Fault Diagnosis Method for Rolling Bearings Based on VMD and SDAE-CNN

An intelligent diagnostic algorithm is proposed to address issues such as poor interpretability and inadequate representation of end-to-end advantages in the smart diagnosis of rolling bearings. Firstly, the Variational Mode Decomposition and Hilbert transform (VMD-Hilbert) method is employed to obtain the envelope spectrum map of the original vibration signal, constructing a feature set for bearing faults. Secondly, the high-dimensional feature set is subjected to dimensionality reduction using the Stack Denoising Autoencoder (SDAE) unsupervised learning model, effectively utilizing the information compression capability of the encoder to obtain low-dimensional features. Finally, the compressed feature set is input into a Convolutional Neural Network (CNN) supervised learning model to achieve intelligent diagnosis of rolling bearings. The results indicate that the proposed method based on VMD and SDAE-CNN exhibits superior diagnostic accuracy compared to the contrastive methods. Furthermore, in comparison to some cumbersome feature extraction processes, this method not only enhances interpretability but also simplifies the fault information extraction process, reducing the skill requirements for diagnostic personnel.