Data-driven discovery of heat release rate markers for premixed NH3/H2/air flames using physics-informed machine learning

The spatial distribution of heat release rate (HRR) is important for flame front identification. However, direct measurement of HRR is impossible using the current experimental technology. Indirect HRR markers have thus been proposed and used in experiments to approximate the heat release profile. Suitable HRR markers have been identified in previous studies for flames with hydrocarbon fuels. However, study of HRR markers for premixed flames with NH3/H-2 blend fuel has rarely been done. Due to the increasing interest in carbon-free fuels, such as NH3/H-2, a generally-valid but accurate HRR marker is in demand. The present study involves data-driven discovery of optimal HRR markers for premixed NH3/H-2/air flames using physics -informed machine learning. The training data are generated by detailed simulations of turbulent flames with detailed chemistry. Several markers are finally identified. The most accurate combinations that can be measured experimentally read [NH](0.952)[OH](0.062) and [NH2](1.039)[OH](0.641). They show very high reconstruction quality for both spatial and temporal evolutions of HRR in premixed NH3/H-2/air flames for a variety of equivalence ratios (ranging from 0.45 to 1.25), fuel compositions (NH3/H-2 ratios ranging from 5/5 to 7/3 by volume), and turbulence intensity (Re-t ranging from 16.5 to 145.5). Their validity is not limited to the specific kinetic mechanism employed to generate the training data.

Automated summary

This study utilizes physics-informed machine learning to discover optimal HRR markers for premixed NH3/H-2/air flames. The identified markers show high reconstruction quality for various flame conditions and are not limited to specific kinetic mechanisms.