You are on page 1of 12

Computers and Structures 85 (2007) 932–943

www.elsevier.com/locate/compstruc

Lagrangian finite element modelling of dam–fluid interaction:


Accurate absorbing boundary conditions
Francesco Parrinello *, Guido Borino
University of Palermo, Dipartimento di Ingegneria Strutturale e Geotecnica, Viale delle Scienze, 90128 Palermo, Italy

Received 19 June 2006; accepted 21 November 2006


Available online 19 January 2007

Abstract

The dynamic dam–fluid interaction is considered via a Lagrangian approach, based on a fluid finite element (FE) model under the
assumption of small displacement and inviscid fluid. The fluid domain is discretized by enhanced displacement-based finite elements,
which can be considered an evolution of those derived from the pioneering works of Bathe and Hahn [Bathe KJ, Hahn WF. On transient
analysis of fluid–structure system. Comp Struct 1979;10:383–93] and of Wilson and Khalvati [Wilson EL, Khalvati M. Finite element for
the dynamic analysis of fluid–solid system. Int J Numer Methods Eng 1983;19:1657–68]. The irrotational condition for inviscid fluids is
imposed by the penalty method and consequentially leads to a type of micropolar media. The model is implemented using a FE code, and
the numerical results of a rectangular bidimensional basin (subjected to horizontal sinusoidal acceleration) are compared with the ana-
lytical solution. It is demonstrated that the Lagrangian model is able to perform pressure and gravity wave propagation analysis, even if
the gravity (or surface) waves are dispersive. The dispersion nature of surface waves indicates that the wave propagation velocity is
dependent on the wave frequency.
For the practical analysis of the coupled dam–fluid problem the analysed region of the basin must be reduced and the use of suitable
asymptotic boundary conditions must be investigated. The classical Sommerfeld condition is implemented by means of a boundary layer
of dampers and the analysis results are shown for the cases of sinusoidal forcing.
The classical Sommerfeld condition is highly efficient for pressure-based FE modelling, but may not be considered fully adequate for
the displacement-based FE approach. In the present paper a high-order boundary condition proposed by Higdom [Higdom RL. Radi-
ation boundary condition for dispersive waves. SIAM J Numer Anal 1994;31:64–100] is considered. Its implementation requires the res-
olution of a multifreedom constraint problem, defined in terms of incremental displacements, in the ambit of dynamic time integration
problems. The first- and second-order Higdon conditions are developed and implemented. The results are compared with the Sommer-
feld condition results, and with the analytical unbounded problem results.
Finally, a number of finite element results are presented and their related features are discussed and critically compared.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Absorbing boundary; Dam–fluid interaction; Lagrangian finite element; Dynamic analysis

1. Introduction but the reservoir itself must also be considered. The dam-
foundation interaction is not investigated in this paper
The dynamic structural response of retaining dams is and the foundation soil is assumed as rigid.
strongly affected by the interaction between the fluid and The fluid–structure interaction play an important role in
the dam wall. As a consequence, the structural analysis the dynamic dam wall response, particularly when sub-
of a such a problem cannot be restricted to the dam wall, jected to seismic excitation, due to the potentially disas-
trous consequences of a seismic event. A number of
*
Corresponding author. Tel./fax: +39 0916568448.
numerical approaches based on finite elements have been
E-mail addresses: parrinel@diseg.unipa.it (F. Parrinello), borino@ proposed in literature for the investigation of the fluid–
unipa.it (G. Borino). structure interaction. The most simple model, known as

0045-7949/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2006.11.004
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 933

