Standard designs (a.k.a. canned designs) always have the same points when built with the same parameters (although different random run orders). Most of them are created with certain properties in mind.
Standard factorial designs are built from combinations of the vertices. They are intended to detect factor effects and their interactions. When there are numeric factors, center points may be added to estimate the amount and direction of curvature. Regular two-level factorial and multilevel categoric designs come in both randomized and split-plot versions.
Standard response surface designs are built to fit full quadratic polynomial models which include: linear, two-factor interactions and quadratic terms. The definitive screening designs fit the full linear, but only subsets of the quadratic model. Central composite designs can also be set up as split-plot experiments.
Standard mixture designs are simplex designs which require that the difference between the maximum and minimum of each component is equal.
Mixture screening designs can be non-simplex. In these cases, the design structure might vary from build to build but are restricted to vertices and center points.
Although standard designs have great properties for analyzing the data gathered during the experiment, make sure they fit the problem before using them.
Use custom designs when…
some combinations of the factors aren’t feasible, requiring additional constraints;
all possible categoric combinations result in an unreasonable number of runs;
the design has mixture components that do not fit a simplex;
discrete settings for numeric factors are necessary for the design; and/or
the necessary model is third-order or higher.