Treatment of expert opinion diversity in Bayesian belief network model for nuclear digital I&C safety software reliability assessment

Abstract

Since digital instrumentation and control systems are expected to play an important role in safety systems in nuclear power plants (NPPs), the need to incorporate software failures into NPP probabilistic risk assessments has arisen. In order to estimate the failure probability of safety software in NPP and incorporate it into a PRA model, a Bayesian belief network (BBN) model was developed which estimates the number of defects in the software considering the software development life cycle (SDLC) characteristics. In the model, due to a lack of sufficient safety software operation experience data, expert opinion was instead used to quantify the distributed node probability tables (NPTs) that are tables of random variables whose probabilistic distributions were aggregated from experts' elicitation. In addition, handbook data on U.S. software developments and V&V as well as the testing results of two example nuclear safety software were used to Bayesian update the BBN distributed NPTs in order to reduce the BBN parameter uncertainty from the diverse expert opinion. Based on the estimated NPTs, the number of defects at each SDLC phase is evaluated for the typical digital protection software (50 function points and Medium development, V&V quality). This study is expected to provide insight on several aspects of BBN model quantification for nuclear safety-related software reliability assessment, including the expert opinion elicitation and aggregation, the representation of the node probabilities using probability distribution, and the Bayesian updating of the NPTs using available software development data.