Remaining useful life prediction method considering multi-degenerate model decision fusion

Abstract Wiener-process-based methods for remaining useful life (RUL) prediction are widely adopted for component degradation modeling. However, the prediction accuracy of these methods is often significantly influenced by the selection of the degradation mode (DMS). To address the uncertainty inherent in DMS results, which can compromise RUL prediction accuracy, this paper proposes a novel RUL prediction framework that integrates multi-degradation model decision fusion. First, a temporal dynamic sensor data fusion method is introduced: data-level fusion enhances component degradation information by jointly considering data consistency and predictability, while feature-level fusion dynamically selects an optimal feature subset to construct a sensitive health indicator. Subsequently, the RUL prediction derived from a single Wiener degradation mode is calibrated using a degradation angle deviation metric. To mitigate the uncertainty of DMS, the corrected RUL estimates from multiple degradation modes are fused in real time, thereby achieving decision-level fusion. Finally, the effectiveness of the proposed method is validated using two bearing vibration signal datasets.