Combined 3D FEA and Machine Learning Design of Inductive Polyphase Coils for Wireless EV Charging

Wireless power transfer (WPT) technologies are currently researched and developed for charging the batteries of electric unmanned air and ground vehicles. This paper presents systems with special polyphase inductive coils, which generate rotating fields and achieve high power density and efficiency. The complex geometry is modeled and studied with 3D electromagnetic finite element analysis (FEA). In order to reduce the substantial computational effort, machine learning techniques are proposed for surrogate modeling. A deep learning algorithm is introduced to capture the physics-based relationships between geometry and electromagnetic properties in inductive coils for wireless charging. Parametric models are systematically gener-ated and analyzed by 3D FEA to create a data base with hundreds of designs, which are then used as training and testing data for the machine learning model. A multi-input univariate output for the mutual inductance between the transmitter and receiver for an example two-phase WPT system is established. The outputs of the deep learning model are satisfactorily validated with 3.3 % NRMSE and a R2 value of 0.985.