Clayey Soil Reinforced with Stone Column

Group: Model Tests and Analyses

J. T. Shahu1 and Y. R. Reddy2
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Fully drained, load-controlled laboratory model tests and their numerical simulations are presented. The tests were performed on
adequately instrumented, small scale physical models of floating stone column group foundations placed in slurry deposited clayey soil beds
with known effective stress states. Effect of various group foundation parameters, such as area ratio, length of columns, relative density, and
moisture content of the column material is evaluated. The numerical analyses consist of three-dimensional, elastoplastic, finite-element
analyses of the model foundation. In the analyses, the clayey soil behavior is represented by the modified Cam-clay model, and the stone
column and mat are represented by the elastic, fully plastic Mohr-Coulomb constitutive relationship. The finite-element analysis was suc-
cessful in predicting the model test results with reasonable accuracy. The results are presented in nondimensional form. The major foundation
parameters affecting the group response were identified as area ratio, normalized column length, Young’s modulus of column, overconso-
lidation ratio, initial geostatic stresses, and clayey soil parameters. DOI: 10.1061/(ASCE)GT.1943-5606.0000552. © 2011 American Society
of Civil Engineers.
CE Database subject headings: Stone columns; Foundations; Model tests; Finite element method; Clays.
Author keywords: Stone columns; Clayey soil; Foundation; Model tests; Finite-element analysis.

Introduction Whereas usually it is the short-term bearing capacity of the

foundation that is critical consideration under the small groups,
For low-rise buildings and structures such as liquid storage tanks, it is the long-term drained settlement response of the large groups
abutments, embankments, and factories that can tolerate some set- that is important. Wood et al. (2000) have conducted model tests on
tlement, stone columns (also known as granular piles or granular groups of floating stone columns to evaluate long-term drained set-
columns) provide an economical method of support in compress- tlement. Although actual loading conditions on prototype founda-
ible and fine-grained soils. Stone columns are either constructed tions resemble more to load-controlled loading, these tests have
as fully penetrating through a clayey soil layer overlying a firm been carried out by using deformation-controlled loading. Simple
stratum or as floating (or partially penetrating) with their tips em- equations, such as equations of radial consolidation, were used to
calculate constant rate of deformation, and the tests were completed
bedded within the clayey soil layer.
in a matter of 8–12 h. Because the mechanism of deformation and
Physical modeling plays a fundamental role in development of
pore pressure dissipation in a floating stone column group founda-
geotechnical understanding. Model tests provide an alternative way
tion is rather complex, the fully drained behavior can only be en-
to directly reflect the behavior of the prototype under simulated sured in load-controlled tests.
conditions and are used to validate theoretical or empirical hypoth- In this paper, the results of load-controlled, fully drained model
eses. In the past, model tests have been mostly conducted on a sin- tests on groups of stone column-mat foundation placed in a slurry
gle, fully penetrating, stone column simulating unit cell behavior deposited clayey soil bed of known effective stress state are pre-
(Hughes and Withers 1974; Lee et al. 1999). Data are rather scarce sented. All tests were conducted in the laboratory at constant tem-
on model tests on unit cells with a floating stone column (Rao et al. perature and humidity conditions, and each test took approximately
1997; Ambily and Gandhi 2007). The unit cell concept assumes an 45–60 days for completion. Effect of various group foundation
infinitely wide-loaded area reinforced with stone columns having parameters, such as area ratio and length of columns is studied
constant diameter and spacing. However, stone columns are often in these tests. The test results are compared with three-dimensional,
used in limited size groups to improve the foundation soils. Very elastoplastic, finite-element analyses of the model foundation.
little data are available on model tests on groups of floating stone A brief parametric study has also been conducted by using the
columns (Wood et al. 2000 is a notable exception). finite-element analyses. The results are presented in a nondimen-
sional form.
1
Associate Professor, Dept. of Civil Engineering, Indian Institute of
Technology, New Delhi, India 110 016 (corresponding author). E-mail:
Model Setup and Test Procedure
[email protected]
2
Ph.D. Student, Dept. of Civil Engineering, Indian Institute of Technol-
ogy, New Delhi, India 110 016. E-mail: [email protected] Model Dimensions, Materials, and Measurements
Note. This manuscript was submitted on March 6, 2010; approved on
April 12, 2011; published online on April 14, 2011. Discussion period open A total of 15 model tests was conducted as shown in Table 1.
until May 1, 2012; separate discussions must be submitted for individual A schematic view of model stone column foundation is shown
papers. This paper is part of the Journal of Geotechnical and Geoenvir- in Fig. 1. The model tests were performed in perspex cylindrical
onmental Engineering, Vol. 137, No. 12, December 1, 2011. ©ASCE, tanks of 300-mm diameter, 600-mm depth, and 10-mm wall thick-
ISSN 1090-0241/2011/12-1265–1274/$25.00. ness. The depth of clayey soil bed was 300 mm. The clay bed was

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011 / 1265

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 Table 1. Summary of Laboratory Model Tests The vertical settlement was measured by using dual type dial
Test Type of p0p gauges (giving electronic signal and also showing mechanical dis-
number Ar (%) N d (mm) l (mm) sand Dr (%) (kPa) play for better control) with a sensitivity of 0.01 mm. Two such dial
gauges were placed on the top of the loading plate in a diametrically
1 10 9 13 100 Dry 50 60
opposite direction. Miniature pressure cells of capacity 200 and
2 20 13 13 100 Dry 50 60
500 kPa, having 34 mm diameter and 5 mm thickness, were em-
3 30 21 13 100 Dry 50 60
ployed to measure applied vertical stresses at the top of the model
4 10 9 13 150 Dry 50 60
foundation. The dial gauges and pressure cells were connected to a
5 20 13 13 150 Dry 50 60
nine-channel portable datalogger for recording the data. Before the
6 30 21 13 150 Dry 50 60
tests, all instruments were duly calibrated.
7 20 5 25 100 Moist 50 60
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

