You are on page 1of 13

Abstract Load displacement analysis of drilled

shafts can be accomplished by utilizing the t-z


method, which models soil resistance along the
length and tip of the drilled shaft as a series of
springs. For non-linear soil springs, the govern-
ing differential equation that describes the soil-
structure interaction may be discretized into a
set of algebraic equations based upon nite
difference methods. This system of algebraic
equations may be solved to determine the load
displacement behavior of the drilled shaft when
subjected to compression or pullout. By com-
bining the nite difference method with Monte
Carlo simulation techniques, a probabilistic
loaddisplacement analysis can be conducted.
The probabilistic analysis is advantageous com-
pared to standard factor of safety design because
uncertainties with the shaftsoil interface and tip
properties can be independently quantied. This
paper presents a reliability analysis of drilled
shaft behavior by combining the nite differ-
ence technique for analyzing non-linear load
displacement behavior with Monte Carlo
simulation method. As a result we develop
probabilistic relationships for drilled shaft design
for both total stress (undrained) and effective
stress (drained) parameters. The results are
presented in the form of factor of safety or
resistance factors suitable for serviceability
design of drilled shafts.
Keywords Drilled shaft Probabilistic analysis
Finite difference t-z methods Monte Carlo
simulation
Notation
A
m
cross-sectional area of drilled shaft, m
2
a curve tting parameter
b curve tting parameter
c constant
D drilled shaft diameter, mm
E
m
drilled shaft elastic modulus, MN/m
2
E
s
tip soil elastic modulus, kN/m
2
E(Q
D
) expected value of dead load
E(Q
L
) expected value of live load
K shear modulus of shaftsoil interface
sub-grade reaction, kN/m
2
K
ep
coefcient of lateral earth pressure
A. Misra (&)
Professor of Civil Engineering, University of
Missouri-Kansas City, 5100 Rockhill Road, 350H
Flarsheim Hall, Kansas City, MO 64110, USA
e-mail: misraa@umkc.edu
L. A. Roberts
Geotechnical Engineer, Terracon Consultants, Inc.,
13910 W. 96th Terrace, Lenexa, KS 66215, USA
e-mail: laroberts@terracon.com
S. M. Levorson
Senior Geotechnical Engineer, Terracon Consultants,
Inc., 13910 W. 96th Terrace, Lenexa, KS 66215, USA
e-mail: smlevorson@terracon.com
Geotech Geol Eng (2007) 25:6577
DOI 10.1007/s10706-006-0007-2
1 3
ORIGINAL PAPER
Reliability analysis of drilled shaft behavior using nite
difference method and Monte Carlo simulation
Anil Misra Lance A. Roberts
Steven M. Levorson
Received: 17 February 2006 / Accepted: 11 April 2006 / Published online: 5 October 2006
Springer Science+Business Media B.V. 2006
K
init
initial tangent shear modulus of
shaftsoil interface sub-grade reaction,
kN/m
2
K
m
drilled shaft axial stiffness, MN
K
mod
modulus number
K
o
in-situ horizontal stress coefcient
K
t
drilled shaft tip soil stiffness, kN/m
K
ti
initial tangent tip soil stiffness, kN/m
L
b
shaft interaction zone length, m
L
d
shaft non-interaction zone length, m
p
f
probability of drilled shaft failure
P
t
drilled shaft tip force, kN
P
utip
ultimate capacity of the tip soil, kN
Q deterministic drilled shaft load, kN
q shear force per unit length, kN/m
q
f
failure strength of the shaftsoil inter-
face, kN/m
q
o
ultimate (asymptotic) strength of shaft
soil interface, kN/m
q
t
unit toe bearing resistance, kN/m
2
R drilled shaft load capacity, kN
R
f
failure ratio
u displacement, mm
u
t
tip displacement, mm
z depth, m
b reliability index
b
ep
side resistance parameter
b
T
target reliability index
c
D
dead load factor
c
L
live load factor
Dz incremental length along the drilled
shaft, mm
d drained friction angle for the shaftsoil
interface, deg
k
QD
bias of the dead load
k
QL
bias of the live load
k
R
bias of the resistance
l
s
tip soil Poissons ratio
r
atm
atmospheric pressure, MPa
r
z

