02 e 7 e 516 e 4580 A 3923000000

You might also like

You are on page 1of 16

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/236177585

A bonded-particle model for cemented sand

Article in Computers and Geotechnics April 2013


DOI: 10.1016/j.compgeo.2012.09.001

CITATIONS READS

40 442

4 authors, including:

Martin Obermayr Klaus G. Dressler


ZF Friedrichshafen Fraunhofer Institute for Industrial Mathemati
15 PUBLICATIONS 122 CITATIONS 63 PUBLICATIONS 270 CITATIONS

SEE PROFILE SEE PROFILE

Univ.-Dr.-Ing. habil. Christos Vrettos


Technische Universitt Kaiserslautern
34 PUBLICATIONS 298 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Soil-Tool interaction View project

All content following this page was uploaded by Martin Obermayr on 30 June 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Computers and Geotechnics 49 (2013) 299313

Contents lists available at SciVerse ScienceDirect

Computers and Geotechnics


journal homepage: www.elsevier.com/locate/compgeo

A bonded-particle model for cemented sand


Martin Obermayr a,, Klaus Dressler a, Christos Vrettos b, Peter Eberhard c
a
Fraunhofer Institute for Industrial Mathematics, FraunhoferPlatz 1, 67663 Kaiserslautern, Germany
b
Division of Soil Mechanics and Foundation Engineering, Technical University of Kaiserslautern, 67663 Kaiserslautern, Germany
c
Institute of Engineering and Computational Mechanics, University of Stuttgart, Pfaffenwaldring 9, 70569 Stuttgart, Germany

a r t i c l e i n f o a b s t r a c t

Article history: We propose an extension of the Discrete Element Method for the numerical simulation of cemented
Received 25 March 2012 sands, in which spherical particles are bonded together by elastic beams connecting the centers of the
Received in revised form 29 August 2012 spheres. The parameters of this model are the strengths and stiffnesses of the bonds and particles. For
Accepted 2 September 2012
small strains, the elasticity of the bond element is equal to the well-known linear nite-element
Available online 15 October 2012
Timoshenko beam element with reduced integration. The nite rotations are represented by unit quater-
nions. An efcient way to compute relative rotations and to decompose them into their components is
Keywords:
presented.
Discrete Element Method (DEM)
Bonding
The results of triaxial compression tests on articially cemented sands are used to verify that the model
Cemented sand can capture the macroscopic behavior of such materials. The results show that peak stress mainly
Soil depends on the strength of the bonds and the number of initially bonded particles in the material. Results
of triaxial tests with different cement contents are reproduced by the analysis. An important parameter of
the model is the strength difference between tension and compression of the bond element. This property
controls the inuence of the conning pressure on peak strength. In the future, the model could be
adapted to other types of bonded materials like asphalt or rock.
2012 Elsevier Ltd. All rights reserved.

1. Introduction A number of different materials are characterized by such a


peak in the shear-stress vs. axial-strain curve followed by a strain
The prediction of reaction forces in soiltool interaction is an softening until the critical state is reached. Examples of such mate-
important task for the design of machinery used in construction rials are heavily overconsolidated soils and cemented or dense
and agriculture. Especially for earth-moving machinery, it enables sands. In the present study, a cemented sand with chemically
the assessment of fatigue life at an early stage of the product devel- bonded particles is taken as representative for this class of materi-
opment, enhancing the reliability of the machines. als [814].
Draft forces in cohesionless soil have already been studied Characteristics of cemented sands are summarized in the fol-
numerically, mostly by applying the Discrete Element Method lowing: The bonding of the grains yields a large initial stiffness.
(DEM) [16]. However, cohesionless materials cover only a part With increased load, the cement bonds progressively break until
of the application eld of such machinery. Another important class a peak state is reached [9]. The post-peak variation with strain de-
of soil materials is characterized by cohesive properties due to pends on the dilatancy of the material. Dilation causes volumetric
bonding between the particles. During the interaction of the buck- work leading to a gradual decrease of the stressstrain curve
et with the soil, the material breaks up into its granular constitu- [11,15]. The bonds can withstand much higher compressive stres-
ents. At the critical state, the strength of the initially cemented ses than tensile stresses [14]. At low conning pressures, the
material is then equal to that of the cohesionless counterpart. In strength due to cementation is most important, while at high pres-
many applications it is sufcient to consider just the critical state. sure levels frictional behavior dominates. As a consequence, the
The additional strength due to the cementation is then used as a peaks in the stressstrain curves disappear at high conning pres-
safety margin [7]. In contrast to that, the prediction of soiltool sures [9,11,16,17]. Depending on the conning pressures during
reaction forces can only be successful considering the shear the test, an apparent angle of internal friction occurs [8] or a
strength at peak of the cemented material. parallel offset of the MohrCoulomb failure curve [11] is observed.
At low conning pressures, the material fails in a brittle manner. At
high conning pressures, failure is mainly ductile as for the unce-
Corresponding author. Tel.: +49 631 31600 4676.
mented material [10,11].
E-mail address: martin.obermayr@itwm.fraunhofer.de (M. Obermayr).

0266-352X/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2012.09.001
300 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

Nomenclature

General Jb second moment of cross-section


e1, e3, ev axial, lateral and volumetric strain kN,kT normal and tangential stiffnesses
qg grain density L0,L initial and current length of the bond
r1, r3 axial and lateral stress mi,mr mass of particle i and reduced mass
sf shear stress at failure q12 relative rotation between particles 1 and 2
/ angle of internal friction qB,0, qB initial and current orientation of bond element
ccm cohesion intercept due to cementation q1,q2 orientations of particles 1 and 2
CC cement content qB1, qB2 relative orientations of bond element
h, h0 actual and initial height of sample ri, rij radius of particle i and average radius
p typical pressure rb ; rb absolute and relative radius of bond element
w deformation vector C strain vector
e components of relative rotation
Discrete element model j curvature vector
ab modication factor of bond strength nT,ij tangential deformation vector
ac compressive strength factor xi angular velocity in the global frame
ar random number a axis of rotation
as shear coefcient b1, b2, b3 basis vectors of the bond system
dij overlap of particles i and j Fs,ax, Fd axial, shear, and damping forces in bond frame
h quaternion angle fi force on particle i in the global frame
l coefcient of friction between particles FT,ij tangential force vector
raxial axial stress in bond element Ii inertia tensor of particle i
sshear shear stress in bond element ti torque on particle i in the global frame
u angle of rotation Tb,t,Td bending, torsional, and damping torque
Ab cross-sectional area of bond element u relative displacement
dN, dT normal and tangential damping coefcients V12 relative velocity in bond frame
D damping ratio vT,ij relative tangential velocity
Eb, Gb Youngs and shear modulus of bond element xi position of particle i in the global frame
Eu Youngs modulus of unbonded interaction x12 vector connecting the sphere centers
FN,ij magnitude of normal force
Ii,Ir inertia of particle i and reduced inertia
Shear
stress
0

cs

Normal stress

Fig. 2. Peak shear-stress envelope for soils resulting from cohesion co, soil tension
Fig. 1. Effects of cementation on the response of a soil: (a) uncemented soil, (b) ct, and cementation ccm (after [8]).
cemented soil, and (c) cemented soil at high normal effective stress (after [8]).

