GeoCongress 2012 © ASCE 2012 1958

Numerical Study of the Effectiveness of

Bentonite Treatment for Liquefaction Mitigation

A. F. Witthoeft1, M. C. Santagata2, and A. Bobet2

1
Geosyntec Consultants, 2100 Main Street, Suite 150, Huntington Beach, CA 92648;
PH (714) 969-0800; FAX (714) 969-0820; email: [email protected]
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

2
School of Civil Engineering, Purdue University, 550 Stadium Mall Drive, West
Lafayette, IN 47907

ABSTRACT

This paper presents the results of a numerical study investigating the

effectiveness at the field scale of a liquefaction mitigation technique based on the
replacement of the soil pore fluid by a thixotropic bentonite suspension. While
previous laboratory experimental work has demonstrated that this method increases
the liquefaction resistance of sands in undrained cyclic triaxial tests, questions remain
with regard to the field-scale performance, and the extent of soil treatment required.
For this numerical study a bounding surface-based constitutive model was
implemented in the finite difference software FLAC. Model parameters were
obtained for the untreated and the bentonite treated sand, using experimental data,
which included monotonic and cyclic triaxial and resonant column results.
The implemented model is used to analyze the problem of a vertically-
vibrating footing founded on a saturated sand. Loss of mean effective stress and
settlement of the footing are used as performance indicators to assess the impact of
soil treatment. For the case examined, the model indicates that a targeted soil
treatment forestalls the onset of liquefaction in the bentonite treated soil.

INTRODUCTION

General. The phenomenon of liquefaction, the loss of effective confining stress in a

saturated loose soil deposit, is a potential cause of severe damage to structures.
Various approaches for liquefaction mitigation have been proposed (e.g., as
summarized by Yasuda, 2007). One approach in particular, the replacement of the
pore fluid with (thixotropic) bentonite slurry, is the focus of this study.

Liquefaction Mitigation using Bentonite Treatment. The liquefaction mitigation

solution investigated here (i.e., bentonite treatment) is based on the concept of pore
fluid engineering, whereby the pore water in-situ is displaced by a purpose-designed
fluid (in this case, bentonite slurry). This liquefaction mitigation measure aims to
reduce the buildup of excess pore pressure that occurs when a loose sand is sheared
under undrained conditions. This excess pore pressure generation is the result of
volumetric contraction of the soil skeleton during shearing. This treatment of
liquefiable soil is thought to provide a restraint against soil particle rearrangement,
thus limiting volumetric contraction during shearing and so inhibiting liquefaction.

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1959

An extensive experimental investigation was undertaken by Clarke (2008) and

El Mohtar (2008) to investigate this technique at the laboratory scale. The goals of
the work were twofold: (i) to evaluate the properties of concentrated bentonite
suspensions and (ii) to investigate the effects of bentonite treatment on liquefaction
resistance in the laboratory. Clarke (2008) analyzed the rheological properties of
concentrated bentonite suspensions treated with sodium pyro-phosphate (SPP) to
evaluate the feasibility of delivery of the bentonite slurry in-situ through permeation,
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

and the time required for the gelation of the slurry once inside the sand pore space.
This study demonstrated that the addition of 0.5% SPP (by mass of the bentonite) to
10% bentonite suspensions yields materials characterized by minimal yield stress and
thixotropy, significantly reduced storage modulus and viscosity, which can be
permeated in laboratory prepared sand columns. The treatment is reversible, in that
with time the gel-like structure typical of untreated suspensions is redeveloped.
El Mohtar (2008) evaluated the mechanical properties of bentonite-treated sand and
the effectiveness of the treatment in preventing liquefaction by performing undrained
cyclic triaxial tests on sand specimens (skeleton relative density below 40%) with 3-
5% bentonite (by dry mass of the sand). The experimental program included tests on
specimens of sand dry-mixed with bentonite and subsequently saturated with water,
as well as on specimens of clean sand permeated with concentrated (10% bentonite
corresponding to the concentration of the clay suspension formed inside the pores
with 3% bentonite by mass of sand) suspensions treated with SPP. Findings of this
portion of the experimental program relevant to the present investigation are: (i)
bentonite treatment increases the number of loading cycles for a given cyclic stress
ratio (CSR) to which a sand specimen under undrained cyclic triaxial conditions may
be subjected before reaching liquefaction; (ii) based on resonant column test results,
the treatment appears to work in part by extending the elastic region of the material
behavior over a greater range of strain values; and iii) the gel-like structure of the
pore fluid is responsible for suppressing volumetric straining during cyclic loading.

