Recent advances in accelerated multi-objective design of high-frequency structures using knowledge-based constrained modeling approach

摘要

Design automation, including reliable optimization of engineering systems, is of paramount importance for both academia and industry. This includes the design of high-frequency structures (antennas, microwave circuits, integrated photonic components), where the appropriate adjustment of geometry and material parameters is crucial to meet stringent performance requirements dictated by practical applications. Realistic design has to account for multiple objectives, which are often conflicting. Identification of available trade-offs (e.g., electrical/field properties vs. physical size and cost), otherwise essential from industry standpoint, requires multi-objective optimization. It is a computationally expensive endeavor as in most cases - for the sake of accuracy - the system evaluation has to be carried out using full-wave electromagnetic (EM) analysis. Attempting to solve EM-driven multi-objective (MO) tasks directly using population-based nature-inspired techniques may be prohibitive in terms of cost. Employing surrogate modeling techniques can lead to mitigation of the cost issue; however, construction of fast replacement models over broad ranges of the system parameters is expensive by itself, especially in higher-dimensional spaces. Recently, several approaches involving knowledge-based surrogate modeling approach have been proposed with the metamodels constructed over small regions of the parameter space containing the Pareto front. The latter are approximated using the sets of preoptimized reference designs and permit a dramatic reduction of the number of training points required to set up a reliable surrogate, thus reducing the overall cost of the MO process. This paper reviews the recent advancements in these methodologies, and demonstrates the benefits of domain confinement using the various techniques such as reference design triangulation, nested kriging, and modeling with explicit dimensionality reduction using spectral analysis of the reference set. Demonstration examples of multi-objective design of antenna and miniaturized microwave components are provided as well. (C) 2020 Elsevier B.V. All rights reserved.