Continuous heterogeneous synthesis of hexafluoroacetone and its machine learning-assisted optimization

Conventional batch synthesis of hexafluoroacetone (HFA), an important pharmaceutical intermediate, suffers from complex catalyst preparation, harsh reaction conditions (up to 200 °C), and low selectivity. In this study, we developed a continuous flow system that employs a micro packed-bed reactor (MPBR) filled with Lewis acid catalysts. After an initial screening of reaction conditions and catalysts in the batch reactor, a Bayesian Optimization model and the multi-objective optimization algorithm qNEHVI were used to find a compromise between conversion and energy efficiency for the reaction in the continuous flow system. After 14 rounds of experiments, BO found the best results with conversion of 98.6%, selectivity of 99.9%, and an energy cost of 0.121 kWh per kg of product at 25.1 °C, atmospheric pressure, and a GHSV of 931.5 h^− 1 reaction conditions. The study demonstrates that BO can be used as an efficient tool for multi-objective optimization of heterogeneous catalysis in continuous flow.

Graphical abstract