Insulator Faults Detection Based on Deep Learning
Mohamed Witti Adou, Huarong Xu, Guanhua Chen · 2019
Electrical insulators are mainly used in transmission system for electrical insulation and mechanical support purpose. Since the insulators are exposed to environment, they can be victim of stresses (electrical, mechanical, environmental) which can lead to bunch-drop of insulators. Bunch-drop of insulators can occur due to aging, overloading, corrosion and so on. So, in this paper we propose a new method which can detect bunch-drop of insulators. We propose an object detection algorithm (YOLO, you only look once) in order to perform insulator localization and its bunch-drop detection. YOLO is the stat-of-the-art object detection method which is based on deep learning. YOLO adopts supervised learning mode. It takes as input labeled dataset. So, we trained YOLOv3 on insulator dataset of 2000 images with corresponding labels for two classes, insulator and the defect. Features are extracted by those images using several convolutional layers (53 layers). Then logistic regression is used for performing classes probabilities and labels predictions. The advantage of YOLO over other object detection method is based on its speed; thus, it is fast and can process 45 frames per second. The experiment results show that the proposed method can effectively localize insulator and detect its bunch-drop.