01 August 2003
Catching the drift
Extending pressure transmitters' calibration intervals aids nuclear plants.
By H.M. Hashemian
This dates back as far as thirty years, and it is no secret that nuclear power plants must calibrate important instruments once every fuel cycle. However, based on calibration data accumulated over these thirty years, officials determined the calibration of some instruments, such as pressure transmitters, do not drift enough to warrant calibration as often as once every fuel cycle. This fact, combined with manpower limitations and reduced maintenance budgets, has motivated the nuclear industry, associated research organizations, and national and international laboratories to develop new technologies for identifying drifting instruments during plant operation. Implementing these technologies allows calibration efforts to focus on the instruments that have drifted out of tolerance, as opposed to current practice, which calls for calibration verification of almost all instruments every fuel cycle.
Experts developed an array of technologies, referred to as "online calibration monitoring," to meet this objective. These technologies identify sensors by comparing a particular sensor's output to a calculated estimate of the actual process the sensor is measuring. The process uses methods ranging from averaging redundant instrument readings, to empirical and physical modeling of plant processes that employ neural networks for process estimation, fuzzy logic for data classification, and pattern recognition for instrument fault detection.
The online monitoring approach for extending calibration intervals of pressure transmitters in nuclear power plants received regulatory approval in France and the U.S. France approved it in 1996, and the Nuclear Regulatory Commission (NRC) approved the plan in 2000.
The NRC approval came through a safety evaluation report (SER) issued in response to a topical report from the Electric Power Research Institute (EPRI). The NRC approval in the U.S. and the French regulatory approval are contingent upon a few stipulations.
Account for common mode drift
Online monitoring for extending transmitter calibration intervals includes the stipulation that at least one transmitter from each group of redundant transmitters be calibrated at each refueling outage. Further-more, this calibration will occur on a rotating basis so every transmitter in the redundant group undergoes calibration at least once every eight years, even if a particular transmitter shows no problems during the online monitoring process. This eight-year limit came about because typically U.S. plants have four redundant transmitters, and the length of an operating cycle is about two years. The same stipulation is also true for plants in France; however some high-quality pressure transmitters in French plants get up to twelve years between calibrations.
The reason for the stipulation is to protect against common mode drift. For example, if four redundant transmitters drift at the same rate in the same direction, and the process drift is in the opposite direction at the same rate, then online monitoring may not detect the drift. By calibrating one transmitter in a redundant group of transmitters at each refueling outage, you can reveal the common mode drift. Another approach is to use physical and/or empirical modeling to track the process independent of the monitored transmitters to distinguish between instrument drift and process drift.
However, NRC did not approve physical or empirical modeling as the sole means of monitoring for common mode drift. The NRC endorsed these modeling techniques as only a supplementary means of accounting for common mode drift. The NRC left it to the user to decide which empirical and/or physical modeling algorithms to implement. In the SER, the NRC described the multivariate state estimation technique (MSET) as an example of an acceptable empirical modeling technique for online calibration monitoring. However, the NRC did not approve or disapprove any particular modeling technique in the SER and pointed out that users are responsible for verifying the validity of any technique used to draw conclusions about whether they have to calibrate a pressure transmitter.
Another important issue in the regulatory approval of the online monitoring approach is "single-point monitoring." More specifically, if you collect online monitoring data only during normal plant operation, the analysis of this data only verifies the calibration of the instruments at the monitored point. To verify the calibration of instruments at other points over their entire operating range, you must collect online monitoring data not only during normal operation, but also during startup and shutdown periods. If this is not possible, the online monitoring approach is still acceptable according to the NRC, but the allowable calibration limits must be cut by a specific allowance for single-point monitoring, as described in the EPRI report.
The NRC imposed a number of other stipulations before it approved the online monitoring approach. Organizations must develop all software modules used for acquisition and analysis of online monitoring data under a formal quality assurance program that includes software verification and validation and formal procedures to handle the online monitoring data and results.
Furthermore, users must verify the calibration of the online monitoring equipment that collects the data using calibration standards traceable to a national organization, such as the National Institute of Standards and Technology (NIST). Also, before implementing online calibration monitoring, the user must examine the historical calibration data for the plant pressure transmitters and demonstrate that the transmitters have had a good history of stable and acceptable calibrations. A plant that has a history of unacceptable drift with a significant number of pressure transmitters may not be able to use online calibration monitoring.
