GVF becomes 98.5 vol%, which corresponds to a liquid fraction of 1.5 vol%. Hence, for the example above, an error in the input configuration gas density of 5% (which is not unusual) results in a measurement error in the liquid fraction (and correspondingly the liquid flow rate) of 200%. The measurement uncertainty of the liquid phases relative to uncertainties in the gas density increases exponentially as the gas fraction in the pipe increases. The same type of argument can be used for dielectric based measurements of liquid or water content when an average mean field approach is being used. This problem has been highlighted by Hans van Maanen at the 2008 North Sea Flow Measurement Workshop [2]. 1.3 PVT Configuration Data All multiphase flow meters need to be configured with the fluid properties of oil, water and gas. The fluid properties of oil and gas can easily be calculated based on the total hydrocarbon composition for the well using Equation of State programs such as Calsep PVTSim. Similarly the water properties can be obtained by sampling. The MPM meter is also capable of measuring the water salinity from which all required water parameters such as density, conductivity and mass absorption can be calculated. In most gas production applications the fluid properties may change significantly over time. If the meter is installed on a test header with many wells from different reservoirs, it is important that the meter is tolerant with respect to variations in the PVT configuration data since it is difficult, if not impossible, to maintain accurate PVT data over time for such installations. Many multiphase flow meters are also used on comingled well streams from subsea tie backs or from wells producing from multiple reservoir zones, and under such circumstances significant variation in the PVT properties can easily occur. In order to obtain accurate measurement over time it is therefore important that a wetgas (and multiphase) meter is able to cope with significant variations in the PVT configuration data. Alternatively, sampling systems and operating procedures have to be put in place for regularly sampling of the well streams. If a frequent update of PVT data is required, the life cost of obtaining PVT data can easily exceed the cost of the wetgas/multiphase flow meter. In addition sampling may also introduce significant HSE issues (high pressure, H2S), complicated logistics and is clearly highly undesirable in subsea installations. 1.4 Uncertainty in PVT Data PVT data is required for configuration of all wetgas and multiphase meters. Reliable PVT data is often hard to obtain and thus errors should be expected in such configuration data. There are two issues which then must be addressed. The first issue is the real uncertainty in the PVT data originating from the sampling and characterisation process.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
4
The second issue is the variations in the PVT data that might occur at a later stage due to changes in the reservoir, wellbores effects, or variations in the instantaneous contributions from individuals sections of a multi-reservoir completion. In a real field application there can be significant errors in the PVT data, which may originate from several sources. Some of the error sources may be due to lack of representative samples of the fluids and errors in the characterisation process. Commonly used Equation of State models are also known to contain errors, which typical give a bias in the calculation of the PVT data. Tests based on gas densities calculated based on Equations of State have shown that a positive bias is quite common at higher pressures and may typical be in the range 1-3% for the gas density. All PVT models for calculating single phase properties at actual conditions rely on input of temperature and pressure. Temperature and pressure inputs may also contain a bias, which introduces shift in the configuration data for the meter. Finally, the fluids of the well may change during the period of the test or installation, further introducing errors in the PVT configuration data. In a real field application, a 2-5% uncertainty in the PVT data would be considered as normal. Even a 10% change (error) in the gas density and gas permittivity can be expected for comingled well applications where the gas composition can change significantly over time. In order to provide reliable measurement of water production of a wetgas well, the measurement system needs to be able to handle uncertainties of 5-10% in the configuration fluid properties like gas density and gas permittivity and still maintain an accurate and repeatable water measurement. 1.5 Flow Effects At high GVF’s, the liquid fraction is marginal. For gamma-ray based systems, the natural variation in the gamma-ray absorption signal, where the measurement is already at the limit of the GVF range, may cause the GVF to instanteously exceed the 100% limit, itroducing a positive bias in the liquid measurement. Averaging can to some extent overcome this; however since gamma ray absorption is a non-linear phenomena, this introduces systemic errors in slug or wave flow conditions. In order to achieve accurate measurements at high GVF, the measurement needs to be able to provide reliable measurements very close to the 100% GVF limit without actually crossing it. Other common flow effects are liquid recirculation introduced by mixing devices. Mixing devices creates turbulence in the gas stream, which can cause liquid re-circulation and unpredictable local holdup of liquids. The amount of liquid re-circulation is also highly dependent on the velocity and viscosity of the fluids and difficult to predict. Liquid re-circulation will have a particularly large influence on the water fraction measurement since it is the smallest fraction within the pipe and hence most vulnerable to unpredictable noise/uncertainty in the measurement signals. For reliable and accurate water measurement in a wetgas, mixing devices for homogenizing the flow should be avoided.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
5
1.6 Measurement of water production in a wet gas The current commercially available wetgas flowmeters capable of measuring water production are either based on permittivity measurements or dual-energy gamma absorption. Some meters are able to realise a full three-phase measurement in wetgas i.e. all three phases present in the wetgas (water, condensate and gas) are invidually metered. Other meters are based on a two-phase measurement and rely on additional input parameters such as a PVT predicted GOR. Some meters use mixers in order to homogenize the flow whereby the mixer is integrated with a common flow measurement device such as a V-cone. Other meters are based on measurement of the natural flow conditions in the pipe. 2 The MPM 3D Broadband™ technology The MPM HighPerformanceFlowmeter uses a combination of a Venturi flow meter, a gamma-ray detector, a multi-dimensional, multi-frequency dielectric measurement system [5] and advanced flow models [1, 4], which are combined to a multi-modal parametric tomographic measurement system. The Venturi is used to create a stable radially symmetric flow condition in the 3D Broadband™ section downstream the Venturi, which would be the natural flow condition if the pipe were infinitely long. These flow conditions are ideal when using tomographic inversion techniques. The technology is marketed as 3D Broadband™ and is used to establish a three-dimensional picture of what is flowing inside the pipe. The basis for the technology is often referred to as ‘process tomography’ which has many parallels to the type of tomography used in medical applications. In the oilfield, the challenges are however far different than in a hospital. Firstly, the meter is measuring fluids and gases under high temperature and high pressure. Secondly, the multiphase mixture can be mowing at velocities of more than 30 meters per second inside the pipe, and the volumes of gas, water and oil are not in thermodynamic equilibrium and do not have constant phase fractions over the cross-sectional area of the pipe. The 3D Broadband™ system is a high-speed electro-magnetic (EM) wave based technique for measuring the water liquid ration (WLR), the composition and the liquid/gas distribution within the pipe cross section. By combining this information with the measurements from the Figure 1 -
3D Broadband™ tomography based meter
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
6
Venturi, accurate flow rates of oil, water and gas can be calculated. The MPM meter is extremely fast. Averaging of measured raw data is limited, to avoid errors due to non-linearity in the flow. The result is a measurement with an unparalleled performance in multiphase and wetgas flowing regimes. With its dual mode functionality, which means that both multiphase and wetgas applications are addressed with the same hardware, and its capability to measure water salinity, the MPM meter bridges the existing measurement gap between multiphase and wetgas meters. In wetgas mode, MPM Meter can either operate in three-phase mode or in two-phase mode. In three-phase mode, the meter measures all the fractions of the flow (oil, water and gas). In two-phase mode, the MPM Meter needs the GOR as an additional configuration parameter which is typically calculated based on the well composition. 2.1 Tomography based fraction and rate measurements for wet gas flow The MPM meter performs measurements on the natural flow conditions occurring in a vertical pipe. It uses the swirl created by the Venturi to obtain natural vertical flow conditions in the tomographic measurement section downstream the Venturi. Established correlations for calculating the flow rate of liquid and gas at wetgas conditions assume that the GVF is known. The GVF in this context means the gas volume flow rate divided by the total volume flow rate. The GVF is a parameter which is extremely challenging to measure since the liquid is distributed partly as liquid film along the wall and partly as droplets in gas phase in the centre of the pipe. Measurement of the GVF is further complicated by the fact that the velocity of the liquid film is significantly lower than the velocity of the average droplet (can be more than 10 times lower). Even at such large differences between the film and droplet velocity, most of the liquid flowing in the pipe may originate from the film since it typically occupies a larger portion of the cross sectional volume of the pipe. Hence, in order to obtain a correct GVF measurement, the film thickness, film velocity, droplet volume, droplet velocity and gas velocity need to be known. The use of a tomographic measurement principle is one solution to this problem. In the MPM meter, the cross section of the pipe is parameterised as liquid film with thickness m along the pipe wall and liquid droplets with a diameter Dd and number of liquid droplets Na. The liquid fraction in the cross section of the pipe then becomes the area of the liquid film divided by the total flow cross section plus the relative volume of liquid droplets present in a certain flow volume element which can be calculated from m, Dd and Na. By multiplying the liquid droplet fraction in the cross section of the pipe with the velocity of the liquid droplets, the liquid droplet flow rate is obtained. Similarly, by multiplying the liquid film fraction in the cross section of the pipe with the velocity of the liquid film, the liquid film flow rate can be obtained. The total liquid flow rate is the sum of the liquid droplet flow rate and liquid film flow rate.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
7
The gas flow rate is obtained by multiplying the gas fraction in the cross section of the pipe with the velocity of the gas. The flow pattern is illustrated in [3] and shown in Figure 2 below for a Venturi with vertical downward flow.
Figure 2 - Flow pattern of a wet gas stream in a Venturi
The various 3D Broadband™ measurements have different sensitivity to the film and droplets in the cross section of the pipe and are also influenced by the amount of water in the liquid phase. This is used together with physical models for the fluid distribution within the cross section of the pipe (i.e. values of m, Dd and Na), the physical properties of oil, water and gas such as surface tension, viscosity and density, measured WLR, measured Venturi dP, and energy conservation equations to calculate the flow rate of liquid and gas. The outcome of these calculations gives the velocity of the liquid film, velocity of the liquid droplets and the gas velocity, hence the GVF can be easily inferred. The GVF, together with the physical properties of liquid and gas are then used together with Venturi-correlations in order to calculate the liquid and gas flow rates. The result from the Venturi correlations is then compared with the outcome of the energy conservation equations for calculating the GVF. If there is a difference between these two approaches then the calculations are repeated in an iterative fashion until the two GVF measurements converge. As a result, the flow rates of oil, water and gas are derived. In addition, the following parameters are obtained:
• Liquid film thickness • Droplet diameter, • Number of droplets • Velocity of liquid film • Velocity of liquid droplets • Velocity of gas.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
8
Using this methodology, the GVF and flow rates of liquid and gas can be determined without use of any mixing device to homogenise the flow or the use of empirical slip correlations. This makes it straight forward to scale the measurement principle over a broad range of sensor dimensions, fluids and operating pressures. This also eliminates measurement errors due to liquid recirculation which is often seen in wetgas meters that use mixing devices. At ultra-high GVF the Droplet Count® functionality is an add-on feature that contributes to significantly improving the measurement resolution in a regime where liquid volumes are infinitesimally small. The Droplet Count® was commercially released in 2009 but has been in field operation in MPM meters since early 2008. By using Droplet Count® , the MPM meter can make precise measurements of minute liquid volumes in a GVF range where and conventional technologies are no longer able to make true three-phase measurements. The method is furthermore highly tolerant towards changes in fluid PVT properties (i.e. gas density and water properties). This is achieved by a patented (pending) methodology with a far higher resolution on mixture density as compared to gamma based density measurements, and for which the liquid metering accuracy actually increases with increasing GVF. Liquid droplets flowing with the gas stream in a pipe cause statistical variations in electromagnetic measurement signals. The statistical variation is primarily a function of the liquid droplet size, the number of droplets and the permittivity of the droplets. Hence the PVT properties of the gas phase (i.e. density and permittivity) are not a part of the measurement loop. This makes the GVF measurement based on droplet counting insensitive to uncertainty and changes in the gas PVT properties.
The method uses the 3D BroadbandTM ultra-fast electromagnetic measurements scaled to the pipe diameter and the permittivity of the material within the pipe. The measurement field is uniformly distributed within the cross-section of the pipe with low sensitivity to the liquid film along the pipe wall, and high sensitivity towards the droplet flow at the centre of the pipe. The measurements are extremely sensitive to small variations in permittivity caused by droplets flowing in a pipe.
Measurement Uncertainty vs GVFThree Phase WetGas Meter
97% 98% 99% 100% [GVF]
Mea
sure
men
t Unc
erta
inty
3D BroadBand Based GVF DropletCount
Gamma Based GVF Measurement
97% 98% 99% 100% [GVF]
Mea
sure
men
t Unc
erta
inty
3D BroadBand Based GVF DropletCount
Gamma Based GVF Measurement For ultra high GVF aplications,
MPM’s Droplet Count (patent pending) method provides extremely sensitive and accurate results
Figure 3 - Uncertainty of Gamma Ray GVF measurement and Droplet Count GVF measurement
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
9
2.2 Combined Wetgas and Multiphase flow meter The MPM meter is a combined multiphase and wetgas flow meter. The meter can be software configured to operate using either its multiphase or wetgas models. These are often referred to as multiphase and wetgas modes. In addition, the Droplet Count® further enhances the range of the MPM meter models to fully cover the 0-100% GVF range.
MPM meter Dual Mode and Droplet Count
Operating range
99,9%
99%
90%
50%G
as Void Fraction (G
VF)
Gas
–flo
w r
ate
Oil – flow rate
* * * * * * * **
* *
*
*** *
***
***
***
** *
* *
* * * * * * * **
* *
*
*** *
***
***
***
** *
* *
Multiphase mode
Wetgas mode
Droplet Count
The MPM meter automatically switches between modes in a fraction of a second.
Covers FULL operating range, and all flow cases
Min
Max
Gas - flow rate
Liq
uid
-flo
w ra
te
Figure 4 : Dual Multiphase and Wetgas mode
The standard MPM meter is delivered either as a multiphase or a wetgas meter. The hardware parts are, however, identical and the difference between the two meter versions is simply the software. 2.3 Water Salinity In wetgas applications, the salinity measurement method implemented in the MPM meter is split into two stages: • First, it is determined whether salt
is present in the stream or not (by a so-called salt water index)
• Second, if salt is present, then the salinity is measured quantitatively.
The reason for this two-step approach is that some measurement plans of the
Multiphase
Multiphase
Wetgas
Effect ofadding 0.1% water
Changingconductivity
Effect ofadding 0.1% water
Changingconductivity
Figure 5 - Water conductivity measurement principle for wet gas applications
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
10
3D Broadband™ are very sensitive to presence of salt, whereas others are useful to determine quantitatively the degree of saltiness. This feature is also used as a type of quality assurance. Another reason for splitting the salinity measurement into two stages is to make the formation water break-through measurement robust with respect to discrepancies in the configuration data such as the dielectric properties of the gas phase. The water salinity measurement is related to the water fraction measurement such that any error in the water fraction measurement will relate to an error in the measured water salinity. As an example, the dielectric properties of the gas is a configuration parameter for the meter and a discrepancy in the dielectric constant for gas may cause a measurement error on the water fraction and hence the water salinity. The salt water index on the other hand, is virtually independent of the gas properties and as a consequence, reliable detection of salt (or formation water) can be achieved irrespective of significant discrepancies in the dielectric constant of gas. The curves in Figure 5 illustrate the basic measurement principles. Using the 3D Broadband™, many cross-sectional planes are measured and analysed simultaneously to determine the liquid and particularly the water content. Some of the measurements are based on frequency sweeps, which are performed in each direction with a step in the phase of the electromagnetic waves. The frequency location of a differential phase shift between two receiving antennas is related to the water fraction of the wetgas and the slope of the phase shift vs. frequency is related to the conductivity of the water fraction. An increase in water conductivity causes a decrease in the slope of the curve. The measurement is based on a differential measurement within the pipe. Hence, any discrepancies in the cables, antennas and electronics are cancelled out. The water salinity is then obtained from the measured conductivity and measured temperature. This measurement can be performed in all the 27 measurement directions used by the 3D BroadbandTM system. In order to maintain a high speed measurement principle, 10-15 of the measurements are performed and 5-10 of the measurements are used. This is considered an appropriate trade-off between speed and number of measurements.
Figure 7 - Raw measurements of salinity factors for different measurement directions of the 3D Broadband sensor
Normalised (inverse) S-Factor MeasurementTest at SwRI - November 2007
0
0.5
1
1.5
2
2.5
0 10 20 30 40 50
Test Point ID
1/(N
orm
alis
ed S
-Fac
tor)
0
0.5
1
1.5
2
2.5
%S
alin
ity
at V
ega
Co
nd
itio
ns
Normalised SFactor-D1 (inv)
Normalised SFactor-D14 (inv)
Normalised SFactor-D7 (inv)
Reference Water Conductivity
Fresh Water Salt Water
Fresh water boundaries
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
11
For each measurement direction a so-called normalised S-factor is calculated. The S-factor is a number which is related to the slope of the frequency sweeps. It is defined to be one for fresh water and deviates from one for increasing salinities. Typical, the normalised factor increases when the salinity increases. Figure 7 shows the inverse normalized S-factor for a few measurement directions. The measurements were logged at South West Research Institute (SwRI) in San Antonio, Texas, October 2007 where a 5” subsea version of the MPM meter was tested in a blind test sponsored by 8 international oil companies. The chart shows all the test points at SwRI with reference to the test ID on the x-axis, with a total of 52 test points. Each test point was logged for a period of 2-4 minutes (to maintain stable conditions in the flow rig). The first 21 test points were performed using fresh water, with maximum variations in the other parameters (95-99.5% GVF and 0-0.2% water fraction). The subsequent test points, from 22 to 52, were performed using saline water. As can be seen in the chart, the water salinity was increased in steps from 0.4% to 0.8% and finally to 1.9%. The saline water test points were run with similar variations in GVF, water fraction and velocity. As the graph demonstrates, the normalized S-factors remain within the fresh water boundaries for all test points with fresh water, irrespective of the GVF, water fraction or velocity. The inverse of the S-factor is shown to better visualise the connection between an increase in the water salinity versus a change in the normalised S-factor. Once salt water is used, the normalized S-factor for measurement direction D1 moves outside of the fresh water boundaries. As the salinity increases, more and more S-factor measurements fall outside of the fresh water boundaries. A clear indication of salt water means that one of the S-factor measurements is outside of the fresh water boundaries. The greater the number of measurements that fall outside of the fresh water boundaries, the higher the confidence level for detection of saline water. At the two highest salinities, almost all the normalized S-factor measurements are outside the fresh water boundaries.
4.1 Salt water Index and Early Detection of Salt Water Production A salt water index is a number between 0 and 100% and is a function of the salt water detection from all measurement directions in the 3D BroadbandTM system. In principle:
If the salt index is 0 then the water is certainly fresh whereas if it is 100 then the water certainly contains salt. Salt water is detected if the salt water index is measured to be above a given threshold value. Hence, the detection of salt water is independent of the water fraction measurement and of the PVT properties of the gas.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
12
Figure 8A illustrates how the salt water index is used for wetgas conditions (GVF of 99.5%).The example shows results for measurements at a WLR of 10% in fresh water performed on a 3” Topside meter installed at K-Lab in November 2006. Here, the index is a very small number and far below the threshold. Figure 8B illustrates a measurement at a WLR of 5%, and a water salinity of 3%. As salt is present, the index jumps to 100, and subsequently the salinity is calculated. The salinity is correctly measured, even at short time intervals. When some averaging is applied, the salinity value will be stable and change only if a real salinity change appears. Note that at GVF of 99.5% and a WLR of 5%, the absolute water fraction of the flow is only 0.025%.
4.2 Quantitative Measurement of Water Salinity Once the salt water index, exceeds its threshold value, corresponding to a reliable detection of salt, functionality for actual quantitative measurement of the water conductivity and salinity is started. This routine is based on the measured water fraction and on the measured S-factors. In this step, accurate water fraction measurements are required as well as information about the gas PVT properties. Measurement error is minimised through an iterative process.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.50
0.2
0.4
0.6
0.8
1
Minutes
%
Water Salinity
0 0.5 1 1.5 2 2.5 3 3.5 4 4.50
20
40
60
80
100
Minutes
%
Salt Water Index
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.50
2
4
6
Minutes
%
Water Salinity
0 0.5 1 1.5 2 2.5 3 3.5 4 4.50
20
40
60
80
100
Minutes
%
Salt Water Index
Figure 8A - Salt Water Index in a Wetgas case with fresh water
Figure 8B - Salt Water Index in a Wetgas case with saline water
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
13
3 SENSITIVITY TO GAS CONFIGURATION PARAMETERS As previously mentioned the gas properties are the most important configuration parameters for a wetgas flow. Since gas is the dominating phase and water is (normally) the smallest phase in the pipe for a wetgas, any error in the gas configuration PVT properties have the largest influence on the water measurement. The MPM meter can operate as either two-phase or a three-phase wetgas meter. In two-phase mode, the PVT predicted GOR is needed as an additional input configuration parameter (similar to other two-phase meters). The GOR input is not needed nor used in three-phase mode. A 3” MPM Meter was tested at K-Lab in 2006 with a second 5” subsea version tested at SwRI (South West Research Institute) in 2007 as a part of an operator driven JIP program involving 9 international oil companies. Data from the tests at K-Lab and SwRI (approx 200 test points) have been used to generate the plots below by re-simulation the tests with a 5% error in the input configuration gas density. The sensitivity plots are for gas flow rate (volume), hydrocarbon mass flow rate and water fraction in both two-phase and three-phase mode (without Droplet Count). When the density is known, the permittivity of the gas can be calculated using the Clausius Mosotti equation which shows that the permittivity for the gas is proportional to the gas density [2]. Hence a 5% error in the gas density will also introduce an error in the gas permittivity. The sensitivity plots are generated at 120 bar based on re-simulation of the raw datafiles logged at K-Lab and SwRI. The error in the measurement is plotted as a contour plot with GVF on the x-axis and WLR on the Y-axis. From figure 9A and 9B it is seen that the error on the gas flow rate due to a 5% error in the gas density is quite small in both two-phase and three-phase mode.
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : % Effect on Gas Measurement for 5% error in Gas Density
% R
elat
ive
Err
or in
Gas
Mea
sure
men
t
3-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : % Effect on Gas Measurement for 5% error in Gas Density
% R
elat
ive
Err
or in
Gas
Mea
sure
men
t
2-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
Figure 9B – 2-Phase Mode Effect on gas rate for 5% error in gas density
Figure 9A – 3-Phase Mode Effect on gas rate for 5% error in gas density
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
14
The hydrocarbon flow rate is slightly more influenced by the error as compared to the gas flow rate as can be seen in figure 10A and 10B below. It is also worth noting that the hydrocarbon mass flow rate is less influenced by error in the gas density in three-phase mode compared to two-phase mode.
Figure 11A and 11B below show the effects on the measured water fraction for a 5% error in the gas density for both two-phase and three-phase mode. Clearly the three-phase mode is more robust towards errors in the gas density. In particularly the zero point of the water fraction measurement is far less influenced by errors in the gas density in three-phase mode operation as compared to two-phase mode. As seen from figure 11B, the zero point of the water fraction measurement is almost unaffected by a 5% change in the gas density in three-phase mode whereas an offset is introduced on the two-phase mode measurement as shown in figure 11A. A stable zero point in the water fraction measurement is essential for full confidence in any indications of formation water break through on a previously water-dry well.
Figure 11A – 3-Phase Mode Effect on gas rate for 5% error in gas density
Figure 11B – 2-Phase Mode Effect on gas rate for 5% error in gas density
Figure 10A – 3-Phase Mode Effect on gas rate for 5% error in gas density
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : % Effect on Hydrocarbon Mass Meas. for 5% error in Gas Density
% R
elat
ive
Err
or in
Hyd
roca
rbon
Mas
s M
easu
rem
ent
2-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : % Effect on Hydrocarbon Mass Meas. for 5% error in Gas Density
% R
elat
ive
Err
or in
Hyd
roca
rbon
Mas
s M
easu
rem
ent
3-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : Effect on Water Fraction [%abs] for 5% error in Gas Density
% A
bs E
rror
in W
ater
Fra
ctio
n M
easu
rem
ent
3-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : Effect on Water Fraction [%abs] for 5% error in Gas Density
% A
bs E
rror
in W
ater
Fra
ctio
n M
easu
rem
ent
2-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Figure 10B – 2-Phase Mode Effect on gas rate for 5% error in gas density
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
15
In the two-phase mode of operation, the PVT predicted GOR is required as an additional configuration parameter. The GOR is typically calculated using a compositional study for the well fluids. Figure 12A and 12B below show the influence on the water fraction measurement and hydrocarbon mass flow measurement for a 10% error in the GOR input. From figure 12A it is seen that the error on the water fraction measurement increases as the GVF falls. An error in the GOR also influences the zero point of the measurement, particularly for lower GVFs. The influence on the hydrocarbon mass flow rate measurement is less pronounced.
4 IN-SITU MEASUREMENT OF GAS PROPERTIES Since June 2008 a dedicated JIP project, involving 9 international oil companies, have targetted towards developing methods for in-situ verification of measurement values. One of the key goals for the project has been to develop and qualify a method for in-situ measurement of the gas density and permittivity in order to eliminate the distorting effects of uncertainty in the configuration data for gas. A patented (pending) method for in-situ measurement of gas properties has been developed and implemented in the MPM meter. The method uses the Droplet Count®
function to detect short periods of time where pure gas flows through the measurement section of the meter. Alternatively the meter can be bypassed and gas filled during a scheduled shut-in of the well or during the passage of long gas slugs. Figure 13 below shows the measured GVF and the liquid detection signal, called the Liquid Index, for a gas filled period in the measurement section. The yellow line is the threshold value for gas detection.
Figure 12B – 2-Phase Mode Effect on Hydrocarbon mass flow rate for 10% in GOR input
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : Effect on Water Fraction [%abs] for 10% error in GOR Input
% A
bs E
rror
in W
ater
Fra
ctio
n M
easu
rem
ent
2-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
GVF[%]
WLR
[%]
SwRI+K-Lab 120 bar : % Effect on Hydrocarbon Mass Meas. for 10% error in GOR Input
% R
elat
ive
Err
or in
Hyd
roca
rbon
Mas
s M
easu
rem
ent
2-PHASE MODE90 92 94 96 98 100
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
Figure 12A – 2-Phase Mode Effect on water fraction for 10% error in GOR input
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
16
Gas TresholdReliable detection of insitu-gas
period at the end of the test
When pure gas is present, the permittivity and the density of the gas are measured using the 3D Broadband section and the gamma densitometer. The Droplet Count® is so sensitive to droplets that it immediately detects when condensation of liquid starts to occur, due to falling temperature, such that the in-situ gas measurement can be halted in due time. Since the 3D Broadband measurement performs measurement of permittivity at multiple frequency and on multiple measurement planes, many different measurements of the gas permittivity and density can be made. These all should give the same result and thus the gas in-situ gas measurement has a built-in quality verification function of the 3D BroadBand measurements. Such measurements can also be usd to verify the integrity of the 3D Broadband sensors during flow. The in-situ measurement can either be used to calculate correction factors to the input configuration gas density and gas permittivity, or to adjust the composition of the well fluid and generate new look-up tables using a sub-service based on the Calsep PVT Sim routines. Two methods for use of the in-situ gas measurement have so far been implemented: a manual procedure and a method based on automatic update. The automatic method is well suited for applications where frequent variations in the gas properties are expected. In the manual version, an in-situ report is then generated where the in-situ measurements are documented together with a calculation of the effect any changes in the gas configuration data may have on historical measurements. A recommendation for potential
Figure 13– Example of detection of gas in the meter
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
17
corrective action is also added to the report before it is issued to the operator for final approval. If the operator approves the corrective actions, the in-situ measured corrections to the gas density and permittivity are implemented in the MPM meter and the date and time of the implementation is noted in the in-situ report. This manual procedure ensures full tractability of any changes performed on the gas configuration data and the procedure is particularly suited for applications where the MPM meter is used for fiscal applications. This procedure is typically used as a part of the commissioning of the meter. Most MPM meters are pre-configured with the field PVT data prior to delivery as a part of the FAT procedure. The MPM meter is then fit for service immediately at start up of the wells. Following successful commissioning of the field and the individual wells, any in-situ measurements made can then be inspected to validate the pre-configured PVT data in the meter. Evaluation of the in-situ gas measurements may also be performed on a regular basis as part of the metering quality assurance plan. Using pre-agreed acceptance limits for the in-situ gas measurements allows the operator to efficiently process the in-situ reports. This procedure also ensures that the operator has full documentation of the validity of the configuration data for the meter. Documentation of the integrity of the measurements from the meter is also obtained by inspecting the historical trend of the multi-frequency and multi-directional measurements from the 3D Broadband sensor in gas. In wetgas, the MPM meter incorporates three different methods for measurement of the fractions and flow rates of the wetgas, namely
1) two-phase mode with GOR Input 2) three-phase mode based on the gamma densitometer 3) three-phase mode based on Droplet Count®
As outlined above, these three methods behave differently when errors are introduced in the input configuration data for gas density and permittivity. By comparing the measurement results from these three different operation modes it is also possible to verify the quality of the input PVT configuration data for gas. This can also be trended as a PVT quality index where 100% means good agreement between the different operation modes and 0% means poor agreement due to errors in PVT input configuration data. 5 TEST RESULTS The MPM Meter has been through an extensive test and qualification program and more than 10 meters are now in continuous operation. This section summarizes some of the test results obtained in field.
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
18
5.1 SwRI (2007) and K-Lab (2006) A 5” subsea meter was tested in a blind test at SwRI in 2007. A year earlier, a 3” topside meter was tested at K-lab. The plots below give an overview of some of the test results in both two-phase and three-phase mode operation. The measurements are performed without the Droplet Count function since this was not commercially released at the time of the test. Figures 14A and 14B shows the difference between the measured and reference gas flow rates in three-phase and two-phase mode operation. From the figures below it is seen that the performance in two-phase and three-phase mode are more or less identical. Figures 15A and 15B shows the difference between the measured hydrocarbon mass flow rate by the reference measurement and the MPM meter.
SwRI vs K-Lab : 3 -Phase WetGas ModeGas Flow Rate Deviation vs. GVF
-10
-8
-6
-4
-2
0
2
4
6
8
10
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]G
as F
low
Rat
e D
evia
tion
[%]
SwRI (5")
K-Lab (3")
SwRI vs. K-Lab : 3-Phase WetGas ModeHydrocarbon Mass Flow Rate Deviation vs. GVF
-10
-8
-6
-4
-2
0
2
4
6
8
10
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]
Hyd
roca
rbon
mas
s flo
w r
ate
devi
atio
n [%
]
K-Lab (3")
SwRI (5")
SwRI vs. K-Lab : 2-Phase WetGas ModeHydrocarbon Mass Flow Rate Deviation vs. GVF
-10
-8
-6
-4
-2
0
2
4
6
8
10
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]
Hyd
roca
rbo
n m
ass
flow
rat
e de
viat
ion
[%]
K-Lab (3")
SwRI (5")
Figure 14B– Difference for Gas flow rate in 2-phase mode
Figure 15A– Difference for hydrocarbon mass flow rate in 3-phase mode
Figure 14A– Difference for Gas flow rate in three-phase Mode
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
19
This measurement follows more or less the same trend as the gas flow rate measurement. Slightly higher spread in the measurement is obtained in 3-phase mode, particularly at SwRI. The spread is partly due to very short test durations at SwRI where many of the points had duration of 2-3 minutes in order to prevent liquid carry over and gas carry under in the test separator. At SwRI many of the test points also were performed at dPs in the range 15-25 mBar giving a larger spread in the data. The graphs in figure 16A and 16B show the difference between the measured water fraction and the reference water fraction as a function of GVF. For the test at SwRI, the average deviation in 3-phase mode operation is -0.008% and in 2-phase mode operation the average deviation -0.007% abs.
SwRI vs K-Lab : 2-Phase WetGas ModeGas Flow Rate Deviation vs. GVF
-10
-8
-6
-4
-2
0
2
4
6
8
10
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]G
as F
low
Rat
e D
evia
tion
[%]
SwRI (5")
K-Lab (3")
SwRI vs. K-Lab : 2-Phase WetGas ModeWater Fraction Deviation vs. GVF
-0.50%
-0.40%
-0.30%
-0.20%
-0.10%
0.00%
0.10%
0.20%
0.30%
0.40%
0.50%
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]
Wat
er F
ract
ion
Dev
iati
on [%
abs]
SwRI (5")
K-Lab (3")
SwRI vs. K-Lab : 3-Phase WetGas ModeWater Fraction Deviation vs. GVF
-0.50%
-0.40%
-0.30%
-0.20%
-0.10%
0.00%
0.10%
0.20%
0.30%
0.40%
0.50%
90.00 91.00 92.00 93.00 94.00 95.00 96.00 97.00 98.00 99.00 100.00
GVF [%]
Wat
er F
ract
ion
Dev
iati
on [%
abs]
SwRI (5")
K-Lab (3")
Figure 15B– Difference for hydrocarbon mass flow rate in 2-phase mode
Figure 16B– Difference for measured water fraction in 2-phase mode
Figure 16A– Difference for measured water fraction in 3-phase mode
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
20
5.2 Test with Droplet Count® Function The first version of the Droplet Count® function was tested out in 2007 based on data from SwRI and the result is shown in figure 17. From the graph it is seen that the GVF measurement performed by the droplet counting function agrees very strongly with the reference GVF. The time scale is in seconds. Prior to using it in a field location, the function had been tested by re-simulating historical measurements based on raw data captured in previous field test. The chart in figure 18 to the right shows the measured water fraction from a K-Lab test in 2006 when the tests have been re-simulated with the Droplet Count function. From this graph it is seen that the difference between the measured and reference water fraction is well within a margin of ± 0.02% abs. 5.2 Test of 10” MPM Meter at K-Lab – 2008 & 2009 A 10” MPM meter has been in continuous operation at K-Lab since April 2008. The meter is equipped with both the Droplet Count® function and in-situ measurement of gas density and permittivity. Since November 2008, the meter has been configured to automatically correct the input configuration data for gas density and permittivity based on in-situ measured density and permittivity for gas. At K-Lab the composition of the gas changes frequently due to loading and re-loading of both gas and condensate in the flow rig. Frequent pressure changes caused by adding and flashing gas also produced significant changes to the
0 500 1000 1500 2000 2500 3000 350099
99.1
99.2
99.3
99.4
99.5
99.6
99.7
99.8
99.9
100Test of Three Phase WetGas Meter
GV
F [
%]
Time
Reference GVF
Measured GVF
Water Fraction Deviation vs. GVF(Resimulation of test at K-Lab in 2006 with Droplet Count)
-0.10%
-0.08%
-0.06%
-0.04%
-0.02%
0.00%
0.02%
0.04%
0.06%
0.08%
0.10%
99.00 99.10 99.20 99.30 99.40 99.50 99.60 99.70 99.80 99.90 100.00
GVF [%]
Wat
er F
ract
ion
Dev
iatio
n [
%ab
s]
WLR > 10
WLR < 10
Figure 17– Droplet Count Measurement
Picture of 10” MPM Meter at K-lab
Figure 18 : Resimulation of test results at K-Lab in 2006 with Droplet Count
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
21
configuration data. The density of the gas may easily change by 5-7 % due to compositional changes of the gas. Without the in-situ measured correction factors to the input configuration data, the configuration parameters for gas would then be expected to incorporate a 5-7 % error. Since November 2008 the original input PVT configuration of the MPM meter at K-lab has remained unchanged and the in-situ gas measurement has automatically adopted the PVT configuration data to the frequent changes in gas properties. StatoilHydro is continuously logging the data from the MPM Meter, and the graph in figure 19 below shows a test of the GVF measurement from the MPM meter performed in May 2009 and presented by StatoilHydro at the MPM user forum in June 2009. As seen, the GVF measurements obtained with a 10” MPM meter at K-Lab in 2009 are markedly similar to the measurements obtained with a 5” subsea meter at SwRI in 2007 (ref figure 17).
Figures 20A and 20B below show the result of a sensitivity test for the liquid and water measurement performed in November 2008. From figure 20A it is seen that the sensitivity of the GVF measurement is better than 0.002% abs. Similarly, the zero point
Figure 19– Test of Droplet Count GVF Measurement at K-Lab
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
22
of the water fraction measurement is approximately 0.001 % abs and a change in the water fraction of only 0.0018% is measured with high precision.
Figures 21A and 21B below show the liquid detection signal and GVF measurement for a short shut down period. The green and blue line of figure 21A is the liquid detection signal and the red line is the pure gas threshold value. In connection with the shut down, there is a short period with pure gas in the meter where an in-situ measurement is performed. After approximately 20 minutes condensation of liquid start to occur and the liquid raw signal moves above the red liquid detection threshold – thus halting the in-situe measurement.
0 500 1000 1500
99.9
99.92
99.94
99.96
99.98
100
100.02
Time [Minutes]
GV
F [
%]
GVF : K-Lab 12. Nov 2008
MPM
K-Lab
0 500 1000 1500
0.005
0.01
0.015
0.02
0.025
Time [Minutes]
Liquid Raw Signal : K-Lab 12. Nov 2008
D7-Measured
D9-MeasuredD7-Gas
D9-Gas
Figure 20B – Water Fraction measurement with Droplet Count
Figure 20A– Sensitivity test of Droplet Count GVF Measurement
Figure 21A–In-situ detection of pure gas
Figure 21B–GVF during in-situ measurement
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
23
Figure 22 below shows the measured water fraction before and after the in-situ gas measurement. The reference water fraction (red line) is approximately 0.002 % abs both before and after the shut down. As seen from the graph of the water fraction measurement, there is a negative bias on the measured water fraction measurement from the MPM meter (blue line) prior to the in-situ measurement is performed. This is due to the error in the input configuration data for gas causing a negative bias on the zero point for the water measurement. After the gas permittivity and density have been automatically corrected with the in-situ measurement, the negative bias is removed and the measurement from the MPM meter (blue line) follows the reference measurement of 0.002% abs water (red line).
0 500 1000 15000
0.005
0.01
0.015
0.02
0.025
0.03
Time [Minutes]
Wat
er F
ract
ion
[%]
Water Fraction : K-Lab 12. Nov 2008
MPM
K-Lab
Insitu measurement of gas densityand dielectric constant performed
Improved water fractionmeasurement after insitu gas properties has been measured
Negativ water fractionmeasured prior to in-situgas measurement
In-situ measurementof gas propererties removesnegative bias on water fraction measurement
0 500 1000 1500
0.005
0.01
0.015
0.02
0.025
Time [Minutes]
Liquid Raw Signal : K-Lab 12. Nov 2008
D7-Measured
D9-MeasuredD7-Gas
D9-Gas
Figure 23 below shows the measured water fraction with in-situ gas measurements for a 14 day period in November 2008. Initially, the water fraction varies in the range from 0 – 0.05% by volume and for the remaining 14 day period the water fraction is mainly well below 0.01 %. The pressure during this period varies from 25 to 55 barg. From the graph it is seen that the water fraction measurement tracks the small variations in the water fraction well and a stable zero point is maintained for the entire period; despite significant changes in the operating pressure (and hence gas properties) of the flow rig.
Figure 22–Water fraction measurement before and after in-situ gas measurement
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
24
Figure 24 below shows the measured fraction in September 2009, after 10 months continuous operation with in-situ gas measurements. The setup configuration data of the meter has been untouched during this entire period despite significant changes to the gas composition due to frequent loading and discharge of the gas in the flow
rig.
The pressure varies from 28 to 74 barg during the test in September without any noticeable effect on the zero point on the water fraction measurement. The water fraction
Figure 23–14 day period with automatic in-situ gas measurements
Figure 24–Water Fraction measurement after 10 months with automatic in-situ gas measurements
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
25
measurement of the meter is generally within ± 0.02% abs of the K-Lab reference measurement for the entire GVF and pressure range with a zero point stability well within 0.002% abs. Figure 25 to the right shows a close up of the water fraction measurement at the start of the test shown in figure 24 above. From the graph it is seen that the zero point, after 10 months of continuous operation, is still well within a 0.002% margin. In fact, the zero point is also within 0.001% in this case. 6 SUMMARY & CONCLUSIONS In this paper, a true three-phase wetgas flow meter, which is in addition capable of measuring the water production of wetgas wells, has been presented. The meter has proven to be insensitive to errors in the initially input fluid properties and is similarly able to cope with changes in fluid properties that generally occur during the production of wetgas.With this design, the challenges in measuring water production of wetgas fields, as pointed out by Hans van Maanen at the North Sea Flow Measurement Workshop in 2008 [2], have been mitigated. The wetgas meter has been designed to handle the naturally occurring flow conditions of wetgas in a vertically upward flow direction without the use of a mixing device. This has been an important design criterion for the meter and is essential in order to avoid turbulence in the gas phase, which would otherwise cause severe liquid re-circulation and local hold up of liquid – thus deteriortating the water measurement. The meter is equipped with functionality for in-situ measurement of PVT gas properties which, when combined with a naturally high tolerance to variations in the gas PVT properties, enable sensitive, accurate and repeatable water fraction measurement over time, with a zero point stability that is better than 0.002% abs. The meter also been shown to be capable of measuring the salinity of the water fraction which can be used for early detection of formation water break through.
Figure 25– Zero point stability of water fraction measurement after 10 months of operation
27th International North Sea Flow Measurement Workshop 20th -23rd October 2009
26
6 ACKNOWLEDGEMENTS The development of the MPM HighPerformanceFlowmeter has been supported technically and financially by the six oil companies; ENI, ConocoPhillips, Hydro, Shell, Statoil, and Total. In addition, Anadarko, Chevron and Gaz de France have participated in the test of the MPM subsea meter at South West Research Institute. Nine oil companies; ENI, Chevron, ConocoPhillips, Gaz de France, Shell, StatoilHydro, Petronas, Total and Woodside are also participating in an ongoing JIP to develop and qualify operational procedures to automatically monitor and verify the measurement integrity of the MPM meter. Flow models implemented in the MPM meter have been developed in collaboration with Onera, Total and Gaz de France. 7 REFERENCES [1] M. van Werven and H. R. E. van Maanen, Modelling Wet-Gas Annular/Dispersed
Flow through a Venturi AIChE Journal, June 2003, Vol. 49, No. 6 [2] Hans R E van Maanen, Shell Global Solutions, Measurement of the Liquid Water
Flow Rate Using Microwave Sensors in Wet-Gas Meters: Not As Simple As You Might Think, North Sea Flow Measurement Workshop 2008.
[3] J.P. Couput, G. Salque, P. Gajan, A. Strzelecki, J.L. Fabre, New Correction
Method For Wet Gas Flow Metering Based on Two Phase Flow Modelling: Validation on Industrial Air/Oil/Water Tests at Low And High Pressure, North Sea Flow Measurement Workshop 2007.
[4] R. de Leeuw, Liquid Correction of Venturi meter Reading in Wet Gas Flow,
North Sea Flow Measurement Workshop 1997 [5] A Wee, H Berentsen, V.R. Midttveit, H. Moestue, H.O. Hide, Tomography
powered multiphase and wetgas meter providing measurements used for fiscal metering, North Sea Flow Measurement Workshop 2007.
1
Operational Experience Roxar Wetgas Meters, Offshore Egypt
By
Mohamed Bydoun, Burullus Gas Corporation Ingar Tyssen, Roxar Flow Measurement
Executive Summary
Customer: Burullus Gas Corporation
Challenge: To accurately measure all produced fluids on a well-by-well basis within
the West Delta Deep Marine (WDDM) concession for production optimization, high
availability, production allocation and history matching purposes.
Solution: Burullus went through a rigorous selection processing assessing the relative
merits of Inference Systems, Differential Pressure Meters, Multiphase Meters, and
Wet Gas Metering. Working on evaluation criteria of reliability, accuracy, water
measurement, and ease of installation, Burullus opted for the Roxar Wetgas meter.
The Roxar Wetgas meter is a compact, state of the art meter for the inline
measurement of wet gas flow. The meter provides real-time, accurate measurement of
hydrocarbon flow rates and water production - highly valuable for reservoir
management, flow assurance and for optimizing the production process.
Results: Installation of the Roxar Wetgas meter has helped Burullus to monitor water
production profiles in real-time. By providing early warnings of the water produced,
the wet gas meters installed on the field have helped Burullus and its partners save
several wells from water breakthrough. It has also provided Burullus with the
necessary information to optimize the performance of its wells, leading to increased
availability. Moving forward, it is hoped that Burullus will be able to create a
dynamic simulation model of the reservoir continually updated with production data.
In this way, production and pressure data from the simulation model can be matched
to historical pressure data and daily production data by adjusting any number of
reservoir parameters. The result would be reduced uncertainty, and a better
understanding of reservoir behavior.
2
The Challenges of Wet Gas Fields
Oil and gas operators today are facing a number of significant challenges in their efforts to optimize production and generate more from their reservoirs – particularly in wet gas fields.
The ability to predict and measure the water production profile for a subsea well has become critical for optimizing production, preventing hydrate, scale and corrosion in pipelines, and ensuring reliability of supply. Water and formation water in particular, can lead to scaling and corrosion of the pipelines and chokes leading to a significant reduction in well production.
The growth in subsea tiebacks has only exacerbated the challenge with it taking longer to detect a water breakthrough in the well which, by the elapsed time, could lead to severe consequences and pipeline damage.
It is therefore essential for operators to be able to measure the early onset of formation-water production in real time in order that preventative or remedial action (adjusting the pH in the MEG/water mixture or controlling the hydrate inhibitor added to the produced fluids, for example) can be taken. Availability Challenges
Operators are also facing availability challenges. They require accurate estimates of the timing of future developments including: infill wells, new field developments, and production facility requirements.
Poor estimates will lead to the premature advances of development phases to
avoid any shortfalls in production, leading to significant potential financial losses. To this end, accurate predictive tools are required. The Rise of Wet Gas Metering
The challenges of wet gas have led to the emergence of specialised wet gas meters, providing real-time, accurate measurement of hydrocarbon flow rates and water production. Yet, how does one opt for a wet gas metering solution and what are the benefits compared to other solutions?
The West Delta Deep Marine (WDDM) Concession
The West Delta Deep Marine (WDDM) concession is situated offshore Egypt in the Mediterranean Sea, about 130 kilometers north east of Alexandria. Burullus Gas Corporation is developing and operating the fields on behalf of its partners.
The WDDM is Egypt’s largest known gas-bearing block offshore and is producing gas for both domestic consumption and to feed the country's growing LNG industry.
3
There are three principle fields in the WDDM concession: the Simian/Sienna
field and the Sapphire field which is tied back to the Scarab/Saffron field. From the Scarab/Saffron field the gas is commingled before transferred to Burullus’ onshore processing facilities at Idku near Alexandria.
Specific production characteristics from the fields’ wells include high gas rate
wells with low water gas ratios (0.3-10 bbl/mmscf) and water depths of between 400 and 1,400 meters.
Formation water production is also expected to increase with both the Simian/Sienna and Sapphire developments anticipated to produce both condensed water and increasing volumes of formation water over their production lifetime. Reservoir Management Challenges Facing Burullus Burullus had two immediate reservoir management challenges - the need to be able to detect water in real-time so that immediate corrections can be made to production operations; and the high availability requirements to meet obligations to provide gas for both export and the domestic Egyptian market.
There were also a number of future reservoir management challenges which Burullus wished to address. This included the need to optimize future development phases focusing on the timing and locations of the next drilling phases.
To address these challenges, it was essential for Burullus to be able to accurately measure all produced fluids on a well-by-well basis for production optimization, production allocation and history matching purposes.
The Decision to Opt for Wet Gas Metering
Burullus had a number of requirements for any metering solutions it selected. The solution needed to be able to continuously measure flow on a well by well basis; have the ability to detect water breakthrough for the controlling of chemical injection; ensure high availability; and result in the accurate measurement of all produced fluids on a well-by-well basis.
4
Burullus assessed a number of metering options - Inference Systems, Differential Pressure Meters, Multiphase Meters, and Wet Gas Meters.
The estimated uncertainty from the inference measurement systems was
considered by Burullus to be well outside the acceptable range. Differential pressure meters were considered to be unsuitable because of the
changing composition especially as a result of water breakthrough. Multiphase flow meters were also considered unsuitable because of high
liquid flow uncertainty in the case of high GVF as exists in the WDDM fields.
The conclusion was that wet gas meters offer the only acceptable solution to the water flow rate measurement of high Gas Void Fraction (GVF) wellstreams in remote subsea locations.
Having made the decision to opt for wet gas meters, Burullus compared
Roxar’s Wetgas Meter with a rival meter. It was determined that the Simian/Sienna and Sapphire flow regimes were within the operating envelope of the Roxar Wetgas Meter except at the highest water production rates. The Simian/Sienna flow regimes were outside the rival meter’s operating envelope except at the highest water production rates (the Sapphire flow regime was within the operating range). The Roxar Wetgas meter
The Roxar subsea Wetgas meter (figure 1) is a compact, state of the art meter, which provides real-time, accurate measurement of hydrocarbon flow rates and water production. The meter has an operating range of 90 to 100 per cent GVF, a hydrocarbon mass flow relative uncertainty of +/- 5%, a water fraction measurement uncertainty of 0.1abs vol. % and water fraction sensitivity < 0,0008 abs. vol.%.
Figure 1
The Roxar Wetgas meter is being utilized in some of the world’s best known
offshore gas fields, including the Independence Hub in the Gulf of Mexico and the Ormen Lange field in the North Sea.
5
There are three key attributes to the Roxar Wetgas meter – its robustness
and low maintenance requirements, its accuracy, and its ability to measure water fraction in a wet gas flow in real-time.
The Roxar Wetgas meter is qualified for high-pressure/high-temperature applications to water depths of 3,000 meters with the meter body constructed of UNS S31803 Duplex. The meter can operate at up to 10,000 psi and 150 °C.
The Roxar Wetgas meter provides online and direct, accurate measurements
of water fraction in a wet gas flow, utilizing microwave-based dielectric measurements. In combination with a V-cone based differential pressure measurement this also generates accurate water, gas and condensate flow rates.
Implementation
Today, eight Roxar Wetgas meters are installed in the Simian-Sienna field and eight on the Sapphire field, providing real-time, accurate measurements of hydrocarbon flow rates and water production.
Water production monitoring started almost immediately. Figure 2 shows the
gas rate of each field with purple corresponding to the Sapphire field, red the Simian Sienna field, and yellow Scarab/Saffron. The blue curve represents the saline water coming from the reservoir, measured on shore.
Figure 2 illustrates the challange of determining sources and causes for
changing salinity levels, as there are no Wetgas meters on the Scarab wells.
Figure 2 Each meter delivered to Burullus for the Simian/Sienna and Sapphire fields was equipped with the formation water detection functionality which in combination with
0
500
1000
1500
2000
2500
1-Ja
n15
-Jan
29-J
an12
-Feb
26-F
eb12
-Mar
26-M
ar9-
Apr
23-A
pr7-
May
21-M
ay4-
Jun
18-J
un2-
Jul
16-J
ul30
-Jul
13-A
ug27
-Aug
10-S
ep24
-Sep
8-O
ct22
-Oct
5-N
ov19
-Nov
3-D
ec17
-Dec
31-D
ec14
-Jan
28-J
an11
-Feb
25-F
eb11
-Mar
25-M
ar8-
Apr
22-A
pr6-
May
20-M
ay3-
Jun
17-J
un1-
Jul
15-J
ul29
-Jul
12-A
ug26
-Aug
9-Se
p
Gas
Rat
e (M
Msc
f/d)
0
7,000
14,000
21,000
28,000
35,000
Salin
ity (p
pm)
6
the high water fraction sensitivity allows for early identification of potential water breakthrough. In Figure 3, the red line shows the water and the blue line the gas flow rates from one specific well. On 25th May 2006, a water breakthrough was detected. Burullus decided to reduce the choke size in order to lower the water produced and recover and generate better production profiles. After approximately three weeks producing at lower rates, Burullus decided to open the upper zone of the smart completion. Another two weeks later the water production started to increase rapidly again. In an effort to stabilize the well, it was shut in for a week in early August 2006. When production started again, however, the water starting to increase dangerously and a decision was taken to decrease the lower inflow control valve position. Water then started to decrease, indicating increased water production coming from the lower zone. Burullus then went into a period of tuning the lower inflow control valve position to find the optimum gas production at an acceptable water rate. By providing early warnings of the water produced, the wet gas meters installed on the field has helped Burullus and its partners to save several wells from water breakthrough in a similar manner. It also gave Burullus the necessary information to optimize the performance of its wells at acceptable and controlled water rates.
Figure 3 Moving forward, it is hoped that Burullus will be able to create a dynamic simulation model of the reservoir continually updated with production data. In this way, production and pressure data from the simulation model can be matched to historical
Water breakthrough detection
0
10
20
30
40
50
60
70
80
90
100
Apr-06 Apr-06 May-06 May-06 Jun-06 Jun-06 Jul-06 Jul-06 Aug-06 Aug-06 Sep-06 Sep-06
Time
Gas
rat
e M
MSC
FD
0
20
40
60
80
100
120
140
160
W
ater
rat
e ST
B/D
Gas MMscfd Water stbd
Start Water Breakthrough
Reduce shock SizeUpper Zone start up
Increase Water Production
Decreased lower ICV Postion
Increased lower ICV one Posion
7
pressure data and daily production data by adjusting any number of reservoir parameters. The result will be reduced uncertainty, and a better understanding of reservoir behavior.
Page 1 of 15
To be presented at the North Sea Flow Measurement Workshop 2009, Tonsberg, Norway.
Calibration errors of ultrasonic meters in the Bernoulli laboratory
due to non-isothermal flow conditions.
Authors:
1. Aernout van den Heuvel, KEMA (previously: Gasunie Engineering & Technology)
2. Frans Doorman, NAM
3. Piet van den Herik, NMi
4. Arjan Stehouwer, Elster-Instromet
5. Robert Kruithof, Gastransport Services
ABSTRACT
The Bernoulli laboratory (“Westerbork”) in the Netherlands, jointly operated by Gasunie,
NMi and KEMA, holds a record of over 30 years in calibrating very large gas meters at high
pressure with natural gas. NAM, a major producer of natural gas in the Netherlands, utilises a
large number of ultrasonic meters. Elster-Instromet is a world wide operating manufacturer of
multi-path ultrasonic meters. Gastransport Services is the national gas transporting company
of the Netherlands, who is connected to the network of NAM via several delivery stations
utilizing ultrasonic meters.
Due to the results of five calibrations of 24-inch ultrasonic meters in the Bernoulli laboratory
early 2008, a quality check procedure (number QC-11) was started by the Bernoulli
laboratory. Four out of five ultrasonic meter showed strongly non-linear behaviour at low
flow rates, with a maximum at about 560m3/h (approximately 2% of Qmax) with errors
peaking as high as +2%. Taking into account, the demands set in ISO/FDIS 17089-1:
2009(E), all four meters would have been rejected had the standard had already been ratified.
Together, the manufacturer, the user and the calibration facility decided to give high priority
to quickly identify and resolve this problem. After excluding all straightforward errors, two
possible causes remained on the short list: a disturbed flow profile at low flow rates, typically
below 2 m/s, and meter problems. A series of experiments and improvements to the
calibration facility were executed in the course of 2008: an additional temperature
measurement at the bottom; dozens of calibrations of one specific 24-inch ultrasonic meter,
made available by the user (this meter was even calibrated in the up side down position);
thermal lagging was improved; and finally, in January 2009, the full measurement section was
insulated. Applying full thermal lagging with 15°C temperature difference between gas and
ambient, temperature differences were largely reduced, as was the error of the ultrasonic
meter at 800m3/h. Also the ultrasonic meter diagnostics showed a large improvement of the
flow profile. Although the thermal lagging had insufficient heat transmission resistance,
which prevented a 100% success, the experiment showed, without doubt, that non-isothermal
flow conditions were the root cause of the calibration errors.
Non-isothermal flow conditions have previously been addressed at the 2005 Flomeko
conference [3], however, the discussion at that time concentrated on the temperature
measurement error and the effect was characterised as “stratification”. In that paper, other
effects of non-isothermal flow: non-axial velocity components and an asymmetric flow
profile, were not recognised as a problem. The majority of gas meters calibrated in Bernoulli
laboratory, turbine meters, are not sensitive to non-isothermal flow conditions. Ultrasonic
meters however, measure velocity components and may be very sensitive to a combination of
non-isothermal flow, depending on their path configuration. Clearly, the Q.Sonic-4C of
Elster-Instromet is sensitive to these non-isothermal flow conditions present in Westerbork.
Page 2 of 15
(Presently, insufficient data on other types of meters are known to the authors to be able to
judge their sensitivity to non-isothermal flow. One of the problems in obtaining this
information is the likeliness of missing the error peak, since it might very well lie between
two calibration points.)
1. INTRODUCTION The facility:
Established in 1978, the Bernoulli laboratory in Westerbork, the Netherlands, provides
research and calibration services with high pressure natural gas under flowing conditions.
Figure 1 - Arial overview of the Bernoulli laboratory in Westerbork, the Netherlands
The facility is among the largest in the world, capable of calibrating 30” gas meters with
120D straight upstream piping. Annually, more than 200 large meters are calibrated.
24"30"
40D 2D
Ball Valve USMReducer
Bottom PT100
Top PT100
3D
Permanent Lagging Figure 2 - Schematic lay out of the meter run with a 24 inch ultrasonic meter; valid as from December 2008.
Note that the initial results presented, were achieved without the depicted “permanent lagging” and without the
depicted “bottom PT100”.
Page 3 of 15
The parties involved:
The Bernoulli laboratory is owned and operated by Gasunie, the Dutch national gas
transmission company, transporting more than 80 billion cubic meters annually. Gasunie
cooperates with NMi-Nederland and KEMA.
NMi-Nederland is accredited according to ISO17025 for flow calibrations in the Bernoulli
laboratory. NMi operates the front- and back office, supervises the calibration, issues the
certificate and is responsible for the quality and the traceability of the reference standards.
KEMA, division GCS (Gas consulting and services), until July 1st 2009 part of Gasunie,
delivers, among others, services to Gasunie in the field of instrumentation, ict and scientific
support.
NAM, is the largest gas producing company in the Netherlands. Almost all gas produced by
NAM is transported by Gastransportservices. The majority of the delivery stations are
equipped with ultrasonic gas meters, which are calibrated in the Bernoulli lab. The theme of
this paper is about some of these calibrations.
Gastransportservices, owned by Gasunie, is on the receiving end of the gas delivered by
NAM, which makes the company equally concerned about the flawless determination of
delivered volumes.
Elster-Instromet is a market leader in the design, manufacture and supply of gas metering
products. In addition to the Ultrasonic meter portfolio, Elster-Instromet offers a full range of
turbine meters, frequently utilised as calibration reference meters, as well as a complete range
of volume correctors, flow computers, pressure reduction & control equipment, gas
chromatographs and energy measurement devices.
A working group was formed with representatives of all parties.
2. PROBLEM DESCRIPTION
In February 2008 a series of five new 24” Q.Sonic-4C Ultrasonic meters were calibrated.
They were intended for Custody Transfer station “Sappemeer”, delivering gas from the
network of NAM to the network of Gasunie. The calibration curves of four of these meters
indicated an unexpected large peak in the calibration curve; see Figure 3.
Page 4 of 15
Calibration results five Q.Sonic-4 in Bernoulli laboratory spring 2008
-1,0
-0,5
0,0
0,5
1,0
1,5
2,0
0 5000 10000 15000 20000 25000
Flowrate [m3/h]
De
via
tio
n [
%]
3643 3644 3645 3646 3647
Figure 3 - Calibration results of five Q.Sonic-4 ultrasonic meters
The peak appeared in a flow rate of 560m3/h in all cases. Curious about the reproducibility of
the peak and the absence of the peak in serial number 3644, it was considered that a sharp
peak, covering a limited flow range, could be overlooked by the choice of the calibration flow
rates. So, with serial number 3646 still installed, the flow region around 560 m3/h was
examined in smaller steps, as is shown in Figure 4.
Calibration Q.Sonic-4C sn 3646 in Bernoulli laboratory spring 2008
-2
-1,5
-1
-0,5
0
0,5
1
1,5
2
0 5000 10000 15000 20000 25000
flow rate [m3/h]
De
via
tio
n [
%]
Limit Limit More flow rates than usual Usual flow rates
Figure 4 - Calibration result of serial number 3646 with additional flow rates below 1600m3/h
Page 5 of 15
Unexpected calibration results concentrate in the region 300-1800m3/h, which equals 1-6% of
Qmax. With a maximum value at 800m3/h of 1,54%, exceeding 1,4%, the calibration result
does not fulfil the requirements of ISO/FDIS 17089-1: 2009(E) [1]; see the relevant table
below.
Were ISO/FDIS 17089-1: 2009(E) [1] already published, the four meters with a peak
exceeding 1,4% would have been declared unfit. Next to that, the meter that does not show
the peak, would leave the participants with the considerable doubt that the peak may have
remained undetected.
3. STEPS TAKEN TO UNDERSTAND THE PROBLEM
3.1. Initial investigations
The question was, whether the peak was a meter malfunction or the result of an error during
calibration. This question triggered the “QC” procedure (Quality Check) of the laboratory.
Due to strong similarities with some irreproducible results even dating back to December
2004, the new question was bundled with the earlier questions into the existing, although
inactive, QC11. The phenomenon-under investigation is referred to as the QC11 problem.
The “standard” facility checks on the five calibration runs were done, which showed no
shortcomings in the set-up of the calibrations, the data acquisition and the calculations.
The only problems that were found, and were thought to be compensated for, were non-
isothermal flow conditions, at that time called “stratified flow conditions”. By applying two
temperature measurements, one of which was at the bottom of the pipe and the other was at
the top of the pipe, the temperature measurement was thought to be corrected for.
Page 6 of 15
Non-isothermal flow conditions in the Bernoulli laboratory were already discussed on the
2005 Flomeko conference [3]. It appeared that with sufficiently low gas velocities, the gas
flowing along the top of the pipe was warmer than the gas flowing at the bottom of the pipe,
due to lack of thermal lagging and temperature differences between the gas and the ambient.
Consequently, the temperature measured at top of the pipe was not representative for the
average temperature of the gas inside the gas meter. Installing an extra temperature
measurement at the bottom of the pipe, allowed the facility to detect non-isothermal
conditions and to correct for it by taking the average value. Temperature differences between
top and bottom of up to 3 °C were found. However, although considered in 2005, thermal
lagging was not introduced, motivated by the demand for user friendliness and unaware of
secondary effects. Note that flow conditions were still in the turbulent region (Re=7*105 @
280 m3/h)
Being a heat transfer process, time, location and weather became relevant parameters:
1. Time: It takes time, after changing the gas flow, to reach a new equilibrium.
2. Location: The amount of heat that is transferred, accumulates in the downstream direction.
3. Weather: Changing weather conditions affect the amount of heat that is transferred.
These parameters, together with the chance of missing the peak by the choice made for the
calibration flow rates, make the appearance and disappearance of the peak in the error curve
seemingly irreproducible.
Therefore, non-isothermal flow conditions were put on the list of possible causes although the
working group did not understand the mechanism at that moment of time.
In addition, Elster-Instromet performed standard health checks on the diagnostic data
collected during the calibrations. No malfunctions were found, but surprisingly, at flow rates
around 800m3/h, the average gas velocity of the axial paths and the average gas velocity of the
swirl paths did not show the expected 1,05 ratio. This ratio, also known as the profile factor,
reduced even to 1,00, making it appear that the flow profile was “flat”.
Top of Pipe
Figure 5 - front view path lay out Q.Sonic-4C Figure 6 - top view path lay out Q.Sonic-4C
(Axial paths omitted)
Combining the findings of the facility and the meter manufacturer, the focus of the
investigations was on the flow profile.
Page 7 of 15
3.2. Further investigations
Before starting any further investigations, the temperature measurement was improved and
pulsation measurements were performed:
1. Until spring 2008, a single bottom temperature measurement was installed in the 20”
section; not shown in figure 2. At restarting QC11, it was completed with a dedicated 24”
bottom temperature measurement, exactly below the top measurement, 2D downstream of
the 24” ultrasonic meter; shown as “Bottom PT100” in figure 2.
2. Pulsation measurements were taken during a relevant calibration, confirming the absence
of pulsations as was found during earlier measurements.
For the purpose of investigating the QC11 problem, NAM provided a spare Q.Sonic-4C
ultrasonic meter with serial number 3015.
First flow profile investigations:
1. 30D upstream of the 24” ultrasonic meter, a full-bore valve is installed. Considering
possible flow profile disturbances from a not fully opened valve, the position of the
opened valve was recorded. Time after time, the valve showed to be opened for the full
100% after the closure and reopening action. So, this valve was no longer considered as a
risk.
2. A reasonably simple experiment that has been performed was the calibration of an
ultrasonic meter in the upside down position. No results different from the previous
calibration were expected if the peaked curve was caused either by a symmetrical flow
disturbance or by a misinterpretation of the flow profile by the meter. A different result
was expected with an asymmetrical flow disturbance, since the ultrasonic meter is not
symmetrical see figures 5 and 6. As a result of the experiment, the calibration curve
showed to be very different; see figure 7. Also the profile factor remained close to the
expected value of 1,05; see figure 8.
Elster-Instromet Q.Sonic-4C sn 3015 in Bernoulli laboratory
-0,8
-0,6
-0,4
-0,2
0,0
0,2
0,4
0,6
0,8
1,0
1,2
1,4
0 5000 10000 15000 20000 25000 30000
Q [m3/h]
Devia
tio
n [
%]
upside down isothermal average of >20 calibrations throughout 2008 and 2009
Figure 7 - Several calibration results with sn 3015
Page 8 of 15
3015 Q.Sonic-4 of NAM at BNL in 2008profile factor = v_axial/v_swirl
1
1,02
1,04
1,06
1,08
1,1
0 5000 10000 15000 20000 25000
flow [m3/h]
pro
file
facto
r
non-isothermal upside down isothermal
Figure 8 - Profile factors of sn 3015 from several calibration runs
At that phase in the investigations, it was clear that the facility was suffering from a flow
profile disturbance, which was asymmetrical, and, from the earlier results, was not
reproducible at all circumstances.