the ‘‘Added Mass method’’, allows the evaluation of the p;i þ bi ¼ €ui q in V ; ð1Þ
hydrodynamic pressure on the dam wall [1]. It does not
where p is the pressure, q is the mass density, ui is the dis-
require any discretization of the reservoir domain, but pro-
placement component and bi is the volume force compo-
vides acceptable results only in the range of restricted
nent. The following constitutive linear relationship is
hypothesis.
considered:
If a detailed description of the fluid–structure interac-
tion is required, the numerical modelling of the water res- p ¼ Kui;i ; ð2Þ
ervoir is necessary and the coupled analysis of the dam–
where K is the bulk modulus, and ui,i = eV is the volumetric
reservoir system must be performed. Several approaches
strain.
are available in literature for the dynamic analysis of fluids,
Under the hypothesis of neglecting body forces, from
namely: the displacement-based formulation [2–5], the
Eq. (1) follows:
pressure- and displacement potential-based formulation
[7] and the velocity potential-based formulation [8,9]. The 1
€ i ¼ eijk €uj;k ¼ 0;
x ð3Þ
displacement-based formulation is the simplest of these 2
for the analysis of fluid–structure interaction problems, where xi is the vorticity component and eijk is the alternate
not having any special coupling condition requirements symbol. By integrating Eq. (3) we obtain
or any specific resolution strategies. If elements for both
the fluid and the solid domains are both displacement 1
x_ i ¼ eijk u_ j;k ¼ const; ð4Þ
based, with the main variable being the nodal displace- 2
ments, only a standard finite element assembling procedure meaning that the fluid motion is defined as a steady state
is required. In this paper the displacement-based fluid ele- vortex motion, i.e. its velocity vorticity motion is constant.
ment developed by Wilson and Khalvati [3] is used to The current model describes a fluid initially at rest, with a
numerically model the domain of the water reservoir. vorticity velocity remaining constantly at zero, and the
Due to the extremely large size of the fluid domain, with fluid displacement field being irrotational.
respect to the dimensions of the dam wall, the discretiza- In [5] it is demonstrated that the frequency analysis of a
tion of the entire domain can be excessively onerous in liquid volume gives two types of analytical solution,
terms of computer processing resources. As such a resolu- namely: rotational modes, which have frequencies of zero,
tive condition for the numerical analysis might be the dis- and irrotational modes, whose frequencies are positive. In
cretization of a limited portion of the reservoir and the the numerical approaches, as the irrotational condition is
adoption of an adequate boundary condition, designed to generally not verified a priori, it must be imposed. Other-
simulate the presence of the neglected fluid domain. These wise the solution may be corrupted by spurious modes
types of boundary conditions are known in literature as: and the frequency analysis may give a number of zero-fre-
‘‘absorbing boundaries’’, or ‘‘non-reflecting boundaries’’. quencies modes.
Inadequate boundary conditions can lead to errors in the One method of overcoming this obstacle is the penalty
numerical results, both in terms of pressure and displace- approach [2–6], which imposes an elastic constraint to the
ment. The study of the asymptotic boundary condition is rotational component
pursued in this paper and implemented in a finite element
mi ¼ bKxi ; ð5Þ
code.
The paper is organized in the following sections: where mi is the vorticity moment. b is a penalty parameter,
which, in order to obtain neglecting values of rotations xi,
– the Lagrangian fluid finite element model is presented; is assumed in the range 10–1000. For such a material the
– the fluid model is validated comparing numerical and equilibrium equation (1) is not adequate, due to the pres-
analytical results; ence of the vorticity moment, such as (can be seen) in a
– the asymptotic boundary condition is analysed; micropolar material [10]. The correct equilibrium condition
– the Higdom high-order boundary condition is applied is
for the finite-element method;
1
– the numerical results of the asymptotic boundary p;i þ ejki mj;k þ bi ¼ €ui q in V : ð6Þ
condition are presented. 2

2.1. Analytical response of a horizontally excited rectangular


2. Fluid governing equations tank

Let a fluid volume Vf with boundary S = S1 [ S2 be con- The motion or oscillation problem of irrotational fluids
sidered, where S1 is the free surface of the fluid. The fluid is in the small displacement and incompressibility hypothesis
assumed to be inviscid and compressible, and the hypothe- is well established [12,13] and the analytical solution is
sis of small displacements is adopted. For such a problem available for simple problems. The sloshing modes are
the continuum dynamic equilibrium equation is not affected by the fluid compressibility, so the analytical
934 F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943

oscillating liquid in a basin, but do occur in cases of oscil-


lating solid bodies in a fluid domain. The problems ana-
lysed in this paper do not present the mentioned
inconveniencies, and the displacement-based approach of
[3] is adopted and implemented in the finite element code
FEAP [11].
The variational formulation of the dynamic fluid prob-
lem is defined by introducing the following potential energy
functional and kinetic energy term:
Z Z
1 1
P¼ ½ev Kev þ xi bkxi  dV þ uy gquy dS;
Fig. 1. Fluid contained in rigid rectangular tank. V 2 S1 2
Z f
T ¼ qu_ i u_ i dV : ð11a; bÞ
solution can be used to validate the numerical approaches, Vf

despite the two methods having different hypotheses. Invoking the Hamilton principle and by applying a finite
The sloshing oscillations are characterised by the pres- element spatial discretization, the following dynamic equi-
ence of gravity (or surface) waves, which behave in a differ- librium equation is obtained:
ent manner to the acoustic waves. Gravity waves are non-
MU € þ KU þ KS US ¼ R: ð12Þ
conservative and their velocity depends on their
wavelength. where M is the mass matrix, K is the volumetric and rota-
Let a rectangular basin of width b and height h (Fig. 1) tional stiffness matrix, KS is the surface stiffness matrix, U is
be considered in a state of rest, at time t = 0. The sloshing the displacement vector, US is the vector containing the free
modes of a such system [12,13] have the following angular surface vertical displacement and R is the nodal force
frequencies: vector.
1 The reported finite element formulation for fluid
hn ¼ ½gk n tanhðhk n Þ2 ; ð7Þ dynamic analysis is developed for quadrilateral elements
being the wave number kn = 2pn/b. in the plane strain condition. Four-, eight- and nine-node
An horizontal excitation is considered in terms of elements are considered and implemented in the FEAP
imposed displacement, as follows: finite element program [11]. In [3] it is shown that the three
elements are very dissimilar, the element type and the
ux ðtÞ ¼ A cosðhtÞ on S 2 ; ð8Þ
numerical integration order radically affecting the dynamic
where A is the forcing amplitude and h is the angular fre- response of the finite element model. If the bulk and rota-
quency. The analytical solution of the excited tank, in tional stiffness are fully integrated, then locking problems
terms of vertical displacement on the free surface, is given arise and the sloshing modes cannot be described.
as The full integration of consistent mass and surface stiff-
"  # ness matrices, and reduced integration of bulk and rota-
Ah2 b X1
4b
uy ðx; h; tÞ ¼ x cosðhtÞ þ F n ðtÞ cosðk n tÞ ; tional stiffness matrices give better results. Four- and
g 2 m¼1
p2 n2 eight-nodes elements, even if the reduced integration
ð9Þ approach is adopted, produce sloshing frequencies that
are strongly affected by the penalty parameter value b.
where The nine-node elements with reduced integration of bulk
1 and rotational stiffness matrices properly reproduce the
F n ðtÞ ¼ ðh2 cosðhtÞ  h2n cosðhn tÞÞ if h 6¼ hn ; sloshing, acoustic and rotational modes.
h  h2n
2

t In the following paragraph the numerical results of fre-


F n ðtÞ ¼ cosðhn tÞ  sinðhn tÞ if h ¼ hn : quency and step-by-step seismic analysis are compared
2
with the analytical solutions proposed in the previous
ð10a; bÞ
paragraph.

