1

JSCE Journal of Earthquake Engineering

Impedance-Based Winkler Spring Method for Soil-Pile group
Interaction Analysis


Hossein TAHGHIGHI
1
, Kazuo KONAGAI
2


1
PhD Candidate, Institute of Industrial Science, University of Tokyo
Tokyo 153-8505, J apan, [email protected]
2
Professor, Institute of Industrial Science, University of Tokyo
Tokyo 153-8505, J apan, [email protected]


A simple numerical model is presented for handling both linear and nonlinear soil-pile interactions. In the
present method, the diagonal terms of the exact soil impedance matrix define the elastic characteristics of
Winkler side soil springs. When its off-diagonal terms are described as a function of the piles active length,
they were found less dependent on the other secondary factors. Despite the simplification, the proposed
nonlinear soil model can reproduce piles behaviors computed by more rigorous finite element methods.


Key words: Soil-Pile group Interaction, Winkler spring, Active pile length, Nonlinear FEM analysis



1. Introduction

Evaluation of soil-pile-structure interaction is
necessary for a rational seismic design of civil-
infrastructures. Such studies can be done
experimentally, theoretically or numerically.
Various tools have been developed for the analysis
of soil-foundation-structure interaction.
As for piles, the static group effect was put on a
rational basis, relying on continuum mechanics, by
Poulos et al. (1971). For describing dynamic pile
group effect, the thin layer element method, TLEM,
was first developed by Tajimi and Shimomura
(1976) etc. Boundary element solutions for soil-pile
systems were formulated by Kaynia et al. (1982)
and Banerjee et al. (1987). Though rigorous, these
methods are only for analyzing elastic behaviors of
soils and structures. With the rapid development of
computer technology, a variety of straightforward
methods are available for solving problems of
increasing complexity. They include Finite Element
Methods (FEM) developed by Kimura et al. (2000),
Wakai et al. (1999) and Yang et al. (2003). Direct
methods, however, require both soils and structures
to be treated with equal rigor, and complex
variations of soil profile in a 3D expanse should be
provided for the analysis. Hence, there yet remains
an important place for simple approaches even in
these days of highly manipulative numerical
solutions to difficult problems. A simple model
developed by Nogami and Konagai (1992)
hypothesizes that a plane strain slice of side soil
determines the side soil stiffness. With this
approach, one can cut a distinct side soil spring in
half, one for the near field, and the other for the far
field that is not strongly affected by the presence of
the pile (See Fig. 1). However, a problem for this
model is that the impedance function for the elastic
side soil converges on zero for static loading.
Moreover, its dynamic stiffness near the ground
surface is overestimated because the effect of stress
free surface is completely ignored assuming a plane
strain condition. However, by virtue of its simplicity,
there remains a strong need for a Winkler model.
In the present method, the diagonal terms of the
exact soil impedance matrix define the elastic side-
soil springs of Winkler type. The effect of the off-
diagonal terms of the matrix is represented by an
2

Fig. 1. Schematic view of the Winkler side soil spring

additional displacement vector, which is applied to
the other ends of the Winkler side-soil springs. The
displacement vector is introduced herein as a
function of the piles active length, which was
found to be less dependent on the other secondary
factors.


2. Modified Winkler Model

Recently, the second author developed a simplified
approach in which a group of piles is viewed as an
equivalent single upright beam [Konagai et. al.
2000], the idea based on the fact that a group of
piles often trap soil among them as observed when
pulled out [Railway Technology Research Institute
1995]. The upright single beam is a composite of
p
n piles and the soil caught among them (Figure 2).
The broken line in Figure 2a circumscribing the
outermost piles in the group determines its cross
section
G
A . The soil-pile composite together with
its exterior soil is divided into
L
n horizontal slices
as shown in Figure 2b. The idea has been verified in
both linear and nonlinear soil-pile interaction
analyses [Konagai et. al. 2002, Konagai et. al. 2003].
When the pile group in Fig. 2 is subjected to
lateral displacements along its depth, the equation
of equilibrium for soil-pile system is written as:
{ }
{ } { } { }
{ }
F 0
ext pile pile soil pile free
+ + =




