Classification and Prediction of Hydrophobicity in Polymeric Insulator using Convolutional Neural Networks

Polymeric insulators are widely used in the power systems mainly due their superior hydrophobic property. These insulators being organic in nature, degrades gradually when they are continuously exposed to external environment. This degradation leads to loss of surface hydrophobicity culminating in a flash over which in-turn can lead to a catastrophic collapse of the power system. Evaluation of hydrophobicity to test the performance of the insulators has proved to be extremely useful. The hydrophobic characteristic of new polymeric insulator surfaces causes discrete water droplets to form on the surface of the insulator. However, when the insulation ages, the surface loses its hydrophobicity, causing continuous water channels to emerge. These channels then cause dry band arcing and even flashover, which has an impact on the insulators' long-term performance. As a result, accurate wettability sensing of polymeric insulators is crucial for trustworthy insulator diagnosis. Taking into account the aforementioned fact, the current study analyses the categorization of polymeric insulators in hydrophobicity classes using convolutional neural networks in accordance with the spray method of IEC Standard 62073. Using an image dataset, convolutional neural networks were trained, validated, and tested to identify the hydrophobicity class of polymeric insulators.