Simplified Plane-Strain Modeling of Stone-Column

Reinforced Ground
S. A. Tan1; S. Tjahyono2; and K. K. Oo3

Abstract: The acceleration of consolidation rate by stone columns was mostly analyzed within the framework of a basic unit cell 共i.e.,
Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

a cylindrical soil body around a column兲. A method of converting the axisymmetric unit cell into the equivalent plane-strain model would
be required for two-dimensional numerical modeling of multicolumn field applications. This paper proposes two simplified conversion
methods to obtain the equivalent plane-strain model of the unit cell, and investigates their applicability to multicolumn reinforced ground.
In the first conversion method, the soil permeability is matched according to an analytical equation, whereas in the second method, the
column width is matched based on the equivalence of column area. The validity of these methods is tested by comparison with the
numerical results of unit-cell simulations and with the field data from an embankment case history. The results show that for the case of
linear-elastic material modeling, both methods produce reasonably accurate long-term consolidation settlements, whereas for the case of
elastoplastic material modeling, the second method is preferable as the first one gives erroneously lower long-term settlements.
DOI: 10.1061/共ASCE兲1090-0241共2008兲134:2共185兲
CE Database subject headings: Plane strain; Columns; Stones; Soil consolidation; Numerical models; Constitutive models; Plastic
deformation; Embankments.

Introduction 10 to 20 times the elastic modulus of soft clay 共Barksdale and

Bachus 1983兲—may contribute to the dissipation of pore water
Stone columns are commonly installed as partial replacement pressure by as much as 40% of the total dissipation by means of
of soil to improve soft clay foundation for rapid embankment vertical stress reduction on the surrounding soft soil, with the rest
construction. Their main purpose is to accelerate the primary con- being attributed to radial drainage 共Han and Ye 2001兲. Han and Ye
solidation of foundation soil by means of the following two 共2001兲 developed a simplified analytical solution for the rate of
mechanisms. First, the high column permeability causes radial consolidation of stone-column reinforced ground that accounted
drainage resulting in faster dissipation of excess pore water pres- for the contributions of the above two mechanisms. This solution
sure and second, the high column stiffness reduces foundation however assumes one-dimensional deformation under loading
load or vertical stress on the soil body and so reduces the genera- that limit its applications to only simple cases of stone-column
tion of excess pore water pressure. The acceleration of consolida- reinforced ground under one-dimensional compression with ho-
tion due to the first mechanism has been well researched, mostly mogenous soil properties and regular stone-column patterns.
in vertical drain studies where the rate of consolidation in a unit In vertical drain studies, several methods have been proposed
cell 共i.e., a cylindrical soil body around a vertical drain兲 has been to convert the axisymmetric unit cell to the equivalent plane-
analytically derived with the assumption of negligible drain stiff- strain model for the purpose of two-dimensional 共2D兲 numerical
ness 共Barron 1948; Hansbo 1981兲. This neglect of drain stiffness, modeling of multidrain field applications 共Hird et al.1992;
however, may not be similarly applied to stone column studies
Indraratna and Redana 1997兲. These conversion methods involved
lest it would lead to the disregard of the second mechanism
the derivations of the equivalent plane-strain permeability or the
and so the underestimation of the rate of consolidation. This is
equivalent plane-strain geometry based on the matching of axi-
especially so as the high column stiffness—typically ranging from
symmetric and plane-strain consolidation analytical solutions. For
1
stone-column applications, such conversion methodology needs
Associate Professor, Dept. of Civil Engineering, National Univ. of to include a procedure for obtaining the equivalent plane-strain
Singapore, 1 Engineering Dr. 2, Singapore 117576. E-mail: cvetansa@
column stiffness so that its effect on the consolidation rate may
nus.edu.sg
2
Research Scholar, Dept. of Civil Engineering, National Univ. of be properly simulated in plane-strain numerical analysis. The ob-
Singapore, 1 Engineering Dr. 2, Singapore 117576. E-mail: sindhu@ jective of the present study is to propose simplified con-
nus.edu.sg version methods to obtain the equivalent plane-strain model of the
3
Ex-Research Scholar, Dept. of Civil Engineering, National Univ. of stone-column unit cell, and to investigate their applicability to
Singapore, 1 Engineering Dr. 2, Singapore 117576. multicolumn reinforced ground for practical plane-strain numeri-
Note. Discussion open until July 1, 2008. Separate discussions must cal analysis.
be submitted for individual papers. To extend the closing date by one In this study, two conversion methods as described earlier are
month, a written request must be filed with the ASCE Managing Editor.
proposed based on a macroscopic approach, i.e., one that consid-
The manuscript for this paper was submitted for review and possible
publication on May 31, 2006; approved on May 22, 2007. This paper is ers the overall effects of radial 共or horizontal兲 drainage and soil
part of the Journal of Geotechnical and Geoenvironmental Engineer- vertical-stress reduction in stone-column reinforced ground on the
ing, Vol. 134, No. 2, February 1, 2008. ©ASCE, ISSN 1090-0241/2008/ rate of consolidation without the need for the complex analytical
2-185–194/$25.00. solution. The validity of these methods is tested in two stages:

