Machine learning enabled high-throughput screening of hydrocarbon molecules for the design of next generation fuels

Abstract Next generation high energy density hydrocarbon (HEDH) fuels are urgently demanded to extend the range of propulsion system and meet additional requirements of new engines. We develop a facile and efficient methodology based on machine learning enabled high-throughput screening to accelerate the design of next generation fuels, and present a proof-of-concept study for discovering new HEDH fuels. This approach screens 319,895 hydrocarbon molecules using the key properties of fuel as the threshold values, and a group of 28 highly potent hydrocarbon molecules with high net heat of combustion, high specific impulse, high density and low melting point has been identified. The as-discovered molecules possess distinctive ring composition and unique spatial structure, which direct the synthetic efforts toward next generation HEDH fuels. This strategy not only discovers a new group of polycyclic molecules as competitive fuel candidates but also accelerates the development of new HEDH fuels.