Repozytorium publikacji - Politechnika Gdańska

Multi-objective optimization is indispensable when possible trade-offs between various (and usually conflicting) design objectives are to be found. Identification of such design alternatives becomes very challenging when performance evaluation of the structure/system at hand is computationally expensive. Compact RF and microwave components are representative examples of such a situation: due to highly compressed layouts and considerable electromagnetic (EM) cross-couplings between the circuit building blocks, high-fidelity EM simulation is the only reliable way of their accurate analysis. Unfortunately, it is a time consuming process. Consequently, direct multi-objective optimization of EM simulation models is impractical. In this work, we investigate strategies for computationally feasible multi-objective optimization of such structures. The presented solution approaches rely on surrogate-based optimization (SBO) paradigm, where the design speedup is obtained by shifting the optimization burden into a cheap replacement model (the surrogate). Two SBO techniques are considered: (i) structure decomposition and surrogate construction through cascading of (global) approximation models of the individual building blocks of the compact structure of interest, and (ii) design space confinement and approximation model construction using coarse-discretization EM simulation data. In both cases, design refinement is executed using space mapping. The presented techniques are illustrated using examples of wideband matching transformers. Numerical results and qualitative comparison of the methods are also provided.

Autorzy

• prof. dr inż. Sławomir Kozieł link otwiera się w nowej karcie ,
• dr hab. inż. Adrian Bekasiewicz link otwiera się w nowej karcie