K u K u u
(1)
where, vector { }
ext
F denotes the external load on
the pile cap from the superstructure and { }
free
u is
the free field ground motions. From the Eq. (1),
lateral soil reaction forces on the pile group is
written in the following form as:



{ } { } { }
{ }
pile soil pile free
=

p K u u (2)
Extracting diagonal terms
j
k ) , 2 , 1 ( n j = of the
side soil stiffness matrix
[ ]
soil
K , Equation (2) is
rewritten as:
{ } { }
{ }
T
k u k u k u
n 1 2 pile pile,1 pile,2 pile,n
(3)
off diag
=

p
p

with
{ }
{ } { }
Off diagonal termsof
off diag
(4)
soil pile soil free
=

p
K u K u

Equation (3) yields the following expression as:
{ } { } { }
{ }
k 0 0
1
0 k 0
2
(5)
pile pile far
0 0 k
n
L
=






p u u

In which;
{ }
{ }
off diag
far
i
p
k

=



u (6)
Equation (5) indicates that a Winkler model can
describe the soil-pile interaction. Differing from the
conventional Winkler model, Eq. (5) shows the
necessity of subtracting a displacement
vector { }
far
u from
{ }
pile
u . The displacement
vector { }
far
u is interpreted as the displacements
given on the other ends of the Winkler side-soil
springs, and therefore, will be referred to as the
far-end displacements. The side soil stiffness
Fig. 2. Assumption for single beamanalogy
(a) Soil-grouped pile system
(b) Sliced elements
3
matrix in equation (4) together
with
{ }
pile
u determines the far-end
displacements { }
far
u . The pile-group stiffness
matrix includes two stiffness parameters,
sway
EI
and
rock
EI for the grouped piles, and shear modulus
of soil [Konagai et. al. 2000]. The first parameter
sway
EI governs the sway motion of the beam is the
product of the bending stiffness of an individual
pile EI
single
and the number of piles
p
n , and
therefore
sway
EI and are crucial parameters for
determining the far-end displacement vector{ }
far
u .
When a pile group is laterally displaced, the
horizontal deflection of the pile group decreases
with increasing depth. In practice, most laterally
loaded piles are indeed flexible in the sense that
they are not deformed over their entire lengthL .
Instead, pile deflections become negligible below an
active length (or effective length)
a
L . The active
length
a
L depends largely on these parameters
sway
EI and . The above consideration leads to an
idea that the far-field displacement vector { }
far
u
will be expressed uniquely in terms of the active
pile length
a
L .

3. Far-end soil displacement{ }
far
u

To verify the present idea for the far-end soil
displacements, { }
far
u , some representative cases
were examined. The soil medium was assumed to
be a horizontally stratified infinite deposit with
material damping of the frequency-independent
hysteretic type. Necessary parameters for the
examined cases are listed in Table 1. The pile
groups examined included 2 2, 3 3 and 4 4
piles (N4, N9 and N16) with the spacediameter
ratio s/d set at 2, 3 and 5.

Young Modulus
E(GPa)
Poisson ratio

Unit weight
3
(KN/ m )
Friction angle
(deg)
Cohesion
C(GPa)
Length
L(m)
Diameter
d(m)
Pile 50 --------- 24.5 -------- ------- 15 1.0
Soil (type 1) 0.05 0.4 17.2 35 0 ------- --------
Soil (type 3) Eq. (7) 0.4 17.2 35 0 ------- --------

Table 1. Material properties for pile and soil

0 0.1 0.2 0.3 0.4 0.5
0
0.5
1
1.5
a
d
=d/v
s
R
e
a
l

[
L
a

/

L
a

(

=

0
)
]
Homogen Soil , S/d=3
N4
N9
N16

0 0.1 0.2 0.3 0.4 0.5
0
0.5
1
1.5
a
d
=d/v
s
R
e
a
l

[
L
a

/

L
a

(

=

0
)

