Engineering Analysis with Boundary Elements 103 (2019) 66–79

Contents lists available at ScienceDirect

Engineering Analysis with Boundary Elements

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

A 3D BEM/FEM formulation for the static analysis of piled rafts and

capped pile groups subjected to vertical and horizontal loads
Endi Samba Luamba∗, João Batista de Paiva
São Carlos School of Engineering, University of São Paulo, São Carlos, SP, Brazil

a r t i c l e i n f o a b s t r a c t

Keywords: In this paper, soil-structure interaction problems are analysed using a three-dimensional formulation obtained by
Soil-structure interaction combining the Boundary Element Method (BEM) and the Finite Element Method (FEM). Structural elements that
Boundary elements interact with the soil are modelled with the FEM. Soil, which is considered as a semi-infinite, homogeneous, linear
Finite elements
elastic and isotropic medium, is modelled with the BEM using Mindlin’s fundamental solution. A pile is discretized
Piled raft
by several three-dimensional finite beam elements. A plate is discretized by finite flat shell elements, making it
Capped pile groups
BEM/FEM coupling possible to consider the bending and membrane effects. The vertical and horizontal loads in the plate-pile-soil
system are directly applied to the plate. This approach allows for a static analysis of the full interaction between
capped pile groups or piled rafts and the soil. The developed formulations can also be extended to dynamic or
non-linear analyses. Coupling between the BEM and FEM is performed through a mixed formulation in which
the matrix of the soil’s influence coefficients obtained by BEM is added to the stiffness matrix of the structural
elements obtained by FEM, resulting in an equivalent stiffness matrix. Numerical examples of the plate-pile-soil
interactions are solved, and their results are compared with those of other formulations.

1. Introduction applied to the analysis of general capped pile groups. Butterfield and
Banerjee [3] proposed a general formulation for the analysis of capped
Piled rafts and capped pile groups are types of deep foundations in pile groups. However, as in [2], the cap is supposed to be rigid. The
which all the individual components, namely, the raft or the cap on the earliest study considering the plate as a flexible medium is by Brown
piles and the surrounding soil, are mutually interacting. Their function and Weisner [4], in which an analysis is performed for a long flexible
is to resist and transfer the superstructure loads to the soil farther down footing that is supported on the soil and on piles. However, in this for-
from the surface. Piled rafts are usually used when a raft foundation mulation, the footing is analysed by classical beam theory, limiting its
alone does not satisfy a design requirement, such as the allowable set- applications.
tlement for a large structure. In capped pile groups, when considering a Many other studies have been published on plate-pile-soil interac-
rigid cap, the displacements at the top of the piles are the same. The soil- tions, such as the one by Fatemi-Ardakani [5] in which the plate is anal-
structure interaction (SSI) remains a subject of growing interest given ysed by BEM and the piles are represented by springs. Hain and Lee
the many different disciplines and aspects it encompasses, and it refers [6] proposed a formulation in which the plate is analysed by FEM and
in general to static and dynamic phenomena mediated by a compliant the contributions of the piles are given by the load-deflection curves
soil and a stiffer structure [1]. presented in [2]. Gazetas and Mylonakis [7] presented a formulation
For the analysis of piled rafts and capped pile groups, in which for analysing grouped piles with rigid caps in a layered soil. In this for-
the soil is generally represented as a linear elastic homogeneous semi- mulation, the soil around the shaft of the pile is empirically approxi-
infinite medium, one of the first studies was published by Poulos and mated by distributed Winkler springs. From a study on the interaction
Davis [2], in which a formulation is presented to study the interaction between two piles, load-displacement factors are obtained and then used
between capped pile groups and the soil. The contribution of the piles for analysing groups of piles. Recently, Ai and Feng [8] presented a
is computed using load-deflection curves obtained from a parametric boundary element formulation for analysing laterally loaded fixed-head
study of the interaction between a capped two-pile group and the soil, pile groups, in which the piles are divided into several segments and
in which the cap is circular and considered rigid. Although elastic su- solved by finite difference method.
perposition is valid only when the piles are located along a circumfer- It should be noted that none of the studies mentioned above con-
ence and subjected to the same load, these load-deflection curves were sidered the full interactions between the plate, the pile and the soil. In

∗
Corresponding author.
E-mail address: [email protected] (E.S. Luamba).

https://doi.org/10.1016/j.enganabound.2019.02.009
Received 10 October 2018; Received in revised form 19 January 2019; Accepted 21 February 2019
0955-7997/© 2019 Elsevier Ltd. All rights reserved.
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

fact, those formulations do not allow for direct coupling with a founda- of structural elements that interact with the soil presents five nodal pa-
tion plate, since the displacements and rotations at the top of the pile rameters (three displacements and two rotations), facilitating the con-
do not appear as unknowns in the system of equations. Some studies nection between the pile and the plate at the common nodes. More-
such as [9–11] presented BEM and BEM/FEM formulations in which all over, a perfect bond condition is assumed over the contact regions of
the interactions between the plate, the pile and the soil were considered the structural elements; hence, the compatibility of the displacements
simultaneously. However, the pile was discretised by only one element, is considered along those contact regions. The matrix of the soil’s in-
and it was not possible to simulate the flexibility of long piles adequately fluence coefficients is also increased, considering the influences on the
when they were subjected to horizontal loads; hence, only vertical loads soil surface and in the piles due to source points placed on both the soil
were applied. Moreover, in formulations such as those in [8,12,13], the surface and the piles.
load applied within the plate-pile-soil system is distributed to the top of During each step of the analyses presented here, the final system
the piles instead of being applied directly to the plate, thus providing a of the governing equations is obtained through a mixed formulation in
simplified approach and not considering the modelling of the plate. In which the influence coefficients matrix from the BEM is added to the
Ribeiro and Paiva [14,15], Kelvin’s fundamental solution was used to stiffness matrix from the FEM, resulting in an equivalent stiffness ma-
analyse the plate-soil interaction, leading to the complex discretisation trix of the problem under analysis. Additionally, the resolution of this
of the analysed problems, as shown in the examples. There are many system allows for the determination of displacements and rotations in
studies, such as [9,15,16] and the present one, in which the raft is mod- the nodes of the finite elements of the pile and in the nodes belonging
elled as a thin plate; others, such as [17,18], model the raft as a thick to the contact region between the plate and the soil surface. The new
plate. The results from both studies tend to be highly consistent when aspects of the present research, apart from the proposed BEM/FEM for-
analysing piled rafts, since the thickness of the raft is generally very mulation, which greatly simplifies the model discretisation, are the new
small relative to the other dimensions. This characteristic is mentioned discretisation of the pile to account for its flexibility and the consider-
in one of the examples of this work. ation of the membrane effect on the plate, allowing the plate-pile-soil
The SSI applications of this study are adequately analysed by a system to also be subjected to horizontal loads. Therefore, the presented
BEM/FEM formulation. As stated in [19] by Zienkiewicz et al., benefits model allows for the analysis of a deep foundation that is subjected
could be obtained by combining the BEM and FEM, thereby yielding the to horizontal and vertical loads, with the raft (or the cap) in contact
‘best of both worlds’. Hence, the stiffness matrix of the Boundary Ele- with the soil. The same model can also readily be extended to analyse
ment region is obtained and assembled with the FEM stiffness matrices. the interaction between the foundation and the superstructure. Numer-
Since many general-purpose programmes allow for the import of a user- ical examples of plate-pile-soil interactions are solved to verify, vali-
defined element stiffness matrix, this approach may be used to extend date and demonstrate the efficiency of the developed and implemented
the capabilities of an FEM code. The soil-structure interaction problems formulations.
of the present work may also be analysed entirely with the FEM, but in
this case, the degrees of freedom that result from the discretisation of the
model are numerous when compared with the BEM/FEM formulations 2. Soil equations
proposed herein.
Therefore, in this study, in a continuation from the study initiated The soil as modelled with the BEM refers to the soil surface and the
in [20], a formulation for the analysis of pile-soil interaction is first de- embedded line loads. The matrix [G] of the soil’s influence coefficients
veloped in which the pile can be short or long, and it is subjected to is made up of the following sub-matrixes:
horizontal and/or vertical loads and bending moments at the top. The
pile is discretised by several three-dimensional finite beam elements, [GSS ]: the matrix generated by placing the source points and the field
and the interaction tractions have a linear distribution for each finite points on the soil surface
pile element. The pile surface is assumed to be rough to prevent any [GSL ]: the matrix generated by placing the source points on the soil
sliding along the pile-soil contact region. The soil is modelled using the surface and the field points along the line loads below the soil
BEM, and it is assumed to be a semi-infinite, homogeneous, isotropic surface
and linear elastic medium. Mindlin’s fundamental solution on displace- [GLS ]: the matrix generated by placing the source points along the
ments is used to obtain the soil’s influence coefficients as calculated at line loads below the soil surface and the field points on the soil
the nodes of the pile and modelled as a line load. Mindlin’s fundamental surface
solution is particularly suitable for the type of problems under analysis, [GLL ]: the matrix generated by placing the source points and the field
viz., problems involving semi-infinite three-dimensional solids, since it points along the line loads below the soil surface
is necessary to discretise only the loaded surface of the soil and/or the
line load and not all the three-dimensional solids. This approach there- In the ensuing subsections, we describe the determination of the ma-
fore reduces the number of the problem unknowns and consequently the trix [G] components.
dimensions of the matrices involved, saving on computational costs.
The interaction between the plate and the soil is then analysed. The
plate is modelled with the FEM and discretised by DKT (Discrete Kirch- 2.1. Matrix [GLL ]
hoff Theory) finite elements, and then it is subjected to vertical loading.
The soil continues to be modelled with the BEM using Mindlin’s funda- For a homogenous, isotropic and linear elastic domain, a boundary
mental solution, which is now obtained on the soil surface. During this integral equation called the Somigliana identity, which represents the
step, only the soil surface that is in contact with the plate needs to be equilibrium of a solid body, is given as follows:
discretised by cells or triangular boundary elements. The cells coincide
with the DKT elements, and the linear distribution of contact tractions in
𝐶 (𝑠)𝑢(𝑠) + 𝑃 ∗ (𝑠, 𝑓 )𝑢(𝑓 )𝑑Γ = 𝑢∗ (𝑠, 𝑓 )𝑃 (𝑓 )𝑑Γ + 𝑢∗ (𝑠, 𝑓 )𝑏(𝑓 )𝑑Ω
each cell is admitted. During the BEM/FEM coupling, in the contact re- ∫Γ ∫Γ ∫Ω
gion, the tractions acting on the soil surface are converted into reactive (1)
nodal loads on the plate.
Finally, the plate-pile-soil interaction is analysed, and the plate can The pile is immersed in a continuous and semi-infinite medium, and
now also receive horizontal loading; hence, for the accurate mechani- it is subjected to interface tractions Q, as shown in Fig. 1.
cal modelling of the membrane effect, CST (Constant Strain Triangle) While considering stress 𝜎 b to be uniformly distributed at the base
finite elements are used. For this step, it is worth noting that each node of the pile and the interface tractions Q, Eq. (1) becomes

