Sei sulla pagina 1di 15

698

Network model for hydraulic conductivity of sand


bentonite mixtures
Tarek Abichou, Craig H. Benson, and Tuncer B. Edil

Abstract: A network formulation was used to model the hydraulic conductivity of sandbentonite mixtures (SBMs) as
a function of bentonite content. The sand particles were assumed to be spheres, and their arrangement was defined using a discrete element model simulating sand particle interactions. Pores between the spheres were approximated as a
network of straight capillary tubes. The space defined by the spheres was divided into a collection of neighboring tetrahedrons, and the geometry of the tetrahedrons was used to define tube diameters and lengths in the network. Hydraulic
heads throughout the network were computed by solving a system of equations describing flow through the tubes. Hydraulic conductivity of the network was calculated as the rate of flow per unit area for a given network of tubes driven
by a one-dimensional hydraulic gradient. Bentonite was introduced into the network in several schemes to simulate
SBMs. SBMs prepared with powdered bentonite were modeled as a packing of sand, where the sand particles are
coated with bentonite (grain coating model and pipe blocking model), whereas SBMs prepared with granular bentonite
were modeled as a packing of sand with bentonite occupying pores between the sand particles (junction blocking
model). The relationship between hydraulic conductivity and bentonite content obtained from the grain coating model
was similar to that measured on sand powdered bentonite mixtures. A comparable relationship was also obtained for
hydraulic conductivities predicted with the junction blocking model using a size-based filling approach and hydraulic
conductivities measured on sand granular bentonite mixtures.
Key words: sandbentonite mixtures, network models, hydraulic conductivity, degree of bentonation, bentonite distribution.
Rsum : On a utilis une formulation de rseau pour modliser la conductivit hydraulique des mlanges de sablebentonite (MBSs) en fonction de la teneur en bentonite. On a suppos que les particules de sable taient sphriques, et
leur arrangement a t dfini au moyen dun modle dlments discrets simulant les interactions entre les particules de
sable. On a considr que les pores entre les sphres se rapprochaient approximativement dun rseau de tubes capillaires droits. Lespace dfini par les sphres a t divis en une collection de ttradres voisins et la gomtrie des ttradres a t utilise pour dfinir les diamtres et les longueurs des tubes dans le rseau. Les charges hydrauliques dans le
rseau ont t calcules en partant de la solution dun systme dquations dcrivant lcoulement travers les tubes.
La conductivit hydraulique du rseau a t calcule comme tant le dbit par unit de surface pour un rseau donn
de tubes soumis un gradient hydraulique unidimensionnel. On a introduit de la bentonite dans le rseau de diffrentes
manires pour simuler des MBSs. Des MBSs prpars avec de la bentonite en poudre ont t modliss comme un
bourrage de sable o les particules de sable ont t enrobes de bentonite (Modle denrobage des grains et Modle de
blocage des tubes), dans lesquels les MBSs prpars avec de la bentonite en grains ont t modliss comme un bourrage de sable avec la bentonite occupant les pores entre les particules de sable (Modle de jonction de blocage). La relation entre la conductivit hydraulique et la teneur en bentonite obtenue du modle denrobage des grains tait
semblable celle mesure sur les mlanges de sable et de poudre de bentonite. On a aussi obtenu une relation comparable pour les conductivits hydrauliques prdites avec le modle de jonction de blocage en utilisant une approche de
remplissage base sur les dimensions et les conductivits hydrauliques mesures sur des mlanges de sable et de bentonite en grains.
Mots cls : mlanges sable-bentonite, modles de rseau, conductivit hydraulique, degr de caractre bentonitique, distribution de bentonite.
[Traduit par la Rdaction]

Abichou et al.

712

Received 15 April 2003. Accepted 20 January 2004. Published on the NRC Research Press Web site at http://cgj.nrc.ca on
25 August 2004.
T. Abichou.1 Department of Civil and Environmental Engineering, Florida State University, Tallahassee, FL 32310, USA.
C.H. Benson and T.B. Edil. Department of Civil and Environmental Engineering, University of Wisconsin-Madison,
2214 Engineering Hall, 1415 Engineering Drive, Madison, WI 53706, USA.
1

Corresponding author (e-mail: abichou@eng.fsu.edu).

Can. Geotech. J. 41: 698712 (2004)

doi: 10.1139/T04-016

2004 NRC Canada

Abichou et al.

Introduction
Almost all studies on the hydraulic conductivity of sand
bentonite mixtures (SBMs) focus on their use as hydraulic
barriers (Lundgren 1981; Chapuis 1981, 1990; Abeele 1986;
Daniel 1987; Garlanger et al. 1987; Kenney et al. 1992;
OSadnick et al. 1995; Gleason et al. 1997; Howell and
Shackelford 1997; Kraus et al. 1997; and others). The objective generally is to determine the quantity of bentonite required to obtain suitably low hydraulic conductivity. The
hydraulic conductivity of SBMs typically ranges from 1
106 to 1 109 cm/s, with lower hydraulic conductivities
generally associated with higher bentonite contents.
The traditional approach to determine the required bentonite content is experimental. A series of specimens is prepared, their hydraulic conductivity is measured, and then the
bentonite content to achieve the target hydraulic conductivity
is obtained from the experimental data. A second approach
is to predict the relationship between hydraulic conductivity
and bentonite content using a model that simulates pores filling with bentonite. A limited number of tests are then conducted to check that the bentonite content predicted from the
model is appropriate. A model of this sort can also be used
to understand how the hydraulic conductivity of clean sand
transitions to that of clay as the pores in the sand fill with
bentonite.
A network model is described in this paper that relates the
hydraulic conductivity of SBMs prepared with clean uniform
sands and powdered or granular bentonite to the bentonite
content. The model was developed to understand how the
hydraulic conductivity of SBMs is related to changes in
microstructure that occur as the pores in the sand are filled
with bentonite. The model consists of extracting a pore network from packings of spheres, introducing bentonite into
the network using different schemes, and computing the hydraulic conductivity of the network.

