The Use of Optimization Algorithms in Flow
Francois-Xavier Felpin, Professor, University de Nantes
Traditional optimization studies are mainly based on the know-how
of chemists and eventually use the principles of Design of Experiment
(DoE) methods. While DoE can be automated to speed-up the screening of
predetermined experimental conditions, it requires a high number of
experiments to locate an optimum. DoE has been also studied with flow
reactors for the efficient optimization of chemical reactions. A much
more appealing strategy consists in integrating a feedback with
optimization algorithms. This approach enables optimization systems to
analyze previous reactions data in order to adapt and modify the
following experiments in a short time frame. The challenge associated
with intelligent optimization systems results in the need to perform a
constrained optimization without a priori on the value of reaction
parameters, making the optimization of complex reactions a challenge
beyond classical optimization techniques. We will discuss
reaction optimizations in flow, using of a modified Simplex algorithm.
The beneficial properties of flow reactors associated to the power of
optimization algorithms for the fine-tuning of experimental parameters,
allowed reactions to proceed in conditions unable to promote the
coupling through traditional batch chemistry. Our modified simplex
algorithm showed great flexibility since the experimental conditions
could be tuned according to the nature of three different objective
functions which could be either the yield, the throughput or the
production cost.
|
|