| Aerospace design can be viewed as an optimization process. A multi-dimensional search is conducted, with the physical description modified to produce the best functional performance. Despite this character and the wide availability of optimization software, conceptual design is rarely performed using formal search algorithms. The failure to exploit the benefits of optimization is examined in this thesis, and three key issues that restrict the success of automatic search are found. New approaches are introduced to address the integration of analyses and optimizers, to avoid the need for accurate gradient information and a smooth search space (required for calculus-based optimization), and to remove the restrictions imposed by fixed complexity problem formulations. Conceptual design of aircraft is a difficult endeavour. The disciplines are tightly coupled and performance assessment must be precise, so a design system is a complex combination of large analysis modules. Optimization should be performed in a flexible, extensible environment that conveniently links these modules and automatically co-ordinates their execution. A quasi-procedural architecture provides such an environment, and its efficient control of a large-scale design task is demonstrated. Calculus-based optimizers use gradient information to search for performance improvement. For many design tasks, gradients cannot be accurately computed, because variables have discrete values, analyses are noisy, or the search space has discontinuous topology. A genetic algorithm that provides a search method for such domains is studied. Applications demonstrate the utility of genetic optimization. The relationship between the genetic algorithm and calculus-based methods is explored. Numerical optimization algorithms use a complete a priori prescription of the search space, which, to a large degree, defines a design concept. The value of each variable is optimized, but the search scheme cannot alter the fixed set of parameters. A variable-complexity genetic algorithm is created to permit more flexible parameterization, so that the level of description can change during optimization. Promising general characteristics are identified in simple designs, and refined by increasing description detail. Structural and aerodynamic applications prove that this optimizer can discover novel designs. |
