Mixed-level column augmented uniform designs (Q2001206)

To explore a nonlinear relationship between factors (inputs) and responses (outputs) a second-order model is often used. Initial designs are good for a rough overview, but the number of runs in the initial design is often smaller than the number of parameters in the second order model. Therefore a follow-up design is needed to collect further information. Initial design together with a follow-up design is called an augmented design. In this paper the authors introduce and investigate several kinds of augmented designs: with additional runs, with additional two- or three-level factors, with an additional blocking factor. When the true relationship between factors and responses is unknown, uniform designs should be preferred. As uniformity criterion the wrap-around \(L_2\)-discrepancy (WD) is used. The authors discuss the best mixed-level column augmented design under the WD-criterion. They present theorems for lower bounds of the WD-criterion. It is further shown that column augmented uniform designs are also optimal under the non-orthogonality criterion \(E(f_{NOD})\). A construction algorithm is presented and some examples show its effectiveness.