Machine-Learning-Guided Discovery of CH₄ Combustion Catalysts Operating in the Presence of SO₂

Methane (CH₄) combustion under lean conditions is a critical reaction for controlling unburned hydrocarbon emissions in natural gas engines. However, the development of highly active and sulfur-tolerant catalysts remains a major challenge due to the severe deactivation caused by sulfur compounds such as SO₂. In this study, we adopted a machine-learning (ML)-guided strategy to accelerate the discovery of CH₄ combustion catalysts that are tolerant to sulfur poisoning. Starting from 16 initial catalysts and conducting 24 cycles of a closed-loop discovery system (ML prediction + experiment), a total of 300 multielemental catalysts were experimentally evaluated under identical conditions in the presence of SO₂. Through this approach, over 30 catalysts exhibiting high CH₄ conversion and excellent sulfur tolerance were identified. Among them, Pd(2)-Ru(0.4)-Ir(0.3)-Pt(0.3)/ZrO₂_JRC3_RC-100 demonstrated the highest catalytic performance. Control experiments and comprehensive characterizations, including in situ/operando spectroscopy, revealed the individual and synergistic roles of each component in enhancing both activity and resistance to sulfur poisoning.