ANN inverse mapping technique applied to electromagnetic design

Abstract

This paper examines the use of inverse mapping for the design of electromagnetic devices whose performance is described by an artificial neural network (ANN). In cases where the function describing the inverse problem is ill-conditioned, it may be very difficult to train a neural network to learn the inverse problem. In these cases, an iterative method based on an inverse mapping technique may be employed. When provided with the desired output vector, this method uses backpropagation to train (solve) for the unknown parameters in the input vector. The results of using the ANN based inverse mapping approach in the design optimization of a hollow transmission line are presented.