A Prognosis Swin Transformer Network for Rolling Bearing Remaining Useful Life Prediction

The research on the remaining useful life (RUL) of rolling bearings based on deep learning plays an important role in the safe and economic operation of rotating machinery. In order to improve the matching degree of network structure and vibration signal data, a novel method named Prognosis Swin Transformer (ProgSwT) is developed for bearing RUL prediction. In the ProgSwT network, the hierarchical structure and sliding window feature extraction are retained, and the number of stages in the Swin Transformer and the number of network layers per level are optimized. Specifically, the input signal first obtains features through the Patch Partition, then transmits into three consecutive stages composed of Linear Embedding and Swin Transformer Block, and finally enters the overall average pooling layer to output the RUL result. As a result, the flexibility and portability of the network structure can be heightened while the computational complexity of the network can be decreased when extracting the characteristics of vibration signals. The experimental results show that the ProgSwT network has good prediction accuracy in the RUL prediction.