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008 / 185

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

 R=B 共2兲
In practical finite-element applications, this method allows an
easy transition between axisymmetric and plane-strain meshed
geometry as both of them may usually be derived from the same
basic 共2D兲 input geometry.
The plane-strain material properties need to be adjusted to ac-
count for the geometrical changes. The plane-strain material stiff-
ness is given by the following relationship based on the matching
of the column-soil composite stiffness
Ec,plas,plEs,pl共1 − as,pl兲 = Ec,axas,ax + Es,ax共1 − as,ax兲 共3兲
Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

where Ec and Es = elastic moduli of the column material and the

surrounding soil, respectively, and subscripts pl and ax denote
plane-strain and axisymmetric conditions, respectively. Area re-
placement ratio as = Ac / 共Ac + As兲, where Ac and As = cross-section
areas of the column and the surrounding soil respectively. In this
study, for simplicity, Es,pl = Es,ax, and hence Ec,pl can be deter-
mined accordingly. The matching of the soil permeability is done
using the following analytical equation as derived by Tan and Oo
共2005兲:
kh,pl F共N兲pl
= 冋
mvsmvc共1 − as兲
kh,ax F共N兲ax mvc共1 − as兲 + mvsas
册冋
pl
mvc共1 − as兲 + mvsas
mvsmvc共1 − as兲
册 B2
ax R
2

共4兲

Fig. 1. Cross sections of unit-cell stone column and plane-strain where kh = coefficient of soil permeability in horizontal direc-
conversions tion; F共N兲 = 关N2 / 共N2 − 1兲兴ln共N兲 − 共3N2 − 1兲 / 共4N2兲; diameter ratio
N = R / rc for axisymmetric condition, whereas N = B / bc for plane-
strain condition; mvs = 共␣vs兲 / 共1 + es兲; mvc = 共␣vc兲 / 共1 + ec兲; ␣vc
and ␣vs = coefficients of compressibility of the column and the
surrounding soil, respectively; and ec and es = void ratios of the
First, by comparison with the numerical results of axisymmetric columns and the surrounding soil, respectively. The plane-strain
and three-dimensional unit-cell simulations and second, by com- coefficient of soil permeability in vertical direction kv,pl is
parison with the field data from an embankment case history in expected to bear little influence on the overall consolidation rate
Malaysia. The effects of smear 共i.e., disturbance on the surround- as water flow in the soil is predominantly radial or horizontal,
ing soil due to column installation兲 and finite column permeability and is simply assumed to follow the axisymmetric value, i.e.,
are excluded in this study. kv,pl = kv,ax.
Method 2: An alternative geometrical transformation is based
on the equivalence of the column drainage capacity in both axi-
Proposed Conversion Methods symmetric and plane-strain conditions, whose concept has been
proposed in a vertical drain study by Indraratna and Redana
Based on a macroscopic approach as described in the earlier para- 共1997兲 to convert vertical drain system into the equivalent plane-
graph, two methods of converting the axisymmetric unit-cell strain drain walls. This method hence preserves the cross-
model to the equivalent plane-strain model are proposed. The sectional areas of the column and the surrounding soil for the
macroscopic approach has an advantage in permitting the formu- same total area in both conditions. The plane-strain column width
lation of simple, and yet rational, conversion methods that can be is given by the following relationship based on the equivalence of
easily applied for practical geotechnical problems. These methods area replacement ratio:
are described as follows.
r2c
Method 1: This method has been described in detail by Tan bc = B 共5兲
and Oo 共2005兲 and is briefly reviewed here. To obtain equal flow R2
path length normal to the column perimeter, plane-strain column which results in smaller plane-strain column widths and larger
共or wall兲 width can be taken equal to the axisymmetric column flow path lengths as compared to the previous method, as
diameter, i.e. seen in Figs. 1共b and c兲. The relationship between R and B
关Figs. 1共a and c兲兴 may be given by the following equation based
on the equivalence of total area for a square pattern of columns
bc = rc 共1兲
共Barron 1948兲:
as shown in Figs. 1共a and b兲. This geometrical transformation has
R = 1.13B 共6兲
been similarly applied by Indraratna and Redana 共2000兲 in the
permeability-matching approach for the plane-strain conversion In practical multidrain applications that involve high area replace-
of vertical drains. Correspondingly, the radius of drainage zone ment ratios 共i.e., large column diameters relative to column spac-
R can be taken equal to the equivalent plane-strain width B ing兲, this method may be preferable to the previous one as the
关Figs. 1共a and b兲兴, i.e. enlargement 共with respect to the axisymmetric geometry兲 of the

