This project couples machine learning with reactive molecular dynamics (ReaxFF MD) to overcome the accuracy and computational limits of conventional empirical force fields in ammonia (NH₃) combustion modeling. Deep neural networks are used to construct high-precision machine-learning potential energy surfaces, replacing empirical potentials and enabling long-timescale simulations of complex reaction networks. ML algorithms are then applied to large reaction-trajectory datasets to identify key intermediates, activation-energy variations, and dominant reaction pathways, with a particular focus on the formation mechanism of NO_x. Extracted micro-kinetic parameters are fed back into the reaction model to globally optimize combustion pathways. Interdisciplinary collaboration with ISEE (deep-learning architecture optimization) and IGSES peers (CFD and engine experiments) builds a multiscale workflow connecting atomistic kinetics to engine-scale combustion.

This project couples machine learning with reactive molecular dynamics (ReaxFF MD) to overcome the accuracy and computational limits of conventional empirical force fields in ammonia (NH₃) combustion modeling. Deep neural networks are used to construct high-precision machine-learning potential energy surfaces, replacing empirical potentials and enabling long-timescale simulations of complex reaction networks. ML algorithms are then applied to large reaction-trajectory datasets to identify key intermediates, activation-energy variations, and dominant reaction pathways, with a particular focus on the formation mechanism of NO_x. Extracted micro-kinetic parameters are fed back into the reaction model to globally optimize combustion pathways. Interdisciplinary collaboration with ISEE (deep-learning architecture optimization) and IGSES peers (CFD and engine experiments) builds a multiscale workflow connecting atomistic kinetics to engine-scale combustion.