Instrument calibration verification

The redundant sensors typically found in nuclear power plants present an opportunity for verifying the calibration of sensors using non-intrusive signal validation techniques. Such techniques allow periodic surveillance (online or offline) that can quantitatively trend the performance characteristics of sensors. This allows I&C personnel to (1) identify incipient calibration failures, (2) identify sensors with chronic calibration drift problems, (3) prioritize maintenance activities, and (4) potentially adjust calibration schedules based on the historical behavior of sensors. A consistency checking algorithm compares the signals from redundant sensors against their respective tolerances to calculate a process variable estimate, and to generate a sensor inconsistency index. A data-driven empirical modeling scheme is used to provide an additional, diverse, point of redundancy, and to check against possible common mode calibration failures. A sequential probability ratio test is applied to generate two additional performance characteristics that aid in tracking and predicting calibration drift. Application of the methodology to plant sensor data with a bias error artificially added to one of the signals demonstrates an ability to detect the error within the prescribed tolerances of the sensors.