Transformer insulation condition monitoring using artificial neural network

Partial Discharge (PD) is the most hazardous phenomena that deteriorates the insulation life of an electrical device in a covert way (and finally with an overt puncture or flashover) in an otherwise healthy insulation. Mostly the conventional EHV and UHV transformers used in the present times employ the oil impregnated paper/pressboard insulation. Recently, a radically new approach to the PD pulse-pattern analysis in oil impregnated pressboard insulation has been established in a relevant study, where an experimental setup was prepared to simulate the commonly encountered wedge shaped oil gaps between the edges of rounded conductors and pressboard spacers inside transformer. The observed PD pulse patterns at different stress levels are classified into five distinct categories with each one having some relevance to the deteriorating condition of insulation health. The interesting results revealed by this study are serving as the underlying principle to carry out an ANN based insulation condition monitoring system. A backpropagation learning model multi-layer perceptron (MLP) paradigm is characterised, which is used as a tool to further automate and achieve a fast classification of an unknown PD pulse pattern to a particular class (without visual or statistical interpretation), thus enabling addressing the insulation health condition or imminent breakdown probabilities. Diverse structures of MLP were implemented using single or double layers with different number of processing elements in each layer and different learning coefficients. The performance is studied and the intricacies involved are addressed in the light of achieving an optimal structural configuration