(Dr. Farrokh Mistree, advisor)
"Design Space Exploration: Sequential Metamodeling and Analysis for Computer Experiments"
Experimentation and approximation are essential for efficiency and effectiveness in concurrent engineering analyses of large-scale complex systems in which designers have to deal with multi-disciplinary and multi-objective analysis using very complicated and expensive-to-run computer analysis codes. This process of experimentation and approximation is called metamodeling in which we need: (a) choosing an experimental design for generating data, (b) choosing a model to represent the data, and (c) fitting the model to the data. Sequential metamodeling and analysis is the development of series of metamodels with different sets of data by realizing these steps sequentially and repeatedly along the design timeline. It helps designers explore the design space to find satisfacing solutions in early design stages.
The primary objective of the proposed research is to develop a systematic yet flexible method in which various metamodeling techniques are utilized in building series of appropriate metamodels for robust design space exploration in accordance with the change in information quality along the design timeline in the early stages of design. Development of the method will be accomplished by (1) studying measures for metamodel validation with deterministic computer experiments, (2) developing methods for sequential experimental design in fixed design spaces, (3) developing methods for design space reduction by screening factors and reducing factor ranges, (4) developing methods for design space definition and shift, and (5) developing methods for model selection along the design timeline. Testing and verification of the approach will be accomplished through several one- or two-variable functions and two industry case studies.