Digital Twin-Based Graph Convolutional Network for Rectifier Fault Diagnosis
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 3519715-3519715
DOI: 10.1109/tim.2026.3714609 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Haocheng Wang und 1 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Power electronic converters, such as rectifiers, are essential in industrial production, yet their fault diagnosis remains challenging in practical applications. The application of conventional deep learning models is restricted by parameter drifts in harsh operating environments and the scarcity of destructive real fault data. Although digital twin (DT) technology provides a solution for label-scarce scenarios, the physical–virtual mismatch remains a persistent application bottleneck. To address rectifier fault diagnosis, this article proposes a novel physical–virtual discrepancy-driven DT graph convolutional network (GCN) architecture. This architecture shifts from attempting to eliminate or conceal the model mismatch to actively utilizing this discrepancy to drive the neural network. First, based on the intrinsic electrical coupling mechanisms of the rectifier, a topological prior is established to rigorously constrain the feature aggregation process. Second, rather than treating the modeling error as a system defect, an interpretable physical–virtual discrepancy observer is introduced to identify the dynamic deviation between the physical entity and the nominal virtual twin in real time. This precisely quantified discrepancy is not treated as a simple concatenated feature; instead, it serves as a macroscopic contextual condition to directly drive the neural network. Furthermore, combining twin data with a metalearning strategy further improves the generalization capability of the architecture. Extensive experiments on a three-phase rectifier platform demonstrate that the proposed framework exhibits excellent diagnostic accuracy and robustness under parameter drifts and label-scarce scenarios, providing a new avenue for label-efficient and interpretable fault diagnosis in industrial applications.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Wang, H., & Yang, Y. (2026). Digital Twin-Based Graph Convolutional Network for Rectifier Fault Diagnosis. IEEE Transactions on Instrumentation and Measurement, 75, 3519715-3519715. https://doi.org/10.1109/tim.2026.3714609
Vancouver
Wang H, Yang Y. Digital Twin-Based Graph Convolutional Network for Rectifier Fault Diagnosis. IEEE Transactions on Instrumentation and Measurement. 2026;75:3519715-3519715. doi:10.1109/tim.2026.3714609
BibTeX
@article{Wang2026,
author = {Wang, Haocheng and Yang, Ying},
title = {Digital Twin-Based Graph Convolutional Network for Rectifier Fault Diagnosis},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {3519715--3519715},
doi = {10.1109/tim.2026.3714609}
}
EUVIMED