Towards a Practicable Process for Automated Safety Analysis

Over the last ten years numerous approaches to automated safety analysis have emerged, motivated mainly by the increasing complexity of safety critical systems, the increased use of computer-based controllers in such systems and, ultimately, increased costs associated with manual safety analysis. These approaches have attracted significant interest (and funding) from industry. Furthermore, numerous safety certification authorities and standardisation bodies have indicated that they would be ready to accept automated analyses as a basis for certification. This paper examines some prominent approaches and outlines problems that need to be solved by research and tool development communities before the state-of-the-art can evolve into a robust industrial process. In particular, the paper focuses on filling a gap between current state-of-the-art proposals that either combine individual component failure characterisations to arrive at system-level fault trees, or automatically extract minimal cut sets directly from detailed design models (e.g. SCADE, Matlab/Simulink, etc), bypassing Fault Tree generation altogether. The former proposals (e.g. the AltaRica modelling language and Y. Papadopolous’s HiP-HOPS method) allow analysts to model of the failure behaviour of a system as design work progresses from the system architecture to the implementation level. The latter approach (investigated, for example, by the ESACS and ISAAC projects) allows truly automated analysis of detailed design models (thus minimising both the possibility of safety analysis errors and the cost of the analysis).