Research on fault diagnosis combining optimized VMD and pyramid transformer

The operating condition of bearings directly determines the operating condition of the system. It is necessary to monitor the signal data of bearings to maintain the stability and reliability of the equipment. Addressing the issues of incomplete feature extraction in traditional diagnostic methods, high deployment costs in actual industrial production, and reduced diagnostic efficiency due to noise interference, a method for fault diagnosis of bearings was proposed that combines optimized Variational Mode Decomposition (OVMD) with Pyramid Transformer. Firstly, the optimized Variational Mode Decomposition parameters are obtained by analyzing the original bearing vibration signals. Subsequently, the signals processed through variational mode decomposition are preprocessed to generate two-dimensional feature maps, which are then input into the Transformer module. Passing through multiple modules of different sizes, the features extracted by different modules are fused to obtain vibration signal features with different receptive fields. After classification calculation, the final result is obtained, and a visual analysis of the results is conducted. The proposed method combines early fault diagnosis methods with deep learning techniques. Experiments have shown that this method achieves structural lightweighting while demonstrating its effectiveness and noise resistance on common bearing datasets.