Objective of the Present Study. Whereas no experimental study to date has

evaluated the effectiveness of bentonite treatment at the field scale, this study intends
to test the effectiveness of bentonite treatment in one field-scale application using a
numerical approach. A bounding surface-based constitutive model was selected from
the literature and implemented in the finite difference software FLAC (Itasca
Consulting Group, 2005). The model parameters were calibrated both for clean (i.e.,
untreated) sand and for sand treated with bentonite. The implemented model was
used to analyze a dynamic problem: a vertically-vibrating footing founded on a
saturated sand. The main objective of the study was to assess the extent of soil
treatment required to avoid liquefaction.

NUMERICAL APPROACH

Constitutive Model Implementation and Calibration. The bounding surface-type

(Dafalias and Popov, 1975, 1976) constitutive model proposed by Papadimitriou and

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1960

Bouckovalas (2002) was implemented in the numerical software and calibrated based
on a database of laboratory test results for Ottawa sand.
The constitutive model selected incorporates Critical State Soil Mechanics
(CSSM, Schofield and Wroth, 1968) and the state parameter (Been and Jefferies,
1985). This model is well-suited for studying conditions leading to liquefaction
because it combines stress-dependent non-linear elastic behavior at relatively small
strains with fabric-dependent plastic behavior at relatively large strains.
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

Numerical Implementation. The Papadimitriou and Bouckovalas (2002) constitutive

model was implemented as a User-Defined Model (UDM) using the FISH
programming language developed for use with the software FLAC (Itasca Consulting
Group, 2005). In general, the UDM follows the equations outlined by Papadimitriou
and Bouckovalas (2002). However, for practical reasons, various modifications (e.g.,
allowable ranges or cutoffs on certain parameter values) were made to the UDM to
facilitate the numerical implementation. These deviations from the Papadimitriou
and Bouckovalas (2002) formulation are described in detail by Witthoeft (2009).

Parameter Calibration for Clean Sand. An extensive calibration process was

performed for clean (untreated) Ottawa sand using test data reported in the literature
(Carraro, 2004; El Mohtar, 2008; Loukidis, 2006; Loukidis and Salgado, 2009; and
Murthy, 2006) and for treated sand. Details of the calibration are presented by
Witthoeft (2009). Calibrated parameter values are listed in Table 1, along with
“typical” parameter values reported by Papadimitriou and Bouckovalas (2002). Note
that only two parameters (yield surface radius, m, and fabric evolution parameter, H0)
were modifed in the case of the treated sand relative to the clean sand. They were
selected based on lab results showing an extended elastic region and different fabric
evolution with cyclic loading.

Table 1. Listing of parameters, typical ranges, and values for Ottawa sand.

Value for Clean Value for Treated

Parameter Typical Range(a)
Ottawa Sand(b) Ottawa Sand
(ecs)a 0.72-0.90 0.697 0.697
λ 0.01-0.03 0.0189 0.0189
Mcc 1.20-1.37 1.22 1.22
Mec 0.86-1.0 1.16 1.16
B 500-900 641 641
a1 0.45-0.85 0.616 0.616
0.65x10-4 to
γ1 1.29x10-4 1.29x10-4
2.4x10-4
ν 0.2-0.4 0.15 0.15
m None Provided 0.145 0.246
κ None Provided 2.0 2.0
kcb 0.5-4.0 1.83 1.83
keb None Provided 1.74 1.74

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1961

kcd 0.1-3.0 2.40 2.40

d
ke None Provided 2.28 2.28
μ None Provided 1.0 1.0
A0 1.0-3.0 0.6 0.6
h0 1,000-10,000 1,000 1,000
H0 50,000-100,000 100,000 200,000
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

ζ 0.5-2.5 1.0 1.0

(a) Papadimitriou and Bouckovalas (2002). (b) After calibration and adjustment
Figure 1 shows a comparison between measured and model-predicted results
for an undrained cyclic triaxial test conducted on ASTM C778 Ottawa sand
(emax=0.78, emin=0.48, 2-5% non plastic fines) with Dr=35%, at initial confining stress
p’0 = 100 kPa and CSR = 0.15. While the comparison highlights some uncertainty in
the predictions, the agreement between calibrated model and measured data is similar
to that reported by Papadimitrou and Bouckovalas (2002) for Nevada sand.

