September 15, 2006

How to conduct a plant performance test

Tina L. Toburen, PE, and Larry Jones, McHale & Associates Inc.

Understanding test uncertainty

Uncertainty is a measure of the quality of the test or calculation result. A pretest uncertainty analysis can be used to design a test to meet predefined uncertainty limits. A post-test uncertainty analysis should be performed to verify that those uncertainty limits were met and to determine the impact of any random scatter recorded in the test data.

Each input to the calculation must be analyzed for its impact on the final result. This impact is identified as the sensitivity of the result to that input. For example, if inlet air temperature changes by 3 degrees F, and the corrected output changes by 1%, the sensitivity is 1% per 3 degrees F or 0.33%/degree F.

The instrumentation information is used to identify the systematic error potential for each input. For example, a precision 4-wire resistance-temperature detector can measure inlet air temperature with an accuracy of +/- 0.18F, based on information provided by the manufacturer and as confirmed during periodic calibrations.

During a test run, multiple recordings are made for any given parameter, and there will be scatter in the data. The amount of scatter in the data is an indication of the random error potential for each input. For example, during a 1-hour test run, the inlet air temperature may be recorded as an average of 75F, with a standard deviation in the measurements of 0.6F.

If more than one sensor is used to measure a parameter, there also will be variances between sensors based on location. These variances may be due to the variances either in the instrumentation or in the actual parameter measured. For example, if air temperature is being measured by an array of sensors, there may be effects due to ground warming or exhaust vents in the area, either of which would affect the uncertainty of the bulk average measurement. These variances will affect the average and standard deviation values for that parameter. Spatial variances are added into the systematic error potential, based on the deviation of each location from the average value for all locations.

Now that we've defined the three separate inputs to the uncertainty determination—sensitivity (A), systematic error potential/uncertainty (B), and random error potential/uncertainty (C)—it's time to put on our statistician's hats.

The terms can be combined in the following equation:

Uncertainty = SQRT[(A x B)² + ( t x A x C)²]

The "t" value on the right side of the equation is known as the Student-t factor and is based on the number of degrees of freedom (or number of data points recorded) in the data set. For a 95% confidence interval and data taken at 1-minute intervals for a 60-minute test run, the value of "t" is 2.0. If data are taken less frequently (such as at 2-minute intervals), fewer recordings are made and therefore either the test run must be longer (which is not recommended, because ambient conditions may change) or the value of "t" will increase.

The example given above is for a single parameter, such as inlet air temperature, and its effect on corrected output. For each correction made, the same process must be carried out to determine the sensitivity, systematic uncertainty, and random uncertainty of the corrected result on that correction parameter (such as barometric pressure and relative humidity).

Once each individual uncertainty has been identified, they can be combined to determine the overall uncertainty of the corrected result. Combining the individual uncertainties is a three-step process:

• Determine the total systematic uncertainty as the square root of the sum of the squares for all the individual systematic uncertainties.
• Determine the total random uncertainty as the square root of the sum of the squares for all the individual random uncertainties.
• Combine the total systematic uncertainty and total random uncertainty as follows: Total uncertainty = SQRT[(systematic_total)² + ( t x random_total)²].

The result of the analysis is an expression stated in terms of the uncertainty calculated for an individual instrument or the overall system. We might normally say, "The inlet air temperature is 75F," but when including an uncertainty analysis of a temperature measurement system, a more accurate statement would be, "We are 95% certain that the inlet air temperature is between 74.6F and 75.4F."

Once again, the value for "t" will depend on the design of the test, including the number of multiple sensors and the frequency of data recordings. Additional information on the Student-t factor as well as a discussion of how to determine uncertainty can be found in ASME PTC 19.1 (Test Uncertainty).

Conduct the test

The performance test should always be conducted in accordance with its approved procedure. Any deviations should be discussed and documented to make sure their impact is understood by all parties. If the test is conducted periodically, it is important to know what deviations were allowed in previous tests to understand if any changes in performance might have been due to equipment changes or simply to the setup of the test itself.

