Professional Documents
Culture Documents
Kenneth D. Elliott
Executive Vice President
Omni Flow Computers Inc.
12620 W. Airport Blvd.
Sugar Land, TX. 77478, USA
(Flowmeter Frequency)
advantages such as self-diagnostic checks and no
rotating parts. The diagnostic data is important
Flowrate
because it can warn of impending failures ‘before’
they have a major impact on the measurement.
These flowmeters however are substantially different
from other primary devices because of their reliance
on the accompanying ‘secondary’ electronics.
Time
Flowmeters used to measure liquids are frequently
calibrated by using a pipe prover or small volume Fig. 1 - Pulse Delay Decreasing Flow
prover (SVP). Proving systems encountered today
These flowmeters perform many indirect
(2005) are designed to count ‘real time’ pulses from
measurements in order to calculate a flowrate, and
a turbine or PD meter, they are not designed to
then must create an output pulse frequency that
count ‘time delayed’ ‘manufactured pulses’ from a
accurately represents this flowrate.
microprocessor. To understand the challenges, we
need to understand the characteristics of the The manufactured flow pulse ‘always’ lags the
manufactured pulse train and how it affects the actual flow in the pipe by some sampling period or
ability of the flowmeter to be proved using current calculation time, T1 (Fig.1 and 2), and may be
proving technology. further delayed or modified by filtering or damping
applied in the pulse output stage of the flowmeter’s
API Forms a Task Group
electronics, T2 (Fig.1 and 2). During a meter
The American Petroleum Institute (API) created a proving, because of this lagging pulse, we are
Task Group (TAG) to investigate how the ‘micro- counting pulses representing a volume that has
processor generated pulses’ produced by these already passed by the detector switches. The
flowmeters interact with the existing proving delayed pulses make the proving process more
technologies in use. The TAG performed controlled sensitive to flowrate changes during the prove
flow testing in a Pasadena, Texas laboratory process, and can cause an incorrect K-Factor
between July 2003 and August 2004. (pulses /unit volume or mass) to be calculated.
Most industry testing of Coriolis and Ultrasonic With delayed flow pulses, a meter calibration can
flowmeters to that date had focused on accuracy, end up introducing bias errors, instead of eliminating
repeatability, and reproducibility, over a wide range them!
of flowrates. This API testing program focused
solely on the errors and uncertainties introduced by
the flowmeter electronics used to calculate flow and
generate flow proportional pulses.
The K-Factor Bias Error outputting pulses representing this disturbance after
the first detector occurred; assuming the disturbance
Fig.1 shows what happens when a decrease in
was close enough to the first detector, and the pulse
flowrate occurs between the prover’s detector
delay was long enough.
switches. Because the flowmeter’s pulse frequency
lags the actual flowrate change, too many flow Prover Interaction With Control Systems
pulses are counted between the detector switches.
The pressure disturbance created when the ball or
The calculated K-Factor will show a positive bias
piston is launched can also interact with other flow
with decreasing flowrate. Likewise, Fig.2 below
or pressure control elements in the measurement
shows the opposite effect. In this case, with
system causing highly repeatable flow changes
increasing flowrate, too few pulses are counted
during each of the prove passes. Even when good
between the detector switches, and the calculated
run-to-run repeatability is obtained, under some
K-Factor will show a negative bias.
disturbance conditions it is possible for the
calculated K-Factor to be in error.
2nd Detector
1st Detector
T1 T2 The API Testing Process Described
(Flowmeter Frequency)
The testing sequence was as follows: In many cases it was possible to reduce the K-
Factor bias error by a factor of 10, by delaying the
1) Steady flowrate conditions detector switches used to gate the test meter pulses.
2) Using the SVP prover, determine the master
turbine meter’s K-Factor. The test sequence 1) through 8) above was
3) By a master meter prove method; determine repeated for each time constant setting tested, and
the test meter’s baseline K-Factor. at two flowrates. The data collected provided K-
4) Using the SVP prover, re-verify the master Factor bias error versus time constant settings, as
turbine meter’s K-Factor. well as providing test meter baseline K-Factors for
each time constant setting used.