Models of SBMs
Models for predicting the hydraulic conductivity of SBMs
generally assume that the mixture is ideal, i.e., bentonite
fills all the pore spaces between the sand particles. The sand
particles are assumed to be nonconductive, and the hydraulic
conductivity of the bentonite controls the hydraulic conductivity of the mixture (Chapuis 1990; Kenney et al. 1992;
Mollins et al. 1996; Sllfors and berg-Hgsta 2002).
Chapuis (1990) describes an ideal-mixture model where
the swollen bentonite is assumed to fill all of the pore space
between the sand particles as water becomes available. A
portion of the water is assumed to be fixed to the surface of
bentonite particles, rendering the water immobile. The remaining water is also associated with the bentonite but is
mobile. The mobile water represents the portion of the pore
space available for flow, which is referred to as the efficient
porosity. Chapuis proposed two empirical equations relating
hydraulic conductivity to efficient porosity for ideal mixtures and compared hydraulic conductivities predicted by the
equations with data for SBMs prepared with mostly uniform
soils having 2%15% nonplastic fines and bentonite.
Kenney et al. (1992) theorize that an ideal SBM is a twocomponent mixture of sand and saturated bentonite where

699

all of the water is associated with the bentonite. The matrix


of the saturated bentonite is assumed to be continuous and
the sand particles are assumed to be impervious inclusions
in the bentonite matrix. The fabric of the bentonite is assumed to be unaffected by the presence of sand particles.
Under these conditions, the hydraulic conductivity of the
mixture is controlled by the hydraulic conductivity of the
bentonite fraction. Kenney et al. also indicate that the hydraulic conductivity of the bentonite is a unique function
of its void ratio. Thus, the hydraulic conductivity of ideal
SBMs can be computed as the product of the total porosity
of the mixture and the hydraulic conductivity of the bentonite at its void ratio in the mixture.
Sllfors and berg-Hgsta (2002) investigate the effects
of degree of compaction on the hydraulic conductivity of
SBMs. They use the ideal-mixture concept to derive a new
index quantity, k1, which combines the dry unit weight of the
mixture and the bentonite content. SBMs were prepared with
powdered bentonite and well-graded medium sand. The parameter k1 was calculated for all mixtures and correlated
with hydraulic conductivity.
Mollins et al. (1996) developed empirical power law relationships between the void ratio of the bentonite in an SBM
and the vertical effective stress. The power laws were calibrated using data from one-dimensional swelling and hydraulic conductivity tests on SBMs containing 5%, 10%, and
20% bentonite at varying effective stress. The SBMs were
prepared with a silty fine angular quartz sand and sodium
bentonite. Mollins et al. indicated that their model cannot be
used for mixtures with low bentonite content because the
bentonite is not evenly distributed in the SBM.

Network model
Ideal-mixture models can predict the hydraulic conductivity of SBMs reasonably well when the SBMs contain
enough bentonite to satisfy the ideal-mixture assumptions.
However, the hydraulic conductivity of SBMs with low bentonite content and uneven bentonite distribution cannot be
predicted using ideal-mixture models. In contrast, network
models can be used to simulate nonideal mixtures associated
with low bentonite content because they explicitly account
for filling of the pores with bentonite. Network models can
also be used to study how the hydraulic conductivity of an
SBM changes as the pores in the sand are filled and how the
hydraulic conductivity depends on the type of pore filling.
A network model was developed in this study to evaluate
how the hydraulic conductivity of SBMs transitions from
that of sand to that of bentonite. The model was used to examine mechanisms responsible for the decrease in hydraulic
conductivity and to identify which of the mechanisms is
associated with powdered and granular bentonites. The network model developed in this study employs four steps:
(i) generating packings of equal-sized spheres to simulate
sand grains, (ii) extracting networks of capillary tubes representing the pore space in the packing of spheres, (iii) calculating the hydraulic conductivity of the entire network using
a capillary tube model, and (iv) evaluating how introducing
bentonite into the network in different schemes affects
hydraulic conductivity. The methods used to introduce the
2004 NRC Canada

700

Can. Geotech. J. Vol. 41, 2004

Fig. 1. Typical tetrahedron (a) and branches of network (b).

bentonite are based on pore-scale observations reported by


Abichou et al. (2002).
Sand pack generation
A three-dimensional discrete element model (DEM) developed by Horner (1997) was used to generate packings of
spherical particles representing sand. The objective was to
develop a reasonably accurate representation of the packing
of sand particles and the pore spaces between these particles.
In the DEM, soil particles are modeled as points in space,
and each carries with it an estimate of the state of the soil
mass within its vicinity (Horner 1997). For the application in
this study, particles were randomly selected for placement
on a lattice, with the spacing between the centers of particles
being large enough to minimize the initial interparticle
forces. A simulation was then conducted where the particles
were rained into a rigid-wall container until the particles
achieved static equilibrium. When denser packings were desired, walls of the simulated container were pushed to new
positions, causing the particles to achieve a denser equilibrium state.
There is a distinct difference between the DEM approach
and an actual SBM which deserves mention. In an SBM, the
sand and bentonite are mixed together and then compressed.
In contrast, the collection of sand particles is compressed in
the DEM, and then bentonite is added. The presence of bentonite may affect friction that develops between the sand
grains and to some degree the packing of sand grains. This
effect is believed to be small, however, unless the bentonite
content is so large that the sand grains become independent
inclusions in the bentonite matrix. Data in Mollins et al.
(1999) and Goodhue et al. (2001) support this supposition.
Both studies show that the shear strength of SBMs remains
equal to that of the sand alone at low bentonite contents.
This suggests that the intergranular friction is not affected
appreciably by the bentonite content until the bentonite content is large.
For simplicity, the DEM code developed by Horner was
used to generate packings of equal-sized spheres, although
the code allows for up to five grain sizes. The output con-