8 30 5 25 100 Moist 50 60 Modeling Considerations

9 10 9 13 150 Moist 50 60
10 20 13 13 150 Moist 50 60 The model tests have not been performed with any particular proto-
11 30 21 13 150 Moist 50 60 type in mind but are shown as a generic study. The model test setup
12 10 9 13 100 Dry 80 60 was designed on similar lines as done by Wood et al. (2000).
13 20 13 13 150 Dry 80 60
Important details related to boundaries, dimensions, materials,
14 10 9 13 100 Dry 50 90
and loading conditions in the model vis-a-vis the actual prototype
were given due consideration as presented subsequently.
15 20 13 13 150 Dry 50 90
Similitude ratio refers to the ratio of any linear dimension of the
model to the corresponding dimension of the prototype. For fully
drained loading conditions, the longer the drainage path, the longer
the duration of the test. To reduce the total test duration, a minimum
possible similitude ratio is desirable owing to extremely low per-
meability of the clayey soil. Also, the bigger the size of the footing,
the heavier the footing load required. A typical prototype stone col-
umn diameter varies from 0.6 to 1.0 m and length from 5–20 m. It
was observed that the model columns can be reliably installed en-
suring proper continuity and integrity if the column diameter is at
least 13 mm. Because of this, 13 mm diameter columns are used in
model tests, giving rise to a similitude ratio between 0.013 and
0.022. Usually, l=d ratio in the prototype stone columns varies be-
tween 5 and 20, in which l and d are the column length and diam-
eter, respectively. Based on this, the l=d ratio in the model tests was
adopted as 8 and 12 (corresponding to the model column lengths of
100 and 150 mm).
The model tank boundaries were determined on the basis of cri-
Fig. 1. Schematic view of stone column foundation terion that induced stresses should be insignificant at the tank boun-
daries. Assuming an equivalent footing located at two-third depth
made-up of slurry deposited Kaolinite clay consolidated at a past of the columns and 2∶1 spread, Fig. 1 shows the stress distribution
maximum consolidation pressure of 60 kPa in most cases. Few for 100 mm diameter footing and 150 mm long columns (represent-
model tests on Kaolinite clay beds, consolidated at 30 and ing the worst case). At a depth equal to twice the width of the foun-
90 kPa, were also carried out. The tests were mostly conducted dation, the induced stresses may be assumed to be approximately
on 13-mm diameter columns. Two tests were, however, also con- equal to 11% of the applied stresses. Thus, Fig. 1 shows that the
ducted on 25-mm diameter columns. The lengths of the columns induced stresses become insignificant at tank boundaries if the tank
were 100 and 150 mm, and the area ratios 10, 20, and 30%. A con- diameter and the tank depth are 300 mm or more. On the basis of
stant footing diameter of 100 mm and mat thickness of 20 mm were this, all tests were conducted in tanks of 300 mm diameter and
used in all model tests. 300 mm depth.
The columns were made-up of angular Badarpur sand, com- Prototype stone columns (diameter d ¼ 0:6–1.0 m) are usually
pacted at relative densities of 50 and 80% and at two different mois- made of stones of particle sizes D ¼ 25–50 mm. Thus, d=D ratio
ture conditions, namely, dry and moist. Badarpur sand is a approximately varies between 12 and 40 (Wood et al. 2000). The
predominantly subangular Quartz sand. The mineralogy of Badar-
column diameters used in the model tests are 13 and 25 mm. The
pur sand as obtained from a petrographic analysis by using an op-
particle sizes of the made-up Badarpur sand used in the model tests
tical microscope is as follows: 97.5% silica (SiO2 ), 1% iron oxide
as the granular material for the column and mat were kept between
(nonmagnetic), 1% clay mineral (primarily Kaolinite) and 0.5%
1 mm and 425 μm. Thus the ratio d=D in the model tests has values
muscovite. On the basis of the grain size distribution, Badarpur
sand can be classified as a poorly graded sand (SP). The original ranging from 13–59, which compare well with the corresponding
Badarpur sand was sieved and modified to get required particle values for the prototype foundation. A relative density of 50% has
sizes for the use in the model tests. The dry density of the Badarpur been used for the columns in most of the cases. A relative density of
sand in the loosest state (γdmin ), as determined by the sand pouring 80% has also been used in few tests for comparison. It is very dif-
method, was 12:93 kN=m3 and in the densest state (γdmax ), as ob- ficult to ensure a uniform stone column diameter after compaction
tained by the shake table method, was 15:75 kN=m3 . to relative density of 80%.

1266 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 Test Procedure
All tests were conducted in the laboratory at constant temperature
and humidity conditions. Each test took approximately 45–60 days
for completion.
Before filling up the test tank, silicon grease was applied on the
inner surface of the tank and then a polythene sheet was wrapped
on the inner side of the tank to minimize the side friction. The bot-
tom of the tank was provided with a 50 mm thick sand filter layer
with drainage outlets capped with porous tips. The clayey soil bed
was prepared by consolidation, starting with Kaolinite clay slurry
under gradually increased step loading under double drainage path.
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

