Estudo de otimização da reconstrução de imagens via tomografia capacitiva elétrica através dos métodos de elementos finitos e enxame de partículas utilizando computação de alto desempenho

• Ano de defesa: 2017
• Autor(a) principal: Souza, Bruno César Bezerra Nóbrega de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Mecânica; Programa de Pós-Graduação em Engenharia Mecânica; UFPB
• Palavras-chave em Português: Tomografia capacitiva elétrica; Problema inverso; Sistemas multifásicos; Método de elementos finitos; Otimização por enxame de partículas; Electrical capacitance tomography; Inverse problem; Multiphase systems; Finite elements methods; Particle swarm optimization; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/12915

This work aims to present the particle swarm heuristic as a viable method of solution for the inverse problem of the technique of Electrical Capacitance Tomography (ECT), with the purpose of applying to the imaging of multiphase flows. It will also be presented the philosophy of parallel programming that aligns with changes that favor the expansion of collective intelligence of swarm, thus reduces computational processing time, critical parameter in the design of the system of reconstruction of images, necessary for the production of images with higher spatial resolution. An alternative to increase the efficiency of the image reconstruction system is the removal of redundant calculations in the code (appropriation of the sparsity and ordering of the matrices associated with the associated mathematical problem) and the migration of the reconstruction algorithm from a sequenced programming to a parallel programming. It will be presented the physical principles of the ECT, the studies of use of the numerical characteristics of the finite element method (FEM), the heuristic algorithms used in the reconstruction process and the concepts of the main paradigms of parallelization (shared memory, distributed memory) and their software and hardware implementation. The results are encouraging because of the agreement between the images obtained and the simulated tomograms and the significant reduction in their processing time, thus stimulating the continuity of research in this topic for the future. Is found in this thesis all the steps to construct a Beowulf cluster and the serial and parallel codes necessary for implementation the algorithms.