You are on page 1of 12


19, 1657-1668 (1983)


University of California, Berkeley, California, U.S.A.

EDS Nuclear, San Franscisco, California, U.S.A

Several new finite elements are presented for the idealization of two- and three-dimensional coupled
fluid-solid systems subjected t o static and dynamic loading. T h e elements are based on a displacement
formulation in terms of the displacement degrees-of-freedom at the nodes of the element. T h e formulation
includes the effects of compressible wave propagation and surface sloshing motion.
The use of reduced integration techniques and the introduction of rotational constraints in the
formulation of the element stiffness eliminates all unnecessary zero-energy modes. A simple method is
given which allows the stability of a finite element mesh of fluid elements to be investigated prior to
analysis. Hence, the previously encountered problems of ‘element locking’ and ‘hour glass’ modes have
been eliminated and a condition of optimum constraint is obtained.
Numerical examples are presented which illustrate the accuracy of the element. It is shown that the
element behaves very well for non-rectangular geometry. The optimum constraint condition is clearly
illustrated by the static solution of a rigid block floating on a mesh of fluid elements.

The dynamic behaviour of a fluid-solid system is a very important problem in both civil and
mechanical engineering. Dam-reservoir interaction during earthquakes and fluid storage
containers subjected to dynamic loads are examples of this class of problem. Applications
involving large displacement fluid flow will not be considered in the formulation presented in
this paper.
Finite elements have previously been developed for the numerical analysis of fluid-solid
systems. These methods can be categorized into the following three basic approaches: added
mass approximations, Eulerian formulation and Lagrangian formulation.
In the added mass approach some fraction of the fluid mass is added to the structural model
at the interface between the two materials.’ Normally the assumption used to evaluate the
added mass is based on a flexible structure and incompressible fluid. This approach neglects
the stiffness effects in the fluid and, in general, leads to conservative results. It is a simple
method which has been used to obtain practical engineering results for both two- and
three-dimensional structure^.^'^ Similar approximate methods have been used for the analysis
of tanks.4
In the Eulerian approach the behaviour of the fluid is formulated in terms of a pressure
. ~ behaviour of the fluid can be expressed in terms of analytic functions (for
p ~ t e n t i a l The
i Professor of Civil Engineering.
$ Senior Engineer.

0029-5981/83/111657-12$01.20 Received 5 July 1982

@ 1983 by John Wiley & Sons, Ltd. Revised 25 September 1982

certain geometries) or a mesh of finite elements with the pressure as the unknown nodal
variable. The solution of the coupled system can be accomplished by solving the two systems
separately with the interaction effects enforced by i t e r a t i ~ n . Both
~ ’ ~ systems can be coupled
and solved as one system; however, these lead to unsymmetrical equation^.^.^ This approach
generally requires a special-purpose computer program. Also, the coupled equations at the
interface cause the finite element equations to have a large bandwidth. This method is currently
the most common and practical approach to many fluid-solid systems.
In the Lagrangian approach, the behaviour of the fluid is expressed in terms of the
displacements at the finite element node points. Hence, compatibility and equilibrium is
automatically satisfied at the nodes along the interfaces between the fluid and solid structure.
The major advantage of this approach is that the element can be readily incorporated into a
general-purpose computer program for structural analysis, since special interface equations
are not required.
Many different Lagrangian fluid elements have been proposed.”.” Most of these elements
suffer from the existence of zero-energy deformation modes. For many problems these modes
are not excited and practical engineering results have been obtained. However, the finer the
mesh the greater is the tendency for the zero-energy ‘hour glass’ or ‘circulation’ modes to be
excited. These zero-energy modes are generated due to zero shear modulus in combination
with the use of reduced integration method^.'^"^ The use of higher order integration tends
to increase the stiffness of the fluid elements and to eliminate modes which are necessary to
capture the low frequency sloshing behaviour. In Reference 14, global rotational constraints
are introduced in order to correct these spurious modes. However, this approach does not
use reduced integration to produce a single element stiffness matrix.
In this paper a new Lagrangian method is introduced which selectively eliminates the
zero-energy modes and produces a fluid element with optimum behaviour. The method involves
the introduction of the ‘constraint’ of zero fluid rotation at the integration points. For a
nine-node Lagrangian fluid element with 2 x 2 integration, the addition of four rotational
constraints at the integration points produces one of the basic elements which is presented in
this paper.
The behaviour of the element is demonstrated by several two-dimensional examples and
the results compared to other solutions or experiments. In addition, the appropriate integration
order and rotational constraints are given which can be used to construct a three-dimensional
fluid element.


