Proceedings of ALGORITMY 2009

pp. 246255
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION
PROBLEMS BY FINITE ELEMENT METHOD

SV

CEK P.

Abstract. This paper is devoted to the topic of mathematical modelling and numerical simu-
lation of the interaction of two dimensional incompressible viscous ow and a vibrating structure.
A solid airfoil with two degrees of freedom is considered. The numerical simulation consists of the
nite element solution of the Navier-Stokes equations coupled with the system of ordinary dierential
equations describing the airfoil motion. The time dependent computational domain and a moving
grid are taken into account with the aid of the Arbitrary Lagrangian-Eulerian (ALE) formulation of
the Navier-Stokes equations. High Reynolds numbers up to 10
6
require the application of a suitable
stabilization of the nite element discretization. Here, the modied Streamline-Upwind/Petrov-
Galerkin(SUPG) together with Pressure Stabilizing/Petrov-Galerkin(PSPG) stabilization is applied
and modied within the context of ALE formulation of Navier-Stokes system of equations. The uid
model is coupled with the nonlinear structure model for the solid airfoil. The method is applied on
several technical problems.
Key words. aeroelasticity, nite element method, Arbitrary Lagrangian-Eulerian method
AMS subject classications.
1. Introduction. In many technical disciplines the interaction of uid ow and
an elastic structure plays an important role. The research in aeroelasticity or hydroe-
lasticity focuses on the interaction between moving uids and vibrating structures
[see, e.g., [6], [16]]. Usually only special problems of aeroelasticity or hydroelasticity
are solved, mainly limited to linearized models. The nonlinear postcritical limit states
usually had not been considered, as the appearance of any aerodynamic instability
is not admissible in normal ight regimes. Recently, the modelling of post-utter
behaviour began to be more important.
Nonlinear uid/structure interaction problems arise in many engineering and
scientic applications. During last years, signicant advances have been made in the
development and use of computational methods for uid ows with structural interac-
tions. The more ecient computational techniques were reached with the increasing
computational power, see for example [1]. As the valuable information coming from
uid-structure interaction analysis need to be performed in many elds of industry
(automobile, airplane) as well as in biomedicine, the analysis of uid-structure inter-
action problems become more eective and more general. Since there is a need for
eective uid-structure interaction analysis procedures, various approaches have been
proposed. In current simulations, arbitrary Lagrangian-Eulerian (ALE) formulations
are now widely used. The ALE method is straightforward; however there is a number
of important computational issues, cf. [7], [13], [9], [2].
In this paper, attention is paid step-by-step to the following aspects: second order
time discretization and space nite element discretization of the Reynolds Averaged
Navier-Stokes equations, GLS stabilization of the FEM, the choice of stabilization

This research was supported under grant No. 201/08/0012 of the Grant Agency of the Czech
Republic and under the Research Plan MSM 6840770003 of the Ministry of Education of the Czech
Republic.

Czech Technical University Prague, Faculty of Mechanical Engineering, Karlovo n am. 13, 121 35
Praha 2, Czech Republic, Petr.Svacek@fs.cvut.cz
246
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION PROBLEMS 247
parameters, discretization of the structural model, numerical realization of the non-
linear discrete problem including the coupling of the uid ow and airfoil motion. The
developed suciently accurate and robust method is applied to a technically relevant
case of ow-induced airfoil vibrations.
2. Fluid model. In order to take into account the deformations of the compu-
tational domain, we start with a short introduction of Arbitrary Lagrangian-Eulerian
(ALE) method, see also [18], [17], [9].
2.1. Arbitrary Lagrangian-Eulerian method. Let us assume that there exists
a mapping = (, t) dened for any
0
and t [0, T] such that for any t
the mapping (, t) is a one-to-one transformation of
0
onto
t
. Let us denote
/
t
= (, t). The mapping /
t
is called arbitrary Eulerian-Lagrangian mapping (ALE
mapping). We assume that for any t I the mapping /
t
denotes C
1
continuous
bijective mapping from the reference (original) conguration
0
onto the domain
t
at time t (the current conguration).
The time derivative of the ALE mapping /
t
yields the domain velocity w
D
=
w
D
(x, t) for x
t
and t [0, T].
w
D
(x, t) =

