Data-driven catalyst design and

optimization: Integrating machine

learning and advanced simulations.
1
Vince James Mbanze, Ofentse Isaac Pholosi, Thabo Pilusa, Khanyisile Masemola
2 3 4

1,2
Sefako Makgatho Health Sciences University 3,4
University of the Witwatersrand:

Catalyst Design
BACKGROUND •Machine-learned potentials, based on
quantum data(ab initio),speed up
•Rapid shift towards clean energy due to climate catalyst structure searches.
•Once stable structures are found,
change and depleting non-renewables. mechanistic analysis and microkinetic
•Hydrogen fuel cells are a key energy alternative, Ab initio Data simulations guide catalyst design.
impacting various sectors. •Catalyst predictions are then verified
•Project emphasis on PGM-free catalysts using through synthesis, characterization,
materials like carbon substrates and metal-based and testing.
Machined-Learned Potentials Catalyst Synthesis •Data from the synthesized catalyst is
compounds. used in quantum calculations,
•Challenges in optimizing these materials for real- completing the workflow cycle.
world use. Expected Outcomes:
•Data-driven approach with machine learning and •Enhanced catalyst performance in
simulations employed. hydrogen fuel cells, leading to more
•Goal: Understand material properties, find efficient and sustainable clean energy
Catalyst Structure Catalyst Prediction solutions, with significant contributions to
optimal catalyst designs, and speed up new
scientific research and innovation.
catalyst discovery. Ethical Considerations:
•Adherence to strict ethical standards in
data handling, respecting privacy,
intellectual property, and environmental
AIM sustainability throughout the research
Reaction Mechanism Microkinetic Simulation process.

•Advancing PGM-Free Catalyst Development: Methods

• Objective: Develop sustainable PGM-free catalysts for hydrogen fuel cells. •Literature Review and Material Selection:
• Rationale: Address PGM scarcity and high cost with alternative materials • Conduct an extensive literature review on promising
for equal or better performance. catalysts.
•Integrating Machine Learning and Computational Simulations: •Data Collection and Database Creation:
• Objective: Use machine learning and simulations to optimize catalyst • Gather data from studies and databases and standardize the
design. information.
• Rationale: Improve understanding of catalytic mechanisms for faster •Machine Learning Model Development:
catalyst discovery and development. • Train models with collected data and validate and optimize
•Understanding Catalytic Mechanisms and Material Properties: them.
• Objective: Investigate material properties, reaction mechanisms, and •Computational Simulations and Ab Initio Calculations:
structural influences on catalysis. • Simulate catalytic reactions and integrate ab initio
• Rationale: Understand material behaviour under varying conditions and calculations.
optimize structures for performance. •Experimental Validation:
•Promoting Sustainable and Cost-Effective Solutions: • Synthesize and test identified catalyst materials.
• Objective: Support transition to sustainable energy with cost-effective, eco- •Iterative Optimization and Feedback Loop:
friendly catalysts. • Refine models with experimental results and adjust material
• Rationale: Decrease reliance on PGMs and boost hydrogen fuel cells as a structures and compositions.
clean energy source. •Documentation and Dissemination:
• Draft reports, papers, and presentations and disseminate
results.
REFERENCES
• Goldsmith, Bryan R.; Esterhuizen, Jacques; Liu, Jin‐xun ; Bartel, Christopher J.; Sutton, Christopher (2018). "Machine learning
for heterogeneous catalyst design and discovery." AIChE Journal 64(7): 2311-2323.

• Kim, C., Chandrasekaran, A., Hwang, J., Park, Y., & Choi, W. (2020). Machine learning-assisted discovery of efficient and stable
single-atom catalysts for electrochemical hydrogen peroxide production. Nature Catalysis, 3(3), 282-290.

• Corma, A.; Serra, J. M.; Serna, P.; Moliner, M., Integrating high-throughput characterization into combinatorial heterogeneous
catalysis: unsupervised construction of quantitative structure/property relationship models. J. Catal. 2005, 232 (2), 335.

• Nørskov, J. K., Bligaard, T., Rossmeisl, J., Christensen, C. H., & Dahl, S. (2011). Towards the computational design of solid
catalysts. Nature Chemistry, 1(1), 37-46.

• Jakab-Nácsa, A.; Sikora, E.; Prekob, Á.; Vanyorek, L.; Szőri, M.; Boros, R.Z.; Nehéz, K.; Szabó, M.; Farkas, L.; Viskolcz, B. (2022).
Comparison of Catalysts with MIRA21 Model in Heterogeneous Catalytic Hydrogenation of Aromatic Nitro Compounds.
Catalysts, 12, 467.

• Chen, L., Ye, W., Zuo, Y., & Zheng, C. (2019). Machine Learning in the Modelling of Catalytic Processes. Chemical Reviews,
119(18), 11020-11071.

Diagram showing flow diagram of catalyst development workflow

[Link]