Reliability Assessment of Hierarchical Systems With Incomplete Mixed Data

Abstract

A complex engineering system typically consists of a group of components/subsystems in a hierarchical structure and the system is monitored at only some, not all, of these hierarchical levels. This paper investigates a Bayesian approach to system reliability prediction using multilevel incomplete data. These data are drawn simultaneously from different component/subsystem levels within the same system, thus need to be analyzed with the consideration of their overlapping nature. In this paper, a Bayesian network model is proposed for modeling the reliability of a multilevel system, where a lower level node can only be connected to one higher level node. Through Bayesian inference, the posterior distributions of lifetime parameters and conditional probabilities in the model are obtained by combining prior beliefs with lifetime data coming from different system levels. This study is also extended to include mixed data types, i.e., both pass/fail data and lifetime data. The effectiveness of the proposed approach is illustrated in a case study.