t
(, t), /
t
() = x,
0
. (2.1)
Furthermore, by D
A
/Dt the ALE derivative is denoted (derivative with respect
to a xed point in the reference domain
0
). The ALE derivative is related to
the time and spatial derivatives as
D
A
f
Dt
(x, t) =
f
t
(x, t) +w
D
(x, t) f(x, t), (2.2)
for any x
t
and t (0, T).
2.2. Reynolds equations. We consider the following ALE form of Reynolds
equations in
t
D
A
v
Dt
(
eff
S(v)) + ((v w
D
) )v +p = 0, (2.3)
v = 0,
where S(v) =
_
v + (v)
T
_
,
eff
= ( +
T
), v denotes the vector of mean part
of the velocity, p denotes the mean part of the kinematic pressure, denotes the
kinematic viscosity, and
T
denotes the turbulent viscosity.
The system is equipped with the boundary conditions prescribed on the mutually
disjoint parts of the boundary
t
:
a) v(x, t) = v
D
(x), x
D
,
b) v(x, t) = w
D
(x, t), x
Wt
, (2.4)
c)
eff
S(v) n +p n = 0, x
O
.
Finally, we prescribe the initial condition
v(x, 0) = v
0
(x), x
0
.
248 P. SV

CEK
2.3. Time discretization. We consider a partition 0 = t
0
< t
1
< < T, t
k
=
k, with a time step > 0, of the time interval [0, T] and approximate the solution
v(, t
n
) and p(, t
n
) (dened in
tn
) at time t
n
by v
n
and p
n
, respectively. For the
time discretization we employ a second-order two-step scheme using the computed
approximate solution v
n1
in
tn1
and v
n
in
tn
for the calculation of v
n+1
in the
domain
tn+1
=
n+1
.
We dene for a xed time t = t
n+1
the function spaces J, A by
J = H
1
(
tn+1
), A =
_
z J : z = 0 on
D

Wtn+1
_
,
and space Q = L
2
(
tn+1
). Furthermore, the ALE velocity w
D
(t
n+1
) is approximated
by w
n+1
D
and we set v
i
= v
i
/
ti
/
1
tn+1
. The vector-valued functions v
i
are dened
in the domain
tn+1
.
The second-order two-step ALE time discretization on each time level t
n+1
yields
the problem of nding unknown functions v
n+1
:
tn+1
R
2
and p
n+1
:
tn+1
R
satisfying the equations
3v
n+1
4 v
n
+ v
n1
2
+
_
w
n+1

_
v
n+1
(
eff
S(v)) +p
n+1
= 0, (2.5)
div v
n+1
= 0,
in
tn+1
, w
n+1
= v
n+1
w
n+1
D
, and the boundary conditions (2.4 a,b). The problem
(2.5) is then weakly formulated. For the weak formulation we shall make use of the
forms
a(U

; U, V ) =
_
3v
2
, z
_
n+1
+
_
n+1
(w v) zdx
+(
eff
S(v), z)
n+1
(p, z)
n+1
+ ( v, q)
n+1
f(V ) =
_

4 v
n
v
n1
2
zdx
_
O
p
ref
z ndS, (2.6)
where U = (v, p), V = (z, q), U

= (v

, p), and where =

n+1
, w = v

w
n+1
D
.
Problem 2.1 (Weak formulation of Navier-Stokes in ALE form). Find U = (v, p)
such that
a(U; U, V ) = f(V ), (2.7)
holds for all V = (z, q) A Q, and v satises conditions (2.4 a,b).
2.4. Stabilized nite element method. In order to apply the Galerkin FEM,
we approximate the spaces J, A, Q from the weak formulation by nite dimensional
subspaces J

, A

, Q

, (0,
0
),
0
> 0, A

= v

; v

[
DWt
= 0.
The couple (A

, Q

) of the nite element spaces should either satisfy the Babuska

Brezzi (BB) condition, cf. [11], [12] or [19], or the BB condition can be violated
provided additional stabilization is applied, cf. [15]. In practical computations we
assume that the domain
n+1
is a polygonal approximation of the region occupied by
the uid at time t
n+1
. The spaces W

