Multisource Domain Transfer Learning for Bearing Fault Diagnosis

Deep Learning based fault diagnosis methods assume that training and testing data share the same distribution, which will not hold in practical scenarios due to the variable working conditions of rotating machineries. By utilizing knowledge learned from the source domain to help target tasks, transfer learning (TL) overcomes this problem. However, most TL-based fault diagnosis studies have focused only on single-source TL, while useful multisource domains with sufficient labeled samples are available. In this work, a novel multisource TL model, called the moment matching-based intraclass multisource domain adaptation network, is proposed. This model uses a feature learner to generate features of each source and target domain data to enable the joint weight classifier to predict target labels. It also introduces a moment matching-based distance metric to reduce distance among all source domains and the target domain. During the training of the model, an intraclass alignment training strategy is applied to match the marginal and conditional distributions of each domain simultaneously. Experiments under four load conditions are performed, whose results validate the proposed model’s reliability and generalizability.