Research On Bearing Fault Diagnosis Methods Based On Improved Generative Adversarial Network

With the development of manufacturing industry,the safety of industrial equipment has attracted more attention.As one of the most important parts of rotating machinery,the rolling bearing has been in a bad working environment for a long time such as high temperature,high speed or heavy load.So the monitoring of the working state and the diagnosis of the fault state of the rolling bearing have become the focus of researchers’ attention,as well as one of the key problems in the prevention and maintenance of mechanical systems.Therefore,this thesis takes bearing failures as the research object,and takes unbalanced datasets,the lack of labeled data,and production environments with variable operating conditions as the practical application background,conducts research on the problem of unbalance sample under supervised conditions,clustering data under unsupervised conditions,and difficult migration of training models.The specific research contents are as follows:(1)Aiming at the problem of unbalanced bearing fault data classes,an unbalanced fault diagnosis method based on the auxiliary classification one-dimensional convolution generative adversarial network is proposed.Firstly,this method introduces one-dimensional convolution neural network(One Dimensional Convolutional Neural Networks,1DCNN)to transform the auxiliary classification generative adversarial network,uses its feature extraction capabilities for one-dimensional time series data to improve the quality of generating samples.Secondly,it blends the generated samples into the original samples to balance a few class samples,and obtains a data set with richer sample characteristics.Finally,the 1DCNN model is used to diagnose the bearing data set of Case Western Reserve University with various balance rates before and after the expansion.The experimental results show that the performance of the model is significantly improved after the data is gradually balanced,indicating the excellent ability of this method in solving the imbalance problem of positive and negative samples.(2)Aiming at the problem of lacking of labels of bearing fault data,an unsupervised fault diagnosis method based on self-attention mechanism classification generative adversarial network is proposed.Firstly,the bearing data is processed by short-time Fourier transform,and the main structure of the design model is 1DCNN.Secondly,the self-attention mechanism module is introduced to improve the perception ability of the model for the characteristics between channels.Finally,the performance of the model is verified using an unlabeled bearing dataset.The experimental results show that the proposed model effectively improves the diagnosis accuracy,and also shows the outstanding anti-noise ability and good anti-migration performance of all the comparison algorithms.(3)Aiming at the difference of bearing fault data feature distribution caused by complex working conditions in industrial scenes and the difficulty of obtaining a large number of labeled data,a one-dimensional convolution subdomain adaptive adversarial neural network based on Wasserstein distance and local maximum mean discrepancy is proposed.Firstly,the network constructs a feature extractor based on 1DCNN for pre-training and learning the domain feature representation.In the adversarial training stage,the adversarial layer introduces Wasserstein distance to measure the difference between the source domain and the target domain,realize the alignment of marginal distribution and solidify the training results.In the feature extraction layer,local maximum mean discrepancy calculation module is introduced to capture the fine-grained information of each category to realize the alignment of conditional distribution.The performance of the model is verified by the bearing fault data sets under two different working conditions.The experimental results show that under unsupervised conditions,the proposed method improves the recognition accuracy by 5.0%and 6.9% respectively compared with the basic domain adversarial network on the target data set,and the performance is better than the existing migration algorithms.