Physicochemical properties of synthetic fuels are important but difficult to measure/predict, especially when complex surrogate fuels are concerned. In the present work, machine learning (ML) models are constructed to discover intrinsic chemical structure-properties relationships. The models are trained using data from molecular dynamics (MD) simulations. The fuel structure is represented by molecular descriptors. Such a symbolic representation of the fuel molecule allows to link important features of the fuel composition with key properties of fuel utilization. The results show that the present approach can predict accurately the fuel properties of a wide range of pressure and temperature conditions.
Keywords fuel properties, molecular dynamics simulations, molecular descriptor, machine learning models