67
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 2. Cell ijk with linearly approximated tractions.

The tractions field is given by

[ ]{ }
𝑄 = Φ𝑛 𝑄 𝑛 (6)

where

[Φn ]: the matrix of shape functions

{Qn }: the vector of tractions

Thus, for each cell, we have the following:

⎧𝑄 ⎫
[ ]⎪ 1 ⎪
𝑄 = Φ 1 Φ 2 Φ 3 ⎨𝑄 2 ⎬ (7)
Fig. 1. Interface tractions acting along the pile and in the soil domain. ⎪𝑄 3 ⎪
⎩ ⎭
The shape functions Φn written in relation to a coordinate system (x,
𝐶 (𝑠) 𝑢(𝑠) + 𝑃 (𝑠, 𝑓 ) 𝑢(𝑓 ) 𝑑Γ =
∗
𝑢 (𝑠, 𝑓 ) 𝑃 (𝑓 ) 𝑑Γ +
∗
𝑢 (𝑠, 𝑓 ) 𝑏(𝑓 ) 𝑑Ω
∗ y) are expressed as
∫Γ ∫Γ ∫Ω
1 [( ) ( ) ( )]
Φ1 = 𝑦2 − 𝑦3 𝑥 + 𝑥3 − 𝑥2 𝑦 + 𝑥2 𝑦3 − 𝑥3 𝑦2
+ 𝑢∗ (𝑠, 𝑓 ) 𝑄(𝑓 ) 𝑑 Γ𝑞 𝐷𝑡
∫Γ𝑞
1 [( ) ( ) ( )]
Φ2 = 𝑦3 − 𝑦1 𝑥 + 𝑥1 − 𝑥3 𝑦 + 𝑥3 𝑦1 − 𝑥1 𝑦3
𝐷𝑡
+ 𝑢∗ (𝑠, 𝑓 ) 𝜎𝑏 (𝑓 ) 𝑑 Γ𝑏 (2)
∫Γ𝑏 1 [( ) ( ) ( )]
Φ3 = 𝑦1 − 𝑦2 𝑥 + 𝑥2 − 𝑥1 𝑦 + 𝑥1 𝑦2 − 𝑥2 𝑦1 (8)
𝐷𝑡
With the use of Mindlin’s fundamental solution and the volumetric
forces that are being neglected, Eq. (2) is reduced to where
(xi ,yi ): the coordinates of cell node i, with i = 1, 2, 3
𝐶 (𝑠)𝑢(𝑠) = 𝑢∗ (𝑠, 𝑓 )𝑄(𝑓 )𝑑 Γ𝑞 + 𝑢∗ (𝑠, 𝑓 )𝜎𝑏 (𝑓 )𝑑 Γ𝑏 (3) ( ) ( ) ( )
∫Γ𝑞 ∫Γ𝑏 𝐷𝑡 = 𝑥1 𝑦2 − 𝑦3 + 𝑥2 𝑦3 − 𝑦1 + 𝑥3 𝑦1 − 𝑦2 (9)
Mindlin’s fundamental solution is used in Eq. (3); therefore, the For simplicity, the shape functions Φn can be written as
source points are applied to the pile domain, which coincides with
boundary Γq . Thus, C(s) is equal to the identity matrix. Moreover, within Φ𝑖 = 𝐴 𝑖 𝑥 + 𝐵 𝑖 𝑦 + 𝐶 𝑖 (10)
the continuous semi-infinite medium, piles are represented by line loads. One can observe that in Eq. (10), the shape functions are equivalent
Thus, Eq. (3) can be written as follows: to the equation of a plane. The loading being applied to the soil surface,
𝑁𝑝 [ ] the source points are located at the cell nodes, and the field points are
∑
𝑢 (𝑠 ) = 𝑢∗ (𝑠, 𝑓 )𝑄(𝑓 )𝑑𝑞𝑖 + 𝑢∗ (𝑠, 𝑓 )𝜎𝑏 (𝑓 )𝑑 Γ𝑏𝑖 (4) placed along the three sides of each cell.
∫Γ𝑞 ∫Γ𝑏𝑖
𝑖=1 𝑖 The displacements of the source points are given by
Np is the number of piles or line loads immersed in a continuous, 𝑁𝑐
semi-infinite medium. ∑
𝑢(𝑠) = 𝑢∗ (𝑠, 𝑓 ) 𝑄 (𝑓 ) 𝑑 Ω𝑞 (11)
Mindlin’s fundamental solution for displacements u∗ (s, f) is ∫
𝑘=1 Ω𝑞
presented in annex A.
For numerical implementation purposes, Eq. (4) can be written in where
matrix form as
[ ] Nc : the number of cells
{𝑢} = 𝐺𝐿𝐿 {𝑄} (5) Ωq : the cell domain
Q: the load applied to the soil surface
2.2. Matrix [GSS ]
The fundamental solution of interest in Eq. (11) is the one that refers
To analyse the interaction between the plate and the soil, the dis- to the vertical displacements, that is,
cretisation of the plate and the one on the soil surface coincide, i.e., { }
𝑟23 3 − 4𝜈 6𝑐𝑧𝑅3
2
8(1 − 𝜈)2 − (3 − 4𝜈) (3 − 4𝜈)𝑅3 − 2𝑐𝑧
2
there is a perfect match between the DKT finite elements of the plate 𝑢∗33 = 𝐾 + + + +
and the triangular boundary elements or cells of the soil surface. A cell 𝑟3 𝑟 𝑅 5 𝑅 𝑅 3

of the soil surface, as well as a DKT element, has three nodes at the ver- (12)
tices. In each cell, the tractions are linearly approximated, as shown in
When the source points and the field points are on the soil surface,
Fig. 2. Therefore, in this study, the plate (cap or raft) is in contact with
Boussineq’s fundamental solution is obtained from Mindlin’s fundamen-
the soil. Other studies, such as the one in Ref. [21], consider off-ground
tal solution, substituting in Eq. (12): x3 (s) = 0 and x3 (f) = 0.
caps.

