To improve nuclear plant safety by learning from accident's experience

The ultimate goal of this study is to produce an expert system that enables the experience (records and information) gained from accidents to be put to use towards improving nuclear plant safety. A number of examples have been investigated, both domestic and overseas, in which experience gained from accidents was utilized by utilities in managing and operating their nuclear power stations to improve safety. The result of investigation has been used to create a general 'basic flow' to make the best use of experience. The ultimate goal is achieved by carrying out this 'basic flow' with artificial intelligence (AI). To do this, it is necessary (1) to apply language analysis to process the source information (primary data base; domestic and overseas accident's reports) into the secondary data base, and (2) to establish an expert system for selecting (screening) significant events from the secondary data base. In the processing described in item (1), a multi-lingual thesaurus for nuclear-related terms become necessary because the source information (primary data bases) itself is multi-lingual. In the work described in item (2), the utilization of probabilistic safety assessment (PSA), for example, is a candidate method for judging the significance of events. Achieving the goal thus requires developing various new techniques. As the first step of the above long-term study project, this report proposes the 'basic flow' and presents the concept of how the nuclear-related AI can be used to carry out this 'basic flow'. (author)