Comput. Methods Appl. Mech. Engrg.

190 (2001) 31713188

www.elsevier.com/locate/cma

Arbitrary LagrangianEulerian formulation for uidrigid

body interaction
Josep Sarrate a, Antonio Huerta a,*, Jean Donea b
a
Departament de Matematica Aplicada III, ETS Ingenieros de Caminos, Universitat Polit
ecnica de Catalunya, Jordi Girona 1,
Edici. C2, E-08034 Barcelona, Spain
b
Aerospace Laboratory-Thermomechanics, University of Liege, rue E. Solvay, 21, B-4000 Li
ege, Belgium
Received 30 July 1998

Abstract

A new formulation for two-dimensional uidrigid body interaction problems is developed. In particular, vortex-induced oscil-
lations of a rigid body in viscous incompressible ow are studied. The incompressible NavierStokes equations are used to describe the
motion of the uid, while it is assumed that the rigid body is mounted on a system consisting of a spring and a dashpot. An arbitrary
LagrangianEulerian formulation (ALE) is used in order to account for large boundary motion. A general formulation for the coupled
problem is obtained by uncoupling the translation motion of the body from its rotational motion and developing a specic algorithm
to eciently handle the nonlinear dependence of the rotations. This general formulation can be easily applied to multi-body problems.
Two numerical examples involving either translations and rotations are presented as an illustration of the proposed methodologies for
uidrigid body interaction. 2001 Elsevier Science B.V. All rights reserved.

Keywords: Fluidstructure interaction; Arbitrary LagrangianEulerian formulation; Vortex shedding; Induced oscillations; Large
boundary motion; Transient NavierStokes

1. Introduction

Over the last decades, technological development in several engineering elds has placed emphasis on
uidstructure interaction [1,3,5,7,9,10]. As a consequence, numerical simulation of coupled hydrody-
namic/structural problems has become one of the most challenging problems in computational uid dy-
namics. In some circumstances, especially when the structure is embedded in a uid and its deformations
are small compared to the displacements and rotations of its center of gravity, it is justied to idealize the
structure as a rigid body resting on a system consisting of springs and dashpots. Typical situations in which
such an idealization is legitimate include the simulation of wind-induced vibrations in high-rise buildings or
large bridge girders, the cyclic response of oshore structures exposed to sea currents, as well as the be-
havior of structures in aeronautical and naval engineering where structural loading and response are
dominated by uid induced vibrations.
In most cases, oscillations are due to a vortex shedding process around a structure that can be considered
as a rigid body. As vortices are shed, the pressure distribution on the body is modied and time-dependent
dynamic forces appear, hence inducing structural vibrations. The rigid body motion interferes with the ow
pattern through a nonlinear interactive process and it is possible that, for a specic range of Reynolds

*
Corresponding author. Tel.: +34-93-401-6916; fax: +34-93-401-1825.
E-mail address: [email protected] (A. Huerta).

0045-7825/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 5 - 7 8 2 5 ( 0 0 ) 0 0 3 8 7 - X
3172 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

numbers, the shedding frequency becomes close to one of the natural frequencies of the immersed body. In
these cases, resonance occurs, both frequencies collapse to the same value and the rigid body motion, as
well as the dynamic forces can become very large. This phenomenon, known as lock-in, can be critical for
the structural design, because the ow-induced dynamic forces have a direct inuence on the possible
catastrophic failure of the structure under consideration. It is, therefore, of prime importance from the
viewpoint of structural safety to determine in which range of Reynolds numbers resonance can appear and
to appraise the ensuing oscillation amplitude.
In the numerical simulation of uidstructure interaction, see for instance [6,7,9,1114], one of the
most critical issues is the treatment of the nonlinear convective terms in the NavierStokes equations
governing the uid motion. However, when considering vortex-induced oscillations of rigid bodies, the
situation is further complicated due to the appearance of two geometric nonlinearities. The rst one
arises from the explicit dependence of the governing equations on the rotation angle of the rigid body.
The second is related to the interface between the uid and the solid domains. The interface is part of
the boundary of the uid domain, its position is unknown a priori, and its motion may be relatively
large. Thus, its location is governed by the incompressible NavierStokes equations with large
boundary motion, the rigid body equilibrium equations, and the restrictions imposed by the fact that
the structure is rigid.
To solve the rst problem, we introduce in the present paper a new algorithm that allows to analyze the
uidrigid body interaction problem by uncoupling the translation part of the rigid body motion from its
rotational part. We also develop a specic algorithm, based on the Sherman and Morrison lemma [4], to
handle properly the nonlinear dependence of the rotations. For solving the second problem above, we
propose using the arbitrary LagrangianEulerian (ALE) formulation [5,7,9]. It allows to combine the ease
of a Lagrangian description for treating the solid surface and for enforcing the compatibility and equi-
librium conditions between the uid and the rigid body. And, at the same time, it exploits the superiority of
an Eulerian description for dealing with the possible distortions of the hydrodynamic domain far away
from the moving boundaries. Furthermore, the freedom oered by the ALE description to move the mesh
of nodal points independently of the material particles represents an easy way of adapting the uid mesh in
response to the solid movement.
The outline of the remainder of the paper is as follows. In Section 2 we introduce the problem
statement by describing the equations governing the transient motion of the uid and of the embedded
rigid body. Then, we specify the compatibility conditions at the uid/structure interface and briey
discuss the algorithm used for displacing the mesh points in the ALE domain. In Section 3 we develop a
new algorithm to properly handle the nonlinearities due to explicit dependence of the ow equations on
the rotation angle of the immersed rigid body. Finally, in Section 4, we present and discuss our numerical
results.

