Domain-Aware Hierarchical Contrastive Learning for Semi-Supervised Generalization Fault Diagnosis

2026-04-24 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

Researchers Junyu Ren, Wensheng Gan, and Philip S. Yu propose a unified framework called Domain-Aware Hierarchical Contrastive Learning (DAHCL) for semi-supervised domain generalization fault diagnosis (SSDGFD). This framework addresses challenges in fault diagnosis under unseen operating conditions with scarce labeled data, specifically focusing on cross-domain pseudo-label bias and inefficient utilization of unlabeled samples. DAHCL introduces a Domain-Aware Learning (DAL) module to capture source-domain geometric characteristics and calibrate pseudo-label predictions, mitigating bias. Additionally, a Hierarchical Contrastive Learning (HCL) module combines dynamic confidence stratification with fuzzy contrastive supervision, allowing uncertain samples to contribute to representation learning without relying on unreliable hard labels. The framework was evaluated on three benchmark datasets (CWRU, PU, JUST) under severe noise (0 dB SNR) and substantial domain shifts, consistently outperforming advanced SSDGFD baselines and demonstrating superior robustness and generalization capabilities. The code is publicly available on GitHub.

Key takeaway

For AI Scientists and Machine Learning Engineers developing fault diagnosis systems in industrial settings with limited labeled data, DAHCL offers a robust approach. Its dual focus on mitigating cross-domain pseudo-label bias and effectively utilizing uncertain unlabeled samples can significantly improve diagnostic accuracy and generalization, especially under noisy conditions. Consider integrating domain-aware learning and hierarchical contrastive supervision to enhance model performance and reliability in real-world deployments.

Key insights

DAHCL improves semi-supervised fault diagnosis by correcting pseudo-label bias and leveraging uncertain unlabeled data through domain-aware and hierarchical contrastive learning.

Principles

• Domain-specific geometry informs pseudo-label calibration.
• Uncertain samples offer valuable weak semantic information.
• Dynamic confidence stratification improves unlabeled data utilization.

Method

DAHCL uses a Domain-Aware Learning (DAL) module to embed domain-specific geometry for pseudo-label calibration and a Hierarchical Contrastive Learning (HCL) module for stratified, fuzzy contrastive supervision of unlabeled samples.

In practice

• Incorporate engineering noise into SSDGFD evaluation protocols.
• Use exponential moving average for stable domain descriptor updates.
• Employ fuzzy proxies for uncertain samples in contrastive learning.

Topics

• Semi-Supervised Domain Generalization
• Fault Diagnosis
• Contrastive Learning
• Pseudo-Label Calibration
• Domain-Aware Learning

Code references

• JYREN-Source/DAHCL

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• YOTOnet: Zero-Shot Cross-Domain Fault Diagnosis via Domain-Conditioned Mixture of Experts — YOTOnet enables zero-shot cross-domain fault diagnosis in mechanical equipment using a domain-conditioned mixture of experts.
• Learning to Label: A Reinforced Self-Evolving Framework for Semi-supervised Referring Expression Segmentation — A reinforced self-evolving framework improves semi-supervised referring expression segmentation by learning to construct reliable pseudo-labels.
• Neural Collapse by Design: Learning Class Prototypes on the Hypersphere — Supervised learning can achieve optimal Neural Collapse geometry by unifying cross-entropy and contrastive methods on the hypersphere.
• Anti-causal domain generalization: Leveraging unlabeled data — In anti-causal settings, utilizing unlabeled data from multiple environments can improve domain generalization by penalizing model sensitivity to covariate shifts.
• RaLMPH: Reliability-aware Learning for Multi-Pathologist Harmonization in Whole-Slide Image Classification — RaLMPH uses a reliability field to harmonize multi-pathologist WSI annotations by identifying trustworthy local opinions.
• Federated Medical Image Segmentation under Real-World Label Noise: A Benchmark Suite for Noisy Label Learning Method Selection — The new benchmark suite enables realistic evaluation and selection of Federated Noisy Label Learning methods for medical image segmentation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.