Sequential approximation in nonhierarchic system decomposition and optimization: A multidisciplinary design tool

This thesis demonstrates the utility of approximation in the optimization of non-hierarchic systems. Two variations of a multidisciplinary design optimization algorithm for optimizing non-hierarchic systems are developed and implemented in this thesis. A third variation is developed for future implementation studies. The algorithm provides for the optimization of non-hierarchic systems by decomposition into reduced design subspaces. The procedure utilizes approximating functions as a basis for the coordination of concurrent subspace design optimizations.; The algorithm imposes the same objective function and constraints at each subspace. Non-local analyses are approximated at the subspaces using global sensitivities. Local analyses are performed using the design tools (analyses packages) available to the subspace design team. The algorithm optimizes the nonlinear subspace problems concurrently allowing for parallel processing. Following each sequence of concurrent subspace optimizations an approximation to the global problem (i.e., the system) is formed using design data accumulated during the subspace optimizations. The solution of the global approximation problem is used as the starting point for subsequent subspace optimizations in an iterative solution procedure. The algorithm is sequential with move limits being imposed at the subspaces and in the coordination procedure to maintain the accuracy of approximations.; The algorithm accommodates existing design procedures where design teams operate concurrently. The algorithm's use of subspace optimization provides for individual design teams working concurrently. Information obtained during the subspace optimizations is used to build a design data base representative of design experience. The coordination procedure of global approximation, built from the design data base, accommodates design tradeoffs and provides robust algorithm coordination in implementation studies.; The use of approximating functions is seen to significantly reduce the number of system analyses required to optimize multidisciplinary design optimization problems. The algorithm effectively optimizes each problem in a test set of non-hierarchic system problems developed as part of this thesis.