You are on page 1of 11

Available online at www.sciencedirect.

com

Computers and Geotechnics 35 (2008) 537547


www.elsevier.com/locate/compgeo

Eect of soil spatial variability on the response of laterally


loaded pile in undrained clay
Sumanta Haldar 1, G.L. Sivakumar Babu

Department of Civil Engineering, Geotechnical Engineering Division, Indian Institute of Science, Bangalore 560 012, India
Received 3 April 2007; received in revised form 11 October 2007; accepted 11 October 2007
Available online 26 November 2007

Abstract
A comprehensive study is performed on the allowable capacity of laterally loaded pile embedded in undrained clay having spatial
variation of strength properties. Undrained shear strength is considered as a random variable and the analysis is conducted using random
eld theory. The soil medium is modeled as two-dimensional non-Gaussian homogeneous random eld using Cholesky decomposition
technique. Monte Carlo simulation approach is combined with nite dierence analysis. Statistics of lateral load capacity and maximum
bending moment developed in the pile for a specied allowable lateral displacement as inuenced by variance and spatial correlation
length of soils undrained shear strength are investigated. The observations made from this study help to explain the requirement of
allowable lateral capacity calculations in probabilistic framework.
 2007 Elsevier Ltd. All rights reserved.
Keywords: Clay; Lateral loads; Piles; Probabilistic analysis; Spatial variability

1. Introduction
Assessment of lateral load capacity of piles embedded in
undrained clay medium, exhibiting spatial variability is of
considerable importance in geotechnical engineering.
Hence, stochastic treatment for analysis of soil spatial variability and probabilistic models for assessment of lateral
allowable load are necessary. Due to inherent variability,
property variations in the in-situ soil normally exhibit a
trend with distance and scatter and that are represented
by the mean value, coecient of variation and correlation
distance. The property values are correlated to each other
at adjacent points, and the distance up to which this significant correlation exists is termed as correlation distance.
The eect of inherent random variations of soil properties
on the response of foundation structures received consider*

Corresponding author. Tel.: +91 80 22933124; fax: +91 80 23600404.


E-mail addresses: sumanta@civil.iisc.ernet.in (S. Haldar), gls@civil.
iisc.ernet.in (G.L. Sivakumar Babu).
1
Tel.: +91 80 22932815; fax: +91 80 23600404.
0266-352X/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2007.10.004

able attention in the recent years. Griths and Fenton [1],


Fenton and Griths [2] examined the response of shallow
foundations; Fenton and Griths [3], Haldar and Babu [4]
analyzed response of deep foundations under vertical load.
The present study focuses on the response of laterally
loaded pile in a spatially varied soil media. There are two
aspects in designing laterally loaded pile foundations: (i)
maximum lateral displacement at pile head and (ii) maximum bending moment in the pile. If these two aspects
are satised, pile is considered to be safe and the load corresponding to allowable lateral displacement can be considered as lateral load capacity of the pile. Several methods
are described in literature to determine the lateral capacity
and failure mechanism of pile in a homogeneous soil. Few
of them are summarized as follows: Broms [5] assumed a
limiting resistance of 9su to determine the ultimate lateral
load. Based on upper bound analysis, Randolph and Houlsby [6] indicated that the value 9su is largely empirical and
suggested values in the range of 9.14su11.94su. Phoon and
Kulhawy [7] indicated that interpretation based on specied displacement limit, rotational limit or moment limit,

538

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

and hyperbolic capacity do not consider the actual soilshaft behaviour and hence, moment limit is a better choice
to compute lateral capacity. Fan and Long [8] showed that
the existing methods for predicting soil resistance like API
method, Hansens method and Broms method give dierent values of prediction of soil resistance for the same soil.
Poulos and Davis [9], Hsiung and Chen [10] indicated that
in designing pile foundations under lateral loads, the maximum lateral displacement controls the design rather than
the ultimate resistance. Zhang and Ng [11] indicated that
many geotechnical structures such as building foundations
are more often governed by allowable displacement
requirements rather than by ultimate limit requirements.
Hence, this paper considers the evaluation of lateral load
capacity based on specied allowable/serviceable lateral
displacement.
The objective of this paper is to investigate the statistics
of allowable lateral capacity of pile for a specied allowable lateral displacement and maximum bending moment,
as inuenced by spatial variation and correlation structure
of soils undrained shear strength. The lateral capacity is
dened as lateral load for a specied lateral allowable displacement and this lateral capacity is termed as allowable
load throughout this paper. The present study is conducted
using random eld theory combined with nite dierence
code, Fast Lagrangian Analysis of Continua, FLAC [12].
Two-dimensional non-Gaussian homogeneous random
eld is generated by Cholesky decomposition technique.
Monte Carlo simulation is conducted to determine the statistics of the pile response. The allowable load is computed
based on generated lateral loaddisplacement curves. The
maximum bending moment corresponding to allowable
load is also determined. Propagation of failure and formation of failure mechanism close to the pile foundation are
examined considering soil stiness and shear strain level
in soil near pile. The following sections present the details
of analysis, typical results obtained and the conclusions
from the study.
2. Method of analysis
2.1. Overview of nite dierence model
The nite dierence program uses the 4-noded quadrilateral grids. The total soil medium is divided into nite difference grids for analysis. Appropriate boundary
conditions are applied in the soil zone. At the bottom plane
of the grid, all movements are restrained. The lateral sides
of the mesh are free to move in downward direction (ve
Y-axis) but not in the X-direction. In order to investigate
the pile response a vertical pile is considered which is
embedded in soil media. The soil is modeled using elastoplastic MohrCoulomb constitutive model and the pile is
modeled using linearly elastic beam elements with interface
properties (termed as pile element). Each element has threedegrees-of freedom (two displacements and one rotation)
at each node. The pile elements interact with the nite dif-

