COMPARISON OF METHODS FOR CONSTRAINED DESIGN OF COMPUTER EXPERIMENTS

Abstract eng:
A design of experiments creates an essential part in any experimentation, development of metamodels, sensitivity analysis or probability calculations. The creation of the design of experiments is determined by the shape of a design domain. In case of dependent input variables the design domain is constrained. Therefore classical designs developed for regular hypercubes cannot be applied here. This contribution presents comparison of several methods for designs in constrained design domains in terms of space-filling properties of the resulting designs and time required for their generation. Examples are focused on a special type of experiments called a mixture experiment, where individual parameters (ingredients) form a unit volume or unit weight. This condition leads to a simplex or a convex polytope shape of the design domain.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket