CBI Colloquium - Prof. Alexei Lapkin - University of Cambridge

Jun 22
22.06.2017 16:15 Uhr bis 17:45 Uhr
Cauerstrasse 4, 91058 Erlangen; KS I, Room 01.421

Colloquium website: https://goo.gl/P2s0Hf

Speaker: Prof. Alexei Lapkin

Affiliation: Department of Chemical Engineering and Biotechnology, University of Cambridge

Invited by: Marco Haumann

Title: Continuous flow synthesis: from DFT models to process design


Continuous flow processes are seen as a future of manufacturing of speciality chemicals and pharmaceuticals. However, continuous processes could give worse performance that batch processes: this depends on the kinetic scheme, on physical properties of the system and on separation of the products and catalysts. Thus, a mixture of scale-up and process model questions must be answered to develop the most optimal process option. We are developing a generic approach to process design, combining first principles modeling with design of experiments (DoE) driven either by models or by statistics. The ultimate objective of our work is to be able to co-develop optimal reaction system configuration and its validated predictive model in the shortest possible time and smallest number of experiments. Thermodynamic and quantum chemistry calculations provide a significant amount of a priori data about a chemical system. These data can be used to develop initial, not very good, models. Existence of a model allows for model-based DoE to be used and for exploration of different hypotheses about the reaction mechanism, approaches to separation (work-up), recycling schemes, and so on. However, in some cases developing such predictive model will take simply too long. This usually is the case when reaction systems are complex: multi-phase, with very few states being observed. In this case statistical DoE and robotic experimentation offer an alternative approach to process development. A very rapid convergence to satisfactory process recipes was shown in several case studies of reactions being optimised by ‘black-box’ machine learning algorithms. The remaining challenge is to combine the experimental and computational efficiency of machine learning DoE with the desire to develop mechanistic predictive models, to ultimately allow model-based economic control and optimisation of most commercial processes.