186 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Finite-element models of a unit cell

plane-strain column area in Method 1 might result in excessively model is shown halved due to symmetry in Fig. 2共a兲, and has a
high plane-strain area replacement ratio and column drainage height of 10 m and an outer radius of 2.550 m. The 0.850 m
capacity. The plane-strain column stiffness may be similarly de- radius stone column gives a diameter ratio of 3, which is within
termined from Eq. 共3兲, which results in equal elastic moduli for the typical range of 1.5–5. The model is horizontally fixed on the
both axisymmetric and plane-strain conditions 共i.e., Ec,pl = Ec,ax兲 vertical sides and fully 共i.e., horizontally and vertically兲 fixed on
given that as,pl = as,ax and that Es,pl = Es,ax as in the previous the base, and has top vertical loading which is applied through a
method. The soil permeability however cannot similarly be rigid plate 关Fig. 2共a兲兴 to enforce equal vertical strain at the top
matched using Eq. 共4兲 because the equation carries the assump- boundary. For realistic simulation of unit-cell consolidation, only
tion of equal flow path length in both axisymmetric and plane- the top boundary is set to be pervious and the phreatic level is set
strain conditions 共Tan and Oo 2005兲. For simplicity, the at the top surface to provide pore water pressure throughout the
plane-strain soil permeability is taken equal to its axisymmetric model material. Figs. 2共b and c兲 show the equivalent plane-strain
counterpart, i.e., kh,pl = kh,ax and kv,pl = kv,ax. This method thus models, also halved due to symmetry, based on Methods 1 and 2,
retains the axisymmetric material properties for the plane-strain respectively. The equivalent outer widths of plane-strain-1
geometry. 关Fig. 2共b兲兴 and plane-strain-2 关Fig. 2共c兲兴 models are based on
Eqs. 共2兲 and 共6兲, respectively, and their boundary, phreatic, and
loading conditions correspond to the axisymmetric case. The
Comparison of Unit-Cell Simulations plane-strain-1 column width follows the axisymmetric column di-
ameter, whereas the plane-strain-2 column width, as obtained
To verify the proposed conversion methods, numerical simula- from Eq. 共5兲, is 0.252 m or approximately 30% of the diameter.
tions of unit-cell models under consolidation process have been The axisymmetric and the plane-strain models were developed
carried out based on the axisymmetric and the equivalent plane- using 15-node triangular elements in Plaxis 2D Version 8.5
strain geometry. Fig. 2 shows four finite-element models of a 共Brinkgreve et al. 2006兲. The last model in Fig 2共d兲 is a three-
stone-column unit cell used for the simulations. The axisymmetric dimensional representation of a unit cell, only one-quarter of

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008 / 187

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

 Table 1. Material Parameters for Unit-Cell Models
MC parameters
␥s = ␥c Es Ec kh共s兲 kv共s兲 cs⬘ ␸s⬘ cc⬘ ␸c⬘
Model 共kN/ m3兲 ␯⬘ 共kPa兲 共kPa兲 共m/s兲 共m/s兲 共kPa兲 共deg兲 共kPa兲 共deg兲
Axisymmetric, plane-strain-2, 15 0.3 3,000 30,000 3.47⫻ 10−9 1.16⫻ 10−9 0.1 22 1 40
and three dimensional
Plane-strain-1 15 0.3 3,000 12,000 2.60⫻ 10−9 1.16⫻ 10−9 0.1 22 1 40

which is shown due to symmetry. The model is similar to the lated with no change in the applied loading. The dissipation was
Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

