Rotating Machinery Bearing Fault Diagnosis via Transfer Learning Based on Subdomain Adaptation

Rotating machinery (RM) is utilized extensively in various industrial applications, primarily due to its dependable operation is essential for optimizing productivity and preventing costly downtime. Accurate and early bearing fault diagnosis in RM is crucial to prevent machinery breakdown and maintain optimal performance. In this study, we propose a novel method for rotating machinery bearing fault diagnosis based on transfer learning with subdomain adaptation. Our method leverages the knowledge acquired from a source domain that contains labeled data for both healthy and faulty conditions. By transferring this knowledge to a target domain with labeled data limitation, our approach addresses the scarcity of labeled data in the target domain. We specifically focus on adapting the domain shift by computing the discrepancy between the source and target domains through subdomain adaptation. To validate the effectiveness of our proposed approach, extensive experiments were carried out on a large-scale CWRU dataset that encompasses various bearing operation conditions. The experimental outcomes reveal that our transfer learning-based fault diagnosis method achieves 93.8% to 100.0% in diagnostic accuracy and outperforms conventional methods.