sisted of the coordinates of the center of each sphere in the


packing. Packings were obtained for spheres with a diameter
of 0.2 mm (fine sand), 1.2 mm (medium sand), and 3.4 mm
(coarse sand). Porosities of the packings were 0.29 (dense
packing), 0.36 (medium packing), and 0.44 (loose packing).
Capillary tube network
Once a sphere packing is generated, the space defined by
the spheres was divided into tetrahedrons using Voronoi
Delaunay (VD) triangulation. Each tetrahedron is formed
by the closest neighboring four spheres as illustrated in
Fig. 1a and has four faces (i.e., each side of the tetrahedron).
Each tetrahedron has four windows (one on each face)
formed by the space between the three spheres on each face
and one internal pore in the center of the tetrahedron. The
window in each face is equivalent to a pore throat, and the
internal pore in the center of the tetrahedron is equivalent to
a pore. This arrangement can be seen as four tubes (one
from each face) forming a junction at the center of the pore
(Fig. 1b). This technique transforms the pore space defined
by the packing into a network of capillary tubes where the
interior of a tetrahedron corresponds to a junction of tubes in
the network (Mason 1971), as shown in Figs. 1a and 1b. A
code developed by P.K. Sweby of the Department of Mathematics at Reading University (Reading, UK) and modified
by Bryant et al. (1993a, 1993b) transforms the centers of
spheres into a network geometry.
Once the geometry of the network of tubes is defined, the
diameter and length of each tube in the network are determined. Bryant et al. (1993a) found that accurate estimates of
flow are obtained when the area of each tube is assigned
using the arithmetic mean of the radius of the largest circle
between the spheres surrounding a pore (rm in Fig. 2a) and
the radius of a circle having equivalent area as the pore
space between the same adjacent spheres (req in Fig. 2b).
This arithmetic mean is referred to as the effective radius
(reff) and was used to define the tube radius in this study.
Since packings were obtained with equal-sized spheres,
the tube diameters are similar in size to the distance between
the centroid and the edge of the tetrahedron. This condition
2004 NRC Canada

Abichou et al.
Fig. 2. Schematic of largest circle (a) and equivalent circle
(b) for pipe sizing.

701
Fig. 3. Distribution of tube radius (a) and length (b) in a network extracted from dense, medium, and loose packings of 1000
spheres with a radius of 0.2 mm.

ings of 1000 spheres with a diameter of 0.2 mm. The distribution of tube radius shifts in the direction of smaller tubes
as the packing becomes denser (Fig. 3a). Also, denser packings tend to have more short tubes (Fig. 3b).

leads to double counting of the tube length in the central region where the tubes intersect if the tube length is assumed
to be equal to the distance between the central pore and the
edge of the tetrahedron. This overestimation was corrected
by applying an algorithm developed by Bryant et al. (1993a)
that shortens the tube length so that the volume of tubes in
the network equals the volume of pores in the packing. The
tube connecting two adjacent junctions of the network is
then comprised of two sections where each has a different
length and radius.
Examples of the distributions of tube length and diameter
are shown in Fig. 3. These distributions were obtained from
the networks extracted from dense, medium, and loose pack-

Network hydraulic conductivity


If flow in the network is laminar, then the flow rate (Q) in
each tube can be described by the Hagen-Poiseuille equation:
[1]

Q=

4
Hreff
8L

where H is the difference in total head between the two


ends of the tube, reff is the effective radius of the tube, L is
the length of the tube, and is the viscosity of the fluid.
The flow rate can also be written in terms of a hydraulic
conductance Kt (units of L2/T) of the tube as
[2]

Q = Kt H
2004 NRC Canada

702

Can. Geotech. J. Vol. 41, 2004

Fig. 4. Conceptual sketch of network with detail showing a junction, four intersecting tubes, and four adjacent junctions. Upper and
lower horizontal surfaces are constant-head boundaries. Vertical surfaces are no-flow boundaries. For clarity, this sketch contains fewer
tubes and junctions than that of an actual network.

Hydraulic conductance was used instead of hydraulic conductivity because conductance is easier to work with numerically. Equating eqs. [1] and [2] yields the hydraulic
conductance Kt as a function of the tube geometry:
[3]

Kt =

4
reff
8L

Pathways between two neighboring tetrahedrons are represented by a tube with two contiguous sections having
different conductances (K1, K2). The conductance of the pathway is represented by an equivalent tube with hydraulic conductance Ke :
[4]

1
1
Ke =
+

K1 K2

tions are applied by assigning hydraulic heads at the ends of


the tubes emanating from the influent and effluent faces of
the network (upper and lower surfaces shown in Fig. 4). The
head at each junction of the network is obtained by solving
the system of equations defined by application of eq. [5] to
each junction. The system was solved iteratively by Gauss
Sidel successive over-relaxation.
Once the heads are determined at each junction, the flow
rate through the entire network (QT) is estimated by summing the flows in the tubes emanating from the up-gradient
or down-gradient faces of the network. The hydraulic conductivity of the entire network (Kn) is estimated by applying
Darcys law to the entire network:

[7]

Kn =

QTLn
Hn A

The hydraulic head at the end of each tube in the network


is determined by applying conservation of mass at each
junction. Conservation of mass requires the net flow rate at
each junction to be zero, i.e.,

where Ln is the linear length of the network (influent face to


the effluent face), Hn is the drop in total head across the
network, and A is the gross area of the soil at the effluent
face (see Fig. 4).

[5]

Introduction of bentonite

Qij + Qik + Qil + Qim = 0

where Qij, Qik, Qil, and Qim are flow rates through the four
tubes (ij, ik, il, and in) that meet at junction i and emanate
from junctions j, k, l, and n as shown conceptually in Fig. 4.
The flow rate through tube Qij is
[6]

Qij = Kij(Hi Hj)

where Hi is the total head at junction i, Hj is the total head at


the other end of the tube (junction j), and Kij is the conductance of tube ij defined by eqs. [3] and [4]. Boundary condi-

Grain coating model (GCM)


Bentonite is introduced into the network in a manner consistent with observations of the microstructure of mixtures of
glass beads and bentonite (powdered and granular) made by
Abichou (1999) and Abichou et al. (2002) using optical and
scanning electron micrographs. For mixtures prepared with
powdered bentonite, bentonite coats the grains at low bentonite content and, in the presence of water, the bentonite
swells and fills some of the pore spaces between the grains.
2004 NRC Canada

Abichou et al.

703

As the bentonite content increases, the thickness of the coating increases and more of the area available for flow is
blocked by swollen bentonite. The pores are filled by a continuous matrix of bentonite only at relatively high bentonite
contents (>8%).
The grain coating model (GCM) was developed to simulate the pore structure of SBMs prepared with powdered
bentonite. Radii of the spheres in the original packing were
increased to account for the bentonite coating without
changing the location of the centers of the spheres, i.e.,
[8]

Fig. 5. Illustration of method used to reduce area of flow in


grain coating model (a) and scanning electron micrograph showing coating of grains with powdered bentonite (b). Unshaded
portion of sphere in (a) corresponds to original sand grain.
Shaded ring around sphere corresponds to bentonite coating.

