Artificial Magnetic Conductor Unit Cell Design Using Machine Learning Algorithms

Abstract

Commercial electromagnetic (EM) simulator tools solve complicated Maxwell’s equations to design and optimize electromagnetic devices, which is computationally expensive and time consuming. There is a dire need to solve complex electromagnetic problems with least amount of computational resources in a short time. This work proposes the application of machine learning techniques in design process of electromagnetic problem. For the proof of concept, we demonstrated an optimum design process of an artificial magnetic conductor, which is a metasurface unit cell, by applying machine learning algorithms namely, artificial neural network (ANN), k-nearest neighbor (KNN), support vector machine (SVM), extreme gradient boosting (XGBoost), and least absolute shrinkage and selection operator (LASSO). The performances of these machine learning optimization models were evaluated on the test data set based on root mean squared error (RMSE) values. To the best of our knowledge, this is the first work that yields an excellent match with the original EM results from a commercial simulator tool with very small training dataset. Thus, it obviates the need of using computationally expensive and time-consuming electromagnetic simulators and massive training datasets for data-driven design approach of complex electromagnetic problems.