Research On Rolling Bearing Fault Diagnosis Method Based On Generative Adversarial Networks

Intelligent fault diagnosis based on deep learning is one of the research focuses in the field of bearing fault diagnosis.In industrial production,due to the lack of fault data,the marking of data samples requires a lot of manpower and time costs,leading to the problem of small data samples and semi supervised learning.In response to the above two issues,this article adopts generative adversarial networks as the model foundation,proposes feasible solutions,and conducts experimental verification by combining public and laboratory datasetsFirstly,for fault data,the feasibility of generative adversarial network for bearing fault diagnosis is verified,and a bearing fault diagnosis method based on conditional generative adversarial network is proposed.Introducing label information into the generator network to achieve data augmentation,the input of the network is transformed from one-dimensional vibration signals into two-dimensional grayscale images,and a stacked convolutional automatic encoder is introduced to extract features from the generated and original samples for fault diagnosis.The validation is conducted on the bearing dataset of Case Western Storage University and the bearing dataset of the mechanical fault comprehensive simulation test bench of Yanshan University.High accuracy was achieved on test samples from different datasets,verifying the feasibility of the proposed method.Secondly,aiming at the small sample problem,a small sample bearing fault diagnosis method based on improved auxiliary classification generative adversarial network is proposed.Wasserstein distance is used as the loss function of the auxiliary classification generative adversarial network,which enhances the ability of data generation,improves the quality of generated samples,and expands the small sample data set.Combining convolution network with attention mechanism to improve the learning ability of complex data.The method proposed in this paper is verified and analyzed by using the data of the rolling bearing of Western Reserve University and the small sample data of Yanshan University under variable conditions.Ablation experiments,comparative experiments,and noise resistance experiments were conducted to verify the feasibility of the proposed method.Finally,a bearing fault diagnosis method based on Triple-GAN in semi supervised mode is proposed for the semi supervised problem.This method mainly improves the structure of the generation of countermeasures network,which can not only enhance the data of fault samples,but also introduce the cross entropy loss of fault classification into the labeled samples to realize the pseudo label prediction of unlabeled samples.The discriminator no longer determines whether the data source is true or false,but rather whether the data label comes from the real sample.The input of the network uses the time-frequency image after the short-time Fourier transform.Experimental verification was conducted on the bearing dataset of Case Western Reserve University,and the feasibility of the proposed method was verified by comparing it with traditional semi supervised methods.