Combining Bayesian Belief Networks and the Goal Structuring Notation to Support Architectural Reasoning About Safety

摘要

There have been an increasing number of applications of Bayesian Belief Network (BBN) for predicting safety properties in an attempt to handle the obstacles of uncertainty and complexity present in modern software development. Yet there is little practical guidance on justifying the use of BBN models for the purpose of safety. In this paper, we propose a compositional and semi-automated approach to reasoning about safety properties of architectures. This approach consists of compositional failure analysis through applying the object-oriented BBN framework. We also show that producing sound safety arguments for BBN-based deviation analysis results can help understand the implications of analysis results and identify new safety problems. The feasibility of the proposed approach is demonstrated by means of a case study.