Rolling Bearing Fault Diagnosis Based on VMD-CNN-SLSTM Model

Convolutional neural networks (CNN) are limited by the local receptive field, making it difficult to capture the shock features of the cross period. In view of this limitation, a VMD-CNN-SLSTM rolling bearing fault diagnosis method is proposed. First, VMD is harnessed to decompose the original signal, and the fitness function is envelope kurtosis. Then the reconstruction was carried out according to the Pearson correlation coefficient and the envelope entropy. Secondly, continuous wavelet variation (CWT) is utilized to convert signals into two-dimensional time-frequency diagrams and divide them into the training set, the verification set and the test set in proportion. The CNN -SLSTM model is used to study the high-dimensional features of the specified data set. Finally, the softmax classifier is utilized for fault diagnosis. The model is validated on the rolling bearing vibration signal dataset from Case Western Reserve University (CWRU) in the United States, demonstrating excellent recognition accuracy of 99.2 %, highlighting its effectiveness in the diagnosis of rolling bearing failure.