An adaptive yolo model for detection of faulty insulators in power transmission network using unmanned aerial vehicle

Abstract

Effective monitoring and problem detection are thought necessary for assuring continuous power supply. This study developed a machine learning approach to detect and diagnose insulator failures in power transmission lines by analyzing photos of cracked, polluted, and flash-overed insulators recorded using a drone. Through numerous rounds of the object recognition model YOLO (You Only Look Once), a diverse dataset of power transmission line photos was used to train and compare the performance of an enhanced YOLOv8x model versus YOLOv5n, YOLOv5x, YOLOv7, YOLOv7x, and YOLOv8n. Furthermore, approaches such as transfer learning and augmentation have been used to improve the model’s performance. The YOLOv8x model outperformed the other YOLO models tested, with an accuracy of 0.94, recall of 0.934, and Mean Average Precision (mAP) at 0.5 of 0.944 for insulator failure detection. The suggested fault detection machine learning approach combined with a dronebased system provides an adaptive fault monitoring system with high precision and low cost.