Challenge: Lubricants are a commercially-important class of chemical which are predominantly mixtures of alkanes. Yet understanding of how to improve some of their key properties is still relatively poor, even after study by a range of physico-chemical and thermodynamic methods. Can Alchemite deep learning help?
Solution: The existing relatively sparse experimental data was combined with results from molecular dynamics simulations. Alchemite was able to exploit property-property correlations in this data to predict the physical properties of known and new alkanes.
Outcome: Alchemite accelerated the identification of optimal hydrocarbons. Properties predicted included boiling point, heat capacity, and vapor pressure as a function of temperature as well as properties that were intractable by other prediction methods including density and shear viscosity. The results were significantly more accurate and consistent than those reproduced by the other methods.