3. Variational formulation and FE model 3.1. Numerical result validation

The numerical solving approach of fluid dynamic prob- The fluid domain of Fig. 1 is modelled with nine-node
lems used in this paper closely follows the approach origi- elements with three different meshes, as shown in
nally proposed by Wilson and Khalvati [3]. Bathe et al. Fig. 2a–c. The following parameters are adopted: b = 2 m,
[5,6,20,21] demonstrated that, with a such method, a num- h = 1 m, q = 1000 kg/m3, K = 2.07 · 1011 N/m2, b = 1000,
ber of inconveniences such as the presence of spurious g = 9.81 m/s2.
modes, may emerge. In those papers is also demonstrated The frequency analysis results obtained by the finite ele-
that the cited problems do not occur in situations such as ment method are compared with the analytical frequencies
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 935

Fig. 2. (a–c) Meshes used in the finite element analysis.

Table 1 face point P in contact with the left wall, see Fig. 1. The
Analytical and numerical angular frequencies of the liquid filled rectan- finite element and the analytical responses are compared
gular tank
in Fig. 3, in terms of vertical displacement at point P versus
Modal Analytical 2·4 4·8 6 · 12 time.
shape sloshing elements elements elements
number frequencies mesh mesh mesh
The results shown in Fig. 3 confirm the suitability of the
adopted model for the analysis of fluid–structure interac-
1 3.76 3.76 3.76 3.76
2 5.54 5.56 5.54 5.54
tion. This model is used in this paper for the study of the
3 6.80 6.91 6.81 6.80 asymptotic boundary condition of fluid dynamic problems.
4 7.85 8.24 7.88 7.86
5 8.78 9.83 8.85 8.79
6 9.62 12.17 9.78 9.65 4. Asymptotic boundary condition
7 10.39 16.58 10.70 10.46
8 11.10 2261.14* 11.66 11.23 Considering that the part of the basin adjacent to the
9 11.78 3.197.73* 12.70 11.98 dam wall has the greatest effects on the fluid–structure
10 12.41 5084.33* 13.90 12.73
interaction, a strategy to reduce the quantity of data
11 13.02 7014.19* 15.35 13.49
12 13.60 7369.64* 17.25 14.28 involved in the simulation, is to consider only a limited
*
portion of the water reservoir, adjacent to the dam wall.
Acoustic frequencies.
The remaining portion of the reservoir is still considered
relevant, but its contribution can be simulated by the so-
obtained by Eq. (7) in Table 1 for the 2 · 4, 4 · 8 and 6 · 12 called ‘‘asymptotic boundary condition’’, which produces
elements discretizations. Table 1 demonstrates a strong satisfactory results. The fluid domain is than bounded by
correspondence between the numerical and analytical the vertical plane S3 (Fig. 4) on which the asymptotic
solutions. boundary condition is imposed.
A second validation test is the step by step analysis of The asymptotic boundary can be considered as a math-
the same rigid tank subjected to the imposed horizontal ematical model, which provides results for the whole fluid
displacements, whose time variation law is defined by Eq. domain, although a significant portion (X2) is excluded.
(8), where A = 0.0036 m and the forcing frequency is one It is applied on the section where the fluid domain is trun-
half of the first sloshing mode frequency, that is cated (S3 in Fig. 4), as a special boundary condition. The
h = 1.88 rad/s. The numerical analysis is conducted using asymptotic model is generally based on the hypothesis that
a Newmark implicit time integration scheme with the time the excluded portion (X2) is infinite and with constant
step Dt = 0.025 s. depth, allowing the assumption that: all acoustic and grav-
The response of the fluid to the horizontal excitation is ity waves travel only from X1 to X2, in a horizontal
recorded in terms of vertical displacement of the free sur- direction.
The latter hypothesis is applied if the boundary surface
S3 absorbs all incoming waves and no reflecting waves are
generated. Such boundary conditions are generally known
as ‘‘absorbing’’ or ‘‘non-reflecting’’.
Several absorbing boundary methods are available in lit-
erature, the majority of which are developed in other
research areas, such as: meteorology, oceanology and naval
engineering. An accurate type of absorbing boundary is
non-local in the time domain [14,15], requiring a convolu-
tion integral, with a large amount of data storage and high
computational cost. Other approaches present non-local
conditions in the space domain, for which the behaviour
of any generic point on the boundary is influenced by the
state of the entire non-reflecting layer. Such an asymptotic
Fig. 3. Analytical and numerical vertical displacement of the point P. boundary condition is very onerous from the computational
936 F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943

Fig. 4. Reservoir domain and asymptotic boundary layer S3.

