UNCERTAINTY ISSUES ASSOCIATED WITH A VERY LARGE CAPACITY FLOW CALIBRATION FACILITY

T. M. Kegel (tkegel@ceesi.com) Colorado Engineering Experiment Station, Inc. (CEESI) Nunn, Colorado, USA

ABSTRACT In the spring of 1999, CEESI began operation of a natural gas calibration facility located in Iowa with a maximum flow capability in excess of 2000 pounds per second. The flow standard is a parallel array of ten turbine meters. The traceability to NIST is achieved through the CEESI Colorado facility. This paper discusses aspects of the flow measurement traceability methodology and associated uncertainty analysis. SCOPE The complete uncertainty analysis comprises many components. These include the measurements of pressure, temperature, and gas composition as well as flowrate. The data acquisition system and calculations of compressibility and flowrate also contribute uncertainty. This paper concerns only those components of uncertainty contributed by the flow measurement. The paper is divided into four sections. The first two sections briefly describe the flowmeter types and initial calibration process. The last two sections describe two measurement assurance programs that are underway. 1. FLOWMETER TYPES

The calibration facility flow standard is a parallel array of ten turbine meters. A turbine meter generates a frequency that is proportional to the angular velocity of the rotor. A K factor (K) relates the frequency (f) and volumetric flowrate (Q): K= f Q [1]

The value of K factor often varies with the flowrate. This variation, when plotted against flowrate, results in a signature curve. The most common type of flowmeter tested in the facility is the multipath ultrasonic meter. An ultrasonic meter operates by sending sound pulses in the upstream and downstream flow directions. The velocity and flowrate are calculated based on the difference between upstream and downstream propagation times, the multipath design is intended to interrogate the entire flow area with multiple transducer pairs. The ultrasonic meter, which is inherently computer based, generates a log file containing tabulated flowrate data. A frequency based output is also available that enables flowrate to be determined from a K factor. This allows the ultrasonic meter data to be handled in the same manner as a turbine meter with flowrate determined from Equation 1. In

they are identified by serial numbers 99-100502 and 99-200194 respectively. These meters are 20” and 24” in size. Two activities have been initiated to ensure that traceability is maintained on a regular basis. The uncertainty (at 95% confidence) in the SwRI results is estimated to be ±0. The uncertainties (at 95% confidence) in individual primary and secondary data points are estimated to be ±0. bearings and frequency pickoff in a self contained module that can be removed from the pressure housing. The mechanics of applying SPC to .3%. The CEESI Colorado facility operates in a “blowdown” mode which means that the air is vented to the atmosphere through the test section. First. This design allows modules to be readily moved for calibration work. 2.5%. individual modules are returned to Colorado on a regular basis for recalibration. The statistical tool applied to quantify the consistency of the check standard meter data is called “statistical process control” (SPC).15% is claimed by the manufacturer to account for slight variations in geometry between any module and pressure housing. This term is selected rather than the more commonly used “flowrate error” so as to not contribute confusion in uncertainty analysis of the system. Three of the units were also calibrated by the Southwest Research Institute4 (SwRI) to ensure that no significant systematic effects were present in the CEESI calibration process. Several calibrations of each type have been performed. The solid lines define an interval with width of ±0. The purpose of these meters is to serve as check standards that confirm consistent performance of the calibration facility. Data are obtained from one of these meters during every calibration performed on client meters. This calibration effort resulted in a total of 37 primary and 716 secondary data points for all ten turbine meters. these data indicate considerable scatter which is attributed to the test facility. the SwRI data are shown as triangles. 2 and 3. These data are well centered within the secondary results indicating that there are no appreciable systematic effects despite the significant random effects. Calibration data for one of the turbine meters (serial number 99-500359) are contained in Figure 1. Operating in this mode introduces unavoidable transients at the high flow rates required to calibrate the turbine meters. In other words. a module is periodically installed in the test section for calibration. Additional information about both meter types is available in References 1. The Iowa facility can maintain steady flow long enough to achieve stable conditions that should significantly reduce random effects. Details of these calibration processes and the associated uncertainty analysis are given in Reference 5. Second.5% respectively. A maximum additional uncertainty of 0. the consistency of a particular set of check standard calibration data is strong evidence for the validity of the corresponding client meter calibration data.the present studies the ultrasonic meter calibration results in a “flowrate offset” which is defined as the difference between indicated and actual flowrate. These transients result in random effects that contribute scatter to the data. The CEESI primary data are shown as circles. The design of turbine meter used in the facility incorporates the rotor.1% and ±0. The CEESI secondary data are shown as pluses. 3. TURBINE METER CALIBRATION The turbine meters were calibrated using air at the CEESI Colorado high flow test facility. STATISTICAL PROCESS CONTROL Ultrasonic flowmeters have been permanently installed in two of the three test sections.

