New health-state assessment model based on belief rule base with interpretability

Health-state assessment is the foundation of optimal-maintenance decision-making for complex systems to maintain reliability and safety. Generating the assessment results in a convincing and interpretable way to avoid potential risks is of great importance. Belief rule base (BRB) as an interpretable model performs well in health-state assessment. However, the interpretability of a BRB-based model may be lost during the optimization process, which is expressed mainly as three problems: expert knowledge is not effectively used in the optimization process; the optimized rules of BRB may be in conflict with real systems; and some parameters get over-optimized, which may affect experts’ initial judgment. Three concepts — “searching intensity”, “interpretability constraint of belief distribution”, and “rule-activation factor” — are defined to address these problems. Using these concepts, we propose a new health-state assessment model based on the interpretable BRB and a new optimization method to improve the accuracy and preserve the interpretability of the new model. To demonstrate the effectiveness of the proposed model, we conducted an aero-engine case study.