Analysis of Pile Foundations under Seismic Loading

Jayram Ramachandran

CBE Institute 2005: Final Report

1 Introduction

Pile foundations are commonly used to transfer loads from a structure to the ground in cases
where the structural loads are very high or at locations where the soil at shallow depths can-
not carry the imposed loads. Under moderate and strong seismic loading, pile foundations
undergo large displacements and the behavior of the pile-soil system can be strongly non-
linear. Such pile-soil systems interact with the supported structure; hence their nonlinear

behavior is extremely important in evaluating the seismic response of pile-supported struc-

tures. In seismic regions, the analysis and design of pile-supported structures requires an
accurate prediction of the pile head response and the load resistance to lateral shaking caused
by earthquake ground motions. In order to predict the response of pile groups under seismic

loading, it is important to understand the interaction between the piles in a group. This
research studies the interaction between piles in pile groups accounting for soil nonlinearity
and pile-soil gapping effects.
Most pile foundations consist of a group of piles rather than a single pile. Thus, an important

component of the analysis of pile supported structures is the ability to perform an accurate
and efficient analysis of pile groups. However, most pile group analysis techniques are an
extension of techniques used to predict the response of single piles. Therefore, Section 2
presents a brief review of single pile analysis techniques. The Beam on Winkler Foundation
model is a computationally efficient tool to estimate the response of single piles accounting

for pile and soil nonlinearity. A number of researchers have used this technique and hence,
this model is briefly described. Section 3 presents an overview of the current techniques
available to perform pile group analysis. These include elasticity based methods, wave prop-
agation based methods and the p-multiplier method. Section 4 presents the results of a

numerical study using 3D finite element analysis to investigate pile-to-pile interaction in a

pile group, accounting for soil nonlinearity and pile-soil gapping effects.

1
2 Single Pile Analysis

Most pile group analysis techniques are an extension of methods employed to compute sin-
gle pile response. Therefore, it is highly instructive to review techniques used to compute
single pile response, including the various numerical and analytical methods that have been
developed. Closed form analytical solutions have been obtained by a number of researchers
for single piles under axial or lateral loading assuming the soil to be linear viscoelastic and

the pile to be a linear elastic beam [15, 17, 18, 24]. Such closed form solutions are hard
to extend to reflect the true nonlinear behavior of the pile-soil system. Some researchers
include soil nonlinearity in the analysis using an equivalent viscous dashpot [14]. The most
direct method of analysis involves 2D/3D modeling of the pile and the soil continuum using

the finite element method. In such an analysis, the soil nonlinearity may be considered ex-
plicitly by using appropriate material models [12, 37]. The finite element method, although
powerful, is computationally very expensive since it requires discretization of the pile and
the large volume of surrounding soil. In contrast, the Beam-on-Winkler-Foundation model

is a versatile and efficient approach to pile foundation analysis that has been adopted widely.
This section reviews the Winkler model.

2.1 Beam on Winkler Foundation Model

The Beam on Winkler Foundation Model model or Winkler model is commonly used to
study the response of single piles. The pile is modeled as a beam while the surrounding

soil is modeled using continuously distributed springs and dashpots. Pile nonlinearity may
be considered in the analysis using an appropriate nonlinear material model. Hysteretic
energy loss in the soil is modeled using nonlinear springs whose parameters are determined
experimentally from load-deformation curves [21, 22] or analytically by defining stiffness
and strength parameters [3, 7, 8, 28]. The dashpots placed in parallel with the nonlinear

2
 Figure 1: Beam on Winkler foundation model for a single pile under lateral loading

springs account for energy loss due to radiation damping, under dynamic loading conditions.

Since radiation damping is a frequency dependent phenomenon, the dashpot coefficients are
usually frequency dependent. The Winkler model is easily implemented for analysis of non-
linear pile-soil systems using most standard 1-D finite element programs. Within the finite
element framework, the pile is modeled using discrete beam elements and the continuous
soil springs/dashpots are replaced by discrete springs and dashpots. Figure 1 shows the dis-

cretized form of the Winkler model for the analysis of a single pile under dynamic (including
seismic) lateral loading. A similar model may be used for the analysis of single piles under
static lateral loads. However, under static loads, the dashpots would be irrelevant. The

3
Winkler model has been shown to provide a good estimate of the pile response by several
researchers [3, 32].

2.2 Model parameters

The Winkler model requires definition of parameters for the nonlinear springs as well as

viscous dashpots. This section defines the parameters for the Winkler model under lateral
loading. The nonlinear spring may be represented using either the bilinear or Bouc-Wen
[35] models. The bilinear model is a very simple representation of the hysteretic behavior of
the soil, but its main attraction is that it can be easily implemented in most finite element

programs. The Bouc-Wen model is much more versatile as it allows a smooth transition from
linear to nonlinear behavior. However, it is not implemented in a number of commonly used
analysis programs. Both the bilinear and Bouc-Wen models require definition of the initial
spring stiffness and strength at large displacements (assuming a zero post yield stiffness [3]).
Based on a number of analytical studies that compare pile head displacements from Winkler

model analyses with 3-D finite element analyses, the stiffness of the soil springs per unit
length under lateral loading can be approximated realistically by [14, 17]:

kx = 1.2Es (z) (1)

where kx is stiffness of the continuously distributed spring and Es (z) is the Young’s modulus
of the soil as a function of depth z. The value in equation 1 is multiplied by the spacing

between two adjacent springs to obtain the discrete spring stiffness. The maximum force in
the spring is equal to the ultimate lateral reaction per unit length of pile. For cohesive soils,
the lateral soil strength based on theoretical studies [3, 8, 28] using the theory of plasticity