3.3. Decisive investigations
Major progress in the investigations came with the consideration that non-isothermal flow
conditions must be accompanied with a mechanism to transport warmer gas to the top and
colder gas to the bottom; this implies non-axial flow. Since it depends on the ambient
conditions whether non-isothermal flow conditions would develop or not, the working group
was able to explain the apparent irreproducibility of the earlier results.
Confirmation was provided by the measurements of August 21st 2008, during which the gas
temperature was the same as the ambient temperature. With non-isothermal flow conditions
absent, no peak in the calibration curve was found, as was the expected profile factor of 1,05.
At this point, the working group was quite convinced that non-isothermal flow, was the root
cause of the QC11 problem.
The working group decided to perform a decisive experiment by using thermal lagging to
suppress the non-isothermal flow conditions and to summon non-isothermal flow conditions
again by removing the thermal lagging. The purpose of this experiment was to demonstrate
the decrease of the peak in the error curve and decrease of the dip in the profile factor.
Therefore, the quite stationary part of the upstream piping, being the 30” piping and the 24”
piping between the 30”piping and the full bore 24” valve was lagged with permanent lagging;
see figure 2. Flexible lagging, not shown in figure 2, was used to lag the remaining 24” piping
including the ultrasonic meter and the temperature measurements. The experiments performed
on January 9th
2009 proved unambiguous, although, the thermal conductivity of the flexible
Page 9 of 15
lagging still appeared to be too large. Figure 9 shows the development of the profile factor
during the application of the lagging.
Development of the profile factor of Q.Sonic-4C sn 3015 during lagging on January 9th
2009
(60s-running average)
0,96
0,97
0,98
0,99
1
1,01
1,02
1,03
1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 1801 1901
time [s]
pro
file
facto
r (V
ax/V
sw
)
Figure 9 - Effect of the application of thermal lagging on the profile factor
Figure 10 shows the results obtained on January 9th
, first without lagging, then with lagging
applied and finally without lagging (lagging removed).
Experiment with thermal lagging on January 1st 2009
0
0,5
1
1,5
2
2,5
without thermal lagging thermal lagging applied thermal lagging removed
time
Tem
pera
ture
dif
fere
nce [
K];
Devia
tio
n [
%]
0,98
0,99
1
1,01
1,02
1,03
1,04
Pro
file
Facto
r [-
]
Temperature difference top-bottom [K] Deviation [%] Profile Factor
Figure 10 - The effect of thermal lagging on
1. the measured temperature difference at the top and at the bottom,
2. the deviation of the meter,
3. the profile factor of the meter
Page 10 of 15
During the calibration with full thermal lagging the peak @ 800m3/h was significantly
reduced.
Figure 11 - Thermal lagging; Jan. 9
th 2009, Figure 12 - Thermal lagging detail; Jan. 9
th 2009
3.4. CFD investigations
In order to gain confidence on the conclusions derived from the experiment on January 9th
Elster-Instromet proposed to perform CFD (Computational Fluid Dynamics) simulations of
the phenomenon. TNO Science and Industry was asked to perform the analysis. Although a
simplified model of the setup and the ambient conditions was used, the results matched the
phenomena that were seen in the Bernoulli laboratory. Initially, two cases were simulated, one
without a temperature gradient across the top and bottom of the pipes and the other case with
a temperature gradient close to the ambient conditions of those of January 9th
. Note: ambient
conditions such as wind and humidity were not taken into account.
Both the measured temperature differences between top and bottom thermo wells, the
established measurement error and the profile factor were predicted quite accurately. The
secondary flow profile (perpendicular to the axial direction of the flow) as can be seen in
Figure 13 shows two counter rotating cells, one on the left side and one on the right side,
transporting warmer gas to the top and colder gas to the bottom.
Figure 13 - Secondary flow at meter inlet; the ambient being colder than the gas
Figure 13 shows the rotation direction of two counter rotating cells under the condition that
the ambient is colder than the gas. Iit is inevitable that the rotation direction of the cells will
reverse when the ambient is warmer than the gas. However, in both conditions, the warmer
gas will flow to the top section, while the colder gas will flow to the bottom section.
Page 11 of 15
Figure 14 - Axial flow pattern at meter inlet 0.8m/s
This behaviour results in an asymmetric profile as seen in Figure 14 and Figure 15. The gas
velocity in the upper (warmer) part of the pipe is higher than the gas velocity in the lower
(colder) part.
2D plot of axial f low pattern
-0,3
-0,15
0
0,15
0,3
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Velocity [m/s]
Po
sit
ion
[m
]
Figure 15 - 2D plot of the axial flow profile (centre line) @ 800 m
3/h
In addition two more cases were simulated, half (0.4m/s) the flow rate and double (1.6 m/s).
With double flow rate the non-isothermal profile almost completely disappeared, again being
consistent with the measured results obtained.
Page 12 of 15
Figure 16 - Axial flow pattern at meter inlet 1.6 m/s
At half flow rate the model showed even more severe non-isothermal flow conditions.
However, the model did not match the measured values anymore. It is expected that other
effects, not considered in the simulation, overwhelm the non-isothermal flow conditions
causing the mismatch between model and reality.
3.5. Concluding the investigations
Our conclusion became final:
1. An asymmetrical flow profile with non-axial flow emerges in the 24” section in the
Bernoulli laboratory when:
a) Flow rates are low (typically<1,6m/s),
b) There is significant heat transfer between the gas and the ambient due to insufficient
lagging, combined with a temperature difference between the gas and the ambient.
2. A Q.Sonic-4C ultrasonic meter under these specific non-isothermal flow conditions
calculates a deviating flow rate with errors up to +2%.
3. An asymmetrical flow profile with non-axial flow cannot be characterised as stratified flow
conditions as was earlier assumed, but it should be characterised as non-isothermal flow or,
flow with natural convection; the flow pattern looks like two counter rotating corkscrews.
Non-isothermal flow is undesirable in any calibration facility, which characterizes the meter
deviation as a calibration error rather than a meter error.
Although non-isothermal flow conditions is a typical low flow rate problem, both the flow
rate at which the peak appears and the height of the peak, is likely to vary with the already
mentioned parameters. At this moment in time it is still not understood why at lower flow
rates, it seems that as if the problem diminishes; usually, problems worsen at lower flow rates.
Undoubtedly, all calibration facilities and metering stations without sufficient thermal
lagging, combined with temperature differences between the gas and the ambient will suffer
from non-isothermal flow at low flow rates.
Page 13 of 15
Prior to discussing the actions that need to be taken by the different parties, the question
should be answered whether it is likely or not that other types, sizes and makes of ultrasonic
meters detect the same phenomenon.
From the nature of the effect, it may be expected that the problem will appear in all diameters
and also in the Q.Sonic-5. This has already been confirmed by the results of the calibration of
a 20”Q.Sonic-5 and a 12”Q.Sonic-4. Some historic calibrations may seem unaffected by this
effect due to the choice of the calibration flow rates. Other calibrations will be unaffected due
to favourable weather conditions. Some historic calibrations however, will be affected by this
effect.
Actions by the Bernoulli laboratory: Measures should be taken to eliminate non-isothermal
flow or to avoid calibrating ultra sonic meters with non-isothermal flow. So, sufficient
thermal lagging should be applied or gas velocities below 1,6m/s should be avoided. Being a
calibration facility on the brink of closure, any investment will be hard to defend from an
economical point of view.
Actions by the meter manufacturer: The manufacturer may consider to improve its meters by
making them (practically) immune to non-isothermal flow or indicate when specific non
isothermal flow may exist as detected by the diagnostics.
4. DEALING WITH SUSPICIOUS CALIBRATIONS
Calibration results achieved with non-isothermal flow conditions, may be considered
“suspicious” with regard to flow rates below 1,6m/s.
The parties dealing with an ultrasonic meter in full service on a delivery station with an
apparently suspicious calibration curve may consider the folliwing actions:
Suspicious curves impact delivered volumes only if:
1. The calibration curve is used as the basis of a calibration curve correction mechanism
2. The meter run has delivered volumes in the flow range affected by the suspicious
calibration flow rate. Generally, all flow rates below the next largest calibration flow rate
will be affected, e.g. 2800 m3/h in the case of the five meters with serial numbers 3643 to
3647.
Possible action to detect whether the calibration curve of an ultrasonic meter is suspicious:
Comparison/monitoring of flow meter diagnostics between calibration and initial field
installation. The problem created by non-isothermal flows can be detected by monitoring
profile factor, asymmetry and other diagnostics from the flow meter. Changes in the flow
meter diagnostics can be used to determine if conditions in the pipe have changed over
time or between calibration and field installation
Possible actions to stop the building up of delivered volumes with a suspicious curve
correction:
1. Close the meter run and recalibrate the meter
2. Do not operate the ultrasonic meter in the affected flow range
3. Correct the calibration curve correction mechanism by altering the contribution of the
suspicious calibration flow rate.
Page 14 of 15
a. Ignore the suspicious flow rate
b. Making the deviation of the suspicious flow rate equal to the deviation of the next
largest flow rate (2800 m3/h in the example).
Possibilities 3a and 3b may also be used as the correction mechanism to recalculate the
delivered volumes by having the suspicious curve as the basis for the automatic correction.
The working group investigating QC11 appreciates the formulations of ISO/FDIS 17089-1:
2009(E) [1]. This applies for:
1. Warning for discontinuities Ch5.5.2
2. The ultrasonic meter requirements in table 6
3. The stability criteria Ch6.3.2.4
4. The way non-isothermal flow conditions are addressed Ch6.3.2.5
The working group has two suggestions for further improvement:
1. The initial calibration shall contain more calibration flow rates than the standard series,
especially at low flow rates
2. Add non-isothermal flow conditions to the type testing section Ch6.4.3
5. CONCLUSIONS 1. At low flow rates and significant heat transfer between the gas and the ambient due to
insufficient lagging, temperature differences between the gas and the ambient generate
non-isothermal flow.
2. At this moment in time, the working group has only studied the Elster-Instromet Q-
Sonic4C, which shows errors up to +2% under non-isothermal flow conditions. However,
due to the nature of an ultrasonic meter, meters with other path layouts are also expected
to be sensitive to non-isothermal conditions albeit to a different extend. In addition, all
types of gas meters sensitive to flow profile distortions are likely to be affected by non-
isothermal flow.
3. At non-isothermal low flow rates, a single temperature measurement for volume
conversion will not measure a representative value for the bulk flow rate, introducing a
measurement error. Even with two temperature measurements it is uncertain whether the
average value is representative.
4. Due to the influences of time, position in the meter run, ambient conditions and the chosen
calibration flow rates, the adverse effects of non-isothermal flow conditions may either
not be present, or may not be detected and therefore may seem irreproducible.
5. The ISO/FDIS 17089-1: 2009(E) [1] has proven itself to be a valuable tool to distinguish
calibration errors with ultrasonic meters.
6. RECOMMENDATIONS 1. Calibration facilities should actively search whether non-isothermal flow conditions are
possible and if so, should avoid calibration at these conditions, preferably by adding
sufficient thermal lagging.
2. Calibration facilities are encouraged to reconsider their temperature measurements.
3. Calibration facilities should monitor and log diagnostic data such as profile factor,
asymmetry and swirl angle. When diagnostics are outside acceptable limits, the
manufacturer should be consulted. It is recommended to include intermediate calibration
points at low flow rates, in order to detect possible non-linearities.
4. Manufacturers of ultrasonic meters are encouraged to make their meters (practically)
immune to non-isothermal flow or make the ultra sonic meter indicate when specific non
isothermal flow conditions exist as detected by the diagnostics
Page 15 of 15
5. Owners are encouraged to prevent further build up of delivered volumes affected by
suspicious calibration curves
6. Owners are also encouraged to calculate the amount of delivered gas that has been
affected by suspicious calibration curves
7. Owners are encouraged to lag meter runs in situations where temperature differences
between ambient and gas are expected in combination with very low flow rates
8. All parties are encouraged to develop meter diagnostics to detect flow conditions
unfavourable for accurate measurement.
9. All standardisation working groups active in flow measurement, like ISO TC30SC5WG1
and CEN TC234WG5, are encouraged to adopt the recommendations of the QC11
working group and to consider the lessons learnt.
REFERENCES
[1] ISO/FDIS 17089-1: 2009(E) “Measurement of fluid flow in closed conduits – Ultrasonic
meters for gas – Part 1: Meters for custody transfer and allocation measurement”
[2] EN 1776 – 1998, “Gas supply – Natural gas measuring stations – functional requirements”
[3] A.F. van den Heuvel, “Flow measurement errors due to stratified flow conditions”,
Flomeko conference 2005, Paper 1.3
[4] S. Kimpton, “Thermal lagging – The impact on temperature measurement”, NSFMW2008
paper 8.3
Note:Decimal comma notation is utilized throughout the document except for table 6 and figures 14 and 16.
Field Experience of Ultrasonic Flow Meter Use in CO2-Rich Applications
Keith Harper, Sandridge Energy, Oklahoma City, Oklahoma, USA
John Lansing, SICK Inc. Houston, Texas, USA
Toralf Dietz, SICK AG, Germany
1 INTRODUCTION
Ultrasonic gas flow meters have gained a wide acceptance in the field of natural gas exploration,
transport, storage and distribution as well as in the process industry. There is one major segment,
which could not be addressed so far with this technology – high attenuating gases like Carbon
Dioxide.
On the other hand exploration of less conventional natural gas sources will lead to more diverse
operating conditions and compositions for natural gas measurement. One significant change
compared to traditional sources is the increased level of CO2 in the gas. While standard
applications deal with levels well below 5 mole percent, this amount may be as high as 20 mole
percent, or even higher at some installations.
Re-injection of CO2 into declining oilfields will require accurate and reliable flow measurement.
Such applications contain up to 60% CO2 and require an accuracy level comparable to custody
transfer for natural gas. While the flow measurement is currently being done primarily using ∆p
devices, such as orifice meters, it would be a significant improvement to use ultrasonic meters
with their increased functionality, larger turn-down ratio reduced maintenance, and diagnostic
capabilities. Applications such as CCS (Carbon capture and storage) with CO2 concentrations
near 100% are even feasible today.
Table 1 gives a short summary of various applications and the typical amount of carbon dioxide in
the gas stream.
Application type CO2 Content Other
components
Pressure Accuracy
requirements
Natural gas, low
CO2 <5%
CH4, N2, higher
HC 150 to 2250 psig Custody transfer
Natural gas, high
CO2 5% to 20%
CH4, N2, higher
HC 150 to 2250 psig Custody transfer
Re-injection Up to 60% CH4 700 to 1480 psig Allocation
Carbon capture
& storage Nearly 100% no 150 to 2250 psig Allocation
Table 1: Typical CO2 content in different applications
Such applications are handled by orifices today, although many users would prefer to use
ultrasonic meters due to there benefits of the technology. Unfortunately high carbon dioxide
content usually causes serious problems for ultrasonic flow meters because its attenuation of
ultrasonic signals is extremely high compared to that of other gases. That is why ultrasonic flow
meter manufacturers must state the maximum amount of CO2 allowed in the gas stream at which
the meter will still function within specifications.
The main purpose of this paper is to show how today’s advanced ultrasonic flow meters can
handle such high amounts of CO2. Based on a systematic study of the attenuation of ultrasound in
CO2-rich mixtures, an application model for ultrasonic flow meters was developed at the
manufacturer’s facility. This model created an estimation of the applicability of ultrasonic flow
meters under different operating conditions and indicated a requirement for special meter design
considerations. Operational limits regarding path length, pressure and frequency where
determined by laboratory examinations and an optimized meter design is concluded.
Field experiments on a typical re-injection application with a CO2 concentration near 60% showed
the successful operation of an ultrasonic flow meter and are presented in this paper.
2 PHYSICAL BACKGROUND OF ULTRASONIC SIGNAL ATTENUATION
Carbon dioxide is well known for the significant attenuation it causes to ultrasonic sound waves.
Table 2 shows the attenuation coefficients and the signal losses (absorption at atmospheric
pressure) of methane and carbon dioxide as compared to dry air. The attenuation coefficient is
defined by the Lambert-Beer-Law:
(1) )exp(0 lpp α−=
• p - acoustic pressure at distance l
• 0p - initial acoustic pressure at source
• α - attenuation coefficient
Operating frequency Dry air Methane Carbon dioxide
α [m-1
] dB cm-1
α [m-1
] dB cm-1
α [m-1
] dB cm-1
80 kHz 0,09 0 5,3 -10,4 33,5 -25,3
135 kHz 0,26 0 9,9 -13,5 39,9 -31
208 kHz 0,62 0 12,3 -15,2 42,6 -33,3
Table 2: Comparison of attenuation coefficients gases at typical operational USM frequencies and
loss of acoustic pressure per length due to attenuation compared to dry air
These numbers show that carbon dioxide is one of the most difficult gases to measure with
ultrasonic gas flow meters. To develop the successful application of USMs in CO2-rich
atmospheres, the basics behind the attenuation and all other influences on performance must be
understood. The acoustic absorption by gases can be divided into two main contributions:
• classical absorption
• thermal relaxation phenomena.
The classical absorption is based on the viscosity and the thermal conductivity of the gas. This
attenuation increases with the square of the frequency. Since the difference between various
gases is quite small, this classical absorption can not reflect the enormous differences in the
behaviour between carbon dioxide and other gases.
Attenuation based on thermal relaxation processes reflects the fact that energy is exchanged
between molecular vibrations and translations. A maximum of normalised attenuation can be
found depending on frequency (see Figure1a). The frequency position and the strength of this
maximum attenuation are typical for the gas or gas mixture to be measured. If this maximum is
located within the range of the operating frequency of ultrasonic transducers, there may be a
strong influence on the performance of the ultrasonic flow meter.
Carbon dioxide is one of the model gases used to study this type of attenuation, and therefore a lot
of literature can be found to describe its behaviour. From that data, a theoretical model of the
attenuation in CO2 can be developed. Figure 1b shows the attenuation coefficient and the
normalized attenuation coefficient for CO2 at room temperature.
1 103
× 1 104
× 1 105
× 1 106
× 1 107
×
0
0.05
0.1
0.15CO2
Air
Frequency/Pressure [Hz/bar]
Att
enu
atio
n*
Wav
elen
gth
[ ]
1 103
× 1 104
× 1 105
× 1 106
× 1 107
×
1 105−
×
1 104−
×
1 103−
×
1 102−
×
1 101−
×
1 100
×
1 101
×
1 102
×
1 103
×
1 104
×
classical attenuation
relaxational attenuation
total attenuation
Frequency [Hz]
Att
enuat
ion c
oef
fici
ent
[m-1
]
Figure 1a: Normalised attenuation coefficient Figure 1b: Attenuation coefficient for CO2
for CO2
The relaxation frequency of CO2 can be found at about 50 kHz and has an extremely high
attenuation coefficient compared to other gases. The thermal relaxation dominates the attenuation
in the kHz-range, and this leads to the difficulties exhibited in the application of ultrasonic flow
meters in carbon dioxide. Classical absorption will contribute significantly at frequencies above 1
MHz.
While pure gases are quite well understood from a theoretical point of view, the influence of
additional gas species in a mixture is rather complicated to describe. Different relaxation
interactions between the various molecules will lead to an effective relaxation frequency different
from that of the pure gas, and will influence the relaxation strength. Unfortunately, this influence
depends on the kind of gas and the concentration in the mixture.
So far, only attenuation has been addressed. In addition, two other affects need to be taken into
account. First, gas composition and pressure will change the density of the gas. This leads to a
change of the acoustic coupling efficiency of the ultrasonic wave from the transducer into the gas.
Higher pressure will result in better coupling and therefor lead to higher signal strength.
Second, there is an influence of the actual speed of sound on the beam characteristic transmitted
from the transducer. The beam width characteristic is dependent on the speed of sound. Reduced
speed of sound, compared to dry air, will sharpen the beam and concentrate more sound energy
into a smaller area. Affects of temperature on the speed of sound, or dispersion from relaxation
processes, have to be considered here also.
Both affects can be described easily and are incorporated into an application model.
3 GENERAL CONCLUSIONS FOR ULTRASONIC METER DESIGN FOR CO2 RICH STREAMS
Looking at formula (1) it can be seen that some general conclusions can be made with respect to
meter design. The values, which can be influenced to generate sufficient signal quality at the
receiver, are:
• Initial acoustic pressure coupled into the gas
• Transducer frequency
• Measuring path length
First, the acoustic pressure can be increased by a higher driving voltage at the transducer.
Unfortunately this voltage might be limited by explosion proof design requirements of the meter
installation, or by the breakdown voltage of the piezoceramics. As such this option is only a
possibility in special cases.
Second, the design of the transducer itself can raise the acoustic sound pressure level (SPL). This
is inherent in modern design principle compared with traditional design utilizing an epoxy
matching layer.
Nevertheless, a matching layer could further increase the acoustic coupling. Such a layer is made
of epoxy resins using hollow glass spheres, and its thickness is dependent on the working
frequency of the ultrasonic sensor. Unfortunately this matching layer could exhibit some
drawbacks like a pressure or temperature dependency, and also lowers the SPL. Therefore, the
acoustic matching layer should be left out if it is possible to achieve sufficient vibration
amplitudes at the sound emitting surface.
A stacked piezoelectric transducer in the form of a resonance converter is a viable alternative. A
metallic spring-mass-system is used to increase the amplitude at resonance (see Figure 2a). By
utilizing numerical optimization of mechanical and electrical parameters, it is possible to produce
sensors which exhibit:
• sufficient bandwidth for short signals with a very high amplitude (SPL)
• a maximum acoustic efficiency (efficiency converting electrical energy into sound
energy).
Figure 2a: Schematic diagram
Figure 2b: Acoustic spectrum of the
sound signal
Figure 2: Stacked ultrasonic transducer
This sensor concept is characterized by pure tone resonance mode and a well-defined working
range (see Figure 2b). There are several advantages:
• the electric energy is efficiently transformed into acoustic energy,
• the transducer is hermetically sealed and has a full metal housing,
• the bandwidth allows relatively short pulse signals.
A set of transducers with different operating frequencies, but identical installation dimensions, are
presented in Figure 3.
Figure 3: Ultrasonic transducer with different operating frequencies (80 kHz, 135 kHz, 208 kHz)
The absorption coefficient of the gas to be measured must be considered by having ultrasonic
transducers with various operating frequencies available to be installed into the same meter body.
Based on Figure 1a, it is obvious the optimum operating frequency pressure ratio should be below
50 kHz/bar for CO2. That means operating frequencies of less than 50 kHz at atmospheric
pressure. Unfortunately, such transducers would be relatively large and not suitable for typical
pipeline sizes less than 16 inch.
If a short path length is considered as a requirement for measuring high attenuating gases, only
direct path layouts should be used. The influence of doubling the path length is shown in Figure 4
which plots signal strength as a function of path length. Assuming a direct path length of 6 inches
(150 mm), it can be seen that the measured signal strength is above the noise level for this path
length. If, however, the path length is doubled to 12 inches (300 mm,) the signal strength is
below the noise level.
Resonator
Piezo-rings
Housing
Bolt
0.1 0.2 0.3 0.4
0
0.02
0.04
0.06
0.08
Noise level
Measurement signal
Path length [m]
Sig
nal
Str
ength
[arb
. u
nit
s]
Figure 4: Influence of path length on signal strength
Based on the knowledge of the transducer design and meter performance parameters, as well as
the attenuation characteristics of the gas or gas mixture under consideration, an application model
can be set up. Such a model will allow prediction of a maximum measuring path length for
operation of the meter within acceptable specifications and accuracy.
4 LABORATORY VERIFICATION OF THE ATTENUATION CHARACTERISTICS OF CO2-RICH
GASES
The primary consideration in the application model is the attenuation characteristics of the gas or
gas mixture. While these values are known for some pure gases, the parameters can not be found
for special mixtures. To extend the knowledge on this topic, a special measurement system was
developed. Measurements were carried out in a special pressure chamber equipped with a
moveable mount (see Figures 5 and 6) equipped with transducers of different operating
frequencies.
The tested transducers were 80 kHz, 135 kHz and 208 kHz standard probes as used in a
FLOWSIC600 ultrasonic gas flow meter. The distance between the transducers can be varied
between 0 and 10.2 inches (260 mm) from outside of the chamber. The gas within the test
chamber can be mixed for a maximum of three different gas components. The gas composition is
defined by mixing rules based on partial pressures corrected for real gas affects. Pressure and
temperature are measured during the tests. While pressure is a significant parameter for the
measurement target, temperature is measured for check purposes (stability of the gas atmosphere)
only.
Figure 5: High pressure chamber – outside view
Figure 6: High pressure chamber – inside view
Various gas mixtures were evaluated using this chamber. First, pure CO2 was measured to
confirm the theory. Afterwards different mixtures of CO2, CH4 and N2 where used in the
measurement tests. Table 3 gives the nominal and actual composition of the mixtures. Mixture 2
is a typical re-injection gas and mixture 3 simulates a natural gas with increased CO2 content.
Mixture CO2 CH4 N2 Balance
Nominal Actual Nominal Actual Nominal Actual Nominal Actual
1 100 99.9 0 0.1
2 60 59.4 40 40.3 0 0.3
3 12.5 13.1 85 84.8 2.5 2 0 0.1
Table 3: Nominal and actual composition of the mixtures
The attenuation coefficients measured for pure CO2 are plotted and compared with theoretical
values shown in Figure 7. There was very good correlation between the theory and the actual
measurements.
0
0.05
0.1
0.15Theoretical
Measured
Frequency/Pressure [Hz/bar(a)]
Att
enuati
on*
Wav
elen
gth
[ ]
1x103 1x104 1x105 1x106 1x107
Figure 7: Measurements and theory for normalised attenuation coefficient in Carbon Dioxide
The measured attenuation coefficients (shown as blue circles in Figure 7) were used to establish a
model of the frequency dependent attenuation as an input for the application model.
5 APPLICATION MODEL FOR AN ULTRASONIC GAS FLOW METER
The application model should result in the ability to predict the maximum working path length
depending on operating conditions (e.g. pressure, temperature, gas composition), and meter
parameters (e.g. operating frequency) based on easy to measure meter characteristics. Besides
attenuation, the affect of pressure on the acoustic coupling and speed of sound related acoustic
beam width of the transducer, have to be considered. Gas parameters, such as viscosity and
thermal conductivity, which are necessary for calculation, can be found in the literature [1] or can
be calculated for gas mixtures [2, 3].
The receiver sensitivity parameter AGC (automatic gain control) of an ultrasonic meter is a
measurement value inversely proportional to the acoustic pressure at the receiving transducer, and
therefore directly linked to the signal attenuation on the acoustic path. This parameter is meter
design specific, due to different path layouts, transducers and electronics design used by the
various meter manufacturers. Therefore, each manufacturer will have different pre-defined values
to estimate the quality of the measurement.
In this test, the FLOWSIC600 by SICK (Figure 8) was evaluated. This meter uses direct path
layout, advanced transducer technology with no epoxy matching layers, low noise receiver
electronics, and appropriate signal processing. Acceptable operation of the meter is ensured up to
an AGC level of 93 dB.
Figure 8: Path Layout for the 4 Path Meter
The model reference point for the acoustic signal strength is the behaviour in ambient air. Driving
voltage is limited in design to allow for operation in an explosion proof (hazardous gas)
installation. Attenuation, speed of sound and, gas pressure is well known. Transducer
characteristics, e.g. AGC levels for the different path lengths, are proven by standard procedures
in each meter. The pressure and speed of sound for the gas in the specific application is used to
calculate changes in sound pressure due to the operating conditions compared to ambient air.
Using the measured frequency dependent attenuation model, the changes in sound pressure due to
different attenuation characteristics can be modelled. Finally a maximum path length can be
determined for the specific application. Figure 9 gives an example for the gas mixture 2, e.g. the
re-injection type gas shown in Table 3. The maximum path length is plotted against the absolute
pressure of the application.
1 2 3 4 50
0.5
1
208 kHz
135 kHz
80 kHz
Pressure [bar(a)]
Cri
tica
l p
ath
len
gth
[m
]
Figure 9: Critical length vs. Absolute Pressure in gas mixture 2 (60%CO2+40%CH4)
80 kHz
135 kHz
208 kHz
For a specific ultrasonic flow meter design, the critical path length can be translated into the
maximum nominal pipe size for each transducer frequency. For each operating pressure a
maximum nominal pipe size can then be calculated. Figure 10 shows correlation for a 4-path
FLOWSIC600 ultrasonic gas flow meter. As an example, for 30 psig (2 bar(g)) operating
pressure, and 80 kHz transducer frequency, a maximum nominal pipe size of 32-inch should
result in acceptable meter operation and performance.
Figure 10: Application limits for a 4path meter in 60%CO2, 40%CH4 (mixture 2) type application
The model can be used to evaluate an application on a systematic basis with regard to the
maximum possible meter size at the given application’s operating conditions. For applications
which do not fall within this criteria, potential solutions such as reduced path lengths by using
protruding transducers, may need to be considered.
6 FIELD EXPERIENCE
Shown here is one of the first test installations, where the theory for using an ultrasonic meter for
high CO2 applications was demonstrated. This is a CO2-reinjection project. The application
utilizes a 4-path (8 inch) meter, operated at 928 psig (64 barg), 95 °F (35°C), and with a typical
gas composition as shown in Table 4.
CO2 CH4 C2H6 C3H8 N2
62.1 37 0.3 0.1 0.5
Table 4: Typical gas composition
The main goal of the installation was to demonstrate how an ultrasonic flow meter can handle
high amounts of CO2 and how the model can predict the performance of the meter in the
application.
To prove the metering concept and the model, a standard 2plex 4+1 meter [4] was installed. This
meter consists of a traditional 4-path custody transfer meter combined with an independent single
path check meter (Figure 11). The 2plex meter is typically used for custody transfer applications.
Example
for 80 kHz
It utilizes the traditional Westinghouse 4-path design, but incorporates a second, single path,
center-line meter in the same body. Changes in operating conditions (e. g. a blocked flow
conditioner or pipeline contamination) can be detected by the single path because it is very
sensitive to profile changes. If this occurs, there will be a difference in reported volumes.
Typically the 4 path meter continues to operate with virtually no impact on accuracy.
Figure 11: 2plex 4+1 Meter Design
For this special test case, the check system (single path) was equipped with a 135 kHz transducer
pair, while all other paths of the main system (custody) operated with 208 kHz transducers. The
data collected and presented in Section 4 of this paper predicted both frequencies should be
suitable. AGC levels of about 25 dB for the 135 kHz transducers and 28 dB (outer paths) / 35db
(inner paths) for the 208 kHz transducers where expected.
The meter was delivered without any high pressure calibration and installed as shown in the field
(Figure 12) without a flow conditioner. The meter’s flowrate was then compared with that of an
existing orifice meter located upstream.
Figure 12: Test installation set up
Figure 13 shows a screen capture of the waveform signals detected on the 135 kHz system and
Figure 14 the waveforms on the 208 kHz system outer path. Both systems run very stable and
show signals with very acceptable SNR and AGC levels.
Figure 13: Waveforms for the 135 kHz single path meter
Figure 13 shows the average AGC of 24 dB (135 kHz) and Figure 14 shows an AGC of 28 dB/34
dB for the 208 kHz. Both are very close to the expected values. While both systems show stable
operation, the signal of the 135 kHz transducer is 1.22 times higher compared to the inner path for
the 208 kHz.
Figure 14: Waveforms for Path 1 (outer) for the 208 kHz 4 path meter
In this example both transducer frequencies worked very well. Depending upon the application
(pressure, temperature and CO2 percentage) the standard 208 kHz transducer frequency would be
suitable for most typical applications.
Figure 15 shows a screen capture of all of the meter’s diagnostics.
Figure 15: Diagnosis session 4 path meter
The Performance chart in Figure 15 shows 100% signal availability. The SNR and AGC levels
are in a stable and normal range of operation. Speed of sound (SOS) differences between paths
show no significant deviations. The Profile indication and Velocity ratio charts indicate a slightly
asymmetric behaviour probably due to some residual swirl in the piping. The increased
Turbulence value (above the normal) on the outer paths is most likely due to the orifice and the
residual swirl upstream of the USM.
The comparison of the ultrasonic meter flow rate with that of the orifice meter (Figures 16 and 17)
shows an excellent correlation between the two in spite of the fact that the ultrasonic meter was
not even flow calibrated. The comparison between the orifice and the ultrasonic meter was within
0.2% during the test period (several weeks). This agreement was considered excellent.
Figure 16: 5 day daily flow rate comparison of ultrasonic flow meter with orifice meter
Figure 17: One month daily flow rate comparison of ultrasonic flow meter with orifice meter
7 CONCLUSION
Modern ultrasonic gas flow meters are capable of handling a broad range of metering applications
with high CO2 levels, and today even measurement of pure CO2 is possible. Advanced ultrasonic
transducer design, with high efficient acoustic energy coupling into the gas, are the basis for this
capability. Choosing transducer frequencies suitable for the application and specific operating
conditions will lead to successful installations. Since the received signal strength in high
concentrations of CO2 is more dependent on path length than in other gases, the use of reflection
type path layout, or any other extension of the path length, is not recommended.
To ensure accurate operation, the meter design and specifics related to the application, will need
to be thoroughly evaluated prior to selecting the meter. Knowledge of gas composition, pressure,
temperature, and meter size (based upon flow rate requirements), will be the basis for an
application model that can predict the performance of the meter in the field. The results predicted
by the model will help to optimize the meter parameters and will give recommendations for the
improvement of the installation design.
This paper confirms that the transducer’s performance (AGC gain) in the field can be compared to
that predicted from the laboratory test conditions. This correlation allows predicting other
applications much more reliably.
With the added diagnostic capabilities of the ultrasonic flow meter, even difficult applications like
custody transfer measurement of CO2-rich natural gas, or CO2 re-injection installations, are
possible.
8 REFERENCES
[1] Verein Deutscher Ingenieure, VDI Wärmeatlas Berechnungsblätter für den Wärmeübergang,
VDI Verlag, 1991, Düsseldorf, Germany
[2] Theo M. Geerssen, Physical properties of natural gases, N.V. Nederlandse Gasunie, 1980,
Groningen, The Netherlands
[3] Wolfgang Wagner, Describtion of the Gas Version of the Software Package for the
Calculation of Thermodynamic Properties from the GERG-2004 Wide Range Eduqation of State
for Natural Gases and Other Mixtures, Ruhr-Universität Bochum, 2007, Bochum, Germany
[4] John Lansing, The relevance of two different path layouts for diagnostic purposes in one
ultrasonic meter, North Sea Flow Measurement Workshop, 2007, Oslo, Norway
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
OIML R 137-1, the first ultrasonic meter to be tested to accuracy class 0.5?
Skule Smørgrav, FMC Kongsberg Metering Atle K. Abrahamsen, FMC Kongsberg Metering
1. INTRODUCTION Over the past 20, 10 and 5 years gas production world wide has been on an ever increasing rise. At the same time measurement of gas has been changing from the traditional turbine and orifice meters to the ultrasonic meters. Multi-path gas ultrasonic meters have by now become the preferred device for custody transfer measurement. The first step in an international acceptance of these state of the art technology based devices was probably the first edition of the AGA Report number 9 which was released in June 1998. This report was updated and the second edition was released in April 2007. The AGA Report No. 9 has since been used all over the world as the reference “standard” when ultrasonic meters have been specified for most allocation and custody transfer projects. In Europe there has been a working group in session for several years working on a corresponding ISO standard for Ultrasonic meters, and the ISO 17089 will hopefully be officially released in 2009. But, there is another standardization organ in Europe which released a gas meter recommendation in 2006 which is also applicable to ultrasonic meters – the OIML R 137-1. This creates a third “standard” which manufactures of gas ultrasonic meters may be asked to follow. Already a number of meters have been tested to the defined accuracy class 1 in OIML R 137-1, but FMC has now fully tested what we have been informed is the first multi-path gas ultrasonic meter to the accuracy class 0.5 together with the PTB of Germany. This paper will describe the key items of OIML R 137-1, point to the differences and similarities between OIML R 137-1 and AGA-9 and highlight certain limitations/shortcomings in OIML R 137-1. The second part of the paper will describe the detailed tests required for accuracy class 0.5, show the results from multiple size meters tested and comment on how the industry and standardization committees can work even closer together to get even more applicable and relevant standards and test methods. The latter to achieve more repeatable, comparable and usable results across the industry, and across the different geographical areas of our “shrinking” planet.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
2
2. OIML R 137-1, AGA 9 and ISO 17089 The AGA Report No. 9, Measurement of gas by Multipath Ultrasonic Meters, Second Edition was released in April 2007. It has now been in use around the world for more than 10 years since the initial release in June 1998, and it has probably been the most influential document for ultrasonic meters. It has been adopted by most countries around the world and many national regulations and oil and gas company specifications have adopted the requirements described in AGA 9. The yet to be released ISO 17089 also has the AGA 9 as its base even though it will differ in certain areas and be more comprehensive when it eventually is released. The intent was to also discuss ISO 17089 in this paper, expecting it to be released at this time, but since it has not been released, and the final version is not ready, it will not be further discussed here. The OIML R 137-1 Edition 2006, International Recommendation Gas Meters, is a more general recommendation which does not target ultrasonic meters directly but sets forth the same requirements to be met by all devices designed to measure “quantities of gaseous fuels or other gases, except gases in the liquefied state and steam” [2]. This recommendation supersedes the previous versions of R 31 (1995) and R 32 (1989) and partially supersedes OIML R 6 (1989). Both these “standards” put forth a number of requirements that gas/ultrasonic meters must meet and many of them are fairly similar but the main difference between the two is that AGA 9 has a section to be followed for each produced and delivered meter, placing requirements on the performance of the meter before, during and after a flow calibration. In addition it has a number of type approval tests taken from OIML R 6 and OIML D 11, specifically for the electronics. The OIML R 137-1 on the other hand is designed to be used for type approval testing and not as a reference for testing every meter produced. It uses the same OIML D 11 for Environmental tests for electronics or devices but has in addition a range of other tests, including flow tests to be performed on a selection of sizes or worst case - most difficult scenarios. 2.1 AGA Report No. 9 The primary performance requirements in AGA 9 are described in chapter 5. It says that “When a meter is flow calibrated, it shall meet the minimum measurement performance requirements detailed below before the application of any calibration-factor adjustment” [1]. Small meters (below 12”):
Maximum error is ± 1.0% for qt ≤ qi ≤ qmax ± 1.4% for qmin ≤ qi < qt Maximum Peak-to-Peak Error is 1.0% for qt ≤ qi ≤ qmax 1.4% for qmin ≤ qi < qt
Large meters (12” and above): Maximum error is ± 0.7% for qt ≤ qi ≤ qmax ± 1.4% for qmin ≤ qi < qt Maximum Peak-to-Peak Error is 0.7% for qt ≤ qi ≤ qmax 1.4% for qmin ≤ qi < qt
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
3
In addition to the specific flow calibration criteria they have a few general requirements for repeatability, resolution, velocity sampling interval, zero-flow reading, speed of sound deviation and maximum SOS path spread. In Appendix B AGA 9 lists all the OIML R 6 and D 11 electronics design testing which is to be performed on one meter. They are: B.1 Static Temperature, Dry Heat B.2 Static Temperature, Cold B.3 Damp Heat, Steady State B.4 Damp Heat, Cyclic B.5 Random Vibration B.6 Sinusoidal Vibration B.7 Mechanical Shock B.8 Power Voltage Variation B.9 Short Time Power Reduction B.10 Bursts (Transients) B.11 Electrostatic Discharge B.12 Electromagnetic Susceptibility 2.2 OIML R 137-1 The R 137 specifies a number of different accuracy classes for different applications. It states that “Gas meters shall be classified into the Accuracy Classes given in Table 2. The errors shall be within the applicable values given in Table 2” [2]. Table 2 is the table given below and the requirements are applicable to all meters produced. Table 2 Maximum permissible errors of gas meters
On type approval and initial verification Accuracy Class
In-service * Accuracy Class
Flow rate Q
0.5 1 1.5 0.5 1 1.5
Qmin ≤ Q < Qt ± 1% ± 2% ± 3% ± 2% ± 4% ± 6%
Q t≤ Q ≤ Qmax ± 0.5% ± 1% ± 1.5% ± 1% ± 2% ± 3%
* Note: National Authorities may decide whether they will implement in-service maximum permissible errors or not. Annex A lists their OIML R 6 and D11 tests to be performed on one type approval sample: A.4.1.1 Static Temperature, Dry Heat (non condensing) A.4.1.2 Static Temperature, Cold A.4.2.1 Damp Heat, Steady State (non condensing) A.4.2.2 Damp Heat, Cyclic (condensing)
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
4
A.5.1 Vibration (random) A.5.2 Mechanical Shock A.6.1.1 Radiated, radio frequency, electromagnetic fields A.6.1.2 Conducted radio-frequency fields A.6.2 Electrostatic Discharge A.6.3 Bursts (transients) on signal, data and control lines A.6.4 Surges on signal, data and control lines A.7.1 DC mains voltage variation A.7.2 AC mains voltage variation A.7.3 AC mains voltage dips, short interruptions and voltage variations A.7.4 Bursts (transients) on AC and DC mains A.7.5 Surges on AC and DC mains lines A.8 Low voltage of internal battery (not connected to the mains power) As can be seen these are essentially the same as specified in AGA 9. In addition to the above tests the type approval verification process listed the below tests to be performed from chapter 7.4:
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
5
Clauses 7.4.9, 7.4.10 and 7.4.11 are not applicable to ultrasonic meters, ref Annex C. Clause 7.4.8, refers to Annex B: Flow Disturbance Tests and consist of testing for mild flow disturbances in accordance with the figures below in B.2 [2]:
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
6
And severe flow disturbances using a half pipe, or half moon, plate installed between the two elbows with the opening toward the outside radius of the first bend as described in B.3 [2] and shown in the figure below.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
7
For temperature there are 3 different options:
We did the testing according to the 2nd method; “monitoring the unsuppressed flow rate output of the meter at no-flow conditions at different temperatures (for electronic meters.)” [2] This results in the following type approval test matrix with responding performance criteria: Test Clause 0.5 Accuracy class 1.0 Accuracy class Error (WME=Weighted mean error)
7.4.2 Err < 0.5% for Q>=Qt Err < 1.0% for Q<Qt WME <= 0.2%
1.0% for Q>=Qt 2.0% for Q<Qt WME <= 0.4%
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
8
Reproducibility 7.4.4 Standard deviation: < 0.075
Standard deviation: < 0.15%
Orientation and flow direction
7.4.5 Same as error test Same as error test
Working pressure 7.4.6 Shift less than 0.25% and Err < 0.5% for Q>=Qt Err < 1.0% for Q<Qt
Shift less than 0.5% and Err < 1.0% for Q>=Qt Err < 2.0% for Q<Qt
Temperature 7.4.7 See above See above Flow disturbance 7.4.8 Shift less than 0.165% Shift less than 0.33% Interchangeable components
7.4.14 Shift less than 0.165% and Err < 0.5% for Q>=Qt Err < 1.0% for Q<Qt
Shift less than 0.33% and Err < 1.0% for Q>=Qt Err < 2.0% for Q<Qt
It should be noted here the very stringent requirements in 7.4.8 of less than 0.165% for all the disturbance tests for the 0.5 Accuracy Class. As opposed to the AGA 9 then one can use the OIML R 137-1 to present an “approval” in the form of a third party certified OIML test report, in the same mould as has been normal practice with for example OIML R 117-1 for liquid meters.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
9
3. The MPU 800 The MPU 800 is a four path gas ultrasonic meter intended for use in custody transfer applications. It is the little brother of the well known MPU 1200 six path master meter and has been designed as the most cost effective custody transfer solution for applications with less complex upstream piping configurations.
The meter uses the well known four path configuration with four paths in four parallel planes as shown in the figure below.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
10
It uses the same transducers, electronics, software and signal processing as the rest of the meters in the FMC gas and liquid ultrasonic meter family, the single path MPU 200, the three path MPU 600 and the six path MPU 1200 for gas, and the four path Ultra4 and six path Ultra6 for liquids. The meter is available in a large range of materials, with different flange types and from 4” to 56” in size. 4. OIML R 137-1 testing of the MPU 800 To ensure a valid test report with results that are representative for the whole range of meter sizes available, three different meters were testes, a 4”, an 8” and a 12”. The 12” was tested at the Advantica facility, now GL Industrial Services, in the UK. The 8” was tested at both the Lintorf and the Pigsar facilities operated by Ruhrgas is Germany. And the 4” was tested at Pigsar. Multiple tests were completed to satisfy both the requirements in OIML R 137-1 and to verify meter performance under other frequently encountered and requested conditions:
• Traditional flow test of all 3 meters • Without Flow Conditioner (FC) • With FC at 3D and at 5D • Mild (low) flow disturbance and severe (high) flow disturbance testing as defined in
Annex B of OIML R 137-1 4.1. Traditional flow test of all 3 meters Below are the results from the traditional flow tests of all the three 12”, 8” and 4” meter at the respective calibration facilities. As can be seen they all fall well within the requirements in OIML R 137-1.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
11
Flow calibration of 8 inch MPU 800 Pigsar - Dorsten Test Facility, 18-Mrz-2009.
S/N 2324-MPU-4501
-2.00
-1.50
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
0.00 5.00 10.00 15.00 20.00 25.00 30.00
Gas velocity (m/s)
Ave
rage
dev
iatio
n (%
)
Flow calibration of 12 inch MPU 800 Advantica - Bishop Auckland Test Facility, 03-Feb-2009.
S/N 2274-MPU-4385
-2.00
-1.50
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
0.00 5.00 10.00 15.00 20.00 25.00 30.00
Gas velocity (m/s)
Ave
rage
dev
iatio
n (%
)
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
12
Flow calibration of 4 inch MPU 800 Pigsar - Dorsten Test Facility, 18-Mrz-2009
S/N 2292-MPU-4501
-2.00
-1.50
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
0.00 5.00 10.00 15.00 20.00 25.00 30.00
Gas velocity (m/s)
Ave
rage
dev
iatio
n (%
)
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
13
4.2 Different flow conditioner setups The three different flow conditioner tests were all performed at Lintorf using the 8” meter. The figure below shows the results from the 3 tests; without FC, with the FC at 5D and with the FC at 3D. As can be seen all the results were well within the OIML R 137-1 requirements.
Flow test Lintorf 8" Different FC setups
-1,50
-1,00
-0,50
0,00
0,50
1,00
1,50
0,00 5,00 10,00 15,00 20,00 25,00 30,00 35,00
Flow velocity (m/s)
Erro
r (%
)
8" with FC 5D8" without FC8" with FC 3D
4.3 Flow disturbance tests The flow disturbance tests were also performed using the 8” meter at the Lintorf facility. As can be seen in the two graphs below a total of 6 tests were done. First we did the tests as described in the OIML R 137-1 in Annex B with the meter installed 5D after the flow conditioner. In addition to the low and high disturbance tests we also tested the high disturbance case with the MPU 800 rotated 90° to verify the performance with the bends originating in any plane.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
14
The MPU 800 performed flawlessly at 5D and the results were all within the 0.165% requirement for Accuracy Class 0.5. As the results were so good with the disturbances placed 5D upstream of the meter we decided to push the meter and to see where the boundaries of the performance was with respect to the low and high disturbances were. As showed in the figure below we found that the MPU 800 still complies with the OIML R 137-1 requirements for Accuracy Class 1.0, within 0.33%, if used: - in a low disturbance application at 10D without a flow conditioner - in a low disturbance application with a flow conditioner at 3D - in a high disturbance application with a flow conditioner at 3D
Disturbance testing 5D
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00
Flow velocity (m/s)
Erro
r shi
ft (%
)
Low disturbance 5D FC
High disturbance 5D FC
High disturbance 5D FC MPU800 90° Rotation
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
15
4.4 Repeatability The below figure show the repeatability numbers for all the flow tests to verify the performance numbers listed in the MPU 800 specifications.
Repeatability 8" MPU 800 Lintorf
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00
Flow velocity (m/s)
Rep
eata
bilit
y
8" without FC
8" with FC 5D
8" with FC 3D
Repeat 8" with FC 5D
Repeat 8" without FC
Repeat 8" with FC 3D
3D FC Low disturbance
3D FC High Disturbance
Low disturbance 5D FC
High disturbance 5D FC
High dise 5D FC 90° RotationLow disturbance 10D w/o FC
Disturbance testing 10D without FC and 3D with FC
-0.5-0.4-0.3-0.2-0.1
00.10.20.30.40.5
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00
Flow velocity (m/s)
Erro
r shi
ft (%
)
Low disturbance 10D without FC
Low disturbance 3D FC
High disturbance 3D FC
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
16
4.5 Interchangeable components To show that the key components of the MPU 800, the electronics and the transducers, can be changed without significant effect on the meter performance the following results were obtained:
• Transducer exchange: – 1 transducer pair changed (Path 4) - Error shift: 0.09% – 2 transducer pairs changed (Path 3 and 4) - Error shift: 0.00% – Back to original transducers – Error shift: 0.06%
• Electronics change UDSP (Computer) and UAFE (analog front end): – Both boards changed - Error shift: 0.00% – Both boards changed back to original - Error shift: 0.00%
4.6 Environmental tests All the electronics tests as specified in AGA 9 were performed on the MPU 1200 several years ago and did not need to be repeated. The additional requirement in OIML R 137-1 was to test the full meter and monitor zero drift for the cold and dry heat conditions.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
17
Low temperature (-25C): Theoretical Flow rate Theoretical error including
at zero flow error at
Qmin error at flow
test Requirement
m3/h % % % Before (+20C) 0.017 0.11 0.31 1.0
After 2 hours at -25C 0.012 0.08 0.28 1.0 1 hour recovery to +20C 0.016 0.11 0.31 1.0
High temperature (+70C): Theoretical Flow rate Theoretical error including
at zero flow error at
Qmin error at flow
test Requirement
m3/h % % % Before (+20C) 0.016 0.11 0.31 1.0
After 2 hours at +70C 0.022 0.15 0.35 1.0 1 hour recovery to +20C 0.009 0.06 0.26 1.0
To make this test relevant for the full size range of MPU 800 meters the most difficult size was chosen, the smallest - a 4” MPU 800. The tests were performed by the PTB at the Institute fur BSFV test facility in Hamburg Germany and as showed in the table above the meter were well within the requirements. Considering that this is the shift in output under essentially zero flow conditions, where any minute change will have a very large percentage influence, the above results support our claim that the FMC MPU ultrasonic meters are fully independent of any temperature influence. This is achieved thru the unique reciprocal design of the electronics and transducers, and the advanced temperature compensation implemented in all the MPU series ultrasonic gas flow meters.
27th North Sea Flow Measurement Workshop 20th – 23rd October 2009
18
5. CONCLUSIONS The current international ultrasonic gas meter “standards” all try to achieve the same result – help users ensure they get high quality measurement from high quality meters. Unfortunately they all try to achieve this with similar but yet different specifications. As has been showed above a four path meter has been tested in various sizes and successfully passed all the test requirements for an OIML R 137-1 accuracy class 0.5 meter. This was achieved using some test results from previous testing based on AGA 9 but a full compliance required further tests to be completed. The world needs standards, this industry needs standards, and manufacturers and users all applaud the existence of standards. However, we encourage further cooperation between the standardization committees, industry stake holders and national organizations because standards in gas ultrasonic today can seem to be on its way to creating conflicting interests and recommendations. We have applications today where we as a manufacturer is being asked to be in accordance with two international recommendations, one national standard and one set of company guidelines – and quite frequently these 4 ask for quite a number of different results – although they all ultimately want the same thing – accuracy! REFERENCES [1] AGA-9, “Measurement of gas by ultrasonic meters”, A.G.A. Report no. 9, American Gas
Association, Transmission Measurement Committee (June 1998 and April 2007). [2] OIML R 137-1, “International Recommendation Gas Meters, Part 1: Requirements”,
Organisation Internationale De Mètrologie Lègale (Edition 2006)
ULTRASONIC METER CONDITION BASED MONITORING – A FULLY
AUTOMATED SOLUTION
George Kneisley, Transwestern Pipeline, Roswell California, USA
John Lansing, SICK Inc., Houston, Texas USA
Toralf Dietz, SICK AG, Germany
1. INTRODUCTION
The customer requires a fiscal meter that measures with highest reliability within the required
accuracy limits throughout the life time. Whenever this requirement isn’t fulfilled due to changed
process/flow conditions or changes to the meter, the user needs to be warned in real-time. To
ensure such warning, the diagnosis parameters implemented into modern ultrasonic flow meter
can be useful. Since the introduction of the global diagnosis concept major improvements in
diagnosing a USM have been achieved. This requires a thorough understanding of the meter’s
operation and also understanding what normal, and non-normal responses of all diagnostic
parameters are in order to insure proper operation. The automated diagnostics will monitor, and
alarm, on all important parameters such as Profile Factor, Symmetry, SNR, Turbulence, etc.
These warnings are today an important factor for driving the condition based maintenance of the
installation. Additionally, it is very important to have a long-term history of these diagnostics in
order to properly determine if a meter is still operating accurately.
Beside this features inherent to every ultrasonic flow meter with a multiple number of paths
additional concepts to compare measurements directly exist. Two main concepts can be realised -
permanent serial metering with two independent fiscal meters or with a combination of a fiscal
and a check meter, introduced by TransCanada Pipeline (TCPL) several years ago. This concept
involves using a single path USM downstream of the fiscal multipath meter. Papers have shown
that single path USM meters are significantly affected by abnormal measurement conditions such
as flow conditioner blockage, pipeline contamination from oil and mill scale, and any other
change in operation that impacts accuracy. Since the single path meter has significantly more
sensitivity, comparing the uncorrected readings of both meters provides a simple solution for
determining if the fiscal meter is still operating accurately. If both are in agreement, then
measurement must be OK. Should the two meters deviate, then more than likely there is some
condition which might impact the accuracy of the fiscal meter. This paper will discuss the results
and benefits in terms of reliability and economic impact of the TCPL-method installed in various
field applications. Data will be presented on dirty vs. clean meters to show that the single path
meter shifts significantly in a dirty environment while the 4-path custody meter is relatively
insensitive to this.
2. AUTOMATED METER DIAGNOSTICS
A fully automated USM requires having all aspects of the meter’s diagnostics monitored on a
real-time basis within the meter. Traditionally users have collected periodic files and performed
manual analysis of these to determine if the meter is operating correctly. That is no longer needed
with today’s technology.
All USMs have a variety of diagnostics to help the user determine if the meter is operating
correctly. The basics for a typical chordal meter include the following:
• Gain for each transducer
• Performance (percent accepted signals by path)
• Signal to Noise Ratio (SNR)
• SOS for each path
• Velocity profile (path ratios or velocities)
Today these 5 are enhanced by additional diagnostics for the user to better understand if the
meter is operating correctly. These include the following:
• Profile Factor
• Symmetry
• Turbulence
Profile factor and Symmetry are methods of analyzing the velocity profile (or path ratios). They
are values which reflect the shape and amount of distortion in the basic velocity profile. They
have been used for years to help understand the velocity profile which is often considered by
many to be the most difficult diagnostic to understand [Ref 2 & 12]. These two are defined as
follows:
• Profile Factor = (Path 2 + Path 3) / (Path 1 + Path 4)
• Symmetry = (Path 1 + Path 2) / (Path 3 + Path 4)
Many consider a meter’s velocity profile the most difficult to understand because it can vary due
to installation and type of upstream piping components, and to some degree, flow rate. Generally
speaking these two diagnostics (Profile Factor and Symmetry) are very stable above a velocity of
approximately 2-3 m/s.
However, this is only true if the piping design incorporates a flow conditioner. Without a flow
conditioner, the velocity profile will have a wide range of both Profile Factor and Symmetry, and
thus very difficult, if not impractical, to monitor and alarm on these value.
Figure 1 is an example of a profile with a flow conditioner (CPA type plate) and thus has an idea
profile. It was collected at the calibration facility and had a flow conditioner upstream (at 10D)
along with perhaps 50D of straight piping. Figure 2 shows a velocity profile of the same meter
without a flow conditioner when it is subjected to piping that had 3 elbows and a tee upstream.
Figure 1: Path Ratios with CPA Conditioner Figure 2: Path Ratios w/o CPA Conditioner
Clearly the profile in Figure 2 is significantly distorted. This amount of profile change will
impact the accuracy of the meter. The intent of velocity profile analysis is to identify if there are
any changes from the baseline. If Figure 1 were the baseline profile in the field, and a future
inspection revealed Figure 2, then obviously there is something wrong internally and would
warrant inspection.
Turbulence is a third advanced diagnostic, and is a measure of flow stability, or variability, of
each path’s velocity reading. It is presented in percentage by path. Values are computed from the
variability of all of the transit-time readings (per path) and updated once per second. More
information is available from previously published papers [Ref 2, 11 & 12].
Typically a chordal meter will have a turbulence value of 2-3% for Paths 2 & 3, and 3-4% for
Paths 1 & 4. The Turbulence values are higher for the outer paths (Paths 1 & 4) as they are
located closer to the pipe wall and thus there is some influence due to the surface of the piping.
Even more advanced diagnostics are provided by some meters and include the following:
• High meter velocity (exceeding programmed limit)
• Power supply voltage too low (below a programmed value)
• Logbook(s) warning when full of un-acknowledged entries
These previously identified advanced diagnostics, and more, can now be fully automated in the
meter by having site-specific programmed values for each diagnostic parameter. There are many
benefits to having these programmed in the meter rather than in the User’s software or in a flow
computer. Knowing within minutes that a problem may exist significantly reduces measurement
uncertainty. Some of the many benefits include:
• The meter monitors all parameters/diagnostics on a real-time basis. When a diagnostic is
approaching a limit, the meter then presents an alarm that can be remotely monitored.
• If diagnostic limits established in meter’s software (User interface), the only time the user
will know there is a problem is when they are connected to the meter. As a result, the
meter can operate for weeks with a problem before a technician might identify it.
• Each location, and meter size, will have different limits due to the station design, meter
size, pressure, flow rates, etc. By having these programmed in the meter, they can be
tailored to the site-specific conditions and thus present an alarm more quickly and
accurately.
• One additional benefit of having the meter do this, rather than the flow computer or RTU,
is that no additional programming is required. Simply monitor a digital output (DO) and
the RTU can then report if there is a problem with a diagnostic. This saves the user time
and money and makes implementing this feature very easy.
To proof the feasibility of the concept to practical installations a special test was performed at
CEESI facilities.
All testing was performed with a significant length of straight piping upstream of the metering
package. This upstream length of straight pipe would present a very symmetrical and non-
swirling profile to the CPA 50E flow conditioner.
Figure 3 shows the 12 inch meter installed at the CEESI facility for the testing.
Figure 3: 12-inch 4+1 CBM Meter CEESI
A CPA was installed 10D upstream for baseline testing. After the baseline testing was complete
the 12-inch CPA was partially covered with duct tape (40% blockage). Figure 4 shows the 40%
blocked flow conditioner prior to testing.
Figure 4: 12-inch CPA with 40% Blockage
Duct tape was used to block the holes as this provides a repeatable method of testing and can
withstand the pressures created by the flow rates used.
The example shows the velocity profile changed from symmetrical to somewhat distorted
towards the bottom of the meter (compare Figure 5 to Figure 6).
Flowing Velocity Ratios at 66.4 ft/s
0.919
1.020
1.017
0.915
0.80 0.85 0.90 0.95 1.00 1.05 1.10
Path 1
Path 2
Path 3
Path 4
Path Ratios
Path Velocity Ratios at 66.5 ft/s
0.885
1.000
1.039
0.945
0.80 0.85 0.90 0.95 1.00 1.05 1.10
Path 1
Path 2
Path 3
Path 4
Path Ratios
Figure 5: Path Ratios without Blocked CPA Figure 6: Path Ratios 40% Blocked CPA
Figure 5 shows the velocity profile in normal conditions with no blockage. The Profile Factor
was 1.111 which is normal for this meter design, and the Symmetry was 1.004, or within 0.4% of
a perfectly symmetrical profile. Meter data was collected at three velocities of approximately 7,
14, and 21 m/s. The data here is from the 21 m/s (64 fps) flow rate, but all three velocities
essentially looked the same.
Figure 6 shows the distortion that occurred as a result of the blockage. The Profile Factor
changed to 1.114, or only about 0.4% shift. This amount of change is relatively small compared
to the baseline Profile Factor of 1.111, and if the technician were only monitoring Profile Factor,
they would assume everything is OK. However, it is apparent these two graphs do not look the
same.
This is the reason it is very important to also monitor the second diagnostic called Symmetry. For
the 40% blocked condition the Symmetry is 0.950. This represents about a 5.5% shift in the
velocity profile towards the bottom of the meter (high velocity average). This magnitude of
change is very obvious and easily identified. By monitoring within the meter on a real-time basis,
it can be reported to the RTU or flow computer within minutes as an abnormal profile.
One can argue that 40% blockage is very significant and not likely to occur often. Let’s look at
the profile with only one hole blocked. Figure 7 is a picture of the CPA prior to the testing.
Figure 7: 12-inch CPA with 1 Hole Blocked
The single blocked hole was located on the bottom of the meter. This location was chosen as the
most likely place for such a blockage to occur. Figure 7 shows the profile with the one hole
blocked at the bottom.
Flowing Velocity Ratios at 66.8 ft/s
0.898
0.992
1.037
0.954
0.80 0.85 0.90 0.95 1.00 1.05 1.10
Path 1
Path 2
Path 3
Path 4
Path Ratios
Figure 8: Path Ratios with One CPA Hole Blocked
Figure 8 shows a similar distortion as with 40% blockage. For this condition the Profile Factor
was 1.096 and the Symmetry was 0.949. In this case the Profile Factor changed about 1.5% and
the Symmetry almost 6%. Thus it is fairly easy to see that these two profiles (blocked vs. not
blocked) create a flow profile (Path Ratios) that can be seen visually, and thus can be alarmed on
via the meter’s firmware.
A typical alarm limit on Profile and Symmetry, based upon these test conditions, and other
testing, would suggest something between 3% and 5% tolerance for both the Profile Factor and
Symmetry values. Thus, when this meter was installed in the field, it would have a baseline
factor of perhaps 1.11 for Profile Factor, and 1.005 for Symmetry, and the Warning alarm limits
would be established at 5% for both. Thus if either the Profile Factor or Symmetry values were
outside of the 5% limit, the meter would then report a Warning alarm.
Another very valuable tool for analyzing flow conditioner blockage is monitoring Turbulence. In
the case of the baseline condition, the Turbulence values are typically around 2-3% for the
middle paths (Paths 2 & 3) and 3-4% for the outer paths (Paths 1 & 4). These numbers are based
upon many tests in the calibration lab and results from field installations.
Figure 9 shows the baseline results at 20 m/s with no CPA flow conditioner blockage, and Figure
10 with 40% blockage of the flow conditioner.
Turbulence at 66.4 ft/s
2
4
6
8
20
40
60
80
1
10
100
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time (sec)
Tu
rbu
len
ce
(%
)
Path 1 Path 2 Path 3 Path 4
Turbulence at 67.2 ft/s
2
4
6
8
20
40
60
80
1
10
100
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time (sec)
Tu
rbu
len
ce
(%
)
Path 1 Path 2 Path 3 Path 4
Figure 9: Turbulence without Blocked CPA Figure 10: Turbulence 40% Blocked CPA
Figure 9 shows turbulence values that average around 2.8% for the outer paths and around 1.9%
for the inner paths. Figure 10 has average values in the range of 7-10% for the outer paths and 5-
6% for the inner paths. Clearly there is a significant difference in the baseline (Figure 9) and the
blocked condition (Figure 10). Besides blockage, Turbulence can also identify a cyclic flow
control valve or significant pulsation. In these cases Turbulence values can exceed 50%
depending on the frequency and amplitude of the pulsation or flow control valve cyclic
operation.
Based upon testing in the lab and field results, a setting of perhaps 5 or 6% for the field appears
to be a viable limit for identifying problems. In the case of the single blocked flow condition,
Figure 11 shows what the Turbulence would be under this condition.
Turbulence at 66.8 ft/s
2
4
6
8
20
40
60
80
1
10
100
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150Time (sec)
Tu
rbu
len
ce
(%
)
Path 1 Path 2 Path 3 Path 4
Figure 11: Turbulence with 1 Hole Blocked CPA
Figure 11 shows the Turbulence doesn’t look much different than Figure 9. Analysis of the data
shows the average for Paths 1 & 4 was about 3.1% and for Paths 2 & 3 the average was about
2.4%. In this example if the technicians were only looking at Turbulence as an indicator for
blockage, it probably would have been over-looked since averages were not that significantly
different. This example shows why it is important to monitor all aspects of the diagnostics.
When a technician connects to the meter using the factory software, it simply polls the meter for
all the diagnostics and presents this in a graphical format. Each manufacturer has developed
software in order to display the diagnostics for the user. In this case the software not only
presents all the diagnostics, but also shows the alarm limits that have been established in the
meter. Figure 12 shows a screen capture of a meter with all diagnostics within normal limits.
Figure 12: Diagnostics for a Meter with No Problems
Figure 12 shows all the diagnostics the user needs to monitor. The Velocity Profile is the graph
that starts at the top left, and from there the graphs, from left to right on the top row, include SOS
difference by path (in percentage), Performance by path (in percentage), AGC (Gains in dB) by
transducer, and on the second row of graphs there is SNR (in dB), Turbulence (in percentage)
and a Profile Indication Graph (more on that later).
The red dotted lines indicate the alarm limits that are programmed in the meter. Once any
diagnostic exceeds (or in the case of Performance and SNR go below) the programmed limit, the
individual graph turns yellow and the meter then reports via Modbus, and a digital output (DO),
that a Warning alarm is activated.
