You are on page 1of 24

How do interparticle contact friction, packing density and degree of polydispersity affect force

propagation in particulate assemblies?


This article has been downloaded from IOPscience. Please scroll down to see the full text article.
J. Stat. Mech. (2006) P09003
(http://iopscience.iop.org/1742-5468/2006/09/P09003)
Download details:
IP Address: 150.216.68.200
The article was downloaded on 05/09/2013 at 08:03
Please note that terms and conditions apply.
View the table of contents for this issue, or go to the journal homepage for more
Home Search Collections Journals About Contact us My IOPscience
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
ournal of Statistical Mechanics:
An IOP and SISSA journal
J
Theory and Experiment
How do interparticle contact friction,
packing density and degree of
polydispersity aect force propagation
in particulate assemblies?
Maya Muthuswamy and Antoinette Tordesillas
1
Department of Mathematics and Statistics, University of Melbourne 3010,
Australia
E-mail: m.muthuswamy@ms.unimelb.edu.au and atordesi@ms.unimelb.edu.au
Received 7 April 2006
Accepted 13 August 2006
Published 7 September 2006
Online at stacks.iop.org/JSTAT/2006/P09003
doi:10.1088/1742-5468/2006/09/P09003
Abstract. Interparticle contact friction, packing density and polydispersity
are known to be major contributors to the macroscopic strength of particulate
assemblies, that is, their bulk resistance to deformation. For example, when a
solid object penetrates a particulate material, the penetration resistance (i.e. the
force opposing the object) increases concomitantly with an increase in the
degree of polydispersity, packing density and interparticle friction. To establish
the underlying mechanisms by which these properties govern the macroscopic
response, we characterize quantitatively force propagation at length scales beyond
that of the interparticle contact region. Using data derived from discrete element
simulations of a two-dimensional granular assembly subject to indentation by a
rigid at punch, we examine the properties of force chains and the force chain
network as they evolve during the course of the deformation. Findings indicate
that increasing interparticle friction, packing density and degree of polydispersity
promotes the formation of straighter chains and a greater degree of branching
in the force chain network. Although the force chain length appears to be
independent of friction and polydispersity, on average, denser systems tend to
favour shorter chains. Thus, straighter and shorter force chains, combined with a
greater degree of branching in the force chain network, result in a macroscopically
stronger granular material.
Keywords: granular matter
1
Author to whom any correspondence should be addressed.
c 2006 IOP Publishing Ltd and SISSA 1742-5468/06/P09003+23$30.00
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Contents
1. Introduction 2
2. Numerical experiment 5
2.1. Contact model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2. Creation of granular specimen of various packing densities . . . . . . . . . 6
2.3. Justication of parameters chosen . . . . . . . . . . . . . . . . . . . . . . . 7
3. Quantication of force chain and force chain network properties 7
3.1. Force chain length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2. Force chain load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3. Force chain curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4. Force chain network branching properties . . . . . . . . . . . . . . . . . . . 9
4. Results and discussion 9
4.1. Loadindentation: macroscale . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2. Mesoscale: force chain and force chain network properties . . . . . . . . . . 12
5. Conclusions 18
6. Acknowledgements 21
References 21
1. Introduction
It is now well known from photoelastic experiments that forces in a granular material are
propagated along chain-like particle groups called force chains (e.g. [1]). This complex
network of chains is highly ramied, and undergoes rapid changes in branch morphology
as deformation proceeds. However, propagation of stress or force along such discrete paths
is not unique to granular systems. In jammed colloids, for example, jamming occurs due
to force chains that form along the compressional direction, giving rise to fundamentally
dierent mechanical behaviour from that displayed in standard solids or uids [2][4].
Similarly, force networks formed in 2D amorphous glassy materials are strongly
inhomogeneous, consisting of regions of weak attractive forces embedded within a strong
skeleton of repulsive forces [5]. In supercooled glass-forming liquids, the presence of force
chains has been linked to local particle motion along string-like paths [6]. More recently,
experimental measurements of contact forces inside three-dimensional piles of frictionless
liquid droplets have revealed long-range chain-like correlations of large forces, the length
of which is approximately ten particle diameters [7]. While the underlying mechanisms
that govern the formation and evolution of such force structures are recognizably dierent
from one material to the next, a key challenge that confronts the modeling of all complex
media is the representation of such highly heterogeneous structures within the connes of
continuum theory [8].
For granular materials, there is now abundant experimental evidence that the spatial
and temporal evolution of these force structures on the mesoscale govern macroscopic
doi:10.1088/1742-5468/2006/09/P09003 2
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
behaviour (see, for example, [9][11], [1], [12][14]). In particular, discrete element
analyses and photoelastic disc experiments have shown that bulk instabilities are induced
by the collapse of jammed clusters developed at multiple length scales within the
mesodomain [15, 10, 9]. The unjamming and jamming transitions of such particle clusters
are inextricably linked to the continual collapse of old, and formation of new, force chains.
A direct eect of unjamming of particle clusters is nonane motiona factor that is
neglected in the majority of existing micromechanical continuum models today [16, 8].
Constitutive formulations which neglect the eects of nonane motion overpredict
the stability of the material, consequently limiting their capacity to capture dening
behaviour, especially strain-softening and associated instabilities like shear bands.
Clearly, to extend the practical applicability of existing modeling toolsboth from
the perspective of the discrete element method and continuum theorytechniques must
be developed that go beyond the qualitative and subjective descriptions of mesoscale force
structures. In a recent paper [17], we provided a brief literature review of past studies on
force chains in an eort to highlight this important research need. Within this context,
we proposed a technique by which force chains in a two-dimensional granular assembly
could be characterized and hence identied quantitatively from contact force data [17].
