A bayesian inference algorithm to identify types of accidents in nuclear power plants

Abstract

In complex systems, it is necessary to model a logical representation of the overall system interaction with respect to the individual subsystems. Operators are allowed to follow Emergency Operating Procedures, when a reactor is tripped because of accidents. But, it's very difficult to diagnose accidents and find out appropriate procedures to mitigate current accidents in a given short time. Even if they diagnose accidents, it also has possibility to misdiagnose. TMI accident is a good example of operators' errors. Methodology using Influence Diagrams has been developed and applied for representing the dependency behaviors and uncertain behaviors of complex systems. An example to diagnose the accidents such as SLOCA and SGTR with similar symptoms has been introduced. From the constructed model, operators could diagnose accidents at any states of accidents. This model can offer the information about accidents with given symptoms. This model might help operators to diagnose correctly and rapidly. It might be very useful to support operators to reduce human error. Also, from this study, it is applicable to diagnose other accidents with similar symptoms and to analyze causes of reactor trip.