Accurate Hydrophobicity Grade Detection of Polymeric Insulators in Extremely Wetted and Humid Environments Using Bi-LSTM Neural Network Classifier

An accurate detection of hydrophobicity grade (HG) is essential for reliable condition monitoring of polymeric outdoor insulators in wetted and humid environments and for increasing their service life as well. With the above context, this paper proposes a novel HG detection methodology by incorporating a local binary pattern texture feature infused with bi-directional long short-term memory (bi-LSTM) neural network classifier. Different experiments were carried out on 11 kV silicone rubber (SiR) polymeric insulators to emulate various hydrophobic conditions, and the images of the water droplets on the insulator surface were captured. After that, texture features were extracted from the images using a suitable pre-processing of the acquired images and the LBP technique. The extracted features were subsequently fed to a bi-LSTM classifier for HG classification, which returned high recognition accuracies in classifying different hydrophobicity grades. The proposed HG detection technique is suitable and can be implemented for remote condition monitoring purposes.