The proposed element is based on a formulation in which the fluid strains are calculated from
the linear strain-displacement equations. The only strain energy considered is associated with
the compressibility of the fluid. The kinetic energy is based on the movement of the fluid
particles as approximated by the finite element displacement functions. The change in the
potential energy due to the low frequency sloshing of the fluid system is included in the
formulation. Also, the displacement field is constrained to be irrotational by the introduction
of a rotational stiffness.
In terms of the three-dimensional elasticity, the pressure volume relationship for a linear
fluid is given by

where the pressure p is equal to the magnitude of the mean stress. The constant CII is the
bulk modulus of the fluid. The volume change e can be expressed in terms of the displacements
by the following strain-displacement equation:
e = U,, x + U,, y + U,, z (2)
where u,,j is the partial derivative of the ith component of displacement with respect to the
j direction.
In order to enforce the rotational constraints the following rotations are defined:

ez = (Ux,Y - U,, x ) / 2
The forces and stiffnesses associated with these rotations are defined by
px = C22exr py = C33ey, and p z = C44ez (4)
where Ciiis a constraint parameter. The selection of Ciiwill be illustrated by an example.
The total strain energy of the fluid system is written as

n,=: I eTCedV

where eT= [ee,eyez]and the diagonal terms of the 4 x 4 C matrix are the previously defined
bulk modulus and constraint parameters. A selection of a large constraint parameter will cause
the rotation and the strain energy associated with the rotation to approach zero.


An important behaviour of fluid systems is the ability to displace without a change in volume.
For reservoirs and storage tanks this movement is in the form of sloshing waves in which the
displacement is in the vertical direction. Those sloshing waves, in a steady-state condition,
involve a harmonic interchange of the kinetic and potential energy of the fluid system.
The low frequency sloshing behaviour of a fluid system involves incompressible modes of
displacements which result in relatively large vertical surface displacements. This sloshing
mode of motion is shown in Figure 1. The weight of a column of fluid of area dA and height
D + Us is given by
df =w ( D + U s )dA (6)

Figure 1. Potential energy of sloshing waves


The average vertical displacement of the fluid column is UJ2; therefore, the increase in the
potential energy of the system is

n,=$ I U,w(D+U,)dA


n, = $ I I
UswUsd A +$ wDU, d A

The first integral will produce surface stiffness terms. The second integral represents the weight

of the fluid which is normally evaluated as an element volume integral rather than as a surface
in te gra].

The kinetic energy of the fluid system is given by

I m(u;+u:+vf)dV

where m is the mass density of the fluid and u1 is the ith component of the velocity field.
Equation (8) can be written in the following matrix form:

T = $ mvTvdV (9)

vT = [ u x v y u z ]


Within a typical element i the displacements are expressed in terms of the nodal displacements
by equations of the form

where h is an 1x n array of interpolation functions for a n node element and H is a 3 x n

array for a three-dimensional system. The application of equations (2) and (3) yield the
following equation for the element volume changes and rotations in terms of the nodal
ei = B,U (11)
The strain energy of the system, equation (3,can now be expressed in terms of element
properties and nodal displacement by
n e
z UTK U (12)

The surface potential energy, equation (7), is expressed in terms of the vertical node displace-
ments at the surface as
n, = ;UTsu, (13)
and the kinetic energy, equation (9), can be written in the form
T = 'ZV MV

where U and V are vectors of nodal displacements and velocities. The symmetric matrices K,
S and M are defined in Table I.
The direct application of Lagrange's equation yields the following set of equations:
Ma+ KU + SU, = R
Ma+ K*U = R
where R is a vector of time-dependent nodal forces.
A summary of all equations required to completely describe the static of dynamic behaviour
of liquid systems idealized as a finite element system is given in Table I. The major advantage

