Classification of Power Transformer Faults Using ANN

This paper presents the use of artificial neural network (ANN) for classifying the faults in power transformers depending on Dissolved Gas Analysis (DGA). In order to identify initial faults on transformers, DGA is a valid method as it gives valuable information on the conditions of the transformers. Using DGA methods, defects that appear at an early stage on transformer insulation can be detected since several gases will be emitted and dissolved within power transformer that can be used for the prediction of transformer failure. As inputs for the multilayer perceptron (MLP), samples data of combustible gases generated within the power transformer tank were used. A neural network was fed by inputs from five gases per the IEC-60599 method to obtain the targeted classification outcome. Training accuracy of 94 % and testing accuracy of 93.3 were obtained.