(472d) Highly Accurate Prediction of Pure Component and Mixture Viscosities of Hydrocarbons Using the Tlvmie Force Field

Many physical properties are needed for chemical process engineering. Experimental measurement of these properties is ideal, but such is often not possible due to a variety of problems including: an inability to obtain samples with sufficient purity, thermal decomposition, safety issues, the phase of the compound at temperatures accessible to the apparatus, and cost. A variety of prediction methods and strategies are used in such cases to provide the needed property data. Many work well, especially those for thermodynamic properties, but the liquid viscosity of a compound or mixture is an important property for which prediction is notoriously difficult. Despite the approach taken, tests show that prediction methods commonly have more than 25-50% error outside of the set of compounds for which the method was trained, with some producing errors of more than 75%.

The recently-developed TLVMie (Transferable Liquid Viscosity Mie) force field has displayed superior capability at predicting liquid viscosity of alkanes and alkylbenzenes. This work describes how this forcefield has been expanded to include branched hydrocarbons and mixtures. The presentation will begin by outlining the basics of the forcefield and the standardization of the simulation methods—the latter of which is crucially important to the transferability of the technique. Next, the prediction capability of the new formalism is evaluated with 49 branched alkanes and 6 branched alkylbenzenes with the results showing an average absolute error of 5.7% for compounds outside the training set. Finally, the performance of the model is tested for 31 different binary, ternary, quaternary, and quinary mixtures containing alkane and alkybenzenes. Despite no new parameterization being done, the experimental data for these systems are reproduced to 4.5%. Taken as a whole, the results indicate that the prediction accuracy of the TLVMie force field is superior to existing techniques for both pure component and mixture viscosities, especially for systems with no experimental data due to the high transferability of the technique.