Classification of Defects in Outdoor Ceramic Insulators using Machine Learning

Outdoor ceramic insulators are widely used in both transmission and distribution overhead lines. Many of these insulators either exceeded or approaching the end of their expected life. The failure of outdoor insulators can lead to significant economic losses. Hence, it is paramount to develop non-intrusive techniques to detect the likely hood of outdoor insulators failures. This paper presents a system combining an ultrasonic sensor with a bandwidth between 20-100kHz and a machine learning classification model to detect various defects in ceramic insulators, including internal defects, pollution, and corona. The approach involves extracting both statistical and spectral features from the time-domain ultrasonic signals. Key features are selected using Causality Analysis and Information Gain, focusing on the top 10 most significant features. K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and XGBoost (XGB) have been used as classifiers. The system demonstrates an overall classification accuracy exceeding 93% across four different machine learning models.