Semi-supervised contrastive fault diagnosis with uncertainty-aware and performance-guided adaptive thresholds under...

Although existing semi-supervised learning (SSL) methods provides promising framework for fault diagnosis with limited labels, they struggle to simultaneously address the dual challenges of label scarcity and class imbalance in real-world scenarios. To bridge this gap, we propose an adaptive semi-supervised contrastive learning fault diagnosis method. The adaptive threshold strategy integrating prediction uncertainty and diagnosis performance is introduced to boost the quality of pseudo-labels and guide consistency regularization. Further, the probabilistic contrastive learning mechanism is integrated to enhance the model's fault discrimination capability under class-imbalanced conditions by enlarging inter-class separability and tightening intra-class compactness. Experiments on two public datasets show that the proposed method achieves significant accuracy improvements of 5.36 % and 3.04 % compared to the state-of-the-art SSL methods in fault diagnosis with severely limited labeled samples (10 %) and highly imbalanced class distribution (20:1 ratio).