Molecular Dynamics And Multiscale Simulations Of Polymeric Flow

Polymeric fluid is a typical non-Newtonian fluid,which has a wide range of applications in daily life,industrial production and engineering science,and has important research value.Different from Newtonian fluid,the constitutive equation of polymeric fluid has considerable complexity and typical multiscale characteristics,and the macroscopic flow characteristics are closely related to the evolution of microscopic molecular morphology,accompanied by complex phenomena such as slip and instability.In this paper,the numerical simulations of polymeric flow with unknown constitutive models are carried out based on molecular dynamics,multiscale algorithm and machine learning.The main work and results are as follows:(1)Molecular dynamics simulations of the impact process of pure solvent droplets and dilute polymer solution droplets on a flat superhydrophobic substrate at nanoscale are carried out.The anti-rebound phenomenon of dilute polymer solution droplets at nanoscale is reproduced.The predominance of polymer-substrate interaction and droplet impact velocity in this phenomenon is first explained from the microscopic point of view.The effects of static wetting behavior,rheological properties and surface tension of different droplets on the impact results are excluded.The results show that the high impact velocity can enhance the stretch behavior of the polymer chains,make the polymer chains closer to the substrate,and increase the possibility of polymer molecules contacting the bottom substrate.Under the strong polymer-substrate interaction,the polymer molecules are adsorbed by the bottom substrate,which inhibits the whole droplet retraction process and eventually leads to the occurrence of anti-rebound phenomenon.(2)A modified multiscale algorithm based on seamless embedded multi-scale method,scale bridging method and leapfrog time coupling is developed,which does not require constitutive equation and has the strength of high parallel efficiency,and is utilized to investigate the microscopic characteristics and macroscopic flow characteristics of polymeric fluids between plates.The entire computational domain is described by a macroscopic model,using molecular dynamics to calculate the required local stresses.The algorithm is verified by creep recovery motion and pressure-driven flow of polymer melt.In particular,in the pressure-driven flow simulations of polymer solutions,This work is supported by ta normalized monotonically decreasing behavior between the maximum or average velocity within the channel and the solution concentration is found.The normalized reference concentration satisfying the power-law relationship is closely related to the overlap concentration of polymer solutions,and the reference velocity corresponds to the corresponding velocity of Newtonian fluid under zero shear rate viscosity.(3)A hybrid multiscale algorithm based on scale bridging method and domain decomposition method is proposed.This algorithm has the strength of no constitutive equation and boundary condition assumption,and its computational complexity is reduced by about two orders of magnitude compared with pure molecular dynamics simulations.It is also the first to be applied to the simulation of interfacial slip of polymer melts at low shear rates.The computational domain far from the wall is described by macroscopic model,and the local stresses required in the bulk regions and near the wall are calculated by molecular dynamics.The algorithm is verified by the non-slip behavior of Newtonian fluid and the shear-slip system of polymer melt.It is also found in the study of interfacial slip behavior that the variation of wall-fluid interaction can lead to completely opposite slippage trend.For weak wall-fluid interaction,due to the competition between fluid viscosity and interfacial friction coefficient,the slip length first remains constant and then increases rapidly with the increase of shear rate.For medium wall-fluid interaction,due to the complete dominance of fluid viscosity,the slip length first remains constant and then decays rapidly with the increase of shear rate.(4)The simulation and prediction of flow field information such as velocity field,pressure field and stress field based on physical equations of fluid mechanics and machine learning are carried out.Firstly,the Navier-Stokes equation and boundary condition are directly coded into the deep learning neural network to complete the direct prediction of the velocity field and pressure field of two-dimensional steady Newtonian flow without spatial discretization.Secondly,the neural network is applied to reconstruct the stress field from the velocity field,and the mapping between strain rate and stress is established.The constitutive models of Newtonian fluid and CarreauYasuda fluid are obtained respectively.The two-dimensional pressure-driven flow and lid-driven flow of Newtonian fluid and Carreau-Yasuda fluid are solved by coupling the obtained constitutive model with the governing equation.The simulation results are compared with the numerical solutions of pure computational fluid dynamics qualitatively and quantitatively to verify the feasibility of the present algorithm.