Figure 13 shows an example of a condition where there may be contamination of the Path 4
transducer pair.
Figure 13: Possible Blockage or Contamination in Front of Path 4
Figure 13 shows the SOS for Path 4 is exceeding the red line (Warning alarm limit), and also the
AGC gains for Path 4 transducers are above the Warning alarm limit (Chart shows warnings in
yellow). In this condition the meter is now reporting a problem, and when the technician
connects to the meter, it is very apparent there is a problem. For this example there may only be a
slight impact on accuracy, but if left un-attended it might mean that Path 4 could eventually fail.
Figure 14 shows a different condition which may be caused by a blocked flow conditioner.
Figure 14: Possible Blocked Flow Conditioner
In Figure 14 Turbulence in the graph is above the alarm point and it is reported as a yellow bar
for Path 4. This graph was developed with the help of a demo model and thus only one path (Path
4) is reporting high turbulence due to the blockage being very close to the meter. In the real
world more than one path would probably be reporting an alarm (many bar graphs would be
yellow). The velocity profile is also distorted as can be seen in the Profile Indication.
Understanding the Path Ratios has perhaps been the most difficult diagnostic to understand.
Previously in this paper two advanced diagnostics were explained. One is Profile Factor and the
second is Symmetry. These two methods of analyzing the Path Ratios greatly simplify
understanding if the meter’s profile is normal. In the example of Figure 14 we can see the line in
the Profile Indication is yellow. The following two graphs show a larger version of a normal
Profile Factor and Symmetry (Figure 15), and an abnormal Profile Factor and Symmetry in
Figure 16 when there is blockage in front of the flow conditioner.
Figure 15: Normal Profile – No Warning Figure 16: Profile Distorted – Warning Reported
For each meter installed in the field the Profile Factor and associated Warning alarm limits can
be programmed for the site-specific conditions. For the example in Figure 15 the normal Profile
factor is 1.11 as shown on the X axis, and 1.00 for Symmetry on the Y axis. This is represented
by the dot in the middle of the red box. This dot stays in this location to indicate what the normal
values are. The size of the box is also programmed in the meter, and in this case a 5% tolerance
(plus and minus for both Profile Factor and Symmetry) from the baseline has been determined as
the limits. Thus the size of the box is ±5% from the baseline values.
The dot at the end of the line (towards the top of the red box in Figure 16) represents the current
average for the Profile Factor and Symmetry. The current value of Profile Factor is 1.133 and the
Symmetry is 1.038 as shown in the dialog boxes to the right of the graph. The dot is near the red
box but inside of it so the meter is not reporting any Warning alarms.
In Figure 16 the dot at the end of the line is outside of the box. This causes the line to turn yellow
and indicates a Warning alarm, and thus is reported via Modbus and the DO. In this example the
Profile Factor is 1.216 or about 10% from normal, and the Symmetry is 1.08 or about 8% from
normal. The velocity profile is perhaps the most important, and most difficult, diagnostic to
understand. This combined graphical representation of Profile Factor and Symmetry (Profile
Indication) greatly simplifies understanding the velocity profile and makes it very easy to
identify if the meter is operating within normal limits or not.
3. DIAGNOSIS ON READINGS OF TWO DIFFERENT METERS CBM 2PLEX
4+1 DESIGN
The first part of the CBM 2Plex 4+1 meter design is a conventional fiscal 4-path chordal
ultrasonic meter. The meter incorporates an additional, independent single-path meter and
associated electronics incorporated into the same body. The purpose of the additional path is for
continuous comparison of volumes to the fiscal 4-path meter’s measurement results.
The transducers for the independent single path are located in such a fashion as to traverse the
meter in the center of the meter body. The transducers for the fiscal 4-path meter are located in
the traditional Westinghouse configuration. The reason for locating the single-path in the middle
(center vertically) is to put it in the most profile-sensitive measurement position of the meter.
This will result in a difference in volumes between the single path and the fiscal 4-path when the
velocity profile changes. That is, the single-path meter, with the sensors located in the center of
the flowing gas, is more sensitive to flow disturbances than the 4-path meter design.
These disturbances (velocity profile changes) can be caused by several external factors including
partially blocked flow conditioners and pipeline contamination. All of these will cause a change
in the velocity profile seen at the meter. This concept works because changes in profiles
significantly impact the reading by the centrally located single path while having very little affect
on the 4-path meter. Figure 17 is an artist drawing of this design. Figure 18 shows an 8-inch 4+1
meter with plastic covers over the transducer mounting area.
Figure 17: CBM 2Plex 4+1 Meter Design
Figure 18: 8-inch CBM 2Plex 4+1-Path Meter
Note that the plastic is simply for a better view and certainly would not be used in the field. The
single-path transducers on the right of the meter are located in the center of the meter (vertical
axis) and do not bounce off of the meter body (direct path configuration). Figure 1 shows this
direct path uses the traditional angle of 60 degrees. Thus the overall path length of this single-
path pair of transducers is only slightly longer than the longest paths (paths 2 and 3) in a 4-path
meter.
Two independent Signal Processing Units (SPUs) are used, one for the 4-path configuration, and
one for the single-path configuration. Both electronics energize their transducers independently
of each other. There is also no communication between the electronics, and no interaction
between the sound pulses from one “meter” to the other.
The comparison of the output of both meters is not on a real-time basis, but rather performed
once per day, or even once per hour. That is, the uncorrected accumulated volume in the 4 path
meter is compared to the single path meter at the daily (or hourly) level. Daily checks help
eliminate the minor differences that can occur on a real-time basis between the two meters due to
their different velocity sampling techniques. This also permits using a tighter tolerance and
increases the reliability of the comparison.
During operation, conditions can change in the piping system that can impact the accuracy of the
meter, even when using a flow conditioner. These changes include blockage of the flow
conditioner with a foreign object, contamination over time from oil and mill scale, unexpected or
unanticipated pulsation of gas, and potential changes within the 4-path meter electronics and
transducers. By incorporating a second independent electronics with an independent path, this
design essentially provides a real-time flow check against the 4-path meter. But why use a single
path design to check the 4-path meter instead of another 4-path design checking the fiscal 4-path
meter?
During the past several years data has shown that single path meters, with the transducers located
to send sound pulses through the middle of the meter body, are more sensitive to profile changes.
In a paper published in 1998 by Terry Grimley [Ref 3], installation effects were measured on two
multi-path meters, and on two single-path meters. A variety of installation effects were tested
including two elbows in and out of plane upstream of the four meters. The multipath meters
performed relatively well with errors attributed to the installation effects on the order of 0.5% or
less. In the same piping configuration the single path meters had errors that were on the order of
2-5%. Clearly the multipath meters could deal with the asymmetrical and swirling profiles far
better than the single-path meters.
Profile changes also occur when contamination develops on the inside of the piping and meter.
As the buildup occurs, the wall friction increases causing the velocity profile in the center of the
meter to be higher relative to the area along the pipe wall. A paper published at the North Sea
Flow Measurement Workshop (NSFMW) in 2005 [Ref 1] discusses how the profile changed over
time due to internal pipeline contamination. This paper shows examples of the meter’s response
when blockage occurs upstream at the flow conditioner. The velocity profile differences between
the 4 path meter and the independent single path meter resulted in significantly different
measurements between the two designs.
Placing the single-path pair of sensors in the center of the meter body was done intentionally as
this is the most sensitive location for flow measurement. That is the center-line path will shift far
more than if located at any other position within the meter. This makes it an excellent check
against the 4-path which experiences much less shift when the profile changes.
Through the use of meter diagnostics, and the associated manufacturer’s software, many of the
above problems can be identified. The problem with the conventional method of identifying
potential measurement errors is that most users only check the meter’s diagnostics on a monthly
basis, and sometimes less often that that. When a problem occurs, it may be weeks before it is
identified, and thus the impact on billing can be substantial.
By using the CBM 2Plex 4+1 method of comparing the output of a single-path meter to that of
the fiscal 4-path chordal meter, the performance of the two meters is validated every day. This
means if a problem occurs, a potential measurement error can be identified by the system within
one day. Once a problem has been identified, technicians can be dispatched to investigate or the
meter can be monitored more closely for further action. In today’s environment where the price
of gas is ever increasing, errors in transportation, buying and selling of natural gas can lead to
more significant financial risk than ever before. Knowing a meter has a potential problem within
a day (or hours) will help reduce unaccounted for gas (UAF).
4. PROVING THE CONCEPT
Does this technique really work when both meters are incorporated into one meter body? To
answer this question, testing was conducted at the CEESI Iowa high flow calibration facility in
Garner, Iowa. For this test a 12-inch 4+1 meter was installed with a CPA 50E flow conditioner
upstream. This type of flow conditioner has been used in many USM applications around the
world.
One of the issues with using a flow conditioner is that debris can collect in front of the flow
conditioner. When this occurs there can be an affect on the USM accuracy. The effect has been
documented in several presentations [Ref 1, 5 & 6] and is demonstrated in the first part of this
paper.
To quantify the benefits of this design, testing with several blockage scenarios was conducted.
Not only were the 40% blockage, but additional testing was done only 1 hole blocked. Three
velocities were used for all of these tests. These were approximately 7 m/s, 14 m/s and 21 m/s.
Figure 19 shows the results of the 4-path meter after baseline calibration (piecewise linearization)
and subsequent results with 40% blockage of the flow conditioner.
12-inch, 4-Path Meter - 40% Blocked Results
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
0 5 10 15 20 25
Meter Velocity (m/s)
% E
rro
r
Un-Blocked CPA 40 Percent Blocked CPA
Figure 19: 12-inch, 4-Path Meter Results with 40% Blockage
Figure 19 shows that the 12-inch, 4-path meter shifted on the order of -0.15%, or less, for all
three velocities tested. Figure 20 shows the results of the single-path during this same time.
12-inch, Single-Path Meter, 40% Blocked Results
-4.0
-3.5
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
0 5 10 15 20 25
Meter Velocity (m/s)
% E
rro
r
Un-Blocked CPA 40% Blocked CPA
Figure 20: 12-inch, single-Path Meter Results with 40% Blockage
The accuracy impact on the single-path meter is on the order of -3.5%. Thus for the same
blockage upstream of the 12-inch meter, the single-path meter shifted more than 20 times as
much as the 4-path meter.
The next step was to test the meter with only one blocked hole. Figure 21 shows the results of the
12-inch, 4-path meter with this blockage and Figure 22 shows the results of the 12-inch, single-
path meter.
12-inch, 4-Path Meter - 1 Hole Blocked Results
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
0 5 10 15 20 25
Meter Velocity (m/s)
% E
rro
r
Un-Blocked CPA Single Hole Blocked - Botton CPA
Figure 21: 12-inch, 4-Path Meter Results with 1 Hole Blocked
12-inch, Single-Path Meter, 1 Hole Blocked Results
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
0 5 10 15 20 25
Meter Velocity (m/s)
% E
rro
r
Un-Blocked CPA Single Hole Blocked - Bottom CPA
Figure 22: 12-inch, Single-Path Meter Results with 1 Hole Blocked
Figure 21 shows there is no impact on the 4-path meter’s accuracy, while the single-path meter,
shown in Figure 22, shifted between -0.6% and -0.85%. Thus, even with one hole blocked, the
single-path meter shift was very significant, and thus the difference between the 4-path and
single-path could be easily identified.
Pipeline contamination, especially over time, is a more challenging problem for the technician.
Many pipelines have some minor amount of oil and mill scale that is being transported down the
pipeline. Although this contamination is generally small, it can accumulate over time and have a
significant impact on a meter’s accuracy.
Several papers have been published over the past 10 years [Ref 4, 7, 8, 9 & 10] which discuss the
impact on the meter’s accuracy. Some meter designs tend to register fast when contamination
coats the meter piping and meter body, while others tend to register slower. The challenge for all
users of USMs is to identify this contamination and then to decide when it is time to clean the
meter run.
The question is: “Can the 4+1 CBM meter design identify contamination in the piping?”. To
answer this question an existing 4-path meter was borrowed from an installation. This particular
meter is an inter-company operational meter, in a bi-directional application installed in the late
1990’s and uses a 19-tube bundle flow conditioner. During the several years of service it had
been cleaned numerous times due to contamination.
For this test the entire meter run (including all piping and flow conditioner) was removed, and
sent to the CEESI Garner calibration facility. As the installed meter was not a 4+1 CBM design,
only the piping was used for this testing to see if the contamination could be identified by the
CBM meter.
Unfortunately the meter run had been cleaned recently and the upstream piping was not as dirty
as was expected. Figure 23 shows the “as-found” condition of the piping between the flow
conditioner (19-tube bundle) and the meter.
Figure 23: 12-inch Dirty Piping
As Figure 23 shows, there was not a lot of contamination remaining due to the recent cleaning of
the meter piping.
Today most designers do not use this type of flow conditioner but instead select a perforated
plate like the CPA 50E. The customer chose to re-install the meter after testing and replace the
19-tube bundle with the CPA unit. For this reason all testing was conducted with a CPA flow
conditioner. To simulate what the flow conditioner may have looked like it had been subjected to
normal pipeline contamination, “texture” paint was applied to the CPA flow conditioner. Figure
24 shows the flow conditioner just prior to being installed for the testing.
Figure 24: 12-inch CPA with “Texture Paint”
Although this coating might not represent the identical contamination to the piping, it was felt at
the time that some type of contamination was needed to at least simulate surface buildup.
Because it was previously decided to re-install the 4+1 CBM meter, after all testing was
completed, for some long-term testing and wanted the CPA to be used during this time. In order
to save some calibration time, rather than conduct testing with the 19-tube bundle, it was decided
to contaminate the CPA for the “as found” dirty testing.
Figure 25 shows the results for the 4-path meter both dirty and clean.
Figure 25: 12-inch, 4-path As-Found Dirty and As-Found Clean
The results of the 4-path meter as-found baseline are shown with the blue dots after the piping
was cleaned (meter was brand new and thus clean). The red dots represent the as-found results
with the upstream piping and CPA dirty. The table in Figure 26 shows the difference between the
two at each flow velocity in m/s.
Velocity % Diff.
23.3 -0.12
15.7 -0.10
7.8 -0.05
Figure 26: 4-path Dirty vs. Clean Differences
Figure 26 shows the meter registered slightly slower with the upstream piping being dirty
compared to the clean piping. This is the expected results since a previous paper [Ref 4 & 7] had
demonstrated that the upstream piping tends to cause the chordal meter to register slightly slower
when dirty.
Figure 27 shows the results for the single-path during the same conditions.
Figure 27: 12-inch, 4-path As-Found Dirty and As-Found Clean
In Figure 27 the single path meter registered faster (red dots) when the upstream piping was dirty
compared to the clean upstream piping as shown with the blue dots. This difference is
summarized in Figure 28.
Velocity % Diff.
23.3 0.57
15.7 0.50
7.8 0.35
Figure 28: Single-path Dirty vs. Clean Differences
As this table shows the single path meter registered faster when dirty and is just the opposite of
what the 4-path meter showed. Although this isn’t as thorough of a test as a uniformly dirty
meter, it does show that the single-path meter behaves differently than the 4-path meter. With
very little contamination on the upstream pipe wall, the change in the single path meter’s
response was easily seen.
5. IMPLEMENTING THIS DESIGN
As discussed earlier in this paper, both electronics operate independently. The output of each
meter needs only to be brought into the same flow computer and volumes stored for both as
would normally be done for two separate meters. To take advantage of this feature, the hourly
uncorrected volumes would then be compared and an acceptable tolerance would be determined
based upon some history established during commissioning. The tolerance may vary somewhat
from site to site, and will depend slightly upon the upstream piping conditions and the symmetry
of the profile downstream of the flow conditioner. However, the typical agreement that has been
seen from some field data is on the order of ±0.5%.
The comparison test probably should not be conducted when meter velocities are below perhaps
3 m/s as the profile effects can become more significant. For this reason the flow computer
should accumulate separate totals for comparison testing since the effective cutoff for the
comparison would be perhaps 3 m/s. Thus it may not be practical to use the absolute uncorrected
volumes through each meter if the flow rate is frequently below this velocity.
For most installations meter velocities are usually always above 3 m/s. This is common in
mainline stations where there is always flow. For these cases a direct comparison of hourly (or
daily) uncorrected volumes would suffice. Many users already have this capability built into their
flow computers. They do comparisons of “run ratios” in order to spot potential problems. For
these users they simply have to connect the meter to the flow computer, set the comparison ratio
to a value, and start monitoring for the alarm. Thus full advantage of this meter design can be
incorporated immediately without special flow computer programming.
Figure 29 is an example of a 2Plex 4+1 meter in a bi-directional application that has been
compared at the daily level for more than 19 months.
-3.00%
-2.50%
-2.00%
-1.50%
-1.00%
-0.50%
0.00%
0.50%
1.00%
1.50%
2.00%
9/1
8/0
7 1
2:0
0 P
M
9/3
0/0
7 1
2:0
0 A
M
10/1
1/0
7 1
8:0
0
10/2
4/0
7 1
6:0
0
11
/4/0
7 6
:00
11/1
7/0
7 1
6:0
0
11/2
6/0
7 1
0:0
0
12
/7/0
7 2
:00 P
M
12/2
0/0
7 9
:00 A
M
12/2
9/0
7 2
:00 A
M
1/6
/08
10
:00 P
M
1/1
8/0
8 1
2:0
0 P
M
2/8
/08 5
:00 P
M
2/2
0/0
8 1
:00 A
M
3/2
/08
8:0
0 A
M
3/1
0/0
8 6
:00 P
M
3/2
1/0
8 3
:00 A
M
3/3
0/0
8 1
:00 P
M
4/1
2/0
8 7
:00 P
M
4/2
3/0
8 1
2:0
0 P
M
5/5
/08 1
2:0
0 A
M
5/1
4/0
8 3
:00 A
M
5/2
7/0
8 7
:00 A
M
6/1
0/0
8 7
:00 A
M
7/1
/08
7:0
0 A
M
7/1
2/0
8 1
2:0
0 P
M
7/2
4/0
8 5
:00 P
M
8/4
/08 2
:00 P
M
8/1
4/0
8 9
:00 P
M
8/2
5/0
8 4
:00 A
M
9/6
/08 1
0:0
0 A
M
9/1
6/0
8 8
:00 A
M
9/2
4/0
8 4
:00 P
M
10/3
/08
11
:00 P
M
10/1
3/0
8 8
:00 P
M
10/2
2/0
8 8
:00 P
M
10/3
1/0
8 1
:00 P
M
11
/9/0
8 5
:00 A
M
11/1
8/0
8 7
:00 P
M
12
/1/0
8 7
:00 A
M
12/1
2/0
8 5
:00 P
M
12/2
3/0
8 1
:00 P
M
12
/31/0
8 1
1:0
0 P
M
1/9
/09 2
:00 P
M
1/2
0/0
9 1
0:0
0 A
M
2/1
/09 1
2:0
0 A
M
2/1
2/0
9 1
1:0
0 P
M
2/2
4/0
9 6
:00 A
M
3/8
/09 1
2:0
0 A
M
3/2
3/0
9 5
:00 A
M
4/5
/09 3
:00 P
M
4/1
4/0
9 1
:00 P
M
4/2
5/0
9 1
:00 P
M
% D
iff.
4-p
ath
vs
1-p
ath
ac
f
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
450,000
500,000
AC
FH
Fwd
Rev
ACFH
Figure 29: 4-Path vs. 1-Path Daily Comparison for 19 Months of Hourly Flow Rates
This graph shows the comparison of the percent difference in dark blue for the forward direction,
and magenta in the reverse with the scale on the left axis. The average flow rate is in orange and
the scale is on the right. This graph is of hourly flow comparisons graphed over 19 months time.
For the most part, all hourly comparisons are all within 0.25%. On a few occasions the difference
exceeds ±0.5%. This is when the flow rate is low and perhaps during this time some minor
contamination occurs. When the flow rate increases, the difference goes back to the normal
(which is approximately -0.15%). This may indicate some clearing out of minor contamination
since the meter difference generally goes positive which is expected if the piping becomes dirty.
Many users today know they have contamination in their metering systems. They periodically
clean the meter in order to minimize the uncertainty effect due to contamination. If the amount of
difference between the 4-path meter and the single-path meter can be used to determine the
cleaning interval, these users will then benefit from extended inspection intervals and thus save
significant O&M expenses.
6. CONCLUSIONS
Today the cost of energy is higher than it was several years ago, and it is not likely this trend will
reverse itself. By implementing ultrasonic metering technology many users have been able to
improve their measurement and reduce their un-accounted for (UAF) gas during the past several
years. One task always remains for the technician and that is to insure the meter is operating
correctly and accurately. This applies to all measurement technologies, not just USMs. The
significant benefit of the USM is the ability to provide diagnostic information for the user to help
determine the meter’s “health.”
Today technicians have software to help understand the operation of their USM. Since each
manufacturer of USMs uses a different velocity integration technique (different path
configurations), it is often difficult for the technician to fully understand whether his USM is
operating correctly or not. Additionally, since most only inspect the meter’s operation once per
month, problems can occur and go undetected for many days or weeks. This can significantly
increase measurement uncertainty during this time.
The CBM 2Plex 4+1 meter design relies on basically two principles. First, the fiscal meter is
chosen to be the least sensitive to any flow profile changes that may occur in normal operation.
And second, the “check” meter design is chosen to be one that is the most sensitive to any flow
profile changes. Ideally any affect from a profile change would not only have a significantly
different impact on accuracy for each path layout, but the affect would be in opposite directions,
making the difference much easier to detect.
The benefit of the CBM 2Plex 4+1 meter design is that the flow computer is used to check the
health of the fiscal 4-path chordal meter by simply comparing it to the single-path meter. If the
velocity profile remains relatively constant, both meters will agree. Should some process
condition upset the normal profile, the single-path meter will respond significantly different than
the 4-path. These upsets can include the following:
• Blockage in front of the flow conditioner
• Contamination due to oil and mill scale buildup over time
• Pulsation in the pipeline due to compressors (sampling rate for the single-path is much
faster than the 4-path and thus less sensitive to pulsation)
• Potential problems with the fiscal meter including transducers and electronics problems
• Full redundancy should there be a failure of the electronics.
When a meter is equipped with automated meter diagnostics, as described in Section 2, it
expands the monitoring of the fiscal meter’s health to an even higher level. The 2Plex 4+1 design
only validates that the velocity profile hasn’t changed. By monitoring all of the remaining
diagnostics on a real-time basis, the meter’s health can be validated on a real-time basis. This is
important should a diagnostic value such as gain for a pair of transducers, or low SNR from a
control valve, approach a value that may cause a path to fail.
By monitoring all aspects of the meter’s diagnostics, both with the 2Plex redundant design, and
with continuous checking of all other diagnostics (for both the 4-path and 1-path meter), the user
can have a much higher degree of confidence that the measurement is accurate.
Today the cost of accuracy has never been more important. Virtually all applications today
require the measurement accuracy be maintained at the highest possible level. The CBM 2Plex
4+1 meter design, combined with automated real-time internal monitoring of all diagnostic
values, provides a complete “health check” on the custody transfer meter (4-path). This can
significantly reduce both operation and maintenance (O&M), and measurement uncertainty, and
thus reduces the cost of doing business.
7. REFERENCES
1. John Lansing, How Today’s USM Diagnostics Solve Metering Problems, North Sea Flow
Measurement Conference, October 2005, Tonsberg, Norway
2. Klaus Zanker, Diagnostic Ability of the Daniel Four-Path Ultrasonic Flow Meter, Southeast
Asia Flow Measurement Workshop, 2003, Kuala Lumpur, Malaysia
3. T. A. Grimley, The Influence of Velocity Profile on ultrasonic Flow Meter Performance,
AGA Operations Conference, May 1998, Seattle, Washington, USA
4. John Lansing, Dirty vs. Clean Ultrasonic Flow Meter Performance, North Sea Flow
Measurement Conference, October 2004, St. Andrews, Scotland
5. Larry Garner & Joel Clancy, Ultrasonic Meter Performance – Flow Calibration Results –
CEESI Iowa – Inspection Tees vs. Elbows, CEESI Ultrasonic Conference, June 2004, Estes
Park, Colorado, USA
6. John Lansing, Features and Benefits of the SICK Maihak USM, CEESI Ultrasonic
Conference, June 2006, Estes Park, Colorado, USA
7. John Lansing, Dirty, vs. Clean Ultrasonic Gas Flow Meter Performance, AGA Operations
Conference, May 2002, Chicago, Illinois, USA
8. L. Coughlan, A. Jamieson, R.A. Colley & J. Trail, Operational Experience of Multipath
Ultrasonic Meters in Fiscal Service, North Sea Flow Measurement Conference, October
1998, St. Andrews, Scotland
9. John Stuart, Rick Wilsack, Re-Calibration of a 3-Year Old, Dirty, Ultrasonic Meter, AGA
Operations Conference, April 2001, Dallas, Texas, USA
10. James N. Witte, Ultrasonic Gas Meters from Flow Lab to Field: A Case Study, AGA
Operations Conference, May 2002, Chicago, Illinois, USA
11. John Lansing, Operation and Maintenance Considerations for Ultrasonic Meters,
Appalachian Gas Measurement Short Course, August 2008, Coraopolis, Pennsylvania, USA
12. John Lansing, Advanced Ultrasonic Meter Diagnostics, Western Gas Measurement Short
Course, May 2007, Seattle, Washington, USA
Three Columns Gas Chromatograph Analysis Using Correlation between Component's Molecular Weight and Its Response Factor
Anwar Sutan Metco Services Ltd
Charles Johnson
Metco Services Ltd
Jason Laidlaw Metco Services Ltd
1
Three Columns Gas Chromatograph Analysis Using Correlation between Component's Molecular Weight and Its Response Factor
Anwar Sutan, Metco Services Ltd Charles Johnson, Metco Services Ltd. Jason Laidlaw, Metco Services Ltd.
1. Introduction Gas Chromatographs (GCs) are delivered from factory with a multilevel calibration already programmed. While this is an effective method to handle the linearity of the detector, it requires many sets of gases at varying concentrations to obtain the multi level calibration parameters. On site when component parts of the GC are changed such as columns, diaphragm, detectors, etc. The GC may require a new set of multilevel calibration parameters. For a number of reasons this is not always practical to do on site or in the field.
Other calibration issues on site can result in the systematic drift of the response factor. This sometimes cannot be detected by the GC automatically as the response factor shift can remain within the tolerance limit from the previous response factor set in the GC controller. Even with the wrong calibration result, the Gas Chromatograph still can give repeatable results, however the results will not be accurate and can increase the uncertainty of the measurement.
The repeatability and reproducibility tests are good to prove that a GC is working within limits which are specified in ASTM D1945:1996 [3] or GPA 2261:1995 [5]. However, due to the wide tolerance on some compounds these tests do not guarantee that the GC is working as intended. Using these tests does not ensure that each of the components goes through its intended column and further it does not confirm that all the valve timings are correct. A further analysis is required to check this functionality and this can be done by analyzing the response factor of each components.
This paper describes a practical method that can be used to overcome these issues by looking at the correlation between component’s molecular weight and its response factor, and by looking at the historical response factor data for each component.
2. Chromatograph Design and Operation In order to get a composition of a natural gas with inert components and hydrocarbon components ranging from C1 to C7+ in a practical time frame and without temperature ramping, a multi column separation technique is required.
The three columns GC design is consistent with ISO 6974-5:2000 [6]. It uses three 6-port chromatograph valves, three columns, a restrictor, reference detector and measuring detector in a controlled temperature chamber. The detectors are thermistors, where resistance changes depend on the temperature. The reference and measuring detectors form a balanced Wheatstone Bridge. Helium is the preferred carrier gas because it has high thermo-conductivity, Nitrogen, Hydrogen and Argon can also be used in special circumstances. With only carrier gas flowing across the two detectors, the Wheatstone bridge is in balance. In the
2
measuring detector, the sample gases passing across the thermistor causes thermo-conductivity changes resulting in a change of thermistor heat exchange rate, this in turn results in an increase of the temperature of the thermistor. The change of temperature results in a change of resistance in the measuring detector and unbalances the Wheatstone Bridge. The magnitude of the voltage created by Unbalance Bridge and the time taken to pass through the detector then forms a response curve proportional to the amount of the component in the carrier gas stream. The area under the response curve is proportional to the mole % of component being measured.
On a single column GC application where pressure, temperature, and flow rate can be maintained constant, there is a very high correlation between the molecular weight of the saturated gas components and their response factor. With a three columns GC restrictor tubing has to be applied to regulate carrier gas flow to maintain and achieve the close-to-constant flow rate during valve actuation operations. With pressure and temperature maintained constant, and with a restrictor tubing in place, slight flow rate differences can occur that will affect the response of thermal conductivity detector. The fluctuations in flow rate reduce the correlation of the molecular weight of each component with its response factor.
The GC tested was configured such that C1, C2 and C6+ went through measurement detector with the same flow rate, and C3, iC4, nC4, iC5, nC5 went through measurement detector with another flow rate. Figure 1 shows flow path for C1, C2 and C6+ where the flow went through column 3.
Figure 1. Danalyzer / 2350A Gas Chromatograph, Hardware Reference Manual [1], Three-
column GC flow path of C1, C2 and C6+
3
Figure 2 shows flow path for C3, through nC5 where the flow goes through restrictor tubing.
Figure 2. Danalyzer / 2350A Gas Chromatograph, Hardware Reference Manual [1], Three-
column GC flow path of C3, C4 and C5
3. Verification Procedures There are some well established procedures available and used to verify Gas Chromatograph operation on site. These procedures ensure that GC is working within the limits specified in the standard. A daily or periodic auto calibration is normally done. The repeatability criteria in D1945:1996 [3] or GPA 2261:1995 [5], is used to ensure that the calibration is valid. Other periodic checks that are usually carried out are:
• Reproducibility test
• Detector bridge balance;
• Oven and detector temperature check;
• Carrier gas pressure check;
• Calibration gas pressure check;
• and flow rate check for both sample and measurement detector vent.
3.1. Periodic Auto Calibration A periodic auto calibration is performed to ensure that the GC is functioning within a defined specification. These periods are determined by the stability of the GC calibration and can be daily, weekly or monthly. A calibration report is generated after each calibration cycle. In the report, there are old response factors from the previous calibration and new response factors from the current calibration. A slight shift in the response factor is acceptable as defined in ASTM D7146-05 [4]. This auto calibration however is not designed to detect systematic shift in these factors and this must be reviewed manually over time and historical calibrations. If the response factor increases or decreases consistently after every calibration, and the response factor deviation from the previous response factor is still within the deviation limit set, the GC will not generate any alarm and will function without reporting any faults.
4
3.2. Repeatability Test Repeatability stated by D1945:1996 [3] section 10.1.1 is the difference between two successive results obtained by the same operator with the same apparatus under constant operating conditions on identical test materials should be considered suspect if they differ by more than the following amounts:
Mol % Lookup Tolerance %0 to 0.1 Mol% 0.01 %0.1 to 1 Mol% 0.04 %1 to 5 Mol% 0.07 %5 to 10 Mol% 0.08 %Over 10 Mol% 0.10 %
Table 1. D1945:1996 [3] section 10.1.1
Repeatability stated by GPA 2261:1995 [5] section 9 is the expected precision within a laboratory using the same equipment and the same analyst. It should be considered suspect if they differ by more than the following amounts:
Component Mol % Range Tolerance (Percent relative)
Nitrogen 1 to 7.7 2CO2 0.14 to 7.9 3Methane 71.6 to 86.4 0.2Ethane 4.9 to 9.7 1Propane 2.3 to 4.3 1Isobutane 0.26 to 1 2n-Butane 0.6 to 1.9 2Isopentane 0.12 to 0.45 3n-Pentane 0.14 to 0.42 3C6+ 0.1 to 0.35 10
Table 2. GPA 2261:1995 [5] section 9
Repeatability test on an on-line GC is performed by analyzing a gas of known composition. A calibration gas that has a known composition gas is used and analyzed a number of times. The analysis result of the sample gas is compared with the gases’ calibration certificate. By carrying out this type of repeatability test, the GC is confirmed to work within the standard specification. If the repeatability test fails, the data can be used as a troubleshooting tool.
There are two sets of criteria to be checked when performing the repeatability test range and precision. Range is the difference between maximum value and minimum value of the measured gas during the test. Precision is the difference between the value measured by GC and the value stated in calibration gases’ certificate. Both have to fall within the stated tolerance (Table1, Table 2).
5
3.3. Reproducibility Test Reproducibility stated by D1945:1996 [3] section 10.1.2 is The difference between two results obtained by different operators in different laboratories on identical test materials should be considered suspect if they differ by more than the following amounts:
Mol % Lookup
Tolerance %
0 to 0.1 Mol% 0.02 %0.1 to 1 Mol% 0.07 %1 to 5 Mol% 0.10 %5 to 10 Mol% 0.12 %Over 10 Mol% 0.15 %
Table 3. ASTM D1945:1996 [3] section 10.1.2 Reproducibility stated by GPA 2261:1995 [5] section 9 is the expected precision when the same method is used by different laboratories using different equipment and different analyst. It should be considered suspect if they differ more than the following amounts:
Component Mol % Range Tolerance (Percent relative)
Nitrogen 1 to 7.7 7CO2 0.14 to 7.9 12Methane 71.6 to 86.4 0.7Ethane 4.9 to 9.7 2Propane 2.3 to 4.3 2Isobutane 0.26 to 1 4n-Butane 0.6 to 1.9 4Isopentane 0.12 to 0.45 6n-Pentane 0.14 to 0.42 6C6+ 0.1 to 0.35 30
Table 4. GPA 2261:1995 [5] section 9
On site or in the field , this is normally achieved by comparing the result obtained from the online GC with a Laboratory GC result.
4. Gas Thermal-Conductivity vs Response Factor The relationship between Thermal Conductivity and Molecular weight in Hydrocarbon and inert gases is well known. The larger the molecular weight the smaller the Thermal Conductivity. The thermal conductivity detector’s resistance changes as the temperature changes. The temperature on the detector changes whenever gas with different thermal conductivity property flows through it. The higher thermal conductivity difference between carrier gas and the component being measured, the more temperature change will occur, and the more change on the thermal conductivity resistance. Therefore the higher the thermal conductivity property of a component, the lower the peak area generated by detector, and the lower response factor for that component (RF = PA/mol%).Applying this to the thermal-conductivity detector, because of the high Thermal Conductivity property of the carrier gas,
6
the higher the thermal conductivity of a component being measured the lower difference on carrier gas – measured component gas thermal conductivity, the lower peak area generated by the detector and the lower the response factor for that component (RF = PA/mol%). The three column gas chromatograph measures components, in order from high thermal-conductivity to low thermal conductivity, Methane; Nitrogen; Ethane; CO2; Propane; i-Butane; n-Butane; neo-Pentane; i-Pentane; n-Pentane; Hexane+; and Heptane+ etc.
As a first verification step the calibration result can be analyzed by monitoring the response factor data. Response factor of methane will be lowest, followed by Nitrogen; Ethane; CO2; Propane; i-Butane; n-Butane; neo-Pentane; i-Pentane; n-Pentane; Hexane+; and Heptane+. If this sequence is not evident, then this would point to some form of incorrect setting, or if everything is ok with GC, the calibration gas could be considered suspect. Typical graphical representation of components response factor is shown in figure 3.
Figure 3. Typical response factor graphic of gas components response factor
5. Flow Rate Effect on Response Factor The Author performed a test on a C7+ GC where there were significance differences with flow rate when valve 3 was on and when valve 3 was off. The initial test had flow a rate of 13.3 cc/min when valve 3 was on (flow through column 3), and it had flow rate of 10.4 cc/min when valve 3 was off (flow through restrictor tubing). The result of the test was as follow:
Comp MW RF Log(MW) Log(RF) C1 16.04 7.91733E+06 1.20520 6.89858 C2 30.07 1.22268E+07 1.47813 7.08731 C3 44.09 1.30256E+07 1.64434 7.11480 nC4 58.12 1.56587E+07 1.76433 7.19476 nC5 72.15 1.78403E+07 1.85824 7.25140 C6 86.18 2.01611E+07 1.93541 7.30451 C7 100.21 4.12349E+07 2.00091 7.61526
Table 5. Calibration result of GC with different restrictor-column3 flow rate
7
Figure 4. Correlation between Response factor and Molecular weight with different flowrate
when valve 3 is on and when valve 3 is off
An adjustment was made to the restrictor tubing to achieve a similar flow rate between the condition when valve 3 is on and when valve 3 is off. The final test had flow rate of 12.9 cc/min when valve 3 was on (flow through column 3), and it had flow rate of 12.7 cc/min when valve 3 was off (flow through restrictor tubing). And the result was as follows:
Comp MW RF Log(MW) Log(RF) C1 16.04 7.59792E+06 1.20520 6.88069 C2 30.07 1.18068E+07 1.47813 7.07213 C3 44.09 1.51142E+07 1.64434 7.17939 nc4 58.12 1.72937E+07 1.76433 7.23789 nc5 72.15 1.96312E+07 1.85824 7.29295 c6 86.18 2.20184E+07 1.93541 7.34279 c7 100.21 2.82900E+07 2.00091 7.45163
Table 6. Calibration result of GC with similar restrictor-column3 flow rate
Figure 5. Correlation between Response factor and Molecular weight with similar flow rate
when valve 3 is on and when valve 3 is off
From the graphic of correlation (R2) for all components there is a big discrepancy in terms of correlation when the flow rates were similar and when the flow rates were significantly different. There was no significant difference in correlation of components where the flow rates were similar, e.g. (C3, nC4, nC5, nC6) and (C1, C2, C7) however there was a significant divergence in components where the flow rates were different.
The restrictor column arrangement makes it very rare to have an absolute match on flow rate through all the columns. Therefore it is difficult to get a high correlation on all components. Good correlation is achievable for each group of components when the flow from valve 3 is either through the column or the restrictor when the component passes the measurement detector.
8
6. Chromatograph Valve Timing There are 3 critical valve timings on the 3 column chromatograph:
1. Back-flush the heaviest component (C6+ on C6+ GC and C7+ on C7+ GC). Back flush is initiated after C5 and lighter (on C6+ GC) or after C6 and lighter (on C7+ GC) are eluted from column 1 to 2, but before the heavy component leaves column 1
2. Trap the light components in column 3. The valve actuation has to be done after all ethane is eluted into column 3, but before any propane leaves column 2
3. Allow lights to leave column 3. Valve actuation has to be done after all the middle components clear the measurement detector
In GC operation, valve-timing errors can result in:
1. Some of heavy components leave column one and flow through column 2
2. Some of the middle components are back-flushed together with the heavy component
3. Some ethane left in column 2 after the valve 3 actuation to trap the lights
4. Some propane goes in to column 3 before valve 3 actuation to trap the lights
5. Splitting of the first and last components on the column resulting in extraneous peaks on the chromatogram
6. If extraneous peaks are within the peak window of calibrated peaks they may be recognized as the calibrated peak,
The impact of these valve timing issues will affect correlation of the components. The more valve timing issues, the less correlation between components. Using the chromatogram we can analyze what has occurred with regard to the valve timing.
Before checking the correlation of components a few criteria have to be met.
• Carrier gas pressure must be stable,
• Oven temperature must be stable,
• Flow rate to be noted and stable when valve 3 on and when valve 3 off,
• Detector-bridge must be balanced,
• And repeatability must be within specified standard.
After all these criteria are fulfilled, the next calibration result can be analyzed based on its component correlation.
7. Components Correlation Analysis To analyze the correlation between components, both graphs need to be utilized. The cause and effect of the valve timing faults are as follow:
9
1. Some of heavy components leave column 1 and flow through column 2. The effect is smaller response factor of heavy component. We can see it from graphical presentation, the heavy component will be lower than it should be and result in bad R2.
Figure 6. RF – MW Correlation Graphical presentation
2. Some of the middle components are back-flushed together with the heavy component. The effect is smaller response factor on the heaviest middle component and bigger response factor on the heaviest component. From the graphical presentation we are able to see that the heaviest component response factor is higher than expected, and heaviest of the middle component is lower than expected, both graphs have bad R2.
Figure 7. RF – MW Correlation Graphical presentation
3. Some ethane left in column 2 after the valve 3 actuation to trap the lights. The effect is a smaller response factor of Ethane. From graphical presentation it can be shown that the ethane will be lower than the line of C1 and the heaviest component. This results in bad R2.
Figure 8. RF – MW Correlation Graphical presentation
10
4. Some propane goes in to column 3 before valve 3 actuation to trap the lights. The effect is smaller response factor of propane. From graphical presentation it can be shown that the propane is lower than expected, results in bad R2.
Figure 9. RF – MW Correlation Graphical presentation
Based on these analyses, the valve timing can be adjusted, and the effect of adjustment can be seen after another calibration.
8. Footprint and Historical Data The correlation method can be utilized more effectively if a footprint of the GC is taken when the conditions and operations were stable and the GC was freshly calibrated. At this time historical data of calibration reports and chromatograms could be acquired. A live test case of the use of correlation method, footprint and historical data is detailed below:
A chromatograph valve diaphragm was changed on a C6+ GC on 24 July 2007. A chromatogram and a calibration report were taken as a footprint on 26 July 2007. The chromatogram below shows data from 26 July 2007, 28 August 2008, and 16 September 2008. On analysis there were a number of issues that could be taken from these chromatograms, these are discussed in detail.
11
Figure 10. Daniel MON2000 Interface [2], Chromatogram of 26 Jul 2007 (blue), 21 Aug 2008 (red), 16 Sep 2008 (Magenta)
Figure 10 shows the chromatogram for the analysis and footprint of the GC tested. The red circle and blue circle is where we can see the detail of the error and they are zoomed in on figure 11 for the red circle and figure 12 for the blue circle.
From the chromatogram (figure 11), the analysis (red) when compared against the footprint (blue), we can see that there were some deviations. This was caused by a port to port leak on the valve 2 diaphragm. The retention times of components were slower (relative move right) than they should have been (figure 12) from the calibration report if the results were compared with the footprint calibration report. The RF of components generally went up, with C6 having the highest shift of RF (figure 15), this was caused by a change of the carrier gas flow rate which changed the response of the TCD to each component. Even with this deviation and slower retention times, the correlation analysis was still good. However, the increase of response factors of some components and the shift of retention times suggested that the GC was moving from being in a healthy to an unhealthy state. By analysing this information, it can be shown when to intervene and perform maintenance before the GC enters an unhealthy state.
Figure 11. Daniel MON2000 Interface [2], Bumps caused by leaking chromatograph valve 2
12
Figure 12. Daniel MON2000 Interface [2], Shift in retention time
A software package is available which can be used to analyze a GC in the way described in this paper. The software uses the footprint information generated when the GC is known to be functioning correctly. Data such as oven temperature, carrier gas pressure, carrier gas flow rate, response factor etc. are recorded and the response factor and correlation between response factor and molecular weight is plotted. These footprint values can be used as a tool to analyze day to day calibration results. This can be used in contrast to comparing calibration data on day to day basis, which has some limitations as described earlier..
The report generated by the software uses the footprint data as the initial configuration. The report compares the current calibration data to the footprint by overlaying plots of the data from the footprint, with the plot from current calibration. With the graphic representation, a drift in response factor from the footprint data can be seen. There are tables for repeatability tests either based on D1945:1996 [3] or on GPA 2261:1995 [5].
A trend is generated of the error between each component response factors for selected dates and the footprint data. Based on this error analysis, the drift in response factor can be trended. Further statistical methods can be used to analyze the error data.
13
A report generated for example above is as follows.
Figure 13. Footprint page, Current test page, and Error analysis page of GC analysis report
9. Correlation Method Application on Calibration Gas Replacement A further example of the use of the correlation method described above is when a calibration gas is to be replaced. The footprint and the historical data can be used to make sure that the certificate and calibration for the new reference is correct. Before a new calibration or reference gas bottle is replaced, the last calibration data of the GC needs to be checked. When the calibration data is good, then the calibration gas to be changed should be analyzed to ensure that the composition determined by the GC is within acceptable deviation limits of the calibration gas certificate. Once the reading has stabilized, a calibration can be performed using the new calibration gas.
The calibration results, the shifts from the footprint data, and the shifts from the last calibration result can be used in the correlation software to confirm the validity of the new calibration. Further investigation can be done based on the analysis of this data if the calibration result is not satisfactory.
10. Summary This section summarizes the conclusions from Correlation Method.
14
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
10.1. Parameters affecting GC Response Factor There are 3 parameters that must be kept constant in order to get good repeatability on GC Response Factor. They are Pressure, Temperature, and Carrier gas Flow rate. All these parameters affect the response of thermal conductivity detectors to each component. A correct pressure, temperature and flow rate must be maintained to get good separation of components and good repeatability of GC.
10.2. Correlation Method A correlation between Molecular weight and response factor has been developed to demonstrate the representivity of the compositions from the GC. The correlation can be used, to troubleshoot GC errors associated with valve timings, flow inconsistencies and calibration drift. A high correlation of Molecular Weight and Response Factor can be achieved when a GC maintains a constant pressure, temperature, and flow rate. Two sets of Correlation data are required on three column GC to compensate the un-avoidable slight difference in flow rate caused by difference of restrictor tubing and column-3.
10.3. Footprint and Historical Data Footprint and historical GC data can be important to analyze and identify symptoms of GC health. This information can be used as a continuous monitoring tool to detect and act on GC symptoms before failure.
10.4. Calibration Gas Replacement The footprint, historical data, and correlation method can be used to confirm the certification of the calibration gas and make sure that the calibration gas composition is as stated on the certificate. It is also an effective double check to ensure that the first calibration after the calibration gas replacement is correct.
15
11. References [1] Danalyzer / 2350A Gas Chromatograph, Hardware Reference Manual, Part Number 3-
9000-537 Revision F, December 2002
[2] Daniel MON2000 Version 2.46 Interface
[3] ASTM D1945, Standard Test Method for Analysis of Natural Gas by Gas Chromatography, 1996
[4] ASTM D7164-05, Standard Practice for On-line/At-line Heating Value Determination of Gaseous Fuels by Gas Chromatography, 2005
[5] GPA 2261, Analysis of Natural Gas and Similar Gaseous Mixtures by Gas Chromatography, 1995
[6] ISO 6974-5, Natural gas – Determination of composition with defined uncertainty by gas chromatography – Part 5: Determination of nitrogen, carbon dioxide, and C1 to C5 and C6+ hydrocarbons for a laboratory and on-line process application using three columns, 2000
Filename: Three Columns Gas Chromatograph Analysis Using Correlation between Component's Molecular Weight and Its Response Factor.doc
Directory: C:\Documents and Settings\anwar.sutan\My Documents Template: Normal.dotm Title: Three Columns Gas Chromatograph Analysis Using Correlation
between Component's Molecular Weight and Its Response Factor Subject: Author: Anwar Sutan Keywords: Comments: Creation Date: 02/10/2009 11:28:00 Change Number: 9 Last Saved On: 09/10/2009 09:28:00 Last Saved By: Anwar Sutan Total Editing Time: 87 Minutes Last Printed On: 09/10/2009 09:29:00 As of Last Complete Printing Number of Pages: 16 Number of Words: 4,661 (approx.) Number of Characters: 26,568 (approx.)
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
1
Validation of the CFD method for determining the
measurement error in Flare Gas Ultrasonic meter installations
Jeff Gibson, TUV NEL
1 INTRODUCTION
This paper discusses the results of an ongoing project assessing the effectiveness of using Computational Fluid Dynamics (CFD) modelling to predict the installation error on a single-path flare gas ultrasonic flowmeter. The work is being funded through the National Measurement Office’s (NMO) Engineering and Flow Programme (www.flowprogramme.co.uk) and will be detailed in TUV NEL report 2008/301 [1]. The CFD simulations were compared with experiments undertaken in TUV NEL’s National Standard Atmospheric Flow Measurement Facility. Tests were performed using a 1.5D-radius single bend placed at various pipe diameters (D) from the inlet flange of the meter. The experimental work was conducted using ultrasonic transducers supplied by GE Sensing. The transducers were installed in a specially-made 12-inch meter spool to allow the error to be assessed at the commonly used diametric and mid-radius positions. The meter was tested from 0.25 m/s to 30 m/s corresponding to a range of Reynolds numbers of 5,000 to 600,000, the Reynolds numbers at the lower end being such that the flow was likely to be in laminar-turbulent transition. Such flows are not uncommon in emergency flare systems during routine day-to-day flaring. The flow simulations were undertaken using the commercial CFD package Fluent
TM.
The work described in this paper demonstrates the capabilities of the TUV NEL low-pressure test facility for determining the installation error in flare gas ultrasonic meters. In addition, CFD modelling has proved to be a very useful tool for determining the installation errors, also helping to interpret and rationalise the experimental data. A follow-up project is underway to further investigate the issues raised by the initial phase of work and this will be briefly discussed in this paper.
2 BACKGROUND
Flaring from UK oil and gas facilities is controlled and regulated through the Flare Consent Scheme operated by the Department of Energy and Climate Change (DECC). Operators are also required to report their calculated CO2 emissions resulting from gas flaring, and fuel usage, into the EU Emissions Trading Scheme (EU ETS). With CO2 credits being a tradable commodity within the EU ETS, the operator has an incentive to reduce the amount of gas flared below their permitted level. However, the importance of obtaining accuracy on the determination of the quantity of gas flared, and therefore the mass of CO2 released to atmosphere, is very clear. To this end the EU ETS prescribes uncertainty levels which must be achieved in order to comply with the regulations (Directive 2003/87/EC [2]). The requirements for measurement and reporting flare are given within Annex II, Section 2.1.1.3 of the accompanying Measurement and Reporting Guidelines 2007 (MRG 2007 [3]).
Fig. 1 – Gas flaring from an offshore oil
and gas platform
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
2
Installations covered by the EU ETS are categorised depending on total CO2 emissions. For the oil and gas sector this is mostly made up of the CO2 emitted from the burning of produced gas as a fuel, with the remainder being made up of flaring and the burning of liquid fuels such as Diesel oil etc. The higher the category (i.e. “Tier Level”), the lower the required uncertainty. Section 5.2 of Annex I of the MRG states that: “The highest tier approach shall be used by all operators to determine all variables for all source streams for all Category B or C installations. Only if it is shown to the satisfaction of the competent authority that the highest tier approach is technically unfeasible, or will lead to unreasonably high costs, may a next tier be used for that variable within a monitoring methodology”. The maximum allowable activity data (volume or mass) uncertainty for flaring under category A, B and C installations is stated 17.5%, 12.5% and 7.5% (corresponding to tiers 1, 2 and 3 respectively). The majority of UK offshore installations fall into Category B. UK operators must therefore be able to demonstrate to the regulator that they are meeting 12.5% and the uncertainty calculations must also be verified by an accredited third-party. There are some larger, Category C platforms in the Norwegian and Danish sectors of the North Sea. Ultrasonic meter manufacturers state the uncertainty on their flare meters based on ideal, fully developed flow conditions where there are no entrained liquids in the gas and the critical dimensions are accurately measured. Therefore, it is important to consider the effect of any deviations from the ideal in subsequent uncertainty analyses of flare meters as installed and used in the field. An important aspect that must be considered in the uncertainty analysis arises from the piping configuration upstream and downstream of the meter. Upstream fittings will disturb the flow and will result in an installation error. An approach to determining this error is discussed in this paper.
3 FLARE GAS ULTRASONIC METERING
Ultrasonic meters are the most developed and widely used technology for flare gas measurement, with thousands of meter installations worldwide. The main advantages are a wide rangeability in velocity, no moving parts and virtually zero pressure drop – a mandatory requirement in an emergency flare line. An additional advantage of ultrasonic transit-time meters is their ability to determine molecular weight, and hence density of the gas, from speed of sound measurement. Relevant standards for general ultrasonic gas flow meters include ISO TR 12765 [4] and BS 7965 [5]; BS 7965 provides some guidance on the application of ultrasonic meters to flare measurement under its Class 4 category of meters. Depending on the meter type and line size, the velocity range of a flare gas ultrasonic meter is quoted as between 0.03 m/s to 100 m/s. Uncertainty for a flare gas ultrasonic meter is typically specified by the manufacturers as 2.5 - 5% on velocity across the range 0.3 – 80 m/s, increasing as velocity reduces. However, these specifications are only strictly applicable under ideal flowing conditions (i.e. relatively stable, fully developed flows that are free of liquid droplets, solids and ultrasonic noise generated by valves etc. with the critical dimensions accurately measured). Any additional uncertainty arising from deviations from the ideal must be added to the quoted baseline uncertainty to arrive at the total uncertainty figure. These additional uncertainty components should be considered when determining the overall figure for emissions reporting purposes. Hydrocarbon Management Committee document HM58 [6], published by the Energy Institute, Aberdeen provides guidance on the determination of flare gas quantity for environmental reporting purposes. It includes a section identifying the sources of uncertainty on a flare gas ultrasonic meter and a methodology of determining and correcting for installation error.
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
3
3.1 Installation of Ultrasonic Flare Gas Meters Operators have the opportunity to fit spool-pieced flare meters into new oil and gas facilities. However, older installations will not have had a flare meter installed during construction. Therefore, unless production can be stopped for a period to allow breaking into the line, meters will likely have to be retrofitted to the existing flare line using either cold- or hot-tap welding techniques. Space on offshore oil and gas platforms is at a premium and, as a result, it is often impossible to meet the upstream and downstream length requirements specified by the meter manufacturers to ensure that there is no additional uncertainty in the measurement. These are most commonly stated as 20 straight pipe diameters (20D) upstream and 5 or 10D downstream.
One of the key issues regarding flare gas meter uncertainty is the effect of upstream installation on flow profile. Fixtures such as bends, valves and reducers tend to disturb the flow such that the velocity profile deviates from the ideal, axi-symmetric shape. An error occurs because the calculation of average velocity (and hence volumetric flow) relies on a correction factor that is based on the assumption of an ideal flow profile. In addition to a change in the shape of the axial flow profile, non-axial velocity components (causing the flow to swirl) will also affect the meter reading. It is also worth noting that increased turbulence generated by some fittings may also cause problems with repeatability, especially when the meter is in close proximity to the fitting. Flare-gas ultrasonic meters generally employ a single, wetted beam-path which means that they are particularly sensitive to flow profile compared with multipath meters. Dual-path designs are also available which provide an improvement in uncertainty. The work summarised in this paper concentrates on the more common single-path design.
3.2 The GE Sensing GF868 Flare Gas Ultrasonic Meter The GF868 ultrasonic flare gas meter comprises two ultrasonic transducers inserted into the flare gas line through bosses welded onto the pipework). The transducers can be installed in various path configurations to allow ease of access, but are often inserted from the top of the pipe (Fig. 2). The transducers may be installed through isolation valves allowing retraction of the transducers for maintenance and inspection purposes without the need to shut down the process.
The flare gas meter measures velocity along the path, which usually at an angle of 45° to the flow. This is then converted to an average velocity via a correction factor; average volumetric flowrate can be calculated through input of the meter internal diameter. Volume flowrate can be calculated at standard reference conditions using a measurement of temperature and pressure. Finally, density and, therefore, mass flow can be calculated from speed of sound measurement, temperature (and in more recent designs, pressure) using a proprietary correlation for average molecular weight. GE Sensing supplied a dual-channel GF868 flare gas ultrasonic meter to TUV NEL for testing in July 2008. This allowed for two separate, single path measurements to be read
by switching between the channels in alternation. The purpose of the dual-path meter was to assess the effect of installation on two different path positions during an installation test. This is of interest as smaller meters (typically < 14-inch) will tend to have a diametric path, whilst larger meters will tend to have a mid-radius path.
Fig. 2 - General Arrangement of the GF868 meter (www.gesening.com)
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
4
3.2.1 Spool piece and meter configuration A spool piece was manufactured by GE Sensing in order to house the ultrasonic transducers at the diametric and mid-radius positions within a 12” Sch. 20 pipe (i.d. 311 mm), as shown in Fig. 3. The transducers are inserted through four, 3-inch (76 mm) i.d. flanged bosses. With the flow direction from left-to-right in Fig. 3, the diametric path is upstream of the mid-radius path, three pipe diameters (3D) from the inlet flange of the meter spool. The mid-radius path is 3D downstream of the diametric path in the same plane, but offset in upper portion of the pipe by half the pipe radius (R/2 = 77.75 mm). Both paths were aligned at 45° to the flow. Pressure and temperature were measured at 2D and 3D downstream of the mid-radius channel respectively.
The meter was set up to take the average of 32 readings of transit time difference, t, from each channel, the electronics periodically alternating between channels. The sequence of obtaining the 32 readings, processing and averaging took approximately 7 seconds for each channel. The output from the two meter channels was read by TUV NEL’s data acquisition system via two separate 4-20 mA signals.
a)
b) c)
Fig. 3 - Dimensioned Drawing of the 12” Spool Housing the two Sets of Ultrasonic Transducers: a) Plan View of Spool, b) End View of Diametric Path, c) End View of Mid-
Radius Path Figure 4 shows a close-up detail of the transducers installed in the bosses for the diametric (Fig. 4a) and mid-radius (Fig. 4b) path positions respectively. The transducers were set such that the path length was the same (329 mm) in both cases, meaning that the mid-radius transducers were retracted into the boss about 30 mm further back than the diametric path.
Flow
Pressure Temp
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
5
a) b)
Fig. 4 - Detail of the Transducers at: a) Diametric, and b) Mid-radius Positions (Dims. in mm) The meter determines the mean flow velocity along the length of the ultrasonic beam-path by measuring the difference in time taken for the ultrasonic pulses to pass between the transducer pair in the upstream (t21) and downstream (t12) directions in rapid succession. When gas is flowing, the beam will take longer to travel in the upstream direction (against the flow) compared with the downstream (with the flow); therefore t21 > t12. It can be shown that the path velocity, vp, is a function of the meter geometry and the transit
time difference, t, and is thus independent of the speed of sound
1221cos2 tt
tLv p (1)
Assuming ideal flow conditions, with a fully developed velocity profile in the meter, the mean
flow velocity across the pipe cross-section, v , and hence the volumetric flowrate, Q, can be
calculated thus:
pkvv (2)
and 2RvQ (3)
The meter factor, k, is used to convert measured path velocity to a mean pipe velocity and is normally set to a value based on the assumption of an ideal, fully developed flow profile. The meter factor is usually set as variable with Reynolds number for a diametric path and constant for a mid-radius path. Since the reference flow was measured during the TUV NEL tests, the meter factor was simply set as 1.0 for both the diametric and mid-radius channels. In doing so this allowed k to be determined by simply dividing the reference velocity through the meter (as calculated from the reference mass flowrate, local density and meter diameter), with the average velocity output by the meter (also based on the same diameter).
4 INSTALLATION EFFECT TESTS Table 1 summarises the tests performed on the flare gas meter. The installation error was determined by comparing the velocity output obtained at the two path positions with the single bend at various locations upstream of the meter with that measured under baseline conditions. Curve fits were applied to the two baseline data sets to ensure that the comparison was undertaken at the same velocity. The installation effect tests were performed with the bend in the vertical orientation relative to the meter. The bend was placed 5, 10, 20 and 45D from the meter inlet flange and the
26
76
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
6
installation error was determined at 0.25, 0.5, 4, 7, 15 and 30 m/s respectively. The equivalent pipe Reynolds numbers are as given in Table 1. The baseline test was repeated at the lower velocity end following unexpected results at such low Reynolds numbers.
Table 1 – Summary of tests carried out on the meter
Nominal
Velocity
(m/s)
Nominal
Volume flow
(m3/hr)
Nominal
Reynolds
numbers
Baseline test 0.25 – 30 68 – 8,200 5,000 – 600,000
Bend at 5D 0.25 0.5 4 8 15 30
70 140 1,000 2,200 4,100 8,200
5,000 10,000 80,000 160,000 300,000 600,000
Bend at 10D as above as above as above
Bend at 20D as above as above as above
Bend at 45D as above as above as above
Repeat
baseline
0.25 – 1.5 70 – 400 5,000 – 30,000
Figure 5 shows a schematic of the Atmospheric Gas Flow Measurement Facility for the baseline and installation test cases. Ambient air under a slight vacuum was drawn into the test line from the gas flow laboratory via a 70 kW, centrifugal fan which vents into a basement area. The fan runs at a constant speed and the flowrate through the test section is set by adjusting a bypass section and/or a set of guide vanes. The fan can deliver up to around 15,000 m
3/hr depending on pressure drop. Figure 6 shows a photograph of the test meter as
installed in the specially made spool-piece. Insulation material was fixed to the crown of the pipe following some issues with noise on the readings from the mid-radius channel. This approach is commonly used in the field to attempt to attenuate ultrasonic waves which may be “leaking” from one transmitter to the other across the surface of the pipe. Although there was variability in the signal-to-noise levels, the magnitude of the noise was not thought to be too serious and the testing was progressed regardless. In the baseline case (Fig. 5a) there was 45D of straight pipe upstream of the meter inlet flange (giving 48D to the diametric and 51D to the mid-radius paths respectively). Figure 5b details the pipework configuration for the case with the bend at 45D from the meter inlet flange. In this case, the bend was simply attached to the end of the pipe that had previously served as the inlet for the baseline tests.
Reference flow rate was measured by one of four orifice plates (of diameter ratio, = 0.2, 0.3, 0.5 and 0.75) installed in an 8-inch holder. There was 30D of straight length upstream of the orifice plate holder and 8D downstream of it. The orifice plates were calibrated over an equivalent Reynolds number range in TUV NEL’s water-flow facility, which is the primary standard for the UK. The expansibility equation in ISO 5167:2 [7] was then used to correct for changes in throat density due to temperature changes. The uncertainty in the reference mass flowrate for the test facility orifice plates is calculated to be 0.5% (95% confidence). However, given the comparative nature of installation effect testing, repeatability is more important than the uncertainty. The repeatability on the measurements was found to be dominated by the scatter on the average velocity obtained by the meter as will be discussed later. Figure 7 shows a photograph of the bend inlet section at 10D from the meter inlet flange. The bend included a 10D inlet length to allow the flow to reattach to the pipe wall on entry before reaching the bend. Care was taken to ensure that the plane of the bend was aligned with the meter.
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
7
a)
b)
Fig. 5 – Schematics of the configuration used to test the flare gas meter: a) baseline tests, b) single bend tests (note: not to scale).
FLOW
FLOW
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
8
Fig. 6 - Detail of the Transducers and Bosses at: a) Diametric, and b) Mid-radius Positions
4.1 Data Acquisition and Instrumentation Used The resolution of the digital reading of transit time was 0.01 m/s. The output signal from the two ultrasonic meter channels was 4-20 mA and these were recorded by two separate channels in an Agilent data logger. The
resolution on the measured current was 4.8 A. The full-scale value at 20 mA was adjusted as velocity decreased to ensure that the resolution error did not exceed 0.25% on velocity at the lowest measured current (6 mA). It was therefore most convenient to adjust the full scale during changeover of the reference orifice plate. All instrumentation was calibrated using equipment traceable to National standards. Static pressure at the orifice plate and ultrasonic meter was measured using two Druck DPI 142 absolute pressure gauges. Temperature was measured at the orifice plate using a 4-wire PRT and at the ultrasonic meter using a 4-wire Emerson PT100 sensor mounted in a thermowell. All temperature measurements were logged via precision resistors within an Agilent data logger. The temperature probes were calibrated in a heated bath over the range 0
to 60°C to an uncertainty of better than 15 mK, which equates to 0.005% at 20°C. Differential pressure across the reference orifice plate was measured using a Yokogawa DP Harp EJX110A digital sensor calibrated against a dead weight tester to better than 0.08% (95% confidence) over the range 100 - 20,000 Pa. The lowest differential pressure measured during the tests was about 500 Pa. The absolute pressure sensors were calibrated to an uncertainty of better than 6 Pa (95% confidence) across the range 750 – 1150 mbar, equating to 0.06% (95% confidence) at atmospheric pressure.
FLOW
Diametric path
Mid-radius path
Insulating material
Fig. 7 – Single bend installation
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
9
Ambient pressure, temperature and humidity close to the inlet of the test line were recorded at the start of a test and assumed to be constant over its duration. The humidity was corrected to the pressure and temperature conditions at the ultrasonic meter and orifice plate respectively so that the air density could be calculated at these positions. The humidity sensor was calibrated to an uncertainty of better than ± 1% of relative humidity (e.g. at RH = 50%, the actual reading will be somewhere between 49% and 51%)
4.1.1 Stability issues Given the comparative nature of installation effect testing, it is the repeatability of the results, rather than the overall uncertainty, that is most important. During testing it was noted that the standard deviation on the meter velocity was much higher than on the reference flow rate and tended to dominate the repeatability in the results. Repeatability became an issue below about 1 m/s. The main reasons for this were:
The resolution on the digital output of velocity was limited to ± 0.01 m/s. This translated to ± 4% on an individual reading at the lowest velocity of 0.25 m/s, reducing to 1% at 1 m/s and 0.03% at 30 m/s. In theory at least, the uncertainty on the average velocity logged during the test point reduces with the square-root of the number of sample points. It should also be noted that GE Sensing quote an uncertainty on velocity of ± 20% over the range 0.03 to 0.3 m/s.
The protruding diametric probes caused instability to the velocity on the mid-radius path, particularly at lower velocities where the flow was straddling the laminar-turbulent transition region.
The logging speed was slowed down by the use of a single flow computer to switch between the two measurement channels in alternation. This, together with the molecular weight calculation (which could not be switched off), incurred a delay of about 7 seconds between readings thus limiting the number of readings which could be taken over a reasonable time period.
Even at zero flow the signal-to-noise level on the mid-radius path, although of a high enough magnitude, was seen to be variable and this may have contributed to the stability of the reading.
The above issues are being addressed in the next phase of the project (as listed in the further work section of this paper).
