Research On Intelligent Fault Diagnosis Method Of Rolling Bearing Based On Partial Domain Adaption

Rolling bearings are the key part of mechanical equipments,a their failure will lead to unstable operation of equipments and increase of maintenance costs.Rolling bearing fault diagnosis is an important link to ensure the normal operation of mechanical equipments.However,due to the differences in operating conditions of equipments,there are differences in distribution and fault types of bearing data,which is detrimental to the accuracy and reliability of bearing fault diagnosis.To address the above problems,this paper conducts research on intelligent fault diagnosis methods for rolling bearings based on partial domain adaptation on the baisi of deep learning and domain adaptation theory.The main work of this article is as follows:(1)Aiming at the problem of large differences in data distribution under different operating conditions,which leads to poor generalization ability of the model,and the target domain label space being a subset of the source domain label space,on the basis of the domain adversarial theory,a partial domain adaptive rolling bearing fault diagnosis method based on the joint domain alignment of domain adversarial and Maximum Mean Discrepancy(MMD)is proposed.This method embeds an attention mechanism in the feature extractor to enhance the attention to shared key features of fault data between two domains.And this method proposes a joint approach of MMD measurement and domain adversarial to jointly improve the model’s ability to extract domain invariant features,thereby promoting the transfer of source domain knowledge to the target domain and achieving fault identification and classification.Two sets of rolling bearing datasets are used for to validate the proposed metod,and the results show that the proposed method can achieve high-precision fault diagnosis and has good robustness and generalization ability.(2)In the case of multiple source domains intersect with the label space of the target domain,a rolling bearing fault diagnosis method based on attention multi-source partial domain adaptation is proposed to address the issue of varying degrees of differences between source domains and between source and target domains.Firstly,this method designs a unique extractor and classifier for each source domain.By reducing the MMD distance between each source domain and the target domain,multiple source domain distributions are aligned with the target domain distribution,which reduces the negative migration impact of the source domain private categories.Secondly,a consistent regularization criterion is proposed,which takes the difference of each source domain classifier’s prediction results for the target domain samples as a loss,and together with the MMD loss and source domain classification loss,an objective function is formed.The model is optimized through back propagation to improve the reliability of the target domain classification.Finally,the prediction labels of the target domainby multiple source domain classifiers are averaged as a fault classification result.The proposed method is analyzed using a publicly available rolling bearing dataset,and the results show that the method can achieve fine fault diagnosis results while ensuring robustness.