, A

, Q

are dened over a triangulation T

of the domain
n+1
, formed by a nite number of closed triangles K T

. We use
the standard assumptions on the system of triangulation, cf. [4]. Here denotes the
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION PROBLEMS 249
size of the mesh T

. In this paper the non-conforming equal order nite elements are

used, i.e for k = 1 or k = 2 we dene
H

= v C(
n+1
); v[
K
P
k
(K) for each K T

,
J

= [H

]
d
, A

= J

A, (2.8)
Q

= v C(
n+1
); v[
K
P
k
(K) for each K T

.
As mentioned above the Galerkin approximation of the weak formulation suer
from two sources of instabilities. One instability is caused by the incompatibility of
the pressure and velocity pair of nite elements. It can be overcome by the use of
pressure stabilizing terms. Further, the dominating convection requires to introduce
some stabilization of the nite element scheme, as, e.g. upwinding or streamline-
diusion method. In order to overcome both diculties, modied Galerkin Least
Squares method is applied, cf. [10]. We start with the denition of the local element
rezidual terms
a
K
and
f
K
dened by

a
K
(w; v, p) =
3v
2t

_
( +
T
)
_
v + (v)
T
__
+ (w ) v +p (2.9)
and

f
K
( v
n
, v
n1
) =
1
2t
(4 v
n
v
n1
). (2.10)
The GLS stabilizing terms are then dened
L

(U

; U

, V

) =

KT

K
_

a
K
(w
n+1
; v, p),
_
w
n+1

_
z +q
_
K
, (2.11)
T

(V

) =

KT

K
_

f
K
( v
n
, v
n1
),
_
w
n+1

_
z +q
_
K
,
where the local element rezidual terms
a
K
( ; , ) and
f
K
(, ) are dened by equa-
tions (2.9) and (2.10), and where the function w
n+1
= v

w
n+1
D
.
Further, the div-div stabilization is introduced
T

(U

, V

) =

KT

K
( v, z)
K
, (2.12)
U = (v, p), V = (z, q).
The following choice of parameters
K
,
K
is used

K
=
K
_
1 +Re
loc
+
h
2
K

K
t
_
,
K
=
h
2
K

K
,
where
K
= [ +
T
[
0,2,K
, h
K
denotes the local element size and the local Reynolds
number Re
loc
is dened as
Re
loc
=
h
K
|v|
K
2
K
.
The stabilized GLS scheme then reads: Find U

= (v, p) J

such that v
satises approximately the Dirichlet boundary conditions (2.4 a,b) and the equation
a(U

; U

, V

) +L(U

; U

, V

) +T

(U

, V

) (2.13)
= f(V

) +T(V

),
holds for all V

= (z, q) A

.
250 P. SV

CEK
3. Spalart-Allmaras turbulence model. The system of equations (2.3) is
coupled with the nonlinear partial dierential equation for an additional quantity .
The one equation turbulence model reads

t
+v =
1

_
2

i=1

x
i
_
( + )

x
i
_
+c
b2
( )
2
_
+G( ) Y ( ), (3.1)
where the functions G( ) and Y ( ) are functions of the tensor (
ij
)
ij
of rotation of
the mean velocity and of the wall distance y. Here, the components of the rotation
tensor are dened by
ij
=
1
2
_
vj
xj

vj
xi
_
. The turbulent viscosity
T
is dened by

T
=

3

3
+c
3
v
, =

. (3.2)
Furthermore we use the following relations (see also [20])
G( ) = c
b1

S , Y ( ) = c
w1

2
y
2
_
1 +c
6
w3
1 +c
6
w3
/g
6
_
1
6
,

S =
_
S +

2
y
2
f
v2
_
,
f
v2
= 1

1 +f
v1
, g = r +c
w2
(r
6
r), r =

S
2
y
2
, S =

i,j

2
ij
,
where y denotes the distance from a wall. The following choice of constants is used
c
b1
= 0.1355, c
b2
= 0.622, =
2
3
, c
v
= 7.1, c
w2
= 0.3, c
w3
= 2.0, = 0.41,
c
w1
= c
b1
/
2
+ (1 +c
b2
)/.
3.1. Time discretization. The equation (3.1) is time discretized with the aid
of -stepping scheme, cf. [14]. We choose the parameter (0, 1) and at every time
step t
n
we approximate (t
n
)
(n)
. We dene

