Related Experiment Video
Updated: Jun 28, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
529
Contrast-Assisted Domain-Specificity-Removal Network for Semi-Supervised Generalization Fault Diagnosis
Summary
This study introduces a novel network for intelligent fault diagnosis that improves model generalization across unseen conditions. It effectively handles domain shift using partially labeled data, enhancing reliability in real-world applications.
Area of Science:
- Machine Learning
- Artificial Intelligence
- Engineering
Background:
- Intelligent fault diagnosis models suffer performance degradation due to unknown domain shift when target domains are unavailable during training.
- Existing domain generalization (DG) methods often require multiple fully labeled source domains and neglect domain-specific information variations.
Purpose of the Study:
- To propose a novel network for reliable generalization fault diagnosis using partially labeled source domains.
- To extract transferable features by disentangling domain-invariant and domain-specific information.
Main Methods:
- Introduced a contrast-assisted domain-specificity-removal network (CDSRN).
- Employed a domain-specific feature removal branch to isolate domain-invariant features.
- Integrated a proxy-contrastive representation enhancement module for improved feature learning.
Main Results:
- The CDSRN effectively extracts generalized information from the domain-invariant feature dimension.
- The network demonstrates improved fault class-discriminative and domain-discriminative feature learning.
- Experimental studies confirm the effectiveness and competitiveness of CDSRN in semi-supervised generalization fault diagnosis.
Conclusions:
- The proposed CDSRN achieves reliable generalization fault diagnosis with partially labeled source domains.
- The method successfully mitigates the impact of domain shift, enhancing model robustness for unseen working conditions.
