Application of Convolution Neural Network in Hydrophobicity Classification

Evaluating silicone rubber outdoor insulators surface condition is crucial to ensure their health conditions. A hydrophobicity classification system first developed by Swedish Transmission Research Institute (STRI) and recently adapted by the IEC 62073 standards classifies the insulators to seven different classes. The system requires certain expertise to be able to implement the classification which may not be available in many utilities. The objective of this paper is use deep learning to classify non-ceramic insulators hydrophobicity. Moreover, the efficiency of deep learning will be compared with the traditional machine learning (ML) approach. A previous dataset will be used as the database for this paper. Different convolutional neural network (CNN) topologies will be investigated. It has been found that the prediction accuracy of using CNN is similar to classical ML algorithms with the advantage of being easier to implement.