100
90 Model
80 Data
70
60
u (kPa)

50
40
30
20
10 (a)
0
0 5 10 15 20
0.20 N
Model
0.15
Data
0.10 εv =0.0
0.05
εv (%)

0.00
-0.05
-0.10
-0.15
(b)
-0.20
0 5 10 15 20
N
Figure 1. Model-predicted and measured (a) u-N response and (b) εv-N response
for an undrained cyclic triaxial test with p0’=100 kPa and CSR=0.15 on clean
Ottawa sand. Data reproduced from test results reported by El Mohtar (2008).
Reproducing the Effects of Bentonite Treatment. Results of undrained cyclic
triaxial testing by El Mohtar (2008) show at least two readily observable effects of

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1962

bentonite treatment: (i) delayed buildup of excess pore pressure and (ii) delayed
buildup of axial strain during cyclic loading. Figure 2 illustrates these effects using
undrained cyclic triaxial data for tests with p’0 = 100 kPa and CSR = 0.125.
These effects are captured within the modeling framework by modifying
parameter values as shown in Table 1. Figure 3 shows sample model results
comparing the behavior for clean sand and bentonite-treated sand.
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

Numerical Analysis. A dynamic analysis was undertaken to assess the field-scale

performance of treated versus untreated foundation soil. The implemented and
calibrated model was used to analyze the classic problem of a vertically-vibrating
footing founded on a saturated sand. Specifically, the model represents a strip
footing supporting a vertically vibrating machine and overlying a saturated layer of
loose sand. Details of the model (e.g., geometry, boundary conditions, and damping)
are derived from the boundary value problem analyzed by Ni (2007). The material is
C778 Ottawa sand with skeleton relative density of approximately 35% (see Table 1).

Bentonite
delays Δu
buildup

(a)

Bentonite
delays εv
buildup
(b)

Figure 2. Measured trends during undrained cyclic triaxial shear at CSR=0.125

of (a) excess pore pressure, Δu, and (b) axial strain, εv. Data reproduced from
test results reported by El Mohtar (2008).

Figure 4 shows the geometry of the numerical grid. Due to symmetry of

loading and boundary conditions, only half the problem was modeled. The problem
was analyzed assuming plane strain conditions. The footing was modeled such that it
was: (i) perfectly rigid, (ii) perfectly rough, and (iii) freely draining.
A static load of 150 kPa was applied to the footing, and a surcharge of 20 kPa
was applied from the end of the footing rightward over the top of the numerical grid.

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1963

During the dynamic simulation, the total load applied to the footing was equal to the
sum of two components: the static load of 150 kPa and the dynamic load of variable
magnitude. This may be expressed as:
qtotal = qstatic + qdynamic (1a)
where q denotes a footing load and the subscript indicates the total value or a
component value. While the static load was held constant at qstatic = 150 kPa, the
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

dynamic load was varied according to the expression:

qdynamic = ( 20 kPa ) sin ( 2πtf ) (1b)
where t is the simulation time and f is the frequency, set to f = 10 Hz for all
simulations. Dynamic loading according to Equation 1 continued until excessive
settlement of the footing began to develop for the untreated case.
(a)

(b)

Figure 3. Model-predicted trends during undrained cyclic triaxial shear at

CSR=0.125 of (a) excess pore pressure, Δu, and (b) axial strain, εv (lines through
bands represent average values of pore pressure and axial strain).

RESULTS OF NUMERICAL ANALYSIS

In order to interpret the model results and assess the impact of soil treatment,
the loss of mean effective stress in the foundation soil and the settlement of the
footing were used as performance indicators. The results of the analysis are reported
in terms of these indicators.

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1964

Loss of Mean Effective Stress. The performance dimension most closely related to
the topic of liquefaction is that of mean effective stress changes under dynamic
loading. This was evaluated by introducing an index variable, Δp’/p0’, where p0’ is
the mean effective stress at the beginning of the dynamic loading phase and Δp’ is the
change of mean effective stress during the dynamic phase (i.e., the ever-changing
current value of p’ minus the fixed value of p0’). The index Δp’/p0’ is physically
meaningful for values in the interval -1≤ Δp’/p0’<∞. When the value reaches the
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

lower bound of the interval (i.e., when Δp’/p0’= -1), a complete loss of effective
confining stress has occurred. While large positive values of Δp’/p0’ might raise
physical concerns of consolidation settlement and/or particle crushing, these are
thought not to be relevant for this particular problem.

