Improving early stage system design under the uncertainty in reliability-wise structure

Abstract

This paper is concerned with system reliability improvement at as early as the system's embodiment design stage. Design decisions made at this stage, including component reliability specifications, will have long-term impacts on the entire lifecycle of the system, but they are also difficult to make because the knowledge of system and/or subsystem performance is most likely incomplete at this moment. Thus the system reliability function cannot be fully defined analytically. Because of this inherent uncertainty at early design stage, the traditional reliability evaluation methods, which assume deterministic reliability-wise structures and component reliability specifications, may not be appropriate in real-world applications. In this paper, a Nonparametric Bayesian Network (NPBN) approach is proposed to tackle such challenges at early design stage. The proposed method can represent the uncertainty in the structure of system reliability by combining any available information of component performance in some similar or past-generation systems. This approach can partially quantify the reliability function of the system under design and propose design change recommendations in order to achieve the targeted system reliability level.