Abstract:  Aiming at the problems that bearing fault diagnosis in mining machinery was easily affected by noise interference and variable working conditions, and was difficult to adapt to different diagnosis tasks, a branch convolutional neural network(B-CNN) hierarchical diagnosis method for idler bearing fault was proposed. Firstly, the hierarchical structure of faults was divided according to the specific diagnostic tasks, and multi-level labels were used to represent health status, fault types and degrees of damage. One-dimension convolutional neural network(1DCNN) feature extraction blocks were constructed by alternating convolutional and pooling layers. Then, a bearing fault hierarchical diagnosis method model based on branch one-dimension convolutional neural network(B1DCNN) was designed by combining the hierarchical structure with the feature extraction blocks. Finally, the simulated idler bearing fault experiment was conducted through using the data from the Case Western Reserve University bearing and self-built belt conveyor idler fault test bed, and the diagnostic performance of the method under noise interference and variable working conditions was verified. The results show that the proposed method successfully achieves the diagnosis of roller bearing fault from coarse to fine, and it exhibits good robustness to noise interference and variable working conditions. Comparing with support vector machine(SVM) and back propagation neural network(BPNN) models, this method has better performance in fault diagnosis.

Key words:  mining machinery; rolling bearing; onedimension convolutional neural network(1DCNN) ; belt conveyor; branch convolutional neural network(BCNN); hierarchical structure; variable conditions