Automated long-term surveillance of a commercial nuclear power plant

A pattern-recognition system is described for monitoring nuclear reactor signals. The system is based on detecting deviations from baseline signatures identified during normal plant operation. The capabilities and limitations of this pattern recognition approach were investigated during a 2-1/2-year series of continuous online experiments at the Sequoyah-1 Nuclear Power Plant. The principal purpose of these experiments were (1) to investigate the data-reduction aspects of the system in the sense that it stored only those signals flagged as being different from their corresponding baseline signatures; (2) to evaluate system behavior over a long period of operation; and (3) to determine the applicability of the discriminants to various types of signals commonly monitored in a nuclear plant.<>