Artificial Neural Network as Part of a Saturation-Level Detector Within the Transformer's Magnetic Core

This paper deals with an algorithm for saturation-level detection within the iron core of a transformer, where the focus of this paper is on an artificial neural network (ANN)-based method. The iron core can be considered as saturated when the value of the ratio between the magnetically nonlinear characteristic flux linkage and the magnetomotive force decreases, thus dropping under the saturation-level value. However, the signal that represents the dynamic inductivity is contaminated with noise, which is substantially increased within the vicinities of the reversal points of the hysteresis, thus making reliable iron core saturation-level detection almost impossible. Therefore, some existing algorithms for iron core saturation detection based on dynamic inductivity criteria often fail when approaching the reversal point on the hysteresis loop. This paper presents an ANN-based method where the different structures of the ANNs are compared and evaluated.