Multi-objective optimisation of split-plot designs

This is the authors' abstract. We don't add key points for this paper.

Modern experiments allow scientists to tackle scientific problems of increasing complexity. Often experiments are characterised by factors that have levels which are harder to set than others. A possible solution is the use of a split-plot design. Many solutions are available in the literature to find optimal designs that focus solely on optimising a single criterion. Multi-criteria approaches have been developed to overcome the limitations of the one-objective optimisation, however they mainly focus on estimating the precision of the fixed factor effects, ignoring the variance component estimation. The Multi-Stratum Two-Phase Local Search (MS-TPLS) algorithm for multi-objective optimisation of designs of experiments is extended, in order to ensure pure-error estimation of the variance components. The proposed solution is applied to two motivating problems and the final optimal Pareto front and related designs are compared with other designs from the relevant literature. Experimental results show that the designs from the obtained Pareto front represent good candidate solutions based on the different objectives.