Evaluation of safety parameters in the presence of uncertainties and design assistance: application to Safety Instrumented Systems.

Summary The introduction of instrumented systems dedicated to safety applications requires the evaluation of their reliability. Generic reliability databases are generally used for this purpose. However, the experience feedback for these systems, which generally present rare failures, is insufficient to validate the results obtained. Moreover, the collection of reliability data and their extrapolation to other components introduces uncertainties. The work of this thesis deals with the problem of taking into account the uncertainties related to the reliability data of components for the evaluation of the safety of systems by the formalism of fuzzy subsets. The proposed methodology is applied to the evaluation of failure probabilities of Safety Instrumented Systems (SIS) in the presence of imprecise reliability data. We introduce two new importance factors to help the designer. In addition, we propose a methodology to assist the design of SIS based on reliability network modeling and optimization by genetic algorithms of the structure of SIS for the respect of safety integrity levels (SIL) required The use of safety related systems imposes to evaluate their dependability.