| Title |
Study on Classification for Phase Resolved Partial Discharge Pattern Based on CNN Algorithm |
| DOI |
http://doi.org/10.5207/JIEIE.2025.39.2.155 |
| Keywords |
Classification; Convolutional neural network; Deep learning; Grad CAM; Phase resolved partial discharge; Visual geometry group |
| Abstract |
Partial discharge (PD) refers to a localized dielectric breakdown occurring in a solid or liquid electrical insulation system under high voltage stress, without fully bridging the space between conductors. It is critical as it can significantly impact the reliability and lifespan of high-voltage electrical equipment such as transformers, switchgear, and cables. This paper presents an artificial intelligence (AI) model designed to classify patterns of various partial discharges. To evaluate the model, training data for each type of partial discharge, generated through UHF(Ultra High Frequency) sensors, were collected. These data were then converted into 2D representations using Phase Resolved Partial Discharge (PRPD) analysis. The proposed models were built on deep learning algorithms, specifically the Visual Geometry Group (VGG), which is a type of Convolutional Neural Network (CNN). The classification accuracies achieved for DI, FE, and PE patterns were 99%, 94.5%, and 90%, respectively. Also, Grad-CAM was employed to provide class-discriminative, high-resolution visualizations, effectively demonstrating the significance of the training data. |