Nuclear plant fault diagnosis using probabilistic reasoning

This paper presents a new approach to nuclear plant fault diagnosis using probabilistic reasoning techniques, specifically, a Bayesian network. The scheme is well suited to the task since the symptoms of certain faults are ambiguous. This approach provides a way to capture the knowledge and reach rational decisions in uncertain domains by casting the decision-making process as computation with a discrete probability distribution represented by a causal network. This scheme, unlike some other learning schemes, supports a mathematical explanation of the results, which is necessary in many critical applications. A brief review of probabilistic reasoning via Bayesian networks is provided. Learning the probability values from expert beliefs and statistical data is discussed. The system design process and architecture are explained, and some performance measurements are presented. This module will be deployed as part of the Situation-Related Operator Guidance system (intelligent hypertext manual)