using hardening plasticity theory in combination with the Finite


The above mentioned effects are summarized in Fig. 1 which Element Method (FEM) [28].
shows idealized stressstrain curves for uncemented and cemen- In the DEM the bonding agent of the cemented material is often
ted soils. The effect of different types of cohesion on the peak represented by breakable bonds between particles. Some of the
shear-stress envelope is shown in Fig. 2. In this paper, we assume bond models are based on a transfer of tensile and tangential forces
that cohesion c0 and soil tension ct are negligible and focus on the until there is bond breakage [15,24,25,27,29,30] while others addi-
cohesion intercept due to cementation ccm. tionally use torque transfer to account for the nite extent of the
To capture such behavior and in order to consider large scale cementation [31,32]. The latter is mostly found in applications to
simulations involving large deformations including material sepa- rock mechanics.
ration in cemented soils, the DEM is applied in this research. The While experimental studies on cemented sand only allow
DEM [18] has already been used for modeling of cohesive materi- assessment of the global behavior of the material, in [25] a series
als, like concrete [1921], asphalt [22,23], hard rock [24], and ce- of tests was combined with an elaborate two-dimensional discrete
mented sands [15,2527]. Cemented soil has also been described element model. It was found that the peak shear strength is not
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 301

just determined by the bond strength. While bond breakages lead theory analogon. The model is then applied to simulate triaxial
to a strength reduction, the associated dilatancy partially compen- compression tests on typical cemented soil. Results are compared
sates for this. to ndings from the literature.
A study of the shear behavior and strain localization in cemen-
ted sands is presented in [26]. This study is based on a very ne 2. Material and methods
two-dimensional discrete element model. Different bond models
are compared, and the correlation between micromechanical ef- 2.1. The discrete element model
fects and the macroscopic response is investigated. It was found
that the size of the shear band depends on the bond model A discrete element simulation consists of a large number of,
assumed. usually rigid, bodies that interact via forces exerted at their com-
In [27], a bond model based on tangential and normal forces is mon contact points. The dynamics of the particles is described by
presented and successfully applied in a two-dimensional simula- the Newton and the Euler equations [35]
tion of weathering-induced failure of a cliff.
A numerical model for bonded particles including a transfer of mi xi f i
torque is presented in [31]. While similarities to the model pre-
1
Ii  x
_ i xi  I i  xi t i
sented in this research exist, their model is not based on beam the-
ory. It is implemented in an incremental fashion. This allows large for the ith body with its mass mi, acceleration x i , inertia tensor Ii,
rotations to be accounted for, as they are decomposed into many angular velocity xi, and acceleration x _ i . The force fi and torque ti
small deformation steps, where each step is geometrically linear. are the sums of all forces and torques acting on that body. As basis
In [32], quaternion algebra is used for the calculation of the rel- for the implementation of the bond element, the Pasimodo software
ative rotation of particles. They point out that this method has [24,36] is used. In this program, Eq. (1) are written in the principal
advantages over an incremental approach as used in [31] with re- body frame such that the inertia tensor Ii decouples into the three
spect to numerical stability and accuracy. principal values. The rotatory dynamics is formulated using quater-
Some authors use non-spherical particles to model the bonded nions [36]. This fact is used later in the formulation of the bond ele-
material [33,34]. An advantage of this approach is the elimination ment in Section 2.3, when the relative rotations of particles are also
of the unrealistically high void ratios that are inevitable in case of described by quaternions. In the present study, only spherical par-
spherical particles. The inuence of the relative density on the re- ticles interacting via linear springs and dampers are considered.
sults of DEM simulations has been studied in [2]. It was found that This choice is made in order to keep the numerical overhead low,
the relative density in the real material and in the numerical sam- as our aim is to simulate large systems on an industrial scale at rea-
ple have to be equal in order to yield similar results. sonable computational time.
The aim of the present study is to formulate a numerically ef-
cient three-dimensional model for cemented sands that can repro- 2.2. The unbonded interaction
duce the macroscopic response of this type of material. The
identication of micromechanical effects observed in [25,26] is The contact law used for the interaction of unbonded particles
not addressed. has already been described in [2]. However for completeness, the
The model adopted in the present study approximates the model is summarized in this section.
grains of the granular material by spherical particles and repre- The shear resistance of granular material depends on the shape
sents the cementation bonds by beam elements connecting the of the grains, their size distribution, consolidation state, and the
centers of adjacent spheres. Beam theory is used to derive the friction between grains. For natural soil, it is virtually impossible
strain components from the displaced position of the particles. to transfer all these properties to the DEM model exactly. It is nec-
Failure of the bond occurs when a given equivalent stress level in essary to simplify particle shape while maintaining the relevant
the beam is reached. The idealization of the grains shape and the bulk behavior.
bond are shown in Fig. 3. It has been reported that particle rolling is a major microscopic
In the sequel, the discrete element model is presented. The deformation mechanism, especially for high inter-particle friction
bond element used to model cementation is derived using a beam [37,38]. With spherical particles, the resulting angle of internal

Fig. 3. Idealization of cemented soil by spherical particles connected by beam elements.


302 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

friction is limited to very low values due to excessive rolling of the 2.3. The bond element
particles, and the local friction has a limited effect on the macro-
scopic shear resistance [37]. This situation can be improved by The bond element obeys a linear elastic material law and allows
using non-spherical particles, using compounds of two or more for nite values for displacements and rotations. These properties
spheres [3941], or by introducing more complex contact laws stem from the geometrically exact beam theory [46]. The equiva-
including an additional rolling resistance with an elastic limit lence of the kinematics between continuum and discrete beam
[4244]. In order to obtain reasonable values of the shear resis- model are taken from the work in [47]. It is shown in Appendix
tance, the present study locks the rotational degrees of freedom A that for small rotations the elasticity of the bond element is equal
of the particles when there are unbonded particle interactions to a linear Timoshenko beam element with reduced integration.
[2]. This compensates for more complex shapes of the particles, Like the work in [32], the relative displacements and rotations
which are known to have a major effect on the shear resistance between adjacent particles are decomposed into axial, tangential,
[45]. bending, and twisting components. While in [32], the relative
When particles get in contact, a repulsive force FN and a tangen- deformations are transferred to reaction forces and torques by di-
tial frictional force FT are calculated at the contact point between rectly introducing stiffnesses, our formulation is based on the mod-
particles. In the case of spherical particles, the overlap dij between el of a Timoshenko beam of circular cross-section connecting the
particles i and j can be directly computed from their positions xi spheres centers. From this assumption, we derive a measure for
and xj and radii ri and rj the strains, stresses, and repulsive forces between the particles.
The independent parameters of the beam are the Youngs modulus
dij r i r j  kxi  xj k 2
Eb, the shear modulus Gb, and the cross-section radius rb. This al-
and a contact force is calculated in case of dij > 0. At the angle nor- lows the formulation of a scale-invariant model, as outlined in Sec-
mal to the contact surface, a linear-spring dashpot model is applied tion 3.6.
Using beam theory also gives access to a lot of experience in
F N;ij kN;ij dij dN d_ ij 3 providing efcient numerical implementations [47]. The use of
with the normal stiffness kN and normal damping dN. In contrast to the expressions for the relative particle rotations in Eq. (23), and
the usual assumption of a xed kN for all contacts, we use here a the selection of the reference frame in the midpoint of the beam
stiffness that is calculated from the deformation of an elastic rod in Eq. (9) make the model numerically efcient.
with cross-section A, Youngs modulus Eu, and length l connecting Two particles which are in contact at the beginning of the sim-
the centers of the particles ulation are bonded by a beam element that connects the centers of
the spheres as shown in Fig. 4. The bond element is a beam with
Eu Aij Eu r ij p p
2
circular cross-section of radius rb rb r 1 r 2 =2, where rb is the
kN;ij Eu r ij 4 relative radius of the bond element (usually set to rb 1:0) and
lij 2r ij 2
r1 and r2 are the radii of the bonded particles. A relative radius
with the average radius rij = (ri + rj)/2. This yields different stiffness rb = 1.0 leads to a bond element with the same axial stiffness as
values for contacts between particles of different size [24]. the unbonded interaction in Eq. (4).
It is impossible to capture the real grain size and grain size dis- In the following, orientations are described by unit quaternions
tribution in the numerical model. Thus, the particles have to be q = (cos(u/2)jsin(u/2)a) with the angle of rotation u around the
scaled in the simulation. In Section 3.6, it is shown that the linear axis of rotation a. If a quaternion has one index, it refers to a rota-
springdashpot model in Eqs. (3) and (4) is scale-invariant, i.e., the tion relative to the inertial system, e.g., q1 is the orientation of par-
results of simulations with identical microproperties, but scaled by ticle 1 in the global frame. Quaternions with two indices represent
a constant factor, are equal. This allows to identify the local contact relative rotations, e.g., qB1 is the relative rotation from the bond
parameters for a given material. The particle size can then be element to particle 1. The initial state is further denoted by an in-
adapted to the needs of the simulation. dex 0, e.g., qB1,0.
The tangential force is computed from
F T;ij kT nT;ij  dT v T;ij 5 2.3.1. Initial state
For all particles that should be connected, the necessary infor-
with the tangential stiffness kT being a constant in this model, the mation is collected at the beginning of the simulation in order to
tangential deformation vector nT,ij connecting the last contact point be able to compute the relative displacements between the parti-
with the current contact point in the tangential plane, the tangen- cles in the subsequent simulation steps.
tial damping dT and the relative tangential velocity of the particles
vT,ij. If the tangential force exceeds the limit of FT,ij = lFN,ij, where l
is the local friction coefcient, slipping friction occurs 2.3.1.1. Initial length. To measure the axial strain, the initial length
of the beam has to be stored. This can be either the actual distance
v T;ij of the connected spheres L0 = kx2  x1kor the sum of their radii
F T;ij lF N;ij 6
kv T;ij k