68
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 3. Tractions in the contact region between the

plate and the soil surface.

We obtain the following: Eq. (14) for the present case is given as
( ) {( )[ ]
1−𝜈 1 𝐴𝑖 𝑅𝑐𝑜𝑠𝜃 + 𝐵𝑖 𝑅𝑠𝑒𝑛𝜃 𝛽
𝑢∗33 = (13)
2𝜋𝐺 𝑟 ℎ=𝛼 𝐹1 (𝑅) + 𝑥23 (𝑠)𝐹2 (𝑅)
∫Γ𝑐 𝑅 𝛼
where [ ]}
G and 𝜈: the shear modulus and the Poisson’s ratio of the soil, 𝛽 𝑛̄ ⋅ 𝑟̄
+ 𝐷𝑖 𝐹3 (𝑅) + 𝑥23 (𝑠)𝐹4 (𝑅) ⋅ ⋅ 𝑑 Γ𝑐 (16)
respectively. 𝛼 𝑅
1
𝑟 = (𝑟𝑖 𝑟𝑖 ) 2 : the modulus of the radius vector. where
√
The integral in Eq. (11) on the cell domain can be transformed into
𝑅 1 = 𝑅 2 + 𝑥 3 2 (𝑠 ) (17)
an integral on the cell boundary. Thus, for one cell, we obtain
( )
1−𝜈 ( ) 𝑅 + 𝑅1
ℎ= 𝐴𝑖 Rcos𝜃 + 𝐵𝑖 Rsen𝜃 + 2𝐷𝑖 ⋅ 𝑛 ⋅ 𝑟 ⋅ 𝑑Γ𝑐 (14) 𝐹0 (𝑅) = ln (18)
4𝜋𝐺 ∫Γ𝑐 𝑥 3 (𝑠 )
1[ ]
where 𝐹 1 (𝑅 ) = 𝑅𝑅1 − 𝑥3 2 (𝑠)𝐹0 (𝑅) (19)
2
𝑛̄ : unit outward normal at field points
𝑅
𝑟̄: unit radius vector 𝐹 2 (𝑅 ) = − + 𝐹0 (𝑅) (20)
𝑅1
R: distance between the source point and the field point
Γc : cell boundary 𝐹 3 (𝑅 ) = 𝑅 1 − 𝑥 3 (𝑠 ) (21)
Di = Ai xs + Bi ys + Ci : shape function calculated at the source point of
coordinates (xs ,ys ) 1 1
𝐹 4 (𝑅 ) = − + (22)
𝑅 1 𝑥 3 (𝑠 )
It should be noted that the integral in Eq. (14) has no singularity.
Eq. (11) can now be written in matrix form as follows: 2(1 − 𝜈)
𝛼= (23)
[ ] 4𝜋𝐺
{𝑢} = 𝐺𝑆𝑆 {𝑄} (15)
1
𝛽= (24)
The determination of matrix [GSL ] is performed using the formula- 4𝜋𝐺
tions presented in Section 2.1, that is, the field points are placed along Considering the overall plate-pile-soil interaction, in determining the
the line loads. The difference is that the source points are placed on the [GSS ] and [GLS ], the influence coefficients are null at the intersection
soil surface, as presented here in this section. node of the line load and the loaded soil surface. In fact, at that node,
as is common between the plate and the pile, the soil reaction is zero
2.3. Matrix [GLS ]
since there is no contact between the soil and the plate. In this region,
only normal stress 𝜎 N is transmitted by the plate to the pile, as shown
To determine matrix [GLS ], we follow the same procedure described
in Fig. 3.
in Section 2.2, with the source points placed along the line loads, and

69
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

After determining all the components of matrix [G] of the soil’s in- I: the area moment of inertia of the pile cross-section, I = Ix = Iy
fluence coefficients, the same can finally be expressed as A: the cross-section area of the pile
[[ ] [ ]] Qx , Qy , Qz : interface tractions along the pile
𝐺𝑆𝑆 𝐺𝑆𝐿
[𝐺 ] = [ ] [ ] (25) ux , uy , uz : displacements of the node at the top of the pile
𝐺𝐿𝑆 𝐺𝐿𝐿
By minimising the total potential energy Π given in Eq. (27), to
In Appendix A, the assembling of matrix [G] is illustrated by a the-
obtain the equilibrium situation, we obtain the following system of
oretical example proposed by the authors.
equations:
3. Pile equations
[𝐾 ]{𝑢} = {𝐹 } + [𝑇 ]{𝑄} (30)

The pile is modelled with the FEM and discretised by several three- where
dimensional finite beam elements with nodes at the ends. At each node,
there are five nodal parameters, as shown in Fig. 4. [K]: the stiffness matrix of the pile
The pile may be subjected, at its top, to horizontal forces Hx and {u}: the nodal displacements vector of the pile
Hy , to vertical force V, and to bending moments Mx and My . There are { }
{𝑢}𝑇 = 𝑢1 𝑣1 𝑤1 𝑢1 𝑣1 𝑢2 𝑣2 𝑤2 𝑢2 𝑣2 . .. 𝑢𝑛 𝑣𝑛 𝑤𝑛 𝑢𝑛 𝑣𝑛
′ ′ ′ ′ ′ ′
a total of ten nodal parameters per finite element, with horizontal dis-
placements u and v, vertical displacement w, and rotations u′ and v′
{F}: the vector of the external forces applied to the top of the pile
around the y-axis and the x-axis, respectively.
{ } { }
{𝑢}𝑒 𝑇 = 𝑢𝑖 𝑣𝑖 𝑤𝑖 𝑢𝑖 ′ 𝑣𝑖 ′ 𝑢𝑗 𝑣𝑗 𝑤𝑗 𝑢𝑗 ′ 𝑣𝑗 ′ (26) {𝐹 } = 𝐻𝑥 𝐻𝑦 𝑉 𝑀𝑥 𝑀𝑦 0 0 0 0 0 . . . 0 0 0 0 0

where [T]: the matrix that transforms pile-soil interface tractions into
T: equivalent nodal forces
{u}e the nodal displacements vector of the finite element e as
{Q}: the vector of pile-soil interface tractions
delimited by nodes i and j.
The total potential energy of the pile is given by The stiffness matrix [K]e of a single finite element of the pile is given
by
Π=𝑈 +Ω (27)
where 𝐿
[𝐾 ]𝑒 = [𝐵 ]𝑇 EI [𝐵 ] dz (31)
∫0
U: the potential energy of deformation or elastic potential energy
Ω: the potential energy of applied loads where

