A Self-Tuning Hybrid Model for RUL Prediction of Rolling Bearings With Two-Stage Degradation

Accurate remaining useful life (RUL) prediction of rolling bearings is critical for the safe operation of mechanical systems. During their lifecycle, rolling bearings typically undergo two stages with distinct degradation rates: a slow degradation stage (SDS) and a fast degradation stage (FDS). Due to the significant uncertainty during the stage transition, existing methods often rely on stage segmentation algorithms to divide the degradation process and perform RUL predictions. However, these methods are limited by their restricted prediction range and the increased risk of model mismatch. To address these challenges, a new self-tuning hybrid model is proposed to predict the RUL of bearings exhibiting two-stage degradation. The proposed model combines adaptively updated linear and nonlinear models to describe the two-stage degradation process of bearings. Additionally, a transition zone is considered to manage the uncertainty in stage segmentation. In addition, adaptive weight updates are directly guided by the parameters of the prediction model, assigning greater weight to the model that best fits the current degradation stage, thereby enhancing the overall prediction accuracy. The proposed model is validated using vibration data collected from accelerated degradation tests of rolling bearings. Experimental results demonstrate that the proposed method achieves more accurate predictions than existing methods.