Hybrid computation of uncertainty in reliability analysis with p-box and evidential networks

This paper presents a method to assess system reliability in the presence of epistemic and aleatory uncertainty. This hybrid method uses belief functions to model and manipulate uncertainty. P-boxes are used to represent basic uncertainties and acyclic directed networks to model the system reliability. These choices allow a flexible modeling of uncertainty, while limiting the computational cost of inferences. In particular, they offer convenient ways of integrating expert opinions and many kinds of uncertainty sources. It also can model complex systems. In this paper, we introduce the modeling method and apply it to a fire detection system with different kinds of input p-boxes for the sake of illustration.