and flowrate. Second. These points are also shown as being OOC in Figure 3. This is accomplished by combining in quadrature the most recent values of the control limits from Figures 2 and 3. the calibration is represented by the data point immediately prior to the OOC condition. Some out of control (OOC) conditions are recorded in the charts. Meanwhile. this case could be the 5% that doesn’t. A calibration process that is operating in a state of statistical control is producing valid data. These data correspond to the calibration of a meter that was too small for the facility. 1999 check meter number 99-100502 has been installed for 26 separate calibrations resulting in 792 data points. It is concluded that the system cannot be operated in a state of control under these conditions. The control charts from check meter number 99-200194. Within the SPC algorithm these two points were “set aside” meaning they were not used in calculating the control limits of Figure 2. The two data points indicated by the black circles represent a potential out of control condition.pressure transducer data are described in References 6 and 7. The results are shown in Figures 2 and 3 in the form of control charts. It is important to note that 95% of the data must fall within the control limits.18% at a 95% level of confidence. The important question is whether the system is truly out of control or the algorithm is generating a false signal. the same techniques are applicable to the present data. The data of Figure 3 indicate a second OOC condition. The pressure and temperature data could identify a problem The unique nature of these tests could affect the statistics of the SPC algorithim for two reasons. The next step in investigating this OOC event is to look at the performance of the unit under test (UUT) as to the source of the false signal. The interpretation of the control charts is as follows: The process of calibrating the check standard is said to be in a state of statistical control if 95% of the data fall within the limits of the control charts. Investigation of this event is continuing. The solid line represents a limit below which 95% of the data points are expected to fall. The “as found” data were in control. First. As of November 15. The data were repeated only because of the OOC condition The next source of additional information is the chart of Figure 3 which does not indicate an OOC condition. . this is an excellent indication of the reproducibility of the Iowa facility. a redesign of the facility is underway that will allow for better control under low flowrate conditions. indicate that the system was operating in control during the entire time period. only eight samples were recorded instead of 50 which is the usual practice. The resulting value is ±0. Valid data for the check standard result in valid data for the client. the data were obtained at a single flow point rather than over a range of flowrates. The solid lines represent limits within which 95% of the data points are expected to fall. temperature. The Figure 2 data indicate a return to control after a period of time when this test section was not used. which is installed in the other test section. a fact that may indicate a false signal. The chart in Figure 3 is used to check for consistency in the random effects observed with a series of calibrations. These data represent the “as left” condition for a meter that had been calibrated and adjusted. The control chart data can be used to estimate the uncertainty associated with random effects present in the calibration process. These observations further support the presence of a false signal. Difficulty in controlling the low flowrate resulted in considerable variations in pressure. The chart in Figure 2 is used to check for consistency in the random effects present during each of the calibrations. In Figure 2 the data identified by the open circles lie within the stated limits. The most important observation is that unit under test (UUT) responded as expected in the as left tests.

