Study of methodology diversification in diagnostics

not registered ; not registered; Yoshikawa, Shinji ; Hasegawa, Makoto

There are several research activities to enhance safety and reliability of nuclear power plant operation and maintenance. We are developing a concept of an autonomous operation system where the role of operators is replaced with artificial intelligence. The purpose of the study described in this report is to develop a operator support system in abnormal plant situations. Conventionally, diagnostic modules based on individual methodology such as expert system have been developed and verified. In this report, methodology diversification is considered to integrate diagnostic modules which performance are confirmed using information processing technique. Technical issues to be considered in diagnostic methodology diversification are; (1)reliability of input data, (2)diversification of knowledge models, algorithms and reasoning schemes, (3)mutual complement and robustness. The diagnostic module utilizing the different approaches defined along with strategy of diversification was evaluated using fast breeder plant simulator. As a result, we confirmed that any singular diagnostic module can not meet accuracy criteria for the entire set of anomaly events. In contrast with this, we confirmed that every abnormality could be precisely diagnosed by a mutual combination. In other words, legitimacy of approach selected by strategy of diversification was shown, and methodology diversification attained clear efficiency for abnormal diagnosis. It has been also confirmed that the diversified diagnostic system implemented in this study is able to maintain its accuracy even in case that encountered scale of abnormality is different from reference cases embedded in the knowledge base.