Numerical models represent more and more complex physics. Such problems, often large dimension, require sophisticated and time-consuming computer codes. The considered approach consists in choosing a limited number of simulations organized according to an experimental design from which a metamodel is built to fit the simulator. Exploratory designs are only considered (when we do not know the true relation between the response and inputs) generated with deterministic codes. Therefore, designs should allow one to fit a variety of models and should provide information about all portions of the experimental region. If we expect the output to depend on only a few of the inputs, then it is preferable that points are evenly spread across the projection onto these factors. Two criteria quantifying the intrinsic quality of designs were developed. The key point of this work is based on the simulation of Markov chains Monte Carlo method (Strauss and Gibbs process) to build experimental designs.
