An entropy-driven method for exploring extreme and unexpected accident scenarios in the risk assessment of dynamic engineered systems

International audienceIn Integrated Deterministic Probabilistic Safety Assessment (IDPSA), the end state reached by the dynamic engineered system under analysis, as outcome of an accident scenario, depends on the failure sequence in the scenario, but also on the magnitude and timing of the failures. Including these additional dimensions can render the analysis infeasible, as the size of the system state space explodes with the large number of components and physical quantities involved in realistic cases. In this paper, we address the problem of au-tomatically and efficiently probing the space of the event sequences in a dynamic system. In particular, an en-tropy-driven technique is proposed for intelligently focusing the exploration efforts preferably on those sce-narios that are most interesting for the scope of the analysis. To this aim, the relevance of a given accident scenario is defined by two original indicators: (i) the criticality of the corresponding system end states, related to their (un)likelihood of occurrence and/or their (extreme) consequences; (ii) the level of “unexpectedness” of the corresponding event sequences, with respect to the prior knowledge of the analyst. The method is applied to a simple case study of a dynamic system for gas transmission