In this technique, the process of identifying force chains proceeds in two steps: rst, a
force chain is dened to be a quasilinear arrangement of three or more particles where
stress is concentrated; second, along the chain, stress concentration within each grain is
characterized by the vector delineating the most compressive principal stress. Groups of
particles whose vectors line up, and whose stress is above average, then constitute a force
chain. The procedure was incorporated into an algorithm which takes contact force data
as the known input, and provides the force chain network as the output. We emphasize
here that this technique is strictly designed to identify force chains based on contact
force dataand not to model or predict the emergence of force chains, as was attempted
previously by other workers [18, 19].
With this new tool for force chain identication, the next step is to obtain
reliable contact force measurements for a granular material, not just at one stage
of its deformation, but throughout its deformation or loading history. In the past,
such data could only be generated from discrete element simulations. However, a
recent breakthrough in the quantitative measurement of interparticle contact forces from
photoelastic disk experiments has now paved the way for the future validation of discrete
element models at the most fundamental level [20]. With this development, the use of
DEM to probe force propagation at multiple spatial and temporal length scales, while
changing various material properties, is now an even more compelling undertaking, given
that the eects of each property may be isolated with relative ease.
Accordingly, in this study, we use discrete element simulations in conjunction with
the technique developed in [17] to generate new knowledge on the origins of macroscopic
strength. In particular, we examine how certain properties, known to promote macroscopic
strength or bulk resistance to deformation, favour the formation of force chains and the
force chain network of a particular geometric morphology. The three properties we have
chosen are
(i) interparticle contact friction (e.g. [21][23]),
(ii) packing density, or packing fraction (e.g. [9], [24][29]), here dened as the area
occupied by particles divided by the total area considered, and
doi:10.1088/1742-5468/2006/09/P09003 3
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 1. Sketch of penetration force as a measure of macroscopic strength of a
granular material subject to an indenting at punch (top), and underlying force
propagation at three length scales, viz. the force chain network and force chain
in the mesodomain and the contact force on the microscale (bottom).
Table 1. Examined properties of force propagation at multiple length scales.
Scale Force propagation property
Contact or particle (microscale) Contact force
Force chain (mesoscale 1) Length, load, curvature
Force chain network (mesoscale 2) Branching
Observable response (macroscale) Loadindentation curve
(iii) degree of polydispersity (e.g. [24, 30, 31]). This property is dicult to study in
isolation, even in DEM, since polydispersity exists in and amongst other complicating
factors.
One of the critical questions we seek to answer is how these properties aect the
nature of force propagation and how this should be quantied. In this paper, we apply
and extend the method of quantitative characterization previously developed in [17], to
investigate the evolution of properties of force chains (i.e. length, load and curvature)
and the force chain network (degree of branching), as shown in table 1. We attempt
to understand how the three propertiesinterparticle contact friction, packing density
and degree of polydispersitypromote the eective transmission of force beyond the
interparticle contact scale, that is, the mesodomain (see gure 1). The system we have
chosen to examine these eects in is that of a granular material subject to an indenting
at punch (see gure 2). Indentation by a at punch is a central problem in soil mechanics
and is of considerable importance in geotechnical engineering, particularly in foundation
design. Thus, analysis of the kind undertaken here not only provides basic insights into
doi:10.1088/1742-5468/2006/09/P09003 4
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
5.08 x 10
-1
m
Punch
Granular assembly g
x
1
x
2
5.08 x 10
-2
m
2
.
5
4

x

1
0
-
1

m
(150 particle diameters)
(
7
5

p
a
r
t
i
c
l
e

d
i
a
m
e
t
e
r
s
)
(15 particle diameters)
Figure 2. The initial setup and dimensions of the at punch system.
Table 2. Denitions of terms used.
Term Denition
Macroscopic strength A measure of the resistance of the material to punch penetration
Macroscopic load Magnitude of the net force exerted by the particles on the punch
Particle load Vector for a particle, whose magnitude and direction are given
respectively by the largest eigenvalue of the force moment tensor
and its associated eigenvector
Force chain load Average of the magnitudes of the particle loads in the force chain
Indentation Distance punch has indented material as a percentage of total
height of undeformed material
force propagation, but may also advance models for predicting the ultimate and allowable
bearing capacity for foundations or footings on rock and soil. (The average bearing
capacity is dened to be the average load per unit area required to produce failure of
the supporting soil or rock.) In the system considered here, the resistance of the material
to the punch penetration provides a measure of the macroscopic strength of the granular
material. The number of particles, size of the punch and dimensions of the granular
assembly are chosen so as to minimize the eects of the boundary on force propagation.
In table 2, we list some denitions used in this paper.
The paper is organized as follows. In section 2, we describe the numerical experiment.
In section 3, we present a method for quantitative characterization of force chains with
respect to their geometry and the load they carry, as well as branching properties of
the force chain network. We present and discuss results in section 4 and summarize our
ndings in section 5.
2. Numerical experiment
Using a discrete element method (DEM) code based on earlier work by Horner et al
[32] and partially validated against photoelastic disc experiments [33], 2D polydisperse
granular assemblies subject to indentation by a rigid at punch were simulated. Figure 2
doi:10.1088/1742-5468/2006/09/P09003 5
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Table 3. Parameters used in the simulations.
DEM parameters and material properties Value
Total indentation (% box height) 10
Time-step increment 2.86 10
6
s
Punch velocity 1.02 10
2
m s
1
Box width, height 5.08 10
1
m, 2.54 10
1
m
Punch width 5.08 10
2
m
Number of particles 10 00013 000
Particle density 2.65 10
3
kg m
3
Average radius 1.69 10
3
m
Ratio of smallest to largest radius 1:3 (1:1.2, 1:2)
Inter-particle friction coecient (f) 0.5 (0.1, 0.8, 0.3, 0.65)
Particle-punch friction coecient 0.0
Rolling friction coecient (m) 0.4 (0.01, 0.2, 0.6, 0.8)
Normal spring constant (k
n
) 1.0 10
5
N m
1
Tangential spring constant (k
t
) 2.0 10
4
N m
1
Rotational spring constant (k
m
) 2.26 10
1
N m rad
1
shows a schematic diagram of the initial setup, and table 3 gives the parameters and
material properties used, where values in brackets indicate the parameter/property was
varied. Please note that unless otherwise mentioned values used are the set of parameters
that are not in brackets.