4
under plane-strain conditions is given by:

Fs,max (z) = λSu (z)d (2)

where Fs,max (z) is the maximum lateral force exerted by the soil on the pile (as a function
of depth, z), d is the pile diameter, Su (z) is the variation of shear strength with depth and
λ is a dimensionless parameter. Considerable research has been done to quantify λ [8, 19].

λ values between 9 and 12 may be appropriate at depths where plane strain conditions are
valid. At shallow depths, plane strain conditions are not valid due to vertical deformation
of the soil during lateral motion of the pile. Hence, λ values of 2 or 3 have been suggested.
The following expression for the variation of λ with depth is recommended [19]:

σx z
λ(z) = 3 + +J (3)
Su d

where σx is the overburden pressure and J is a co-efficient that is obtained by calibrating

the model against known experimental data. In the absence of such data, the recommended
value of J = 0.5 may be used.
For cohesionless soils, the following expression for the limiting force of the soil on the pile
[7] is used:
1 + sin φ
Fs,max (z) = µγs d z (4)
1 − sin φ

where Fs,max (z) is the maximum lateral force exerted by the soil on the pile (as a function
of depth, z), γs is the specific weight of the soil, φ is the angle of internal friction and µ is

a dimensionless parameter. A value of µ=3 has been suggested [7]. The maximum force (or
strength) of the spring is obtained by multiplying the maximum reaction force of the soil in
equations 2 and 4 with the spacing between two adjacent springs, s.
A number of expressions are available for the dashpot coefficient [14, 34]. The expression

5
used in this study defines a frequency dependent dashpot coefficient as used by [3]:

c(a0 , z) = Qa−0.25
0 ρs Vs ds (5)

where c is the dashpot coefficient, a0 is a non-dimensional frequency parameter (= ωd/Vs),

ω is the angular frequency, ρs is the soil density, Vs is the shear wave velocity in the soil
medium, d is the pile diameter, and s is the spacing between two adjacent dashpots. The

coefficient Q is given by the expression:



3
 if z < 3d,
Q= (6)
3.4
]1.25 ( π4 )0.75

2[1 +
 if z > 3d.
π(1−νs

where νs is the Poisson’s ratio of the soil.

6
3 Pile Group Analysis

Most pile foundations consist of a group of piles rather than a single pile. Thus, an important
component of the analysis of pile supported structures is the ability to perform an accurate
and efficient analysis of pile groups. Just as for single piles, various numerical and analytical
methods have been developed for the analysis of pile groups. Currently, the most accurate
method of pile group analysis may be the 3D finite element analysis method. Such analyses

have been performed under static [11, 38] as well as dynamic (including seismic) [16] loading
conditions. However, solving any problem with the 3D finite element method is extremely
time consuming and hence numerous attempts have been made to find more computationally
efficient solution procedures. A quasi-three-dimensional method is presented in [36], where

nonlinear dynamic analysis of single piles and pile groups is performed in the time domain
using strain-dependent moduli and damping, and yielding at failure. Since most pile groups
are made up of identical piles and the soil profile at the locations of all piles in the group
may also be assumed to be identical, one of the most common techniques of pile group

analysis involves modifying the response of a single pile by a suitable factor to obtain the
response of the group [13, 15, 17, 27]. The response of the single pile may be obtained
using any of a wide range of techniques, but, most commonly a Winkler model is used, as
discussed in Section 2. The resistance provided by the group under vertical or lateral loading
is generally not equal to the sum of the resistances of the individual piles. Most often, the

group resistance is less than the sum of the individual pile resistances and is a function of
the pile group configuration as well as pile spacing. A method for analysis of pile groups
using interaction factors and based on the theory of elasticity was proposed by Poulos and
Davis [27]. This method is applicable to pile groups in elastic soils under static loading

conditions. Approximate analytical closed form solutions for pile-to-pile interaction factors
under dynamic loads, based on frequency domain solutions are also available [13, 15, 17].

7
These solution procedures have also been modified to accommodate time domain analysis
procedures for nonlinear problems [4]. This section reviews some of the current procedures
used to analyze pile groups.

