Estudo avaliativo de algoritmos genéticos aplicados a problemas de identificação em Elastodinâmica

• Ano de defesa: 1999
• Autor(a) principal: Silva, Luciano Afonso da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Mecânica
• Palavras-chave em Português: Algoritmos genéticos; Ajuste de modelos; Elementos de suporte; Acoplamento de sub-estruturas; Problemas inversos; Genetic algorithms; Parametric identification; Inverse problems; Model updating; Sub- structure coupling; Supporting elements; CNPQ::ENGENHARIAS
• Link de acesso: https://repositorio.ufu.br/handle/123456789/28906; http://doi.org/10.14393/ufu.di.1999.13

The objective of this work is to perform an assessment study of a optimization methodology, know as genetic algorithms, when applied to some parameter Identification problem in elastodynamics. These algorithms are based on natural selection principies, created by Darwin, and unlike the classical algorithms, which seek the solution of the problem starting from a single point of the search space, the genetic algorithms operate simultaneously with a great number of points. Thus, the chances that the global minimum of the objective function is reached are increased. The parameter Identification problems are formulated as optimization problems where the objective functions represent the differences between the experimentally observed dynamic behavior and the previsions of the analytical models. The basis of genetic algorithms and also a brief comparative study to classical methods of optimization are presented, where robustness to noise and the presence of local minima is evaluated. The main focus is given to the following inverse problems: finite element model adjustment, evaluation of structural damage, and Identification of physical parameters - mass, stiffness and damping - of linear and non-linear support elements. Several numerical applications are presented. The first deals with the problem of model adjustment, and the objective is the location and extension of modeling errors. These errors are treated as stiffness variations, and their assessment is made by comparing the eigenvalues and eigenvectors of the structure with and without modeling errors. This methodology is applied to numerically simulated Systems and also to a simple structure, tested in laboratory. For the Identification of support parameters, two methodologies are used: in the first one, linear and non-linear support parameters are identified from the time-domain responses. In the second one, the parameters are identified considering a sub-structure coupling technique using frequency response functions. Applications to numerically simulated structures are performed. Based on numerical simulation examples, the genetic algorithms are appraised in terms of accuracy of the obtained Solutions and robustness to random noise present in the used data.