Abstract: Gearbox bearings had many faults under complex operating conditions, and the various faults affected each other, it was difficult to meet the high-precision and intelligent classification requirements by relying on traditional fault diagnosis methods; therefore, a fault diagnosis method combining symmetrized dot pattern (SDP) technology and dilated-grouped convolution residual network (DG-ResNet) was proposed. Firstly, the onedimensional bearing vibration data was converted into a two-dimensional image through the SDP algorithm. Without reducing the original data, the image could clearly show the original characteristics of the vibration data. Then, the image was input to DG-ResNet for fault classification, the feature extraction and classification of bearing fault were carried out, the expansion-group convolution residual block increased the number of convolutions and the size of the receptive field, allowing the network to extract high-level image features. Finally, the accuracy of fault diagnosis was compared with this method and a variety of classic convolutional neural network algorithms. The experimental results show that, comparing with a variety of classic convolutional neural network algorithms, the average accuracy of the proposed method reaches 93%, which is much higher than other comparison networks. It can efficiently classify bearing faults and further verify the effectiveness of the proposed method. It can be used for fault classification of gearbox bearings in practice.

Key words:  gearbox bearings; fault diagnosis;symmetrized dot pattern (SDP); dilated-grouped convolution residual network (DG-ResNet)

HAN Chun-lei, WU Bing, XIONG Xiao-yan, et al. Fault diagnosis of gearbox bearings based on SDP and DG-ResNet［J］.Journal of Mechanical & Electrical Engineering, 2021,38(11):1395-1401.