Looking at accessing new performance regions via formulation from a small number of synthesized materials, a mixture DOE was identified as the best approach. Upon initial discussion with the team, it was shown that DOE was deemed infeasible as traditional  “random” point selection (based on target component ranges) could lead to large numbers of “un-processable” runs, resulting in an output data set that could not provide a useful model.

The team knew the rules that needed to be applied to maximise the chances that all samples would be processable but were unsure how to implement these in a DOE. After trying a number of approaches to address this issue, the successful approach proved to be use of a candidate file to select runs; the runs to be included were generated using a pseudo-model, the factor grid tool and the grid point generator. The presentation reviews how this approach was carried out and the impact on the DOE of this approach.