.png){kind=icon}
.png){kind=icon}
Return
Zero-Shot Open Circuit Fault Diagnosis of Converter Based on Prior-Knowledge Injected Dual Bayesian Network With Attention Mechanism
.png){kind=icon}
PRE
AI
G
S
J
F
H
J
L
DOI：10.1109/tie.2026.3677610.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Power electronic systems serve as core infrastructure for modern energy conversion and power control. Their reliability is crucial for the stable operation of key fields, such as industrial power systems, transportation, and aerospace. Open-circuit faults are difficult to detect, which can lead to serious consequences. While effective in data-rich environments, existing data-driven approaches’ reliance on extensive labeled data hinders their practical deployment. The reason is that acquiring comprehensive fault data in real-world settings is often prohibitively expensive. To address this, this article proposes a dual Bayesian zero-shot fault diagnosis approach based on prior-knowledge injected with attention mechanism (PKIDBAAM). This approach constructs a fault attribute space with clear physical meaning by analyzing the physical relationship between the Vienna rectifier topology and fault modes. It employs a multihead attention mechanism to dynamically fuse multidomain fault features from three-phase current signals. A dual Bayesian inference framework combining naive Bayes and integrated multiple Gaussian Bayesian algorithms is designed to achieve high accuracy recognition of inaccessible open-circuit faults. The verification results on a Vienna rectifier experimental platform show that it significantly outperforms conventional classical machine learning models, which provides a new solution for intelligent fault diagnosis in power electronic systems.
Keywords:
Dual Bayesian network
fault diagnosis
prior-knowledge
power converter
zero-shot learning (ZSL)
Journal
IF:
7.2
Papers:
1.8W
Citations:
9.8W
Organization
Cited Papers