Intelligent diagnosis method for mechanical and electrical equipment faults in coal mines under lightweight deep learning

This paper proposes a lightweight parallel deep network fault intelligent diagnosis method to address the problem of weak fault features being submerged by strong background noise in coal mines, and the limited computing power and storage of mining edge equipment, which makes it difficult to deploy high-precision models in real time. Design an energy gradient adaptive wavelet packet threshold function to dynamically adjust the denoising intensity based on local signal energy, while suppressing noise and preserving transient impact of faults; Constructing a parallel lightweight dual branch model of 1D-CNN and GRU, integrating local impact and long-term degradation features in the channel dimension, to achieve accurate identification of composite faults. The experiment shows that the confidence entropy of this method is superior to the comparison method in all fault states, verifying its diagnostic effectiveness under resource constraints.