Table I. Summary of fluid element formulation

Governing equation in matrix form

M a ( t ) + K u ( t ) + S u , ( t=) R ( t )
where u ( t )= vector of node point displacements
u , ( t )=vector of vertical displacements at surface nodes
a(t) = vector of node point accelerations
R ( t )=vector of node point loads (gravity, wind or floating objects)
S =symmetric matrix of surface potential terms which is defined by

S = 1 S, in which S, = w
I hTh, dA,

Summation j is over horizontal surface elements only;

w is the unit weight of the fluid
K = symmetric matrix of compressibility and rotational constraint terms

K = 1 K, in which K, =

= symmetric
I BTC,B, dV,

matrix of node point mass terms which is defined

M = 1 M , in which M , = rn
J HTH, d V ,

Summation i is over all fluid elements;

rn is the unit mass of the field
Numerical integration
Four-node quadrilateral-1 point on all terms
Nine-node q u a d r i l a t e r a l 4 point ( 2 x 2) on all terms
Eight-node 3D fluid element-1 point on all terms
Twenty-seven 3D fluid element-8 point (2 x 2 x 2 ) on all terms

of this displacement-based formulation is that it is compatible with the direct stiffness formula-
tion used for the analysis of solids. Therefore, liquid-solid systems can be modelled with the
same type of elements.

The equations given in Table I represent a correct formulation for the small displacement
dynamic behaviour of a compressible, irrotational fluid with surface sloshing. As the finite
element mesh is refined and as the rotational stiffness is increased the solution will approach
an exact solution of the fluid mechanics problem. If the integrals are evaluated exactly, the
finite element solution represents a lower bound solution; therefore, the finite element
displacements (and frequencies) will be lower than the exact solution. Because of the slow
rate of convergence this approach has not been found to be practical.
The technique of reduced integration can be used to reduce the stiffness of finite element
systems. However, care must be taken not to introduce zero-energy modes. For one-point
integration the four-element fluid model shown in Figure 2 has five zero-energy modes if
surface sloshing and the rotational constraints are eliminated. The addition of one-point
integration on the rotational constraint terms eliminates four zero-energy modes and the
addition of surface potential terms will convert the last zero-energy mode to a low frequency
sloshing mode. Therefore, the system is stable and zero-energy modes d o not exist.
For each integration point one non-zero frequency is produced due to the compressibility
of the fluid. Also, one high frequency is produced for each rotation constraint at each integration
point. Therefore, the remaining possible frequencies will be zero or associated with surface
sloshing modes.

0 Nine Nodes-18 Possible NINE MODESHAPES

I sloshing mode
0 9 Boundary Conditions 4 volume chonge modes
@4 Integration Points 4 high frequency rotational modes

@ @
3) 0 C

@ @
P n

For exampre, consider a three-dimensional tank idealized by eight 8-node elements. The
3 x 3 x 3 mesh has 27 nodes or 81 displacements. The four sides and the bottom will introduce
45 (5 x 9) boundary restraints. The 8 integration points will produce 8 compressive modes
and 24 (3 x 8) high frequency rotational modes. Therefore, four low frequency sloshing modes
will exist, or

4 sloshing modes
8 compressive modes
24 rotational constraints
45 boundary conditions
81 nodal displacements
It is clear that this simple approach allows the stability of two- and three-dimensional fluid
meshes to be established prior t o analysis.

The theory presented in this paper was tested using the two-dimensional nine-node element.
T3e major advantages of the nine-node element over the four-node plane element are:
1. Improved accuracy with very little increase in numerical effort.
2. Compatibility with the 9- and 27-node solid elements which can be used to model bending
in plates, beams and shells.
The examples in this section are intended to illustrate the basic behaviour and accuracy of
this element when subjected to static or dynamic loading.

