Professional Documents
Culture Documents
253262 (2007)
1. INTRODUCTION
Aeroelasticity is the interaction between
aerodynamics and a flexible body. Flutter is a
typical aeroelastic problem. Because of the needs of
flights in the transonic regime or at high angle of
attack, and because of the inefficacy of traditional
methods in solving these non-linear problems, some
non-linear fluid-structure coupled numerical
methods are developed for these non-linear
aeroelastic problems. Euler and Navier-Stokes
equations based Computational Fluid Dynamics
(CFD) techniques are suitable for solving the nonlinear aeroelastic problems since they make the
fewest assumptions about the characteristics of the
flows. Since the computational cost of the method
is very high, exploration for better computational
efficiency and quality is now a hot topic that
attracts much research efforts.
An implicit monolithic algorithm to solve the fluidstructure interaction problems has been documented
in literature (Alonso and Jameson, 1994; Liu et al.,
2001; Sadeghi and Liu, 2001). The coupling
operation is carried out in every finite cell of the
flow field. So it needs to modify the existing
unsteady flow field solvers and it is not convenient
to upgrade the subsystems.
A strongly coupled algorithm is presented in the
works of Yang, Obayashi and Nakamichi (2003)
and Allen and Djayapertapa (2002). Subiteration is
Received: 9 Apr. 2007; Revised: 7 Jun. 2007; Accepted: 9 Jun. 2007
253
Monolithic algorithm
Conventional scheme
IA - R - K scheme (present)
Loosely coupled algorithm
Explicit scheme
Fig. 1
2. GOVERNING EQUATIONS
A convenient and economical representation of the
dynamic characteristics of a flexible structure is the
concept of the generalized coordinates. In this
concept, the physical deformations of an elastic
structure are represented by a generalized
coordinate transformation, where the generalized
coordinates are given by eigen modes or assumed
modes of the structure. In general, the eigen modes
are most prefered. The time dependent deformation
vector is:
254
+
+ B F((E n + k 1/ 2), ( t + t/ 2)))
k
t
A
E
k
/
2
(
(
)
2
n
1
k = t ( A (E + k / 2) + B F((E + k / 2), ( t + t/ 2 )))
n
2
n
2
3
k 4 = t ( A (E n + k 3 ) + B F((E n + k 3 ), ( t + t )))
w (x , y, z, t ) = w x i + w y j + w z k
N
(x, y, z) (t )
i
(1)
i =1
(4)
In the Eq. (4), the function F(E, t ) needs to be
calculated four times at each physical time-step.
F(E, t ) is not explicit. It is solved by CFD based
unsteady flow field solver. In addition, the
boundary conditions are difficult to deal with for
unsteady flows. So the function such as
F((E n + k1 / 2 ), ( t + t / 2 )) cannot be easily obtained.
The operation of conventional single-step
scheme
freezes
the
aerodynamic
parts
F at each physical time-step. That is to say, the
F((E n + k1 / 2 ), ( t + t / 2 ))
,
functions
F((E n + k 2 / 2 ), ( t + t / 2 )) and F ((E n + k 3 ), ( t + t )) are
replaced by the function F (E n , t ) at each physical
(2)
where
F = p( x , y, z, t ) i ( x , y, z )ds = q C p ( x , y, z, t ) i ( x , y, z )ds
i s
s
E = f (E, t ) = A E + B F(E, t )
where
A=
O
M 1K
I
1
,B=
(3)
O
M 1 .
E n +1 = E n + (k 1 + 2k 2 + 2k 3 + k 4 ) / 6
k = t ( A E + B F ( t ))
n
1
=
k
t
(
A
(E
2
n + k 1 / 2) + B F ( t + t / 2))
k = t ( A (E + k / 2) + B F ( t + t / 2))
n
2
3
k 4 = t ( A (E n + k 3 ) + B F ( t + t ))
255
(5)
F((E n + k 3 ), ( t + t )) F( t + t )
= F( t 2t ) 3F( t t ) + 3F( t ) + O t 3
Fn +1 = 2Fn Fn 1
( )
( )
E = E + t (3 f f )
n
n
n 1
n +1
2
t
t
= E n + (3A E n A E n 1 ) + (3B Fn B Fn 1 )
2
2
t
E n +1 = E n + ( f n + f n +1 (E n +1 , t + t ))
2
= E n + t (A E n + A E n +1 ) + t (3B Fn B Fn 1 )
2
2
(10)
Similarly, by a fourth-order extrapolation using four
points:
Fn +1 = 4Fn 6Fn 1 + 4Fn 2 Fn 3
= En +
(11)
(9)
E
n +1
E n +1
t
(3 f n f n 1 )
2
t
(3A En A En 1 ) + t (3B Fn B Fn 1 )
2
2
(7)
It can also be integrated by the second-order
implicit Adams linear multi-step scheme (IALM)
using predictor-corrector procedure.
