Greedy Algorithm for Black-Box Parameterized Modeling of Electromagnetic Structures
thesisposted on 27.11.2018 by Elisa Fevola
In order to distinguish essays and pre-prints from academic theses, we have a separate category. These are often much longer text based documents than a paper.
The use of parameterized macromodels has become more and more popular in a wide range of applications, from Electromagnetic Compatibility to microwave engineering, where they can be employed for the analysis, design and optimization of microwave structures. It is quite common, in fact, that for design optimization purposes or sensitivity analysis some geometric properties and physical quantities are left unde ned and become variables of the system. Usually data upon which the macromodel is built come from first-principle solvers. This implies that, in order to simulate all con gurations that are necessary to build the model, an extremely large amount of full-wave analyses must be performed. With the increase in the number of dimensions, the quantity of data samples increases exponentially, making the construction of the macromodel too expensive in terms of computational time and resources. One of the possible solutions to overcome this problem is the use of adaptive sampling algorithms targeting the identi cation of a quasi-minimal distribution of data samples sufficient to characterize the system, and thus to build a sufficiently accurate model. This thesis presents a class of sampling algorithms for a fully-automated generation of parameterized macromodels, based on a given electromagnetic solver. The presented framework not only implements a combination of greedy and adaptive algorithms for the optimal choice of points in the parameter space, but it also connects one of the most diffused commercial EM solvers (Keysight EMPro) to a MATLAB tool for the extraction of parametric macromodels in a fully automated way. During model creation, the choice of those data which will be used to fit the model, and those left for validation is made by a Vectorial Kernel Orthogonal Greedy Algorithm (VKOGA). The iterative nature of the algorithm allows to obtain a compact, robust model with a minimum number of points, and so of solver simulations, reducing drastically computational time. The tool developed for this thesis, moreover, provides a perfect integration between field solvers and the macromodeling tool. The algorithm has been trained and tested on a number of electromagnetic structures, and its effectiveness has been demonstrated in all the aforementioned cases.