EI: bending rigidity of the element

[B]: strain-displacement interpolation matrix

⎡ 12𝐸𝐼 0 0 6𝐸𝐼
0 − 12𝐿𝐸𝐼 0 0 6𝐸𝐼
0 ⎤
⎢ 𝐿 𝐿2 𝐿2
3 3
12𝐸𝐼 6𝐸𝐼
⎥
⎢ 0 𝐿3
0 0 𝐿2
0 − 12𝐿𝐸𝐼
3 0 0 6𝐸𝐼
𝐿2
⎥
⎢ 𝐸𝐴 ⎥
⎢ 0 0 𝐿
0 0 0 0 − 𝐸𝐴
𝐿
0 0 ⎥
⎢ 6𝐸𝐼 4𝐸𝐼 ⎥
⎢ 𝐿2 0 0 𝐿
0 − 6𝐿𝐸𝐼
2 0 0 2𝐸𝐼
𝐿
0 ⎥
⎢ 6𝐸𝐼 4𝐸𝐼 2𝐸𝐼 ⎥
𝑒 ⎢ 0 𝐿2
0 0 𝐿
0 − 6𝐿𝐸𝐼
2 0 0 𝐿 ⎥
[𝐾 ] = ⎢ 12𝐸𝐼 ⎥ (32)
⎢− 𝐿3 0 0 − 6𝐿𝐸𝐼
2 0 12𝐸𝐼
𝐿3
0 0 − 6𝐿𝐸𝐼
2 0 ⎥
⎢ ⎥
⎢ 0 − 12𝐿𝐸𝐼
3 0 0 − 6𝐿𝐸𝐼
2 0 12𝐸𝐼
𝐿3
0 0 − 6𝐿𝐸𝐼
2 ⎥
⎢ ⎥
⎢ 0 0 − 𝐸𝐴
𝐿
0 0 0 0 𝐸𝐴
𝐿
0 0 ⎥
⎢ 6𝐸𝐼 2𝐸𝐼 ⎥
⎢ 𝐿2 0 0 𝐿
0 − 6𝐿𝐸𝐼
2 0 0 4𝐸𝐼
𝐿
0 ⎥
⎢ 0 6𝐸𝐼 2𝐸𝐼
− 6𝐿𝐸𝐼 4𝐸𝐼 ⎥
⎣ 𝐿2
0 0 𝐿
0 2 0 0 𝐿 ⎦

For the pile of length L, and with modelling according to Fig. 4, we

4. Plate equations
find the following:
𝐿( 2 )2 𝐿( 2 )2
EI 𝜕 𝑢 (𝑧 ) EI 𝜕 𝑣 (𝑧 ) In this study, “plate” indicates the structural elements of rafts and
𝑈 = dz + dz
2 ∫0 𝜕𝑧 2 2 ∫0 𝜕𝑧 2 rigid caps, which are subjected to bending and membrane effects. Thus,
𝐿( )2 the modelling of those elements is performed by FEM using flat shell
EA 𝜕𝑤(𝑧)
+ dz (28) finite elements. The flat shell finite element considered here is the result
2 ∫0 𝜕𝑧
of the superposition of the DKT finite element and CST finite element,
as shown in Fig. 5.
𝐿
𝜕 𝑢̄ 𝜕 𝑣̄ There are five nodal parameters per node and a total of fifteen nodal
Ω = − 𝐻𝑥 𝑢̄ − 𝑀𝑥 + 𝑄𝑥 (𝑧)𝑢(𝑧)𝑑𝑧 − 𝐻𝑦 𝑣̄ − 𝑀𝑦
𝜕𝑧 ∫0 𝜕𝑧 parameters for the flat shell finite element.
𝐿 𝐿 { }
+ 𝑄𝑦 (𝑧)𝑣(𝑧)𝑑𝑧 − 𝑉 𝑤̄ + 𝑄𝑧 (𝑧)𝑤(𝑧)𝑑𝑧 (29) {𝑢}𝑒 𝑇 = 𝑢1 𝑣1 𝑤1 𝑢1 ′ 𝑣1 ′ 𝑢2 𝑣2 𝑤2 𝑢2 ′ 𝑣2 ′ 𝑢3 𝑣3 𝑤3 𝑢3 ′ 𝑣3 ′ (33)
∫0 ∫0
where where

E: Young’s modulus of the pile {u}e T : the nodal displacements vector of the flat shell finite element

70
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Formulations on the CST finite element can be found in Ref. [22].

The stiffness matrix of the DKT finite element is explicitly derived in
Ref. [23], which can be used directly in computer programming.
The stiffness matrix of a flat shell finite element is then given by
[[ ]𝑒 ]
𝐾𝐷𝐾𝑇 0
[𝐾 ]𝑒 = [ ]𝑒 (34)
0 𝐾𝐶𝑆𝑇
where

[KDKT ]e : the stiffness matrix of the DKT finite element

[KCST ]e : the stiffness matrix of the CST finite element

As we can observe in Eq. (34), in a flat shell finite element, the DKT
and CST finite elements are decoupled, i.e., the bending and membrane
effects occur independently of one another. It should also be noted that
each node of the three-dimensional finite beam element and of the flat
shell finite element has five nodal parameters; therefore, the connection
between both finite elements is straightforward.

5. BEM/FEM coupling

BEM/FEM coupling is performed by combining the systems of equa-

tions arising from the BEM (soil equations) and FEM (pile and plate
equations). The subscripts “p” and “q” below indicate “pile” and “plate”,
respectively. Considering the plate-pile assembly, similar to Eq. (30), the
system of equations is given by
[ ]{ } [ ]{ }
𝐾𝑝𝑞 𝑢𝑝𝑞 = {𝐹 } + 𝑇𝑝𝑞 𝑄𝑝𝑞 (35)
Fig. 4. Model of the pile: (a) Loads applied at the top of the pile and (b) nodal where
parameters.

Fig. 5. Flat shell finite element. (a) DKT finite element and (b) CST finite element.
71
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

For a single finite element of the pile, the vector of equivalent nodal
forces is given by

⎛ 𝐹𝑥𝑖 ⎞ ⎛ 𝑄𝑥𝑖 ⎞
⎜ 𝐹𝑦𝑖 ⎟ ⎜ 𝑄𝑦𝑖 ⎟
⎜ ⎟ ⎜ ⎟
𝐹
⎜ 𝑧𝑖 ⎟ ⎜ 𝑄𝑧𝑖 ⎟
⎜ 𝑀𝑥𝑖 ⎟ ⎜ 0 ⎟
⎜ ⎟ [ ]⎜ ⎟
⎜ 𝑀𝑦𝑖 ⎟ = 𝑇 𝑒 ⎜ 0 ⎟ (39)
𝑝
⎜ 𝐹𝑥𝑗 ⎟ ⎜𝑄𝑥𝑗 ⎟
⎜𝐹 ⎟ ⎜𝑄 ⎟
⎜ 𝑦𝑗 ⎟ ⎜ 𝑦𝑗 ⎟
𝐹
⎜ 𝑧𝑗 ⎟ ⎜𝑄𝑧𝑗 ⎟
⎜𝑀 ⎟ ⎜ 0 ⎟
⎜ 𝑥𝑗 ⎟ ⎜ ⎟
⎝𝑀𝑦𝑗 ⎠ ⎝ 0 ⎠

where

Fig. 6. Tractions to equivalent nodal forces for a single finite element of the ⎡ 7𝐿 0 0 0 0 3𝐿
0 0 0 0⎤
⎢ 20 7𝐿
20
3𝐿 ⎥
pile. ⎢ 0 20
0 0 0 0 20
0 0 0⎥
⎢ 0 0 𝐿
0 0 0 0 𝐿
0 0⎥
⎢ 2 3 6 ⎥
[Kpq ]: the stiffness matrix of the plate-pile assembly; ⎢ 𝐿 0 0 0 0 𝐿2
0 0 0 0⎥
⎢ 20 𝐿2
30
𝐿2
⎥
{upq }: the vector of nodal displacements of the plate-pile assembly; [ ]𝑒 ⎢ 0 0 0 0 0 0 0 0⎥
{F}: the vector of equivalent nodal forces arising from external loads; 𝑇𝑝 = ⎢ 3𝐿 20
7𝐿
30
⎥ (40)
⎢ 20 0 0 0 0 20
0 0 0 0⎥
[Tpq ]: the matrix that transforms element tractions to equivalent nodal ⎢ 0 3𝐿 7𝐿
0 0 0 0 0 0 0⎥
forces; and ⎢ 20
𝐿
20
𝐿 ⎥
{Qpq }: the plate-pile-soil interface traction vector ⎢ 0 0 6
0 0 0 0 3
0 0⎥
⎢ 𝐿2 ⎥
− 𝐿20
2
⎢− 30 0 0 0 0 0 0 0 0⎥
The stiffness matrix of the plate-pile assembly is given by ⎢ 0 − 𝐿30 − 𝐿20 0⎥⎦
2 2
⎣ 0 0 0 0 0 0

