Gestão técnica de isoladores poliméricos utilizando aprendizado de máquina.

Abstract

This work proposes a methodology for the technical management of polymeric insulators based on the correlation between attributes, obtained by the detection of infrared radiation, ultraviolet light radiation and ultrasonic acoustic emission. Therefore, 60 insulators of class 138 kV removed from operation were used in laboratory tests for inspection and obtaining the attributes. The attributes were initially analyzed using boxplot graphics with the aim of identifying and removing outliers. Next, the k-means algorithm was used to divide the database in order to divide the insulators into three groups with different operating patterns. These groups were used as a reference in the creation of a classification model by artificial neural networks of the operational state of insulators that enabled the classification of samples in which the operational state is unknown. The developed methodology was effective in classifying the operational state of polymeric insulators in a non-invasive way, through the joint application of inspection techniques associated with machine learning algorithms in a non-supervised way. The methodology proved capable of providing the technical management of polymeric insulators, providing the greatest possible use of the insulators' useful life without compromising the safety of the electrical system, thus increasing the reliability, continuity and availability of the transmission lines.