Hi, in the case of experimental parameters I often use screening parameters to aproximately estimate trend of their effect (althoug not often must be linear), at the same time I try to obtain maximum informations from colleagues which have experiences with specific machine. So it is crutial to study maximum describe (known) variability of machine and materials. So then you can choose optimisation design, where more than 2 levels of variable are used= higher order polynomials for nonlinear effects. Usually center experiments are used to describe random variability and lack of fit. I suppose three replicated measurement is minimum. If the event, that your model describe varibility weel, the residuals have Gaussian distribution and predicted and observed effect values are linear. 

Thanks Petr.

Hi

Check this for little help it might;

International Journal of Pharmaceutical Sciences and Research .04/2014; 5(4):1208-1227.

http://www.researchgate.net/publication/272503301_A_STUDY_ON_FEASIBILITY_OF_OPTIMIZATION_TECHNIQUE_IN_FORMULATION_OF_DISPERSIBLE_TABLETS_BY_FACTORIAL_DESIGN

Article A STUDY ON FEASIBILITY OF OPTIMIZATION TECHNIQUE IN FORMULAT...

Very interesting Ugandar. Thanks!