A Rolling Bearing Fault Diagnosis Method Based on Multi-Sensor Information Fusion and Effective Diagnostic Vision Transformer

Rolling bearings are crucial components of rotating machinery. Accurate and reliable fault diagnosis is vital for ensuring the smooth operation of modern industrial equipment. To address incomplete feature extraction, low diagnostic efficiency and reliability in rolling bearing fault diagnosis, this paper proposes a method based on multi-sensor information fusion and an Effective Diagnostic Vision Transformer (EDViT). This approach aims to enhance diagnostic accuracy and reliability. Initially, a kurtosis-based weighted fusion strategy merges sensor information, ensuring its integrity and relevance. Subsequently, Short-Time Fourier Transform (STFT) converts these fused signals into time-frequency images as inputs to the EDViT network, capturing their time-frequency characteristics. The Dual Attention Convolutional Module and Dual-branch Patch Vision Transformer Module of EDViT then sequentially extract local and global features, and finally, a classifier performs fault classification. Experimental validation was conducted on bearing datasets from Case Western Reserve University (CWRU). The experimental results demonstrate that multi-sensor information fusion improves the accuracy and reliability of diagnostic results. The EDViT model exhibits excellent feature extraction capabilities, fast convergence speed, and high diagnostic accuracy. Comparisons with other methods reveal that the EDViT model possesses strong generalization capability, achieving over 95% average cross-domain diagnostic accuracy on the CWRU dataset.