Predicting NOx emissions from porous media burners using physics-informed graph neural networks

The combustion of ammonia is accompanied by the release of NO𝑥, which refers to both NO and NO2 i.e., the nitrogen oxides contributing to air pollution. A potential technology for complete combustion
of ammonia and reduced NO𝑥
emissions is combustion in porous media. Multiple test cases are defined
using 2D burner configurations that are investigated computationally by means of Direct Numerical
Simulations (DNS). The main objective of this work is to study the exhaust gas products for a wide
range of burner configurations. However, simulating a large number of burner setups using DNS is
time-consuming and costly. To reduce the number of DNS computations, a physics-informed deeplearning model based on Graph Convolution Neural Networks (GCNNs) is employed. In a first step,
the effectiveness of data-driven GCNNs is validated for reactive flows in porous media and preliminary
data is shown here. Subsequently, GCNNs augmented with constraints from combustion chemistry are
implemented to train on sparse data and to predict the combustion characteristics of the porous media
burner with high efficiency.