A data-driven multi-level simulation framework for ammonia-syngas combustion

• A comprehensive multi-level modelling framework was developed to investigate ammonia-syngas combustion. • Machine learning models were coupled with a genetic algorithm to screen optimised operating conditions. • An optimal ammonia–syngas blend was identified, simultaneously reducing NO, CO, and CO 2 emissions. • Reaction pathways underlying emission reductions were revealed by reactive molecular dynamics simulations. • Application of the modelling framework was exemplified by simulations at molecular and device scales. Zero-carbon ammonia and renewable low-carbon syngas fuels offer a promising pathway to a cleaner and more sustainable energy future. Combustion of ammonia-syngas blends may address the limitations of each fuel, enabling simultaneous reductions in nitrogen and carbon emissions. However, the complex fuel composition and operating conditions pose significant challenges for traditional methods of optimising the combustion performance of such blends due to the high-dimensional parameter space and conflicting performance objectives. In this study, we develop a data-driven, multi-level simulation framework that combines machine learning-based surrogate modelling with the non-dominated sorting genetic algorithm for multi-objective optimisation. This framework identifies the optimal ammonia-syngas fuel composition to minimise emissions, ensuring flame stability while maintaining a laminar burning velocity similar to methane combustion and acceptable flame temperatures. Quantitatively, the optimised blend achieves substantial emission reductions compared to the baseline syngas mixture, with NO, CO, and CO 2 decreased by 23.8 %, 29.0 %, and 14.8 %, respectively. To comprehensively assess the performance of this optimal composition, we utilise multi-level combustion simulations, including laminar premixed flame simulation, Reynolds-Averaged Navier-Stokes (RANS) simulation, and reactive molecular dynamics (RMD) simulation. RANS simulations examine the spatial distribution of combustion outputs in real-world scale applications, while RMD simulations provide insights into detailed reaction pathways at the molecular level. Through the application of the multi-level modelling framework, this study offers understanding, prediction and guidance for the effective utilisation of low-carbon gaseous fuel mixtures, revealing the intrinsic relationship between macroscopic combustion properties and the underlying microscopic reaction mechanisms responsible for pollutant formation.