| The design of a complex product requires multiple analyses, many of which trade information. In some cases, some information is required before it is available, as a result an estimate must be made of that input for the analysis to proceed. Once the input becomes available, additional calculations must be made to ensure that the result of the analysis reflects the changed input. This iteration creates a considerable computational cost in the design process.; This iteration may be eliminated through the reordering of the analyses, or through the temporary or permanent removal of the couplings between the system analyses. This elimination of couplings is termed system reduction. Until relatively recently, the trade of accuracy and efficiency that enables system reduction was only carried out in the context of local sensitivities, how much one analysis output impacts another output. Recent developments expanded this capability to include a system level measure of introduced error into the objective function and constraints, allowing the development of a selection subproblem that trades accuracy and efficiency in the system reduction context.; The current state of the trade-off between accuracy and efficiency is far form complete. The initial subproblem formulation resulted in potentially destabilizing sets of couplings being selected, which could result in system analysis convergence errors. Additionally, the technique employed to select couplings for suspension is extremely limited in scope. The designer's intuition and experience had been eliminated from the selection process, replaced by a simple selection algorithm.; First, the current coupling suspension problem is augmented, incorporating additional constraints to improve performance in system analysis stability, improved convergence characteristics, and more accurate error growth modeling.; The second issue this dissertation addresses is the development of a methodology that allows a designer to interactively manage the system topology. This development will provide a previously unavailable capability—real time interaction with the system topology. Through the use of visual cues, the method condenses the large quantity of information that a designer must consider into an intuitive format. Through the use of a platform and geographic interface, the methodology developed allows a designer to alter the structure of the system analysis, accounting for both quantitative and experiential issues that may arise from suspending couplings between analyses. This trade-off activity represents an implementation of the Visual Design Steering paradigm, allowing a designer to improve the efficiency of the design process through the use of a visual tool.; The third advance that this dissertation provides is the investigation of the suitability of various surrogates for designers in the selection of couplings for suspension. This dissertation investigated the feasibility of Discrete Optimization, Simulated Annealing, and various implementations of the Genetic Algorithm technique of robust optimization. (Abstract shortened by UMI.)... |