L: the length of the finite pile element

(36) For a single finite element of the plate, the vector of equivalent nodal
forces is given by
where
⎛ 𝐹𝑖 ⎞ [ ] ⎛ 𝑄𝑖 ⎞
⎜ 𝐹 ⎟ = 𝑇𝑞 𝑒 ⎜ 𝑄 ⎟ (41)
[Kp ]: the stiffness matrix of the pile(s) ⎜ 𝑗⎟ ⎜ 𝑗⎟
[Kq ]: the stiffness matrix of the plate given by ⎝𝐹𝑘 ⎠ ⎝𝑄 𝑘 ⎠
[[ ] ]
[ ] 𝐾𝐷𝐾𝑇 0 where
𝐾𝑞 = [ ] (37)
0 𝐾𝐶𝑆𝑇 ⎡2 1 1⎤
[ ]𝑒 𝐴⎢
The transformation matrix [Tpq ] is given by 𝑇𝑞 = 1 2 1⎥ (42)
12 ⎢ ⎥
⎣1 1 2⎦

A: the area of the plate finite element

The soil equation considering the plate-pile-soil interaction system

(38)
is given by
where { } { }
𝑢 𝑠 = [𝐺 ] 𝑄 𝑠 (43)
[Tp ]: the matrix that transforms tractions along the interface between where
the pile and the soil to equivalent nodal forces, as shown in Fig. 6;
and {us }: the vector of nodal displacements for the soil
[Tq ]: the matrix that transforms tractions in the contact region between [G]: the matrix of the soil’s influence coefficients, as described in
the plate and the soil to equivalent nodal forces, as shown in Section 2
Fig. 7 {Qs }: the vector with the nodal values of the soil interface tractions

Fig. 7. Tractions to equivalent nodal forces for a single finite element of the plate.

72
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 8. Loaded area.

Reordering Eq. (43) results in the following: Table 1

{ } { } Vertical displacement (m x 10−2 ).
𝑄𝑠 = [𝐺]−1 𝑢𝑠 (44)
Point A Error (%) Point B Error (%)
To account for the force equilibrium that must hold along the inter-
face between the piles, the plate and the continuum medium, viz. Exact 2.2444 − 1.1222 −
{ } { } Ref. [25] 2.2520 0.7 1.1298 0.4
𝑄𝑠 + 𝑄𝑝𝑞 = 0 (45) Ref. [14] No IBEs 2.1410 4.6 1.0371 7.6
With IBEs 2.2114 1.5 1.1159 0.6
Eqs. (35) and (44) can be combined to give This work 2.2444 0.0 1.1222 0.0
[ ]{ } { }
𝐾𝑝𝑞 𝑢𝑝𝑞 = {𝐹 } − [𝑀 ] 𝑢𝑠 (46)
where
[ ] 6.1. Semi-infinite medium with a square load
[𝑀 ] = 𝑇𝑝𝑞 [𝐺]−1 (47)
In Eq. (46), the vectors {us } and {upq } are of different orders. Thus, Over the surface of a homogeneous, isotropic and linear elastic
to perform the coupling of the two systems of equations, {us } must be semi-infinite medium, a square load of Q = 100 kN/m2 with an area of
rewritten by incorporating the nodal displacement derivatives of vector 2 m × 2 m is applied, as shown in Fig. 8. The Young’s modulus of the
{upq }. Then, matrix [M] must be augmented to be of the same order as material is ES = 10,000 kN/m2 and the Poisson’s ratio is 𝜈 = 0. Point A
matrix [Kpq ]. Hence, columns of zeroes are added to [M], which stands is placed at the centre of the square, and point B coincides with one of
for the neglected rotations around the x-axis and y-axis in the soil. The the vertices.
enlargement of matrix [M] is denoted by the addition of a bar, [𝑀 ̄ ]. As This problem is analysed using a mesh containing only 4 cells and 5
a consequence, displacement vector {us } in the soil equation must also nodes, as illustrated in Fig. 9. In fact, because the purpose is to determine
be augmented. the displacements at the centre and one of the vertices of the loaded
The displacement compatibility yields area, the generated mesh is sufficient to yield the exact solution for
{ } { } { } the problem, as presented in Table 1. The analytical (exact) solution
𝑢𝑠 = 𝑢𝑝𝑞 = 𝑈̄ (48) presented here was extracted from Ref. [24], and it is used as a reference
Eq. (48) in Eq. (46) results in for the error calculation.
[[ ] [ ]]{ } As in Eq. (15), the displacements are calculated by the following
𝐾𝑝𝑞 + 𝑀 ̄ 𝑈̄ = {𝐹 } (49)
expression:
Finally,
[ ]{ } {𝑢} = 𝐾 [𝐺]{𝑄}
𝐾̄ 𝑈̄ = {𝐹 } (50)
where K is the coefficient that appears in Eq. (14).
where
1−𝜈 1 10−4
[𝐾̄ ]: the final BEM/FEM coupling stiffness matrix; 𝐾= = = = 1.59155 × 10−5 m2 ∕kN
4𝜋𝐺 4 𝜋 5000 2𝜋
{𝑈̄ }: the vector that encompasses all the nodal displacements and ro-
tations of the plate-pile-soil system; and
{F}: the vector of the external loads applied to the plate
It should be noted that the BEM/FEM coupling formulation for the
analysis of the plate-pile-soil interaction presented in this section can be
particularized easily to analyse the pile-soil interaction and the plate-soil
interaction. The results for the different types of interaction analysis are
discussed in this work.

6. Examples

The computational code is written in Fortran®, in a 64-bit Win-

dows® environment, in an Intel® CPU 3.10 GHZ with 8 GB of RAM.
Each pile is discretized into 20 three-dimensional finite beam ele-
ments. The computational times required, by order of the following
applications, are 0.016, 5.7, 0.9, 19.6, 127.5 and 168.3 s. Fig. 9. Generated mesh.

73
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 10. Raft with the applied load and generated mesh.

Table 2
Fig. 11. Capped single pile and applied load.
Vertical displacement (m x 10−8 ).

Point Ref. [16] Ref. [26] Ref. [15] This work

A 2.88 3.00 3.40 3.50

B 3.26 3.63 3.75 3.90

Thus, we obtain

𝑢1 = 𝑢2 = 𝑢3 = 𝑢4 = 7.0510 × 100 × 𝐾 = 1.1222 × 10−2 m

𝑢5 = 14.1020 × 100 × 𝐾 = 2.2444 × 10−2 m Fig. 12. Generated mesh.

