Quantitative software reliability assessment methodology based on Bayesian belief networks and statistical testing for safety-critical software

Abstract This study proposes an overall methodology that provides in-depth evidence on software reliability. It is used to quantitatively assess the reliability of nuclear power plant (NPP) safety-critical software for the incorporation of digital instrumentation and control systems into NPP probabilistic risk assessment (PRA). The methodology consists of three parts: (1) the relationships among the software development life cycle (SDLC) phases, the number of remaining faults in the software, and the probability of failure on demand (PFD) are modeled by a Bayesian belief network, which can provide a prior distribution of the software PFD; (2) a reliability model for the PFD is used to calculate the number of no-failure tests needed to meet the expected reliability target according to the prior distribution; (3) the software statistical testing (SST) based on PRA is used as a reliability validation test method to assess reliability, when the required no-failure tests are completed, it is considered that the software meets the expected reliability target. The main contribution of this methodology is that it fully considers the factors that affect software reliability, i.e. the quality of development activities and verification & validation (V&V) activities of the SDLC processes, software operational profile and software operational environment when assessing software reliability. This is done such that the methodology overcomes the subjectivity of separate quality assessments of the SDLC processes. It also solves the problem that occurs because an individual SST using an uninformative prior distribution is conservative.