Healthy Research Rewards
ResearchHub is incentivizing healthy research behavior. At this time, first authors of open access papers are eligible for rewards. Visit the publications tab to view your eligible publications.
Got it
YZ
Yuteng Zhang
Author with expertise in Anomaly Detection in High-Dimensional Data
Achievements
This user has not unlocked any achievements yet.
Key Stats
Upvotes received:
0
Publications:
2
(0% Open Access)
Cited by:
0
h-index:
6
/
i10-index:
6
Reputation
Biology
< 1%
Chemistry
< 1%
Economics
< 1%
Show more
How is this calculated?
Publications
0

Integrating intrinsic information: A novel open set domain adaptation network for cross-domain fault diagnosis with multiple unknown faults

Yuteng Zhang et al.Sep 1, 2024
Existing popular domain adaptation approaches typically assume that the source and target domains share the same label set. However, in industrial scenarios, the equipment may encounter unknown fault modes because of the harsh operating environment, which limits the application of existing diagnostic methods. To address the above problem, an intrinsic information-guided open set domain adaptation network is proposed for cross-domain fault diagnosis with unknown faults. First, a similarity-based discrimination framework is constructed to enhance the robustness of the model for unknown samples, which learns the similarity between samples and fault prototypes to enhance the classification performance. Then, a multi-information integrated weighting module is designed to quantify the transferability of samples through enhanced domain similarity information learning and prediction uncertainty information learning methods. Additionally, a self-supervised neighborhood clustering learning method is constructed, which enables the model to learn structural information about the target domain and encourages the target samples to cluster closely for better separability. Finally, the weighted open set adversarial training framework effectively facilitates diagnostic knowledge transfer and unknown fault recognition. Comprehensive experimental results on two datasets demonstrate the effectiveness of the proposed method in addressing the open set cross-domain diagnosis problem, which achieves promising performance over the comparison methods.