vertical effective stress, kPa


s
u
ultimate shear strength of shaftsoil
interface, kN/m
2
/ resistance factor
/ drained friction angle of soil, deg
W
QD
coefcient of variation of dead load
W
QL
coefcient of variation of the live load;
and
W
R
coefcient of variation of the resistance
Introduction
Drilled shafts are being widely used as foundation
elements in bridges and buildings. Compared to
other conventional foundation systems, drilled
shafts can be cheaper, easier to construct, and less
intrusive to the environment. Most designs for
drilled shafts are conducted using the so-called
allowable stress method and only consider the
ultimate limit state, assuming full resistance along
the length of the shaft and the tip. The service
limit state, or load capacity of the drilled shaft at
an allowable head displacement, is often not
considered in the design. However, the displace-
ment of a drilled shaft can be an important design
consideration, especially when foundation settle-
ments are critical to the operation of a structure.
Therefore, the ability to quantify the drilled shaft
load capacity at the service limit state becomes
extremely important and should be considered as
part of the drilled shaft design.
Empirical methods exist for determining the
load capacity of a drilled shaft based upon the
ratio of mobilized resistance to ultimate resistance
(both side friction resistance and tip resistance)
versus the ratio of allowable displacement to shaft
diameter. Normalized curves, based upon re-
search conducted by ONeill and Reese (FHWA
1999), can be found in most foundation engi-
neering textbooks. Because this method was
developed using specic diameter drilled shafts
that were installed in particular strata, extrapola-
tion to other design conditions is cautioned
(FHWA 1999). It is, therefore, benecial to utilize
a procedure that can explicitly consider the soil-
drilled shaft interaction in the analysis. This paper
utilizes the t-z method, which models the soil
resistance along the length and at the tip of the
drilled shaft as a series of springs. The shaftsoil
interface is assumed to be either homogeneous or
linearly increasing with depth, depending upon
the soil prole. The shaftsoil interface properties
are given in terms of the shear modulus of shaft
soil interface sub-grade reaction, K, the ultimate
shaftsoil interface shear strength, s
u
, and the
modulus of tip soil sub-grade reaction, K
t
. For this
paper, the loaddisplacement behavior of the
shaftsoil interface and the tip soil was assumed to
66 Geotech Geol Eng (2007) 25:6577
1 3
be hyperbolic (non-linear soil springs) for both
undrained and drained soil proles. Due to this
assumption, the governing differential equations
describing the shaftsoil interaction cannot be
solved in closed form, but instead must be solved
numerically using, say, nite difference method.
By combining the nite difference scheme with
Monte Carlo simulation techniques, a probabilis-
tic loaddisplacement analysis can be performed.
The advantage of the probabilistic loaddis-
placement analysis is that it rationally incorpo-
rates a reliability based design approach.
Uncertainties with shaft-soil interface properties,
due to soil inherent variability and construction
techniques, are thereby directly quantied. In an
allowable stress design approach, uncertainty
with the design is indirectly accounted for by
applying a factor of safety to the calculated ulti-
mate capacity. In reliability based design, how-
ever, the shaftsoil interface properties are
considered as random variables and are described
statistically with non-Gaussian probability distri-
bution functions. Consequently, a probabilistic
loaddisplacement relationship can be obtained,
resulting in an improved technique for design of
drilled shafts at the service limit state.
This paper extends the previous work by Misra
and Roberts (2006) to develop an approach for
reliability analysis of drilled shafts by combining
the nite difference technique for analyzing non-
linear loaddisplacement behavior with Monte
Carlo simulation method. As a result, probabilistic
relationships are obtained for drilled shaft design
assuming both total (undrained) and effective
(drained) stress parameters. These relationships
are analyzed to develop methods for determining
the probability of drilled shaft failure at the ser-
vice limit state. The results are presented in the
form of load capacity and factor of safety cumu-
lative histograms, as well as resistance factors, that
can be utilized for service limit state design.
Drilled shaftsoil interaction model for total
and effective stress analysis
In the t-z method for loaddisplacement anal-
ysis, the shaft-soil interaction along the length of
the soilconcrete interface of the drilled shaft is
represented by the spring-slider system in Fig. 1.
Similar assumptions are commonly made for
analytical and numerical models for loaddis-
placement behavior of piles and drilled shafts (cf.
Scott 1981; Kraft et al. 1981; Reese and ONeill
1987; FHWA 1999; Misra and Chen 2004).
For this paper, the drilled shaftsoil interface
was assumed to behave as a non-linear material.
Figure 2 shows the forcedisplacement behavior
for the non-linear spring-slider system, as de-
picted by the shear force per unit length q, versus
displacement u, curve. In Fig. 2, K
init
is the initial
tangent shear modulus of shaftsoil interface sub-
grade reaction, and q
o
is the ultimate strength of
the shaftsoil interface, given by the product of
the shaft perimeter and the ultimate shear
strength of the shaftsoil interface, denoted by s
u
.
In general, the shaftsoil interface properties are
related to the construction techniques and the
properties of the soil strata. The shaftsoil inter-
face properties will typically follow the strati-
cation of the natural soil; therefore, interface
parameters must be calculated for each soil layer.
However, for simplicity in this presentation, we
have assumed a single soil stratum along the en-
tire drilled shaft length.
Shaftsoil interface properties
For a total stress analysis with undrained param-
eters, the ultimate shear strength of the shaftsoil
interface, s
u
, and the initial tangent shear modu-
lus of shaftsoil interface sub-grade reaction,
Spring
Slider
Spring
Slider
Fig. 1 Spring and slider model for drilled shaftsoil
interface
Geotech Geol Eng (2007) 25:6577 67
1 3
K
init
, were assumed to be constant with depth.
This soil model is often used to describe the shear
strength behavior of ne-grained soils, such as
clays and silts, which develop excess pore pres-
sures during loading. For an effective stress
analysis, it is assumed that both of the shaftsoil
interface parameters, s
u
and K
init
, increase with
an increase in conning stress (i.e. increase with
depth). The ultimate shear strength of the shaft
soil interface, s
u
, for effective stress conditions,
can be described with the following equation
given by Kulhawy (1991):
s
u
r
0
z
K
ep
tan d 1
where, K
ep
is the coefcient of lateral earth
pressure, r
z