Steps 1) through 4) are referred to in this
document as a ‘master meter transfer prove’. Non-ideal Flow Profile Test
The criteria used to determine the master meter With a non-ideal flow profile presented to the test
prove volume was as follows: a) The volume flowmeter, the K-Factor of the test meter was
used had to allow at least 15,000 pulses from determined at two significantly different time
master and test meter to be counted. b) To constant settings using the ‘master meter transfer’
minimize the impact of the delayed pulses, method described above. This test was performed at
prove run time was in the order of 100 times that the high flowrate condition.
of the anticipated pulse delay. The master meter
prove of the test flowmeters consisted of two 30 Satisfying the API TAG’s Testing Requirements
barrel prove runs. A 1 inch Coriolis meter was In order to address each of the API TAG’s areas of
proved using two 5 barrel prove runs. interest it was necessary to perform a large number
of SVP flowmeter provings while manipulating the
5) Determine the test meter K-Factor using an flowrate in a consistent manner during the process.
SVP with a nominal 5% step-increase in The decision was made to use an SVP for three
flowrate occurring while the displacer was reasons:
between the detectors.
6) Determine the test meter K-Factor using an a) While the TAG members were aware that
SVP with a nominal 5% step-decrease in SVP’s were not the ideal type of prover to be
flowrate occurring while the displacer was proving these flowmeter technologies, many
between the detectors. users around the world were trying to do just
7) Determine the test meter K-Factor using an that, especially users of Coriolis meter
SVP with a nominal 5% ramp-increase in technology.
flowrate occurring while the displacer was b) Because the K-Factor bias error due to the
between the detectors. pulse delay is inversely proportional to the
8) Determine the test meter K-Factor using an prove time between the detectors, proving
SVP with a nominal 5% ramp-decrease in using an SVP provides shorter prove times
flowrate occurring while the displacer was and larger more easily seen error values,
between the detectors. albeit with more uncertainty.
c) The test protocol required many proves at a
number of different timing settings and
flowrate conditions. It would not have been control valve. Temperature and pressure taps were
practical or economically feasible to use a located in a pipe section located between the
large pipe prover. reference turbine meter and the test meter, and also
at the outlet of the SVP.
The Flow Test Loop A bypass ball valve fitted with a fast acting actuator
A flow test loop was constructed as shown in Fig.3 was connected in series with a hand operated
using water as the circulating fluid. Water from a throttle valve between the pump outlet and the
large supply tank was pumped into the test loop, return line to the tank. The flow through the flow loop
passing first through the reference or master turbine was step increased or step decreased during each
With ball valve open, With the ball valve closed, adjust
adjust throttle valve for 5% control valve for required flowrate
decrease in flowrate
Flow
Control
Valve
Throttle
Valve Opening/closing the ball valve
decreases/increases flow
through the meter run
Ball T
Valve
P
1500 bbl/hr. T P
Tank Small Volume Prover
Turbine Test
Meter 20 Gallon
Pump
10D up, 5D down with tube bundles, each meter
Upstream strainer not shown.
Flow
Control
Valve
Throttle
Valve
Non ideal flow profile
created by 3 out of T
Ball plane elbows
P
Valve
Test Meter
disturbance to be applied during an SVP prove. this delay is. In one case the user selects the
One flow computer was used to prove the reference flowmeter’s time constant as ‘Short’, ‘Medium’, or
turbine meter, archive its prove data, and operate ‘Long’. In another case the time constant setting is
the bypass ball valve. A second flow computer was ‘0’ through ‘4’, and it isn’t linear! Clearly the
used to prove the test meter and archive its prove equipment manufacturers need to provide this
data. A custom programmed PC based acquisition information in a more scientific manner. Referring to
system was used to monitor the flowmeter pulse Table 1, it can be seen that in all cases, the actual
signals, the SVP launch control, and detector switch observed pulse delay was different from what the
signals. Instantaneous pulse frequency measure- manufacturers time constant settings (TC1 and TC2)
ments for both the test meter pulses, and the turbine lead you to believe it was.
Table 1 – Manufacturers Time Constant Settings versus Observed Delay
Note the variety of ways that the pulse delay is specified !
