Simulation of electrically large structures in EMC studies : application to automotive EMC

The aim of this thesis is the study of the simulation of electrically large structures and the application of the results to automotive Electromagnetic Compatibility (EMC). The theoretical and experimental work carried out has led to the development of computational tools and to the further understanding of the mechanisms involved in the representation of solid surfaces by means of wire-grid simplifications. The work was done in the context of a European project GUIDELINES FOR ELECTROMAGNETIC COMPATIBILITY MODELLING FOR AUTOMOTIVE REQUIREMENTS (GEMCAR). The first two chapters of the thesis contain a description of the GEMCAR project, a brief overview of some of the existing numerical methods for electromagnetic simulations (particularly, the ones used in GEMCAR), and the explanation of efficient, general simulation strategies that can be applied to different methods. The concept of adaptive sampling and its application are also introduced there. The main original contributions of this thesis are presented in Chapters 3 through 6. They consist of theoretical and experimental work as follows. We present, in Chapter 3, a modified version of the Numerical Electromagnetics Code (NEC). This version, which we have called Parallel NEC, has been adapted to run on parallel supercomputers, taking advantage of the combined processing power and memory of several processors working as a team. Parallel NEC has been implemented in two different supercomputing architectures to test the portability of the code. The original NEC routines in charge of the calculation and filling of the interaction matrix have been modified to work in a parallel environment. The matrix is now distributed among the available processors and the elements of the matrix are locally and individually calculated by their "owners". Thus the number of integrals carried out to build the complete matrix equation gets shared, diminishing the necessary runtime for this time-consuming operation. The system of equations is also solved using a parallel version of the Gauss-Doolitle algorithm. However, the most important feature of Parallel NEC is the possibility to use the distributed memory of the processors. This allows the calculation of problems of a size never achieved before using this numerical method without the need of using disk-space as swap memory. The code has been tested with models containing over 20.000 segments, exhibiting execution times comparable to those obtained with a single-processor PC calculating models of one tenth of that size in terms of the number of segments. Parallel NEC is also able to adapt itself automatically to its environment. It will detect the number of available processors and will take advantage of all available memory and calculating resources. The validation of Parallel NEC has been carried out in two steps. First, it was validated using simulation results obtained with other numerical methods. Then, it was validated by using experimental data from the GEMC