Finite Elements in Analysis and Design 44 (2008) 537 – 551

www.elsevier.com/locate/finel

Full-order and multimode flutter analysis using ANSYS

X.G. Hua1 , Z.Q. Chen ∗,2
Wind Engineering Research Center, College of Civil Engineering, Hunan University, Changsha, Hunan 410082, PR China

Received 2 November 2006; received in revised form 16 January 2008; accepted 18 January 2008
Available online 7 March 2008

Abstract
This paper presents the full-order and multimode methods for analyzing coupled flutter of long-span bridges using commercial finite element
(FE) package ANSYS. In the full-order method of flutter analysis, a novel FE model is developed to model the coupled wind-bridge system, in
which a specific user-defined Matrix27 element in ANSYS is adapted to model the aeroelastic forces and its stiffness or damping matrices are
parameterized by wind velocity and vibration frequency. Variation of complex eigenvalues of the coupled system with wind velocity is then
depicted by using this model together with complex eigenvalue analysis, and flutter instability can be determined from the variation diagram. In
the multimode method, equations of motion for the coupled wind-bridge system are first represented using a modal superposition technique. This
formulation leads to a single-parameter searching technique without iteration to determine the conditions of flutter instability when structural
damping is not considered in solution. The contribution of participating modes to flutter instability is given in terms of modal amplitude and
modal energy in the multimode method. Numerical studies are provided to validate the developed methods as well as to demonstrate both
the procedures for flutter analysis using ANSYS. The proposed methods enable the bridge designers and engineering practitioners to analyze
bridge flutter in commercial FE package ANSYS.
䉷 2008 Elsevier B.V. All rights reserved.

Keywords: Long-span bridges; Coupled flutter; Instability; Matrix27; ANSYS

1. Introduction critical flutter wind velocity as well as the corresponding flutter

frequency.
Owing to large flexibility and low structural damping, many Since the collapse of the old Tacoma Narrow Bridge in
flexible and slender structures such as long-span bridges, high- 1940, considerable efforts have been made to develop pro-
rise buildings and chimneys are susceptible to a variety of cedures for analyzing coupled flutter of long-span bridges
wind-induced vibrations [1]. Among them, wind-induced flut- by integrating finite element (FE) techniques with the flutter
ter instability is the most dangerous one in which the bridge derivatives determined either from Thoedorsen’s theoretical
oscillates in a divergent and destructive manner at some critical formulation or from wind tunnel testing. Bleich [2] was among
wind velocity. As a result, flutter instability is one of the major the first to perform the coupled flutter analysis of suspension
concerns in the design and construction of long-span bridges, bridges using the theoretical flutter derivatives. The coupled
and the lowest wind velocity inducing flutter instability of a flutter analysis using the measured flutter derivatives from the
bridge must exceed the maximum design wind velocity of that spring-mounted bridge sectional model testing in wind tun-
bridge. The objective of flutter analysis is to predict the lowest nel was pioneered by Scanlan and his co-workers [3–5]. At
present there are two general approaches for coupled flutter
analysis of bridges: (i) the full-order flutter analysis approach
∗ Corresponding author. Tel.: +86 731 8821424x203;
where the aeroelastic loadings are applied directly to the
fax: +86 731 8823923.
E-mail addresses: [email protected] (X.G. Hua), [email protected]
physical coordinate of structures [6–9] and (ii) the multimode
(Z.Q. Chen). flutter analysis approach where the equations of motion for
1 Associate Professor. structures are represented using a modal superposition tech-
2 Professor. nique [9–19].
0168-874X/$ - see front matter 䉷 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.finel.2008.01.011
538 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