3.1 Use of pile-to-pile interaction factors in pile group analysis

3.1.1 Elasticity based solution

Poulos and Davis [27] developed a method of analysis of pile groups using single pile analysis
results along with pile-to-pile interaction factors. These solutions are based on elasticity and
apply to pile groups under static loading. Separate but similar procedures were developed for

the analysis of pile groups under both axial and lateral loading conditions. In this method,
the response of each pile in a group is a sum of the response of the pile if it were a single
pile and the additional displacements induced in the pile due to its interaction with each of
the other piles in the group. In computing the response of the single pile under lateral loads,

the pile is assumed to be a thin rectangular vertical strip of constant flexibility and the soil
is considered to be a continuum. The pile is divided into a number of elements, each acted
upon by a uniform horizontal stress, which is assumed to be constant across the width of the
pile. The soil is assumed to be an ideal, homogeneous, isotropic, semi-infinite elastic material

with properties unaffected by the presence of the pile. A solution is obtained by imposing
compatibility between the displacements of the pile and the soil for each element of the pile.
The soil displacement influence factors, relating the soil displacements with the horizontal
stress, are evaluated by integration over a rectangular area of the Mindlin equation for the
horizontal displacement of a point within a semi-infinite mass caused by a horizontal point

load within the mass. This solution is also available for piles in soils with Young’s modulus
varying with depth. The soil displacement influence factors are a function of the pile flex-
ibility factor which measures the relative pile to soil stiffness, pile length to diameter ratio

8
and the Poisson’s ratio of the soil. These influence factors have been computed considering
the effect of the different parameters [25, 27]. The procedures used in the elasticity based
analysis of single piles may be found in great detail in the references cited.
A major advantage of the elasticity based method of analysis is that since the soil is treated

as a continuum, it is possible to consider the effects of interaction between piles. The

method employed to compute the pile group response is an extension of the procedure used
to compute the response of single piles [26]. The interaction between two identical piles is
considered first and the analysis is then extended to general pile groups. The two identi-
cal, equally loaded piles are each divided into a certain number of elements. While elastic

conditions prevail, the horizontal displacements of the soil and pile at each element may
be equated, and together with the equilibrium equations may be solved for the unknown
pressures. In this analysis, the only interaction effect that is considered is the horizontal
movement of one pile that results from loading on another pile. The interaction factors

are expressed as a ratio of the additional displacement caused by the adjacent pile to the
displacement of the pile caused by its own loading. Interaction factors as a function of the
spacing between piles, pile flexibility factor, departure angle (for laterally loaded piles – the
angle between the horizontal line joining the two piles and the line of action of the load),

pile head fixity condition, pile length to diameter ratio and the Poisson’s ratio of the soil are
available [27]. The method of analysis of pile groups with two piles is extended to a gen-
eral pile group using the principle of superposition. The additional horizontal displacement
caused in a particular pile in the group, due to the loading on every other pile in the group is
computed by considering the interaction between piles in pairs and summing the additional

displacements caused by each of the adjacent piles in the group.

One of the major disadvantages of this method is that it is hard to extend this method
to compute the load-deformation response of pile foundations with significant nonlinear de-
mands. It is also hard to analyze the foundation under complex loading scenarios such as

9
seismic or other dynamic loads.

3.1.2 Wave propagation based solution

Wave propagation based methods use wave propagation theory to compute the interaction
between piles in groups under dynamic loads. These interaction factors are combined with the
single pile response computed using any procedure (usually Winkler model type procedure)

to compute the response of pile groups. These procedures were originally developed in
the frequency domain and applied to foundation systems in which the pile response may be
considered elastic and the soil is modeled as a linear viscoelastic medium. Dobry and Gazetas
developed a simple analytical solution for the dynamic stiffness and damping of floating pile
groups [13] subjected to harmonic loading at the pile head. The vibration of each pile causes

the generation of waves that propagate in all directions, causing additional displacements in
the adjacent piles. The vibrating pile is referred to as the active pile, while the adjacent pile is
referred to as the passive pile. Dobry and Gazetas introduced some simplifying assumptions
on the nature of the generated wave field and the consequent displacement of adjacent piles.

They used closed form expressions for the generated wave field (in frequency domain) and
assumed that the additional displacement caused in the passive pile is equal to this dynamic
displacement field. For piles under axial loading, the interaction factor is derived from the
asymptotic cylindrical wave expression [23] that was assumed for the generated wave field

around a vibrating pile:

d 0.5 −βωS −iωS

αv (S) ≈ ( ) exp( ) exp( ) (7)
2S Vs Vs

where αv (S) is the pile interaction factor, S is the pile spacing, d is the pile diameter, β is
the soil damping, ω is the angular frequency of the input loading, and Vs is the shear wave
velocity in the soil medium. For laterally loaded piles, the interaction factor is a function of

10
the angle θ between the line of the two piles and the direction of the horizontal force applied.
However, it is sufficient to compute the interaction factor for θ = 0◦ and 90◦ , and then use
the following expression to compute the interaction factor for any angle θ [26]:

αh (S, θ) ≈ αh (S, 0◦ ) cos2 θ + αh (S, 90◦ ) sin2 θ (8)

