Abstract: Abstract:With With the development of data analysis technology and the improvement of coal mine intellectualization level, the unmanned and fewer-person inspection of underground belt conveyors has become a key link in the intelligent upgrading of the main coal flow transportation system. Aiming at the problems that the driving unit of coal mine belt conveyors requires manual inspection for a long time, and the existing fault diagnosis architecture is highly dependent on massive data interaction between underground and ground platforms, making it difficult to achieve independent edge-side fault diagnosis, a vibration signal preprocessing method combining the improved Ensemble Empirical Mode Decomposition (EEMD) and Denoising Autoencoder (DAE) is proposed. In the processes of vibration signal decomposition, decoding, encoding and reconstruction, this method adds white noise for multiple times, which weakens the interference of underground noise on the vibration data of monitoring equipment. Secondly, the fault characteristics of the driving unit of coal mine belt conveyors are analyzed through mechanism analysis, and a fault diagnosis method based on the dimensionality-increasing convolutional neural network is integrated to extract detailed features that are difficult to cover by mechanism analysis, thus improving the accuracy of fault diagnosis. Finally, a fault diagnosis model lightweight method based on the importance of channels and neurons is proposed, which effectively prunes the non-critical structures of the fault diagnosis model and enhances the feasibility of deploying the algorithm model on the underground edge side. Experimental verification shows that the comprehensive fault diagnosis accuracy of the original model of the proposed method reaches 98.61%. After lightweight processing, the redundant structure of the original model is reduced by 27.5%, while the comprehensive fault diagnosis accuracy only decreases by 0.69 percentage points. This research provides a feasible solution for edge-side fault diagnosis in coal mines.