Interpreting The API Test Results Coriolis Flowmeter - K-Factor Bias Versus Pulse
Results were presented as follows: Delay and Prove Time
Referring to the data in the ‘Avg’ columns of Tables
• Data for flowmeters utilizing the same 2, 3, and 4, it can be seen that:
technologies were grouped.
• The K-Factor bias error was proportional to the
• Data within a technology were grouped response time of the flowmeter.
according to the flowmeter response time.
• The bias error was approximately the same for
• Individual flowmeters may have multiple sets of ramp or step changes in flowrate.
data included in a result set according to their
time constant settings during the tests i.e. data
from a flowmeter may be included in a data • The polarity of the bias error was consistent, and
group twice because tests were performed at in the opposite direction to the flowrate step or
multiple time constant settings falling within the ramp change.
group criteria.
• While mostly symmetrical, there was some slight
• K-Factor bias errors are normalized as a percent asymmetry in bias error versus direction of
error per percent change in flowrate, i.e. API flowrate change.
standards currently require flowrate to be held
within 5% during a prove, in that instance, the
bias errors in these tables have to be multiplied • Longer prove times appeared to equate to less
by 5 to arrive at the estimated bias error. bias for the faster response settings (Tables 2
and 3). This was not the case for the slower
The data was presented in ‘Min’, ‘Max’, ‘Average’, response settings (Table 4) where the bias error
form specifically to show trends over a range of appeared to be unaffected by double the prove
flowmeter devices. The API TAG wanted to avoid time.
singling out any individual flowmeter as being better
or worse than any other flowmeter.
The percentage values in these tables are normalized to error percent versus percent flowrate change. Max,
Min, Avg columns represent the worst case, best case and average K-Factor bias error for each flowmeter
tested in that group.
Long Filtering/Damping Time Constants (Table 4) was therefore not unexpected. Note also
Examining all of the test flowmeter data, it was found that the response time of the flowmeters being
that the K-Factor bias results became much less tested (Table 4) are similar in magnitude to the
predictable as the amount of filtering and damping actual time it took the displacer to travel between the
was increased. Configuration settings in the detector switches! It could be argued that the fluid
flowmeters that provided long filtering time represented by the flow pulses counted for these
constants, distorted the shape of the flowmeter’s proves had already passed through the prover and
response to flowrate changes as observed by therefore the prove result was highly questionable.
examination of the timing charts. As long as the
manufactured pulse from the test flowmeter was Longer Prove Times/ Larger Prove Volumes
simply a delayed version of the turbine meter Based on the results in Table 3 and 4 above, it can
reference signal, the amount of bias in the K-Factor be predicted that significantly longer prove times
resulting from the prove could be accurately between the detector switches, i.e., by increasing
predicted. The lack of a bias error reduction with the prover volume, would provide a reduction in K-
increased prove time at the slower response settings Factor bias error.
Coriolis Flowmeter - Run-to-run Repeatability improved when there is no flow disturbance
Tables 5, 6, and 7 show typical run-to-run taking place.
repeatability and also equivalent random uncertainty
values calculated per API MPMS Chapter 13 for • There is a correlation between faster flowmeter
three groups of test meters with different response responses settings and an improvement in
characteristics. Taken individually, the results of any repeatability when prove times are longer. There
one prove may have an uncertainty greater than API is little or no improvement in run repeatability
currently requires, but when the same trend is when longer response times are selected (Table
observed in over 900 proves, a higher confidence 7). It should be noted however that while the
level in the data is appropriate. prove time was approximately double at the
Referring to the values in the ‘Avg’ columns of lower flowrate, the number of pulses output by
Tables 5, 6, and 7: the flowmeter during that time remained the
same.
• Notice that the average repeatability for all three
flowmeter response settings is only marginally
Table 8 - Coriolis Meter – Baseline K-Factor Shift With Disturbed Flow Profile
Coriolis Flowmeter – Baseline K-Factor Versus Table 9 below shows very small shifts from the
Flowmeter Response Time baseline K-Factor (less than 0.01 %) for three of the
A baseline K-Factor was developed for each test flowmeters, the fourth meter showing a 0.033% shift
meter using the master meter transfer method, at the for a time constant change of 1:5.
higher test flowrate. Flowmeter time constant On a Coriolis flowmeter at least, the time constant
settings were then changed significantly and a new settings do not appear to impact the flowmeters
K-Factor developed. performance significantly when proved using the
master meter transfer method.