The consolidation process took approximately 20–25 days. After

consolidation, the clay bed was completely unloaded for installa-
tion of stone columns and mat. Because the water remained avail- Fig. 2. Comparison of finite-element and model test results for 13 mm
able during unloading in the holes of the top plate and at the bottom diameter columns with different Ar (l=d ¼ 7:7)
sand filter, effective vertical stresses after unloading were estimated
to be the same as overburden stresses. In a few cases, laboratory
vane shear test (vane diameter ¼ 12 mm) was performed at the bed may thus vary between 2.5 and 6.5 kPa. Because the total depth
center of the clay bed at the location where a stone column would of the model is small, a uniform average value of p0i ¼ 5 kPa (after
be later installed. The undrained shear strength of the clay bed var- rounding off) has been adopted for all model tests. However, the
ied between 7 and 9 kPa. confining stresses at the center of the column may vary between 1.2
The stone columns were installed in a square grid pattern by a and 3.0 kPa and at the bottom of the column between 1.8
replacement method. For 13 mm diameter columns, the three area and 4.8 kPa.
ratios result into the following three groupings (Table 1): nine-
column group (Ar ¼ 10%), 13-column group (Ar ¼ 20%), and
21-column group (Ar ¼ 30%), in which Ar ¼ πðr=sÞ2 with r and Model Test Results
s as the radius and spacing of columns. For 25 mm columns, both
Ar ¼ 20 and 30% result into the five-column group. Most of the To get a better qualitative representation and to overcome the scale
columns in each group was located within the footing area except effect to some extent, it was decided to normalize the results. In the
four corner columns that were located half inside the footing and literature, either undrained strength, cu , or initial effective geostatic
half outside it. This was done in accordance to the field practice stress, p0i , has been used for normalizing stress parameters (Wood
wherein some percentage (≅ 10–20%) of the total columns is al- et al. 2000; Shahu et al. 2000). In this study, p0i has been used for
ways cast outside the loading area to achieve a confining effect. normalization because the model test results have been compared
with finite-element analysis results, and the finite-element analysis
After marking column locations, the columns were installed by
uses p0i as a foundation parameter and not cu . The settlement was
manually pushing thin perspex casing pipes of 10 mm wall thick-
normalized by the length of the column following Shahu et al.
ness and 13 mm inner diameter into the soil up to required depth.
(2000) which presents a mechanistic model for settlement analysis
The soil from inside the casing pipes was then taken out with the
of the stone column foundation.
help of an auger of 10 mm diameter. The holes thus formed were
The major parameters influencing the response of the stone col-
filled up with made-up Badarpur sand at a relative density of 50 or
umn group foundation are given as follows:
80%. A 20 mm thick and 100 mm diameter granular mat was then
compacted at a relative density of 80% on the top of the stone
δ ¼ f ðσv ; s; l; d; p0i ; p0p ; λ; κ; M; E s ; NÞ ð1Þ
column group. A footing plate in the form of a 100 mm diameter
suitably-stiffened perspex plate, with a large number of small holes
dotted over the whole plate area for drainage of water, was placed in which δ = settlement of footing; σv = applied vertical stress on
over the mat. footing; s = spacing of columns; l = length of column; d = diameter
The footing load was applied in 10–14 equal loading increments of column; p0i = initial mean effective geostatic stress; p0p = mean
of 15 kPa each with the help of a lever-arm mechanism under dou- effective preconsolidation stress; λ = slope of virgin consolidation
ble drainage path until the failure occurred. Each load increment line; κ = slope of unloading-reloading line; M = critical state ratio;
was applied until the rate of the settlement became less than Es = secant modulus; and N = number of columns.
1 mm=day. It took nearly 20–25 days for completion of this stage In the present study, these parameters are nondimensionalized
of the test. After completion of the test, the failed column group following Buckingham’s theorem of dimensional analysis given
was cast by using plaster of Paris by exhumation technique (Wood as follows:
et al. 2000). However, the casting was successful only for a few  
cases, such as one shown in the inset of Fig. 2; for most of the δ σ l Eλ
¼ f v0 ; Ar ; ; R; s0 ; M; N ð2Þ
cases, some columns in the group were broken and the attempt l pi d pi
was not successful.
As previously discussed, initial effective vertical stresses in which Ar = area ratio = ðπ=4Þðd=sÞ2 and R = overconsolidation
in the clay bed may be assumed as equal to overburden stresses. ratio = p0p =p0i . κ is generally assumed as one-fifth of λ. The reasons
Assuming total unit weight of mat ¼ 15 kN=m3 and saturated for choosing these ratios more than other forms of groupings are
unit weight of clayey soil ¼ 18 kN=m3 , effective vertical stress given at appropriate sections in which the influence of these ratios
at the center of the clay bed may be calculated as 1.5 kPa. Assum- are presented. Minor parameters having little influence over settle-
ing K 0 varying between 2 and 6 (Bowles 1997), initial effective ment of the footing are thickness of the mat and the dilation angle
geostatic stress, p0i [¼ ðσv þ σh1 þ σh2 Þ=3] at the center of the clay and the angle of shearing-resistance of sand.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011 / 1267

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Comparison of finite-element and model test results for 13 mm Fig. 5. Comparison of finite-element and model test results for
diameter columns with different Ar (l=d ¼ 11:54) different l=d ratios (Ar ¼ 10%)

Figs. 2 and 3 present model test results in terms of normalized

The effect of the normalized column length, l=d, on the normal-
applied vertical stress, σv =p0i versus normalized settlement, δ=l, for
ized stress-settlement relationship is shown in Figs. 5–7 for area
different area ratios for column lengths of 100 mm (l=d ¼ ratios of 10, 20, and 30%, respectively (Tests 1–6). Both the stiff-
7:7; Tests 1–3) and 150 mm (l=d ¼ 11:54; Tests 4–6), respectively. ness and the strength of the foundation increase with the increase in
As the normalized vertical stress increases, the normalized settle- the normalized column length. Hence, at any given normalized ap-
ment of the footing also increases for both l=d ¼ 7:7 and 11.54. plied vertical stress, the normalized settlement decreases with the
The settlement of the footing increases nonlinearly with the in- increase in the normalized column length. Also, the failure stress
creasing vertical stress up to a certain critical stress level. After this increases with the increase in the normalized column length.
critical stress level, the settlement becomes excessive with further Fig. 8 shows the effect of sand physical state. The model tests
increase in the vertical stress, indicating the failure. A versatile ul- have been conducted by constructing stone columns both with
timate load criterion defines the ultimate load as the point in which dry and moist sand (Tests 9–11). The same trends as previously
the slope of the load-settlement curve first reaches zero or a steady,
minimum value (Vesic 1963). This failure criterion has been used to
determine normalized failure stress, (pmax =p0i ). The more the area
ratio, the more the strength of the foundation and hence, the more
the failure stress (Figs. 2 and 3). For a given normalized applied
vertical stress, the settlement decreases as the area ratio increases
(Figs. 2 and 3), owing to the corresponding increase in the stiffness
of the foundation. Thus, more area ratio is beneficial in reducing the
settlement and increasing the failure stress.
The model tests have been also conducted on two different
groups of 25 mm diameter columns (Tests 7–8). The area ratio
for these groups is 20 and 30%; the total number of columns in
each group is five; and the footing diameter is 100 mm in both
cases. It was not possible to accommodate more than five columns
even for the group with area ratio = 30%. The same trends as pre-
viously discussed for 13 mm diameter column groups (Figs. 2 and Fig. 6. Comparison of finite-element and model test results for
3) were also observed for 25 mm diameter column groups (Fig. 4). different l=d ratios (Ar ¼ 20%)