5 CFD MODELLING OF THE FLARE GAS ULTRASONIC METER
The CFD modelling was undertaken using Fluent v6.2.3 solver and post-processing package and Gambit v2.4.6 meshing software [8]. The default parameters for air were used to model the flow: Density 1.225 kg/m
3
Viscosity 1.784 ×10-5
kg/m-s
The flow was assumed to be steady and incompressible throughout. The realisable k- turbulence model was used in most cases. More information on the CFD method, and turbulence modelling, can be found in standard texts [9]. The CFD modelling of the ultrasonic meter was undertaken in three separate stages: 1. Bend model 2. Baseline model 3. Meter model The bend and baseline models are run in order to generate the inlet boundary conditions for the meter model at each given flowrate. The velocity in the x, y and z directions, and turbulence parameters, at the relevant locations are then transferred from these two models to the relevant meter model via so-called “profile files”.
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
10
The grid geometry of the bend model is shown in Fig 8. The flow at the inlet was modelled using a cube-shaped pressure boundary. This allowed air to be drawn into the pipe from all directions and along the outer surface of the pipe. The grid employed a total of 2.5 million computational cells. The majority of these were triangular prisms which were concentrated inside the pipe. The cross sectional grid (inset of Fig. 8) was made up of triangular elements that exactly matched the grid of the meter model (Fig. 9) so as to minimise interpolation errors. The baseline model was similar to the bend model except that the bend was replaced with the 50D length of straight pipe. This model had a total of 1.7 million cells. The pipe wall and inlet flange were set as smooth and a no-slip boundary condition (i.e. velocity is zero at the wall in all directions) was assumed. The standard wall function model was used to calculate the flow properties in the cell immediately adjacent to the pipe wall. The flange at the inlet of the 10D pipe was also included in the model. Figure 9 shows the computational mesh on the walls of the flare gas ultrasonic meter with the wetted portion of the transducers included. The meter model used just under 2 million cells with the grid being concentrated around the transducers and along the beam paths (indicated by the dotted lines). The mesh is refined local to the transducers and was structured and uniform along the beam paths. To minimise interpolation errors - and to ensure a smooth, regular velocity curve was obtained from each path - the grid (and beam paths) were constructed such that the path passed directly through each of the 66 node points on the line. Several integration methods were used to determine an average velocity along the beam paths, but it was found that a step-wise, area weighted-average was adequate (this is a standard reporting option within Fluent).
Fig. 8 – Mesh and geometry used to simulate the upstream bend installation
5.1 Method of Determining Installation Error Using CFD Figure 9 details the computational grid used to model the flare gas meter with the beam paths superimposed onto it. The average velocity in the flow direction, v , is calculated from the
average velocity in the direction of the path, vp, using
Pressure inlet
Velocity outlet
10D
50D
Grid used in pipe cross-section
vp
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
11
cos
pvv , (4)
where is the beam angle (45°).
The meter factor, k, is given by the ratio of the mean pipe velocity, v , to the axial velocity by
v
vk (5)
The meter error is then expressed as:
100%i
ib
k
kkError (6)
where ki is the meter factor calculated for the relevant installation and kb is the meter factor calculated for the baseline case.
6 COMPARISON OF CFD RESULTS
WITH TEST DATA Figure 10 compares the results of the installation error plotted against distance from the bend to the beam path for the diametric path at 0.25, 0.5, 4 and 30 m/s respectively. These equate to approximate pipe Reynolds numbers of ReD = 5,000, 10,000, 80,000 and 600,000. The CFD predicts the correct general trend for the diametric path starting off as negative and (but for one point on the 0.25 m/s curve) reducing in magnitude towards zero as distance from the bend increases. The CFD results are within 2 - 4% of the test data for
velocities above 4 m/s and between 4 - 6%, for velocities below 0.5 m/s. The CFD model also correctly predicts the more rapid flattening-off of the error trend with distance from the bend which occurs at the lower velocities. However, it is noted that the test data generally lies below the CFD data (i.e. a higher negative error is evident) and is more velocity-dependent. The CFD does not agree well with the test data for the mid-radius path, both in terms of the magnitude of the error and the general trend (Fig. 11). The error is of the opposite sign and the shapes of the curves are almost the mirror-image of the test data. The reason for this is the subject of ongoing research but it is thought to be largely due to non-ideality in the tests. It is anticipated that the mid-radius path is much more sensitive to swirl and flow profile than the diametric path and may have been adversely affected by the disturbance to the flow caused by the upstream diametric path. Although some deviation would have been expected at the lowest velocity of 0.25 m/s, the resulting error trend was unexpected – the error actually increasing with distance from the bend. A repeat of the baseline results revealed that there was an issue with reproducibility of the baseline results at 0.5 m/s and, especially, at 0.25 m/s, which is most likely due to laminar-turbulent transition occurring and, therefore, instability in the flow profile. There was also high scatter observed on the mid-radius data at this flowrate which was likely due to a combination
Fig. 9 – Grid on the wall of the flare meter model including transducer tips
(dotted lines = beam paths).
vp dia
vp mid rad
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
12
of the flow regime and the disturbance caused by the protruding transducers of the diametric path. GE Sensing were consulted during the project and provided the following feedback:
The mid-radius path is not normally used on flare lines of less than 14-inch diameter with the result that there would have been increased interaction between the transducer and the wall during the tests in 12-inch line
The disturbance from the protrusion of the diametric transducers would likely have caused problems with the mid-radius path.
The diametric transducers could have been retracted to the wall and did not need to be set such that the distance between them was the same as for the mid-radius path.
GE Sensing are aware of the issues at low Reynolds number where stratification will occur; a dual-path meter configuration would be recommended in such circumstances.
-16
-14
-12
-10
-8
-6
-4
-2
0
0 20 40 60
No. of Pipe Diameters to Path
Ins
talla
tio
n e
rro
r (%
)
4 m/s (CFD)
30 m/s (CFD)
4 m/s (test)
30 m/s (test)
a) b)
Fig. 10 - Comparison of Installation Error Determined by tests and CFD for the diametric Path (ReD = 80,000 and 600,000)
-10
-8
-6
-4
-2
0
2
4
6
8
10
0 10 20 30 40 50 60
No. of Pipe Diameters to Path
Ins
talla
tio
n e
rro
r (%
)
4 m/s (CFD)
30 m/s (CFD)
4 m/s (test)
30 m/s (test)
a) b)
Fig. 11 - Comparison of Installation Error Determined by tests and CFD for the mid-radius path
(ReD = 80,000 and 600,000)
6.1 Comparison of CFD velocity profiles with published LDV data
The question arises as to whether or not the CFD correctly predicts the velocity profiles through the single bend and in the downstream pipe. In order to address this issue, it was decided to compare the predicted velocity profiles with available velocity profiles measured by Laser Doppler Velocimetry (LDV). These data were kindly supplied to TUV NEL by the
-25
-20
-15
-10
-5
0
5
0 10 20 30 40 50 60
No. of Pipe Diameters to Path
Ins
talla
tio
n e
rro
r (%
)0.25 m/s (CFD)
0.5 m/s (CFD)
0.25 m/s (test)
0.5 m/s (test)
-20
-10
0
10
20
30
40
50
0 10 20 30 40 50 60
No. of Pipe Diameters to Path
Ins
talla
tio
n e
rro
r (%
)
0.25 m/s (CFD)
0.5 m/s (CFD)
0.25 m/s (test)
0.5 m/s (test)
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
13
National Institute of Standards and Technology (NIST) [10]. The profiles were taken across the horizontal and vertical axes downstream of a single bend tested in water at ReD = 10,000 and 100,000 and at about 3, 6, 11, 15, 19, 23D from the outlet of a 2-inch single bend. Figure 12 compares the horizontal and vertical velocity profiles at 3D and 23D downstream of the single bend at ReD = 10,000 (the results at ReD = 100,000 were similar). The CFD is in moderately good agreement with the LDV data close to the bend (especially on the horizontal axis), but deviates further downstream, there being more asymmetry in the CFD profiles compared with the LDV data. Additional CFD modelling revealed that it would require a total of about 75D downstream of the bend before the predicted flow profile recovered back to fully developed.
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-0.5 -0.25 0 0.25 0.5
x/D
No
rma
lis
ed
ve
loc
ity
CFD - FD
CFD
LDV - FD
LDV
a)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-0.5 -0.25 0 0.25 0.5
x/D
No
rmali
sed
velo
cit
y
CFD
LDV
b)
Fig. 12 - Predicted axial velocity on the horizontal (x) and vertical (y) axes compared with LDV data [10]: a) 3D and b) 23D from the bend outlet (ReD = 10,000)
-0.5
-0.25
0
0.25
0.5
0.6 0.8 1 1.2 1.4
Normalised velocity
y/D
CFD
LDV
-0.5
-0.25
0
0.25
0.5
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Normalised velocity
y/D
CFD
LDV
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
14
However, in comparing the fully developed profiles (plotted in Fig. 12a) it is noted that the CFD profile is slightly rounder than the LDV profile. Therefore it might be expected that the velocity profiles downstream of the bend would also differ. It is therefore more appropriate to calculate the percentage shift between the average velocity determined for the disturbed flow case with that of the baseline case. This then gives an approximation to the installation error which can then be compared with the CFD and test data results. Figure 13 compares the results of analysis of the LDV data with the CFD and meter test data for a velocity of 0.5 m/s (ReD = 10,000). The LDV data set was generated by taking the average velocity along the horizontal diameter in fully developed and disturbed flow conditions and calculating the percentage difference between them. It was only possible to do this for the diametric path position. There is excellent agreement between the CFD and LDV data, which is an encouraging result as it validates the CFD to some degree. However, although the trends in the ultrasonic flowmeter test data are similar, they show a slightly larger shift than the CFD and LDV results perhaps indicating that the flow profiles were not quite the same in the tests on the ultrasonic meter as in the case of the CFD or the LDV data.
6.2 Discussion of CFD results A single bend generates an asymmetric flow profile that also has non-axial velocity components present. Figure 14 details the velocity contours for the cases of fully developed flow entering the meter (Fig. 14a) and with the meter immediately downstream of the bend (Fig. 14b). The curvature of the bend causes the flow to be thrown to the outside of the pipe creating a skewed velocity profile which will then slowly develop down the pipe. The asymmetry is accompanied by non-axial velocity components (i.e. swirl).
-20
-15
-10
-5
0
5
0 10 20 30 40 50
% E
rro
r
No of diameters between bend and beam path
Analysis of NIST's LDV data
CFD analysis
FG meter Test data
Figure 13 – Comparison of installation error determined from LDV data and CFD analyses with test meter data (ReD = 10,000)
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
15
6.2.1 Effect of swirl A single bend tends to generate a double, counter-rotating vortex pattern (as shown by the vectors superimposed on Fig. 14b), whilst an out-of-plane double bend tends to generate a single vortex
†. In
theory, provided the swirl patterns are centred on the axis of the pipe, the diametric path should be less sensitive to swirl than the mid-radius path. This demonstrates the importance of using the method described in Section 5.1 to model installation effect which takes into account both asymmetry and swirl.
6.2.2 Effect of the transducer tips It was of interest to assess the effect of ignoring the transducer tips in the models. Figure 15 shows how the velocity contours are locally disturbed by the presence of the transducer tips within the CFD model. Figure 16 details the effect on the installation error if the transducers are not included in the model. The percentage difference between the installation error determined using models without transducers and those with transducers included is plotted against normalised distance from the bend (i.e. number of diameters from the bend to the meter inlet). Apart from the mid-radius beam at 30 m/s, the effect is generally limited to 1%, but it does appear that the intruding transducer tips influence the error more as the bend gets closer to the meter.
† Note: in reality the flow pattern will probably contain elements of both patterns depending on
bend separation, Reynolds number, upstream fittings etc.
Fig. 15 - Velocity contours through meter model at 30 m/s (fully developed flow at inlet)
a)
b)
Fig. 14 – Predicted velocity distribution at the meter inlet: a) fully developed flow, b) bend immediately upstream of meter
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
0 10 20 30 40 50
No. of Pipe Diameters from inlet
Dif
fere
nce (
%) Diametric, u=30 m/s
Mid-radius, u=30 m/s
Diametric, u=0.5 m/s
Mid-radius, u=0.5 m/s
Fig 16 - Effect of modelling the installation error
without transducers at 0.5 m/s and 30 m/s
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
16
7 CONCLUSIONS The CFD simulations agreed well with the test data for the diametric path position and predicted the correct trend with distance from the bend. The predictions were within 2 - 6% of the measurements in all cases. Further confidence is given by the fact that there was excellent agreement between the CFD and published LDV data based on analysis of the velocity profiles along the horizontal diameter. However the errors at the mid-radius position were significantly different to the test data, especially at the lowest velocities where very large errors were recorded during the tests (i.e. between 5% and about 44%). The error was also seen to increase with distance from the bend, which was somewhat unexpected and counter-intuitive.
There was significantly more scatter on the output from the mid-radius path than on the diametric which may be attributed, in some part, to a disturbance created by the diametric transducers which were only 3 pipe diameters upstream of the mid-radius path. It is also possible that the flow profile was unstable at the lowest velocity tested owing to the flow undergoing laminar-to-turbulent transition. The problems seen at the lowest flow rates are a reflection of the difficulties faced in measuring such low Reynolds number flows – especially considering application to offshore flares where the flare line can be subject to pulsations from winds, temperature differentials etc. Further technical discussions with GE Sensing have raised the possibility that the signal noise on the mid-radius channel (seen even at zero flow) could have been the result of reflections of the ultrasonic beam occurring due to the transducer tip being so close to the pipe wall. It is possible that the electronic noise could have caused the large variation in meter error at low flow. The influence of the transducer tips on the resulting installation error was also examined using the CFD. The results were more sensitive to the inclusion of the transducer tips as the bend was moved closer to the meter. This is perhaps as would be expected since the swirl is more intense closer to the bend outlet creating increased interaction of the flow with the transducer tips. There was no obvious trend with velocity.
8 FURTHER WORK
The issues raised in this paper are being examined under an ongoing follow-up project. Some of these are listed below:
Further testing will be carried out with a modified meter arrangement (i.e. removing the diametric path to observe if there is any effect on mid-radius path). This phase will involve close working with GE Sensing to ensure there are no issues with signal noise etc. resulting from set-up.
GE Sensing will be supplying a GC868 meter with an additional digital signal processing board that allows a much faster response time (~5 Hz as compared with 0.2 Hz when using the GF868 logging from two channels).
Inclusion of the transducer bosses in the CFD models.
CFD modelling of meters with dual- or multi-path transducers to examine the effect on the installation error.
9 ACKNOWLEDGEMENTS The author wishes to thank Cliff Probert of TUV NEL for carrying out the testing, GE Sensing for the supply of the GF868 flare gas meter and Jed Matson, Hilko den Hollander, Gordon Mackie and colleagues at GE Sensing for their advice and technical support during this project.
27th International North Sea Flow Measurement Workshop 20
th – 23
rd October 2009
17
10 NOTATION D Pipe diameter k Meter factor L Path length Q Volumetric flowrate R Pipe radius t12 Transit time in upstream direction t21 Transit time in downstream direction
t Transit time difference v Velocity
Beam angle
11 REFERENCES [1] TUV NEL Report 2008/301 Installation effects on a flare gas ultrasonic meter (Draft) [2] Directive 2003/87/EC of the European Parliament and of the Council of 13 October
2003 establishing a scheme for greenhouse gas emission allowance trading within the Community and amending Council Directive 96/61/EC (http://ec.europa.eu/environment/climat/emission/implementation_en.htm)
[3] Establishing Guidelines for the monitoring and reporting of greenhouse gas emissions
pursuant to Directive 2003/87/EC of the European Parliament (2007) (http://eurlex.europa.eu/LexUriServ/site/en/oj/2007/l_229/l_22920070831en00010085.pdf)
[4] BRITISH STANDARDS INSTITUTE. Guide to the selection, installation, operation and
calibration of diagonal path transit time ultrasonic flowmeters for industrial gas applications, BS 7965 2008.
[5] INTERNATIONAL STANDARD ORGANISATION. Measurement of fluid flow in
closed conduits - methods using transit-time ultrasonic meters, BSI ISO/TR 12765, 1998.
[6] Hydrocarbon Management Committee. Determination of flare quantities from
upstream oil and gas facilities, HM58, Energy Institute, London, May 2008. [7] INTERNATIONAL STANDARD ORGANISATION. Measurement of fluid flow in
closed conduits - Part 2: Orifice plates. ISO 5167-2, Geneva: International Organization for Standardisation, 2003.
[8] ANSYS Fluent website http://www.fluent.com/ [9] Versteeg, H.K., and Malalasekera, W. An Introduction to Computational Fluid
Dynamics - The Finite Volume Method. Longman Scientific and Technical Publications, New York, 1st Ed., 1995.
[10] T.T. Yeh, National Institute of Standards and Technology, Gaithersburg, MD. Private
communication, 2000.
Nitrogen subtraction on reported CO2 emission using ultrasonic flare gas meter Kjell-Eivind Frøysa, Christian Michelsen Research AS, Bergen, Norway Henning Ekerhovd, StatoilHydro ASA, Kollsnes, Norway Atle A. Johannessen, Fluenta AS, Bergen, Norway 1 INTRODUCTION The CO2 emission from flaring is typically measured by ultrasonic flare gas meters. In order to reduce the CO2 emissions, nitrogen purging is often utilized in situations of low flow in the flare. At such purging conditions, a significant amount of the gas flow in the flare is nitrogen. The CO2 emission data are to be reported to authorities. In order to get a realistic report of the CO2 emissions, the nitrogen purging should be subtracted from the total combustible gas flow. Ultrasonic flare gas meters measure primarily the flow velocity through the flow meter. From this, the volumetric flow rate at line conditions can be calculated using dimensions of the pipe, and by using measured pressure and temperature, the volumetric flow rate at a reference condition (for example 15 °C and 1.01325 bar) can be calculated. Such flow meters also measure the velocity of sound. From this measured velocity of sound, in combination with pressure and temperature, the density of the flare gas is estimated, and also the mass flow rate can thus be found. In the models relating the velocity of sound to the density, there are underlying assumptions regarding the gas composition. Typically, the assumption is that the gas contains hydrocarbons, in addition to up to some few percents of inert gases like nitrogen and carbon dioxide. Through the measured velocity of sound there is also a potential for estimation of nitrogen molar fraction in cases where nitrogen purging is a significant part of the flow. In the present paper, tests of such an algorithm in real flow tests at StatoilHydro’s process plant at Kollsnes is reported. At StatoilHydro’s process plant at Kollsnes outside Bergen, Norway (see Fig. 1), wet components are separated from the natural gas from the fields Troll, Kvitebjørn and Visund. For the three flaring systems at Kollsnes, a nitrogen subtraction algorithm has been implemented based on the traditional measurements of the ultrasonic flare gas meter. In the present paper, flow tests addressing nitrogen subtraction will be reported. The flare gas meter is tested in series with a fiscal multipath ultrasonic gas flow meter, and the gas quality is monitored by gas chromatography and manual sampling and laboratory analysis. In addition, new functionality for nitrogen estimation has been implemented in the ultrasonic flare gas meter, and a new test with this new functionality has been carried out. Section 2 of the paper gives some more background on the problem. Section 3 contains a closer description of the Kollsnes process plant, focused on issues of relevance for the present paper. In Section 4, the flow tests are described. Section 5 contains a general discussion of the problem, before the conclusions are made in Section 6.
Figure 1 Photo of StatoilHydro’s process plant at Kollsnes. 2 BACKGROUND CO2 emission in flaring system has according to the MRG [1] to be reported as activity data (quantity of flare gas) and CO2 emission factor. The CO2 emission is then the product of these two quantities. There are two commonly used alternatives for these quantities. These are (i): activity data in mass and CO2 emission factor in mass CO2 per mass flare gas, and (ii): activity data in standard volume and CO2 emission factor in mass CO2 per standard volume flare gas. In a nitrogen purging case, there are two ways of determining the CO2 emission via the activity data and CO2 emission factor. These are (i): to subtract the nitrogen purging from the activity data, and to use a CO2 emission factor representative for the quality of the flare gas in the case of no extra nitrogen purging, or (ii): to include the nitrogen purging in the activity data and to apply a CO2 emission factor that reflects the extra nitrogen content due to nitrogen purging (i.e. a smaller value for the CO2 emission factor than in the first case). The activity data in a flaring system is typically measured by an ultrasonic flare gas meter. This meter initially measures the flow velocity and thus the volumetric flow rate at line conditions. In order to convert to volume at standard pressure and
temperature, measurement of pressure and temperature close to the flare gas meter have to be carried out. In order to convert to mass flow rate (when relevant due to activity data in mass), the density can be obtained from an internal algorithm in the flare gas meter that calculates the density from pressure, temperature and measured velocity of sound. There are several methods for estimating the CO2 emission factor for a flare gas system. These include:
• A standard value in order not to underestimate the value • A value based on calculated gas composition from process simulations • A value based on the gas composition of some few gas samples taken from the
flare gas and analyzed in a laboratory • A value based on the gas composition of regular gas samples on e.g. daily,
weekly or monthly basis, taken from the flare gas and analyzed in a laboratory • Online gas chromatography
All these methods have advantages and disadvantages, both in flaring systems with and without nitrogen purging. These issues will not be addressed in detail here. However, two of the issues related to online gas chromatography in flaring systems are the time response and the representativity of the sample. In a flaring system where rapid changes of gas quality may happen, the uncertainty of the average CO2 emission factor over a time period may therefore be large. As an alternative to installing online gas chromatography, a metering regime with the potential of subtraction of nitrogen purging from the activity data is tested out. This system estimates the amount of nitrogen in the flare gas, and carries out the subtraction. The system is based on the measured quantities like flow rate, velocity of sound, pressure and temperature. All these quantities are measured online and continuously in the flow. 3 PRESENT SITUATION AT KOLLSNES At the StatoilHydro plant at Kollsnes, there are three flare gas lines, in addition to three export lines each equipped with gas chromatographs. This means that there is a good control over the typical natural gas that is present on the plant. The flare gas lines include (i): the high pressure flaring system, (ii): the low pressure flaring system and (iii): the maintenance flaring system. Each of these lines is equipped with an ultrasonic flare gas meter, in addition to double pressure and double temperature sensors. The metering system is monitored according to condition-based maintenance scheme. This also includes the ultrasonic flare gas meter, where internal quality parameters are monitored. The ultrasonic flare gas meters have traditionally been set up to also provide the density of the flare gas. A typical flare gas metering station is illustrated in Fig. 2.
Figure 2 Typical set-up of an ultrasonic flare gas metering station. From the gas density provided by the ultrasonic flare gas meter, the molar fraction of nitrogen is calculated in the flow computer. This calculation is based on an assumption that the flare gas consists of a natural gas part and nitrogen. In a normal set-up, the natural gas part of the flare gas is assumed to be equal to one of the export gases that are measured by online gas chromatography. The flow computer is set up and followed up by a fiscal metering system vendor. A typical example of an on-line report under nitrogen purging conditions is shown in Fig. 3. The density algorithm, calculating the density from the measured velocity of sound, pressure and temperature is therefore a crucial part of the system. In 2009 this has been up-graded. There are at present two different options in the density algorithm. These are (i): flaring option and (ii): purging option. Flaring option is typically used for high flow rates, where the flaring consists mainly of hydrocarbons. This option is similar to the traditional way of calculating the density in a flare gas meter. However, the measurements have been made more robust with respect to variations of gas composition. This is done by opening up for specification of a typical gas composition for the flaring at the specific plant. In particular, the specification of CO2 and N2 content can be relevant for the uncertainty of the estimated density. In cases where gas composition knowledge is not possible to provide, the algorithm uses default values and thereby works similar to the traditional density algorithms.
Figure 3 Snap-shot of the on-line report on estimated gas composition (Beregnet
Gasskomposisjon) for the 3 flare gas lines at Kollsnes. Purging option is typically used at low flow rates, where nitrogen may be a major part of the flare gas. In this option, the flare gas is assumed to consist of natural gas and nitrogen. The measured velocity of sound determines the fraction of each component. The natural gas composition has to be specified in order to use this option. The precision of this input is discussed in Section 5. 4 FLOW TESTS The uncertainty of activity data and CO2 emission factor in a flaring system with ultrasonic flare gas meter and subtraction of nitrogen has several contributions. Among these are the uncertainty of the measured flow rate, the uncertainty of the calculated density (in purging or flaring mode) and the uncertainty of the estimated nitrogen fraction (in purging mode). These three issues have been tested in dedicated flow tests at Kollsnes. The flow test addressing flow rate had to be carried out before the test addressing density and nitrogen fraction. This was because there was only a short time window where the flow test could be carried out (explained below). At that time, the upgraded density algorithms were not yet ready for installation.
4.1 Flow rate In 2009, a new high pressure pipe line from Kollsnes to the oil refinery at Mongstad was set in operation. This pipe line will provide Mongstad with Troll gas for the new gas power plant that will soon be set in operation. At Kollsnes the pipe line is equipped with two 6-path ultrasonic fiscal flow meters installed in parallel. Before the pipe line was set in operation, but after the metering station was installed, there was a possibility to route gas through one of the 6-path ultrasonic flow meters and then further to the high pressure flare equipped with an ultrasonic flare gas meter. This means that the same amount of gas was measured by both the 6-path fiscal flow meter at about 75 bar pressure, and by the ultrasonic flare gas meter at about 1 bar pressure. The pipe distance between the two meters was about 250 metres. This is accounted for in the analysis. The composition of this gas was in addition measured by online gas chromatography. In addition, gas samples were taken from the flare gas. These were analyzed in the laboratory at Kollsnes to provide gas composition. It should here be commented that in addition to the Troll gas through the fiscal flow meter, the flaring also consists of a more or less constant background flaring of about 400 Sm3/h. This background flaring is related to nitrogen purging and it consists therefore of significant amounts of nitrogen. It was not possible to stop this background flaring during the flow test. Therefore it should be expected that the ultrasonic flare gas meter measures a flow rate about 400 Sm3/h larger than the 6-path ultrasonic fiscal flow meter. The flow test was carried out at a flow rate of about 7500 Sm3/h. This flow rate was held for about 30 minutes in order to simultaneously stabilize the flow through both meters in series. The flow rates measured in each flow meter during the test period are shown in Figure 4. It can be seen from that figure that the flow through the 6-path fiscal flow meter was fairly stable over the entire flow test period, while the flare gas flow rate oscillated more in the first part of the flow test period. In the time period from 10:03 to 10:18 the flow through both flow meters was quite stable. Average flow rate measured by the 6-path fiscal flow meter in this period was found to be 7561 Sm3/h. For the ultrasonic flare gas meter, the similar flow rate was found to be 7917 Sm3/h. This last flow rate has to be corrected for the background flaring in order to be compared to the flow rate measured by the 6-path fiscal flow meter. The background flaring can be estimated from the measured flow rate by the flare gas meter before the flow test. In principle also the time period just after the flow test could have been used. However, in this period, there is a possibility that there are still some remaining natural gas from the flow test left in the system, and therefore, the flow rate measured by the flare gas meter just after the flow test may therefore not be representative for the background flaring. It can be seen in Figure 4 that there is a stable background flaring in the initial data in the plot, from 09:31 to 09:37. The average background flaring in this period is 381 Sm3/h. If the whole period from 09:31 to 09:46 (just before the flow test) is used, an estimate of 447 Sm3/h is found for the background flaring. It is thus expected that the background flaring is between 381 and 447 Sm3/h. This means that the measured flow rate by the flare gas meter during the stable flow period from 10:03 to 10:18 is between 7470 and 7536 Sm3/h
after correction for the background flaring. This should be compared to the measured flow rate by the 6-path fiscal flow meter of 7561 Sm3/h. This means that the deviation between the measured flow rate by the ultrasonic flare gas meter and the 6-path fiscal flow meter was between -0.3 % and -1.2 %, after correction for the background flaring. The specifications of the ultrasonic flare gas meter are that the flow rate is measured with an uncertainty of 2.5 - 5 %. The fiscal 6-path flow meter was flow calibrated prior to installation at Kollsnes, and the uncertainty is therefore expected to be below 1 %. This means that the deviation between the meters (0.3 - 1.2 %) is well inside the expected deviations, when the uncertainty specifications of the flow meters are taken into consideration.
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
09:21 09:36 09:50 10:04 10:19 10:33 10:48 11:02Time [hh:mm]
Stan
dard
vol
umet
ric ra
te [
Sm³/h
] 6-path fiscal meterFlare gas meter
Figure 4 Measured flow rates by the 6-path ultrasonic fiscal flow meter and the
ultrasonic flare gas meter during the flow test. Background flaring is not accounted for, and thus the flare gas meter is expected to measure a higher flow rate than the fiscal meter.
4.2 Density and nitrogen fraction, first test In the flow test described in Section 4.1, also tests for measurement of the gas quality was carried out. As discussed in Section 4.1, the Troll gas was sent through the 6-path fiscal ultrasonic flow meter at high pressure. Thereafter, the same gas was sent through the flare gas meter at low pressure. When flowing through the flare gas meter, the Troll gas was mixed, about 7500 Sm³/h Troll gas and about 400 Sm3/h background flaring with nitrogen-rich gas.
6-path fiscal flow meter: The pressure was here about 75 bara and the temperature about 3 °C. The average axial flow velocity was about 0.5 m/s. Typical gas composition measured by on-line gas chromatography during the test was:
• C1: 93.01 % • C2: 3.67 % • C3: 0.61 % • C4: 0.43 % • C5: 0.08 % • C6+: 0.12 % • N2: 1.67 % • CO2: 0.40 %
The fiscal flow meter measures the velocity of sound, in addition to the flow velocity. The velocity of sound measured on each of the 6 acoustics paths of the meter was reported. There was typically a span in velocity of sound (highest minus lowest velocity of sound measured by the 6 paths simultaneously) of about 0.9 m/s. The calculated velocity of sound agrees with the measured (averaged over the 6 paths) within +/- 0.7 m/s. Flare gas meter: As mentioned above, the gas is here a mixture of about 7500 Sm3/h Troll gas and 400 Sm3/h nitrogen rich background flaring. Before the Troll gas was led through the flare gas meter, the flare gas meter measured the background flaring only (see Section 4.1). At that time, the flare gas meter measured a velocity of sound of about 342 – 343 m/s, typically about 3 m/s larger than the velocity of sound in pure nitrogen. This indicates that there are small amounts of hydrocarbons mixed into the background flaring. For example, a gas of 95 % nitrogen and 5 % methane will have about the same velocity of sound as measured by the ultrasonic flare gas meter. During the flow test, with Troll gas through the two flow meters, the flow velocity through the flare gas meter was about 3 m/s. The pressure was here about 1 bara and the temperature about 3 °C. During the test, a gas sample was taken of the flare gas. This was analyzed in the laboratory, and the following gas composition was then found:
• C1: 87.61 % • C2: 3.44 % • C3: 0.61 % • C4: 0.45 % • C5: 0.09 % • C6+: 0.39 % • N2: 6.98 % • CO2: 0.38 %
The velocity of sound is calculated to be 405.96 m/s from this composition and the pressure and temperature (about 1 bara and 4°C) measured at the time of the gas
sampling. At that time the flare gas meter measured 405.59 m/s. This means that the deviation between measured and calculated velocity of sound was about 0.4 m/s. The flare gas meter also estimated the density of the flare gas. It should be emphasized that this was the traditional density algorithm in that type of flare gas meter, where no assumption of gas composition had been made. The algorithm is optimized for low molar fractions of nitrogen and carbon dioxide. From this estimate, a nitrogen fraction was calculated from the assumption that the gas consisted of a mixture of Troll gas and pure nitrogen. It was then estimated a nitrogen molar fraction in the flare gas of 7.8 %. This has to be compared to the measured nitrogen fraction (of 7.0 %). It can thus be seen that the flare gas meter predicted the nitrogen molar fraction with a deviation from reference of 0.8 % (abs). 4.3 Density and nitrogen fraction, second test After the first test (described in Sections 4.1 and 4.2), the ultrasonic flare gas meter was upgraded with new density algorithm. As described in Section 3, the new algorithm has two options:
• Flaring option • Purging option
The flaring option is similar to the traditional density algorithms of ultrasonic flare gas meters, where the measured velocity of sound is used as basis for estimation of the density of the flare gas under the assumption that the gas to a large extent consists of hydrocarbons. The main difference from the previous versions is that it is now possible to specify a typical gas composition for the flare gas when available, including nitrogen and carbon dioxide, in order to reduce the uncertainty of the density estimate. The purging option can be used in nitrogen purging conditions with large nitrogen content. In this case, the flare gas is assumed to be a combination of a specified natural gas and nitrogen. From the measured velocity of sound, the density and the molar fraction of nitrogen is estimated. For both the flaring and the purging options the typical natural gas composition for this test was specified as the Troll gas composition measured 6 months earlier (see section 4.2). The flow test was carried out during a planned event on the high pressure flare, where a compressor had to be de-pressurized, and therefore a high-flare situation took place. As in earlier flow tests, there was a nitrogen rich background flaring in addition. During the high-flaring situation and in the low-flaring situation after the high-flaring, gas samples were taken and analyzed on the laboratory. For the flow rate, there was no reference instrumentation during this test. In the low flaring conditions before the depressurizing, a background flaring of about 370 Sm3/h was measured. This number is taken as an average over a period of 2 minutes of stable flaring shortly before the high flaring case took place. The measured flow velocity during this background flaring was about 0.16 m/s. As discussed above,
the background flaring has high nitrogen content. By using the new purging algorithm, the nitrogen fraction was calculated to be 95.2 %. This is in good agreement with similar estimates half a year earlier (see Section 4.2). During the high flaring period, a sample of the flow was taken and analyzed at the laboratory. At this time, the ultrasonic flare gas meter gave the following parameters:
• Line pressure: 1.6100 bara • Line temperature: 15.9744 °C • Measured velocity of sound: 412.2596 m/s • Volumetric flow rate at standard ref. cond.: 24192.98 Sm3/h • Measured flow velocity: 6.73 m/s • Estimated line density, old model: 1.2087 kg/m3 • Estimated line density, new flaring model: 1.2203 kg/m3 • Estimated line density, new purging model: 1.1753 kg/m3. • Estimated molecular weight, old model: 18.0466 g/mole • Estimated molecular weight, new model flaring: 18.2198 g/mole • Estimated nitrogen fraction: 3.1165 %
The laboratory analysis of the gas sample gave the following results:
• Methane: 89.137 % • Ethane: 6.247 % • Propane: 0.243 % • I-Butane: 0.083 % • N-Butane: 0.032 % • I-Pentane: 0.015 % • N-Pentane: 0.008 % • Hexane+: 0.032 % • Nitrogen: 1.316 % • Carbon dioxide: 2.859 % • Argon/oxygen: 0.019 %
From the results of the laboratory analysis in addition to the pressure and temperature, the velocity of sound can be calculated as 414.7 m/s. This is 2.5 m/s above the measured value. Similarly, the density can be calculated from the gas composition as 1.2121 kg/m3. The flare gas meter (new flaring algorithm) estimated the density as 1.2203 kg/m3. This means that there is a deviation of 0.7 % between the two densities. In the interpretation of these deviations (in velocity of sound and in density), one should bear in mind that also the gas sample and laboratory analysis contain challenges and therefore also uncertainties. After the high flaring period, there is a period that is dominated by nitrogen flaring. However, there are still small residues of the high flaring gas. This means that the molar fraction of nitrogen is expected to be somewhat lower than the 95 % that was found before the high flaring period. In this period, a sample of the flare gas was taken and thereafter analyzed on the laboratory. The laboratory analysis of the gas sample gave the following results:
• Methane: 18.732 % • Ethane: 2.300 % • Propane: 0.395 % • I-Butane: 0.067 % • N-Butane: 0.088 % • I-Pentane: 0.031 % • N.Pentane: 0.029 % • Hexane+: 0.074 % • Nitrogen: 76.847 % • Carbon dioxide: 0.754 % • Argon/oxygen: 0.679 %
At the time of the gas sampling, the following data were read from the flare gas meter:
• Line pressure: 1.0210 bara • Line temperature: 17.8482 °C • Measured velocity of sound: 354.8365 m/s • Volumetric flow rate at standard ref. cond.: 463.24 Sm3/h • Measured flow velocity: 0.2045 m/s • Estimated line density, old model: 0.9976 kg/m3 • Estimated line density, new flaring model: 0.9990 kg/m3 • Estimated line density, new purging model: 1.1037 kg/m3. • Estimated molecular weight, old model: 23.6384 g/mole • Estimated molecular weight, new model flaring: 23.6718 g/mole • Estimated molecular weight, new model purging: 26.6503 g/mole • Estimated nitrogen fraction: 87.4278 %
Note that at this time, the flow rate was down to 463.24 Sm3/h and the flow velocity down to 0.20 m/s. With such a low flow rate (and flow velocity), the purging model is expected to be describing the gas quality better than the flaring algorithm. From the gas composition measured at the laboratory from the gas sample, in addition to the pressure and the temperature, the velocity of sound was calculated to 355.0 m/s. This agrees within 0.2 m/s with the measured velocity of sound. Similarly, the density was calculated from the gas composition, pressure and temperature to be 1.1041 kg/m3. This agrees well with the value estimated from the purging algorithm of the flare gas meter (1.1037 kg/m3). With respect to estimation of the nitrogen molar fraction, the agreement between the purging model of the flare gas meter and the gas analysis is not as good as for the velocity of sound and density. Here, the gas analysis indicates 76.8 %, while the purging model indicates 87.4 % nitrogen. One explanation for this deviation can be related to the model where the gas is expected to be a mix of the Troll gas and pure nitrogen. It can easily be seen that the gas sample cannot be such a mixture. For example, the molar fraction of CO2 is larger than for Troll gas. Also the molar fraction of ethane in the gas sample is much higher than expected. The explanation of
these deviations has not been found, but they should anyway be kept in mind when interpreting the estimated nitrogen molar fraction. 5 DISCUSSION AND PERSPECTIVES The flow results presented in Chapter 4 are valuable input to the more general discussion related to flare gas metering, and in particular the nitrogen subtraction prior to CO2 emission reports. In the nitrogen purging situations, the flow rates are generally low, and the nitrogen content can be quite high. If this purging gas is measured and interpreted as basically hydrocarbons, the CO2 report will overestimate the emissions. How large this over-estimation is depends on how much of the total accumulated flaring during the reporting period that is carried out as purging. This will vary from installation to installation, and possibly also from period to period. In general there are today no industrially accepted solutions for on-line composition measurements, and thus for nitrogen subtraction today. The methods that have been tried may have large uncertainties. On this back-ground, the possibility of using the on-line measurements carried out in the flare gas meter itself, which follow the changing flow rates and gas compositions in the flare is seen as an interesting alternative. The requirements from the authorities depend on the quantity of flare gas and thereby the total CO2 emission from the flaring system. Under the strictest requirements, the activity data (accumulated mass or standard volume of the flare gas) shall be determined with a documented relative expanded uncertainty of 7.5 % with 95 % confidence interval. Depending on the installation, and following the specifications of the flare gas meters, there is a potential for determining the activity data when no nitrogen subtraction is addressed, with relative expanded uncertainty of between 3 and 6 %, depending on the actual installation. This uncertainty is to be combined in an un-correlated way with the uncertainty contribution related to nitrogen subtraction, in order to obtain the total uncertainty budget for the activity data after nitrogen subtraction. To be formal, the activity data after the nitrogen subtraction, Atot, can be written in the following way:
nsubtractioNmeasuredtot AAA 2−=
where Ameasured is the activity data as measured by the flare gas meter (without nitrogen subtraction) and AN2 subtraction is the nitrogen subtraction. By assuming uncorrelated uncertainties, the uncertainty model for Atot can be written as follows:
( ) ( ) ( )
( ) ( ) ( ) 2
2
22
22
22
22
⎟⎟⎠
⎞⎜⎜⎝
⎛+⎟⎟
⎠
⎞⎜⎜⎝
⎛=⎟⎟
⎠
⎞⎜⎜⎝
⎛
+=
nsubtractioN
nsubtractioN
tot
nsubtractioN
measured
measured
tot
measured
tot
tot
nsubtractioNmeasuredtot
AAu
AA
AAu
AA
AAu
AuAuAu
where u denotes standard uncertainty. In table 1, the relative expanded uncertainty (95 % confidence level) that can be tolerated on the nitrogen subtraction before the relative expanded uncertainty of the activity data after nitrogen subtraction becomes as high as 7.5 % is shown. Numbers are given for four different relative expanded uncertainties of the activity data as measured by the flare gas meter (before nitrogen subtraction), 3 %, 4 %, 5 % and 6 %, and for four different fractions of the accumulated nitrogen purging relative to the total flaring (including the nitrogen flaring), 10 %, 20 %, 40 % and 60 %. When “Not possible” is stated in the table, even 0 % uncertainty of the nitrogen subtraction will lead to an uncertainty higher than 7.5 % for Atot. However, in many cases, with an uncertainty of the nitrogen subtraction of 10 – 20 %, the relative expanded uncertainty of the activity data, Atot, will still be below 7.5 %. It should also be commented that also in cases where the relative expanded uncertainty of Atot is above 7.5 %, the method is still of interest. In such cases the method must be evaluated against other possible on-line methods for nitrogen subtraction. Table 1 Uncertainty that can be tolerated on the nitrogen subtraction for keeping
the total uncertainty of the activity data below 7.5 %, as a function of uncertainty of measured activity data by the flare gas meter, and the fraction of N2-purging of the total flaring. All uncertainties are relative expanded uncertainties with 95 % confidence level.
Uncertainty of measured activity data, before nitrogen subtraction
Quantity of nitrogen purging in percent of total flaring
3 % 4 % 5 % 6 %
10 % 60 % 54 % 45 % 30 % 20 % 26 % 22 % 16 % 0 % 40 % 8.3 % 5 % Not possible Not possible 60 % 0 % Not possible Not possible Not possible
The uncertainty of the estimated molar fraction of nitrogen depends on several effects, including:
• Specified natural gas in the purging situation • Value of molar fraction of nitrogen • Measured velocity of sound • Measured temperature and pressure
For the flow test described above in Section 4.3, the natural gas is specified as Troll gas. This is reasonable for the Kollsnes plant. However, as part of a more general discussion, the effect of specifying non-optimal natural gases is briefly discussed. For
illustration of this issue, two other gases are selected in addition to the Troll gas. These are methane, and a more heavy gas that is purely theoretical. The gas composition of this “heavy gas” is as follows
• C1: 85 % • C2: 5 % • C3: 1 % • C4: 0.5 % • C5: 0.3 % • C6+: 0.2 % • N2: 4.0 % • CO2: 4.0 %
This means that the three gases in question have the following molar masses:
• Methane: 16.0 g/mole • Troll: 17.4 g/mole • Heavy: 19.1 g/mole
In Fig 5, the molar fraction of nitrogen as a function of velocity of sound is shown for mixtures of nitrogen and each of the three natural gases, for a pressure of 1 bar and a temperature of 20 °C. As expected, for smaller nitrogen molar fractions, the spread between the curves is large. For example, for a measured velocity of sound of 400 m/s, predicted molar fraction of nitrogen is from 10 % to 35 %, depending on which gas that is specified. However, the purging method is developed primarily for high nitrogen molar fractions. In Fig 6, a close-up of Fig. 5 is shown, for the more typical purging conditions. It can there be seen that for example for a measured velocity of sound of 360 m/s, the estimated molar fraction of nitrogen is from 74 % to 83 %. It should here be kept in mind that the span in quality between the three gases selected here may be larger than what is the case in practice. It can also be seen that at high nitrogen fractions, typically a change in velocity of sound of 1 m/s causes a change in predicted nitrogen molar fraction of about 2 % (abs), for the gases considered here. This means that the uncertainty of the measured velocity of sound is important to keep control of in order to use such an approach.
0
10
20
30
40
50
60
70
80
90
100
340 360 380 400 420 440 460
Velocity of sound [m/s]
Mol
ar fr
actio
n of
nitr
ogen
[%
]
MethaneTroll gasHeavy gas
Figure 5 Calculated data for molar fraction of nitrogen as a function of velocity of
sound for a mixture of nitrogen and each of three natural gases. Pressure: 1 bara, temperature 20 °C.
70
75
80
85
90
95
100
345 350 355 360 365 370
Velocity of sound [m/s]
Mol
ar fr
actio
n of
nitr
ogen
[%
]
MethaneTroll gasHeavy gas
Figure 6 Same as Fig. 5, but here a blow-up of the high nitrogen range.
6 CONCLUSIONS Nitrogen purging under low flaring conditions is increasingly taken into operation. In such situations, the nitrogen purging should be subtracted from the total amount of flare gas (activity data) that is reported to the authorities as part of the CO2 emission report. In this paper, nitrogen subtraction based on the measurements carried out by an ultrasonic flare gas meter is discussed. The ultrasonic flare gas meter has two options of operation, with respect to the estimation of the flare gas quality: (i) purging option and (ii) flaring option. Purging option is typically used under low flow conditions. The molar fraction of nitrogen is then estimated, in addition to the gas density. The nitrogen molar fraction can be high. Flaring option is typically used under high flow conditions. This is the traditional option, where the meter estimates a gas density under the assumption that the flare gas to a large extent consists of hydrocarbons. The verification of a system for nitrogen molar fraction estimation and thereafter nitrogen subtraction based on ultrasonic flare gas meters, is here considered to be based on gas samples and laboratory analyses. Such a verification leans on a representative manual point for gas sampling. This is not trivial, but will here just be addressed as an issue that must be considered. Flow tests have been carried out at StatoilHydro’s gas processing plant at Kollsnes outside Bergen, Norway. Under these tests, both purging and flaring conditions have been tested. The flare gas was measured by the ultrasonic flare gas meter, and at the same time also sampled and thereafter analyzed at the laboratory. Under purging conditions, the nitrogen molar fraction has been estimated by the purging method of the flare gas meter with a deviation of about 10 % or less from a reference based on laboratory analysis of a gas sample. The uncertainty of the nitrogen subtraction depends on the actual site. However, the results are promising with respect to applying this methodology for nitrogen subtraction in flaring systems with nitrogen purging. 7 ACKNOWLEDGEMENT Useful help on set-up and follow up of the fiscal metering system by Oussama Tlili and Reidar Høgvoll, FMC Technologies, is greatly acknowledged. 8 REFERENCES [1] Guidelines for the monitoring and reporting of greenhouse gas emissions
pursuant to Directive 2003/87/EC of the European Parliament and of the Council, 2006.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
COMPARISON OF MULTIPATH ULTRASONIC METER CALIBRATION DATA FROM TWO LIQUID HYDROCARBON
FACILITIES AND ONE WATER FACILITY
Gregor J Brown, Cameron, UK Terry Cousins, Cameron, USA Bobbie Griffith, Cameron USA
Donald R Augenstein, Cameron, USA 1 INTRODUCTION The comparison exercise presented here started as a bilateral inter-laboratory comparison of the Cameron calibration facility at the Caldon Ultrasonics Technology Centre in Pittsburgh with the oil flow facilities of TUV NEL Ltd in Scotland. The primary driver for the intercomparison was to provide results of proficiency testing in support of the laboratory’s ISO 17025 accreditation. NEL was selected as the second laboratory for a number of reasons, including their ability to cover an overlapping flow and Reynolds number range, the low uncertainty of their facilities and their position holders of the UK national standards. Most importantly, NEL also has ISO 17025 accreditation for their facilities and regularly participates in international intercomparison exercises, thus ensuring a high level of confidence in the validity of the comparison. The facilities at NEL are based on gravimetric (weighing) systems, traceable to the UK’s primary mass standards. The Cameron calibration laboratory uses a volumetric prover, which is in turn calibrated using a traceable volumetric tank. Good comparison results produced by these two different methods would therefore also demonstrate that the results obtained are independent of the calibration method used. When designing the transfer package to be used at the two laboratories it became clear, as with any intercomparison, that the meters should be as immune as possible to installation effects. Also as it was necessary to transport the package across the Atlantic, it would need to be compact and robust. To meet these requirements, and bearing in mind that the Cameron calibration facility is used primarily for calibration of ultrasonic meters, the decision was made use the Caldon 280Ci eight path ultrasonic in the package. The transfer package was assembled in the Cameron lab and included two 280Ci flowmeters, with an upstream straight run and a perforated plate (CPA) flow conditioner. The 8-path configuration of the Caldon 280Ci usually negates the need for a flow conditioner, but in this case the CPA plate was included in the package as additional insurance against installation effects, as the requirement was to reduce any possible meter related differences to an absolute minimum. The package was calibrated at the Cameron laboratory against the ball prover and then transferred by ship to NEL where the oil intercomparison tests were completed. The opportunity was then taken to also perform a water calibration at NEL, which would add a further data set, against another independent calibration system, and with a fluid having differing properties. Subsequently, further tests have been carried out at Cameron Caldon with the package, using a small volume prover and turbine meter combination for calibration and compared with calibrations obtained using the ball prover. By virtue of the different flow lines used, the package has been shown to have changes in calibration that are within the combined uncertainty of the measurement methods with 5 different installations. In addition to
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
validating the traceability and uncertainty of the facilities, this also demonstrates the robustness of the package and the metering technology. 2 THE CALIBRATION FACILITIES Both the Cameron and TUV NEL calibration facilities have been described in detail elsewhere, but a brief description is given here for information. 2.1 The Cameron Calibration Facilities The Cameron calibration facilities are comprised of oil storage tanks, pumps, test sections, a ball prover and a piston prover. Three types of oil are available, allowing for calibration at different viscosities. The chosen oil is pumped into the system until full and then shut off from the loop. The calibration facility is then operated as closed re-circulating loop; oil is pumped around the loop, through the selected test section, prover and reference meters. Control of temperature is carried out by passing the fluid through an in-line shell and tube heat exchanger, which is cooled by glycol passing through a chiller. The primary references in the Caldon facility are a 10m3 unidirectional ball prover and a 0.11 m3 Brooks small volume prover (SVP). Meters above 10 inches in diameter are usually calibrated using two 10-inch 8-path ultrasonic reference meters, while meters between 6 and 10 inches are calibrated against the ball power. Small meters are calibrated against the small volume prover, normally in combination with a turbine master meter. The ball prover is equipped with multiple switches allowing different calibration volumes to be chosen to suit the meter and flow rate for a test, two volumes were used for the inter-comparison calibration, the larger being a nominal volume of 10m3 and the smaller being 3.3m3. For the intercomparison between NEL and Cameron the ball prover was used to calibrate the package, using a combination of the 10 m3 and 3.3 m3 volumes. A similar process was used when comparing the ball prover with the small volume prover. The calibration facility as a whole has a range of 20 m3/hr up to 3800 m3/hr. Above 2200m3/hr the ultrasonic master meters are used, and below 100m3/hr the SVP is normally used. The ISO17025 accredited uncertainty for the ball prover is +/- 0.065% for the 3.3m3 volume, and +/- 0.04% for the large 10m3 volume (at 95% confidence limits). The uncertainty for the SVP and turbine master meter method has been certified by NMI to be 0.04% (these tests are part of the exercise for obtaining ISO 17025 for the SVP). These uncertainties are quoted with a coverage factor of k = 2. A photograph of the Cameron calibration facility is shown in Figure 1.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 1 The Cameron calibration facilities 2.2 The NEL Calibration Facilities The NEL oil calibration facility has oil storage tanks, pumps and a choice of test sections where meters are installed. Oil is circulated from the storage tanks, through the test section and either returned to the storage tank or collected in gravimetric weighing tanks prior to return to the storage tank. The temperature of the oil in the storage tanks is controlled by means of a conditioning circuit with hot and cold heat exchangers. Test meters can be calibrated either directly against the primary gravimetric standard or against a choice of reference turbine and positive displacement flowmeters. The gravimetric calibration method is a standing start and finish method, where the required flowrate is established in the test line, the flow stopped, the collection tank drained, the test started, the tank filled and then the flow stopped again and the vessel weighed.. The largest collection tank in the gravimetric facility has a capacity of 6 tonnes and the maximum flowrate for this system is 360 m3/hr. For flowrates greater than 360 m3/hr, secondary reference meters are used. The reference flowmeters are regularly calibrated at their conditions of use against the primary gravimetric standards. The ISO 17025 accredited uncertainties of the NEL oil facilities are +/- 0.03% for the gravimetric system and +/- 0.08% for calibration against reference meters (these uncertainties also being stated with a coverage factor k = 2). The specification for the 6-inch Caldon 280Ci flowmeter covers a range of 74 to 740 m3/hr. Therefore, in order to achieve the maximum flowrates required, the NEL turbine reference meters were used for the comparison. NEL offers a range of oils, and combined with good temperature control this made it possible to reproduce very similar conditions to the calibration carried out in the Cameron laboratory. A picture of the liquid flow laboratory area is shown in Figure 2, and a schematic of one of the circuits of the oil facilities is shown in Figure 3.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 2 The NEL oil (blue, right) & water (silver, left) calibration facitlites
Figure 3 A schematic of one of the NEL oil facility circuits
NEL’s water flow calibration facility is very similar in principle to the oil flow facilities, with the primary references being a series of gravimetric collection tanks. The facility has four separate flow lines, covering a wide range of flowrates in different line sizes. The main difference between the oil and water facilities at NEL is that the water facility uses a knife-edge diverter to switch the flow between the return to the sump tank and the diversion to the collection tank. As a result, the calibration points are taken with a ‘flying start and finish’, i.e. the flow through the test meter is not stopped. The largest weightank in the water facility is a 12 tonne tank which can be used for flowrates up to with 720 m3/hr. Above 720 m3/hr, up to a maximum of 1440 m3/hr, parallel reference meters can be used. The uncertainty of
Pumps
Test Lines
Weigh Tank Control
valves
Tank
Turbine reference meters
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
the gravimetric calibration system in the water lab is +/- 0.1%. A schematic of the facility is shown in Figure 4.
Figure 4 A schematic of the NEL water calibration facility 3 THE TRANSFER PACKAGE The meters used for this exercise were 6-inch Caldon LEFM 280Ci meters. These meters have been described in detail in previous publications [1], but briefly they are meters with two planes each of four paths, the paths being at right angles to one another. The basis of this design is that the pairs of crossed paths cancel the effects of non-axial (swirling) flow on the axial velocity measurement, and hence enable the 4-chord Gaussian integration technique to accurately integrate the axial velocity profile. An illustration of the 8-path meter design is shown in Figure 5.
Figure 5 An illustration of the Caldon 280Ci 8-path ultrasonic meter
Weigh Tank
Diverter
Tanks Reference meters
Test lines
Pumps
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Testing of the 280Ci meter design has shown that even with fairly extreme upstream installation conditions the calibration is not significantly affected, and hence a flow conditioner is not normally required. However, in this case a CPA plate was included in the package as additional insurance against installation effects, as the requirement was to reduce any possible meter related differences to an absolute minimum. The CPA plate was chosen because it has a relatively low pressure drop compared to other plates, and because it has performed well in tests in combination with Caldon ultrasonic meters. A 5 diameter pipe section upstream of the CPA was included as a settling length to minimize any interactions between the plate and the pipe work of the calibration facilities. As a further safeguard to ensure that moving from one lab to another would not introduce any hydraulic changes, such as might be caused by protruding gaskets or misaligned flanges, the whole package was kept bolted together for all tests at Cameron and NEL. A schematic of the transfer package is shown in Figure 6. Two 8-path meters were included in the package in order that any fault with either of the meters might be readily identified and ensure that and differences in results might be more readily traced to either the meters or the calibration facilities. The two meters were separated by 3 pipe diameters of straight pipe of matching schedule.
A and G – 5D straight length (matched schedule with internal welds ground flush) B – Perforated plate flow conditioner (CPA) C – 15D straight length (matched schedule with internal welds ground flush) D & F – 280Ci ultrasonic flow meters E – 3D 150# to 300# flange adaptor spool
Figure 6 A schematic of the transfer package
Photographs of the intercomparison package installed in the Cameron and NEL laboratories are shown in Figure 7.
B C D E F G A
Flow
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 7 The transfer package installed in the Cameron (left) and NEL (right) labs 4 THE METHOD OF COMPARISON The method of comparison used, was to leave the meters uncharacterized and compare the calibrations on the basis of pipe Reynolds number, i.e. only geometric and time constants were used in the meters, no empirical corrections were applied. The nominal Reynolds numbers chosen for the test were evenly distributed on a log scale to lie inside the normal operating flowrate range of the meters. The oil product used for the calibrations was an Exxon kerosene product with a nominal viscosity of 2.4cSt at 20oC, as this was available at both laboratories. The nominal flow range for the intercomparison was 100 - 600 m3/hr. The nominal Reynolds numbers chosen selected to give equal spacing on a logarithmic scale are given below:
Nominal Reynolds Numbers 92 829 113 278 138 231 168 681 306 514 205 839 251 182 374 034 456 429 556 973
At each Reynolds number, for both the tests at NEL and at Cameron, repeat points were carried out, in accordance with the API Manual of Petroleum Measurement Standards, Chapter 5.8, Table B-1. This table provides the spread of repeats that will give an uncertainty of the mean of better than +/- 0.027%. The meter factor used for the comparisons is defined as the ratio of reference volume (volume obtained from the facility) to the indicated volume from the meter. The differences
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
shown in the comparison graphs are defined as the percentage difference between the meter factor obtained in the NEL facility and the value obtained at the Cameron facility. A similar approach is used for the comparison between different calibrations carried out in the Cameron facility. For the initial testing at Cameron, the baseline is taken as the tests carried out in line 1. The acceptance limits for each comparison is taken as the total uncertainty obtained by calculating the combined uncertainties due to the each facility and the uncertainty due to the repeatability of the data. Thus:
Total Uncertainty = Where: U1 = Uncertainty in the first facility (e.g. Cameron ball prover) U2 = Uncertainty in the second facility (e.g. NEL oil or NEL water facility or Cameron SVP) U1R = Uncertainty due to meter repeatability in the first facility U2R = Uncertainty due to meter repeatability in the second facility 5 INITIAL TESTING IN THE CAMERON LABORATORY Initially, to check the quality of the package design, it was calibrated at the Cameron facility in two different test lines to confirm the reproducibility of the results would be satisfactory. The package was first calibrated on line 1. This has a nominal 8” line coming directly from the 24” header, and so the meter has an 8” to 6” reduction upstream of the package. After the calibration the package was moved to line 2. Line 2 has a 24” outlet from the header, this was coned down to the 6” package. The comparison of the results from these two installations are are shown in Figures 8 and 9 below. The data plotted, is the difference in the mean meter factors at the specified Reynolds numbers. As can be seen both meters fall well within the combined uncertainty of the two calibrations, and so the conclusion was that the package appeared to be stable and capable of handling substantial change in installation conditions.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 8 Difference in calibration between lines 1 and 2 in the Cameron lab – Meter 1
Figure 8 Difference in calibration between lines 1 and 2 in the Cameron lab – Meter 2 6 OIL LABORATORY INTERCOMPARISON RESUTLS The package was shipped as a complete assembly to NEL in Scotland, and installed in line B of the oil flow facility. As stated previously, the primary method of calibration is different, using a mass based system rather than the volume based system used by Cameron. In order to cover the flowrates above 360 m3/hr and to avoid changing between references, the full range of the oil calibration was carried out using turbine master meters. As the viscosity of the kerosene at NEL was slightly lower than that in the Cameron facility (2.45 cSt vs 2.67 cSt), the flowrate used at NEL was adjusted to match the Reynolds numbers as closely as possible, as illustrated in Figure 10 below.
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Meter 1 Combined uncertainty acceptance limits
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Meter 2 Combined uncertainty acceptance limits
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 10 Comparison of calibration flowrates used at Cameron and NEL
As can be seen from the graphs Figures 11 & 12, the differences for both meters fall between the uncertainty acceptance limits. It should be noted, that both meters exhibit a similar difference curve peaking at a Reynolds number of approximately 230,000, with a non-linearity of around +/- 0.05%. As the curve lies inside the uncertainty bands, it is difficult to attribute this characteristic to any particular source but it suspected to arise from residual errors the characterization of the turbine meters. This is discussed further in the following section.
Figure 11 Difference in calibration between Cameron and NEL oil labs – Meter 1
0
100
200
300
400
500
600
10,000 100,000 1,000,000
Flowrate (m
3/hr)
Reynolds number
Cameron laboratory test points
NEL oil laboratory test points
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Meter 1 Combined uncertainty acceptance limits
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 12 Difference in calibration between Cameron and NEL oil labs – Meter 2 7 WATER FACILITY CALIBRATION RESULTS Following completion of the oil intercomparison, the package was tested in the NEL water flow facility. As the main intercomparison had been completed successfully, meter 2 was now used for some R&D testing and meter 1 was left unaltered. The flowrates for the water calibration were chosen to overlap with the oil calibration but also to remain within the normal volumetric flowrate/velocity range of the meters. The relationship between the oil and water test points in terms of Reynolds number and flowrate are shown in Figure 13 below. Note that only the four lowest flowrates were used for the comparison between water and oil.
Figure 13 Flowrates and Reynolds numbers used for the comparison between oil and water
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Meter 2 Combined uncertainty acceptance limits
0
100
200
300
400
500
600
700
10,000 100,000 1,000,000 10,000,000
Flowrate (m
3/hr)
Reynolds number
Cameron laboratory test points
NEL oil laboratory test points
NEL water laboratory test points
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 14 below shows the results from the water calibration of meter 1 compared against the oil calibration in the Cameron facility. Also shown on the graph for reference is the NEL/Cameron oil comparison data. It is interesting to note that the comparison appears better (almost constant at around + 0.02%) in the case of the NEL water calibration. The Cameron calibration was carried out against a volumetric prover. The water calibration was carried out against NEL’s weightank system (i.e. a primary facility). The kerosene calibration at NEL however was carried out using calibrated turbine meters (i.e. secondary standards), as the oil tanks cannot be used directly over the flow range of the 6-inch meter. As such is appears that the non-linearity in kerosene comparison may its origin in the calibration curve of the turbine meters. This is also suggested in the previous data showing both meters calibrated on oil, where the difference curves of the two meters have very similar humps at a Reynolds number of around 230,000.
Figure 14 Difference in calibration between the Cameron oil calibration and the NEL water calibration – Meter 1
Two months after the initial intercomparison exercise, both meters were tested in both oil and water as part of an ongoing R&D project. The same calibration methods were used as before, i.e. turbine meters for the oil calibration and the weigh tank for the water calibration. The results of these calibrations are compared in Figure 15, this time in the form of the difference between the NEL oil and NEL water facilities. Again it can be seen that the results from both meters fall within the expected uncertainty bounds for this comparison.
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
NEL oil data compared to Cameron oil calibration
NEL Water data compared to Cameron oil calibration
Combined uncertainty acceptance limits
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
Figure 15 Difference in calibration between NEL oil facility and the NEL water facility 8 REPEAT TESTS ON RETURN TO THE CAMERON LAB Some nine months after the transfer package was first calibrated it was returned to the Cameron facility and re-tested. Inadvertently, the transfer package had not be properly secured in it crate before shipping and showed some minor damage to the paint on the head-mounted electronics. Fortunately the damage was only superficial and when the meters were recalibrated it was found that they were both still in good agreement with the preceding oil calibration data, as shown in Figure 16 below.
Figure 15 Difference in calibration between NEL oil facility and Cameron repeat calibration
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Combined uncertainty acceptance limits
Meter 1
Meter 2
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.9 5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7
Diff
eren
ce (%
)
Log Reynolds number
Meter 1Meter 2Combined uncertainty acceptance limits
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
9 SMALL VOLUME PROVER CALIBRATIONS A final series of tests were carried out in the Cameron lab using the small volume prover in combination with a 6” turbine meter. The API master meter method of proving was used, where the turbine is calibrated against SVP, then the ultrasonic meter is calibrated against the turbine at the same rate and then the turbine is calibrated against SVP again. To reduce the uncertainty of the turbine meters, they were calibrated using 10 runs within a spread of 0.05%, bringing the meter factory uncertainty down to +/- 0.012%. The meter factors for the turbine taken before and after the calibration run must remain within 0.02% of one another resulting in an overall uncertainty in the master metering of method of +/- 0.04%. For these tests the ball prover was operated using the 10 m3 volume, so that the uncertainty remained constant over the full flow range. As can be seen from Figure 16 the differences are well within the combined uncertainty of the calibration methods. This verifies the uncertainty and traceability of the small volume prover by comparison with the large prover, and by virtue of using the transfer package, also directly links the comparison with the facilities at NEL.
Figure 16 Difference in calibration between the Cameron ball prover and SVP
9 NEL’S PARTICIPATION IN INTERNATIONAL INTERCOMPARIONS As mentioned earlier in the paper, NEL regularly participates in international intercomparison exercises. Of particular relevance to this paper is the BIPM International Key Comparison of Liquid Hydrocarbon Facilities (CCM-FF-K2). The intercomparison initially involved nine laboratories each designated as national standard calibration laboratories. The comparison was carried out using the BIPM guidelines for key comparisons and is included in the BIPM database to support the capability statements of the participating institutes. The key comparison was carried out using light liquid hydrocarbon across a flow range 5 to 30 l/s. Two meters, a Kral positive displacement meter and a turbine meter, were used in the
-0.15%
-0.10%
-0.05%
0.00%
0.05%
0.10%
0.15%
4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8
Diff
eren
ce (%
)
Log Reynolds number
Meter 1Meter 2Combined uncertainty acceptance limits
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
intercomparison package; however, the primary comparison used the Kral positive displacement meter. Six laboratories finally provided results to allow the calculation of a Key Comparison Reference Value (KCRV) and all six sets of results were consistent with the KCRV. The deviations from the KCRV using the Kral meter lay within a band of ±0.026% as illustrated in Figure 17.
