Orthogonal Space Filling Design for Computer Simulated Experiments

Abstract

At present, computer simulated experiment (CSE) has been widely used in many engineering applications, in order to explore the complex phenomenon that is hard or unable to perform physical experiment. Hence, a relationship between input variables and output response in these complex systems can be explored and constructed as a metamodel. CSE, in nature, is deterministic and time-consuming when running for a single value of output response.  Therefore, an experimental design for CSE plays an essential role and is based on developing a statistical predictive model with a handful of runs.  Latin hyper cube design (LHD) is one of popular design class in this context. The design with spread filling property or orthogonal property has been proposed to use in CSE. This paper proposes a combined criterion namely  and correlation coefficient criterion called multi-objective criterion along with simulated annealing search algorithm to construct an optimal LHD design. The results show that the multi-objective criterion can produce optimal LHD designs with orthogonal and space filling properties. Moreover, the predictive model conducted from these designs has good performance in terms of prediction accuracy.