Fig. 4. Normalized vertical stress versus settlement relationship for Fig. 7. Comparison of finite-element and model test results for
25 mm diameter column groups (l=d ¼ 4) different l=d ratios (Ar ¼ 30%)

1268 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. Effect of sand moisture condition on stress-settlement Fig. 10. Settlement versus vertical stress relationship for different re-
relationship lative density of sand for Ar ¼ 10% and l ¼ 100 mm; and Ar ¼ 20%
and l ¼ 150 mm (d ¼ 13 mm, dry sand)

mentioned for the dry sand (Fig. 2) were also noted for the moist
sand (Fig. 8). For a given normalized applied stress, the settlement
is higher for the case when the column is constructed with moist
sand than that for the case when the column is cast with dry sand. In
the case of moist sand, the clayey soil around the periphery of the
column would be softened owing to casting of the column, and the
stiffness of the clayey soil would decrease, leading to the higher
settlement.
Fig. 9 gives the relationship between the area ratio and the nor-
malized settlement of footing for different values of normalized
vertical stress for l=d ¼ 7:7 (Tests 1–3). As the area ratio increases
from 10 to 30%, the settlement of footing decreases for any given
value of the vertical stress. However, the decrease in the settlement
is much larger for the increase in the area ratio from 10 to 20% than
the decrease in the settlement for the increase in the area ratio from Fig. 11. Settlement versus vertical stress relationship for different
20 to 30%. For any given area ratio, the settlement becomes in- R for Ar ¼ 10% and l ¼ 100 mm; and Ar ¼ 20% and l ¼ 150 mm
creasingly larger for the same increment of applied vertical stress (d ¼ 13 mm, Dr ¼ 50%, dry sand)
(¼ 15 kPa) as one approaches failure stress.
Fig. 10 shows the relationship between the normalized vertical
stress and the normalized settlement of footing for two different relative densities becomes appreciable only near failure stress
relative densities of columns for the following two cases (Tests levels.
12–13): Ar ¼ 10% and l=d ¼ 7:7; and Ar ¼ 20% and Fig. 11 shows the normalized vertical stress versus the normal-
l=d ¼ 11:54. In both cases, as the relative density of stone columns ized settlement relationship for two different values of overconso-
increases, the stiffness of the stone column group would increase, lidation ratio (R) for Ar ¼ 10% and l=d ¼ 7:7; and Ar ¼ 20% and
and hence, the settlement of footing decreases. However, this ben- l=d ¼ 11:54 (Tests 14–15). The more the overconsolidation ratio,
eficial effect of more density is negligible at low applied stress lev- the more the stiffness of the clay bed; and the stiffer the clay bed,
els. The difference in the settlement of footing for the two different the more the stiffness of the group foundation. Thus, with the in-
crease in R, the stiffness of the group foundation increases and as a
consequence, the settlement decreases (Fig. 11). The results for
Tests 16–17 could not be compared with any other tests because
more than one variable were varying in these tests.

Finite-Element Analysis

Details of Finite-Element Simulation

For finite-element analyses of the laboratory model tests, three-
dimensional finite-element models of exactly the same size as the
laboratory foundation models were prepared [Figs. 12(a)–12(c)]
and analyzed by using ABAQUS software. The material properties
of different foundation components were described with a few well-
Fig. 9. Normalized settlement versus area ratio relationship for differ- established constitutive relationships: the clayey soil by modified
ent applied vertical stress for l ¼ 100 mm (d ¼ 13 mm; p0p ¼ 60 kPa; Cam-clay model (Roscoe et al. 1958) and the stone columns and
Dr ¼ 50%; and dry sand)
mat by Mohr-Coulomb’s elastic-perfectly plastic model.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011 / 1269

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 consolidation stage under elevated value of backpressure ranging
from 350–450 kPa for 24 h and then measuring B-value as
0.97– 0.99. The parameter p0p was assumed as preconsolidation
stress for slurry deposited clayey soil samples. Poisson’s ratio μ
was assumed as 0.33 on the basis of the available literature on
Kaolinite clay.
The angle of shearing resistance, ϕ0 , and dilatancy, ψ, for made-
up Badarpur sand compacted at a relative density of 50% were ob-
tained by conducting drained triaxial tests on saturated samples.
The secant modulus E s at peak, at confining stresses of 20 and
50 kPa, was determined from deviator stress versus axial strain
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

