Otimização global eficiente aplicada a projetos de laminados de rigidez variável.

Abstract

The efficient global optimization (ego), presented by Donald R. Jones at the end of the 90s, propose to solve optimization problems where, due to the high computational burden of the cost function, the number of evaluations is severely limited. ego is based on the concept of surrogate models, that are fast-to-compute functions that adjust the “experimental” data (physical, analytic or simulations). In the current work, as well as in the Jones’s one, the Kriging metamodel is used. The efficiency and robustness of the method lays at the balance between exploration and exploitation of the surrogate response. Here, ego is used to find the optimized parameters for curved fiber structures with relatively complex geometry. The number of variables are moderately (in general 12) for the various problems presented and the geometries vary from simple square plates with a central hole to an airplane fuselage-like section. Trough these problems, single objective optimizations with and without constraints and an unconstrained multiobjective optimization are performed. For the constraints and multiobjectives, adaptations to the original framework were made. For most of the problems, the computational budget is set to just 60 evaluations of the high fidelity model (finite element model built using ANSYS software). The results showed that the framework is capable of finding good designs with a limited computational budget. Considering the variable stiffness composites, the use of curved fibers presented substantial structural improvements. For example, for the fuselage-like structure, it was observed an improvement of 47% in the buckling load when compared to straight fibers optimal solution.