Date of Award

Fall 1997

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Arkadan, Abdul-Rahman A.

Second Advisor

Hock, Jeffrey L.

Third Advisor

Josse, Fabien J.

Abstract

In electromagnetics, frequently we deal with problems that require identification or determination of a system given some measure of its performance. In such a problem, it is required to predict the system that yields the known or measurable performance. Such problems represent the class of problems known as inverse problems. In contrast, there are direct problems in which a fully known system is analyzed to calculate its performance. A direct problem is well-defined so that a unique and a stable solution always exists. Inverse problems, on the other hand, are typically ill-defined implying that less is known about the problem as compared to direct problem. Hence, the inverse problem methodology for the solution of system identification problem is typically more complex and involves effectively addressing various issues. In recent years, research in solving system identification problems has given birth to several new inverse problem methodologies, each possessing varying degrees of applicability and effectiveness. However, traditional deterministic methods fail to accurately and efficiently solve such inverse problems when the measure of system performance, known as the object function, demonstrates a complex and, in many cases, a discontinuous variation with respect to the parameters of the unknown system. Hence, it is required to develop a new methodology to overcome any inherent difficulties and solve the inverse problem of system identification. The highlight of the work presented in this thesis is the introduction of a methodology to solve inverse problems that presented insurmountable difficulties to the methods applied earlier. More specifically, this work extends the applicability of a relatively new breed of algorithms, known as Genetic Algorithms, to authoritatively solve system identification problems that could not be solved effectively using traditional methods. Genetic Algorithms are employed along with Finite Element Analysis to accurately identify unknown or inaccessible electromagnetic systems. The intent is to combine the powerful search methodology of Genetic Algorithms with the power of analysis of the finite element method for system identification problems in electromagnetics.

Share

COinS

Restricted Access Item

Having trouble?