As expected, the largest displacement occurs in the centre of the

loaded area. Note that the coefficients 7.0510 and 2.2444 are the sum same as that indicated in Ref. [14] under Example 6.1. It can be ob-
of the soil’s influence coefficients in [G] at the corresponding line of the served that the results are highly consistent when a mixed BEM/FEM
source point (or node of interest). technique is used, and the discrepancy is due to differences between the
The results in Ref. [25] were obtained using BEs (boundary elements) techniques and the assumptions under consideration. Except the mesh
and IBEs (infinite boundary elements) with quadratic shape functions. used in Ref. [26], Ref. [16] and Ref. [15], the discretisation of the plate
In Ref. [14], the author simulated the problem by employing a 73-node is identical to the one shown in Fig. 10. The advantage of the proposed
mesh with 128 BEs and 32 IBEs. The shape functions were linear and model is that the mesh is simpler than the one in Ref. [15], since using
Kelvin’s fundamental solution was used. IBES is not necessary.
Upon comparing the meshes used in the different formulations and
the results, the efficiency of the formulation proposed in this work 6.3. Capped single pile under a uniform vertical load
is clear with regard to problems involving a continuous semi-infinite
medium. The interaction between a piled cap and the soil is analysed, as
shown in Fig. 11. The Young’s modulus and Poisson’s ratio of the soil
6.2. Raft resting on a semi-infinite medium are Es = 3000 kN/m2 and 𝜈 s = 0.5, respectively. The pile has a diame-
ter D = 0.8 m and a length L = 8 m, while the plate is a 2 m × 2 m square
The problem considered here is illustrated in Fig. 10. A square raft with thickness h = 0.5 m. Both the pile and the plate are admittedly rigid
with a 12 m edge and a 0.1 m thickness is modelled using the FEM with with a Young’s modulus Ep = 106 Es and a Poisson’s ratio 𝜈 p = 0.2. Over
a Young’s modulus of Ep = 9.783 × 1013 N/m2 and a Poisson’s ratio of the entire area of the plate, the vertical and uniformly distributed load
𝜈 p = 0.3. The vertical uniform load g = 1 N/m2 is applied to the raft. The Q = 200 kN/m2 is applied.
midpoint of one edge is denoted as A, and point B represents the geo- The top of the pile presents a vertical displacement of 4.20 cm. This
metrical centre of the raft. The Young’s modulus and Poisson’s ratio of result is highly consistent with the vertical displacement of 4.13 cm ob-
the soil are Es = 2.6 × 108 N/m2 and 𝜈 s = 0.3, respectively. tained in Ref. [3] and Ref. [10]. Considering the s-axis shown in Fig. 12,
The values in Ref. [16], where a pure BEM formulation was em- the displacement along the plate is practically constant. It is also ver-
ployed, are the lowest (Table 2). As in this study, the other formulations ified that there are no changes in the results if the distributed load
used a mixed BEM/FEM technique. In Ref. [26], a mesh with 100 quadri- Q = 200 kN/m2 is replaced by concentrated force P = 800 kN. In addi-
lateral finite ACM elements was used. The mesh used in Ref. [15] is the tion, the variation in the displacements along the nodes of the pile and

74
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 13. Capped four-pile group.

Fig. 14. Generated mesh.

over the plate is insignificant. This behaviour is expected due to the high Table 3
stiffness of the plate and the pile and the symmetry of the applied load. Lateral displacement (in).
In this particular case, the length/thickness ratio is 4 for the plate, At the top of the pile
and the DKT element is tested and yields good results in comparison
Ref. [13] 1.53
with the ones obtained from the literature. In other examples, the di-
Ref. [12] 1.20
mensions of the plate are at least 10 times higher than the thickness. This work 1.47
Since the Kirchhoff thin plate theory is used, a limitation of the pro-
posed model occurs when the shear strain (and consequently, the shear
stress) becomes considerable during the analysis. This is the case, for
In this example, the presented model performs as well as the pre-
example, in foundation blocks in which the height of the block is signif-
vious formulations when vertical loads are applied. Since the pile was
icant in relation to the other dimensions.
modelled by only one element in the previous formulations developed
The DKT element is chosen to model the bending behaviour of the
by the co-author and co-workers, the consideration of horizontal loads,
plate because, according to ref. [23], it was found to be an effective 9-
such as the ones applied in the next two examples, was not attempted or
DOF triangular element for the bending analysis of thin plates. It can
it was performed inaccurately. In fact, the pile’s flexibility was poorly
be formulated and implemented very efficiently, and it has few storage
accounted for, such as in Ref. [9,11], and the full interaction between
requirements. Its effectiveness has been observed over a variety of static
the structural elements was not considered, as in Ref. [12].
and dynamic analyses.

6.4. Strip footing under a uniform vertical load 6.5. Capped pile group with a horizontal load

The problem considered in this example is shown in Fig. 13. The In this example taken from Ref. [13] and illustrated in Fig. 16, the
strip footing is a 25 m × 2.5 m rectangle with thickness h = 0.079 m; lateral displacement at the top of the piles is calculated. The capped
the pile has a diameter D = 0.5 m and a length L = 50 × D = 25 m. The pile group is subjected to horizontal load HG = 100 kips. Each pile has
Young’s modulus and Poisson’s ratio of the soil are Es = 2 × 106 N/m2 a diameter Dp = 1 ft and a length Lp = 25 ft. The Young’s modulus and
and 𝜈 s = 0.5, respectively. Both the pile and the strip footing are ad- Poisson’s ratio of the soil are Es = 500 psi and 𝜈 s = 0.5, respectively. The
mittedly flexible with Young’s modulus Ep = 2 × 1010 N/m2 and Pois- space between piles is s = 3 ft. The cap is considered rigid and the sys-
son’s ratio 𝜈 p = 0.2. Over the entire area of the strip footing, as shown in tem’s coefficient of flexibility is KR = Ep Ip /Es Lp 4 = 10−3 .
Fig. 14, vertical and uniformly distributed load g is applied. Because the cap is rigid, the displacements at the top of the piles
As plotted in Fig. 15, the results of this study are highly consistent are equal (Table 3). The discrepancy between the results is due to the
with those in Ref. [9], where a BEM formulation was used while consid- different techniques and assumptions under consideration. Broadly, in
ering a thin plate. The same example is also analysed in Ref. [18] with Ref. [12] and in Ref. [13], the interactions between the plate, the pile
the raft modelled using the direct boundary element formulation for and the soil are not fully considered. Assumptions about the restrictions
thick plates, and the results obtained here are also highly consistent at the top of the piles were made, and the modelling of the plate was
with the ones from the present study. not taken into account. In this study, the rigid cap is modelled as a very

75
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Este
Thistrabalho
work Paiva
Ref. (1993)
[9]
1.9
1.8
1.7
1.6
[w/g]*106 (m3/kN)

1.5
1.4
1.3
1.2
1.1
1.0
0.9
0.8
0 1.5625 3.125 4.6875 6.25 7.8125 9.375 10.9375 12.5
Axis AB (m)

Fig. 15. Vertical displacement along the AB-axis.

Fig. 17. Piled raft subjected to a uniform lateral load and the generated mesh.

6.6. Piled raft with a horizontal load

A piled raft containing 9 piles is subjected to uniformly dis-

Fig. 16. Capped six-pile group. tributed lateral load Q = 100 kN/m2 , as shown in Fig. 17. The lateral
displacements at the top of the piles are analysed by considering two
cases, a flexible and a rigid piled raft. The data for this problem are
displayed in Table 4.
stiff plate. The connection between the plate and the piles, as normally Case 1: Rigid piled raft
performed between structural elements analysed by FEM, takes place in Considering a rigid piled raft in comparison to the soil flexibility, the
the stiffness matrix of the plate-pile assembly by adding stiffness coeffi- displacements at the top of the piles are the same (Table 5). This case is
cients at common nodes. similar to the problem analysed in Example 6.5.

76
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

Fig. 18. Lateral displacement along the s-

0.060 axis.

0.059
0.058
Lateral displacement (m)

0.057
0.056
0.055
0.054
0.053
0.052
0.051
0.050
0 2.5 5 7.5 10 12.5 15 17.5 20
Axis s

Fig. 19. Lateral displacement for rigid and

Estaca flexible piles.
Pile 1 1rigid
rígida Estaca
Pile 1 flexível
1 flexible
0.060
0.055
Lateral displacement (m)

0.050
0.045
0.040
0.035
0.030
0.025
0.020
0.015
0.010
0 1.25 2.5 3.75 5 6.25 7.5 8.75 10 11.25 12.5
Depth (m)

Table 4 Table 6
Physical properties of the model. Displacements for Case 2 (m x 10−2 ).

