International Journal of Engineering & Technology IJET-IJENS Vol:15 No:02 67

On The Use of the Scaled Boundary Finite Element

Method for Dynamic Analysis of Plates
Nivaldo Benedito Ferreira Campos1, Jos Maria Campos dos Santos2
1
Department Of Structural Engineering, School Of Civil Engineering, Architecture And Urbanism, UNICAMP, Av. Albert
Einstein, 951 - 13083-852 - Campinas SP BRAZIL
2
Department of Computational Mechanics, College of Mechanical Engineering, UNICAMP,
Emails: ncampos@fec.unicamp.br1; zema@fem.unicamp.br2

Abstract-- Predicting the dynamic behavior of complex model the dynamic behavior of plates under plane stress
structures is possible by the use of high accurate numerical proposed and evaluated.
models. Generally, structural systems are modeled using methods
such as finite element method (FEM), finite difference method 2. GEOMETRY TRANSFORMATION
(FDM) or boundary element method (BEM). The scaled The concept of scaled boundary finite element method
boundary finite element method (SBFEM) is a semi-analytical
can be explained by using a section of a 2D linear elastic
boundary method that does not requires the knowledge of a
fundamental-solution to make possible its formulation and it is bounded medium (Fig. 1). A scaling center O , located inside
based on FEM. Using in-plane and out of plane motion structural the domain, is chosen in a position from which the whole
dynamic problems and discretizing the boundary with three boundary must be reached. Only the boundary needs to be
nodes quadratic line finite element, in this work the derivation of discretized with line finite elements. A radial coordinate ,
the frequency response function (FRF) by SBFEM is presented. from the scaling center O to the boundary surface, and two
From the FRFs, the natural frequencies and mode shapes can be local curvilinear coordinates , , in the circumferential
obtained by performing a theoretical modal analysis. Numerical directions of boundary surface are defined. The side-face Ae,
studies considering various assumptions for the scaled boundary
finite element method are presented and the results obtained are
obtained by using straight lines to connect finite element
compared with FEM. surface to scaling center O and surface Se compose a pyramid
with volume Ve, which is defined by the scaled boundary
Index Term-- Scaled boundary finite element method, plate, coordinates , , with = 0 in the scaling center and = 1 on
vibration, frequency response function. the boundary. This dimensionless radial coordinate can be
seen as a scaling factor or a characteristic length. The
1. INTRODUCTION geometry transformation from Cartesian coordinates x, y to
The knowledge about a particular structure is contained scaled boundary coordinates , , is performed, which is
in a theoretical model that can be constructed using a equivalent to represent the surface in polar coordinates where
numerical method. A large number of papers written on the the value of the radial coordinate at the boundary is constant.
subject have used the finite element method (FEM) as the base All triangles are assembled by connecting their side-faces,
for numerical model. The boundary element method (BEM) which corresponds to enforcing compatibility and equilibrium,
has been successfully applied to numerical solution of many resulting in the total medium with area A and the closed
engineering problems too, and for some classes of problems boundary S.
has been proved competitive with the FEM (Banerjee and
Butterfield, [1]).
The scaled boundary finite element method (SBFEM)
has been applied to many fields of engineering (Wolf & Song
[2]). This method is based entirely on finite elements, but with
the discretization only of the boundary. It combines the
advantages of BEM and FEM. No fundamental solution is
required, the spatial dimension is reduced by one, the radiation
condition is satisfied exactly at infinity, and it yields a
O
symmetric dynamic-stiffness matrix. Moreover, the boundary
placed in the radial direction to the scaling center does not
need to be discretized. Therefore a correct choice of the
scaling center yields, for example, that cracks and the free
surface of a foundation embedded in half-space can be
modeled without discretization. In the following, a brief
theoretical review of the scaled boundary finite element
formulation for two-dimensional elastodynamics, according to
Song and Wolf [3], will be presented, and a method to use it to Fig. 1. Scaled boundary coordinate system for a 2D domain.

1510802-7575-IJET-IJENS April 2015 IJENS

IJENS
 International Journal of Engineering & Technology IJET-IJENS Vol:15 No:02 68

J x y, y x, ; N u N1 N2 (14)
Denoting points on the boundary with x, y the geometry
in a 2D domain is described as
4. DYNAMIC STIFFNESS MATRIX ON BOUNDARY
x , N x The dynamic stiffness matrix on a surface with constant
(1)
y , N y is defined as
where N() is a shape function, and 0 1 from de scaling R S , u (17)
center to the boundary. where R( ) are the amplitudes of nodal forces and S() is the
dynamic stiffness matrix.
3. ELASTODYNAMIC EQUATIONS
For any surface S with a constant , the virtual work is
The matrix differential equation in the frequency domain
formulated as
for in plane elastodynamics problems can be expressed in
function of stresses and displacement amplitudes u as: w R w T t d S (18)
S
L u 0
T 2
(2)
where t are the amplitudes of the surface tractions, dS the
where is the mass density, and infinitesimal surface and w is the weighting function.
For an arbitrary w( ) and introducing E0, E1 and the
u u x, y
T
ux u y
(3) displacements u( ) in (18), results in
R E0 2 u , E1 u
T T

x y x y
(4) (19)

