Ultrasonic flowmeters undergo accuracy, repeatability tests

Terrence A. Grimley
Southwest Research Institute
San Antonio

The tests were conducted in baseline and disturbed-flow installations to assess baseline accuracy and repeatability over a range of flowrates and pressures.

Results show the test meters are capable of accuracies within a 1% tolerance and with repeatability of better than 0.25% when the flowrate is greater than about 5% of capacity. The data also indicate that pressure may have an effect on meter error.

Both meter types showed shifts of 0.4% over a 650-psi static pressure variation but remained within a 1% tolerance. Flow calibration of each meter at or near the field operating conditions may be necessary for reducing any measurement bias.

More understanding of the relationships between various operating parameters and meter design is needed before a dry calibration will achieve bias errors on the order of 0.2%.

Results further suggest that both the magnitude and character of errors introduced by flow disturbances are a function of meter design. Shifts of up to 0.6% were measured for meters installed 10D from a tee (1D = 1 pipe diameter).

Better characterization of the effects of flow disturbances on measurement accuracy is needed to define more accurately the upstream piping requirements necessary to achieve meter performance within a specified tolerance.

Reduced station costs

Ultrasonic flowmeters are of interest because of their flow capacity and rangeability, with reported accuracies similar to other traditional metering methods such as orifice and turbine meters.^{1 2}

Because the gas may be metered over a bulk velocity range from 1 to 100 fps, a single ultrasonic flowmeter can replace parallel orifice meter runs or can be installed in place of a turbine meter to extend the metering range. This provides a potential for reduced capital cost for meter stations.

The relationship between the measured transit time of an ultrasonic pulse and the average velocity along the pulse path has been well described.^3-5

Although this basic relationship is common to all transit-time ultrasonic flow meters, considerable variation exists among path configuration, transducer type and placement, transit time measurement algorithm, and flow calculation method used by the different commercially available meters.

These differences result from the use of different strategies to achieve the meter's target accuracy, typically stated as 0.5-1.0%.

One multipath ultrasonic meter, designed around the use of four-parallel chordal paths, exploits a numerical integration technique to form a weighted average of the path measurements without an assumption of the velocity profile.⁶

Another meter arrangement utilizes the measurements from three single-reflection diametral paths and two double-reflection chordal paths to form an average velocity based on a combination of theoretically and experimentally determined weighting factors.⁴

Differences in meter configuration and data processing methods can affect meter accuracy, rangeability, repeatability, and susceptibility to error as a result of less-than-ideal installation configurations.

Numerous tests have been conducted elsewhere^{7 8} on a variety of upstream disturbances and on different meter types, but the need remains for more information on flowmeter performance. The purpose of these tests was to expand the information available on the performance of ultrasonic flowmeters over a range of operating conditions and configurations.

Test methods

Tests for this program were conducted in the GRI MRF High Pressure Loop (HPL) at Southwest Research Institute, San Antonio. Test meters were installed in the 12-in. reference flow leg of the MRF and tested with transmission-grade natural gas.

Data were collected simultaneously on the ultrasonic meters and on the HPL critical flow nozzle bank, which served as the flow reference. The five binary-weighted sonic nozzles were calibrated in situ at different pressures against the HPL weigh tank system.⁹

An on-line gas chromatograph and equations of state from AGA Report 810 were used to determine gas properties for all calculations. Static pressure and temperature were measured at the meters.

The volumetric flowrate reported by the ultrasonic meter was acquired with different methods, depending on the meter options available from the manufacturer.

For Meters A1 and A2, a "calibration mode" was used, whereby the meter internally totalized the gas volume and the time during which a specific register was toggled. The average flowrate was then calculated from the totalized numbers.

For Meter B, reported values of actual flowrate (which were provided at a rate of one/sec) were averaged to determine the average volumetric flowrate. Speed-of-sound measurements taken by both meters were also recorded.

A typical test sequence consisted of recirculating gas through the flow loop long enough to allow the gas temperature to stabilize. Steady flow was established by selecting and choking different nozzle combinations.

A test point consisted of the average values of flow rate and other variables computed over 90-120 sec. Test points were repeated a minimum of five to ten times to calculate an average value and standard deviation.

Data were collected simultaneously from other flow-measurement devices in the flow loop (typically one 10-in. orifice meter and two 12-in. turbine meters), which aided in establishing the validity and consistency of the data.

The ultrasonic meters were tested as received from the manufacturers, and all tests were conducted without flow conditioners. Meters A1 and A2 were of the four-chordal path design and "dry calibrated" by the manufacturer.

The dry calibration included measuring the various lengths required for the calculations and characterizing the timing delays for the ultrasonic transducer pairs. These meters had not been exposed to flowing gas before being installed at the MRF.

Meter B was a five-path design (two double-reflecting chordal paths and three single-reflecting diametral paths) that had previously been flow calibrated and tested at several European laboratories.

