Numerical Simulation of NOx Emissions in a Lean Premixed Hydrogen Combustor Using Different Levels of Fidelity

This work presents a comprehensive computational framework for the prediction of NOx emissions in lean premixed hydrogen combustion systems for aviation applications, using the swirl-stabilized AHEAD combustor as the reference test case. Three modeling levels are assessed across inlet air temperatures from 313 to 693 K, mass flow rates from 80 to 295 kg/h, and equivalence ratios from 0.30 to 1.00: a 2D axisymmetric RANS with the k-omega SST closure, a 3D RANS with the Explicit Algebraic Reynolds Stress Model, and a Chemical Reactor Network calibrated via the Covariance Matrix Adaptation Evolution Strategy. Turbulence-chemistry interactions are described through the Eddy Dissipation Concept, and both the Naik and CRECK kinetic mechanisms are tested. Given the inherent availability of both 2D and 3D CFD datasets, a lightweight modified U-Net is introduced to correct the computationally efficient 2D fields toward their corresponding 3D planar references. Validation across the operating envelope indicates that while the models exhibit differences in predictive accuracy, the Chemical Reactor Network successfully captures the fundamental emission trends, providing NOx predictions with an accuracy comparable to the CFD models but at a computational cost several orders of magnitude lower. Furthermore, the U-Net effectively bridges the spatial fidelity gap, reducing the 2D-3D discrepancies for velocity, temperature, NO, and NO2 fields while adding only a marginal computational overhead compared to full 3D computations.