is the vertical effective stress, and d is


the drained friction angle for the shaftsoil
interface.
We recognize that K
ep
is not only a function of
the original in-situ horizontal stress coefcient,
K
o
, it is also dependant upon the stress changes
caused by the construction of the drilled shaft,
loading, and desiccation. Recently, Rollins et al.
(2005) back-calculated the value of K
ep
from
drilled shaft pullout tests conducted in gravel,
gravelly sands, and sands in Utah and other sites.
Utilizing this data, the back-calculated values of
K
ep
were plotted versus depth. An exponential
curve was t to the data, thereby providing a
method for determining K
ep
at a given depth. It
was observed that the value of K
ep
reduces with
an increase in depth and will eventually asymp-
tote to the normally consolidated at-rest earth
pressure coefcient, K
oNC
, at a given depth (see
Rollins et al. 2005). This depth is often referred to
as the critical depth and is dependant upon the
diameter of the drilled shaft (Terzaghi et al.
1996). This depth is typically given as 15 to 20
times the drilled shaft diameter.
Alternatively, ultimate shear strength of the
shaftsoil interface, s
u
, can be determined from
the widely used b-method (see Chandler 1968;
Burland 1973; and Meyerhof 1976 among others).
In the b-method, the product of the parameters
K
ep
and tan(d) in Eq. (1) are replaced with the
parameter, b, and Eq. (1) is simplied to:
s
u
b
ep
r
0
z
2
Now utilizing the back-calculated values of
K
ep
and an average value for tan(d) by assuming
d/ /=1 based upon data from Rollins et al.
(2005), back-calculated values for b
ep
were
plotted with respect to depth for gravel, gravelly
sand, and sand in Fig. 3. The following exponen-
tial function was t to the depth prole for the
mean value of b
ep
:
b
ep
2:2e
0:14z
0:30 3
where, z is the depth, in meters.
For the shaftsoil interface, the non-linear
force-displacement behavior has been success-
fully described using a hyperbolic model (see
Kondner 1963; Duncan and Chang 1970):
Fig. 2 Shaftsoil interface forcedisplacement relation-
ship for a non-linear spring-slider system
0
2
4
6
0 5 10 15 20 25
Depth (m)

e
p
Fig. 3 b
ep
versus depth for a drained soil prole. Solid line
is mean t of back-calculated data. Dashed lines are upper
and lower bound ts to the data
68 Geotech Geol Eng (2007) 25:6577
1 3
q
u
1
K
init

uR
f
q
o
h i 4
where, the factor R
f
, described as the failure ratio,
relates the ultimate (asymptotic) strength, q
o
, of
the shaft-soil interface to the failure strength, q
f
,
as follows (Duncan and Chang 1970):
R
f

q
f
q
o
5
According to Duncan and Chang (1970), very
often the asymptotic value of strength is larger
than the failure strength of the soil by a small
amount. This is expected due to the fact that the
hyperbola will remain below the asymptote at all
nite values of strain. The failure ratio, R
f
, has
been found to have a value between 0.75 and 1.00
(Duncan and Chang 1970).
Experimental studies conducted by Janbu
(1963) show that the initial tangent shear modulus
of shaftsoil interface sub-grade reaction, K
init
, is
also proportional to the conning stress in an
effective stress analysis and may be determined
by the following:
K
init
K
mod
r
0
z
K
o
r
atm

x
6
where, x is a exponent that determines the rate of
variation of K
init
with respect to r
z

K
o
and K
mod
is
a modulus number. Following the typical practice,
the horizontal earth pressure is normalized with
respect to atmospheric pressure, r
atm
, expressed
in the same pressure units as K
init
(see, for
example, Duncan and Chang 1970; Duncan et al.
1980).
Governing equation
For drilled shafts, the load transfer occurs via the
shaftsoil interface within the interaction zone
length, L
b
. The upper 0.31.5 m of the shaftsoil
interface, called the non-interacting zone length,
L
d
, is considered to have negligible shear resis-
tance due to ground disturbance, ll placement
and construction technique utilized. Based upon
the t-z methodology, we have previously
presented the mathematical model that describes
the loaddisplacement behavior of drilled shaft
foundations accounting for shaftsoil interaction
(Misra and Roberts 2006; Misra and Chen 2004).
Here we state the governing equation, given in
Eq. (7), and the relevant boundary conditions for
completeness of our discussion.
K
m
d
2
u
dz
2
Kuz 0 7
In Eq. (7), u(z) is the shaft deformation at that
location, K
m
is the shaft axial stiffness, and K is
the tangential shear modulus of shaftsoil inter-
face sub-grade reaction. At the shaft head (z=L),
the boundary condition is the applied load P,
while at the shaft tip (z=0), the boundary condi-
tion is the developed tip force, P
t
. The tip force,
P
t
, develops proportional to the tip displacement,
u
t
, given by:
P
t
K
t
u
t
8
where, K
t
is the tip soil stiffness. The magnitude
of the tip displacement is dependent on the pro-
portion of the applied load that reaches the tip.
Based upon theories for rigid punch bearing upon
an elastic half-space, the initial tangent tip soil
stiffness, K
ti
, may be related to shaft diameter and
the elastic properties of tip soil as follows (John-
son 1985):
K
ti

0:3pDE
s
1 l
2
s

9
where, E
s
is the tip soil elastic modulus, l
s
is tip
soil Poissons ratio, and D is the shaft diameter.
For the tip soil, Eq. (4) can be written in a slightly
different format:
P
t

u
t
1
K
ti

u
t
R
f
P
utip
h i 10
where the failure ratio relates the ultimate
(asymptotic) capacity, P
utip
, of the tip soil to the
failure strength, P
ftip
, in a manner similar to Eq.
(5).
As the compressive load on the drilled shaft
increases, the shaftsoil interface will yield
Geotech Geol Eng (2007) 25:6577 69
1 3
zcompletely and the soil at the tip of the drilled
shaft will carry the entire additional load. The
ultimate capacity of the tip soil can be determined
from standard formulas, such as the following
(see Coduto 2001; Das 2004):
P
utip
q
t
A
m
11
where, q
t
is the unit tip bearing resistance and A
m
is the cross sectional area of the drilled shaft.
When the load at the tip of the drilled shaft, P
t
,
reaches the tip bearing capacity given by Eq. (11),
the drilled shaft will fail by plunging.
Finite difference method for loaddisplacement
calculations
For non-linear loaddisplacement behavior of the
shaft-soil interface and tip soil, the governing Eq.
(7) cannot be solved in closed-form. Therefore, a
nite difference technique is utilized. In the nite
difference technique, the drilled shaft is discret-
ized into a series of equidistant nodes along the
length of the shaftsoil interface beginning at the
tip and proceeding to the head. Using the central
difference methodology, a set of algebraic equa-
tions is written to solve for the displacement at
each node using the governing Eq. (7) (Hilde-
brand 1974):
u
i1

