A New Approach to Multidisciplinary Design Optimization via Internal Decomposition
Andrew Lambe, Joaquim R. R. A. Martins · 13th AIAA/ISSMO Multidisciplinary Analysis Optimization Conference · 2010
Traditional approaches to MDO problem decomposition have shown poor performance when solving problems with strong interactions between disciplines. We present a new decomposition strategy for MDO that aims to overcome this diculty. The method, called Block Approximation with Krylov Renement, or BAKR, decomposes the solution of the linear system present at each iteration of an interior point algorithm. By decomposing the linear system inside the optimization algorithm, rather than the original design problem, we maintain the strong global and local convergence properties of the interior point algorithm while reducing the overall computational cost of the solution. Preliminary test results show reductions in both computational cost and the number of function evaluations, demonstrating strong potential for future application in large MDO problems.