Optimal experimental design for parameter estimation of steady-state and dynamic models: from theory to experimental application

The chemical industry faces the challenge of meeting the growing demand for chemical products with a limited supply of conventional petroleum - the primary energy and feedstock source for the chemical industry. This requires the development of new processes based on unconventional feedstocks such as biomass. Since process development is effectively done through model-based simulation and optimization, it is essential that models describing reaction kinetics or phase separation are accurate. In addition, it is important to account for the uncertainty in the predictions of such models when designing equipment or optimizing operating strategies. The estimation of model parameters and their uncertainty is done by Parameter Estimation (PE), while the experimental design can be efficiently computed by solving the model-based Optimal Experimental Design (OED) optimization problem. Despite a significant amount of research in the areas of PE and OED, there are still challenges that need to be overcome. In this thesis, three challenges have been formulated and addressed, resulting in three main contributions. First, the applicability of the sequential OED method is validated theoretically and experimentally for the NRTL model used to describe the vapor-liquid equilibrium of binary mixtures. Second, the applicability of the PE is extended to dynamic models with discontinuities by using a new method to describe the phase disappearance, which is shown using the example of an esterification reaction where one of the separated liquid phases disappears. Third, the applicability of the OED methods to dynamic models by non-experts is improved by proposing a classification of nonlinear program (NLP) formulations and providing guidelines for the selection of NLP formulation and solution strategies.