]
Inhomogen Soil , S/d=3
N4
N9
N16

(a) (b)
Fig. 3. Active pile length distribution against frequency factor,
d
a , for different Pile-group configuration with S/d=3
located in Homogeneous (a) and Inhomogeneous (b) soil
Following the definitions taken by [Konagai et. al.
2003], the point where 3% of the pile head
deflection is reached determined the active length.
Fig. 3 describes the variations with increasing
frequency of the dynamic active pile length
a
L
normalized by its static value. Frequency, , is
4
normalized with the single pile diameter, d, and the
soil deposit shear wave velocity,
s
v .
Fig. 3 depicts that the curves are less sensitive to the
change in number of piles,
p
n . Fig. 4 shows
variations of far-end displacements with respect to
the depth, where I
f
is the far-end displacement
far
u normalized by its value at the ground surface
level, ) 0 ( = z u
far
. Figure 5 combines all the results
for the examined soil pile-group cases. This
confirms that the far end displacement { }
far
u can
be uniquely described as a function of the
normalized depth
a
L z / for a monotonic loading at
the pile cap.
Figure 6 elucidates far-end displacements for the
dynamic loading, the upper three and the lower
three for the homogeneous and heterogeneous soil
profiles, respectively. As the non-dimensional
frequency increases, the displacements at deeper
locations became larger. This tendency is clearer for
narrower pile spaces. Curves for narrower pile
spaces were plotted on one figure frame (Fig. 7).
Frequency here was normalized by
a
L and
s
v in
such a way that the non-dimensional frequency
L
a
indicates the ratio between the active pile length and
the shear wave length in soils. This figure thus
provides a perspective on the limitation of the
proposed idea for describing the dynamic far-end
displacement distribution with depth. However for
the most important range of frequency ( 2 <
L
a ), far
field soil displacement
far
u can be practically
expressed as a unique function of the active pile
length
a
L .
Fig. 7-c and 7-d describe imaginary part of the far-
end displacement. Though with increase in non-
dimensional frequency, the imaginary part at deeper
locations became bigger but their values are almost
small for the upper locations close to the ground
surface. Since the main concern in seismic pile
design goes into its upper part, therefore it is
possible to ignore the effect of imaginary part in the
far-end displacement with an enough approximation.


4. Nonlinear soil-pile interaction analysis

Taking the advantage of the modified Winkler
model, a pushover analysis of a pile foundation
subjected to lateral loading was conducted, and the
result was compared with a rigorous solution
obtained from an object-oriented OpenSees Finite
Element Platform.
Material parameters of the pile and the soil are
given in Table 2. The top 2.4m of the 13.7m long
aluminum pile with a 0.429m * 0.429m square
cross-section was assumed to stick out above the
ground surface (see Fig. 8).
The elastic modulus E of the medium dense sand
was assumed to vary with confining pressure as:
p/ n E E ( p )
0 a
=



(7)
where;
'
p / 3 [(1 2k ) (z)]/ 3
ii 0 v
= = + (8)
0
E is the elastic modulus of soil under the
atmospheric pressure (
0
E =17.4 GPa,
a
p =98 kPa),
) ( ' z
v
is an effective overburden stress at the depth
z,
0
k is the coefficient of earth pressure at rest
estimated by a typical empirical equation
sin 1
0
= k with as the internal friction angle,
and n is constant for a given void ratio, which was
set at 2.0.
The pile model consists of 16 beam-column
elements. Its bottom end was assumed to be fixed
upright, while two boundary conditions were
discussed for the top end of the pile. Elasto-plastic
features of the side soil springs were assumed. The
linear stiffness of each spring was identical to the
corresponding diagonal component of the rigorous
soil stiffness matrix described in the previous
section, while its ultimate strength was given by:
Ka
Pu (Kp Ka) z (1 )Kp z Kp z
p
Kp
= = =
(9)
where, is the unit weight of soil, Ka and Kp are
the active and passive earth pressure coefficients,
is a modification factor to represent three
dimensional effects in the effective interaction zone
in subsoil layers . Based on numerical results of
various cases, 4
p
= was chosen (Shirato, M.
2004).
The effect of off-diagonal terms was taken into
account just by adding the far-end soil
displacements as has been mentioned above.
5