As received, the meter was set up for approximately 400-psi operating conditions by the specification of density and viscosity values that are used for a Reynolds number calculation, which is part of the algorithm for calculation of the flowrate.

Baseline tests

The baseline tests were conducted with the meters installed approximately 68 diameters (D) downstream of an in-plane tee, at meter location No. 3 (Plan view, Fig. 1 [49978 bytes]).

The pipe 17D upstream of the meter had a 12-in. ID, which matched the diameters for Meters A1 and A2 and was slightly smaller than the 12.059-in. ID of Meter B.

The baseline testing for Meters A1 and A2 was conducted with the two meters in series. The first meter was located at Location 3 (Fig. 1 [49978 bytes]); the second, 10D downstream from the first. The meters were oriented so that the chords were aligned in a horizontal plane. Baseline testing on Meter B was conducted while Meters A1 and A2 were installed at Locations 1 and 2 (Fig. 1).

Fig. 2 [14648 bytes] shows the baseline performance of Meters A1, A2, and B over the range of flowrates achievable in the MRF (30-40% of full scale for a 12-in. meter). The error percentages shown are calculated relative to the nozzle bank reference flowrate.

It is apparent from the curves that all the meters are well within a 1% error tolerance, and for all but the lowest velocity, the points fall within a 0.5% band. The error bars shown on the data represent two standard deviations calculated from the data scatter at each velocity.

The repeatability is similar for all the meters, having a value of less than 0.25% above approximately 5 fps.

At low velocities there tends to be more scatter in the data, which is likely an effect of the resolution of the transit-time measurements.

Fig. 2 also shows that for Meter A1, a small zero (0) offset in the meter is indicated by an increase in meter error as the velocity approaches zero. The error changes from 0.3% when the velocity is greater than 5-10 fps to -0.5% as the velocity approaches 1 fps.

It is important to note that the zero offset in this meter, which would normally be eliminated during the dry calibration at the factory, was knowingly left in.

Effect of pressure

Baseline testing was conducted at line pressures of 250, 400, and 900 psia to assess any effect of pressure on the meter calibration. Figs. 3 [14474 bytes] and 4 [14584 bytes] show the effect of pressure on the performance of Meters A1 and B. (Meter A2 showed similar trends.)

While the repeatability appears to be independent of pressure, the data reflect a shift in the average error of about 0.4% over the 650-psi range of pressures tested.

The calculation method employed by Meter B uses corrections that depend on the velocity profile and therefore on the Reynolds number. Because the density and viscosity values used by the meter were set to values appropriate for 400-psi operation, a portion of the pressure shift for Meter B can be attributed to the fixed values preventing the meter from making proper corrections.

Additional tests conducted with Meter B at 250 and 900 psia, with values for the density and viscosity consistent with the pressures, showed shifts of roughly 0.1% towards the 400-psia line for the 250-psia case and essentially no change for the 900-psia case.

The results suggest that the Reynolds number-based profile correction does not completely account for the effect of pressure on the meter error.

Calculations have shown that the theoretical performance of the integration method used by Meters A1 and A2 is largely independent of velocity profile over the Reynolds number range (and therefore the pressure range) involved in these tests.5 The calculations show a theoretical variation of less than 0.05%.

The change in the path length and meter body diameter as the pressure is increased also introduces an error into the measurement. The estimated change in calibration as a result of deformation, however, is less than 0.05% for the 650-psi change in pressure.

These results run counter to the findings of other testing in which no consistent dependence on pressure was identified (although there were variations in the mean error at different pressure levels).^{7 8} The effect of pressure needs to be investigated further to fully explain the data.

Speed of sound

Since transit-time ultrasonic flow meters measure the speed of sound independent of the flowing velocity, the speed of sound can be used for diagnostic purposes in field applications without the need for an independent flow reference.

Comparisons can be made between the value of sound speed measured by the meter and that determined from pressure, temperature, and gas-composition measurement. Agreement between multiple paths and the value calculated from the gas conditions can be used to verify consistency in the meter setup and proper operation of the paths.

It should be recognized, however, that because of the independence of the measurements, good agreement between the calculated speed of sound and that reported by the meter is insufficient for accurate flow measurement.

Fig. 5 [14653 bytes] shows the percent error between the speed of sound reported by the meters and that derived from AGA Report 8 (1985) density calculations. The figure shows that the error is typically less than 0.2% for both test meter types, with each meter type having its own bias relative to the calculated value.

The variation in speed of sound for the data shown in Fig. 5 was limited and ranged from 1,360 to 1,410 fps. The data for Meter A1 show more scatter than the Meter B data. Although not presented, data for Meter A2 were similar to those for Meter A1.

There appears to be no large dependence on pressure, with about 0.1% shift between the 900-psi data and the results for the other pressures.

Disturbance tests

Fig. 1 [49978 bytes] showed that the location of the meters for the disturbance tests was 10D downstream of a tee (that is, meter Location 1 on Fig. 1) that was located just downstream from a 12-in. diameter full-port ball valve. Tests were conducted with the ball valve open, and at two partially closed positions (to increase the level of flow disturbance entering the test meter).