and a new contact point is dened. Thus, l, Eu, kT, damping dN, and
damping dT are the model parameters.
The particles in the unbonded state have their rotations locked,
while in the bonded state the particle rotations are allowed. To en-
able a transition between these two states during the simulation, a
mechanism to transfer bonded particles to unbonded particles is
needed. To this end, the current number of bonds for each particle
is stored during the simulation. As soon as a threshold is reached
for the number of bonds, the particle rotations are locked and
the remaining bonds are removed. An obvious choice for the
threshold number would be zero. However, we used the number Fig. 4. Initial state of bonded particles with the connecting beam element as a bold
one instead, in order to avoid the formation of two-sphere clusters. line.
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 303

L0 = r1 + r2. If particles have an initial overlap, the second option al- Eq. (9) represents a spherical linear interpolation (slerp) be-
lows a pre-stressed structure with a higher dilation to be modeled. tween the two frames at the ends of the bond. The slerp is de-
scribed in [48] and the special case of the interpolation at the
2.3.1.2. Initial orientation. During the simulation, the relative rota- center point in [49]. It would also be possible to set qB equal to
tions of the connected particles have to be computed and trans- either qB1 or qB2. Dening qB the way it is done here has a compu-
ferred into the material conguration. To achieve this, the tational advantage over other options. A relative rotation between
orientations of the particles relative to the bond element in their particles leads also to a shear deformation. This coupling between
initial states are stored. First, a local frame with its x-axis aligned the deformation modes is automatically taken into account for our
with the bond element is constructed from choice of qB, see Appendix A for details.
The calculation of the relative displacements can be divided into
x2  x1
b1 a relative translational deformation as shown in Fig. 5 and a rela-
kx2  x1 k
tive rotation as shown in Fig. 6. The bonding forces resulting from
s  s  b1 b1 both deformation modes are computed separately and are then
b2 7
ks  s  b1 b1 k superposed.
b1  b2
b3
kb1  b2 k 2.3.2.1. Relative translation. First, a translational deformation be-
tween the particles is considered as shown in Fig. 5.
where xi are the particle center positions and s can be any vector as
The deformation of the actual bond element which is aligned
long as it is not co-linear with b1. From the local axis, a rotation ma-
with the vector connecting the sphere centers
trix R = [b1 b2 b3] is constructed and transferred into its quaternion
representation qB,0, see Appendix B for details on this transforma- x12 x2  x1 11
tion. The quaternion qB,0 transforms from the inertial system to
from its current undeformed deformation qB is measured in the
the bond system at the beginning of the simulation.
bond system qB
The quaternions q1;0 and q2;0 are the complex conjugates of the
particles orientation at the beginning of the simulation. The rela- u qB  x12  qB ux uy uz T 12
tive rotations between the bond element and the particles then fol-
The deformation u has three components that correspond to the ac-
low from
tual length of the bond element ux and the shear deformations uy,
qB1;0 q1;0  qB;0 uz. To be consistent with case of unbonded particles, the elongation
8 deformation is computed as ux = kx12k  L0. This differs slightly
qB2;0 q2;0  qB;0
from the above calculation for large deformations. The strains then
and are stored for the rest of the simulation. follow from
1
2.3.2. Current state C u 13
L
At any point in time, the current deformations of the bond ele-
ment, its strains and the resulting contact forces can be computed with the actual length of the bond element L. The repulsive forces
from the relative positions and orientations in the initial congura- between particles are calculated from the axial stiffness EbAb and
tion and the current conguration. The current state of the parti- the shear stiffness asGbAb with Youngs modulus Eb, shear modulus
cles is dened by their positions xi and orientations qi. Gb, the cross-section area Ab and the shear coefcient as
For the further analysis, the current orientation of the bond ele- 2 3
E b Ab
ment qB is dened as 6 7
F s;ax 4 as Gb Ab 5C: 14
qB1 qB2 as Gb Ab
qB 9
jqB1 qB2 j
Additionally, a viscous damping
p
force is calculated. From the
where the current orientations of both ends of the beam element damping ratio D d=2 kmr , with the stiffness k and the reduced
qB1 and qB2 are determined from mass
qB1 q1  qB1;0 m1 m2
10 mr 15
qB2 q2  qB2;0 m1 m2
for each component, the damping force Fd = dV12 is calculated
with the particles orientations q1 and q2 and the relative orienta-
tions of the bond qB1,0 and qB2,0.

Fig. 5. Shear displacement of bonded particles. Fig. 6. Bending deformation of bonded particles.
304 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

2 q 3
Eb Ab
mr Re(q12) ? 1. Note that the singularity is only due to the form of
L0
6 q 7 the scalar part of
6 7
F d 2D6
6
as Gb Ab
L0
m r
7V 12
7 16 arccosn
4 q 5 zn p 26
as Gb Ab
L0
mr 1  n2
with n = Re(q12). In the limit for n ? 1, we get
with the relative velocity vector V12 expressed in the local coordi-
nate system of the bond qB. The use of the initial length L0 instead limzn 1 27
n!1
of the actual length L in Eq. (16) allows to compute the damping
matrix only once at the beginning of the simulation. This reduces without a singularity. Another drawback of the original form of Eq.
the computational time drastically. If more than one bond is con- (25) is the non-effective implementation of the trigonometric func-
nected to the same particle, this denition of the damping matrix tions and the square root. Thus, Eq. (26) is expanded in a Taylor ser-
can lead to overdamping even for D < 1. ies around the undeformed state at n0 = 1
The forces Fs,ax and Fd are transferred into the global coordinate X
1 X
1
1
system to get the forces on the particles f1 and f2 zn zn n0 n  n0 n C n n  n0 n 28
n0
n! n0
f 1 f 2 qB  F s;ax F d  qB : 17
For completeness, the rst six terms of the Taylor expansion are gi-
Further, from the equilibrium of moments around the midpoint of
ven in Table 1. From the relative rotation, the torsion and curvature
the beam
of the bond element are
X 1 1
t x12  f 1 x12  f 2 t 1 t 2 1
2 2 j e 29
!
L
x12  f 1 t 1 t 2 0 18
The torque between the connected particles is then
and by distributing the resulting torque equally on both particles 2 3
2Gb J b
t1 = t2 then follow the torques on both particles resulting from a 6 7
shear deformation
T b;t 4 Eb J b 5j 30
Eb J b
1
t 1 t 2  x12  f 1 19
2 with bending stiffness EbJb and torsional stiffness 2GbJb, where Jb is
the second moment of the cross-section area of the beam. It has
been applied in Eq. (30) that the cross-section of the bond element
2.3.2.2. Relative rotation. When two connected particles change
is circular.
their orientation during the simulation such that their relative ori-
Damping is calculated as before for the translational displace-
entation is different from the initial state, then the bond element is
ments from
bended or twisted. Fig. 6 demonstrates this for the planar case.
2 q 3
The relative rotation between both ends of the beam at any time 2Gb J b
Ir
L0
during the simulation follows from 6 q 7
6 7
q12 qB1 qB2 20 Td 6
6
Eb J b
L0 r
I 7Xrel
7 31
4 q 5
Eb J b
If both particles move together as a rigid body, then q1;0 q1 q2;0 q2 L0 r
I
and Eq. (20) reduces to
with the relative angular velocity vector Xrel in the bond system qB
q12 qB1 qB2 qB1;0 q1 q2 qB2;0 qB;0 q1;0 q1 q2 q2;0 qB;0 and the reduced moment of inertia
qB;0 qB;0 1; 0 21 I1 I2
Ir 32
The calculation of the relative rotation is thus invariant against rigid I1 I 2
body motions of the bonded particles. where I1 and I2 are the moments of inertia of the connected
The current relative orientation of the particles is again a unit particles.
quaternion The torques are converted into the global system and applied as
q12 coshj sinha cosu=2j sinu=2a 22 t1 and t2 on particles 1 and 2, respectively

