An AI-driven framework for property-oriented fuel design toward sustainable engine combustion.

Co-optimizing renewable synthetic fuels and internal combustion engines is critical for decarbonization, with fuel design as a crucial approach. However, the combinatorial explosion of fuel compositions renders composition-driven design infeasible. While artificial intelligence (AI) offers optimization tools, data scarcity hinders its application in engine research. Therefore, we propose an AI-driven fuel design framework using physicochemical properties to bridge fuel composition and engine performance. Treating properties as direct design variables, this framework reduces design dimensionality and enables standardized integration of data from different studies. Based on a literature-mined dataset, we trained a predictive model and applied it in a proof-of-concept demonstration to identify a tailored property combination. Relative to diesel, this combination improved thermal efficiency by 6.62% and reduced NOx and soot emissions by 19.28% and 67.76%, respectively. This framework provides a pathway for AI-driven engine research and facilitates property-oriented design, which could instruct upstream synthesis and help reduce trial-and-error in fuel development.