Fault Diagnosis Method of a Rolling Bearing Under Varying Loads Based on Unsupervised Feature Alignment

Aiming at the problem of data lack for rolling bearing under certain load in actual work, which causes great difference between source domain data and target domain data distribution, and the absence of labels in target domain samples, a fault diagnosis method of a rolling bearing was proposed based on multi-domain feature construction and unsupervised feature alignment. Variational mode decomposition and singular value decomposition were combined to obtain time-frequency features of vibration signals. And then combining time-domain and frequency-domain features of vibration signals to construct multi-domain feature sets. The subspace alignment (SA) algorithm which can realize unsupervised domain adaptation in transfer learning, was introduced and improved, and it was proposed to combine the kernel mapping method with SA algorithm. The training data and test data were mapping to the same high-dimensional space, and feature alignment was carried out in the subspace of the high-dimensional space to increase the discrimination between data class, and the alignment of source domain features to target domain features under different loads can be realized. The experimental results show that, compared with partial dimensionality reduction methods and unsupervised transfer learning methods, the proposed method can recognize the corresponding states of other load data using the known load data of rolling bearings without labels in the target domain, and has a higher fault diagnosis accuracy. © 2020 Chin. Soc. for Elec. Eng.