Computers and Geotechnics 59 (2014) 112126

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Proposed nonlinear 3-D analytical method for piled raft foundations

Sangseom Jeong, Jaeyeon Cho
Department of Civil Engineering, Yonsei University, Seoul 120-749, Republic of Korea

a r t i c l e

i n f o

Article history:
Received 1 February 2013
Received in revised form 5 February 2014
Accepted 25 February 2014

Keywords:
Piled raft
Soilstructure interaction
Numerical analysis
Field measurement
Load transfer approach

a b s t r a c t
The load distribution and deformation of piled raft foundations subjected to axial and lateral loads were
investigated by a numerical analysis and eld case studies. Special attention is given to the improved analytical method (YSPR) proposed by considering raft exibility and soil nonlinearity. A load transfer
approach using py, tz and qz curves is used for the analysis of piles. An analytical method of the
soilstructure interaction is developed by taking into account the soil spring coupling effects based on
the Filonenko-Borodich model. The proposed method has been veried by comparing the results with
other numerical methods and eld case studies on piled raft. Through comparative studies, it is found
that the proposed method in the present study is in good agreement with general trend observed by eld
measurements and, thus, represents a signicant improvement in the prediction of piled raft load sharing
and settlement behavior.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
In recent years, a number of huge construction projects, such as
high-rise buildings and long span bridges, are being undertaken.
The piled raft foundations are especially being recognized as an
economical foundation system for high-rise buildings. Here, piles
as settlement reducers have been discussed for over a quarter of
a century [2] and some signicant applications have been reported
[12,38,42]. Optimized design strategy is a major importance for an
economic construction to be achieved. An optimized design of a
piled raft can therefore be dened as a design with minimum costs
for the installation of the foundation and satisfactory bearing
behavior for a given geometry and raft loading [35]. The piled raft
is a composite foundation system consisting of three bearing elements: raft, piles and subsoil. Therefore, the behavior of a piled raft
is affected by the 3D interaction between the soil, piles and raft,
thus, a simple and convenient analytical method is needed to evaluate these interactions.
Much work has been done to study load sharing and settlement
behavior of piled raft by many researchers. Numerical methods
have been developed widely in the last two decades because
numerical methods are less costly and may be used to consider
many kinds of different soil and foundation geometries compared
to eld and model tests. Although these methods make slightly
different modeling techniques, they can generally be classied into
three groups: (1) simplied calculation methods [30,32], (2)
Corresponding author.
E-mail address: jaeyeon82@hotmail.com (J. Cho).
http://dx.doi.org/10.1016/j.compgeo.2014.02.009
0266-352X/ 2014 Elsevier Ltd. All rights reserved.

approximate computer-based methods [5,9,14,15,37] and (3) more

rigorous computer-based methods [12,17,18,45,48].
The rst type of method is based on the linear elastic analysis of
piled raft subjected to axial loading. Generally, the simplied
calculation methods are most commonly used procedure for the
preliminary design of a piled raft foundation. However, it is noted
that these analytical methods are limited to elastic problem.
Because this calculation procedure is developed for rigid raft and
is assumed that the soil is perfectly elastic. Thus, it may not represent the nonlinear behavior of actual piled raft in the eld: it does
not take into account the actual behavior of nite exible raft and
pilesoil interaction, etc.
The second type of method has been used to investigate the
piled raft system, which is analyzed as a continuous elastic
medium using nite element formulation. In these methods, the
research by Poulos [29], Clancy and Randolph [5], Poulos [30]
and Russo [37] also have some disadvantages. It did not predict
the membrane behavior of raft because the raft is generally modeled as plate element. Therefore, the raft used in these methods
may not reect the displacement due to membrane action of large
size raft foundations for high-rise buildings. In addition, most of
the previous research is related to piled rafts subjected to vertical
loading and only semi-innite homogeneous single soil layer was
considered. The consideration of various loading condition and soil
layer will be more realistic in design practice.
The third type of method is based on the three-dimensional
nite-element or nite-difference techniques. Poulos [31] noted
that the most feasible method of analysis was the threedimensional linear/nonlinear FE method. However, a rigorous

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

113

Fig. 1. Flat-shell element.

Fig. 2. Modeling of pile element.

numerical approach of the piled raft system is computationally

expensive and requires extensive training because of the threedimensional and nonlinear nature of the problem. Therefore, a nite element analysis is more suitable for obtaining benchmark
solutions against which to compare simpler analysis methods, or
for obtaining solutions of a detailed analysis for the nal design
of a foundation, rather than as a preliminary routine design tool
[15].
In this study, an improved analytical method (YSPR) for the design of piled raft has been proposed to overcome some limitations
of the existing methods. It is intermediate in complexity and theoretical accuracy between the second and third type of method. In
the present method, a numerical technique is used to combine
the soil and pile head stiffness with the stiffness of the raft. In order

to examine the validity of the proposed method, the analysis results are compared with the available solutions from previous researches. In the eld case study, comparative analyses between
YSPR and a eld measurement data are carried out for the pile load
and settlement behavior.

2. Method of analysis
2.1. Modeling of exible raft
Finite element techniques have often been used for the analysis
of raft by different researchers such as Clancy and Randolph [5],
Zhang and Small [49], Kitiyodom and Matsumoto [14]. According

114

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Fig. 3. Soilstructure interactions in piled raft foundation [13].