(n+)
= (1 )
(n)
+
(n+1)
and for every n = 0, 1, . . . solve the nonlinear equation of Spalart-Allmaras turbulence
model coupled with the Reynolds equations (2.3), i.e.

(n+)

(n)
t
+ (v )
(n+)
=
1

_
( +
(n+)
)
(n+)
_
+c
b2
_

(n+)
_
2
_
+c
b1

S
(n+)
c
w1
_

(n+)
_
2
y
2
_
1 +c
6
w3
1 +c
6
w3
/g
6
_
1
6
. (3.3)
Once
(n+)
has been computed, the value
(n+1)
is obtained by

(n+1)
=
1

(n+)
+
_
1
1

_

(n)
.
3.2. Weak formulation and linearization. The numerical solution of the
Spalart-Allmaras problem is performed with the aid of nite element method. The
turbulence model is described by one partial dierential equation of the convection-
diusion-reaction character with dominating convection, strong nonlinear behaviour
and with abrupt changes of the source functions.
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION PROBLEMS 251
Now, we choose the space 1 = H
1
0
(), take a test function 1, multiply
equation (3.3) by , integrate over and apply Greens theorem. Thus we get the
weak formulation of the problem: Find 1 such that B
sa
( , ) = L
sa
() for all
1, where
B
sa
( , ) = ( , )

+
_

t
+ (v ) +s ( )
2

c
b2

( )
2
c
b1

S ,
_

,
L
sa
() =
_

(n)
t
,
_

and we set
s = c
w1
1
y
2
_
1 +c
6
w3
1 +c
6
w3
/g
6
_
1
6
, = ( ) =
+

. (3.4)
The weak formulation of Spalart-Allmaras turbulence model is nonlinear and requires
application of a linearization procedure. We use the following linearization:
s
_

(n+)
_
2
s (
n
)
2
+ 2s
n
(
(n+)

n
),
_

(n+)
_
2

(n)

(n+)
.
Thus the linearized problem reads: Find 1 such that
B( , ) = L(), (3.5)
for all 1, where
B( , ) =
_

t
+
_
v
c
b2

(n)
_
+ 2s
(n)
,
_

+ ( , )

, (3.6)
L() =
_
s
n

n
+

(n)
t
+c
b1

S
(n)
,
_

. (3.7)
The viscosity parameter is taken as = (
n
) =
+e
n

.
3.3. Space discretization of the turbulence model. In order to aproximate
the problem (3.5) the space 1 is approximated by the nite element subspace 1

1
1

=
_
C
0
() : [
K
P
1
(K) K T

_
,
and Galerkin approximations are sought by solution of the problem: Find

such that (3.5) holds for any

.
Due to large mesh Peclet numbers the Galerkin approximations are unstable. In
order to obtain admissible solution, the SUPG stabilization is applied:
B
SUPG
( , ) = B( , )+ (3.8)
+

KT

K
_

t
+b + 2s
(n)
+ ( ) , (b )
_
K
L
SUPG
() = L() +

KT

K
_
s
n

n
+

(n)
t
+c
b1

S, (b ))
_
K
, (3.9)
252 P. SV

CEK
where the vector b is dened locally on every element K T

as b =
_
v
c
b
2

(n)
_
,
and the parameters
K
are dened by

K
=
_
4[[
0,K
h
2
K
+
2[b[
0,K
h
K
+[s[
0,K
_
1
.
Nevertheless, the use of SUPG/GLS stabilization still does not avoid local os-
cillations near sharp layers, which can lead to patological situations with negative
viscosity. In order to solve this problem we shall make use of the discontinuity cap-
turing techniques (or shock capturing techniques). These stabilization techniques
introduce additional dissipation in crosswind direction, cf. [15], [5]. The nonlinear
stabilization problem reads: Find 1

