Research On CNN-based Domain Adaptive Rolling Bearing Fault Diagnosis Method

Rolling bearing is an important part of rotating machinery.It is widely used in modern industrial production,and its fault diagnosis will effectively ensure the normal and stable operation of mechanical equipment and prevent major accidents.The development of signal acquisition system and deep learning technology promotes the bearing fault diagnosis algorithm based on data-driven technology.However,in rolling bearing fault diagnosis,the following problems restrict the application of convolution network: it is difficult to obtain fault data labels,the actual condition changes are complex,and there are differences between acquisition and test equipment.Therefore,how to use labeled data or auxiliary domain data to establish mathematical model and carry out fault diagnosis for target domain with different data distribution is a problem to be solved.The core of transfer learning is to find the similarity between different domains,and improve the performance of the classification or regression model of the target domain with the help of the knowledge learned in the source domain,which provides a basic idea for solving such problems.Therefore,this paper combines deep learning with transfer learning for rolling bearing fault diagnosis.The main research work of this paper includes the following points:(1)In order to solve the problem that the difference of the extracted migration features is not comparable due to the inconsistent setting of the convolution layer parameters in the convolutional neural network(CNN),a domain adaptive rolling bearing fault diagnosis model based on the maximum mean difference(MMD)is established.Firstly,the CNN model framework with consistent convolution layer parameters is established to extract the multi-level feature representation of rolling bearing fault vibration signal;secondly,the regularization of MMD is introduced to impose constraints on CNN parameters to reduce the mismatch of feature distribution between source domain and target domain,forming a domain adaptive rolling bearing fault diagnosis method based on MMD;finally,the proposed method is verified by data sets under different working conditions Compared with the domain free adaptive technology,the diagnostic accuracy of this method is improved by 12.56%.(2)In view of the fact that the traditional CNN domain adaptive method only considers the difference of the highest layer domain,but ignores the difference of all the multi-layer domains,a new multi-layer domain adaptive rolling bearing fault diagnosis method based on MMD is proposed.Firstly,CNN is used to extract the migration features of the original vibration data;secondly,multi-layer domain adaptation and weight regularization term are used to constrain CNN parameters to further reduce the distribution difference of the migration features,so as to solve the multi-layer domain displacement problem;finally,the migration diagnosis experiment is constructed by using the rolling bearing data sets of different working conditions to verify the adaptability,superiority and diagnosis results of the proposed method The reliability of the results was analyzed and verified.(3)Aiming at the problems of low utilization of fault data features and high cost of MMD calculation in traditional CNN model,a domain adaptive rolling bearing fault diagnosis method based on Wasserstein distance is proposed by introducing the difference of Wasserstein distance measurement domain.Firstly,the Wasserstein distance between the source domain and the target domain is used as the measurement standard to complete the confrontation training between the feature extractor and the domain discriminator,and construct the CNN domain adaptation model;secondly,t-distributed stochastic neighbor is used Finally,through experiments and comparative analysis of rolling bearing vibration data under different working conditions,it shows that the domain adaptive model can effectively extract domain invariant features.