Use of Bayesian Networks for System Reliability Assessment

Abstract

Probabilistic Safety Assessment (PSA) is a technique to quantify the risk associated with complex systems like Nuclear Power Plants (NPPs), chemical industries, aerospace industry, etc. PSA aims at identifying the possible undesirable scenarios that could occur in a plant, along with the likelihood of their occurrence and the consequences associated with them. PSA of NPPs is generally performed through Fault Tree (FT) and Event Tree (ET) approach. FTs are used to evaluate the unavailability or frequency of failure of various systems in the plant, especially those that are safety critical. Some of the limitations of FTs and ETs are consideration of constant failure/repair data for components. Also, the dependency between the component failures is handled in a very conservative manner using beta factor, alpha factors, etc. Recently, the trend is shifting toward the development of Bayesian Network (BN) model of FTs. BNs are directed acyclic graphs and work on the principles of probability theory. The paper highlights how to develop BN from FT and how it can be used to develop a BN model of the FT of Isolation Condenser (IC) of the advanced reactor and incorporate the system component indicator status into the BN. The indicator status would act like evidence to the basic events, thus updating their probabilities.