Fractal-ANN Tool for Classification of Impulse Faults in Transformers

Transformers are impulse tested in laboratory to assess their insulation strength against atmospheric lightning strikes. Inadequate insulation may cause transformer winding to fail during such tests. Detection of such faults is an important issue for repair and maintenance of such transformers. This paper describes the application of the concept of fractal geometry to obtain the features inherent in the impulse response of transformers subjected to impulse test. Fractal features such as fractal dimension (calculated by Higuchi, Kartz, Petrosian and Box counting methods), lacunarity and entropy has been used for collection of proper features from the current waveforms. Artificial Neural Network (ANN) has been used to classify the patterns inherent in the features extracted from Fractal analysis. The complex nature of transformer winding and its impulse response gives rise to a complex non-linear pattern of fractal features. In this regard, the application of ANN for pattern classification has greatly reduced the complexity and at the same time increased the accuracy in the fault localization and identification system. The proposed tool has been tested to identify the type and location of faults by analyzing experimental impulse responses of analog and digital transformer models.