Failure Propagation Modeling for Safety Analysis using Causal Bayesian Networks

Bayesian Networks (BN) have in previous literature been recognized as a powerful tool for safety analysis, with several advantages over traditional methods such as fault trees. The construction of BNs for safety analysis is however cumbersome; no easier than construction of fault trees. The paper therefore presents a systematic method for construction of BNs for analysis. It is recognized that a special kind of BNs is required, namely Causal BNs. The basic principle to construct these Causal BNs is to utilize specifications of services, or requirements, and their relationships. The approach is especially attractive in the context of safety standards (e.g. ISO26262) where specification and traceability of requirements is already mandatory. The framework in the paper also provides a theoretical link between requirements engineering and the dependability theoretical definitions of fault and failure