Data-driven prediction of laminar burning velocity for ternary ammonia/hydrogen/methane/air premixed flames

Zero-carbon fuels such as hydrogen and ammonia play a pivotal role in the energy transition by offering cleaner alternatives to natural gas (methane), especially in industrial combustion systems. Binary and ternary blends of these fuels offer a transitionary, low-carbon solution in the near future. Laminar burning velocity (LBV), as a fundamental combustion property, is significantly different for ammonia, hydrogen, and methane. Although the LBV of binary blends of these fuels is well-studied, ternary blends have not been extensively studied. In this study, the primary objective is to employ a simple ensemble learning method to predict the LBV of ternary ammonia/hydrogen/methane/air mixtures. The training dataset consists of experimental data sourced from a large number of publications (3,846 data points), as well as synthetic data generated by 1D freely propagating premixed flame simulations in Cantera using a detailed chemical kinetic model. Three machine learning algorithms, namely artificial neural networks, gaussian process regression, and extreme gradient boosting trees are trained and optimised. Then, a simple ensemble averaging method is used to reduce overfitting and improve robustness. The ensemble model achieves coefficient of determination (R2) of 0.991 on the test set with an inference time that is approximately 8,000 times faster than the 1D simulation run time. The ensemble model is capable of predicting LBVs of ammonia/hydrogen/methane/air mixtures for T=295–756K, P=1–10bar, ϕ=0.5–1.8 across all possible blending ratios.