Soil Pile Plate Lateral displacement

2 3 9 9 Pile 1 Pile 2 Pile 3 Pile 4 Pile 5 Pile 6 Pile 7 Pile 8 Pile 9
Young’s modulus [kN/m ] Case 1 3 × 10 3 × 10 3 × 10
Case 2 3 × 103 3 × 105 3 × 105
5.571 5.368 5.255 5.215 5.255 5.368 5.571 5.529 5.322
Poisson’s ratio 0.5 0.2 0.2
Dimensions [m] Semi-infinite D = 1.0 A = 20 × 20
L = 12.5 h = 0.50
coefficient of flexibility. When load Q is applied to the left of the plate,
the leftmost piles (piles 1, 7 and 8) receive the greatest portion of Q, thus
Table 5 showing the largest displacements. One can also note the symmetry in
Displacements for Case 1 (m x 10−2 ). the results in which equal displacements occur for piles 1 and 7, 2 and 6,
and 3 and 5. Lastly, the displacements of the flexible system are roughly
Lateral displacement
twice those of the rigid system (Fig. 19).
All the piles 2.823
7. Conclusions

Case 2: Flexible piled raft This paper presented an efficient BEM/FEM formulation for
In this case, the displacements at the top of the piles vary (Fig. 18 and analysing piled rafts and capped pile groups. The pile is discretized by
Table 6), and the differences may be large or small, depending on the several three-dimensional finite beam elements, allowing for the consid-

77
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

eration of piles of any size that are subjected to any type of loading. The
flexibility of the pile is consequently well accounted for. The plate is dis-
cretized by flat shell finite elements, making it possible to consider the
bending and membrane effects. Under this approach, vertical and hori-
zontal loads are directly applied to the plate, and not, as performed in
some studies, distributed on the top of the piles. The problem is there-
fore analysed in a more realistic way, considering the full interaction
of the plate-pile-soil system, and it is possible to analyse both capped
pile groups and piled rafts of any stiffness. The soil is modelled with
the BEM using Mindlin’s fundamental solution, a particularly suitable
fundamental solution for a consistent analysis of problems involving
semi-infinite mediums. In fact, with this solution, it is only necessary
to model the loaded areas, which, in the present study, are the soil sur-
face discretized by triangular boundary elements or cells and the piles
discretized by line loads, making the analysis of the problem computa-
tionally cost-effective. Lastly, using BEM/FEM coupling, the examples
demonstrated the fast convergence of the numerical solution with rela- Fig. 20. Nodes and elements of the discretisation with the applied load.
tively simple meshes, and the results were highly consistent with those
from other formulations.

Acknowledgements

This study was financed in part by the Coordenação de Aperfeiçoa-

mento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001.

Appendix A. Matrix of the soil’s influence coefficients

A uniform vertical load Qs is applied over an area of the soil surface,

as shown in Fig. 20. The area is discretized into 4 cells or triangu-
lar boundary elements, and the line load is discretized with 2 linear
boundary elements. The objective of the problem is to illustrate the com-
position of matrix [G] of the soil’s influence coefficients, as in Eq. (25).
For the purpose of presenting an example, in considering the vertical
displacement component, the matrices that compose the final matrix [G]
are given as follows:
⎡𝑎ℎℎ 1+4 𝑎ℎ𝑖 1+2 𝑎ℎ𝑗 2+3 𝑎ℎ𝑘 3+4 0⎤
⎢ 𝑎 1+4 𝑎𝑖𝑖 1+2 𝑎𝑖𝑗 2+3 𝑎𝑖𝑘 3+4 0⎥
[ ] ⎢ 𝑖ℎ 1+4 ⎥
𝐺𝑆𝑆 = ⎢ 𝑎𝑗ℎ 𝑎𝑗𝑖 1+2 𝑎𝑗𝑗 2+3 𝑎𝑗𝑘 3+4 0⎥ Fig. 21. Mindlin’s problem.
⎢𝑎𝑘ℎ 1+4 𝑎𝑘𝑖 1+2 𝑎𝑘𝑗 2+3 𝑎𝑘𝑘 3+4 0⎥
⎢ 1+4 ⎥
⎣ 𝑎𝑙ℎ 𝑎𝑙𝑖 1+2 𝑎𝑙𝑗 2+3 𝑎𝑙𝑘 3+4 0⎦
Annex A. Mindlin’s fundamental solution for displacements
⎡𝑏ℎ𝑙 5
𝑏ℎ𝑚 5+6
𝑏ℎ𝑛 ⎤
6
⎢𝑏 5 𝑏𝑖𝑚 5+6 𝑏𝑖𝑛 6 ⎥ Mindlin’s fundamental solution, Ref. [27], for the displacements at
[ ] ⎢ 𝑖𝑙 5 ⎥ point s due to a unit load acting at point f, as shown in Fig. 21, is given
𝐺𝑆𝐿 = ⎢ 𝑏𝑗𝑙 𝑏𝑗𝑚 5+6 𝑏𝑗𝑛 6 ⎥
⎢𝑏𝑘𝑙 5 𝑏𝑘𝑚 5+6 𝑏𝑘𝑛 6 ⎥ by
⎢ 5 ⎥ { ( )
⎣ 𝑏𝑙𝑙 𝑏𝑙𝑚 5+6 𝑏ln 6 ⎦ 3 − 4𝜈 1 𝑟2 (3 − 4𝜈)𝑟21 2𝑐𝑧 3𝑟2
𝑢∗11 = 𝐾 + + 1 + + 1− 1
[ 1+4 ] 𝑟 𝑅 𝑟3 𝑅3 𝑅3 𝑅2
[ ] 𝑐 𝑐𝑚𝑖 1+2 𝑐𝑚𝑗 2+3 𝑐𝑚𝑘 3+4 0
𝐺𝐿𝑆 = 𝑚ℎ1+4 [ ] }
𝑐𝑛ℎ 𝑐𝑛𝑖 1+2 𝑐𝑛𝑗 2+3 𝑐𝑛𝑘 3+4 0 4(1 − 𝜈)(1 − 2𝜈) 𝑟2
+ 1− ( 1 )
[ 5 ] 𝑅 + 𝑅3 𝑅 𝑅 + 𝑅3
[ ] 𝑑 𝑑𝑚𝑚 5+6 𝑑𝑚𝑛 6
𝐺𝐿𝐿 = 𝑚𝑙5 { }
𝑑𝑛𝑙 𝑑𝑛𝑚 5+6 𝑑𝑛𝑛 6 1 3 − 4𝜈 6𝑐𝑧 4(1 − 𝜈)(1 − 2𝜈)
𝑢12 = 𝐾 𝑟1 𝑟2
∗
+ − − ( )2
𝑟3 𝑅3 𝑅5 𝑅 𝑅+𝑅
where xij k + l : is the sum of the influence coefficients at node j common 3
to elements k and l, with the source point at node i { }
𝑟3 (3 − 4𝜈)𝑟3 6𝑐𝑧𝑅3 4(1 − 𝜈)(1 − 2𝜈)
By superimposing the matrices above, [G] is finally determined as 𝑢∗13 = 𝐾 𝑟1 + − + ( )
follows: 𝑟3 𝑅3 𝑅5 𝑅 𝑅 + 𝑅3
⎡𝑎ℎℎ 1+4 𝑎ℎ𝑖 1+2 𝑎ℎ𝑗 2+3 𝑎ℎ𝑘 3+4 𝑏ℎ𝑙 5 𝑏ℎ𝑚 5+6 𝑏ℎ𝑛 6 ⎤ 𝑢∗21 = 𝑢∗12
⎢ 𝑎 1+4 𝑎𝑖𝑖 1+2 𝑎𝑖𝑗 2+3 𝑎𝑖𝑘 3+4 𝑏𝑖𝑙 5 𝑏𝑖𝑚 5+6 𝑏𝑖𝑛 6 ⎥
⎢ 𝑖ℎ 1+4 ⎥ { ( )
⎢ 𝑎𝑗ℎ 𝑎𝑗𝑖 1+2 𝑎𝑗𝑗 2+3 𝑎𝑗𝑘 3+4 𝑏𝑗𝑙 5 𝑏𝑗𝑚 5+6 𝑏𝑗𝑛 6 ⎥ 𝑟22 (3 − 4𝜈)𝑟22 2𝑐𝑧 3𝑟22
3 − 4𝜈 1
[𝐺] = ⎢𝑎𝑘ℎ 1+4 𝑎𝑘𝑖 1+2 𝑎𝑘𝑗 2+3 𝑎𝑘𝑘 3+4 𝑏𝑘𝑙 5 𝑏𝑘𝑚 5+6 𝑏𝑘𝑛 6 ⎥ 𝑢∗22 =𝐾 + + + + 1−
⎢ 1+4 ⎥ 𝑟 𝑅 𝑟3 𝑅3 𝑅3 𝑅2
⎢ 𝑎𝑙ℎ 𝑎𝑙𝑖 1+2 𝑎𝑙𝑗 2+3 𝑎𝑙𝑘 3+4 𝑏𝑙𝑙 5 𝑏𝑙𝑚 5+6 𝑏ln 6 ⎥
[ ]}
⎢𝑐𝑚ℎ 1+4 𝑐𝑚𝑖 1+2 𝑐𝑚𝑗 2+3 𝑐𝑚𝑘 3+4 𝑑𝑚𝑙 5 𝑑𝑚𝑚 5+6 𝑑𝑚𝑛 6⎥
4(1 − 𝜈)(1 − 2𝜈) 𝑟 2
⎢ 1+4 ⎥ + 1− ( 2 )
⎣ 𝑐𝑛ℎ 𝑐𝑛𝑖 1+2 𝑐𝑛𝑗 2+3 𝑐𝑛𝑘 3+4 𝑑𝑛𝑙 5 𝑑𝑛𝑚 5+6 𝑑𝑛𝑛 6 ⎦ 𝑅 + 𝑅3 𝑅 𝑅 + 𝑅3