such that
B
DC
( , ) = L
SUPG
(), 1

, (3.10)
where
B
DC
( , ) = B
SUPG
( , ) +

KT
_
K

K
dx+
+

KT
_
K
(max(
K

K
, 0)
K
)
_
b b
|b|
2
0,,K
_
dx.
Here

K
is the additional diusion from SUPG terms

K
=
K
|b|
0,,K
and
K
is the diusion of the shock capturing method, see [5], [14]. The additional
diusion is based on the local element reziduals
rez( ) =

t
+b + 2s
(n)
+ ( ) s
n

n

(n)
t
c
b1

S.
We set

K
=
_
1
2
A
K
h
K
rez(e )0,2,K
e 0,2,K
if | |
0,2,K
,= 0,
0 elsewhere,
where A
K
= max
_
0, 0.7
2
a10,2,KhK
_
, a
1
=
rez(e )
e
2
0,2,K
, and h
K
is the characte-
ristic length of the element K
4. Structure model. Here, a solid exibly supported airfoil is considered. The
airfoil can be vertically displaced and rotated. Figure 4.1 shows the elastic support
of the airfoil on translational and rotational springs. The pressure and viscous forces
acting on the vibrating airfoil immersed in uid result in the lift force L(t) and the
torsional moment M(t). The governing nonlinear equations are written in the form
(see [6], [8])
m

h +S

cos S

2
sin + d
hh

h +k
hh
h = L(t), (4.1)
S

hcos +I

+d

+k

= M(t),
where k
hh
and k

are the bending stiness and torsional stiness, respectively, and

m is the mass of the airfoil, S

is the static moment around the elastic axis EA, I

is
the inertia moment around the elastic axis EA.
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION PROBLEMS 253
h
L(t)
M(t)
U
T
EA

Fig. 4.1. The elastic support of the airfoil on translational and rotational springs.
t [s]
0 0.2 0.4 0.6 0.8 1
-2
0
2

U =

5 m/s
t [s]
h
[
m
m
]
0 0.2 0.4 0.6 0.8 1
0
1
U =

5 m/s
Fig. 5.1. The aeroelastic response (h, ) for the far eld velocity U = 5m/s.
5. Numerical results. For the aeroelastic simulations we compare the pre-
sented approach to the results of the method published previously in [8], where the
laminar incompressible Navier-Stokes equations were employed and discretized by the
conforming nite element method (Taylor-Hood family of nite elements) together
with SUPG and grad-div stabilization. We use the modied parameter values taken
from [3], where the critical velocity determined by NASTRAN was 30.4 m/s. We
present comparison of laminar and turbulence results for the far eld velocity in the
range U

= 10 32m/s. The parameters of the structural model was set as

m = 0.086622 kg, S

= 0.000779673 kg m, I

= 0.000487291 kg m
2
,
k
hh
= 105.109 N m
1
, k

= 3.695582 N m rad
1
, l = 0.05 m, c = 0.3 m.
The elastic axis is located at 40% of the airfoil, = 1.225 kg m
3
, = 1.5
10
5
m s
2
. The numerical computations were performed for airfoils NACA 0012.
Figures 5.1, 5.2, 5.3 shows the comparison of the aeroelastic response for dierent
values of the inlet velocity.
254 P. SV

CEK
t [s]
0 0.2 0.4 0.6 0.8 1
-2
0
2

U =

15 m/s
t [s]
h
[
m
m
]
0 0.2 0.4 0.6 0.8 1
0
1
U =

15 m/s
Fig. 5.2. The aeroelastic response (h, ) for the far eld velocity U = 15m/s.
t [s]
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
U =

31 m/s
t [s]
h
[
m
m
]
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-10
0
10
U =

31 m/s
Fig. 5.3. The aeroelastic response (h, ) for the far eld velocity U = 31m/s.
REFERENCES
[1] K. J. Bathe, editor. Computational Fluid and Solid Mechanics 2007. Elsevier, 2007.
[2] D. Bo and L. Gastaldi. Stability and geometric conservation laws for ALE formulations.
Computational Methods in Applied Mechanical Engineering, 193:47174739, 2004.
[3] J.