2. Problem statement

2.1. Fluid motion equations

Consider a two-dimensional rigid body immersed in an incompressible uid. The motion of the uid is
governed by the incompressible NavierStokes equations which are given as follows in the ALE descrip-
tion:

ov
v ^v
rv rp mr2 v f b ;
ot
r
v 0;

where v is the uid velocity, ^v the uid mesh velocity, p the uid pressure divided by the density, m the
kinematic viscosity, and f b represents the body forces. Appropriate initial and boundary conditions are
assumed to be given.
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3173

After spatial discretization by the nite element method, the unsteady NavierStokes equations de-
scribing the motion of a viscous incompressible uid are expressed as follows in the ALE formulation [7]:

Ma gv ^ K l v
v Gp f ; 1
T
G v 0; 2

where M is the mass matrix, g represents the nonlinear convective terms depending on the relative velocity
v v ^ between uid velocity and mesh velocity, K l the uid viscosity matrix, G the gradient matrix and G T
the divergence matrix, f the global vector of external loads applied on the uid, v, a and p the global vectors
^ is the global nodal
listing the nodal values of velocity, acceleration and pressure, respectively, and nally, v
vector listing the components of the uid mesh velocity.
Vectors a, v and f are decomposed in a similar manner to [14],
n o n o n o
aT af ; as ; ~
ad ; vT vf ; vs ; v
~d ; f T f~f ; f s ; f d ;

where superscript (f ) indicates values related to nodes placed in the uid or on the Neumann portion of the
boundary, superscript (s ) means values related to uid nodes on the rigid body surface, superscript (d )

indicates values related to nodes on the Dirichlet portion of the boundary and the tilde symbol denotes
the prescribed values. According to this decomposition, the momentum conservation equation (1) can be
rewritten as
0 18 9 0 ff 18 9 8 9 0 1 8 f9
M ff M fs M fd < af = K K fs K fd < vf = < gf = Gf < f~ =
@ M sf M ss M sd A as @ K sf K ss K sd A vs gs @ G s AP f s ; 3
: d; : d; : d; : d;
M df M ds M dd ~
a K df K ds K dd ~
v g Gd f

while the mass conservation equation (2) takes the following form:
8 9
 < vf =
T T T
G f G s G d : vs ; 0: 4
~d
v

Eq. (3) clearly shows that the coupling between the uid and the rigid body has to be achieved through
variables dened on the rigid body surface, namely: as , vs and f s .

2.2. Rigid body motion equation

In the two-dimensional case, the motion of the rigid body has three degrees of freedom dened at the
center of gravity of the body: two translational displacements, d1 and d2 , and one rotational displacement,
h. The rigid body is assumed to be mounted on a system composed of elastic springs and dashpots (see
Fig. 1) and its equation of motion is given by

M a Cm Kd u; 5

where M , C and K are the mass, damping and stiness matrices, respectively. If each individual system is
uncoupled from the others, the above matrices are diagonal and possess constant coecients. For the
subsequent developments, we also dene the following vectors (see Fig. 1):

dT fd1 ; d2 ; hg; mT fd_1 ; d_2 ; hg;

_ aT fd1 ; d2 ; hg;
 uT fu1 ; u2 ; u3 g;

where d, m and a are the center of gravity displacement, velocity and acceleration, while u contains the force
and the momentum resultants. In particular cases, some of the above degrees of freedom are xed and
system (5) is therefore reduced to one or two equations for the remaining degrees of freedom.
3174 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Fig. 1. Two-dimensional rigid body model.