Rib = Ri + Ri

where Ri is the radius of the ith sphere, Rib is the radius of


the ith sphere coated with bentonite, and Ri is the thickness
of the swollen bentonite coating (Fig. 5a). As the sphere size
increases because of the bentonite coating, the area available
for flow in each window of each tetrahedron decreases. The
similarity between the observed structure and the idealized
scenario assumed in the model is shown in Fig. 5b. After adjusting the sphere radii using eq. [8], a new tube is fitted to
each window of each tetrahedron as described in the previous section.
The total volume of coating in the network equals the
total volume of hydrated bentonite in the mixture, and the
ratio of the total volume of hydrated bentonite to the volume
of the pore space in the original packing (i.e., pore space
without bentonite) is defined as the degree of bentonation.
The degree of bentonation (B) is a measure of how much of
the original pore space is occupied by hydrated bentonite,
whereas conventional bentonite content is the ratio of weight
of dry bentonite to weight of the sand particles. As grains
are coated with more bentonite, B increases. The upper
bound on B is 1, indicating that the entire pore space in the
sand matrix is occupied by bentonite.
The coated tubes are composed of two zones. An equivalent hydraulic conductance (Keq) of the coated tube is calculated by equating the flow through the coated tube, Q, to the
sum of the flow rate through the coating (Qc) and the flow
rate through the remainder of the tube (Qp):
[9]

Q = kc

H
Ac + K H = c Ac + K H = Keq H
L

where kc and Ac are the hydraulic conductivity and crosssectional area, respectively, of the bentonite coating; K is the
conductance of the inner uncoated portion of the tube (as
calculated in eq. [4]); and H is the difference in total head
across the tube.
As the bentonite content increases, the area Ac increases
until it reaches the area of the original tube (i.e., without
bentonite). When this occurs, the tube is said to be
blocked with bentonite. When four intersecting tubes are
blocked, the junction where they meet is blocked. Tube and
junction blocking are a direct result of the coating of the
sand grains with bentonite and are not independent mechanisms, as is the case in other models described in the next
sections.
In the GCM, and in all other models, the hydraulic conductivity of bentonite was assumed to be constant. This assumption was made for simplicity. In reality, the hydraulic
conductivity of bentonite will be affected by the size of the

pores and the degree of bentonation. Both will affect the


amount of swelling that can occur in the pore space and the
void ratio of bentonite. Given the lack of knowledge on how
pore size and bentonite content affect the void ratio and the
hydraulic conductivity of the bentonite fraction in SBMs, the
simplification of constant hydraulic conductivity was considered to be warranted.
Tube blocking model (TBM)
In the GCM, tube blocking is a direct result of increasing
the thickness of the bentonite coating around each sphere.
Another alternative to model SBMs prepared with powdered
bentonite is to assume that bentonite fills the window between three adjacent spheres, resulting in tubes that are either blocked or unblocked. When a tube is blocked with
bentonite, the tube is assumed to be entirely filled with bentonite.
2004 NRC Canada

704

Can. Geotech. J. Vol. 41, 2004

Fig. 6. Scanning electron micrograph of SBM prepared with glass beads (diameter = 1.2 mm) and granular bentonite (bentonite content = 5%).

The tube blocking model (TBM) was developed to simulate this condition. The probability of being blocked is assumed to be independent of the size and location of the tube,
and a tube cannot be partially blocked. A Bernoulli random
number generator was used to define whether a tube was
blocked or unblocked. The Bernoulli generator returns a 0
(empty tube) or 1 (blocked tube) depending on the probability of blockage (p) for a given tube. If a network has Nt
tubes and Ft is the number of tubes blocked with bentonite,
then the probability of success (p) of the Bernoulli random
variable is Ft/Nt. The volume of hydrated bentonite in the
network is calculated as the sum of the volume of the tubes
containing bentonite. The ratio of the volume of tubes containing bentonite to the volume of all tubes in the network is
the degree of bentonation (B).
Junction blocking model (JBM)
The junction blocking model (JBM) was developed to
simulate SBMs prepared with granular bentonite. Based on
observations reported by Abichou et al. (2002), bentonite
granules occupy the space between the grains (Fig. 6) and
then swell to fill the available space when hydrated. As the
bentonite content increases, the number of granules increases, leading to more spaces being filled with bentonite.
The GCM and the JBM do not provide good physical representations of mixtures with granular bentonite because the

bentonite does not coat the sand grains as reported by


Abichou et al.
The JBM simulates pore filling by granular bentonite by
filling the four tubes that meet at a junction with bentonite.
Therefore the conductance of each tube connected to a
blocked junction is assigned the hydraulic conductance of a
tube filled with bentonite. The degree of bentonation (B)
equals the volume of the tubes filled with bentonite divided
by the total volume of tubes in the network, which is equivalent to the volume of blocked junctions divided by the original volume of pore space.
Two scenarios were considered. In the first scenario, the
bentonite is assumed to occupy larger pores first and then
fills smaller and smaller pores. This scenario is referred to as
the size-based filling scenario (JBM-SB). This scenario can
also represent the case where the bentonite granules are
large and cannot fit in smaller pores. The pore volume in
each tetrahedron is calculated, and then the pores are sorted
based on their volume. The largest pore (i.e., junctions in the
network) is blocked first, followed by sequentially smaller
pores. The second scenario assumes that the distribution of
bentonite is independent of the size of the pore and that the
probability of a pore being filled with bentonite is equal for
all pores. This scenario is referred to as the random filling
scenario (JBM-R). The Bernoulli random number generator
was used for this model as well. Tubes meeting at the junc 2004 NRC Canada

Abichou et al.

705

Fig. 7. Hydraulic conductivity of clean sands (no bentonite) versus median sand grain diameter. Predictions made with network model
are shown as solid circles. Hydraulic conductivities computed with Hazen, Harleman, and KozenyCarmen equations are shown as
lines.

tions where the Bernoulli random number generator returns


a 1 were blocked with bentonite and their conductance was
assigned to that of a tube filled with bentonite. Tubes meeting at junctions where the Bernoulli random number generator returns a 0 were free of bentonite.

measured hydraulic conductivities (Fig. 7). As the size of


the particles gets smaller, however, hydraulic conductivities
predicted by the model are slightly lower than the measured
hydraulic conductivities and the hydraulic conductivities
predicted using the equations from the literature.

