Substructuring for structural optimization in a parallel processing environment

Design optimization of large structures can be attempted through a substruc ture strategy. In this strategy the structure is divided into smaller subst ructures that are clustered to obtain a sequence of subproblems. Solution t o the large problem is obtained iteratively through repeated solutions to t he modest subproblems. Substructure strategies, in sequential and parallel computational environments on a Cray-YMP computer have been implemented in a design test bed CometBoards. The issues encountered during substructure s olution and their resolution are discussed under (I) coupling and constrain t formulation, (2) differences in optimal solutions, and (3) amount of comp utation. Coupling between subproblems and separating constraints into local and global sets promote convergence of the iterative process. The substruc ture strategy can converge to different local optimal designs with equal mi nimum weight. Substructure optimization can be computation-intensive. Howev er in a parallel computational mode, it can effectively use assigned proces sors.