Research On Fault Diagnosis Method Of Rolling Bearing With Small Sample Based On Improved Generating Countermeasure Network

Rolling bearings are key components of rotating machinery,and evaluating and diagnosing their health status is an important technical means to ensure the safe operation of mechanical equipment.However,the working environment of the bearing is complex and harsh,and it is in normal operation most of the time.As a result,in the process of fault diagnosis of rolling bearings,it is difficult to obtain fault samples and the number of samples is limited,making it difficult to achieve accurate fault diagnosis.To this end,this paper conducts research on small-sample fault data enhancement and intelligent fault diagnosis methods for rolling bearings through the design,improvement and application of the Deep Convolutional Generative Adversarial Network(DCGAN)structure.The main research content and results are as follows:(1)When using the symmetrical point pattern(SDP)method for bearing signal conversion,the parameters of the method have an important impact on the recognizability of the formed image,and it is necessary to optimize it.First,the value of the SDP method is selected through image comparison analysis,and then the d and g values of the SDP method are selected using the Pearson correlation coefficient,and the validity of the selected parameters is verified by two image similarity indexes and graphic visualization results.The results show that the selected parameters can effectively improve the resolution and recognition of the image,and the two-dimensional images of different faults are significantly different,realizing the effective representation of different fault states of rolling bearings.(2)Aiming at the problem of low quality samples generated by traditional DCGAN for data enhancement,a small-sample fault data enhancement method based on SDP and W-DCGAN(SW-DCGAN)is proposed.Firstly,the SDP method with the optimal parameters is used to convert the one-dimensional vibration signal into a two-dimensional symmetrical petal image,and according to the characteristics of the two-dimensional petal image,the hyperparameters of the network are designed to build a model with better generation performance and stability;Secondly,in order to enhance the stability of DCGAN network training,the Wasserstein distance is introduced to improve the objective function and spectral normalization processing is added to the discriminant network to improve the quality of the samples generated by the model.The experimental verification results show that this method can effectively improve the quality of the generated samples and realize the data enhancement of small samples of bearing fault status.(3)Aiming at the problem of low recognition rate when the shallow neural network diagnoses small sample faults of bearings,a rolling bearing fault diagnosis method based on SW-DCGAN and C-Res Net18 is proposed.First,improve the fault feature extraction ability of the model by improving the residual unit of the Res Net18 model,and at the same time introduce the convolution attention mechanism and integrate it with the deep residual network to form a C-Res Net18 model,and use the Adam optimization algorithm to optimize the C-Res Net18 model the parameters are optimized;then the SW-DCGAN model is used as the front part of C-Res Net18 to form a complete rolling bearing fault diagnosis method.Experimental results show that this method can identify different fault types of rolling bearings in the case of small samples,and can achieve a high fault recognition rate.