Eigenvalues of single nine-node element

The purpose of this example is to illustrate the basic deformation modes of the element
and to present a rational method for the selection of the rotational constraint parameters C22,
C 3 3 and C44.The eigenvalues and vectors of a square water element (18 degrees-of-freedom)
without boundary restraints were calculated using two values of the constraint parameter C22.
Positive surface potential terms at the top surface and negative terms at the bottom surface
were introduced; therefore, a vertical rigid body translational mode exists and the sloshing
modes are relative to the bottom of the element. The eight zero-energy modes are shown in
Figure 3(a). Note that these modes d o not have rotation at the integration points. The next
two modes are low frequency, involving surface displacements, and are shown in Figure 3(b).
The next four modes are shown in Figure 3(c) and involve volume change deformations. The
modes associated with the four highest frequencies are shown in Figure 3(d). All these modes
involve rotations at the integration points.
Table I1 summarizes the frequencies of the various modes with the rotation parameter, C 2 2 ,
set t o 5 and 100 times the bulk modulus, Clt. We note that large value causes all rotation
frequencies to be an order of magnitude larger than the volume change modes; therefore,
they should not be excited by the low frequency or wave propagation behaviour of the fluid.
Because a larger value of rotational constraint may cause numerical problems, it is recom-
mended that the rotational stiffness term be set to 100 times the bulk modulus for all cases.
More experience may be required before a general recommendation is made on the selection
of the value of the rotation parameter, C2*.The best approach may be to have the user specify

Figure 3(a). Nine-node element-ight zero-energy modes

Figure 3(b). Nine-node element-two low frequency sloshing modes

Figure 3 ( c ) .Nine-node element-four volume change modes


Figure 3(d). Nine-node element-four high frequency rotational modes

it as a material property. It then would be possible for the user to systematically increase its
value until rotation and ‘hour glass’ modes are eliminated.

Table 11. Frequencies of one nine-node element

Mode no. c2,=5.c,, c22 = 100. c,,

1-8 0 0
9 2.046 rad/sec 2.046 rad/sec
10 2.046 2.046
11 980.831 1,387.11
12 1,083.02t 1,600.80
13 1,083,021 1,680.80
14 1,387.11t 1,961.66
15 1,387.11 1,3871.061
16 1,776.57 16,998.03t
17 1,776.57 16,998.03t
18 1,961.66 19,616.641

+ Rotational frequencies.
Bulk modulus CI1= 300,000 psi; density p = 0.000093552 slugs/in3.

Rectangular rigid tank

The mode shapes and frequencies of a rectangular tank of fluid was evaluated. Meshes of
1, 2, 4, 9 and 16 nine-node fluid elements were used. The dimensions of the tank and the
nine-element mesh are shown in Figure 4(a). The first sloshing mode is shown in Figure 4(b).
Note that the mode has zero volume change and very small rotation at the integration points.
Figure 4(c) indicates the percentage errors in the first sloshing frequency and the first impulsive
frequency. Note that only for one nine-node element the error in sloshing frequency is less
than 8 per cent and the error in the first impulsive frequency is very small.

Frequency analysis of wedge-shaped tank

A wedge-shaped tank was idealized in order to illustrate the behaviour of the nine-node
element in a distorted configuration. The fluid was modelled by four nine-node elements, as

- k 4-

(0 1 Dimensions ond Nine Element Mesh

( b ) First Sloshing Mode - W, = 2.237 r o d l s e c

8 0

First sloshing frequency

First impulsive frequency

( c ) Error in Frequencies Versus Number o f Elements

Figure 4. Frequency analysis of rectangular water tank

shown in Figure 5 . Also shown is the fundamental sloshing mode shape and a comparison of
the finite element solution with the closed form solution given in Reference 14. The error of
approximately 1 per cent is excellent, considering the very coarse mesh used.

