Model-based Design of Experiments for the Identification of Kinetic Models of Amide Formation

Model-based design of experiments (MBDoE) techniques have been applied to various process systems in the scientific community to optimally determine a minimum number of informative experiments to enable identification of a kinetic model structure with precisely determined parameters. The effectiveness of MBDoE techniques is however deeply affected by parametric uncertainty. To evaluate the effect of parametric uncertainty on model effectiveness, this work compares and evaluates two different MBDoE approaches: 1) LHS-MBDoE, where the Latin-hypercube sampling (LHS) precedes MBDoE application; and 2) robust MBDoE, where MBDoE techniques apply ab-initio via either the expected value or worst-case approach. Using experimental and in-silico data, MBDoE methodologies were tested on a pharmaceutically relevant reaction system involving homogeneous amide formation, which can be described using reversible chemical kinetics. The performances of the two MBDoE approaches were assessed using i) the χ² lack-of-fit test; ii) the Student's t-test; and iii) the determinant of Fisher information matrix (FIM) as posterior scalar measure of information.