t
(55 f n 59 f n 1 + 37 f n 2 9 f n 3 )
24
t
= En +
(55A E n 59A E n 1 + 37 A E n 2 9A E n 3 )
24
t
+
(55B E n 59B E n 1 + 37B E n 2 9B E n 3 )
24
t
(9 f n +1 (E n +1 , t + t ) + 19 f n 5 f n 1 + f n 2 )
= En +
24
t
(9A E n +1 + 19A E n 5A E n-1 + A E n 2 )
= En +
24
t
(55B Fn 59B Fn 1 + 37B Fn 2 9B Fn 3 )
+
24
= En +
(12)
The corrector-step of equation (10) shows that with
the introduction of the extrapolation of the
generalized aerodynamic loads, the implicit scheme
of the aerodynamic parts in Eq. (8) turns out to be
an explicit scheme of the same order. The structural
parts retain the characteristics of an implicit scheme
and therefore, the fluid-structure interaction
solution method presented in this paper is called
implicit-explicit hybrid linear multi-step scheme
(HLM), in which only one time aerodynamic
solution is needed at each time-step.
The error analysis of the HLM scheme is as
follows:
E = E + t (3 f f )
n
n
n 1
n +1
2
t
t
= E n + (3A E n A E n 1 ) + (3B Fn B Fn 1 )
2
2
t
E n +1 = E n + ( f n + f n +1 (E n +1 , t + t ))
2
= E n + t (A E n + A E n +1 ) + t (B Fn + B Fn +1 (E n +1 , t + t ))
2
2
(8)
Equation (7) shows that one-time aerodynamic
solution is needed for the explicit scheme and
Eq. (8) shows that two-time aerodynamic solution is
needed for the implicit scheme at each physical
time-step.
For the corrector step of Eq. (8), the aerodynamic
loads of the n+1th time-step ( Fn+1 ) is extrapolated
by Fn1 and Fn .
256
t
A
E
n + 1 = E n + 2 t E n + E n + 1 + 2 3B Fn B Fn 1
A
t
= E + t E + E +
n n 2 3 f n f n 1 +
n
2
[3(A E n + B Fn ) (A E n 1 + B Fn 1 )
3A E + A E
n
n 1
t
A
= E + t 2E +
3f f
n
n
n
n 1
2
2
t
+
3A E + A E
3f f
n
n 1
n
n 1
2
t
A
)
=E +
2 A E + t (3 f f
n
n 1
n 2
n
2
+ 3f f
3A E + A E
n
n 1
n
n 1
3
1
=E +
+ t f t f
A E + E
n
n 1 2
n 2
n 1
n
2
+ 3f f
n
n 1
R n ,t =
(16)
Thus it can be seen that the truncation error of
equations in (12) is a fifth-order one. So the scheme
is of fourth-order accuracy.
(13)
Assuming that E(t) is approximated by including up
to the third derivative, thus:
4 2 '' 8 3 '''
t E + t E + .....
2
6
2
3
t
t
E n = E n 1 + tE ' +
E '' +
E '''
2
6
t 2 '''
f n = E n + E ' n 1 + tE '' +
E
2
( )
9
251 5 (5)
At 5 E ( 4) ( t n ) + O t 6
t E ( t n ) +
24
720
E n +1 = E n 1 + 2tE ' +
(14)
2
3
t
t
E n 1 + tE ' +
E '' +
E '''
2
6
t
7
2 ''
3 '''
+
At E + At E
2
12
3
'
''
+ 2E + 3tE + t 2 E '''
2
5 3 ''' A 3 ''
=
t E t E +O t 4
12
2
5
This paper
Liu et al.
