Fault diagnosis of shearer rocker gear based on deep residual network

Abstract

The traditional shearer rocker gear fault diagnosis methods cannot extract features autonomously, resulting in poor gear fault diagnosis accuracy and efficiency. In order to solve the above problems, a fault diagnosis model of shearer rocker gear based on deep residual network (ResNet) is constructed. By pre-activating the residual unit module, the complexity of the model is reduced so as to make the model converge faster. By reorganizing the vibration signal data, the data input method is optimized so as to improve the model's ability to identify the fault of shearer rocker gear. Model verification tests are carried out on the rocker gear loading test bench of shearer to collect vibration signals of the rocker spur gears under five states of normal, worn, fractured, pitting and cracked. It is concluded that there are significant differences in their characteristics. The visual analysis of the confusion matrix of the test set verifies that the ResNet model can realizeshearer rocker gear fault classification well. Moreover, the comparison results with the DNN model and LeNet-5 model show that the ResNet model has higher fault diagnosis accuracy and efficiency. The comprehensive recognition rate and F-score reach 99.19% and 99.05% respectively. The t-SNE technology is used to reducedimension and visualize the high-dimensional features output from the maximum pooling layer, the pre-activated residual unit module and the fully connected layer of the ResNet model, which verifies that the ResNet model has strong feature extraction capability.

Keywords

• shearer rocker
• gear
• fault diagnosis
• deep residual network
• deep learning