Flatness-Based Design of Experiments for Model Selection

High product standards and cost-efficient robust process designs are key issues in the pharmaceutical industry and require thorough system understanding of the underlying physical phenomena. Model-based approaches have been proven favorable to gain valuable system insights. However, qualitative inferences can be drawn only if the model is adequately derived and validated by - at best - optimally designed experiments. In this paper, we present a new model inversion strategy to solve model-based design of experiments (MBDoE) problems for model selection; i.e., we maximize the difference of competing model candidates by using the differential flatness concept. Differentially flat systems can represent their states and controls analytically by the so-called flat output and its derivatives. By maximizing the difference of the flat outputs of all considered model candidates, we efficiently obtain analytical model-revealing control actions for the MBDoE problem. When these optimized control actions are implemented, the resulting experimental data are expected to invalidate incorrect models more reliably. We demonstrate the effectiveness of the proposed MBDoE concept with potential model candidates that describe different realizations of a mass action reaction system.