Temperature-Dependent Density and Viscosity Prediction for Hydrocarbons: Machine Learning and Molecular Dynamics Simulations

P Panwar and QP Yang and A Martini, JOURNAL OF CHEMICAL INFORMATION AND MODELING (2023).

DOI: 10.1021/acs.jcim.3c00231

Machine learning-based predictivemodels allow rapidand reliableprediction of material properties and facilitate innovative materialsdesign. Base oils used in the formulation of lubricant products arecomplex hydrocarbons of varying sizes and structure. This study developedGaussian process regression-based models to accurately predict thetemperature-dependent density and dynamic viscosity of 305 complexhydrocarbons. In our approach, strongly correlated/collinear predictorswere trimmed, important predictors were selected by least absoluteshrinkage and selection operator (LASSO) regularization and priordomain knowledge, hyperparameters were systematically optimized byBayesian optimization, and the models were interpreted. The approachprovided versatile and quantitative structure-property relationship(QSPR) models with relatively simple predictors for determining thedynamic viscosity and density of complex hydrocarbons at any temperature.In addition, we developed molecular dynamics simulation- based descriptorsand evaluated the feasibility and versatility of dynamic descriptorsfrom simulations for predicting the material properties. It was foundthat the models developed using a comparably smaller pool of dynamicdescriptors performed similarly in predicting density and viscosityto models based on many more static descriptors. The best models wereshown to predict density and dynamic viscosity with coefficient ofdetermination (R (2)) values of 99.6% and97.7%, respectively, for all data sets, including a test data setof 45 molecules. Finally, partial dependency plots (PDPs), individualconditional expectation (ICE) plots, local interpretable model-agnosticexplanation (LIME) values, and trimmed model R (2) values were used to identify the most important static anddynamic predictors of the density and viscosity.

Return to Publications page