Efficient modeling and design optimization of computationally intensive high frequency structures

计算机辅助设计（CAD）的主要元素之一是建模过程。 建模使我们能够预测结构对给定系统参数值的响应，而不必实际建造和测量。另一个同样重要的过程是设计优化。 使用优化算法（优化器），我们能够确定设计下的系统参数的最佳值，这将导致所需的响应。 优化过程涉及针对不同参数值对系统的时间密集型重复模拟。 在这项工作中，我们的目标是开发新的技术，有效和准确的建模计算密集型高频结构。 这些技术将利用传输线的保形网格，能够精确地建模具有任意形状和材料特性的结构，并尽可能少地存储内存。 我们还将应用新的优化方法，利用电磁（EM）求解器的独特属性。 有效的伴随灵敏度分析技术将扩展到色散材料的结构，其材料特性随频率变化。 我们将开发用于构建“代理”模型的方法，这些模型能够准确地预测参数空间的广泛区域内的EM模型响应，但评估速度要快得多。 我们开发的算法将达到所需的最佳值，只使用几个EM模拟。 我们预计，所提出的研究将有显着的影响，在减少建模和优化时间的许多高频组件，如大型天线阵列，电磁波和人体的相互作用的建模，以及建模的电气大的光子和光电器件。 这将缩短产品的上市时间，并为高频结构的设计者增加有价值的工具。

One of the main Computer Aided Design (CAD) elements is the process of modeling. Modeling allows us to predict the response of a structure for given values of the system parameters without having to actually build it and measure it. Another procedure of equal importance is that of design optimization. Using an optimization algorithm (optimizer), we are able to determine the optimal values of the parameters of the system under design that would result in the desired response. The process of optimization involves time intensive repeated simulation of the system for different parameter values. We aim in this work to develop novel techniques for efficient and accurate modeling of computationally intensive high frequency structures. These techniques will utilize conformal meshes of transmission lines capable of accurately modeling structures with arbitrary shapes and material properties with as little memory storage as possible. We will also apply novel optimization approaches that exploit the unique properties of electromagnetic (EM) solvers. Efficient adjoint sensitivity analysis techniques will be extended to structures with dispersive materials whose material properties vary with frequency. We will develop methodologies for constructing "surrogate" models that are capable of accurately predicting the EM model response over a wide region of the parameter space but are much faster to evaluate. Our developed algorithms will reach the desired optimal values using only few EM simulations. We expect that the proposed research will have a significant impact in reducing the modeling and optimization time of many high frequency components such as large antenna arrays, the modeling of the interaction of EM waves and the human body, and the modeling of the electrically large photonic and optoelectronic devices. This will reduce the time-to-market time of products and add valuable tools to the designer of high frequency structures.