½B=2m
Footing load
(variable)

Constant 20 kPa surcharge load

Horizontal boundary free in x and y and

fixed u=0 (both static and dynamic phases)

phase) and quiet (dynamic phase)

Vertical boundary fixed in x (static
Bentonite-treated
Vertical boundary fixed in x (both
static and dynamic phases)

Structural node fixed in x

and tied in y (typical)

H = 10 m
Saturated soil mass

Horizontal boundary fixed in y (static

+y, +j phase) and quiet (dynamic phase)

+x, +i
L = 10 m

Figure 4. Boundary value problem geometry and boundary conditions.

Figure 5(a) shows a plot of Δp’/p0’ contours at a simulation time of t = 5 s (or

50 cycles) for the case of untreated foundation material. Because increases in p’ are
not of interest for this study, only negative values of Δp’/p0’ are differentiated in the
figure. Significant decreases of mean effective stress within about 5 m (or 1¼ B) of
the ground surface are observed. Negative values of Δp’/p0’ extend almost to the far-

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1965

end boundary. The magnitude of mean effective stress loss suggests that the material
in the vicinity of the footing is at or near liquefaction for the untreated case.
In contrast, Figure 5(b) shows that with treatment, the mean effective stress
reduction is largely confined to the area near the corner of the footing. The maximum
magnitude of Δp’ (i.e., 40%-60% of p0’) is significantly lower than that observed for
the untreated case (i.e., 80%-100% of p0’), suggesting that the bentonite treatment
prevents or significantly delays the onset of liquefaction.
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

-0.2

-0.9
-0.4
-0.3
-0.1
-0.8 -0.2

-0.7 -0.1
-0.6
-0.5
-0.4
-0.3
-0.2

-0.1

(a) (b)
Figure 5. Contours of change in mean effective stress, Δp’, normalized by initial
mean effective stress, p0’ = p’|t=0, at simulation time t = 5 s (a) untreated and (b)
bentonite treated soil. Positive values of Δp’ are shown as -0.2 < Δp’/p0’ < 0.0.

Vertical Displacement of the Footing. Figure 6 shows the time history of vertical
footing displacement during the dynamic phase of the simulation. Four points along
the left boundary were tracked: one at the ground surface (i.e., the footing) and three
more at depths of 2 m, 4 m, and 6 m below the ground surface.
As Figure 6(a) indicates, the average movement of the footing with time was
relatively constant for the first two seconds of simulation time (~20 cycles).
Thereafter, the accumulation of vertical displacement accelerated, and a footing
settlement of 25 mm was reached around t = 3.5 s (~35 cycles). It is also observed
that the magnitude of the ground displacement decreases with increasing depth. The
displacement at a depth of 2 m followed closely that of the footing. In contrast, at a
simulation time of t = 2 s (when footing settlement began to accelerate) the
displacement at 4 m depth was dramatically less (by a factor > 3) than that of the
footing, and the displacement at 6 m depth was smaller still. This suggests that the
dynamic footing load is limited in its depth of influence.
Figure 6(b) shows the history of footing settlement for the treated case. The
displacement of the footing for the untreated case is shown for reference. Although
the same four points were tracked as for the untreated case, for clarity only two (i.e.,
at the footing and at 6 m depth) are plotted in the figure. It is observed from the
figure that the treatment regime is effective in delaying the settlement of the footing.

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1966

Whereas for the untreated case the footing reached a settlement of approximately 25
mm at 3.5 s of simulation time, the application of bentonite treatment held footing
settlement to approximately 1 mm in the same amount of simulation time.

0.005

0.000 Depth = 6 m

Footing Settlement (m)

Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

-0.005
Depth = 4 m
-0.010

-0.015
Ground Depth = 2 m
-0.020
(a) Surface
-0.025
0 1 2 3 4 5
Simulation Time (s)
0.005
Depth = 6 m
0.000
Footing Settlement (m)

Ground
-0.005 Surface
-0.010

-0.015 Ground
Surface
-0.020 (Untreated)
(b)
-0.025
0 1 2 3 4 5
Simulation Time (s)
Figure 6. Footing settlement with simulation time for (a) untreated and (b)
bentonite treated sand. Note that, since the dynamic load is applied at a
frequency of 10 Hz, each 1 s of simulation time corresponds to 10 cycles.