axisymmetric case except that the geometry is rectangular instead allowed to progress until the remaining excess pore water pres-
of circular, and is similar to the plane-strain case in that they both sure was less than 0.01 kPa, after which the simulation ended.
have rectangular drainage zone. The purpose of this model is to Two constitutive models were considered for the representa-
simulate three-dimensional effects arising from the use of rectan- tion of the unit-cell materials, giving the following two cases of
gular geometry that involves lesser degree of symmetry than the simulations. In the first case 共Case 1兲, the soil and column mate-
axisymmetric circular geometry. This would hence enable the rial were represented by the linear-isotropic-elastic model, whose
measurement of potential errors resulting from the conversion of behavior was fully defined by Young’s modulus and Poisson’s
axisymmetric to rectangular geometry in the earlier plane-strain ratio. In the second case 共Case 2兲, the materials were idealized as
models. The widths of its square column and drainage zone are the MC model with the characteristic linear-elastic-perfectly plas-
0.753 and 2.26 m, respectively, as obtained from Eq. 共6兲, and its tic behavior and the failure criteria defined by the strength param-
boundary, phreatic, and loading conditions follow the axisym- eters in Table 1.
metric case. The model was developed using 15-node wedge A mesh convergence analysis was conducted using the axi-
elements in Plaxis 3D Tunnel Version 2 共Brinkgreve and Broere symmetric model 关Fig. 2共a兲兴 with MC soil model to obtain the
2004兲. optimum mesh size. Four mesh coarseness as defined in Plaxis
The material parameters for the unit-cell models are presented was considered: 共1兲 coarse mesh 共160 elements兲; 共2兲 medium
in Table 1; where ␥ = unit weight; ␯⬘ = Poisson’s ratio in terms of
effective stress for both soil and column material; c⬘ = effective
stress cohesion; ␸⬘ = effective stress angle of friction; and sub-
scripts s and c denote soil and column material, respectively. The
soil was soft clay and modeled as undrained material in Plaxis.
The stone column was idealized as a homogenous drained mate-
rial having certain characteristic stiffness and strength parameters.
The unit weight of soft clay is assumed to be equal to that of
stone column so that for the assumption of uniform K0, no excess
pore water pressure is generated due to difference in the two
materials’ mean effective stresses during the initial-stress genera-
tion. This would hence allow the top 100 kPa loading to be the
sole generator of excess pore water pressure in the model. The
elastic modulus of column material is assumed to be ten times of
that of soil, which is within the typical range of 10–20. The per-
meability parameters for the stone column are arbitrarily taken to
be 10,000 times of the corresponding soft clay parameters given
in Table 1. In the plane-strain-1 model, the elastic modulus of
column material is 12,000 kPa and the clay’s permeability in hori-
zontal direction is 2.60⫻ 10−9 m / s, as determined from Eqs. 共3兲
and 共4兲, respectively. The clay’s permeability in vertical direction
is assumed to be one-third of its horizontal counterpart. The
strength parameters given in Table 1 correspond to the Mohr-
Coulomb 共MC兲 failure criteria, of which the effective stress
cohesion is given small nonzero values to avoid numerical
complications.
Simulation procedures and results: The simulation of each
unit-cell model was initialized by applying initial stresses in the
two materials based on uniform K0 = 0.7, and activating instanta-
neous 共i.e., undrained兲 vertical loading of 100 kPa on the
top surface of the model. Equal vertical strain condition was
enforced by applying the loading on a rigid plate 共Fig. 2兲
having normal stiffness of 1 ⫻ 107 kN/ m and flexural rigidity
of 1 ⫻ 104 kN m2 / m. The instantaneous loading implied that there
was no dissipation of the generated excess pore water pressure in
the model. This was followed by a consolidation procedure
whereby dissipation of the excess pore water pressure was simu- Fig. 3. Effects of mesh sizes on excess pore water pressure

188 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Simulated unit-cell surface settlement and excess pore water Fig. 5. Simulated unit-cell surface settlement and excess pore water
pressure with time for Case 1 pressure with time for Case 2

