An Evaluation of Alternative Techniques for Monitoring Insulator Pollution

Abstract

Electrical utility companies are constantly seeking predictive techniques that indicate the appropriate moment for maintenance intervention, while aiming toward a continuous increase in service indices. This paper proposes the use of pattern-recognition techniques to build classifiers that diagnose the operational state of the insulation structure in an online application. Important results were achieved during the study, mainly related to the types of sensors and features that need to be applied during the diagnosis process. Ultrasound and current leakage sensors, very-high frequency antenna, and thermovision instruments were employed to acquire signals and images in order to construct recognition systems. A number of specific features were applied to verify their importance during the classification process. Features were obtained in the time, frequency, and wavelet domains. Two groups of pattern-recognition techniques were applied: linear (Fisher and Karhunen-Loeve) and nonlinear (artificial neural network). The results indicated that pollution deposit can be evaluated by the proposed techniques, especially when a combination of sensors is employed.