A Field Reconstruction Method for EMI Near-Field Scanning Based on Wave Equation-Informed Neural Network

In this article, a near-field reconstruction method based on neural network technique is introduced for advancing the near-field technology in electromagnetic interference (EMI) measurement and diagnosis. A deep neural network (DNN) scheme is first proposed and further enhanced by embedding the wave equation into the training process. The electric/magnetic field measured by probes is approximated by minimizing a loss function to reconstruct a radiation source. Different to DNN, the wave equation-informed DNN (WE-DNN) incorporates both the residuals of the measured field data and the wave equation. Therefore, the WE-DNN is capable of efficiently interpolating and extrapolating the scanned field. The performance and efficiency of the proposed DNN and WE-DNN are investigated through numerical and experimental examples. The relative errors observed in the proposed WE-DNN in both numerical and experimental studies substantiate its superior accuracy compared to the DNN. Furthermore, the proposed WE-DNN also outperforms traditional source reconstruction (TSR) and cubic spline interpolation (CSI) methods. The proposed WE-DNN method holds great potential for electromagnetic field reconstruction.