2.3. Fluidrigid body compatibility relationship

Compatibility equations are developed in order to relate variables dened at the center of gravity of
the rigid body with those dened on its surface, see [14,15] for details. If the origin of the spatial
coordinates is chosen at the center of gravity, the displacements of a point located at the solid surface
Cs , d c d1 ; d2 , are related to the displacements of the center of gravity, d d1 ; d2 , by d c d
Rc hxc , where xc xc1 ; xc2 denotes the surface nodes coordinates, and Rc h is a matrix that explicitly
depends on the rotation angle h
 
cos h 1 sin h
Rc h :
sin h cos h 1

By time dierentiation it is possible to relate the velocity and the acceleration of the surface points, vc and
ac , to their counterparts m and a dened at the center of gravity

vc T c hm;
6
^ c hm2 ;
ac T c ha T

where matrices
   
1 0 Lc2 h ^ c h 0 0 Lc1 h
T c h and T
0 1 Lc1 h 0 0 Lc2 h

contain the following angle-dependent coecients: Lc1 h xc1 cos h xc2 sin h, and Lc2 h xc1 sin h
^ c has been introduced in order to develop the new formulation properly.
xc2 cos h. Notice that matrix T  c T
^ 2
However, the term Tc hm in (6), in fact, only contains the product of the vector L1 ; Lc2 times the
scalar h2 .
An equivalent relation between the dynamic uid force acting at a point located on the solid surface, f c ,
and its counterpart at the center of gravity, uc , can be determined

T Tc hf c uc : 7
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3175

Relationships (6) and (7) are dened for each point on the rigid body surface. Velocities, accelerations and
forces of uid and rigid body do coincide on the interface because the uid is viscous. Thus, the following
assembled vectors, which are needed in Eqs. (3) and (4), are produced:
8 9 0 . 1
> .. > .
< . >
> = B . C
B C
vs vc B T c h Cm Thm; 8
>
> >
> @ A
: .. ; ..
. .
8 9 0 . 1 0
..
1
> .. > .
< . >
> = B . C B . C
B C B^ C 2 ^
as ac B T c h Ca B T c h Cm Tha Thm ;
2
9
>
> > @ A @ A
: .. > ; .. ..
. . .
8 9
>
> .. >
>
< . >
> = X X
T T hf s . . . T Tc h . . . f c T Tc hf c uc u: 10
>
> >
>
> . ;
: > c c
..

These expressions are the compatibility equations. They relate the variables dened at a node of the uid
boundary with those dened at the center of gravity. Notice that the last equation is in fact a statement of
equilibrium in the solid. It states that the total force and momentum acting at the center of gravity are the
^
resultant of all the actions exercised on the solid boundary. Th and Th are the assembled transformation
matrices. They explicitly depend on the rotation angle h, i.e. Eqs. (8)(10) are nonlinear. It has to be noted
that the transformation matrices are rectangular and of order NS  NG , where NS is the number of un-
knowns on the rigid body surface and NG is the number of unknowns dened at the center of gravity
(NG 6 3).

2.4. Mesh motion description

Since the only moving boundary considered here is the rigid body surface, the mesh motion is dened as
follows: (1) a Lagrangian description is imposed on the rigid body surface in the sense that at the uid/solid
interface the nodes of the uid mesh are constrained to remain attached to the solid nodes during the whole
calculation. The mesh velocity of the uid nodes at the interface is imposed equal to the velocity of the
corresponding uid/solid nodes: v ^ vs Thm; (2) in the outer part of the hydrodynamic domain an
Eulerian description is imposed by stating: v ^ 0; and (3) in an area around the rigid body the mesh
velocity decreases along with the distance to the rigid body surface according to a prescribed law, see [15]
for details.
Note that, if a free surface exists, a Lagrangian description will be imposed on the nodes belonging to it,
so that the computational mesh follows the motion of the uid particles. Consequently, the algorithm for
computing the mesh motion has to be changed accordingly. However, this change will only aect the mesh
motion algorithm, and not the general formulation of the problem.

3. Proposed nonlinear algorithm

3.1. General formulation

In order to simplify the notation, the explicit dependence on the rotation angle will be dropped in the
rest of the present paper. The general formulation for the uidrigid body interaction problem can be
derived after substitution of the compatibility conditions (8) and (9) into Eq. (3), where the component
related to the prescribed Dirichlet values is omitted. This yields the following system of dierential
equations,
3176 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

    ff  f   f   
M ff M fs af K K fs v g Gf
^ 2 p
M sf M ss Ta Tm K sf K ss Tm gs Gs
( )    fd 
f~f M fd d
~
K
~d :
a v 11
fs M sd K sd

From the previous equation, an expression for the forces acting on the rigid body surface, f s , is obtained.
Using this expression of f s and Eq. (10), the rigid body equilibrium Eq. (5) can be rewritten as,
2 8 f 9 8 f9 3

< a = <v =
M  a C m Kd T T 4 M sf M ss M sd ^ 2 K sf K sd K ss 0
Tm gs G s p5;
: d ; : d;
~a ~
v
12
where we have introduced the modied mass and damping matrices
M  M T T M ss T;
C C T T K ss T:

It is important to notice that the order of the matrices M  and C is NG 6 3 (the number of degrees of
freedom of the rigid body). Also, these matrices are no longer diagonal, but they remain symmetric.
In conclusion, the uidrigid body interaction problem can be described by a system of ordinary dif-
ferential equations composed of the rst in equation (11), Eq. (12) and mass conservation, Eq. (4). Re-
arranging the rst two equations into a single system, we obtain:
  f    f    
M ff M fs T a K ff K fs T v 0 ^
M fs T 0
T sf  T sf  T ss ^ 2
T M M a T K C m 0 T M T m
    f    ( )    
0 0 0 g Gf f~f M fd
~a d K fd
~d
P v 13
0 K d T T gs T TG s 0 T T M sd T T K sd

and
  
vf T
G fT sT
G T Gd v
~d ; 14
m

where the velocity compatibility condition (8) has also been included in (14). It has to be noted that the
mass, viscosity and divergence matrices depend on the rotation angle h.

3.2. Fluidrigid body interaction with translation only

As previously noted, if a certain degree of freedom of the rigid body is xed, the compatibility and the
assembled transformation matrices have to be regarded as a reduced system. In particular, if the xed
degree of freedom is the rigid body rotation, then the equation of motion (5) can be written as:

M tt at Ctt m t Ktt dt ut ; 15

where subscript t means translation. Furthermore, the nodal transformation matrices in Eqs. (6) and (7) do
not depend anymore on the rotation angle h. Their order is NG 6 2, and they are equal to identity and null
^ c 0).
matrices, respectively (T c I d and T
Therefore, the assembled transformation matrices are also constant coecient matrices. In fact, matrix
^
Th, see (9), is null and the compatibility conditions for velocity and acceleration are identical

vs T t m t ; as T t at ; T Tt f s u;
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3177

where matrix T t is the assembled transformation matrix for translations, and it is composed by 2  2
identity matrices.
The uidrigid body interaction problem is now described by the following system of ordinary dier-
ential equations, which is similar to (13) and (14). T is replaced by T t and the m2 term has disappeared.
!  !     ( f ) !
M ff M fs T t af K ff K fs T t vf 0 0 0 g Gf
p
T Tt M sf M tt at T Tt K sf Ctt mt 0 Ktt d T Tt gs T Tt G s
( ) ! !
f~f M fd d K fd
~
a ~d ;
v 16
0 T Tt M sd T Tt K sd

  
vf T
G fT sT
G Tt Gd v
~d ; 17
mt

where the modied mass and damping matrices for translation problems are:
M tt M tt T Tt M ss T t ;
Ctt Ctt T Tt K ss T t :

The time integration algorithm used in Eqs. (16) and (17) is standard in viscous incompressible uid
structure interaction, see [2,9]. The basic characteristic of our treatment of problems with translations only
is the particular computational implementation of Eqs. (16) and (17). This implementation is based on a
proper interpretation of three kinds of operations that appear very often in (16) and (17). Such operations
are classied as follows: (1) to pre-multiply a matrix or a vector by the T Tt matrix, (2) to post-multiply a
matrix or a vector by the T t matrix and (3) to pre-multiply a matrix by T Tt and then post-multiply the result
by the T t matrix.
To this end, consider a general block square matrix Ann composed of 2  2 boxes, Aij i; j 1; . . . ; n.
Since matrix T t is also composed of T c matrices of order 2  2 (recall that T c I d ), then
0 10 1
A11 A12 ... A1n Id
B CB C
BA A22 ... A2n C B C
B 21 CB I d C
B CB C
T Tt AT t I d Id ... I d B
B .. .. .. C B C
CB . C
B . ... . . CB .. C
@ A@ A
An1 An2 ... Ann Id
0 1
P
n

B j1A1j C
B C
B C
BP C
B n C
B A2j C
B j1 C
Id Id ... I d B
B
C
C
B . C
B . C
B . C
B C
BP C
@ n A
Anj
j1
!
X
n X
n
Aij :
i1 j1

Therefore, this operation of pre- and post-multiplication is equivalent to add all the rows and columns
corresponding to nodes on the rigid body surface. Similarly, to pre-multiply by the T Tt matrix amounts to
adding all the rows of A, and post-multiply by the T t matrix is equivalent to adding all the columns of A.
3178 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Hence, the implementation of the above three kinds of operation is extremely easy. It is equivalent to
assigning the same equation to each degree of freedom dened on each node on the rigid body surface. In
this case, the assembly process will automatically do all the work.