Fig. 4. Interactions between raft, piles, and subsoil in present method.

to the former methods [5,49], the raft can be treated as a plate and
the soil can be treated as a series of interactive springs by using a
Mindlins solutions [22], in which the contact pressure at any point
on the base of the raft is proportional to the deformation of the soil
at that point or as an elastic half-space in which the behavior of the
soil can be obtained from a number of closed-form solutions. In the
later method, the raft is modeled as thin plates and the piles as
elastic beams and the soil is treated as interactive springs [14].
The interactions between structural members are made by the
use of Mindlins solutions. The primary limitation of these methods
is that the membrane behavior of the exible raft cannot be considered because the nodal displacements (in the x- and y-direction)
for the membrane action are not included. This limitation can be
overcome by using a at-shell element. An improved four-node
at-shell element proposed by the authors [48], which combines
a Mindlins plate element and a membrane element with torsional
degrees of freedom, is adopted in this study. The at-shell element
can be subjected to the membrane and bending actions that are
shown in Fig. 1. The displacement due to the membrane action is
considered independent of the displacement due to the bending

action, therefore it can be considered separately. For the bending

action, the displacement eld for an individual element can be described in terms of the vertical nodal displacement and the rotations about the x and y axes. For the membrane action, the
displacement eld can be described in terms of the nodal displacements in the x and y directions.
2.2. Modeling of single and pile groups
In this study, piles are treated as beam-column elements. The
behavior of soil surrounding the individual piles is represented
by loadtransfer curves (tz, qz, and py curves), and the interaction between piles is represented by p-multiplier (fm) and group
efciency factor (Ge). The loaddeformation relationship of individual pile heads may be derived by a single pile analysis based on
beam-column method. In this method, a pile member is described
as a series of beam column elements with discrete springs to represent the soil support condition as shown in Fig. 2. The governing
differential equations for the axially loaded and laterally loaded
pile can be expressed as:

115

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

2

Axially loaded pile : EA

Load increment

Iteration

dz

 Cbz w 0

(k t) 2

Laterally loaded pile : EI

P2

(k s) 2
1

P1

(kt) i : tangential slope

(k t) 1

P1

(ks) i : secant slope

0

u1 u1

P2

d w

u2

u2

(a)
(k i) j : i = load increment
j = Iteration number

j=1 : tangential stiffness

j>1 : secant stiffness

Fu
(k i) 1
f((u i) i)
Fu=f(u )

(ki) j

d y
4

dz

d y
2

dz

q  Ksy 0

where EA, EI are the axial stiffness and the exural rigidity the pile,
w is the vertical deection of the pile at point z, bz is the stiffness/
circumference for the axial reaction represented by the modulus
of the soil-response (tz or qz or both), which depends on the
depth z and pile movement w, and C is circumference of the pile
at point z. Q is the axial load on the pile, q is the distributed load
along the length of the pile, and KS is the stiffness for the lateral soil
reaction represented by the modulus of the soil-response (py)
curve.
In the next step, nite difference technique is used to solve the
differential equations governing the compatibility between the pile
displacement and the load transfer along a pile. These techniques
are generally based on load tests on full-scale and parametric nite
element analyses of pilesoil interactions, which are represented
by loadtransfer curves (tz, qz, and py curves).

f((u) i- 1)

2.3. Soilstructure interaction

uj

(ui) j

(u) i- 1

(b)
Fig. 5. Increment secant modulus method [48]. (a) Concept of increment secant
modulus method. (b) Estimating stiffness at ith load increment.

The load-bearing behavior of a piled raft is characterized by

complex soilstructure interaction between the piles, raft and the
subsoil, as shown in Fig. 3 [13]. The present method makes use
of pilesoilpile and raftsoilpile interaction to simulate the real
piled raftsoil response under lateral and vertical loadings. Additionally, for the raftsoilraft interaction, this study uses a semiempirical parameters proposed by many researcher [7,39,40] as
the modulus of soil reaction below the raft. The use of these
parameters as assumed in the derivation procedure, may be a limitation. However, these interactions are incorporated in a calculation procedure that is computationally very efcient.

Fig. 6. Modeling of piled raft (YSPR).

116

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Fig. 7. Flow chart of YSPR.

Piles in such groups interact with one another through the surrounding soil, resulting in the pilesoilpile interactions. In this
study, a set of nonlinear py curves which can be modied by
reducing all of the p-values on each curve by a p-multiplier (fm)
are used as input to study the behavior of the laterally loaded piles.
The p-multiplier can be calculated for each pile in the group
[3,6,19]. For each pile i in the group, the p-multiplier can be expressed as:

fmi b1i b2i b3i    bji

where bji is the p-reduction factor due to the effect of pile j on pile i.
In a group of closely-spaced piles, the axial capacity of group is
also dominated by variation in settlement behavior of individual
piles due to pilesoilpile interaction. The most reliable data

concerning the efciency of the piles in a group is derived by many

researchers [11,21,41]. In this study, loadtransfer curves in side
resistance (tz curve) and in end bearing resistance (qw curve)
which can be modied by reducing all of the t- and q-values on
each curve by a group efciency factor (Ge) are used as input to
study the behavior of the vertically loaded piles.
In classical solution, the Winkler model [46] is used for analyzing raft foundation. However, the Winkler model could not predict
accurately the displacement of some solids, e.g. soil. The Winkler
model ignores the important interaction existing between adjacent
points in the soil continuum. In other words, the soil springs are
considered as isolated foundation elements. In order to overcome
a limitation, much work has been performed to propose some
improved or rened models [8,10,27,43]. For the raftsoilpile

117

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

present method proposed an improved raftsoilpile system by

connecting the top ends of soil springs and pile elements with an
elastic at-shell element including membrane action. By using
at-shell element, a realistic representation of the subgrade reaction can be established directly in terms of coupled soil resistance
in which the response at any point on the interface affects other
points. The authors believe that a combination of the soil spring
and the elastic at-shell element may be used to overcome the
restrictions associated with conventional methods, and thereby
also used to analyze appropriately axially loaded piled raft, in soil
deposits. Consequently, the proposed analytical method should be
based on the concept of soilstructure interaction under the lateral
and vertical loadings.
2.4. Global stiffness matrix
The stiffness matrix of a at-shell element (Kat-shell=raft) in local
coordinate system was constructed through combining separately
the stiffness matrix of a plate element (Kplate) and that of a membrane element (Kmembrane) as followings:

K flat-shell

K plate
0

K membrane

The stiffness matrix of a plate element Kplate is represented in

the following form:

K plate

Z
V

BTb Db Bb dV

Z
V

BTs Ds Bs dV

where Bb is the bending strain matrix and Bs is the shear strain matrix. For an isotropic material, Db and Ds are given as follows:

Db

Fig. 8. Schematic diagram of vertical and lateral loaded piled raft. (a) Pile
conguration. (b) Section-view.

interaction, in this study a membrane-spring system originally

proposed by Filonenko-Borodich [8] was incorporated to involve
the soil spring-coupling effects. This system can provide a mechanical interaction between the individual soil spring and pile
elements by using the at-shell element. As shown in Fig. 4, the

Ds

0
0

Et
6
7
4m 1
5
121  m2
0 0 1  m=2


WEt

1 0

21 m 0 1


;

6a

5
6

6b

where E is Youngs modulus, m is Poissons ratio, and t is constant

thickness of the plate. On the other hand, the stiffness matrix of a
membrane element Kmembrane is represented in the following form:

K membrane

Z
v

Bm GRT  C  Bm GRdV

Fig. 9. Soil spring constant for linear analysis of a single pile.

cV

hh

7a

118

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

IwV
0.02

0.03

0.04

CsH

0.05

-0.1

0.2

Depth from G.L.(m)

PRAB
FEM (K&M, 2003)
PLAXIS 3D
YSPR

PRAB
FEM (K&M, 2003)
PLAXIS 3D
YSPR

10

10

(a)

(c)

IuH
-0.1

CbH
0.1

0.2

-0.4

Depth from G.L.(m)

Depth from G.L.(m)

0.1

Depth from G.L.(m)

PRAB
FEM (K&M, 2003)
PLAXIS 3D
YSPR

10

-0.2

0.2

PRAB
FEM (K&M, 2003)
PLAXIS 3D
YSPR

10

(b)

(d)

Fig. 10. Comparison of analysis result for piled raft: (a) Settlement and (b) lateral displacement, (c) shear force; and (d) bending moment.

g dV;
bg b

7b

E
21 m

for pile groups can be formed by sum of n single pile stiffness matrix (Eq. (10)).

2
7c

where C is the constitutive modulus, c is taken as the shear modulus. Bm, G, R are the strain matrices representing the relationship between the displacements (the membrane displacement, the
rotation, and midside incompatible displacement respectively)
 and g are also the strain matrices for the
and the strains. b, g, b,
innitesimal rotation elds.
The pile head stiffness (K11  K66) is assumed to be constant
within each load increment and each iteration and then superposition can be applied in order to develop a pile head stiffness matrix
(Eq. (8)) in individual piles. Using loaddisplacement relationships
representing pile behaviors according to pile head movements
[34], the relationship between the nodal force and nodal displacements can be expressed in Eq. (9). In addition, the stiffness matrix

K pile

K 11

6 0
6
6
6 0
6
6 0
6
6
4 K 51
0

K 15

K 22

K 24

K 33

K 42

K 44

K 55

Kpilei fdgi fF i g

K pilegroups

n
X
K pilei 

0 7
7
7
0 7
7
0 7
7
7
0 5
K 66

10

i1

where [K]pile(i) is an individual pile head stiffness matrix, {di} a displacement or rotation, and {Fi} force or moment at the ith pile head.

119

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

22 mm
22 mm

Fh
(variable)

2.5D and 5.0D

2.5D, 5.0D,
and 7.5D

Fh
(variable)

Pile cap

Pile cap

15 mm

15 mm

600 mm

Sand

600 mm

Sand

Pile

Pile

Rock

Rock

(a)

(b)

Fig. 11. Test pile group congurations [4] (a) 2  2 pile groups (b) 3  3 pile groups.

A component (K11  K66) of pile head stiffness matrix is changed at

each load increment and iteration stage.
The soil support at various nodes of raft foundation is simulated
by a series of equivalent and independent springs in three directions (x, y and z directions). The spring behavior can be linear or
nonlinear. In linear case, soil behavior is dened by soil stiffness
(K11  K33) which is assumed to be constant within each load
increment and each iteration. The soil reactions at any point can
be expressed as

2
6
6
6
6
6
6
6
6
4

8
9
38 9
du >
Fu >
>
>
>
>
>
>
>
>
>
>
> dv >
>
>
>
>
Fv >
07
7>
>
>
>
>
>
>
>
>
>
>
>
7>
<
<
=
7
dw
Fw =
0
7

> Mu >
>
07
>
>
>
7>
> au >
>
>
>
>
>
>
7>
>
>
>
>
>
>
>
a
M
05 >
>
>
>
>
v
v
> ;
>
>
>
:
:
;
Mw i
0 i aw i

k11

0 0 0

k22

0 0

k33

0 0

0 0

0 0

0 0

Ksoili fdgi fF i g

11

12

where [K]soil(i) = individual soil stiffness matrix, {di} = displacement

or rotation, and {Fi} = force of soil at point i. In nonlinear case, spring
behavior is dened by giving pairs of loadrelative displacement
values. At this point, soil stiffness is calculated by nonlinear solution
procedure.
Finally, the stiffness matrix of a piled raft can be dened by the
combination of the foundation system and the supporting soil.
Therefore, the stiffness matrix formulations of a piled raft system
can be written as the following:

K piled

raft 

K raft  K soil  K pilegroups 

13

2.5. Nonlinear solution procedure

To consider the nonlinear loaddisplacement relationship at
each pile head and soil (below the raft), an incremental secant
modulus method developed by Won et al. [48] is used. When this
incremental secant modulus method is used, the displacement u2
corresponding to load P2 is increased to u02 as shown in Fig. 5(a), so
that point (P2, u02 ) will be located on the curve and consequently
the displacement will be close to the exact solutions.

