The use of finite elements and neural networks for the solution of inverse electromagnetic problems

A method that combines a neural network (NN) and the finite-element method is introduced for solving inverse electromagnetic field problems. This forms the basis for design synthesis. A two-layered NN with one-pass training is used in this scheme. It uses the information from the finite-element analysis for training and is very efficient and stable. The one-pass training of the NN leads to a time efficient scheme. The finite-element method is used to produce the training patterns and to analyze the optimized solution, and the neural network is used to optimize the parameters. With the use of the trained NN for optimization, the solution time for design optimization is reduced. An example of its use in the optimization of a permanent-magnet rotor configuration is presented