2.1. Contact model
Similar to [34], the contact model used in the code is the following.
Linear spring (spring constant k
n
) and dashpot for the normal contact force.
Linear spring (spring constant k
t
), slider (coecient of friction f) and dashpot for
the tangential contact force.
Linear spring (spring constant k
m
), slider (coecient of friction m) and dashpot for
the contact moment (i.e. rolling resistance).
2.2. Creation of granular specimen of various packing densities
The initial packings were generated by randomly creating particles on a grid with no
contacts and allowing them to fall into the box under gravity. Once particles had settled
(negligible kinetic energy), particles that fell above the line x
2
= 0 (see gure 2) were
deleted, to create a relatively at surface on which to move the punch. The system, once
again, was allowed to settle.
To create systems with varying packing densities, interparticle friction was either
turned o or set to 0.5 during this period. This resulted in two sets of systems: a more
densely packed set (referred to as dense) and a less densely packed set (referred to as
loose). The names of the dierent systems that will be used from here on and their
corresponding initial packing densities are given in table 4. Note that packing densities
are calculated based on the initial area of the granular assembly as in gure 2.
doi:10.1088/1742-5468/2006/09/P09003 6
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Table 4. Test names and initial packing densities.
Test name Average packing density
1:1.2 loose 0.864
1:2 loose 0.861
1:3 loose 0.862
1:1.2 dense 0.905
1:2 dense 0.894
1:3 dense 0.898
2.3. Justication of parameters chosen
A drawback of DEM is the slightly random way in which parameters are chosen. The
normal spring drives the time step (stier spring requires smaller time step), and here the
value chosen has been used by Luding et al [35, 36]. Typical ratios of tangential to normal
spring constants range between 1:1 and 1:10. Here, the ratio is 1:5 [35, 36].
Following the work of Sakaguchi and co-workers [37], we include the eects of rolling
resistance that would be present in assemblies of non-circular particles. However, since
rolling resistance is a less common feature of DEM codes, and so far is not able to be
measured experimentally, the value for rolling stiness (k
m
) was based on k
m
/k
t
r
2
[38],
where r is the (average) particle radius. The value for rolling friction (m) was varied.
The dimensions of the box and ratio of punch width to box width (1:10) were chosen
in order to reduce boundary eects (e.g. [39]), particle density is that of sand, the time
step is calculated based on ratios of stiness to mass and the largest polydispersity (1:3) is
as in [36]. Since we are considering shallow indentation only, the punch indents to a total
of 10% of the height of the box (the height of the granular assembly in its undeformed
conguration; see gure 2).
3. Quantication of force chain and force chain network properties
This section describes and quanties the four properties of force chains and the force
chain network that will be examined. The denition of force chains, quasilinear particle
assemblies carrying the majority of the load, is used to quantitatively identify force chains
in accordance with a modied version of the method proposed in [17]. In the original
method, the particle stress tensor was used to calculate an estimate of the load carried by
a particle [17]. However, since the particle stress tensor is normalized by particle volume,
this produced biased statistics (small particles were preferred). Here we use an improved
approach, recently proposed in [40], in which a force moment tensor for each particle is
calculated as follows:

ij
=
1
|r|
N

c=1
f
c
i
r
c
j
(1)
where N is the number of contacts of the particle, f
c
is the contact force and r
c
the
vector from the particles center to the point of contact. The magnitude and direction
of the particle load vector are then given respectively by the largest eigenvalue of the
force moment tensor and its associated eigenvector, with the main direction of force
doi:10.1088/1742-5468/2006/09/P09003 7
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
1
2
3
4
Figure 3. Force chain branching: choosing the best branch (top) and allocating
a branch number to each particle (bottom).
transmission represented by the direction of the eigenvector. Groups of particles whose
eigenvectors line up, and whose particle load magnitude is above average, constitute a
force chain. More details are given in [17].
3.1. Force chain length
An improvement to the algorithm was made as a rst step towards incorporating
branching. In the case where there are two possibilities for choosing the next particle,
the particle candidate with a higher load is chosen. Each chain is allocated a lengththe
number of particles in the chain. Figure 3 (top) shows the higher load pathway being
chosen, resulting in one chain of length 5 and one of length 3.
3.2. Force chain load
We dene the force chain load to be the average of the magnitudes of the particle loads
of the particles in the force chain.
3.3. Force chain curvature
As well as length and load, we wish to quantify the geometrical structure of a chain. To
do this, we propose a measure of the force chain curvature as the ratio of the sum of
the magnitudes of branch vectors linking each particle to the next in the chain to the
magnitude of the vector linking the rst to last particle (also known as the end-to-end
vector in polymer rheology [41]):
s =

N1
p=1
|x
p+1
x
p
|
|x
N
x
1
|
(2)
doi:10.1088/1742-5468/2006/09/P09003 8
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
(a)
(b)
Figure 4. Force chain curvature: (a) a straighter chain, curvature value 1, and
(b) a more curved chain, curvature value > 1.
where N is the number of particles in the chain, and x
p
is the position vector of particle
p. Therefore, a curvature value of unity indicates a perfectly straight chain, and values
greater than unity indicate the chain is more curved (see gure 4).
