Computational aspects in modelling electromagnetic field parameters in microstrips.

Abstract

This thesis introduces the results of a detailed investigation and analysis into the key aspects of the Vector Finite Element Method, the critical relationships between the VFEM zone discretization and the associated solution matrix equations, the discovery of the original fill-in laws and the fill-in prediction methods, and their impact on the computational aspects and processes. Furthermore, the work included the design and implementation of accelerated matrix solution models, and their successful implementation and application to various electromagnetic problems and the presentation of the excellent performance results obtained. Also, one of the problem examples used in the investigation contributed to the validation of a method to reduce the characteristic impedance of a microstrip by means of grooves. Finally, the modelling of the electromagnetic problems by Artificial Neural Networks, and the successful investigation of an original concept---training by decimation---led to the validation of Artificial Neural Networks as a real-time modelling tool which completed the work of this thesis.