Machine learning of lubrication correction based on GPR for the coupled DPD–DEM simulation of colloidal suspensions

Hydrodynamic interactions have a major impact on the suspension properties, but they are absent in atomic and molecular fluids due to a lack of intervening medium at close range. To reproduce the correct hydrodynamic interactions, lubrication correction is essential to compensate the missing short-range hydrodynamics from the fluids. However, lubrication correction requires many simulations in particle-based simulations of colloidal suspensions. To address the problem, we employ an active learning strategy based on Gaussian process regression (GPR) for normal and tangential lubrication corrections to significantly reduce the number of necessary simulations and apply the correction to the coupled multiscale simulation of monodisperse hard-sphere colloidal suspensions. In particular, a single-particle dissipative particle dynamics (DPD) model with parameter correction is used to describe the solvent–solvent and colloid–solvent interactions, and a discrete element method (DEM) model to depict the colloid–colloid frictional contacts. The lubrication correction results demonstrate that only six and four independent simulations (observation points for GPR training) are required to achieve accurate normal and tangential lubrication corrections, respectively. To validate the machine learning of lubrication correction based on GPR, we investigate the self-diffusion coefficients of colloids, suspension rheology and microstructure using the coupled DPD–DEM model with GPR lubrication correction. Our simulation results show that the machine learning of lubrication correction based on GPR is effective and the lubrication corrected DPD–DEM model is indeed capable of accurately capturing hydrodynamic interactions and correctly reproducing dynamical and rheological properties of colloidal suspensions. Moreover, the machine learning of lubrication correction based on GPR is not limited to the coupled DPD–DEM simulation of colloidal suspensions presented here, but can be easily applied to other particle-based simulations of particulate suspensions.