Shear strength and stress distribution in wet granular media

Vincent Richefeu∗ , Farhang Radjaï† and Moulay Saïd El Youssoufi†

∗
Laboratoire 3S-R, Domaine Universitaire, B.P. 53, 38041 Grenoble Cedex 9, France
†
LMGC - UMR CNRS 5508, Université Montpellier 2, C.C. 048, 34095 Montpellier Cedex 5, France

Abstract. We investigate the shear strength and stress distribution properties of wet granular media in the pendular state where
the liquid is mainly in the form of capillary bonds between particles. This work is based on a 3D discrete-element approach
(molecular dynamics) with spherical particles enriched by a capillary force law. We show that the capillary force can be
expressed as an explicit function of the gap and volume of the liquid bridge. The length scales involved in this expression are
analyzed by comparing with direct integration of the Laplace-Young equation. In the simulations, we consider a maximum
number density of liquid bonds in the bulk in agreement with equilibrium of each liquid bridge. This liquid bond number is a
decisive parameter for the overall cohesion of wet granular materials. It is shown that the shear strength can be expressed
as a function of liquid bond characteristics. The expression proposed initially by Rumpf is thus generalized to account
for size polydispersity. We show that this expression is in good agreement with our experimental data that will be briefly
described. At low confining stress, the tensile action of capillary bonds induces a self-stressed particle network organized in
a bi-percolating structure of positive and negative particle pressures. Various statistical descriptors of the microstructure and
bond force network are used to characterize this partition. Two basic properties emerge: (i) The highest particle pressure is
located in the bulk of each phase (positive and negative particle pressures); (ii) The lowest pressure level occurs at the interface
between the two phases, involving also the largest connectivity of the particles via tensile and compressive bonds.
Keywords: capillary cohesion, discrete element approach, shear strength, self-stresses
PACS: 45.70.-n, 81.05.Rm, 83.80.Fg

INTRODUCTION MODEL DESCRIPTION

Cohesive granular media are of particular interest to the We used the MD method with spherical particles and a
processing (compaction, granulation. . . ) of fine powders capillary force law. The total normal force f n at each con-
or geomaterials. Although cohesive materials present tact is the sum of a repulsive force f nr and an attractive
strong similarities with dry granular media due to their capillary force f nc . The latter is a function of the liquid
common granular microstructure, the presence of cohe- bond parameters, namely the gap ! n , the liquid bond vol-
sion induces new mechanisms such as particle aggrega- ume Vb , the liquid surface tension " s , and the particle-
tion affect the static and dynamic properties of the mate- liquid-gas contact angle # ; see inset in Fig. 1. The cap-
rial. illary force can be calculated by integrating the Laplace-
In this paper, we use 3D molecular dynamics method Young equation [1, 2, 3]. However, for efficient MD sim-
in which capillary attraction between spherical particles ulations, we need an explicit expression of f nc as a func-
is implemented as a force law expressing the capillary tion of the liquid bond parameters.
force as a function of the distance, water volume, and We used an analytical form for the capillary force
particle diameters. We first describe our model of cap- which is well fitted by the data from direct integration
illary cohesion. The shear strength at low confinement of the Laplace-Young equation both for polydisperse
pressures is then analysed by means of experimental particles [4]. At leading order, the capillary force f 0 at
and numerical direct shear tests. We show that the shear contact, i.e. for ! n ≤ 0, is
strength can be expressed as a function of liquid bond
characteristics. In the last section, we introduce particle f0 = −$ R, (1)
stresses and we analyze their distribution to characterize
the organization of particle pressures which take positive where R is a length depending on the particle radii R i and
or negative values according to their positions in the net- R j and $ is given by [5, 6, 7]
work of self-equilibrated bond forces. We will see that
$ = 2%"s cos # . (2)
this organization involves a genuin partition of the parti-
cles in two phases of negative and positive pressures. The adhesion force f 0 at contact (independent of the
bond liquid volume Vb ) is the highest value of the cap-
illary force. The latter declines as the gap ! n increases.
The capillary bridge is stable as long as ! n < !nmax , where
 In the simulations, the total liquid volume is dis-
tributed among all eligible particle pairs (the pairs with
a gap below the debonding distance, including the con-
tact points) in proportion to the reduced diameter of each
pair. We also assume that the particles are perfectly wet-
Ri
Rj table, i.e. # = 0. The choice of the water volume has no
θ
influence on the value of the largest capillary force in the
δn pendular state [9]. For our simulations, we chose a gravi-
metric water content lower than 0.05 so that the mate-
rial is in the pendular state. The coefficient of friction is
µ = 0.4 in all simulations.