The procedure for nonlinear solution in this study includes the

following step. In total, 10 (ten) loaddisplacement curves (axial 1;
lateral 8; torsional 1) are estimated per each pile head. Fig. 5(b)
shows the estimation method of stiffness at an ith load increment.
In this method, external forces are rst divided by N (number of
load increment). The stiffness at ith load increment and jth iteration is represented (ki)j. In each load increment, tangential slope
is adopted at rst iteration (j = 1) and the secant modulus at j > 1
for the stiffness of pile head, which is expressed as Eqs. (14) and
(15), respectively.

ki j



df u
du uui1

ki j

f ui j  f ui1
ui j  ui1

ui j ui1 Duj

j 1

j > 1

14

15

16

where (u)i1 is an accumulated nal displacement at a previous

load increment and (ui)j is an accumulated displacement at the ith
load increment and jth iteration.
At each load increment, displacements (Duj) are calculated
through structural analysis and then accumulated displacements
(ui)j are estimated using Eq. (16). If the convergence criteria,
DujDuj1 < e is satised, the accumulated nal displacements
(u)i are calculated and continue to the next load increment. This
process iterates until the load increment number reaches N. In
the structure analyses, the tangential slope (df(u)/du) and load
(f(u)) of individual piles are estimated using cubic spline method
[1]. The procedure described above is iterated until the error
between the assumed and calculated displacements falls within a
tolerance limit.
As a nal outcome, an improved numerical method (YSPR) was
proposed to analyze the response of a raft and a piled raft considering raft exibility and soil nonlinearity (Fig. 6). Fig. 7 shows the
ow chart of present method.

120

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

3. Verication of proposed method with previous studies

0.04

Lateral load (kN)

3.1. Kitiyodom and Matsumoto [14]

0.03

0.02

measured (2.5D)
measured (5.0D)
measured (7.5D)
predicted (2.5D)
predicted (5.0D)
predicted (7.5D)

0.01

0
0

0.004

0.008

0.012

0.016

Displacement (m)

(a)

Lateral load (kN)

0.08

0.06

0.04

0.02

measured (2.5D)
measured (5.0D)
predicted (2.5D)
predicted (5.0D)

0
0

0.004

0.008

0.012

0.016

0.02

A series of linear piled raft analyses were performed to verify

the present method by comparison with other numerical methods
which have been used in the preliminary design of piled raft. A
schematic diagram of a 2  2 piled raft is shown in Fig. 8. This
structure consists of a raft, and four identical vertical piles, which
are spaced by 1.5 m (=3.75D, where D is the pile diameter). The
piles have an embedded length of 10 m, a diameter of 0.4 m. Pile
head conditions are xed. A square raft of size 3  3 m with a
thickness of 0.9 m is rested on a homogeneous soil. The Youngs
modulus and Poissons ratio of the soil are 12,500 MPa and 0.3.
The raft and piles, with a Youngs modulus and Poissons ratio of
125,000 MPa and 0.3 respectively, is subjected to a vertical and
lateral load. Fig. 9 shows the spring constants were used for the
linear soil condition. The same axial spring constants were used
along the pile depth, with a constant value of 7,527,867 kN/m2,
which includes the pile perimeter. The end-bearing spring was
8,692,180 kN/m2, and the tension part was neglected. The constants of the horizontal springs were increase from 0 to
4,682,274 kN/m2 along the pile depth. Since the soil is assumed
to be an elastic model, the p-reduction and group efciency factor
of unity were used [6,18,41].
The response of piled raft is presented in settlement, lateral displacement of pile, and in shear force and bending moment distribution at various depths. Fig. 10(ad) shows representative
results from the proposed method. In addition, these results were
tested by comparing them with well-known three existing numerical methods: the PRAB [15]; the nite element method performed
by Kitiyodom and Matsumoto and PLAXIS 3D [28]. The results are
shown in terms of dimensionless parameters of IwV for the settlement, IuH for the lateral displacement of a pile respectively, CsH,
CbH for the shear force, and the bending moment along the pile
respectively. These parameters can be calculated by Eqs. (17)(20).

IwV

Es Dw
qz Br Lr

17

IuH

Es Du
qx B r L r

18

C sH

S
qx Br Lr

19

C bH

B
qx DBr Lr

20

Displacement (m)

(b)
Fig. 12. Lateral loaddisplacement curves at pile head. (a) 2  2 pile groups. (b)
3  3 pile groups.

Table 1
Material parameters used for this study (case studies).
Case

Material properties
Type

Japan case [15]

Pile
Raft
Soil

Germany case [34]

Pile
Raft
Soil

Korea case

Pile
Raft
Soil

Depth (m)
Steel pipe
Concrete
Sandy silt
Silty clay
Concrete
Concrete
Sand
Frankfurt clay
Concrete
Concrete
Gneiss

E (MPa)

0 to 5.5
2.1E08
0 to 2.2
30,000
0 to 1.7
13
1.7 to 13.5
15
5.5 to 25.5
23,500
3 to 5.5
34,000
3 to 8
75
8 to 113
47a
0 to 30
28,000
0 to 6.0
33,234
Soil spring stiffness (kPa/m)
0 to 204,250

c (kN/m3)

/ ()

c (kPa)

Modela

0.2
0.2
0.3
0.3
0.2
0.2
0.25
0.15
0.2
0.15

75
25
18
18
25
25
18
19

0
0

32.5
20

25
29.64

0
20

L.E.
L.E.
M.C.
M.C.
L.E.
L.E.
M.C.
M.C.
YSPR

Note: M.C. is Mohr Coulomb elasto-plastic model, L.E. is linear elastic model used in PLAXIS 3D Foundation Frankfurt clay: E = 45 + [tanh((z  30)/15) + 1]  0.7z.

121

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Fig. 13. Field test of piled raft [16]. (a) Plan-view and (b) section-view.

Table 2
Properties used for estimating load transfer curves (Japan case).

