Rolling Bearing Fault Diagnosis Based on SVD-GST Combined with Vision Transformer

Aiming at rolling bearing fault diagnosis, the collected vibration signal contains complex noise interference, and one-dimensional information cannot be used to fully mine data features of problem. This paper proposes a diagnosis method based on SVD-GST combined with Vision Transformer. Firstly, is preprocessed reduce using singular value decomposition (SVD) obtain more accurate useful signal. Then, generalized S-transform (GST) convert processed into two-dimensional time–frequency image make full use advantages deep learning in classification higher recognition accuracy. In order avoid problem limited sensory fields CNN need for an RNN compute step by over time when processing sequence data, Transformer model pattern proposed. Finally, experimental platform bearings built. The experimentally validated, achieving average accuracy 98.52% multiple tests. Additionally, compared SVD-GST-2DCNN, STFT-CNN-LSTM, SVD-GST-LSTM, GST-ViT models, proposed has diagnostic stability, providing new diagnosis.