point of view and is generally not used in the seismic analysis 4.1. Higdom’s non-reflecting boundary condition (NRBC)
of dams. The most widely used method is the so-called Som-
merfeld condition [16], which can be defined by the following The Sommerfeld condition can be considered as the first
equation approximation of a more general non-reflecting boundary
p ¼ qcu_ x þ pst on S 3 ; ð13Þ condition, initially developed by Higdom [17].
Assuming S3 the asymptotic boundary location, whose
where pst is the hydrostatic pressure, q is the mass density normal is x and for a generic variable field /(x, y) (displace-
and c is the wave velocity. Eq. (13) is applied by a forces ment, pressure, etc.), Higdom’s absorbing boundary of
distribution statically equivalent to the hydrostatic pressure order J is defined as
pst and by a layer of dampers with viscous parameter qc " #
(Fig. 5). YJ
ðot þ cj ox Þ /ðx; yÞ ¼ 0 on S 3 ; ð17Þ
The Sommerfeld condition adequately reproduces the j¼1
outgoing wave problem, particularly when the velocity c
is known. But a fluid dynamic problem involving free sur- where cj are parameters and o is the partial derivative oper-
face is characterized by the contemporaneous presence of ator with respect to the index.
acoustic and gravity waves, the first of which presents a The Higdom condition was successfully adopted by
univocally defined velocity Givoli et al. [18,19] for the shallow water equation problem
sffiffiffi using the finite element method and the finite differences
k method. The Higdom condition of order J = 1 for pres-
c¼ : ð14Þ
q sure-based formulation coincides with the Sommerfeld
condition and c1 represents the wave velocity. In fact, for
The gravity waves are dispersive and their velocity depends
J = 1 and /(x, y) = p(x, y), Eq. (17) gives
on frequency and water depth h. More accurately, the grav-
ity wave velocity in an unbounded domain is p_ þ c1 p;x ¼ 0 on S 3 : ð18Þ
h The equilibrium equation (1), for neglecting horizontal
c¼ ; ð15Þ
k body force bx, is equivalent to
where k is the wave number and the angular frequency h is p;x ¼ €ux q on S 3 ; ð19Þ
1
h ¼ ½gk tanhðhkÞ2 : ð16Þ
and substituting in Eq. (18) gives
Moreover, Eq. (13) is obtained assuming true the equilib-
p_  c1 q€ux ¼ 0 on S 3 : ð20Þ
rium equation (1) however, for the displacement-based for-
mulation, the correct form is given by Eq. (6). By integration in time Eq. (20) the Sommerfeld condition
(13) is obtained.
The object of this paper is the study of an alternative
asymptotic boundary condition, with respect to the Som-
merfeld approach, which can be applied to dispersive wave
problems, considering more than one wave velocity. The
proposed NRBC is based on the Higdom approach, but
applies to the displacement components, instead of the
pressure.
The Higdom boundary condition, for a displacement
based formulation, applies Eq. (17) to both the compo-
nents u and v, obtaining an exact response if the set of J
parameters cj contains all the wave speeds si, i = 1, . . . , n.
In fact, as shown in [17], the wave reflection coefficient is
YJ  
 cj  si 
Ri ¼ c þ s  ð21Þ
j¼1 j i
Fig. 5. Sommerfeld condition representation.
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 937

and it can be easily observed that, if ci = si, then Ri = 0; in X9   


1 _
b 2a
other words, no reflected waves are generated in the bound- N i;x ðU i þ DU i Þ þ N i U i þ DU i ¼ 0; ð26Þ
i¼1
c1 bDt
ary by a wave of speed si and the absorbing boundary is
exact. which defines a linear relationship between displacement
In the following section, the Higdom condition is devel- increments DUi and the terms of the previous time step
oped in the first and in the second order for the finite b_ i . The latter equation must be verified on the entire
U i; U
element formulation, considering the Newmark time inte- boundary surface S3, represented in a finite element model
gration method [11], which can be written in the following of Fig. 6, and may be rewritten in the following simplified
form for the horizontal displacement component form:
uðtnþ1 Þ ¼ uðtn Þ þ Du; X
9
2a ai DU i ¼ gx ðtn Þ; ð27Þ
_ nþ1 Þ ¼ ^
uðt u_ þ Du;
bDt ð22a–cÞ i¼1

2 where ai are constant parameters and gx is a value depen-


€uðtnþ1 Þ ¼ €

^ Du;
bDt2 dent on the previous time step solution. When Eq. (27) is
evaluated at a generic point of the element side in contact
where
with the boundary surface S3, then all the parameters ai
2 1b are not equal to zero and a linear equation of the nine de-
^_ ¼ 
u _ nÞ 
uðt uðtn Þ;

bDt b grees of freedom DUi is obtained. Moreover, considering
  ð23a; bÞ
2a 1a that the nodes 7, 8, 9 and 1, 2, 3 also belong to other ele-