Alonso&Jameson
4.5
(degree)
4
3.5
3
2.5
2
( )
1.5
1
(15)
0.5
0
0.7
Fig. 2
257
0.75
0.8
0.85
Ma
0.9
0.95
CA-R-K
IA-R-K
Second order EALM
Second order IALM
Second order HLM
Forth order EALM
Forth order IALM
Forth order HLM
/ rad
0.01
0.005
/ rad
0.015
N=100
N=200
N=600
0.005
0
-0.005
10
20
-0.005
0
30
40
50
0.01
0.005
10
0.005
/ rad
(a) CA-R-K
40
60
/rad
20
N=30
N=100
N=200
0.005
30
40
50
th
30
40
50
60
-0.005
0
Fig. 6
-0.005
0
10
20
30
40
50
(b) IA-R-K
Fig. 4
60
N=75
N=200
N=600
-0.005
0
10
20
30
40
50
30
40
50
30
40
50
60
N=30
N=200
N=600
0.005
-0.005
0
10
20
nd
60
N=30
N=200
N=600
0.005
-0.005
0
10
20
nd
10
20
60
0.005
/ rad
th
/rad
20
N=40
N=200
N=600
-0.01
/rad
60
-0.005
-0.005
Fig. 5
0.01
50
0.005
N=40
N=200
N=600
N=3600
-0.015 0
40
N=40
N=200
N=600
60
0.015
/ rad
20
/rad
Fig. 3
10
30
th
60
0.01
/ rad
/ rad
0.01
-0.01
-0.01
0
20
40
60
10
15
0.02
0.01
/ rad
/ rad
0.01
-0.01
-0.01
-0.02
0
10
20
40
60
0.02
0.01
/ rad
/ rad
0.01
-0.01
0
Fig. 7
-0.01
-0.02
10
10
15
Table 1 Flutter velocities of Isogai Wing computed by various schemes and various N.
ND
40
75
100
200
600
1800
3600
CA-R-K
0.68
0.72
0.74
0.77
0.796
0.806
0.81
IA-R-K
0.814
0.81
0.81
0.81
0.81
0.81
0.81
0.813
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.812
0.81
0.81
0.81
0.81
0.81
ORDER EALM
ORDER HLM
0.82
TH
ORDER HLM
0.815
0.81
0.81
0.81
0.81
0.81
0.81
0.83
0.816
0.815
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
0.81
/COMPUTING
STEP(MIN)
PER TIME-
40
75
100
200
600
1800
3600
120
145
153
213
346
473
480
241
291
305
424
673
930
1155
259
Present methods(Dt=0.002)
Farhat and Lesoninne(Dt=0.002)
Lee-Rausch and Batina(Dt=0.0001)
Gupta(Dt=0.000425)
Experiment
0.8
0.7
0.6
Vf
0.5
0.4
0.3
0.2
0.5
0.6
0.7
0.8
0.9
1.1
0.8
0.6
0.7
0.5
0.4
0.3
0.2
0.5
0.6
0.7
0.8
0.9
1.1
CA-R-K(N=1040)
IA-R-K(N=35)
Second order EALM(N=83)
Second order HLM(N=35)
Forth order EALM(N=104)
Forth order HLM(N=52)
Fig. 8
4E-05
2E-05
2E-05
1
2
3
4
-2E-05
0
0.2
0.4
0.6
0.8
-2E-05
2E-05
-4E-05
-2E-05
0
0.2
0.4
0.6
0.8
0.1
0.2
t/s
0.3
0.4
0.5
5E-05
0
-5E-05
0
Fig. 9
0.2
0.4
0.6
0.8
t/s
260
1
2
3
4
4E-05
2E-05
2E-05
-2E-05
-2E-05
-4E-05
-4E-05
0
0.02
0.04
0.06
0.01
0.02
0.03
0.04
t/s
1
2
3
4
4E-05
REFERENCES
2E-05
2E-05
0.08
t/s
1
2
3
4
4E-05
ACKNOWLEDGEMENTS
1
2
3
4
4E-05
-2E-05
-2E-05
-4E-05
-4E-05
0
0.02
0.04
0.06
0.08
t/s
0.02
0.04
0.06
0.08
t/s
6. CONCLUSIONS
This research shows that the present loosely
coupled schemes (IA-R-K and HLM) for solving
aeroelastic problem in time domain are not inferior
to implicit strongly coupled scheme or fully
monolithic schemes in terms of computation
accuracy and numerical stability. At the same time,
the two schemes presented in this paper achieve
high order accuracy by solving the flow field only
one time at each time step. These characteristics not
only improve computational efficiency but also
provide convenience for the fluid-structure
interaction solution. The unsteady flow solver as a
module can be used to solve the aeroelastic problem
without any modifications. It also makes it
convenient to upgrade the computation structural
subsystem or fluid subsystem.
The error of the HLM scheme is analyzed
theoretically. The numerical results validate the
accuracy and efficiency from both the convergences
of aeroelastic responses and flutter velocities. The
numerical stability of the IA-R-K or HLM schemes
is also very strong. To achieve satisfactory results,
only 3050 samples per flutter period are needed.
The IA-R-K and HLM schemes presented in this
paper provide high efficiency, high-order accuracy,
strong stability and simple operation.
261
262