Analysis on the important indicators in Bayesian belief network model for RPS software reliability assessment

Abstract

Bayesian belief network model was developed in authors' previous research that quantifies the number of software faults based on software development life cycle (SDLC) characteristics of nuclear power plant (NPP) safety-related software. In a nuclear application, in order to effectively reduce the number of software defects in a target digital safety system, it is important to analyze the SDLC phases or related activities that are the major contributors to the final number of residual faults in the software. First, in order to identify the software development activities (attributes) as strong or weak indicators for the overall development or V&V quality of specific software development lifecycle (SDLC) phase, the indication measure of an attribute for development and V&V quality is proposed and the contribution of attributes' states to the quality nodes were analyzed. Secondly, the contribution analysis of quality nodes on the number of residual software defects is conducted considering the improvement of development and V&V quality from Medium to High quality in each SDLC phase. Furthermore, the cost of fixing detected defects passed from previous phases when the development and V&V quality is improved from Medium to High is assessed. This study is expected to provide an insight on analyzing the important SDLC phases and related software attributes to be considered when one desires to effectively optimize the SDLC for targeting fewer residual software defects in NPP digital safety-related system.