Source Reconstruction of Array Antenna based on Physics Informed Neural Networks

초록

This paper presents a novel two-stage deep learning framework based on Physics-Informed Neural Networks (PINNs) for reconstructing surface current distributions on antenna arrays from complex-valued near field electric field measurements. Traditional source reconstruction methods suffer from high computational cost and limited physical consistency. To overcome these limitations, the proposed PINN model directly incorporates Maxwell's equations into the training process, enabling the prediction of physically consistent current distributions. In the first stage, the vector potential is estimated using the predicted electric field, while the second stage reconstructs the current distribution from the predicted vector potential. A physics-driven loss function guides both stages to ensure consistency with electromagnetic theory. Simulation results on 4x4 patch antenna demonstrate that the proposed approach significantly outperforms classical methods in term of fidelity to ground truth current distributions. This work highlights the potential of PINN-based inverse modelling for reliable and accurate fault diagnosis in complex electromagnetic systems. © 2025 IEEE.

키워드