Solving Electromagnetic Scattering of 3D PEC Targets Based on Graph Neural Networks

In this study, a deep learning-based approach is proposed to address electromagnetic (EM) scattering of 3D perfect electric conductor (PEC) targets. Message passing graph neural networks (GNNs) are employed to build a surrogate model capable of predicting the surface currents of 3D PEC targets. Triangle meshes are converted into graph-structured data, allowing the 3D PEC targets to be represented as graphs that can be processed by GNNs. Training and testing datasets are generated by applying the method of moments to solve the combined-field integral equations (CFIE) of 3D PEC targets. Three basic types of 3D target geometries are considered, including spheroids, conical frustums, and hexahedrons. With the unique capability of adaptively managing unstructured data and varying unknown quantities, the proposed GNN model demonstrates good numerical precision. This study reveals the great potential of GNNs for 3D EM modeling.