Belt conveyor idler fault diagnosis method based on multi-scale feature fusion and residual mask convolution attention

Existing idler fault diagnosis methods have problems in failing to fully obtain global context information and providing poor diagnostic accuracy. To address these problems, this paper investigates a new method for diagnosing faults in belt conveyor idlers, based on analysis of their acoustic signals. The method is also applied to existing databases of bearing fault data. Firstly, an eight-element microphone array sound signal collector is designed to suppress environmental noise and raise the signal-to-noise ratio of the idler sound signal. Secondly, a multi-scale feature fusion (MSFF) module is constructed to learn complementary information between features at different scales. Then, a residual mask convolutional attention (MCA) module is designed to raise the modelling capability of local features and global contextual information. Finally, the structure of the ResNet-18 network is optimised to improve model fitting performance. Experimental results on self-made and public datasets show that the suggested method outperforms other comparative methods, achieving real-time accurate detection and classification of belt conveyor idler faults and typical bearing faults.