Since the 1970s a number of commercial FE packages such Lae

as ANSYS, ABAQUS and ADINA have emerged and received Mae
wide applications in various disciplines due to the advancement Dae
of FE methods and computing technologies. These FE pack-  
U
ages have friendly graphical user interface and powerful com- Wind attack angle
putational capability. However, the general purpose commercial B
FE packages commonly used in civil engineering community
Fig. 1. Aeroelastic forces acting on bridge deck.
cannot be directly used for flutter analysis of bridges due to
lack of the capability of calculating motion-dependent aeroe-
lastic loads. Although it is possible to develop special purpose
ment, velocity and acceleration vectors, respectively; and Fae
FE packages to tackle bridge flutter analysis such as ANSUSP
denotes the vector of nodal aeroelastic forces.
[11] and NACS [14], the incorporation of functions or modules
The motion-dependent aeroelastic forces distributed on unit
capable of flutter analysis into general purpose commercial FE
span of bridge girder are expressed as a linear function of nodal
packages provides an alternative way.
displacement and nodal velocity [4,15]
This paper presents two alternative methods, namely the full-
order method and the multimode method, for analyzing cou- 
1 ḣ B ˙
pled flutter of long-span bridges using ANSYS, with the main Lae = U (2B) KH ∗1
2
+ KH ∗2 + K 2 H3∗ 
2 U U
purpose of providing practical tools for researchers and engi- 
neering practitioners to analyze bridge coupled flutter problem 2 ∗ h ∗ ṗ 2 ∗ p
+K H4 + KH 5 + K H6 , (2a)
using ANSYS. In the development of the first method, the cou- B U B
pled wind-bridge system is first modeled by a hybrid FE model 
1 ṗ B ˙
which incorporates structural FE model with fictitious specific Dae = U 2 (2B) KP ∗1 + KP ∗2 + K 2 P3∗ 
user-defined Matrix27 elements used to represent the motion- 2 U U

dependent aeroelastic forces. The stiffness or damping matrices 2 ∗ p ∗ ḣ 2 ∗ h
+K P4 + KP 5 + K P6 , (2b)
of Matrix27 element are expressed in terms of wind velocity B U B
and vibration frequency. The complex eigenvalues of the low-

order modes at varying wind velocities are then determined 1 ḣ B ˙
from this hybrid FE model together with complex eigenvalue Mae = U (2B ) KA∗1
2 2
+ KA∗2 + K 2 A∗3 
2 U U
analysis, and the real and imaginary parts of the eigenvalues are 
h ṗ p
the logarithm decay rates and damped vibration frequencies of +K 2 A∗4 + KA∗5 + K 2 A∗6 , (2c)
B U U
these modes, respectively. Flutter instability will occur when
real part of any eigenvalue becomes positive. While in the mul- where  is air mass density; U is wind velocity; B is the width
timode method, equations of motion for the structure subjected of bridge deck; K = B/U is the reduced circular frequency;
to aeroelastic forces are first reformulated with some selected h, p and  are the vertical, lateral and torsional displacements,
low-order natural modes. A single-parameter searching tech- respectively; A∗i , Hi∗ and Pi∗ (i = 1, . . . , 6) are flutter deriva-
nique without the need of frequency iteration is then described tives which are expressed in terms of reduced wind velocity
to determine the critical conditions of flutter instability when Ũ = U/(fB) and f is the natural frequency. The aeroelastic
structural damping is not taken into account in solution. Contri- forces on bridge deck are illustrated in Fig. 1.
bution of the selected participating modes to flutter instability By converting the distributed aeroelastic forces of element e
is also provided in terms of modal amplitude and modal energy of bridge girder into equivalent nodal loadings at member ends,
in the multimode method. The flutter analysis of both a simply one obtains the equivalent nodal loadings for that element as
supported line-like bridge with the theoretical flutter derivatives
e e e
and a real suspension bridge with the theoretical and measured Fae = Kae X + Ceae Ẋe , (3)
flutter derivatives is carried out to validate the developed pro-
where Kae e and Ce are the elemental aeroelastic stiffness and
cedures and demonstrate the full-order and multimode flutter ae
analysis of cable-supported bridges using ANSYS. damping matrices for element e, respectively. Similar to the
general procedures in formulating elemental mass matrix, both
2. Full-order flutter analysis a lumped formulation and a consistent formulation can be used
to derive the elemental aeroelastic stiffness and damping ma-
2.1. Novel FE model for flutter analysis trices [20]. When using the lumped formulation, the elemental
stiffness and damping matrices are
The equations of motion for a bridge in the smooth flow can  e 
be expressed as e Kae1 0
Kae = e , (4a)
0 Kae1
MẌ + CẊ + KX = Fae , (1)
 
Ceae1 0
where M, C and K are the global mass, damping and stiffness Ceae = , (4b)
matrices, respectively; X, Ẋ and Ẍ represent the nodal displace- 0 Ceae1
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 539

