Enhancing Software Reliability Estimation Using Bayesian Networks and Fault Trees

A majority of mission-critical or safety-critical systems are complex computer-controlled systems, which are increasingly relying on software to provide added functionality. Although improved design techniques and development methodologies have reduced the software engineering cycle time, the complexity of software design and analysis has increased. The probabilistic assessment of software safety and reliability is important due to the pervasive use of software in such systems and since more system failures are being attributed to software failures. Moreover, software reliability is an important metric in determining overall system reliability. However, this is a difficult task involving uncertain predictions. We describe a novel approach to estimate software reliability by using bayesian belief networks (BBNs) and fault trees. The goal is to provide a more realistic estimate of software reliability that accounts for the causal dependence between processes in the software engineering cycle and software reliability.