Exploring the fuel structure dependence of laminar burning velocity: A machine learning based group contribution approach

The laminar burning velocity (LBV) is a fundamental property of a fuel/oxidizer mixture with high impact on combustion processes in practical engines. Profound knowledge of its dependence on the underlying molecular structures of hydrocarbon and oxygenated hydrocarbon fuels is of high interest. In the present work, a quantitative structure-property relationship model is developed for the first time to predict the LBVs of a wide range of fuels. For this purpose, an artificial neural network is trained based on a training set consisting of both the experimental LBV values of 124 fuel compounds and additional data obtained from numerical simulations with a detailed kinetic model. Twelve molecular groups as well as pressure, temperature, and fuel/air equivalence ratio serve as input features to the model. Cross-validation reveals a mean absolute error of 3.3 cm/s when applying the model to fuels, whose LBV datapoints were not used for training. In order to gain insights into the underlying fuel structure dependence of LBV, the model is then applied to analyze the functional group effects at unified conditions by means of sensitivity analysis and detailed fuel comparisons. It is found that unsaturation increases the LBV, while methyl substitution consistently has a negative effect for the wide range of fuel structures considered, which confirms similar findings in the literature. More interestingly, while carbonyl groups in ketones and aldehydes, ether groups in ethers, acetals, furanics, and oxygenated benzenoids, as well as hydroxy groups in n-alcohols tend to increase the LBV compared to corresponding non-oxygenated fuels of similar structures, ester and carbonate functional groups have a clearly negative impact. Overall, the results demonstrate that a group contribution approach in combination with a machine learning methodology is capable of predicting the LBVs of a wide range of fuel structures with acceptable accuracy, which can be useful for future fuel design.