Condition Sensing of Overhead Line Insulator Strings Using an Advanced CNN-Based Classifier Based on Infrared Thermography

Abstract

This letter proposes an advanced convolutional neural network (CNN)-based classifier for detecting the contamination level of in-service insulator strings. The goal is to enhance condition monitoring of insulators and ensure safe and reliable power system operation under adverse weather conditions and polluted environments. All possible partial and full contamination cases of a string of three disc insulators have been considered. Infrared thermography images taken from a safe distance have been cropped to consider the effective portions before being fed into a convolution-based deep neural network. The classifier has been trained with a total of 1248 thermal images across 12 contamination classes achieving an accuracy level of 99.04%. The proposed classifier has also been compared with two other benchmark CNN models.