This process continues until a number of. The effects of improvements can be monitored. At low flowrates this change in mass can be significant relative to the total amount that passes through the meters. The check meters are located in the test section and therefore subject to this potential error. The third and fourth columns identify the number of data points and number of standard turbine meters involved in a given test. The availability of SPC data provides a good solution to address a concern that occasionally arises in flow calibration work. With the SPC data in hand it can be stated that the mass storage effect contributes no more than is ±0. statistical process control is proving to be a very powerful tool in the operation of the Iowa facility. the transfer meter has a long . The pipe network that connects the turbine array to the test section contains approximately 2000 cubic feet of gas. In summary. The result of this analysis is a set of data points (K factor and flowrate) that are treated as if they were calibration data. One or more different flowrates are then established and the process repeated.The basic principle of flow calibration is conservation of mass. A “standard turbine meter” is one of the ten meters used as flowrate standards. Under certain conditions the mass of fluid within this volume can change during a calibration. A “check meter” is sometimes installed in series with the transfer meter as an independent check of the data. either the system or the meter under test. 4. the design of test number 15 is different from the rest and is not discussed in this paper. The data are analyzed as if the transfer meter was used to calibrate each of the standard turbine meters. The transfer and check meters can be either turbine or ultrasonic meters.18% at a 95% level of confidence. It states that the mass flowrates through two flowmeters installed in series are equal if there is no mass storage or leakage between the two meters. Ideally. Is the shift due to the facility or the meter under test? A different scenario involves flow conditioners that are frequently installed upstream of flowmeters to provide a degree of isolation from upstream flow disturbances. the presence of the check meter during the test allows a numerical value to be assigned to the significance of any observed shifts in performance. 1999 is summarized in Table 1. standard meters have been tested. Observed random effects can be attributed to the proper source. After obtaining at least three data points the flow is directed through a different standard turbine meter. The test program as of November 15. This statement is valid only under the conditions where statistical control is maintained which excludes some low flowrate conditions. One scenario is as follows: a meter returned for recalibration exhibits a shift in performance relative to the last calibration. TURBINE SUBSTITUTION TESTING The discussion contained in this section can be confusing. The test involves establishing a steady flow through one of the standard turbine meters and the transfer meter which is installed in the test section. the following term are defined in an attempt to reduce the confusion. A turbine substitution test is based on comparisons made to a well characterized transfer meter. One unavoidable design feature of the Iowa facility is a large physical separation between the turbine meter bank and test section. Steady operation is confirmed. Testing with and without flow conditioners may result in systematic shifts in meter performance. and preferably all. The first two columns uniquely identify each test. A “transfer meter” is installed in the test section for the purpose of obtaining data on the performance of the various standard turbine meters.

