Surrogate modeling techniques and heuristic optimization methods applied to design and identification problems

• Ano de defesa: 2008
• Autor(a) principal: Viana, Felipe Antonio Chegury
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Engenharia Mecânica; Engenharias; UFU
• Palavras-chave em Português: Métodos heurísticos de otimização, Meta-modelagem; Otimização de projeto; Problemas inversos; Modelagem por fidelidade variável; Otimização matemática; Vibração; Heuristic optimization; Surrogate modeling; Design optimization; Inverse problems; Variable-fidelity modeling; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/14667

Advances in computer throughput have helped to popularize numerical optimization as an engineering tool. However, they also favor an increase in complexity of the state-of-the-art simulation. As a result, the computational cost of complex high-fidelity engineering simulations often makes it difficult to rely exclusively on simulation for optimization. This doctoral research presents an effort in combining global optimization and surrogate modeling techniques as a way to rationally use the computer budget and increase the information level obtained during the optimization task. The above mentioned techniques were used in the solution of the continuousdiscrete problems of the optimal design of a vehicular structure and aircraft structural components; identification of aircraft longitudinal stability and control derivatives and non-linear landing gear model and the improvement of surrogate models through extra simulations. Besides, the solution of the combinatorial problem of the optimal Latin Hypercube has been implemented. At the end of the research, the main learning is that, as it also happens with classical optimization algorithms, the success in using heuristic methods is highly dependent on a number of factors, such as the level of fidelity of the simulations, level of previous knowledge of be problem, and, of course, computational resources. This way, the use of variable fidelity and surrogate models together with heuristic optimization methods is a successful approach, since heuristic algorithms do not require gradient information (i.e., resources are directly used for the search, and there is no propagation of the errors due to the computation of the gradients); and they have the trend to find the global or near global solution. In some cases, a cascade-type combination of heuristic and classical optimization methods may be a suitable strategy for taking advantage of the global and local search capabilities of the individual algorithms.