Figure 17 Results of the BIPM liquid hydrocarbon Key Comparision
The intercomparison was led by NEL as the designated pilot laboratory. The other participants named in the graph were SP (Sweden), NMi (NMi Van Swinden Laboratory - Netherlands), FORCE (Denmark), NMIJ (Japan) and CMS (Chinese Taipei). 10 CONCLUSIONS From the comparision results given in section 6 above, it can be concluded that the Cameron calibration laboratory and the UK National Standards operated by NEL are equivalent in terms of their reference measurements. Underpinned by the inclusion of NEL in international intercomparisons of hydrocarbon facilities, this result validates the traceability of the Cameron laboratory and its associated statement of uncertainty. The comparison results link together the Cameron lab, the NEL oil facility and the NEL water facility and act to demonstrate the validity of Reynolds number based calibration using dissimilar fluids. This form of Reynolds number calibration is appropriate when the velocities and the acoustic properties of the fluids can be shown to have little influence on the uncertainty of calibration. The transfer package has also been utilized to provide validation of the traceability and uncertainty of the Cameron small volume prover reference standard. The performance of the transfer package throughout this series of tests has demonstrated that the Caldon 280Ci flowmeters are stable and robust, both in terms of metrology and
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
practicality, even after being inadvertently shipped back across the Atlantic without being properly secured in the packing crate. In the process of this exercise the transfer package reproduced its calibration in three completely separate facilities, with four different methods of calibration, giving further credence to the use of such systems for master metering. REFERENCES [1] G J Brown, D R Augenstein, H Estrada and T Cousins, “Metering of Liquefied
Natural Gas Using 8-Path Ultrasonic Meters” NSFMW, 2008 [2] “Bi-lateral Intercomparison of Flow Facilities Between TUV NEL and Cameron
Measurement Systems” TUV NEL report No: 2008/316
High viscosity hydrocarbon flow measurement,
a challenge for ultrasonic flow meters?
Jankees Hogendoorn, Karsten Tawackolian, Peter van Brakel,
Jeroen van Klooster and Jan Drenthen
Summary
The time of easy recoverable oil is fading away while at the same time the demand for refined
products is rising. To meet this demand, oil has to be recovered from fields, such as tar sands,
that are much more difficult to explore. Consequently there is also a large increase in the
variety of products to be measured as well, including many high viscosity products.
Whereas for conventional mechanical meters the measurement of high viscosity flows is
limited by the force on the bearings, one of the few meters that can successfully be applied in
these is the ultrasonic flow meter. However, that is not an easy task either.
Major issues leading to an increasing measurement uncertainty of ultrasonic meters are:
• the attenuation of the acoustical signal
• the strong dependency of the viscosity on the temperature
• the unsteady flow profile in the laminar-turbulent transition region
In order to improve on this, a development project was started, specifically aimed at the
measurement of high viscosity flows. As a result, a new meter has been designed, equipped
with new powerful transducers and algorithms capable of solving the fluid dynamic
challenges in the transition region.
The present paper describes:
• the meter design
• the design of the new transducers
• CFD calculations simulating the sensitivity of the 5-beam ultrasonic flowmeter to
changing viscosity in the boundary layers due to for thermal effects at high viscosity
applications.
• Test results of boundary layer disturbance test.
• And the test results obtained with a series of 24” flow meters tested at SPSE at 400 cSt
and some field experience.
Contents
1. Introduction .................................................................................................................... 3
2. Critical factors ................................................................................................................ 4
2.1. Acoustic attenuation of the ultrasonic signals............................................................ 4
2.2. Cross talk.................................................................................................................... 5
2.3. Low Reynolds numbers.............................................................................................. 5
2.3.1. Laminar range ........................................................................................................ 5
2.3.2. Transition range...................................................................................................... 5
2.3.3. Turbulent range ...................................................................................................... 6
2.4. Impact of temperature on viscosity ............................................................................ 6
2.5. Calibration of flow meters for High Viscosity products ............................................ 7
2.6. Long term stability ..................................................................................................... 8
3. New developments ......................................................................................................... 9
3.1. Transducer design ...................................................................................................... 9
3.2. Test Results plus certification .................................................................................. 10
4. CFD simulations on thermal effects and boundary layer disturbance test ................... 11
5. Customer experience .................................................................................................... 17
5.1. Application in Norway Snorre-Vigdis [7]................................................................ 17
5.2. Application in Brazil on different FPSO’s [8] ......................................................... 18
6. Conclusions .................................................................................................................. 19
7. References .................................................................................................................... 19
1. Introduction Ultrasonic flow meters for custody transfer application were introduced in the industry in
1996. Supported by a significant number of national and international approvals, ultrasonic
measurement techniques have been adopted by oil and gas industries and frequently used for
custody transfer measurement of hydrocarbon products worldwide. After the introduction of
the API standard 5.8 “Measurement of liquid hydrocarbons by ultrasonic flow meters using
transit time technology” in February 2005 the confidence of industries rose and resulted in
higher acceptance of this technology for custody transfer crude oil applications.
During the introduction of the first ultrasonic flow meters approved for custody transfer
applications, manufacturers focussed on the generic applications where most of the
applications were liquids with viscosities up to 140 cSt. Analysing the present crude oil
exploration and production developments it is evident that highly viscous crude oils are being
increasingly produced and make up a significant part of global crude oil production.
Definitions for the different types of crude do vary by the different institutes; but, the
following descriptions are defined by the U.S. Geological Survey in an article dated August
2003 [1].
Light oil also called conventional oil with an API gravity of min. 22° and a viscosity < 100
cP.
Heavy oil is an asphaltine, dense (low API gravity), and viscous oil that is chemically
characterized by its content of asphaltenes (very large molecules incorporating most of the
sulfur and perhaps 90 percent of the metals in the oil). Although variously defined, the upper
limit for heavy oil has been set at 22° API gravity and a viscosity of 100 cP.
Extra-heavy oil is that portion of heavy oil having an API gravity of less than 10° and a
viscosity of above 100cP.
Natural bitumen, also called tar sands or oil sands, shares the attributes of heavy oil but is
yet more dense and viscous. Natural bitumen is oil with a viscosity greater than 10.000 cP.
An estimation of the known global reserves was presented in an article from “Highlighting
Heavy Oil” in the summer of 2006 as:
conventional oil
oil & sand bitumen
extra heavy oil
heavy oil
Global known reserves in percent
conventional oil
heavy oil
extra heavy oil
oil sand & bitumen
The information above implies that the demands of industries for the custody transfer
measurement of crude oils changed and manufacturers are requested to develop products that
will fulfil the industry demands for increasing the measuring capabilities for highly viscous
crude oils.
2. Critical factors
The critical factors with ultrasonic flow measurement techniques using transit time on highly
viscous products are [2]:
• Acoustic attenuation of the ultrasonic signals
• Cross talk
• Effect of low Reynolds numbers , i.e. laminar profiles
• Effect of temperature deviations on the liquid viscosity
2.1. Acoustic attenuation of the ultrasonic signals
The receiving signal will attenuate by means of:
1. Acoustic attenuation
2. 1/r - law
Acoustic attenuation
Acoustic waves generate micro movements in the fluid. These micro movements are
attenuated by molecular friction which is directly related to viscosity.
Due to attenuation in the medium the amplitude of the acoustic pressure (P) decreases
exponentially with the distance (L): L
o ePP ⋅−⋅= α (1)
in which α is the attenuation coefficient calculated as follows:
+⋅⋅⋅⋅
= bulkdync
ηηρω
α3
4
2 3
2
(2)
The attenuation is a function of frequency (ω), density (ρ), dynamic viscosity (ηdyn), speed of
sound (c) and the bulk viscosity (ηbulk). When we introduce the bulk factor Kv, we obtain:
vdyn Kc
⋅⋅⋅⋅
= ηρω
α3
2
2 in which
+=
dyn
bulkvK
ηη
3
4 (3)
1/r - law
Due to the fact that the acoustic energy radiated by the transducer is radiated in ‘all
directions’, the amplitude of the acoustic pressure decreases inversely proportionally with
distance (L):
oPL
P ⋅=1
(4)
This is called the 1/r - law.
Due to high acoustic attenuation of viscous crudes the measurement of these crudes with
acceptable custody transfer uncertainty is complicated. The damping of the acoustic signal
received will result in a reduced ratio between the emitted and received signals, and can in
some applications result in complete loss of signal. The damping of the acoustic signal is
directly related to the flow meter diameter and will increase for big sizes.
2.2. Cross talk
Cross talk is defined as the ultrasonic acoustic signal transported via the pipeline wall. The
more viscous the medium is, the stronger will be the cross talk signal compared to the
received signal. This will have a significant effect on the overall uncertainty.
The time measurement is based on the summation of the received signal and the cross talk
signal, for viscous crudes the cross talk signal will have a higher contribution than the
received signal and as a result an error will be introduced into the time measurement.
2.3. Low Reynolds numbers
Transit time ultrasonic technology is based on the measurement of the liquid velocity on
different positions in the pipe. The average velocity is proportional to the volume throughput.
Different flow profiles can influence the uncertainty of ultrasonic flow meters. The
identification of flow profiles is based on the dimensionless Reynolds number. API Ch 1
defines Reynolds as follows:
Where: D = inside diameter of the pipe
V = mean flow velocity
ρ = fluid density
η = fluid dynamic viscosity
2.3.1. Laminar range
For Reynolds numbers < 1000 the flow profile is laminar and
has a stable parabolic shape. Here, the flow velocity in the
middle is twice as high as the average flow velocity.
2.3.2. Transition range
Above Reynolds 1,000 the laminar flow becomes unstable.
Cross talk
ηυρ D
Re
××=
Turbulent plugs start to arise. These turbulent plugs are carried along with the flow and start
to grow in size. This causes an intermittent laminar-turbulent flow which is very unsteady.
This region of intermittent flow is called the transition area and runs up to a Reynolds number
of about 5,000. Above this Reynolds number the flow is fully turbulent.
Where and when these turbulent plugs occur and their frequency of occurrence is completely
stochastic and among others dependent on external factors such as local sharp edges, local
temperature differences, pipe vibrations, etc.; in other words, it is also installation dependent.
In the transition range, the shape of the flow profile is unpredictable and this results in an
increased uncertainty in the measurement.
.
2.3.3. Turbulent range
In turbulent flow, unsteady vortices appear on different axes and interact with each other
causing an exchange of energy in the radial direction; in other words the high and low
velocities average out. Due to this effect the flow velocity profile will become much flatter
than for a laminar flow. The flow range above Reynolds 5.000 is called the turbulent region.
2.4. Impact of temperature on viscosity
Because highly viscous crude oils are transferred at high temperatures, temperature
fluctuations are commonplace. A direct consequence of these temperature variations is
viscosity change, specifically for highly viscous crude oils, and as a result the Reynolds
number i.e. flow profile will change gradually.
Another effect that easily occurs is a temperature profile over the pipe cross section. This
causes a gradually varying viscosity with the pipe radius leading to an undefined velocity
profile. This effect shall be addressed in chapter 4.
Laminar profile
Turbulent profile
2.5. Calibration of flow meters for High Viscosity products
Another challenge is the calibration of highly viscous products. Thus far, no (laboratory) test
facilities in the world exist where large size flow meters could be tested or calibrated with
highly viscous products against a reliable reference. This necessitated the development of an
alternative calibration method.
A solution has been found in calibrating ultrasonic flow meters using the so-called Reynolds
calibration method.[3],[4].
First of all, it is essential to demonstrate that the flow meter linearity is a function of Reynolds
only. This is clearly seen in Figure 1. This figure shows that there is a very clear correlation
and overlap between the individual linearity curves for different products.
Figure 1 Linearity curve for different products with different viscosities. A clear correlation
and overlap between the individual linearity curves for different products is observed.
Because ultrasonic flow meters are velocity measurement devices, and the performance
depends on Reynolds, simulation techniques can be use that to simulate a highly viscous
product using an alternative calibration medium by manipulating the flow rate such to achieve
the requested Reynolds range.
In the table below an sample Reynolds calculation is shown for the following process
conditions:
Pipeline size 24 inch
Product # 1 viscosity 150 cSt
Product # 2 viscosity 400 cSt
Product # 3 viscosity 600 cSt
Product 0.5 m/s 1 m/s 2 m/s 3 m/s
#1 150 cSt 2000 4000 8000 12000
#2 400 cSt 750 1500 3000 4500
#3 600 cSt 500 1000 2000 3000
It is evident that Reynolds numbers for highly viscous crude oils are significant lower than
calculated for the more familiar products. However based on above table it is clear that, when
above mentioned flow meter i.e. 24”, are used on High Viscosity oil of 600 cSt at 3 m/s the
same Reynolds will be achieved as with a product of 150 cSt @ 0.75 m/s.
Based on above it is possible to calibrate ultrasonic flow meter for High Viscosity application
by using a lower viscosity product and by adjusting the flow rate such that the required
Reynolds is established.
Figure 2 This graph shows a 24” flow meter calibrated on three products, Naphtha, Oural
c.o. and water. It is clearly shown that the calibration with water, although a complete
different product, shows the same errors as for naphtha and oural c.o. at the same Reynolds.
2.6. Long term stability
The long term stability of flow meters needs to be checked at pre-defined intervals by
performing regular verifications. However based on the issues described earlier these actions
are labour-intensive demanding and will result in down-time of the installation. Ultrasonic
flow meters have proven to be extremely stable over time.
The characteristics which contribute to the long term stability of the ultrasonic flow meter
include no moving parts and thus no wear and tear. The condition of the measurement section
inside the meter body is not deteriorated by the medium and there is no shift of the k-factor
due to changes in process properties. In this respect ultrasonic flow meters differ from, for
example, mechanical flow meters where the internal parts are affected by process conditions
and therefore there are shifts in the k-factor shifts and regular re-calibrations are required for a
stable performance.
The long term stability or reproducibility of ultrasonic flow meters has been monitored
several times [4]. An example of this is shown in Figure 3. The flow meter was calibrated in
1999 and verified in 2009. No significant deviation in performance was observed. In between
no maintenance have been carried out and all original settings were used.
Linearity comparison 1999/2009 - 20" Altosonic V
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0 500 1000 1500 2000 2500 3000 3500
Q [m3/h]
Err
[%
]
Oural 17-3-1999 Initial calibration Body temp corr. ON
Oural 29-1-2009 Client w itness2 Body temp corr. ON
OIML - R117
Figure 3 Long term stability data of a 20inch ALTOSONIC V. This instrument has been
initially calibrated in 1999 and verified in 2009. No maintenance has been carried out in
between. The original setting have been used during verification in 2009.
3. New developments
3.1. Transducer design
To reduce the influence of cross talk, there has been specific R&D focussed on reducing this
effect. Because cross talk is a constant parameter that depends on the size and material of the
flow meter tube, two approaches have been investigated:
1. Increasing the strength of the ultrasonic signal to reduce the ratio between cross talk
and the received acoustic signal
2. Acoustically decoupled transducers.
The strength of the acoustic signal can be increased by decreasing the generated frequency:
The lower the frequency the higher the strength of the received signal. Because ultrasonic
transit time technology is based on accurate time measurement a lower frequency will
automatically result in lower performance. For larger diameters > 24” this effect will be less
significant because the delta time compared to smaller diameters is higher, and errors in delta
time measurement are negligible.
The most significant reduction in the effect of cross talk can be expected from acoustically
decoupled transducers. By isolating the transducers from the flow meter tube and reducing
direct metallic contact, the cross talk effect is reduced to a minimum. Tests were performed
on a flow meter tube using crude oil with a viscosity of 1500 cSt and a density of 950 kg/m³
(see Figure 4). A welded and decoupled transducer were compared simultaneously both at a
frequency of 1 MHz.
Figure 4 Test carried out during the development of the high viscosity transducer.
R&D test results prove that the signal to cross talk ratio is significantly lower for the
decoupled transducers. For the welded transducer the ratio was maximum 1% and for the
decoupled 0.18%, which is an improvement of a factor of 5.
3.2. Test Results plus certification
In January 2006 a number of ultrasonic flow meters were calibrated on highly viscous heavy
fuel oil. The calibration was performed using a unidirectional ball prover with a base volume
of 15 m³, and witnessed by the Dutch authorities (NMi). The results (errors and repeatability)
of one of these flow meters are shown in the graphs below. The horizontal line shows the
requirements as stated in the OIML R-117 recommendation. The error band shown of +/-
0.2% is specified in OIML R-117 Class 3.0. The associated maximum repeatability is 2 x
0.06% = 0.12%.
It is clear that the calibrated flow meter is well within the specifications. The flow rates used
during the calibration are in the range of 250 to 2600 m³/hr, which is equivalent to a Reynolds
range of 400 to 5400.
Figure 5 Test results on a 24” ALTOSONIC V running at viscosities of about 400 cSt.
Based on above calibrations the Dutch metrological institute (NMi) issued an OIML approval
in which the maximum viscosity limit was given as 400 cSt for diameters up to DN 600 (24”).
4. CFD simulations on thermal effects and boundary layer
disturbance test The viscosity of highly viscous oil is strongly dependent on temperature. The higher the
viscosity, the stronger the dependency. This is clearly illustrated in Figure 6 where the
kinematic viscosity is shown as a function of temperature.
Figure 6 Relationship between viscosity and temperature of an extra-heavy-oil sample.
This temperature dependency has implications for practice.
Usually the oil temperature in the system is different from the environmental temperature.
Consequently, heat transfer is induced, leading to a thermal boundary layer. This thermal
boundary layer does affect the local viscosity, which in its turn does affect the flow profile.
This effect is more pronounced in applications with high viscosity.
In this chapter CFD calculations are described that have been carried out to try to get a feel
how strong these effects are and how the 5-path ultrasonic flowmeter responds [5].
Efforts are made to find a calculation example which is straightforward and which represents
reality as much as possible.
The following assumptions have been made:
• The flow is always laminar because of the high viscosity of the medium.
• The laminar flow at the inlet is fully developed and has an uniform temperature. The
assumption has been made that the pipeline system was subject to constant conditions for
a long run, leading to an uniform temperature and fully developed laminar flow.
• Directly after the inlet, the heat exchanging process starts to play a role.
• In order to obtain a rather well developed thermal boundary layer, a relatively long
distance of 50D has been chosen. Shorter distances shall lead to a flow situation which is
less affected. In that respect this calculation example can be considered as a worse case
situation.
• Furthermore, the temperature at the outside pipe wall has been prescribed. In practise the
outside pipe wall temperature almost equals the fluid temperature inside the pipe. This is
clearly illustrated in paper [6]. In this paper several situations have been simulated. An
extreme example is the situation where gas at a pressure of 60 bar, a temperature of 37.7
°C, a flow speed of 10 m/s in an uninsulated 24 inch pipe was subject to a 5 m/s cold
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 10 20 30 40 50
Temperature [degC]
Vis
cosity [cSt]
400 cSt @ 20ºC
1850 cSt @ 0ºC
wind at -10°C. The resulting outside wall temperature is only 1 to 2 °C cooler than the gas!
The (insulating) thermal boundary layer of the air outside the pipe has been neglected in
the calculation example presented in this paper (see Figure 7). This leads to much higher
heat exchange rate, and consequently a stronger thermal effect, than it would be the case
in practice. This could be considered as a worst case condition as well.
Figure 7 Left hand figure: temperature profile as been used in this CFD simulation. Right
hand figure: temperature profile as it is in practice.
The boundary conditions and geometrical configuration for the CFD simulation have been
defined in Figure 8 and Table 1.
Figure 8 Sketch of the geometrical configuration and some boundary conditions which are
used for the CFD simulation.
Table 1 Numerical values that have been used for the CFD simulation.
air
oil
Pipe wall
air
oil
Pipe wall
Tamb.=Twall out.
Tamb.
Calculated
Situation
Situation in
practice
A rotational symmetric problem is investigated. All thermodynamic properties (density, heat
capacity, thermal conductivity) except from viscosity are assumed to be constant. Buoyancy
effects are neglected in this calculation. The relationship between fluid stress S and strain σ is described by the Newtonian model σ=ηS where η is the dynamic viscosity. All non-Newtonian effects are not accounted for and the fluid dynamic properties are described by the
single viscosity parameter ν in dependence of temperature as given by Figure 9. Viscous heating effects are included in the calculations. This leads to some self-heating, especially for
high Reynolds numbers.
0
500
1000
1500
2000
2500
10 15 20 25 30 35 40 45 50 55 60
Temperature [degC]
Kin
em
atic v
iscosity [cSt]
Figure 9 The relation between kinematic viscosity [ m
2/s] and temperature [°C] which has
been used for the numerical simulation described in this paper.
ANSYS CFX 12 is used to solve the incompressible Navies-Stokes equations on a V-shaped
slice of a pipeline section. The grid has 80 cells in the radial direction for the case Re=1500.
For the other Reynolds number cases, a grid with 40 cells in the radial direction is used. At
the inlet of the geometry, a parabolic velocity profile is prescribed. Convergence of the
calculation typically requires about 10,000 iterations.
In total 15 test cases have been calculated. They are named by the Reynolds number and
ambient temperature according to Table 2.
Table 2 Coding scheme for the calculated test cases. Re1: Re=100, Re2: Re=500, Re3:
Re=1500, t1: tamb.=15°C, etc. The initial oil temperature is 35 °C.
500 cSt @ 35ºC
2250 cSt @ 15ºC
The velocity profiles at two positions have been studied being the meter inlet and the
measuring section of the flowmeter. These positions are illustrated in Figure 10.
Figure 10 Locations where the velocity profiles have been investigated.
The boundary layer profiles are shown in Figure 11 for the meter inlet position and in Figure
12 for the measuring section position. In all investigated cases, the thermal boundary layer is
found in the region r/R>0.85.
Figure 11 The velocity profile (left hand figure) and thermal boundary layer at the Meter inlet
section (z/D=49) for the 15 different test cases.
Figure 12 The velocity profile (left hand figure) and thermal boundary layer at the measuring
section of the flowmeter for the 15 different test cases.
The curves that corresponds to different ambient temperatures have not been labelled since
they can be identified by the steepness at the wall. The steepest curve belongs to the situation
with the highest wall temperature. The thermal effects on the velocity profiles are less
pronounced for higher Reynolds numbers because the thermal boundary layers are thinner.
It is remarkable to see from Figure 11 (left hand figure) that there seems to be one point
where all velocities come together (around r/R1≈0.7) regardless the changing viscosity in the thermal boundary layer.
Inside the conical section, the fluid is accelerated. This leads to thinning of the velocity
boundary layers. As expected, it is found that the higher the Reynolds numbers, the thinner
the boundary layers. The flow in the central region becomes more homogeneous. This could
also be observed from Figure 13 which depicts the temperature distribution in the meter for
two different Reynolds numbers and two different ambient temperatures.
Figure 13 Temperature distribution [K] in the measuring section (Re1: Re=100, Re3:
Re=1500, T1: Tamb.=15°C, T5:Tamb.=55 °C). Note that the figures have strongly been
compressed in horizontal direction.
The CFD data have been used as input for the ALTOSONIC V measuring algorithm. This
enables us to simulate the behaviour of the ALTOSONIC V for these kind of profile
disturbances.
The profile factors that have been obtained correspond to the profile factors that are observed
in practice. The simulated ‘reading’ of the ALTOSONIC V has been compared with the
volume flow which is precisely known. This enables us to express the sensitivity of the
ALTOSONIC V for these kind of disturbances in a percentual error (see Figure 14).
-0,40%
-0,30%
-0,20%
-0,10%
0,00%
0,10%
0,20%
0,30%
0,40%
0 500 1000 1500 2000
Reynolds number [-]
Err
or [%
]
dT=-20 degC
dT=-5 degC
dT=0 degC
dT=5 degC
dT=20 degC
Figure 14 The response of an ALTOSONIC V to changing boundary layer viscosity as result
of heat transfer to and from the environment. dT is the difference between fluid and
prescribed outside pipe wall temperature.
Several conclusions could be drawn from these simulations:
• As expected, thermal boundary layer development for these Reynolds numbers (100 < Re
< 1500) is a very slowly process. Even after 50D the thermal boundary layer is relatively
thin: about 10% of the pipe radius.
• The conical measuring section has a positive effect on both the thermal and velocity
boundary layer. The thicknesses are reduced, the profiles are more homogeneous.
• Temperature differences between outer pipe wall and fluid up to 5 °C doesn’t lead to a significant measuring effect. This holds for the entire laminar flow regime that have been
simulated.
• For temperature differences equals 20 °C, the effects become significant. However, it should be emphasized that this is an extreme situation simulating e.g. falling snow on a
warm non-insulated pipe.
• When thermal insulation is being applied, the effect of ambient temperature is expected
to be not significant at all.
• The results shown in this paper may be applied to even higher viscosities, taking into
account reduced temperature differences such that the percentual viscosity changes stay
within the limits as presented in this paper.
• Buoyancy effects have been neglected in this paper. We like to take these effects into
account in the next simulations. Buoyancy could lead to non-symmetric flow situations.
• These results confirm the experience that we have with comparable situations in the field.
We don’t observe significant meter reading changes due to changing ambient conditions.
• These results also correspond to the experimental results described in the next paragraph:
tests with protruding gaskets disturbing the boundary layer.
Tests with protruding gasket
In addition to the above mentioned numerical investigations, another boundary layer
disturbance test have been carried out. By using a protruding gasket, the boundary layer just
in front of the flowmeter has been disturbed (see Figure 15). The response of the
ALTOSONIC V to this disturbance type has been tested.
Two 12 inch flowmeters have been used for this test. The viscosity of the oil during this test
was about 270 cSt. The gasket was protruding about 5 mm.
Figure 15 Schematic overview of a test with protruding gasket. The gasket protrudes about
5mm. The viscosity of the oil is about 270 cSt.
The performance of the flowmeters have been tested with and without this protruding gasket
using unchanged settings. The result of this test is shown in Figure 16.
HV ALTOSONIC V, 12inch, #1
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0 1000 2000 3000 4000 5000 6000 7000
Reynolds number [-]
Err
or
[%]
normal
protuding gasket
HV ALTOSONIC V, 12inch, #2
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0 2000 4000 6000 8000
Reynolds number [-]
Err
or [%
]
with protuding gasket
normal
Figure 16 Response of 2 x 12inch ALTOSONIC V flowmeters to protruding gaskets at high
viscosity applications.
No systematic effect could be observed. This holds for the laminar, transitional and turbulent
region. The outcome of this test support the conclusion from the previous paragraph that the
5-beam design using a conical measuring section is not really sensitive to small disturbances
in the boundary layer. This holds for high viscosity as well as for low viscosity applications.
5. Customer experience
5.1. Application in Norway Snorre-Vigdis [7]
Two identical ALTOSONIC V’s 8” (DN 200) have been in use since 1999 for the fiscal oil
transfer between the Snorre and Vigdis process trains on the Snorre tension-leg platform since
October 1997. One flow meter functions as the duty (transfer) meter and the other as the
master meter. For calibrations the flow meters can be run in line. The customer required a
5 mm
Protruding
gasket
12” ALTOSONIC V
Flow
flow meter that complied with NPD requirements, had a long-term repeatability, and required
a minimum of maintenance.
Conclusions on this application:
• The k-factors of the duty ultrasonic flow meter, determined from comparison with the master ultrasonic meter after initial set-up adjustment, were all within +/- 0.10%. In
similar applications, such good results could never be achieved with turbine meters even
with frequent washing and cleaning. The average bias in k-factor from the pre-established
curve in 1997, with all the registered data points, was within 0.02%.
• In addition, over for ten years no maintenance was carried out on the KROHNE ALTOSONIC V meters. The diagnostic tools built into the ALTOSONIC V have proven
to be reliable, and a strong guide to whether the processing and piping installation is
satisfactory.
5.2. Application in Brazil on different FPSO’s [8]
Quote
”All the petroleum production in the Marlin Asset, Campos Basin, after treatment, is stored in
the production platform tanks for some days, until is offloaded. During this period, the
residual water is partially stratified, resulting in petroleum layers near the bottom of the tank
with higher water content, more than what is allowed. In order to meet the specifications of
the National Agency for Petroleum, Natural Gas and Biofuels (ANP), refineries or even
international export market requirements, the high water content petroleum is removed from
the treated oil tanks, being directed to another treatment process, and, therefore, being mis-
accounted for the second time in the petroleum fiscal metering before tank storage.
In order to overcome this problem all the fiscal metering of the petroleum production in the
Marlin Asset, Campos Basin is made in the offloading lines of the FPSO’s.”
Unquote
For this ultrasonic flow meters were installed and metrological certification of the flow meters
required tests with high viscosity in an international laboratory and analysed by the Dutch
Board for Weights and Measures (NMi).
The long term stability of ultrasonic flow meters was monitored resulting in the following
data and is based on re calibrations at a certified calibration facility:
Meter size No drift observed over a
period of
6 inch 23 months
6 inch 9 months
6 inch 20 months
12 inch 8 months
24 inch 46 months
6. Conclusions
• Crude oil production will in the near future increasingly use high-viscosity oils and bitumen and therefore the industry is demanding ultrasonic flow meters capable of
measuring such crude oils.
• The critical factors for using ultrasonic flow meters on High Viscosity products, as specified in this paper required extensive Research and Development efforts. These
critical factors have been solved and as a result a NMi approval has been obtained for
viscosities up to 400 cSt for diameters up to DN 24”
• Presently Ultrasonic flow meters can handle 500 cSt for diameters up to DN 24”, higher viscosities are feasible for smaller diameters.
• Calibration of High Viscosity ultrasonic flow meters based on Reynolds, by using lower viscosity products is well established and proven to provide excellent results
• The reproducibility and long term stability has been proven by internal tests and evaluations performed by KROHNE, many end-users, and independent authorities which
resulted in NMi confirmation that the performance of the KROHNE ultrasonic flow meter
only needs verification in intervals of 5 years.
• Thermal boundary layers does affect the viscosity in the boundary layer significantly. This leads to a changing laminar flow profile. However, temperature differences between outer
pipe wall and fluid up to 5 °C doesn’t lead to a significant measuring effect. This holds for the entire laminar flow regime that have been simulated.
• For temperature differences equals 20 °C, the effects become significant. However, this is a very extreme situation simulating e.g. falling snow on a warm non-insulated pipe.
• When thermal insulation is being applied, the effect of ambient temperature is expected to be not significant.
• The conical measuring section has a positive effect on both the thermal and velocity boundary layer. The thicknesses are reduced, the profiles are more homogeneous.
• Buoyancy effects have been neglected in this paper. We like to take these effects into account in the next simulations. Buoyancy could lead to non-symmetric flow situations.
• These results confirm the experience that we have with comparable situations in the field. We don’t observe significant meter reading changes due to changing ambient conditions.
• These results also correspond to the experimental results described in the next paragraph: tests with protruding gaskets.
7. References
[1] U.S. Geological Survey in an article dated August 2003.
[2] Jankees Hogendoorn et. al.,“High viscosity hydrocarbon flow measurement: A
challenge for Ultrasonic Flow Meters?” 8th South East Asian Flow Measurement
Workshop, March 2009.
[3] Jankees Hogendoorn, et. al.,“An Ultrasonic Flowmeter for Custody Transfer
Measurement of LNG: A challenge for Design and Calibration”, 25th International
North Sea Flow Measurement Workshop, October 2007.
[4] Jankees Hogendoorn, “Flow Measurement in Nuclear Power Plants”, Tokyo,
KROHNE, May 2008.
[5] Karsten Tawackolian,“Flow profile disturbance caused by viscosity effects in non-
isothermal media”, Internal communication, PTB Berlin, September 2009.
[6] Sarah Kimpton and Ali Niazi, “Thermal Lagging – The Impact on Temperature
Measurement, 26th International North Sea Flow Measurement Workshop, October
2008.
[7] Maron Dalström,“KROHNE ALTOSONICV with Master Meter Approach”, Paper 18
present at the NSFMW 2003, Statoil ASA Norway, 2003.
[8] Mr. Josapaht Dias da Mata (Petrobras), et. al.,“Petroleum Measurement And
Diagnostics Of Ultrasonic Flow Meters In High Flowrates And Viscosities”, Paper 8.2
presented at the American Workshop 2008.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
1
A MULTIPATH ULTRASONIC METER WITH REDUCING NOZZLE FOR IMPROVED PERFORMANCE IN THE LAMINAR/TURBULENT
TRANSITION REGION
Gregor J Brown, Cameron, UK Terry Cousins, Cameron, USA
Donald R Augenstein, Cameron, USA Herbert Estrada, Cameron, USA
1 INTRODUCTION With rising worldwide energy demands and depletion of existing conventional oil reserves the production of heavy oil is becoming increasingly common. The high viscosity of heavy oils presents measurement challenges for most types of flow meter. For example it limits the maximum flow of PD meters, reduces the turndown of turbine meters and can result in measurement errors in Coriolis meters. Ultrasonic meters can be used for measurement of high viscosity oils. However, in order to do so with high accuracy they have to cope with increased signal attenuation and changing velocity profiles through the transition from turbulent to laminar flow. This paper explains the technical challenges faced when using ultrasonic meters for high viscosity/low Reynolds number flows and shows how these conditions can adversely affect the performance of some designs of ultrasonic meter. Modelling using velocity profile data and analysis of meter diagnostic data is presented in order to illustrate the physical processes that are at work. Test data is presented to demonstrate the performance of conventional and improved ultrasonic meter designs. The improved ultrasonic meter design incorporates a reducing nozzle to flatten and stabilise the velocity profile in the transition region. The impact of this design feature on permanent pressure loss is also evaluated. 2 BACKGROUND In 2001 US Department of Energy figures classified only 30% of the estimated 9 -13 trillion barrels of total world oil reserves as conventional oil, with heavy oil and extra heavy oil/bitumen accounting for 15% and 55% respectively. Production of heavy oils is more technically demanding than conventional production, requiring new and improved recovery methods and increased crude sales values to ensure its economic viability. The depletion of existing conventional reserves, rising worldwide energy demand, and new technology developments are therefore leading to increased activity in heavy oil production. In the terminology of the American Petroleum Institute (API), heavy oil is classified as having API gravity of less than 22.3 (equivalent to a density of greater than around 920 kg/m3 at 60 ºF). In terms of flow measurement, the relatively high density of the liquid is not problematic in itself. Consider for example that water has a density of around 1000 kg/m3 and can be measured relatively easily by a wide variety of techniques. The main challenge
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
2
comes from the fact that the higher the density of a crude oil is, the more viscous it is likely to be. This is illustrated in the viscosity versus temperature data shown in Figure 1 below for crude oils with densities between 830 and 930 kg/m3 at 20 ºC. Another important property of viscous oils is evident in Figure 1; the rate of change in viscosity with temperature increases with increasing density/viscosity, e.g. for a reduction in temperature from 30 to 20 ºC, the increase in viscosity for the three oils in Figure 1 is as follows:
832 kg/m3 oil, viscosity changes from 4.6 to 5.6 cSt, an increase by a factor of 1.22 890 kg/m3 oil, viscosity changes from 17.2 to 25.8 cSt, an increase by a factor of 1.5 930 kg/m3 oil, viscosity changes from 124 to 240 cSt, an increase by a factor of 1.9
Figure 1 Viscosity versus temperature for crude oils of different densities
In addition to having high viscosity, heavy oils also tend to have higher than normal levels of contaminants such as wax, asphaltenes, sand, water, heavy metals and sulphur. Combined with potentially high process temperatures, these issues combine to produce a challenging environment for high accuracy flow measurement. 3 METERING OF VISCOUS OILS USING MECHANICAL AND MASS FLOW
METERS Some of the problems associated with measurement of viscous oils using conventional mechanical meters have been known for a long time. In the case of turbine meters, the increased drag associated with higher viscosity oils is known to make the meters more non-linear and restrict their operating range. Design modifications such as use of helical blades can reduce the sensitivity of turbine meters to viscosity changes, but even these modifications have a limited effect in terms of the increase in range of viscosity that turbine meters can easily handle. When it is taken into consideration that they also have moving parts and that for custody transfer they normally require regular in-situ proving, it is clear that turbine meters are not ideally suited for high accuracy measurement of highly viscous oils.
1
10
100
1000
0 10 20 30 40 50 60 70
Temperature (deg C)
Visc
osity
(cSt
)
930.29
890.35
832.14
DENSITY AT 15 deg C
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
3
The relatively limited applicability of turbine meters in terms of viscosity range is illustrated in Figure 2 below, from the API Manual of Petroleum Measurement Standards [1]. The superimposed red line in Figure 2 shows the max viscosity limits taken from the FMC Smith Meter MV Series helical turbine meter data sheet [13].
Figure 2 API selection guide for displacement and turbine meters Positive displacement (PD) meters are recognised as being more robust and much less sensitive to viscosity changes. However, even these meters have limitations in high viscosity applications. They are also relatively expensive and have a lower flowrate capacity than turbine or ultrasonic meters of the same nominal pipe diameter, as illustrated in Figure 3 below. This means an application that could be served by one turbine or ultrasonic meter might require two or more PD meters in parallel.
Figure 3 Flowrate capacity of PD, turbine and ultrasonic flowmeters
0
2,000
4,000
6,000
8,000
10,000
12,000
2 4 6 8 10 12 14 16 18 20 22 24
Max
flow
rate
(m3/
hr)
Pipe diameter (inches)
Double case rotary vane PD meter
Helical bladed turbine meter
Caldon LEFM 2xxCi ultrasonic flowmeter
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
4
The response of PD meters to changes in viscosity is dependent on the amount of ‘slippage’ that occurs; that is the amount of fluid that leaks past the clearances in the internals of the meter. It is generally recognised that as the viscosity increases the amount of slippage reduces; the meter becomes more linear and the meter factor tends towards a constant value. However, when the viscosity gets very high it creates another problem, in that the increased shearing stresses may make it necessary to use larger clearances (thus reintroducing slippage) or reduce the maximum flowrate capacity of the meter. For example, a 16-inch PD meter might require increased clearances or reduced capacity for operation at viscosities above 200 cP. Increasing the clearances means that the same meter would be less accurate when measuring lower viscosity products of say 10 cSt. Coriolis mass flow meters are now used regularly for custody transfer of crude oil. They are also used extensively in process applications where extremely high viscosities can be encountered, for example in the measurement of food products. Early experimental investigations with relatively low viscosity hydrocarbon products concluded that there was no obvious influence of liquid viscosity on Coriolis meters [2]. However, a recent study carried out by NEL has shown that Coriolis meter performance is affected by changing fluid viscosity. The NEL tests were carried out using two commercial Coriolis meters and showed errors in excess of 0.4 to 1% when the viscosity was increased to around 200 cSt [3]. 4 METERING OF VISCOUS OILS USING ULTRASONIC FLOW METERS Ultrasonic meters are now used regularly for custody transfer measurement of crude and refined oils. Significant potential benefits of ultrasonic meters are related to the fact that they have no moving parts and can be both non-intrusive and non-invasive. It may appear at first glance that ultrasonic meters are ideally suited to measurement of viscous oils. However, when the principles of operation and the characteristics of viscous flows are examined in more detail it becomes apparent that measurement of viscous oils presents two issues that must be addressed:
The response of the meter to extreme changes in fluid velocity profile Attenuation of the ultrasonic signals transmitted through the fluid
4.1 Hydraulic Effects in Low Reynolds Number Flows Since the early studies of Osborne Reynolds in 1883 it has been known that the character of pipe flow changes from laminar conditions to turbulent conditions at Reynolds numbers greater than ~2,000 [4]. Laminar flow is characterised by the fluid moving in a direction parallel to the pipe axis and the velocity profile being parabolic in shape. Turbulent flow is chaotic with the flow at any point in the pipe having a mean velocity with superimposed random velocity components in three dimensions. Turbulent flow has a flatter profile shape that varies as a function of Reynolds number. In the transition region between laminar and turbulent flows the profile continuously switches back and forth between states similar to laminar and turbulent flows, being similar to
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
5
turbulent flow for a larger fraction of the time at higher Reynolds numbers in the transition region and being similar to laminar flow for a larger fraction of the time at lower Reynolds numbers in the transition region. As multipath ultrasonic meters measure velocity on discrete paths and then combine these to obtain an estimate the average axial velocity in the pipe it is well known that these meters can exhibit some sensitivity to changing profiles. The sensitivity of different path configurations to velocity profile changes in turbulent flows (including distorted flows) has been the subject of numerous studies, e.g. [5]. A number of papers have also been published that include some data on the response of ultrasonic meters to changing velocity profiles in the transition region e.g. [6, 7]. However, there has been very little systematic study in this area. The following subsections explore the issues of the influence of velocity profile in the transition region in more detail. 4.1.1 Reynolds number and viscosity limits for transitional flow Before reviewing the performance of ultrasonic meters in the transition region and the hydraulic features of transitional flow, it is useful to consider the Reynolds number range in which transitional flow occurs, and the restrictions that this might place on standard ultrasonic meters in terms of liquid viscosity. Numerous text books and papers suggest that transitional flow occurs in a range of Reynolds numbers between 2,000 and 5,000. Schlichting [8] states that numerous experiments show a lower bound for transitional flow of around 2,000 but that some researchers succeeded in maintaining laminar flows up to Reynolds numbers of 20,000 and 40,000 by minimising the disturbance to the flow entering the pipe. For industrial flow metering applications it would seem consistent that transition is most likely to occur in the lower region between 2,000 and 5,000 Re, as inlet conditions are generally subject to significant disturbance. However, a number of other factors also affect the transition to turbulence, including wall roughness, vibration and heat transfer. For application of the established range of full-bore Caldon ultrasonic meters, we at Cameron recommend a lower Reynolds number limit of 10,000. This is based on both field and laboratory experience with multipath ultrasonic flowmeters. Independent of the flow measurement performance of the device, Caldon meters are capable of providing diagnostic information that can be used to determine if the flowmeter is operating in laminar, transitional or turbulent flow. The diagnostic parameter most useful in this respect is termed the flatness ratio, FR, which is simply the sum of the outside path velocity measurements divided by the sum of the inside path velocity measurements, or with reference to Figure 4,
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
6
Figure 4 Schematic of a 4-path Caldon meter Analysis and experiment show that in the case of 4-chord Gaussian path arrangements for turbulent flow the flatness ratio is greater than 0.75 and for laminar flow is less than 0.45. Figure 5 below is a graph of experimental flatness ratio data showing the transition from laminar to turbulent flow as a function of Reynolds number. The flatness data was taken directly from the ultrasonic meters and the Reynolds number was calculated independently based on measurements of flowrate and viscosity performed by the test laboratories. One meter was a 6-inch Caldon meter tested at NEL in the UK and the other was a 12-inch Caldon meter tested at SPSE in France. It can be observed that in the case of the 6-inch meter the transition occurred at Reynolds numbers between 3,000 and 5,000 but in the case of the 12-inch meter the transition occurred between 6,000 and 9,000 Re. Although pipe diameter may be one influencing factor in this case, it does not seem likely that pipe diameter alone is responsible for the difference in the critical Reynolds number (as this would be contrary to the whole notion of Reynolds number similarity). What is clear however, is that the critical Reynolds number is subject to various influences, hence this is why we at Cameron apply a conservative lower limit of 10,000 Re for our standard Caldon ultrasonic meters. It can also be seen from the data in Figure 5 that the transition occurs over a relatively narrow range of Reynolds number (and hence flowrate), meaning that the flow profile change is quite abrupt.
Figure 5 Flatness ratio data showing the transition from laminar to turbulent flow
0.300
0.400
0.500
0.600
0.700
0.800
0.900
100 1,000 10,000 100,000
Reynolds Number
Flat
ness
Rat
io
12-inch meter, 100-140 cSt, 28-35 deg C
6-inch, 100-200 cSt, 20-30 deg C
V1
V2
V3
V4
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
7
We can examine the impact of imposing a low Reynolds number limit on the maximum viscosity for an ultrasonic flow meter very simply. Reynolds number is given by
where U is the mean velocity, D is diameter and is kinematic viscosity. Rearranging we can write
Figure 6 below shows the viscosity limits calculated for minimum Reynolds numbers of 5,000 and 10,000, assuming a minimum flow velocity of 1 m/s. It can be seen that the maximum viscosity in centistokes at 10,000 Re works out to approximately 2.5 times the pipe diameter in inches, and 5 times the pipe diameter for 5,000 Re. Clearly these Reynolds number based limits are very restrictive if we wish to measure heavy oils.
Figure 6 Viscosity limits for 1 m/s min. velocity as a function of diameter and Reynolds no.
4.1.2 Performance of conventional ultrasonic flowmeters in transitional flow There are a number of papers and test reports that provide examples of degraded ultrasonic meter performance in the transition region. In a similar fashion to the improvements that multipath designs offer in turbulent flows [5], we find that multipath meters with four or more paths perform better than single-path and two-path meters. For example, Figure 7 below shows data from a test carried out by the author while working at NEL in 1997 on a 6-inch two-path (Krohne) ultrasonic meter [7]. The viscosity of the oil was approximately 24 cSt. It can be observed that the meter shows average errors of more than 2% at a Reynolds number of 3,000 but less than 0.5% at Reynolds numbers both above and below this value.
0
20
40
60
80
100
120
140
0 4 8 12 16 20 24
Visc
osity
(cSt
)
Meter diameter (inches)
Reynolds numbers less than 5,000
Re between 5,000 and 10,000
Reynolds numbers greater than 10,000
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
8
Figure 7 Two-path meter performance through the transition region Figure 8 shows data from a 6-inch 4-path Caldon 240C full-bore flowmeter tested through the transition region at NEL in 2006. It shows that linear performance within +/- 0.15% can be achieved above and below the transition region, but that in transition the average errors can increase to around +/- 0.5%. Recent tests obtained at NEL using a multipath meter from another manufacturer show broadly similar results [3]. The NEL data was corrected offline by as a function of Reynolds number whereas Caldon meters perform any required linearization correction internally.
Figure 8 Four-path meter performance through the transition region It can be observed in Figure 8 that the data ‘scatter’ appears to be greater in the transition region between 2,000 and 5,000 Re. It has been observed in transition, over the same range as the rapid variation in the shape of the mean velocity profile, that the short term repeatability of a standard meter is poorer than normal. This is illustrated in Figure 9 below,
‐1.00
‐0.50
0.00
0.50
1.00
1.50
2.00
2.50
1000 10000 100000
Error (%)
Reynolds number
-1.0%
-0.8%
-0.6%
-0.4%
-0.2%
0.0%
0.2%
0.4%
0.6%
0.8%
1.0%
100 1,000 10,000 100,000Pipe Reynolds Number
Erro
r (%
)
220 cSt, 20 deg C
115 cSt, 30 deg C
55 cSt, 45 deg C
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
9
which plots the standard deviation of the path velocity measurements versus Reynolds number. This data is from the same four-path meter that was used to produce the data in Figure 8. It can be seen that the standard deviations peak in the transition region at a Reynolds number of around 3,000, with paths 1 and 4, which are closer to the pipe wall, being affected the most. Above the transition, in turbulent flow, the standard deviations decrease, and as should be expected they are lower still in the laminar region below 2,000 Re.
Figure 9 Path velocity standard deviations through the transition region
From the test data we can conclude that multipath designs can perform better than more simple designs in the transition region by virtue of better flow integration and the availability of additional information for correction purposes. However, it is clear that their performance is still degraded relative to how they behave in turbulent flow. Given that good performance can be achieved in turbulent flow conditions, and in laminar flow conditions, it remains to be explained exactly why performance is poor and the meter less repeatable in the transition region, where the flow profile is changing between its laminar and turbulent forms. 4.1.3 The behaviour of transitional flow Experimental studies of laminar to turbulent transition have been carried out by a number of researchers [e.g. ref 9] and have included detailed measurement of velocity profiles using hot-wire anemometry. These studies have been complimented more recently with computational simulations of transitional flows using large eddy simulation methods [10]. Both the experimental and computational studies reveal some of the important characteristics of transitional flow. Of particular relevance to our work with ultrasonic meters, these studies of transitional flow show that in the process of switching back and forth between laminar and turbulent forms, two things happen that result in adverse effects on the performance of ultrasonic meters:
0%
5%
10%
15%
20%
25%
30%
1,000 10,000
Path Velocity Standard Deviation
Pipe Reynolds number
Path 1
Path 2
Path 3
Path 4
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
10
1) In the process of changing from laminar to turbulent and back again the axial velocity profile takes on forms that are like neither turbulent nor laminar flow
2) During the process of rapidly changing from one axial profile form to the other, the fluid must move in toward or out from the axis of the pipe and this generates non-axial flow components
Figure 10 below reveals the structure of transitional flow. Figure 10(a) shows a schematic illustrating the profile changing from a laminar (parabolic) form to a flatter profile and back again, and the generation of non-axial flow components at the leading and trailing edges of the turbulent zone that accompany this process [9]. Figure 10(b) shows a particle tracing simulation illustrating the initial laminar condition, the leading edge of a turbulent ‘slug’ with large vortices near the wall, and the turbulent zone that follows [10]. It is this behaviour that is responsible for the increased variability of the velocity measurements in the transition region as illustrated in Figure 9.
(a)
(b)
Figure 10 The structure of transitional flow
Figures 11 and 12 show the detailed form of the axial velocity profile at various stages during the velocity profile switching from its laminar-like state to the turbulent zone and back again. The data is plotted as the velocity, u, normalised by the velocity in the centre of the pipe, uc, versus the normalised radial position, r/R. These figures represent the two different forms of turbulent feature that have been observed in the transition region, one termed a slug and the other a puff. The velocity profiles in each case are shown for five locations relative to the turbulent feature, x* = 0 being the velocity before the turbulent zone, x* = 2 being in the centre of the turbulent zone, and x* = 4 being downstream of the turbulent zone. Hence the profiles at x* = 1 and x* = 3 are located at points where the velocity profile is rapidly changing. Observe that although the two graphs are not identical they do bear a close similarity to one another.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
11
Figure 11 Average axial velocity profiles for a turbulent slug
Figure 12 Average axial velocity profiles for a turbulent puff
The availability of such detailed velocity profile data allows an assessment of the impact of these velocity profile shapes on the accuracy of average velocity measurements made by multipath ultrasonic meters. 4.1.4 Velocity profile analysis Previous analytical evaluations of the ability of the 4-chord Gaussian integration techniques used in Caldon meters show that path positions and weightings used in these meters are highly accurate for fully developed flow profiles in both turbulent and laminar conditions [e.g. ref 5]. These results are reproduced in Figure 13 below. However, it is obvious from Figures 11 and 12 above, that some of the velocity profiles observed in transitional flow are very different to either laminar or turbulent profiles, particularly those at x* = 1.
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
r/R
u/uc
Slug x* = 0
Slug x* = 1
Slug x* = 2
Slug x* = 3
Slug x* = 4
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.0 0.2 0.4 0.6 0.8 1.0
u/uc
r/R
Puff x* = 0
Puff x* = 1
Puff x* = 2
Puff x* = 3
Puff x* = 4
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
12
The velocity profiles in Figures 11 and 12 were modelled by fitting discrete data [11] with a function in the form:
′ 1 ′ 1 ′ 1 ′ ′ 1 ′
where r’ is the normalised radius r/R, and j, n, s, t, g and k are constants that change for each profile. This function can then be integrated over the whole cross-section to yield the true average velocity and along paths representing the ultrasonic flowmeter to yield the flowmeter’s estimate of the average velocity. The results of this exercise for a 4-path Gauss-Jacobi integration are shown in Figure 13 alongside the results for fully developed laminar and turbulent profiles as represented by parabolic and powerlaw equations. The data is plotted in the form of the velocity profile factor kh (which is the true average velocity divided by the measured velocity) versus the velocity profile flatness ratio. It can be seen that in both turbulent and laminar conditions the velocity profile factor is very close to 1; the errors are predicted to be less than +/- 0.12%. In the transition region however, the results for the instantaneous transitional velocity profiles span from 0.994 to 1.005, i.e. +/- 0.6%. Clearly this offers an explanation for the magnitude of errors shown in Figure 8. The data of Figure 13 suggests that a solution might be found by correlating the meter factor with the flatness ratio in a similar fashion to the methodology used by Caldon for very high Reynolds numbers [12]. However, practical experience has show that doing this alone is of limited use owing to factors that affect the reproducibility of flow profiles in the transition region (and hence the relationship between flatness ratio and meter factor).
Figure 13 Profile factor versus flatness ratio for various forms of velocity profile
0.992
0.994
0.996
0.998
1.000
1.002
1.004
1.006
1.008
1.010
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Pro
file
fact
or, k
h
Flatness ratio
Laminar - parabolic profileShan slug profilesShan puff profilesTurbulent - power law profiles
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
13
4.1.5 A solution to the problem of transitional flow The preceding sections of this paper demonstrate that the poorer performance of ultrasonic meter in the transition region is related to the forms of velocity profile that occur when the flow is switching back and forth between laminar and turbulent states. Avoiding operation in the transition region is difficult as often in the flow processes that we wish to measure there are severe constraints placed on the ability to adjust flowrate or viscosity. Experiments show that the bounds of transitional flow can be moved by altering flow conditions, but only to a limited extent, such that in practice any application range that crosses a Reynolds number range of 2,000 – 10,000 must experience transitional flow. It has now been clearly demonstrated that it is the unusual velocity distributions that occur in transitional flow that cause problems with ultrasonic meters in this region, and this suggests that a solution may be found by artificially altering the velocity profiles upstream of the point of measurement. Over recent years, Cameron has performed extensive experiments to evaluate the effectiveness of various forms of flow conditioning in transitional flows. Some forms of conventional flow conditioner that are normally intended to reshape the axial velocity profile and remove axial velocity components in turbulent flow can be shown to be of some use in limiting the effects of transitional flow. This is contrary to the implied conclusion of the NEL paper which stated that “the inclusion of an artificial flow conditioner provided no obvious improvement to the USM response under the elevated test viscosity conditions applied here” [3]. The apparent lack of improvement in the NEL experiments is a result of the type and location of flow conditioner, which was a tube bundle, presumably placed between 7 and 10 diameters upstream of the meter. This specific result does not disagree with our findings; tube bundles do relatively little to reshape the axial flow profile; but we found several other forms of flow conditioning to have quantifiable benefits in terms of limiting the effects of transitional flow. One of the methods of altering the flow profile that we have thoroughly tested, and since incorporated into the Caldon product line, is the use of a reducing nozzle immediately upstream of the ultrasonic measuring section. This is illustrated in Figure 14 below in the form of an integrated nozzle with 8-path metering section and downstream expansion cone. The diameter ratio, β (throat diameter over pipe diameter) for this meter design is ~0.63, and was selected as a practical compromise between meter performance and pressure loss. Values of beta greater than 0.64 are not used for this purpose, as the flattening effect on the profile becomes less effective with increasing beta. A photograph taken looking into the throat of a 6-inch 4-path meter with integrated reducing nozzle is shown in Figure 15.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
14
Figure 14 A schematic of an 8-path Caldon meter with reducing nozzle inlet and conical outlet
Figure 15 The a 6-inch test meter with reducing nozzle inlet The contour of the nozzle has been designed so as to avoid negative pressure gradients along the nozzle surface. Such gradients could induce boundary layer separation and hence turbulence. The reducing nozzle is beneficial in the transition region owing to the way it acts upon the fluid entering the measurement section. In effect what the nozzle does is to flatten and stabilise the velocity profile owing to the motion toward the centre of the pipe that is imposed on the fluid. Importantly this means that not only are non-axial flow components suppressed, but the shape of the axial velocity profile in both laminar and turbulent conditions become more alike. This is shown in Figure 16 below, where the normalised path velocities are shown for (a) a normal 6-inch full-bore meter and (b) a 6-inch meter with a reducing nozzle at Reynolds numbers of 1,000 and 7,000 (i.e. either side of the transition region).
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
15
(a) (b)
Figure 16 Velocity profiles for (a) a standard 4-path meter; and
(b) a 4-path meter including a reducing nozzle with a diameter ratio of 0.63 Not only does the reducing nozzle bring the laminar and turbulent profiles closer together, but it also causes the profile to change more gradually as Reynolds number is reduced. This is shown in Figure 17 below, where the profile flatness measured by the meter is plotted versus Reynolds number. It is clear that in the case of the meter with integrated reducing nozzle the transition is much more gradual when compared with the abrupt profile transition that is seen by the full-bore meter.
Figure 17 Flatness ratio versus Reynolds number for a conventional full-bore meter and a meter employing a reducing nozzle with a diameter ratio of 0.63
As suggested above, the ‘smoothing’ of the velocity profile accomplished by use of the reducing nozzle is accompanied by suppression of the very strong non-axial flow components normally found in transitional flows. The end result is that meter repeatability is also
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
‐1 ‐0.8 ‐0.6 ‐0.4 ‐0.2 0 0.2 0.4 0.6 0.8 1
Norm
alised
Path Velocity
Path Radial Location
Re = 1,000 Re = 7,000
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
‐1 ‐0.8 ‐0.6 ‐0.4 ‐0.2 0 0.2 0.4 0.6 0.8 1
Norm
alised Path Velocity
Path Radial Location
Re = 1,000 Re = 7,000
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
100 1,000 10,000
Reynolds Number
Flat
ness
Rat
io
6-inch, 100-200 cSt, 20-30 deg C
6-inch meter with reducing nozzle
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
16
improved in the transition region relative to a standard full-bore meter design. This is illustrated in Figure 18 below, which plots the standard deviation of the path velocity measurements versus Re for the 6-inch meter with reducing nozzle of β = 0.63. By comparison with Figure 10 it can be observed that the effect of the reducing nozzle is to reduce the standard deviations in the transition region dramatically; from a peak value of than 25% in the case of the full-bore meter to less than 6% when the reducing nozzle is employed. It can also be observed that the reducing nozzle has a beneficial effect in turbulent flow.
Figure 18 Path velocity standard deviations through the transition region for a meter employing a reducing nozzle of β = 0.63
Overall, the impact of the reducing nozzle on performance of the flow meter is to extend the performance that is normally achieved in turbulent flows right through the range of pipe Reynolds number for transitional flow. This is demonstrated in Figures 19 and 20, which show data for meters tested at NEL through the transition region. Figure 19 shows the performance of a 6-inch meter with reducing nozzle tested at a viscosity of ~270 cSt. This data straddles the range of Reynolds number where transition region effects would be seen in the case of a standard full-bore meter.
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
10%
1,000 10,000
Path Velocity
Stand
ard Deviatio
n
Pipe Reynolds number
Path 1
Path 2
Path 3
Path 4
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
17
Figure 19 Error vs Re for a 6-inch meter with reducing nozzle at a viscosity of ~270 cSt
Figure 20 shows data from a 12-inch meter with reducing nozzle tested at three different temperatures resulting in viscosities of 50, 80 and 120 cSt. This result is important as it demonstrates stability of the meter factor within +/- 0.15% with changing temperature and viscosity through the transition region. The reason to stress the importance of this result is that when other ultrasonic meter types are tested and transitional flow does have an influence on the meter factor, a single product calibration over a limited temperature range may be insufficient to show the effects. It is variation in process conditions in the transition region that result in the lack of reproducibility shown for the full-bore meter in Figure 8 – the effects of transition could be calibrated out for any one condition, but not for all three.
Figure 20 Error vs Re for a 12-inch meter with reducing nozzle
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
100 1000 10000
Pipe Reynolds number
Erro
r (%
)
-0.5%
-0.4%
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
0.4%
0.5%
1,000 10,000 100,000
Reynolds number
Ave
rage
Err
or (%
).
50 cSt, 40 deg C
80 cSt, 30 deg C
120 cSt, 20 deg C
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
18
Figure 21 shows the data from Figure 19 presented as a function of flowrate rather than Reynolds number. This demonstrates that the meter design with reducing nozzle can achieve +/- 0.15% over a range of 10:1, and +/- 0.2% over a range of 15:1. It should also be noted that although this meter design is well suited to low Reynolds number applications, it can also be used at much higher Reynolds numbers and has been tested successfully up to Reynolds numbers of 900,000.
Figure 21 Error vs flowrate for a 6-inch meter with reducing nozzle at ~270 cSt 4.2 VISCOSITY AND ATTENUATION OF ULTRASOUND Having established that the hydraulic problems in the transition region can be overcome, it is now required that we review the effects of attenuation in high viscosity applications. The amplitude A of an ultrasonic signal that has been transmitted through a fluid is given by the exponential function:
)exp(0 xAA Where A0 is the amplitude of the signal in the absence of attenuation, x is the distance travelled and α is the attenuation coefficient. For viscous absorption the attenuation coefficient is given by
23
2
38
fc
K
f
vis
where is the kinematic viscosity, cf is the sound velocity of the fluid, f is the frequency of the ultrasound, and K is a constant that is dependent on the nature of the fluid.
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
0 100 200 300 400 500 600
Flowrate (m3/hr)
Erro
r (%
)
Turndown based on nominal max flow of 500 m3/hr
15:1 10:1
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
19
The equations above show that the attenuation coefficient is proportional to viscosity and frequency squared. This allows us to examine the effect of viscosity on signal strength for various combinations of path length and transducer frequency typical of those used in ultrasonic flowmeters. Figure 22 below shows the relative signal amplitude versus path length for three different combinations of viscosity and frequency. Taking first the case of a 5 cSt oil and a signal frequency of 1 MHz we can see that the signal is attenuated by less than 10% over a path length of 40 inches. If the oil viscosity is increased to 500 cSt, the 1 MHz signal would now attenuate much more rapidly, with the signal being reduced to less than 10% when the path length is just 10 inches long. Using a lower frequency of 0.5 MHz, the attenuation for a given viscosity is reduced, and now at 500 cSt with a 10-inch path length the attenuation is less than 50%.
Figure 22 Signal amplitude vs distance for three combinations of frequency and viscosity Given an understanding of how frequency and viscosity affect signal attenuation, it is important to consider how much attenuation can be allowed before the accuracy of the ultrasonic transit time measurement is affected. This is dependent on a number of factors, including the quality of the electronics, the method of transit time detection employed, the size of the meter and the minimum flowrate that we wish to measure. In terms of the quality of the electronics, the ratio of the measured signal to the coherent (stationary or correlated) noise is important. If, for example, a signal-to-coherent noise ratio (SNRc) of 200:1 is required for accurate transit time measurement, and the flowmeter has baseline SNRc of 1000:1 , then we can estimate that the maximum allowable attenuation would be -14 dB (i.e. a five-fold reduction in signal strength). Then with knowledge of the meter geometry and signal frequency, it is possible to use the equations above to estimate the maximum viscosity corresponding to this amount of attenuation. From the equations and the data presented in Figure 22, it is clear that viscous attenuation is more severe over long distances. However this potential limitation is offset by the fact that the transit times also increase in large pipe diameters, thus placing lower demands on the
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 5 10 15 20 25 30 35 40
Distance (inches)
Sign
al a
mpl
itude
A/A
o.
5 cSt, 1 Mhz500 cSt, 1 MHz500 cSt, 0.5 MHz
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
20
timing accuracy, and hence lower requirements in terms of the signal-to-noise ratio of the received signals. It can be shown that to achieve the same relative uncertainty in the transit time measurements at a given velocity, the required signal-to-coherent-noise ratio is inversely proportional to the path length (or pipe diameter). This is illustrated in Figure 23, which shows a simple example of the required value of SNRc for a single diameter path at an angle of 45º, sound velocity equal to 1250 m/s, a minimum velocity of 1 m/s and the requirement for +/- 0.15% maximum uncertainty in the transit time measurements.
Figure 23 Example calculation of required SNRc for a single diametric path
Reducing the signal frequency requires a higher signal-to-noise ratio to maintain the same timing uncertainty, but also reduces the viscous attenuation. To examine the trade off between these opposing requirements, we can use the baseline and minimum required SNRc to calculate the allowable attenuation as a function of pipe diameter, and then translate this into a maximum viscosity. Figure 24 below shows the result of performing this calculation for the single diametric path example that was also used to generate the data in Figure 23. A baseline SNRc of 1250 has been assumed for this example. It can be seen that the reducing demands on transit time measurement accuracy with increasing size do indeed act to offset the increased attenuation and the resulting maximum viscosity is fairly constant over a wide range of meter size. It can also be observed that the lower signal frequency of 0.5 MHz significantly increases the viscosity limit for meters of 12-inch diameter and above.
0
100
200
300
400
500
600
700
800
900
1000
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
6 8 10 12 14 16 18 20 22 24
Req
uire
d si
gnal
to c
oher
ent n
oise
ratio
Req
uire
d tr
ansi
t tim
e un
cert
aint
y (n
s)
Pipe diameter (inches)
Required transit time uncertainty (ns)
Minimum SNRc
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
21
Figure 24 Example calculation of maximum viscosity for a single diametric path
When we now consider the multipath meter with reducing nozzle we find that there are further benefits of this design when considering high viscosity applications. Firstly the path lengths are obviously shorter than the paths in a full-bore meter of the same pipe diameter, so less attenuation is incurred. Secondly, although the overall transit times are also shorter, the velocity in the throat is much higher, and hence for the same velocity in the upstream pipe, the transit time difference is higher and the demands in terms of signal-to-noise ratio are lower. The result is that the meter with reducing nozzle can tolerate much higher viscosity than a full-bore meter of the same pipe size. This is illustrated in Figure 25 where the calculated maximum viscosity limit is compared for the single diametric path full-bore meter, and the multipath meter with reducing nozzle. As before the baseline SNRc and minimum pipe velocity are assumed to be 1250 and 1 m/s respectively, and the uncertainty limit is set at +/- 0.15%. Figure 25 also shows the maximum viscosity calculated for the multipath meter with reducing nozzle when the velocity is increased to 5m/s. It can be observed that with the increase in the velocity, the maximum viscosity also increases. This means that in applications where a limited turndown is acceptable (such as batch offloading from floating production storage and offloading vessels where low velocities constitute only a very small part of the batch) even higher viscosities can be tolerated.