with the rotation angle u = 2h. The components of the relative rota- t 1 t 2 qB  T b;t T d  qB 33
tion ua follow from
h h 2.3.3. Stresses
e u a 2h a 2 sinha 2 Imq12 23
To model an irreversible cracking of the soil material, the bond
sinh sinh
elements are removed when the stresses exceed a given limit.
The quaternion angle h is obtained from the real part of q12 by
h arccosReq12 24
Table 1
From Eq. (23), then follow the components of the relative rotation First six coefcients for the Taylor series.
arccosReq12 C0 1
e2 Imq12
sinarccosReq12 C1  13
arccosReq12 C2 2
15
2 q
Imq12 25 C3 2
 35
1  Req12 2 C4 8
315
C5 8
 693
For an effective implementation of Eq. (25), it is important to avoid C6 16
3003
numerical problems, when h ? 0 which is equivalent to
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 305

According to beam theory, the stresses in the shearing deformation Table 2


mode and the bending mode are Maximum deection of the cantilever beam.

2 3 Model No. of elements Deection


Eb
rax;s 6 7 Bond element 2 0.312486
4 Gb 5C 34 Bond element 10 0.330992
Gb BEAM188 19 0.332490

2 3
Gb
rt;b 6
4 Eb
7
5jr b 35 appropriate model calibration procedure, e.g., triaxial compression
Eb tests.

Summing the axial components, we obtain


2.4. Verication
q
raxial Eb Cx Eb jy r b 2 Eb jz r b 2 36
In this section, a simple numerical test is used to check the cor-
and for the shearing components rectness of the model and its implementation. Although the test
q has no relation to cemented sands, it is used here, because a refer-
sshear Gb jx rb Gb Cy 2 Gb Cz 2 37 ence solution is available for comparison.
A simple cantilever beam with a circular cross-section of radius
Taking into account that a bonding agent in cemented material is r = 0.1 mm and a length of l = 2 mm under a vertical load of
much more sensitive in tension than in compression, the normal F = 100 N is modeled by a number of particles that are connected
stress raxial is only considered further if it is a tensile stress. The by bond elements. Two adjacent particles are connected by one
equivalent stress is computed following the von Mises criterion bond element. With a Youngs modulus of E = 107 Pa and a Poisson
8 q ratio of m = 0.3, a maximum deection of umax = 0.332490 mm is
>
< r2axial 3s2shear if raxial > 0 calculated with 19 quadratic and geometrically non-linear
reqv q 38 BEAM188 elements in ANSYS as a reference. Fig. 7 shows a solution
>
: 3s2
shear else with 10 particles, connected by bond elements. The relative error
to the reference solution is about 0.5%. The bending stress in the
It was found that the model produces much more realistic results rst bond element next to the xed end is 2.37  105 Pa compared
for different levels of conning pressure if the bond element is to 2.501  105 Pa in ANSYS. Table 2, compares the results and gives
weaker in tension than in compression. This will be called the bond a solution with only two particles that are connected with a single
compressive strength factor ac that is applied to the admissible bond element.
stress in the case of compression. Another modication consists of
introducing a uniformly distributed random number ar = U(ar,min,
2.5. Triaxial compression test
ar,max), where ar,min and ar,max bound ar, and U(a, b) is the uniform
distribution function over the interval [a, b]. The factor ar mimics
The triaxial compression test is a standard test in soil mechanics
a natural distribution for the bond strength. Being a strength reduc-
and it is also applied as a method for the identication of the local
tion factor, the range of ar must have 0 < ar,min and ar,max < 1. At the
parameters in the DEM [5052]. Others have used this test to gain
beginning of the simulation for each bond element, the value of ar is
insight into shear behavior and strain localization in cemented
computed from a random number generator algorithm. The
sands [25,26]. Simulations of triaxial compression tests are used
strength of an individual bond element is obtained by
here to show the effects of the bond elements on the force response

ar ac if Cx < 0 of the material.
ab 39
For the numerical model of the triaxial compression test, some
ar else
authors introduce exible walls to simulate the latex membrane of
The criterion for bond failure is dened by means of the equivalent the physical test. Instead of modeling the membrane itself, this is
stress: done by introducing membrane particles surrounding the speci-
reqv > ab rbreak 40 men. The membrane particles are radially loaded to apply the con-
ning pressure [15]. Others replace the latex membrane by servo
Other failure criteria would be possible and could be easily imple- controlled rigid walls [41,44,51]. This approach is reported to pre-
mented. Values for ac, ar, and rbreak have to be identied by an vent localized effects, e.g., the development of shear bands, while

Fig. 7. Deformation of the cantilever beam.


306 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

100 Table 3
Parameter set A for the simulations of triaxial compression tests without a strength
distribution.
80
Finer by weight [%]

Parameter Value

60 Properties of the bond elements


Youngs modulus of bond Eb = 200 MPa
Shear modulus of bond Gb = 77 MPa
40 Relative radius of bond r b 1:0
Compressive strength factor ac = 10
Uniform random number ar = 1
20
Percentage of bonded particles 100%
Breakage stress of bonds rbreak = 300 kPa
0 Properties of the unbonded model
1 2 3 4
Unbonded Youngs modulus Eu = 800 MPa
Grain diameter [mm] Tangential stiffness kT = 108 N/m
Local friction coefcient l = 0.32
Fig. 8. Particle size distribution in the numerical sample.
Youngs modulus at walls Ew = 200 MPa
Friction coefcient at walls l = 0.0

still faithfully reproducing the macroscopic response of the mate-


rial [53]. As the focus of this study is on the macroscopic behavior
of the material, the latter approach is applied here.
The simulation of the triaxial test is performed in a cubic cell
with a side length of 50mm and rigid walls, which are illustrated
with opaque surfaces in Fig. 9. The lateral walls are controlled to
keep the lateral pressure r3 constant. The contact between parti-
cles and the walls is frictionless. The axial stress r1 is determined
at the top surface.
The initial particle arrangement is created by rst dropping the
particles into the container with no friction between particles.
Then, the particles are inated until a hydrostatic pressure of
p = 100 kPa is reached. The sample contains a total of 1400 parti-
cles. The particle size distribution is shown in Fig. 8.
During the simulation, the top surface moves down at a con-
stant axial strain rate of e_ 0:1, which is small enough to maintain
quasi-static conditions (see Fig. 9).

2.5.1. Number of bonds


Fig. 9. Initial state of a sample of polydisperse particles for the simulation of a The amount of bonds in the initial setup of the material is con-
triaxial compression test.
trolled by setting a given percentage of the particles as cohesive
particles and the remaining particles as non-cohesive particles.
Only between cohesive particles bond elements are created. This
is like having two phases of different materials mixed. Fig. 10
shows a sample of a triaxial test with 50% of cohesive particles.