Z formulated. Because one Matrix27 element can only model ei-

ther an aeroelastic stiffness matrix or an aeroelastic damping
matrix instead of both of them simultaneously, a pair of Ma-
Ze trix27 elements are attached to each node in a structural ele-
ment to simulate the aeroelastic forces. For a deck element e as
K , C , or M shown in Fig. 3, Matrix27 elements e1 and e3 are attached after
i j defining a fictitious node k to represent the aeroelastic stiffness
Ye e
and damping at node i, while Matrix27 elements e2 and e4 are
Xe attached after defining a fictitious node l to represent the aeroe-
Y
lastic stiffness and damping at node j. The pair of Matrix27
elements attached to each structural node share the same nodes.
In the case when the length of each bridge deck element is
X the same, the element matrices of the four Matrix27 elements
are simplified as
Fig. 2. Geometry configuration of Matrix27 element.
Ke1 = 2Kae
e
, (5a)

e Ce3 = 2Ceae , (5b)

i j

e1 e3 e2 e4 Ke2 = 2Kae
e
, (5c)
k l
Ce4 = 2Ceae . (5d)
Fig. 3. Hybrid finite element model for flutter analysis in ANSYS.
Assembling all elemental matrices into global aeroelastic
⎡0 0 0 0 0 0⎤
stiffness and damping matrices leads to
⎢0 P6∗ P4∗ BP∗3 0 0⎥ Fae = Kae X + Cae Ẋ, (6)
⎢0 H6∗ H4∗ BH∗3 0⎥
e ⎢ 0 ⎥
Kae1 =a⎢ ⎥, (4c)
⎢0 BA∗6 BA∗4 B 2 A∗3 0 0⎥ where Kae and Cae denote the global aeroelastic stiffness and
⎣ ⎦ damping matrices, respectively.
0 0 0 0 0 0
0 0 0 0 0 0 Substituting Eq. (6) into Eq. (1) results in the governing
⎡0 equation of motion for the coupled wind-bridge system and
0 0 0 0 0⎤ expressed as the structure FE model incorporated with Matrix27
⎢0 P5∗ P1∗ BP∗2 0 0⎥ elements, namely
⎢0 H5∗ H1∗ BH∗2 0⎥
⎢ 0 ⎥
Ceae1 = b ⎢ ⎥, (4d)
⎢0 BA∗5 BA∗1 B 2 A∗2 0 0⎥ MẌ + (C − Cae )Ẋ + (K − Kae )X = 0. (7)
⎣ ⎦
0 0 0 0 0 0
0 0 0 0 0 0 Eq. (7) represents the mathematical model of an integrated
system with the effect of aeroelasticity and it is parame-
where a = U 2 K 2 Le /2 and b = UBKLe /2; Le is the length terized by wind velocity and vibration frequency. With this
of element e. equation, complex eigenvalue analysis can be carried out to
The user-defined element in ANSYS, Matrix27, is a versa- determine the eigenvalues of the parameterized system at a
tile element with two nodes each having six degrees of freedom specific wind velocity and vibration frequency. As flutter in
and with its local coordinate system being parallel to the global the real bridges occurs at the lowest flutter wind velocity cor-
coordinate system [21]. Fig. 2 illustrates the geometry config- responding to the low-order modes, as will be shown later,
uration of Matrix27 and its local coordinate system (Xe , Ye , only the first several eigenvalues will be required in complex
and Ze ). Compared with other structural elements in ANSYS, eigenvalue analysis. Assume the conjugate pairs of complex
Matrix27 possesses some unique features: (i) user-specified en- eigenvalues j = j ± ij (j = 1, 2, . . . , n) and the conjugate
tries of mass, stiffness and damping matrices instead of physi- pairs of complex eigenvectors j = pj ± iqj have been ob-
cal parameters such as mass density and Young’s modulus and tained, the system will be dynamically stable if the real parts
(ii) accommodating both symmetric and asymmetric elemen- of all the eigenvalues is negative and dynamically unstable if
tal matrix formulations. These features enable the modeling of the real part of any eigenvalue is positive. The condition for
the aeroelastic matrices in Eqs. (3) and (4) using Matrix27 as onset of flutter instability is then stated as follows: at a certain
follows. wind velocity Uf the system has only one complex eigenvalue
The first step for full-order flutter analysis using ANSYS is f with zero real part, the corresponding wind velocity Uf be-
to represent the elemental stiffness and damping matrices due ing the critical flutter wind velocity and the imaginary part f
to motion-dependent aeroelastic forces by element Matrix27. of the complex eigenvalue f becoming the flutter frequency.
To achieve this, a hybrid FE model incorporating one structural In practice, the lowest wind velocity that induces the flutter
element with four Matrix27 elements as illustrated in Fig. 3 is instability is the most important and of concern.
540 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

