THE USE OF BAYESIAN BELIEF NETS IN SAFETY ASSESSMENT OF SOFTWARE BASED SYSTEMS

This report describes an investigation of methods to perform a reliability and safety assessment of the software in programmable safety relevant systems. It emphasises in particular how disparate information sources and different quantitative and qualitative methods should be combined in such an assessment. It starts with a general discussion of probabilistic methods and their applicability to software. Various models for reliability and safety assessment are outlined. An overview is made of different evidences and information sources which are of relevance for such an assessment. One method for combining different evidences in a reliability and safety assessment is pinpointed, viz, the Bayesian Belief Net methodology. This method is described in some detail followed by an illustration on how it may be applied in a reliability assessment. In realistic cases this methodology requires fairly heavy computations, so there is a need for computer aided tools. One such tool is described with a demonstration on how it can be applied. Finally the report makes a summary and draws some conclusions, and outlines plans for further work in this area, viz, an experimental investigation of the BBN method on a realistic test case. (author)