plots obtained from the drained triaxial tests by drawing a straight

line from the origin to the peak failure stress. Secant modulus E s is
highly dependent on confining stress. As the confining pressure
increases, the stiffness of the specimen increases, and hence, E s
increases. Because confining stresses are of the order of 5 kPa
in the model tests, Es should be evaluated at the relevant confining
stress for use in model test analyses. Because it is difficult to con-
duct triaxial tests at such a low confining stress, Es value at the
confining stress of 5 kPa (¼ 1;250 kPa) was calculated by the lin-
ear extrapolation from E s values at 20 and 50 kPa. Poisson’s ratio μ
was assumed as 0.35 on the basis of available data in the literature
(Bowles 1997). The constitutive material parameters for the granu-
Fig. 12. Clayey soil components and mesh discretization lar mat were kept the same as that for the stone columns. A small
parametric study indicated that virtually no difference is obtained in
the response owing to this.
The back-predicted stress-strain response of the triaxial tests, on
The material properties used in the analyses for the clayey soil both Badarpur sand and Kaolinite clay by the finite-element analy-
and the column and mat are listed in Table 2 as the nominal value. λ ses by using the previously discussed constitutive parameters
and κ were determined from the correlations with compression in- (Table 2), matched reasonably well with the actual test data, thereby
dex C c and swelling index C s (Wood 1990), which were deter- establishing the accuracy of adopted models and evaluated param-
mined from consolidation tests. Based on this, λ and κ were eters (Reddy 2010).
obtained as 0.11 and 0.025, respectively. The correlation of com- The stone column-mat foundation components were simulated
pression index with liquid limit for remolded soils as given in the by using 20-noded quadratic hexahedral, reduced integration ele-
literature (Lambe and Whitman 1979) gives C c ¼ 0:322 and from ments. Each component of the stone column-mat foundation was
this value, λ ¼ 0:14 is obtained. This value matches fairly well meshed individually. The mesh was created automatically for the
with the λ-value, previously evaluated. Because κ is usually as- stone columns and mat, but the clay bed required three partitions
sumed as one-fifth of the λ-value, the evaluated κ-value is also rea- perpendicular to one other [Figs. 12(a)–12(c)], two in the vertical
sonable. The slope of critical state line, M (¼ 0:703), was obtained direction and one in the horizontal direction at the level of the bot-
from the angle of shearing resistance, ϕ0 , which was evaluated by tom of the columns, before the mesh could be automatically gen-
conducting slow Consolidated Undrained triaxial tests with pore erated. Tie constraints were used at all interfaces (i.e., no slip or
water pressure measurements (CU) on slurry deposited separation was allowed at the interfaces). Because the interface
Kaolinite samples. The full saturation was ensured by performing is the mixed zone of granular particles and the clayey soil, the shear
properties of the interface are expected to be more than that of the
clayey soil (Ambily and Gandhi 2007). By adopting the tie con-
Table 2. Nominal Set of Group Parameters and the Range of Variation
straint, these higher interface properties are neglected in this analy-
Foundation instances Parameter Nominal value Range adopted sis on a conservative side. Roller supports were used on the lateral
Clayey soil λ 0.11 0.073–0.22 (vertical) faces of the clay bed and mat. The bottom face of the clay
bed was considered as fixed. The default value of most of the com-
κ 0.025 0.025–0.055
putation control parameters as suggested by ABAQUS was used
M 0.703 0.703–1.2
except for some parameters that were changed to achieve the maxi-
p0p (kPa) 60 16–150
mum possible iterations (Reddy 2010).
μ 0.33 0.33 Fig. 12(d) shows the mesh discretization adopted for the nine-
p0i (kPa) 5 3–60 column model group foundation. A detailed mesh sensitivity analy-
e0 1.09 1.09 sis was carried out before adopting mesh discretizations for various
Column and mat ϕ0 43.42° 43.42° group configurations by varying the number of elements in each
ψ 8° 8° component of the group foundation and comparing the results
Es (kPa) 1250 250–2,000 of different combinations (Reddy 2010). However, owing to a pe-
μ 0.35 0.35 culiar nature of the group foundation, the increase in the fineness of
Geometry Ar (%) 30 10–30 the clay bed mesh does not automatically lead to a usual type of
N 21 9–21 convergence of the results. Owing to a small spacing between the
stone columns, the dimensions of clayey soil elements between two
l (mm) 100 100–150
neighboring stone columns are very small [Fig. 12(d)]. With refine-
d (mm) 13 13
ment in the mesh, this zone becomes too dense, resulting in poor

1270 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 13. Response contours of group foundation at failure

and skewed elements, which many times result into a premature Table 3. Comparison of Model Tests and FEM inPmax and ρmax
termination of computer runs. Hence, a less fine clayey soil Pmax (kPa) ρmax (mm)
mesh [Fig. 12(d)], which gave consistent results for all cases, was
adopted. The maximum error in the ultimate vertical stress and the l Ar p0i Model % Model %
(mm) (%) (kPa) FEM tests change FEM tests change
corresponding settlement owing to this has been estimated to be
less than 25%. The predicted stresses are expected to be on the 100 10 3 79 75
5 20 19
5
higher side and the settlement on the lower side. 20 90 120 25 29 28
4
Fig. 13 shows the response contours of 21-column group foun- 30 95 135 30 27 25
8
dation at failure. Fig. 13(a) demonstrates that as one moves away 150 10 3 106 120 12 31 26
19
from the center of the group, the bending of the columns increases. 20 108 150 28 35 29
21
Central columns almost do not bend and remain vertical. The col-
30 105 165 36 26 32 19
umns along the periphery bend more than the columns that lie in-
100 10 5 77 75
3 12 19 37
side the group. Among the peripheral columns, the corner columns
bend more than the columns on the edges. The vertical deformation 20 105 120 13 24 28 14
is maximum at the center column and decreases as one moves away 30 106 135 21 18 25 28
from the center. Similarly, the vertical deformation is maximum on 150 10 5 120 120 0 25 26 4
the top of the column and decrease as one moves down the column. 20 120 150 20 26 29 10
Active yielding is observed in all the three components, namely, 30 135 165 18 27 32 16
stone columns, mat, and clayey soil [Fig. 13(b)]. The plastic strain
is high in the clayey soil zone below the footing area up to the
bottom of the stone columns. The plastic strain in the stone col- Pmax and ρmax by the finite-element analysis vary between
5 and
umns and the mat is very small. 36% and
21 and 19%, respectively, as compared with the mea-
Comparison of Analysis Results sured results when the finite-element predictions are made by using
with Model Test Results p0i ¼ 3 kPa (Table 3). The variations for both Pmax and ρmax de-
crease when the predictions are made by using p0i ¼ 5 kPa. For this
The results of the finite-element analysis have been calibrated case, the predicted values of Pmax and ρmax vary between
3 and
by a detailed comparison with the results of the laboratory model 21% and 4 and 37%, respectively.
tests. Whereas Figs. 2 and 3 compare the results in the form of The differences in the finite-element predictions and the model
normalized stress-settlement relationship for different area ratios,
test results may arise because of the following issues in the finite-
Figs. 5–7 compare the corresponding results for different column
element analyses: mesh convergence issues, uncertainties in the
lengths. A close agreement has been obtained between the finite-
evaluation of constitutive parameters, and inappropriateness of
element predictions and the measured-model test results.
The results are quantitatively compared in terms of the ultimate the constitutive model used for the granular material. The error in-
vertical stress Pmax and the settlement ρmax corresponding to Pmax duced because of mesh convergence issues has been previously dis-
in Table 3 for different area ratios and column lengths. In view of cussed. The model parameter E s has some uncertainty because it
some subjectivity involved in the estimation of initial geostatic has been calculated as the secant modulus at the required confining
stresses p0i in the model tests, the finite-element predictions have stress by extrapolation. Some doubts also exist regarding estima-
been made using both p0i ¼ 5 and 3 kPa. E s ¼ 1;250 kPa has been tion of initial geostatic stress in the model foundation on account
used for the predictions with p0i ¼ 5 kPa, and Es ¼ 750 kPa has of uncertainties in estimation of K 0 . Next, the constitutive model
been used for the predictions with p0i ¼ 3 kPa on the basis of for the granular material in terms of the constant E s value is not
the linear extrapolation as explained earlier. The ultimate vertical appropriate. E s is nonlinearly dependent on confining stress. Dur-
stress for both model tests and finite-element predictions was as- ing the model test, Es is expected to remain small during initial
sumed as the load at which a constant final rate of penetration was stages of loading because of low confinement. As the test pro-
achieved following Vesic (1963) criterion. The predicted values of gresses, Es will increase with the increase in confining stresses.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011 / 1271

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 This is, however, not possible to simulate with a constant Es model
used in the present analysis, thus introducing some error in the
calculations.

Parametric Study
A sensitivity analysis of some important parameters is necessary to
evaluate the error that may be caused owing to uncertainties in the
evaluation of these parameters. Also, although the effect of certain
parameters, such as area ratio and length of columns, on group re-
sponse has been previously discussed, still several important var-
iables such as clayey soil parameters, E s , and R, are remaining
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

whose effect on the group response has not been evaluated. Hence,
a parametric study of the group response for these parameters has
been carried out. Fig. 16. Effect of stress ratio M on load-settlement behavior for Ar ¼
The parametric study has been conducted by first choosing a 30% and l ¼ 100 mm (E s ¼ 1;250 kPa; p0i ¼ 5 kPa; p0p ¼ 60 kPa; and
nominal set of values of model group foundation parameters, which ϕ0 = 43.8°)
was kept fixed, and then varying the value of each group parameter
within a practical range with respect to the nominal set (Table 2).
The chosen nominal set is representative of the 21-column group columns increases with the increase in p0i (and decrease in R).
model foundation with area ratio of 30% and column length of Increase in lateral confinement makes the group foundation stiffer,
100 mm. The measured test result for the 21-column group foun- resulting in more applied vertical stress for a given normalized
dation has also been superimposed on the parametric variation for settlement. However, the increase in the applied vertical stress
comparison (Figs. 14–16). σv is proportionately smaller than the increase in p0i , thus resulting
Fig. 14 depicts the effect of the overconsolidation ratio, R in less σv =p0i with decrease in R.
(¼ p0p =p0i ), on the normalized stress-settlement relationship. In this Several researchers have used the parameter E s =Ec (e.g.,
figure, R has been varied by keeping p0p constant but changing p0i . Bergado et al. 1996) for normalization of stone column responses,
First, at any given normalized applied vertical stress, the settlement in which Ec is the Young’s modulus of the clayey soil. Shahu et al.
decreases as R increases. The lateral confinement to the stone (2000) have used a nondimensional parameter Rs as
 
Cc Es
Rs ¼ 0:434 ð3Þ
ð1 þ e0 Þ p0i

Also, it was experienced during the analyses that λ and κ of the

clayey soil and Es of the granular material were somehow related:
for the given λ and κ, the computations could only be run up to a
certain maximum value of Es ; when λ and κ were reduced (indi-
cating that the soil had become stiffer), the upper limit of E s to
which computations could now be run increased. Hence, it was de-
cided to normalize the stiffness of soil and column by a single
parameter: E s λ=p0i . Fig. 15 depicts the normalized stress-settlement
relationships for different values of Es λ=p0i for two different R val-
ues for the nominal case. At any given normalized applied vertical
stress, the settlement decreases as E s λ=p0i increases for both R val-
ues. E s λ=p0i is indicative of relative stiffness of the composite
Fig. 14. Effect of R on load-settlement behavior for Ar ¼ 30% ground. For example, an increase in E s λ=p0i is indicative of increase
in column stiffness, keeping λ and p0i constant. Thus, with the in-
crease in relative stiffness Es λ=p0i , the reinforced ground becomes
stiffer and the settlement decreases. This trend is similar to Bergado
et al. (1996) wherein as Es =E c increases, the settlement decreases.
Similarly, Shahu et al. (2000) have observed that as Rs increases,
the settlement decreases.
Different combinations of λ, E s , p0p , and p0i have been tried to
further investigate the normalization behavior of Es λ=p0i . The val-
ues of E s λ=p0i ¼ 6:9 and 27.5 were obtained by two different com-
binations of λ, E s , p0i , and p0p , and the corresponding responses were
plotted in Fig. 15 as curves A (λ ¼ 0:11; Es ¼ 500 kPa;
p0i ¼ 8 kPa; and p0p ¼ 32 kPa) and B (0.11, 250, 4, and 16), and
curves F (0.073, 1,875, 5, and 60) and G (0.11, 1,250, 5, and
60), respectively. In both cases, different combinations resulted
in approximately the same response. This is indicative of the val-
idity of the normalization by E s λ=p0i.
Fig. 15. Effect of secant modulus E s λ=p0i on load-settlement behavior
Fig. 16 gives the variation in the stress-settlement relationship
for Ar ¼ 30%
for different stress ratio M of the clayey soil bed for the nominal

1272 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 used for checking the estimation of long-term drained settle-
ment of prototype foundations by other methods. An example
of settlement estimation by using the nondimensional charts is
given in Appendix A.
2. In general, a practical analysis or design should require as
few materials and stress parameters as possible. The present
study identifies major and minor parameters affecting the re-
sponse of a floating stone column group foundation. This para-
meter identification, along with the suggested nondimensional
groupings, can be used in the development of a rational ana-
lysis for the settlement estimation of the stone column founda-
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

tion. Such a rational analysis can then be calibrated by using

the model test results presented in this study.
Fig. 17. Comparison of present model test results with Wood et al. 3. The details of the finite-element analysis of the model founda-
(2000) data tion conducted in this study can be directly used in the analysis
of any particular case of a prototype floating stone column
group foundation. Beside this, a general methodology can also
case. As the value of M increases, the clay bed becomes stronger be formulated for the analysis of the prototype foundation by
and the ultimate failure stress of the group foundation increases. performing finite-element analyses of prototype foundations of
The normalized settlement is practically unaffected by the criti- different sizes on similar lines as described here for the model
cal state ratio M of the clayey soil as long as the normalized applied foundations.
stress is less than 10 (Fig. 16). M is representative of the failure
stress condition; and hence, the effect of M on the normalized set-
Conclusions
tlement begins to show up only at higher values of normalized ap-
plied stress when a significant volume of the foundation soil has 1. The major parameters affecting the response of a group of
reached the failure state. On the other hand, the normalized settle- floating stone column foundations are area ratio Ar , normalized
ment is significantly affected by the overconsolidation ratio R column length l=d, overconsolidation ratio R, relative stiffness
(¼ p0p =p0i ) even at low values of the normalized applied stress E s λ=p0i , and stress ratio M of the clayey soil. At any given nor-
(Fig. 14). The less the p0p (and thus R for a given p0i ), the less malized vertical stress σv =p0i , normalized settlement δ=l de-
the applied stress value at which the plastic deformation begins. creases as Ar , l=d, R, E s λ=p0i , and M increase. The failure
Hence, large settlements can occur even at low values of applied stress increases as Ar , l=d, R, E s λ=p0i , and M increase. Thus,
stress if p0p (or R) is small. a foundation with more area ratio and more l=d ratio is ben-
Most of the available experimental or analysis data in the liter- eficial in reducing the settlement and increasing the failure
ature pertains to the response of fully penetrating stone columns by stress.
using unit cell simulation. The only relevant data available are 2. The minor parameters affecting the group foundation response
strain controlled model test results on groups of floating stone col- are thickness of mat in addition to dilation angle and angle of
umns by Wood et al. (2000). These data have been compared with shearing resistance of granular material.
the present test data in form of applied vertical stress versus settle- 3. The bending of a column is dependent on the position of the
ment relationship in Fig. 17. The major differences between the two column in the group. The bending increases as one moves from
sets of data pertain to the stress history of the clayey soil and the the center column to a peripheral column and to a corner col-
time duration of the tests. In Wood et al. (2000), the clayey soil was umn. The plastic strain at failure is highest in the clayey soil
consolidated to 120 kPa and then unloaded to 30 kPa (R ¼ 4), and zone below the footing extending up to the bottom of the col-
the tests were completed in 8–12 h. The Wood data exhibit more umns. The stone columns and mat show very small plastic
initial stiffness and sudden failure at relatively smaller ultimate load strain.
than the present test data. The Wood data for Ar ¼ 30% have also 4. A close agreement has been obtained between the finite-
been compared with the present test and analyses results in non- element predictions and the measured-model test results.
dimensional form in Fig. 14. The difference in soil stress history The predicted values of ultimate load Pmax and the correspond-
can be accounted for in this figure by comparing the Wood data ing settlement ρmax , by using p0i ¼ 5 kPa by the finite-element
with analyses results for R ¼ 4. Still, some differences remain be- analysis, lie within
3 to 21% and 4 to 37%, respectively, as
tween fully drained analyses results and the Wood data that can compared with the measured results.
only be accounted for considering test duration. Given the short 5. The inaccuracy in the finite-element results may arise because
duration of the Wood tests, it is possible that these tests may of mesh convergence issues, the subjectivity/uncertainties in
not be fully drained. This would also explain for smaller ultimate the evaluation of the constitutive model parameters, and inap-
load for the Wood data in Fig. 14 in spite of more p0i and p0p as propriateness of the constitutive model employed for the gran-
compared with the present data. ular material.

Application Appendix. Practical Application

Some practical applications of the findings of the present study are Consider a design of a floating stone column foundation for a 10 m
listed as follows: diameter water tank. The tank height is 15 m, and a uniform load
1. The results of model analysis have been presented in the form intensity (σv ) on the stone column foundation is 150 kN=m2 .
of nondimensional charts. Although the charts are based on the The clayey soil extends to a great depth. The clayey soil properties
model analyses because they are nondimensional, they may be are assumed as follows: γtotal ¼ 18 kN=m3 ; λ ¼ 0:11; κ ¼ 0:025;

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011 / 1273

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.