Simulation results

Grain coating model


The general relationship between hydraulic conductivity
and degree of bentonation for a packing with a sphere diameter of 0.2 mm and a porosity of 0.36 (i.e., porosity of
spheres without bentonite) is shown in Fig. 8a. The percentage of blocked tubes and junctions in the network is also
shown in Fig. 8a as a function of degree of bentonation B.
The same relationships for packings with sphere diameters
of 0.2 mm (fine sand) and 3.4 mm (coarse sand) and porosities ranging from 0.29 (dense sand) to 0.44 (loose sand) are
shown in Fig. 8b.
There are three distinct zones in Fig. 8a. In zone 1, as the
volume of bentonite coating increases, the effective size of
the tubes in the network decreases, leading to a slight reduction in hydraulic conductivity. All the tubes are still unblocked in zone 1, however, and the hydraulic conductivity
remains high. In zone 2, tubes start to become blocked at
B = 0.5, and the hydraulic conductivity begins decreasing
appreciably as more tubes become completely filled with
bentonite. Zone 3 corresponds to the region where B > 0.8.
At this high degree of bentonation, tubes continue to get
blocked with bentonite at a higher rate. At the same time,
the additional coating starts to block entire junctions, i.e.,
four tubes get blocked at the same time.

Particles in the simulations were chosen to simulate fine,


medium, and coarse sands based on the mean grain diameter
for these grain sizes in the Unified Soil Classification System (Holtz and Kovacs 1981). Porosities of the packings
were chosen to simulate loose, medium, and dense sand as
described by Mitchell (1993).
Packings without bentonite
Simulations were first performed on packings with identical porosities but comprised of 10, 100, and 1000 spheres.
The hydraulic conductivity predicted using the model was
the same for each of these packings, indicating that a packing with 1000 spheres was more than adequate to represent
the network. All subsequent simulations were conducted
with packings of 1000 spheres.
Hydraulic conductivities of 1000 sphere packings without
bentonite are shown in Fig. 7 for several sphere diameters.
Also shown in Fig. 7 are hydraulic conductivities of glass
spheres (measured by Chu and Ng 1989) and uniform fine
sand (measured by Abichou 1999). Hydraulic conductivities
predicted by the Hazen, Harleman, and KozenyCarmen
equations are also shown in Fig. 7 as smooth curves. At
large grain size the model predictions are very similar to

2004 NRC Canada

706

Can. Geotech. J. Vol. 41, 2004

Fig. 8. (a) Hydraulic conductivity, percent blocked tubes, and percent blocked junctions predicted with grain coating model (GCM)
versus degree of bentonation for a packing with porosity of 0.36 and sphere diameter of 0.2 mm. (b) Variation of hydraulic conductivity with degree of bentonation for packings with porosity varying from 0.29 to 0.44 and sphere diameter of 0.2 and 3.4 mm.

The faster rate of tube blocking and the start of junction


blocking (both caused by grain coating) cause the hydraulic
conductivity to drop dramatically in zone 3 as more and
more of the permeable pathways through the network are
eliminated. When B > 0.8, only a few continuous flow paths
exist in the network that are not blocked with bentonite and
the hydraulic conductivity becomes very low (dropping to
7.1 1010 cm/s) when complete filling occurs (B = 1.0).
This hydraulic conductivity is lower than that of the benton-

ite alone because a portion of the gross area of flow is


blinded by the spheres.
Figure 8b shows similar trends of decreasing hydraulic
conductivity with increasing degree of bentonation B. Packings of smaller spheres and lower porosity have lower hydraulic conductivities for a given B. Denser packings also
reach low hydraulic conductivity at a slightly lower degree
of bentonation. Packings of smaller spheres also reach low
hydraulic conductivity at a lower degree of bentonation. At
2004 NRC Canada

Abichou et al.

707

Fig. 9. (a) Hydraulic conductivity and percent blocked tubes predicted with tube blocking model (TBM) versus degree of bentonation
for a packing with porosity of 0.29 and sphere diameter of 0.2 mm. (b) Variation of hydraulic conductivity with degree of bentonation
for packings with porosity varying from 0.29 (dense packing) to 0.44 (loose packing) and sphere diameter of 0.2 (simulating fine sand)
and 3.4 mm (simulating coarse sand).

full bentonation (B = 1.0), all hydraulic conductivities are


equal to that of bentonite corrected for the blinding spheres.
Figure 8b also shows that the hydraulic conductivities of
SBMs with coarse packings are almost four orders of magnitude higher than those of fine packings, when the degree of
bentonation is below 0.6. The difference in hydraulic conductivity between coarse and fine packings decreases, however, as

the degree of bentonation increases beyond 0.6. The hydraulic


conductivities of all packings are in the same order of magnitude when the degree of bentonation approaches 1.0.
Tube blocking model
The relationship between hydraulic conductivity and degree of bentonation predicted with the TBM is shown in
2004 NRC Canada

