Solving Very Large-Scale Structural Optimization Problems

摘要

Large-scale structural parametric optimization problems still impose a challenge to computer hardware. A three-step approach to reduce the run time and memory requirements is proposed. It is based on the observed sparsity of the matrix of partial derivatives in structural optimization. The approach includes a differentiation scheme to compute smaller gradient matrices in parallel and assemble them employing the chain rule, an adaptive filtering framework for an effective selection of active constraints based on numerical value and engineering knowledge and the pairing of a known sequential convex programming algorithm with a preconditioned conjugate gradient solver for the internal matrices. Software has been developed and successfully applied to the optimization of the outer wing panels of a large military transporter aircraft.