3.4. Force chain network branching properties
Individual force chains provide some meso-level information; however, information on the
global structure of the force chain network is also desirable. Hence, we dene a branch
number for each particle in the system as follows:
0, not part of a force chain;
1, end-point, transmitting load to one other particle (the rst or last particle in a
chain);
2, segment-point, transmitting load to two other particles in force chains (one particle
in each directionin a straight segment);
3 or more, branch-point, transmitting load to three or more particles in force chains
(at a junction point of three or more chain segments).
This is represented in gure 3 (bottom).
4. Results and discussion
In this section, we examine the eects of sliding friction, degree of polydispersity
and packing density on the nature of force propagation. As noted earlier, we have
also introduced rolling friction in these simulations to account for the resistance to
rolling that would arise in assemblies of non-circular particles [37]. In all systems
examined, irreversible particle rearrangements occur even under innitesimal loads. Thus,
the deformation of the granular assembly caused by the indention of the punch is
predominantly plastic.
4.1. Loadindentation: macroscale
Figure 5 shows the loadindentation curve, for the ve dierent values of interparticle
sliding (left) and rolling (right) friction used. Here, the macroscopic load is dened to be
the magnitude of the net force exerted by the particles on the punch, and indentation is
measured as the distance the punch has indented the granular material as a percentage of
the total height of the granular material in its undeformed conguration. The inset of each
doi:10.1088/1742-5468/2006/09/P09003 9
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 5. Loadindentation curve, for (1:3, loose, vary f, m = 0.4) (left) and
(1:3, loose, f = 0.5, vary m) (right). The insets show the gradient of the line of
best t versus friction.
graph shows the macroscopic strength plotted against the respective friction coecients.
We calculate the macroscopic strength for each friction value by performing a linear
regression to obtain the least-squares t line (so called line of best t) for the load
indentation curve, and then use its gradient as a measure of the macroscopic strength.
The test case used for varying friction is the 1:3 loose system (see table 4).
As sliding friction is increased, the macroscopic strength also increases, although not
linearly. The increase of the macroscopic strength with increasing sliding friction at the
interparticle contact has also been found in experiments examining the force on a piston
vertically pushing up a granular column of monodisperse beads [23], as well as in granular
fault gouge layers [21, 22]. Higher values for sliding friction mean that a greater amount of
tangential force can be sustained at the contact without resulting in plastic interparticle
slip. Thus, the increased bulk resistance of the material to the indenting punch is due
to the greater resistance to relative motion developed at the individual interparticle
contacts.
Interestingly, the macroscopic strength saturates beyond sliding friction values of
approximately 0.6. This phenomenon of saturation of macroscopic properties has been
found in three dimensional (3D) discrete element simulations measuring the angle of
repose [42], as well as the mobilized angle of internal friction, critical void ratio, average
coordination number and ratio of sliding contacts [43]. Also, microscopic quantities
exhibit this qualityuniaxial compression experiments of smooth and rough glass beads,
in amorphous and crystalline packings, measured the probability distribution of normal
forces at the boundary of the container and found it to be unaected when interparticle
friction increases by a factor of three [44]. This trend may signal the onset of a dierent
mechanism beyond the contact scale. This is currently being investigated.
doi:10.1088/1742-5468/2006/09/P09003 10
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 6. Loadindentation curve for ve realizations of (1:2, loose, f = 0.5,
m = 0.4) (top left) and (1:2, dense, f = 0.5, m = 0.4) (top right). Load
indentation curve, averaged over realizations, for (vary polydispersity, vary
packing density, f = 0.5, m = 0.4) (bottom).
Since the macroscopic strength also increases with rolling friction, this demonstrates
that rolling friction is an equally important microscale particle property that cannot be
neglected. Notably, for very small rolling friction (0.01), which corresponds to very little
rolling resistance, the granular materials ability to resist the punch is weak, even with
a moderate sliding friction value (0.5). This further reiterates the need for models of
granular materials to incorporate rolling resistance in order to produce more realistic
results [45, 46, 34]. Finally, the gradient of the line of best t of the loadindentation
curve also saturates past rolling friction values of 0.4.
Figure 6 (top row) shows the loadindentation curves for ve realizations of the
1:2 polydispersity, for less densely packed (1:2 loose, left) and more densely packed
(1:2 dense, right). In all cases, there is signicant variation between realizations. To
remove uctuations, we present the average over realizations in gure 6 (bottom), for all
polydispersities and both packing densities. Note that in all further results, data presented
are the averages over realizations unless otherwise noted.
Figure 6 (bottom) shows that in all polydispersities the more densely packed material
is stronger, consistent with experiments (see, for example, [9], [24][29]). This is because
constituent grains in a denser specimen are more interlocked (i.e. jammed) and thus
doi:10.1088/1742-5468/2006/09/P09003 11
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 7. Sample force chains found using particle loads (left) and scaled particle
loads (right), before the punch has indented, coloured by average force chain load,
for (1:3, loose, f = 0.5, m = 0.8).
constrain each others motion. On the other hand, the presence of larger interstitial voids
in looser specimens simply means that the particles have greater room to move and can
thus accommodate greater relative particle motion and rearrangements.
Varying the degree of polydispersity in the more densely packed assemblies had
little eect on the macroscopic strength, with the loadindentation graphs being nearly
coincident over the entire range of indentation. For the less densely packed specimens,
however, our data indicate that a greater degree of polydispersity leads to an overall
increase in the macroscopic strength. While Zhou and co-workers experimentally observe
that the macroscopic drag force for a monodisperse mixture is lower than for polydisperse
mixtures, they attribute this to packing density [24]. Material-point method simulations
of dense granular material under shear also nd that monodisperse material is weaker
than polydispersehowever, this is for identical packing densities [31].
