1 April 2018 in Computer Methods in Applied Mechanics and Engineering
2 June 2025 in IEEE Transactions on Antennas and Propagation
Data-driven deep learning techniques have made notable advancements in modeling electromagnetic scattering problems. However, its accuracy on the testing dataset can be heavily reduced when data availability is constrained. To address the overfitting problem caused by limited data, we propose a novel hybrid physics-data-driven neural network (HPDNN), which incorporates both data loss and physical loss. Specifically, Maxwell’s equations are used as the physical constraint to evaluate the prediction accuracy of neural network, and the physical loss is combined with the data loss in the backpropagation for extrapolation tasks. The results show that the proposed method can achieve accuracy enhancement up to 15.7% over the typical data-driven method, and the field profile of prediction can be significantly improved with the aid of physical constrain. Besides, this work reveals that the convergence speed and accuracy of HPDNN are better than those of purely physics-informed neural network (PINN), exhibiting the importance of loss hybrid on accurate modeling of electromagnetics.
4 March 2020 in IEEE Transactions on Antennas and Propagation
An efficient and accurate spectral-element spectral-integral method is developed for rigorously modeling arbitrary inhomogeneous objects with doubly periodicity embedded in a layered medium background. It is an improvement over the traditional finite-element boundary-integral method. In the proposed approach, the domain decomposition between the interior spectral-element domain and the outer surface boundary-integral domain enables the boundary-integral domain to be solved by a far more efficient spectral integral method using a uniform mesh; without any loss of generality, it leaves the interior spectral-element domain with flexibility to model any inhomogeneous objects. For the boundary-integral domain, with a uniform mesh and rooftop basis functions, we can exploit the Toeplitz structure of the boundary-integral matrices with the fast Fourier transform algorithm, and apply the Calderón preconditioner to achieve a high iteration convergence rate that will not slow down as the problem gets larger. With the periodic layered medium Green’s functions, the layers without objects can be treated as background and the discretization is avoided. We show that this method is very promising for large-scale problems, because the CPU time and memory required with respect to the number of unknowns grows slowly. We validate this method with a commercial finite element solver and show its ability by solving a large-scale optical lithography problem.
28 March 2022 in IEEE Transactions on Antennas and Propagation
Deep transfer learning has a great success story in computer vision (CV), natural language processing, and many other fields. In this communication, we are going to push forward the deep transfer learning to the hydraulic fracture imaging problem by proposing a two-step approach: 1) train a convolutional neural network (CNN) to reconstruct target geometries by a relatively large amount of approximated field patterns generated from a simplified model and 2) fine-tune the top layers of transferred CNN by a small amount of true field patterns generated through a full model. The advantages include the rapid generation of large amount of data through the simplified model and the high reconstruction accuracy through a careful design of the deep transfer learning. The CNN trained through the deep transfer learning can accurately reconstruct the lateral extent and direction of fractures with unseen conductivity and white Gaussian noise, showing a notable acceleration/accuracy enhancement over the previous CNN trained by mixing a small number of true data into the approximated data as data augmentation.
20 September 2021 in IEEE Transactions on Antennas and Propagation
Thin dielectric or conductive structures are very common in many applications. Modeling such structures in the finite-element method can require a very dense mesh near the structures, which increases the computation cost and degrades the mesh quality. In this work, the surface current boundary condition is incorporated into the framework of the spectral-element spectral-integral method to efficiently model thin structures with double periodicity embedded in layered media. The thin structures are viewed as zero-thickness surfaces, and no dense mesh is required. Furthermore, a fast and convenient formulation is developed to evaluate the tangential field in the background of doubly periodic structures. Numerical experiments demonstrate the accuracy and efficiency of the proposed method by comparing with various reference solvers.
3 April 2023 in IEEE Transactions on Antennas and Propagation
The Wigner-Smith (WS) time delay matrix relates a system’s scattering matrix to its frequency derivative and gives rise to so-called WS modes that experience well-defined group delays when interacting with the system. For systems composed of nondispersive and lossless materials, the WS time delay matrix previously was shown to consist of volume integrals of energy-like densities plus correction terms that account for the guiding, scattering, or radiating characteristics of the system. This study extends the use of the WS time delay matrix to systems composed of dispersive and lossy materials. Specifically, it shows that such systems’ WS time delay matrix can be expressed by augmenting the previously derived expressions with terms that account for the dispersive and lossy nature of the system, followed by a transformation that disentangles effects of losses from time delays. Analytical and numerical examples demonstrate the new formulation once again allows for the construction of frequency stable WS modes that experience well-defined group delays upon interacting with a system.