Condition Assessment of High Voltage Insulator using Convolutional Neural Network

An insulator can be characterized as a device used for insulating and holding conductors or equipment which are subject to the difference in electric potential as stated in IEC 60050-471-01-10. Airborne contamination and weathering are the major factors that influence the operating condition and service life of the power line insulator. Due to outdoor harsh environment conditions and aging power line insulators deteriorate rapidly and fails before the estimated life. Various defects intended to the severity of outdoor insulation is essential for the prediction and life expectancy of in-service insulators. The application of Deep Learning (DL) which is a subgroup of Machine Learning (ML) helps to predict the power line insulator conditions such as defected insulator, surface corrosion, material degradation, and hydrophobicity loss state of the polymeric insulators etc. This work primarily focused on the image classification (binary and multi-class) of the insulator dataset with the help of Convolutional Neural Network (CNN) by varying different optimizers like Adagrad, Adadelta, Adam, Adamax, Nadam, RMSprop, and Ftrl to find out with which combination of neural network and optimizer will classify more accurately. In this work for the binary image classification case, CNN with Adamax optimizer provided overall accuracy of 95.27% and for the multi-class image classification case, CNN with the same Adamax optimizer gave overall accuracy of 98.20%.