A hybrid muti-dimension normalization layers improved ResNet based fault diagnosis method of rolling bearing

Abstract

CNN, a kind of deep learning method, has been widely used in fault diagnosis. It requires a large number of training samples, but it is difficult to obtain abundant samples under different conditions. Aiming at insufficient fault samples, an improved ResNet (IResNet) is proposed in this paper. Firstly, order spectrum is computed from raw data as pre-processed samples, which will be further augmented to improve the generalization ability of the model. Secondly, IResNet is constructed by several hybrid residual building blocks fused from multi-dimensional normalization layers, which can be adopted to enhance the feature extraction ability of the model. Then, the parameters of IResNet in the source domain are transferred to identify the health status of rolling bearing in the target domain. Finally, experimental data under different working conditions are used to verify the performance of the proposed method. The experimental results indicate that the recognition accuracy of the proposed method is higher than other methods and that the proposed method can identify the health status of rolling bearing with small training samples.