Closed-loop optimization

Marçal Capdevila-Cortada

DOI: https://doi.org/10.1038/s41929-024-01124-5

IF: 37.8

2024-02-28

Nature Catalysis

Abstract:Now, Paco Laveille and colleagues combine Bayesian optimization, automated synthesis and high-throughput catalytic experiments in a closed-loop approach to design cost-effective heterogeneous catalysts with high activity and selectivity for the hydrogenation of CO 2 to methanol. In their study they considered six metals, namely cerium, cobalt, copper, iron, indium and zinc, potassium as a promoter, and four supports, Al 2 O 3 , SiO 2 , TiO 2 and ZrO 2 . Each catalyst contains up to three metals with a total metal content between 2.5 and 5 wt% and can contain or not potassium promoter with a maximum loading of 1 wt%. The catalysts are prepared by a robot via incipient wetness impregnation and calcined, and catalytic tests are performed at 275 °C, 50 bar, H 2 /CO 2 ratio of 3 and a space velocity of 10,000 mL g –1 h –1 . The outcomes to optimize after each iteration are maximizing CO 2 conversion and methanol selectivity while minimizing methane selectivity and cost of supported metal. A total of 24 catalysts are prepared in each iteration and 5 iterations are performed. The best-performing catalyst in the first generation contains 1.22 wt% Zn, 1.16 wt% Cu, 0.5 wt% In on ZrO 2 and achieves 6.2% CO 2 conversion and 45% methanol selectivity. From the second generation onwards, the algorithm turns into mostly Cu-based catalysts supported onto ZrO 2 . The optimal catalyst is obtained in the fourth generation, consisting of 1.85 wt% Cu, 0.69 wt% Zn and 0.05 wt% Ce on ZrO 2 , which achieves 10.2% CO 2 conversion and 41.8% methanol selectivity, with a much-reduced cost than those derived from the previous iterations. Interestingly, if the cost minimization is removed as an objective after the first generation, in the second generation the catalysts become mostly In-based and the methanol selectivity increases substantially.

chemistry, physical