Application of Support Vector Machines to Locate Minor Short Circuits in Transformer Windings

Abstract

Turn-to-turn short circuit fault is one of the most important failures in transformer windings. Turn-to-Turn fault can be occurred due to removal of insulation between winding turns. Early detection of Turn-to-Turn fault (high impedance short circuit) can prevent a direct short circuit. Frequency Response Analysis (FRA) as a well-known method, introduced for transformer winding mechanical defect recognition, is able to recognize the turn-to-turn fault potential but its interpretation to detect exact location of high impedance short circuit fault is difficult. In this regard, producing a mapping between the frequency response and the exact location of each fault can be helpful. In this paper, this mapping was made using support vector regression (SVR). Extracted features from frequency responses are used to train and test SVR. The results show that this method is able to detect turn-to-turn faults in the transformer winding even in their early stages.