The substitute test data based on a turbine transfer meter are symbolized by circles. . Two standard deviations are calculated for each data set.990 < K < 5. each has been calibrated with at least 50 data points. They are within ±0.15% over the 300-2500 acfm range. When a pair of meters are available as calibration standards there are two choices as to how the data should be interpreted. the turbine meter was selected as the transfer meter.consistent calibration history. The eleven ultrasonic meters identified as USM A to USM K are customer meters. therefore the results are independent of any meter changes between tests. The meters identified as TRB A and TRB B (serial numbers 98-070103 and 98-070100) have long consistent calibration histories. In the three subsequent tests. In reality. The first option is to average the results from both meters. The second option. The results of the integrity tests are contained in Table 2. The analysis does not identify variations from one test to the next. The mean values of the two sets of standard deviations are contained in the third and fifth columns. which was selected for the present study. CEESI Colorado and SwRI. including correlating with other data. The standard deviations of the mean values are contained in the fourth and sixth columns. The fifth column identifies the transfer meter or meters used for the test. The final column in Table 1 identifies the flowrates that were investigated expressed as percentages of the full scale capacity (2500 acfm) of a standard turbine meter. flow control. A quantitative analysis. the abbreviations “USM” and “TRB” refer to ultrasonic and turbine meters respectively. is required to gain further insight as to the meaning of the observed turbine meter trending. An integrity test was developed in order to help designate standard and check meters for each of the substitution tests. an opportunity often arises to use the meter under test as a transfer meter when time is not available to install the well characterized transfer meter. Areas for investigation include pressure fluctuations. are symbolized by pluses. The integrity test results have thus far been used only qualitatively to select from the two meters. The meter exhibiting the lowest standard deviations is designated as the calibration standard for that test. The exception. This test compares the two meters based on the standard deviations of multiple data points taken at a common flowrate. For eight of the tests. A data set consists of data points taken sequentially at the same flowrate with flow going through the same standard turbine meter. the same two meters were used. The meters identified as USM E and USM F were available for more extensive testing. The selection of flowrate values for a particular test represents a compromise between facility availability and ideal conditions. Once again it is clear that the turbine data exhibit less scatter than the ultrasonic data. Typical turbine substitution test results are contained in Figure 4 for turbine meter number 99500359. The three data points from substitution test eleven predict 5. It is clear from the data of Table 2 that the turbine meter is more repeatable than the ultrasonic. the turbine meter repeatability improves steadily. the values in the first column are the same as those in Table 1. the interval defined by the solid lines. test number eleven. and operator variability. The second column identifies data sets. indicates nearly an order of magnitude increase in turbine meter non-repeatability. It is tempting to dismiss the significance of this observation because the ultrasonic results are based on six different meters while only two turbine meters were used. data are available from both ultrasonic and turbine meters. The data for flowrates lower than 300 acfm exhibit considerably greater scatter. The Figure 1 data. is to designate one of the two as the transfer meter and use the second as a check meter. For all of the tests except one.997 for Q ≈1250 acfm. For three of these tests. one each for the turbine and ultrasonic. the ultrasonic data are symbolized by triangles.

Finally. 2. This paper is not intended to be a complete uncertainty analysis. these values were obtained at low flowrates where the turbine meter is subject to greater scatter. REFERENCES 1. some level of agreement. Three important conclusions are made based on these analyses. Does this shift represent a physical change in the meter itself. 1981. Transmission Measurement Committee Report 7. R. 1998. There are unresolved issues regarding the shifting of calibration data and the underlying physical phenomena that may cause such shifts. and gas composition have not been determined in detail. VA. These results cast some doubt on the validity of substitution test eleven. Miller. There is.3% value is equal to the uncertainty in the calibration. Transmission Measurement Committee Report 9. only a refinement that is a result of better data. uncertainty in determining pressure. Substitution testing allows for the separation of individual turbine meter performance. Therefore. some of the uncertainty components of the statistical process control and turbine substitution testing have not been evaluated. SUMMARY A detailed description of two measurement assurance activities has been presented. Finally. the ultrasonic data are within ±0. “Measurement of Fuel Gas by Turbine Meters. the 0. Flow Measurement Engineering Handbook. while the turbine based substitution data indicate a systematic effect. Further insight is gained from the ultrasonic based substitution data.18%. These data are a demonstration of basic repeatability of the calibration facility. McGraw-Hill. While the scatter band is wider with the ultrasonic data the mean values agree with the turbine data to within 0. they are consistent with the scatter observed with the original calibration. Such programs are important to asses systematic effects. Arlington. It is anticipated that these issues can be more clearly resolved upon applying similar analysis to a number of other standard turbine meters. Second.” American Gas Association.3%. the Iowa facility exhibits smaller random effects than the Colorado facility. In the case of the primary data. . In summary. Second. the difference is approximately 0. First. First. temperature. The primary and SwRI data lie outside the interval of Figure 4. the turbine substitution test provides data with less scatter than has been previously available. It is interesting to note that if the turbine had been used as the transfer meter in substitution test eleven.If these points are neglected. measurement comparison programs with other facilities8 have been performed but are not discussed. 3. “Measurement of Gas by Multipath Ultrasonic Meters. These observations support a proposal that there is no shift in the calibration data. In the case of the SwRI data. several important components are absent. or availability of calibration data with less scatter? One relevant observation is that the interval of Figure 4 is fully contained within interval of Figure 1. the resulting data points would have fallen outside the bands shown in Figure 4. Statistical process control (SPC) tracks the performance of groups of standard turbine meters. Arlington. VA. The data of Figure 4 indicate an apparent shift in the performance of this turbine meter. 1983.03%. the system is capable of operating with reproducibility of ±0. the turbine meters exhibit better repeatability than ultrasonic meters. therefore.” American Gas Association. W.30% over the 300-2500 acfm range. This was determined by two independent methods using different meter types.