78
E.S. Luamba and J.B.d. Paiva Engineering Analysis with Boundary Elements 103 (2019) 66–79

𝑟2 ∗
𝑢∗23 = 𝑢 [4] Brown PT, Weisner TJ. The behaviour of uniformly loaded piled strip footings. Soil
𝑟1 13 Fndns 1975;15(4):13–21.
[5] Fatemi-Ardakani B. A contribution to the analysis of pile-supported raft foundations
{ } Ph.D. thesis. Southampton University; 1987.
𝑟3 (3 − 4𝜈)𝑟3 6𝑐𝑧𝑅3 4(1 − 𝜈)(1 − 2𝜈)
𝑢∗31 = 𝐾 𝑟1 + + − ( ) [6] Hain SJ, Lee IK. The analysis of flexible pile-raft systems. Géotechnique
𝑟3 𝑅3 𝑅5 𝑅 𝑅 + 𝑅3 1978;28(1):65–83.
[7] Gazetas G, Mylonakis G. Settlement and additional internal forces of grouped piles
𝑟2 ∗ in layered soil. Géotechnique 1999;48(1):55–72.
𝑢∗32 = 𝑢 [8] Ai ZY, Feng DL. BEM analysis of laterally loaded pile groups in multi-layered trans-
𝑟1 31
versely isotropic soils. Eng Anal Bound Elem 2014;44:143–51.
{ [9] Mendonça AV, Paiva JB. A boundary element method for the static analysis of raft
𝑟23 3 − 4𝜈 6𝑐𝑧𝑅3 8(1 − 𝜈)2 − (3 − 4𝜈)
2
foundations on piles. Eng Anal Bound Elem 2000;24:237–47.
𝑢∗33 =𝐾 + + + [10] Trondi RR, Paiva JB. Boundary element analysis of capped and uncapped pile groups.
𝑟3 𝑟 𝑅5 𝑅
Adv Eng Softw 1999;30(9–11):715–24.
} [11] Mendonça AV, Paiva JB. An elastostatic FEM/BEM analysis of vertically loaded raft
(3 − 4𝜈)𝑅3 − 2𝑐𝑧
2
+ and piled raft foundation. Eng Anal Bound Elem 2003;27:919–33.
𝑅3 [12] Matos Filho RF, Mendonça AV, Paiva JB. Static boundary element analysis of piles
submitted to horizontal and vertical loads. Eng Anal Bound Elem 2005;29:195–203.
where [13] Poulos HG, Davis EH. Pile foundation analysis and design. New York: John Wiley
and Sons; 1980.
( )1
𝑟 = 𝑟𝑖 𝑟𝑖 2 [14] Ribeiro DB, Paiva JB. Analyzing static three-dimensional elastic domains with a new
infinite boundary element formulation. Eng Anal Bound Elem 2010;34:237–47.
[15] Ribeiro DB, Paiva JB. An alternative BE–FE formulation for a raft resting on a finite
( )1 soil layer. Eng Anal Bound Elem 2015;50:352–9.
𝑅 = 𝑅𝑖 𝑅𝑖 2 [16] Paiva JB, Butterfield R. Boundary element analysis of plate–soil interaction. Comput
Struct 1997;64:319–28.
[17] Rashed YF, Aliabadi MH, Brebbia CA. The boundary element method for thick plates
𝑟 𝑖 = 𝑥 𝑖 (𝑓 ) − 𝑥 𝑖 (𝑠 )
on a Winkler foundation. Int J Numer Methods Eng 1998;41:1435–62.
[18] Shehata OE, Farid AF, Rashed YF. Practical boundary element method for piled rafts.
( )
𝑅 𝑖 = 𝑥 𝑖 (𝑓 ) − 𝑥 𝑖 𝑠 ′ Eng Anal Bound Elem 2018;97:67–81.
[19] Zienkiewicz OC, Kelly DW, Bettess P. Marriage a la mode – The best of both worlds
(Finite elements and boundary integrals) Ch 5, 82-107. Energy methods in finite
𝑐 = 𝑥 3 (𝑠 ) > 0 element analysis. Glowinski R, Rodin EY, Zienkiewicz OC, editors. John Wiley and
Sons; 1979.
[20] Luamba ES, Paiva JB. A BEM/FEM formulation for the analysis of piles submitted
𝑧 = 𝑥 3 (𝑓 ) > 0 to horizontal loads. Eng Anal Bound Elem 2017;81:12–20.
[21] Zhang HH, Small JC. Analysis of capped pile groups subjected to horizontal and
1+𝜈 vertical loads. Comput Geotech 2000;26:1–21.
𝐾=
8𝜋𝐸 (1 − 𝜈) [22] Cook RD, Malkus DS, Plesha ME, Witt RJ. Concepts and applications of finite element
analysis. 4th ed. New York: John Wiley & Sons; 2002.
[23] Batoz JL, Bathe KJ, Ho LW. A study of three-node triangular plate bending elements.
Int J Numer Methods Eng 1980;15:1771–812.
References [24] Timoshenko SP, Goodier JN. Theory of elasticity. NewYork, USA: McGraw-Hill;
1970.
[1] Kausel E. Early history of soil-structure interaction. Soil Dyn Earthq Eng [25] Moser W, Duenser C, Beer G. Mapped infinite elements for three-dimensional mul-
2010;30:822–32. ti-region boundary element analysis. Int J Numer Methods Eng 2004;61:317–28.
[2] Poulos HG, Mates NS. Settlement and load distributions analysis of pile groups. Aust. [26] Messafer T, Coates LE. An application of FEM/BEM coupling to foundation analysis.
Geomech. Jnl. 1971;G1(1):18–28. In: Brebbia CA, Connor JJ, editors. Advances in boundary methods. Southampton:
[3] Butterfield R, Banerjee PK. The problem of pile group-pile cap interaction. Géotech- CMP; 1993. p. 211–21.
nique 1971;21(2):135–42. [27] Mindlin RD. Force at a point in the interior of a semi-infinite solid. Physics
1936;7:195–202.

79