(335d) Machine Learning Boosted Design of Plasma-Enhanced Catalytic Ammonia Cracking

Hydrogen (H₂) is an energy carrier with high energy content (142 kJ/g) and a non-polluting energy source. Hydrogen's implementation is hindered, however, by delivery and storage issues. Ammonia (NH₃) is being studied as a storage medium for H₂ due to its high H₂ density, ability to easily condense, storage and distribution. However, thermal NH₃ cracking to generate H₂ usually requires the reaction temperatures above 773 K and an earth-rare metal catalyst (i.e., Ru) to achieve complete conversion.

One strategy to activate ammonia at mild conditions over earth-abundant metal catalysts is by applying non-thermal plasma since the vibrationally excited species produced in non-thermal plasmas lower the temperature of NH₃ cracking. However, modelling interfacial plasma-surface interaction is a multi-scale problem and building such simulation for a complex reaction mechanism over a large search space of catalyst materials is computational immense.

To solve the above challenge, it is important to identify the key descriptor that impacts the overall rates of ammonia cracking under non-thermal plasma conditions and apply such descriptor to screen the large search space of catalysts. Thus, we first built the microkinetic model for ammonia cracking over various metallic surfaces (e.g., Ni, Co, Ru, Fe). Our theoretical results indicate that plasma lowers the energetics of the first N-H bond cleavage in NH3 and improves the overall rates by creating vibrationally excited state. Then we performed the sensitivity analysis to identify the key descriptor of N₂ dissociative desorption from surface N^*. We then applied this descriptor to scan the catalyst materials and found that earth-abundant Ni-Co alloy showed satisfactory catalytic performance compared to that of earth-rare Ru. AI-predicted catalysts will be further validated by experiments using a dielectric-barrier discharge reactor. The integration of multi-scale modeling, machine learning algorithms and experiments will facilitate materials development in interfacial plasma-surface reaction rather than trial-and-error.