Direct Inverse Modeling for Electromagnetic Components Using Gaussian Kernel Regression
Yuki Sato, Kenji Kawano, Hajime Igarashi · IEEE Transactions on Magnetics · 2022
This article proposes a novel direct inverse modeling (DIM) of electromagnetic (EM) devices using Gaussian kernel regression. In this method, the nonlinear multivariate relationship between design and electrical properties is represented by the Gaussian kernels. Once the regression function is built, the device parameters that lead to the required electrical properties can be directly computed using the Newton method. It is shown that the proposed DIM can find the multiple solutions in case of both 2-D and 3-D inductor design with less computing cost compared with the conventional modeling method.