mesh 共282 elements兲; 共3兲 fine mesh 共561 elements兲; and 共4兲 very- with the plots following the typical consolidation trend. The ini-
fine mesh 共1,270 elements兲. The effects of the mesh coarseness on tial point on each curve signifies the undrained response of the
the excess pore water pressure during consolidation are shown in soft clay under the 100 kPa applied loading. The initial excess
Fig. 3. In the consolidation time of less than 0.01 day, some pore water pressure is hence expected to be approximately
variation is seen as the coarse mesh gives an unusual jump in 100 kPa, which is well reflected in the results except for the 5%
excess pore water pressure. The size of the jump decreases with overestimation in the plane-strain-1 model. The settlement curve
finer mesh and is unnoticeable for the very-fine mesh. The exis- of the axisymmetric model is in reasonable agreement with the
tence of the jump is caused by the use of a very small first time analytical solution of Han and Ye 共2001兲 and the numerical solu-
step 共approximately 0.0001 days兲 which falls below the critical tion of Balaam and Booker 共1981兲, with the maximum differences
time step in Plaxis’ consolidation numerical procedures, leading limited to approximately 20 and 8%, respectively. The settlement
to oscillating excess pore water pressure and loss of accuracy in and excess pore water pressure curves for the three-dimensional
the beginning of consolidation. The critical time step is propor- model differ from the axisymmetric case by up to 4 mm and
tional to the square of the height of the finite element 共Brinkgreve 6 kPa, respectively, which implies that the geometrical conver-
et al. 2006兲, and hence the accuracy is higher for the finer mesh, sion from axisymmetry to rectangularity in the three-dimensional
which explains the smoother consolidation curves respectively. and plane-strain models accounts for only marginal errors for this
After time of 0.1 day, the consolidation curves are identical re- case of modeling. The results of the plane-strain-2 model also
gardless of mesh sizes. This implies that for consolidation analy- follow closely the axisymmetric case as the deviation is limited to
sis with time domain greater than 0.1 day, the use of coarse mesh only 8 mm. The results of plane-strain-1 model however exhibit a
suffices. In this study however, the time domain is set to be significant discrepancy from the norm, with as much as 20 mm
0.0001 day onwards, and hence very-fine mesh is adopted for all larger settlement. This is attributed to the larger cross-sectional
subsequent 2D unit-cell simulations for greater accuracy in the area of stone column in the model, whose drained behavior is less
beginning of consolidation process. stiff than the undrained soft clay at the start of consolidation,
The simulation results were expressed in terms of the surface resulting in a higher proportion of the top loading carried by the
settlement and the excess pore water pressure during consolida- soft clay at the start of consolidation as compared with the other
tion, the latter of which was measured at the bottom right corner models. This leads to larger overburden stresses in the soft clay
of the model geometry for slowest dissipation of excess pore and so a larger initial settlement of the model. The larger over-
water pressure. Fig. 4 shows the results for Case 1 simulations burden stress is also reflected in the initial 5%-overestimation of

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008 / 189

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. Cross section of embankment case history through centerline

of stone columns

the final settlement where they differ by 8%. The corresponding

excess pore water pressure is well matched with the axisymmetric
case. As in Case 1 simulations, the plane-strain-1 model presents
significant deviation from the norm with as much as 21 mm and
5% larger settlement and excess pore water pressure, respectively,
in the initial consolidation period for the same reason described
earlier. More importantly, its long-term settlements considerably
underestimate the norm leading to a final settlement that is 25%
lower than 190 mm 共the average final settlement of the other
models兲. This implies that the model may give erroneous under-
predictions of long-term settlement when MC soil model is used.
To explain the reason for this error, the material stress state of
Fig. 6. Simulated plastic stress points in unit-cell models at end of each model at the end of the simulations is analyzed and pre-
consolidation for Case 2 sented in Fig. 6. The axisymmetric model 关Fig. 6共a兲兴 is seen to
sustain some plastic yielding in the vicinity of the column mate-
rial, as indicated by the numerous plastic stress points that are
excess pore water pressure for the model. Despite these discrep- concentrated within and slightly beyond the column periphery. A
ancies, in period longer than 10 days—defined here as long term plastic stress point defines a stress state that lies on the MC failure
for convenience—the models are all in agreement with settle- envelope corresponding to perfectly plastic behavior. Similar
ments of approximately 140 mm and excess pore water pressure yielding pattern, albeit with less scatter beyond the column pe-
of nearly zero. This has an important implication in that the plane- riphery, is observed in the plane-strain-2 model 关Fig. 6共c兲兴,
strain models based on Methods 1 and 2 may both represent the whereas hardly any yielding is present in plane-strain-1 model
axisymmetric case with reasonable accuracy for long-term 关Fig. 6共b兲兴. This explains the differential responses of the two
settlements—often the most critical design outcome in practical models: The plastic yielding in the former model allows itself to
applications—insofar as the linear-elastic constitutive model is simulate the settlements of axisymmetric model, while the latter
used. model’s elasticity gives a stiffer response leading to a lower final
Fig. 5 shows the corresponding results of Case 2 simulations. settlement that is indeed comparable to that in Case 1 simulations.
The three-dimensional and the axisymmetric curves again differ The sustained elastic behavior in the latter may be due to its larger
marginally with less than 4 mm or 6 kPa in settlement and excess cross-sectional column area with higher elastic capacity in both
pore water pressure deviations, respectively, thereby signifying shearing and bending. Thus, for conditions involving plastic
that the conversion of axisymmetric to rectangular geometry also yielding with the MC model, the model based on Method 1 may
yields minimal errors for Case 2 modeling. The plane-strain-2 give incorrect representation of the axisymmetric case and so the
settlement curve follows closely the axisymmetric case except for conversion based on Method 2 is preferable.