To compute the interaction factors for θ = 0◦ and 90◦ , it was assumed that the 90◦ pile

is affected only by S-waves which emanate from the active pile with a velocity Vs , while
the 0◦ pile is affected only by compression-extension waves that emanate from the active
pile and travel with a velocity VLa . VLa is the so called Lysmer’s analog velocity, VLa =
3.4Vs /[π(1 − ν)]. Thus, the following expressions were proposed:

d 0.5 −βωS −iωS

αh (S, 90◦ ) ≈ ( ) exp( ) exp( ) (9)
2S Vs Vs
d −βωS −iωS
αh (S, 0◦ ) ≈ ( )0.5 exp( ) exp( ) (10)
2S VLa VLa

The work by Dobry and Gazetas was carried forward by Gazetas and Makris [15, 17]. They

retained the expressions for the wave field generated by the vibration of the active pile.
However, they assumed that the adjacent pile modifies the arriving wave field (us (ω, z))
depending upon its relative flexural rigidity and the vertical fluctuations of the arriving
wave field. The additional displacements in the passive pile are approximately equal to

the arriving wave field for long flexible piles and smoothly varying us (ω, z). The additional
displacements are nearly zero for rigid piles and rapidly varying us (ω, z). However, in general,
the additional response will be something in between these two extremes. These additional
displacements may be computed by applying us (ω, z) as a support excitation to the passive

pile. This may be accomplished using the Winkler model [3, 17].
The use of this technique as suggested above is restricted to linear systems, where nonlinear

11
effects may be included indirectly in the form of equivalent viscous damping to account for
energy dissipation. However, the wave propagation based technique has been modified to
obtain time domain expressions for the wave field generated by the active pile. This allows
the use of time domain procedures in computing pile groups response, thereby allowing

nonlinear effects to be included directly using nonlinear material models [3, 4]. Although
the time domain procedures allow the solution of a larger class of problems, they still have
some limitations. The consideration of nonlinear effects is restricted to computing the single
pile response. The pile-to-pile interaction factors are based on Fourier transforms of the
closed form expressions obtained in frequency domain. Thus, the effect of soil nonlinearity

on pile-to-pile interaction is not considered.

3.2 Pile group analysis using p-y multipliers

Another class of pile group analysis techniques involves the use of p-y multipliers. These
procedures are an extension of the Winkler model technique used for single piles (Section 2).
In the Winkler model, the soil is replaced by discrete springs (Figure 1) and soil nonlinearity

is represented using a nonlinear load-deformation curve referred to as the p-y curve for lateral
load response. These p-y curves may be generated using analytical expressions for stiffness
and strength of the p-y curve along with standard hysteretic models such as the bilinear
and Bouc-Wen models, as suggested in Section 2. However, p-y curves most commonly are

obtained from back calculations based on experimental or 3D finite element response data
[12, 19, 21, 30, 31, 37]. The Winkler model for single piles may be extended for analysis of
pile groups. In such a model, each pile is modeled as a beam supported on horizontal springs
that account for soil response. The nonlinear response of each soil spring is represented by a

p-y curve. The p-y curves used in pile group analysis are obtained by modifying p-y curves
for single pile analysis using p-y multipliers. The effect of pile-to-pile interaction on soil p-y
curves has been studied by a number of researchers [9, 10, 11, 38]. This technique of pile

12
group analysis is widely used and is also incorporated into software programs. However, the
major drawback of this pile foundation model is that though it is computationally much
more efficient than 3D finite element models, it still requires extensive modeling and may
take considerable analysis time, particularly for large pile groups under dynamic loads. Thus,

the use of such models may not be very efficient for certain applications such as vulnerability
analysis.

13
4 Pile-to-Pile Interaction Analysis

Some of the currently available techniques of pile group analysis were reviewed in the previ-
ous section. Pile group analysis using pile interaction factors is common. These procedures
are applicable to elastic systems as they use the frequency domain. Although these proce-
dures were extended to the time domain in order to incorporate pile and soil nonlinearity
effects, they only consider nonlinearity effects in the computation of the single pile response.

Currently available techniques for computing pile interaction factors do not account for non-
linearity. Since soil and pile nonlinearity effects do have a significant effect on pile foundation
response, it is necessary to understand the effect of these nonlinearities on pile-to-pile inter-
action, particularly in groups with closely spaced piles. Pile-to-pile interaction studies were

conducted using 3-D finite element analysis. A single pile model was first calibrated against
experimental response data. The parameters used in this model were then used to perform
pile group analysis. Soil nonlinearity and pile-soil gapping effects were considered in the
analysis.

4.1 3D Finite Element Model for Pile Foundation Analysis

Two sets of experimental tests performed at the same site were used to calibrate the finite
element model. Lateral load tests were performed on a group of nine steel pipe piles and
also on an isolated single pile [10]. The location of the tests was an overconsolidated clay
site in Houston, TX. In another experimental program at the same site, dynamic lateral load

tests were conducted on a cantilevered mass supported by a single steel pipe pile embedded
in stiff, overconsolidated clay [6]. 3D models were created using the program ABAQUS [1].
Due to the symmetry of the problem, only half the mesh was generated (Figure 2). The
pile was assumed to respond elastically. The steel pipe pile was modeled as an equivalent
cylinder. Thus, the moment of inertia and the mass density were modified so that flexural