u¼ 1
^ _ nÞ þ
uðt Dt€ uðtn Þ;
b b ments, then the equations obtained by applying Eq. (27)
to all the elements of the surface S3 are coupled.
in which Dt is the time increment and a, b are the Newmark The resultant linear equation system must be solved at
integration parameters. In the spatial finite element discret- each time step, with a number of computational difficulties.
ization, where the classical nodal interpolation procedure In order to overcome this inconvenience, Eq. (26) can be
of the displacement field is assumed, the terms of Eqs. evaluated at the nodes 3, 6 and 9. In fact, assuming the
(22a–c) can be written as hypothesis of unwarped elements (rectangular, and with
X
9 the midside nodes at the midpoints) and evaluating Eq.
uðn; g; tnþ1 Þ ¼ ½N i ðn; gÞðU i ðtn Þ þ DU i Þ; (26) at node 3 (Fig. 6), the only shape functions whose first
i¼1 derivates do not vanish are N1,x, N2,x and N3,x. More gen-
X
9 erally, the first derivate of the shape functions Ni,x evalu-
u;x ðn; g; tnþ1 Þ ¼ ½N i;x ðn; gÞðU i ðtn Þ þ DU i Þ; ated on the node j, with j = 3, 6, 9 (following the
i¼1 numeration of Fig. 6), do not vanish only for i = j, j  1,
ð24a–dÞ
X9   
_
b 2a j  2. In mathematical form:
_ g; tnþ1 Þ ¼
uðn; N i ðn; gÞ U i þ DU i ;
i¼1
bDt ðjÞ
N i;x ðnj ; gj Þ ¼ N i;x 6¼ 0; i ¼ j; j  1; j  2;
X9    ð28Þ

b a ðjÞ
uðn; g; tnþ1 Þ ¼
€ N i ðn; gÞ U i þ DU i ; N i;x ðnj ; gj Þ ¼ N i;x ¼ 0; i 6¼ j; j  1; j  2:
i¼1
bDt2
The first-order Higdom condition provides a set of uncou-
where Ni, i = 1, . . . , 9 are the shape functions, n, g are the
pled equations, which can be easily solved at each time
natural coordinates of the isoparametric element, Ui(tn),
i = 1, . . . , 9 are the nodal displacements obtained at the pre-
vious time step and DUi, i = 1, . . . , 9 are the nodal incre-
ments of displacement. Finally, U b_ i ; U

b i ; i ¼ 1; . . . ; 9 are
the nodal values of the terms ^ _ €
u defined in Eqs. (23a,b).
u; ^

4.2. First order Higdom NRBC

Eq. (17) is written for the horizontal displacement com-


ponent with order J = 1, giving
u_
u;x þ ¼0 on S 3 : ð25Þ
c1
By substituting the time and spatial discretized form of the
displacement and velocity field, reported in Eq. (24a–d),
the following relation is obtained: Fig. 6. Finite element discretization of the asymptotic boundary surface
S3.
938 F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943

step. Eq. (26) applied at a generic node of the surface S3 N i;x ðn; gÞ ¼ N i;n ðn; gÞJ 1 1
11 ðn; gÞ þ N i;g ðn; gÞJ 12 ðn; gÞ;
gives ð35Þ
N i;y ðn; gÞ ¼ N i;n ðn; gÞJ 1 1
21 ðn; gÞ þ N i;g ðn; gÞJ 22 ðn; gÞ;
 
ðjÞ ðjÞ ðjÞ 2a
N j2;x DU j2 þ N j1;x DU j1 þ N j;x þ DU j ¼ gðjÞ
x ðt n Þ; where J 1
ij is the inverse of the Jacobian matrix.
bDt
The second derivatives of the shape functions require
j ¼ 3; 6; 9; ð29Þ knowledge of the derivatives of the inverse of the Jacobian
where gðjÞ matrix. However, assuming hypothesis of an unwarped ele-
x ðt n Þ is a term which depends on displacement and
velocity at the previous time step ment, the Jacobian matrix is constant on the whole element
domain, and its derivative vanishes. As such the second
ðjÞ ðjÞ ðjÞ
b_ j
U partial derivates of the shape functions are defined in the
gðjÞ
x ðt n Þ ¼ N j2;x U j2  N j1;x U j1  N j;x U j  : ð30Þ following compact form:
c
   
The procedure followed for the horizontal displacement N i;xx N i;xy N i;nn N i;ng T
¼ J1 J : ð36Þ
may be similarly applied to the vertical component, obtain- N i;yx N i;yy N i;gn N i;gg
ing the following equation:
  By virtue of the assumed hypothesis of unwarped elements
ðjÞ ðjÞ ðjÞ 2a beside the surface S3, if Eq. (34) is applied on the boundary
N j2;x DV j2 þ N j1;x DV j1 þ N j;x þ DV j ¼ gðjÞ
y ðt n Þ;
bDt nodes 3 (Fig. 6), the only shape functions whose first and
j ¼ 3; 6; 9; ð31Þ second partial derivates with respect to x that do not van-
ish are N1, N2 and N3. More generally, considering the
where example of Fig. 6, the firsts two derivates, Ni,x and Ni,xx,
_ evaluated on the node j = 3, 6, 9, do not vanish only for
ðjÞ ðjÞ ðjÞ Vb j i = j, j  1, j  2. The analogous conditions of Eq. (24)
gðjÞ
y ðt n Þ ¼ N j2;x V j2  N j1;x V j1  N j;x V j  : ð32Þ
c can be written for the second derivate:
Eqs. (29) and (30) state a non-homogeneous multifreedom ðjÞ
N i;xx ðnj ; gj Þ ¼ N i;xx 6¼ 0; i ¼ j; j  1; j  2;
boundary condition in terms of incremental displacements, ð37a; bÞ
ðjÞ
which can be solved in a finite element program in exact N i;xx ðnj ; gj Þ ¼ N i;xx ¼ 0; i 6¼ j; j  1; j  2:
form.
The proposed procedure verifies Eq. (25) only at the Considering Eqs. (28) and (37a,b) and observing that
ðjÞ
nodes of the boundary surface, so that the Higdom NRBC N i ¼ dij , with dij the Kronecker delta, Eq. (34) can be
is consistently applied in a discrete finite element form. rewritten as
X j   
ðjÞ c2 þ c1 ðjÞ b_ 2a
4.3. Second order Higdom NRBC N i;xx ðU i þ DU i Þ þ N i;x U i þ DU i
i¼j2
c2 c1 bDt
 