K
K
m
u
i
Dz
2
2u
i
u
i1
12
where, u
i
is the nodal displacement and Dz is
the distance between each node. The subscript i
refers to the i-th node along the shaftsoil
interface, and the nodes are numbered sequen-
tially from the tip of the drilled shaft (node 0)
to the head (node x). As a result, exactly (x+1)
nodes exist, enabling us to write (x+1) equations
for the nodal displacements. However, it is
evident from Eq. (12) that two additional un-
known displacements are introduced into the
system (u
1
and u
x+1
). Therefore, to solve for
the nodal displacements, it is necessary to
incorporate two boundary conditions. Again,
using a central difference formulation, the
boundary conditions can be written in terms of
nodal displacements for tip force and applied
shaft load, respectively:
u
1
u
0
K
t
K
m
2Dz u
1
13
u
x1
u
x1

P
K
m
2Dz 14
Once these boundary conditions are intro-
duced, (x+3) equations can be written. When the
head of the drilled shaft is subjected to the initial
load, the set of algebraic equations is solved for
displacement at each discrete node. The nite
difference formulation requires that nodal dis-
placements be solved at small incremental load-
ing steps. This is necessary for two reasons. First,
the value of the shear modulus of shaftsoil
interface sub-grade reaction, K, and the tip soil
stiffness, K
t
, must be updated at each load incre-
ment. Second, as the load is incrementally in-
creased on the drilled shaft, the shaftsoil
interface will begin to yield from the head of the
drilled shaft to the tip. By substituting q
o
=p Ds
u
,
into Eqs. (4) and (5) and then dividing by the
displacement at node i, u
i
, the following expres-
sion is developed:
K
i
K
init
q
o
q
o
u
i
R
f
K
init
15
where, K
i
, is the secant shear modulus of shaft
soil interface sub-grade reaction at the node of
interest corresponding to the node displacement,
u
i
. A similar equation can be written for the tip
soil. By dividing Eq. (10) by the tip displacement,
u
0
, yields the following:
K
t
K
ti
P
utip
P
utip
u
0
R
f
K
ti
16
where, K
t
, is the secant stiffness of the tip soil
corresponding to the tip displacement, u
0
.
Therefore, at each load increment of the nite
difference technique, K
i
can be determined for
the i-th node based upon the nodal displacement
calculated from the previous load step. Corre-
spondingly, K
t
can be determined based upon the
tip displacement from the previous load step. It is
70 Geotech Geol Eng (2007) 25:6577
1 3
advantageous to represent the decrease in stiff-
ness of the shaftsoil interface and tip soil with a
line secant to the hyperbolic curve rather than a
line tangent to the curve. Unlike a tangent line,
the slope of a line secant to a hyperbolic curve
will never be negative, even for strain softening
soils. A negative value for K
i
or for K
t
in the nite
difference matrix calculations will result in a sin-
gular matrix and an unsolvable system of equa-
tions.
It is important to observe that in a total stress
analysis using undrained parameters, the magni-
tude of q
o
, along with K
init
, is constant in Eq. (15)
for each node along the length of the drilled shaft.
However, in an effective stress analysis, the
magnitude of q
o
and K
init
are both dependant
upon the conning stress. By performing a nite
difference analysis, the dependence of the shaft
soil interface parameters with conning stress is
easily handled by the discretization of the drilled
shaft into equidistant nodes. Closed form rela-
tionships for shaft displacement are not possible
in an effective stress analysis, even when simple
linear loaddisplacement behavior of the soil
interface is assumed (see Misra and Roberts
2005); hence, the nite difference method be-
comes an efcient means to handle this circum-
stance as well.
Once K
i
for the i-th node of the shaftsoil
interface and K
t
for the tip soil node are known,
the displacements at each node are calculated. In
addition, the tip force, P
t
, at each load increment
can be calculated using nite difference formulas
based upon the tip displacement as follows:
P
t
K
m
u
1
u
1

2Dz
17
The nite difference technique is continued by
increasing the applied shaft load incrementally,
updating the load vector and the stiffness vector,
and solving for the new nodal displacements. The
process is completed for each node along the
length of the shaftsoil interface and at the tip. As
the displacement of each shaftsoil interface node
increases with an increase in the applied shaft
load, K
i
for each node will approach zero and the
shaftsoil interface will fail. The shaftsoil inter-
face nodes will fail from the head of the drilled
shaft to the tip, and the tip soil will carry a larger
portion of the total load at each load increment.
Eventually, the shaftsoil interface will yield at all
nodes and all additional load on the drilled shaft
will be taken completely by the tip soil. Once the
tip force, P
t
, calculated using Eq. (17) reaches the
tip bearing capacity given by Eq. (11), the drilled
shaft will fail by plunging.
It is pertinent to note that the central difference
method can diverge, producing meaningless
results, in the presence of numerical round-off if
the distance increment, Dz, is not small enough.
For this paper, the distance increment was taken
as 100 mm in the nite difference analysis. The
specic requirement for stability is (Hildebrand
1974):
Dz60:1