Load (MN)
0

py curves [19,20]

Subgrade reaction
modulus

Sandy silt

Silty
clay

Ultimate skin friction, s

(kPa)
Initial shear modulus, Gi
(kPa)
Poissons ratio, m
Ultimate bearing
capacity, Qf (kN)

40

40

5000

5769

0.3

0.3
250

Undrained shear strength

(kPa)
Unit weights (kN/m3)
py modulus, k (kN/m3)

25

29.64

18.0
27,150

18.0
27,150

Kx, Ky (kN/m3)
Kz (kN/m3)

27,150
5291

1.5

2.5

10

Settlement (mm)

tz, qz curves [39]

Contents

0.5

where w, u are the settlement and lateral displacement at the pile

head, qz and qx are uniform vertical and lateral load, the breadth,
Br and length, Lr, S and B are the shear force and the bending moment along the pile.
The calculated results of the proposed analysis method closely
approach the computed data from the other numerical methods.
It should be noted that the present method provides a very satisfactory prediction of the shear force and the bending moment in individual piles, when the exibility of the raft is considered by using
the combination of the membrane and bending actions. Although
a reasonably good agreement between the proposed and the existing methods was obtained, the proposed method has a larger settlement those of the existing methods at the same load. Conclusively,
it is thought that YSPR can be used with some condence in the preliminary design of axially and laterally loaded piled raft.
3.2. Chung and Jeong [4]
In this section, the verication of lateral response of the present
method against laboratory load test is discussed. By Chung and

20

30

40

Measured (K&I, 1967)

Calculated (R & E, 2006)
YSPR
PLAXIS 3D

50
Fig. 14. Computed and measured response of piled raft settlement.

Jeong [4], a series of small scale model tests were carried out to
study the behavior of pile groups subjected to lateral loadings on
sand. The test soil used in this study was: the unit weight
15.3 kN/m3, cohesion 0 kN/m2 and drained friction angle 37. The
model piles made from PVC tubes were 0.6 m in embedded length,
22 mm in diameter and 2.5 mm wall thickness and 28,265 kN m2
exural rigidity(EI). Fig. 11 shows an idealization of the subsurface
prole and pile embedment for test piles.
Using present method the behavior of pile groups are predicted
with different group congurations and different center-to-center
pile spacing: 2.5D, 5.0D, and 7.0D. Back-tted hyperbolic py
curves that are calculated at 5, 10, and 20 cm along the pile depth
in model test of single pile are implemented. Initial tangent

Table 3
Calculated stiffness of single pile and piled raft (Japan case).

Single pile
Piled raft (w/o Ge)
Piled raft (w/Ge)

K11 (kN/m)

K22 (kN/m)

K33 (kN/m)

K44 (kN/rad)

K55 (kN/rad)

K66 (kN/rad)

0.4052E+02
0.2735E+05
0.2735E+05

0.4052E+02
0.2735E+05
0.2735E+05

0.3877E+05
0.3453E+06
0.2492E+06

0.3434E+03
0.2730E+06
0.2208E+06

0.3434E+03
0.2730E+06
0.2208E+06

0
0
0

122

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Fig. 16. Pile load: (a) pile 1, 2, 3 and (b) pile 4, 5, 6.

Fig. 15. Torhaus Der Messe: (a) prole view and (b) conguration of pile.

Table 4
Properties used for estimating load transfer curves (Germany case).

tz, qz curves [39]

py curves [24,33]

Subgrade reaction
modulus

Contents

Quaternary
silt

Frankfurt
clay

Ultimate skin friction, sf

(kPa)
Initial shear modulus, Gi
(kPa)
Poissons ratio, m
Ultimate bearing
capacity, Qf (kN)
Internal friction angle ()
Unit weights (kN/m3)
py modulus, k (kN/m3)
Kx, Ky (kN/m3)
Kz (kN/m3)

143

91.6

30,000

20,434

0.25

0.15
90

32.5
18
16,300
16,300
294,000

20
19
136,000
136,000

stiffnesses (Ks) of the py curves at the depths of 0.05, 0.1, and

0.2 m are 11, 14.3, and 50 kN/m2, respectively. Also ultimate capacities (Pu) of the py curves at the same depths are 0.0011, 0.0033,
and 0.0033 kN/m, respectively.
To consider the detailed group effect, p-multipliers calculated
from the Chungs experiment are implemented. For the 2  2
group, p-multipliers are 0.86 for lead row and 0.45 for trail row
at 2.5D pile spacing; 0.95 for lead row and 0.67 for trail row at
5.0D; 1.0, 0.83 for lead, trail row at 7.5D. For the 3  3 group,
p-multipliers are 0.8, 0.3 and 0.4 for lead, middle, and trail rows
at 2.5D pile spacing; 0.93, 0.48, and 0.6 at 5.0D pile spacing.
Fig. 12 shows the predicted and observed lateral loadsettlement curves. The analysis of pile groups was performed for a xed
head condition and spacing-to-diameter ratios varying from 2.5 to
7.5. The present method considering pilesoilpile interaction relatively well predicts the general trend of the measured lateral
loads for the pile groups studied if the measured deections are
relatively small (say less than 15 mm).

Table 5
Calculated stiffness of single pile and piled raft (Germany case).

Single pile
Piled raft (w/o Ge)
Piled raft (w/Ge)

K11 (kN/m)

K22 (kN/m)

K33 (kN/m)

K44 (kN/rad)

K55 (kN/rad)

K66 (kN/rad)

0.3979E+03
0.1118E+08
0.1117E+08

0.3979E+03
0.1138E+08
0.1137E+08

0.3020E+06
0.1300E+08
0.1242E+08

0.4482E+05
0.2583E+09
0.2548E+09

0.4482E+05
0.2115E+09
0.2078E+09

0
0
0

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Load (MN)
0

50

100

150

200

250

Settlement (mm)

0
20

PLAXIS 3D
YSPR (w/o Ge)
YSPR (w/ Ge)