0
20
40
60
80
100
120
140
160
180
200
6 8 10 12 14 16 18 20 22 24
Max
imum
vis
cosi
ty f
or 1
m/s
min
vel
ocity
(cS
t)
Pipe diameter (inches)
1 MHz signal frequency
0.5 MHz signal frequency
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
22
Figure 25 Example calculation for full-bore and reducing nozzle meters at 0.5 MHz
In reality the relationships between viscosity, meter diameter, received signal frequency and signal-to-noise ratio are more complex than we have described here. For example, the level of coherent noise can actually reduce with increasing viscous attenuation, with the result that the signal-to-noise ratio does not reduce as much as expected. Furthermore, it is possible to allow different signal frequencies for different sized meters and gain further improvements. For Caldon meters a database of results has been compiled by performing experiments and varying each of the relevant parameters, and this allows empirical evaluation of the maximum viscosity limits for each meter size. However the analytical results in this section do reflect the geneneral trends and serve to highlight the following facts:
High viscosities can be tolerated without compromising measurement uncertainty The limit of viscosity increases with increasing flow velocity Use of appropriate signal frequencies can extend operation to higher viscosities The meter design with reducing nozzle can tolerate higher viscosities than a full-bore
meter design 4.3 Pressure Loss Comparisons One of the potential benefits of ultrasonic meters is that they can be non-intrusive and therefore create no more pressure loss than a standard piece of pipe of the same length. By adding a reducing nozzle to the meter the permanent pressure loss in increased, so it is of interest to compare this with other metering technologies to evaluate if the pressure loss advantage is retained. Estimated pressure loss has been calculated for a nominal pipe diameter of 6-inches and viscosity of 100 cSt. The calculation for the turbine meter was taken from the Smith MV Series turbine meter data sheet [13]. The data for the tube bundle and one of the calcs for the Coriolis meter (calc 1) were obtained by assuming pressure drop proportional to velocity squared and interpolating the data published by NEL [3]. The PD meter data is taken from the data sheet for the Smith Meter 6" Steel Model G6 PD Rotary Vane Meter, and is actually
0
500
1000
1500
2000
2500
3000
3500
6 8 10 12 14 16 18 20 22 24
Max
imum
vis
cosi
ty (
cSt)
Pipe diameter (inches)
Single diametric path full-bore meter @ 1 m/s
Multipath meter with 0.63 beta reducing nozzle @ 1 m/s
Multipath meter with 0.63 beta reducing nozzle @ 5 m/s
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
23
for a low viscosity product of 2 cSt [13]. The second Coriolis meter calculation is taken from the E&H Promass F technical manual [14]. It can be seen from Figure 26 that the pressure loss increases in the following order:
Full bore ultrasonic meter Ultrasonic meter with reducing nozzle and recovery cone Turbine meter Tube bundle PD meter Coriolis meters
It is clear therefore that the pressure loss advantage of the ultrasonic meter is retained, even when a diameter ratio of less than 0.64 is used to ensure an effective flattening of laminar velocity profiles. The calculations for the tube bundle show a higher pressure loss than would normally be expected for this type of device. This is because the relatively high viscosity results in laminar or transitional conditions in the tubes themselves and increases the frictional losses relative to the results that would be obtained using the loss coefficient that is normally applied in turbulent flow. The fact that the tube bundle alone creates a significant pressure loss is worthy of further discussion. This result emphasises the fact that an ultrasonic meter has the greatest pressure loss advantage when used without flow conditioning. Given the known sensitivity of typical multipath meter designs to swirl, dispensing with upstream flow conditioning requires the use of a meter design that is insensitive to swirl, such as the Caldon 280C 8-path meter design [5]. The Caldon meter with reducing nozzle can be made in both 4-path and 8-path formats, and therefore can be used in most circumstances without requiring additional upstream flow conditioning.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
24
Figure 26 Permanent pressure loss comparison for meters in a 6-inch pipe
5 CONCLUSIONS Metering of viscous oils presents challenges for many types of flowmeter. Positive displacement meters are well suited for high accuracy measurement of viscous oils but suffer from being fairly expense, having limited flow capacity and using moving parts that are susceptible to damage and wear. Turbine meters are sensitive to viscosity changes, are normally limited to lower viscosity products and also have moving parts. Coriolis meters appear promising for measurement of viscous oils but generate relatively high pressure losses and are reported to be susceptible to errors as viscosity increases. Ultrasonic meters also have to overcome a number of issues in order to be used successfully for measurement of viscous oils at low Reynolds numbers. Firstly they must tolerate increased signal attenuation and still operate with sufficiently good signal-to-noise ratio to permit accurate transit time measurements to be made. Secondly, they must be able to accurately measure in the laminar/turbulent transition region where fluid mechanics dictate that rapid changes in velocity profile occur. A method for improving ultrasonic meter performance in the transition region has been presented whereby a reducing nozzle with a diameter ratio of less than 0.64 is used to stabilise and flatten the velocity profile. It has been shown that this not only brings the laminar and turbulent profiles closer together but also results in a more gradual transition leading to improved meter performance in terms of both repeatability and reproducibility. Furthermore, in terms of attenuation this design extends the maximum viscosity that can be tolerated relative to a conventional full-bore meter of the same pipe size. The resulting new meter design extends the turbulent flow performance of Caldon ultrasonic meters through the transition region whist still maintaining lower permanent pressure loss than turbine, PD or Coriolis meters.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0 100 200 300 400 500
Pres
sure
Dro
p (b
ar)
Flowrate (m3/hr)
Full bore ultrasonic meterUltrasonic meter with reducing nozzle and recovery coneTurbine meterTube bundlePD meterCoriolis meter (calc 1)Coriolis meter (calc 2)
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009
25
REFERENCES [1] API (1995), Manual of Petroleum Measurement Standards – Chapter 5 – Metering,
Section 1 – General Considerations for Measurement by Meters
[2] Nicholson, S (1994) Coriolis Mass Flow Measurement Proceedings of Flomeko 93, Glasgow, Scotland, June 1994
[3] Miller, G J and Belshaw, R (2008) Measurement of Flow in Viscous Fluids 7th South East Asia Hydrocarbon Flow Measurement Workshop, Kuala Lumpur, Malaysia
[4] Reynolds, O (1883) An experimental investigation of the circumstances which determine whether the motion of water in parallel channels shall be direct or sinuous and of the law of resistance in parallel channels.
[5] Brown, G J, Augenstein, D R and Cousins, T (2006) The relative merits of ultrasonic meters employing between two and eight paths 5th South East Asia Hydrocarbon Flow Measurement Workshop, Kuala Lumpur, Malaysia.
[6] Fronek, V (1978) Ultrasonic measurements of oil flow in a laminar flow-turbulent flow transition region Proceedings of FLOMEKO ’78, Groningen, The Netherlands, 11-15 September 1978, H H Dijstelbergen & E A Spencer (eds), North-Holland Publishing Co, Amsterdam, pp. 141-146.
[7] Brown, G J (1999) Long Term Evaluation of Ultrasonic Meters, Project No: OR7 (OSDC57), National Engineering Laboratory Report No. 315/99, September 1999.
[8] Schlichting, H (1968) Boundary Layer Theory, McGraw-Hill, New York.
[9] Wygnanski, I J and Champagne, F H (1973) On transition in a pipe, Part 1: The origin of puffs and slugs and flow in a turbulent slug J Fluid Mechanics, Vol 59, Part 2, pp 281 - 335.
[10] Shan, H, Ma, B, Zhang, Z and Nieuwstadt, F T M (1999) Direct numerical simulation of a puff and a slug in transitional cylindrical pipe flow J Fluid Mechanics, Vol 387, pp 39 - 60.
[11] Shan, H (2006) Personal communication.
[12] Brown, G J, Estrada, H, Augenstein, D R and Cousins, T (2007) Ultrasonic metering of liquefied natural gas for allocation and custody transfer, 6th South East Asia Hydrocarbon Flow Measurement Workshop, Kuala Lumpur, Malaysia
[13] http://info.smithmeter.com/literature/online_index.html
[14] http://www.endress.com: Technical Information Proline Promass 80F, 80M, 83F, 83M Coriolis Mass Flow Measuring System
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
1
Cost Benefit Analyses in the Design of Allocation Systems
Phillip Stockton, IMASS (formerly Smith Rea Energy Ltd)
1 INTRODUCTION
This paper considers issues pertaining to the costs of the implementation of
measurement systems, e.g. to reduce uncertainty, against the benefits accrued by a
reduction in exposure to loss of revenue in allocation systems. Statistical based
techniques are presented to assess the risk of loss of revenue.
In Section 2 these issues and methods are discussed and illustrated with simplified
theoretical examples. The discussion is principally in terms of flow meters but the
issues can equally be extended to any measurements used as inputs to allocation.
Indeed in Section 3 the techniques are applied to data from a real system in which the
cost savings accrued from a reduction in compositional sampling frequency were
compared with the potential impacts on the allocation system.
2 COST BENEFIT ANALYSIS ISSUES AND CONCEPTS
2.1 Introduction
This section explores the basis underlying cost benefit analysis associated with
measurement requirements for allocation.
In Section 2.2 a simple example system is introduced which is utilised to illustrates
some of the ensuing concepts explored in Sections 2.3 to 2.7:
Systematic allocation bias resulting from differences in meter uncertainty
Impact of meter uncertainty on the ability to detect meter bias
Impact of meter uncertainty on allocation uncertainty
Exposure to loss (risk aversion)
Techniques to evaluate exposure to loss
These issues are explored and the considerations serve to inform the approach to cost
benefit analysis presented in this paper. Finally in Section 2.8 a simplified cost benefit
analysis is performed, to illustrate some of the concepts.
2.2 Example System
Consider the simplified example below.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
2
Figure 1 – Simplified Schematic Two Streams Commingling
M
Field A
Field B
Commingled Stream
M
M
M Meter
40,000 bbl/d
(≡5,000 te/d)
20,000 bbl/d
(≡2,500 te/d)
20,000 bbl/d
(≡2,500 te/d)
The flow rates are presented in barrels/day and tonnes/day. For the purposes of the
example, the Fields produce similar oils such that their densities are the same and
when commingled their standard volumes are additive1. This renders the cost benefit
analysis calculations more transparent since oil revenue is generally in terms of
$/barrel.
The allocation system is a proportional one in which the metered export product is
allocated in direct proportion to the metered flow from each Field.
In this example consider the case when the commingled (or Export) stream and Field
A’s flow are measured to fiscal accuracy: ± 0.25%. What level of accuracy is required
for the Field B meter? In reality there is not a continuum of meter uncertainties that
could be installed but a number of distinct meter alternatives that can be compared –
this is the approach adopted in the simplified cost benefit analysis in Section 2.8.
It might be argued that since the uncertainties in the meters are normally distributed,
an individual meter could be over or under reading with equal probability and
therefore it doesn’t matter how good Field B’s meter needs to be. Any gains and
losses will even themselves out over a period of time and the cheapest meter should
be installed irrespective of its quality. This assumption is not strictly true and is
discussed in the next section.
1 In a real allocation system it would be desirable (generally) to allocate on a mass basis as liquid
volumes are not normally additive. Mass is conserved whereas volume generally isn’t.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
3
2.3 Allocation Bias
The field with the poor quality meter (higher uncertainty) will be systematically
under-allocated product. At first sight this may appear counter-intuitive but consider
the example above where the Export and Field A stream meters have a very low
uncertainty, negligible in fact when compared to the Stream B meter. Say that each
field is actually producing consistently 20,000 bbl/d each. We would observe that
Stream A would meter almost exactly 20,000 bbl/d and the total measured Export
would be almost precisely 40,000 bbl/d. However Stream B’s meter say has a ±10%
uncertainty so it is measuring flows in the typically in the range 18,000 to 22,000
bbl/d but over a period of time it is averaging 20,000 bbl/d. On a day when it
measures 18,000 bbl/d the allocation to A and B would be:
Stream A allocated 21,053 bbl
Stream B allocated 18,947 bbl
However, the next day, B’s meter reading swings to 22,000 bbl/d (average over the 2
days is 20,000 bbl/d) and the allocation is:
Stream A allocated 19,048 bbl
Stream B allocated 20,952 bbl
So totalising over the 2 days:
Stream A allocated 40,100 bbl
Stream B allocated 39,900 bbl
Stream B has been under allocated by 0.25%. This may not appear to be a large
percentage but it is systematic and refutes the claim that “things will even themselves
out over a period of time”. Interestingly justice has been done as the stream that has
invested in the better quality metering as at an advantage.
In reality the meter readings would be over-under reading with probabilities dictated
by the normal distribution and degree in accordance with their uncertainty. Based
upon these distributions an expected value of the allocated oil to Field A and B may
be calculated.
In effect this calculation takes every possible value that the Field A and B meter
readings could have, calculates the allocated quantities, and multiplies the result by
the probability of its occurrence. The sum of these probability weighted values is the
expected allocation result based on the probabilities. This is presented pictorially in
Figure 2 for Field B’s allocation:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
4
Figure 2 –Field B Probability Weighted Allocated Oil
19,8
75
19,8
91
19,9
07
19,9
23
19,9
38
19,9
54
19,9
70
19,9
86
20,0
02
20,0
18
20,0
34
20,0
50
20,0
65
20,0
81
20,0
97
20,1
13
15,000
16,746
18,492
20,238
21,984
23,730
0
10
20
30
40
50
60
70
80
All
oc
ate
d O
il x
Pro
ba
bil
ity
Meter A
Meter B
The horizontal axes are the range of A and B meter readings, covering ± 5 standard
deviations around the average value of 20,000 bbl/d. This covers in excess of
99.9999% of all the possible values each meter could read consistent with its
uncertainty. (It should be noted that the scales of these axes are different).
Each point on the surface represents a quantity allocated to Field B corresponding to
the values of Meter A and B readings multiplied by the probability of those two meter
readings occurring. The total volume under the surface is the “expected” allocated
quantity and this represents the average allocated quantity over a period of time.
The most probable allocation occurs when both meters read 20,000 bbl/d and this is
where the apex of the surface occurs. As the readings move away from the average the
probability diminishes to almost zero within 2 standard deviations. The surface is not
actually symmetrical in all planes and is weighted towards an under-allocation to Field
B.
For the case where Field A meter is ±0.25% and Field B’s is ±10% uncertainty, the
expected allocation to the two Fields is:
Stream A allocated 20,012.5 bbl
Stream B allocated 19,987.5 bbl
Section 4.2 presents the mathematical derivation of the expected under/over allocation
as a result of differences in metering quality. To illustrate the impact of Field B meter
uncertainty on both Fields’ expected oil allocation, the under/over allocations are
plotted as a function of its meter B uncertainty in Figure 3.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
5
Figure 3 – Expected Oil Allocation – Variation in Meter B Uncertainty
19,950
19,960
19,970
19,980
19,990
20,000
20,010
20,020
20,030
20,040
20,050
0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00% 14.00% 16.00% 18.00% 20.00%
Field B Uncertainty (+/-%)
All
oc
ate
d O
il (
bb
ls)
Field A Field B
This bias in the allocation is small but it is systematic and occurs as a result of the
mathematics of the equations. In the above example, at 10% meter uncertainty for
Field B, the expected under-allocation is 12.5 bbl/d which is worth approximately
$625/d (assuming 50 $/bbl oil price). Over a year this translates into nearly a quarter
of a million dollars.
Though a small effect it is important to understand when apparently reasonable
assumptions are not true.
2.4 Meter Bias Detection
A second reason meters with relatively high levels of uncertainty are not desirable is
that they can mask systematic bias as illustrated in Figure 4.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
6
Figure 4 – High and Low Quality Meter Uncertainty Distribution in Realtion to
Bias
17000 18000 19000 20000 21000 22000 23000
Oil Flow (bbls/d)
High Uncertainty
Low Uncertainty
True flow
The true meter reading is 20,000 bbl/d but both meters under read by 300 bbls/d. With
the lower quality meter (higher uncertainty) the problem is not so easy to detect since
the bias is still located in the central broad peak of its probability envelope. In
contrast, the bias lies beyond the standard uncertainty confidence level and would be
more apparent.
2.5 Impact on Allocation Uncertainty
In proportional allocation systems the uncertainties in the contributing stream flows
have impact on each others allocated quantity uncertainty. Consider the example
above, where Field A and the Export meter uncertainty are ±0.25%. Field B’s meter
uncertainty is ±10%, at the example flowrates the uncertainties in the allocated
quantities are:
Stream A allocated oil uncertainty ±1,000 bbl, ±5%
Stream B allocated oil uncertainty ±1,000 bbl, ±5%
These uncertainties were calculated analytically using propagation of uncertainties as
described in the GUM [1].
Both streams have the same allocation uncertainty but for the case of Stream A, the
uncertainty is much greater than its individual meter uncertainty. The variation in
allocation uncertainty with Stream B meter uncertainty is presented in Figure 5.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
7
Figure 5 – Field Allocation Uncertainty with Variation in Meter B Uncertainty
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00% 7.00% 8.00% 9.00% 10.00%
Field B Meter Uncertainty (+/-%)
Fie
ld A
Allo
cati
on
Un
cert
ain
ty (
+/-
%)
Another option would be not to install a meter at all for Field B and allocate by
difference. This reduces Field A’s allocation uncertainty to ±0.25% (i.e. just equal to
the Field A meter uncertainty) and Field B’s uncertainty is ±0.56%. However, there
are other issues with by difference allocation which have to be considered. For
example there is no quality check built into the allocation in that if Field A meter
develops a problem this might not be detected so easily compared with the
proportional allocation system in which the sum of the Field meter flows should be
close to the total metered export (within the uncertainties). Also if Field B flow
reduces then its uncertainty rises sharply, at 1000 bbl/d it is over ±7% and at 100 bbl/d
it rises to over ±70%.
2.6 Exposure to Loss (Risk Reduction)
One of the key drivers in selection of meter is the risk of exposure to loss.
For one moment, if the issues surrounding meters with higher uncertainties (raised
above in Sections 2.3 to 2.5) are put to one side, then from a pure cost perspective
shouldn’t the cheapest meter always be installed?
From a purely probabilistic viewpoint the increased uncertainty introduced by
installing low quality metering is just as likely to result in a gain as a loss to one Field
(compensated by an equal and opposite gain or loss in the other Field) because the
meter is just as likely to under-read as over-read.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
8
This might be a reasonable approach if the Field owners had many such systems and
the value of the product was low since on the average over all the systems it is most
probable that the allocation results would even themselves out. In oil and gas systems
the value of the product is normally high compared with the meter costs and it is
unlikely that an investor in a Field would have such a vast portfolio of systems that he
would be relaxed that all the gains and losses would even themselves out.
In the simplified example above, 10,000 bbl/d represents a revenue stream of
$500,000/d so an improvement in allocation uncertainty of 1% would reduce the
uncertainty in allocated revenue by $5,000/d. This could be a gain or a loss though to
an individual Field but that would be compensated by an equal gain or loss to the
other Field(s). The extra cost of better quality metering is a cost to the system as a
whole and therefore a guaranteed expense but it buys more certainty in the allocated
quantities. In effect the investment in the improved metering is buying insurance
against loss of revenue. The study of attitude to risk in decision making is termed
Preference or Utility, Theory. An example serves to illustrate the concept:
Imagine you are presented with the choice of being given $1,000 or being entered into
a lottery, the outcome of which depended on the toss of a coin, in which you stood to
win $2,000 or nothing with equal probability. Based on the probabilities the expected
average outcome of both choices is $1,000. However, most people would tend to
select the guaranteed $1,000.
However, if the game was changed slightly so that the guaranteed quantity was
reduced to $900 then on probabilistic grounds alone you should enter the lottery.
However, it is likely a lot of people would still take the guaranteed $900. In effect
these people are giving up $100 to insure against receiving nothing.
If the guaranteed quantity was reduced further then there would come a value when
you decide that it was worth risk to play the lottery. The reduction in the guaranteed
quantity to this point is the cost you are prepared to pay to insure against a loss.
The essential point is that some money has effectively been traded to insure against a
loss. Translating this to the cost benefit analysis, how much are you willing to invest
in a meter for Field B to reduce the exposure to loss of revenue caused by the
uncertainty in the meter. The difficulty is that the amount to be invested is subjective
and depends on a number of factors:
The amounts involved; if in the above lottery example the guaranteed amount
was $1 and the maximum win was a $2, it is much likely you would gamble
since you could probably afford to lose $1. If the guaranteed amount was
$1,000,000 it is unlikely you would be willing to gamble that for the chance of
winning $2,000,000.
Your attitude to risk, some people and organisations are more less risk averse
than others.
The next section describes two possible methods to evaluate the exposure to loss.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
9
2.7 Evaluating Exposure to Loss
Allocation Uncertainty Approach
This involves calculating the Field allocation uncertainty associated with the different
meter options. The minus side of this uncertainty will be the exposure to loss at the
97.5% confidence level, i.e. there is a 2.5% probability that the loss will lie beyond
the minus uncertainty band. This is illustrated in Figure 6.
Figure 6 – Distribution in Allocation Uncertainty
-4 -3 -2 -1 0 1 2 3 4
Standard Deviations
~95% Probability
result will lie within 2
standard deviations, ~
equivalent to quoted
uncertainty
~2.5% Probability
result will lie lower than
minus 2 standard
deviations
The allocation uncertainty can be calculated for each meter option and converted into
an equivalent revenue quantity. The difference in revenue uncertainty associated with
the two options then represents the reduction in exposure to loss (over Field life) and
this can then be compared against the difference in meter costs. This is illustrated in
Figure 7.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
10
Figure 7 – Reduction in Loss Exposure Based on Allocation Uncertainty
17000 18000 19000 20000 21000 22000 23000
Allocated Oil (bbls)
High Uncertainty
Low UncertaintyReduction in
exposure to
loss of
allocated oil
These figures were calculated at the 97.5 % confidence level. The answer could be
different based on different confidence levels.
The next approach integrates the risk of loss over a range of confidence levels.
Integrated Risked Exposure Approach
By analysing the impact of the variations in Field B’s metered flow, due to its
uncertainty, on the allocation results the mis-allocation of revenue can be calculated.
Using the example above the impact of variations in the meter reading on the
allocated revenue has been analysed using a Monte Carlo simulation. In each run of
the simulation, the meter flow was varied in accordance with appropriate standard
deviation figures and the allocated revenues collated. The results of such a Monte
Carlo simulation for Field B are presented in Figure 8.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
11
Figure 8 – Distribution in Field B Revenue
0
50
100
150
200
250
300
-$125,000 -$100,000 -$75,000 -$50,000 -$25,000 $0 $25,000 $50,000 $75,000 $100,000 $125,000
Over/Under Allocated Revenue ($)
Nu
mb
er
of
Ru
ns
1000 runs were used to generate the data in the above chart. The histogram bars
represent the number of runs in which the allocated revenue lay in the bandwidth
indicated on the horizontal axis – (the figures on the x-axis are the midpoints of the
bands). The values on the axis refer to the difference between daily allocated revenue
in the run compared with the average value, i.e. the over/under allocation of revenue
compared with the average.
The blue line is a plot of the normal distribution curve with the same average and
standard deviation; the curve demonstrates that the allocated revenue is normally
distributed.
On any individual run there is a chance that the allocated revenue could be anywhere
along the x-axis but the probability diminishes the further from the mean. Because the
revenue is normally distributed the probability or risk associated with any individual
under- or over-allocation of revenue figure can be calculated.
It is possible to multiply each lost revenue figure by its individual probability of
occurrence. These can then be summed to give a total risked lost revenue figure. This
is, in effect, the Integrated Risked Exposure to mis-allocated revenue and is calculated
by the following equation:
8
BB
URL (1)
The derivation of this equation is presented in Section 4.1.
The difference in the risked loss exposure for the two metering options can then be
compared with the difference in meter costs as described above for the Allocation
Uncertainty approach.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
12
The two methods give different answers, with the Allocation Uncertainty approach
being more conservative. There is no right or wrong method because the approach
adopted depends on attitude to risk aversion. They are just two possible methods that
provide a degree of auditability but the final decision should be based on engineering
judgement, considering all the factors discussed in the above sections.
2.8 Simplified Cost Benefit Analysis
The two methods are applied to a simplified example in this section. Using the
throughputs and meter uncertainties presented in Section 2.2 a comparison is made
between installing an equivalent fiscal quality meter for Stream B (± 0.25%) versus an
allocation standard meter with an uncertainty of (± 5%).
The figures presented are fictitious but intended to be roughly representative.
The cost of a fiscal quality meter, with attendant proving facilities, etc., has been
estimated to be $5,000,000. The yearly maintenance costs associated with maintaining
its accuracy have been assumed to be $50,000 per year.
The cost of the allocation quality meter has been assumed to be $500,000. The meter
maybe the same type as the fiscal quality meter but it won’t have all the proving,
spares, etc. The yearly running costs have been assumed to be negligible.
The analysis is based on a 10 year life at an oil price of $50/bbl.
The figures for the case where a fiscal quality meter is installed for Field B is
presented in Table 1.
Table 1 – Cost Benefit Analysis Field B Fiscal Meter Field A Field B Export
Flow bbl/d 20,000 20,000 40,000
Field B Fiscal Meter
Meter Uncertainty (Relative) ±% 0.25% 0.25% 0.25%
Allocation Uncertainty (Relative) ±% 0.31% 0.31%
Loss exposure at 95% confidence level (Field Life) $ $11,175,797 $11,175,797
Loss exposure integrated (Field Life) $ $2,229,249 $2,229,249
Meter Installation Cost $ $5,000,000
Meter OPEX (Field Life) $ $500,000
Total Cost $ $5,500,000
The allocation uncertainty for both fields is ±0.31% and at the quoted flows this
equates to a loss exposure of over $11 and $2 million for the allocation uncertainty
and integrated risked approaches respectively.
Performing the same analysis for the allocation quality meter produces the results
presented in Table 2.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
13
Table 2 – Cost Benefit Analysis Field B Allocation Quality Meter Field B Allocation Meter Field A Field B Export
Meter Uncertainty (Relative) ±% 0.25% 5.00% 0.25%
Allocation Uncertainty (Relative) ±% 2.52% 2.52%
Loss exposure at 95% confidence level (Field Life) $ $91,818,541 $91,818,541
Loss exposure integrated (Field Life) $ $18,315,149 $18,315,149
Meter Installation Cost $ $500,000
Meter OPEX (Field Life) $ $0
Total Cost $ $500,000
The allocation uncertainties for both Fields have now increased to ±2.52%. The
poorer quality Field B meter is having a deleterious effect not only on Field B’s loss
exposure but also on Field A’s despite the investment in its fiscal quality meter.
The reduction in meter costs are compared with the increased loss exposure for the
two approaches in Table 3.
Table 3 – Cost Benefit Analysis Comparison of Field B Allocation Quality Meter Cost Benefit Analysis
Meter Cost Saving $ $5,000,000
Increase in Loss Exposure at 95% Conf Level $ $80,642,744
Increase in Loss Exposure integrated $ $16,085,900
Since the exposure to loss is experienced by both fields it could be argued that the
values should be doubled when considering the impacts on the system as a whole and
not just Field B.
The two methods produce seemingly largely different loss exposure figures but in fact
the two methods provide similar conclusions. To illustrate this, the uncertainty in the
Field B meter would need to be reduced to 1.9% to reduce the integrated risk loss
exposure to $5,000,000 compared with 0.7% for the 95% confidence level exposure.
Similarly a reduction in the Field B meter flow would reduce the exposures and for
the Integrated Risk Exposure approach the flow would need to be 3,500 bbl/d to reach
the break even point and to around 600 bbl/d for the Allocation Uncertainty approach.
The reduction in exposure is not directly proportional to either the flow rate or the
meter uncertainty because of the non-linearity in the Field allocation uncertainty.
For this simplified example both methods illustrate the exposure to loss is
significantly in excess of the cost of the meter upgrade.
3 COST BENEFIT ANALYSIS OF SAMPLING OPEX IN AN OIL SYTEM
The example presented concerns an oil terminal allocation system fed by a number of
offshore facilities via a subsea pipeline see Figure 9.
Oil is fed from various offshore facilities (Fields) into a common pipeline and
delivered to the onshore terminal. The products from the terminal are allocated to the
various Fields based on the quantities they have delivered to the terminal. This
allocation is performed at a mass based, hydrocarbon component/fraction level.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
14
Figure 9 – Oil Pipeline Schematic
Field D
Field C
Field A
Field B
Field F
Field E
Crude
Ethane
LPG
HP Fuel
LP Fuel
Flare
Oil
Terminal
1
2
1
1
2
3
4
1
2
2
3
4
5
6
7
8
9
1
5
6
1
The compositions of the various streams, associated with each Field, vary
considerably ranging from heavier oils to light condensate streams.
An integral part of the allocation system is the sampling and analysis of fluids, which
incurs OPEX in the region of £350,000 per annum. With a decline in production
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
15
through the oil terminal reductions in OPEX associated with the system were being
sought.
The Fields’ feed Streams are all metered and sampled to determine their composition
and flow, as are the onshore products. Weekly samples are taken from each Stream
and analysed; these weekly samples are also combined into a monthly sample upon
which a detailed assay is performed. However, it was observed that some of the
compositions of the Streams delivered into the pipeline appear to remain relatively
constant.
The objective of cost benefit analysis was to analyse options to reduce weekly and
monthly sampling (and analysis) and hence OPEX associated with the allocation
system whilst ensuring the integrity of the system is not compromised and
safeguarding against unacceptable exposure to mis-allocation of the terminal products.
For the purposes of this illustration the OPEX savings associated with a reduction in
the weekly samples is presented. A similar analysis (not presented) was applied to the
monthly assays.
3.1 Sampling and Analysis
For each Stream, ideally samples are collected over a week using flow proportional
auto-samplers. This means that for practical purposes the samples collected are
completely representative of the fluid that has been produced in that Stream over the
sample period. This means that the current integrity of the Allocation system is high
as is reasonably practical.
The weekly analysis measurements are used to allocate the crude, LPG and ethane
products along with fuel gas and flare at the Oil Terminal.
The cost of sampling was approximately $400 per sample for the weekly samples.
Approximately 50 samples per month were taken with a total monthly OPEX of
approximately $20,000 per month.
The OPEX was split between the fields in proportion to their throughput:
Table 4 –Allocation of Sampling OPEX
Field Sampling OPEX ($)
Allocated
OPEX
Monthly
Throughput
(tonnes)
Percentage of
Production
A 2,290 64,985 11%
B 5,488 155,746 26%
C 8,820 250,281 42%
D 728 20,665 3%
E 397 11,271 2%
F 3,090 87,692 15%
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
16
The cost benefit analysis considered the impact of reducing OPEX by obtaining
samples on a discontinuous basis; e.g. collect the weekly samples only one week out
of four (omitting three weeks of samples). This approach relies on the sample
obtained being sufficiently representative of the period over which it is applied in the
allocation; hence this introduces risk of mis-allocation of Oil products.
The approach involved a fundamental change in the philosophy associated with the
Oil Allocation system, in that sampling would be performed on a discontinuous rather
than the current continuous basis. This introduces an element of risk, since it is
possible that the assumed constant composition may in fact vary in the un-sampled
period; this would result in a mis-allocation of Oil products associated with the Field
whose sampling frequency was reduced. Any gain enjoyed by that Field would be
exactly balanced by a corresponding loss distributed across the other Fields and vice
versa. Hence, any mis-allocation would impact all Fields in the allocation system.
Sources of variation in individual Stream compositions and properties can be
attributable to a number of factors, which include:
Changes in the relative flows of wells from different reservoir zones
Changes in the composition of the hydrocarbons in the reservoirs from which
the Stream is produced, e.g. caused by falling reservoir pressure
New wells brought on Stream
Changes in offshore process operating conditions
Uncertainty in the laboratory sample measurements.
With regard to reducing sampling frequency the desirable behaviour of the measured
composition would be:
The underlying average value is constant for sustained periods, i.e. there is no
systematic shift in the value over the period
Any fluctuations are random about the average value
The fluctuations are sufficiently small that the savings in reduced sampling
and analysis OPEX outweigh any mis-allocation in revenue incurred
Or, the size of the fluctuations is within the legitimate measurement
uncertainty (i.e. within measurement tolerance).
An important aspect of this approach is determining, based on historical data, whether
a Stream’s composition does remain essentially constant within acceptable levels of
variability. The level of confidence in this assumption increases with the number of
consecutive historical analyses over which the composition or property was deemed to
be stable. A reduction in sample frequency can then be justified once the desired
confidence in the stability of Stream has been established.
The next section (Section 3.2) is concerned with establishing the degree of variability
in compositions and examining the associated impact on allocated revenue. This was
accomplished by simply reviewing the data and by the application of some statistical
methods.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
17
3.2 Compositional Variability
Some statistical analysis was employed to detect the presence of trends in the data
along with some sensitivity analysis to estimate the impact of the changing
compositions and properties on allocated revenue.
General observations associated with the compositional data were:
There are two types of Streams: oil and condensate/NGL
Oil Streams consist principally of C5P (>90%)
Lighter condensate/NGL Streams have a more evenly distributed spread of
components
The variability in the composition of the lighter condensate Streams is
considerably greater than that observed with the oil Streams.
For the oil Streams, any increase in C5P generally results in corresponding
decreases in the majority of the other components and vice versa .
In order to condense the data and directly compare the variability of the various
Streams, the deviations associated with all the components in a Stream were pooled to
produce a figure that represented the overall variability of a Stream’s composition2. It
is not the absolute pooled standard deviations themselves that are of interest but rather
the relative values between the Streams. The pooled standard deviations have been
used as a mechanism to determine the most and least compositionally variable
Streams and are presented in Figure 10:
Figure 10 – Weekly Compositional RMS Variability
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Stream
A1
Stream
A2
Stream
B1
Stream
B2
Stream
B3
Stream
C2
Stream
C3
Stream
C4
Stream
C5
Stream
C6
Stream
C7
Stream
C8
Stream
C9
Stream
D1
Stream
D2
Stream
E1
Stream
F1
Stream
C1
RM
S S
td D
Ev
Streams E1 and C1 were focussed on in the statistical analyses because of their low
compositional variability. The rationale adopted was that if a case to reduce sampling
2 The pooled standard deviation was obtained from the root mean square (RMS) of the component
standard deviation figures.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
18
frequency cannot be demonstrated for these apparently stable Fields and Streams then
the case cannot currently be made for any of the Fields or Streams.
In order to focus on longer-term changes in the various Streams’ compositions the
dominant C5P component was considered in further detail. To remove some of the
scatter observed and identify systematic trends, the 8-week rolling average of the C5P
content has been plotted in Figure 11 for several of the streams.
Figure 11 – C5P Component Moving Average
C5P Moving Average
Variation
92
93
94
95
96
97
98
99
23-Jun-00 12-Aug-00 01-Oct-00 20-Nov-00 09-Jan-01 28-Feb-01 19-Apr-01 08-Jun-01 28-Jul-01
Date
Wt
%
Stream A1 Stream B1 Stream B2 Stream B3 Stream D1 Stream E1 Stream F1 Stream C1
Streams C1 and E1 (emboldened in the plot) do appear comparatively stable but even
these streams show evidence of compositional drift. The Stream C1 C5P content (not
averaged) is plotted on an expanded scale in Figure 12:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
19
Figure 12 – Stream C1 C5P Variation June '00 to July '01
97
97.05
97.1
97.15
97.2
97.25
97.3
97.35
97.4
97.45
97.5
23-Jun-00 12-Aug-00 01-Oct-00 20-Nov-00 09-Jan-01 28-Feb-01 19-Apr-01 08-Jun-01 28-Jul-01
Date
Wt
%
Weekly Composition 3rd Qtr Ave 4th Qtr Ave
The red line depicts the actual data values, the blue dotted line is the average value in
the third quarter and the purple line the average in the 4th quarter of the period
considered. There appears to be a systematic rise in the C5P content of approximately
0.15 wt% between the final two quarters. The question of whether this shift is a
systematic change or whether it could have arisen by chance alone, consistent with the
typical scatter in the data, has been tested statistically3. The results of the test show
that the change in composition is statistically significant.
This Stream in particular had been identified as one in which the composition was
stable. However, the analysis illustrates that even in this apparently stable stream there
are detectable variations in composition and it is not safe to assume any of the
streams’ composition remained constant.
However, these small changes, though real, may result in little impact on the
allocation results. This has been analysed for Field E, which also exhibits relative
compositional stability. Field E was selected as it comprises only one Stream (E1) and
hence the impact on its allocated quantities was relative simple to determine. This
Field has the lowest throughput (see Table 4) and hence impacts on allocated revenue
would be the smallest.
3 The method employed to determine if there is a change in the average value of the C5P content
between the 3rd and 4th quarter was a comparison of means for independent samples using a two-tailed
Student’s t-test. An F test was initially used to determine if the statistical variances in the data for the
two periods could be pooled to calculate the standard error in the means. If not, then Satterthwatite’s
approximation was used to determine the degrees of freedom in the calculation of the standard error.
The value of the standard error was compared against the observed difference in the means from the
two periods to determine the t statistic value. The confidence level that the two means are different can
then be determined from standard reference tables of the t statistic.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
20
To gain an understanding of the impact of reducing the sample frequency, the
allocation results were recalculated substituting each individual weekly sample
composition, in place of the flow-weighted combined composition, as inputs to the
allocation model. The Field E revenue, based on the flow-weighted analysis over the
four weeks, was $2,541,018. The flow-weighted analysis comprised three samples
taken over the period. Figure 13 shows the impact on allocated revenue of basing the
allocation on the individual samples compared with the averaged composition:
Figure 13 – Field E - Impact of Weekly Sample Composition
Highlander - Impact of Weekly Sample Composition
-$1,000
$0
$1,000
$2,000
$3,000
$4,000
$5,000
$6,000
06-May-01 13-May-01 20-May-01Over/
Un
der
All
ocati
on
of
Reven
ue (
$)
Total
OPEX
Saved
The histogram bars represent the difference in allocated revenue associated with the
individual weekly compositions compared with the combined composition. Hence, the
first histogram bar shows that Field E would have been allocated nearly $6,000 dollars
more in revenue terms if the allocation had been based on the sample drawn from the
first week only. This also means that the other five Fields would collectively have lost
$6,000 revenue. Though Field E appears to gain $6,000 based on the first weekly
sample, the allocation could have resulted in a loss as can be seen from the chart if the
allocation had been based on the 20th May sample.
The OPEX associated with each Stream’s weekly sample is $400, hence if two out of
the three Field E samples were omitted the total OPEX saved would have been $800
dollars over the month - this value is indicated by the dashed blue lines on the chart.
This provides some indication of the relative value of the total OPEX savings in
comparison to the potential mis-allocated revenue for the system as a whole.
The impacts presented in Figure 13 are associated with the Field E, whose throughput
is the smallest and whose composition is relatively stable. Larger impacts are
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
21
observed for other Fields, for example a similar reduction Field F’s weekly sampling
frequency results in mis-allocated revenue of the order of $60,000.
Returning to Stream C1, the 0.15 wt% rise in C5P content described above, incurs an
increase in allocated revenue of approximately $15,000 based on its typical
throughput of 100,000 tonnes per month.
These figures indicate the significant level of mis-allocation of revenue potentially
incurred for even the most apparently stable compositions, compared with modest
savings in OPEX afforded by reducing their sampling frequency.
3.3 Integrated Risked Exposure to Lost Revenue
It appeared apparent in Section 3.2 that even the most compositionally stable Stream
exhibits sufficient variation to warrant the full sampling OPEX. However, that was
just a snapshot of the data and this section analyses the data using a more statistically
structured approach.
In order to evaluate Integrated Risked Exposure to loss (as described in Section 2.7)
the standard deviation of the Field E allocated revenue is required. Because of the
complexity of the allocation, the standard deviation was calculated from a Monte
Carlo analysis.
In each run of the Monte Carlo simulation, the weekly sample composition of Field E
was varied in accordance with appropriate standard deviation figures and the allocated
revenues collated. The results of such a Monte Carlo simulation for Field E are
presented in Figure 14.
Figure 14 – Field E – Allocated Revenue Monte Carlo Simulation Results
0
50
100
150
200
250
300
-$125,000 -$100,000 -$75,000 -$50,000 -$25,000 $0 $25,000 $50,000 $75,000 $100,000 $125,000
Over/Under Allocated Revenue ($)
Nu
mb
er
of
Ru
ns
In each run only the Field E composition was varied according to standard deviations
typical of the Field E weekly analyses over the period June 2000 to July 2001. The
histogram bars represent the number of runs in which the allocated Field E revenue
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
22
lay in the bandwidth indicated on the horizontal axis – (the figures on the x-axis are
the midpoints of the bands). The values on the axis refer to the difference between
allocated revenue in the run compared with the average value, i.e. the over/under
allocation of revenue compared with the average.
The mean allocated revenue was calculated as $2,541,000 and the associated standard
deviation in the revenue figure slightly in excess of $3,000. The blue line is a plot of
the normal distribution curve with the same average and standard deviation; the curve
demonstrates that the allocated revenue is normally distributed.
For the example above, the risked exposure to lost revenue was calculated to be
$1,270 using Equation (1). The cost of performing the Field E weekly sample is $400
per week, approximately $1,600 per month if four samples are taken. If the sampling
is reduced to one week in four then the OPEX savings would be $1,200.
Collectively to the system the risk of mis-allocation is comparable with the total
OPEX savings and a reduction in sampling frequency may marginally be justified in
this case.
These impacts are associated with the Field E, whose throughput is relatively small
and whose composition is relatively stable and hence the most likely candidate for a
reduction in sampling frequency. Much larger impacts are observed for other Fields;
for example the same analysis applied to Field F produced a risked exposure to lost
revenue in excess of $20,000.
In general, there appears to be little opportunity to justify the risk of mis-allocation
associated with sampling frequency reduction and these findings concur with the
analyses described in Section 2.8.
4 MATHEMATICAL DERIVATIONS OF EQUATIONS PRESENTED
4.1 Risked Exposure to Lost Revenue
The allocated revenue is normally distributed and the probability is therefore
described by the locus of the standard normal distribution presented in Figure 15.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
23
Figure 15 – Standard Normal Distribution
Standard Normal Curve
0.0
0.1
0.2
0.3
0.4
0.5
-4 -3 -2 -1 0 1 2 3 4
Standard Deviations
The equation that represents the locus of the blue line is given by:
2
2
2
1Z
ey (2)
Where Z is given by:
meanxxZ (3)
The total area under the standard curve (blue line) is equal to 1 and the area under any
section of it represents the probability that the value of X will fall between the two
values of Z. In the above figure the area under the purple shaded region represents the
probability that Z will lie between –1 and –2.
When perturbing input variables randomly the calculated revenue will vary about the
mean value. The probability of being allocated revenue above or below the mean
diminishes as the value moves from the average value. Hence in calculating the risked
revenue it is necessary to multiply the revenue by the probability that that revenue
value would be allocated. For example if the average revenue allocated was $100,000
and the standard deviation was $1,000 the probability of being under-allocated $1,000
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
24
(i.e. $99,000) is approximately represented by the area under the thin purple vertical
line in Figure 16.
Figure 16 – Standard Normal Distribution
Standard Normal Curve
0.0
0.1
0.2
0.3
0.4
0.5
-4 -3 -2 -1 0 1 2 3 4
Standard Deviations (Z)
dZ
To calculate the risked mis-allocation exposure the under/over allocation of revenue
(x-xmean) needs to be multiplied by the area under the curve:
dZyxxR mean )( (4)
Where R is the risked exposure to mis-allocation of revenue and has a negative value
for under-allocation (i.e. loss). Substituting from (2) into (4):
dZexxR
Z
mean
2
2
2
1)(
(5)
Substituting from for (x-xmean) from (3) and forming the integral to calculate the
risked lost revenue between Z1 and Z2:
2
1
2
2
2
1Z
Z
Z
dZeZR
(6)
This integrates to:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
25
2
1
2
2
2
1
Z
Z
Z
eR
(7)
Hence, R is given by:
2
1
2
2 22
2
1ZZ
eeR (8)
Expressed in terms of x:
2
1
2
222
2
1meanmean xxxx
eeR
(9)
The total exposure to lost revenue is found between x1 equal to minus infinity and x2
equal to xmean. R then reduces to:
2R
(10)
The value of L is the negative of R and can be expressed in terms of the uncertainty
(equal to twice the standard deviation).
8
UL (11)
4.2 Expected Value of Field Allocation
The allocation to Field B is given by:
E
BA
AB M
MM
MAL (12)
The expected quantity allocated to Field A is calculated by integrating the allocation
equation with respect to its probability measure:
2
1
2
1**
B
B
A
ABABAE
BA
AB dMdMPPM
MM
MEL (13)
The probabilities of the range values of MA and MB are described by the normal
distribution described by equation (2) in the Section 4.1. Substituting these in (12)
produces:
2
1
2
1
*2*22
2
2
2
*2
*B
B
A
ABA
B
U
MM
A
U
MM
E
BA
AB dMdM
U
e
U
eM
MM
MEL
B
aveBB
A
aveAA
(14)
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
26
This equation has to be integrated numerically over a suitable range of meter values.
The uncertainty in the export meter could also be included and the integration would
be in three dimensions.
NOTATION
AL Allocated quantity
L Cumulative risked lost revenue
M Metered quantity
Mave
Average metered quantity
P Probability density function
R risked exposure to mis-
allocation of revenue
U Uncertainty in metered quantity
UR Uncertainty in allocated
revenue
x Revenue
xmean Mean value of revenue.
y Standard probability density
function
Z number of standard deviations
from mean
σ Standard deviation of revenue
Subscripts
A Field A
B Field B
E Export
5 REFERENCES
[1] Guide to the Expression of Uncertainty in Measurement, International
Organisation for Standardisation, ISO/IEC Guide 98:1995.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 1 of 20
Software for evaluation of uncertainty in liquid hydrocarbons flow
measurement systems
Verónica Mejía Gallardo
Diego Nelson Moncada Benavides
Centro de Tecnología Avanzada CIATEQ, Calzada del Retablo No. 150, Col. Constituyentes
Fovissste, C.P. 76150, Santiago de Querétaro, Qro., México, Teléfono: +52(442) 211 2600.
ABSTRACT
Flow measurement of liquid hydrocarbons and its resulting economic impact, creates
the need to make the correct execution of measurements. With this, competitiveness and
transparency of measurements operations are ensured and can help demonstrating
through the measurement associated uncertainty, accuracy and reliability of achieved
results, no matter if its end user or intermediate.
Even though many documents have been developed, in which methodology is described
and in some cases, they simplify the evaluation problem and the measurement
expression uncertainty, it is difficult for flow measurement systems users to develop this
kind of tasks, as support in their activities, due not only knowledge and experience is
required in the measurement systems operation, but also it is important the management
of statistical and mathematical concepts (some basics some advanced). Because this
situation, it is necessary to generate tools that can make easier the metrology application
to the daily measurement activities, which is one of the objectives pursued with this
software development. Using this application, there will be enough capacity to make
calculations in order to obtain the uncertainty of liquid hydrocarbons flow measurement
systems and which results, among other aspects, will serve as mechanism to evaluate
decision making risks, and also to determine the systems conformity grade against
established operations tolerance and standards accepted in commercial (fiscal) business.
A main goal of this tool is to facilitate end user practical interpretation of uncertainty
results without a very deep knowledge of statistical or mathematical concepts, but with
basics concepts to identify risks, make decisions and to have a reliable data for support
its daily operation.
Nomenclature
Xi Input
xi Input best estimate
u(xi) Standard uncertainty associated to xi of input Xi
uc(xi) Combined uncertainty
U Expanded uncertainty
k Coverage factor
u(xi,xj) Associated covariance of xi and xj of inputs Xi and Xj
r(xi,xj) Associated correlation coefficient to xi and yi
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 2 of 20
PDF Probability density function
a Difference between upper limit a+ and a-
ci Sensitivity coefficient
veff Degree of freedom effective
vi Degree of freedom
r Random number
Y Output
Vn Net Volume
Cd Discharge coefficient
GSV Gross volume
IV Indicated volume
CCF Combined correction factor
CTL Correction factor due temperature
CPL Correction factor due pressure
MF Meter factor
Pulses Accumulated pulses
KF K-Factor
Tm Measured liquid temperature
Pm Liquid pressure
RHO Observed density
Tobs Observed temperature
GUM Guide to the expression of uncertainty in measurement
MCM Monte Carlo method
ISO International Organization for Standardization
API American Petroleum Institute
MPMS Manual of Petroleum Measurement Standards
IP Petroleum Institute
VIM International Vocabulary of Metrology
IMNC Instituto Mexicano de Normalización y Certificación A.C.
OIML International Organization of Legal Metrology
NMX-CH-140-IMNC-2002 Mexican Standard “Guía para la expresión de la incertidumbre en las mediciones”
Introduction
Uncertainty evaluation has been considered as an activity only for calibration purposes
which require a deep knowledge in statistical and mathematical concepts. This situation
can be associated with absence of friendly computational tools that help end users to
determine, understand and apply in its process information related with uncertainty.
This paper describes methodology considered in GUM and Monte Carlo method in
order to validate a spreadsheet calculation and group information in friendly screens.
Some concepts directed related with uncertainty aren’t developed in a deep way, but its
reference documentation is indicated where it’s necessary.
However, its necessary to develop more robust applications, that have international
standards support for uncertainty evaluation, but also validate and guarantee reliability
of obtained results through alternative methodologies, due nowadays trends require a
deep knowledge of uncertainty associated to measurements, not only for quality system
requirements, but also to improve performance and guarantee competitiveness of
measurement process.
Methodology
Activities considered in scope of this job can be divided in next:
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 3 of 20
a. Definition of characteristics of flow measurement system
b. Uncertainty estimation of volume flow measurement system for liquids
hydrocarbons using GUM Methodology:
Variable specification and measurement procedure
Determination of estimate values for input quantities
Uncertainty sources quantification
Consideration of correlations between uncertainty components
Calculation of measurement results
Calculation of measurement result standard uncertainty
Definition of coverage factor
Measurement result report
c. Spreadsheet elaboration
d. GUM Methodology validation using Monte Carlo methodology
e. Definition of characteristics to include in software development
f. Flow diagrams development for programming
g. Software Development
h. Software validation
a. Definition of characteristics of flow measurement system
Type of volume flow meters for hydrocarbons selected for purposes if this job is turbine
meter. This was considered due is the type of meter of most applications for flow
measurement of liquids hydrocarbons.
After this selection, proceeds to define: operation and performance characteristics,
secondary instrumentation required for flow volume measurement and calculation
equations to determine volume quantity. This information was obtained from
international standards related to measurement of liquids hydrocarbons which is
indicate in next figure:
Figure 1. Standards related with flow hydrocarbons measurement
API MPMS 12.2 API MPMS 12.2.1
API MPMS 21.2
API MPMS 11.1 Volume X
API MPMS 11.2.2M
Volume
API MPMS 5.1 API MPMS 5.3
ISO 2715
Operation characteristics
Performance flow meters
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 4 of 20
A flow measurement system with turbine meter it’s covered by chapter 5, section 3 of
API MPMS, which recommends installation and operation topics of meter arrangement.
Its consider that flow measurement systems comply with recommendations established
referring to erection and installation of elements for flow conditioning, valves, filter
systems, protection dispositive for meter, accessories location, and others, assure an
adequate performance which no affects exactitude of measurement system. Points
considered for evaluation of uncertainty in a volume flow measurement system referring
to installation is instruments location for temperature, pressure and density
determination, due variations of this magnitudes most affect performance of turbine
meter and therefore define exactitude of measurements.
Algorithms and equations for flow volume calculation using turbine meter are defined
in chapter 12, section 2 and Appendix B of part 1 same chapter of API MPMS. This
standard contents support for algorithms and equations for base density, pressure
correction factors and liquid temperature determination. For determination of density
and correction factor for temperature at reference conditions IP Report 3 was used.
An electronic flow measurement system is considered in order to compensate in real
time effects of pressure and temperature, therefore determination of CTL and CPL is
made during flow measurement. For this, flow measurement system include a tertiary
measurement element (flow computer), which collect data from primary and secondary
elements, store it and calculate flow.
A flow measurement system generally consists in a paralleled meter arrangement where
total volume is obtained by sum of individual volumes. For purposes of this document,
an arrangement of not more than four turbine meter run and all instrumentation
(pressure, temperature and density) and equipment associated for flow measurement.
Total volume is defined by:
1 2 3 4NV Vn Vn Vn Vn
b. Uncertainty estimation of volume flow measurement system for liquids
hydrocarbons using GUM Methodology.
I. Variable specification and measurement procedure
For flow volume liquid hydrocarbon measurement, API MPMS establish calculation
procedures, measurement methods, calculation equations, operational recommendations
and practices. For this reason in this paper proposed models by API were considered
based on its experience and fundaments which assure exactitude and reliability of
measurements.
Therefore, for definition of mathematical model begins form equations established in
standards for flow volume calculation at reference conditions that can be described in
next figure:
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 5 of 20
N
KF RHO
Tobs
K0
K1
Tb
API MPMS 21.2
IP Paper No. 3
Tm
API MPMS 11.1 VOL XRHO20
RHO15
Tr
VIAPI MPMS 11.2.1M
A
B
C
D
F
Pm
API MPMS 12.2.1
CPLAPI MPMS12.2.1CCFAPI MPMS12.2.1
Vn
MF
Value calculated
Constant values
Data operation
IP Paper No. 3
CTL
Figure 2. Volume at reference conditions for turbine meter
According this figure, input magnitudes which define Net Volume are:
N = Pulse emitted by turbine during transference
KF = K - Factor
MF = Meter factor
Tm = Fluid temperature
Pm = Fluid pressure
RHO = Fluid density at observed temperature Tobs
Tobs = Observed temperature
K0, K1 = Constants according fluid density
A, B, C, D = Constants
Once identified input magnitudes, an analysis in order to identify some other factor that
affects measurement results was developed.
Pulses, N
Since some factors as: electromagnetic interferences, energy supply variances, noise,
wiring problems, among others, can promote false pulses which can cause errors in
measurement results.
K-Factor, KF
When in the mathematic model for net volume flow calculation is affected by meter
factor, MF, so K-Factor, KF, doesn’t affects net volume uncertainty, it mean, a constant
value without variation possibilities is considered.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 6 of 20
Meter factor, MF
As mentioned above, MF is used to correct volume registered by turbine meter to same
value of reference meter, and it’s obtained during meter calibration. For this, always it’s
important to consider calibration uncertainty and indicated in calibration report.
Sometimes, laboratories provide a calibration curve of MF, and if electronic system can
made calculation on real time, so, coefficients for calibration curve (which generally is
lineal) can be loaded, to correct MF according flow measured. It’s important to consider
this equation has an additional uncertainty source: adjust graph uncertainty due residual
errors that have to be considered.
Temperature
Exact temperature determination its essential to determine flow volume of liquids
hydrocarbons at reference conditions. Temperature measurement generally is affected
by instrument resolution, temperature gradient, stability and instrument calibration.
First factor to consider is uncertainty obtained from calibration of measurement
instrument. Always it’s necessary to consider in order to evaluate uncertainty of any
measurement obtained with this instrument.
According NMX-CH-140-IMNC-2002, sometimes and under some circumstances
corrections of systematic errors are not applied to results of measurements, nevertheless,
it’s a practice that has to prevent and consider it during uncertainty estimation. One way
to estimate it is increasing uncertainty associate to result, it mean, to expanded
uncertainty in correspond measurement have to add arithmetically maximum error not
corrected. This result in an increased uncertainty which compensate correction doesn’t
make. Also, it’s common that measurements of secondary instruments associated to
flow measurement (temperature, pressure and density) don’t be affected by corrections
of systematic errors obtained during calibration process; again it’s important to consider
this during calibration uncertainty process.
One of difficulties founded during any measurement process is the limited resolution of
instrument, which causes no exact knowledge of measurement result. Usually,
resolution its not considered as uncertainty source due it was included in instrument
calibration process, but some authors (23) have determine that its correct consider
resolution as uncertainty source. As well, GUM identifies finite resolution of instrument
as source of measurement uncertainty.
Objective of determine temperature of fluid measured, to correction of thermal effects
in liquid, is obtain an exact temperature of liquid inside body of meter. Generally,
temperature sensor can’t be installed in a flow meter for constructive characteristics, so
according API MPMS 7.2 recommends install a temperature sensor downstream turbine
meter. But, strictly, under this conditions a temperature gradient exists due temperature
determination and volume metering are obtained in different points. In other hand, some
factor relates with installation of temperature sensor can increase that gradient caused
by an incorrect insertion length, bad heat transfer in instrument or bad lecture of fluid
temperature.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 7 of 20
All measurement instruments change its characteristics along time; therefore it’s
important to consider effects of time and frequency of usage, it mean, stability. Some of
tools used to determine stability are control charts, which show variation between
controls limits established.
Pressure
Effect of pressure change in flow volume for liquids hydrocarbons is less than effect
due to temperature; nevertheless, it’s important to consider effect of magnitudes that
affect pressure determination, which are similar with those defined for temperature.
As well as temperature, it’s important to consider uncertainty of calibration of pressure
meter, in order to evaluate uncertainty of any measurement obtained with this
instrument. Also, its necessary increase uncertainty if corrections to systematic error
aren’t made.
Its necessary to considerer also, resolution of pressure instrument, as measurement of
limit knowledge of pressure value.
Performance of pressure instrument is affected by incorrect installation which can cause
a pressure gradient, for example pressure sensor and meter location which can cause an
increase o decrement of pressure value. Sometimes it’s difficult to assure location of
sensor and meter, for what this pressure gradient shall be considered. Pressure meter
also is affected by stability, due to time and frequency time usage.
Density
Density of liquid at base conditions have to be known exactly to calculate correction
factors CTL and CPL, due errors caused by an incorrect density measurement will have
a considerable effect in CPL and CTL.
If density is obtained, for example using a densimeter, so all magnitudes that affect this
instrument have to be considered.
Models for RHOb and F determination
Others magnitudes that are important to consider are the associated with mathematical
model for determination of RHOb and F, which are determined by calculation
procedures established in API MPMS. Chapter 11, section 2 part 1M, established
compressibility factor for liquid hydrocarbons can be estimated form equation:
2 2
C DTA BT
RHO15 RHO15F e [1]
Uncertainty associated to this mathematic model is 6.5% F, with a confidence level of
95%.
II. Determination of estimated values for inputs
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 8 of 20
For flow volume metering systems of liquid hydrocarbons, estimated values of inputs
referring to number of pulses, temperature, pressure and density aren’t obtained form
repeated observations, simply values of magnitudes obtained from operation of systems
are considered, it mean, field operational conditions. And, estimated value for MF is
determined from measured flow and on basis information located in calibration report.
III. Uncertainty sources quantification
Uncertainty components are evaluated considering type B method, because none one of
them are obtained form a series of repetitions; they are obtained form only one
observation because it belongs to a dynamic metering system characterized by
variations of pressure, temperature, density and, as result, flow volume or similarly
change of properties of fluid being measured.
According described above, quantification of uncertainty source consists in
determination of its distribution, PDF, depending available information. Once PDF is
assigned, standard uncertainty can be determined, which is calculated basis on type of
distribution selected and considering next criteria:
Available Information Probability Density Function Standard Uncertainty
The quoted uncertainty is given as
a multiple of a standard deviation
Coverage factor
Gaussian
i
Uu x
k
It may be possible to estimate only
the bounds (upper and lower limits)
to state that the probability that the
value of Xi lies within the interval
a- to a+.
There may not be enough
information available.
Rectangular
12i
a au x
Values near the bounds are less
likely than those near the midpoint. Triangular
6i
au x
Values near the bounds are less
likely than those near the midpoint. Trapezoidal
1
6i
au x
Table 1. Probability functions
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 9 of 20
IV. Consideration of correlations between uncertainty components
According defined characteristics of metering system and for a typical installation with
multiple meters, some of the uncertainty components are considered correlated in next
situations:
When meters are calibrated using same reference meter
Frequently, for a metering system composed by multiple meter runs, calibration of
metering instruments is developed using only a reference meter for each magnitude. In
this way, correlation depends only by reference meter used for calibration.
When only a meter (temperature, pressure, density or observed temperature) is
used to determine magnitude and only this found value is used for calculation of
corrected volume of all flow meters installed.
This situation generally happens because exist only one density meter and observed
temperature for whole metering system, so calculation for base density and flow volume
are made using same values for each meter run.
When same equation is used for calculation associated with each meter.
Mathematical models used to estimate base density and compressibility factor are
established in API Chapter 11 and are used in same way for flow volume calculation in
each meter run.
If equations for calculation of flow volume indicate in API Chapter 11 are observed, it
can be determined a correlation between CTL and CPL, due a third magnitude, RHO,
affect both of them. Nevertheless it’s recommended, in those situations where
applicable, redefine mathematical model in order to eliminate correlation due method
described in GUM for management of uncertainty components correlated is complex.
V. Calculation of measurement result
As was described above, net volume of metering system is determined adding
individual volumes so:
5
1
N i i
i
V VI CCF
Where:
ii
i
NVI
KF
i i i iCCF MF CTL CPL
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 10 of 20
VI. Standard uncertainty determination of measurement result
In order to estimate standard uncertainty of net volume, it’s necessary to combine
standard uncertainties of inputs estimations:
4 3 4
2
1 1 1
2 ,C N i i i j i j
i i j i
u V c u x c c u x x
Sensitivity coefficients Ci, are determined by means of partial derivative of VN with
regard to each one of influence magnitudes defined as:
i
i
fc
x
Sometimes sensitivity coefficients can be difficult to obtain using algebraic methods,
mainly when mathematical models are complex. In these cases, alternative
methodologies which consider other techniques can be used, for example numerical
analysis.
ISO 5168 propose a numerical technique for sensitivity coefficients determination,
which quantify effect of a small change in input variable xi, on result value y, keeping
constant other variables. So, sensitivity coefficients are calculated according next
expression:
i
i
yc
x
Its recommended increasing value used will be as small as possible, and not larger
than uncertainty of parameter xi. Process can initiate with a value equal to
uncertainty in xi and reduce progressively until Ci value corresponds to a result between
established tolerances.
In this way, to determine sensitivity coefficient for CvarN, first net volume is calculated,
VN using N value, and its recalculated using N + ΔN, where ΔN is a small increased
value in N. Result can be expressed as VN + ΔVN, where ΔVN is the incremented valued
caused by ΔN.
Figure 3. VN Variation
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 11 of 20
Once standard uncertainties of uncertainty components, sensitivity coefficients and
estimated covariance are obtained, it can be determined standard uncertainty of result,
Net Volume, applying next:
4 3 4
2
1 1 1
2 ,C N i i i j i j
i i j i
u V c u x c c u x x
VII. Coverage factor definition
Standard uncertainty Uc obtained previously generally implies to content a true value
with a probability p of 68% more or less, but usually a better probability is desired so a
coverage factor k is used to expand this interval to an upper confidence level.
In order to obtain a rigorous estimation of expanded uncertainty of the flow volume
meter system, degrees of freedom for each one of uncertainty components defined was
obtained using: 2
1
2
i
i
i
u xv
u x
Basis on available information, and finally obtain expanded uncertainty
c p effU u t v
Results
VIII. Measuring result report
The way result of a measurement process is expressed is VN = VN +/- U with
corresponding units. Also its included relative expanded uncertainty and k factor used to
obtain U.
Generally, number of significant digits used in uncertainty expression is one according
literature. Besides, number of significant digits of flow volume has to be consistent with
same number of uncertainty value.
c. Spreadsheet elaboration
As first stage in this paper, a spreadsheet was developed to register al required
information for flow volume calculation and for associated uncertainty evaluation:
models, equations, etc. This was developed in order to define required models and
equations, apply GUM methodology, validate if its adequate use of this methodology
for this specific application, define parameters to use in software and as previous
support to develop software.
Preliminary spreadsheet to evaluate uncertainty measurement of metering systems for
liquid hydrocarbons, was developed in excel, following recommendations established in
GUM.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 12 of 20
d. Validation of GUM methodology through Monte Carlo
For uncertainty evaluation its recommendable applies uncertainty propagation law
(methodology used by GUM) as well as Monte Carlo method, and later to compare both
results. If comparison is favorable, uncertainty propagation law can be used later to
solve similar problems. Otherwise, Monte Carlo simulation can be applied.
As was mentioned previously, Monte Carlo method (MCM) permits combine using
numerical simulation PDF of inputs defined to flow volume metering systems and
obtain a PDF of result, it mean: volume.
Procedure considered during MCM applying for uncertainty evaluation (distribution
propagation) conforms Supplement 1 of GUM and it’s described next:
Select number of Monte Carlo tests, M, to be made.
For each defined input (influence magnitude) for flow volume metering system,
generate M random numbers according PDF assigned and considering this
criteria:
Available information PDF Uncertainty Standard
baR ,
Lower and upper limits a, b Rectangular
raba
baT ,
Lower limit 21 aaa
Upper limit 21 bbb
Triangular
212
rrab
a
,,baTrap
Lower limit 21 aaa
Upper limit 21 bbb
ababab 2211
Trapezoidal
21 112
rrab
a
Best estimate x
Associated standard uncertainty u(x) Gaussian
zxux
Table 2. Random number generation
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 13 of 20
For each generated vector (group of each random data of each input), flow
volume is obtained according mathematical model defined. By this way, M
volume values are obtained.
Organize M volume values in descendent order
Estimate y and its standard deviation associated u(y), it means, average and
deviation of M volume values, respectively.
Calculate coverage interval for a determinate coverage probability p (confidence
level). One option to obtain this is using position measurement of position,
percentile. In this way, variable values under a determined percentage can de
obtained.
In order to compare both techniques, steps indicate in Supplement 1 of GUM were
followed:
Apply GUM methodology (uncertainty propagation law) and obtain y +/- Up
in a coverage interval 100 p% for result value.
Apply Monte Carlo simulation and obtain standard uncertainty u(y) associated
to estimate value of result value and limits Ylow and Yhigh of coverage interval
100p% for result value.
