Recognition of Partial Discharge Signatures using Convolution Neural Network

This research work mainly focuses on classifying the different types of faults located in high voltage (HV) transformers using a convolution neural network (CNN). The study mainly focused on classifying the types of faults in HV transformers using partial discharge (PD) signatures without human-interpreted feature extraction approaches. In this proposed study, two different types of the database with six different types of faults include needle tip fault (S1), free metal particle fault (S2), and floating electrode fault (S3). PD can happen in single or several signatures in reality. A measured PD signal from the laboratory is transformed into 3-Dimensional given fed as input to deep learning approaches. Totally 3000 signatures are considered in this work for recognition. Using the two databases (DB1 and DB2), three different scenarios are considered to strengthen the proposed algorithm. The performance of the CNN is increased by tuning the hyperparameters. The findings demonstrate the viability of the suggested strategy for identifying the various fault in the HV transformer issue.