• Except for two instances (‘Min values’ of Table • Longer prove times generally produce a smaller
10, 600bbl/hr, step and ramp increases), the bias error.
polarity of the bias error is consistent, and in the
opposite direction to the flowrate step or ramp
change. The two non-complying values in Table
10 were re-checked and found to belong to the
flowmeter with the fastest response.
The percentage values in these tables are normalized to error percent versus percent flowrate change. Max,
Min, Avg columns represent the worst case, best case and average K-Factor bias error for each flowmeter
tested in that group.
Ultrasonic Meter - Run-to-Run Repeatability Referring to the values in the ‘Avg’ columns of
Tables 13, 14, and 15 show typical run-to-run Tables 13, 14, and 15:
repeatability and the equivalent random uncertainty
calculated per API MPMS Chapter 13 for three sets • Like the Coriolis meters tested, the average
of Ultrasonic flowmeters, data is grouped by repeatability for all flowmeter response settings
response time. was only marginally improved when there was
When comparing devices with similar response no flow disturbance taking place.
characteristics, test results showed that the
Ultrasonic flowmeters tested, exhibited run-to-run • There was a slight improvement in repeatability
repeatability values of approximately twice that of when prove times were longer. Note also that
results obtained when testing the Coriolis meters while the prove time was approximately double
(see Tables 5, 6, and 7). It should be noted that the at the lower flowrate; the number of pulses
Ultrasonic meters were larger in size (4 and 6 inch output by the flowmeter during that time
versus 3 Inch), and tested at a flowrate more than remained the same, there were however twice
50% higher than the Coriolis meters. the number of sonic measurements made.
This higher uncertainty may have impacted the
values appearing in Tables 10, 11, and 12, providing
less predictable trends.
Removing Outliers – Would it Help? A flow profile disturbance was then introduced and
While no data is presented on this subject in this K-Factors were again developed using the master
document, during the API testing attempts were meter transfer method. Referring to Table 16,
made to improve the run-to-run repeatability by flowmeter ‘A’ had fixed time constants so it was not
removing outliers, to better match the K-Factor result possible to complete this test, it did however show a
to the baseline K-Factor. This was done for a significant negative shift in K-Factor when the flow
random selection of twelve provings for Coriolis and profile disturbance was added. Flowmeters ‘B’ and
Ultrasonic flowmeters. In only one case did the ‘C’ show significant shifts in K-Factor when the flow
resultant re-calculated K-Factor better match the profile disturbance was introduced and these shifts
baseline K-Factor. The run-to-run prove count were not improved by increasing the flowmeter’s
scatter appears to be truly random and better left as time constant settings. Flowmeter ‘F’ was mistakenly
a set. tested with the tube bundles out, but interestingly
showed a relatively small shift in K-Factor at its fast
Ultrasonic Meter - Baseline K-Factor Shift With response settings. The shift increased significantly
Disturbed Flow Profile when more sonic samples were added to its timing
The possibility had been raised that, given more settings. Flowmeter ‘M’ was something of a puzzle,
measurement samples, an Ultrasonic flowmeter The number of samples used to calculate the
could better predict the flow profile in the flowrate was changed by over 50% but the observed
measurement section and therefore provide better response time did not change to match the
accuracy. To verify this, baseline K-Factors were adjustments as expected (80mS to 90mS). The K-
developed for each Ultrasonic flowmeter at two Factor shift for this flowmeter was relatively small for
different time constant settings at the higher test both time constant settings.
flowrate, using the master meter transfer method.
Table 16 - Ultrasonic Meter – Baseline K-Factor Shift With Disturbed Flow Profile
Ultrasonic Flowmeter – Baseline K-Factor Versus shifts from the baseline K-Factor (less than 0.1 %)
Time Constant Settings for flowmeters ‘B’ and ‘E’. Flowmeters ‘F’ and ‘M’
A baseline K-Factor was developed for each meter showed a significant negative shift -0.26% to –
at the higher test flowrate using the master meter 0.36% when the time constants were changed.
transfer method. Flowmeter time constant settings Flowmeter ‘M’ was tested twice at two different time
were then changed significantly and a new K-Factor constant settings, both K-Factor shifts from the
developed. Because time permitted, a third ‘K-Factor baseline are quite similar being –0.3584% versus -
versus time constant’ test was performed on .03239%.
