| Mixture-Process experiments are commonly found in many industrial settings. At times, some of the process variables are uncontrollable or difficult to control and are therefore considered to be noise variables. The purpose of this research was to first develop models for analyzing and for evaluating prediction variance properties of the mixture-process-noise setting, including models in the presence of correlation among the noise variables. Models for both the mean and slope were developed. Graphical tools, Three-Dimensional Variance Dispersion Graphs and Fraction of Designs-Space Plots, were developed to assess and compare the variance properties of designs generated by standard software. The final part of the research used Genetic Algorithms to create designs with superior prediction variance properties than designs generated by standard alphabetic optimality criteria, such as the D criterion. The Genetic Algorithm was able to successfully find designs with lower maximum predication variance values than the standard approaches. |
