Improved CNN-Based Fault Diagnosis Method for Rolling Bearings under Variable Working Conditions

Aiming at the worse fault diagnosis of rolling bearing and poor generalization ability in a strong noise environment and variable working conditions, an improved CNN-based fault diagnosis method for rolling bearing under variable working conditions is proposed. A multi-scale feature extraction module is designed, and convolutional layers of different scales are adopted to extract features from the input data to maximize the extraction of feature information in the fault data. The channel attention mechanism is then introduced to extract the more important and critical components from this module. A convolution module with skip connection lines is designed to prevent the extracted rich features from being lost when the convolutional layer is forwarded. Regarding softmax cross entropy as the loss function, the Adam optimization algorithm is chosen to realize the fault diagnosis for rolling bearing. The proposed method is verified by experiments on the bearing dataset and gearbox dataset from Case Western Reserve University. The results show that in the variable noise experiment on the bearing dataset from Case Western Reserve University, the proposed method achieves an average diagnostic accuracy rate of 96.49%, and the diagnostic accuracy rate is beyond 90% in variable working conditions, which are obviously higher than the competing methods. On the gearbox bearing data set, the diagnostic accuracy rate of the proposed method with better noise resistance and generalization ability reaches 99.54%. © 2021, Editorial Office of Journal of Xi'an Jiaotong University. All right reserved.