Finally, determine if coverage interval obtained using GUM methodology and MCM
are between established tolerances, it mean, if absolute differences of limits of both
coverage intervals aren’t higher than established tolerance, so use of GUM methodology
is valid for this application.
e. Definition of characteristics to include in software
Once spreadsheet for uncertainty evaluation is developed, not only characteristics that
software have to comply are defined in order to evaluate uncertainty according GUM
methodology, but also characteristics of a tool to obtain in a practical manner an
uncertainty of a flow volume metering system for liquids hydrocarbons. By this way
next was defined:
As generally a metering system include several meter runs, it was decided the
software can evaluate uncertainty of a system conformed by maximum four
meter runs.
When an equation is provided in calibration report to determine MF at different
flows generally isn’t indicate associated uncertainty of meter model. For this
reason, software will have capability to obtain calibration curve and its
associated uncertainty.
According type of available information regards variability of influence
magnitudes, a distribution function will be associated, but with option to select
other type of PDF.
Availability to input process data and uncertainties in different units according
influence magnitude.
As mathematical model considered for evaluation of volume uncertainty is
defined basis on applicable standards, an option to recalculate volume at
reference conditions (20 °C) and correction factors will be include in order to
determine if an error exists in calculation.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 14 of 20
Graphs to show uncertainty components that have major contribution to total
uncertainty of volume metering system are included.
f. Flowcharts for programming
Once characteristics to include were defined, flowcharts were developed about
procedures for uncertainty evaluation in order to programming tasks were simple and
understandable. Next figure shows main diagram and functionalities to include in
software for uncertainty evaluation:
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 15 of 20
Measurement System
Measurement system configuration
Name
System description
Date
§ Fluid
Number de meters §
Data operation
§
Enter Input data: N, MF, KF, Tm, Pm,
RHO, Tobs.
Computer Volume to 20°C
§
Uncertainty Evaluation
for each Individual meter
§
Identification of the uncertainty sources
Determination of the nature of the errors
Assume probability distributions
Obtain standard uncertatinty of each component
Determine the sensitivity coefficients
Combine the numerical values to give a numerical
value for the uncertainty
Calculation of effective degrees of freedom
Definition of coverage factor
Expanded Uncertainty
§
Uncertainty Budget
for each meter
Overall Uncertainty Evaluation
Measurement System
Uncertainty Budget
Calibration results
§
Volume, temperature, pressure and density
§
§
Uncertainty budget
§
§
Determination of correlations between uncertainty
contributions
Determine the combined standard uncertainty
Definition of coverage factor
Give an expanded Uncertainty
§
§ Generar presupuesto de incertidumbre
§
Start
End
Figure 4. Software functionalities diagram
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 16 of 20
g. Software Development
Initial stage for software development was to select programming platform, which main
consideration was programming language was simple, mainly due to number of
mathematical model to programming. Visual Basic was selected due is a relatively easy
to learn and use programming language, because of its graphical development features
Visual Basic can create executables (EXE files), and also is used to develop Windows
applications and to interface database systems
Also, it was decided to use a database due quantity of variables and parameters to
handle. Option to store generated information regarding uncertainty evaluation to
recover later if it’s required for post analysis or similar.
Once platform was selected for software development, a task related with windows
creation starts according design characteristics defined previously, following this to
implement algorithms defined.
Final task of software development was its validation / verification. This was developed
using spreadsheet as reference to software. The fact to validate spreadsheet calculation
using Monte Carlo method, alternative methods to sensitivity coefficient determination,
review of generated codes (functions, algorithms and parameters) and calculation
validation of correction factor for liquid flow volume using proposed examples in
different standards, guarantee reliability and quality of obtained results.
By this way if software generated results correspond to spreadsheet results for testing
exercises, can be affirmed software developed is adequate for uncertainty evaluation of
volume metering systems for liquid hydrocarbons.
Some screens of software are showed:
Figure 5. Input data screen and volume calculate at 20°C
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 17 of 20
Figure 6. Uncertainty sources selecting and its associated distribution screen
Figure 6. Calibration curve and its associated uncertainty screen
Figure 7. Uncertainty Budget associated to meter run 1 of the metering system
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 18 of 20
Figure 8. Components contribution graph for uncertainty determination
Conclusion
In Mexico, end users of flow liquid hydrocarbon metering systems aren’t yet
familiarize with measurement uncertainty concept and, mainly, with its importance to
determine and know uncertainty of their measurements. Nevertheless, quality systems
requirements and related standards will require offer better guarantee of their
measurements results, therefore they will have to validate, assure traceability of their
measurements to national and international standards and indicate associated
uncertainty of each one.
There is a lot of information related with uncertainty evaluation, however its is
presented in a complicated language which results that any person not familiarize with
metrology language can’t assimilate this important information being this data difficult
to access to improve a measurement process. For the reason above mentioned
importance of generation of knowledge and tools as described in this paper, which
facilities metrology application in all aspects of our life and promote use of uncertainty
when a measurement concept is declared, as a data that indicate quality of measurement
reported.
This software will facilitate evaluation process of flow volume metering systems of
liquid hydrocarbon due only require to input system operational data, choice from la list
sources which contribute to uncertainty according characteristics of metering systems
and finally to select and input available information related to variability of each
component. Once this information is loaded, software can automatically determine
uncertainty of metering system and report components which have more participation in
measurement uncertainty. These results can be useful to visualize fulfillment of
requirements and established tolerances for operation of metering system, and also as
metrological control mechanism of measurements which permits minimize, in some
cases, effect of a specific component in a total uncertainty of metering system.
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 19 of 20
In one hand, analysis made for uncertainty evaluation of flow volume metering systems
which is base for this software development, guarantee reliability of results. In other
hand, MCM application to validate GUM methodology for uncertainty evaluation in
this specific case, permits guarantee reliability results obtained with this software and
can be use to demonstrate quality of measurement result in a effective way.
References
1. International Organization for Standardization. Guide to the Expression of
Uncertainty in Measurement. Switzerland : ISO, 1993. GUM.
2. IMNC. NMX-CH-140-IMNC-2002. Guía para la expresión de incertidumbre en las
mediciones. 2003.
3. JCGM. International vocabulary of metrology - Basic and general concepts and
associated termns (VIM). s.l. : JCGM, 2008. JCGM 200:2008.
4. EUROLAB. Guide to the Evaluation of Measuremente Uncertainty for Quantitative
Test Results. Paris, France : EUROLAB, 2006. Technical Report No. 1.
5. European co-opeation for Accreditation. Expression of the Uncertainty of
Measurement in Calibration. s.l. : EA, 1999. EA-4/02.
6. Metodología para el cálculo de incertidumbre. Moreno, J. Angel. México :
CENAM, 2005.
7. Cox, M. G. y Harris, P. M. Uncertainty Evaluation. United Kingdom : National
Physical Labotory, 2004.
8. Aplicación de la guía ISO/TAG4/WG3/GUM para la estimación de incertidumbre en
la mediicón de flujo de hidrocarburos. Arias Romero, Roberto, Alvarez Vargas,
Rogelio y Negrete García, Salvador. México : Congreso Internacional de Ductos,
2001.
9. International Organization of Legal Metrology. International Recommendation,
Dynamic measuring systems for liquids other than wate. Dynamic measuring systems
for liquids other than water, Part 1 Metrological and technical requirements. Paris,
France : OIML, 2007. Vol. Part 1 Metrological and technical requirements, OIML R
117-1. OIML R 117-1.
10. American Petroleum Institute. Manual of Petroleum Measurement Standards,
Chapter 5 - Metering, Section 3 - Measurement of Liquid Hydrocarbons of Turbine
Meters. Washington, D.C. : API, 1995. API MPMS 5.3.
11. —. Manual of Petroleum Measurement Standards, Chapter 12- Calculation of
Petroleum Quantities using Dynamic Measurement Methods and Volumetric Correction
Factors, Section 2 - Electronic Liquid Volume Measurement using Positive
Displacement and Turbine Meters. Washington, D.C. : API, 1995. API MPMS 12.2.1.
12. Comité de Normalización de Petróleos Mexicanos y Organismos Subsidiarios.
Medición ultrasónica para hidrocarburos fase líquida. s.l. : PEMEX, 2009. NRF-240-
PEMEX-2008.
13. International Organization for Standardization. Measurement of fluid flow -
Procedures for the evaluation of uncertainties. Switzerland : ISO, 2005. ISO 5168.
14. Influence of the Temperature, Pressure and Viscosity on the oil Measurement with
Turbine Type Measurers. Araújo Damasceno, Magalí, y otros. Brazil : Brazilian
Archives of Biology and Technology, 2006.
15. American Petroleum Institute. Manual of Petroleum Measurement Standards,
Chapter 21- Flow Measurement Using Electronic Metering Systems, Section 2-
Software for evaluation of uncertainty in liquid hydrocarbons flow measurement systems Page 20 of 20
Electronic Liquid Volume Measurement using Positive Displacement and Turbine
Meters. Washington, D.C. : API, 1998. API MPMS 21.2.
16. Why Spreadsheets are Inadecuate for Uncertainty Analysis. Castrup, Suzanne.
Lancaster, CA : s.n., 2004|.
17. A Comprehensive Comparison of Uncertainty Analysis Tools . Castrup, Suzanne.
Anahemim, CA : Measurement Science Conference, 2004.
18. Teknologisk Institute. GUM Workbench. [En línea] Danish Techology Institute, 8
de June de 2005. http://www.gum.dk.
19. Uncertainty Manager. [En línea] ValiTrace GmbH, 2008.
http://uncertaintymanager.com.
20. Integrated Sciences Group. [En línea] Integrated Sciences Group.
http://www.isgmax.comm.
21. Colorado Engineering Experiment Station Inc. [En línea] Richard Estabroock, Mayo
de 2007. http://www.ceesi.com.
22. ema, CENAM. Guía técnica sobre trazabilidad e incertidumbre en metrología
dimensional. 2008.
23. Schmid, Wolfgang A. La simulación numércia como una herramienta en la
estimación de la incertidumbre de medición. Querétaro, México : CENAM.
24. American Petroleum Institute. Manual of Petroleum Measurement Standards,
Chapter 7 -Temperature Determination, Section 2 - Dynamic Tempeature
Determination. Washington, D.C. : API, 1995. API MPMS 7.2.
25. Kegel, Tomas M. Uncertainty Analysis of Meter Volume Measurements - Part 3.
s.l. : Colorado Engineering Experiment Station (CEESI).
26. Cox, M. G. y Harris, P. M. Uncertainty Evaluation. United Kingdow : National
Physical Laboratory, 2004.
27. Metas & Metrólogos Asociados. Evaluación de Incertidumbres con Método Monte
Carlo en Excel. Cd. Guzmán, Jalisco : La Guía MetAs, 2007.
28. Joint Committee for Guides In Metrology. Evaluation of measurement data -
Supplement 1 to the "Guide to the expression of uncertainty in measurement" -
Propagation of distributions using a Monte Carlo Method. s.l. : JCGM, 2008. JCGM
101:2008.
29. ema, CENAM. Guía técnica sobre trazabilidad e incertidumbre en la calibración de
medidores de flujo de líquidos empleando como referencia un patrón volumétrico. s.l. :
ema, CENAM, 2008.
30. Schmid, Wolfgang A. y Lazos Martínez, Ruben J. Guía para estimar la
incertidumbre de la medición. Querétaro, México : CENAM, 2004.
31. entidad mexicana de acreditación. Trazabilidad e incertidumbre de mediciones.
México : ema, 2005.
32. Importancia de la metrologíaen la aplicación de la norma ISO 9000 y la guía 25
ISO/IEC. Torres Guzmán, Jorge C. y Cederborg Almeyda, Bruno. Querétaro,
México : CENAM, 1997.
33. Incertidumbre en la Medición. Subcomisión de Mediciones de Gas de la
Comisión de Mediciones y Extracciones de Muestra. Argentina : Petrotecnia, 2003,
Vol. Junio 2003.
34. Castelazo Sinencio, Ismael. Incertidumbre en las mediciones: Impactos
económicos y sociales. Querétaro, México : CENAM.
35. American Petroleum Institute. Manual of Petroleum Measurement Standards,
Chapter 11.1 Volume Correction Factors. Washinton D.C. : API, 1980. Vols. Volume X
- Background, Development, and Program Documentation, Tables 5B. API Standard
2540.
Page 1 of 26
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Realistic Pipe Prover Volume Uncertainty
Paul Martin, IMASS (formerly Smith Rea Energy Limited).
ABSTRACT
Traceability for liquid turbine metering systems is generally achieved via a calibrated pipe prover volume used to verify the meter K-factor in situ. This paper demonstrates how incompatibility between an arbitrary tolerance set for the calibration of a pipe prover and the achievable uncertainty in measurement when determining the prover volume can lead to practices which may result in measurement bias.
This paper presents a robust method of estimating the pipe prover volume uncertainty determined using the master meter/master pipe prover calibration method. The individual uncertainty components used in the estimate, and the method of combining them, are included along with a comparison of the gravimetric and volumetric calibration methods for determining the compact prover volume. The traceability chain relating to the calibration of the pipe prover and the importance of accreditation for measurements are also discussed.
The paper concludes by examining the tolerances in place in the North Sea and how the practices which have evolved to meet the tolerances may compromise good metrology and lead to measurement bias. 1. INTRODUCTION
Fiscal oil metering stations typically consist of two or more meter runs with turbine metering and associated secondary instrumentation such as temperature, pressure, density and sampling systems to allow the fluid properties and liquid composition to be determined. The oil measurement system is also usually equipped with a permanently installed pipe prover or small volume prover used as a volume calibration reference for the turbine meters.
When the pipe prover volume is too great to perform a volumetric water-draw, to directly compare the volume of the pipe prover and volumetric standard measure (or proving tank), calibration is generally carried out by the master meter/master pipe prover method. Using a compact prover as the master pipe prover, the calibrated volume of the compact prover is used to calibrate the master meter which is then used as a transfer standard for determining the volume of the (permanently installed) pipe prover. This calibration method is undertaken with the compact prover, master meter and pipe prover connected in series and allows both the master meter and pipe prover to be calibrated on operating fluids at the same pressure, temperature and flowrate.
Figure 1. Master Meter/Master Prover Method
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 2 of 26
This paper investigates the pipe prover volume uncertainty determined using the master meter/master pipe prover calibration method and also compares the volumetric and gravimetric water-draw methods for compact prover volume calibration.
In the UK sector of the North Sea the regulatory DECC guidelines [1] on pipe prover volume calibration define the requirements for repeatability as ±0.01% with a year on year tolerance band as 0.02% and highlight that the operator must seek approval before any shift in excess of 0.02% is accepted. 2. TERMS AND DEFINITIONS
2.1 General Abbreviations
API American Petroleum Institute BIPM International Bureau of Weights and Measures (translation from French) CIPM International Committee for Weights and Measures (translation from French) CTE Coefficient of Thermal Expansion DECC Department of Energy and Climate Change GUM Guide to the Expression of Uncertainty in Measurement ISO International Organisation for Standardisation (translation from French) MPMS Manual of Petroleum Measurement Standard NASA National Aeronautics and Space Administration NML National Measurement Laboratories NMS National Measurement System SI International System of Units (translation from French) SVP Small Volume Prover TUR Test Uncertainty Ratio UK United Kingdom UKAS United Kingdom Accreditation Service VIM Vocabulary of International Metrology
2.2 Pipe Proving Terms
A short list of pipe proving terms specific to this paper is given below, although complete lists of precise and rigorous definitions are listed within ISO 7278 [2] and API MPMS Chapter 4 [3]:
Calibrated Volume – also known as the ‘Base Volume’ of a pipe prover between detectors or calibrated standard measure or volumetric prover tank at standard conditions. Compact Prover – typically a small volume prover with a piston displacer installed in a precision bored cylinder. Detectors – optical sensors or electronic switches placed at either end of the calibrated volume section of the prover and actuated by the displacer to start and stop the pulse counters. Displacer – generic name for the sphere or piston used to sweep the calibrated pipe prover volume. K-Factor – number of pulses generated by a meter in relation to the volume passed. Master Meter – a meter that serves as the reference for the proving of another meter or pipe volume. Pass – a single movement of the displacer between detectors. Pipe Prover – the generic name for provers either conventional pipe provers or small volume provers in which a sphere or piston is displaced to measure the passed volume. Pulse Interpolation – technique to enhance the meter pulse count resolution, typically ‘double chronometry’. Run – a set of passes deemed necessary to derive a single K-factor suitable for reporting.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 3 of 26
Small Volume Prover (SVP) – a pipe prover producing less than 10,000 meter pulses per pass although capable of achieving the required repeatability and accuracy due to installation of high precision detectors and use of pulse interpolation techniques. Water Draw – term for the operation of calibrating a pipe prover with water into a volumetric or gravimetric tank.
2.3 Uncertainty of Measurement Terms The precise and rigorous definitions of the following terms are defined within the ISO documents, International Vocabulary of Basic and General Terms in Metrology (VIM) [4] and Guide to the Expression of Uncertainty in Measurement (GUM) [5]:
Combined Standard Uncertainty – standard uncertainty of the result of the combination of standard uncertainty components. Covariance – measure of mutually dependent uncertainties where correlations among the input estimates affect the combined standard uncertainty of the output estimate. Coverage Factor – numerical factor used to multiply the combined standard uncertainty to give the expanded uncertainty at a specified level of confidence. Expanded Uncertainty – an interval about the measurement result that may be expected to encompass a large fraction of the distribution of values that could reasonably be attributed to the measurand. Measurand – particular quantity subject to measurement. Measurement Accuracy – closeness of the agreement between a measurement result and the true value. As the true value is not known, accuracy is a qualitative term only (not quantitative). Precision – the closeness of agreement between independent test results obtained under stipulated conditions. Repeatability – precision under conditions where the results of successive measurements of the same measurand are carried out under the same conditions of measurement within short intervals of time. Reproducibility – precision under conditions where test results are obtained with the same method on identical test items under changed conditions of measurement over a longer interval of time. Probability Distribution – a function giving the probability that the random variable takes any given value or belongs to a set of values. i.e. Gaussian (normal), rectangular (uniform), triangular, etc. Sensitivity Coefficient – the differential change in the output value generated by the differential change in one input value divided by the change in that input. Standard Uncertainty – the uncertainty of the result of a measurement expressed as a standard deviation. Tolerance – the limiting or permitted range of values of a defined quantity. Type A – evaluation of uncertainty of measured values by statistical methods. Type B – evaluation of uncertainty by means other than statistical analysis Uncertainty of Measurement – parameter associated with the result of a measurement that characterises the dispersion of the values that could reasonably be attributed to the measurand. 3. PIPE PROVERS
3.1 History
The first pipe provers were used in the early 1950’s [6]. One of the first pipe provers was the ‘mile of pipe’ which used the predetermined volume of the pipe length and also tracked the position of a tightly fitted piston down the pipe flowing full of oil to increase the measurement accuracy of flow
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 4 of 26
meters. Pipe provers were first documented in the API Standards 1101 (1960) [7] for Positive Displacement Meters and later in API Standards 2531 (1963) [8] for Mechanical Displacement Meter Provers.
Pipe provers operate by displacing a known volume of liquid within a calibrated section of pipe. Repeatable displacement of fluids is achieved by an oversized sphere or piston travelling through the pipe between detectors. The first conventional pipe provers standardised by API achieved the required measurement resolution of 0.01% pulse resolution by generating no less than 10,000 meter pulses during a proving pass. The pipe prover design is such that the full flow through the metering stream being proved will pass through the pipe prover.
In 1967, the Apollo manned space program exhibited a need for precision test equipment to test altitude control rocket motors as part of the NASA test program [9]. The requirement to calibrate small flow meters to an accuracy of ±0.05% was met by a small manufacturer of flow meters who developed a Small Volume Prover (SVP) device which utilised an electronic pulse-counting technique know as ‘double chronometry’. It was the late 1970’s before modification and testing of this initial SVP design was conducted with the aim of moving the design from the laboratory into the field. The initial design used magnetic reed switches and a compressed air actuator system which comprised an air compressor and tank storage unit. These limiting components were replaced with high accuracy optical switches and a nitrogen/hydraulic system respectively. A further development was the introduction of an Invar rod with low coefficient of expansion on which the optical switches are attached and spring loaded to allow more accurate measurement of volume. These modifications made to the SVP created a portable device for use in industry similar to the present day compact prover. 3.2 Operating Principle
Pipe provers are an important part of a turbine meter fiscal oil metering station and are used to calibrate the meter K-factor periodically when flowrates or conditions change. The pipe prover is used to ‘prove’ the accuracy and repeatability of flow meters on actual operating fluids by measuring the volume of fluids passing through the meter in relation to the number of pulses generated by the meter.
The basic operating principle of the pipe prover is to meter the fluids swept by the displacer through a calibrated volume of pipe by counting the number of meter pulses between the start and stop detectors at either end of the calibrated volume. The displacer, either a piston or sphere, actuates the start and stop detectors and is designed to form a sliding seal which moves at the same rate as the flowing liquid. Temperature and pressure corrections are required to convert the calibrated volume at standard conditions to process conditions. The volume indicated by the meter is compared to the calibrated volume to determine a meter K-factor. Generally, meter K-factor calibration must achieve repeatability over five successive runs to within a band of 0.1% to meet the UK regulatory requirement for overall dry mass uncertainty of ±0.25% at the fiscal oil metering station.
International standards ISO 7278 parts 1 to 4 [2] and American standards API MPMS Chapter 4 sections 1 to 9 [3] are the current publications standardising pipe prover design and operation.
3.2.1 Conventional Pipe Prover
The conventional positive displacement pipe prover generates no fewer than 10,000 pulses for each proving pass to achieve a measurement resolution of 0.01% as defined in the API MPMS standards [3]. Conventional pipe provers can be constructed in a number of configurations such as uni-directional or bi-directional pipe provers with piston or sphere displacers.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 5 of 26
The bi-directional spheroid pipe prover shown in Figure 2 utilises a four-way diverter valve to direct the fluid flow in both directions along the calibrated section of pipe and allow measurement in both directions during a proving run. Bi-directional pipe provers require 20,000 pulses for each prover round trip (ie. 10,000 pulses for each pass).
Figure 2. Bi-directional Spheroid Pipe Prover Design
The uni-directional pipe prover channels fluid in only one direction through the calibrated section of the pipe prover. The meter pulses are recorded when the displacer travels in one direction with the flow across the calibrated section of pipe. Uni-directional pipe provers also require a method of returning the displacer to its starting position.
The displacement systems in a pipe prover are either an oversized sphere which forms a seal and travel with the flow or a sealing piston within the pipe. The design of a pipe prover with sphere displacer must incorporate chambers to launch and receive the sphere. Spheroid pipe prover design is more common as the sphere can move through bends in a pipe. Uni-directional piston pipe prover design uses a piston and poppet valve to allow the fluid to pass when the piston is returning to the start position.
Since conventional pipe provers are generally large permanent constructions, traceability for the volume of the calibrated section of pipe cannot be obtained in a laboratory so must, therefore, be calibrated using a transfer standard such as a volumetric proving tank or against a master prover and master meter.
3.2.2 Small Volume Prover
As small volume provers do not have sufficient volume to generate 10,000 unaltered meter pulses, pulse interpolation techniques are employed to increase the meter pulse count resolution and can therefore operate with less than 10,000 pulses as defined in the API MPMS standards [3]. However, pulse interpolation must achieve the required resolution of 0.01%. Pulse interpolation techniques interpolate fractional meter pulses or mathematically interpolate partial pulses with the most widely used method being double chronometry.
Small volume prover design incorporates a precision bore cylinder; a displacer with means of positioning and launching the displacer upstream of the calibrated section; displacer detectors that allow fluid flow while the displacer is travelling and temperature and pressure measurement devices with meter pulse counting instrumentation (timer, counters and pulse interpolation).
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 6 of 26
Figure 3. Uni-directional Compact Prover Design
The ‘Compact Prover’ is the name given to a typical style of mobile small volume prover. The basic design features are shown in Figure 3, and are comprised of a uni-directional pipe prover with piston and poppet valve (or flow-through valve).
Due to the compact size and portability of the small volume prover, the calibrated volume section of pipe within the precision bore chamber can be verified in a laboratory using a water draw procedure. 4. UNCERTAINTY OF MEASUREMENT
4.1 Uncertainty Evaluation
The result of a measurement is only an estimate of the value of a measurand, which means that any measurement result is only complete once a statement of uncertainty accompanies the result. Uncertainty of measurement is the doubt that exists regarding the result of a measurement and defines the level of confidence in that particular measurement. Measurement uncertainty can be estimated by quantifying the possible spread of measurements and provides the range of dispersion of results that can be reasonably associated with the measured value. Estimation of the measurement uncertainty is useful to help understand the parameters affecting the measurement; helps to define good quality measurements and allows meaningful comparison of results. A measurement result is only complete once a statement of uncertainty accompanies it.
Measurement uncertainty evaluation involves the use of a mathematical model for measurement and statistical techniques to determine the uncertainty associated with the best estimate of the value of the measurand. Each quantity significantly influencing the measurand value within the model also has prescribed uncertainties which must be accounted for within the uncertainty evaluation.
The GUM and several other documents stemming from GUM provide guidance on uncertainty evaluation. GUM enables measurements to be compared between different laboratories and provides a common approach for estimating measurement uncertainty. The GUM uncertainty framework has become the internationally accepted method for uncertainty calculation since its initial publication in 1993 but it should be noted that GUM is a guide and not a standard.
The main stages of uncertainty evaluation are given as follows:
(i) Define the output quantity Y, to be measured. (ii) Determine all input quantities Xi on which the output quantity Y depends.
(iii) Develop the mathematical model relating the input quantities Xi to the output quantity, Y = f (X1, X2,…, XN).
(iv) The values determined for the input and output quantities are defined as x1, x2,…, xN (input estimates) and y (output estimates), respectively.
(v) Assign probability density functions (PDF) to the values of the input estimates xi.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 7 of 26
(vi) Evaluate the standard uncertainties u(xi) by determining the estimated standard deviations for the input estimates, either by statistical means (Type A evaluation of standard uncertainty) or by other means (Type B evaluation of standard uncertainty).
(vii) Evaluate the covariance of input estimates that are correlated. For each pair i, j for which the values of Xi and Xj are mutually dependent, calculate the estimated correlation coefficient u(xi, xj) associated with xi and xj.
(viii) Calculate the model sensitivity coefficients ci by forming partial derivatives �f/�xi describing how the output estimate y varies with changes in the values of the input estimates xi.
(ix) Calculate the best estimate y of the output quantity value by evaluating the model using the input estimates xi.
(x) Determine the combined standard uncertainty uc(y) of the output estimate by combining u(xi), u(xi, xj ) and the model sensitivity coefficients ci. The combined standard uncertainty uc(y) is the positive square root of the combined variance u2
c(y) obtained from:
),()(2)()(1
1
1 1
222ji
N
i
N
i
N
ijijiiic yxrxuccxucyu � � �
=
−
= +=+=
(xi) Calculate � (effective degrees of freedom) associated with u(y) using the Welch–Satterthwaite formula (GUM G.4).
(xii) Multiply the combined standard uncertainty uc(y) by a coverage factor k to obtain the expanded uncertainty U = k uc(y). The coverage factor k is chosen on the basis of the level of confidence required of the interval. Coverage factor would normally be k = 2, giving a confidence level of approximately 95%.
(xiii) Express the result of the measurement as y ± U stating the level of confidence in the interval. Partial Derivatives
The sensitivity coefficient ci describes how the output estimate y varies with changes in the values of the input estimates xi. Using the analytical method the sensitivity coefficient ci can be obtained by partial differentiation.
ii x
yc
∂∂=
The analytical method involves differentiating the output parameter with respect to each of the input parameters in turn. The input values are then substituted into the resultant functions and each answer provides the sensitivity coefficients appropriate to each input parameter. This is the most mathematically correct method for evaluating sensitivity coefficients.
Finite Difference
If analytical determination of the partial derivatives is complicated where the mathematical model is complex, the actual derivative can be approximated by ‘Finite Difference’, which provides a robust method for uncertainty evaluation. The partial derivative �y/�xi can be approximated numerically by the finite difference expression [10],
i
iii
ii x
xyxxq
xy
xy
∆−∆+
=∆∆≈
∂∂
where the value �xi is as small as is practical, initially applying an increment which is equal to the value of uncertainty in the parameter xi.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 8 of 26
Where the mathematical model relates the input estimates xi to the output quantity, y = f (x1, x2,…, xN), the value of variation ui(y) can be taken as Zi (GUM 5.1), with corresponding sensitivity coefficient ci as Zi/u(xi).
)]),...,(,...,()),...,(,...,([21
11 NiiNiii xxuxxfxxuxxfZ −−+=
Combined standard uncertainty uc(y) can be defined numerically as:
),()(1 1
2ji
N
i
N
jjic yxrZZyu � �
= ==
Monte Carlo Simulation
Monte Carlo simulation involves making a large number of calculations of the output estimate y, by assigning different values to each of the input estimates xi. Each input value is generated at random from the distribution for each input quantity to form the corresponding value and distribution of the output quantity. The Monte Carlo method is a numerical approach to uncertainty evaluation and is defined in a supplement [11] to GUM published in 2006. 4.2 Traceability
The VIM defines traceability as ‘the property of the result of a measurement or the value of a standard whereby it can be related to stated references, usually national or international standards, through an unbroken chain of comparisons all having stated uncertainties’. The traceability chain involves the calibration of a measurement artefact or measurement equipment against a reference standard of greater accuracy.
Le Système International d’Unités (International System of Units), commonly referred to as the SI System is a universally adopted self consistent international system of measurement. The SI system base units are the meter, kilogram, second, ampere, Kelvin, candela and mole, respectively, for length, mass, time, electric current, thermodynamic temperature, luminous intensity and amount of substance.
National Measurement Laboratories (NML) hold the national primary measurement standard derived from the internationally recognised standard. The SI base unit for mass is the kilogram (kg) and the international standard artefact is held by the Bureau International des Poids et Mesures (BIPM, International Bureau of Weights and Measures) at Sèvres in France. The artefact is a cylinder of iridium alloy which is the primary measurement standard for mass and is the only remaining artefact as all other SI base unit standards are derived from physical properties such as the wavelength of light in vacuo which can be reproduced in laboratories throughout the world.
Most countries around the world have National Measurement Laboratories which hold national measurement standards to ensure confidence and accuracy of measurement. A National Measurement System (NMS) is also maintained and forms the technical and organisational infrastructure that ensures a consistent and internationally recognised basis for measurement. Calibrations undertaken against the national or regional standards form the traceability chain which links back to the SI base units. The NMS ensures accuracy and traceability of measurement for use in trade, industry, academia and government. For any measurement it should be possible to demonstrate traceability to international standards via an unbroken chain of calibrations. 4.3 Measurement Accreditation
Measurement standards in the UK are managed via traceability through United Kingdom Accreditation Service (UKAS) accredited laboratories. Measurement accreditation is important as it
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 9 of 26
ensures standards, accuracy and consistency of measurement which enables consumers to compare products for sale and make informed decisions and also facilitate trade to Europe and worldwide.
Accreditation assessment is conducted by an assessor with the expertise to cover the scope of accreditation. Accreditation is verified on an annual basis by surveillance visits (audits), with a full reassessment every four years. Those laboratories which have been assessed and approved by UKAS as meeting the requirements of ISO/IEC 17025 [12] may be granted UKAS accreditation. The Laboratories meeting the ISO/IEC 17025 standard requirements for calibration and testing activities also comply with the relevant requirements of the ISO 9001 [13] standard.
The five essential requirements for measurement accreditation backed by an efficient measurement audit are (1) Staff, (2) Equipment, (3) Accommodation and Environment, (4) Documentation (Quality Manual and Measurement Procedures) and (5) Traceability.
5. UNCERTAINTY EVALUATION EXAMPLES
The five stages (gravimetric water draw) or six stages (volumetric water draw) of the traceability chain that links pipe prover volume calibration to a standard weight set are shown in Figure 4. The standard weight shown here in stage 1 is deemed to be a company standard weight although there are obviously a few stages prior to this when tracing back to regional, national and international standards.
Figure 4. Pipe Prover Volume Uncertainty Traceability Chain
In stage 1, the standard weight set is calibrated in a UKAS accredited laboratory against a regional mass standard traceable to the national measurement system. Stages 2 and 3 are conducted simultaneously with the standard weights being used to calibrate the electronic weighing instrument, know as a mass comparator, by comparative methods along with the calibration of the compact prover volume by water draw. The calibration of the weighing instrument and compact prover volume is conducted in the laboratory under controlled conditions. The repeatability of 5 calibration runs should lie within a band of 0.02%. In Stage 4, the compact prover calibrated section of pipe is used to calibrate the turbine master meter k-factor. The repeatability of 5 calibration runs should again lie within a band 0.02%. Stage 5 is the final stage in which the pipe prover volume is calibrated by using the master meter. Using the K-factor obtained for the master meter, the number of pulses counted between detectors on the pipe prover calibrated section can be converted to volume. The repeatability of 5 calibration runs should again lie within a band of 0.02%. Temperature and pressure corrections are used in the calibration of the compact prover, master meter and pipe prover volume to adjust the process conditions to standard conditions of 15°C and 1.01325 bara.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 10 of 26
The volumetric water draw traceability chain is also shown in Figure 4, although an additional stage is included whereby the volumetric standard measure is first calibrated (3a) using a mass comparator, then (3b) the standard measure is used to calibrate the compact prover. Stage 3 is broken into two which increases the number of stages to six.
An example uncertainty evaluation for pipe prover volume calibration is presented in this document. The uncertainty evaluations for the calibration of pipe prover volume using hydrocarbon fluids and water are compared. The uncertainty evaluation accounts for the correlated uncertainties that occur between successive stages. The uncertainty is evaluated numerically using the ‘Finite Difference’ method for sensitivity coefficient calculation as described in GUM.
Also presented in this document are the measurement uncertainty evaluations for compact prover volume calibration by gravimetric and volumetric methods. 5.1 Gravimetric Water Draw Compact Prover Volume Calibration
The procedure for gravimetric water draw calibration is to displace the water volume between detectors from the compact prover at a controlled flowrate into a container located on an electronic weighing instrument. The switches operate a solenoid valve which is used to divert the volume of water between optical switches into the container. The mass of the water obtained from weighing is then divided by the density of the water to obtain the volume of water between switches.
In the example a 60 litre compact prover is calibrated by gravimetric water draw method. Temperature of the water is 16°C with a pressure of 5 barg in the compact prover.
Mathematical Model
The main components of the mathematical model are shown below; all other equations are given in the Appendix.
Gravimetric Water Draw Calibration: plppsptsp
tdwRb CCC
CVV
×××=
Where: Vb Compact prover base volume at 15°C and zero barg [14, 15, 16] VR Compact prover indicated volume at observed temperature and atmospheric
pressure Ctdw Correction factor for the thermal expansion of water Ctsp Correction factor for the effect of temperature on the steel of the prover Cpsp Correction factor for the effect of pressure on the steel of the prover Cplp Correction factor for the effect of pressure on liquid at the prover
Uncertainty of Weighing, U(M)
Uncertainty in the balance weighing method U(M) gives an expanded uncertainty [k=2] of ±2g. This uncertainty is the result of a weighing conducted by substitution method using an F1 weight set calibrated and traceable to national standards by a mass comparator weighing instrument with resolution of 0.05g. For the purpose of this uncertainty example the weighing uncertainty evaluation is not presented, although guidance on gravimetric water draw uncertainty is given in PD ISO/TR 20461 [17]. Guidance on the uncertainty of mass comparator balance measurement can also be found in OIML D28 [18], OIML R111-1 [19] and OIML R111-2 [20]. The uncertainty components to consider in relation to the calibration of the weighing instrument are: calibration of standard weight; drift of standard weight; comparator linearity; repeatability; buoyancy correction; drift of standard; indicator resolution; temperature sensitivity; eccentricity and other influencing factors.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 11 of 26
Uncertainty of Temperature Measurement, U(T)
The combined expanded uncertainty of the temperature measurement has been estimated as U(T) = ±0.13°C, assuming rectangular distribution. Scale resolution is 0.1°C and it is expected that any instruments under the same conditions will differ at any time by greater than ±0.1°C. All temperature indicators and probes are checked pre and post calibration by a 2 point spot check to ensure no drift in temperature measurement between instruments. The uncertainty is regarded as a constant value throughout the measuring range. Temperature calibration uncertainty has been deemed as negligible due to the covariance that exists between the temperature measurement instruments calibrated from the same reference.
Uncertainty of Pressure Measurement, U(P)
The combined expanded uncertainty of the pressure measurement has been estimated as U(P) = ±0.13 barg, assuming rectangular distribution. Scale resolution is 0.1 barg and it is expected that any instruments under the same conditions will not differ at any time by greater than ±0.1 barg. The expanded uncertainty is regarded as a constant value throughout the pressure measuring range. Pressure calibration uncertainty has been deemed as negligible due to the covariance that exists between pressure measurement instruments calibrated from the same reference.
Uncertainty of the air buoyancy correction, U(BC)
The combined uncertainty within the buoyancy correction U(BC) is calculated from the air buoyancy correction functional model by applying uncertainty estimates to the individual input values. The uncertainty components associated with buoyancy correction are: uncertainty in the density of air U(�a) calculated using the CIPM air density formula [21]; uncertainty in the density of reference weights U(�sm); uncertainty in the density of the water sample U(�w). Uncertainties are given below:
Uncertainty of sample density measurement, U(�w)
The expanded uncertainty in the sample density (water) is U(�w) = ±0.0468 kg/m³ [k=2]. The uncertainty is evaluated separately for water density measurement by high precision density meter.
Uncertainty of air density measurement, U(�a)
The expanded uncertainty in the air density measurement U(�a) is the combined uncertainty of the laboratory air density variation and the air density calculation uncertainty given in section C.6.3.6 of the OIML R 111-1 [19]. The sensitivity coefficients are provided as partial derivatives for the uncertainty components for air humidity U(H), air temperature U(T) and air pressure U(B) within the CIPM air density formula [21], although further uncertainty for draught, electrical interference and temperature variations must be considered. U(�a) = ±0.0031kg/m³.
Uncertainty of reference weights density, U(�sm)
The expanded uncertainty of the reference weights density of U(�sm) should be known from the calibration certificate U(�sm) = ±600 kg/m³. Standard weights density limits are specified within OIML R 111-1 [19].
Uncertainty of Water Compressibility, U(Fw)
The expanded uncertainty of the water compressibility is stated by Kell [22] as being U(Fw) = ± 0.0003 x 10-6 bar-1, assuming rectangular distribution.
Uncertainty of Steel Area Expansion Coefficient, U(Gcp)
The expanded uncertainty of the area coefficient of thermal expansion for steel is set as a default value of U(Gcp) = ±10%, assuming a rectangular distribution.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 12 of 26
Uncertainty of Invar Linear Expansion Coefficient, U(GI) The expanded uncertainty for the Invar linear thermal expansion coefficient is set as a default value of U(GI) = ±10%, assuming a rectangular distribution.
Uncertainty of Young’s Modulus of Elasticity, U(Ecp)
The expanded uncertainty for the Young’s Modulus of steel is set as a default value of U(Ecp) = ±10%, assuming rectangular distribution.
Uncertainty of Internal Diameter Measurement, U(IDcp)
The expanded uncertainty of internal diameter measurement of the flow tube calibrated section is U(IDcp) = ±1mm, assuming rectangular distribution.
Uncertainty of Wall Thickness Measurement, U(WTcp)
The expanded uncertainty of flow tube wall thickness measurement is U(WTcp) = ±1mm, assuming rectangular distribution.
Repeatability of Calibration Process, U(R)
For five successive calibration runs the maximum acceptable repeatability must be within a band of 0.02%. U(R) = ±0.01 % (0.02%), assuming rectangular distribution.
Uncertainty of Optical Switching Unit, U(SR)
Filling of the container is controlled by two optical detectors and a solenoid valve. Both switches are repeatable to within ±0.013mm. The uncertainty is proportional to the volume of a cylinder, U(SR) = ±0.0014 litres, assuming rectangular distribution.
Uncertainty of Ctdw Correction, U(Ctdw)
The combined uncertainty U(Ctdw) is calculated from functional model Ctdw = �1/ �2, correction factor for the thermal expansion of water. Uncertainty estimates are applied to each of the individual input values. The density of water in the container U(�1) and the density of water in the compact prover U(�2) are both calculated using the Tanaka [23] equation, where the expanded uncertainty of the water density calculation is given as ±0.00084 kg/m³. The terms in the Ctdw correction are correlated; therefore the covariance must be evaluated. Since the models present in the Ctdw correction are equal and the values of the input quantities are almost equal in magnitude and uncertainty, the correlation coefficient r(�1, �2) may be considered to be 1.
Uncertainty of Ctsp Correction, U(Ctsp)
The combined uncertainty U(Ctsp) is calculated from the functional model, Ctsp correction for the effect of temperature on prover steel, by applying uncertainty estimates to the individual input values. Since the Ctsp functional relationship is well known it may be assumed that the model uncertainty is confined within the uncertainty of the thermal expansion coefficients, therefore no model uncertainty is applied for the Ctsp correction factor.
Uncertainty of Cpsp Correction, U(Cpsp)
The combined uncertainty U(Cpsp) is calculated from the functional model, Cpsp correction for the effect of pressure on prover steel, by applying uncertainty estimates to the individual input values. Since the Cpsp functional relationship is well known it may be assumed that the model uncertainty is confined within the uncertainty of the modulus of elasticity, therefore no model uncertainty is applied for the Cpsp correction factor.
Uncertainty of Cplp Correction, U(Cplp)
The combined uncertainty U(Cplp) is calculated from the functional model, Cplp correction for the effect of pressure on the liquid at the prover, by applying uncertainty estimates to the individual input
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 13 of 26
values. Since the Cplp functional relationship is well known it may be assumed that the model uncertainty is confined within the uncertainty of water compressibility, therefore no model uncertainty is applied for the Cplp correction factor.
Table 1. Compact Prover Volume Uncertainty (Gravimetric Water Draw Method)
Uncertainty of Compact Prover Volume, U(Vbm-grav)
Table 1 shows the result of the compact prover gravimetric water draw volume calibration uncertainty evaluation. The relative expanded uncertainty U(Vbm-grav)= ±0.0134%. 5.2 Volumetric Water Draw Compact Prover Volume Calibration
The procedure for gravimetric water draw calibration is to displace the water volume between detectors from the compact prover at a controlled flowrate into volumetric prover.
Mathematical Model
The main components of the mathematical model are shown below; all other equations are given in the Appendix.
Volumetric Water Draw Calibration: plppsptsp
tsttdwRb CCC
CCVV
××××
=
Where: Vb Compact prover base volume at 15°C and zero barg [14, 15, 16] VR Compact prover indicated volume at observed conditions Ctdw Correction factor for the thermal expansion of water Ctst Correction factor for the thermal expansion of proving tank metal Ctsp Correction factor for the effect of temperature on the steel of the prover Cpsp Correction factor for the effect of pressure on the steel of the prover Cplp Correction factor for the effect of pressure on liquid at the prover
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 14 of 26
Unless stated below, all uncertainty estimates are identical to those in section 5.1.
Uncertainty of Proving Tank Volume, U(Vm)
The expanded uncertainty in the proving tank volume is taken from the certificate of calibration from a regional measurement laboratory, U(Vm)= ±0.01% [k=2].
Uncertainty of Volume Expansion Coefficient, U(Gm)
The expanded uncertainty of the coefficient of volume thermal expansion for proving tank metal is set as a default value of U(Gm) = ±10%, assuming a rectangular distribution.
Uncertainty in Scale Reading, U(RD)
The expanded uncertainty of volume reading is calculated from the scale reading resolution which has a neck gauge with 1 millilitre increments. U(RD) = ±1ml, assuming a rectangular distribution.
Uncertainty in Wetting Variance, U(W)
The wetting and drip variance is influenced by the liquid properties, construction of the proving tank, drip time and method. The expanded uncertainty of the wetting variance of the proving tank metal is U(W) = ±0.001%, assuming a rectangular distribution.
Uncertainty of Ctsm Correction, U(Ctsm)
The combined uncertainty U(Ctsm) is calculated from the functional model, Ccts correction for the effect of temperature on volumetric prover steel, by applying uncertainty estimates to the individual input values. As the Ctsm functional relationship is well known it may be assumed that the model uncertainty is confined within the uncertainty of the modulus of elasticity, therefore no model uncertainty is applied for the Ctsm correction factor.
Table 2. Compact Prover Volume Uncertainty (Volumetric Water Draw Method)
Uncertainty of Compact Prover Volume, U(Vbm-vol)
Table 2 shows the result of the compact prover volumetric water draw calibration uncertainty evaluation. The relative expanded uncertainty U(Vbm-vol)= ±0.0159%.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 15 of 26
It is worth noting that any comparison between the uncertainty of the volumetric and gravimetric water-draw methods must be linked by traceability chain to the same mass reference. In the previous example, the volumetric uncertainty of the standard measure is ±0.01% [k=2] as determined by a regional measurement laboratory; therefore no comparison of uncertainty is undertaken. 5.3 Pipe Prover Volume Calibration on Water
The procedure for pipe prover calibration by master meter/master prover method is to displace a known water volume between detectors of the compact prover at a controlled flowrate and calibrate the master meter K-factor whilst simultaneously displacing the volume between the pipe prover detectors. Using the K-factor obtained for the master meter, the number of pulses counted between detectors on the pipe prover calibrated section can be converted to volume.
Uncertainty due to non-linearity and repeatability on master meter is negligible since the meter and the pipe prover are calibrated consecutively at the same stable conditions.
In the example a 60 litre compact prover and turbine master meter are used to calibrate a 1776 litre pipe prover with water as the calibration fluid. The master meter generates 23,137 meter pulses per pass and the temperature of the water is 14°C at pressure of 7 barg within the system.
Mathematical Model
The main components of the mathematical model are shown below; all other equations are given in the Appendix.
Pipe Prover (by Master Meter): ���
����
�
××××××
=CplpCtlpCpspCtspK
CplmCtlmNV
mb
2
Master Meter (by Compact Prover): ���
����
�
××××××
=CplcpCtlcpCpscpCtscpV
CplmCtlmNK
bmm
1
Where: Vb Pipe prover base volume at 15°C and 0 barg [15, 16, 24]
Vbm Master pipe prover base volume at 15°C and 0 barg Km K-factor at master meter N1 Meter pulses counted during meter K-factor calibration N2 Meter pulses counted during pipe prover calibration
Unless stated below, all uncertainty estimates are identical to those in section 5.1. Uncertainty of Compact Prover Volume, U(Vbm)
The uncertainty is taken from the result of the compact prover gravimetric water draw volume calibration uncertainty evaluation in section 5.1. The relative expanded uncertainty U(Vbm-grav)= ±0.0134 % [k=2].
Uncertainty of Water Density, U(�15)
The expanded uncertainty of the water standard density is taken as, U(�15) ±5.0 kg/m³. It is expected that any density measurement over the 5 passes will not differ at any time by greater than ±5.0 kg/m³, assuming rectangular distribution.
Uncertainty of Ctl Correction, U(Ctl)
The combined uncertainty U(Ctl) is calculated from the functional model Ctl = �t/�15, correction factor for the effect of temperature on the liquid, by applying uncertainty estimates to the individual input values. As the Ctl functional relationship is well known it may be assumed that the model uncertainty is confined within the uncertainty of water density calculation by Tanaka [23], therefore no model
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 16 of 26
uncertainty is applied for the Cplp correction factor. The expanded uncertainty of the water density calculation is given as ±0.00084 kg/m³.
The Ctl correction terms in both stages of the calibration K-factor and pipe prover volume) are correlated; therefore the covariance must be evaluated. As the models present in the Ctl correction are equal and the values of the input quantities are almost equal in magnitude and uncertainty, the correlation coefficient r(Ctlcp , Ctlm) and r(Ctlm , Ctlp) may be considered to be 1.
Table 3. Pipe Prover Volume Uncertainty on Water Uncertainty of Compact Prover Volume, U(Vb-water)
Table 3 shows the result of the water calibration pipe prover volume uncertainty evaluation. The relative expanded uncertainty U(Vb-w)= ±0.0190% [k=2]. 5.4 Pipe Prover Volume Calibration on Hydrocarbon Fluids
The procedure is identical to the method described in section 5.3 with the exception that hydrocarbon fluid is used as the calibration medium.
Uncertainty due to non-linearity and repeatability on the master meter is negligible since the meter and the pipe prover are calibrated consecutively at the same stable conditions.
In the example a 60 litre compact prover and turbine master meter are used to calibrate a 7051 litre pipe prover with hydrocarbon fluid as the calibration fluid. The master meter generates 105,338 meter pulses per pass and the temperature of the water is 18°C at pressure of 9 barg within the system.
Mathematical Model
The main components of the mathematical model are shown below; all other equations are given in the Appendix.
Pipe Prover (by Master Meter): ���
����
�
××××××
=CplpCtlpCpspCtspK
CplmCtlmNV
mb
2
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 17 of 26
Master Meter (by Compact Prover): ���
����
�
××××××=
CplcpCtlcpCpscpCtscpVCPLmCTLmN
Kbm
m1
Where: Vb Pipe prover base volume at 15°C and 0 barg [15, 16, 24]
Vbm Master pipe prover base volume at 15°C and 0 barg Km K-factor at master meter N1 Meter pulses counted during meter K-factor calibration N2 Meter pulses counted during pipe prover calibration
Unless stated below, all uncertainty estimates are identical to those in section 5.1.
Uncertainty of Hydrocarbon Fluid Density, U(�15)
The expanded uncertainty of the hydrocarbon fluid standard density is taken as, U(�15) ±10.0 kg/m³. It is expected that any density measurement over the 5 passes will not differ at any time by greater than ±10.0 kg/m³, assuming rectangular distribution. Uncertainty of Cpl Correction, U(Cpl)
The combined uncertainty U(Cpl) is taken from API MPMS 11.2.1M [25]. The Ctl model relative expanded uncertainty in volume is ±0.03% [k=2] up to 34.5 bar.
The Ctl correction terms in both stages of the calibration (K-factor and pipe prover volume) are correlated; therefore the covariance must be evaluated. As the models present in the Ctl correction are equal and the values of the input quantities are almost equal in magnitude and uncertainty, the correlation coefficient r(Ctlcp , Ctlm) and r(Ctlm , Ctlp) may be considered to be 1.
Uncertainty of Ctl Correction, U(Ctl)
The combined uncertainty U(Ctl) is taken from API MPMS 11.1 [26]. The Ctl model uncertainty in volume is ±0.15% [k=2] up to 65°C.
The Cpl correction terms in both stages of the calibration (K-factor and pipe prover volume) are correlated; therefore the covariance must be evaluated. As the models present in the Ctl correction are equal and the values of the input quantities are almost equal in magnitude and uncertainty, the correlation coefficient r(Cplcp , Cplm) and r(Cplm ,Cplp) may be considered to be 1.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 18 of 26
Table 4. Pipe Prover Volume Uncertainty on Hydrocarbon Fluid
Uncertainty of Compact Prover Volume, U(Vb-hc)
Table 4 shows the result of the hydrocarbon fluid calibration pipe prover volume uncertainty evaluation. The relative expanded uncertainty U(Vb-hc)= ±0.0296% [k=2].
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 19 of 26
Table 5. Pipe Prover Volume Uncertainty on Hydrocarbon Fluid (High Temperature)
Uncertainty of Compact Prover Volume, U(Vb-hc.ht)
Table 5 shows the result of the hydrocarbon fluid calibration pipe prover volume uncertainty evaluation. The relative expanded uncertainty U(Vb-hc.ht)= ±0.0396% [k=2], at Temperature 65°C.
When a higher temperature of hydrocarbon fluids is observed during calibration the sensitivity of the cubical thermal expansion coefficient of steel at the pipe prover increases significantly. Pressure has also been increased in the example although the pressure sensitivities are not significantly increased. The increase in area expansion coefficient of steel at the pipe prover highlights that the uncertainty must reduce to maintain an uncertainty in the order of ±0.03% of volume. 6. HISTORIC CALIBRATION RECORDS
The results of 198 pipe prover calibrations from 14 locations have been used to compile the histogram shown in Figure 5. From the calibration results two standard deviations are calculated as ±0.042% of volume. The calibration data compiled in the histogram includes all results for both oil and water calibrations. It should also be noted that there has been no filtering of the data to remove volume shifts due to switch changes or equipment failures which may be related to the larger shifts in volume.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 20 of 26
Figure 5. Pipe Prover Volume Calibration Distribution
7. RESULTS
7.1 Compact Prover Volume Calibration
Gravimetric water-draw method: U(Vbm-grav)= ±0.013 %.
Volumetric water-draw method: U(Vbm-vol)= ±0.016 %.
No comparison of uncertainty can be undertaken as the volumetric and gravimetric water-draw uncertainty evaluations in this document are not linked by traceability chain to the same mass reference. The volumetric uncertainty of the standard measure is ±0.01% [k=2] as determined by a regional measurement laboratory.
The gravimetric and volumetric water-draw methods both have their advantages and disadvantages. The significant difference in the methods is that the gravimetric method is mainly confined to the laboratory whereas the volumetric method can be applied on site. The gravimetric calibration eliminates the possibility of drift by allowing continual calibration of the compact prover. Also, the gravimetric method reduces the number of steps in the traceability chain, therefore having a slightly better uncertainty. 7.2 Pipe Prover Calibration Pipe Prover Calibration with Water: U(Vb-w)= ±0.02% [k=2]. Pipe Prover Calibration with Hydrocarbon Fluid: U(Vb-hc)= ±0.03% [k=2] , 18°C and 9 bar.
Pipe Prover Calibration with Hydrocarbon Fluid: U(Vb-hc.ht)= ±0.04% [k=2], 65°C and 20 bar. The significant increase in uncertainty of volume with higher temperature calibrations is a result of the increase in the sensitivity of the cubical thermal expansion coefficient of the pipe prover steel. The usefulness of uncertainty evaluation can be seen from this result as it highlights the need for a reduction in the coefficient of thermal expansion (CTE) uncertainty to maintain a pipe prover volume uncertainty in the order of ±0.03%. In the example a default value of ±10% is used for CTE whereas a material certificate or material testing results may reduce the uncertainty estimate of CTE.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 21 of 26
8. DISCUSSION
Uncertainty Evaluations
From the uncertainty evaluations provided in this document the greatest sensitivity is noted for temperature measurement uncertainty, highlighting the need for accurate temperature measurement and stable operating conditions throughout the calibration.
The uncertainty evaluation for pipe prover volume calibration uncertainty with water U(Vb-w)= ±0.0190% [k=2], which suggests using water as the calibration fluid it may be possible to calibrate the pipe prover volume to ±0.02% as indicated within industry documentation [27, 28, 29, 30].
A difference in the uncertainty obtained between water and hydrocarbon fluid calibrations was also noted from the uncertainty evaluations. The difference is mainly due to greater sensitivity in relation to temperature measurement uncertainty which is greater in the hydrocarbon fluid calibration model.
In the uncertainty evaluations presented, it is clear that pipe prover cubical thermal expansion uncertainty is significantly affected by the increase in oil temperature highlighting the need for accurate determination of the coefficient of thermal expansion and hence a low uncertainty. Pipe Prover Volume Uncertainty
There have been a number of documents published, which provide a range of what is deemed to be a sensible or achievable level of uncertainty when determining a prover volume. Alan T.J. Hayward [27] indicates that prover volume calibration accuracy can be between ±0.05% and ±0.02%. API MPMS 4.9.1 [28] discusses the frequency of calibration of pipe provers and indicates a range from ±0.05% to ±0.02% on volume. API MPMS 4.1 [29] Table 3 provides some hypothetical uncertainty values within a hierarchy (traceability chain) for prover base volume as ±0.03%, calibrated using a field standard test measure with ±0.015% on volume. Another source of data was found within the NFOGM Handbook of Uncertainty Calculations [30] which uses a pipe prover base volume uncertainty of ±0.038% (0.011m³ in 28.646 m³) within its fiscal turbine meter station uncertainty calculations.
From the information above it is clear that pipe prover volume uncertainty is the direct result of measurement traceability and no single generic measurement uncertainty can therefore be used to define pipe prover volume uncertainty. The measurement method and uncertainty estimates for reference instruments within the traceability chain will influence the estimate of the combined measurement uncertainty. Pipe Prover Volume Calibration Tolerance
Within the UK sector of the North Sea pipe prover volume calibration repeatability limit is set as ±0.01% of the mean (or within a band of 0.02%) [1] for a set of 5 successive calibration runs. Also defined is the year to year volume shift tolerance requirement of ±0.02% from the previous year. Although the repeatability limit is fully achievable, the volume shift tolerance seems very narrow in relation to the uncertainty evaluations contained within this document and information presented in industry documentation [27, 28, 29, 30].
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 22 of 26
Figure 6. Tolerance Intervals
For a case where the best possible pipe prover volume measurement uncertainty is taken as ±0.02% and year to year tolerance interval is set as ±0.02%, no acceptance interval [31] is available to assess the measurement system conformance, see Figure 6. As the pipe prover volume uncertainty is equal to the tolerance, conformance to specification is not possible. In fact, in some instances, the pipe prover volume calibration uncertainty may be greater than the tolerance interval.
The measurement uncertainty should generally be smaller than the tolerance in order to ensure that the tolerance is met for a given measurand. ISO 12001-1: 1993 [32] suggests minimum requirements for test uncertainty ratio (TUR) of 3:1 as being acceptable for any measurement process, although some kind of compromise is required for pipe prover calibration as it is obvious that permanently fixed pipe provers cannot be calibrated in the laboratory.
The challenge arises when a series of measurements result in values which are scattered around the tolerance limit. We know that small changes in process and environmental conditions can have a significant effect so how do we apply a rigorous method of ensuring that the final result is both eliminating errors due to the bad measurement conditions whilst maintaining objectivity by not selecting the results and possibly introducing a bias.
Statistical methods for determining the true average calibration factor may in some cases provide improvements to the conventional proving method of five successive runs as varying proving results are normally due to variations in process conditions rather than the inherent repeatability of the meter.
Figure 7. Theoretical Calibration Results
Consider the case given in Figure 7 where the first set of calibration runs returns a shift calculated as –0.021%. As this value is outwith the tolerance band of ±0.02% a second set of calibration runs are undertaken to yield a result of –0.016% shift in volume. As the verification limit between the 1st and 2nd result is set to ±0.01%, this means the ‘as left’ calibration from the second set of runs is actually within the tolerance limit leaving considerable doubt in the measurement result.
From the information presented in this paper is may be possible to calibrate a pipe prover volume within the uncertainty limits but actually be outwith the tolerance specification. Widening the tolerance limit may be a more robust approach to pipe prover volume calibration.
Another point to consider is, why are the UK (±0.02%) [1] and NPD (±0.04%) [33] year to year pipe prover volume tolerance limits different for the same quality of measurement equipment ?
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 23 of 26
9. REFERENCES
[1] “DTI Guidance Notes for Petroleum Measurement”, DECC – Department of Energy and Climate Change (formerly DTI & BERR), Issue 7, 2003.
[2] ISO 7278, “Liquid Hydrocarbons – Dynamic Measurement – Proving Systems for Volumetric Meters”, Parts 1 to 4, International Organisation for Standardisation, Geneva.
[3] API MPMS Chapter 4, “Manual of Petroleum Measurement Standards Chapter 4 – Proving Systems”, Sections 1 to 9, American Petroleum Institute. Washington, D.C.
[4] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP and OIML, “International Vocabulary of Basic and General Terms in Metrology”, 2nd Edition, 1993.
[5] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP and OIML, “Guide to the Expression of Uncertainty in Measurement”, 2nd Edition, 1995.
[6] Hayward, Alan T. J., “Pipe Provers: a User’s Manual”, ISBN 0 521 33113 7, Cambridge University Press, 1991.
[7] API Standard 1101, “Measurement of Petroleum Liquid Hydrocarbons by Positive Displacement Meter”, American Petroleum Institute, Washington DC, 1st Edition 1960.
[8] API Standards 2531, “Manual of Petroleum Measurement Standards – Mechanical Displacement Meter Provers”, API, Washington DC, 1st Edition 1963.
[9] Graves, Frank D., “Operational Experience with Small Volume Provers”, International School of Hydrocarbon Measurement, 1989.
[10] Dunn, Patrick F., “Measurement and Data Analysis for Engineering and Science”, ISBN 0 07 282538 3, McGraw-Hill, 2005.
[11] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, and OIML, “Evaluation of measurement data – Supplement 1 to the Guide to the expression of uncertainty in measurement – Propagation of distributions using a Monte Carlo method”, 2006.
[12] ISO/IEC 17025: 2005, “General Requirements for the Competence of Testing and Calibration Laboratories”, International Organisation for Standardisation, Geneva, Switzerland, 2005.
[13] ISO 9001: 2008, “Quality management systems – Requirements”, International Organisation for Standardisation, Geneva, Switzerland, 2008.
[14] EN ISO 4267-2: 1995, “Petroleum and liquid petroleum products – Calculation of oil quantities – Part 2: Dynamic measurement”, ISO, Geneva, Switzerland, 1995.
[15] API MPMS Chapter 12.2.4, “Calculation of Base Prover Volumes by the Water Draw Method”, American Petroleum Institute. Washington, D.C, 1997.
[16] IP PMM Part X Section 3, “Code of Practice for the Design, Installation and Calibration of Pipe Provers”, Energy Institute, London, 1988.
[17] PD ISO/TR 20461:2001, “Determination of uncertainty for volume measurements made using the gravimetric method”, ISO, Geneva, Switzerland, 2001.
[18] OIML D 28, “Conventional value of the result of weighing in air”, International Organization of Legal Metrology, Edition 2004.
[19] OIML R111-1, “Weights of Classes E1, E2, F1, F2, M1, M1–2, M2, M2–3 and M3, Part 1 – Metrological and Technical Requirements”, International Organization of Legal Metrology, Edition 2004.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 24 of 26
[20] OIML R111-2, “Weights of Classes E1, E2, F1, F2, M1, M1–2, M2, M2–3 and M3, Part 1 – Test Report Format”, International Organization of Legal Metrology, Edition 2004.
[21] Picard et al, “Revised formula for the density of moist air (CIPM-2007)”, Metrologia 45 (2008) 149–155, February 2008.
[22] Kell, G. S., “Density, Thermal Expansivity, and Compression of Liquid Water from 0° to 150°C”, J. Chem. Eng. Data, 20(1), 97–105, 1975.
[23] Tanaka et al , “Recommended table for the density of water between 0 °C and 40 °C based on recent experimental reports”, Metrologia 38 No 4, 301-30.9, 2001.
[24] API MPMS Chapter 12 – Section 2 – Part 5, “Calculation of Base Prover Volumes by Master Meter Method, American Petroleum Institute. Washington, D.C, 2001.
[25] API MPMS Chapter 11.2.1M, “Compressibility Factors for Hydrocarbons: 638–1074 Kilogams per Cubic Meter Range”, 1984, American Petroleum Institute. Washington, D.C.
[26] API MPMS Chapter 11 – Section 1, “Temperature and Pressure Volume Correction Factors for Generalized Crude Oils, Refined Products, and Lubricating Oils”, 2004, American Petroleum Institute. Washington, D.C.
[27] Hayward, Alan T. J., “Flowmeters: A basic guide and source-book for users”, ISBN 0 333 21920 1, The MacMillan Press Ltd, 1979.
[28] API MPMS Chapter 4 – Section 9 – Part 1, “Introduction to the Determination of the Volume of Displacement and Tank Provers”, American Petroleum Institute, Washington, D.C., 2005
[29] API MPMS Chapter 4 – Section 1, “Determination of the Volume of Displacement and Tank Provers by the Waterdraw Method of Calibration”, American Petroleum Institute, Washington, D.C., 2005.
[30] NFOGM, NPD, CMR, NIF, “Handbook of Uncertainty Calculations – Fiscal Orifice Gas and Turbine Oil Metering Systems”, Revision 2, 2003.
[31] ISO 14253-1:1998, “Geometrical Product Specifications (GPS) – Inspection by measurement of workpieces and measuring equipment – Part 1: Decision rules for proving conformance or non-conformance with specifications”, ISO, Geneva, 1988.
[32] ISO 10012-1: 1993, “Quality assurance requirements for measuring equipment – Part 1: Metrological confirmation systems for measuring equipment, ISO, Geneva, 1993.
[33] NPD, “Regulations Relating to Measurement of Petroleum for Fiscal Purposes and for Calculation of CO2-Tax (The Measurement Regulations)”, Norwegian Petroleum Directorate, 2001.
[34] EN ISO 8222: 2002, “Petroleum measurement systems – Calibration – Temperature corrections for use when calibrating volumetric proving tanks”, ISO, Geneva, 2002.
[35] API MPMS Chapter 11.2.3M, “Water Calibration of Volumetric Provers”, American Petroleum Institute. Washington, D.C., 1984.
[36] Schoonover, R. M. & Jones, F.E., “Air Buoyancy Correction in High-Accuracy Weighing on Analytical Balances”, Analytical Chemistry, Vol. 53. pp 900-902, 1981.
[37] NORSOK I-105, “Fiscal measurement systems for hydrocarbon liquid”, NORSOK Standard I0501, Edition 3, Standards Norway, August 2007.
[38] API MPMS Chapter 13 – Section 2, “Methods of Evaluating Meter Proving Data”, American Petroleum Institute, Washington, D.C., 2006.