14
 Property Value
Pile properties
Pile diameter 0.273m
Pile length 13.1m
Flexural stiffness 1.34X104 kN − m2
Lineal pile density 60.3kg/m
Soil properties
Young’s modulus 100M P a
Density 2100kg/m3
Poisson’s ratio 0.49
Undrained shear strength 47kP a

Table 1: Pile and soil properties used in the 3D finite element model

stiffness (EI) and total mass of the cylinder were equal to the flexural stiffness and total
mass respectively of the steel pipe. The soil was assumed to be homogeneous. The model
was meshed using 3D brick (denoted C3D8 in ABAQUS) elements. A number of laboratory
and in-situ field tests were performed during both the mentioned experimental programs, to

determine soil properties. More details may be found in the cited references. However, the
tests showed considerable scatter in the results. The value of Young’s modulus for the soil
was back calculated so that the stiffness of the load deformation curve in an elastic analysis
matched the initial stiffness of the experimental curve. The obtained value was found to fall

within the range of experimentally determined values. A constant value of undrained shear
strength was assumed based on experimental evidence. Soil and pile properties used in the
model are listed in Table 1.
Checks were performed to verify the finite element mesh, assuming the soil to be linear
elastic. The soil-pile model was fixed at the base (and free on all sides) and subjected to a

lateral load at the ground level. The obtained response was compared with the deflection
of a cantilever beam subjected to a tip load, as obtained from beam theory. The difference
between the two results was about 8% (Table 2), which is acceptable since beam theory is not
exact in this case due to the presence of shear deformations. The second check performed,

15
Figure 2: 3D finite element mesh for single pile analysis

16
 Analysis procedure Applied load (kN) Pile head displacement (m) Error (%)
Beam theory 50 0.0144 0.00
Beam theory 300 0.0863 0.00
3D FEA 50 0.0155 7.73
3D FEA 300 0.0933 8.08

Table 2: 3D FE model check: verification of pile head response as a cantilever beam compared
with beam theory

Analysis procedure Pile head response (m)

3D model 0.00128
Analytical solution using Winkler model [29] 0.00204

Table 3: 3D FE model check: comparison of elastic response with analytical solution

compared the response of the 3D model under a static lateral pile head load of 100kN,

with solutions predicted by a beam-on-elastic-foundation (Winkler model) approach (Table

3), obtained analytically from the solution of the differential equation [29]. The observed
difference between the 3D model and Winkler model solution is due to the fact the Winkler
model assumes that the modulus of subgrade reaction provided by the soil is k = 1.2Es

(equation 1) while the 3D model explicitly considers the Young’s modulus of the soil.

4.1.1 Modeling of soil nonlinearity

Since the aim of this study is to understand the effect of nonlinearity on interaction factors,
nonlinear soil models were used in the 3D finite element analyses. Abaqus provides a large
number of plasticity models to incorporate soil nonlinearity in the analysis. These include
the Extended and Modified Drucker-Prager models, the Mohr-Coulomb plasticity model and

the Critical state (clay) plasticity model [1]. These are sophisticated plasticity models that
require calibration based on experimental data. In this study, the Mohr-Coulomb model is
used since it can be calibrated based on the standard Mohr-Coulomb parameters. It uses

17
 1 60

0.8 50

Shear stress, τ (kPa)

40
0.6

G/Gmax
30
0.4
20

0.2
10

0 −3 −2 −1 0 0
10 10 10 10 0 0.2 0.4 0.6 0.8 1
Strain, γ (%) Shear strain, γ (%)

(a) Modulus reduction curve (PI=30) [33] (b) Shear stress-strain curve

Figure 3: Nonlinear soil model for cohesive soil used in the 3D FEM

the classical Mohr-Coulomb yield criterion:

τ = c − σ tan φ (11)

where τ is the shear stress, σ is the normal stress (negative in compression), c and φ are the

Mohr-Coulomb parameters. The definition of material behavior using the Mohr-Coulomb

model in Abaqus includes the elastic properties (Young’s modulus and Poisson’s ratio),
density (for dynamic analysis), angle of friction, angle of dilation (which governs the flow
potential), and cohesive yield stress vs. plastic strain. Since the soil being modeled in the

presented set of analyses was cohesive, the angle of friction (φ) was assumed to be zero. The
dilation angle (ψ) was also assumed to be zero. The cohesive stress-strain response (Figure
3(b)) was derived from the modulus reduction curve for cohesive soils with plasticity index,
P I = 30 [33] (Figure 3(a)). The parameters eccentricity () and deviatoric eccentricity (e),
which determine the shape of the flow potential function, assumed default values:  = 1 and

e = (3 − sin φ)/(3 + sin φ).

18
4.1.2 Modeling of pile-soil interface

Modeling of the pile-soil interface is critical under strong loads that cause separation between
the pile and the soil. Abaqus provides a number of advanced models for contact behavior.

