Transformer incipient fault diagnosis based on probabilistic neural network

This paper proposes a technique for transformer fault diagnosis. The proposed technique is a four-layer probabilistic neural network (PNN). The proposed diagnostic technique has faster training capability because it is build with a single pass of exemplar pattern set and without any iteration(epochs) for weight adaptation. This diagnostic technique uses normalized parts per million values of gases (hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2)) as an input to detect partial discharge, low energy discharge, high energy discharge, low & medium temperature fault, high temperature transformer faults. The effectiveness of the proposed diagnostic approach is verified on the basis of the experiments on transformer oil dissolve gas samples. The results indicate that the PNN approach can be successfully used for transformer faults diagnosis.