抄録
This paper presents an Automatic External Superelement Optimization (AESO) procedure for efficiently and accurately solving large-scale optimization tasks. In the procedure, a full model is automatically (i.e., without any work on the user's part) partitioned into a residual structure that includes the designed part and an upper stream superelement containing the non-designed part. Then, the upper stream superelement part is reduced into boundary matrices with Component Mode Synthesis (CMS) techniques and these matrices are stored on the database while a new user input file is created from the designed part. The original optimization task is now performed in a separate assembly run that assembles the boundary matrices into the residual structure for solving the system solutions. This approach is very efficient since the residual model can be solved quickly when the size of the residual model is much smaller than the size of the analysis model (say 10% or less). To further improve the performance, Matrix Domain Decomposition techniques are applied to both component and residual solution procedures. Furthermore, the residual vector techniques based on applied loads are also applied for improved accuracy. Examples are presented to show the effectiveness and efficiency of the approach with speedups greater than five demonstrated.