ference grids via shear and normal coupling springs which


are represented by appropriate stiness values. The coupling springs are similar to the load/displacement relations
provided by py curves. However, py curves are
intended to capture the interaction of the pile with the
whole soil mass, while in the present analysis nonlinear
springs represent the local interaction of the soil and pile
elements [12]. The interaction of pile with grid is represented by four parameters: (i) kn = normal stiness, (ii)
ks = shear stiness, (iii) cohesive strength of shear spring
that prescribes the limiting shear force at pilesoil interface,
(iv) cohesive strength normal spring which prescribes limiting normal force. The values of the interface parameters
can be derived from the undrained cohesion/shear modulus
of soil. The shear and normal stiness (expressed in stressper-distance units) are obtained as [12]


K 43 G
k n or k s 10  max
1
Dzmin
where K and G are the bulk and shear modulus of soil zone,
respectively; and Dzmin is the smallest dimension of an
adjoining zone in the normal direction. The cohesive
strength of shear coupling spring (expressed in force-perdistance units) can be taken as the pile perimeter times
the undrained cohesion of the soil (e.g., for a circular pile,
2p times the radius). Cohesive strength of the normal coupling spring (expressed in force-per-distance units) can be
considered as limiting lateral resistance and it can be computed based on Broms solution [5] as 9 su Dp, where Dp
is the pile diameter. Lateral load is applied incrementally at
the top of the pile head.
In a numerical analysis, Donovan et al. [13] suggest that
the linear scaling of material properties is the convenient
way of distributing the discrete eect of elements over a
regularly spaced distance between the elements. A threedimensional pile with regularly spaced interval can be
reduced to two-dimensional problem considering averaging
over the distance between the elements. The relation
between actual properties and scaled properties can be
described by considering the strength properties for regularly spaced piles. For a pile element, the following properties are to be scaled: (i) elastic modulus, (ii) stiness of the
interface springs and (iii) pile perimeter. The input parameters are given as the actual values, divided by the spacing
of the piles. The actual pile responses (forces and moments)
are determined by multiplying the spacing value.
2.2. Validation of numerical scheme
The validation of the nite dierence code is conducted
with reference to a pile test result (lateral load verses lateral
ground line displacement) reported in Ref. [9]. The pile test
data is presented in Table 1. A 10 m 10 m size eld is considered and the entire eld is divided into 900 elements (30
numbers row-wise and 30 numbers column-wise). As indicated earlier, the soil is modeled using elasto-plastic Mohr
Coulomb constitutive model and the pile is modeled as lin-

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

539

approach in obtaining the response of laterally loaded pile


in soft clay.

Table 1
Pile and soil data [9]
Pile data

Value

Soil data

Value

Pile diameter, Dp (m)

0.038

14.40

2.3. Deterministic analysis

Pile length, Lp (m)


Bending rigidity (EI)
(kN m2)

5.25
31,600

Undrained shear strength


(kN/m2)
Bulk density (kN/m3)
Shear modulus, G (kN/m2)

19
925.71

Bulk modulus, K (kN/m2)

4320

The focus in this study is to understand the response of a


free head concrete bored pile of 1.0 m diameter (Dp) and
length of 10 m (Lp). Hence, the analysis uses a dierent
eld size (20 m 20 m) in the following sections to avoid
any boundary eect. Total soil medium is discretized into
900 numbers of nite dierence elements in 30 rows and
30 columns of equal size with each side dimension of
0.67 m. A uniform value of su = 20 kPa is employed to nd
out the deterministic allowable load for homogeneous soil.
The same value of undrained shear strength is considered
as mean undrained shear strength for the probabilistic
analysis which is explained in the later section. Pile is eccentric (e) over the ground by 1.0 m. The pile element is
divided into 21 equal segments. Pilesoil interface properties are scaled to represent plane strain condition. Fig. 2
illustrates the schematic diagram of the model. The actual
values (not scaled) of the pile stiness parameters are given
in Table 2. Pile loaddisplacement curve for pile head is
obtained and allowable load for homogeneous soil is computed corresponding to the allowable displacement of
0.0508 m, which is the upper limit for lateral displacement
allowed by AASHTO [14]. The maximum bending moment
is also obtained corresponding to the allowable load. The
random eld is generated in terms of undrained shear
strength values assigned for each grid location. The details
of random eld model and the approach used in merging
the random soil properties in grid locations in the nite difference code are described in the following section.

early elastic pile element and is discretized into 20 equal


segments. The values of bulk and shear modulus, bulk density of soil, pile properties namely diameter, length and
exural rigidity are given in Table 1. The shear and normal
coupling springs properties are obtained as follows:
The values of normal (k
 n) and shear stiness (ks) are
obtained as 10 times of K 43 G =Dzmin . The value of
Dzmin = 0.33 m (10 m/30 = 0.33 m). Hence value of kn
and ks are set to 1.68 105 kN/m/m. The cohesive strength
of shear coupling spring is the pile perimeter times
undrained shear strength; hence, value is p 0.038
14.4 = 1.72 kN/m. According to Broms [5], the value of
cohesive strength of the normal coupling spring can be estimated as 9 su Dp = 9 14.4 0.038 = 4.928 kN/m. The
input parameters are linearly scaled as the model is reduced
to 2D problem. The lateral load is applied incrementally.
The unbalanced force of each node is normalized by gravitational force acting on that node. A simulation is considered to have converged when the normalized unbalanced
force of every node in the mesh is less than 103 and the
results are obtained. Fig. 1 shows the experimental and predicted loaddisplacement curves. The results show that
predicted values are in good agreement with the experimental values. This validation lends credence to the use of the

e=1.0 m
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
DP=1.0 m
X
X
Y
X
X
X
X
X
X
X
X
X
X
X
BYY YYYY YYY YYYY YYYY YYYY YYYY YYYY

LP=10 m

Lateral load (kN)

Measured (Poulos and Davis, 1980)


Predicted
3

0
0

0.005

0.01

0.015

Ground line lateral displacement (m)

0m
Fig. 1. Comparison of measured and calculated loaddisplacement
curves.

X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
B

0m

20 m

20 m
Fig. 2. Schematic diagram of the nite dierence model.

540

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

Table 2
Pile and soil data for the deterministic analysis
Pile data

Value

Pile diameter, Dp (m)


Pile length, Lp (m)
Bending rigidity (EI) (kN m2)
Section modulus, Z (m3)
Yield strength of pile, Fy (kN/m2)
Yield moment of pile section, My (kN m)
Interface spring normal stiness, kn (kN/m/m)
Interface spring shear stiness, ks (kN/m/m)
Cohesive strength of shear coupling spring (kN/m)
Cohesive strength of normal coupling spring (kN/m)

1.0
10.0
1.1 106
0.0981
1.11 104
1088
1.15 105
1.15 105
62.83
180

Soil data
Undrained shear strength (kN/m2)
Shear modulus, G (kN/m2)
Bulk modulus, K (kN/m2)

20
1285.71
6000

2.4. Random eld model


The property variations of the in-situ soil represented by
the mean value, coecient of variation and correlation distance inuence the likely parameters for design. In the present study, soil undrained shear strength su is considered as
random variable and assumed to be a Log-Normally distributed value represented by parameters mean lsu , standard deviation rsu and spatial correlation distance dz. Use
of Log-Normal distribution is appropriate as the soil properties are non-negative and the distribution also has a simple relationship with normal distribution. A Log-Normally
distributed random eld is given by
su ~xi expflln su ~x rln su ~x  Gi ~xg

where ~x is the spatial position at which su is desired. G~x is


a normally distributed random eld with zero mean with
unit variance. The values of lln su and rln su are determined
using Log-Normal distribution transformations given by
!


r2su
2
rln su ln 1 2 ln 1 COV2su
3
lsu
1
4
lln su ln lsu  r2ln su
2
The correlation function is considered as exponentially
decaying correlation function as given by


2s
qsu s exp 
5
dz
where s j~x1  ~x2 j is the absolute distance between the two
points and dz is the spatial correlation distance. The correlation matrix is decomposed into the product of a lower triangular and its transpose by Cholesky decomposition,
L  LT qsu

i
X
j1

Lij Z j ;

2.5. Implementation of random eld


The correlation matrix is generated considering Eq. (5).
The value of the lag distance (s) is considered to be the center to center distance of the consecutive grids. Fig. 3
explains the evaluation of correlation matrix considering
the discretization of nite dierence grid. For an example,
if the center to center distance between grids 1 and 2 is dx,
the correlation between these two grids can be calculated
by putting the value s = dx in Eq. (5). Similarly correlation
between grid 1 with 3, 4, 5 can be established by placing
s = 2 dx, 3 dx, and 4 dx andqbetween
grid 1 with

p
2
2
2
31, 32, 33 are dy, dx dy and 2dx dy 2 , so on.
Therefore values in the 1st row of the correlation matrix
are the correlation coecients between grid 1 and other
grids, and leads to 900 values in a row (as number grids
is 30 30 in the present study). Hence considering all the
grids, the size of the correlation matrix is 900 900. Once
the correlation matrix is established, it is decomposed into
lower and upper triangular matrices using Cholesky decomposition technique. The correlated standard normal random eld is obtained by generating a sequence of
independent standard normal random variables (with zero
mean and unit standard deviation) and decomposed correlation matrix by Eq. (7). The correlation distance (dz) is
utilized to prepare correlation matrix, whereas COVsu is

dx
dy

30

31

32

33

34

35

36

60

61

62

63

64

65

66

90

91

92

93

94

95

96

120

121

122

123

124

125

126

150

151

152

153

154

155

156

180

Given the matrix L, correlated standard normal random


eld is obtained as follows (e.g. [15,16]):
Gi

where Zj is the sequence of independent standard normal


random variables. Typical values of COVsu for undrained
shear strength lie in the range of 1050% [17]. In the present study, results are presented assuming that the soil has
isotropic correlation structure; therefore the correlation
distance is the same in both horizontal and vertical
directions.

i 1; 2; . . . ; n

7
Fig. 3. Discretization of nite dierence grid.

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

bending moment are evaluated by means of Monte Carlo


simulations are shown in Fig. 4 as a function of sample
size. Observing the gure, 100 realizations are considered
because of the following reasons: (i) the uctuation falls
within a tolerable range (between 5% and 10%) after 100
samples, (ii) the calculation represents the worst case i.e.
at highest COVsu 50% and dz = 50 m, where possibility
of uctuation in expected values and standard deviations
are maximum, (iii) with the sample size of 100, estimated
expected value and standard deviation indicates satisfactory stability. Popescu et al. [18,19] also performed probabilistic analysis using 100 simulations.
Each realization produces a dierent lateral loaddisplacement curve. Values of Qia lst i.e. allowable lateral load
corresponding to lateral displacement of 0.0508 m from
ith loaddisplacement curve and, M imax i.e. maximum bending moment in pile corresponding to ith allowable lateral
load Qia lst are obtained (i = 1, 2, . . . , 100). Statistics of the
responses are obtained by ensemble averaging. Results
are examined in terms of dimensionless spatial correlation
distance given by dz/Lp, where Lp is the pile length. A owchart presented in Appendix I gives the scheme used for
statistical numerical analysis.

Table 3
Assumed ranges of probabilistic descriptors for soil undrained shear
strength
Probabilistic descriptor

Range

Coecient of variation, COVsu (%)


Correlation distance, dz (m)
Normalized vertical correlation distance, dz/Lp
Probability distribution function, pdf

10, 30, 50
1.5, 5.0, 15.0, 50.0
0.15, 0.5, 1.5, 5.0
Log-Normal

utilized to determine the standard deviation of the


undrained shear strength (Eq. (3)). Realizations of LogNormally distributed undrained shear strengths at each
grid location are obtained by transformation presented in
Eq. (2) for a specied mean, standard deviation of su.
The total computation process is conducted by developing
a subroutine in FISH code in FLAC. Monte Carlo simulation approach is used in the generation of sample functions of 2D Log-Normal random eld. In the present
study, values of COVsu in the range of 1050% and correlation distance 1.550 m are used (Table 3). Mean
undrained shear strength (lsu ) is taken as 20 kPa. For each
set of statistical properties given in Table 3, Monte Carlo
simulation is performed and totally 100 realizations of
shear strength random eld are generated for the analysis.
The code is run hundred times and in each run the FISH
program (subroutine) which assigns dierent realizations of
random eld in nite dierence grids is executed. The validity of number of realizations in Monte Carlo simulation is
examined. The statistical uctuation of the expected values
and standard deviations of allowable load and maximum

90

70
60
50

Qa

: Mean of lateral

50

100
150
Sample size

160
140

M max

20
Qa

: Standard deviation of

100
150
Sample size

200

100
150
Sample size

200

60
40
20

lateral allowable load.


50

50

80

(kNm)

30

: Mean of maximum
bending moment.

120
0

Mmax

(kN)
Qa

180

40

0
0

200

200

50

10

allowable load.

40
0

COVsu and dz have physical signicance as they reect the


nature (erratic or homogeneous) of a random eld. The
eect of COVsu and dz can be observed in Fig. 5ac, which
shows the gray scale representation of possible realizations

Mmax

Qa

(kN)

80

3. Stochastic verses deterministic analysis results

(kNm)

541

0
0

M max

: Standard deviation of
maximum bending
moment.
50

100
150
Sample size

200

Fig. 4. (a) Mean of lateral allowable load as a function of Monte Carlo (MC) sample size. (b) Standard deviation of lateral allowable load as a function of
MC sample size. (c) Mean of maximum bending moment developed in pile with respect to MC sample size. (d) Standard deviation of maximum bending
moment developed in pile with respect to MC sample size.

542

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

Stronger
zone

Weaker zone

su (kPa)

su (kPa)

Stronger
zone

su (kPa)

Weaker zone

Stronger
zone
Weaker zone

Fig. 5. One realization of random eld, (a) COVsu 10%, dz/Lp = 0.15; (b) COVsu 30%, dz/Lp = 0.15; (c) COVsu 30%, dz/Lp = 1.5.

of the undrained shear strength. Darker shades denote


weaker zones having lower undrained shear strength. Realizations of random eld generated for COVsu 10%, 30%
and dz = 1.5 m are presented in Fig. 5a and b. The results
indicate that increase in COVsu contributes to the erratic
nature of soil and hence more number of weak zones are
present. In the deterministic analysis, a uniform soil is generally considered and a value of su = 20 kPa (present study)
is realized in the soil medium, whereas in the real eld due to
spatial variability eect, shear strength varies within a range.
A higher range of su is generally observed for a eld of high
COVsu (say 30%) compared to a eld of low COVsu values
(say 10%) and this aspect is clearly evident from Fig. 5a
and b. A range of su = 1626 kPa is observed for a realization of random eld having COVsu 10%, where range of
su = 1040 kPa is observed for COVsu 30%. Realizations
of random eld generated for dz = 1.5 m, 15.0 m and
COVsu 30% presented in Fig. 5b and c show the eect of
correlation distance. At low values of dz (say 1.5 m) domain
is similar to an erratic eld and as scale of uctuation
increases (say 15.0 m), it can be noted that the cohesion eld
becomes more homogeneous.
For various combinations of COVsu and dz/Lp as given in
Table 3, the Monte Carlo simulation are conducted and the

pile lateral loaddisplacement curves are obtained. Typical


realizations of loaddisplacement curves for a combination
of COVsu 30% and dz = 1.5 m are presented in Fig. 6.
Ensemble average of loaddisplacement curves is taken as
the mean loaddisplacement curves for spatially varied soil.
The loaddisplacement curve based on uniform shear
strength all over the soil zone (the deterministic loaddisplacement curve) is also shown in the same gure. It is
observed that the resulting mean allowable load corresponding to allowable lateral deection of 0.0508 m of spatially varying soil, lQa lat is signicantly lower than the
corresponding value for the homogeneous soil (Qdet
a lat ). The
goodness of t test is conducted using well known KolmogorovSmirnov (KS) test for COVsu 30% and
dz = 1.5 m and the results are presented in Fig. 7. The KS
test compares the observed cumulative frequency and the
cumulative distribution function (CDF) of allowable load
with an assumed theoretical distribution. It is apparent that
the Log-Normal distribution represents reasonably well.
Fig. 8a and b illustrates typical lateral displacement contours for a certain level of lateral loading (not up to ultimate
state) in homogeneous medium and spatially varied soil corresponding to COVsu 30%, dz = 1.5 m (dz/Lp = 0.15). It is
observed that lateral displacement has a regular pattern for

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547


120

543

Deterministic curve (homogenous soil)


Mean curve (spatially varying soil)

100

Original position of pile

100 realizations of MC analysis

Deflected position of
Deflected
position of
pile

Allowable displacement = 0.0508m

Lateral load (kN)

Q a latdet =83 kN

80
Q a lat = 65 kN

60

40

20
COVsu = 30%,

z/Lp

= 0.15

0
0

0.01

0.02

0.03

0.04

0.05

X-displacement
(m)
X-displacement
contours
-1.10E-02
-8.00E-03
-5.00E-03
-2.00E-03
1.00E-03
4.00E-03
7.00E-03

Contour interval= 1.00E-03

0.06

Pile lateral head displacement (m)

Fig. 6. Loaddisplacement curves for homogeneous soil and spatially


varied soil for COVsu 30% and dz/Lp = 0.15.

X-displacement
(m)
X-displacement contours

-1.90E-02
-1.50E-02
-1.10E-02
-7.00E-03
-3.00E-03
1.00E-03
5.00E-03
9.00E-03

Cumulative probability

Actual distribution
Log-Normal fit
0.8

0.6

Fig. 8. (a) Lateral displacement contours for homogeneous soil and


(b) lateral displacement contours for spatially varied soil (COVsu 30%,
dz/Lp = 0.15).

0.4

0.2
Mean load=65 kN
Deterministic load = 83 kN
0

Contour interval= 1.00E-03

40

60
80
Allowable load (kN)

100

Fig. 7. KS test on allowable load for Log-Normal distribution for


COVsu 30%; dz/Lp = 0.15.

the homogeneous soil whereas, irregular pattern is observed


for spatially varied soil due to presence of weaker zones.
4. Inuence of coecient of variation and correlation
distance
4.1. Inuence on pile responses
The results presented in Figs. 9 and 10 indicate the two
important probabilistic characteristics of the soil variabil-

ity, coecient of variation and correlation distance of


undrained shear strength. They have signicant eect on
the lateral allowable load and maximum bending moment.
Both the characteristics control the amount of weak zones
in the soil mass. Fig. 9 shows the inuence of dz/Lp and
COVsu on the estimated mean allowable load, lQa lat. The
deterministic solution, for homogeneous soil (su = 20 kPa
in all zones), the allowable capacity is also presented in
the same gure which is termed as deterministic solution.
The load corresponding to allowable lateral displacement
of 0.0508 m is obtained as 83 kN which is termed as
deterministic allowable capacity Qdet
a lat . Similarly the maximum bending moment is also observed at the same load
level as 180 kN m M det
max which is presented in Fig. 10.
Fig. 9 shows that there is signicant reduction of mean
allowable load for spatially varied soil compared to deterministic allowable load. At low values of COVsu , mean
allowable load is greater than the value obtained for higher
value of COVsu . For an example, a mean value of lQa lat =
65.1 kN is observed for COVsu 30% and dz = 1.5 m

544

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

Deterministic solution

80

worst case

Mean
ultimatelateral
lateralload
load, Qa lat (kN)
Mean
allowable

85

75
70
65

COVsu =10%

60

COVsu =30%
COVsu =50%

55
50
0

design of laterally loaded pile based on spatially varied soil.


At dz = 5.0 m, there is maximum reduction of allowable
load, hence it can be represented as worst case. This behaviour is also noted by Fenton and Griths [21] for shallow
foundation and Niandou and Breysse [22] for piled raft. It
is noted that generally the worst correlation distance is
problem specic and the value varies within the size of
structure. When dz is of intermediate value (i.e. within
structure size), the structure is sensitive to uctuation in
the soil properties [22]. For the present case worst correlation distance is the half of the length of pile dworst
0:5Lp .
z
The mean allowable load increases with the increase in dz
value. Similar observation is made for maximum bending
moment. Fig. 10 shows similar trends for the maximum
bending moment corresponds to allowable load for homogeneous and spatially varied soil. It shows mean maximum
bending moment increases at low values of COVsu .

z /LP

This section describes the eect of spatial variability on


the maximum shear strain level induced in the soil near pile
and the stiness. The formation of failure surface can be
described in terms of accumulation of shear strains in the
soil near pile. It is due to the applied lateral load in pile head
and it can be noted that it relates to the number of weaker
zones present in the soil. However in the present analysis pile
is not loaded up to ultimate state, but the development of
maximum shear strains controls the likely failure mechanism. To determine the eect of spatial variability on the
maximum shear strain in soil, an analysis is conducted considering dierent values of COVsu and dz and for uniform
soil. Fig. 11a and b shows the maximum shear strain contours using deterministic and probabilistic analysis.

240
220
200

COVsu =10%

Deterministic solution
180

4.2. Inuence on formation of failure

COVsu =30%
COVsu =50%

160
140
120
100

a
0

(dz/Lp = 0.15), whereas for the same correlation distance,


the lQa lat value becomes 60.2 kN, if the COVsu 50%.
The inuence of dz (or, dz/Lp) is also pronounced with
the value of lQa lat. It is observed that the value of lQa lat
increases gradually as the value of dz (or, dz/Lp) reduces
to zero. The reason behind this phenomenon is that as dz
becomes vanishingly small, weakest path becomes more
erratic which means that the formation of failure path is
correspondingly longer and as a result, the weakest path
starts to nd shorter routes cutting through higher strength
material as described by Griths and Fenton [1], Griths
et al. [20]. The value of lQa lat marginally decreases up to
dz = 5.0 m (dz/Lp = 0.5) and then gradually increases. It
represents an important phenomenon with respect to the

-3

Max. strain 7 10

Fig. 10. Mean maximum bending moment with variation in COVsu and
dz/Lp.

-2

z /Lp

Max. shear strain increment


Contour interval= 1.00E-03

Max. strain 1.4 10

Mean maximum moment, Mmax (kN-m)

Fig. 9. Mean allowable lateral load with variation in COVsu and dz/Lp.

Max. shear strain increment


Contour interval= 2.00E-03

Fig. 11. (a) Incremental strain contours for homogeneous soil; (b)
incremental strain contours for COVsu 30%, dz/Lp = 0.15.

(1) An erratic formation surface is observed, passing


mainly through weaker soil zones. As correlation distance increases, the soil becomes more homogeneous
and this aspect is evident from Fig. 11a and b.
(2) It is also important to observe that maximum strain
level where pile undergoes maximum lateral movement, also changes due to the eect of soil variability.
The maximum shear strain corresponding to ultimate
load in the homogeneous soil is about 0.007
(Fig. 11a) whereas in the spatially varied soil, it is
observed that the maximum shear strain value is in
the range of 0.014 for COVsu 30% (Fig. 11b).
(3) The eect of correlation distance can be discerned
from Fig. 12a and b, which shows sample realizations
of spatially varied soil and maximum shear strain contours for COVsu 50%, dz = 1.5 m (dz/Lp = 0.15) and
COVsu 50%, dz = 50 m (dz/Lp = 5.0) respectively. It
can be noted that the eect of correlation distance is
signicant. Higher value of maximum strain level
(0.02) is observed for low correlation distance (say
dz = 1.5 m) and a lower value (0.008) is observed in
higher correlation distance (say dz = 50 m).
The observations from above results are also reected in
soil stiness values and the corresponding mean load
deection curves. The stiness values can be obtained by
analyzing loaddisplacement curves. The stiness values
(K) are computed as incremental applied load divided by

Max. shear strain increment


Contour interval= 2.00E-03

Max. strain
10-3
Max.
strain1.0
8 10

Max. strain 2.0 10

-2

-2

Max. shear strain increment


Contour interval= 1.00E-03

Fig. 12. (a) Incremental strain contours for COVsu 50%, dz/Lp = 0.15;
(b) incremental strain contours for COVsu 50%, dz/Lp = 5.0.

40
30

545

COVsu = 10%
COVsu = 30%

20

COVsu = 50%

10

z/Lp=0.15

K : Soil Stiffness

0
0

b
Soil stiffness, K (kN/m)

Fig. 12a and b represents the maximum shear strain contours for COVsu 50%, dz = 1.5 m (dz/Lp = 0.15) and
COVsu 50%, dz = 50 m (dz/Lp = 5.0) respectively. The
practical signicance of the results can be noted as follows:

Lateral load (kN)

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

0.01
Pile lateral head displacement (m)

0.02

7000
6000
5000
4000
COVsu = 10%

3000

COVsu = 30%
COVsu = 50%

2000
0

z /Lp
Fig. 13. (a) Soil stiness values considering variability of soil parameter
form mean loaddisplacement curve; (b) eect of soil spatial variability of
soil on soil stiness.

incremental lateral displacement. The mean stiness values


are obtained from Monte Carlo simulated loaddisplacement curves. Fig. 13a shows that the mean loaddisplacement curves become at as COVsu increases, which
indicate that soil becomes less sti as variability increases.
Fig. 13b shows a relationship of stiness values with correlation distance and COVsu . Further increase in soil stiness
is observed as correlation distance increases. This is due to
the fact that strain level decreases as correlation distance
increases. Worst correlation distance is also pronounced
at dz = 5 m as from Fig. 13b. At this level soil encounters
least stiness values.
5. Probabilistic interpretation of failure
From the design considerations, a laterally loaded pile is
unserviceable if applied load (Q) is more or, equal to the
allowable load (Qa lat) and failure of it can be interpreted
as maximum bending moment (Mmax) is higher or equal
to the yield moment carrying capacity (My) of the pile section. That means the laterally loaded pile can be considered
to be in a serviceable state of failure if,
Qa lat 6 Q

and

M y 6 M max

where My is the yield moment of the pile section. The


allowable capacity Qa lat and Mmax are considered as ran-

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

where U() is the cumulative normal function. To show the


inuence of COVsu and dz/Lp on P(Qa lat 6 Q), a deterministic load, Q = 60 kN is considered and P(Qa lat 6 Q) is calculated for dierent values of COVsu and dz/Lp. Fig. 14
shows that the probability i.e. P(Qa lat 6 Q) increases as
COVsu increases. For an example, when dz/Lp = 0.15 and
COVsu 30%, P(Qa lat 6 Q) is 0.37, indicating a 37% probability that the pile may subject to unserviceable condition
due to applied lateral load of 60 kN. The value of
P(Qa lat 6 Q) becomes 0.3, when dz/Lp = 5.0 and
COVsu 30% which represents probability of unserviceable condition decreases as correlation distance increases.
This behaviour is also indicated by Phoon et al. [23]. The
highest probability of unserviceable condition is observed
at dz/Lp = 0.5 and further increase in dz/Lp decrease the
probability and this aspect is evident in the same gure.
The probability of failure is also computed for the case
of maximum bending moment Mmax greater than the yield
moment of the pile section (My) for an applied lateral load
Q and is given by


ln M y  lln M max
P M y 6 M max U
10
rln M max
Fig. 15 presents comparison between failure probabilities
due to maximum moment exceeding moment carrying
capacity and COVsu for dierent values of dz/Lp. Similar
to Fig. 14, failure probability increases as COVsu increases.
The value of P(My 6 Mmax) for an applied load Q of 60 kN
is 7.5 109 if dz/Lp = 0.15 and COVsu 30%. The failure
probability is 4.8 1010 for dz/Lp = 5.0 and COVsu
0.5

Q)

0.3

P(Qa lat

0.4

0.2

z /LP=0.15

z /LP=0.50

. z /LP=1.50

0.1

. z /LP=5.00

0
10%

20%

30%

40%

50%

COVsu
Fig. 14. Relationship between P(Qa lat 6 Q) and COVsu for dierent dz/Lp.

0.00012
. z /LP=0.15

0.00010

. z /LP=0.50

Mmax)

dom variables. Considering Qa lat as Log-Normal variable,


the probability that the computed allowable capacity is less
than the deterministic applied load (Q) can be stated as


ln Q  lln Qa lat
P Qa lat 6 Q U
9
rln Qa lat

P(My

546

. z /LP=1.50

0.00008

. z /LP=5.00

0.00006
0.00004
0.00002
0.00000
10%

20%

30%
COVsu

40%

50%

Fig. 15. Relationship between P(My 6 Mmax) and COVsu for dierent
dz/Lp.

30%. The results indicate that probability due to


P(Qa lat 6 Q) is critical for design as the other failure mode
gives low probability of failure. An important observation
from Figs. 14 and 15 is that the correlation distance is relevant to the probabilistic interpretation at higher values of
COVsu . High value of correlation distance is benecial as it
gives lower probability of failure.
6. Concluding remarks
The following conclusions from the present study can be
made:
(1) The major contribution of the present study is the relevance of spatial variability of soil undrained shear
strength in laterally loaded pile design. It is observed
that there is a signicant change in allowable load,
maximum bending moment due to the eect of
COVsu and dz. Marginal increase in allowable load
is observed at low correlation distance. At high value
of correlation distance, the soil becomes almost
homogeneous and allowable load increases.
(2) The propagation of failure in soil near pile is
described in terms of accumulated shear strain in soil.
It is observed that the correlation distance and coefcient of variation signicantly inuence the development of maximum shear strain values in soil near pile.
At higher values of COVsu and lower value of correlation distance, the strain level is likely to be high and
hence soil will have lesser allowable load. With further increase in the correlation distance, strain level
decreases and hence allowable load increases.
(3) Idealisation of the number of weak zones in each realization of random eld is useful to understand the
allowable load. Monte Carlo simulation technique
combined with numerical analysis is a very useful
approach in this regard as demonstrated in this study.

S. Haldar, G.L. Sivakumar Babu / Computers and Geotechnics 35 (2008) 537547

547

Acknowledgements

References

The authors thank the reviewers for their critical comments which have been very useful in improving the work
presented in this paper.

[1] Griths DV, Fenton GA. Bearing capacity of spatially random soil:
the undrained clay Prandtl problem revisited. Geotechnique
2001;51(4):3519.
[2] Fenton GA, Griths DV. Three-dimensional probabilistic foundation settlement. J Geotech Geoenviron EngASCE 2005;131(2):
2329.
[3] Fenton GA, Griths DV. Reliability based deep foundation design.
In: Proceedings of GeoDenver2007: new peaks in geotechnics
ASCE, Reston. Paper no. GSP 170; 2007.
[4] Haldar S, Babu GLS. Ultimate capacity of pile foundation on
spatially random cohesive soil. In: Proceedings of 10th international
conference on applications of statistics and probability in civil
engineering, Tokyo, Japan; 2007.
[5] Broms BB. Lateral resistance of piles in cohesionless soils. J Soil
Mech Found DivASCE 1964;90(3):12356.
[6] Randolph MF, Houlsby GT. The limiting pressure on a circular pile
loaded laterally in cohesive soil. Geotechnique 1984;34(4):61323.
[7] Phoon KK, Kulhawy FH. Characterisation of model uncertainties for
laterally loaded rigid drilled shafts. Geotechnique 2005;55(1):4554.
[8] Fan CC, Long JH. Assessment of existing methods for predicting soil
response of laterally loaded piles in sand. Comput Geotech
2005;32:27489.
[9] Poulos HG, Davis EH. Pile foundation analysis and design. Wiley;
1980.
[10] Hsiung Y, Chen Y. Simplied method for analyzing laterally loaded
single piles in clays. J Geotech Geoenviron EngASCE 1997;123(11):
101829.
[11] Zhang LM, Ng AMY. Probabilistic limiting tolerable displacements
for serviceability limit state design of foundations. Geotechnique
2005;55(2):15161.
[12] Itasca Consulting Group Inc. FLAC, Fast Lagrangian analysis of
continua. Users manual, version 5.0. Minneapolis, USA; 2006.
[13] Donovan K, Pariseau WG, Ceepak M. Finite element approach to
cable bolting in steeply dipping VCR slopes. In: Geomechanics
application in underground hard rock mining. New York: Society of
Mining Engineers; 1984. p. 6590.
[14] Eloseily KH, Ayyub BM, Patev R. Reliability assessment of pile
groups in sands. J Struct EngASCE 2002;128(10):134653.
[15] Shinozuka M, Yamazaki F. Stochastic nite element analysis: an
introduction. Stochastic structural dynamics, Progress in theory and
applications. London and New York: Elsevier Applied Science; 1988.
[16] Fenton GA. Probabilistic methods in geotechnical engineering. In:
Workshop presented at ASCE GeoLogan97 conference, Logan,
Utah; 1997.
[17] Phoon KK, Kulhawy FH. Evaluation of geotechnical property
variability. Can Geotech J 1999;36:62539.
[18] Popescu R, Deodatis G, Nobahar A. Eects of random heterogeneity
of soil properties on bearing capacity. Probabilist Eng Mech
2005;20:32441.
[19] Popescu R, Prevost JH, Deodetis G. Eects of spatial variability on
soil liquefaction: some design recommendations. Geotechnique
1997;47(5):101936.
[20] Griths DV, Fenton GA, Manoharan N. Bearing capacity of rough
rigid strip footing on cohesive soil: probabilistic study. J Geotech
Geoenviron EngASCE 2002;128(9):74355.
[21] Fenton GA, Griths DV. Bearing capacity prediction of spatially
random c/ soils. Can Geotech J 2003;40(1):5465.
[22] Niandou H, Breysse D. Reliability analysis of a piled raft accounting
for soil horizontal variability. Comput Geotech 2007;34(2):7180.
[23] Phoon KK, Quek ST, Chow YK, Lee SL. Reliability analysis of pile
settlement. J Geotech Geoenviron EngASCE 1990;116(11):
171735.

Appendix I. Flowchart for statistical numerical analysis by


nite dierence technique

Input Data File

Generate correlation matrix for a correlation


distance

Decompose correlation matrix by Cholskey


decomposition

Generate normally distributed set of random


numbers

Generation of lognormal random field i.e.


material properties by transformation

(1) Generate grids and dimension of the field


(2) Assign material properties in each grid
from randomly generated values.
(3) Generate Pile element
(4) Assign boundary conditions
(5) Establish initial equilibrium

(1) Run analysis


(2) Save output file

Obtain lateral load-displacement curve for pile


top and calculate allowable load and maximum
bending moment corresponding to lateral
deflection of 0.0508m

Last
Simulation
Yes
End of each input file

No

You might also like