Remaining Useful Life (RUL) prediction of rolling bearings is crucial for ensuring the reliability and safety of rotating machinery in industrial applications. Deep learning-based RUL prediction methods generally adopt a unidirectional time-series feature extraction mechanism, which prevents the utilization of future time-series data. To address this issue, a Bidirectional Temporal Convolutional Network (BiTCN)–Bidirectional Gated Recurrent Unit (BiGRU)–Multi-Head Cross-Attention (MHCA) method for the rolling bearing RUL prediction based on parallel bidirectional feature fusion is proposed. The membership function parameters of slope entropy are fuzzified, and multiscale analysis is performed to construct Refined Composite Multiscale Fuzzy Slope Entropy (RCMFuSE) for characterizing the bearing degradation trend. A parallel dual-branch structure based on BiTCN and BiGRU is established, enabling the model to capture both the long-term cumulative effects of the degradation and the complexity of short-term fluctuations. An MHCA model is employed for adaptive weighted fusion of multidimensional features to improve the prediction accuracy. Experimental verification demonstrates that compared with other state-of-the-art prediction methods, the proposed method can accurately predict the RUL of rolling bearings with smaller prediction errors, verifying its effectiveness and its generalization capability.
Wang et al. (Tue,) studied this question.