K
m
K
init
r
18
For the values used in the nite difference tech-
nique, discussed later in the paper, it is discovered
that the stability requirement is met.
Monte Carlo simulation and probabilistic analysis
Reliability based design differs from the factor of
safety approach typically used in geotechnical
design. Using the factor of safety approach,
nominal values of the shaftsoil interface proper-
ties are used to obtain a representative loaddis-
placement curve as a basis for design of a drilled
shaft foundation. A factor of safety is applied to
the design in order to account for the uncertainty
in the shaftsoil interface properties that can arise
from four primary sources: inherent variability,
measurement error, and transformation uncer-
tainties (Phoon and Kulhawy 1999), and con-
struction techniques. Unfortunately, a factor of
safety approach can only indirectly account for
shaftsoil interface variability because the mag-
nitude of uncertainty is not quantied. Therefore,
to improve the factor of safety approach, it is
desirable to consider the shaftsoil interface
properties as random variables and describe the
properties statistically using non-Gaussian prob-
ability distribution functions. The shaftsoil
interface properties, K and s
u
, are assumed to be
Geotech Geol Eng (2007) 25:6577 71
1 3
randomvariables and are described statistically by
the lognormal probability distribution function.
For this paper, the tip stiffness, K
t
, was not con-
sidered as a random variable. The tip stiffness
predominately affects the post yield displacement
behavior of the drilled shaft installed in deep
deposits and is often less sensitive to variability in
construction techniques. In addition, introducing a
third random variable into the probabilistic cal-
culations would result in an unacceptable increase
in computational time and was therefore not
considered in this paper.
Since the loaddisplacement relationship for
the drilled shaft are highly nonlinear and have to
be obtained using a numerical technique, proba-
bilistic relations cannot be calculated using closed
form relations, such as the rst-order second
moment (FOSM) method (Misra and Roberts
2006). Consequently, the Monte Carlo simulation
method is utilized to obtain the probabilistic
loaddisplacement solutions for the drilled shaft.
During each trial in a Monte Carlo simulation, a
random number is generated from within the
described probability distribution function for
each random variable. The solution is processed
using these random variables, and the process is
repeated for a set number of trials. The required
number of Monte Carlo trials is based upon
achieving a particular level of reliability (Harr
1996; Baecher and Christian 2003). For the results
reported in this paper, Monte Carlo simulations
with 5000 trials were conducted as described in
Misra and Roberts (2006). These simulations
were performed using a computer program writ-
ten utilizing Mathcad (2002) software for both
total stress analysis and effective stress analysis.
Parameter variability models
For the total stress analysis model with undrained
parameters, the shaftsoil interface properties,
K
init
and s
u
, were taken to be independent of
conning stress and were assigned mean values
that were back-calculated from pullout tests
conducted by Phoon et al. (1995) on drilled shafts
installed in clay and silts. Using this data, a mean
of 90 kPa was assigned to s
u
, and a mean of
89 MPa was assigned to K
init
. Based on the range
of the data, the coefcient of variation, or COV,
for each random variable was set at 0.30. It should
be noted that a COV of 0.30 signies that a ran-
dom variable can assume values 90100% lower
or higher than the mean value (Lacasse and
Nadim 1996). This range can be quite large, pro-
ducing values of the shaftsoil interface property
that may not be representative of the magnitude
that the interface property may vary. Therefore, a
COV of 0.30 may not only account for inherent
variability, but can also incorporate other sources
of variability, such as model uncertainty or mea-
surement errors.
For the effective stress analysis, the shaftsoil
interface properties are dependent on conning
stress as given by Eqs. (2), (3) and (6). For further
discussion, Eq. (3) is rewritten in the form:
a
b
ep
c
e
bz
19
As seen from Eq. (19), the range of possible
values for b
ep
, at a given depth z may be de-
scribed by adjusting the value of the curve tting
parameter, a, while the variables, b and c, are held
constant. In Fig. 3, we have shown the average t
as well as the 95% condence limit upper and
lower bounds to the data by varying parameter a.
Finally, we observe that for the mean value of
parameter, a, given in Eq. (3), the magnitude of s
u
falls within the practical limits given by Olson
(1990) for various types of piles in sands. These
limiting values resulted in small scatter in the
measured to computed capacity for the piles.
Based upon the density of the material, Olson
(1990) calculated limiting values of 55 kPa to
190 kPa for loose to dense sands, respectively.
For example, using a nominal unit weight of soil
equal to 18.85 kN/m
3
and a depth of 30m, we
obtain a maximum value of s
u
equal to 188 kPa
based upon Eqs. (2) and (3).
The variability of the shear modulus of soil-
structure interface sub-grade reaction, K
init
, was
determined from the loaddisplacement data
provided in Rollins et al. (2005) for pullout load
tests conducted in Utah and other sites. From the
loaddisplacement data for each test, the value of
K
init
was back-calculated and normalized with
respect to atmospheric pressure and plotted ver-
sus the normalized conning stress, as shown in
72 Geotech Geol Eng (2007) 25:6577
1 3
Fig. 4. A power law was t to the plotted values in
order to determine the value of the modulus
number, K
mod
, and the exponent constant, x, for
use in Eq. (6). From Fig. 4, we observe that the
value of K
mod
was determined to be 792 and the
exponent constant, x, was found to be 0.15.
Load displacement curves
For all loaddisplacement curve simula-
tions, additional shaft and soil parameters were
as follows: D=910 mm; L
b
=10 m; L
d
=1 m;
E
s
=100000 kPa; l
s
=0.40; E
m
=26300 MPa (drilled
shaft elastic modulus); and q
t
=2500 kPa. From the
values given, it can be observed that the tip soil
was assumed to be a much stiffer medium than
the shaftsoil interface, typical of most drilled
shaft installations that bear on intermediate geo-
materials or rock. In addition, the value of the
failure ratio, R
f
, for the shaftsoil interface and
the tip soil was taken as 1.0 for simplicity.
In Fig. 5, we plot the loaddisplacement curves
for the rst twenty Monte Carlo simulations using
the back-calculated mean values of K
init
and s
u
for the effective stress analysis, the shaft/soil
parameters as given above, and a COV of the
shaftsoil interface properties equal to 0.30.
Classic hyperbolic loaddisplacement curves can
be observed, characterized by an initial tangent
modulus that decreases with an increase in load, a
pseudo yield point where the shaftsoil interface
yields completely, and a point of ultimate failure
where the tip of the drilled shaft fails by plunging.
Similar results are obtained for a total stress
analysis and are only dependent on the mean of
the shaft-soil interface properties chosen in each
analysis such as those shown in Misra and Roberts
(2006).
Reliability based design analysis
Service limit state design is controlled by the
allowable displacement of a structure. If the dis-
placement of a structure exceeds an allowable
value, the structure may still be structurally ade-
quate, but may cause user discomfort. The chal-
lenge for the engineer is often determining the
magnitude of the allowable displacement and can
only be accomplished if sufcient and relative
eld data is available (FHWA 1982). For bridge
structures, angular distortion is often the con-
trolling factor, and is a function of vertical dis-
placement of the foundation system and bridge
span lengths. Thus, a single allowable displace-
ment value that encompasses all types of struc-
tures or all design situations is difcult to dene.
A review of the literature does provide some
guidance, including NCHRP Report 343, which
provides a table of allowable displacements for
bridges expressed in terms of settlement magni-
tude (Barker et al. 1991). Settlement criterion in
terms of angular distortion is also provided and is
expressed in terms of tolerable movements for
single span and multiple span bridges.
The loaddisplacement data obtained from the
Monte Carlo simulations can be utilized to de-
velop a load capacity frequency diagram based
upon an allowable displacement. In Fig. 6, we
y = 792 x
0.15
0
500
1000
1500
0.00 0.20 0.40 0.60 0.80
K

