Efficient Design of Large Lightweight Composite Structures: A Parameter Free Optimization Approach

Abstract

Listen

Translate article

This contribution presents a computational framework and the algorithmic techniques for simulation and gradient-based optimization of g eometrically nonlinear and large-scale finite element models of composite structures. Focus is put on efficient adjoint sensitivity analysis to deal with very large design spaces. Par allelization is applied to all computation phases: simulation, sensitivity analysis, as well a s optimization. The modular framework also allows for mixing the design variable types in one optimization step. This enables combined shape and sizing optimization. Several application examples illustrate the methods, show the applicability to large problems, and prove the high parallel efficiency. I. Introduction imulations for the design and analysis of flexible mechanical structures are commonly based on the finite element method. In the case of thin and lightweight structures the aim of the computational design pro cess is often to minimize weight while maintaining or improving stress, stiffness and deformation criteria. Co mposite materials are a good alternative in this context: I n addition to the high strength-to-weight and stiff ness to weight ratios, composite materials also allow the engineer to design both material and structure by tailoring the directional stiffness and strength as required 1 . Numerical structural optimization can be used as ge neral tool in the structural design process 2 . However, the world of computational simulation has to face growi ng multidisciplinary challenges to satisfy the dema nds for realistic prediction in virtual design. Highly spec ialized methods from different disciplines, like co mputer-aided geometrical design, computational mechanics and nonlinear mathematical programming are combined to solve these optimization problems. This results in a crucial de mand for high performance and flexibility of the co mputational environment.