Machine Learning Based Wavelet Decomposition Aided Surface Contamination Prediction of Outdoor Silicone Rubber Insulators

Condition monitoring of overhead line insulators by surface leakage current (SLC) study is an effective technique. The present work uses an experimental setup to acquire the leakage current signal from various contaminated silicon rubber insulators (SiR) surfaces. Some important features associated with contamination levels are extracted from the recorded SLC data. Further, these characteristics are used to identify contamination classes of SiR insulators. A wavelet decomposition method and four machine learning classifiers (namely, Random Forest (RF), Support Vector Machine (SVM), Gaussian Naïve Bayes (GNB), and k-Nearest Neighbor (k-NN)) are employed for feature extraction and feature classification, respectively. A comparative analysis of individual classifiers' execution time and accuracy level is studied. This paper also identified the most appropriate characteristics for predicting the contamination class of overhead line insulators.