Research On Small Sample Bearing Fault Diagnosis Based On Meta-learning And Improved Generative Network

As a necessary part of mechanical equipment,it is important to monitor the condition and timely fault diagnosis of rolling bearings.In recent years,with the introduction of deep learning theory,data-driven class intelligent fault diagnosis methods have been developed rapidly,and a variety of deep neural network class methods have achieved better diagnosis results.However,in actual industrial production,due to various complex factors,it is impossible to collect enough sample data for each bearing state,which leads to the poor effect of the theoretical model in practical application.The introduction of small sample learning provides a new idea to solve the fault diagnosis in the case of sample scarcity.This paper focuses on the problem of bearing fault diagnosis based on small-sample learning.Firstly,to address the problem that there are very few labeled fault data in practice,this paper investigates a small sample bearing fault diagnosis method based on improved metalearning.Firstly,we combine the convolutional attention module on the basis of residual network to improve the network feature extraction ability;introduce task weights in the meta-learning method to differentiate the task importance,reduce the difference of data distribution between the training and test sets at the task level,and improve the classification performance of the model on the test set;finally,we use the fault data to fine-tune the deep network parameters to achieve small-sample fault diagnosis.Secondly,in practical applications,bearings mostly work in healthy conditions,and the collected fault data is much less than normal data.In this chapter,starting from the perspective of data enhancement,to address the sample imbalance in fault diagnosis,the study uses conditional variational self-encoder to improve the generative adversarial network to increase the number of fault class samples and mix the generated samples with the original samples to reduce the sample imbalance in the training set;then,by adding biased loss to the classification network,the fault classification accuracy of the model is improved,and finally,fault diagnosis is performed on the test set.The method studied in this paper uses several public fault datasets to verify the effectiveness of the method,and also analyzes the influence of the main parameters in the proposed method;in addition,the method studied in this paper is compared with other smallsample learning methods,and the experimental results show the effectiveness of the method in this paper for small-sample fault diagnosis scenarios.