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 25 of 26
APPENDIX GRAVIMETRIC WATER DRAW:
2
1
ρρ=tdwC
( ) ( )[ ]TtTtCtsp −+−+= 211 γα
wtD
Ep
C psp ×+=1
[ ]pC pl ∆+= ).(1 β
���
����
� −
���
����
�−
=
ρρρρ
o
c
o
sx MM1
1
α Coefficient of linear expansion of Invar rod � Compressibility factor of water [22] γ Coefficient of area expansion of steel �p Differential pressure �1 Density of water at measure or mass comparator (observed conditions) [23] �2 Density of water in the compact prover (observed conditions) [23] � Density of the sample being weighed �c Density of reference weight �o Density of moist air [21] Ctdw Correction factor for the thermal expansion of water [34, 35] Cpl Correction factor for the effect of pressure on liquid [23] Cpsp Correction factor for the effect of pressure on the steel of prover [14, 15] Ctsp Correction factor for the effect of temperature on the steel of the prover [14, 15] D Internal pipe diameter E Modulus of elasticity Mx Object mass [18, 36] Ms Mass of the sample in air wt Wall thickness of pipe t Any temperature T Base temperature
VOLUMETRIC WATER DRAW:
( )TtCts −+= α31
α Co-efficient of linear expansion Cts Correction factor for the effect of temperature on the steel of the prover [14, 15] t Any temperature T Base temperature
27th International North Sea Flow Measurement Workshop 20 – 23 October 2009, Tønsberg, Norway
Page 26 of 26
PIPE PROVER WATER CALIBRATION:
vN
K =
( )TtCts −+= α31
)]t 0.8 (1 t [- xpeVV VCF TT
T
t
t
T ∆+∆=== ααρρ
T
ttl C
ρρ
=
α Co-efficient of linear expansion ρ Water density Ctl Correction factor for the effect of temperature on the liquid [26] Cts Correction factor for the effect of temperature on steel [14, 15] K K-factor [3] N Meter pulses t Any temperature T Base temperature v Unit volume VCF Volume correction factor [26]
PIPE PROVER OIL CALIBRATION:
))]Tt 0.8 (1 )Tt [- exp C TTtl −+−= (( αα
210
2K
KK
TTT ++=
ρρα
)(11
epl PPF
C−−
=
T.D
+
C + TB + A exp = F
2T
2T
���
����
�
ρρ.
α Coefficient of thermal expansion of the liquid [26] ρT Hydrocarbon liquid standard density A, B, C, D Constants Cpl Correction factor for the effect of pressure on liquid [26] Ctl Correction factor for the effect of temperature on liquid [26] F Compressibility factor of the hydrocarbon liquid [26] Kn Hydrocarbon liquid specific constants Pe Vapour pressure equilibrium Pm Pressure at the meter t Any temperature T Base temperature
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
1
Inferential Chemometric Allocation
Phillip Stockton, IMASS (formerly Smith Rea Energy Ltd)
1 INTRODUCTION
What is “Inferential Chemometric Allocation”?
Chemometrics is the science of extracting information from chemical systems by data-
driven means. The chemical data used in this approach to allocation are the
compositions of the feed and product streams in a commingled system. These are used
to infer an allocation of the product stream between two or more contributing streams
without measuring the flow of the feed streams.
A simple example will illustrate the concept. Consider two streams: Field A and Field
B, each of different compositions. The composition of each stream is known and
remains constant. These streams are mixed together and the commingled product’s
composition is measured as shown in Figure 1.
Figure 1 – Two Streams of known Composition Commingled
M
C
M
C
Field A
Field BMeter
Chromatograph
Commingled Stream
The commingled stream’s composition varies in accordance with the relative
contributions of each of the feed streams.
For example if Field A is 70 wt% methane and Field B 60% wt methane, then if the
measured composition has 65 wt% methane we can infer that half the feed can be
attributed to Field A and half to Field B. Similarly, if the commingled composition
was 67.5 wt% methane we would infer that three quarters of the flow is from Field A
and one quarter from B. The product composition offers a means of allocating to the
feed streams without the need for individual flow measurement of the contributing
fields.
In simple terms this is the basis of inferential chemometric allocation, i.e. it utilises
the composition of a commingled product stream to allocate between two or more
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
2
feed streams. This is a viable means of allocation if the feed streams’ compositions
are essentially constant or known and sufficiently dissimilar from one another.
However, a number of questions arise:
In the example, if a component different to methane had been selected would
the answer have been different?
How accurate is the method when subject to the uncertainties in the measured
compositions?
How dissimilar do the compositions need to be?
Is it possible to use this method to allocate between more than two fields?
Would we know if one of the assumed constant compositions starts to drift?
Or, in summary:
Is this a viable means of allocation in the real world?
This paper attempts to explore and answer these questions.
Section 2 presents various Inferential Chemometric Allocation methods and illustrates
the concepts using simplified examples.
As this is a novel allocation methodology and is in the development stage it has not
been implemented in any real systems. Hence in order to test its accuracy and
practicability it has been necessary to construct theoretical but representative data. In
addition, by utilising data from a real system in which the feed flows are known, it is
possible to test the methods in what may be considered real world scenarios for both
gas and liquid systems. Section 3 presents the results of this analysis and explores the
questions of allocating to more than two fields and detection of compositional drift.
Having stated all this, there is a final question to be answered:
Why would we want to do this?
Section 4 discusses possible applications and developments and Section 5 summarises
the conclusions.
2 CONCEPTS
2.1 Simple Compositional Based Split
Consider Example 1, which comprises two fields, Field A and Field B, of differing
compositions, being commingled in a 50:50 mixture. Imagine that the true
composition and flow data are known and the commingled stream is a perfect mixture
of the two fields. The true measurement data for this perfect system is presented in
Table 1.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
3
Table 1 – True Compositions and Flows of Example Perfect System Field A Field B Commingled
wt % wt % wt%
C1 60% 70% 65.0%
C2 30% 25% 27.5%
C3 10% 5% 7.5%
Flow (te) 500 500 1,000
If the true compositions of all streams but only the flow of the commingled stream
were known, we could still infer the flows of Field A and B by examining any of the
component weight fractions in the commingled stream.
More formally, the equation for calculating the contribution of Field A, based on the
weight fraction of component C1 in the three streams, is given by:
11
11
CC
CCA
BA
BCS (1)
The derivation of (1) is presented in Section 6.1. A similar equation can be written for
any of the components and these will give the same answer for SA if the data is
perfectly consistent.
However, in the real world the compositions and flows would not be known perfectly
and the compositional analyses would be subject to measurement uncertainties.
Applying randomly generated measurement errors to the compositions in Table 1, the
actual measured data may typically look like that presented in Table 2:
Table 2 - Compositions of Measured Streams Field A Field B Commingled Relative
Measurement
Uncertaintieswt % wt % wt% wt%
C1 59.89% 69.88% 64.93% ± 0.6 %
C2 30.01% 25.17% 27.42% ± 1 %
C3 10.09% 4.96% 7.65% ± 3 %
Using the above data it is possible to infer the split of the two Fields using Equation
(1). Based on the C1 weight fractions the contribution from (or split to) Field A is
calculated to be 49.58% and therefore 50.42% is from Field B. These are reasonably
close to the true split of 50%.
If we had chosen C2 to determine the Field A split, the answer would have been
slightly different and if C3 had been chosen different again; the calculated splits based
on each of the three components are presented in Table 3:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
4
Table 3 – Calculated Field Splits
Split Error Split Error
Perfect 50.00% 50.00%
C1 49.58% -0.42% 50.42% 0.42%
C2 46.49% -3.51% 53.51% 3.51%
C3 52.50% 2.50% 47.50% -2.50%
Field A Field B
Each component gives a slightly different answer and in a real system with 10+
components there will be many more choices. Which one is the best estimate?
C1 appears the obvious choice here, as it is the major component and its measurement
uncertainty in relative terms is the lowest. However, what would happen if the C1
content of the two Fields was closer?
Consider Example 2, which is the similar to Example 1, except that Field B’s true C1
content is changed to 61% and C3 to 14%. The inferred allocated quantities now
become:
Table 4 – True and Measured Compositions
Field C1 Content Similar
Field A Field B Commingled
Perfect Perfect Perfect
wt% wt% wt%
C1 60.0% 61.0% 60.5%
C2 30.0% 25.0% 27.5%
C3 10.0% 14.0% 12.0%
Field A Field B Commingled
wt% wt% wt%
C1 59.99% 61.36% 61.09%
C2 30.1% 25.0% 27.4%
C3 9.9% 13.7% 11.6%
True Values
Measured Values
Table 5 – Calculated Field Splits
Field C1 Content Similar
Split Error Split Error
Perfect 50.00% 50.00%
C1 19.84% -30.16% 80.16% 30.16%
C2 46.42% -3.58% 53.58% 3.58%
C3 56.08% 6.08% 43.92% -6.08%
Field A Field B
The inferred allocation based on C1 has produced a split to Field A that is grossly
below the true value, clearly a poor result, though the splits based on the other
components remain reasonable.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
5
If the concentrations of a component are similar in the two Fields then the uncertainty
in the results becomes large. The uncertainty in the calculated Field Split for
component “i” is given by1:
2
22
,
22
,
22
,
ii
iiiCiiiBiiiA
SABA
BAUACUBCUU (2)
Inspection of (2), reveals that the uncertainty in Field A’s split is inversely
proportional to the square of the difference of the component’s concentration in the
two Field streams. Indeed as Ai approached Bi, the uncertainty tends to infinity. Field
A’s calculated split uncertainty based on C1 is plotted against the C1 content in Field
B in Figure 2:
Figure 2 – Uncertainty in Calculated Field A Split as a Function of Field B’s C1
Content
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
20%
40.0% 45.0% 50.0% 55.0% 60.0% 65.0% 70.0% 75.0% 80.0%
Field B C1 Content (wt %)
Ab
so
lute
Un
ce
rta
inty
in
Sp
lit
At 60%, both Fields’ C1 contents are the same and the uncertainty becomes infinite.
2.2 Optimised Split
Is there a methodology, which utilises all the components and therefore maximises the
use of the data, that is not be subject to the uncertainty issues encountered with single
components, encountered in Example 2 above?
Consider a mass balance across component C1:
1111 *)1(* CCACAC CBSAS (3)
1 (2) is derived in accordance with the Propagation of Uncertainties described in the GUM [3].
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
6
“δ” is termed the residual of the mass balance. Similar balances may be written for all
components. As such, unless we have perfectly matching compositions, whatever
value we use for the Field split there will always be a violation in the mass balances of
some or all the components. Using the data in Example 1, the component mass
balances residuals are:
Table 6 – Mass Balance Residuals SA
Field A Field B Comm'ed δ
wt % wt % wt% wt%
C1 59.89% 69.88% 64.93% 0.0000%
C2 30.01% 25.17% 27.42% 0.1499%
C3 10.09% 4.96% 7.65% -0.1499%
49.58%
C1’s mass balance is satisfied but C2 and C3’s aren’t. Now is there a value of SA that
would minimise these mass balance residuals (which can be both positive and
negative)? The “Optimised Split” methodology, proposed in this paper, determines the
value of SA that produces the minimum of the sum of the squares of these residuals.
Figure 3 plots the sum of square of the residuals as a function of Field A split.
Figure 3 – Plot of Sum of Square of Residuals versus Field A Split
0
0.0005
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Split S
Su
m S
qu
are
Resid
ual
Σδ
2
The minimum value of the sum of square of the residuals occurs just below 0.5 and is
more evidently visible in the amplified version of the plot around this area presented
in Figure 4:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
7
Figure 4 – Plot of Sum of Square of Residuals versus Field A Split (Magnified)
0
0.000002
0.000004
0.000006
0.000008
0.00001
0.000012
0.000014
0.000016
0.000018
0.00002
47% 48% 49% 50% 51% 52% 53%
Split S
Su
m S
qu
are
Resid
ual
Σδ
2
Minimum occurs at
SA = 49.61%
This minimum value of can be found analytically by evaluating the derivative of Σδ
2
(with respect to with SA) when it is equal to zero and thereby the optimised value of
SA:
toNi
ii
toNi
iiiiii
ABA
CACABB
S
1
2
1
*)
(4)
Equation (4) is derived in Section 6.2.
At first sight this may appear to be a formidable looking equation. However, it may be
broken down into more tractable elements and is easily coded on a spreadsheet or in
software code. So applying Equation (4) to Example 1:
Table 7 – Calculation of Optimised Field Split Field A Field B Commingled Numerator Terms Denominator Terms
Cpt (i) Ai Bi Ci Bi*(Bi -Ai - Ci) + Ai*Ci (Ai - Bi)2
C1 59.89% 69.88% 64.93% 0.4940% 0.9963%
C2 30.01% 25.17% 27.42% 0.1091% 0.2348%
C3 10.09% 4.96% 7.65% 0.1385% 0.2638%
Sum 100.00% 100.00% 100.00% 0.7416% 1.4949%
49.61%Field Split (SA) = 0.7416% / 1.4949% =
As can be observed, the inferred split is slightly improved over the individual
component based splits. However, this is a single randomly generated snapshot of
fictitious data and a more rigorous analysis is required to compare the relative
accuracies of the various approaches and this is discussed with more meaningful data
in Section 3.3.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
8
2.3 Uncertainty Based Optimised Split
The Optimised Split takes advantage of all the data but it does not account for the
relative variation in its accuracy (or uncertainty). We might believe the C1 mass
balance residual has a lower relative uncertainty than the other components and
therefore should carry more weight in the minimisation. An alternative formulation,
based on the Optimised Split, has been developed, termed the “Uncertainty Based
Optimised Split”. In effect, this approach is a variation on uncertainty based allocation
(see [4], [5] and [6]) and employs principles utilised in data reconciliation (see [1] and
[2]).
Consider the uncertainty in each residual (δ in equation (3)), which may be calculated
using the Propagation of Uncertainties [3] from:
2
,
2
2
,
2
2
,
2
, iC
i
iiB
i
iiA
i
ii U
CU
BU
AU (5)
Residuals are then weighted according to their uncertainty. This is performed by
dividing δi by its uncertainty calculated according to (5). SA corresponding to the
minimum value of the sum of squares of these weighted residuals is then calculated.
The weighted sum of squares of the residuals (J) is given by:
toNi iCiBAiAA
iiAiA
toNi i
i
UUSUS
CBSAS
UJ
12
,
2
,
22
,
2
2
12
,
2
*1*
*)1(* (6)
The variation of J with SA is presented in Figure 5:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
9
Figure 5 – Plot of Weighted Sum of Square of Residuals versus Field A Split
-200
-150
-100
-50
0
50
100
150
200
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Split SA
J
-600
-400
-200
0
200
400
600
dJ/d
S
J dJ/dS
Again the optimum value of SA and can be calculated when the derivative of J with
respect to SA is zero. The equation that describes this is:
toNi
iCiBAiAA
iBAiAAiiAiA
iCiBAiAA
iiiiAiA
A
UUSUS
USUSCBSAS
UUSUS
BACBSAS
dS
dJ
1
22
,
2
,
22
,
2
2
,
2
,
2
2
,
2
,
22
,
2
*1*
*1*2**2**)1(*
*1*
**)1(**2
(7)
The derivative is also plotted on Figure 5 against the right hand axis and it equals zero
at the minimum value of J. The full derivation of (7) is presented in Section 6.3.
SA cannot be obtained directly from (7) and must be calculated using numerical
techniques such as binary chop, Newton Raphson, secant, etc. The Uncertainty Based
Optimised Split (SA) for Example 1 occurs at 50.12%. This is the closest inferred split
to the true value in this example. However, as mentioned Section 2.2 a more rigorous
analysis is required to compare the relative accuracies of the various approaches and
this is in the next section.
3 APPLICATION WITH REALISTIC DATA
3.1 Introduction
Having illustrated the concepts and the potential for these Chemometric Inferential
approaches to allocation in Section 2, we have only examined theoretical models with
a small number of components and only for a couple of cases – these were just
random snapshots of what is, in any case, fictitious data. In order to determine if the
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
10
approaches are viable means of allocation and compare their performance, more
meaningful data is required.
As this is a novel allocation methodology, in the development stage, it has not been
implemented in any real systems. However, the use of real data from system in which
the feed flows are also measured provides a test of how well the inferential methods
perform.
First the methods are tested using data obtained from a gas plant in Section 3.2 and
with some limited oil analysis data in Section 3.4.
In addition, to perform more exhaustive analysis in Section 3.3, the real data is also
utilised to generate a data set with which theoretical tests can be performed that we
might expect to encounter in further real systems hence the use of the term “Realistic”
in the title.
3.2 Actual Gas Plant Data
All feed streams and product streams (except removed CO2) are measured both in
terms of flow and composition2. Using the compositional data alone it is possible to
infer the split of the feed using the various inferential techniques described in Section
2 and compare these against the actual metered split of feed flows. Typical
compositions of the flows and compositions used are presented in Table 8.
Table 8 – Typical Gas plant Feed and Product Streams
Field A Field B Products
wt% wt% wt%
N2 1.13% 1.09% 1.12%
C1 66.81% 72.35% 70.19%
C2 13.54% 11.17% 12.15%
C3 9.85% 7.19% 8.27%
IC4 1.73% 1.38% 1.52%
NC4 3.42% 2.84% 3.00%
IC5 0.99% 0.90% 0.97%
NC5 1.09% 1.10% 1.06%
C6+ 1.44% 1.99% 1.71%
Fraction of Feed 53% 47%
Feed Streams
The above data provides an indication of the difference in the feed stream
compositions and the relative contribution from each Field. In reality the compositions
fluctuate from day to day and the relative flow rates of the feed streams vary more
considerably.
Daily data was available for a period exceeding 200 days. The daily compositions
were used to infer the contribution from each field and this was compared against the
field split calculated from the actual metered flows.
2 The data had to be conditioned to exclude CO2, since this was removed in the process and the
discharge stream wasn’t measured.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
11
In order to calculate the Uncertainty Based Optimised Splits however, component
measurement uncertainties are required. Rather than use published standard
uncertainties, which may be appropriate for laboratory conditions, uncertainties based
on the daily component mass balance residuals were calculated. By analysing the
variance in the mass balance residuals for each component (calculated in accordance
with Equation (3)), over the 200 days of data, the relative uncertainty associated with
the measurement of that component could be estimated.3 The values thus calculated
are presented in Table 9:
Table 9 – Calculated Component Uncertainties Relative Uncertainty
± wt%
N2 2.4%
C1 0.5%
C2 1.4%
C3 3.6%
IC4 6.3%
NC4 4.3%
IC5 11.6%
NC5 9.5%
C6+ 30.9%
For the 200 days of data, the inferred Field A splits, based on all the methods
described in Section 2, were calculated. The results for the Optimised, Uncertainty
Based Optimised and C1 Component based splits are compared with the actual Field
A split in Figure 6:
Figure 6 – Inferred versus Actual Field A Splits – Real Gas Plant Data
Calculated Vs Actual Splits
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 169 175 181 187 193 199
Day
Sp
lit
to F
ield
A
Actual Split Optimised Split UB Opt C1 Based Split
3 Using the propagation if uncertainties (as described in [3]), it is possible to infer an average relative
uncertainty for a component that is consistent with observed variance in that components residual.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
12
All three methods agree relatively well with the actual splits over the full 200 days.
The results data is summarised in Table 10 and this also includes the remaining
component based splits.
Table 10 – Inferred Field A Split Average Difference from
Actual Split
Uncertainty
wt% wt%
Optimised Split -2.2% 4.5%
Uncertainty Based Optimised Split -2.7% 4.8%
Component Based Splits N2 1.5% 426.4%
C1 -2.0% 4.9%
C2 -3.5% 6.9%
C3 -2.1% 11.2%
IC4 3.7% 100.1%
NC4 -5.7% 30.1%
IC5 61.2% 174.2%
NC5 -159.8% 4025.5%
C6+ -0.1% 76.3%
The difference between the inferred split for each day was calculated for each day and
the average of these differences is shown in the table. The uncertainties were
calculated based on the standard deviations of the differences between the inferred
and actual splits.
The two optimised splits and the C1 component based split performed the best. The
minor components performed poorly, notably nC5, this is a result of both Fields
having similar concentrations of this component (see Table 8) and is to be expected
based on the discussion in Section 2.1.
Most of the more viable methods tend to under-predict the Field A Split by between 2
and 4 %. The uncertainties were calculated as double the standard deviation in the
daily calculated differences. The consistent under-prediction possibly indicates a small
bias in either: the actual compositional data or the metered flows.
In summary, the differences between the inferred and actual splits and the levels of
uncertainties encountered, indicate that the inferential chemometric methods are
viable options to allocate the feed contributions.
3.3 Realistic Gas Plant Data
In order to explore the accuracy and robustness of the various methods further, the gas
plant data was conditioned to form a theoretical data set. This has the advantage that it
allows us to test various scenarios using Monte Carlo methods to generate random
uncertainties in the measurements. It also allows numerical uncertainties associated
with the methods to be estimated.
In the first simulation the typical feed compositions in Table 8 were combined in a
50:50 mix to provide a perfectly balance product composition. In effect this is
analogous to Example 1, but with realistic data. This approach provides a controlled
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
13
environment, in which the true measurement values are known (and systematic errors
eliminated), that allows allocation uncertainties to be calculated whilst retaining an
authentic data set.
The component measurement uncertainties presented in Table 9 were applied to each
of the feed and product stream compositions to generate measurement errors
randomly. The inferred Field splits were then calculated and compared with the
known true 50:50 split. The above random generation of measurement errors was
repeated over a number of trials and the results are summarised in Table 11.
Table 11 – Uncertainties Inferred Allocation Methods, 50:50 Field Split Analytical Uncertainty Monte Carlo
Uncertainty
± wt% ± wt%
Optimised Split 4.3% 5.3%
Uncertainty Based Optimised Split 3.8%
Component Based Splits N2 9.6% 9.9%
C1 5.2% 5.8%
C2 7.6% 7.3%
C3 11.0% 10.6%
IC4 18.4% 18.8%
NC4 16.5% 16.9%
IC5 54.0% 80.0%
NC5 101.8% 23682.1%
C6+ 190.0% 7396.2%
Both analytical and numerical uncertainties (calculated based on the Monte Carlo
simulation results) are provided. Equations for the analytical uncertainties are
presented in Section 6.4. The differences between the analytical and numerical
uncertainty values for some components is thought to be due to the analytical
uncertainty equations becoming highly non-linear, strictly requiring the use of higher
order derivatives in their calculation. The results agree reasonably with those
encountered with real data presented in Table 10.
The distribution of the differences in the inferred splits from the true splits is
presented in the bar chart below for the two optimised and C1 based component splits:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
14
Figure 7 – Distribution of Errors of Inferred Allocation Methods
0
50
100
150
200
250
<-10% -9% -8% -7% -6% -5% -4% -3% -2% -1% 0% 2% 3% 4% 5% 6% 7% 8% 9% >10%
Nu
mb
er
wit
hin
% b
an
d o
f tr
ue v
alu
e
C1 Opt UB Opt
The above table and figure show that the Uncertainty Based Optimised method
produces the most accurate estimates of the Field splits.
A similar analysis was performed with a true Field Split of 90:10 and the results
presented in Table 12.
Table 12 – Uncertainties Inferred Allocation Methods 90:10 Field Split Analytical Uncertainty Monte Carlo
Uncertainty
± wt% ± wt%
Optimised Split 4.9% 5.3%
Uncertainty Based Optimised Split 2.5%
Component Based Splits N2 9.2% 9.5%
C1 6.0% 6.0%
C2 7.6% 7.6%
C3 9.9% 9.5%
IC4 16.4% 16.7%
NC4 15.5% 16.0%
IC5 52.5% 7928.5%
NC5 106.4% 547.9%
C6+ 227.3% 43899.8%
The results are similar to those for the 50:50 case.
To test the impact of the C1 content of the two fields being similar (analogous to
Example 2) the 50:50 split was repeated but with Field B’s C1 content reduced to
69% and the remaining components increased proportionately. The results are
presented in Table 13.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
15
Table 13 – Uncertainties Inferred Allocation Methods 50:50 Field Split, C1
Content Similar Analytical Uncertainty Monte Carlo
Uncertainty
± wt% ± wt%
Optimised Split 5.0% 4.7%
Uncertainty Based Optimised Split 2.7%
Component Based Splits N2 9.6% 9.7%
C1 86.3% 41608.4%
C2 5.1% 4.6%
C3 11.0% 10.0%
IC4 18.4% 18.7%
NC4 16.5% 16.4%
IC5 54.0% 68.8%
NC5 101.8% 9156.9%
C6+ 190.0% 9341.4%
In agreement with the analysis presented in Section 2, the C1 Component based split
becomes very large but the optimised approaches remain robust.
3.4 Oil Samples
The data in examples considered so far has all pertained to gas systems. The
techniques are equally applicable to liquid systems also. Consider the compositions of
two liquid streams (from Fields X and Y) and the commingled blend of the two
presented in Table 14.
Table 14 – Oil Stream Compositions
BP Fraction
(oC)
Field X Field Y Commingled
Product Z
45 4.1% 1.1% 7.3%
60 0.6% 0.5% 2.8%
75 4.2% 0.8% 7.6%
90 6.2% 0.8% 5.9%
105 12.3% 1.7% 13.3%
120 8.0% 1.7% 6.6%
135 10.3% 2.0% 9.4%
150 11.5% 1.6% 8.2%
165 7.9% 2.0% 4.7%
200 13.1% 4.8% 10.2%
250 12.7% 9.0% 7.9%
250+ 9.2% 74.3% 16.1%
100.0% 100.0% 100.0%
These are samples from a real system. The Optimised and component based
inferential allocation methods were applied to this data and the results are presented in
Table 15.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
16
Table 15 – Inferred Field Split of Oil
Field X Split
Optimised Split 88.7%
Component Based Splits
BP
Fraction
(oC)
45 202%
60 1359%
75 200%
90 95%
105 110%
120 77%
135 90%
150 67%
165 45%
200 65%
250 -29%
250+ 89.3%
In this instance the true split was unknown, except that it was expected to be within
85% to 90% for Field X. As can be seen from the results the Optimised Split and the
250+ fraction based split both fall in this expected range.
The true split was unknown because of meter failure and this case serves as an
example of how Inferential Allocation techniques could serve as secondary means of
allocation – this is discussed further in Section 4.
3.5 Three Feed Streams
On of the questions posed in Section 1 was whether it was possible to infer the
contribution of more than two Fields. The answer to this is yes and equations are
presented below.
The component based split requires the use of two components to infer the split of
three fields (in general M + 1 components are required to infer the contribution from
M fields). The approach is similar to that developed for the two fields derived in
Section 6.1. The final equation for three Fields (A, B, D) mixing to form a
commingled stream (C), based on C1 and C2, is:
22111122
22111122
**
**
CCCCCCCC
CCCCCCCCA
BDBABDBA
BDBCBDBCS (8)
Similarly, the Optimised Split can be used to allocate between three fields and the
analogous equation is:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
17
toNi
iiii
toNi
ii
toNi
ii
toNi
ii
toNi
iiii
toNi
iiiiiiii
toNiA
BABDBDBA
BDBACBBABDBDCB
S
11
2
1
2
1
2
111
**
*****
(9)
The gas plant, from which the data presented in Section 3.2, also experienced periods
when three fields were producing. Typical compositions and Field splits are presented
in Table 8.
Table 16 – Typical Gas Plant Feed (3 Fields) and Product Streams
Field A Field B Field C Products
wt% wt% wt% wt%
N2 1.13% 1.09% 1.46% 1.12%
C1 66.81% 72.35% 81.86% 70.19%
C2 13.54% 11.17% 9.17% 12.15%
C3 9.85% 7.19% 3.18% 8.27%
IC4 1.73% 1.38% 0.97% 1.52%
NC4 3.42% 2.84% 1.18% 3.00%
IC5 0.99% 0.90% 0.60% 0.97%
NC5 1.09% 1.10% 0.44% 1.06%
C6+ 1.44% 1.99% 1.15% 1.71%
Fraction of Feed 48% 46% 6%
Feed Streams
The figures below plot the inferred allocation to each of the three fields using the two
methods described in Equations (8) and (9).
Figure 8 – Inferred versus Actual Field A Splits – Real Gas Plant Data for Three
Fields
Calculated Vs Actual Field A Splits
0%
10%
20%
30%
40%
50%
60%
70%
80%
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156 161
Day
Sp
lit
to F
ield
A
Actual Split Optimised Split C1 C2 Based Split
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
18
Figure 9 – Inferred versus Actual Field B Splits – Real Gas Plant Data for Three
Fields
Calculated Vs Actual Field B Splits
0%
10%
20%
30%
40%
50%
60%
70%
80%
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156 161
Day
Sp
lit
to F
ield
A
Actual Split Optimised Split C1 C2 Based Split
Figure 10 – Inferred versus Actual Field C Splits – Real Gas Plant Data for
Three Fields
Calculated Vs Actual Field C Splits
0.000%
5.000%
10.000%
15.000%
20.000%
25.000%
30.000%
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156 161
Day
Sp
lit
to F
ield
A
Actual Split Optimised Split C1 C2 Based Split
There is more scatter in the predicted splits and a reduction in accuracy compared to
when two fields are present. The average deviation between the predicted and actual
split and calculated uncertainty (based on twice the standard deviation) are presented
in Table 17.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
19
Table 17 – Summary of Inferred Field Splits vs Actual
Average Difference from
Actual Split
Uncertainty Average Difference from
Actual Split
Uncertainty
wt% ± wt% wt% ± wt%
Field A -3.5% 16.6% -5.2% 16.7%
Field B 4.4% 25.8% 7.3% 25.2%
Field C -0.9% 9.5% -2.1% 11.8%
Optimsed Split C1, C2 Based Split
Overall, the Optimised Split performs slightly better than the Component Based split.
3.6 Compositional Drift
Another question posed in Section 1 was: “Would we know if one of the assumed
constant compositions starts to drift?”
A feature of the optimised split and uncertainty based optimised split is that they
provide a methodology to determine if the commingled mixture is genuinely a mix of
our two assumed feed compositions.
A relatively easy metric to monitor to determine if there has been compositional drift
is to plot the sum of the squares of the component mass balance residuals as given by
equations (15) and (6). Plotting this metric each day provides a figure that may be
monitored and if it increases then this would indicate compositional drift.
As an example, using the 50:50 split case described in Section 3.3, a bias was
introduced part way through the simulation. A plot of the sum of squares of the
residuals for the Optimised Split, filtered to smooth the data, is presented in Figure
11:
Figure 11 – Plot of Sum of Squares of Residuals 50:50 Split, Compositional Bias
Introduced on Day 600
0.000008
0.0000085
0.000009
0.0000095
0.00001
0.0000105
0.000011
0.0000115
1 23 45 67 89 111 133 155 177 199 221 243 265 287 309 331 353 375 397 419 441 463 485 507 529 551 573 595 617 639 661 683
Day
Su
m o
f S
qu
are
s o
f R
esid
ua
ls Bias introduced into
Field B Composition
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
20
As can be seen the point at which the bias occurred is evident.
Though a useful metric to monitor from a practical viewpoint in some putative
“imagined” allocation system, there are more rigorous approaches. The above
approach does not provide any information on what is a significant level of the sum of
squares number is. It simply looks for a rise in the value and infers that something has
changed.
Alternatively, it is possible to determine what a significant value for the sum of
squares figure is, above which systematic errors are indicated. This would provide a
test statistic to determine, given the uncertainties in the compositions, whether the
residuals could have arisen reasonably by chance or whether there is a systematic
shift. Such techniques are used to detect gross errors in data reconciliation [1] and [2].
4 APPLICATIONS
Having shown the potential of these inferred allocation methodologies, the final
question to be answered is: what use can they be for allocation?
For the case of the gas plant described above these techniques would be of limited use
since all the streams are measured. This data was merely used to test the
methodologies. However for the case of the oil sample example the contribution of the
two streams was not known and this is where the inferential techniques could be
usefully employed to allocate the product.
To be used as a primary method of allocation the confidence in the measured
compositions would have to be high. Possible applications could include systems
where the feed streams are known to have stable compositions.
The methods could be used as a secondary, back-up, means of allocation where there
is a risk of the primary measurement failing. For example, in a system relying on a
multi-phase flow meter, especially when located subsea, the inferential technique
could be used as an alternative method of allocation should the meter fail.
Indeed even if the meter was working correctly, the secondary inferential allocation
could be used as a quality control metric. These techniques provide alternative
methods of allocation which can be used to monitor how well the primary
measurement and allocation systems are performing.
5 CONCLUSIONS
This is paper has presented three methods of inferring allocated quantities from
compositional data alone:
Single component based split, using ratios of single components
Optimised Split, calculated by minimising component mass balance residuals
Uncertainty Based Optimised Split, calculated by minimising the uncertainty
in component mass balance residuals
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
21
These have been demonstrated to be viable means of allocation, using both theoretical
examples and real data.
Their accuracy is dependent on the relative flows of the streams and the dissimilarity
of their compositions.
The optimised split approaches were generally found to be the most robust and
accurate.
Though mainly analysed in terms of allocating between two Fields the methods have
been demonstrated to work with three fields with diminished accuracy.
The mass balance residuals provide a metric with which to monitor the consistency of
the compositional data and thereby detect systematic changes.
A number of possible applications have been suggested: as a primary allocation
method, a secondary back-up method to mitigate the impact loss of metering on the
allocation system and as a quality assurance check.
6 MATHEMATICAL DERIVATIONS OF EQUATIONS PRESENTED
6.1 Field Split Based on a Single Component
Consider a mass balance for component i:
iCiBiA CFBFAF *** (10)
Contribution or split of A is given by:
BA
AA
FF
FS (11)
SB can be similarly defined and noting that FA + FB = FC, (10) becomes:
iiBiA CBSAS ** (12)
Also since the splits must sum to 1, SA + SB = 1, and rearranging to obtain SA (13)
becomes:
ii
iiA
AC
BCS (13)
6.2 Optimised Field Split
Consider a mass balance for component i:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
22
iiAiAC CBSAS *)1(*1 (14)
The sum of the squares these residuals, K, is calculated:
toNi
iiAiA CBSASK1
2*)1(* (15)
K is differentiated with respect to SA:
toNi
iiiiAiA
A
BACBSASdS
dK
1
**)1(**2 (16)
The minimum value of K is obtained by setting the derivative to equal zero and
rearranging to obtain SA:
toNi
ii
toNi
iiiiii
ABA
CACABB
S
1
2
1
*)
(17)
6.3 Uncertainty Based Optimised Field Split
Consider a mass balance for component i:
iiAiAi CBSAS *)1(* (18)
The uncertainty in the residual δi for component i is calculated from the following:
2
,
2
2
,
2
2
,
2
, iC
i
iiB
i
iiA
i
ii U
CU
BU
AU (19)
The partial derivatives are:
1;1;i
iA
i
iA
i
i
CS
BS
A (20)
The residuals are then weighted according to their uncertainty. This is performed by
dividing δi by its uncertainty calculated according to (19).
2
,
2
,
22
,
2
2
2
,
2
*)1(*
*)1(*
iCiBAiAA
iiAiA
i
i
UUSUS
CBSAS
U (21)
The weighted sum of squares of the residuals (J) is given by:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
23
toNi iCiBAiAA
iiAiA
toNi i
i
UUSUS
CBSAS
UJ
12
,
2
,
22
,
2
2
12
,
2
*1*
*)1(* (22)
The optimum value of SA corresponding to the minimum value of J can be calculated
when the derivative of J with respect to SA is zero. The equation that describes this is:
toNi
iCiBAiAA
iBAiAAiiAiA
iCiBAiAA
iiiiAiA
A
UUSUS
USUSCBSAS
UUSUS
BACBSAS
dS
dJ
1
22
,
2
,
22
,
2
2
,
2
,
2
2
,
2
,
22
,
2
*1*
*1*2**2**)1(*
*1*
**)1(**2
(23)
SA cannot be obtained directly from (23) and must be calculated using numerical
techniques such as binary chop, Newton Raphson, secant, etc.
6.4 Inferential Method Uncertainties
Single Component Based Split
From equation (13), the partial derivatives are calculated:
iii
A
ii
ii
i
A
ii
ii
i
A
BAC
S
BA
AC
B
S
BA
CB
A
S 1;;
22 (24)
And the uncertainty is given by:
4
2
,
22
,
22
,
2
ii
iCiiiBiiiAii
SABA
UZBUACUBAU (25)
Optimised Split
Equation (17), has to be re-expressed in terms of independent variables and hence N-1
components. One component is not independent since the component mass fractions
must sum to 1.
2
111111
2
11111111111111
1*111*)
toNi
i
toNi
i
toNi
ii
toNi
i
toNi
i
toNi
i
toNi
i
toNi
i
toNi
i
toNi
iiiiii
A
ABBA
CABCABCACABB
S
(26)
Abbreviating (26) to:
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
24
Denom
NumSA (27)
Defining,
;111toNi
id AA (28)
And similar terms for B and C, the partial derivatives of SA are:
2
**2
Denom
ABABNum
Denom
BCBC
A
S ddiiddii
i
A (29)
2
**2*2*2
Denom
ABABNum
Denom
CABCAB
B
S ddiidddiii
i
A (30)
Denom
BABA
C
S ddii
i
A (31)
With these partial derivatives for each of N-1 and the associated absolute component
uncertainties the uncertainty of SA calculated using the propagation of uncertainties.
27th
International North Sea Flow Measurement Workshop
20th
– 23rd
October 2009
25
NOTATION
A Mass fraction Field A
B Mass fraction Field B
C Mass fraction commingled
stream
D Mass fraction Field D
Denom Denominator of Equation 26
F Stream mass flow
J Weighted sum of squares of
residuals
K Sum of squares of residuals
M Number of meters
N Number of components
Num Numerator of Equation 26
S Field split
U Absolute uncertainty
δ Mass balance residual
Subscripts
A Field A
B Field B
C Commingled stream
C1 Component C1
C2 Component C2
d Component as defined in
Equation 28
D Field D
i component
SA Field A split
7 REFERENCES
[1] Data Processing and Reconciliation for Chemical Process Operations, J
Romagnoli and M Sanchez, Published by Academic Press (2000), ISBN 0-
12-594460-8.
[2] Data Reconciliation and Gross Error Detection: An Intelligent Use of Process
Data, S Narasimhan and C Jordache, Published by Gulf Publishing (2000),
ISBN 0-88415-255-3
[3] Guide to the Expression of Uncertainty in Measurement, International
Organisation for Standardisation, ISO/IEC Guide 98:1995.
[4] Use of Subsea Wet Gas Flowmeters in Allocation Measurement Systems,
API RP 85, First Edition, August 2003.
[5] Determination of Measurement Uncertainty for the Purpose of Wet Gas
Hydrocarbon Allocation Robert A. Webb, BP, Winsor Letton, Letton-Hall
Group. Martin Basil, FLOW Ltd North Sea Flow Measurement Workshop,
22nd -25th October 2002.
[6] Experiences in the Use of Uncertainty Based Allocation in a North Sea
Offshore Oil Allocation System, Phillip Stockton & Alan Spence, Smith Rea
Energy Ltd, Production and Upstream Flow Measurement Workshop,
Houston, 12-14 February 2008.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 1
Metering Atlas
A portal to create transparency in
production and fiscal measurement data
by
Lex Scheers (Shell Global Solutions International),
Oi-Mee Voon (Brunei Shell Petroleum)
and Tjidde Boers (Magion)
Summary
The paper describes the development and implementation of a tool that will optimize work
processes between the Metering Data Consumers (Reservoir Engineers, Production
Technologists, Operators, Programmers, Metering Engineers, etc) and Metering Data
Providers (Facility or Metering Engineers, Maintenance/Production Operators and
Production Chemistry). The connecting link between the Metering Data Consumers and the
Metering Data Providers is a web-based portal, called Metering Atlas. This portal collects
data related to measurement equipment from the various software applications and data
sources that are used in the oil and gas business. Examples of these software applications and
data sources are an instrument and meter engineering master database, data historian,
hydrocarbon oil and gas production administration system, laboratory information systems,
financial, resource and work planning systems, etc. Although all these software applications
are in principle independent and they all have their own workflow processes, their own data
flow and their own custodian, it is obvious that combining the information from these
independent software applications will results in valuable additional information regarding the
status of a metering system. Combining all this information it is possible to judge the health
status of a metering system, whether it can be trusted or distrusted. By having all this
information available in an easy accessible web-based system, with easy navigation, similar to
an atlas, also creates transparency of the company’s metering system with the Metering Data
Consumers and might initiate discussions regarding improvements (or relaxations) of
requirements of metering facilities. Ultimately this will lead to an optimal focus on metering
equipment and delivering production data of adequate quality.
1) Introduction
The production measurement process is more than just measurement hardware in the field but
is the entire chain from data collection with meters in the field up to the final production
reporting. It includes all intermediate steps such as measurement and sampling guidelines,
operational procedures like maintenance and calibration procedures, data processing (pVT
algorithms), data transmission and reconciliation/allocation algorithms. In this entire chain,
many software applications and databases are used to gather, validate and store information
about individual flow rates of oil, water and gas and the quality of production streams.
Generally, fiscal or custody transfer measurements show the lowest degree of uncertainty, i.e.
the best that is technically achievable. That is why the fiscal measurements, traditionally are
well covered with extensive procedures to maintain and calibrate the measurement equipment.
Simple reason is that a 1% systematic error in a fiscal measurement results in a 1% systematic
error in the payments between two companies (either loss or gain). However, it can also be
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 2
argued that the uncertainty in far upstream measurements (well tests measurements, platform
or station discharge measurements) are linked to uncertainty in money flow (see Fig. 1).
Although, it is not as straight forward as in a fiscal or custody transfer measurement, also here
systematic errors will eventually influences the economics of a development. Upstream
production measurements have an economic impact on the business as they not only costs
money (both Capital and Operational Expenditures), but they also deliver data that is used in
decision-making processes such as production optimisation, reservoir modelling and in
measuring the economic returns (see Fig. 2).
GA
SG
AS
OIL
OIL
WA
TE
RW
AT
ER
RESERVOIR
GA
SG
AS
WA
TE
RW
AT
ER
WA
TE
R
WA
TE
R
DIS
POS
AL
DIS
POS
AL
SALES GASSALES GAS
SALES OILSALES OIL
FLARE GAS, FLARE GAS, OWN USEOWN USE
PRODUCTION FACILITYfor each phase
ΣΣΣΣin = ΣΣΣΣout
$
$$
$
$ $$ $ $
Fig. 1
Although most focus is on
fiscal or sales allocation
measurement (because of their
direct impact on money
transfer between companies),
also the far upstream
measurements (production as
well as injection or disposal
streams) are impacting the
economics of a project
These economics then not only set the uncertainty requirements for the various production
measurements but often also indicate what the most critical measurements are. This could be
oil flow rate, gas flow rate, GOR or even water flow rate or watercut in water-constrained
facilities. The customers of the measurement and allocation process, i.e. the “Metering Data
Consumers”, are generally spread over several disciplines in the oil and gas companies, their
partners or located in government bodies. Examples are reservoir engineers, petroleum
engineers, facility and process engineers, operators, finance, legal and contract staff.
Moreover, each of these Metering Data Consumers has their own requirements regarding the
measurement process.
TheTheValueValueLoopLoop
Interpretation&
Modeling
Data Gathering
Decisions&
Actions
HydrocarbonAssets
You can’t manage
what
you can’t measure
Fig.2
The value loop, wrong data leads to
poor modelling and ultimately leads
to sub-optimal or poor decisions
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 3
During the last decade a large number of new measurement concepts have been introduced,
examples are Coriolis meters for mass flow rate and net-oil measurement, Ultrasonic gas flow
meters, MultiPhase Flow Meters (MPFM) and Wet Gas Meters (WGM). With the
introduction of all this new and more advanced measurement equipment in the upstream area
of the oil and gas business one should ask the question whether we can still manage the
production measurement chain with the resources we were using in the old days of orifice
plates and positive displacement meters. The introduction of more advanced electronics,
sophisticated fluid flow models, wet gas over-reading correlations, the number of additional
fluid parameters required to properly run the modern measurement equipment also makes it
necessary to adapt the skills of the staff in the field. Moreover, the organisation should be
adapted and process tools should be made available, such that proper management
(custodianship) of this "production measurement chain" can be done. Further below it is
explained that “Metering Atlas” is one of the tools, which will create transparency on the
metering data quality and improves the communication between Metering Data Providers and
Metering Data Consumers.
As every field development has its own specific requirements for the production measurement
process it has been demonstrated in earlier publications1 that an "engineering phase" and
"operations phase" should be established. In the engineering phase the requirements from the
Metering Data Consumers are investigated and, together with the input from Meter Data
Providers, compiled into a measurement and allocation philosophy. Subsequently, this results
in detailed design and description of the system. Once the operation phase has started the
measurement process should be managed through proper custodianship (see Fig. 3). This
management should be transparent and auditable and Metering Atlas is a tool that is
specifically designed to create that transparency for all the Metering Data Consumers. In other
words, Metering Atlas moves the measurement process from a “trust me” to a “show me”
process.
Updates M&A Philosophy, Manual
and Logbooks
Updates M&A Philosophy, Manual
and Logbooks
IDENTIFY & ASSESS SELECT DEFINE EXECUTE OPERATE
Basisfor
Design
VAR 5
Operate
VAR 3
ConceptSelection
VAR 4
ProjectSpec
DecisionGate 4Final
Investm.
VAR 1
ProjectInitiation
DecisionGate 1Case forActions?
VAR 2
FeasibilityStudy
DecisionGate 2Feasible?
Commiss.Perf.Test
Hand-overAsset
Ownership
Design,Construct,Pre-Comm.
Start-UpTransferAsset
Pre-StartUpAudit
Detailed M&A Manual
Detailed M&A Manual
Final M&A Philosophy
Final M&A Philosophy
High level M&A Philosophies
High level M&A Philosophies
DecisionGate 3SelectConcept
Fig. 3
Various stages for Metering and Allocation in de conceptual and final design phase and in
the operating phase.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 4
2) Metering Data Quality (current situation)
The quality of production data collected in the field, whether done with conventional metering
equipment like orifices, Venturies, positive displacement meters (PD), turbine meters or with
the more advanced multi-phase and wet-gas flow metering equipment, is often with doubts.
All metering systems do have uncertainties and often these uncertainties are specified by the
manufacturer or determined either through standard calculations or through tests in a
dedicated flow or calibration loop. However, this in many cases the uncertainty is calculated
with simplified and limited assumptions, i.e. often with a constant fluid (fixed fluid
parameters) or without any disturbing factors like gas presence (in case dealing with liquid
streams), or the other way around liquid presence (in case dealing with gas streams), wax
deposition, sand, hydrates, etc. etc. Furthermore, one can ask the question what is the
uncertainty of a cumulative product stream if it is only metered intermittently, i.e. using a test
separator with its auxiliary measurement equipment that is only used one day every month to
execute a well test. What actually happens in between two well tests can be followed by
monitoring several other parameters but definitely not as accurate as using a continuous
flowrate measurement. Note that Metering Atlas will not deliver actual uncertainty figures for
a measured stream but Metering Atlas is able to roughly indicate the status of a metering point
in the form of a traffic light, green, amber or red.
What Metering Data Consumers see is often just numbers in a production report (see Fig. 4).
As an example a monthly report showing the following: 1,000 bbl/day of oil, 500,000
Sm3/day of gas and 35% watercut. What you can’t see from these reports is whether these
numbers are highly accurate numbers or just a bit of a guess? Are the numbers obtained
through a continuous measurement process or is it the result of a one-day well test that is
extrapolated to a month. Is the watercut just a yearly well head sample or is it a continuous
measurement with properly calibrated equipment ?
Production Net Oil Gas Water Watercut
Wells [days] [Sm3] [k Sm3] [Sm3] [%]
XX-122 30.43 77,774.7 7,451.7 0.0 0.0%
XX-123 30.43 28,385.8 3,002.9 800.0 2.7%
XX-124 16.93 18,467.3 4,455.9 8,234.0 30.8%
XX-125 30.43 30,255.3 3,220.9 0.0 0.0%
XX-126 0.00 0.0 0.0 0.0 NA
XX-127 30.43 71,432.4 6,252.1 0.0 0.0%
XX-128 21.47 23,126.9 3,334.9 8,924.0 27.8%
XX-129 30.43 54,469.5 5,616.7 0.0 0.0%
XX-130 30.43 56,935.1 6,424.6 345.0 0.6%
XX-131 12.90 14,392.9 3,449.0 0.0 0.0%
XX-132 29.44 17,194.6 3,047.3 6,789.0 28.3%
XX-133 27.71 16,453.9 2,230.8 545.0 3.2%
XX-134 8.90 5,839.0 1,364.5 4,567.0 43.9%
Total 414,727.4 49,851.4 30,204.0
How is this informationdetermined ?
1) Spot sample once a year(month) at well head
2) Average of well test separator reading
3) Continuous with a Multi-Phase Flow Meter
Production Net Oil Gas Water Watercut
Wells [days] [Sm3] [k Sm3] [Sm3] [%]
XX-122 30.43 77,774.7 7,451.7 0.0 0.0%
XX-123 30.43 28,385.8 3,002.9 800.0 2.7%
XX-124 16.93 18,467.3 4,455.9 8,234.0 30.8%
XX-125 30.43 30,255.3 3,220.9 0.0 0.0%
XX-126 0.00 0.0 0.0 0.0 NA
XX-127 30.43 71,432.4 6,252.1 0.0 0.0%
XX-128 21.47 23,126.9 3,334.9 8,924.0 27.8%
XX-129 30.43 54,469.5 5,616.7 0.0 0.0%
XX-130 30.43 56,935.1 6,424.6 345.0 0.6%
XX-131 12.90 14,392.9 3,449.0 0.0 0.0%
XX-132 29.44 17,194.6 3,047.3 6,789.0 28.3%
XX-133 27.71 16,453.9 2,230.8 545.0 3.2%
XX-134 8.90 5,839.0 1,364.5 4,567.0 43.9%
Total 414,727.4 49,851.4 30,204.0
How is this informationdetermined ?
1) What equipment has been used ?
2) How often is the datagathered ?
3) Is equipment calibratedand operating OK ?
4) What is uncertainty indication ?
Fig. 4.
Monthly production report just shows numbers but do not reveal any information how these
numbers are obtained, i.e. extremes could be a continuous measurement with high accuracy
or a spot measurement once a year.
The Metering Data Consumers need to know what this data quality is in order to judge the
risks they take when they use these data to run their business. Examples are reservoir
engineers optimizing their reservoirs, operators optimizing their production or making
abandonment decisions but also finance and sometimes contract staff judging the risks in sales
allocation processes. In order to bring all relevant information related to the quality of
production measurement data closer to the Metering Data Users and create some transparency
on production measurement data, Metering Atlas was developed. With the creation of this
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 5
transparency we also give the Metering Data Consumers the status of measurement data and
the possibility to challenge data collection, i.e. they can request better or more frequent data if
that is required or even suggest a relaxation in the data gathering process.
3) Metering Atlas Objectives
With the arguments from the previous sections Metering Atlas has been developed and has
been implemented in one of the Royal Dutch Shell Operating Units in Brunei (Brunei Shell
Petroleum). The ultimate objective for BSP is to provide a means to improve the quality of
production data, which is used by so many in BSP to make important business decisions. Note
that Metering Atlas will not cure metering problems, it just provides information on the
quality and status, users need to take action to cure. The objective can be further split in two
sub-objectives:
1) Metering Atlas shall be a portal to make the entire chain of fiscal and production data
gathering and the subsequent calculations transparent. The aim is to provide a friendly
and easy accessibility for users who have an interest in metering data. Preferred way is a
web-based system with an “atlas look”, showing quality tagged values, in relation to the
infrastructure and the metering/sampling points. Hence, the name Metering Atlas.
2) Metering Atlas shall optimise the work processes between the Meter Data Consumers and
the Meter Data Providers. Therefore Meter Data Consumers shall view all metering
related data via a single portal. This greatly improves the changes that incorrect or
missing data is challenged with the Metering Data Providers.
Metering Atlas, is an internal Shell development funded by Brunei Shell Petroleum and
developed in close co-operation with Shell Global Solutions. It is a portal that retrieves data
from various existing and independent data sources like an oil and gas accounting system, an
instrument master engineering database, a laboratory database, a real time data historian, a
maintenance and work-order tool and uses all the collected data in a health checking process
and presents that data in a coherent and user-friendly form. Metering Atlas has been
developed with an “atlas-look” allowing the users to easily navigate through the various
facilities and finally zoom in to specific metering, sampling or analyzer equipment to reveal
all basic information, including specifications, drawings, photos, calculation routines and
other meter trivia. Two additional modules have been added to Metering Atlas. One is the
health module, which determines the quality based on meter type templates and the various
source data. The second one is a collaborative module to initiates workflow processes based
on quality changes or user entry is also included. Both will be discussed further below. As a
last add-on, Metering Atlas also provides a framework for company wide metering
engineering calculations such that consistency throughout the company is achieved. For this
used is made of the Kelton’s iFloCalc package.
Metering Atlas is built on Shells Production Portal architecture that uses OsiSoft’s Analysis
Framework (AF2) to collect, integrate and analyze data, and Microsoft Sharepoint Services to
visualize the data. Webservices are used to communicate data between the portal components.
The various software applications that will provide data into Metering Atlas are briefly
discussed below.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 6
3.1) Geographical Information System (ArcGis)
Metering Atlas will be a sub-application of Shell’s Production Portal. The latter is a web-
based tool that allows staff to easily access production related data for monitoring and
analyses. All data is accessible through the Shell network and the Production Portal is
equipped with various levels of security and with standard reporting functionality. The reason
to combine Production Portal and Metering Atlas is very simple; Production Portal reveals the
information on various production, injection and disposal streams where as Metering Atlas
provides background on how that information is obtained. Another big advantage to integrate
Production Portal and Metering Atlas is that is becomes feasible to use the same navigation in
both systems. ArcGis is the Shell corporate Geographical Information System (GIS) suite and
refers to an organized repository or database of geographically referenced information.
ArcGis enables the visualization of data in a geographic way. Currently it only runs on Vista
Operating System. Basically, it lets you query or analyse a database and receive the results of
the query in the form of some kind of map. In a GIS, geographic information is described
explicitly in terms of geographic coordinates (latitude and longitude or some national grid
coordinates). With ArcGis it becomes very simple to zoom into production facilities,
platforms, separators and finally select metering equipment (see Fig. 5).
Fig. 5
Example of the “atlas-type” screen for easy navigation and selection of Metering Points.
3.2) Systems Applications and Products (SAP)
Systems Applications and Products (SAP) is the brand name of the company, which provides
web-based integrated business solutions (so called Enterprise Resource Planning systems).
Within Royal Dutch Shell, this integrated business system/solution and has been rolled out
globally, which resulted in standardized and simplified businesses, processes and IT systems.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 7
In principle a number of modules are available in SAP all with the aim to monitor and record
business processes, e.g. resource planning, inventory management, procurement of good and
services, financial control, human resources, etc. Specifically for measurements it is the
maintenance and calibration activities (scheduling and results) that are monitored and
recorded in SAP. The interface between SAP and Analysis Framework (AF) will be via
Facility Status Report (FSR) holding table
3.3) Document Management System (LiveLink)
LiveLink, an electronic document management system, is a 3rd party product from Opentext
and is used for storing and sharing files on a Web-based platform. LiveLink offers
functionality like applying metadata to documents, auditing all document events and
managing version history. Main benefits of LiveLink are the sharing of data files with
controlled access and version control. LiveLink is the Royal Dutch Shell standard Group
platform for facilitating the global exchange of knowledge and information throughout Shell.
Specifically for Metering and Allocation LiveLink is the place to store information like
metering installation and operating manuals, reconciliation or allocation procedures, sales
allocation contracts and possibly other trivia regarding the meters like photographs,
publications, Maintenance Job Routines (MJR), etc.
3.4) Data Historian (PI)
A Data Historian is a database that collects Real Time data from systems like SCADA, DCS
and flow computers. Typical this data is collected at intervals of 1 sec to 10 minutes. The data
historian can store large amounts of data and save the information for many years. The Data
Historian is used for data trending and is interfaced with Real Time Operations applications
and with the Office Domain applications. Today data historians are gaining their own
processing capability and are able to perform more and more complex calculations within the
process historian. The global Shell standard process historian is OSI-PI. With respect to
measurement equipment it is obvious that the output of many meters and analyzers can be
made available in PI, this can be flow rates, pressures, temperatures, composition, etc. Quite
often we also see that data, which is processed by other tools, will be put back into the PI
system and make it available for storage and further accessibility.
3.5) Hydrocarbon Allocation System (Energy Components or EC)
Production or Sales Allocation is used to describe a process that apportions the bulk flow (or
individual oil, water and gas flow) of a hydrocarbon evacuation system to the respective
contributors to that bulk flow. These contributors can be wells, reservoirs, concessions or
companies. There are many different algorithms (on the basis of the volumetric, mass or
composition) that can be used to allocate production but all have in common that they try to
close the product balance over a production facility (or parts thereof). It also covers volumes
that are recycled (e.g. gas lift volumes), volumes for own use (e.g. compressors), volumes
flared, volumes disposed off, line packing, etc. Energy Components (EC) is an application
suite within EP, that comes in 4 modules, EC Production, EC Transport, EC Sales and EC
revenue) and is delivered by TietoEnator. Energy Components is the Royal Dutch Shell
corporate tools for the Hydrocarbon Allocation process.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 8
3.6) Laboratory Information Management System (LIMS or Sample Manager)
Sample manager is the Shell corporate tool used to manage the analysis of samples, i.e. from
sample entry into the laboratory through to the completion of the analyses and reporting of the
results. It contains various calculation routines, ISO, AGA or other standards, to determine the
various fluid properties of the sampled oil, gas and water.
3.7) Instrument Engineering Tools (INTools)
INtools (or now called SmartPlant Instrumentation) is an Instrument Engineering Master
Application/Database that facilitates the design and management of instruments. It is designed
for instrumentation specialists involved in the definition and specification of field
instrumentation and process control systems. INtools manages instrumentation data during
design, engineering, commissioning, operation, systems maintenance, upgrades and re-vamps
or during expansion projects in a variety of facilities. It covers information about the
instrument index, process data, operating envelopes, instrument specification sheets,
calculations, specifications, hook-ups, etc.
4) Metering Atlas
In Fig. 6 below, a schematic is presented with all the above-mentioned tools feeding data into
Metering Atlas. The core of Metering Atlas is OSIsoft Analysis Framework (AF). It acts as an
integrator for all data that is used in Metering Atlas. Note that OSIsoft also delivers the PI
system (Data Historian) for storing real-time data. Another component is Microsoft’s
SharePoint that enables easy viewing and searching of data. Once data is available in
Metering Atlas, data can only be viewed, there is no additional functionality in Metering Atlas
that “modifies” data and hence there is no data transfer back from Metering Atlas into one of
the data sources. Next to making all the data visible, the data will also be used in two
additional modules that have been added to Metering Atlas:
1) Health Monitoring Module; this module contains high value business functions of the
Metering Atlas. It compares the input from the various sources and provides a
conclusion in the form of a traffic light and thus indicate whether a metering system is
healthy or not. In particular this overview is relevant for higher-level management.
2) Collaborative Module; the purpose of the module is to stimulate knowledge sharing and
metering related discussions between Shell employees usually not in daily contact with
each other. The module also gives background information on metering related
decisions, so these discussions can be held at a high functional level, eliminating much
of the introductions usually needed.
One of the key objects in Metering Atlas is a Metering Point. This is defined as a metering
installation that provides data to the Data Consumers. As an example an orifice arrangement,
including all its auxiliary equipment like differential pressure, pressure or temperature
measurements or density measurement is considered as a metering point. The data consumer,
in principle, is only interested in the outcome of a flowrate of x Sm3 of gas and not so much in
the pressure or temperature measurement. However, the temperature and pressure
measurement are important in assessing the quality of that flowrate of x Sm3. A wrong
pressure measurement results in a wrong flowrate. As such the information related to the
pressure measurement contributes to the health of the system and is taken into account in
setting the Metering Point traffic light to green, amber or red. Similar considerations apply to
all other metering Points.
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 9
MeteringAtlas Portal
MeteringAtlas Portal
HEALTH MODULE
COLLABORATIVEMODULE
DATACONSUMERS
SECURITY
INTERFACE
INTERFACE
Energy Components
ArcGis
PU
Data Historian (PI)
InTools
FieldWare
LOTAM ++
LIMS
SAP
Metering Calc’s
LiveLink
Production PortalProduction Portal
HUMAN -MACHINE
INTERFACE
Fig. 6
Schematic of Metering Atlas, information is pulled from various existing independent systems
that are used to manage specific tasks in the business. If all this information is combined,
better indications on the health of a metering point can be obtained
Hence, a Metering Point is defined by
1. Location, this could be any location in the production process, i.e. well flow lines,
separator outlets, platform discharge meters or fiscal flow meters. But also other
locations like disposal and flare meters.
2. Meter type
Currently the following meter types are implemented in Metering Atlas but new meter
types can easily be added:
- DP devices like; Orifice, Venturi and Cone flow meters
- Turbine and Positive Displacement (PD) flow meters,
- The more advanced flow meters like Coriolis or UltraSonic flow meters,
- Electromagnetic flow meters en Vortex flow meters,
- Multi-Phase Flow Meters and Wet Gas flow Meters
3. Type of fluid
- Gross Oil (oil and water emulsions)
- Net Oil (dry crude oil with small amounts of water)
- Gas (gas without any liquids present, either from separator or sales gas)
- Water,
- MultiPhase Fluid,
- Wet Gas
4. Shell's Metering Class, note that each class does have it’s own requirement for
uncertainty.
- Class I for Fiscal measurement
- Class II for Sales Allocations Measurements
- Class III for Upstream Production Measurement
- Class IV for Environmental Measurements
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 10
For each of the above combination of meter type, fluid and class there is a dedicated health
template defined.
In Fig. 7 below an overview is given of information that is pulled from the various
applications and nicely listed on one single web page.
Fig. 7,
A typical example of how all information for one particular metering point, retrieved from the
various IT systems, can be displayed in one single overview screen
4.1) Health Module
In the health module the various input is compared based on a set of rules. In fact each group
of meters have one health evaluation template assigned. As an example, all ultrasonic meters
of series “A” from manufacturer “B” for class II gas flow measurement will be assigned to
one specific health template with a dedicated set of rules. The Metering Custodian will design
and administer the health templates, while the Metering Atlas custodian (or programmer)
build the template into the Metering Atlas. When no heath template is assigned to a metering
system, Metering Atlas will indicate a warning. Templates can easily be configured or
modified; in fact a metering engineer without any specific IT knowledge can configure these
templates. Examples of possible rules that can make up a template are given below and an
example of the a health screen for a metering point is given in Fig. 8.
• Comparison actual flow rates (PI) and design operating envelope (InTools)
• Comparison actual flow rates (EC) and design operating envelope (InTools)
• Comparison watercut (EC) and watercut samples (LIMS)
• Calibrations overdue by xx days (SAP)
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 11
• Sampling overdue by xx days (LIMS)
• Comparison p and T measurement (PI) and their operating envelope (InTools)
• Comparison measured density (PI) and density in samples (LIMS)
• Comparison of speed of sound in US Gas Flow Meter (PI) and calculated from gas
composition with AGA 10 algorithm (LIMS)
• Comparison composition from GC (PI) and sample (LIMS)
• Comparison density in Coriolis meter (PI) with sample (LIMS)
• ……. and many more that can be implemented.
xxxxx
xxxxx
xxxxx
xxxxx
xxxxx
Fig. 8,
The health screen indicating the status of a metering point, in this case samples are out of
date, hence fluid properties might be wrong and the final reading should be questioned.
4.2) Collaborative Module
This module comprises two separate functionalities:
1. A module with forum-like functionality where multiple users share knowledge and where
users can start discussions or ask for certain actions.
2. A system for automated notifications on meters or meter types where a user can subscribe
to via personalized subscriptions.
Regarding the forum-like functionality the user can view discussions by selecting a metering
point or the user can start a discussion by selecting a metering point. There are also
North Sea Flow Measurement Workshop
20-22 October 2009
LS, 12-09-2009 12
possibilities to start a “freeform” topic. In principle any user can contribute to that discussion,
however, a user should ask the administrator to be subscribed to certain discussion topics.
Regarding the actions, everybody can re quest an action, but only the discussion leader can
acknowledge and start the action. This starts a human workflow that assigns an action to
another user. The completion of the workflow can be tracked.
Metering Atlas is also able to “publish” events that a user is subscribed to (automatic events).
For this a user can be a “listener” by subscribing to those events. Per Metering Point a user
can subscribe to a pre-defined set of events, examples are:
• On change of a health status.
• On change of engineering value, a user subscribes to be notified that the value has
changed; he cannot specify an actual value.
• On change of physical property, same as above, a user subscribes to notification that
the value has changed; he cannot specify an actual value.
• On demand, the system can generate up to date lists of events where a user subscribed
(RSS feed).
5) Conclusion
Metering Data Consumers, like reservoir engineers, petroleum engineers but also operators,
have a clear requirement to be able to judge the quality of the production data that they are
using in their day-to-day business. In the ideal world the metering equipment is specified in
the engineering phase and equipment should be maintained such that it is kept in its original
specifications during the operations phase. However, in the real world this seldom happens,
metering hardware will change, fluid parameters will change, pipeline configurations might
change, samples will not be taken or calibrations and maintenance jobs will be postponed or
even cancelled. Metering Data Consumers then can go back to the various systems that are in
use to manage the various aspects of a metering set-up (InTools, EC, SAP, PI, LIMS, etc).
However, these systems often require dedicated training and are far from user-friendly for a
“once-a-week” or “once-a-month” user. Metering Atlas has the ability to retrieve all data from
these independently data applications/sources and present the data through a web-based
system to the various Metering Data Consumers. Hence, in a glance they can judge whether
the production data that they use are adequate enough for their business. With the integration
of the Metering Atlas in the Production Portal, which is the web-based view on the production
environment, not only the absolute values of production data are displayed but with just one
click away also information of the underlying measurement system are revealed.
1 Scheers, A.M. Production Measurement Management, North Sea Flow Measurement Workshop,
Oct 2002, St. Andrews, Scotland