40

Measured
settlement

60
80
100

123

predicted the general trend of the measured values reasonably

well. However, the calculated results by Roberto and Enrico [36]
have a relatively smaller settlement as the applied load increased
than the results of the proposed solution. This clearly demonstrates
that for analysis result, YSPR gives more exible results for nonlinear behavior of soil, because the Roberto and Enrico [36] use soil
exibility matrix(based on linear elastic analysis of pile groups)
for soilpile interaction and the proposed method does so using
nonlinear load transfer curves and solution algorithm. These discrepancies between predicted and measured behavior at the high
load levels are because the assumptions of raftsoil relative stiffness and group efciency factor are inuenced on the settlement
behavior of piled raft. In addition, computational time to run this
case saves 57 min of computer time, and is about 20 times faster
than the 3D FE analysis.
4.2. Germany case

120
Smax = 124mm

140
Fig. 17. Settlement behavior of large piled raft foundation.

4. Comparison with other case histories

The validity of the proposed method was examined by comparing the results from the present approach with some of the eldmeasured results. The pile and soil properties employed with the
YSPR and PLAXIS 3D analyses for the case histories were the same
properties mentioned in their research. In the eld, the soil stiffness signicantly depends on the stress level, indicating that the
stiffness generally increases with depth. To account for the increase of the stiffness with the depth, the Youngs modulus of soil
(Eincrement) value which is the increment of stiffness per unit of
depth was used in FE analyses. Table 1 summarizes the material
properties used in the case studies.

4.1. Japan case

The settlement behavior of axially loaded piled raft reported by
Koizumi and Ito [16] are compared with the predicted values of the
proposed method. This test site was located near the 1-chome, Otemachi in Tokyo. A fully instrumented piled raft was installed in the
clay soil, which consists of sandy silt with gravel and organic silty
clay. Fig. 13 shows the subsurface prole and pile congurations of
the test piled raft. All of the test piles are 300 mm in dia. and 5.5 m
in length. The soil and material properties were determined by
back-analysis of eld load test results using PLAXIS 3D Foundation.
From full-scale tests in clay soil presented by ONeill [23] and Whitaker [47], the group efciency factor, Ge, was set at 0.7 for the
reduction of side resistance (tz curve) and end bearing resistance
(qw curve) of piles. The input parameter of soil used to generate
the load transfer curve and soil-spring are summarized in Table 2.
Table 3 shows the estimated stiffness of single pile and piled
raft when a vertical load of unity is applied. Compared to the stiffness in which the group efciency factor was 1.0, the stiffness of
piled raft showed a signicant decrease in K33 of about 28%. This
is because the decrease of the pile resistance due to the pile
soilpile interaction (i.e. group efciency factor), change the global
stiffness of piled raft.
The proposed analysis method (YSPR) and a nite element
program analysis (PLAXIS 3D) results were compared with the
measured loadsettlement curves in Fig. 14. All the methods

The settlement and load sharing behavior of instrumented,

large, piled raft installed in stiff clay was compared with the predicted values of the proposed and the FE analyses. Constructed between 1983 and 1986, the 130 m high Torhaus was the rst
building in Germany with a foundation designed as a piled raft. A
total number of 84 bored piles with a length of 20 m and diameter
of 0.9 m are located under two 17.5  24.5 m large rafts. The bottom of the 2.5 m thick raft lies just 3 m below ground level
(Fig. 15(a)). The subsoil comprises quaternary sand and gravel up
to 2.5 m below the bottom of the rafts, followed by the Frankfurt
clay [34]. And a schematic diagram of 7  6 piled raft structure is
shown in Fig. 15(b). The maximum load of Peff = 200 MN for each
raft [37] minus the weight of the raft is successively applied by
means of a uniform load over the whole raft area. In the present
method (YSPR), the soil around individual pile is modeled with
nonlinear load transfer curves. The axial load transfer curves (tz,
qz curves) are estimated using the equation developed by Wang
and Reese [44], the lateral load transfer curve (py curve) is used
as an API model [25,32]. The group efciency factor, Ge, was set
at 0.73 for the average value of pile spacing: 3D  4D [23,47].
The input parameter of soil used to generate the load transfer curve
and soil-spring are summarized in Table 4.
Table 5 summarizes the calculated stiffness for the single pile
and the piled raft foundation. A decrease in the group efciency
factor from 1.0 to 0.73 results in about 4.5% decrease in stiffness
of piled raft. It is also noted that the stiffness of piles inside the
group varied with a group effect. Fig. 16(ab) shows a comparison
of the measured and calculated pile loads. The prediction of the
present method is much more conservative than that of 3D FE
analyses and the measured one. However the proposed method
is in good agreement with general trend of pile load which increase
from a center pile (pile1) to the edge (piles 2, 4 and 6) and to the
corner pile (piles 3 and 5). The computed results for the center,
side, and corner piles show that the load distribution of the individual piles in a group is highly inuenced by the exibility of
the raft. This nding was similar to what Won et al. [48] discussed
about correlation between the pile member force and the exibility of pile cap for a pile groups.
Fig. 17 shows a settlement behavior of the piled raft. The measured maximum settlement is about 124 mm, the calculated settlements using YSPR and PLAXIS 3D are 106.7 (with Ge; 111.5)
mm and 117 mm respectively. This curve demonstrates the effect
of pilesoilpile interaction by considering group efciency. The
proposed method with an interaction factor is more appropriate
and realistic to represent a pilesoilpile interaction for closelyspaced piles than on that of no-interaction analysis. In Both values
of YSPR and 3D FE analyses are smaller than the measured one.
However, these two numerical methods provide an acceptable

124

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

Fig. 18. Preliminary design case of large piled raft: (a) plan view and (b) prole view.