Furthermore, in our case, the monodisperse mixture has a slightly higher packing
density (see table 4) but still provides a noticeably lower amount of resistance to
indentation. A possible reason for this is the dierence in the distribution of coordination
numbers for the monodisperse and polydisperse systems (not shown): while monodisperse
assemblies are limited in their range of coordination numbers (the maximum is six for
1:1.2 systems), polydisperse assemblies can accommodate a wider range of coordination
numbers (up to nine for a 1:3 loose system), hence providing more possible pathways for
stress transmission.
4.2. Mesoscale: force chain and force chain network properties
In the previous section, we established that interparticle friction, packing density, and
polydispersity promote bulk resistance to deformation. In this section, we relate these
observations made on the macroscopic scale to the properties of force propagation on the
mesoscale, viz., force chains and the force chain network. Our aim is to identify what
characteristics of force chains and force chain networks lead to a macroscopically stronger
material.
Figure 7 (left) shows an example force chain conguration found using the method
in [17], before the punch has indented. Dark blue particles are not part of a force chain,
and the colour of all other particles represents the average load of the chain (as dened
in 3.2), from light blue (low) to red (high). Note that the same scale has been used for all
doi:10.1088/1742-5468/2006/09/P09003 12
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 8. Evolution of average force chain length (top row) and curvature
(bottom row), for (1:3, loose, vary f, m = 0.4) (left) and (1:3, loose, f = 0.5,
vary m) (right).
force chain pictures that follow. Similar to photoelastic experiments on grain piles [47],
force chains have already formed due to gravitational head and the method of deposition.
However, we wish to isolate the eects of the indenting punch, hence gure 7 (right)
shows the same conguration, but force chains have been found instead using the particle
load (as dened in section 3) divided by the depth of the particle within the assembly.
Again, particles are coloured by the average force chain load (force chain load as dened
in section 3.2 except using the scaled particle loads). The resulting force chains are now
all approximately equal in load. Thus, on indentation, force chains directly under the
punch will be the highest in load. Note that from now on this scaled particle load (and
hence force chain load) will be used.
The evolution of the average load of force chains (not shown) mimics the load
indentation curves almost exactly: that is, as the macroscopic strength increases, so does
the average force chain load. Also, the number of force chains present decreases with
indentation (not shown) and follows the loadindentation curve in reverse. These two
ndings demonstrate that, as the punch indents, the chains that remain must be those
that are able to sustain the highest load.
Figure 8 (top) shows the evolution of the average chain length, for varying both
rolling and sliding friction. There is no noticeable change in average length with either
doi:10.1088/1742-5468/2006/09/P09003 13
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 9. Evolution of average force chain length (left) and curvature (right),
for (vary polydispersity, vary packing density, f = 0.5, m = 0.4) (averaged over
realizations).
indentation or friction. However, from examination of gure 7 (and also gures 10 and 11),
the spatial variations in length are evident. By averaging over the deforming granular
material, any spatial uctuations are smeared out, hence the average fails to capture the
spatial distribution of lengths and also their evolution with indentation. This highlights
the need for future studies to establish other suitable quantities that take into account
these uctuations, and thus retain more information on particulate behaviour.
Figure 8 (bottom) shows the evolution of the average chain curvature, for varying
both kinds of friction. In all cases, the average curvature decreases with indentation,
which indicates that the chains that prevail during indentation are straighter. Average
curvature mimics the loadindentation curve (gure 5), but in reverse: for high friction,
the macroscopic strength is higher, while the average curvature is smaller. A correlation
between each chains load and its curvature (not shown) conrms that the chains that
are able to sustain higher loads are straighter. This suggests that the materials ability
to form straighter force chains is one contributing factor to macroscopic strength.
Figure 9 (left) shows the evolution of the average chain length, for varying
polydispersity and packing. Ignoring at this point the increase in force chain length
in the 1:1.2 dense case (induced by local recrystallization, as explained later), we again
observe no noticeable change in average length with either indentation or polydispersity,
suggestive of a characteristic length. Also, the densely packed systems have shorter chains.
This is in agreement with experiments using a slowly sheared 2D assembly of photoelastic
disks examining the eect of packing density on the force network [1], and with results
from changing gravity [48]. These ndings suggest that a material with shorter chains
is macroscopically stronger. Intuitively, we would expect longer chains to be less stable.
Consider, for example, the task of lifting a row of balls resting on a table by compressing
them together between two hands: it is relatively easy to pick up three balls but far more
dicult to pick up four or more. Interestingly, in order to explain observed macroscopic
doi:10.1088/1742-5468/2006/09/P09003 14
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 10. Force chains after the punch has indented 10%, showing eects of
sliding friction, top left (1:3, loose, f = 0.1, m = 0.4) and top right (1:3, loose,
f = 0.8, m = 0.4), and rolling friction, bottom left (1:3, loose, f = 0.5, m = 0.01)
and bottom right (1:3, loose, f = 0.5, m = 0.8).
dierences in sheared fault gouges, Anthony and Marone [21] propose the same hypothesis:
shorter chains can support greater stress since there are fewer potential failure points.
Figure 9 (right) shows the evolution of the average chain curvature for varying
polydispersity and packing. Again, in all cases, the average curvature decreases with
indentation, reinforcing the fact that chains that are able to sustain higher load and are
more stable are straighter. However, while the polydisperse systems are macroscopically
stronger than the almost monodisperse (gure 6), the almost monodisperse systems have
straighter chains. This suggests that a materials ability to form straight chains is only
one factor that governs macroscopic strengthother competing factors are at play.
On examination of the kinematics of the material at the microscale in the almost
monodisperse systems, local recrystallization is evident. Clusters of particles arrange
themselves into regions exhibiting regular hexagonal (i.e. crystalline) packing. Note that
a comparison of the dense and loose 1:1.2 systems shows a greater degree of crystallization
for the dense system. These clusters eventually break down and new ones are formed as
the punch indents. This phenomenon of local recrystallization was also found in material-
point method simulations of sheared, monodisperse dense granular material [31]. Local
recrystallization provides an explanation as to why the almost monodisperse systems have
straighter force chainsin the crystallized regions, chains are almost perfectly straight.
