Comparative study of mixture designs for complex phenomena

Abstract

In the field of formulation, the behaviour of phenomena studied are often complex, including non-linear or chaotic zones which make classical strategies based on polynomial models inappropriate. In these situations, uniform experimental designs and kriging models are recommended. Nonetheless, when researchers have to set up an experimental strategy, they must choice the class of designs to be constructed, the number of experiments to be performed and the modeling kriging parameters to be used. To help with these choices, we propose a catalogue combining various uniform designs characterized by several intrinsic quality criteria and to establish decision-making rules linking these criteria to the quality of the information obtained from the design.