Equating the right-hand sides of (17) and (19) yields

0
S , u E0 2 u , E1 u
T
x (20)

L 0 (5) Differentiating (20) and adding with (6) results in
y

S , , u S , E1 u , (21)

y x E 2 u 2 M 0 2 u 0
Applying the scaled boundary transformation to the 2D Solving (21) for u(), and substituting the result in (21)
domain geometry, the governing equations can be expressed leads to

S , E
as:
S , E E
1 T

u
1 0 1

E u E E E
T
0 2 0 1 1 (22)
E2 S , , 2 2 M 0 0

E E u M u 0
(6)
T
1 2 2 0 2
being
where S , , S , S , , (23)

and substituting in (22) results in

1
E0
1

1
B1 T D B1 J d (7)
S , E E
1 0
1
S , E E
1
T
2

(24)
E1 B2 T D B1 J d (8) S , S , , 2 3M 0 0
1

1
E2 B2 T D B2 J d (9) For the boundary (=1), the plate differential equation in
1
terms of the dynamic stiffness matrix is given by

S E E S
1 T
N u N u J d S E E
1
M0
T
1
(10) 1 0 1 2
,
2 M0 0 (25)
where D is the elasticity matrix, and 5. STATIC STIFFNESS AND MASS MATRICES ON
B1 b1 Nu (11) BOUNDARY
The static stiffness matrix for the boundary is defined
B2 b2 Nu (12) from (25) by setting = 0:

1
y, 0

y 0
1
K E E
1 0
1
K E E K
1
T
2
, 0 (26)
b1 0 x, ; b2 0 x
J
(13)
J where K = S() is the static stiffness matrix
x, y, x y

1510802-7575-IJET-IJENS April 2015 IJENS

IJENS
 International Journal of Engineering & Technology IJET-IJENS Vol:15 No:02 69
Identifying (26) as a Ricatti equation, it can be solved for
[K ] with the introduction of the Hamiltonian matrix [Z]
K 2

M 0 (37)
defined as which produces the eigenvalues (squared natural frequencies),
1 2, and eigenvectors (mode shapes), .



-1 T 1
E0 E1 E0 As will be shown in the numerical example, the use of
Z (27)
2 1 equation (37) with the mass matrix from equation (36) is valid

1 T
E E E E E
1 0 1 1 0
E only for very low frequency ranges. As the frequency range
increase, the accuracy of the last natural frequencies
Appling a real orthogonal transformation V to Z calculated decrease severely. This precision loss becomes this
S S12 solution inadequate to structural analysis applications.
Z V V S V 11
S 22
(28) In order to overcome this problem the authors suggest
0 another approach to obtain more accurate natural frequencies
and rearranging S in a way that the real parts of the in a larger frequency range. The method is based on the
eigenvalues of S11 are negative and that of S22 are positive and system FRFs obtained directly from the SBFEM dynamic
partitioning conformably V as stiffness matrix S( ) (Eq. 25), instead of M and K matrices.
By re-arranging equation (25) we obtain
V V12
V 11 S
V22
S E M (38)
(29)
1 2
E
1
V21 E S E
T
1 0 1 2 0

the matrix solution of eq. (25) will be
Equation (38) is a Ricatti first order differential equation.
K V21 V11
1
(30) Then, to obtain S() a numerical solution based on a fourth
To determine the mass matrix M, it is convenient to order Hunge-Kuta integration method was used.
express the dynamic stiffness matrix as The frequency response functions matrix between the
displacement response points and the force excitation points
S K 2 M (31) can be obtained from
Substituting (31) in (25), an approximation of M for low 1
H (39)
frequencies can be obtained from: K 2M

K E E I M M E K E I M 0 (32)
1 0
1
0
1
1
T
0 By substituting equation (31) in equation (39), the FRF matrix
will be
H S
Equation (32) can be solved as a Lyapunov equation. 1
(40)
Alternatively, the mass matrix can be determined through the
previous results used to obtain K. Post multiplying (28) by V11 From this numerical FRFs using SBFEM, it is possible to
leads to extract the modal parameters by using well-known modal
E 0
1
K E I V S1
T
11 11 V11
1
(33)
parameter identification methods (Mesquita Neto et al, [4]).
The features and application results of both discussed
approachs are presented in the next section.
Substituting (33) in (32) and pre-multiplying by (V11)T
and post-multiplying by V11 results in
7. NUMERICAL EXAMPLE
I S11
T
m m I S11 V11 M 0 V11
T
(34) In this section the vibration of a clamped-free plate in
plane stress (fig. 2), as presented by Nardini & Brebbia [5], is
where examined. Adopting the dimension rate L/d = 4, E / = 104
m V11 M V11
T
(35) (Youngs modulus E, and mass density ) and Poisson's ratio
= 0.2, the natural frequencies are determined by SBFEM
Using back substitution (34) can be solved for m and the using the eigenproblem solution approach, Eq. (30), (35), and
mass matrix is obtained from (35) as (37). For several boundary discretization, the natural

