Condition monitoring of oil-impregnated paper bushings using extension neural network, gaussian mixture and hidden markov models

In this paper, a comparison between Extension Neural Network (ENN), Gaussian Mixture Model (GMM) and Hidden Markov Model (HMM) is conducted for bushing condition monitoring. The monitoring process is a two-stage implementation of a classification method. The first stage detects whether the bushing is faulty or normal while the second stage classifies the fault. Experimentation is conducted using dissolve gas-in-oil analysis (DGA) data collected from bushings based on IEEEc57.104; IEC60599 and IEEE production rates methods for oil-impregnated paper (OIP) bushings. It is observed from experimentation that there is no major classification discrepancy between ENN and GMM for the detection stage with classification rates at 87.93% and 87.94% respectively, outperforming HMM which achieved 85.6%. Moreover, HMM fault diagnosis surpasses those of ENN and GMM with a classification of 100%. However, for diagnosis stage HMM outperforms both ENN and GMM with 100% classification rate. ENN and GMM have considerably faster training and classification time whilst HMM's training is time-consuming for both detection and diagnosis stages.