Table 2. Material Parameters for Embankment Models

Unsaturated ␥ Saturated ␥ E kh kv c⬘ ␸⬘
Material 共kN/ m3兲 共kN/ m3兲 ␯⬘ 共kPa兲 共m/s兲 共m/s兲 共kPa兲 共deg兲
Embankment fill 18 20 0.3 15,000 1.16⫻ 10−5 1.16⫻ 10−5 3 33
−7
Crust 17 18 0.3 15,000 3.47⫻ 10 1.16⫻ 10−7 3 28
Soft clay 15 15 0.3 1,100 3.47⫻ 10−9 a 1.16⫻ 10−9 1 20
Stiff clay 18 20 0.3 40,000 3.47⫻ 10−9 1.16⫻ 10−9 3 30
Stone column 19 20 0.3 30,000a 1.16⫻ 10−4 1.16⫻ 10−4 5 40
a
Values are only for plane-strain-2 and three-dimensional models; correspondingly for plane-strain-1 model, E = 8 , 670 kPa, and kh = 1.68⫻ 10−9 m / s.

190 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 9. Simulated embankment settlement profiles

surface. Two settlement plates 共SP1 and SP2兲 as shown in Fig. 7

were installed in situ to measure the settlements at the center of
the embankment and at 8 m from its edge.
Three finite-element models of the embankment were consid-
ered: plane-strain-1 共by Method 1兲, plane-strain-2 共by Method 2兲,
and three-dimensional models. The plane-strain modeling was
possible as the embankment spanned a distance of more than
200 m with approximately uniform cross-sectional geometry in
the direction normal to the plane in Fig. 7. As in the unit-cell
modeling, the plane-strain models were developed using 15-node
triangular elements in Plaxis 2D Version 8.5, whereas the last
model were formed with 15-node wedge elements in Plaxis 3D
Tunnel Version 2. The widths of stone columns 共or walls兲 in the
plane-strain-1 and plane-strain-2 models were 0.80 m 共i.e., equal
to the actual column diameter兲 and 0.21 m, respectively, the latter
of which was obtained from Eq. 共5兲. The column spacing of 2.4 m
was used uniformly in all embankment models. Owing to soft-
ware limitations, the stone columns in the three-dimensional
Fig. 8. Comparison between simulated and field settlements at SP1 model were given equivalent rectangular geometry with square
and SP2 cross-sectional area, whose width was given by Eq. 共6兲, in place
of the actual circular geometry. In all models, the materials were
modeled as MC soils, which were more realistic approximations
Comparison with Data from Case History of the real soils than the linear-elastic model, with parameters as
listed in Table 2. The parameters are identical for all models ex-
In the second stage of verification, the proposed conversion meth- cept for the converted Young’s modulus and the coefficient of soil
ods were applied for the modeling of an embankment construc- permeability in horizontal direction for the plane-strain-1 model,
tion for Penchala Toll Plaza project at New Pantai Expressway, whose values are determined from Eqs. 共3兲 and 共4兲, respectively.
Malaysia, in 2003. A brief description of the project was given by Simulation procedures and results: The project involved rapid
Tan and Oo 共2005兲. The embankment geometry with the stone- embankment construction which was simulated as follows. The
column reinforced soil profile is shown in Fig. 7 having a line of stone columns were first installed by partial soil replacement, fol-
symmetry on the left boundary. The 20 m wide and 1.8 m high lowing which the construction of embankment was done in three
embankment is filled by sandy material. The stone columns, ar- equal stages 共i.e., 0.6 m increment in embankment height in each
ranged in a square grid, cover a distance from the embankment stage兲. Each stage of embankment construction is simulated with
base to a depth of 6 m above the layer of stiff clay. The upper 3 day consolidation period, giving altogether 9 days of consoli-
crust layer is 1 m thick fill of hard soil, which was provided as a dation for the simulated construction. The consolidation process
replacement of soft-clay surface to improve ground for stable was then continued with no change in loading condition until the
construction platform as well as drainage of water during consoli- remaining excess pore water pressure fell below a specified near-
dation. The groundwater level is one meter below the ground zero value 共1 kPa兲, which triggered the end of the simulation.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008 / 191

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. Simulated excess pore water pressure in embankment models

Figs. 8共a and b兲 show the simulated settlements at the loca- the column–soil interface area in the former model resulting from
tions of SP1 and SP2 共Fig. 7兲, respectively, with respect to time the conversion of discrete columns into plane-strain walls, which
for comparison with the measured field settlements. At both loca- in turn causes a reduction in the total interface friction and so less
tions, the settlements of the plane-strain-2 model slightly overes- stiffening effect on the soil mass. In spite of this disparity, both
timate the three-dimensional counterparts by a magnitude of less models predict the field data with reasonable accuracy as the de-
than 6 mm, or less than 9% difference, at any given time. The viation is limited to 15% for approximately 90% of the field data.
former model’s tendency to overestimating settlements here is Better agreement is seen for time greater than 60 days where the
parallel with the results of the earlier unit-cell simulations and deviation is less than to 8%. On the other hand, the settlement
may imply a slight disparity in the column–soil composite stiff- curves for plane-strain-1 model diverge from the norm from ap-
ness in the two models. This may be attributed to the reduction in proximately 9 days onwards and eventually reach final settle-

