Machine Learning-Assisted Development of High-Performance Ethanol Synthesis Catalysts via CO2 Hydrogenation

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• High-performance catalyst discovery is vital but traditionally slow.
• Machine learning (ML) shows potential for accelerating catalyst development.
• Previous ML applications in catalysis have had limited success in discovering truly novel catalysts.

Purpose of the Study:

• To develop an ML approach for discovering novel, high-performance catalysts for CO2 hydrogenation to ethanol.
• To incorporate previously unstudied elements into the catalyst design pool.
• To demonstrate the efficacy of a closed-loop ML-driven discovery system.

Main Methods:

• Utilized a closed-loop system with 24 iterations of ML predictions and experimental validation.
• Tested a total of 555 catalysts, building a large experimental dataset.
• Employed advanced characterization techniques including in situ/operando X-ray absorption spectroscopy (XAS), ambient-pressure X-ray photoelectron spectroscopy (AP-XPS), and DRIFTS.

Main Results:

• Discovered over 50 catalysts with superior activity for ethanol synthesis.
• Identified a highly effective multielemental Pd-Au/K-Sr-Fe-Zn-Cd-Yb-Re/CeO2-ZrO2 catalyst.
• Achieved an ethanol space-time yield of 8.2 mmol gcat−1 h−1 with 57.6% CO2 conversion and 23.2% ethanol selectivity.

Conclusions:

• The ML approach successfully identified novel and highly efficient catalysts beyond traditional element pools.
• The study highlights the critical roles of individual elements in the optimized multielemental catalyst.
• This data-driven methodology significantly advances the discovery of catalysts for CO2 utilization.