Abstract
Simulations completely replaced physical tests in some product segments. Despite its acceptance as a design tool, optimal design seems yet to gain mainstream popularity. Statistical models called metamodels empirically that capture the input-output relationship of the analyses for evaluating objective functions and constraints are sometimes used for engineering optimal design. A variety of metamodels are proposed and put into practical use. It is important to select the accurate metamodeing techinque for cutting down the cost ant shoretning the development period. In the paper, we systematically compare popular metamodeling techniques to help constructing metamodels promptly and correctly. 8 metamodeling techniques and 5 sampling techniques are selected and evaluated by means of some example functions. Accordingly to these comparisons, it is demonstrated that some metamodeling and sampling techniques are superior to another techniques and that sampling techniques are very important if the sample size is small relative to the number of the design variables.