V
T frequencies are obtained and compared with that from FEM
m V11
1 1
M 11 (36) (Table 1). It is clear that a good agreement is obtained only for
the first frequencies.
6. MODAL PARAMETERS ESTIMATION
Modal parameters can be determined by solving the
dynamic eigenproblem equation

1510802-7575-IJET-IJENS April 2015 IJENS

IJENS
 International Journal of Engineering & Technology IJET-IJENS Vol:15 No:02 70

y
L

P d
x

Fig. 2. Example of discretization with 16 SBFEM 3 points linear elements for a clamped-free plate

Table I
Natural Frequencies of a structure subject to plane stress from SBFEM and FEM

Scaled Boundary Finite Element Method FEM

10 elem 30 elem 60 elem 120 elem (I) 288 elem (II) Error (I) (II) (%)
0.1616 0.1614 0.1613 0.1613 0.1613 0.00
0.8421 0.8297 0.8293 0.8292 0.8268 0.29
1.0554 1.0550 1.0550 1.0549 1.0430 1.14
2.0079 1.9444 1.9424 1.9421 1.9100 1.68
3.4151 3.2374 3.2306 3.2298 3.0973 4.28
3.4542 3.4126 3.4123 3.4122 3.1222 9.29
5.2052 4.6829 4.6640 4.6621 4.3310 7.64
6.2916 5.9616 5.9351 5.9327 5.1685 14.79

Better results are obtained by using the FRFs through the using four FRFs. In fig. 3 it is presented FRFs from SBFEM
dynamic stiffness matrix, Eq. (38). The natural frequencies are and FEM with excitation at point P direction-y, and
extracted by performing a theoretical modal analysis scheme displacement response at the same point and direction, and in

Fig. 3. FRFs using FEM and SBFEM for point P, force excitation and displacement response at y direction.

1510802-7575-IJET-IJENS April 2015 IJENS

IJENS
 International Journal of Engineering & Technology IJET-IJENS Vol:15 No:02 71

Fig. 4. FRFs using FEM and SBFEM for point P, force excitation at y direction and displacement response at x direction.

In table II it is compared the extracted frequencies from methods, with the advantages that are usual for boundary
SBFEM and the calculated values from FEM. It can be methods.
observed a good agreement between them.
REFERENCES
Table II [1] Banerjee, P. K. and Butterfield, R., 1981,Boundary element methods
Natural Frequencies of a structure subject to plane stress from SBFEM and in engineering science, London, McGraw-Hill
FEM [2] Wolf, J. P., and Song, C., 1996, Finite-Element Modeling of
SBFEM FEM Unbounded Media, John Wiley & Sons, Chichester
[3] Song, C. and Wolf, J. P., 1997, The Scaled Boundary Finite-Element
36 Error (I) 288 Error (I) Method Alias Consistent Infinitesimal Finite-Element Cell Method
14
elem. - (II) elem. - (III) For Elastodynamics, Computer Methods in Applied Mechanics and
elem.(I)
(II) (%) (III) (%) Engineering, 147, p. 329-355
0.1615 0.1616 0.06 0.1613 0.12 [4] Mesquita Neto, E. at al , Acoustic eigenvalues analysis by boundary
element methods In: Boundary Element Acoustics, Edited by T. W.
0.8327 0.8322 -0.06 0.8268 0.71 Wu, London, WITpress, 238p., 2000.
1.0432 1.0433 0.01 1.0430 0.02 [5] Nardini, D., Brebbia, C. A., 1983, A new approach to free vibration
1.9355 1.9379 0.12 1.9100 1.34 analysis using boundary elements, Appl. Math. Modelling, Vol. 7, p.
157-162.
3.1232 3.1251 0.06 3.0973 0.84
3.1627 3.1856 0.72 3.1222 1.30
4.4959 4.5454 1.09 4.3310 3.81
5.1741 5.1914 0.33 5.1685 0.11

8. FINAL REMARKS
A review of plane stress elastodynamics scaled boundary
finite element method formulation was presented, and applied
to obtain the FRFs and the natural frequencies for in-plane
motion of a plate. The results shown that the solution using
static stiffness and mass matrices do not produces accurate
natural frequency results for the whole frequency band,
however the use of the dynamic stiffness matrix was proven
capable to determine the FRFs from which the natural
frequencies could be extracted more precisely. Due to the
singularities of the dynamic equation derivatives, the
numerical determination of the FRFs was very time
consuming. It was shown that the SBFEM is suitable to
model the dynamic behavior of plates and to obtain their
FRFs, been an alternative to the traditional numerical

1510802-7575-IJET-IJENS April 2015 IJENS

IJENS