Surface-based contact, which allows modeling of contact between two deformable bodies that
undergo small or finite sliding, was used in the analyses presented here. Abaqus uses the
concept of contact pairs - a master surface and a slave surface. The pile surface was chosen as
the master and the surrounding soil surface was chosen as the slave. The mechanical contact

simulation of the interaction between two bodies includes a constitutive model for the contact
pressure-overclosure relationship (i.e. behavior normal to the contact surfaces) and a friction
model that defines the force resisting relative tangential motion of the surfaces. The most
common contact pressure-overclosure relationship is “hard” contact. When surfaces are in

contact, any contact pressure can be transmitted between them. The surfaces separate if the
contact pressure reduces to zero. Separated surfaces come into contact when the separation
between them reduces to zero. In addition to the pressure-overclosure relationship, frictional
behavior may or may not be included to model the shear stress transmitted across the
interface. Abaqus has a wide range of friction models to choose from. The classical isotropic

Coulomb friction model, which defines a friction coefficient relating shear stress to the contact
pressure, was used. A value of 0.7 was assumed for the coefficient of friction, as used by
Bentley and Naggar [5], based on the recommendations of the American Petroleum Institute
[2]. The 3D finite element model of the single pile was compared with the experimental

response with and without the friction model.

4.1.3 Response of single pile using 3D FE model

The experimental response of the single pile [10] under a lateral pile head load was compared
with the static response of the 3D finite element model (Figure 4) assuming linear elastic
pile response and nonlinear soil response using the Mohr-Coulomb model. Three different

19
 100

80

Pile head load (kN)

60

40

20 Experimental data
FEM (perfect soil−pile contact)
FEM (hard contact, no friction)
FEM (hard contact, friction)
0
0 0.01 0.02 0.03 0.04
Pile head displacement (m)

Figure 4: Comparison of single pile response using 3D FEM with experimental response

pile-soil contact models were considered: (i)perfect contact (ii)“hard” contact with friction
(iii)“hard” contact without friction. The analyses showed that the use of the Mohr-Coulomb
model for soil nonlinearity along with the “hard” contact model with no friction gives the

best match with experimental data. Therefore, it was decided to use this model in the pile
group analysis. The 1D FE model (or Winkler model) for the single pile was also compared
with the 3D model and experimental response. Figure 5 shows that by using the p-y material
for soft clay [19] to model the nonlinear spring in the Winkler model, it is possible to obtain

a good match with the experimental response. This model has been implemented in the
current version of Opensees [20]. However, using the bilinear and Bouc-Wen model for the
nonlinear springs does not compare quite as well with the experimental response.

4.2 Development of Improved Pile-to-Pile Interaction Factors

To study the effect of soil nonlinearity and soil-pile gapping on the pile interaction factors, 3D

finite element analyses were performed on two-pile groups. The Mohr-Coulomb model used
in the single pile analysis was used in the pile group model, along with the “hard” contact

20
 100

80

Pile head load (kN)

60

40

Experimental data
20 3D FE model
Winkler (soft clay p−y model)
Winkler (bilinear)
Winkler (Bouc−Wen)
0
0 0.01 0.02 0.03 0.04
Pile head displacement (m)

Figure 5: Comparison of single pile response using 1D and 3D FEM with experimental
response

(no friction) model used for pile-soil contact. The soil and pile properties are listed in Table

1. The factors affecting pile-to-pile interaction include the pile to soil relative stiffness,
departure angle and pile length to diameter ratio [26]. However, the most important factor
is the pile spacing. Two finite element models were created with pile spacing, S = 2d and
S = 5d, where d is the pile diameter. The finite element meshes for the pile groups are
shown in Figures 6 and 7. The pile groups were subjected to static lateral pile head loads

and the interaction factors at the pile head level were computed by comparing the pile group
response with the single pile response. The single pile model was similar to what was used
in section 4.1, with the only difference that the pile head was assumed to be fixed. This was
achieved by restraining the vertical degrees of freedom on the top surface of the single pile.

The pile-to-pile interaction factors were computed as:

∆uplgrp
αh = (12)
usglpl

21
Figure 6: 3D finite element mesh for pile group analysis (S = 2d)

22
Figure 7: 3D finite element mesh for pile group analysis (S = 5d)

23
 2

1.8

1.6

Interaction factor, αh
1.4

1.2 Elastic soil

Nonlinear soil w/o pile−soil gapping
1 Nonlinear soil with pile−soil gapping

0.8

0.6

0.4

0.2
0 50 100 150 200
Pile group load (kN)

Figure 8: Effect of soil nonlinearity and pile-soil gapping on interaction factors

where αh is the interaction factor, ∆uplgrp is the additional displacement caused in one
pile due to the loading on the other pile, and usglpl is the single pile displacement. Since
the system is nonlinear, αh is a function of the applied load. To observe the effect of pile

nonlinearity and the pile-soil contact model on the interaction factors, for the case of S = 2d,
αh was computed for three different cases: (i)assuming the soil to be linear elastic and soil-
pile contact to be perfect, (ii)modeling soil nonlinearity using the Mohr-Coulomb model with
no soil-pile gapping (iii)using the Mohr-Coulomb model for soil nonlinearity and the “hard”

contact (no friction) model for the pile-soil interface. The results of the analysis are shown
in Figure 8. Figure 9 shows the effect of pile spacing on the interaction factors. The figure
clearly shows that as the spacing increases, the interaction between piles decreases. If the
piles in a group are very closely spaced, then interaction between the piles is larger. This
implies that the induced additional displacements are also larger. Thus, as the pile spacing

is decreased, the pile group undergoes larger deformations for the same load level. At very
large spacings, there would be no interaction between piles, in which case, the pile group
would carry twice the load of a single pile. Figure 10 shows the effect of pile spacing on the