Figure 10 shows the force chains after the punch has indented 10%, for the limiting
cases of small and large sliding and rolling friction. For large friction, the force chain
doi:10.1088/1742-5468/2006/09/P09003 15
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 11. Force chains after the punch has indented 10%, showing eects of
packing density and polydispersity: top left (1:1.2, loose, f = 0.5, m = 0.4), top
right (1:1.2, dense, f = 0.5, m = 0.4), middle left (1:2, loose, f = 0.5, m = 0.4),
middle right (1:2, dense, f = 0.5, m = 0.4), bottom left (1:3, loose, f = 0.5,
m = 0.4), bottom right (1:3, dense, f = 0.5, m = 0.4).
network is denser and carries more load underneath the punch. Chains are straighter
(conrmed by gure 8) and more radially oriented. Here, particles can resist more shear
force without sliding, so chains can develop to a greater extent and prevail. Hence, the
macroscopic strength is higher. For small friction, the force chain network is sparser and
chains are less radially oriented. Particles can slide (or roll) past each other more easily.
Rather than sticking together to resist the applied load, particles simply rearrange and
move up the sides of the punch (more dilatation). This means force chains cannot develop
to the same extent (cannot sustain as much load) as those in the more frictional material,
resulting in a smaller macroscopic strength.
doi:10.1088/1742-5468/2006/09/P09003 16
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 12. Evolution of distribution of force chain loads (left) and lengths (right),
for (1:2, dense, f = 0.5, m = 0.4).
Figure 11 shows the force chains after the punch has indented 10%, for varying
polydispersity and packing density (for a single realization). Local crystallization is
evident for the 1:1.2 dense system, and to a lesser extent, the 1:1.2 loose system. This
explains the increasing length in gure 9 (left): as the punch indents, the crystallized
structure is already close to maximum packing. In this system, particles cannot rearrange
easily and accommodate a reconguration of the force chain network. With forces being
transmitted only in certain discrete directions, the force chains simply get longer, engaging
more and more of the particles to share the load, as indentation increases. It is interesting
to note that these unique mesoscale events are not evident on the macroscopic scale: there
is no noticeable dierence in the loadindentation curve of this specimen compared to the
other densely packed assemblies.
Also, the general trends observed in gure 6 of denser materials and polydisperse
materials (for less densely packed) having greater macroscopic strength are demonstrated
by the higher loads of force chains evident in gure 11.
The evolution of the distribution of force chain loads in all cases follows the same
pattern as in gure 12 (left): from a tall, narrow humped distribution with a smaller mean
to shorter and wider with a larger mean. The extent to which the distribution translates
and changes shape governs the macroscopic strength. The fact that the distribution does
change shape (and does not simply translate) is indicative of chains able to sustain much
more load than the average.
Figure 12 (right) shows a sample force chain length distribution. Interestingly, there
is no signicant change in the distribution of force chain lengths with either indentation
or with changing friction, polydispersity or packing density (not shown). This, combined
with the uniformity of the average chain length, is suggestive of a characteristic length of
force chains, dependent on packing density but independent of indentation, friction and
polydispersity. Figure 12 (right, inset) shows a log plot, demonstrating an exponential
distribution, consistent with experiments [1].
doi:10.1088/1742-5468/2006/09/P09003 17
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 13 shows the evolution of the force chain curvature distribution for varying
polydispersity and packing. Ignoring the 1:1.2 dense case, we observe that as indentation
proceeds the proportion of straighter chains increases, while the proportion of curved
chains decreases. As noted earlier, the 1:1.2 dense case shows signicant local
recrystallization. For this system, we see a marked dierence in the distribution shape
(also note the dierent vertical scale). The peak has shifted and become much smaller,
with the local minimum coinciding with the peaks in all the other graphs in gure 13.
This is due to the crystallized nature of the material and the fact that only a discrete set
of directions can be chosen for force transmission.
Figure 14 shows the evolution of the three network branching properties dened in
section 3.4. A comparison of the left and right columns demonstrates that, while rolling
friction does not seem to have an eect on the branching properties of the force chain
network, polydispersity and packing density do. Examining each row reveals that, apart
from the crystallized case, while the proportion of segment-points does not vary much with
indentation, branch-points slightly decrease, and end-points slightly increase. Similarities
between the graphs of proportions of segment points (middle row of gure 14) and average
chain length (gures 8 and 9) suggest a possible correlation between numbers of segment
points and average chain length; however, this is outside the scope of the paper.
Focusing on the left column, we see that with increasing polydispersity the proportion
of branch-points increases and segment- and end-points decreases. This is expected, since
the distribution of coordination numbers for the polydisperse system is more spread,
with some particles having up to nine contacts, and hence more possible branch-points.
In contrast, the maximum number of contacts is six for monodisperse systems, which
limits possible directions for force transmission. Previously, we observed that for the
less dense systems (1:1.2 loose, 1:2 loose and 1:3 loose) the polydisperse system is
macroscopically stronger than the almost monodisperse, while the almost monodisperse
system has straighter chains. Hence, one possibility is that bulk macroscopic strength is
linked to the degree of branching possible in the force chain network as well as the shape
of individual chains. Specically, a force chain network with more branch-points means
greater capacity for the material to support more load, since there are more pathways
available for stress transmission. Once an existing pathway for stress transmission is no
longer viable, force may be simply transferred to a dierent branch of the force chain
network, without the need for particles to rearrange.
