Pathways to sustainable fuel design from a probabilistic deep learning perspective

Abstract

To achieve net zero CO₂ emissions by 2050–2060, decarbonising the hard-to-abate sectors such as long-distance, heavy-duty transport is a top priority worldwide. These sectors are particularly challenging to decarbonise due to the use of high-energy-density liquid fossil fuels. In this context, designing low-carbon alternative fuels compatible with existing engines and fuel infrastructures is essential. This work presents an advanced fuel design framework to develop sustainable fuels that meet the high energy density requirements of heavy-duty vehicles. The fuel design approach is built upon a probabilistic perspective by considering a conditional generative model to predict the physicochemical properties of pure compounds and fuel blends with confidence bounds required for decision-making tasks. The probabilistic model is then integrated into an inverse design framework to design fuels with specific requirements. Finally, the fuel design framework is employed to develop new diesel fuel compositions according to the desired targets: ignition quality (cetane number) and sooting tendency (yielding sooting index). The AI-assisted fuel design approach can potentially lead to sustainable liquid fuels that are fully compatible with the existing utilisation equipment and can satisfy the requirements of different application sectors.