Data-driven representative models to accelerate scaled-up atomistic simulations of bitumen and biobased complex fluids

Complex molecular organic fluids such as bitumen, lubricants, crude oil, or biobased oils from biorefineries are intrinsically challenging to model with molecular precision, given the large variety and complexity of organic molecules in their composition. Large scale atomistic simulations have been historically limited by this complexity, which has hampered the bottom-up molecular design of these materials, something especially relevant given the current surge of biobased fluids for sustainable applications and the cost of trial-and-error experimental developments. To address this limitation, we have developed an author-agnostic computational framework to generate data-driven representative models of any complex mixture of organic molecules directly from Gas Chromatography-Mass Spectrometry (GCMS) experimental characterisation, thus reducing human biases in model creation and providing a platform for self-driven digital development of molecular organic fluids. The method proposed generates statistically representative molecular samples that simplify the complexity of the fluid in a limited group of molecules, while capturing the critical chemical features needed to describe the overall properties of the mixture. As a case study, we generated a showcase of data-driven representative models from the GCMS characterisation of a bio-oil from the pyrolysis of pine bark, specially produced for this study. Pyrolytic biomass processing into bio-oils provides a waste valorisation route with applications in biorefinery products like asphalt additives and biofuel precursors. Our case study focuses on complex fluids such as bio-oils for asphalt rejuvenators for self-healing purposes or biofuel upgrading. Nevertheless, the general computational framework developed in this manuscript provides a platform for generating data-driven representative models of any bitumen or biobased organic fluid.