3. Results and discussion

Triaxial tests are simulated in order to show the effects of the


bonding on the stressstrain behavior of the soil and to demon-
strate the capability of the bond element to capture the behavior
of real cemented sand. For the stressstrain diagrams, the deviator-
ic stress (r1  r3) is plotted against the axial strain.
Two sets of parameters are used throughout the simulations,
where in the rst one all bonds have the same strength and all par-
ticles are bonded. These parameters are listed in Table 3 and will be
referred to as parameter set A. The second set of parameters is char-
acterized by a distribution of strength in the bonds. The random fac-
tor ar lies in the interval [0.01, 1]. Parameter set B is shown in Table
4 and is also used to reproduce the laboratory tests in Section 3.8.
For each simulation in this section, it is stated which set of
parameters is applied and, if applicable, what modications are
additionally introduced.

3.1. Bonding effects

Fig. 10. A sample for triaxial tests consisting of two phases: cohesive particles Fig. 11 shows stressstrain curves for simulations of triaxial
(dark) and non-cohesive particles (opaque). Here, 50% of the particles are cohesive. tests with and without bond elements representing cemented
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 307

Table 4 800
Parameter set B for the simulations of triaxial compression tests with a strength

axial stress difference [kPa]


distribution. 3=100 kPa
Parameter Value 600
3=200 kPa
Properties of the bond elements
Youngs modulus of bond Eb = 200 MPa
Shear modulus of bond Gb = 77 MPa
400
Relative radius of bond r b 1:0 3=50 kPa
Compressive strength factor ac = 4
Uniform random number ar = [0.01 . . . 1] 200
Percentage of bonded particles for CC = 0% 0%
Percentage of bonded particles for CC = 2% 25%
Percentage of bonded particles for CC = 4% 45% 0
Breakage stress of bonds rbreak = 1700 kPa 0 1 2 3 4 5
Properties of the unbonded model 1 [%]
Unbonded Youngs modulus Eu = 800 MPa
Tangential stiffness kT = 108 N/m Fig. 12. Axial stress difference between cemented and uncemented material vs.
Local friction coefcient l = 0.32 axial strain for r3 = 50, 100 and 200 kPa.
Youngs modulus at walls Ew = 200 MPa
Friction coefcient at walls l = 0.0

3000
and uncemented materials, respectively. The simulations are based
on parameter set A with a bond strength of rbreak = 100 kPa and =600kPa

(13)/2 [kPa]
break
lateral pressures of r3 = 50, 100, and 200 kPa. The reduced break-
2000
age stress is used for a better visualization of the results. The same break=300kPa
behavior is achieved for higher strength of the bond elements, e.g.
for rbreak = 300 and 600 kPa in Fig. 14. The cemented material has a
peak strength at low strains and then converges to the stress 1000
strain curve for the uncemented material. This behavior is also
reported in [11] for real cemented sands. In contrast to the peak state
measurements, the breakage of the bonds and the associated loss residual state
of shear strength appears immediately in the simulation. 0
0 1000 2000 3000 4000
When the stressstrain curve of the uncemented material is
(1+3)/2 [kPa]
subtracted from the curve of the cemented material, only the addi-
tive strength of the bonding agent is remaining. The contribution of
Fig. 13. Peak strength and residual strength for different values of the breakage
the bond strength to the overall strength is equal for the different stress rbreak and parameter set A.
lateral pressures, see Fig. 12. This effect is also observed in the
laboratory tests in [11].

3.2. Bond-element breakage stress 4000


unbonded
=100kPa
break
For a series of simulations with parameter set A from Table 3 =300kPa
3000 break
and two different bond strengths rbreak of 300 kPa and 600 kPa,

13 [kPa]

=600kPa
the peak and residual stresses at different lateral pressures r3 of break
50, 100, 200, 400, and 600 kPa are computed. Fig. 13 shows the 2000
peak strength and the residual strength for these simulations. A
higher breakage stress of the bond rbreak increases the peak stress
of the material. The residual state of the material is unaffected by 1000

1200 0
=200 kPa cemented 0 2 4 6 8 10
3
1000 uncemented [%]
1

3=100 kPa Fig. 14. Deviatoric stress (r1  r3) vs. axial strain for simulated triaxial tests with
1 3 [kPa]

800
r3 = 100 kPa of unbonded material (dashed line) and bonded material with
600 breakage strengths of rbreak = 100, 300, and 600 kPa (solid lines) for parameter set
A.
=50 kPa
3
400
the higher peak shear stresses, i.e., the residual state and the re-
200 sponse of the uncemented material are equal. The angle of internal
friction at peak shear stress /peak is the same as for the residual
0 state /res. The effect of the breakage stress is a shift in the failure
0 1 2 3 4 5
envelope or, in other words, a cohesive intercept c. The stress
1 [%]
strain curves for r3 = 100 kPa are given in Fig. 14. As before, the
Fig. 11. Deviatoric stress (r1  r3) vs. axial strain for simulated triaxial tests with
material response is approximately linear elastic until the peak
r3 = 50, 100 and 200 kPa of uncemented material (dashed lines) and cemented state. After the peak, the stressstrain curves converge to the unce-
material (solid lines) for parameter set A. mented curves.
308 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

3.3. Compressive strength factor 1000

Tension tests on cemented sand have shown a much lower 800


strength in tension than in compression [14]. This is due to the fact

13 [kPa]
that, in compression, the load is partially carried by inter-grain
600
contacts, while in tension the load has to be supported by the
bonding agent only. In order to mimic this behavior, a compressive
strength factor ac was introduced in the bond element, see Eq. (39). 400
The results of our own simulations show that the effect of ac is not
only important for the difference between tension and compres- 200
sion. It also affects the peak strength of the material for different
conning pressures. For low values of ac, the bonds can break 0
0 2 4 6 8 10
either in tension or compression. High values of ac lead to failure
only in case of local tension. The tension is dominated by the devi- [%]
1
atoric stress in the sample. The parameter ac alters the peak
Fig. 16. Deviatoric stress (r1  r3) vs. axial strain for simulated triaxial tests with
strength of the material, as it switches between failure modes.
different values of the bond stiffness Eb and parameter set B.
This effect is investigated for a series of simulations with
parameter set A from Table 3 and different values of the compres-
sive strength factor ac = 2, 4, and 10. The peak and residual stresses
at different lateral pressures of r3 = 50, 100, 200, 400, and 600 kPa and 200 MPa. All other parameters are from parameter set B in Ta-
are computed. Results of these simulations are shown in Fig. 15. ble 4. As for the unbonded model, a higher bond stiffness leads to a
Higher ac makes the material insensitive to hydrostatic pressure. higher stiffness of the sample. It also affects the post peak behavior.
The angle of internal friction / remains unchanged between the Lower values of the stiffness cause a more ductile behavior of the
peak state and the residual state (curve for ac = 10). Notice that sample. Then, the decrease of the deviatoric stress is slower but
the residual state of the material again is unaffected by the bond- also converges to the residual state. There is neither an effect of
ing. The bonding inuences the cohesion of the material but not the bond stiffness on the shear-stress at peak nor on the residual
the angle of internal friction for high values of ac. This behavior shear strength of the material. Stressstrain curves for different
is also reported in [11]. settings of Eb are shown in Fig. 16.
For low values of ac, the material fails not only in tension, but
also in compression. The higher the conning pressure, the more
3.5. Bond-element cross-section
bonds break in this way and the peak shear strength converges
to the residual state, see Fig. 15.
The bond element is a beam of diameter rb rb r1 r2 =2. The
The strength of the bond element together with the compres-
higher the relative radius r b , the higher the stiffness of the bond.
sive strength factor allows to t the model to MohrCoulombs
The bending stiffness grows faster with the radius than the axial
frictional law
stiffness. The effect of r b on the stressstrain curve is investigated
sf ccm rn tan/ 41 by simulations with parameter set B from Table 4 and different val-
ues of the relative radius rb 0:5, 0.75, 1.0, and 1.5. Due to the
for cemented soils, where ccm is called the cementation strength compressive strength factor ac = 4 and the insensitivity of the bond
and / is the apparent friction angle. to tensile stress according to Eq. (38), the bonds break in tension
rather than in compression. If bonds fail in compression, it has to
3.4. Bond-element stiffness be taken into account that a relative radius r b 1 may cause differ-
ent axial stiffnesses in the bond element and in the unbonded
In case of the uncemented material, the normal stiffness of the model, causing a non-smooth transition between both models.
unbonded particle contact also have an impact on the initial stiff- Figs. 17 and 18 show stressstrain and volumetric curves for these
ness of the sample [2], but the peak stress remains the same. The simulations, respectively. Higher values of rb cause higher peak
inuence of the bond stiffness is investigated by comparing simu- stress. Also the volumetric curve is altered by the radius of the
lations with different values of the Youngs modulus Eb = 50, 100, bonds. The larger rb is, the more dilation.