The piping between the tee and the meter was 12-in. Schedule 80, nominally 113/8 in. ID. Therefore, when located at the position 10D from the tee, the meter was subjected to a combination of the effects of the 5/8-in. concentric step in diameter (an 11% increase in area), the tee, and for some tests, a partially closed valve.

Tests were conducted at the same three pressures used for the baseline calibrations. A set of scoping tests was also performed with the meter located 23D downstream from the tee (meter Location 2 on Fig. 1.)

Meter A1 was tested in two different orientations relative to the tee. The 0° position had the chords aligned in a horizontal plane along with the tee. The meter was also rotated 90°, so that the chords were aligned in a vertical plane. Meter B was tested in only one orientation.

Fig. 6 [15872 bytes] shows the results for Meter A1, when located 10D and 23D downstream of the tee with the valve fully open. Although the data appear to differ from the baseline, additional data collected just before the disturbance tests began revealed that a shift in the baseline data had occurred. (It was later discovered that this shift may have resulted from a failing chord.)

The data for both the 10D and 23D positions lie within 0.1-0.2% of the revised baseline. When the chords are vertical, the data shift by -0.5% relative to the revised baseline.

The fact that there are differences that depend on the chord orientation relative to the disturbance is not surprising, and these differences have been measured by others.⁷

Since the transit time measured by the meter for a particular path is proportional to the average velocity along the path, the paths cutting across the disturbance tend to average out the effect of the disturbance.

When the paths are vertical, and the primary direction of velocity disturbance is horizontal, the ability of the integration method accurately to resolve the average velocity has more of an effect on the results.

Fig. 7 [16089 bytes] shows the effect of the partially closed valve on the meter performance at 10D for both chord orientations. The valve position had little effect on the average error of the repeated runs, but there was a small increase in the data scatter, suggesting that turbulence levels were increased as a result of closing the valve.

Table 1 [8056 bytes] lists, for velocities greater than 9 fps, the average percent error and twice the average standard deviation (2s) of the data for the different axial installation locations, meter orientations, and valve positions.

Fig. 8 [14269 bytes] gives the results for Meter B installed 10D down from the tee. The results show a shift of about 0.6% from the baseline with the valve fully open and a shift of about 0.4% from the baseline when the valve is partially closed.

The result is surprising because the partial blockage by the valve should increase the flow distortion. Table 2 [5935 bytes] shows that although the valve position did influence the absolute value of the error, it did not have a significant effect on the data scatter.

Some data showed a deviation of 0.3% relative to the baseline when the meter was installed 10D down from an elbow.⁷ The effect of the step change in diameter may account for the difference in results; however, time constraints on the availability of the meter prevented additional testing to isolate the effects.

Disturbance tests for both meters were also conducted at 250 and 400 psia and yielded results similar to those obtained at 900 psia. Meter B was not tested farther downstream of the tee or at other orientations.

References

1. Beeson, J., "Ultrasonic Metering-A Field Perspective," AGA Operations Conference, 1995, Las Vegas.

2. Sakariassen, R., "Why We Use Ultrasonic Gas Flow Meters," North Sea Flow Measurement Workshop, 1995, Lillehammer, Norway.

3. Freund, W.R., Jr., and Warner, K.L., "Performance Characteristics of Transit Time Ultrasonic Flow Meters," Third International Symposium on Fluid Flow Measurement, 1995, San Antonio.

4. Drenthen, Jan G., "The Q.Sonic Ultrasonic Gas Flowmeter for Custody Transfer," Third International Symposium on Fluid Flow Measurement, 1995, San Antonio.

5. Van Dellen, K., "Multipath Ultrasonic Gas Flow Meters," Daniel Industries.

6. Groupe Europen de Recherches Gazires, "Present Status and Future Research on Multi-path Ultrasonic Gas Flow Meters," GERG Technical Monograph No. 8, 1995.

7. Van Bloemendaal, K., and van der Kam, P.M.A., "Installation Effects on Multi-Path Ultrasonic Flow Meters: The 'ULTRAFLOW' Project," Third International Symposium on Fluid Flow Measurement, 1995, San Antonio.

8. Vulovic, F., Harbrink, B., and van Bloemendaal, K., "Installation Effects on a Multipath Ultrasonic Flow Meter Designed for Profile Disturbances," North Sea Flow Measurement Workshop, 1995; Lillehammer, Norway.

9. Park, J.T., Behring, K.A. II, and Grimley, T.A., "Uncertainty Estimates for the Gravimetric Primary Flow Standards of the MRF," Third International Symposium on Fluid Flow Measurement, 1995, San Antonio.

10. American Gas Association, "Compressibility of Natural Gas and Other Related Hydrocarbon Gases," A.G.A. Transmission Measurement Committee Report No. 8, 1985.

The Author