A Topology Optimization of Electromagnetic Devices Based on Kernel Ridge Regression as a Variant of Gaussian Network-Based Shape Representation

This article presents an evolutional topology optimization method for electromagnetic devices based on kernel ridge regression (KRR). For this method, device shapes are represented through a spatially smooth function, whose output is determined on the basis of KRR. The present method makes it easy to find some structures with alternating air region, such as multibarriers in synchronous reluctance motor (SynRM). It is shown that the present method is a variant of the shape expression method based on the normalized Gaussian network (NGnet)-based approach. Nevertheless, it is shown from optimization results of magnetic shield and SynRM that the present method can find smooth but more complicated shapes in comparison with the conventional NGnet-based method.