Related Experiment Videos
Multi-source weighted domain adaptation guided mechanical cross-domain diagnosis method with cross-layer hybrid
Yun Kong1, Jie Zhang2, Qinkai Han3
1School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China; Tangshan Research Institute, Beijing Institute of Technology, Tangshan 063015, China; State Key Laboratory of Mechanical Transmission for Advanced Equipment, Chongqing University, Chongqing 400044, China.
ISA Transactions
|April 24, 2026
Summary
This study introduces a multi-source weighted domain adaptation (MSWDA) framework to improve cross-domain fault diagnosis. The MSWDA framework enhances diagnostic accuracy by prioritizing high-weight source domains and learning domain-invariant features.
Area of Science:
- Machine Learning
- Artificial Intelligence
- Fault Diagnosis
Background:
- Domain shifts in partial source-target domains significantly impede domain adaptation.
- This degradation negatively impacts transfer fault diagnosis performance.
Purpose of the Study:
- To propose a novel multi-source weighted domain adaptation (MSWDA) framework for enhanced cross-domain fault diagnosis.
- To address challenges posed by significant domain shifts in transfer learning scenarios.
Main Methods:
- A multi-source domain weighting strategy based on subspace similarity was developed to prioritize high-weight source domains.
- A cross-layer hybrid attention network was employed to strengthen domain-invariant features.
- A multi-objective collaborative optimization strategy was implemented to boost cross-domain diagnostic capabilities.
Main Results:
- The proposed MSWDA framework achieved the highest diagnostic accuracies of 98.48% and 96.08% on two mechanical transmission datasets.
- Performance was superior when compared against existing advanced methods.
Conclusions:
- The MSWDA framework demonstrates superior capabilities for cross-domain diagnostics.
- The proposed methods effectively enhance feature learning and diagnostic accuracy in the presence of domain shifts.