3.3. Nonlinear algorithm for interactions with rotations

As will become apparent from the developments in Section 3.4, the analysis of uidrigid body inter-
action problems with rotations, which corresponds to the numerical resolution of Eqs. (13) and (14), will
involve the solution of a nonlinear system of the form
    
B 11 b12 h x1 r1 x1 ; h
; 18
bT21 h b22 h h r2 x1 ; h

where x1 is the vector associated with the nodal unknowns in the whole uid domain including the
translation motion of the solid, while h represents the unknown associated with the angular motion of the
solid. The basic idea of the algorithm proposed here relies on the following considerations: (1) the basic
nonlinear equation in Eq. (18) is the scalar equation corresponding to the rotation angle h, and (2) the B 11
matrix, which does not depend on the rotation angle, is well approximated by its lumped form [7,9,13,14].
The rotation angle h can be written, using the second equation of (18), as

1  
h r2 x1 ; h bT21 hx1 19
b22 h

then, substituting in the rst equation of (18), one obtains,

 
b12 hbT21 h r2 x1 ; h
B 11 x1 r1 x1 ; h b12 h: 20
b22 h b22 h

Note that, in the previous equation, the matrix on the left-hand side does not depend on x1 . Thus, the left-
hand side in Eq. (20) is linear on x1 . In fact, a similar structure is also encountered in the algorithm used for
uid and uidstructure interaction problems in [2,9] (i.e. a system of equations with a linear left-hand side
and a nonlinear right-hand side). Our goal here is to obtain an algorithm with an equivalent accuracy and
eciency to the one proposed in [2,9], but dealing now with the extra nonlinearities due to rotations.
Therefore, if it is possible to invert eciently the matrix in (20), the resolution of (18) only incorporates an
extra scalar nonlinear equation, (19), which can be solved efciently.
The system matrix in (20) can be inverted via the Sherman and Morrison lemma [4] giving
  1  
b12 hbT21 h 1 1 b21 hbT21 h 1
B 11 B 111 B B 11 ;
b22 h r 11 b22 h

where B 111 must exist and the condition r 1 bT21 hB 111 b21 h 6 0 must hold.
Under these conditions an algorithm can be devised which uses the same simplication as the one
proposed in [2,9] (i.e. B11 is well approximated by its lumped form, B Lump11 ). Then, the solution of Eq. (20) is
  

1 1 Lump
1 b21 hbT21 h Lump
1 r2 x1 ; h
x1 BLump
11 B B 11 r1 x1 ; h b12 h
r 11 b22 h b22 h

where, from the computational point of view, only vectorvector and vectorscalar products are required.
This is also valid for the computation of r.
It is important to notice that this algorithm can be easily extended to problems with several rigid bodies.
In this case, the scalar equation containing the rotation angle becomes a linear system with as many un-
knowns as rigid bodies. Consistently, the r constant becomes a matrix of the same order.
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3179

3.4. Fluidrigid body interaction with rotations

In order to uncouple the translation motion of the rigid body from its rotational motion, the local
matrices in (6) and (7) are separated into two parts,
   
1 0 Lc2 0 0 Lc1

Tc ^
T ct tcr and Tc 0 ^tcr ;
0 1 Lc1 0 0 Lc2

where the subscript (t ) means translations, the subscript (r ) means rotations. Note that T ct is the nodal
transformation matrix extensively used in Section 3.2. We have also introduced the following vectors
 c  c
L2 ^ L1
tcr and tcr :
Lc1 Lc2

Consequently, the compatibility conditions (8)(10) can be written as:

8 9 0 . ..
1
> .. > .
> . >
> > B . . C
< >
> = B C
B C
vs vc B T ct tcr Cm T t tr m; 21
>
> > B C
> >
> > @ A
: .. >; .. ..
. . .
8 9 0 . ..
1 0
.. ..
1
> .. > .
>
> . >
> B . . C B . . C
< >
> = B C B C
B C B C 2
2
as ac B T ct tcr Ca B 0 ^tcr Cm T t tr a 0 ^tr m ; 22
>
> >
> B C B C
>
> > @ A @ A
: .. >
; .. .. .. ..
. . . . .
8 9
> . >
! !> > .. >
> >
>
T Tt . . . T ct . . . < >
T > = X X
fs fc T Tc f c uc u: 23
tTr . . . tTcr . . . > >
> . >
>
>
>
> c c
: . >
> ;
.