The second order of Eq. (17) applied for the horizontal 1 €
b 2
þ Uj þ DU j ¼ 0; ð38Þ
displacement gives c2 c1 bDt2
 
1 1 1 and in compact form
u;xx þ þ u_ ;x þ u ¼ 0; on S 3 ;
€ ð33Þ
c1 c2 c1 c2
aj2 DU j2 þ aj1 DU j1 þ aj DU j ¼ gðjÞ
x ðt n Þ; j ¼ 3; 6; 9:
and assuming the same spatial and time discretization tech-
ð39Þ
nique presented previously, the finite element form of Eq.
(32) is where aj, aj1, aj2 are constant coefficients defined as
X 9      ðjÞ c2 þ c1 2a ðjÞ 1 2
1 1 b_ i þ 2a DU i ai ¼ N i;xx þ N i;x þ dij ; i ¼ j  2; j  1; j;
N i;xx ðU i þ DU i Þ þ þ N i;x U c2 c1 bDt c2 c1 bDt2
i¼1
c1 c2 bDt
  ð40Þ
1 €
b 2
þ Ni U i þ DU i ¼ 0; ð34Þ
c2 c1 bDt2 whereas gðjÞ x ðt n Þ depends on displacement, velocity and
acceleration at the previous time step
which must be imposed on the boundary surface S3. In or-
X j  
der to generate uncoupled equations, as proposed in the ðjÞ ðjÞ c2 þ c1 ðjÞ b_ 1 b €
gx ðtn Þ ¼ N i;xx U i þ N i;x U i þ U j: ð41Þ
previous section, Eq. (34) is evaluated at the nodes of the i¼j2
c c
2 1 c c
2 1
surface S3, which are represented by the nodes 3, 6 and 9
in Fig. 6. Following the same procedure, an analogous relationship
The second derivative of the shape functions Ni,xx is gen- can be obtained for the vertical displacement component:
erally not considered in the finite element formulation and aj2 DV j2 þ aj1 DV j1 þ aj DV j ¼ gðjÞ
y ðt n Þ; j ¼ 3; 6; 9;
special care must be taken. In an isoaparametric element
the first partial derivatives are ð42Þ
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 939

tionships given by evaluating the same equations respec-


tively on node 9 of element e1 and on node 1 of element
e3. In conclusion, the second-order Higdom condition
provides a set of as many linear uncoupled equations as
there are nodes on the boundary surface S3, which can be
Fig. 7. Semi-infinite horizontal water reservoir.
solved at each time step by a master-slave elimination
approach.
where gðjÞ
y ðt n Þ is

Xj  
ðjÞ ðjÞ c2 þ c1 ðjÞ b_ 1 b € 4.4. Master-slave elimination approach
gy ðtn Þ ¼ N i;xx V i þ N i;x V i þ V j: ð43Þ
i¼j2
c c
2 1 c c
2 1
Both the first- and second-order Higdom conditions can
The relationships obtained evaluating Eqs. (39) and (42) on be considered as multifreedom boundary conditions and
nodes 3 and 9 of element e2 (Fig. 6) are identical to the rela- solved by the master-slave elimination method. For a

Fig. 8. Finite element meshes adopted in the numerical simulations: (a) 40.0 m basin; and (b) 2.0 m truncated basin.

Fig. 9. Hydrodynamic pressure on the left wall. The graphs compare the 40.0 m and 2.0 m basin results: (a) without any asymptotic BC; (b) SBC; (c) first-
order HBD; and (d) second-order HBD.
940 F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943

generic element in contact to the boundary surface S3, the ment, the master-slave elimination approach provides the
following equalities hold: following terms:
aj2 a
DU j ¼  aj
DU j2  aj1 DU j1 þ a1j gðjÞ
x ðt n Þ Ke ¼ TT Ke T
j
j ¼ 3; 6; 9 ð47a; bÞ
DV j ¼
aj2 aj1
 aj DV j2  aj DV j1 þ a1j gðjÞy ðt n Þ
Re ¼ TT ðRe  Ke Ge Þ;
ð44a; bÞ which are, respectively, the stiffness matrix and the residual
vector of the constrained elements. The elements modified
and in matrix form
2 3
by the Higdom boundary condition are assembled as if
2
DU 1
3 1 0 0 0 0 ... ... ... ... 2 3 2
0
3 they were ordinary element.
6 0 1 0 0 0 7 DU 1
6 DV 1 7 6 6 7
76 DV 1 7 6 7 The proposed procedure provides the exact solution, at
6 7 76 7 6 0 7
6 DU 7 6 6
0 0 1 0 0
76 DU 7 6 0 7 each time step, of the first- and second- order Higdom
6 27 6 7 6 7
7 6 0 76 2
6 0 0 1 0 76 DV 2 7 6 7 NRBC. Its application does not require any special solver
6 DV 2 7 6 6 a1 a2 7 7 6 0 7
6 7 6 0  0 0 76 7 6 ð3Þ 7
6 DU 3 7 ¼ 6 a3
6 7 a3 76 DU 4 7 6
7 þ 6 gx 7;
7 and may be implemented in an ordinary finite element
6 DV 7 6 a1 a2 76
6 DV 7 6 gð3Þ 7
6 37 6   7 code. The only requirement is the development of the fluid
6 7 6
0 0 0 76 47 6 y 7
6 ... 7 6
a3 a3 76 7 6 7
6 7 6 . .. 76
6
... 7 6 . .. 7
7 6 7 finite element including the special multifreedom boundary
4 ... 5 6 6 . ..
74
7 ... 5 4 . .. 5 condition presented in this paragraph.
4 a8 5
DV 9 . .. ... ... ... ... ... ... ...  DV 8 gð9Þ
y The proposed Higdom NRBC is implemented in the
a9
finite element code FEAP [11] and the results of a dynam-
ð45Þ
ically excited fluid system are presented in the next
which in compact form is paragraph.
Ue ¼ Te Ue þ Ge ; ð46Þ
5. Numerical simulation
where Ue is the vector of the independent degree of free-
dom of the finite element. Calling Ke and Re respectively In order to test the capability of the proposed asymp-
the stiffness matrix and the residual vector of the finite ele- totic boundary condition, based on the Higdom formulation

