(705a) Interpretable Machine Learning-Guided Plasma Catalysis for H2 Production from NH3

Hydrogen (H₂) is a key energy carrier for decarbonizing the transportation sector, but its large-scale adoption is hindered by storage and transport challenges.¹ Ammonia (NH₃) offers an energy-dense and carbon-free alternative for H₂ storage and transportation.² However, conventional NH₃ decomposition requires high temperatures (>773 K) and noble metal catalysts such as Ru, raising cost and sustainability concerns.³ To overcome these limitations, we integrated non-thermal plasma (NTP) catalysis with interpretable machine learning (ML) and multi-scale simulations to design cost-effective, earth-abundant catalysts for NH₃ decomposition under mild conditions. Our study demonstrates that under NTP conditions, Co exhibits superior catalytic activity (Figs. 1a-b) compared to Ru under conventional heating. By employing density functional theory (DFT)-based microkinetic modeling and ML-assisted catalyst screening, we identified nitrogen binding energy (E_N ) as the key descriptor governing NH₃ decomposition rates. Using this insight, we screened over 3,300 bimetallic compositions and discovered highly active catalysts such as Fe₃Cu, Ni₃Mo, Ni₇Cu, and Fe₁₅Ni, which outperformed their monometallic counterparts in experimental validations. The Random Forest model (Fig. 1c) has the highest accuracy, optimizing at 48 trees with a mean absolute error of 0.16 eV. SHAP value analysis (Fig. 1d) indicated the d-band filling of alloy surface atoms was the most influential factor in E_N predictions. Our techno-economic analysis further supports the viability of plasma catalysis for on-site hydrogen production, achieving a hydrogen production cost below $1/kg H₂ and a reduced carbon footprint of ~0.91 kg CO₂ per kg H₂. These results establish machine learning-guided plasma catalysis as a scalable and cost-effective strategy for sustainable hydrogen generation.

References

1. Griffiths, S. et al., Global Energy, 2022, 28, 46-58.
2. Hansgen, D. A. et al., Nature Chemistry, 2010, 2, 484-489.
3. Bayer, B. N. et al., Plasma Chemistry and Plasma Processing, 2024, 1-18.