Fig. 11. Simulated plastic stress points in embankment models at end of consolidation

192 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 12. Simulated horizontal displacements in plane-strain-2 model at end of consolidation

ments of less than 50 mm, which grossly underpredict the field means that the excess pore water pressure discrepancies are only
measurements of approximately 80 mm. This error is parallel confined within a distance of several meters from the stone col-
with the under-prediction of long-term settlements in the unit-cell umns. The excess pore water pressure here takes much longer
simulations using the MC model 共Case 2兲. than 120 days to dissipate and hence the acceleration of consoli-
The embankment settlement profiles as simulated by the dation by the stone columns is hardly evident at this location.
plane-strain-2 and the three-dimensional models at the 20th and In order to explain the main discrepancy in the settlements of
90th days are illustrated in Fig. 9 for further comparison. The the plane-strain-1 and the plane-strain-2 models as shown in
settlements are measured at the same level as SP1 and SP2 for Fig. 8, the material stress state of the models at the end of
comparison with the field measurements taken from Fig. 8. Fig. 9 consolidation are compared and presented in Fig. 11. Numerous
presents better similarity in the simulated and the field settlements plastic stress-points are seen in the plane-strain-2 and the three-
at the 90th than the 20th days, reiterating an earlier finding dimensional models 关Figs. 11共b and c兲兴 mostly within the column
that better agreement between the settlements is obtained in the periphery, signifying plastic yielding in the column material.
later stages of consolidation. The settlement profiles of both mod- More plastic stress-points, and so greater extent of yielding, are
els are almost identical, except for the slight deviation in the present in the columns near the embankment edge 共or outer col-
region between 5 and 15 m from the centerline where the differ- umns兲 than the inner ones. This results in larger deformation,
ence is limited to only 6 mm. Small heaving is noted at distances especially the horizontal displacements, in the outer columns than
beyond the embankment width 共i.e., greater than 20 m from the the inner ones, which is shown in Fig. 12 for the plane-strain-2
centerline兲 as a result of stress distribution in soil around the model. The differential horizontal displacements are because the
embankment. Periodic humps in settlement profiles are also seen inner columns experience larger lateral restraint provided by the
for distances less than about 15 m from the centerline, which neighboring columns as compared to the outer ones. This is in
correspond to the locations of the stone columns and the resulting agreement with the experimental finding of McKelvey et al.
smaller surface settlements directly above them. 共2004兲 for small groups of sand columns underneath a strip foot-
Fig. 10 shows the simulated excess pore water pressure with ing in soft clay. On the contrary, the column material in the plane-
time at the locations of A, B, and C as shown in Fig. 7. At A and strain-1 model as shown in Fig. 11共a兲 does not yield at all and so
B, as shown by Figs. 10共a and b兲, respectively, the excess pore remains elastic, which is parallel with the simulated stress state in
water pressure in the models has an initial peak value of approxi-
the earlier unit-cell model 关Fig. 6共b兲兴 and may be likewise attrib-
mately 17 kPa due to the embankment construction and then
uted to its larger cross-sectional column area with higher elastic
dissipates with different rates to nearly zero after 100 days. Faster
capacity in both shearing and bending. Hence the model gives a
dissipation of excess pore water pressure is seen for the plane-
stiffer response with erroneously smaller settlements than the oth-
strain-1 model as it underestimates the results of the three-
ers. This reinforces the earlier finding that Method 1 may not be
dimensional model by as much as 4 kPa during the consolidation
correctly applied in cases involving plastic yielding, and that
and reaches full dissipation at shorter periods. This discrepancy is
Method 2, which presents reasonably accurate predictions of the
parallel with the results of the earlier unit-cell simulations 共Fig. 5兲
consolidation behavior in the embankment case history, is the
and may be attributed to the larger drainage capacity in the former
preferred conversion method.
model as compared with the others. For the plane-strain-2 model,
reasonable agreement with the three-dimensional model is pre-
sented for excess pore water pressure at A, whereas slight dis-
parity manifested in as much as 2 kPa lower excess pore water Conclusion
pressure 共with respect to the latter model兲 is seen at B. On the
other hand, at C 关Fig. 10共c兲兴, 2 m away from the embankment This paper has proposed two simple conversion methods,
edge, the excess pore water pressure has significantly lower peak Methods 1 and 2, to obtain the equivalent plane-strain models of
values due to the diminished effects of the embankment loading stone-column reinforced ground. The main differences between
and dissipates almost identically in all the three models, which the methods lie in the conversion of column geometry and the

