A lifting contrastive learning method for rotating machinery fault diagnosis

The datasets in the field of machinery fault diagnostics typically contain a significant number of unlabeled data alongside a small amount of labeled data, and there are category imbalances, which all lead to low fault diagnosis accuracy. To cope with this issue, a lifting contrastive learning method for rotating machinery fault diagnosis is proposed. First, the SE-ResNet50-based contrastive learning network is self-supervised trained using all data in the dataset (including labeled and unlabeled data). Second, to address the effect of category imbalance on the classifier, Focal Loss is added to the downstream task. When compared to prior methods for planetary gearbox fault diagnosis, experimental results show that the proposed method achieves higher diagnostic accuracy in limited and imbalanced labeling situations.