Also, we see that the more dense systems (1:1.2 dense, 1:2 dense and 1:3 dense) have
more branch-points and fewer segment-points than their respective less dense systems
(1:1.2 loose, 1:2 loose and 1:3 loose). This nding is consistent with observations from
experiments that examined the eect of packing density: the force chain network was
observed as changing from intermittent, long, radial chains for less densely packed, to
being more tangled, with many shorter intersecting chains for more densely packed [1],
where tangled is quantied here by the degree of branching.
5. Conclusions
Presented herein is an application of the method for the quantitative identication and
characterization of force chains based on contact force data recently developed in [17].
The system considered is a 2D granular material subject to indentation by a rigid at
doi:10.1088/1742-5468/2006/09/P09003 18
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 13. Evolution of distribution of force chain curvatures, for top left (1:1.2,
loose, f = 0.5, m = 0.4), top right (1:1.2, dense, f = 0.5, m = 0.4), middle left
(1:2, loose, f = 0.5, m = 0.4), middle right (1:2, dense, f =0.5, m = 0.4), bottom
left (1:3, loose, f = 0.5, m = 0.4), bottom right (1:3, dense, f = 0.5, m = 0.4).
The legend shown in the top right corner is for all graphs. The arrows show the
general trend with evolution. Note that a curvature value of unity corresponds
to a perfectly straight chain.
doi:10.1088/1742-5468/2006/09/P09003 19
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
Figure 14. Evolution of end-points (top row), segment-points (middle row)
and branch-points (bottom row), for (vary polydispersity, vary packing density,
f = 0.5, m = 0.4) (left column) and (1:3, loose, f = 0.5, vary m) (right column).
Legends are only shown on the bottom row, and scales on the extreme bottom
and left.
doi:10.1088/1742-5468/2006/09/P09003 20
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
punch. It is well known that interparticle friction, packing density and polydispersity are
major contributors to the macroscopic strength of a granular material. In this paper, we
have attempted to give insights into the mechanisms by which these particle properties
promote force propagation via an analysis of the force chains and force chain network.
A materials ability to form straighter force chains inuences its macroscopic strength:
chains that are able to sustain higher loads are straighter.
The shape of chains, however, is not the only factor at play. The degree of branching
also governs macroscopic strength: systems with more branching in their force chain
network are macroscopically stronger.
Force chain length appears to be independent of friction and polydispersity, but more
dense systems have shorter chains than less dense. This suggests that materials with
shorter chains are macroscopically stronger than those with longer chains.
6. Acknowledgements
The support of the US Army Research Oce (grant number DAAD19-02-1-0216) and
the Australian Research Council (DP0558808) through a grant to AT is gratefully
acknowledged, as is the University of Melbourne Advanced Research Computing section
through a grant of computer time. MM gratefully acknowledges the support of the Pratt
Foundation Scholarship for postgraduate study.
References
[1] Howell D, Behringer R P and Veje C, Stress uctuations in a 2d granular couette experiment: a continuous
transition, 1999 Phys. Rev. Lett. 82 5241
[2] Cates M E, Wittmer J P, Bouchaud J-P and Claudin P, Jamming, force chains, and fragile matter, 1998
Phys. Rev. Lett. 81 (9)
[3] Farr R S, Melrose J R and Ball R C, Kinetic theory of jamming in hard-sphere startup ows, 1997 Phys.
Rev. E 55 7203
[4] Brujic J, Edwards S F, Grinev D V, Hopkinson I, Brujic D and Makse H A, 3d bulk measurements of the
force distribution in a compressed emulsion system, 2003 Faraday Discuss. 123 207
[5] Tanguy A, Wittmer J P, Leonforte F and Barrat J-L, Continuum limit of amorphous elastic bodies: a
nite-size study of low-frequency harmonic vibrations, 2002 Phys. Rev. B 66 174205
[6] Lacevic N and Glotzer S C, Dynamical heterogeneity and jamming in glass-forming liquids, 2004 J. Phys.
Chem. B 108 19623
[7] Zhou J, Long S, Wang Q and Dinsmore A D, Measurement of forces inside a three-dimensional pile of
frictionless droplets, 2006 Science 312 1611
[8] Walsh S D C, Tordesillas A and Peters J F, Development of micromechanical models for granular media:
the projection problem, 2006 Granul. Matter at press
[9] Geng J and Behringer R P, Slow drag in two-dimensional granular media, 2005 Phys. Rev. E 71 (1)
[10] Albert I, Tegzes P, Albert R, Barabasi A-L, Vicsek T, Kahng B and Schier P, Stick-slip uctuations in
granular drag, 2001 Phys. Rev. E 64 (3)
[11] Behringer R P, Howell D, Kondic L, Tennakoon S and Veje C, Predictability and granular materials, 1999
Physica D 133 1
[12] Howell D W and Behringer R P, Fluctuations in granular media, 1999 Chaos 9 559
[13] Miller B, OHern C and Behringer R P, Stress uctuations for continuously sheared granular materials,
1996 Phys. Rev. Lett. 77 3110
[14] Liu C-H, Nagel S R, Schecter D A, Coppersmith S N, Majumdar S, Narayan O and Witten T A, Force
uctuations in bead packs, 1995 Science 269 513
[15] Albert R, Pfeifer M A, Barabasi A-L and Schier P, Slow drag in a granular medium, 1999 Phys. Rev. Lett.
82 (1)
doi:10.1088/1742-5468/2006/09/P09003 21
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
[16] Kuhn M R, Are granular materials simple? an experimental study of strain gradient eects and localization,
2005 Mech. Mater. 37 607
[17] Peters J F, Muthuswamy M, Wibowo J and Tordesillas A, Characterization of force chains in granular
material , 2005 Phys. Rev. E 72 041307
[18] Blumenfeld R, Stresses in isostatic granular systems and emergence of force chains, 2004 Phys. Rev. Lett.