design prediction. Despite the approximate assumptions involved

(i.e., loading condition, construction process, consolidation of clay),
the present method when used in nonlinear analysis is useful for
predicting the settlement behavior of a piled raft foundation taking
account of soil nonlinearity, the exibility of the large raft, and the
pile arrangement. The time taken for the computer to run this case
saves 115 min of computer time, and is about 24 times faster than
the 3D FE analysis. For large problems this computational saving
can be very signicant.
4.3. Korea case
As shown in Fig. 18, preliminary design case of a piled raft (OO
super tower) conducted at high-rise building construction sites in
Korea were representatively selected for the design application.
The construction site is comprised mainly of normally banded
gneiss, brecciated gneiss and fault core zones. Based on the results
of pressure meter, Goodman Jack and plate load tests carried out in
the eld, a nonlinear elastic modulus design line is established to

represent the stiffness of the ground. A schematic diagram of a raft

foundation with piles is shown in Fig. 18(b). This structure consists
of a raft, and 112 of ground strengthen piles. The piles have an
embedded length of 30 m, a diameter of 1.0 m. A large raft size
71.7  71.7 m with a thickness of 6.0 m is resting on a banded
gneiss. The raft and ground strengthen piles, with a Youngs modulus of 30 GPa and 28 GPa respectively, is subjected to a vertical
load (Ptotal = 6,701 MN).
Fig. 19(ad) shows the raft settlement at different sections predicted by GSRaft [26] and YSPR. Agreement between the GSRaft
and YSPR of settlement is generally good; however there is a slight
difference in prediction of settlement in the faulting zone where
the sudden drop of the magnitudes were occurred. This can be
attributed to the inappropriate assumption of material properties
due to no accurate ground investigation data on this section. The
calculated raft settlement has some difference between the
proposed method and the existing solution, based on the same
analysis conditions. This is because the conceptual methodology
of the present method is completely different from that of general

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

125

Fig. 19. Raft settlement distribution: (a) section 1, (b) section 2, (c) section 3; (d) section 4.

structural models. The raft is modeled as a grillage and the piles are
treated as bar element with axial stiffness only in GSRaft while
YSPR is adopted at-shell element and 6  6 pile head stiffness.
Although there are no measured proles of raft settlement, the
proposed analysis method showed reasonably good correspondence with well-known in-house program.
5. Conclusions
The primary objective of this study was to propose an improved
analytical method for a pile raft foundations. The conceptual methodology of the proposed method is completely different from that
of general continuum method. A series of analytical studies were
conducted. Through comparisons with case histories, it is clearly
demonstrated that the proposed method was found to be in good
agreement with measurement data. From the ndings of this
study, the following conclusions can be drawn:
1. By taking into account the raft exibility and soil nonlinearity,
the proposed analytical method is an appropriate and realistic

representation of the settlement and load sharing behavior of

piled raft foundation. It provides results that are in good agreement with the eld measurement and numerical analyses.
2. Proposed analytical method produces a considerably larger settlement of piled raft than the results obtained by the linear elastic analysis. Additionally, the analytical method is intermediate
in theoretical accuracy between general three-dimensional FE
analysis and the linear elastic numerical method. The settlement of piled raft obtained by the present method is similar
to that obtained by the PLAXIS 3D, while it shows smaller values than those obtained by existing method based on linear
elastic analysis of pile groups.
3. From the example case histories, the proposed method is shown
to be capable of predicting the behavior of a large piled raft.
Nonlinear loadtransfer curve and at-shell element can overcome the limitations of existing numerical methods, to some
extent, by considering the realistic nonlinear behavior of soil
and membrane action of exible raft.
4. Additionally, the comparative studies demonstrated that the
present method, when used in analysis of large scale piled raft,

126

S. Jeong, J. Cho / Computers and Geotechnics 59 (2014) 112126

is useful for computational saving and improving performance

in engineering practice.

Acknowledgements
This work was supported by the National Research Foundation
of Korea (NRF) grant funded by the Korea government (MSIP) (No.
2011-0030040).
References
[1] Allen MB, Isaacson EL. Numerical analysis for applied science. John Wiley &
Sons; 1998.
[2] Burland JB, Broms BB, De Mello VFB. Behaviour of foundations and structures.
In: State-of-the-Art Rep., Proc., IX Int. conf. of soil mechanics and foundation
engineering (ICSMFE). Rotterdam, The Netherlands: Balkema; 1977. p.
495546.
[3] Brown DA, Reese LC, ONeill MW. Cyclic lateral loading of a large-scale pile
group. J Geotech Eng, ASCE 1987;113(11):132643.
[4] Chung SH, Jeong SS. Analysis of pile groups considering pile-cap interaction.
M.S. thesis. Yonsei Univ; 2001.
[5] Clancy P, Randolph MF. An approximate analysis procedure for piled raft
foundations. Int J Numer Anal Meth Geomech 1993;17(12):84969.
[6] Cox WR, Dixon DA, Murphy BS. Lateral load test of 25.4 mm diameter piles in
very soft clay in side-by-side and in-line groups. Laterally loaded deep
foundations: analysis and performance, ASTM, SPT835; 1984.
[7] Daloglu AT, Vallabhan CVG. Values of k for slab on Winkler foundation. J
Geotech Geoenviron Eng, ASCE 2000;126(5):46371.
[8] Filonenko-Borodich M. Some approximate theories of the elastic foundation.
Uchenyie Zapiski Moskovskogo Gosudarstvennoho Universiteta Mekhanica,
vol. 46; 1940. p. 318.
[9] Hain SJ, Lee IK. The analysis of exible raftpile systems. Geotechnique
1978;28(1):6583.
[10] Hetenyi M. Beams on elastic foundations. Ann Arbor, Michigan: University of
Michigan Press; 1946.
[11] Jeong SS, Kim SI, Briaud JL. Analysis of downdrag on pile groups by nite
element method. Comput Geotech 1997;21(2):14361.
[12] Katzenbach R, Arslan U, Gutwald J, Holzhauser J, Quick H. Soilstructure
interaction of the 300-m-high Commerzbank Tower in Frankfurt am Main.
Measurements and numerical studies. In: Proc, 14th ICSMFE, vol. 2; 1997. p.
10814.
[13] Katzenbach R, Arslan U, Moormann C. Piled raft foundations projects in
Germany, design applications of raft foundations. In: Hemsley JA, editor,
Thomas Telford; 2000. p. 32392.
[14] Kitiyodom P, Matsumoto T. A simplied analysis method for piled raft and pile
group foundations with batter piles. Int J Numer Anal Meth Geomech
2002;26:134969.
[15] Kitiyodom P, Matsumoto T. A simplied analysis method for piled raft
foundations in non-homogeneous soils. Int J Numer Anal Meth Geomech
2003;27:85109.
[16] Koizumi Y, Ito K. Field tests with regard to pile driving and bearing capacity of
piled foundations. Soils Found 1967;7(3):3053.
[17] Lee IK. Analysis and performance of raft and raftpile foundations in a
homogeneous soil. In: Proceedings of 3rd international conference on case
history in geotechnical engineering, St Louis (also Research Report R133,
ADFA, University of New South Wales, Australia); 1993.
[18] Lee JH, Kim YH, Jeong SS. Three-dimensional analysis of bearing behavior of
piled raft on soft clay. Comput Geotech 2010;37:10314.
[19] Lieng JT. Behavior of laterally loaded piles in sand-large scale model test. Ph.D.
thesis, Department of civil engineering, Norwegian institute of technology;
1988.