Cecrdle and J. Malecek. Verication FEM model of an aircraft construction with two and
three degrees of freedom. Technical Report R-3418/02, Aeronautical Research and Test
Institute, Prague, Let nany (in Czech), 2002.
[4] P. G. Ciarlet. The Finite Element Methods for Elliptic Problems. North-Holland Publishing,
1979.
[5] R. Codina. A discontinuity capturing crosswind-dissipation for the nite element solution of the
convection diusion equation. Computational Methods in Applied Mechanical Engineering,
110:325342, 1993.
[6] E. H. Dowell. A Modern Course in Aeroelasticity. Kluwer Academic Publishers, Dodrecht,
1995.
[7] Ch. Farhat, T. Lieu, and V. Kongara. A discussion of key concepts and methodologies for the
CFD-based solution of a class of nonlinear uid/structure and thermouid/thermostructure
problems. In E. Onate, M. Papadrakakis, and B. Schreer, editors, Computational Methods
for Coupled Problems in Science and Engineering II, pages 2932, Barcelona, Spain, 2007.
International Center for Numerical Methods in Engineering (CIMNE).
[8] M. Feistauer, J. Hor acek, and P. Sv acek. Numerical simulation of ow induced airfoil vibrations
with large amplitudes. Journal of Fluids and Structure, 23(3):391411, 2007. ISSN 0889-
9746.
[9] Ch. F orster, W. A. Wall, and E. Ramm. On the geometric conservation law in transient
ow calculations on deforming domains. International Journal For Numerical Methods In
Fluids, 50:13691379, 2006.
[10] T. Gelhard, G. Lube, M. A. Olshanskii, and J.-H. Starcke. Stabilized nite element schemes
with LBB-stable elements for incompressible ows. Journal of Computational and Applied
Mathematics, 177:243267, 2005.
[11] V. Girault and P.-A. Raviart. Finite Element Methods for the Navier-Stokes Equations.
Springer-Verlag, Berlin, 1986.
[12] P. M. Gresho and R. L. Sani. Incompressible Flow and the Finite Element Method. Wiley,
Chichester, 2000.
[13] H. Guillard and Ch. Farhat. On the signicance of the geometric conservation law for ow com-
putations on moving meshes. Computational Methods in Applied Mechanical Engineering,
190:14671482, 2000.
NUMERICAL SOLUTION OF FLUID-STRUCTURE INTERACTION PROBLEMS 255
[14] G. Houzeaux. A Geometrical Domain Decomposition Method in Computational Fluid Dynam-
ics. PhD thesis, UPC, 2003.
[15] T. J. R. Hughes, L. P. Franca, and M. Balestra. A new nite element formulation for computa-
tional uid dynamics: V. Circumventing the Babuska-Breezzi condition: a stable Petrov-
Galerkin formulation of the Stokes problem accomodating equal order interpolation. Com-
puter Methods in Applied Mechanical Engineering, 59:8589, 1986.
[16] E. Naudasher and D. Rockwell. Flow-Induced Vibrations. A.A. Balkema, Rotterdam, 1994.
[17] F. Nobile. Numerical approximation of uid-structure interaction problems with application to
haemodynamics. PhD thesis, Ecole Polytechnique Federale de Lausanne, 2001.
[18] T. Nomura and T. J. R. Hughes. An arbitrary Lagrangian-Eulerian nite element method
for interaction of uid and a rigid body. Computer Methods in Applied Mechanics and
Engineering, 95:115138, 1992.
[19] R. Verf urth. Error estimates for mixed nite element approximation of the Stokes equations.
R.A.I.R.O. Analyse numerique/Numerical analysis, 18:175182, 1984.
[20] D. C. Wilcox. Turbulence Modeling for CFD. DCW Industries, 1993.