/

a
t
m
K
o v
/
atm

Fig. 4 Determination of modulus number, K
mod
, and
exponent constant, x, for a drained soil prole
0
1000
2000
3000
4000
5000
6000
7000
0 20 40 60 80 100 120 140
u (mm)
L
o
a
d

(
k
N
)
Fig. 5 Loaddisplacement curves for the rst 20 Monte
Carlo simulations for an effective stress analysis, assuming
a lognormal distribution and COV of 0.30 for the shaftsoil
interface properties
Geotech Geol Eng (2007) 25:6577 73
1 3
have plotted the cumulative distribution fre-
quency diagram for the drilled shaft load capacity
for the effective stress analysis based upon an
allowable displacement of 10 mm. From the
probability of failure, one can obtain the reli-
ability index, b, with the following equation:
b U
1
p
f
20
where, F() is the cumulative standard normal
distribution function.
The reliability index is an alternative method
for presenting the probability of service limit state
failure on a more convenient scale since the
probability of failure is difcult to assess when its
value is very small (Kulhawy and Phoon 1996). It
is also a way to remove the negative connotation
associated with the word failure, especially for
service limit state design. In a reliability based
design procedure, it is desirable to obtain a target
reliability index, b
T
, that corresponds to a pre-
determined probability of failure. A target reli-
ability index equal to 3.2 at the ultimate limit
state corresponds well with global factor of safety
designs and empirical rates of failure for foun-
dations adjusted to theoretical failure rates
(Kulhawy and Phoon 1996). However, because
the consequences of exceeding an allowable dis-
placement are much less than exceeding an ulti-
mate capacity, the target reliability index for the
service limit state is set much lower; usually
around 2.6 (see Phoon et al. 1995), which corre-
sponds to a probability of failure of about 0.5%.
If the load on the drilled shaft is assumed to be
deterministic, the probability of drilled shaft
failure, p
f
, at the service limit state can be ob-
tained directly from Fig. 6, and, further, the factor
of safety, FS, against service limit state failure
may be calculated as:
FS
R
Q
21
where, Q = deterministic load on the drilled shaft
and R = drilled shaft load capacity. When the
factor of safety is calculated to be less than unity,
failure of the drilled shaft at the service limit state
will occur. If the factor of safety histogram is
assumed to follow a lognormal distribution, a
common assumption made in geotechnical engi-
neering, the probability of drilled shaft failure, p
f
,
at the service limit state can be calculated as:
p
f
pFS\1 U
ln1 l
ln
FS
r
ln
FS

22
where, l
lnFS
is the log mean of the factor of
safety, and r
lnFS
is the log standard deviation of
the factor of safety. Assuming a deterministic
load acting on the drilled shaft of 2000 kN, for
example, a cumulative distribution histogram for
factor of safety can be obtained using Eq. (21)
based upon the load capacity histogram from
Fig. 6. The factor of safety histogram is plotted in
Fig. 7 for an effective stress analysis. Similar re-
sults are obtained for a total stress analysis.
Hence, from Fig. 7 or from Eq. (22), the
probability of drilled shaft failure at the service
limit state for a deterministic load of 2000 kN is
calculated to be approximately 2.1%, corre-
sponding to a reliability index of about 2.0. This is
smaller than the target reliability index of 2.6
assumed for the service limit state. Therefore, for
this design scenario, the drilled shaft design must
be revised in order to satisfy the failure criterion
for the service limit state.
The development of the load capacity and
factor of safety histograms was based upon using
a deterministic value for the drilled shaft load.
Using the standard load and resistance factor
design (LRFD) methodology, however, requires
that the load also be dened in terms of a prob-
Fig. 6 Cumulative distribution histogram for drilled shaft
load capacity at an allowable displacement of 10 mm for
the effective stress analysis. Total stress analysis yields
similar results
74 Geotech Geol Eng (2007) 25:6577
1 3
ability distribution function. Hence, by assuming
that the load capacity histogram follows the log-
normal distribution and knowing the COV of the
load capacity from the simulation data, it is pos-
sible to calculate the resistance factor, /, for
service limit state design from the following for-
mula given by Baecher and Christian (2003):
/
k
R
c
D
EQ
D