708

Fig. 9a for spheres with a diameter of 0.2 mm and a porosity


of 0.36. The percent blocked tubes in the network is also
shown in Fig. 9a as a function of degree of bentonation B.
The same relationships for packings with sphere diameters
of 0.2 mm (fine sand) and 3.4 mm (coarse sand) and porosities ranging from 0.29 (dense sand) to 0.44 (loose sand) are
shown in Fig. 9b.
Three distinct zones exist in Fig. 9a. In zone 1 (B = 0.0
0.25), tubes in the network begin to become blocked with
bentonite, leading to a gradual decrease in hydraulic conductivity. Most of the pathways through the network are unblocked, however. As B increases in zone 2 (B = 0.250.45),
more tubes are blocked with bentonite and the number of
unblocked flow paths decreases dramatically, leading to a
sharp decrease in hydraulic conductivity. For B > 0.45 (zone
3), the hydraulic conductivity approaches that of bentonite
because only a few flow paths through the network exist that
are unblocked. For B > 0.45, the hydraulic conductivities are
of the same order of magnitude and gradually decrease to
that of bentonite corrected for blinding by the spheres (7.1
1010 cm/s). In the limiting case (B = 1.0), all the flow paths
are filled with bentonite and the hydraulic conductivity is
7.1 1010 cm/s. Similar trends of decreasing hydraulic
conductivity with increasing B are shown in Fig. 9b for
packings with different porosities and grain sizes. Denser
packings or packings with smaller spheres reach low hydraulic conductivities at slightly lower degrees of bentonation
than the other packings.
Comparison of Figs. 8 and 9 shows that bentonite is used
more efficiently in the TBM than in the GCM. The hydraulic
conductivity predicted using the TBM reaches low values at
B = 0.45, whereas B > 0.8 was required to reach this condition with the GCM. The presence of bentonite always induces pipe blocking in the TBM. In contrast, the bentonite
coats the grains in the GCM but does not block tubes only
until the coating becomes adequately thick.
Junction blocking model
Hydraulic conductivity, percent blocked junctions, and degree of bentonation obtained from the JBM with size-based
filling are shown in Fig. 10a for spheres with a diameter of
0.2 mm and a porosity of 0.36. The percentage of blocked
tubes and junctions in the network is also shown in Fig. 10a
as a function of B. A similar graph for random filling is
shown in Fig. 10b. Results obtained using packings with different porosities and grain sizes are not shown for this model
but are presented in Abichou (1999). The results for other
porosities and grain sizes follow the same general trends as
those shown in Figs. 8 and 9.
As with the TBM, the relationships between hydraulic
conductivity and degree of bentonation B have three zones.
The hydraulic conductivity decreases slightly in zone 1 as B
increases to 0.5. Blocking the large pores does not affect the
hydraulic conductivity greatly because the hydraulic conductivity is controlled by smaller tubes in the network. In zone
2 (B = 0.50.8), the hydraulic conductivity decreases more
rapidly with increasing B (about two and one half orders of
magnitude). In zone 3 (B > 0.8), an abrupt decrease in hydraulic conductivity occurs as enough junctions controlling
the hydraulic conductivity are blocked to prevent continuous
pathways without bentonite through the network. Ultimately,

Can. Geotech. J. Vol. 41, 2004

at B = 1.0, the hydraulic conductivity reaches that of the


ideal mixture (7.1 1010 cm/s).
Three zones exist for random filling as shown in Fig. 10b.
In zone 1 (B = 0.00.32), the hydraulic conductivity
decreases more rapidly than in size-based filling because
smaller junctions in the network, which are bottlenecks to
flow, are being blocked at lower B with random filling. With
size-based filling, B must be >0.5 before these bottlenecks
are filled. In zone 2 (B = 0.51), the hydraulic conductivity
abruptly drops four orders of magnitude. It appears that, at
B = 0.51, the network becomes devoid of pathways that are
not blocked with bentonite. In zone 3, the hydraulic conductivity stays constant until the degree of bentonation reaches
0.8 and then gradually reaches the limiting case (B = 1.0)
where the hydraulic conductivity is equal to 7.1 1010 cm/s.

Comparison with experimental data


A comparison was made between hydraulic conductivity
predicted with the model and measured hydraulic conductivities of SBMs to ascertain whether the trends observed in
Figs. 810 are realized for SBMs prepared with powdered
and granular bentonite. The mixtures were prepared with
CETCO SS-100 (American Colloid Co., Arlington Heights,
Ill.), a powdered sodium bentonite ground to 70% passing
the No. 200 sieve (0.075 mm), and Benseal (Baroid Corp.,
Houston, Tex.), a granular sodium bentonite with a median
particle size of 1.1 mm. The free swell, as defined by American Society for Testing and Materials (ASTM) Standard
D5890, is 25 mL for the powdered bentonite and 32 mL for
the granular bentonite. Two uniformly graded sands were
used. One was a fine sand with a median particle diameter of
0.2 mm and a coefficient of uniformity (Cu) of 4. The other
was a medium sand with a median grain diameter of 1.2 mm
and Cu of 4.5.
Prior to mixing the sands and bentonite, the sands were
lightly sprayed with water to simulate the natural moisture
that commonly exists in the field. This moisture also caused
the bentonite to stick to the surfaces of the sand grains. A
known mass of bentonite (powdered or granular) was then
added to achieve the desired bentonite content (ratio of the
weight of dry bentonite to that of dry sand). The bentonite
was added in increments to the sand as the mixture was
blended in a large container using a hand trowel. Once all of
the bentonite was added, the entire mixture was mixed thoroughly in the same container until it appeared uniform. The
specimens were compacted in rigid-wall permeameters
(100 mm in diameter and 50 mm high). The desired dry unit
weight (sand porosity of 0.36) was achieved by compacting
a known weight of the mixture into the permeameter mold
(having known volume) in two equal lifts using a standard
Proctor hammer.
The hydraulic conductivity tests were performed directly
in the compaction mold using rigid-wall permeameters. The
falling-head method was used in accordance with ASTM
Standard D5856. The inflow and outflow burettes were covered with thin plastic film to minimize evaporation. Tests
were terminated when no trend existed in the hydraulic conductivity data, the last four hydraulic conductivity values
were within 25% of the mean, and inflow equaled outflow.
Hydraulic gradients were maintained between 28 and 32,
2004 NRC Canada

Abichou et al.

709

Fig. 10. Hydraulic conductivity and percent blocked junctions versus degree of bentonation predicted with junction blocking model
(JBM) for a packing having and original porosity of 0.36 and sphere diameter of 0.2 mm: (a) size-based filling scenario; (b) random
filling scenario.

and tap water was used as the permeant liquid. The permeameters did not include swell rings, so bentonite in these
mixtures could only swell into the pore space in the sand
matrix. Details of the hydraulic conductivity tests can be
found in Abichou et al. (2002).
After the hydraulic conductivity tests were terminated, a
100 cm suction was applied by a hanging column to drain
off excess water held by capillarity (i.e., water not adsorbed

by either sand or bentonite). The remaining volume of water


was assumed to be associated with the swollen bentonite.
Degree of bentonation was calculated as the sum of the volume of dry bentonite and the volume of water remaining in
the specimen divided by the original volume of pores in the
sand matrix. This assumption tends to overestimate the
amount of swell because some of the water that remains in
the specimen may be associated with the sand matrix and
2004 NRC Canada