CONCLUSIONS

Previous research has shown bentonite treatment to increase soil liquefaction

resistance in undrained cyclic triaxial testing. Extrapolation of the laboratory
experimental results suggests that bentonite treatment might mitigate or even prevent
liquefaction at the field scale. However, to date, no pilot study has been performed to
confirm that bentonite treatment performs as expected at the field scale, or to
establish the volume of soil required to be treated. Therefore, the goal of this study
was to use a numerical modeling approach to demonstrate proof-of-concept for
bentonite treatment as a field-scale liquefaction mitigation technique.
For the case of a vertically-vibrating footing founded on a saturated sand, the
model indicates that liquefaction is prevented in the bentonite-treated soil. Model
results suggest that a targeted soil treatment – in this case, limited to a depth of 2
times the footing width and a width of 1.5 times the footing width – may be used to
achieve the desired performance.
While additional research is required to establish the effectiveness of the
treatment in the field, the results of this work, combined with the previous laboratory-
scale experiments, support the use of bentonite treatment for liquefaction mitigation.

GeoCongress 2012
 GeoCongress 2012 © ASCE 2012 1967

ACKNOWLEDGEMENT

The authors wish to thank Neven Matasovic of Geosyntec Consultants,

Huntington Beach, California, for his peer review and valuable comments.

REFERENCES
Downloaded from ascelibrary.org by Universidad Militar Nueva Granada on 03/04/17. Copyright ASCE. For personal use only; all rights reserved.

Been, K., & Jefferies, M. (1985). A State Parameter for Sands. Géotechnique , 35 (2),
99-112.
Carraro, J. (2004). Mechanical Behavior of Silty and Clayey Sands. Ph.D. Thesis,
Purdue University.
Clarke, J. P. (2008). Investigation of Time-Dependent Rheological Behavior of
Sodium Pyrophosphate – Bentonite Suspensions. West Lafayette, IN, USA:
MS Thesis, Purdue University.
Dafalias, Y., & Popov, E. (1975). A Model of Nonlinearly Hardening Materials for
Complex Loading. Acta Mechanica , 21 (3), 173-192.
Dafalias, Y., & Popov, E. (1976). Plastic Internal Variables Formalism of Cyclic
Plasticity. Journal of Applied Mechanics , 97, 645-651.
El Mohtar, C. (2008). Pore Fluid Engineering: An Autoadaptive Design for
Liquefaction Mitigation. Ph.D. Thesis, Purdue University.
Itasca Consulting Group. (2005). FLAC - Fast Lagrangian Analysis of Continua.
Version 5.00. Minneapolis, Minnesota, USA: Itasca Consulting Group, Inc.
Loukidis, D. (2006). Advanced Constitutive Modeling of Sands and Applications to
Foundation Engineering. Ph.D. Thesis, Purdue University.
Loukidis, D., & Salgado, R. (2009). Modeling Sand Response using Two-Surface
Plasticity. Computers and Geotechnics , 36, 166-186.
Murthy, T. (2006). Study of the Undrained Static Response of Sandy Soils in the
Critical State Framework. Ph.D. Thesis, Purdue University.
Ni, B. (2007). Implementation of a Bubble Model in FLAC and its Application in
Dynamic Analysis. Ph.D. Thesis, The University of Auckland.
Papadimitriou, A. G., & Bouckovalas, G. D. (2002). Plasticity Model for Sand under
Small and Large Cyclic Strains: A Multiaxial Formulation. Soil Dynamics
and Earthquake Engineering , 22, 191-204.
Papadimitriou, A. G., Bouckovalas, G. D., & Dafalias, Y. F. (2001). Plasticity Model
for Sand Under Small and Large Cyclic Strains. Journal of Geotechnical and
Geoenvironmental Engineering , 127 (11), 973-983.
Schofield, A., & Wroth, P. (1968). Critical State Soil Mechanics. New York, NY,
USA: McGraw-Hill.
Witthoeft, A. F. (2009). Modeling of Liquefaction Mitigation using Bentonite. West
Lafayette, IN, USA: MS Thesis, Purdue University.
Yasuda, S. (2007). Remediation Methods against Liquefaction which Can Be Applied
to Existing Structures. In K. Pitilakis (Ed.), Earthquake Geotechnical
Engineering (pp. 385-406). Springer.

GeoCongress 2012