Power transformer Internal winding faults detection and classification using AI based Classifiers

Abstract

Despite the widespread use of FRA as a winding deformation and diagnostic tool, no interpretation criteria have been established. As a result, many researchers are exploring potential methods. The most challenging task of FRA Interpretation is to identify and classify them for distinctive classes. In this study, popular AI based classifiers (Radial basis function, support vector machine, ensemble, K-NN) are implemented to classify the faults. In this context measurements on the transformer model under the healthy and various fault conditions (axial, radial deformation, disc fault, turn short circuits) are required. The frequency range of the transient signal is divided to obtain more information. A new feature of wavelet-based magnitude squared coherence (MSC) is proposed and other numerical, statistical indices are also validated with training and testing to enumerate suitable classifiers. The classification accuracy is assessed and compared with performance parameters that are well-known.