This research aims at establishing the predictive capability of the discrete element method
(DEM), and using DEM simulation to investigate the behavior of sand, including the aging
mechanism in sands and the pore space evolution in granular media in response to the triaxial
shearing.
To establish the predictive capability of DEM, the simulated responses of granular
systems were validated against the experimental observations. A published biaxial shearing test
on a 2D assembly of randomly packed elliptical rods was chosen as the benchmark test. In the
corresponding DEM simulations, the contact model was derived and then validated using finite
element analysis; the associated parameters were calibrated experimentally. The flexible
(membrane) boundary was modeled by a bonded-particle string with experimentally calibrated
parameters. An iteration procedure was implemented to replicate the initial packing and also to
satisfy the boundary conditions. Overall, the DEM simulation is found effective in reproducing
the stress-strain-volumetric response, the statistical observation on the fabric anisotropy, and the
strain localization. Furthermore, the closer the numerical packing is to the experimental one, the
closer the response is reproduced, demonstrating the significance of the initial packing
reconstruction. Still, there are some minor differences between the experiment and simulation,
reflecting the limitations associated with the particle number and the measurement resolution
used in the experiment when reproducing the initial packing.
The shape modeling method in DEM simulation, i.e., the clump method, was examined.
When a clump that consists of a cluster of rigidly tied sub-spheres is used to model a target
particle of varying shape in DEM simulation, multiple sub-contacts on the clump can occur,
giving rise to over-stiff and over-damped responses. The sub-contact stiffness and damping
coefficient therefore have to be reduced; both the theoretical and empirical approaches, which
are applicable to the general cases of modeling non-spherical particles and using nonlinear
contact models, were proposed for this goal. In the theoretical approach, the stiffness and
damping coefficient of sub-contacts in the clump are reduced to 1/N
c of those values adopted in
the contact model at the target-particle contact, where N
c is the number of sub-contacts. In the
empirical approach, the reduction of these parameters is described using the Taylor series that
are obtained by curve fitting to the observational data. The validity of these two approaches were
examined using the cases of dynamic collision and quasi-static biaxial shearing. Both approaches
were found to be effective in reproducing the responses that fairly match the reference ones, i.e.,
comparable contact forces and particle velocities versus contact duration responses in the
collision cases, and similar stress-strain-volumetric responses in the biaxial shearing.
A novel method modeling flexible membrane boundary in 3D DEM simulation was
proposed. The membrane boundary is established by a network of bonded membrane particles;
this particle network is further partitioned into finite triangular elements. The associated
algorithm can accurately distribute the applied confining pressure onto the membrane particles
and determine the sample volume. This method enables an investigation on the mechanical
behavior of artificially cemented sands with strong, intermediate, and weak bond strengths using
both DEM simulation and experimentation. The focus was on the features of bond breakage and
the associated influences on the stress-strain response. Under triaxial shearing, the acoustic
emission rate captured in the experiment and the bond breakage rate recorded in the simulations
show resemblance to the stress-strain response, especially for strongly and intermediately
cemented samples. The simulations further reveal the shear-band formation coincides with the
development of bond-breakage locations due to the local weakness caused by the bond breakages.
Strain softening and volumetric dilation are observed inside the shear band while the region
outside the shear band undergoes elastic unloading. The weakly cemented sample exhibits a
strain-hardening response instead; bond breakages and the associated local weaknesses are
always randomly formed such that no persistent shear band is observed.
The DEM simulation was applied to investigate the aging process. The free volume was
found irreverent to the increase in the rigidity (given by small-strain modulus); therefore, the
conventional jamming diagram is modified to consider two order parameters of jammed
structures to elucidate the aging process. During aging, the geometrical order remains almost
unchanged. However, the contact-force order (similar to contact-force homogeneity)
continuously increases, thereby lowering the potential energy of the packing to become more
stable and rigid.
The density effect on aging-induced stiffness increase was also reported; the DEM
simulations on dense, medium dense, and loose samples were carried out for the examination.
Like the experimental observations, among these three samples, the medium dense sample
exhibits the highest aging rate, in terms of the increase in the small-strain shear modulus. This
result comes from the competition between two opposite effects, i.e., contact-force
homogenization and unjamming events that are relevant to the packing density and coexist
during aging. The loose sample has a less homogeneous contact-force distribution before aging
and therefore allows a higher capacity for contact-force homogenization during aging, which
gives rise to a greater aging rate. On the other hand, the unjamming event accompanied by a
sudden increase in the number of sliding contacts occasionally takes place during aging to
destabilize the soil structure, especially for loose sample. This in turn destroys the aging-induced
homogenization of contact forces and then demolishes the aging-induced stiffness increase.
A novel method for the pore network extraction in granular media is reported, which
unifies both the Delaunay tessellation (DT) and maximal ball (MB) methods in a complementary
way. This unified method therefore can retain the advantages of both methods, and, most
importantly, eliminate the disadvantages while either the DT or the MB method is used
individually. As a result of such a combination, the merging process of the Delaunay tetrahedra
can be carried out in an objective way by utilizing the clustering process of maximal balls.
Meanwhile, the problems generated in using the maximal balls method, i.e., underestimating the
throat size and producing tiny and incorrect throats, can be eliminated by incorporating the
Delaunay tessellation, which enables precisely partitioning the pore space and well defining the
throat size. The examination of this unified method, based on the results of pore network
extraction on a sample before and after the triaxial shearing using DEM simulations, corroborates this method is objective and effective. The morphology of the pore space therefore
can be accurately described in terms of the pore size, throat size and the associated distributions.
This unified method also enables the feasibility to characterize another dimension of fabric in
granular media, i.e., the throat-based fabric.
Post a Comment