2.2. Incorporation of structural damping for the full-order flutter analysis in ANSYS is summarized in
the following steps:
The formulation above is exact in the case when structural
damping is not considered (namely C = 0 in Eq. (7)), minor (1) Establish the FE model for the original structure without
modification is required to accommodate the structural damp- Matrix27 elements and compute the first m natural frequen-
ing. The damping of a structure is generally given in terms of cies 0i (i = 1, 2, . . . , m).
modal damping ratios or by assuming the Rayleigh damping (2) Establish the hybrid FE model by attaching Matrix27 el-
matrix as ements to the structural FE model, in which the flutter
derivatives are inputted through the command TABLE in
C = M + K, (8) ANSYS, and specify the range of wind velocity studied.
(3) Determine the present wind velocity U.
where  and  are proportionality coefficients of Rayleigh (4) Let the initial oscillation frequency 0 be the frequency 0i
damping. When the modal damping ratios for the ith and jth of each natural mode in turn.
modes are measured or assumed, the proportionality coeffi- (5) Determine the reduced frequency K and the aeroelastic stiff-
cients can be obtained by [22] ness and damping matrices in Matrix27 elements in Eq. (4)
i j at the present iteration, and then carry out complex eigen-
=2 (j i − i j ), (9a) value analysis.
2j − 2i
(6) Compare the imaginary part of the ith computed complex
i j eigenvalue i with 0 . If |(Im(i ) − 0 )/Im(i )| > 10−3 ,
i j
=2 − + , (9b) let 0 = Im(i ) and repeat steps 5 and 6, otherwise go to
2j − 2i j i
step 7.
(7) Loop steps 4–6 over all the m computed natural modes to
where i and i are the circular frequency and damping ratio
obtain all m pairs of complex eigenvalues at the present
for the ith mode; and j and j are the circular frequency and
wind velocity U.
damping ratio for the jth mode.
(8) Repeat steps 3–7 for the specified range of wind velocity
After incorporating structural damping through the propor-
to obtain the variation of m pairs of complex eigenvalues
tionality coefficients, the equations of motion for the coupled
with wind velocity.
wind-bridge system is modified as

MẌ + (C − Cae )Ẋ + (K − Kae )X = 0, (10) The commercial FE package ANSYS provides three tools
for users to customize and expand its existing capabilities.
where C is the modified Rayleigh damping matrix for the sys- These are ANSYS Parametric Design Language (APDL), User
tem, and Cae is the aeroelastic damping matrix modified to Interface Design Language (UIDL) and User Programmable
account for change in the Rayleigh damping matrix. They are Features (UPFs) [21]. The APDL is a scripting language that
expressed as enables users to automate common tasks or even build the FE
model in terms of parameters (variables). It also encompasses
C = M + (K − Kae ), (11a) a wide range of other features such as repeating a command,
macros, if-then-else branching, do-loops, and scalar, vector and
Cae = Cae − Kae . (11b) matrix operations. Making use of the tool APDL, all the above
steps can be readily implemented in ANSYS.
2.3. Algorithm implementation in ANSYS
3. Multimode flutter analysis
As shown in Eq. (4), the aeroelastic stiffness and damp-
ing matrices in Matrix27 elements are expressed in terms of 3.1. Basic theory
three unknown parameters, namely wind velocity, vibration fre-
quency and reduced frequency, but only two of them are inde- Alternative approach to the full-order flutter analysis is the
pendent. Thus the search for flutter instability involves a sweep multimode method of flutter analysis. In the past years, a num-
and iterative procedure. In this study wind velocity and vibra- ber of multimode flutter analysis methods have been developed.
tion frequency are selected as independent variables in solu- In this study, the method developed by Chen [14] and Chen
tion, and the investigation of flutter instability involves a sweep and Agar [23] is employed and implemented in ANSYS. This
through a range of wind velocity in conjunction with an it- method is commonly referred to as the multimode-participation
eration with respect to vibration frequency, as detailed in the and single-parameter searching method (M–S method). The
following. most attractive merit of the M–S method is the avoidance of the
The eigensolution of Eq. (7) or (10) yields the first m (m>n) iteration with respect to vibration frequency to obtain flutter so-
complex conjugate pairs of eigenvalues in the form of i = lution in the case when the structural damping is ignored. Ad-
i ± ii (i = 1, 2, . . . , m). In order to provide variation of each ditionally, it also provides quantitative descriptions on modal
complex eigenvalue with wind velocity, a mode-by-mode trac- participation factors in terms of modal amplitude and modal
ing method [8] is employed. The algorithmic implementation energy, as will be described later.
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 541

