Power transmission fault detection using hyperparameter-tuned YOLOv11 with XAI

Abstract

The secure and stable operation of power transmission lines is vital for maintaining uninterrupted electricity supply. Faults in these lines, caused by complex natural conditions or aging infrastructure, can lead to significant disruptions, highlighting the need for efficient fault detection. Existing methods face challenges such as complex backgrounds, multi-target scenarios, varying lighting conditions, and limited fault categories. This study addresses these limitations by introducing an improved fault detection framework based on YOLOv11, enhanced with hyperparameter optimization using a genetic algorithm. The incorporation of XAI provides interpretable model outcomes, offering actionable insights to support maintenance and monitoring tasks. Additionally, data augmentation techniques were employed to improve detection in underrepresented fault categories. Unlike previous approaches that focus on specific faults, such as insulator defects, this work identifies a wider range of fault scenarios. The proposed framework achieves an mAP@0.50 of 0.94 and an F1 score of 0.92 on the PTL-AI Furnas dataset, exceeding the benchmark result, demonstrating its effectiveness in addressing various fault detection challenges, and highlighting its potential for practical applications in maintenance and monitoring of power transmission lines.