A high-accuracy rolling bearing fault diagnosis method based on enhanced variational mode decomposition with an adaptive adjustment factor and ConvTransNet

Abstract

Rolling bearings are essential components of rotating machinery, and their operational conditions directly affect the safety and stability of the equipment. However, in complex industrial environments, fault signals are often non-stationary, weak, and susceptible to noise interference, which significantly degrades the performance of traditional feature-extraction and classification methods. To address these challenges, this study proposes an adaptive fault diagnosis method that integrates variational mode decomposition (VMD) with deep learning (DL). First, an improved grey wolf optimizer (IGWO) with an adaptive adjustment factor was employed to optimize the key parameters of VMD, thereby enhancing the quality of signal decomposition and fault feature extraction. Then, a new WPEK index was constructed, which combined weighted-permutation entropy and kurtosis. The WPEK and correlation coefficient were used to select information-rich intrinsic mode functions (IMFs) for signal reconstruction, effectively suppressing redundant components and noise. Finally, a ConvTransNet model, incorporating convolutional neural networks and transformers, was designed to extract and fuse local and global features for accurate fault classification. Experimental results demonstrate that the proposed GVMD-ConvTransNet method achieves superior diagnostic performance and robustness across multiple datasets and noise conditions, offering a reliable solution for practical industrial fault diagnosis applications.