Abstract

A new artificial neural network (ANN) technique is described in this paper to detect and localize turn-to-turn short fault in power transformers by analyzing of transfer function (TF). The ANN is trained by using back propagation learning algorithm. Frequency and amplitude of TF in resonance points are used as ANN inputs and fault location is used as an output. An especially designed double-disk transformer winding with power rating 1.2 MVA is used as a test object to verify the proposed method. All disks of this winding were accessible via a tap in order to form a variety of turn-to-turn short faults.