In the M–S method, the elemental aeroelastic forces in Eq. wind-bridge system in the modal coordinate, where the entries
(3) is rewritten as of matrix M are expressed in terms of the reduced wind velocity
and vibration frequency. Likewise, the flutter analysis reduces
e Ẋe
Fae =  2 A e X e +  2 Be , (12) to searching for a certain values of reduced wind velocity and
 vibration frequency, at which the damping part of one complex
where Ae and Be are referred to as elemental aerodynamic force eigenvalue  is zero. In recognizing that the second term of
matrices. It is straightforward to see that Ae and Be have similar imaginary part in matrix M is zero, a single-parameter sweep of
expressions with Kae e and Ce expect that a = b = B 2 L /2 in reduced wind velocity within a predefined range is sufficient to
ae e
this case. It is worthy to note that both matrices Ae and Be are determine the critical flutter wind velocity for C̃=0. Otherwise,
expressed in terms of flutter derivatives which are functions of an additional frequency iteration procedure is needed to obtain
only the reduced wind velocity. Similarly, the global aeroelastic the exact flutter frequency. By assuming that one eigenvalue f
forces are then obtained as has zero real part at some reduced wind velocity Uf , the flutter
frequency and flutter critical wind velocity are obtained as
Ẋ
Fae = 2 AX + 2 B , (13)
 f = f , (18a)
where A and B are global aeroelastic stiffness and damping Ũf f B
matrices, respectively. Uf = . (18b)
2
By using the modal superposition technique, the displace-
ment vector X can be approximated by the first m mode 3.2. Behaviors of bridge motion at flutter critical state
shapes as
X = Y, (14) In this section, the contribution of the selected participating
modes to flutter is given in terms of modal amplitude and modal
where  is the n × m matrix consisting of the mass-normalized energy to identify the most dangerous natural modes to flutter.
mode shapes of selected m participating modes and Y is the m After presenting conjugate pairs of the eigenvectors pj ± qj
generalized modal coordinates. into Eqs. (16) and (14) and then making some mathematical
Substituting Eqs. (13) and (14) into Eq. (1), pre-multiplying manipulation, the characteristic motion of the bridge at critical
T , and then applying the orthogonality condition give the flutter state is expressed as [14]
representing modal-motion equations as
m

Ẏ X= j rj sin(f t + j ), (19)
Ÿ + C̃Ẏ + Y = 2 ÃY + B̃ , (15) j =1

where j = arctan(−pj /qj ) denotes the phase angle of mode
where Ã=T A is the generalized aeroelastic stiffness matrix;
j at the critical flutter state and rj = 2 pj2 + qj2 represents the
B̃ = T B is the generalized aeroelastic damping matrix; C̃ =
diag(21 1 22 2 · · · 2m m ) is the generalized Rayleigh modal amplitude for mode j.
damping matrix; and =diag(21 22 · · · 2m ) is the diagonal The total energy of the wind-bridge system can be expressed
matrix of eigenvalues. as the summation of kinetic energy and potential energy as
The vibration of the bridge becomes harmonic with a single E(t) = 21 ẊT MẊ + 1
XT KX. (20)
frequency at critical flutter wind velocity as other frequency 2

