CFD–ML Synergistic Modeling of NH3–CH4 Combustion in a Model Gas Turbine

Abstract This study reports an integrated CFD and Machine Learning (ML) framework to study the combustion characteristics of ammonia/methane blends in a model gas turbine. The CFD model uses 242 species and 1769 reactions chemical kinetics mechanism as well as a steady laminar flamelet model to simulate turbulence-chemistry interaction. A total of 48 different CFD simulations were performed to characterize the gas turbine performance over a range of equivalence ratios, ammonia/methane blend ratios, and two fuel injection configurations. Gas turbine performance was evaluated in terms of exhaust gas temperature, combustion efficiency, and emissions of CO, CO2, NOx, and NH3 slip. Overall, the CFD simulations show that the 45° fuel injection configuration leads to higher combustion efficiency and lower emissions compared to in-line fuel injection. Subsequently, a Gaussian Process Regressor (GPR) was developed based on the data derived from these 48 CFD simulations. The efficacy of the ML was evaluated by performing blind tests on data generated from two fresh CFD simulations within the parameter space, and percentage errors between the predictions from the ML model and CFD simulations were reported. Most of the ML-predicted values were within the 10% error margin, with a 45° injection configuration showing smaller errors. Finally, the dominant parameters in the prediction of combustion emissions and performance were determined through Global Sensitivity Analysis using the developed GPR models.