Calibrated Self-Training for Cross-Domain Bearing Fault Diagnosis

Fault diagnosis of rolling bearings is a crucial task in Prognostics and Health Management, as rolling elements are ubiquitous in industrial assets. Data-driven approaches based on deep neural networks have made significant progress in this area. However, they require collecting large representative labeled data sets. However, in industrial settings, assets are often operated in conditions different from the ones in which labeled data were collected, requiring a transfer between working conditions. In this work, we tackle the classification of bearing fault types and severity levels in the setting of unsupervised domain adaptation (UDA), where labeled data are available in a source domain and only unlabeled data are available in a different but related target domain.We focus on UDA with self-training methods, based on pseudo-labeling of target samples. One major challenge in these methods is to avoid error accumulation due to low-quality pseudo-labels. To address this challenge, we propose incorporating post-hoc calibration, such as the well-known temperature scaling, into the self-training process to increase the quality of selected pseudo-labels.We implement our proposed calibration approach in two self-training algorithms, Calibrated Pseudo-Labeling and Calibrated Adaptive Teacher, and demonstrate their competitive results on the Paderborn University (PU) benchmark for fault diagnosis of rolling bearings under varying operating conditions.

Keywords: Bearing fault diagnosis, Domain adaptation, Self-training, Pseudo-labels, Confidence, Calibration.

Download PDF