Remaining Useful Life Prediction of Rolling Bearings Based on Multi-Feature Fusion and TCN-RSA

Abstract

Remaining useful life (RUL) prediction of rolling bearings is important for condition-based maintenance and early fault warning of mechanical equipment. However, practical bearing vibration signals usually show nonlinear, nonstationary, and noise-contaminated characteristics, making stable degradation representation difficult. A single degradation feature often has difficulty in describing the bearing degradation process, which may affect RUL prediction accuracy. To address this issue, this paper proposes a rolling bearing RUL prediction method based on multi-feature fusion and temporal convolutional network with residual self-attention (TCN-RSA). First, VDE, RMS, and kurtosis are fused to construct a comprehensive degradation representation of rolling bearings. Then, a trend-enhanced health indicator is introduced, and the TCN-RSA model is used to establish the mapping relationship between degradation feature sequences and RUL. Experiments on the XJTU-SY rolling bearing dataset show that the proposed method can effectively track the degradation trend of bearing RUL and achieve relatively accurate prediction results. These results verify the effectiveness of the proposed framework for rolling bearing RUL prediction.