93 (10)
[19] Bouchaud J-P, Claudin P, Levine D and Otto M, Force chain splitting in granular materials: a mechanism
for large-scale pseudo-elastic behaviour, 2001 Eur. Phys. J. E 4 451
[20] Majmudar T S and Behringer R P, Contact force measurements and stress-induced anisotropy in granular
materials, 2005 Nature 435 1079
[21] Anthony J L and Marone C, Inuence of particle characteristics on granular friction, 2005 J. Geophys. Res.
110 (B8)
[22] Mair K, Frye K M and Marone C, Inuence of grain characteristics on the friction of granular shear zones,
2002 J. Geophys. Res. 107 (B10)
[23] Kolb E, Mazoz T, Clement E and Duran J, Force uctuations in a vertically pushed granular column, 1999
Eur. Phys. J. B 8 483
[24] Zhou F, Advani S G and Wetzel E D, Slow drag in polydisperse granular mixtures under high pressure, 2005
Phys. Rev. E 71 061304
[25] Yamamuro J A and Lade P V, Drained sand behavior in axisymmetric tests at high pressures, 1996
J. Geotech. Eng.-ASCE 122 109
[26] Anandarajah A, Sobhan K and Kuganenthira N, Incremental stress-strain behavior of granular soil , 1995
J. Geotech. Eng.-ASCE 121 57
[27] Buttereld R and Andrawes K Z, Investigation of a plane strain continuous penetration problem, 1972
Geotechnique 22 597
[28] Koerner R, The behavior of cohensionless soils formed from various minerals, 1968 PhD Thesis Duke
University
[29] Sylwestrowicz W, Experimental investigation of the behaviour of soil under a punch or footing, 1953
J. Mech. Phys. Solids 1 258
[30] Geminard J C and Losert W, Frictional properties of bidisperse granular matter: eect of mixing ratio,
2002 Phys. Rev. E 65 041301
[31] Bardenhagen S G, Brackbill J U and Sulsky D, Numerical study of stress distribution in sheared granular
material in two dimensions, 2000 Phys. Rev. E 62 3882
[32] Horner D A, Peters J F and Carrillo A, Large scale discrete element modeling of vehiclesoil interaction,
2001 J. Eng. Mech.-ASCE 127 1027
[33] Hafez H S, 600-311 research report: a discrete element simulation of a 2d assembly of bi-disperse circular
disks subjected to pure shear, 2005 Technical Report The University of Melbourne
[34] Oda M and Iwashita K, Study on couple stress and shear band development in granular media based on
numerical simulation analyses, 2000 Int. J. Eng. Sci. 38 1713
[35] Luding S, Stress distribution in static two-dimensional granular model media in the absence of friction,
1997 Phys. Rev. E 55 4720
[36] Luding S, Molecular dynamics simulations of granular materials, 2004 The Physics of Granular Media ed
H Hinrichsen and D Wolf (Weinheim: WileyVCH) p 299
[37] Sakaguchi H, Ozaki E and Igarashi T, Plugging of the ow of granular materials during discharge from a
silo, 1993 Int. J. Mod. Phys. B 7 1949
[38] Iwashita K and Oda M, Rolling resistance at contacts in simulation of shear band development by dem,
1998 J. Eng. Mech.-ASCE 124 283
[39] Potts D and Zdravkovic L (ed), 1999 Finite Element Analysis in Geotechnical Engineering (London:
Thomas Telford)
[40] Bosko J T and Tordesillas A, Unjamming transitions in dense granular assemblies subjected to biaxial
compression, 2006 Phys. Rev. E submitted
[41] McNaught A D and Wilkinson A (ed), 1997 IUPAC Compendium of Chemical Terminology Royal Society
of Chemistry
[42] Ji S and Shen H H, Contrasting terrestrial and lunar gravity: angle of repose and incline ows, 2006 Proc.
10th ASCE Aerospace Division Int. Conf. on Engineering, Construction and Operations in Challenging
Environments (Earth & Space 2006) ed R B Malla, W K Binienda and A K Maji (Reston, VA:
Aerospace Division of the American Society of Civil Engineers)
[43] Thornton C, Numerical simulations of deviatoric shear deformation of granular media, 2000 Geotechnique
50 43
doi:10.1088/1742-5468/2006/09/P09003 22
J
.
S
t
a
t
.
M
e
c
h
.
(
2
0
0
6
)
P
0
9
0
0
3
Force propagation in particulate assemblies
[44] Blair D L, Mueggenburg N W, Marshall A H, Jaeger H M and Nagel S R, Force distributions in
three-dimensional granular assemblies: Eects of packing order and interparticle friction, 2001 Phys.
Rev. E 63 (4)
[45] Iwashita K and Oda M, Micro-deformation mechanism of shear banding process based on modied distinct
element method, 2000 Powder Technol. 109 192
[46] Tordesillas A, Peters J F and Muthuswamy M, Role of particle rotations and rolling resistance in a
semi-innite particulate solid indented by a rigid at punch, 2005 CTAC-2004: Proc. 12th Computational
Techniques and Applications Conf. (April 2005) vol 46, ed R May and A J Roberts, pp C26075
[47] Geng J, Longhi E, Behringer R P and Howell D W, Memory in two-dimensional heap experiments, 2001
Phys. Rev. E 64 060301
[48] Muthuswamy M and Tordesillas A, Multiscale analysis of the eects of changing gravity on stress
propagation underneath a rigid at punch, 2006 Proc. 10th ASCE Aerospace Division International
Conference on Engineering, Construction and Operations in Challenging Environments (Earth & Space
2006) ed R B Malla, W K Binienda and A K Maji (Reston, VA: Aerospace Division of the American
Society of Civil Engineers)
doi:10.1088/1742-5468/2006/09/P09003 23

You might also like