基于DGC-SwinT的轴承故障诊断轻量化方法OA

In order to solve the problems of insufficient feature extraction and generalization ability and excessive number of model parameters in strong noise environment of traditional rolling bearing fault diagnosis methods,this paper proposes a lightweight bearing fault diagnosis method based on Dilated Group Convolution(DGC)and Swin Transformer(SwinT)from the perspective of time-frequency feature enhancement and depth feature decoupling.Firstly,aiming at the transient mutation characteristics of fault shock in the time-frequency domain,the synchronous compression wavelet transform is used to construct the time-frequency sensitive feature space,which can effectively highlight the key features in the signal and suppress the secondary information and interference terms.Secondly,a multi-scale feature fusion module is designed,which fully extracts the low-level features from the time-frequency signal and reduces the computational cost of the model through DGC.Then,through the cross-scale dependencies generated by fault modulation,a high-order information interaction block is constructed to weight the important features,and the long-range periodic fault information is deeply mined by relying on the advantage of the SwinT model to capture the global dependencies.Finally,the experiments are carried out by the CWRU bearing dataset and the SEU bearing dataset.Experiments show that the proposed method can achieve an accuracy of up to 99%under the condition of noise interference signal,and it is highly robust.This achievement provides a new technical path for the online monitoring system of rotating machinery and is particularly suitable for early fault warning under complex working conditions such as wind power gearboxes and high-speed rail traction motors.