DoE.MIParray: An R Package for Algorithmic Creation of Orthogonal Arrays

Ulrike Grömping

doi:10.5334/jors.286

Abstract

The R package DoE.MIParray uses mixed integer optimization for creating well-balanced arrays for experimental designs. Its use requires availability of at least one of the commercial optimizers Gurobi or Mosek. Investing some effort into the creation of a suitable array is justified, because experimental runs are often very expensive, so that their information content should be maximized. DoE.MIParray is particularly useful for creating relatively small mixed level designs. Balance is optimized by applying the quality criterion “generalized minimum aberration” (GMA), which aims at minimizing confounding of low order effects in factorial models, without assuming a specific model. For relevant cases, DoE.MIParray exploits a lower bound on its objective function, which allows to drastically reduce the computational burden of mixed integer optimization. Funding statement: The work on this software is the outcome of research that was funded by Deutsche Forschungsgemeinschaft (grant GR 3843/2-1).

Highlights

The R package DoE.MIParray uses mixed integer optimization for creating well-balanced arrays for experimental designs
Balance is optimized by applying the quality criterion “generalized minimum aberration” (GMA), which aims at minimizing confounding of low order effects in factorial models, without assuming a specific model
(1) Overview Introduction The R package DoE.MIParray creates well-balanced experimental designs for experimentation with a set of experimental factors, for which suitable off-the-shelf plans are not readily available. It is useful for relatively small experimental situations for which the experimental factors have different numbers of levels

Summary

Introduction

The R package DoE.MIParray uses mixed integer optimization for creating well-balanced arrays for experimental designs. DoE.MIParray is useful for creating relatively small mixed level designs.

Results

Conclusion