FIGURE 1. Scaled plot of the capillary force as a function SHEAR STRENGTH

of the gap between two particles for different values of the
local liquid volume Vb and size ratio r according to the model For the shear strength analysis, we performed both ex-
proposed in this paper. Inset: Geometry of a capillary bridge. perimental and numerical direct shear tests. The exper-
iments were carried out with glass beads and sand by
means of a setup especially designed for weak confin-
!nmax is the debonding distance given by [8] ing stresses (below 1 kPa) in order to enhance the capil-
! " lary effects [9]. Both experimental and numerical results
max # 1/3
!n = 1 + Vb . (3) were analyzed in the Mohr-Coulomb space. We found
2 that the internal angle of friction was not sensitive to wa-
Between these two limits, the capillary force falls off ter content w and the Coulomb cohesion increased in a
exponentially with ! n : nonlinear fashion with w to reach a limit value at a well-
defined level cm of cohesion independent of water con-
fnc = f0 e−!n /& , (4) tent. A similar behavior was observed in numerical sim-
ulations. The cohesion c m was found to be quite close
where & is a length scale which should be a function of
between the experiments and numerics for the samples
Vb and the particle radii. The asymmetry due to unequal
of the same particle size distribution (c m = 120 Pa in nu-
particle sizes is taken into account through a function of
merics vs. cm = 150 Pa in experiments).
the ratio between particle radii. We set
Let us consider the general expression of the stress
r = max(Ri /R j ; R j /Ri ). (5) tensor ' in a granular material. This is an average quan-
tity with an expression involving contact forces f k and
Dimensionally, a plausible expression of & is branch vectors ! k :
! "1/2 1
V 'i j =
& = c h(r) b$ , (6) V ( fik !kj , (8)
R k∈V

where c is a constant and h is a function only of r. When where i and j refer to the components and V is the control
introduced in Equations (6) and (4), this form yields volume. The set of contact points in a wet material is
a nice fit for the capillary force obtained from direct extended to include gap capillary bridges. From Eq. 8,
integration of the Laplace-Young equation by setting we get the following expression for the stress p in the
R$ = 2Ri R j /(Ri + R j ), h(r) = r−1/2 and c % 0.9. direction of extension:
Figure 1 shows the plots of Eq. 4 for three different
values of the liquid volume Vb and size ratio r together p = nw ' fn !( , (9)
with the corresponding data from direct integration. The
forces are normalized by $ R and the lengths by & . The where nw is the number of bonds per unit volume, f n is
data collapse on the same plot, indicating again that the the normal force and ! is the distance between particle
force $ R and the expression of & in Eq. (6) characterize centers. The symbol '· · · ( designs averaging over all
correctly the behavior of the capillary bridge. Finally, the bonds in the control volume V .
capillary force can be expressed in the following form: Our simulations show that the fraction of liquid bonds
 in the range ! n > 0 (gap liquid bridges) is always below
 −$ R for !n < 0 15%. This means that most tensile stresses occur at con-
fnc = −$ R e−!n /& for 0 ≤ !n ≤ !nmax . (7) tact bonds and, as a result, the capillary failure threshold