2000 1800
rb=0.5
c =10
rb=0.75
1500
(13)/2 [kPa]

rb=1.0
1200 rb=1.5
3

1000
c =4
1

600
500
peak state
c =2
residual state
0 0
0 500 1000 1500 2000 2500 3000 0 2 4 6 8
(1+3)/2 [kPa] 1 [%]

Fig. 15. Peak strength and residual strength for different compressive strength Fig. 17. Deviatoric stress (r1  r3) vs. axial strain for triaxial tests with different
factors ac for parameter set A. relative radius r b and parameter set B.
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 309

8 8
r =0.5 r=0.003m
b
r =0.75 r=0.03m
b
6 6 r=0.3m
r =1.0
b
rb=1.5
v [%]

v [%]
4

2 2

0 0
0 1 2 3 4 5 6 7 8 0 2 4 6 8 10
1 [%] 1 [%]

Fig. 18. Volumetric strain vs. axial strain for the triaxial tests with different values Fig. 20. Volumetric strain vs. axial strain for the triaxial tests with scaled models of
of relative radius r b and parameter set B. cemented material with particle radii of r = 0.003, 0.03, and 0.3 m.

values are kept unchanged. Figs. 19 and 20 show the stressstrain


1200 and volumetric curves for the original and the scaled models, con-
r=0.003m
r=0.03m rming the above mentioned scale invariance numerically.
r=0.3m
800 3.7. Number of bonds
3

The parameters for the bond model have to be identied by trial


1

and error. For a given peak shear stress of the real cemented sand,
400
one could nd different combinations of breakage stress and num-
ber of initially bonded particles that yield a given peak stress. A
parametric study is conducted as follows: After setting the number
0 of bonds, the breakage stress rbreak is adjusted to reach the dened
0 2 4 6 8 10 peak stress. Parameter set B in Table 4 is used. The properties for
1 [%] the three samples are: (a) 100% bonded particles, rbreak = 3.7 
105 Pa, (b) 45% bonded particles, rbreak = 1.1  106 Pa, (c) 25%
Fig. 19. Deviatoric stress (r1  r3) vs. axial strain for simulated triaxial tests with bonded particles, rbreak = 2.7  106 Pa. Fig. 21 shows the resulting
scaled models of cemented material with particle radii of r = 0.003, 0.03, and 0.3 m. stressstrain curves of these models.
The more bond elements in a sample, the higher the initial stiff-
ness. At post peak, the curve for the least number of bonds and the
3.6. Particle scaling
highest bond strength decreases slower than the others.
For natural materials, a slower decrease of the stressstrain
In the numerical sample, the real grain size and grain size dis-
curve after failure is caused by a higher dilation of the material
tribution cannot be approximated due to inherent numerical limi-
[11,25]. The same occurs in the simulation. In case of a higher bond
tations. Thus particles have to be scaled by some factor in the
strength (curve c), more bonds survive the yielding of the material,
simulation. If the numerical model is robust with respect to such
see Fig. 22. The remaining clumps of spheres hinder the rearrange-
scaling, the identication of the model parameters will be valid
ment of the particles thus causing a higher dilation of the material
for different particle sizes, i.e., the particle size can be modied
as shown in Fig. 23.
during the analysis without the need to identify a new set of local
Bonds start breaking from the beginning of the deformation,
contact parameters.
but after the peak state, their number decreases strongly. At the
According to [54], a contact law of the general form
a
F kR db ; 42
with stiffness k, particle radius R, and overlap d, is said to be scale
invariant for three-dimensional problems, if 600 c
a b 2: 43 b
[kPa]

While this condition is fullled for Hertzian contacts, with a = 1/2 400
and b = 3/2, the usual form of the linear springdashpot model with
3

a constant stiffness kN is not scale invariant. In contrast to that, the


1

linear model used here for unbonded particles from Eqs. (3) and (4),
200 a
is scale invariant, since a = 1 and b = 1.
The bond element is formulated in terms of local stresses and
strains and thus is also expected to be scale invariant. In order to
verify this, three simulations with identical parameters from the 0
0 2 4 6 8
parameter set B in Table 4 and a lateral pressure of r3 = 100 kPa 1 [%]
are performed with scaling factors of 1, 10, and 100, respectively.
All geometric quantities are scaled by this factor, i.e., the particle Fig. 21. Deviatoric stress (r1  r3) vs. axial strain for triaxial tests with different
radii, their positions, and the dimensions of the container. All other numbers of bonds but same peak shear-stress.
310 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

5000 900
CC=4% 3=103kPa
number of intact bonds

4000
a

13 [kPa]
600
3000 CC=2%

2000
b 300
c
1000

CC=0%
0 0
0 2 4 6 8 0 2 4 6 8 10
1 [%]
1 [%]
Fig. 22. Number of intact bonds vs. axial strain for the triaxial tests with different
Fig. 24. Deviatoric stress (r1  r3) vs. axial strain for simulations (solid lines) and
numbers of bonded particles.
measurements (dashed lines) of triaxial tests at a lateral pressure of r3 = 103 kPa
and cement contents of CC = 0%, 2%, and 4%.

4
1200
3
3=414 kPa
[%]

c
2
13 [kPa]
b
v

800 3=207 kPa


1 a

0
400
1
0 2 4 6 8
[%] 3=103 kPa
1
0
Fig. 23. Volumetric strain vs. axial strain for the triaxial tests with different 0 2 4 6 8 10
numbers of bonded particles. 1 [%]

Fig. 25. Deviatoric stress (r1  r3) vs. axial strain for simulations (solid lines) and
residual state, the few remaining bonds do not contribute to the measurements (dashed lines) of triaxial tests on material with a cement content
overall strength of the material any more, see Fig. 22. CC = 2% at lateral pressures r3 = 103, 207, and 414 kPa.

3.8. Comparison to laboratory tests

In this section, the results of simulated triaxial compression 1800


tests are shown and compared to laboratory tests from [11] for dif-
ferent conning pressures (r3 = 35,103 and 414 kPa) and cement
contents (CC = 0%, 2% and 4%). The aim is to demonstrate the ability 3=414kPa
13 [kPa]

of the bond element to reproduce realistic stressstrain behavior of 1200


cemented sand.
The local contact parameters that are used to reproduce the 3=207kPa
measured data are listed in Table 4. For each cement content a
600
different set of local contact parameters could be identied. It is
3=103kPa
chosen here, to change only the number of bonds to attain the
stressstrain curves of the different cement contents. The parame-
3=35kPa
ters were identied by trial and error and are not necessarily 0
unique. Other combinations of breakage stress and the number 0 2 4 6 8 10
of bonds etc. could be found with the same peak shear strength 1 [%]
and residual strength.
Fig. 24 shows the results of simulations and measurements for Fig. 26. Deviatoric stress (r1  r3) vs. axial strain for simulations (solid lines) and
measurements (dashed lines) of triaxial tests on material with a cement content
different cement contents but the same lateral pressure of
CC = 4% at lateral pressures r3 = 35, 103, 207, and 414 kPa.
r3 = 103 kPa. With increasing cement content, the peak strength
increases, while the residual strength remains unchanged.
Figs. 25 and 26 show the results of triaxial tests with cement the gradual drop of stressstrain curve at post-peak. It was found
contents of CC = 2% and CC = 4%, respectively. The results at differ- in [15] that strain softening at post-peak is associated with the vol-
ent conning pressures are shown and compared to the corre- umetric work caused by shear banding. The simulations here do
sponding measurements. not develop shear bands and, as a consequence, the drop of the
While all simulations very well reproduce the peak and the shear stress after the peak is steeper than in the measurements.
residual strength of the laboratory tests, they fail to predict Possible reasons for the absence of strain localization and shear
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 311

