Research On Adversarial Learning Based Methods For Rolling Bearing Fault Diagnosis

Rolling bearings as important mechanical parts are widely used in various types of industrial equipment.In the actual production and operation process,the working conditions of high-speed rotating machinery and equipment are bad,resulting in the multiple effects of different loads on the rolling bearings,which is easy to produce various forms of defective faults,causing incalculable losses to industrial production,it has great industrial application significance and scientific research value for how to quickly and accurately diagnose the status of rolling bearings.With the rapid development of industrial Internet of Things and distributed control systems,huge amount of real-time data is analyzed and collected,however,the current data-driven fault diagnosis research is faced with practical problems such as strong noise interference,category imbalance,shortage of label information,and even sudden changes in system working conditions.Therefore,this paper uses adversarial learning techniques to investigate the problems of poor noise immunity of fault diagnosis models with labelled samples,low diagnostic accuracy under unbalanced sample categories and poor migration performance of unlabelled samples with variable operating conditions.The specific research work is as follows:(1)Aiming at the problem that the traditional discriminative bearing fault diagnosis algorithm relies on artificial feature extraction and has poor diagnosis effect under noise interference working conditions,it is proposed to use auxiliary classifier generative adversarial network for bearing fault diagnosis research in generative model.Firstly,the bearing vibration signal is converted into a two-dimensional frequency-domain feature grayscale image through fast Fourier transform,the convolution neural network is designed as the main structure of the model,and batch normalization and Leaky Relu activation function are added to alleviate the problem of gradient disappearance.Secondly,the self-attention mechanism is introduced to correlate the features far away from each other in the data,and a new model is established to realize the effective learning of the original data distribution features in the multi classification scene.Finally,the model is applied to the motor bearing for comparison and verification.The results show that the fault diagnosis accuracy of the proposed method is as high as 99.7%,and has good generalization and robustness.(2)For the diagnosis of rolling bearing faults under unbalanced conditions in sample categories,it is proposed a Wasserstein distance-auxiliary classifier generative adversarial network model based on gradient penalty.The original time domain bearing vibration signal is first transformed into frequency domain samples by Fourier transform,and the convolutional neural network is used as the main structure of the model,incorporating the labeling information of the data for training,so as to exert its powerful image generation capability.The quality of the pseudo-samples generated by the model is evaluated through the Frechette starting distance and Pearson coefficient,and the pseudo-samples whose similarity exceeds the specified threshold are saved,and then the unbalanced data set is gradually expanded.The final results show that the proposed method can synthesize data highly similar to the real samples,overcome the problem that the original network is prone to mode collapse or gradient disappearance,and the accuracy of bearing fault diagnosis is effectively improved as the imbalanced dataset is gradually expanded to equilibrium.(3)Aiming at the problem of unlabeled samples and mismatched model data distribution in the case of sudden change of system conditions,a domain adaptive fault diagnosis model based on adversarial learning is proposed.Combining adversarial learning and cross-domain migration learning technology,the time-frequency characteristic information map of non-stationary bearing vibration signal is constructed using wavelet packet transform,and the cross-domain matching process of features is added to the adversarial learning model.The tag information is matched through multiple linear mapping.The improved maximum mean difference measurement algorithm is used as the measurement standard for features with category information,and the joint distribution distance is calculated with the help of false tags,Reduce the difference of data distribution and realize the simultaneous alignment of edge distribution and condition distribution.The experimental results show that the unsupervised cross-domain rolling bearing fault diagnosis model based on time-frequency information fusion can not only achieve cross-domain fault diagnosis without labels,but also outperform other algorithms in experimental accuracy and feature visualization.