Optimization of batch reactions using data-driven & knowledge-driven models: The case of asymmetric catalytic hydrogenation

The goal of this research work is to develop a systematic methodology that enables the rapid calculation of the optimal operation conditions for batch reactions. The primary objective is to optimize the process through Data-Driven models. However, their efficiency is compared with this of the Knowledge-Driven approaches.;The new methodology of Design of Dynamic Experiments (DoDE) (Georgakis, 2008, 2009) is the Data-Driven method applied to optimize the important pharmaceutical reaction of Asymmetric Catalytic Hydrogenation, a system provided by Sepracor, Inc.. The DoDE approach enables the discovery of optimal time-variant operating conditions that are better than the optimal time-invariant conditions discovered by the classical Design of Experiments (DoE) approach. Only composition measurements at the end of each batch are used to develop Response Surface Models (RSMs).;The Knowledge-Driven approach entails the development of a systematic method for the identification of the appropriate form of kinetic models. This approach is the Generalized Tendency Modeling Optimization and Control (GTeMoC) approach (Makrydaki & Georgakis, 2007), which is a generalization of the Tendency Modeling Optimization and Control (TeMoC) approach (Fotopoulos, Georgakis, & Stenger, 1994) where an approximate kinetic model of the reaction is developed in a systematic fashion. A novel pseudo-linearization and Stepwise Regression (SWR) are used to identify plausible kinetic structures. Once the preferred kinetic form(s) are identified, Non-Linear parameter estimation of the kinetic form(s) completes the modeling task of GTeMOC. If more than one model is identified, the statistically most accurate is selected.;The GTeMOC method is first tested in two simulated cases, the hydrogenation of D-glucose (Crezee et al., 2003), and the epoxidation of Oleic Acid (Rastogi, 1991; Rastogi et al., 1992; Rastogi et al., 1990). For the experiments of the industrial example, online Raman spectra and a PLS model, which relate spectra to the reaction mixture compositions, are used to develop the generalized tendency kinetic model using both DoE and DoDE data.;Finally, the two methodologies (data-driven and knowledge-driven) combined with the DoE or DoDE data are used to optimize the operational conditions of asymmetric hydrogenation reaction. The optimization maximizes the reactant conversion with a minimum diastereoselectivity constraint. Using data-driven models, the DoDE approach has a definitive advantage over the DoE approach. The knowledge-driven GTeMOC approach achieves similar process optimization to the data-driven one but requires a larger effort and time for the model development and the process optimization.