On the accuracy and eficiency of cross-entropy method for structural optimization

• Ano de defesa: 2019
• Autor(a) principal: Issa, Marcos Vinicius dos Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade do Estado do Rio de Janeiro; Centro de Tecnologia e Ciências::Faculdade de Engenharia; BR; UERJ; Programa de Pós-Graduação em Engenharia Mecânica
• Palavras-chave em Português: Mechanical Engineering; Entropy; Genetic algorithms; Finite element method; Structural optimization; Nonlinear optimization; Cross entropy method; Optimization experiments; Engenharia mecânica; Entropia; Algoritmos genéticos; Método dos elementos finitos; Otimização estrutural; Otimização não linear; Entropia cruzada; Experimentos em otimização; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: http://www.bdtd.uerj.br/handle/1/11678

This dissertation has the objective to evaluate the Cross-entropy method (CE) in structural optimization. Trusses made of tubular structures are used as benchmark tests and it is sought to minimize its mass considering some criteria of structural integrity. The optimal values found by the CE are compared with other results obtained by Sequential quadratic programming (SQP) and Genetic algorithm (GA). Numerical experiments demonstrate that the CE offers a solution for structural optimization in terms of accuracy and eficiency. This dissertation has four structural models where the optimization methods are used considering constraints like yield stress, buckling, natural frequencies, and maximum displacement. For each model, a finite element analysis (FEA) is done to verify the structural integrity criteria which is then used for an optimization problem evaluating the constraints, considering the values found by the three optimization procedures (SQP, GA, CE). In some cases, the optimal values found by the CE are close to those found by the SQP, being SQP a first-order method and CE a zero-order method. The SQP, using the gradient (first-order) in its computational process, is more eficient (better results) and faster than the CE, considering convex problems. When comparing CE with another zero-order method, GA, it is noted that in most cases the CE is faster and has better results than the GA, making the CE quite interesting for application in structural optimization.