Adaptation Regularization Based on Transfer Learning for Fault Diagnosis of Rotating Machinery Under Multiple Operating Conditions

Abstract

Greater difference in data distribution of rotating machinery under multiple operating conditions, which increases the difficulty of fault diagnosis. To solve this problem, a multi-condition fault diagnosis method for rotating machinery using adaptation regularization based on transfer learning (ARTL) was proposed. First, a base classifier is trained in the source domain to predict pseudo-labels of the target domain, and the two-domain data distribution is adapted by joint distribution to reduce the difference of data distribution. Then, the local neighborhood relationship of the samples is learned through manifold regularization, and the local geometric characteristics of multi-condition data are mined. Finally, following the framework of structural risk minimization (SRM), the kernel function is selected to build the classifier, which iteratively updates the pseudo-labels in the target domain and obtains the coefficient matrix to complete the fault diagnosis of rotating machinery. Experimental results on bearing and gearbox failure datasets show that the method has good feasibility and effectiveness under multiple operating conditions.