components will be eventually damped out. Thus, the modal Substituting Eq. (19) into Eq. (20) leads to
response corresponding to this undamped frequency  is of
m
interest and it takes a harmonic form as 1  2 2
E(t) = rj [f cos2 (f t + j)
2
Y = Y0 exp(it). (16) j =1
+ 2j sin2 (f t + j )]. (21)
Substituting Eq. (16) into Eq. (15) leads to
  The mean energy for a motion period is then obtained by inte-
1
Y0 =  (I + Ã) + i B̃ − C̃ Y0
2
(17a) grating Eq. (21) over the period as

m
or 1  2 2
E= rj (f + 2j ). (22)
4
Y0 = MY0 , (17b) j =1

where M = [(I + Ã) + i(B̃ − 1/C̃)] with the order of m × m; The ratio of modal energy for mode j to the total energy of the
I is the m × m identity matrix and  = 2 gives the eigenvalue system is evaluated as
at the critical flutter state.
rj2 (2f + 2j )
Similar to Eq. (7), Eq. (17) represents a parameterized and e j = m 2 2 . (23)
i=1 ri (f + i )
2
generalized complex eigenvalue formulation for the coupled
542 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

FE Model Input Essential Parameters

Modal Analysis Formation of Aerodynamic

t Flutter Derivatives
n Forces Matrix
le me
E
eck
in gD
ad
Re
Reading Modal Data
ANSYS Formation of Characteristic
Database Wr Equation
itin
gR
esu
lts

Post Processing Analysis Results

Fig. 4. Flowchart of multimode flutter analysis using ANSYS.

Fig. 5. Customized ANSYS menu for flutter analysis: (a) main menu and (b) dialog.

Through Eq. (23), the modal contribution factor for each mode and (4) post-processing analysis results, such as listing the crit-
can be evaluated and the mode with the maximal value of ej ical flutter wind velocity and flutter frequency as well as vi-
is the most dangerous mode to flutter. sualizing the characteristic motion of the bridge at the critical
flutter states.
3.3. Algorithm implementation in ANSYS Fig. 5 shows two customized menu interfaces of flutter anal-
ysis for illustration. In Fig. 5(a), a new submenu flutter analysis
The multimode method of flutter analysis is implemented
is appended in the main menu of ANSYS, and Fig. 5(b) shows
through the tools of UPF and UIDL in ANSYS. UPF enables
the dialog for inputting the essential parameters needed for flut-
users to develop their own routines and tailor the ANSYS pro-
ter analysis. By using these menus and dialogs, it would be
gram to special kinds of analysis. UIDL is a programming lan-
straightforward to perform flutter analysis in ANSYS in graph-
guage that allows users to customize and configure ANSYS
ical user interface.
menu. In this study, we employ UPF to implement the M–S
method of flutter analysis in ANSYS and utilize UIDL to de-
velop menu for facilitating the flutter analysis. 3.4. Comparison between full-order method and multimode
Fig. 4 outlines the flowchart of multimode flutter analysis method
using ANSYS. The procedures are detailed as follows: (1) es- The advantages and disadvantages for full-order and multi-
tablishing FE model of the bridge using ANSYS; (2) executing mode flutter analysis methods are summarized as follows:
natural modal analysis; (3) developing computer codes based
on Visual FORTRAN and UPF to implement the M–S method, (1) Full-order flutter analysis is performed in the physical co-
capable of accessing (reading and writing) ANSYS database; ordinate, but multimode flutter analysis is carried out in
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 543

the modal coordinate. Thus the multimode method is much moment of inertia Im =4.5×106 kg m2 /m and air mass density
more computationally efficient because the number of equa-  = 1.248 kg/m3 . The structural damping is not considered in
tions in modal coordinate generally is very few; in other this example to make a theoretical comparison. The unsteady
words, m>n. aeroelastic forces acting on a thin-airfoil cross section in smooth
(2) Both methods require a sweep-and-iteration procedure to flow were first analytically derived by Theodorsen [24]. The
identify the flutter critical wind velocity when the structural relation between the flutter derivates and the unsteady aeroe-
damping needs to be considered. The full-order method lastic forces has been well documented in literature (e.g., [8]).
selects wind velocity and vibration frequency as the two Fig. 6 shows the variation of flutter derivatives versus reduced
independent variables, and instead the multimode method wind velocity U/(f B).
uses the reduced wind velocity and vibration frequency.
In the case when the structural damping is ignored, only 4.1.1. Full-order method of flutter analysis
sweep through a range of reduced wind velocity will be Natural modal analysis of the bridge FE model without Ma-
required to identify the critical flutter wind velocity in the trix27 elements is first conducted, where the lumped mass for-
multimode method. mulation is used to construct the mass matrix. The bridge is
(3) The implementation of the full-order method resorts to discretized by 30 deck elements as shown in Fig. 7. The two-
APDL which is much easier than UPF and UIDL required node beam element Beam4 in ANSYS is used to represent the
for the implementation of multimode flutter analysis. deck elements, and the element Mass21 in ANSYS is used to
model the mass moment of inertia. The first 10 natural modes
4. Case studies are extracted using ANSYS and summarized in Table 1.
Matrix27 elements are then incorporated into the structural
4.1. Simply supported thin airfoil bridge FE model to perform full-order flutter analysis. A total of 58

