Bearing fault diagnosis based on cross image multi-attention mechanism

Abstract

Bearings are crucial components of rotating machinery, and fault diagnosis is essential for ensuring the safe operation of mechanical systems. Neural networks, commonly used in bearing fault diagnosis, are effective in extracting deep features from fault signals but often fail to emphasize critical information. We propose a fault diagnosis method that integrates a cross-image multi-attention mechanism with a residual neural network. The collected vibration signals are first preprocessed using VMD-GAF and then fed into the network for fault detection. The results demonstrate that the CIMAM-ResNet18 model significantly enhances the robustness of signal processing, achieving an accuracy of 98.00% when tested on the experimental platform.