Interpretable fault diagnosis of coal-mine conveyor-belt idler bearings using multi-source fusion diagrams and mobile residual soft threshold–MobileViT

The fault diagnosis of conveyor belt idler bearings plays a crucial role in coal mine safety. In practical, inevitable challenges such as high-noise interference and small-sample conditions. Existing methods exhibit inadequate performance under the above conditions. This paper proposes a technique for fault diagnosis that integrates multisource fusion diagrams with a lightweight MRST-MobileViT model. To enhance fault feature representation, GADF, MDF, RP and CWT were employed to convert vibration signals into multi-source fusion diagram with richer fault information. A mobile residual soft thresholding module is incorporated into the MobileViT architecture to enhance the feature extraction capability. Verification through public datasets and self-constructed test data demonstrate that this method has a better diagnostic performance under high-noise and small-sample conditions. Furthermore, Grad-CAM heatmap analysis was performed to validate the interpretability of the model. This study provides an effective and reliable solution for the fault diagnosis of idler bearings in coal mines.