EQ
L

c
L


1X
2
QD
X
2
QL
1X
2
R
r
k
QD
EQ
D

EQ
L

k
QL

e
b
T

ln 1X
2
R
1X
2
QD
X
2
QL

p
23
where, k
QD
,k
QL
, k
R
are the bias of the dead load,
live load, and load capacity, respectively, c
D
and
c
L
and the load factors for the dead load and live
load acting on the drilled shaft, respectively, W
QD
,
W
QL
, W
R
, are the COV for the dead load, live
load, and load capacity, respectively, and E(Q
D
)
and E(Q
L
) and the expected values of the dead
load and live load, respectively. Suggested values
for the bias of the dead load and live load, along
with the COV of the dead load and live load, are
given by Baecher and Christian (2003). The dead
and live load factors, with values greater than or
equal to 1, were assigned based upon the
AASHTO LRFD Bridge Design Specications
(2004) for the Service Limit State I. It is observed
that the ratio of the expected dead load to the
expected live load does not signicantly affect the
value of the resistance factor (Baecher and
Christian 2003; Paikowsky et al. 2004) and can be
set to a value such as 2.0, assuming a typical
highway bridge, for example. From the loaddis-
placement data used to create Fig. 6, the COV of
the load capacity is found to be 0.132 Therefore,
utilizing Eq. (23) with a target reliability index,
b
T
, of 2.6 and assigning values to the remaining
variables as described above, the resistance factor
is calculated to be 0.51 for service limit state de-
sign assuming effective stress parameters. It
would be appropriate to use a value of 0.50, as
resistance factors are typically rounded to the
nearest 0.05 (Baecher and Christian 2003).
The use of resistance factors in geotechnical
design has several advantages. First, most engi-
neers are familiar with using resistance factors
and can directly gauge how the factor of safety of
the design is accounted for by the resistance fac-
tor. Second, by using resistance factors, the
engineer does not need to perform the probability
computations or create the load capacity or factor
of safety histograms for each design. Most
importantly is the fact that resistance factors are
calibrated to produce designs that consistently
achieve the desired level of reliability (Phoon
et al. 1995). To that end, it is observed that using
a probabilistic design approach and dening
shaftsoil interface properties in terms of random
variables, provides a robust method for deter-
mining service limit state failure for drilled shafts.
Conclusions
The loaddisplacement analysis of drilled shafts
subjected to compression loads was modeled
based upon the t-z method. The t-z method
models the shaftsoil interface and tip soil as a
series of springs. Since the springs were assumed
to be non-linear, a nite difference technique was
used to solve the governing differential equation
that describes the soil-structure interaction. The
nite difference solution requires the discretiza-
tion of the shaftsoil interface into a series of
equidistant nodes. In the case of a drilled shaft
subjected to a compression load, the load at the
head of the drilled shaft and the resistance of the
soil at the tip are known, and are applied as
boundary conditions in the system. As the drilled
Fig. 7 Cumulative distribution histogram for drilled shaft
factor of safety against service limit state failure based
upon a deterministic load of 2000 kN
Geotech Geol Eng (2007) 25:6577 75
1 3
shaft is loaded incrementally, the boundary con-
ditions and spring stiffnesses are updated, and the
displacements are solved at each node. The nodes
will begin to fail, starting at the top of the drilled
shaft and progressing to the tip, and eventually
the tip soil will carry all additional load. Once the
load on the tip soil reaches the ultimate tip
capacity, the drilled shaft will fail by plunging.
With the view of developing methods for reli-
ability based design, the nite difference tech-
nique was combined with the Monte Carlo
simulation method to create a probabilistic load
displacement analysis. The Monte Carlo simula-
tion method was used, in lieu of other closed-form
probabilistic techniques, due to the complexity of
the loaddisplacement analyses. The shaftsoil
interface properties, given in terms of the shear
modulus of shaft-soil interface sub-grade reac-
tion, K, and the ultimate shaft-soil interface shear
strength, s
u
, can be developed in terms of total or
effective stresses.
Using the loaddisplacement data generated
from the Monte Carlosimulation, several solutions
are possible to describe the probability of drilled
shaft failure at the service limit state. In this paper,
we have created a cumulative distribution histo-
gram of the drilled shaft load capacity given an
allowable head displacement of 10 mm. Consid-
ering the load on the drilled shaft as a determin-
istic value, a factor of safety cumulative
distribution histogram was also created. Standard
statistical methods were used to calculate the
probability of service limit state failure assuming
the factor of safety histogram followed a lognor-
mal distribution. Most importantly, however, a
procedure to calculate resistance factors for ser-
