This study aims to propose a fault diagnosis method for partial discharge (PD) in high-voltage (HV) potential transformers (PTs) by combining discrete wavelet transform (DWT), scatter plot (SP), and a residual neural network (ResNet) deep learning model for feature extraction and classification. First, models of HV PTs under normal conditions and three internal fault types were established, including coil eccentricity, voids between the primary winding and the core, and voids between the primary and secondary windings. After measuring the PD signals, DWT filtering was applied to process the signals, and the filtered PD signals, together with the fundamental voltage signals, were transformed into an image-based feature SP to represent the characteristics of each fault. Finally, the SPs were trained using the ResNet model to identify four different defect types in HV PTs. Experimental results showed that the proposed method achieves a fault identification accuracy of 98%. Additionally, compared to other deep learning models, the proposed method significantly improves diagnostic efficiency and accuracy. This study also developed an intelligent online fault monitoring and predictive maintenance system for HV PTs to enhance the stability of power grids and equipment.
Hung et al. (Fri,) studied this question.