0 for !n > !nmax f0 is far more important for the failure of a wet material
than the de-bonding distance ! nmax . We may thus get a manifests itself mainly through the average number of
good estimation of the tensile strength for a wet particle bonds per particle z. In particular, the cohesion c m at
assembly by replacing f n in Eq. 9 by the capillary force saturation corresponds to the saturation of z as the water
threshold f 0 and by assuming, for simplicity, that the ma- content w is increased. In fact, when a certain amount of
terial is isotropic (a correction due to anisotropy can then water is homogeneously distributed in the whole sample
easily be made): within the de-bonding distance, one finds that z increases
with w and saturates beyond a certain amount of water.
nw
' th = ' f0 !(, (10)
D
where D is space dimension. The bond density n w is BI-PERCOLATING STRUCTURE OF
equal to half the average number of bonds per particle SELF-STRESSES
divided by the free volume, i.e. the mean volume of a
Voronoi cell surrounding the particle. The latter is simply In an unconfined assembly of dry rigid particles, no self-
the average particle volume (1/6)% 'd 3( divided by the stresses occur and the forces vanish at all contacts. How-
solid fraction ) . Introducing these expressions in Eq. 10 ever, we have seen that the presence of liquid bonds in
and using Eq. 2, we get a wet granular material induces tensile and compressive
√ forces although the average force is zero. In other words,
3$ ) z '(R1 + R2) R1 R2 )( 3 $)z the grains keep together to form a self-sustained struc-
' th = 3
= s , (11)
%D 'd ( 2% D 'd( ture in the absence of confining stresses. In general, var-
ious loading histories such as consolidation or differen-
where z is the average number of bonds per particle, and tial particle swelling can induce self-stresses in a cohe-
we have sive packing [11, 12]. In our system, the self-stresses ap-
'd 1/2 ('d('d 3/2 (
s= . (12) pear during relaxation. This is obviously a consequence
'd 3 ( of the tensile action of capillary bonds bridging the gaps
In derivation of the expression of s, it is assumed that between neighboring particles within the debonding dis-
the particle radii R 1 and R2 are not correlated. It is easy tance. We focus here on the structure of self-stresses in-
to see that for a uniform size distribution, s varies from duced by capillary bonds.
8/15 to 1 as the smallest particle size increases from 0 For a local description of self-stresses we need to
to the mean particle size. Eq. 11 is similar to the expres- characterize the stress transmission at the particle scale
sion proposed first by Rumpf for monodisperse materi- as the smallest scale at which the force balance condition
als without the s prefactor, and recently derived from the is defined for rigid particles. Although the stress tensor is
stress tensor by Gröger et al. [10]. Our equation 11 ac- by definition a macroscopic quantity, it can be shown that
counts for polydispersity and the correlation between the an equivalent quantity ' i , called ‘particle stress’, can be
capillary force threshold f 0 and the particle size d. defined for each particle i of a granular packing in static
As in experiments, our simulations show that the in- equilibrium [13, 14]:
ternal angle of friction µ is practically independent of
1
water content. Hence, from Eq. 11 we get the following ('i )*+ =
Vi ( f*i j r+i j , (14)
expression for the Coulomb cohesion c th = µ' th : j*=i

3 )z where ri j is the position of the contact-point of the force

cth = µ$s . (13) fi j of particle j on particle i, and * and + design the
2% D 'd(
Cartesian components. Vi is the free volume of particle i,
We found that the Coulomb cohesion calculated from the the sum of the particle volume and a fraction of the pore
simulations of polydisperse materials is in good agree- space Vi = % di3 /6, , where di is the particle diameter, and
ment with the theoretical estimate by Eq. 13. It is worth , is the solid fraction of the packing. The sum of particle
noting that important discrepancy would arise if the poly- stresses 'i weighted by the corresponding relative free
dispersity were ignored; i.e. with s = 1. On the other volumes Vi /V tends to the Cauchy stress tensor as the
hand, Eq. 13 suggests that the Coulomb cohesion in- number of particles in a control volume V increases.
creases nonlinearly with water content as the average From the particle stresses we get the particle pressures
bond connectivity z (not shown in this paper; for details pi = 13 (3* =1 ('i )** . Each particle can take on positive
see [9]). The simulations are in good agreement with this or negative pressures according to the forces exerted
prediction. by neighboring particles. The pdf of particle pressures
It is important to note that the water content does not is displayed in Fig. 2 for the unconfined sample. The
enter the above expression of c th . The only parameter pressures have been normalized by a reference pressure
related to water is $ . This suggests that the water content p0 = f0 /'d(2 . The distribution is symmetric around and
FIGURE 2. Probability density function of particle pressures
normalized by the reference pressure p0 (see text) in the uncon-
fined wet packing.

