Multiscale modeling large deformation in granular media

Abstract

Computational modeling of large deformation in granular media has to tackle two concurrent challenges: (a) the continuous change of topology and boundary conditions during the loading process and (b) the loading history and state dependency of mechanical behavior of granular media. Accurate and efficient simulations of large deformation in granular media require not only advanced material models, but also robust computational algorithms and tools. The focus of this dissertation is placed on developing a novel, unified multiscale modeling framework to address both challenges for effective simulation of large deformation in granular media. A multiscale modeling scheme is proposed by hierarchical coupling of material point method (MPM) and discrete element method (DEM), wherein MPM is employed to solve large deformation engineering scale problems of granular media under complex loading conditions, whereas the DEM solves a mesoscale granular assembly which serves as representative volume element (RVE) to produce nonlinear, loading-history dependent material response for each material point of the MPM. The proposed coupling framework not only inherits the advantages of MPM (e.g. Lagrangian description) in tackling large deformation of bulk mass, but also helps avoid the need for complicated, phenomenological assumptions on constitutive models for granular media that have to account for high nonlinearity at finite strain. The predictive capacity of the proposed framework is exemplified by 2D simulations of several geotechnical problems, including biaxial compression test, rigid footing, soil-pipe interaction, and soil column collapse. It is further used to systematically investigate the pull-out of anchors in sand in offshore geotechnics. An effective, scalable Message Passing Interface (MPI) parallel scheme is further developed to implement within the proposed framework for high-performance super-computing of large scale 3D engineering problems. The key idea of the parallelization is to bind individual RVE to corresponding MPI process and conduct subsequent DEM computation locally thereafter. This treatment helps bypass the repeated, random distribution of DEM packings in conventional multiscale modeling, and significantly reduces the cost on RVE (data) formatting and transferred among distributed nodes. The collapse of a three-dimensional granular column is simulated to showcase the performance of the multiscale framework enhanced by the new parallel scheme. Two further enrichments are made for the proposed multiscale framework. (1) A semi-implicit-explicit extension for MPM is further proposed to enhance the capability of the proposed framework for the simulation of saturated porous media with incompressible interstitial fluid constituents. The algorithm features an implicit treatment of the pore pressure field and solving the strongly coupled PDE with the fractional step method, where an intermediate acceleration field is introduced to decouple the variables and to advance the computation to next time instance by multiple substeps. The semi-implicit-explicit extension for MPM helps reduce the pressure oscillation and avoid restrictions on time step related to fluid incompressibility and soil permeability that commonly associated with explicit MPM. The multiscale framework further takes advantage of the effective stress principle by extracting the effective mechanical responses of a saturated granular media by DEM simulations at each RVE. The behavior of interstitial fluid is only described and treated at the macroscale of a domain. The hydro-mechanical coupling multiscale approach is firstly benchmarked by 1D consolidation before being applied to simulate 2D wave propagation and soil column collapse problem of saturated granular media. (2) The DEM part of the multiscale framework is further enriched by considering realistic particle shape. The use of poly-superellipsoid particles is demonstrated for efficient modeling of non-spherical grains. This feature enables our flexibility for modeling naturally anisotropic granular media, which is exemplified by a simulation of rigid footing resting on a inherent anisotropic soil foundation.

Date of Award	2020
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology