Magnetic Pole Shape Optimization Using an IIR-Type Neural Network

This paper presents a method for a magnetic pole shape optimization using a neural network. An IIR type neural network is proposed for this optimization. This neural network is characterized of delay elements, which exist at inputs and outputs of each neuron. At outputs, the signal is connected to input via delay elements in feedback form. These delay elements have a roll of memory action and stabilization due to feedback connection during BP learning process calculation. Hence, the computation for learning process based on back propagation algorithm converges faster than that of the ordinary feed forward type neural network, and is less in time. Finally, the conclusions were obtained that the proposed neural network could reduce the iteration process comparing with that of the ordinary feed forward type neural network, and useful for a magnetic pole shape optimization.