This paper is about a new method of generating optimal split-plot design. An advantage of using split-plot design is that, no need to specify the candidate set in prior. This can be more useful when the candidate set is too large.
The word split-plot design comes from agricultural experiment. A factor that changes between separate plots of land is a whole-plot factor and in sub-plot factor levels varies within each plot. Though split-plot implementation is economic we prefer over randomized design in some cases.A usage of split-plot designs mainly focuses on sample size, whole plot size and priori model.
The algorithms of Goos and Vandebroek choose combinations at factor level and arrange in whole parts so that D-optimality criterion is maximum but it becomes a problem when number of experimental factors are large or when experimental region is highly constrained. Because a candidate set which covers the entire region requires a large number of factor level combinations.
There are two ways by which practitioner’s money can be saved. First, the algorithm provides an opportunity to reduce sample size when compared to the use of fractional factorial designs. Secondly, it provides efficient solutions for challenging practical problems than can be found using available methods.
The main aim of Split-plot experiments is non-specification of the candidate set in prior. The performance of candidate-set-free algorithm is implemented using proof-of-concept example and wood products experiment.