Abstract:
|
Due to various resource restrictions, there has been increased needs to measure multiple responses in a single experiment. Central composite designs have been broadly used for estimating response surfaces from experiments with multiple responses, which typically require a large number of runs across the super space of all relevant design factors. However, in many of these experiments, each individual response is often affected by only a subset of design factors, and such information might be obtained from earlier screening experiments. We propose a more cost-efficient design selection strategy based on utilizing this prior knowledge and the Pareto front approach to select D-optimal designs with balanced performance on multiple responses. A Pareto aggregate coordinate exchange algorithm has been adapted to efficiently identify the Pareto front based on D-efficiencies measured for multiple responses. The method is illustrated with two examples and compared with existing methods on a variety of design characteristics.
|