The same decomposition is applied to the rigid body mass, damping and stiness matrices
     
M tt Ctt Ktt
M ; C ; K ; 24
M33 C33 K33

where M tt , Ctt and Ktt are the mass, damping and stiness matrices used in (15) for the translation case,
and the scalar variables M33 , C33 and K33 contain the rigid body rotation parameters. The same decom-
position is applied to the rigid body acceleration, velocity and displacement unknown vectors
8 9 8 9 8 9
< a1 =   < m1 =   < d1 =  
a mt dt
a a2 t ; m m2 ; d d : 25
: ; h : _; h_ : 2; h
h h h

Eqs. (21)(25) must be introduced in the general expressions (13) and (14). Moreover, the equations as-
sociated to the uid motion and the rigid body translation, see Eqs. (16) and (17) in Section 3.2, are ex-
plicitly separated from the rigid body rotation equation. The objective of this decomposition is to identify,
and then treat in a dierent manner, the terms associated with translations and those related to rotations.
The following systems of ordinary dierential equations describe the uidrigid body interaction with
rotations (see [15] for details)
3180 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

      !  !     
M tt mtr at K tt ktr vt ^ tr
0 m 0 ^ tt
K 0 dt gt G tt
p
mTtr mrr h kTtr krr h_ ^ rr
0 m h_2 0 ^
krr h gr gTr
  ! !
ft M dtt K dtt
T
ad
~ T ~d
v 26
0 mdr kdr

and
 
vt T
G Ttt gr Gd v
~d ; 27
h_

where several matrices, vectors and scalars have been introduced:

 f  f   ( )  
a v 0 f~f M ff M fs T t
at  ; vt  ; dt  ; ft  ; M tt  ;
at mt d 0 T Tt M sf M tt
       
M fd K ff K fs T t K fd ^ 0 0
M dtt ; K tt  ; K tt d
; Ktt  ;
T Tt M sd T Tt K sf Ctt T Tt K sd 0 Ktt
   f   
Gf g  T s M fs tr T
G tt  ; g  ; g T
 t G ; m tr  ; mdr  tTr M sd ;
T Tt G s t
T Tt gs r r
T Tt M ss tr
   
K fs tr dT t sd M fs^tr
ktr  ; kr  tr K ; m ^ tr  ; gr  tTr gs ; mrr  M33 tTr M ss tr ;
T Tt K ss tr T Tt M ss^tr

^ rr  tTr M ss^tr ;
m krr  C33 tTr K ss tr ; k^rr  K33 :

Finally, note that: (1) at this point it is trivial to prove that, if the rigid body rotation is xed (no rotations
exist), Eqs. (16) and (17) are a particular case of Eqs. (26) and (27); and (2) the generalized mass and uid
viscosity matrices in (26) and (27) have the same structure as the matrix in Eq. (18). Therefore, the algo-
rithm introduced in Section 3.3 can be applied to this problem.
The details of the practical implementation of time integration algorithm are presented in Appendix A.

4. Numerical examples

Two examples are presented in this section to illustrate the algorithms for uidrigid body interaction
discussed in the present paper. In both cases, we have used an upwind PetrovGalerkin [2] formulation for
solving Eqs. (26) and (27). Bilinear velocity constant pressure (Q1P0) elements were employed for the
spatial discretization.
In the rst example, we analyze the attenuation of the rotational motion of a rectangular cylinder
submerged in a viscous uid (see Fig. 2). The structural damping is ignored in this case. Initially, the uid
and the rectangular cylinder are at rest. The cylinder is then released
pfrom an initial angular displacement
h0 5. The rigid body torsional frequency is taken as xr kr =I 0:266 s 1 . The problem is scaled
using half of the rigid body base: L b=2 1:25 as the characteristic length. The velocity scale is the
maximum linear velocity of point A rotating without damping: V h0 xr sinxr T =2L 2:90245  10 2
(see Fig. 2). Since there is no damping in the system, the potential energy of the spring has to be absorbed
by the uid forces. Therefore, the rotation amplitude of the rectangular cylinder has to decrease with both
viscosity and time.
In order to analyze the rigid body motion dependence on the uid viscosity three values of the kinematic
viscosity, and hence three values of the Reynolds number, were used:
m 0:01 ! Re 3:628
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3181

Fig. 2. Geometry and boundary conditions for the rst example.

Fig. 3. Detail of the mesh for the rst example.

m 0:001 ! Re 36:28

m 0:0001 ! Re 362:8

The mesh employed consists of 3680 nodes and 3535 elements (see Fig. 3).
Fig. 4 shows the evolution in time of the angular displacement, velocity and acceleration of the rigid
body for the above values of the Reynolds number. The solid oscillates around its equilibrium position
and the amplitude of the oscillatory motion decreases with time due to the eect of the hydrodynamic
forces acting on the boundary of the solid. Note a strong amplitude attenuation for Re 3:628
(m 0:01).
Fig. 5 shows the streamlines for the two limit cases: Re 3:628 and Re 362:8. These results present a
reasonable agreement with those obtained in [7] for translational motion: for high Reynolds numbers most
of the shearing occurs in a small layer surrounding the rigid body and the vortices remain attached to the
rigid body surface.
3182 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Fig. 4. Rigid body angular displacements: Re 362:8: ; R 36:28: ; Re 3:628: .

