Rolling Bearing Fault Diagnosis Based on Improved VSDP Image and Multihead Attention Mechanism

Rolling bearing fault diagnosis is critical for ensuring equipment operational safety. An intelligent fault diagnosis method is proposed, integrates an improved Symmetrized Dot Pattern (SDP) image with an AlexNet-based model augmented by a multi-head wavelet attention block. Vibration signals are first decomposed using Variational Mode Decomposition (VMD). The resulting modal data from multiple sensors are transformed into VMD SDP (VSDP) images and then integrated into a composite Multi-channel VSDP (MVSDP) image, which significantly enriches fault feature representation. This image is processed by a customized AlexNet model incorporating a multi-head wavelet attention mechanism, which collectively solves the challenges of feature clustering, multi-angle attention, and feature fusion to extract global fault information comprehensively. Validation on the CWRU bearing dataset confirms that our approach significantly outperforms existing methods in both recognition accuracy and operational stability.