Neural network simulation of a pulse magnetiser for magnetising permanent magnets

The paper is concerned with a design and a validation of a neurocontroller for a pulse magnetiser for magnetising permanent magnets. The goal is to register the peak time and crest current in order to pick up an optimal intermittent duty conditions regime for the magnetiser. This is usually done by solving a set of coupled ordinary differential equations describing current waveforms and the temperature rise in the magnetising winding. The neurocontroller is based on a one-layer feedforward neural network which is trained using the Levenberg-Marquardt learning rule. We present the results produced by the neurocontroller and we compare them with the numerical and measurement results. The neurocontroller is intended to serve later as a part of a global optimising algorithm.