Bayesian Belief Network Model Quantification Using Distribution-Based Node Probability and Experienced Data Updates for Software Reliability Assessment

Abstract

Since digital instrumentation and control systems are expected to play an essential role in safety systems in nuclear power plants (NPPs), the need to incorporate software failures into NPP probabilistic risk assessment has arisen. Based on a Bayesian belief network (BBN) model developed to estimate the number of software faults considering the software development lifecycle, we performed a pilot study of software reliability quantification using the BBN model by aggregating different experts' opinions. In this paper, we suggest the distribution-based node probability table (D-NPT) development method which can efficiently represent diverse expert elicitation in the form of statistical distributions and provides mathematical quantification scheme. Besides, the handbook data on U.S. software development and V&V and testing results for two nuclear safety software were used for a Bayesian update of the D-NPTs in order to reduce the BBN parameter uncertainty due to experts' different background or levels of experience. To analyze the effect of diverse expert opinions on the BBN parameter uncertainties, the sensitivity studies were conducted by eliminating the significantly different NPT estimates among expert opinions. The proposed approach demonstrates a framework that can effectively and systematically integrate different kinds of available source information to quantify BBN NPTs for NPP software reliability assessment.