Data-driven combustion kinetic model generation with optimized pathway sampling: A case study for fuel ammonia

• Automatic bottom-up model generation provides promising alternative to model design • MA-DOPS samples reactions and rate parameters from reliable literature models • Generated models evolve systematically by expanding and refining reaction pathways • The approach produced accurate model, outperforming source models for diverse data • Micro flow reactor experiments provided valuable extension to validation conditions A novel data-driven algorithm, MA-DOPS (Model Assembly with Data-driven Optimized Pathway Sampling), is introduced for the automated generation of combustion kinetic models. The method systematically assembles models by sampling and evolving reaction pathways from published models to minimize simulation error against experimental data. MA-DOPS builds models in a bottom-up manner, incorporating reactions and parameterizations from validated sources, ensuring physically meaningful model generation. The performance of the method was evaluated in three test cases with increasing complexity of the test conditions. First, algorithm control parameters were systematically analysed, identifying configurations that balance computational cost and model accuracy. Second, the constraining potential of different experimental data types was assessed, including micro flow, jet-stirred and tubular flow reactor speciation data, shock tube ignition delay times, and laminar burning velocities. This analysis showed that micro flow reactor measurements provide valuable complementary information for experimental validation conditions, demonstrating their efficiency as target for model development. Finally, a full-scale model generation task for ammonia combustion was carried out. The resulting model was compiled based on rate parameters from 72 source models and it outperformed all individual sources, showing good agreement with a comprehensive reference dataset, particularly excelling in micro flow reactor simulations. These results highlight the utility of the MA-DOPS approach in generating accurate combustion models by combining literature data with systematic algorithmic refinement.