Combination of mathematical indices and probabilistic neural network to detect the type of winding fault in transformers

Abstract

This paper presents a new method for determination the type of transformer winding fault through transfer function (TF) analysis. For this purpose, probabilistic neural network (PNN) is used. Outset of all, the required measurements are carried out on two groups of transformers, under both intact and faulted conditions of different degrees in axial displacement, radial deformation, disc space variation and short circuit on the winding. Then, using algorithms based on mathematical methods, appropriate indices from frequency responses are extracted with the required accuracy. The extracted features are finally used as the inputs to PNN classifier in order to perform the multi-category fault classification. The obtained results reveal the ability of proposed method in comparison with two other distinguished methods.