[20] Matlock H. Correlation for design of laterally loaded piles in soft clay. In: Proc.
Offshore technology conference, OTC 1204; 1970.
[21] Meyerhof GG. Ultimate bearing capacity of footing on sand layer overlaying
clay. CGJ 1974;11(2):2239.
[22] Mindlin RD. Force at a point in the interior of a semi-in-nite solid. Physics
1936;7:195202.
[23] ONeill MW. Group action in offshore piles. In: Proc specialty conference on
geotechnical engineering in offshore practice. ASCE; 1984.
[24] ONeill MW, Dunnavant TW. A study of effect of scale, velocity, and cyclic
degradability on laterally loaded single piles in overconsolidated clay. Rep. No.
UHCE 84-7, Dept of Civil Engineering, Univ of Houston, Houston, TX; 1984.
[25] ONeill MW, Murchison JM. An evaluation of py relationship in sands. A report
to the American Petroleum Institute, PRAC 82-41-1, University of Houston,
Texas; 1983.
[26] Arup Ove et al. GSRAFT as part of GSA user manual. London: Oasys Ltd.; 1996.
[27] Pasternak PL. On a new method of analysis of an elastic foundation by means
of two constants. Gosudarstvennoe Izdatelstvo Literaturi po Stroitelstvui
Arkhitekture, Moscow; 1954 [in Russian].
[28] PLAXIS 3D Foundation. PLAXIS 3D foundation user manual, version 2.0.
Brinkgreve, R.B., Swolfs, W.M., PLAXIS Inc.; 2008.
[29] Poulos HG. Analysis of piled strip foundations. In: Proceedings of conference
on computer methods and advances in geomechanics. Rotterdam: Balkema;
1991. p. 18391.
[30] Poulos HG. An approximate numerical analysis of pileraft interaction. Int J
Numer Anal Meth Geomech, London 1994;18(2):7392.
[31] Poulos HG. Piled raft foundations: design and applications. Geotechnique
2001;51(2):95113.
[32] Randolph MF. Design of piled foundations. Research Report Soils TR143,
Cambridge: Cambridge University Engineering Department; 1983.
[33] Reese LC, Cox WR. Field testing and analysis of laterally loaded piles in stiff
clay. In: Proc. offshore technology conference, OTC 2312; 1975.
[34] Reese LC, ONeill MW, Smith RE. Generalized analysis of pile foundations. J Soil
Mech Found Div, ASCE 1970;96(1):23550.
[35] Reul O, Randolph MF. Design strategies for piled rafts subjected to nonuniform
vertical loading. J Geotech Geoenviron Eng, ASCE 2004;130(1):113.
[36] Roberto C, Enrico C. Settlement analysis of pile groups in layered soils. Can
Geotech J 2006;43:788801.
[37] Russo G. Numerical analysis of piled rafts. Int J Numer Anal Meth Geomech
1998;22(6):47793.
[38] Sommer H. Entwicklung der Hochhausgrundungen in Frankfurt/Main Festkoll
oquium 20 Jahre Grundbauinstitut. In: Prof. Dr. -Ing. H. Sommer und Partner,
Germany; 1991. p. 4762.
[39] Terzaghi K. Evaluation of coefcients of subgrade reaction. Geotechnique
1955;5:297326.
[40] Vesic AS. Bending of beams resting on isotropic elastic solid. J Eng Mech Div,
ASCE 1961;87:3553.
[41] Vesic AS. Experiments with instrumented pile groups in sand. Performance of
deep foundation. ASTM, special technical publication; 1969. 444, p. 172222.
[42] Viggiani C. Pali come riduttori di cedimento; un esempio. In: Proc., Atti
XIX Convegno Nazionale Geotecnica, 2, Pavia, Italy, Ptron, Bologna; 1995.
p. 5236.
[43] Vlasov VZ, Leontiev UN. Beams, plates, and shells on elastic foundation. Israel
Program for Scientic Translations, Jerusalem (translated from Russian); 1966.
[44] Wang ST, Reese LC. COM624P laterally loaded pile analysis for the
microcomputer.ver. 2.0, FHWA-SA-91-048, Springeld, VA; 1993.
[45] Wang A. Three dimensional nite element analysis of pile groups and piled raft, Ph.D. dissertation, University of Manchester, U.K., 1996.
[46] Winkler E. Die Lehre von der Elasizitat und Festigkeit. Dominicus; 1867.
[47] Whitaker T. Experiments with model piles in groups. Geotechnique
1957;7(4):14767.
[48] Won JO, Jeong SS, Lee JH, Jang SY. Nonlinear three-dimensional analysis of pile
group supported columns considering pile cap exibility. Comput Geotech
2006;33:35570.
[49] Zhang HH, Small JC. Analysis of axially and laterally loaded pile groups
embedded in layered soils. In: Proceedings of 8th Australia NewZealand Conf.
on Geomechanics, vol. 1. Hobart; 2000. p. 475483.