 M ¼ 1:0; and p0p ¼ 120 kPa. The initial geostatic stress, p0i PE33 = vertical plastic strain component;
(¼ 30 kPa), is constant from the top to the bottom. The gravel used p0i = effective mean initial geostatic stress;
for the stone columns and mat has the following properties: p0p = effective mean preconsolidation stress;
Young’s modulus = 7.5 MPa, angle of internal friction = 43° R = p0p =p0i = overconsolidation ratio in effective mean stresses;
and dilatancy angle = 8°. The design steps are given as follows: Rs = nondimensional parameter;
Step 1: Choose stone column diameter d between 0.5 and 1 m. r = radius of stone column;
Adopt d ¼ 1 m. s = center-to-center spacing of the stone columns;
Step 2: Choose l=d ratio between 5 and 20. Adopt l=d ¼ 7:7. U3 = vertical displacement component;
Therefore, the length of columns l ¼ 7:7 m. δ = settlement of footing;
Step 3: Choose area ratio Ar between 15 and 35%. Adopt ϕ = friction angle;
Ar ¼ 30%. The spacing (s) between the columns is obtained from γdmax = dry density in densest state;
Ar ¼ πðr=sÞ2 . Therefore, s ¼ 1:618 m γdmin
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/17/14. Copyright ASCE. For personal use only; all rights reserved.

= dry density in loosest state;

Step 4: For s ¼ 1:618 m, the number of columns that can be κ = slope of unloading-reloading line;
accommodated in a square grid pattern covering a 10 m diameter λ = slope of virgin consolidation line;
circular area is 7 × 7. Thus, the actual area covered by these col- μ = Poisson’s ratio;
umns is 10:71 × 10:71 m, which represents an area extending more ρmax = settlement corresponding to Pmax ;
than 10% of the total tank area beyond the footprint of the tank. σh1 = horizontal stress in X-direction;
Thus, the number of columns = 49 may be adopted. σh2 = horizontal stress in Y-direction;
Step 5: The settlement of the stone column group foundation (δ) σv = applied vertical stress on footing; and
can be determined from Fig. 15. Thus, for R ¼ p0p =p0i ¼ ψ = dilatancy angle.
120=30 ¼ 4; Es λ=p0i ¼ 27:5; σv =p0i ¼ 5; and M ¼ 0:703, Fig. 15
gives δ=l ¼ 0:025. Next, Fig. 16 can be used to approximately ex-
tend the δ=l value obtained for M ¼ 0:703 to that for M ¼ 1:0 by References
extrapolation. Fig. 16 shows that for σv =p0i ¼ 5, the δ=l value for
M ¼ 0:703 remains the same as that for M ¼ 1:0. Ambily, A. P., and Gandhi, S. R. (2007). “Behavior of stone columns based
on experimental and FEM analysis.” J. Geotech. Geoenviron. Eng.,
Thus, for M ¼ 1, δ=l ¼ 0:025.
133(4), 405–415.
Therefore, δ ¼ 0:193 m. Bergado, D. T., Anderson, L. R., Miura, N., and Balasubramaniam, A. S.
In this example, nominal values of all design parameters were (1996). Soft ground improvement in lowland and other environments,
used to simplify the design. For other values of design parameters, American Society of Civil Engineers, New York.
an interpolation or extrapolation is suggested. Bowles, J. E. (1997). Foundation analysis and design, 5th Ed.,
McGraw-Hill, New York.
Hughes, J. M. O., and Withers, N. J. (1974). “Reinforcing of soft cohesive
soils with stone columns.” Ground Eng., 7(3), 42–49.
Lambe, T. W., and Whitman, R. V. (1979). Soil mechanics, SI version,
Notation Wiley, New York.
Lee, K. M., Manjunath, V. R., and Dewaikar, D. M. (1999). “Numerical and
model studies of strip footing supported by a reinforced granular fill-
The following symbols are used in this paper:
soft soil system.” Can. Geotech. J., 36(5), 793–807.
Ar = area replacement ratio; Rao, S. N., Reddy, K. M., and Kumar, P. H. (1997). “Studies on groups of
C c = compression index of clayey soil; stone columns in soft clays.” Geotech. Eng., 28(2), 165–182.
C s = swelling index of clayey soil; Reddy, Y. R. (2010). “Soft soil reinforced with granular pile-mat system:
cu = undrained shear strength of clayey soil; analysis and model tests.” Ph.D. thesis, IIT Delhi, New Delhi, India.
D = particle size of sand; Roscoe, K. H., Schofield, A. N., and Wroth, C. P. (1958). “On the yielding
Dr = relative density; of soils.” Geotechnique, 8(1), 22–53.
d = diameter of stone column; Shahu, J. T., Madhav, M. R., and Hayashi, S. (2000). “Analysis of soft
E s = secant modulus of sand; ground-granular pile-granular mat system.” Comput. Geotech., 27(1),
E c = secant modulus of clayey soil; 45–62.
Vesic, A. S. (1963). Bearing capacity of deep foundations in sans, Vol. 39,
e0 = initial void ratio;
National Academy of Sciences, National Research Council, Highway
K o = coefficient of earth pressure at rest; Research Record, Washington, DC, 112–153.
l = length of stone column; Wood, D. M. (1990). Soil behavior and critical state soil mechanics,
M = critical state ratio; Cambridge University, London.
N = number of columns; Wood, D. M., Hu, W., and Nash, D. F. T. (2000). “Group effects in stone
Pmax = failure stress; column foundations: Model tests.” Geotechnique, 50(6), 689–698.

1274 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2011

J. Geotech. Geoenviron. Eng. 2011.137:1265-1274.