Fig. 10. Responses of the 40.0 m basin, the 2.0 m basin with SBC and the 2.0 m basin with second-order HBC: (a) hydrodynamic pressure on the left wall;
(b) vertical displacement of point P1; (c) horizontal displacement of point P2; and (d) vertical displacement of point P2.
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 941

and applied to a Lagrangian fluid finite element model, the are chosen considering the Fourier transformed of the ver-
results of a number of dynamic simulations are shown here. tical and horizontal displacement of the points P1 and P2
The application regards a bi-dimensional reservoir rep- (Fig. 8a), obtained by the 40.0 m basin analysis. The most
resented in Fig. 7, closed on the left side by a vertical rigid relevant frequencies present in the dynamic response of the
wall, and unlimited on the opposite side, with h = 1 m. The fluid are considered and, by Eqs. (15) and (16), the corre-
basin is subjected to a horizontal dynamic excitation, con- sponding wave velocities are evaluated and adopted for
sisting of a sinusoidal acceleration of the foundation the parameter ci. The best results for the Sommerfeld BC
are obtained assuming c = 2.91 m/s, which is the wave
€ug ¼ €
u1 cosðh1 tÞ þ €
u2 cosðh2 tÞ: ð48Þ speed corresponding to the frequency of 2 rad/s. For the
second-order Higdom BC, the best results are obtained
The first analysis is conducted with mono-frequency forc-
assuming the parameters: c1 = 2.91 m/s and c2 = 2.41 m/
ing, assuming € u1 ¼ 1 m=s2 , h1 = 5 rad/s, €
u2 ¼ 0. The analy-
s, which correspond to the frequencies 2 rad/s and
sis is first conducted for a 40.0 m wide basin closed on both
3.7 rad/s. Fig. 10a compares the responses of the 40.0 m
sides with vertical walls. Because this problem presents an
basin and the 2.0 m basin with the Sommerfeld BC and sec-
anti-symmetrical condition, with respect to the median ver-
ond-order Higdom BC, in terms of Hydrodynamic pressure
tical axis, only one half of the problem is analysed. The
on the left wall. In Fig. 10b the same comparison is
analysis is conduced for a period of 10.0 s, in which period
expressed in terms of vertical displacement of the point
the reflected waves do not affect the recorded results. The
P1, and in Fig. 10c and d respectively in terms of horizontal
same analysis is conduced for a 2.0 m wide truncated basin,
and vertical displacement of the point P2.
for different boundary conditions on the right vertical side,
The latter results show that, in the case of multi-fre-
namely: Sommerfeld (SBC), first- and second-order Hig-
quency analysis, the second order Higdom BC produces
dom (HBC) and the anti-symmetrical boundary condi-
higher quality solutions with respect to the Sommerfeld
tions. Fig. 8a and b shows the finite element
BC. This condition is quite evident with low frequency
discretizations used in the analysis, which are defined with
the same element density.
Fig. 9a compares the responses of the reservoir, in terms
of hydrodynamic pressure acting on the left wall of the
40.0 m basin and of the truncated basin with ordinary
boundary conditions, showing that the two responses are
radically different. Fig. 9b–d compare the same result but
for the three different asymptotic boundary conditions.
These figures demonstrate the high quality of the results
with the three adopted asymptotic BCs, where the improve-
ment of the proposed approach is not immediately evident.
The second application is obtained assuming a dual-fre-
quency horizontal excitation, which is governed by Eq. (48)
2
with the parameters € u1 ¼ € u2 ¼ 1 m=s , h1 = 2 rad/s and
h2 = 3.7 rad/s. The analysis is conducted considering the
response of: a 40.0 m basin, 2.0 m basin with SBC and sec- Fig. 12. Finite element mesh used for the analysis of the fluid–structure
ond-order HBC. The parameters ci of Eqs. (13) and (33) interaction problem.

