Much of today’s engineering analysis work consists of running complex computer codes (simulation programs), in which a vector of responses are obtained when values of design variables are supplied. To save time and effort in simulation, sampling (design of experiments) techniques are applied to help develop metamodels (empirical models or surrogate models) that can be used to replace the expensive simulations in future design stages. The usage of metamodels also helps designers to integrate inter-disciplinary codes and grasp the relationship between inputs and outputs. In this paper, we focus on a very important topic in studies of sampling and metamodeling techniques, i.e., the sequential design of experiments and metamodeling; the research question is: How to design sequential computer experiments to get accurate metamodels? After discussion of design and metamodeling strategies, a Sequential Exploratory Experimental Design (SEED) method is developed to help identify data points at different stages in metamodeling. Given limited resources, it is expected that more accurate metamodels can be developed with SEED. A single-variable example is used to help illustrate the SEED method.

This content is only available via PDF.
You do not currently have access to this content.