Domain Adaptation Learning Methods For Fault Diagnosis Of Rolling Bearings

Rolling bearings,as one of the key components of rotating machinery in many fields,are affected by many factors in their operating condition,and as the structure of modern industrial equipment becomes more and more complex,it also poses higher challenges to the safety and reliability of equipment operation.Therefore,bearing fault diagnosis is essential to ensure the reliability of rotating machinery operation.With the further advancement of intelligent manufacturing and industrial big data,data-driven intelligent fault diagnosis methods have become one of the most popular research directions in the field of fault diagnosis.Machine learning theory can provide new ideas and strategies for intelligent fault diagnosis with its powerful modeling,analysis and data representation capabilities.This paper studies the key problems existing in the fault diagnosis of rolling bearings,and designs some new intelligent fault diagnosis models and algorithms.The main research work of the paper includes the following points:1.In view the problem of insufficient adaptation of existing methods in cross-domain fault diagnosis of rolling bearings,this article puts forward the model of fault diagnosis based on joint subdomain adaptation network.A double level adversarial domain adaptation algorithm for fault diagnosis is designed to solve the problem that the training and test datasets do not obey the same distribution,as well as the domain adaptation global adaptive problem,in order to maximize the intra-domain similarity and minimize the inter-domain discrepancy.This allows the knowledge learned in the source domain to be well transferred to the target domain and to identify and classify faults effectively.Finally,experimental results demonstrate the effectiveness of the proposed method.2.Aiming at the problem that the single-dimensional bearing vibration signal has insufficient ability to characterize the feature,a fault diagnosis model based on tensor data domain adaptation is proposed,and a domain adaptation fault diagnosis algorithm based on invariant tensor space learning is designed to solve the problem that the vector representation cannot fully reflect and retain the important structural information of multi-dimensional signals,and when the domains are very different,only aligning the source domain to the target domain will cause the problem of excessive data distortion.It realizes the unified representation of tensors for multi-dimensional signals and fully exploits the hidden features of the original signals,so that the trained fault diagnosis model has good performance in cross-domain diagnosis tasks.The experiments show that the proposed tensor domain adaptation method can fully consider the correlation between data and extract high-quality feature information to achieve system adaptation of multi-dimensional information.3.Aiming at the problem that fault category information is easy to cause misclassification in the case of multiple source and target domains do not match each other,a fault diagnosis model based on multi-source ensemble domain adaptation is proposed,and a multi-source ensemble adversarial domain adaptation fault diagnosis algorithm considering domain correlation is designed.It breaks through the limitations of single source domain and target domain fault diagnosis,so that the constructed and adapted fault diagnosis model can show good performance.Experiments show that the proposed domain adaptation fault diagnosis algorithm can enhance the positive effect of similar domains and suppress the negative effects of different domains.