Magnetostatic Analysis of SRM Supported by Hybrid FEM-ANN Method

In this article the hybrid method combining the FEM with artificial neural network is depicted. The goal of the method is to obtain the magnetostatic characteristics of the switched reluctance motor. The method focuses on maximal computation time reduction and simultaneously on keeping the solution accuracy. The model of the artificial neural network is based on reduced FEM simulation. For neural network training and testing selected data from FEM simulation were applied. The model of ANN was used to obtain full magnetostatic characteristics of the SRM. Comparison with numerical magnetic field solver proved the hybrid model reliability. To create the ANN model we applied APLAC programming language and embedded libraries of this software, for FEM simulation we used the FEMM tool. The novelty of the paper is the code, which automates date transfer between APLAC and FEMM software. With the use of the hybrid FEM-ANN method we obtained significant accelerations of the field computations.