A Data-Driven Fault Diagnostics Approach for Dual Active Bridge Converters

This paper presents a comprehensive study on open-circuit fault (OCF) diagnosis in dual active bridge (DAB) converters, employing both classical machine learning classifiers and deep learning techniques. The work addresses two hierarchical classification tasks: a 5-class fault pairing scenario and a more granular 9-class switch localization problem. The selected features were extracted from converter measurements and used to train various models, including decision trees, support vector machines (SVM), and a one-dimensional convolutional neural network (CNN1D) enhanced with an attention mechanism. Among the evaluated approaches, the deep learning models, particularly the CNN1D architecture with attention mechanism, consistently outperformed classical methods across both scenarios. The proposed model achieved up to 99.74% accuracy in the fault pair classification and over 82% accuracy in the switch localization task, all while maintaining a compact model size. These results demonstrate the effectiveness of lightweight deep learning architectures for accurate, efficient, and practical fault diagnosis in power electronic systems.