|dc.description.abstract||With the increasing reliance upon advanced computational methods for engineering analyses, there is a need for the efficient exploration of experimental design spaces. Contemporary methods such as traditional Design of Experiments and space-filling techniques have long enabled intelligent resource allocation throughout the design space. However, such methods assume that the feasible space which they sample, defined by limits placed upon the design variables, can be generalized to a d-dimensional hypercube. Due to the presence of features such as embedded constraints, correlated
design variables and numerical failures, such an assumption can be suspect.
This dissertation acknowledges that features present within an experimental design space may yield feasible regions that are non-hypercubic. To address this, a
decision support methodology is presented to provide guidance for the exploration of general design spaces. From an initial sample, this methodology provides hypercubic
classification and informed strategies for further sampling. Additionally, a Set-Based Bounded Adaptive Sampling process, enabled by machine learning techniques, is provided for the identification and exploitation of non-hypercubic feasible spaces.
An application of the methodology is used to provide design space exploration for a Hybrid Wing-Body aircraft. Utilizing the Experimental Design Space as a modeling and simulation test-bed, the methodology allowed for a more complete understanding of the feasible design space of interest. Ultimately, such an approach likely enables improved regression generation, optimization, visualization and additional exploration
for this design problem as well as others of similar characteristics.||