In this section, two examples are provided to verify both Table 1

methods as well as to demonstrate the procedures for flutter Description of the first 10 modes
analysis. The first application illustrated here is the flutter analy-
sis of a simply supported line-like bridge with thin-airfoil cross Mode Frequency Mode Mode Frequency Mode
section. Since the theoretical solution of flutter frequency and no. (Hz) shape no. (Hz) shape
flutter wind velocity for this structure is available, this example 1 0.1788 S–V 6 1.5030 S–T
serves as a verification of both methods. The structural param- 2 0.5028 S–T 7 1.6096 S–V
eters are as follows: span l = 300 m; width of the bridge deck 3 0.5236 S–L 8 1.9976 A–T
B = 40 m; vertically bending rigidity EIz = 2.1 × 106 MPa m4 ; 4 0.7154 A–V 9 2.0944 A–L
5 1.0043 A–T 10 2.4867 S–T
laterally bending rigidity EIy = 1.8 × 107 MPa m4 ; torsional
rigidity GIt = 4.1 × 105 MPa m4 ; mass m = 20, 000 kg/m; mass Note: S—symmetric; A—asymmetric; V—vertical; L—lateral; T—torsional.

4
1
A1* 0
A2*
-1
3
-2
A3* -3
Flutter derivatives

Flutter derivatives

A4* -4
2
-5
-6
-7
1
-8
-9
H1*
-10
0
-11 H2*
-12 H3*
-1 -13
*
H4
-14
-15
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
U/fB U/fB

Fig. 6. Flutter derivatives for thin airfoil.

Fig. 7. Finite element model for natural modal analysis.

544 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

Fig. 8. Finite element model for full-order flutter analysis.

0.02
2.5
0.00
2.0

Imaginary part
-0.02
Real part

Complex mode 1 1.5

-0.04 Complex mode 2
Complex mode 3
Complex mode 4
-0.06 Complex mode 5
1.0
Complex mode 6
-0.08 Complex mode 7
Complex mode 8 0.5
Complex mode 9
-0.10 Complex mode 10
0.0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Wind velocity (m/s) Wind velocity (m/s)

Fig. 9. Variation of complex eigenvalues versus wind velocity: (a) real part and (b) imaginary part.

Table 3
1.0 Comparison of flutter analysis results for thin airfoil bridge

Method Flutter velocity (m/s) Flutter frequency (Hz)

