Graph Partitioning-Based Coordination Methods for Large-Scale Multidisciplinary Design Optimization Problems

Abstract

This paper presents an algorithmic framework that expedites the overall convergence of multidisciplinary design optimization problems through a three-step process: partitioning, reordering and coordination. Calculations of any complexity, as well as coupling information are represented by digraphs with weighted nodes and edges. A partition algorithm divides the MDO problems into multilevel nested clusters of calculations with minimal loss of information while maintaining balanced loads among assigned CPU nodes. The feedback loops within the clusters are further minimized by a reordering algorithm. A new multilevel hybrid multidisciplinary design feasible and individual design feasible coordination method is presented that allows a dynamic MDO architecture based on the computational cost and coupling strength. In addition, this paper describes a real-time visualization for MDO problems with weighted design structure matrix and graph representation. The results of a scalable multidisciplinary design analysis and optimization problem are presented.