Research Of Generative Adversarial Network Based Methods For Rolling Bearing Fault Diagnosis

At present,the bearing fault diagnosis method based on deep learning has achieved remarkable results under the premise of sufficient labeled samples.The larger the amount of labeled sample data,the stronger the discernibility of the features extracted by the trained model,and the better the diagnosis effect.However,in actual industrial production,manual labeling of a large amount of collected data is time-consuming,labor-intensive and requires prior knowledge of expert experience.Therefore,it is usually impossible to obtain a large number of labeled samples,which leads to semi-supervised and unsupervised problems.In this paper,the generative neural network is used to systematically study the fault diagnosis of rolling bearings in the case of fewer labeled samples and unlabeled samples.The feasibility and effectiveness of the proposed method are experimentally verified.The main research contents of this paper are as follows:1)The generative adversarial network is introduced into the field of rolling bearing fault diagnosis,and a supervised fault diagnosis method based on GAN is proposed.The convolutional layer is selected as the layer structure of the proposed model,and to meet the requirements of the input form,the original one-dimensional bearing signal is converted into two-dimensional images through the signal-to-image method.The experimental results show that GAN can reach a recognition rate of 99.45% on the public bearing dataset of Case Western Reserve University,which verifies the feasibility and effectiveness of GAN for fault diagnosis tasks.2)Aiming at the problem of semi-supervised fault diagnosis of rolling bearings,this paper proposes a 1D-SGAN diagnosis model that can realize semi-supervised learning and effectively extract bearing vibration signal fault features by combining one-dimensional convolutional layer and semi-supervised generative adversarial network.The proposed method can make good use of unlabeled samples and get rid of the dependence on a large number of labeled samples.It not only has a high diagnostic accuracy on fault diagnosis tasks,but also can maintain a certain accuracy when there is strong noise interference.3)Aiming at the problem of unsupervised fault diagnosis of rolling bearings,a diagnosis model based on STFT and categorical generation adversarial network is proposed.The method first uses the signal processing technology STFT to preprocess the bearing signals to reduce the interference of ineffective noise and highlight the fault information;then combines the unsupervised clustering algorithm Cat GAN for model training.And after a reasonable structure design,it finally realizes the end-to-end fault diagnosis.Through comparative experiments,the ST-Cat GAN model not only has a high diagnostic accuracy,but also can effectively maintain its diagnostic performance when the bearing workload changes,which shows the strong robustness of ST-Cat GAN against load migration.