BRN: A belief rule network model for the health evaluation of complex systems

Abstract

A belief rule base (BRB) expert system provides a generic inference framework for approximating the complicated nonlinear relationships between inputs and outputs. Such systems have been widely applied in the system health management community. However, limited by the method of construction rules, existing methods encounter the combinatorial explosion problem in the modeling of complex systems. Thus, determining how to effectively use the collected signals, system mechanism, and expert knowledge to realize the health evaluation of complex systems has become a key issue limiting the development of BRB systems. To solve this problem, a modified micro belief rule structure is proposed, and a new belief rule network (BRN) model is developed. Then, the cautious conjunctive rule is introduced to realize the fusion and reasoning of the nonindependent nodes in the BRN. In addition, the structure and parameter learning strategies of the proposed BRN are given, thereby providing a systemic mechanism to enhance the capability of the BRN when both expert knowledge and observed data are available. The effectiveness of the BRN model is verified in an aerospace relay experiment.