| The size and complexity of aircraft design problems has limited the application of multidisciplinary design optimization (MDO) to the earliest stages of the design process where simple analyses are used. The emergence of new aircraft types, with a higher degree of interdisciplinary coupling and a lack of similar previous design on which to base design decisions, motivates research on the design optimization process itself. Accurate analyses must be integrated in a way that is easily manageable and computationally efficient, while maintaining the validity of the numerical optimization process. This dissertation describes an architecture that meets these requirements, discusses implementation issues, and demonstrates the approach on aircraft design problems.; The analysis is decomposed into independent subproblems that can be developed, verified, and maintained separately and executed in parallel. This arrangement is described by carefully considering the relationship between the optimizer and the analysis. The decomposed and original versions of the problem are shown to have the same optimal solutions. Efficiency, including the benefits from parallel execution of the subproblems, is discussed.; Two tools, developed to aid in implementation of the current approach, are described. A generic executive system allows decomposed problems to be run in parallel on a heterogeneous network of computers. A system management tool uses a genetic algorithm to determine the optimal decomposition of a problem into subproblems. The objective function used is an explicit estimate of computation time for the decomposed system.; Finally, aircraft sizing is combined with mission optimization for a transport aircraft. Lift, drag, and wing loads and deflections are calculated at several locations along the flight path, as well as at maneuver conditions used for wing structural sizing. Static aeroelasticity is taken into account at several conditions with the assistance of the optimizer. This final problem shows how analyses that would be difficult to integrate with conventional methods can be implemented easily using the new architecture. |
