A rolling bearing fault diagnosis method based on CAEVMD and CNN

Abstract Rolling bearings inevitably generate noise during operation, which reduces diagnostic accuracy. In this paper, we propose a center-frequency-learning Variational Modal Decomposition (VMD) combined with a Convolutional Autoencoder (CAE) (CAEVMD) to process noisy signals and extract fault features, using a multiscale Convolutional Neural Network (CNN) for bearing fault diagnosis. First, the noisy signal is fed into the VMD for decomposition. The decomposed modal signal is sparsely penalized to suppress the modal noise. The sparsely penalized modal component is encoded, compressed, decoded, and reconstructed. The reconstructed output signal shows reduced noise. Secondly, the fault features are extracted using a multiscale dilated Convolutional Neural Network, combined with the Spatial Attention Mechanism (SAM) and the Channel Attention Mechanism (CAM), for fault diagnosis. Validated on the Western Reserve University bearing dataset and the simulated bearing experimental dataset, the experimental results show that the proposed network model achieves high diagnostic accuracy across different noise conditions, outperforms the comparative methods, and demonstrates high accuracy, robustness, and good generalization.