Experimental design in a multicriteria optimization context: An adaptive scheme

The identification of a promising region in design space where strategies to obtain an optimal experimental design can be applied is crucial in practical applications. In this contribution, starting from a model adjusted to previously conducted experiments, a computationally efficient multicriteria optimization scheme is used to identify the Pareto boundary, where minimization of the prediction errors of the objective functions is included as additional objective. This guarantees that best compromises are found between the process-relevant objectives, like cost and quality criteria, while simultaneously quantifying the trade-off between those objectives and their prediction errors. In a real-time navigation procedure, this allows to narrow down the most promising region in design space, where then strategies of model-based experimental design are applied. The entire workflow is illustrated with an intuitive example which shows that an unacceptably high prediction error of Pareto points can be efficiently reduced by only a few additional experiments.