Multiscale Analysis of Lubricating Grease: Molecular Self-Assembly, Shear Behavior, and Machine Learning-Assisted Viscosity Prediction

Lubricating grease is a semisolid material widely used in various mechanical systems, composed of base oil, thickeners, and additives. The network structures formed by thickeners offer grease special rheological properties. The viscosity of grease, a key indicator of its lubricating performance, is affected by the combined influence of temperature, shear rate, and thickener ratio. In this research, via molecular dynamics simulations and quantum chemistry calculations, combined with machine learning methods, we first demonstrate the self-assembly behavior of the three-dimensional network structure of lithium-based lubricating grease and identify three structural components crucial for network formation: COO^––Li⁺–COO^–, COO^––Li⁺–OH, and OH–OH. Electrostatic interactions mainly drive the self-assembly of thickeners, with hydrogen bonds also playing a role. Nonequilibrium molecular dynamics simulations are conducted to calculate viscosities under different shear rates, temperatures, and thickener ratios. The results show significant shear thinning with increasing shear rate and temperature, and the viscosity increases with increasing thickener ratios. Machine learning algorithms are applied to predict grease viscosities, with ensemble models using the boosting method providing the most accurate prediction performance (coefficients of determination over 0.985). Feature importance and Shapley additive explanation analysis indicate that the order of feature importance is shear rate > temperature > thickener ratio, in which shear rate and temperature have negative effects on the predicted values, whereas thickener ratio has a positive effect. This research offers molecular insights into the formation of lithium-based grease networks, helps us understand its rheological behavior, which is affected by multiple factors, and provides guidance for designing lubricating grease.