In the second example, the proposed formulation is applied to a uidrigid body interaction problem
with cross-ow and rotational oscillations. Fig. 6 shows the geometry of the problem and the applied
boundary conditions. The computational mesh consists of 3680 nodes and 3535 elements (see Fig. 7). The
non-dimensional parameters are obtained from the rigid body height and inow velocity. For Re 1000
the non-dimensional rigid body parameters are:
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3183

Fig. 5. Streamlines for (a) m 0:01 (Re 3:628) and (b) m 0:0001 (Re 362:8). The snapshots correspond to t p=4, t p=2,
t 3p=4 and t 2p.

mt 195:57; ct 0:0325; kt 0:7864;

I  105:94; cr 0:0; kr 17:05:
Fig. 8 shows the vertical and rotational rigid body displacements, velocities and accelerations corre-
sponding to Re 1000. As can be seen, after the initial transient, both motions only contain one main
3184 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Fig. 6. Geometry and boundary conditions for the second example.

Fig. 7. Unstructured quadrilateral mesh for the second example.

harmonic. In Fig. 9 the stationary streamlines and the pressure eld corresponding to the ultimate instants
are presented.

5. Conclusions

In this paper a new ALE algorithm for the numerical simulation of uidrigid body interaction problems
has been developed. The basic characteristics of the proposed methodology are that: (1) the equations
governing the uid motion and the rigid body translations are uncoupled from the scalar equation gov-
erning the rigid body rotation; and (2) a specic algorithm has been developed for eciently solving the
nonlinear system that appears when rotations of the rigid body are included. It is important to note that the
resulting algorithm possesses the same structure as the standard predictormulti-corrector method used for
the unsteady incompressible NavierStokes equations [2,7,9,14]. Furthermore, it maintains the same
truncation error in time indicating that no loss in the time accuracy has resulted from the coupling between
the uid and a rigid solid. Finally, it is stressed that the present algorithm can be easily extended to deal
with problems incorporating several rigid bodies.

Acknowledgements

This work was partially sponsored by the Ministerio de Educaci

on y Cultura (under grant TAP98-0421)
and by the CICYT (under grant 2FD97-1206).
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3185

Fig. 8. (a) Angular displacement and (b) vertical displacement of the rigid body.

Appendix A. Time marching algorithm

In order to solve the systems of dierential equations (26) and (27) a time discretization algorithm needs
to be introduced. As already indicated, nowadays a standard time integration algorithm used in viscous
3186 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Fig. 9. (a) Stationary streamlines and (b) pressure eld around the rigid body.

incompressible uidstructure interaction is employed, see [2,9]. In this predictorcorrector method [8],
which belongs to the Newmark family, calculations start with the given initial data and the unknowns are
then updated according to:
 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188 3187

am1
t amt Dam1
t ; hm1 hm Dhm1 ;
vm1
t vm cDtDam1 ; h_m1 h_m cDtDhm1 ;
t t
A:1
d m1
t d mt bDt2 Datm1 ; hm1 hm bDt2 Dhm1 ;
p_ m1 p_ m Dp_ m1 ; pm1 pm aDtDp_ m1 ;

where the superscript m is the iteration counter, and Dt is the time step. By introducing (A.1) into (26) and
(27) the following system is obtained,
 ( m1 )    
B c Dat G tt m1 r1
aDt Dp_ ;
eT s Dhm1 g Tr r2
( )
T

Dam1t
cDt G tt gr rp ;
Dhm1

where the following matrices, vectors and scalars have been dened

B  M tt cDtK tt bDt2 K^ tt ;
   2
c  mtr cDt ktr 2h_m m ^ tr hm ;
^ tr c2 Dt2 m
eT  mTtr cDtkTtr ;
   2
s  mrr cDt krr 2h_m m ^ rr hm ;
^ rr bDt2 k^rr c2 Dt2 m A:2
 2
r1  f t M dtt ~
ad K dtt v
~d M tt amt K tt vmt K^ tt d m
t mtr hm ktr h_m ^ tr h_m
m gmt G tt pm ;
 2
mrr hm krr h_m ^ rr h_ k^rr hm
T T
m
r2  mdr ~
ad kdr v
~d m mTtr amt kTtr vmt gmr gTr pm ;
8 9
m
 < vt >
> =
h_m :
T
rp  G Ttt gr Gd
: ~d >
> ;
v

As in the standard method, the acceleration increments are decomposed in the form
Dam1
t Dat Dat ;
Dhm1 Dh Dh ;

where Dat and Dh can be interpreted as the acceleration increments due to the increment of the time de-
rivative of the pressure, while Dat and Dh correspond to the increments due to the residual of the previous
iteration. After some matrix manipulation, see [15] for details, it can be proved that three systems must be
solved at each time step. The rst one reads
     
