Research On Intelligent Fault Diagnosis Method Of Rolling Bearing Under The Condition Of Small Sample

Rolling bearings play a pivotal role in the safe operation of equipment,and any form of bearing failure may lead to potential hidden troubles and unexpected safety problems in the equipment.Therefore,it is necessary to use advanced fault diagnosis technology for condition assessment and quality monitoring of rolling bearings to minimize economic losses.However,in actual production,the fault signal of rolling bearing is difficult to obtain,and there are problems such as few marked samples and data imbalance,and manual marking of unlabeled data will consume a lot of manpower and material resources.Moreover,the speed and load of the bearing will also change during operation,and it is difficult to extract effective features from the signal for fault diagnosis only by relying on expert experience.Therefore,how to effectively extract the fault features in the bearing and complete the bearing fault diagnosis under the condition of small samples and multiple working conditions is the focus of current research in the field of fault diagnosis.Based on generative adversarial network(GAN),this paper adopts the idea of data generation and training expansion,and studies the fault diagnosis method of rolling bearing under the condition of small sample and multiple working conditions,and conducts experimental verification.The main work of the paper is as follows:Aiming at the problems of bearing fault data prone to lack of samples and lack of fault data,a small-sample bearing intelligent fault diagnosis method based on 2D grayscale map and auxiliary classification generative adversarial network(ACGAN)was proposed.First,convert the one-dimensional vibration time series signal into a 2D grayscale image and input it into ACGAN for adversarial training to obtain auxiliary training samples;then,the expanded training set is used as the input of the deep convolutional neural network(DCNN);finally,Softmax is used The classifier outputs fault identification results.The research shows that the proposed method has better advantages and stability compared with other fault diagnosis methods,and has practical application feasibility.Faced with the problem of difficult fault feature extraction encountered in the fault diagnosis process of rolling bearings,an improved Mobile Netv3 convolutional neural network is proposed,using a self-attentive mechanism to replace the original selfattentive mechanism in the network,making the model gain in diagnostic accuracy.The improved Mobile Netv3 network mainly consists of a stack of inverse residual structures and convolutional blocks,which enhances the non-linear variation of features and realizes the fusion of features after layer-by-layer convolution,and effectively extracts fault features.Considering the problems of less fault data and working condition changes encountered in the process of bearing data collection,an intelligent fault diagnosis of rolling bearings based on structural similarity generative adversarial network(SSGAN)and improved Mobile Netv3 convolutional neural network is proposed.method.First,the wavelet transform(WT)is used to preprocess the signal,and the wavelet twodimensional image with the time-frequency characteristics of the signal is input into SSGAN to complete the expansion of the training set.The expanded training set is sent to the improved Mobile Netv3 convolutional neural network for learning,the selfattention mechanism is used to extract the fault features,and the trained model is used to test the fault data under different working conditions.The experimental results show that SSGAN has good sample generation ability,which can effectively improve the fault classification accuracy of the improved Mobile Netv3 convolutional neural network in bearing datasets under various working conditions.