0 1 2
0
0.5
1
1.5
I
f
HS - S/d=2
0 1 2
0
0.5
1
1.5
HS - S/d=3
0 1 2
0
0.5
1
1.5
HS - S/d=5
0 1 2
0
0.5
1
1.5
Z/L
a
I
f
IS - S/d=2
0 1 2
0
0.5
1
1.5
Z/L
a
IS - S/d=3
0 1 2
0
0.5
1
1.5
Z/L
a
IS - S/d=5
N4
N9
N16
N4
N9
N16
N4
N9
N16
N4
N9
N16
N4
N9
N16
N4
N9
N16

Fig. 4. Off-diagonal effect in terms of normalized depth for Homogenous soil, HS, and Inhomogeneous soil, IS, in Static state for
different spacing ration among piles in the group



0 0.5 1 1.5 2
0
0.5
1
1.5
Piles in Group=[4, 9, 16] , S/d =[2, 3, 5]
Z/L
a
I
f

=

R
e
a
l

[

U
f
(
z
)
/
U
f
(
z

=

0
)

]

Fig. 5. All Off-diagonal representative effects together in terms
of normalized depth For Homogenous and Inhomogeneous soil
in Static state


5. Finite element model

A complete 3D soil-pile model was created in
OpenSees finite element framework. Taking
account of its symmetric geometry, only half of the
soil-pile system was realized (Fig. 9). The sand was
assumed to follow the simple Drucker-Prager
criterion with non-associated flow rule, and the soil
strength was pressure-dependent. A self-weight
analysis, in which, both the soil and the pile mass
are subject to static gravitational loading, was thus
necessary to realize the initial stress condition in the
model. Then a push over analysis followed. In this
analysis, lateral load applied to the pile head was
increased stepwise.
Excluding the symmetric boundary, where only the
Y components of displacement were confined, three
sides and the bottom of the soil medium were fixed.
The interface between the aluminum pile and the
surrounding soil was expressed by covering up the
pile with thin layers of frictional brick elements. All
the interface elements were assumed to follow the
Drucker-Prager criterion with a friction angle of 25
degrees. Its dilation angle was set at zero.
The cross-section of the square pile was divided
into four elements.
Fig. 10 compares the results from the present
Winkler model and TLEM. A very good agreement
in the initial elastic parts verifies the present
approach. Fig. 11 shows p-y curves obtained from
three different numerical models, which include, in
addition to those aforementioned, the result from
the OpenSees Py-Simple-1 model based on back
bone curve proposed by the American Petroleum
Institute (API), for sand materials (Boulanger RW

6
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
I
f

=

R
e
a
l

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
HS , S/d=2
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
HS , S/d=3
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
HS , S/d=5
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
Z/L
a
I
f

=

R
e
a
l

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
IS , S/d=2
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
Z/L
a
IS , S/d=3
0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
Z/L
a
IS , S/d=5
=0(c/s)
=10
=20
=30

Fig. 6. Far-field representative effect in terms of normalized depth for Homogenous soil, HS, and Inhomogeneous soil, IS, in several
frequencies and different spacing ration for a 3 by 3 pile-group

0 0.5 1 1.5 2 2.5 3
-1
-0.5
0
0.5
1
1.5
Z/L
a
I
f

=

R
e
a
l

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
Homogeneous Soil - Dynamic case
Piles in Group =[4, 9, 16]
Spacing Ratio,S/d, =[2, 3]
a
L
=L
a
/v
s
=0
a
L
=L
a
/v
s
=1
a
L
=L
a
/v
s
=2
a
L
=L
a
/v
s
=3

