Remaining Useful Life Prediction for Bearings Based on Game Theory Feature Selection and Channel Attention-Enhanced TCN

The prediction of Remaining Useful Life (RUL) for rolling bearings is a critical step in the Prognostics and Health Management (PHM) of rotating machinery. However, RUL prediction continues to face challenges, such as the difficulty in extracting vibration-sensitive features of rolling bearings from condition monitoring signals containing substantial redundant information, and the neglect of bearings' multi-stage degradation characteristics in RUL prediction, resulting in reduced accuracy of prediction methods. In response to these challenges, this study proposes a novel data-driven approach. Firstly, a dynamic weight determination method based on game theory combination weighting is implemented to construct comprehensive evaluation indicators, thereby identifying feature sets sensitive to the degradation process. Secondly, a Temporal Convolutional Network (TCN) integrated with the Squeeze-and-Excitation Network (SEN) attention mechanism is employed for multi-stage life prediction. The incorporation of SEN not only enhances the model's ability to capture complex temporal patterns but also improves its adaptability and robustness. Finally, the effectiveness of the proposed method is validated using publicly available bearing datasets.