The two-stage wiener process degradation model has been extensively researched in remaining useful life (RUL) prediction. However, during the prediction process, it is challenging to use feedback from prediction information to optimize the prediction model. Additionally, sudden change in the drift term at change point (CP) affect the model accuracy in describing the degradation path, thereby reducing prediction accuracy. To address these issues, this paper first proposes a feedback multi-feature fusion (FMFF) method. A series of normal numbers not exceeding the threshold, known as fraction threshold (FT), are introduced to segment the degradation state dimension. Moreover, the time from the current moment to the first hitting time (FHT) of the FT is known as the fraction remaining useful life (FRUL), which is used to construct and update fusion factor. In addition, a novel two-stage degradation model is proposed to address the issue of sudden change by retaining partial parameter information from the slow degradation stage, enabling a smooth transition between models. New state value and threshold are constructed to achieve RUL prediction. Finally, the proposed fusion features and model are validated using the XJTU-SY dataset and laboratory datasets, with results demonstrating the superiority of the proposed methods.

Remaining useful life prediction,Wiener process,Feature fusion,Rolling bearing,Prognostics and health management