710

Can. Geotech. J. Vol. 41, 2004

Fig. 11. Hydraulic conductivity ratio versus degree of bentonation predicted with grain coating model (GCM) and tube blocking model
(TBM) and measured for SBMs prepared with powdered bentonite and fine and medium sands (sand porosity = 0.36).

not available for swelling of bentonite. Additionally, some of


the water may also be adsorbed onto the bentonite without inducing any volume change. The magnitude of the overestimation is believed to be small, however.
SBMs prepared with powdered bentonite
A comparison of the predicted and measured relationships
between hydraulic conductivity and degree of bentonation is
shown in Fig. 11. The results are presented in terms of a hydraulic conductivity ratio (Kr), which is defined as the ratio
of the hydraulic conductivity of the SBM to the hydraulic
conductivity of the sand without bentonite. The measured
hydraulic conductivity of the samples prepared with fine sand
varied from 0.045 cm/s when B = 0.0 to 1.5 108 cm/s when
B = 1.0. The measured hydraulic conductivity of the samples
prepared with medium sand varied from 0.68 cm/s when B =
0 to 2 108 cm/s when B = 1.0. Predictions are only shown
for the GCM and TBM because these network models were
developed for powdered bentonite.
Similar relationships exist between hydraulic conductivity
and degree of bentonation for the actual SBMs and the predictions made with the GCM. The degree of bentonation at
which the hydraulic conductivity reaches lower values is
slightly overestimated by the GCM for the fine sand and is
estimated closely for medium sand. In contrast, the hydraulic conductivity predicted with the TBM reaches a low value
at a degree of bentonation significantly lower than occurs for
the actual SBMs. The closer agreement obtained with the
GCM is consistent with the microstructural observations
made by Abichou et al. (2002), i.e., powdered bentonite

coats the grains in an SBM rather than directly blocks the


pores as assumed in the TBM.
The effect of sand size in both models may reflect an artifact of the model formulation, which assumes that bentonite
has constant hydraulic conductivity regardless of the size of
the pores or the degree of bentonation. In reality, the bentonite may have lower conductivity in fine sand because smaller
pores will restrict swelling, resulting in a bentonite fraction
with lower hydraulic conductivity. As a result, the fine and
medium sands could have comparable Kr at a high degree of
bentonation. Moreover, if higher bentonite contents had been
used in the tests with fine sand, the measured Kr of the fine
sand may have dropped below that of the medium sand at a
similar degree of bentonation.
SBMs prepared with granular bentonite
Predicted hydraulic conductivities obtained with the JBM
(size-based and random filling) and the measured hydraulic
conductivities of the SBMs prepared with fine and medium
sand mixed and granular bentonite are shown in Fig. 12 versus degree of bentonation. The measured hydraulic conductivity of the samples prepared with fine sand varied from
0.045 cm/s when B = 0 to 1.9 108 cm/s when B = 1.0.
The measured hydraulic conductivity of the samples prepared with medium sand varied from 0.68 cm/s when B = 0
to 1.8 108 cm/s when B = 1.0. The relationship between
hydraulic conductivity and degree of bentonation predicted
with the JBM using size-based filling is similar to those
measured on the SBMs (especially the SBM with fine sand).
The comparison is poorer for the medium sand. Neverthe 2004 NRC Canada

Abichou et al.

711

Fig. 12. Hydraulic conductivity ratio versus degree of bentonation predicted with junction blocking model (JBM) with size-based filling and random filling along with measured hydraulic conductivities of SBMs prepared with granular bentonite and fine and medium
sand (porosity of sand = 0.36).

less, the similarity of the predicted and measured hydraulic


conductivity ratios suggests that the JBM with size-based
filling captures key mechanisms controlling the hydraulic
conductivity of SBMs prepared with granular bentonite.
Much poorer agreement is obtained with random filling.
The JBM with random filling predicts a large drop in hydraulic conductivity at B = 0.51, whereas the hydraulic conductivity of the SBMs only dropped appreciably for B >
0.75. Apparently, assuming an independent distribution of
bentonite does not reflect the bentonite distribution in SBMs
prepared with granular bentonite. Rather, the distribution of
bentonite in SBMs prepared with granular bentonite probably depends on the size of the granules, the size of the grains
of the sand, and maybe the bentonite content.

Summary and conclusions


A network model was developed to predict the hydraulic
conductivity of SBMs prepared with powdered and granular
bentonite. The model is based on a packing of equal-size
spheres created using a three-dimensional discrete element
model. The space occupied by the spheres is divided into a
collection of neighboring tetrahedrons, and the geometry of
the tetrahedrons is used to define tube diameters and lengths
in a pore network. The hydraulic conductivity of the network
is calculated by applying conservation of mass at each junction in the network. Bentonite was introduced into the network in several schemes to simulate SBMs prepared with
powdered and granular bentonite. The quantity of bentonite
in the network is characterized by the degree of bentonation,

B, which is the volume of swollen bentonite divided by the


volume of pores between the sand grains.
The schemes used to introduce the bentonite were selected
to mimic microstructural observations reported previously.
Mixtures of sand and powdered bentonite were modeled as a
packing of sand with each sand grain coated with a layer of
bentonite (grain coating model or GCM) or as a random distribution of bentonite occupying the pore throats between the
sand particles (tube blocking model or TBM). Mixtures of
sand and granular bentonite were modeled as a packing of
sand with bentonite granules occupying the pores between
the sand particles (junction blocking model or JBM).
Predictions made with the GCM show that bentonite reduces the hydraulic conductivity of an SBM in three steps.
Initially, the tube diameters decrease as the thickness of the
bentonite coating increases, leading to a gradual reduction
in hydraulic conductivity. However, all tubes remain unblocked. In the second step (B > 0.5), the hydraulic conductivity decreases appreciably as B increases because the grain
coatings are thick enough to cause blockage of some tubes.
In the third step (B > 0.8), additional coating induces blocking of entire junctions (blocking four tubes at a time), which
decreases the hydraulic conductivity abruptly. Predictions
made with the TBM also show that bentonite reduces the hydraulic conductivity of an SBM in three similar steps. First,
the hydraulic conductivity decreases slightly as B increases
to 0.5. As the degree of bentonation increases from 0.5 to
0.8, the hydraulic conductivity suddenly decreases by two
and a half orders of magnitude. At a degree of bentonation
greater than 0.8, enough junctions controlling the hydraulic
2004 NRC Canada

712

conductivity are blocked to prevent continuous pathways


(without bentonite), leading to a steep decrease in hydraulic
conductivity.
Predictions made using the GCM compare reasonably
well with hydraulic conductivities measured on SBMs prepared with fine and medium sand and powdered bentonite.
In contrast, the TBM predicted decreases in hydraulic conductivity at much lower B than was measured. Predictions
obtained with the JBM with size-based filling compare reasonably well with hydraulic conductivities measured on
SBMs prepared with granular bentonite. The favorable comparisons obtained with the GCM and the JBM (with size-based
filling) suggest that these models capture key mechanisms controlling the hydraulic conductivity of SBMs prepared with
granular and powdered bentonite.

