基于图时频双流加权对抗网络开放集故障诊断方法
By using a graph-time-frequency dual-stream weighted adversarial network, combined with multiple loss functions and cross-attention mechanisms, the problem of difficulty in identifying new faults in the target domain by closed-set domain adaptive methods is solved, and the accuracy and stability of open-set fault diagnosis are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KUNMING UNIV OF SCI & TECH
- Filing Date
- 2025-09-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing closed-domain adaptive methods cannot identify new fault types when they appear in the target domain, and collecting the complete dataset is time-consuming and expensive. They cannot effectively identify unknown fault categories in the target domain, thus limiting their practical application in engineering.
A graph-based time-frequency dual-stream weighted adversarial network is adopted. By combining the time-frequency dual-stream network, extended classifier, auxiliary domain discriminator, auxiliary classifier and domain discriminator, with cross-attention mechanism and multiple loss functions, the known fault classification and unknown fault identification in the target domain can be achieved.
In open set scenarios, the model achieves accurate classification of known faults and identification of unknown faults in the target domain, improves the generalization ability and stability of the model, reduces interference from unknown categories, and ensures the stability of diagnostic performance.
Smart Images
Figure CN121188570B_ABST