Data-driven enhanced belief rule base for complex system health state assessment

Abstract

In complex systems, assessing the health state is crucial to ensuring safety and reliability. However, due to the complexity of these systems, acquiring a sufficient amount of useful data poses significant challenges. As a knowledge-based modeling approach, the belief rule base (BRB) utilizes expert knowledge to address these challenges. Nonetheless, in many engineering practices, obtaining sufficient expert knowledge can be equally difficult. To address this problem, this study proposes a data-driven enhanced BRB (DDE-BRB) method for initial model generation, which enhances the modeling capability when expert knowledge is insufficient. First, an antecedent attribute reference value initialization method based on fuzzy clustering is proposed. Second, a method using the Gaussian membership function is introduced to initialize the belief degrees. Finally, optimization algorithms are employed to fine-tune the remaining parameters, and evidence reasoning (ER) technology is used to infer the model. In two case studies, the results demonstrate that DDE-BRB can effectively complete the modeling process and achieve accurate assessment results even under conditions of insufficient expert knowledge.