Fig. 11. Hydrodynamic pressure on the left wall: (a) 2.0 m basin with SBC compared to the 40.0 m basin and (b) 2.0 m basin with second-order HBC
compared to the 40.0 basin.
942 F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943

Fig. 13. Response of the dam–reservoir interaction problem with extended domain, truncated domain with SBC and truncated domain with second-order
HB, in term of: (a) horizontal reaction of the dam foundation and (b) vertical displacement of the fluid free surface point in contact with the dam.

forcing, but is less relevant in problems with higher fre- tion proposed by Higdom, to the Lagrangian fluid finite
quency, as shown in the third analysis, which is conducted element formulation. The proposed formulation is derived
2
with the parameters: € u1 ¼ €u2 ¼ 1 m=s , h1 = 10 rad/s, as a multifreedom boundary condition, which is applied to
h2 = 30 rad/s. The results of the third analysis for the the elements containing the truncating surface, and is
2.0 m basin with the two considered boundary conditions, solved linearly at each time step. The finite element numer-
compared with the 40.0 m basin response, are shown in ical simulations have given satisfying results, even when
Fig. 11a and b in terms of hydrodynamic pressure. compared to the well known Sommerfeld condition, espe-
The fourth numerical simulation regards the fluid–struc- cially when the fluid system is excited with low-frequency
ture interaction problem consisting of a retaining concrete forcing.
dam with a fluid reservoir and a height of 100.0 m. The
dam dimensions are: 110.0 m in height, with a base width References
of 20.0 m and a top width of 9.0 m. The concrete is
assumed to be a linear elastic material with Young Modu- [1] Westergaard HM. Water pressure on dams during earthquake. Trans,
ASCE 1933;98:418–33.
lus E = 25,000 MPa, mass density q = 2500 kg/m3 and
[2] Bathe KJ, Hahn WF. On transient analysis of fluid–structure system.
poisson ratio m = 0.2. Comp Struct 1979;10:383–93.
The dam–reservoir system is subjected to horizontal [3] Wilson EL, Khalvati M. Finite element for the dynamic analysis of
ground acceleration with the following temporal law: fluid–solid system. Int J Numer Methods Eng 1983;19:1657–68.
[4] Hamdi MA, Ousset Y, Verchery G. A displacement method for the
ug ¼ €u1 sinðh1 tÞ þ €
€ u2 sinðh2 tÞ ð49Þ analysis of vibration of coupled fluid–structure system. Int J Numer
Methods Eng 1978;13:139–50.
assuming u u2 ¼ 1 m=s2 , h1 = 50 rad/s, h2 = 30 rad/s.
€1 ¼ € [5] Bathe KJ, Nitikitpaiboon C, Wang X. A mixed displacement based
The fluid and dam domains are discretized by nine node finite element formulation for acoustic fluid–structure interaction.
elements in plane strain condition. Com Struct 1995;56:225–37.
[6] Olson LG, Bathe KJ. A study of displacement based finite elements
The problem is analysed considering two different
for calculating frequencies of fluid and fluid–structure system. Nucl
domains of the reservoir: an extended domain with a length Eng Des 1983;76:137–51.
of 1000.0 m and a truncated domain with a length of [7] Morand H, Ohayon R. Substructure variational analysis of the
200.0 m. The extended problem is analysed with the anti- vibration of coupled fluid–structure system. Finite element results. Int
symmetrical BC on the right vertical side. The truncated J Numer Methods Eng 1979;17:741–55.
[8] Everstine GC. A symmetric potential formulation for fluid structure
problem is analysed with the second-order Higdom BC
interaction. J Sound Vib 1981;79(1):157–60.
and the Sommerfeld BC at the right vertical side, and the [9] Olson LC, Bathe KJ. Analysis of fluid–structure interactions. A direct
applied mesh is represented in Fig. 12. The results are com- symmetric coupled formulation based on the fluid velocity potential.
pared in terms of horizontal reaction of the dam founda- Com Struct 1985;12(1,2):21–32.
tion in Fig. 13a, and in terms of vertical displacements at [10] Cosserat E, Cosserat F. Theorie des corps deformables. Paris: Her-
man et fils; 1909.
the point of the free surface in contact with the dam wall
[11] Zienkiewicz OC, Taylor RL. The finite element method. 5th
in Fig. 13b. edition. Oxford: Butterworth-Heinemann; 2000.
[12] Miles J, Henderson D. Parametrically forced surface weaves. Annu
Rev Fluid Mech 1990;22:143–65.
6. Conclusions
[13] Frondsen JB. Sloshing effect in periodically and seismically exited
tanks, WCCM V, 2002, CD Procd.
In this paper we have presented a finite element imple- [14] Bennet AF. Open boundary condition for dispersive waves. J Atmos
mentation of the high-order asymptotic boundary condi- Sci 1976;33:176–82.
F. Parrinello, G. Borino / Computers and Structures 85 (2007) 932–943 943

[15] Feltrin G, Wepf D, Bachmann H. Seismic cracking of concrete dams. [19] Givoli D, Neta B. High order non-reflecting boundary conditions for
Dam Eng 1990;1:276–89. dispersive shallow water equations. J Comp Appl Math 2003;158:
[16] Sommerfeld A. Partial differential equation in physics. New York, 49–60.
USA: Academic Press; 1949. [20] Wang X, Bathe KJ. On mixed elements for acoustic fluid–structure
[17] Higdom RL. Radiation boundary condition for dispersive waves. interactions. Math Models Methods Appl Sci 1997;7(3):329–43.
SIAM J Numer Anal 1994;31:64–100. [21] Wang X, Bathe KJ. Displacement/pressure based mixed finite
[18] Givoli D, Neta B, Platashenko I. Finite element analysis of time element formulations for acoustic fluid–structure interaction prob-
dependent semi-infinite waves-guides with high order boundary lems. Int J Numer Methods Eng 1997;40:2001–17.
treatment. Int J Numer Methods Eng 2003;58:1955–83.

You might also like