65 / 2023-08-30 22:28:39

A few-label contrastive transfer fault diagnosis method for rolling element bearings

rolling bearings, fault diagnosis, few-label scenarios, transfer learning, contrastive predictive coding

Special Sessions > SS09 - Advanced Acoustic and Vibration Measurement Techniques and Applications in the Fault Diagnosis of Industrial Equipment

In industrial production, rotating machinery extensively employs rolling bearings as vital components, fulfilling a crucial role. These bearings operate in harsh environmental conditions for extended periods and are susceptible to failure. Vibration monitoring sensors are commonly used to collect data for fault diagnosis. While supervised fault diagnosis methods have shown success with fully labeled bearing vibration datasets, a significant amount of data collected in industrial settings remain unlabeled. Manual labeling is time-consuming and labor-intensive, resulting in only a limited number of typical data points being labeled for each bearing failure category. Additionally, obtaining fault data becomes challenging when the operating conditions change, often leaving only normal state data available. To overcome these challenges, this paper proposes a fault diagnosis method based on contrastive transfer learning using few-label data. By leveraging the contrastive encoding model and self-supervised representation contrastive learning, discriminative features can be extracted from a large amount of unlabeled data. The method incorporates transfer learning between the source and target domains, extracting domain-invariant features from normal data and training a feature extractor. Finally, a classifier is trained with a small amount of labeled data from the source domain to construct the model. Experimental results on benchmark datasets validate the effectiveness of the proposed approach, offering a promising solution for fault diagnosis in industrial settings with limited labeled data and changing operating conditions.