SUMMARY
The goal of robust design is to improve the quality of an engineering system by minimizing the effects of variation on the performance without eliminating the sources of uncertainty. This uncertainty usually stems from variations in noise factors (Type I - uncontrollable parameters) and variations in control factors (Type II - variables that can be controlled by the designer). Robust solutions can be found by formulating design problems as compromise Decision Support Problems (cDSP). This approach allows for multi-objective robust design decision-making, where the goals for performance (mean) and robustness (variance) can be formulated as separate objectives. In order to obtain satisficing solutions from the exploration or optimization algorithm, accurate estimations of the performance mean and variance are crucial. In this thesis, a current robust design method, the Robust Concept Exploration Method (RCEM), is analyzed and limitations in the goal formulations as well as the applicability of the response surface method are discovered. Addressing the first problem, new goal formulations are suggested that can be integrated to eliminate some limitations, found in the original RCEM, when handling nonlinear design problems. These formulations account for nonlinearity and stationary points within the design space and are applicable to design problems, for which the performance functions are known explicitly. In the second part of the thesis, the limitations of global response surfaces that were found in RCEM are addressed. It is argued that a local response approach would be preferable for nonlinear design problems. An approach is introduced that utilizes the Probabilistic Collocation Method (PCM) within a cDSP formulation for robust concept exploration. This method allows for accurate performance mean and variance estimations while requiring a minimum number of simulation evaluations.