A fuzzy ARTMAP fault classifier for impulse testing of power transformers

Abstract

The work presents an artificial intelligence (AI) based impulse test technique for oil filled power transformers. Determination of exact nature and location of faults, during impulse testing of large power transformer is of practical importance to the transformer manufacturers as well as designers. The presently available impulse test techniques more or less depend on expertise of the test personnel, and in many cases lead to ambiguity and controversy. The new AI approach presented in the paper overcomes the limitations of conventional test methods. This new technique relies on high discrimination power and excellent generalization ability of fuzzy neural networks in complex pattern classification problem. The proposed method employs a fuzzy ARTMAP pattern recognition technique to recognize the frequency responses of the winding admittance of high voltage transformers under healthy and different faulty conditions of winding insulation, and learns to establish the correlations between the nature and physical location of occurrence of an internal insulation fault in a transformer winding and its associated frequency response. The technique was tested on the winding model of typical high voltage transformer and yielded high diagnostic accuracy by successful detection and discrimination of faults of different nature and different site of occurrence in the high voltage winding.