Short-circuited turn fault detection in electrical transformers based on frequency domain features

Short circuits in windings are a major factor contributing to the damage observed in electrical transformers. Therefore, early detection during the initial stages is vital to prevent more extensive damage. This paper proposes an approach for detecting short circuits through vibration analysis. The proposed methodology enables the analysis of various conditions, ranging from a healthy state to six levels of short-circuit turns in an unloaded transformer. A combination of frequency domain features extraction is employed after applying the FFT, principal component analysis (PCA), and classification with techniques such as K-nearest neighbors (KNN) or Support Vector Machine (SVM). The proposed approach accurately determines the extent of damage present in the windings, and the results show the effectiveness of this method in precisely identifying the severity of damage in the transformer.