[This article belongs to Volume - 54, Issue - 01]
Gongcheng Kexue Yu Jishu/Advanced Engineering Science
Journal ID : AES-15-06-2022-219

Title : Inter-class Repulsive Slack Discriminant Transfer Learning Based on Rolling Bearing Fault Diagnosis
LI Feng, WANG Teng, TANG Baoping, TIAN Daqing,

Abstract :

In view of the problem that it is difficult or even impossible to obtain the class labels of new working condition samples under actual variable working conditions, which leads to the low fault diagnosis accuracy, a novel fault diagnosis method based on inter-class repulsive slack discriminant transfer learning (IRSDTL) is proposed. In the proposed IRSDTL, a nonnegative extended slack matrix is constructed to transform the strict binary label matrix into an extended slack label matrix for increasing the distance between different class label vectors in source domain and making the common subspace dimension no longer limited to the number of class labels. As a result, the classification error in source domain is reduced, and the generalization ability of IRSDTL is improved. Moreover, the joint distribution difference is introduced to reduce the difference between auxiliary and target domains, which can better realize the cross-domain transfer learning between two domains; the inter-class repulsive force term is constructed to promote the discriminative learning effect by increasing the distance between one class subdomain samples and the other class subdomains in the two domains. Finally, the whole framework of IRSDTL is optimized by the alternating direction multiplier (ADM) method to easily obtain the optimal parameter values of IRSDTL. The labeled samples under historical working conditions can be used by IRFDTL to perform high-precision class discrimination on the testing samples under new working conditions when there are no class labels of testing samples. Thus, precise fault diagnosis of the testing samples under new working conditions can be achieved by the proposed IRSDTL-based fault diagnosis method. The experimental results of rolling bearing fault diagnosis show that the diagnosis accuracy of the proposed method is higher than that of the other four transfer learning-based methods, and its misdiagnosis rates of misdiagnosing the three types of faults as normal state and the normal state as three types of faults are low, and the effectiveness and practicability of the proposed method are verified