Automated surveillance of reactor coolant pump performance

An artificial intelligence based expert system has been developed for continuous surveillance and diagnosis of centrifugal-type reactor coolant pump (RCP) performance and operability. The expert system continuously monitors digitized signals from a variety of physical variables (speed, vibration level, motor power, discharge pressure) associated with RCP performance for annunciation of the incipience or onset of off-normal operation. The system employs an extremely sensitive pattern-recognition technique, the sequential probability ratio test (SPRT) for rapid identification of pump operability degradation. The sequential statistical analysis of the signal noise has been shown to provide the theoretically shortest sampling time to detect disturbances and thus has the potential of providing incipient fault detection information to operators sufficiently early to avoid forced plant shutdowns. The sensitivity and response time of the expert system are analyzed in this paper using monte carlo simulation techniques