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008 / 193

J. Geotech. Geoenviron. Eng. 2008.134:185-194.

 matching of soil permeability. In Method 1, the plane-strain col- References
umn width is taken to be same as the actual column diameter and
the soil permeability is matched according to Eq. 共4兲; whereas in Balaam, N. P., and Booker, J. R. 共1981兲. “Analysis of rigid rafts sup-
Method 2, the column width is matched according to Eq. 共5兲 and ported by granular piles.” Int. J. Numer. Analyt. Meth. Geomech.,
the plane-strain soil permeability is taken to be the same as the 5共4兲, 379–403.
axisymmetric case. Barksdale, R. D., and Bachus, R. C. 共1983兲. “Design and construction of
The validity of the methods has been tested in two stages: stone columns.” FHWA/RD-83/026, Federal Highway Administration,
comparison of unit-cell simulations and comparison with the field Washington, D.C.
data from an embankment case history. In the first stage, for the Barron, R. A. 共1948兲. “Consolidation of fine-grained soils by drain
case of linear-elastic material modeling, both methods produce wells.” Trans. Am. Soc. Civ. Eng., 113 共Paper No. 2346兲, 718–742.
good agreement with the benchmark results of the axisymmetric Brinkgreve, R. B. J., and Broere, W. 共2004兲. Plaxis 3D tunnel—Version 2,
and the three-dimensional models in terms of the long-term Plaxis bv, Delft, The Netherlands.
Downloaded from ascelibrary.org by UNIVERSITY OF BIRMINGHAM on 10/20/13. Copyright ASCE. For personal use only; all rights reserved.

consolidation settlements, although Method 1 gives an overesti- Brinkgreve, R. B. J., Broere, W., and Waterman, D. 共2006兲. Plaxis 2D—
mation in the consolidation rate. For the case of elastoplastic Version 8, Plaxis bv, Delft, The Netherlands.
共MC兲 material modeling however, Method 1 gives an erroneously Han, J., and Ye, S.-L. 共2001兲. “Simplified method for consolidation rate
lower long-term settlement as compared to Method 2 which gives of stone column reinforced foundations.” J. Geotech. Geoenviron.
good agreement with the benchmark results. The error in Method Eng., 127共7兲, 597–603.
Hansbo, S. 共1981兲. “Consolidation of fine-grained soils by prefabricated
1 has been attributed to its inability to simulate plastic yielding
which occurs mostly in the column material in the process of drains.” Proc., 10th Int. Conf. Soil Mechanics and Foundations Engi-
consolidation under loading. In the second stage where the em- neering, Vol. 3, Balkema, Rotterdam, The Netherlands, 677–682.
bankment construction is simulated using the MC model, a simi- Hird, C. C., Pyrah, I. C., and Russell, D. 共1992兲. “Finite element model-
ling of vertical drains beneath embankments on soft ground.” Geo-
lar discrepancy is obtained: Method 1 gives erroneously 38%
technique, 42共3兲, 499–511.
lower final settlements at two measurement points as compared
Indraratna, B., and Redana, I. W. 共1997兲. “Plane-strain modeling of smear
with the field data, whereas Method 2 presents reasonable agree- effects associated with vertical drains.” J. Geotech. Geoenviron. Eng.,
ment, albeit with slight deviation from the field settlements in the 123共5兲, 474–478.
first 60 days of consolidation. Method 2 also achieves good Indraratna, B., and Redana, I. W. 共2000兲. “Numerical modeling of vertical
agreement with the results of the three-dimensional model, de- drains with smear and well resistance installed in soft clay.” Can.
spite slight disparity in the settlement curves 共within 9% differ- Geotech. J., 37共1兲, 132–145.
ence兲 and the excess pore water pressure dissipation rate at one of McKelvey, D., Sivakumar, V., Bell, A., and Graham, J. 共2004兲. “Model-
the measurement points. The inaccuracy in Method 1 has been ling vibrated stone columns in soft clay.” Geotech. Eng., 157共3兲,
again shown to be attributed to its failure to simulate plastic yield- 137–149.
ing mainly in the column material. Thus, Method 1 would give Tan, S. A., and Oo, K. K. 共2005兲. “Stone column FEM modeling—2D
incorrect plane-strain modeling of stone-column reinforced and 3D considerations illustrated by case history.” Proc., Int. Symp.
ground when used with elasto-plastic materials and so Method 2 on Tsunami Reconstruction with Geosynthetics, ACSIG, Bangkok,
is the preferred conversion method. Thailand, 157–169.

194 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / FEBRUARY 2008

J. Geotech. Geoenviron. Eng. 2008.134:185-194.