Enhancing Power Grid Reliability: Comparative Analysis of YOLO Models for Insulator Fault Detection

This paper presents the development of an insulator fault detector aimed at enhancing the reliability of power grid infrastructure. The study leverages deep learning techniques to train and evaluate distinct object detection models, YOLOv8-M, YOLOv7 and YOLOv10-M, the primary objective is to assess the precision of these models in identifying insulator faults. The YOLOv8-M model achieves an impressive precision rate of 94.9% in identifying insulator faults within the test dataset. This performance underscores the effectiveness of YOLOv8-M in accurately detecting insulator defects, indicating its potential utility in real-world power grid maintenance scenarios. To provide a comprehensive benchmark, the study also evaluates the YOLOv7 model on the same dataset. The YOLOv7 model, while demonstrating commendable performance, achieves a slightly lower precision rate of $\mathbf{9 3. 6 \%}$ in identifying insulator faults within the test dataset. The comparative analysis between these two YOLO variants sheds light on their respective capabilities for insulator fault detection. The results indicate that YOLOv8-M holds a slight precision advantage while offering faster training, making it a compelling candidate for enhancing the fault detection mechanisms crucial for ensuring the uninterrupted operation of power distribution systems. This research contributes to the growing body of knowledge in the field of computer vision applications for power grid infrastructure maintenance and paves the way for more robust and reliable fault detection solutions.