Domain-Aware Contrastive Learning for Fault Diagnosis

A new method for semi-supervised domain generalization fault diagnosis (SSDGFD) is proposed. The approach, called domain-aware hierarchical contrastive learning (DAHCL), addresses two key limitations in existing methods: pseudo-label bias from neglecting domain-specific geometric discrepancies, and imbalanced sample utilization from hard accept-or-discard strategies. DAHCL uses hierarchical contrastive learning to generate better pseudo-labels and handle uncertain samples. The paper is available on arXiv with ID 2604.20928.