Flowmeters ‘E’ and ‘M’. Table 17 below shows small
Table 17 – Ultrasonic - Baseline K-Factor Shift Versus Flowmeter Response Settings
The answer to both these questions is no. There is • Do the test flowmeters react in a
currently no standard way to adjust the symmetrical way to flowrate changes?
responsiveness of these flowmeters. The
manufacturers also need to consider why they The flowmeters were mostly symmetrical with the
provide such a wide range of time constants. They Coriolis meters configured for fast response being
may wish to collaborate and devise a standard way more symmetrical. In 2005 Dr. Zaki Husain of
of informing the user what they mean in real terms. Chevron/Texaco was investigating how the
When asked, “why would a user select a 2 second flowmeters react to increasing versus decreasing
time constant?” ‘Process Control’ is often the flowrates, by analyzing ‘rate of change’ of flowrates,
answer. Is it a good idea to filter the input data into versus ‘rate of change’ of pulse frequency response.
a control system, or is that what the PID tuning Preliminary analysis on one flowmeter appears to
controls are for? For analog readout use, many of show some asymmetry under some flowrate
the flowmeters provided separate time constant conditions.
adjustments.
• Is the bias error affected by flowrate (i.e.
• Is the K-Factor bias error proportional to because of more or less time between
the pulse delay? the detectors)
Yes, the bias error results for both Coriolis and The bias error is less when the prove time is longer
Ultrasonic flowmeters are, broadly speaking, and when the flowmeter reacts faster.
proportional to the pulse delay.
• Does the baseline K-Factor change when
• Is the K-Factor bias error the same for the flowmeter time constant settings are
ramp or step flowrate changes? changed?
In most cases yes, the bias error results for both This was not the case for four out of five Coriolis
Coriolis and Ultrasonic flowmeters are similar no meters tested; shifts were less than 0.01%. The fifth
matter if the flowrate is stepped or ramped. What Coriolis meter shifted only 0.03% with a 1:5 change
matters is, “is the flowrate at the first detector the in time constant. The Ultrasonic flowmeters showed
same as the flowrate at the second detector?” If the more sensitivity to changes in time constants, shifts
answer is ‘No’, then a change took place and a bias ranged from 0.086% to –0.358%.
error is possible.
• Do longer time constant settings enable Make sure that the contact person you are
the flowmeter to better handle non-ideal consulting is aware of the issues.
flow profiles?
• If a Tc2 timing component is available, set this to
No improvement was observed when testing either zero or minimum. Filtering of the output pulse
Coriolis or Ultrasonic flowmeters. Longer time has been found to degrade performance.
constants, or more sonic samples just seemed to
slow responsiveness and increase sensitivity to
flowrate changes. • Minimize flowrate variations during the prove
process. API’s MPMS Chapter 4
• Is the run-to-run repeatability dependent recommendation of 5% allowable variation may
on flow stability or prove time? not be conservative enough for these
technologies.
Overall, repeatability for both Coriolis and Ultrasonic
technologies seemed relatively unaffected by • When designing a proving system, eliminate or
significant flowrate disturbances. Repeatability at reduce the effect of flow disturbances which
steady flowrate conditions was only marginally better occur before the first detector switch, provide
versus when flowrate was unstable during a prove. sufficient pre-travel volume to ensure that any
delayed pulse response to such a disturbance
Practical Recommendations for Users of the has occurred before the first detector.
Technologies
As of spring 2005 members of the TAG were
currently studying the results of the testing program • Use as large a prover as is practical. The bias
and will in due course draft recommendations and error introduced by the delayed manufactured
language that will be included in future API pulse is inversely proportional to the prove time
standards. between the switches.
Practical steps that can be taken by users until then • If a large prover is not an option, consider
are: proving the flowmeter using the master meter
transfer method described here, using a turbine
• Optimize the responsiveness of the flowmeter by meter as a master meter.
reducing the Tc1 timing component to the
minimum recommended by the manufacturer.