Development of an on-line expert system for alarm filtering and diagnosis in nuclear power plants

Abstract

An on-line expert system, called AFDS, is developed for the purpose of supporting the operator during an abnormal state, which performs alarm filtering, overall plant-wide diagnosis, and alarm prognosis. The alarm filtering and diagnostic knowledge bases are organized on the basis of object-oriented concepts. For the filtering of multiple alarms, the knowledge base is organized with alarm relationship network using functional relations among the alarms. The alarm filtering rules define the dynamic prioritization among the fired alarms. For the alarm diagnostics, the knowledge base is organized by using fuzzy relation for the knowledge representation. The alarm diagnostic rules using fuzzy logic use fired alarms and generated trend alarms as symptoms, and the compositional inference scheme is used for the inference mechanism of AFDS. For the case of the alarm prognosis, Levinson algorithm and MDL model order criterion are used according to the results of comparative study. Additionally, AFDS performs a trend analysis using fuzzy membership to monitor some important critical safety parameters and to generate trend alarms when the parameters are deviated from the normal condition. AFDS has been simulated with on-line data which was obtained from the full-scope simulator for several abnormal cases. The results indicated that AFDS can provide the operator with useful informations for the earlier termination and mitigation of an abnormal state.