24
 2
S/d = 2
1.8 S/d = 5

1.6

Interaction factor, αh
1.4

1.2

0.8

0.6

0.4

0.2
0 50 100 150 200
Pile group load (kN)

Figure 9: Effect of pile spacing on interaction factors

200

180

160

140
Pile head load

120

100

80

60

40
2Xsingle pile response
20 Pile group response: S/d = 2
Pile group response: S/d = 5
0
0 0.005 0.01 0.015 0.02
Pile head displacement (m)

Figure 10: Lateral load response of a two-pile group

25
lateral load-deformation response of pile groups. As the pile spacing decreases, the efficiency
of each pile in the group decreases.
Further analyses need to be conducted to study the effect of other factors on the interaction
factors. These include the effect of departure angle, relative pile-to-soil stiffness, soil strength

and dynamic loading.

26
5 Conclusions

A number of techniques currently used to analyze pile foundations were reviewed. The
Beam on Winkler Foundation model, commonly used in the analysis of single piles, was
reviewed. Two major classes of pile group analysis techniques—the pile-to-pile interaction
factor method and the p-multiplier method were also reviewed. The p-multiplier method
while highly efficient (when compared to 3D finite element analysis) and widely used may

still be very time consuming for large pile-supported structures which have a number of large
pile groups. The pile-to-pile interaction factor method is based on modifying the response of
a single pile to account for interactions among different piles in a group. The available tech-
niques to evaluate these interactions are generally applicable to linear elastic systems and

do not incorporate the effect of soil and pile nonlinearity. 3D finite element analyses were
performed to compute pile-to-pile interaction factors accounting for soil nonlinearity and
pile-soil gapping effects. Interaction factors were computed for two-pile groups under lateral
static loading. The analyses showed that both nonlinearity in the soil and gapping between

the pile and the soil have a very significant effect on the interaction. The pile spacing is the
most important parameter influencing the interaction factors. The interaction factors were
computed for two different pile spacings. A number of other factors may have a significant
effect on the interaction—departure angle, relative pile-to-soil stiffness, soil strength, pile
nonlinearity, and load amplitude and frequency (under dynamic loading). A very extensive

set of finite element analyses need to be conducted to understand the full effect of all these
factors.

27
 Acknowledgment

This work was supported by the National Science Foundation through the Mid-America
Earthquake Center under award number EEC-9701785. This work was partially supported
by National Computational Science Alliance under BCS040006N and utilized the IBM P690
at the National Center for Supercomputing Applications.

28
References

[1] Abaqus Inc., Pawtucket, RI. ABAQUS Analysis User’s Manual, 2003.

[2] American Petroleum Institute, Washington, D.C. Recommended Practice for Planning,
Designing and Constructing Fixed Offshore Platforms. API Recommended Practice 2A
(RP 2A), 1991.

[3] D. Badoni and N. Makris. Nonlinear response of single piles under lateral inertial and

seismic loads. Soil Dynamics and Earthquake Engineering, 15:29–43, 1996.

[4] D. Badoni and N. Makris. Pile-to-pile interaction in the time domain – non-linear axial
group response under harmonic loading. Géotechnique, 47(2):299–317, 1997.

[5] K. J. Bentley and M. H. El Naggar. Numerical analysis of kinematic response of single

piles. Canadian Geotechnical Journal, 37:1368–1382, 2000.

[6] G. W. Blaney and M. W. O’Neill. Measured lateral response of mass on single pile in
clay. Journal of Geotechnical Engineering, ASCE, 112(4):443–457, 1986.

[7] B. B. Broms. Lateral resistance of piles in cohesionless soils. Journal of the Soil Me-

chanics and Foundations Division, ASCE, 90:123–156, 1964.

[8] B. B. Broms. Lateral resistance of piles in cohesive soils. Journal of the Soil Mechanics
and Foundations Division, ASCE, 90:27–63, 1964.

[9] D. A. Brown, C. Morrison, and L. C. Reese. Lateral load behavior of pile group in sand.

Journal of Geotechnical Engineering, ASCE, 114(11):1261–1276, 1988.

[10] D. A. Brown, L. C. Reese, and M. W. O’Neill. Cyclic lateral loading of a large -scale
pile group. Journal of Geotechnical Engineering, ASCE, 113(11):1326–1343, 1987.

29
[11] D. A. Brown and C.-F. Shie. Numerical experiments into group effects on the response
of piles to lateral loading. Computers and Geotechnics, 10:211–230, 1990.

[12] D. A. Brown and C.-F. Shie. Three dimensional finite element model of laterally loaded

piles. Computers and Geotechnics, 10:59–79, 1990.

