Rolling Bearing Remaining Life Prediction Based on Multi-Feature Extraction and Transformer-BiTCN

With the rapid development of high-end manufacturing and intelligent maintenance, predicting the Remaining Useful Life (RUL) of rolling bearings is crucial for enhancing equipment reliability and reducing maintenance costs. This paper proposes a prediction model based on multi-domain feature extraction and a hybrid Transformer-BiTCN architecture. The model combines the Transformer's multi-head attention for global temporal modeling with BiTCN's bidirectional convolutions for capturing local dependencies. A comprehensive feature selection strategy is also introduced to enhance feature relevance. Experimental results show that the proposed method achieves superior accuracy, stability, and cross-condition generalization compared to several mainstream deep learning models, demonstrating strong potential for engineering applications.