Diagnosis of operational failures and on-demand failures in nuclear power plants: An approach based on dynamic Bayesian networks

摘要

Successful diagnosis of system failures in nuclear power plants plays a central role in emergency response. Existing research focuses on diagnosis of operational failures that initiate an incident. During the incident progression, on-demand failures may occur and lead to severe consequences. However, little attention has been paid to this subject. In this paper, an approach for diagnosis of both operational and on-demand failures based on dynamic Bayesian networks is proposed. The general method of developing the dynamic Bayesian network model of a plant, the application of fuzzy sets theory in transforming real valued sensor signals to discrete states, and the procedure for real-time inverse inference based on evidence is introduced. A case study of the proposed approach is conducted with the high temperature gas-cooled reactor nuclear power plant. The results from complete evidence and incomplete case that loses the most valuable sensor signals both demonstrate the effectiveness of the proposed approach. (C) 2019 Elsevier Ltd. All rights reserved.