[13] R. Dobry and G. Gazetas. Simple method for dynamic stiffness and damping of floating
pile groups. Géotechnique, 38(4):557–574, 1988.

[14] G. Gazetas and R. Dobry. Horizontal response of piles in layered soils. Journal of
Geotechnical Engineering, ASCE, 110(1):20–40, 1984.

[15] G. Gazetas and N. Makris. Dynamic pile-soil-pile interaction. Part I: Analysis of axial

vibration. Earthquake Engineering and Structural Dynamics, 20:115–132, 1991.

[16] B. K. Maheshwari, K. Z. Truman, M. H. El Naggar, and P. L. Gould. Three-dimensional

finite element nonlinear dynamic analysis of pile groups for lateral transient and seismic

excitations. Canadian Geotechnical Journal, 41:118–133, 2004.

[17] N. Makris and G. Gazetas. Dynamic pile-soil-pile interaction. Part II: Lateral and
seismic response. Earthquake Engineering and Structural Dynamics, 21:145–162, 1992.

[18] N. Makris and G. Gazetas. Displacement phase differences in a harmonically oscillating

pile. Géotechnique, 43(1):135–150, 1993.

[19] H. Matlock. Correlations for design of laterally loaded piles in soft clay. In Proc., Second
Offshore Technology Conference, pages 577–594, Houston, TX, 1970.

[20] S. Mazzoni, F. McKenna, M. H. Scott, G. L. Fenves, and B. Jeremić. Command Lan-

guage Manual: Open System for Earthquake Engineering Simulation (OPENSEES).

Pacific Earthquake Engineering Research Center. website: http://peer.berkeley.edu/ sil-

via/OpenSees/manual/doc/OpenSeesManual.doc.

30
[21] M. C. McVay and L. Niraula. Development of p-y curves for large diameter piles/drilled
shafts in limestone for FBPIER. Technical report, Florida DOT, 2004.

[22] P. J. Meymand. Shaking table scale model tests of nonlinear soil-pile-superstructure

interaction in soft clay. PhD thesis, University of California, Berkeley, 1998.

[23] P. M. Morse and K. U. Ingard. Theoretical Acoustics. McGraw-Hill, New York, 1968.

[24] G. Mylonakis and G. Gazetas. Kinematic pile response to vertical P-wave seismic excita-
tion. Journal of Geotechnical and Geoenvironmental Engineering, ASCE, 128(10):860–

867, 2002.

[25] H. G. Poulos. Behavior of laterally loaded piles: I – single piles. Journal of the Soil
Mechanics and Foundations Division, ASCE, 97(SM5):711–731, 1971.

[26] H. G. Poulos. Behavior of laterally loaded piles: II – pile groups. Journal of the Soil
Mechanics and Foundations Division, ASCE, 97(SM5):733–751, 1971.

[27] H. G. Poulos and E. H. Davis. Pile Foundation Analysis and Design. John Wiley and

Sons, New York, 1980.

[28] M. F. Randolph and G. T. Houlsby. The limiting pressure on a circular pile loaded
laterally in cohesive soil. Géotechnique, 34(4):613–623, 1984.

[29] L. C. Reese. Behavior of piles and pile groups under lateral load. Technical report,
FHWA, 1983.

[30] L. C. Reese, W. R. Cox, and F. D. Koop. Analysis of laterally loaded piles in sand. In
Proc., Fifth Offshore Technology Conference, pages 473–483, Houston, TX, 1974.

[31] L. C. Reese, W. R. Cox, and F. D. Koop. Field testing and analysis of laterally loaded

piles in stiff clay. In Proc., Seventh Offshore Technology Conference, pages 672–690,
Houston, TX, 1975.

31
[32] A. Trochanis, J. Bielak, and P. Christiano. Simplified model for analysis of one or two
piles. Journal of Geotechnical Engineering, ASCE, 117(3):448–466, 1991.

[33] M. Vucetic and R. Dobry. Effect of soil plasticity on cyclic response. Journal of Geotech-
nical Engineering, ASCE, 117(1):89–107, 1991.

[34] S. Wang, B. L. Kutter, M. J. Chacko, D. W. Wilson, R. W. Boulanger, and A. Abghari.

Nonlinear seismic soil-pile structure interaction. Earthquake Spectra, 14(2):377–396,
1998.

[35] Y.-K. Wen. Method for random vibration of hysteretic systems. Journal of the Engi-

neering Mechanics Division, ASCE, 102(EM2):249–263, 1976.

[36] G. Wu and W. D. Liam Finn. Dynamic nonlinear analysis of pile foundations using
finite element method in the time domain. Canadian Geotechnical Journal, 34:44–52,
1997.

[37] Z. Yang and B. Jeremić. Numerical analysis of pile behavior under lateral loads in
layered elastic-plastic soils. International Journal for Numerical and Analytical Methods
in Geomechanics, 26:1385–1406, 2002.

[38] Z. Yang and B. Jeremić. Numerical study of group effects for pile groups in sands.
International Journal for Numerical and Analytical Methods in Geomechanics, 27:1255–

1276, 2003.

32