“Uncertainty Analysis of the CEESI Iowa High Flow Test Facility.21 52 9 9. M. 6220 Culebra Rd. 5. T. TRB B Flowrate [%FS] 31. 1999.TRB A USM J USM K. Instrument Society of America. 1996. 94 58 80 61.TRB A USM I.TRB A USM F. T. Denver. San Antonio.” Proceedings of the 45th International Instrumentation Symposium. Recommended Practice RP-15. TX. Southwest Research Institute.. 75 5. 6. 78238-5166.TRB A USM B USM C USM D USM E USM F. “Statistical control of a pressure instrument calibration process. 1999. 12 . vol 35. Instrument Society of America... Kegel.TRB A USM E USM G. CO.” ISA Transactions. T. CO. Boulder. M. 49 50 34 12 27 80 . 7. M. Kegel.TRB A USM F.” National Conference of Standards Laboratories. 28 28.. “Guide for Interlaboratory Comparisons. 8.4. Kegel. Table 1: Turbine Substitution Test Program Test Number 2 3 4 5 6 7 8 9 10 11 12 13 14 16 Test Date 12-Jul-99 30-Jul-99 10-Aug-99 23-Aug-99 02-Sep-99 09-Sep-99 10-Sep-99 11-Sep-99 22-Sep-99 28-Sep-99 05-Oct-99 06-Oct-99 13-Oct-99 12-Nov-99 Data Points 81 26 17 32 62 54 55 45 45 28 16 32 39 13 Turbine Meters 9 7 7 4 5 9 9 4 5 9 5 7 9 4 Transfer Standard(s) USM A.” Proceedings of the 4th International Fluid Flow Symposium. 1999. “Statistical control of a differential pressure instrument calibration process.TRB A USM H. 60 .

043 0. Dev.040 0.Table 2: Turbine Substitution Testing – Integrity Statistics Test Number 2 7 8 9 11 12 13 16 Data Sets 28 15 18 15 9 5 8 4 Turbine Mean [%] 0.111 0.015 0.103 0. [%] 0. [%] 0.02 Secondary SwRI Primary K-Factor [pul/cu ft] 5.134 0.025 0.046 0.98 5.039 0.135 0.086 0. Dev.041 0.144 0.084 0.059 0.94 5.024 Ultrasonic St.90 0 5 10 Volume Flow [acfh] Hundreds 15 20 .084 0.055 0.140 0.056 0.336 0.032 0.145 0.424 0.048 0.020 0.010 Figure 1: Initial Calibration Data for Turbine Meter 99-500359 6.095 0.021 0.010 Ultrasonic Mean [%] 0.012 Turbine St.033 0.

18 0.09 0.06 05-Sep 22-Sep 09-Oct Date 26-Oct 12-Nov Figure 3: Statistical Process Control Chart 0.4 Between Standard Deviation [%] 0.Figure 2: Statistical Process Control Chart 0.15 0.2 -0.2 0.4 05-Sep 22-Sep 09-Oct Date 26-Oct 12-Nov .12 0.0 -0.21 Within Standard Deviation [%] 0.

Figure 4: Turbine Substitution Test Data for Turbine Meter 99-500359 6.98 5.94 Air & SwRI 5.02 K-Factor [pul/cu ft] 5.90 0 5 USM Trb 10 15 Volume Flow [acfh] Hundreds 20 25 .