| This dissertation develops and demonstrates a methodology for creating alternative solutions prior to the detailed design of a system. These alternative solutions should all meet expected performance targets. The methodology is domain-independent and can be applied at any stage of the design process, including conceptual design, with the use of a proper representation space. Optimization in the classical sense is not used; adaptive search is used. Adaptive search is a form of numerical optimization using heuristics. Adaptive search is sufficient to find our designs because we relax the strong assumptions that are necessary to make classical mathematics (i.e., gradient-based search) work. We satisfice our desires, finding designs that are good enough for our purpose.;Since we use multiple objectives to qualify our designs, some objectives may be incommensurable or even be in opposition to one another. We do not have a single optimum but a set of solutions because the objectives are generally not collinear.;The approach taken is unconventional; we design systems "backwards". Starting with solutions, we search for the designs that produced them. Doing so, we must realize that our solutions may not be unique; several designs may produce the same solution. Using our knowledge of a particular domain, we know the values that some design variables must have in order to obtain a feasible design. We use that knowledge to reduce our search as we explore our design space. By parameterizing one or more of the objective functions we focus our attention on the remaining objectives. For the purpose of this dissertation, the terms solution, design, and system can be used interchangeably since the approach can be applied at any stage of the design process.;The results show that the isoperformance approach to design can be extended to cover nonperformance objectives. The assumptions required for the isoperformance approach can be relaxed; the isoparametric approach does not need real variables, continuous functions, and convex spaces. |
