Abstract:
Probabilistic modeling is widely used in engineering practices to assess the safety and dependability of complex systems to control their risks. The study of the reliability of complex systems is a relatively old line of research, but it is still under development and continuous improvement. After having defined the types of uncertainties and the theories for dealing with uncertainties in dependability, we present an overview of the existing probabilistic approaches for the modeling and evaluation of epistemic uncertainty, and we highlight the interest of the evidence theory. The main difficulties in the reliability and availability studies of complex systems are integrating functional and dysfunctional aspects such as independence between components, failure behavior, dynamic aspect, uncertainty and common cause failure. It becomes necessary to use new approaches for estimating reliability and availability. This thesis aims to present a methodology for evaluating the predictive reliability and availability of a complex system, taking into account the above aspects, using modeling by bayesian and evidential networks. The proposed models were applied to case studies and examples taken from the literature.
