Comparison of FEM and PINN-based prediction of the power transmission of ERS with small data sets

By combining a Finite Element model with a Deep Neural Network (DNN), various surrogate models are developed. These models predict the power transmission efficiency of an electric road system (ERS). The ERS consists of two coils which allow for inductive charging of electric vehicles (EV). To increase the power transmission efficiency, ferrite structures near the receiver coils are used to guide the magnetic field. The design of the ferrite and the exact positioning of the coils with respect to each other significantly impacts the transmission efficiency. Six corresponding parameters are varied to generate training data for arbitrary geometrical arrangements of the ERS. Different surrogate DNN models - generated with COMSOL and Python - are trained with these data and compared with respect to their prediction accuracy. The hyperparameters of the most suitable DNN are further improved.