The Application of Machine Learning for Computational Electromagnetics Solver Selection

The field of Computational Electromagnetics (CEM) consists of several techniques used to solve the electromagnetic response of arbitrarily shaped conducting objects. Each of these solver techniques requires specific geometrical and electromagnetic characteristics within the simulation setup to provide efficient and accurate results. Selecting the most efficient and accurate solver technique usually requires sufficient user experience. Similar to work done by [1] , this work aims to simplify this process by conducting different simulation setups, measuring the accuracy and efficiency each one provides and training a machine learning model to predict whether a given solver technique would be sufficiently accurate and efficient. The implication of this work is that great amounts of time could be saved by CEM engineers both in simulation and time required to gain the experience to choose the optimal solver technique themselves.