Smart Surfactant Discovery with Machine Learning and Molecular Dynamics

In this study, we combine classical molecular dynamics (MD) and machine learning (ML), to optimize the selection of surfactants for reduced interfacial tensions (IFT). The applications of surfactants are vast, and the choice of a suitable one for the desired application is a demanding task due to the large number of available molecules. We have employed MD simulations to create an oil model representative of complex hydrocarbon mixtures found in pre-salt crude oil reservoirs. The ML models were trained using the IFT obtained from a small number of simulations, with features that were not from the simulation itself, but from molecular properties. Hence, we could predict the IFT values for nonsimulated surfactants, with the inferred results, we chose new molecules to simulate. The new results are then incorporated into the ML model and new inferences are made. This process happens iteratively, until we identified a surfactant with an exceptional IFT reduction, which accounted for six interactions. This is the so-called active learning process. Subsequently, the proposed surfactant was experimentally tested, which confirmed the simulated results. These findings hold significant promise for a smart process to find the suitable surfactant, as the number of available surfactants is large, and the combinations of fluid interfaces is as well vast.