In this paper, disk space variation, radial deformation, short circuit, and axial displacement as four common transformer winding faults, in different locations and with various extents, were practically applied to 20 kV winding of a 1.6 MVA distribution transformer. For classification of the fault type, its location, and extent, a new calculation method for transfer function (TF) comparison indices called windowed calculation was proposed. The whole frequency range of the TF is scanned in this method. It was shown that the presented method increases the visual fault detection ability even for fault-type detection and significantly enhances the accuracy of the classification winding faults type, location, and extent. In this study, Fisher discriminant analysis (FDA) was utilized for dimension reduction and feature extraction. The ability of the FDA for maximizing the between-class separability while minimizing their within-class variability was utilized for this application so as to decrease the dimension and extract appropriate features from a lot of calculated features by the proposed method.