There is no specific requirement for the sampling frequency or the type of equipment. The options range from very infrequent data collection (i.e., once a cycle near the end to demonstrate that the transmitters are still within their allowable calibration band) to continuous sampling using the plant computer or a dedicated data acquisition system. However, if a company uses any modeling technique, computer data acquisition at relatively high sampling rates would be required. Furthermore, the signals modeled together may have to undergo sampling simultaneously as the plant operates.
Sizewell nuclear plant provided AMS Research with a report of manual calibration data for 76 pressure transmitters. The report contained transmitter drift statistics compiled from analyzing historical calibration data dating back to January 1995 when Sizewell began commercial operation. The transmitter drift characteristics came from the single calibration regression methodology (SCRM) on the historical "As-Found" and "As-Left" data the Sizewell calibration technicians recorded.
The As-Found and As-Left data records are typically in the form of nine manual calibration points (five in the direction of increasing pressure [up] and four in the direction of decreasing pressure [down]) from 0% span to 100% span in increments or decrements of 25% span. The SCRM for a single transmitter proceeds as follows:
1. The As-Found and As-Left data normalizes into percent span to allow comparable results.
2. Regression lines fit to the normalized As-Left and As-Found data using a least-squares method.
3. The slope of the As-Left line subtracts from the slope of the As-Found line and multiplies by 100 to provide any change in the transmitter span over the plant operation cycle analyzed.
4. The intercept of the As-Left line subtracts from the intercept of the As-Found line to provide any change in the transmitter's zero that might have occurred over the analyzed operating cycle.
The procedure provides a method for quantifying drift from manual calibration of Sizewell transmitters. As for online calibration monitoring, a similar method came about to compare drift information from manual calibrations with comparable drift information from online monitoring.
Officials identified a transmitter as "good" if its calibration comes within allowable limits of manual calibrations or online calibration monitoring. Otherwise, they identified the transmitter as "bad." Officials identified the allowable limits for manual calibrations and online calibration monitoring from plant set point methodology. In doing so, the allowable limits reduced by an amount equal to the overall uncertainty of the calibration method.
Excellent agreement between results of manual calibration and online calibration monitoring came about for many pressure, level, and flow transmitters at the Sizewell nuclear power plant. This effort proved the validity of online calibration monitoring to identify calibration drift in Sizewell transmitters. P
Behind the byline
H.M. Hashemian is president of Knoxville, Tenn.–based Analysis and Measurement Services Corp. His e-mail is email@example.com .
Who did the R&D?
Organizations sponsored or performed research and development (R&D) projects over the past two decades to provide new technology for online calibration monitoring and other applications in nuclear power plants. These organizations include the Electric Power Research Institute (EPRI), the University of Tennessee (UT), the U.S. Department of Energy (DOE), Argonne National Laboratory (ANL), and Analysis and Measurement Services Corp. (AMS). A summary of their efforts follows.
UT and ANL
ANL-Chicago research expanded on the SSA and produced a method called the multivariate state estimation technique (MSET). MSET is software that analyzes online monitoring data records and identifies drifting instruments. The MSET analysis compares present data patterns with data patterns obtained during normal conditions when the sensors were performing within specifications. The technique uses past and present instrument readings to produce an estimate for the process parameter, referred to as an MSET process estimate, or simply, process best estimate. The MSET process estimate undergoes an SPRT evaluation to compare the MSET process estimate with the present reading from a particular instrument. If the two values are different by more than a specific tolerance or acceptance band, it issues an alarm. The MSET/SPRT analysis is especially useful for handling Gaussian data.
The French regulatory authority approved the EDF approach in 1996. In addition, the Halden Reactor Project in Norway performed considerable research. Halden is an international laboratory, funded by over 20 countries, that conducts applied R&D in support of nuclear power plants. At Halden, developers created two software products, "PEANO" and "TEMPO," to apply to online calibration monitoring.
Process Evaluation and Analysis by Neural Operators (PEANO) uses a fuzzy classification technique to divide online monitoring data into regions or clusters. PEANO may use several different neural networks in analyzing data records from a plant. This helps provide more reliable results from neural networks compared to other approaches that fit the whole data record to a single neural network. The other product of Halden, the Thermal Performance Monitoring and Optimization system (TEMPO) is an object-oriented physical modeling tool for process estimation.
Return to Previous Page