Static and frequency analysis of floating block

The static analysis of a floating block, with one-half the density of the fluid, is presented
to illustrate the stability and accuracy of the fluid element. The fluid is modelled by the mesh
shown in Figure 6(a). The nodes at the edge of the block have one horizontal displacement


W = 2.003 radlsec W = 1.9803 rad/sec


Figure 5. Frequency analysis of wedge-shaped tank

1 ( a 1 Finite Element Model o f Fluid

( b ) Static DisplacementsduetoWeight of Block

( c ) Fundamental Mode of Fluid-Block System

Figure 6. Analysis of floating block

and two independent vertical displacements. A static analysis of the fluid due to the weight
of the block indicates that the total volume of the fluid is not changed, as shown in Figure
6(b); or, the decrease in potential energy of the block is equal to the increase of the surface
potential energy of the fluid.
The stiffness and mass of block is then added to the surface nodes and a frequency analysis
performed. The fundamental mode of the fluid-block system is shown in Figure 6(c). This
class of problems has not previously been solved by finite element techniques.

Existing computer programs for the finite element analysis of solids can be easily modified
for the analysis of both fluids and solids. The integration formula must be replaced with the
appropriate reduced integration equations; and, the shear modulus equations are replaced
with the rotational constraint parameter and the strain-displacement equations for rotation
must be introduced. The surface potential terms can be modelled by the addition of vertical
stiffness terms at the surface nodes.
The fluid elements presented in this paper are based on a pure displacement formulation.
The previously encountered problems with stability have been eliminated by the introduction
of rotational constraints. With this element it now appears to be possible to solve many new
fluid-solid interaction problems which were not previously possible to solve in the time domain.

1. H. M. Westergaard, ‘Water pressures o n dams during earthquakes’, Trans. A m . Soc. Ciu. Eng. (1931).
2. H. N. Abramson, (Ed.), ‘The dynamic behavior of liquids in moving storage containers’, Report No. SF-106,
National Aeronautics and Space Administration (1966).
3. 0. C. Zienkiewicz, Finite Element Method in Continuum Mechanics, McGraw-Hill, 1967.
4. R. W. Clough, A. Niwa and D. P. Clough, ‘Experimental seismic studies of cylindrical tanks’, Strucf. Eng. J.,
ASCE ST12, December (1979).
5 . J. T. Oden, 0. C. Zienkiewicz, R. H. Gallagher and C. Taylor, Firrife Elements in Fluids, vols. I and 11, Wiley,
New York, 1975.
6 . J. F. Hall and A. K. Chopra, ’Two-dimensional dynamic behavior of concrete and embankment dams including
hydrodynamic effects’, Earthquake Eng. Struct. Dynam. 10, 305-332 (1982).
7. A. K. Chopra and P. Chakrabarti, ‘Earthquake analysis of concrete gravity dams including dam-water-foundation
rock interaction’, Earthquake Eng. Struct. Dynam. 9, 363-383 (1981).
8. R. H. McNeal, R. Citerly and M. Chargin, ‘A new method for analyzing fluid-structure interaction using
MSC/NASTRAN’, Fifth Int. Conf. on Structural Mechanics in Reactor Technology, Paper B4/9, Aug. 13-17
9. A. Combescure, R. J. Gilbert, F. Jeanipierre, A. Hoffman and M. Livolant, ‘Fluid-structure interaction-a
general method in the Ceasant programs’, Int. Conf. on Engineering Applications of the Finite Element Method,
Hovik, Norway, May (1979).
10. E. L. Wilson and A. K. Chopra, ‘Earthquake analysis of reservoir-dam systems’, Fourth World Conf. of
Earthquake Engineering, Santiago, Chile, Jan. (1969).
11. 0. C. Zienkiewicz and P. Bettes, ‘Fluid-structure dynamic interaction and wave forces, an introduction to
numerical treatment’, Int. J. n u n . Meth. Engng, 13, 1-16 (1978).
~ . Des. 42, 41-52 (1977).
12. T. Belytschko, ‘Methods and programs for analysis of fluid-structure systems’, N I A CEng.
13. K. J. Bathe and V. Sonnad, ‘On effective implicit time integration in analysis of fluid-structure problems’, Int.
J. num. Meth. Engng, 15, 943-948 (1980).
14. M. A. Hamdi and T. Ousset, ‘A displacement method for the analysis of vibrations of coupled fluid-structure
systems‘, Int. J. num. Meth. Engng, 13, 139-150 (1978).