0 0.5 1 1.5 2 2.5 3
-1
-0.5
0
0.5
1
1.5
Z/L
a
I
f

=

R
e
a
l

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
Inhomogeneous Soil - Dynamic case
Piles in Group =[4, 9, 16]
Spacing Ratio,S/d, =[2, 3]
a
L
=L
a
/v
s
=0
a
L
=L
a
/v
s
=1
a
L
=L
a
/v
s
=2
a
L
=L
a
/v
s
=3

(a) (b)

0 0.5 1 1.5 2 2.5
-0.6
-0.4
-0.2
0
0.2
0.4
Z/L
a
I
f

=

I
m
a
g

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
Homogeneous Soil - Dynamic case
Piles in Group =[4, 9, 16]
Spacing Ratio,S/d, =[2, 3]
a
L
=L
a
/v
s
=0
a
L
=L
a
/v
s
=1
a
L
=L
a
/v
s
=2
a
L
=L
a
/v
s
=3

0 0.5 1 1.5 2 2.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Z/L
a
I
f

=

I
m
a
g

[

U
f
(
z
)

/

U
f
(

z
=

0
)

]
Inhomogeneous Soil - Dynamic case
Piles in Group =[4, 9, 16]
Spacing Ratio,S/d, =[2, 3]
a
L
=L
a
/v
s
=0
a
L
=L
a
/v
s
=1
a
L
=L
a
/v
s
=2
a
L
=L
a
/v
s
=3

(c) (d)

Fig. 7. Influence of frequency factor,
L
a , on far-field representative parameter in terms of normalized depth for Homogeneous (a) and
Inhomogeneous (b) soil
7


Fig. 8. Layout of Single Pile (Zhang et. al.1999)



Fig. 9. Three dimensional FE mesh



et. al. 1999). A good correlation shows the potential
of the proposed Modified Winkler Model for non-
linear soil-pile group interaction analysis.


6. Conclusion

A new perspective for the soil-pile interaction
analysis was provided in a way that the classical
continuum mechanics theory yields a Winkler type
expression of side soil stiffness. With the advantage
of distinct expression of side soil stiffness at a
particular depth, one can easily incorporate the
effect of nonlinear soil behavior in the vicinity of a
0 0.02 0.04 0.06 0.08
0
500
1000
1500
2000
Fixed head pile
Pile Cap Displ.(m)
F
o
r
c
e

(
K
N
)
NLinear P-Y Model
Rigourous TLEM approach (Elastic)

Fig. 10. Force-Displacement distribution of pile without
free length on the ground


0 0.05 0.1 0.15 0.2
0
100
200
300
400
500
Free head pile
Pile Cap Displ.(m)
F
o
r
c
e

(
K
N
)
NLinear P-Y Model
OpenSees PySimple1(API Sand)
OpenSees 3D FE Model

Fig. 11. Force-Displacement distribution of pile with
free length on the ground


pile. With the present approach, obtained results
showed good agreements with rigorous 3D

Young modulus
E (GPa)
Poisson
ration,
Unit weight
3
(KN/ m )
Friction angle
(deg)
Pile 69 0.33 26.5 ------
Soil Eq. (7) 0.35 14.5 37.1

Table 2. Material properties for pile and soil
(McVay et al. 1998)

Lateral load
Uniformsand
2.4 m
11.3 m
13.7 m
3.0 m
0.429 m
8
solutions in both linear and nonlinear pushover
analyses. The method still needs to be verified in
cyclic loading cases, where gap creation among
piles will affect the overall behaviors of soil-pile
systems. The discussion will appear in future
publications.


Acknowledgement

This work was primarily supported by the Ministry
of Science, Research and Technology, MSRT, of
the Iranian government under the Award of
complete PhD scholarship for study on abroad.