It is widely used for wave propagation problems and for beam


bending of moderately slender beams.
Reduced integration of the shear deformation is important to
avoid shear locking. Instead of using the discretized rotation DOF
u = N1u1 + N2u2 (where Ni are the shape functions and ui are the
rotations at node i), only a constant ansatz is used
1
c @ x w  u @ x w  u1 u2 A:1
2
The stiffness matrix of this element is compared to the bending part
Fig. 27. Shear deformation from relative rotations. of the bond element in order to demonstrate that both elements are
identical for small rotations. For simplicity, the 2D case is used in
bands are the rigid boundaries and the small number of particles in this section. For a static analysis, the solution is given by means
the simulation. of the nite element procedure
Kw f A:2
4. Conclusions
with the nodal displacements w and forces f
In this paper, a numerical model for the bonding of particles in
w w1 w2 u1 u2 T f f1 f 2 t 1 t2 T : A:3
the DEM is presented. It adopts Timoshenkos beam theory and is
shown to be equivalent to a linear beam element for small deforma- In [55], the element stiffness matrix K is given by
tions. The reference frame at the midpoint of the beam element, " #
K ww K wu
and the efcient method to compute the relative rotations of the K A:4
particles lead to a numerically efcient model. The model is scale K Twu K uu
invariant, i.e., scaling of all lengths by a constant factor does not
alter the macroscopic material response. The use of absolute dis- with
placements and rotations, in contrast to an incremental formula-
   
Eb J b 1 1 as Gb Ab 1 1
tion, and the use of quaternion algebra causes the model to be K uu K uu1 K uu2 A:5
L 1 1 4 1 1
numerically robust, even with large deformations and timesteps.
The model parameters are the strengths and stiffnesses of the  
as Gb Ab 1 1
beam elements. An inverse calibration procedure is needed to nd K ww A:6
L 1 1
a valid set of model parameters for a given cemented sand. For this
purpose, the triaxial compression test may be used. Local parame-  
ters are varied by trial and error, until the peak and residual
as Gb Ab 1 1
K wu A:7
strength, initial stiffness and dilation are close to the reference. 2 1 1
The result of such a parameterization is not necessarily unique, For a cantilever beam, the displacement and rotation at the xed
i.e., different sets of local parameters may exist that allow the cap- end are zero, i.e. w1 = 0 and u1 = 0. Expanding Eq. (A.2) for this case
ture of material properties. gives
Applying the model to triaxial compression tests in cemented 2 3
sand shows that: 2 3  as GLb Ab as Gb Ab
f1 2
6 7 
6 f 7 6 as Gb Ab  as G2b Ab 7 w
 The bonding adds an additional shear strength to the material
6 2 7 6 L 7 2
6
7 6 as Gb Ab L 7
A:8
4 m1 5 6  as Gb Ab  L 4 7
Eb J b u
with a peak in the stressstrain curve at low strains. 4 2 5 2
 At critical state, the stressstrain curve attains that of its unce- m2  as G2b Ab Eb J b
as Gb Ab L
L 4
mented counterpart.
 The post-peak variation with strain depends on dilation. The terms (asGbAb)/L and (EbJb)/L in Eq. (A.8) refer to the pure
 Dilation is higher if some bonds remain intact during shearing. shearing and bending stiffnesses in Kww and Kuu1. Note that
 The compressive strength factor controls the inuence of the f1 = f2. The moment from the shearing forces f1 and f2 are
conning pressure on the strength envelope. t1 = (f2L)/2 and t2 = (f1L)/2. These torques are the terms (asGbAb)/
 The model is capable of reproducing the stressstrain curves of 2 and (asGbAbL)/4 in Eq. (A.8). They have the same sign and sum
laboratory tests at different degrees of cementation. up balancing the system. So far, all terms appear also in the bond
 Besides the properties of the bond element, the initial number element. The remaining terms (asGbAb)/2 will be examined in
of bonds of the material is found to have an important impact more detail now. In case of a pure rotational deformation
on the results. (w1 = w2 = u1 = 0, u2 0), i.e. when particle 2 is rotated relative
to particle 1 then these terms cause a shearing force, as shown
in Fig. 27. In the bond element, the shear deformation C is mea-
Acknowledgments sured in the bond frame qB which is rotated by u2/2. The shear
deformation is due to rotation of the particles is then
Special thanks go to Holger Lang and Joachim Linn for sharing
their profound knowledge of beam theory. This work was funded L
w1 w2 sin u2 A:9
by the Fraunhofer Innovation Cluster Digital Engineering for Com- 2
mercial Vehicles. For small rotations sin u2  u2. The reactions are thus
as Gb Ab L a s G b Ab
Appendix A. Equivalence to a Timoshenko beam f1 f2 u2 u2 A:10
L 2 2
In FEM, the linear Timoshenko beam element with reduced which are exactly the upper right terms in Eq. (A.8). For small rota-
integration of the shear deformation is computationally efcient. tions, the bond element is equivalent to a linear Timoshenko beam.
312 M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313

