Topics on the design and analysis of computer experiments
Peter Z. G. Qian, Youngdeok Hwang · 2012
This dissertation addresses several issues on the design and analysis of computer experiments. First, we propose a new type of design, called a sliced orthogonal array based Latin hypercube design, intended for running multiple computer experiments. The proposed designs achieve both one- and two- dimensional stratification while each slice possesses univariate uniformity. Sampling properties of the proposed designs are derived. Second, we develop two procedures for randomizing a new class of nested space-filling designs. Third, we propose a statistical approach to building an accurate metamodel by exploiting the quality of high-accuracy simulation data and the abundance of low-accuracy simulation data of a mechanical dynamics system. It makes use of Gaussian processes and natural cubic splines. The effectiveness of the proposed methodology is illustrated with an example for studying the dynamics of a slider-crank system.