Rapid Multi-Criterial Antenna Optimization by Means of Pareto Front Triangulation and Interpolative Design Predictors

Modern antenna systems are designed to meet stringent performance requirements pertinent to both their electrical and field properties. The objectives typically stay in conflict with each other. As the simultaneous improvement of all performance parameters is rarely possible, compromise solutions have to be sought. The most comprehensive information about available design trade-offs can be obtained through multi-objective optimization (MO), typically in the form of a Pareto set. Notwithstanding, MO is a numerically challenging task, in a large part due to high CPU cost of evaluating the antenna properties, normally carried out through full-wave electromagnetic (EM) analysis. Surrogate-assisted procedures can mitigate the cost issue to a certain extent but construction of reliable metamodels is hindered by the curse of dimensionality, and often highly nonlinear antenna characteristics. This work proposes an alternative approach to MO of antennas. The major contribution of our work consists in establishing a deterministic machine learning procedure, which involves sequential generation of Pareto-optimal designs based on the knowledge gathered so far in the process (specifically, by triangulation of the already obtained Pareto set), and local surrogate-assisted refinement procedures. Our methodology allows for rendering uniformly-distributed Pareto designs at the cost of a few hundreds of antenna EM simulations, as demonstrated by means of three verification case studies. Benchmarking against state-of-the-art MO techniques is provided as well.