APPLYING BAYESIAN BELIEF NETWORK METHOD TO QUANTIFYING SOFTWARE FAILURE PROBABILITY OF A PROTECTION SYSTEM

摘要

An approach for quantifying the probability of failure on demand of a safety-related protection system due to software failures is presented. It is being applied to an example system. The approach uses checklists to evaluate how good a job was done in different phases of software development and develops two Bayesian belief network (BBN) models for quantifying software failure probability, a complex one and a simple one. It is based on the assumption that the quality in carrying out the software development activities determines the reliability of the software. The more complex BBN model uses the quality evaluation and debugging data to estimate the number of faults injected and the number of faults detected and removed in each phase of the development process. The model then converts the estimated number of faults remaining in the software into a software failure probability by using the concept of fault exposure ratio. The simpler BBN model is based on the idea that software reliability can be calculated as a function of the weighted average of factors related to the quality of software development. This is similar to the Failure Likelihood Index Method of human reliability analysis (HRA). That is, the quality factors of software development play a role in software reliability similar to that played by performance shaping factors in HRA.