Calibrated temporary instrumentation should be installed in the predetermined locations, and calibration records for any plant or utility instrumentation should be reviewed. Check any data collection systems for proper resolution and frequency and do preliminary test runs to verify that all systems are operating properly.

The performance test should be preceded by a walk-down of the plant to verify that all systems are configured and operating correctly. It's important to verify that plant operations are in compliance with the test procedure because equipment disposition, operating limits, and load stability affect the results. Data can then be collected for the time periods defined in the test procedure and checked for compliance with all test stability criteria. Once data have been collected and the test has been deemed complete, the results can be shared with all interested parties.

Because the short preliminary test may be the most important part of the process, be sure to allow sufficient time for it in the test plan. The preliminary test must be done during steady-state conditions following load stabilization or when the unit is operating at steady state during the emissions testing program. The preliminary test has three objectives: to verify all data systems, to make sure manual data takers are reading the correct instruments and meters, and to have the data pass a "sanity check."

After the test data have been collected, the readings should be entered into the correction model as soon as possible and checked for test stability criteria (as defined by the test procedure). At this point, depending on the correction methods, the test director may be able to make a preliminary analysis of the results. If the numbers are way out of whack with expected values, a good director will start looking for explanations—possibly, errors in the recorded data or something in the operational setup of the unit itself. Though everyone is concerned when a unit underperforms, a unit that performs unexpectedly well may have problems that have been overlooked. For example, a unit that corrected test results indicate has a 5% capacity margin may need to have its metering checked and rectified, or it may have been mistuned to leave it in an overfired condition.

Although an overtuned gas turbine may produce more megawatt-hours during initial operations, the gain comes with a price: increasing degradation of the unit's hot section, shortening parts life and increasing maintenance costs. The most common mistake in testing is acceptance of results that are too good. If results are bad, everyone looks for the problem. If the results are above par, everyone is happy—especially the plant owner, who seems to have gotten a "super" machine. However, there's a reason for every excursion beyond expected performance limits—for better or worse.

If all the pretest checks are done properly, the actual performance test should be uneventful and downright boring. It should be as simple as verifying that test parameters (load, stability, etc.) are being met. This is where the really good performance testers make their work look easy. They appear to have nothing to do during the test, and that's true because they planned it that way. Having done all the "real" work beforehand, they can now focus on making sure that nothing changes during the test that may affect the stability of the data.

Analyze the results

Almost immediately after the performance test (and sometimes even before it is complete), someone is sure to ask, "Do you have the results yet?" Everyone wants to know if the unit passed. As soon as practical, the performance group should produce a preliminary report describing the test and detailing the results. Data should be reduced to test run averages and scrutinized for any spurious outliers. Redundant instrumentation should be compared, and instrumentation should be verified or calibrated after the test in accordance with the requirements of the procedure and applicable test codes.

The test runs should be analyzed following the methods outlined in the test procedure. Results from multiple test runs can be compared with one another for the sake of repeatability. PTC 46 (Overall Plant Performance) outlines criteria for overlap of corrected test results. For example, if there are three test runs, a quality test should demonstrate that the overlap is well within the uncertainty limits of the test.

Once test analysts are satisfied that the results were proper, the test report can be written to communicate them. This report should describe any exceptions to the test procedure that may have been required due to the conditions of the facility during the test. In the event that the results of the performance test are not as expected, the report may also suggest potential next steps to rectify them.

For sites where the fuel analysis is not available online or in real time, a preliminary efficiency and/or heat rate value may be reported based on a fuel sample taken days or even weeks before the test. Depending on the type and source of the fuel, this preliminary analysis may be significantly different than that for the fuel burned during the test. It's important to understand that preliminary heat rate and efficiency results are often subject to significant changes. Once the fuel analyses are available for the fuel samples taken during the test, a final report can be prepared and presented to all interested parties.

—Tina L. Toburen, PE, is manager of performance monitoring and Larry Jones is a testing consultant for McHale & Associates. Toburen can be reached at 425-557-8758 or tina.toburen@mchale.org; Jones can be reached at 865-588-2654 or larry.jones@mchale.org.