Fast Parametric Prediction of Coil Coupling in Wireless Power Transfer Systems Using Physics-Informed DeepONet

Evaluation of efficiency and safety is an important requirement for the design of Wireless Power Transfer systems. Numerical methods such as the Finite Element Method offer high accuracy but incur in high computational costs and are limited for real-time optimization and large-scale designs; in addition, uncertainties in the geometrical configuration of the system, for instance due to lateral misalignment and vertical offset, would require specific parametric design. In this work, a Physics-Informed Deep Operator Network (PI-DeepONet) framework offers fast parametric modeling of the systems. This approach, by integrating physics constraints with FEM-generated data, enables efficient prediction of spatial magnetic flux density distributions across a wide range of misalignment and air-gap conditions and reduces the computational effort. A truncated Fourier series expansion of the magnetic vector potential is employed, and the zeroth-order (axisymmetric) component is enforced to satisfy Maxwell’s equations, while higher-order harmonics are learned from Finite Element Method data to reconstruct non-axisymmetric field components. Numerical results show that the proposed PI-DeepONet achieves a relative (L2) error below 2.6% in the prediction of the magnetic vector potential. The implementation of the proposed method is publicly available at https://github.com/HajianMahshad/PI-DeepONet-WPT.