As mentioned before, due to the choice of qB, this coupling of bend- [15] Wang Y-H, Leung S-C. A particulate-scale investigation of cemented sand
behavior. Canadian Geotech J 2008;45:2944.
ing and shear is automatically accounted for.
[16] Vrettos C. Shear strength investigations for a class of extra-terrestrial analogue
soils. J Geotech Geoenviron Eng 2011;1:375.
Appendix B. Transformation between a quaternion and a [17] Lade PV, Overton DD. Cementation effects in frictional materials. J Geotech Eng
1989;115:137387.
rotation matrix [18] Cundall PA, Strack ODL. A discrete numerical model for granular assemblies.
Gotechnique 1979;29:4765.
A rotation can be described in different ways. Here, unit quater- [19] Tran V, Donz F-V, Marin P. A discrete element model of concrete under high
triaxial loading. Cem Concr Compos 2011;33:93648.
nions of the form [20] Camborde F, Mariotti C, Donz F-V. Numerical study of rock and concrete
behaviour by discrete element modelling. Comput Geotech 2000;27:22547.
q cos hj sin ha q0 q1 q2 q3 B:1 [21] Hentz S, Donz F-V, Daudeville L. Discrete element modelling of concrete
submitted to dynamic loading at high strain rates. Comput Struct
are used, with the axis of rotation a, the quaternion angle h, and the 2004;82:250924.
angle of rotation u = 2h. A quaternion can be converted into a rota- [22] Abbas A. Simulation of the micromechanical behavior of asphalt mixtures
tion matrix that describes the same rotation by [48] using the discrete element method. Ph.D. thesis, Washington State University;
2004.
2 3 [23] Buttlar W, You Z. Discrete element modeling of asphalt concrete: Microfabric
1  2q22  2q23 2q1 q2 2q0 q3 2q1 q3  2q0 q2 approach. Transport Res Record: J Transport Res Board 2001;7571:1118.
6 7
R 4 2q1 q2  2q0 q3 1  2q21  2q23 2q2 q3 2q0 q1 5 B:2 [24] Ergenzinger C, Seifried R, Eberhard P. A discrete element model to describe
failure of strong rock in uniaxial compression. Granul Matter 2010;13:124.
2q1 q3 2q0 q2 2q2 q3  2q0 q1 1  2q21  2q22 [25] Wang Y, Leung S. Characterization of cemented sand by experimental and
numerical investigations. J Geotech Geoenviron Eng 2008;134:992.
For the inverse conversion from the rotation matrix to a quaternion, [26] Jiang M, Yan H, Zhu H, Utili S. Modeling shear behavior and strain localization
in cemented sands by two-dimensional distinct element method analyses.
the trace of the rotation matrix t = R11 + R22 + R33 is used [56]
Comput Geotech 2011;38:1429.
p [27] Utili S, Nova R. Dem analysis of bonded granular geomaterials. Int J Numer
t1 R32  R23 Anal Methods Geomech 2008;32:19972031.
q0 q1 p B:3 [28] Vatsala A, Nova R, Murthy BRS. Elastoplastic model for cemented soils. J
2 2 t1 Geotech Geoenviron Eng 2001;127:67987.
R13  R31 R21  R12 [29] Jiang M, Yu H-S, Leroueil S. A simple and efcient approach to capturing
q2 p q3 p bonding effect in naturally microstructured sands by discrete element method.
2 t1 2 t1
Int J Numer Methods Eng 2007;69:115893.
[30] Rojek J, Oate E, Labra C, Kargl H. Discrete element simulation of rock cutting.
This algorithm may fail in case negative values of t. Then an alterna- Int J Rock Mech Min Sci 2011;48:9961010.
tive formulation is given by [31] Potyondy D, Cundall P. A bonded-particle model for rock. Int J Rock Mech Min
Sci 2004;41:132964.
t Rkj  Rjk [32] Wang Y, Alonso-Marroquin F. A nite deformation method for discrete
qi q0 B:4 modeling: particle rotation and parameter calibration. Granul Matter
2 2t 2009;11:33143.
Rji Rij Rki Rik [33] Galindo-Torres S, Pedroso D, Williams D, Li L. Breaking processes in three-
qj qk dimensional bonded granular materials with general shapes. Comput Phys
2t 2t
Commun 2012;183:26677.
p [34] DAddetta GA, Kun F, Ramm E. On the application of a discrete model to the
with t Rii  Rjj  Rkk 1 where i is the index of the largest diag-
fracture process of cohesive granular materials. Granul Matter 2002;4:7790.
onal entry in R and the other indices are [35] Schiehlen W, Eberhard P. Technische Dynamik: Modelle fr Regelung und
Simulation (in German). B.G. Teubner, Wiesbaden; 2004.
j i mod 3 1; and k j mod 3 1 B:5 [36] Fleissner F. Parallel object oriented simulation with lagrangian particle
methods. Shaker, Aachen; 2010.
[37] Oda M, Konishi J, Nemat-Nasser S. Experimental micromechanical evaluation
respectively.
of strength of granular materials: effects of particle rolling. Mech Mater
1982;1:26983.
References [38] Bardet J. Observations on the effects of particle rotations on the failure of
idealized granular materials. Mech Mater 1994;18:15982.
[39] Lu Y, Frost D. Three-dimensional DEM modeling of triaxial compression of
[1] Shmulevich I, Asaf Z, Rubinstein D. Interaction between soil and a wide cutting
sands. In: Soil behavior and geo-micromechanics (GSP 200), ASCE. 2010; p.
blade using the discrete element method. Soil Tillage Res 2007;97:3750.
2206.
[2] Obermayr M, Dressler K, Vrettos C, Eberhard P. Prediction of draft forces in
[40] Ng T-T. Particle shape effect on macro- and micro-behaviors of monodisperse
cohesionless soil with the discrete element method. J Terramechanics
ellipsoids. Int J Numer Anal Methods Geomech 2009;33:51127.
2011;48:34758.
[41] Salot C, Gotteland P, Villard P. Inuence of relative density on granular
[3] A. Oida, M. Momozu, Simulation of soil behavior and reaction by machine part
materials behavior: DEM simulations of triaxial tests. Granul Matter
by means of DEM, in: International commission of agricultural engineering IV.
2009;11:22136.
Manuscript PM 01 004; 2002.
[42] Iwashita K, Oda M. Rolling resistance at contacts in simulation of shear band
[4] Nezami EG, Hashash YMA, Zhao D, Ghaboussi J. Simulation of front end loader
development by DEM. J Eng Mech 1998;124:28592.
bucketsoil interaction using discrete element method. Int J Numer Anal
[43] Jiang M, Yu H-S, Harris D. A novel discrete model for granular material
Methods Geomech 2007;31:114762.
incorporating rolling resistance. Comput Geotech 2005;32:34057.
[5] Coetzee C, Els D. The numerical modelling of excavator bucket lling using
[44] Plassiard J-P, Belheine N, Donz F-V. A spherical discrete element model:
DEM. J Terramechanics 2009;46:21727.
calibration procedure and incremental response. Granul Matter
[6] Coetzee C, Els D, Dymond G. Discrete element parameter calibration and the
2009;11:293306.
modelling of dragline bucket lling. J Terramechanics 2010;47:3344.
[45] Cavarretta I, Coop M, OSullivan C. The inuence of particle characteristics on
[7] Kolymbas D. Geotechnik: Bodenmechanik, Grundbau und Tunnelbau (in
the behaviour of coarse grained soils. Gotechnique 2010;60:41323.
German). Heidelberg: Springer; 2011.
[46] Simo J. A nite strain beam formulation. the three-dimensional dynamic
[8] Budhu M. Soil mechanics and foundations. New York: John Wiley & Sons;
problem. Part I. Comput Methods Appl Mech Eng 1985;49:5570.
2007.
[47] Lang H, Linn J, Arnold M. Multi-body dynamics simulation of geometrically
[9] Coop MR, Atkinson JH. The mechanics of cemented carbonate sands.
exact Cosserat rods. Multibody Syst Dyn 2011;25:285312.
Gotechnique 1993;43:5367.
[48] Shoemake K. Animating rotation with quaternion curves. ACM SIGGRAPH
[10] Collins B, Sitar N. Geotechnical properties of cemented sands in steep slopes. J
Comput Graph 1985;19:24554.
Geotech Geoenviron Eng 2009;135:135966.
[49] Parent R. Computer animation: algorithms and techniques. San
[11] Clough G, Rad N, Bachus R, Sitar N. Cemented sands under static loading. J
Francisco: Morgan Kaufman Publishers; 2002.
Geotech Eng Divis 1981;107:799817.
[50] Cui L, OSullivan C, ONeill S. An analysis of the triaxial apparatus using a mixed
[12] Airey DW. Triaxial testing of naturally cemented carbonate soil. J Geotech Eng
boundary three-dimensional discrete element model. Gotechnique
1993;119:137998.
2007;57:83144.
[13] Schnaid F, Prietto P, Consoli N. Characterization of cemented sand in triaxial
[51] Belheine N, Plassiard J-P, Donz F-V, Darve F, Seridi A. Numerical simulation of
compression. J Geotech Geoenviron Eng 2001;127:85768.
drained triaxial test using 3D discrete element modeling. Comput Geotech
[14] Das BM, Dass RN. Lightly cemented sand in tension and compression. Geotech
2009;36:32031.
Geol Eng 1995;13:16977.
M. Obermayr et al. / Computers and Geotechnics 49 (2013) 299313 313

[52] Hanley KJ, OSullivan C, Oliveira JC, Cronin K, Byrne EP. Application of Taguchi [55] Cook R, Malkus D, Plesha M, Witt R. Concepts and applications of nite
methods to DEM calibration of bonded agglomerates. Powder Technol element analysis. New York: John Wiley & Sons; 2002.
2011;210:23040. [56] Shoemake K. Quaternion calculus and fast animation. SIGGRAPH Course Notes
[53] OSullivan C. Particulate discrete element modelling: a geomechanics 1987;10:10121.
perspective. Taylor & Francis; 2011.
[54] Feng Y, Han K, Owen D, Loughran J. On upscaling of discrete element models:
similarity principles. Eng Comput 2009;26:599609.

View publication stats

You might also like