An a posteriori assessment of AI-designed diesel blends: Molecular dynamics simulation of pollutant formation

To achieve net zero CO2 emissions by 2050–2060, decarbonising hard-to-abate sectors, such as long-distance heavy-duty transport, is a top priority globally. These sectors are particularly challenging to decarbonise because of their reliance on high-energy-density liquid fossil fuels. In this context, it is crucial to design low-carbon alternative fuels that are compatible with existing engines and fuel infrastructures. This work presents an advanced framework for fuel design to develop sustainable fuels that satisfy the high energy density requirements of heavy-duty vehicles by integrating machine learning models with physics-based models. Firstly, we build a detailed quantitative mapping between the chemical composition of the fuel and the key properties of fuel utilisation using a conditional generative model. After training the conditional generative model that maps the molecular structure of the fuel to key combustion characteristics, we leverage it in a formal inverse design framework to screen novel fuel blends that match the required performance properties. Finally, a posteriori assessment of pollutant formation of the designed fuel blends is performed using molecular dynamics. In particular, using the reactive force field parameters in molecular dynamics simulation, the detailed pathways of soot formation examined include the thermal decomposition of fuels, the formation of aromatic rings, the mechanism of nucleation, and the mass growth of nascent soot. The results show that AI-designed fuels provide a high performance in terms of harmful emissions, while keeping key combustion properties of conventional diesel fuels, allowing them to act as a drop-in.