Fault diagnosis of rolling bearings based on adaptive optimization of VMD and deep feature extraction

To address the issues of insufficient signal feature extraction and inadequate diagnostic accuracy in rolling bearing fault diagnosis, this study proposes an intelligent diagnostic model integrating Variational Modal Decomposition (Variational Mode Decomposition, VMD) optimized by an improved thornless bee optimization algorithm (ITGCOA) with a multiscale deep convolutional neural network (MDCNN). First, the TGCOA is enhanced through multi-strategy improvements, introducing adaptive weight calculation and elite individual selection mechanisms. This accelerates algorithm convergence and escapes local optima. ITGCOA is then employed to adaptively optimize the modal decomposition layer count and penalty factor parameters of VMD, enabling adaptive decomposition of bearing vibration signals and effective extraction of fault feature information. Second, a dual-branch MDCNN architecture is constructed, where multi-scale convolutional kernels synergistically extract both local subtle features and global trend characteristics from the signals. Finally, experimental validation is conducted using the Case Western Reserve University bearing dataset. Results demonstrate a diagnostic accuracy of 99.722% on the test set, confirming the model's robust feature learning capability and generalization performance. This approach provides an effective solution for rolling bearing fault diagnosis.