FIGURE 3. The unconfined wet packing with negative

(white) and positive (black) particle pressures.
peaked on zero pressure, and each part is well fit by an
exponential form.
Zero particle pressure corresponds to a state where a
For wet granular media with a homogeneous distribu-
particle is balanced under the combined action of ten-
sile and compressive forces. Such particle states are not tion of liquid, we showed the nontrivial organization of
marginal here and they reflect a particular stress trans- particle pressures in the form of two separate percolating
phases of tensile and compressive particle pressures with
mission in a wet packing. The positive and negative
particle pressures may form separate phases mixing to- an interphase at zero pressure.
gether at large scales or mix together at the particle scale.
Fig. 3, displaying the packing where the positive and
negative pressures are represented in different colors, REFERENCES
credits rather the first scenario. We observe that the parti-
1. M. A. Erle, D. C. Dyson, and N. R. Morrow, AIChE Journal
cles of either positive or negative pressure appear as two 17, 115 (1971)
separate phases each percolating throughout the system. 2. T. Mikami, H. Kamiya, and M. Horio, Chemical
The morphology of each phase is approximately filamen- Engineering Science 53, 1927 (1998)
tary with variable thickness and a large interface between 3. F. Soulié, F. Cherblanc, M. S. El Youssoufi, and C. Saix,
them. A detailed analysis of this structure shows that the Int. J. Numer. Analyt. Meth. Geomech. 30, 213 (2006)
particles at the interface between the two phases have a 4. V. Richefeu, M. S. El Youssoufi, R. Peyroux and F. Radjaï,
Int. J. Numer. Anal. Meth. Geomech. 32(11), 1365 (2007)
weak pressure and the largest negative or positive pres- 5. C. Willett, M. Adams, S. Johnson, and J. Seville, Langmuir
sures are located at the heart of each phase [14]. 16, 9396 (2000)
6. L. Bocquet, E. Charlaix, F. Restagno, C. R. Physique 3,
207 (2002)
CONCLUSION 7. S. Herminghaus, Advances in Physics 54, 221 (2005)
8. G. Lian, C. Thornton, and M. J. Adams, Chem. Eng. Sci.
53, 3381 (2004)
In this paper, we presented an analytical expression of 9. V. Richefeu, M. S. El Youssoufi and F. Radjaï, Phys. Rev.
the capillary force as a function of geometrical and ma- E 73, 051304 (2006)
terials parameters of a liquid bridge, and it was shown 10. T. Gröger, U. Tüzün, and D. Heyes, Powder Technology
that it provides excellent fit for the data from direct [133], 203 (2003)
integration of Laplace-Young equations. By means of 11. F. Radjaï, I. Preechawuttipong, R. Peyroux, in Continuous
three-dimensional discrete element simulations includ- and Discontinuous Modelling of Cohesive-Frictional
Materials, edited by P. A. Vermeer, S. Diebels, W. Ehlers,
ing our capillary law, we analyzed shear strength and H.J. Hermann, S. Luding, E. Ramm (Springer, Berlin,
stress transmission in wet granular packings submitted 2001) p. 149
to very low confinement pressures. 12. M. S. El Youssoufi, J.-Y. Delenne, and F. Radjaï, Phys.
Starting with the expression of the stress tensor, we Rev. E 71, 051307 (2005).
introduced a novel expression for the Coulomb cohesion 13. L. Staron, F. Radjaï, and J.-P. Vilotte, Eur. Phys. J. E 18,
as a function of material and structural parameters. This 311 (2005)
14. V. Richefeu, F. Radjaï and M. S. El Youssoufi, Eur. Phys.
expression extends the classical model of Rumpf to poly- J. E 21, 359-369 (2006)
disperse materials.