vice limit state design was demonstrated. Resis-
tance factors are calibrated to produce a design
that consistently achieves a desired level of reli-
ability against failure. Because the probabilistic
methods presented can be somewhat cumbersome
and time-consuming, the use of resistance factors
for standard designs is found to be benecial.
References
AASHTO (2004) LRFD bridge design specications. 3rd
edn. American Association of State Highway and
Transportation Ofcials, Washington, DC
Baecher GB, Christian JT (2003) Reliability and statistics
in geotechnical engineering. Wiley, West Sussex, UK
Barker RM, Duncan JM, Rojiani KB, Ooi PSK, Tan CK,
Kim SG (1991) NCHRP report 343: manual for the
design of bridge foundations, Transportation Re-
search Board, National Research Council, Washing-
ton, DC
Burland J (1973) Shaft friction in piles in clay: a simple
fundamental approach. Ground Eng 6(3):3042
Chandler RJ (1968) The shaft friction of piles in cohesive
soils in terms of effective stress, Civil Eng Public
Works Rev, 63:4851
Coduto DP (2001) Foundation design: principles and
practices. Prentice-Hall, New Jersey
Das BM (2004) Principles of foundation engineering.
Thomson, New York
Duncan JM, Chang CY (1970) Nonlinear analysis of stress
and strain in soils. J Soil Mech Found Div, ASCE,
96(SM 5):16291653
Duncan JM, Byrne P, Wong KS, Marby P (1980) Strength,
stressstrain, and bulk modulus parameters for nite
element analyses of stresses and movements in soil
masses. Report No. UCB/GT/80-01, College of
Engineering Ofce of Research Services, University
of California Berkeley, Berkeley, CA
FHWA (1982) Tolerable movement criteria for highway
bridges, vol. 1 Interim Report. Report No. FHWA-
RD-81-162, Federal Highway Administration, U.S
Department of Transportation, McLean, VA
FHWA (1999) Drilled shafts: construction procedures and
design methods. Report No. FHWA-IF-99-025, Fed-
eral Highway Administration, U.S. Department of
Transportation, Mclean, VA
Harr ME (1996) Reliability-based design in civil engi-
neering. Dover Publications Inc., Mineola, New York
Hildebrand FB (1974) Introduction to numerical analysis.
Dover Publications Inc., Mineola New York
Janbu N (1963) Soil compressibility as determined by
oedometer and triaxial tests, vol. I, European Con-
ference on Soil Mechanics and Foundations Engi-
neering, Wiesbaden, Germany, pp 1925
Johnson KL (1985) Contact mechanics. Cambridge Uni-
versity Press, London UK
Kondner RL (1963) Hyperbolic stressstrain response:
cohesive soils, J Soil Mech Found Div, ASCE,
89(SM 1): 115143
Kraft LM, Ray RM and Kagawa T (1981) Theoretical t-z
curves. J Geotech Eng, ASCE 107(11):15431561
Kulhawy FH (1991) Drilled shaft foundations. In: Fang
HY (ed) Foundation Engineering Handbook, 2nd
edn. Van Nostrand Reinhold, pp 537552
Kulhawy FH, Phoon KK (1996) Engineering judgment in
the evolution from deterministic to reliability-based
foundation design, Uncertainty 96, Geotechnical
Special Publication No. 58, ASCE, New York, 1, pp
2948
Lacasse S, Nadim F (1996) Uncertainties in characterizing
soil properties, Uncertainty 96, Geotechnical Special
Publication No. 58, ASCE, New York, 1, pp 4975
Mathcad version 11 (2002). Mathsoft Engineering &
Education, Inc., Cambridge, MA
76 Geotech Geol Eng (2007) 25:6577
1 3
Meyerhof GG (1976) Bearing capacity and settlement of
pile foundations. J Geotech Eng, ASCE, 102(GT3):
197228
Misra A, Chen C-H (2004) Analytical solutions for
micropile design under tension and compression.
J Geotech Geol Eng 22(2):199225
Misra A, Roberts LA (2005) Probabilistic axial load
displacement relationships for drilled shafts, Pro-
ceedings Geofrontiers 2005, Austin, TX, ASCE,
Arlington, VA
Misra A, Roberts LA (2006) Axial service limit state de-
sign of drilled shafts using probabilistic approach.
J Geotech Geol Eng 24(5):120
Olson RE (1990) Axial load capacity of steel pipe piles in
sand. Proceedings Offshore Technology Conference,
Houston, TX, pp 1724
Paikowsky SG, Birgisson B, McVay M, Nguyen T, Kuo C,
Baecher G, Ayyub B, Stenersen K, OMalley K,
Chernauskas L, ONeill M (2004) NCHRP report 507:
load and resistance factor design (LRFD) for deep
foundations, Transportation Research Board,
National Research Council, Washington, DC
Phoon K-K, Kulhawy FH, Grigoriu MD (1995) Reliability-
based design of foundations for transmission line
structures, Report TR-105000, Electric Power
Research Institute, Palo Alto, CA, July 1995, p 380
Phoon KK, Kulhawy FH (1999) Characterization of geo-
technical variability. Can Geotech J 36(4):612624
Reese LC, ONeill MW (1987) Drilled shafts: construction
procedures and design methods, in Report No.
FHWA-HI-88-042, Federal Highway Administration,
U.S. Department of Transportation, McLean,
Virginia
Rollins KM, Clayton RJ, Mikesell RC, Blaise BC (2005)
Drilled shaft side friction in gravelly soils. J Geotech
Geoenviron Eng, ASCE 131(8):9871003
Scott RF (1981) Foundation analysis. Prentice-Hall, New
Jersey
Terzaghi K, Peck RB, Mesri G (1996) Soil mechanics in
engineering practice. John Wiley and Sons Inc., New
York
Geotech Geol Eng (2007) 25:6577 77
1 3

You might also like