Acknowledgments
The DEM simulations were performed by Dr. David Horner. Dr. Steve Bryant provided assistance in developing the
codes for the original network model (no bentonite). Their
efforts are greatly appreciated. Financial support for the
study described in this paper was provided by the State of
Wisconsin Solid Waste Research Program (SWRP). The
findings described in this paper are solely those of the authors. Endorsement by the SWRP is not implied and should
not be assumed.

References
Abeele, W.V. 1986. The influence of bentonite on the permeability of
sandy silts. Nuclear and Chemical Waste Management, 6: 8188.
Abichou, T. 1999. Hydraulic properties of foundry sands and their
use as hydraulic barriers. Ph.D. thesis, Department of Civil and
Environmental Engineering, University of Wisconsin, Madison,
Wisc.
Abichou, T., Benson, C., and Edil, T. 2002. Micro-structure and
hydraulic conductivity of simulated SBMs. Clays and Clay Minerals Journal, 50(5): 537545.
Bryant, S., King, P., and Mellor, D. 1993a. Network model evaluation of permeability and spatial correlation in a real random
sphere packing. Transport in Porous Media, 11: 5370.
Bryant, S., Mellor, D., and Cade, C. 1993b. Physically representative network models of transport in porous media. AIChE Journal, 39(3): 387396.
Chapuis, R. 1981. Permeability testing of soilbentonite mixtures.
In Proceeding of the 10th International Conference on Soil Mechanics and Foundation Engineering, Stockholm, 1519 June
1981. A.A. Balkema, Rotterdam, The Netherlands. pp. 744745.
Chapuis, R. 1990. Sandbentonite liners: predicting permeability
from laboratory tests. Canadian Geotechnical Journal, 27: 4757.
Chu, C., and Ng, K. 1989. Flow in packed tubes with a small particle diameter ratio. AIChE Journal, 35: 148158.

Can. Geotech. J. Vol. 41, 2004


Daniel, D. 1987. Earthen liners for land disposal facilities. In
Geotechnical Practice for Waste Disposal 87. Edited by R.D.
Woods. ASCE Geotechnical Special Publication 13, pp. 2139.
Garlanger, J., Cheung, F., and Bishar, S. 1987. Quality control testing for sandbentonite liners. In Geotechnical Practice for Waste
Disposal 87. Edited by R.D. Woods. ASCE Geotechnical Special Publication 13, pp. 488499.
Gleason, M., Daniel, D., and Eykholt, G. 1997. Calcium and sodium bentonite for hydraulic containment applications. Journal
of Geotechnical and Geoenvironmental Engineering, ASCE,
123(5): 438445.
Goodhue, M., Benson, C., and Edil, T. 2001. Interaction of foundry
sands with geosynthetics. Journal of Geotechnical and Geoenvironmental Engineering, ASCE, 127(4): 353362.
Holtz, R., and Kovacs, W. 1981. An introduction to geotechnical
engineering. Prentice-Hall, Inc., Englewood Cliffs, N.J.
Horner, D. 1997. Application of DEM to micro-mechanical theory
for large deformation of granular media. Ph.D. thesis, University
of Michigan, Ann Arbor, Mich.
Howell, J., and Shackelford, C. 1997. Hydraulic conductivity of
sand admixed with processed clay mixtures. In Proceedings of
the 14th International Conference on Soil Mechanics and Foundation Engineering, Hamburg, 612 September 1997. A.A.
Balkema, Rotterdam, The Netherlands. pp. 307310.
Kenney, T.C., Van Veen, M.A., Swallow, M.A., and Sungaila, M.A.
1992. Hydraulic conductivity of compacted bentonitesand mixtures. Canadian Geotechnical Journal, 29: 364374.
Kraus, J., Benson, C., Erickson, A., and Chamberlain, E. 1997.
Freezethaw cycling and the hydraulic conductivity of bentonitic barriers. Journal of Geotechnical and Geoenvironmental
Engineering, ASCE, 123(3): 229238.
Lundgren, T. 1981. Some bentonite sealants in soil mixed blankets.
In Proceedings of the 10th International Conference on Soil Mechanics and Foundation Engineering, Stockholm, 1519 June
1981. A.A. Balkema, Rotterdam, The Netherlands. pp. 349354.
Mason, G. 1971. A model for the pore space in a random packing
of equal spheres. Journal of Colloidal Interface Science, 35:
279287.
Mitchell, J. 1993. Fundamentals of soil behavior. John Wiley and
Sons, Inc. New York.
Mollins, L., Stewart, D., and Cousens, T. 1996. Predicting the hydraulic conductivity of bentonitesand mixtures. Clay Minerals,
31: 243252.
Mollins, L.H., Stewart, D.I., and Cousens, T.W. 1999. Drained
strength of bentonite enhanced sands. Gotechnique, 49(4):
523528.
OSadnick, D., Simpson, B., and Kasel, G. 1995. Evaluation and
performance of a sandbentonite liner. In Geoenvironment 2000.
Edited by Y.B. Acar and D.E. Daniel. ASCE Geotechnical Special Publication 46, pp. 688701.
Sllfors, G., and berg-Hgsta, A. 2002. Determination of hydraulic conductivity of sandbentonite mixtures for engineering purposes. Geotechnical and Geological Engineering, 20: 6580.

2004 NRC Canada

Potrebbero piacerti anche