Prediction of Sooting Index of Fuel Compounds for Spark-Ignition Engine Applications Based on a Machine Learning Approach

A joint consideration of potential combustion and emission performance in spark-ignition engines is essential for designing gasoline fuel replacements and additives, for which the knowledge of the fuels’ characteristic properties forms the backbone. The aim of this study is to predict sooting tendency of fuel molecules for spark-ignition engine applications in terms of their yield sooting indexes (YSI). In conjunction with our previously developed database for gasoline compounds, which includes the physical and chemical properties, such as octane numbers, laminar burning velocity, and heat of vaporization, for more than 600 species, the identification of fuel replacements and additives can thus be performed jointly with respect to both their potential thermal efficiency benefits and emission formation characteristics in spark-ignition engines. For this purpose, a quantitative structure-property relationship (QSPR) model is developed to predict the YSI of fuel species by using artificial neural network (ANN) techniques with 21 well-selected functional group descriptors as input features. The model is trained and cross-validated with the YSI database reported by Yale University. It is then applied to estimate the YSI values of fuels available in the database for gasoline compounds and to explore the sensitivity of fuel’s sooting tendency on molecular groups. In addition, the correlation of YSI values with other properties available in the gasoline fuel database is examined to gain insights into the dependence of these properties. Finally, a selection of potential gasoline blending components is carried out exemplarily, by taking the fuels’ potential benefits in thermal engine efficiency and their soot formation characteristics jointly into account in terms of efficiency merit function and YSI, respectively.