Open circuit fault localization in dual active bridge based simultaneous battery charging systems using multi label classification

This paper presents a multi-label fault localization framework for open-circuit fault diagnosis in a three-port Dual Active Bridge (DAB) converter using deep learning. The proposed method leverages time-frequency features extracted from midpoint voltage signals through the continuous wavelet transform (CWT) to generate multi-channel scalogram images. These images are then used to train a ResNet-18 convolutional neural network (CNN) for simultaneous detection and localization of single and multiple switch faults. The dataset was generated under various state-of-charge (SOC) and fault timing conditions to ensure comprehensive coverage of the converter. Simulation results demonstrate high diagnostic accuracy, with both micro- and macro-F₁ scores exceeding 99% on unseen test data. Compared to an equivalent multi-class ResNet-18 classifier, the proposed multi-label network achieved approximately 3% higher macro-F₁ score and better generalization to simultaneous fault conditions. Moreover, the model generalizes effectively to three-switch faults, achieving a micro-F₁ score of approximately 85% despite being trained only on single- and two-switch cases. Robustness analyses further confirm stable performance under dead-time variations, measurement noise, and sensor reduction, maintaining over 92% F₁-score accuracy with only two voltage sensors. In addition, real-time performance evaluation shows that the proposed framework achieves an online diagnosis latency of 136.8 ms. These findings highlight the model's effectiveness for fault diagnosis in multi-port power converters.