Abstract

Flashover due to contamination of glass insulators is common in electrical systems. An alternative to prevent such failures is the use of silicone-coated glass insulators. Silicone is hydrophobic, which prevents humidity from being deposited on the surface; therefore, the adhesion of environmental contaminants on the insulator is reduced. This prevents breakdowns and enables less frequent maintenance. However, in certain areas, silicone can lose its hydrophobic property, which causes contaminants to accumulate on the insulator surface. Consequently, evaluating the condition of silicone-coated insulators is essential to ascertaining their need for replacement. This requires a method that distinguishes between contamination and aging of the insulators. This paper presents the design of a machine learning classifier using a support vector machine (SVM) model that uses leakage current parameters to classify the condition of the silicone. As a distinguishing feature, the SVM classifier employs the root mean square (RMS) impedance magnitude and harmonic impedances as input parameters. This achieves an accuracy of 99% compared to 65% with data that lacked impedance measurements. The tests were conducted with insulators removed from a 500 kV line and evaluated within an artificial fog chamber.