The risk of severe accidents occurring in a nuclear power plant is quantitatively evaluated utilizing the probabilistic risk assessment methodology. Given that nuclear power plants consist of complex systems with numerous facilities, equipment, piping and electrical components, the scope of potential accident scenarios requiring assessment is vast. The combinations of failures that could lead to a severe accident, as identified through the probabilistic risk assessment, are validated by verifying the plausibility of the events based on failure modes inferred from the design, operation and experience of the nuclear power plant. While the number of possible failure combinations may be enormous, the numerical outcomes can be deemed credible when the dominant contributors are thoroughly analyzed. The credibility of the assessment may be further enhanced if more of the combinations are subject to the validation. To achieve this, we have employed natural language processing, a branch of artificial intelligence to develop a method that facilitates the efficient and comprehensive validation of the probabilistic risk assessment.
