Intelligent fault diagnosis of rolling bearing based on EMDPWVD time-frequency images and improved ViT network

Rolling bearing is one of key components of mechanical equipment, and its fault diagnosis is crucial for the safe and stable operation of equipment. Here, aiming at non-stationary characteristics of vibration signals in rolling bearing, a new intelligent fault diagnosis method based on empirical mode decomposition & pseudo-Wigner-Ville distribution (EMDPWVD) time-frequency images and improved Vision Transformer (ViT) network model was proposed. Firstly, aiming at actual signals, 3 time-frequency analysis methods of short-time Fourier transform (STFT), continuous wavelet transform (CWT) and EMDPWVD were studied. Considering STFT and CWT being unable to simultaneously realize high time resolution and frequency resolution, EMDPWVD was selected as time-frequency image construction method for intelligent fault diagnosis network. Secondly, the classic ViT was taken as the basic model for fault diagnosis, time-frequency images were divided into blocks according to predetermined sizes and linearly mapped into input sequences. Global information of images was integrated with self-attention mechanism, and network transmission was completed using stacked Transformer encoder to realize fault diagnosis. To further improve the accuracy of fault diagnosis, pooling layer was taken as the preprocessing network of ViT to obtain an improved Pooling ViT (PiT) model, realize spatial feature extension of time-frequency images and promote the sensitivity of model to input images. The results showed that the proposed method can have high diagnostic accuracy for different types of rolling bearing faults; the accuracy of PiT is 4. 40% higher than ViT, so adding pooling layer into ViT can improve the diagnostic effect of rolling bearing faults. © 2024 Chinese Vibration Engineering Society. All rights reserved.