B c Dat r1
: A:3
e T
s Dh r2

Then, once Dat and Dh are obtained, we have to solve for the increments of the temporal derivative of the
pressure
(    1 )
1 1 1 1 1 
g gT G Ttt g eT B ceT G tt cg T Dp_ m1 r; A:4
s r r s r s s r acDt2 p

where the following residual for the pressure has been introduced:
 
 Dat
rp rp cDt G tt gr :
Dh
3188 J. Sarrate et al. / Comput. Methods Appl. Mech. Engrg. 190 (2001) 31713188

Third, and nally, we have to solve the system

     
B c Dat rat
 ; A:5
eT s Dh rh

where the following residuals have been introduced for the accelerations
 
1 T T
rat  aDt cg r G tt Dp_ m1 ;
s
rh  aDtg Tr Dp_ m1 :

At this point, several aspects of the time-stepping procedure should be highlighted. If a rst-order
approximation is chosen in (A.2), then B M tt . Hence, matrix B can be substituted by the lumped
approximation of M tt , but more importantly, one has c eT and the matrices in Eqs. (A.3)(A.5) are
symmetric matrices.
Moreover, the systems of equations that must be solved in (A.3) and (A.5) have the same structure as the
one in (18). Therefore, the solution technique presented in Section 3.3 can be applied.
 That is, the Sherman
and Morrison lemma [4] can also be applied to invert the matrix B 1s ceT in system (A.4). Actually,
system (A.4) can be rewritten as
    
1 T T 1 T h 1 ^ 1i 1 T 1 
g g G tt ge B B G tt cg Dp_ m1 r;
s r r s r s r acDt2 p
where we have dened

^ 1 1 1 1 T 1
B B ce B :
r s

References

[1] T. Belytschko, J.M. Kennedy, D.F. Schoeberle, Quasi-Eulerian nite element formulation for uidstructure interaction, ASME
J. Pressure Vessel Technol. 102 (1980) 6269.
[2] A.N. Brooks, T.J.R. Hughes, Streamline upwind/PetrovGalerkin formulations for convection dominated ows with particular
emphasis on the incompressible NavierStokes equation, Comput. Meths. Appl. Mech. Engrg. 32 (1982) 199259.
[3] F. Casadei, J.P. Halleux, An algorithm for permanent uidstructure interaction in explicit transient analysis, Comput. Meths.
Appl. Mech. Engrg. 128 (1995) 231289.
[4] J.E. Dennis, J.J. More, Quasi Newton methods motivation and theory, SIAM Rev. 19 (1977) 4689.
[5] J. Donea, S. Giuliani, J.P. Halleux, An arbitrary LagrangianEulerian nite element method for transient dynamic uidstructure
interactions, Comput. Meths. Appl. Mech. Engrg. 33 (1982) 689723.
[6] H. Hirano, A. Maruoka, M. Kawahara, Finite element analysis of uidstructure interaction problem and its application to wind
resistance design of box girder bridges, in: M. Cecchi et al. (Eds.), Proceedings Ninth International Conference on Finite Elements
in Fluids,Venecia, Italy, 1994, pp. 235244.
[7] A. Huerta, W.K. Liu, Viscous ow structure interaction, ASME J. Pressure Vessel Technol. 110 (1988) 1521.
[8] T.J.R. Hughes, The Finite Element Method, Prentice-Hall, Englewood Clis, NJ, 1988.
[9] W.K. Liu, J. Gvildys, Fluid structure interaction of tanks with an eccentric core barrel, Comput. Meths. Appl. Mech. Engrg. 58
(1985) 5177.
[10] W.K. Liu, D. Ma, Computer implementation aspects for uid structure interaction problems, Comput. Meths. Appl. Mech.
Engrg. 31 (1982) 129148.
[11] S. Mittal, T.E. Tezduyar, Massively parallel nite element computation of incompressible ows involving uid-body interactions,
Comput. Meths. Appl. Mech. Engrg. 95 (1992) 10731118.
[12] S. Mittal, T.E. Tezduyar, Massively parallel nite element computation of incompressible ows involving uid-body interactions,
Comput. Meths. Appl. Mech. Engrg. 112 (1994) 253282.
[13] T. Nomura, ALE nite element computations of uidstructure interaction problems, Comput. Meths. Appl. Mech. Engrg. 112
(1994) 291308.
[14] T. Nomura, T.J.R. Hughes, An arbitrary LagrangianEulerian nite element method for interaction of uid and rigid body,
Comput. Meths. Appl. Mech. Engrg. 95 (1992) 115138.
[15] J. Sarrate, Modelizaci on numerica de la interacci on uido-s olido rgido: desarrollo de algoritmos, generaci
on de mallas y
adaptividad, Doctoral Thesis, Universitat Politecnica de Catalunya, Barcelona, 1996.