References:
1. American Petroleum Institute (API) Recommended Practice
for Planning, Designing, and Constructing Fixed Offshore
Platforms, 1993, Washington, D.C.: API Recommended
Practice 2A (RP2A).

2. Banerjee, P. K. and Sen, R. Dynamic Behavior of Axially
and Laterally Loaded Piles and Piles Group, 1987,
Developments in Soil Mech. Found. Eng., Vol.3: 95-133.

3. Boulanger RW, Curras CJ , Kutter BL, Wilson DW, Abghari
A. Seismic Soil-Pile-Structure Interaction Experiments and
Analyses, J ournal of Geotechnical & Geoenvironmental
Engineering, ASCE, 125(9), 1999: 750-759.

4. J eremic B. and Yang Z. Template Elastic Plastic
Computations in Geomechanics, Int. J . for Numerical and
Analytical Methods in Geomechanics, 2001, 01:1-6.

5. Kaynia A, Kausel E. Dynamic Stiffness and Seismic
Response of Pile Groups, NSF report, NSF/CEE-82023, 1982.

6. Kimura M, Zhang F. Seismic Evaluations of Pile
Foundations with Three Different Methods Based on Three-
Dimensional Elasto-Plastic Finite Element Analysis, Soils and
Foundations, 2000, 40(5): 113-132.

7. Konagai K. and Ahsan R. Simulation of nonlinear SSI on a
shaking table J ournal of Earthquake Engineering, Vol. 6, No.
1(2002) 31-51.

8. Konagai K, Ahsan R, Maruyama D. Simple expression of
the dynamic stiffness of grouped piles in sway motion, J .
Earthquake Eng. 2000; 4(3): 35576.

9. Konagai K., Yin Y., Murono Y. Single beam analogy for
describing soil-pile group interaction Soil Dynamic and
Earthquake Eng. 2003, 23: 1-9.

10. Limin Zhang, M. M., and Lai, P. Numerical analysis of
laterally loaded 3x3 to 7x3 pile groups in sands, J ournal of
Geotechnical and Geoenvironmental Engineering, 1999, 125,
11: 936-946.

11. McVay, M., Zhang, L., Molnit, T., and Lai, P. Centrifuge
testing of large laterally loaded pile groups in sands, J ournal of
Geotechnical and Geoenvironmental Engineering, 1998, 124,
10: 1016-1026.

12. Nogami T., Otani J , Konagai K, Chen HL Nonlinear Soil-
Pile Interaction Model for Dynamic Lateral Motion, J ournal of
Geotechnical Engineering, ASCE, 118(1), 1992: 89-106.

13. OpenSees: Open System for Earthquake Engineering
Simulation, 2004, Nov.30, (http://opensees.berkeley.edu/)

14. Poulos H. G. Behavior of laterally loaded piles, part two
J ournal of the Soil Mechanics and Foundations Division, ASCE,
97(SM5), 1971: 733-751.

15. Railway Technology Research Institute Report of Field
Tests of Prototype Pile Foundations, Tokyo, J apan, 1995, In
J apanese.

16. Shirato, M. Computational seismic performance
assessment of a pile foundation subjected to a sever
earthquake, PhD dissertation, Department of Civil Engineering,
The University of Tokyo, 2004.

17. Tajimi H, Shimomura Y. Dynamic Analysis of Soil-
Structure Interaction by the Thin Layered Element Method (in
J apanese), 1976, Transactions of the Architectural Institute of
J apan, Vol. 243: 41-51.

18. Wakai A, Gose S, Ugai K 3-D Elasto-Plastic Finite
Element Analyses of Pile Foundations Subjected to Lateral
Loading, Soils and Foundations, 1999, 39 (1): 97-11.

19. Yang Z., J eremic B. Numerical study of group effects for
pile groups in sands Int. J . for Numerical and Analytical
Methods in Geomechanics, 2003, 27:12551276.