Fault diagnosis method of rolling bearing based on Xception-CNN model

A fault diagnosis method based on Xception-CNN model was proposed to accurately diagnose the fault types of rolling bearings of belt conveyor transmission rollers under noise interference and variable load conditions. Firstly, based on Xception module, a CNN model body which integrates feature recognition and feature classification is designed, which can effectively reduce the dependence on expert knowledge. Secondly, an MP module is designed to replace the traditional maximum pooling layer, which can not only reduce the model training parameters, but also extract more feature information. Then, by introducing attention mechanism and residual learning mechanism into the model, the attention module can screen out more critical feature information, and the residual module can avoid gradient disappearance and network degradation to a certain extent. Finally, the proposed method and related deep learning methods are simulated and compared on the bearing data set of Case Western Reserve University. The experiments show that the average accuracy rate of the proposed model is 97.39% in the variable noise experiment and 94.58% in the variable load experiment, both of which are higher than other comparison methods, and have good universality and generalization ability.