0.8 Full-order method 135.1 0.3940
Generalized mass

Vertical Multimode method 134.3 0.3936

Torsional Exact solution 136.3 0.3914
0.6

0.4
Following the computational steps described in Section 2.3,
damped complex eigenvalue analysis is conducted for the hy-
0.2
brid FE model under wind velocities varying from 0 to 180 m/s.
The first 10 conjugate pairs of complex eigenvalues and com-
0.0 plex eigenvectors are obtained, and the variation of these com-
0 20 40 60 80 100 120 140 160 180 plex eigenvalues versus wind velocity is plotted in Fig. 9. It
Wind velocity (m/s)
is observed that, in the considered wind velocity range, (i) the
Fig. 10. Variation of generalized mass versus wind velocity.
vibration frequencies (i.e. the imaginary part of complex eigen-
values) of vertically bending modes exhibit a slight increase
with the increase in wind velocity while the real part decreases
Table 2 with increasing wind velocities; (ii) for laterally bending modes,
Results of two possible critical flutter states both the real and imaginary parts of complex eigenvalues re-
Flutter state Ũf f (Hz) Uf (m/s) Mainly contributing modes
main unchanged with the increase in wind velocity and (iii) the
imaginary part of complex eigenvalues for torsional modes de-
1 5.946 0.9085 216.1 4, 5 creases with the increase in wind velocity, while the real part
2 8.529 0.3936 134.3 1, 2
firstly decreases and then increases with increasing wind ve-
locities. As shown in Fig. 9(a), the real part of the second com-
plex mode becomes zero at a wind velocity of 135.1 m/s, and
Matrix27 elements are employed to formulate 29 aeroelastic the corresponding imaginary part of the complex eigenvalue
stiffness matrices and 29 aeroelastic damping matrices. The becomes 0.3940 Hz, identifying the onset of flutter instability.
hybrid FE model incorporating structural FE model with Ma- To illustrate the characteristics of coupled flutter, Fig. 10
trix27 elements for flutter analysis is illustrated in Fig. 8. provides the variation of the generalized mass in the vertical and
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 545

Fig. 11. Illustration of Humen suspension bridge.

2.0 2.0
1.5 1.5
1.0 1.0
0.5 0.5
Flutter derivatives

Flutter derivatives
0.0 0.0
-0.5 -0.5
-1.0 -1.0
-1.5
-1.5
-2.0
-2.5 -2.0
A*2 A2*
-3.0 -2.5
A3* -3.0 A3*
-3.5
-4.0 H1* -3.5 H1*
-4.5 -4.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 3 4 5 6 7 8 9 10 11 12
U/fB U/fB

Fig. 12. Measured flutter derivatives of bridge deck: (a) 0◦ and (b) +3◦ .

different combinations of reduced wind velocity and vibration

frequency, which indicates the susceptibility of multiple mode
combinations to flutter. But the combination that yields the
lowest critical flutter velocity is of practical value. In recog-
nizing that the lowest reduced wind velocity cannot guarantee
the lowest wind velocity as seen from Eq. (18), it is necessary
to search for all the possible critical flutter states within the
specified range of reduced wind velocity for selecting out the
lowest flutter wind velocity.
Following the computational steps described in Section 3,
two possible flutter states are found and given in Table 2. The
first flutter state is that of asymmetric modes (high-order modes)
where the characteristic motion of structure is dominated by
the asymmetric modes; while the second flutter state is that
Fig. 13. FE model of Humen Bridge for natural modal analysis.
of symmetric modes (low-order modes) where the motion of
structure at critical flutter state is governed by the symmetric
torsional directions for the second complex mode versus wind modes. In practice the second flutter state corresponding to
velocity. As expected, the second complex mode is a purely the lowest critical flutter wind velocity is of interest. Thus, the
torsional mode with the generalized mass in torsional direction flutter wind velocity and flutter frequency for the multimode
being unity when wind velocity is zero, and then it becomes method are 134.3 m/s and 0.3936 Hz, respectively.
a vertically and torsionally coupled mode with the increase in Table 3 gives a comparison of the results obtained by differ-
wind velocity. ent methods. The exact solution of flutter frequency and critical
wind velocity is obtained by using a two-mode classical flutter
theory [2,24]. It is seen that results of both full-order method
4.1.2. Multimode method of flutter analysis
and multimode method coincide well with the exact solutions.
Multimode flutter analysis is also performed to obtain
the critical flutter wind velocity and flutter frequency. Dur-
ing the multimode flutter analysis, the first 10 natural modes 4.2. Humen suspension bridge
for the original bridge structure are selected for participating
modes; the structural damping is also assumed to zero; and the The second application is the flutter analysis of Humen (Tiger
range of reduced wind velocity is from 1 to 15. In multimode Gate) suspension bridge, China. The Humen Bridge, as illus-
flutter analysis, multiple critical flutter states may occur at trated in Fig. 11, is a suspension bridge with a main span
546 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

Table 4
Modal frequencies and modal shapes of girder for Humen Bridge

Mode no. Frequency (Hz) Mode shape Mode no. Frequency (Hz) Mode shape

1 0.0901 S–L 6 0.2831 A–L

2 0.1144 A–V 7 0.2840 A–V
3 0.1612 A–V 8 0.3158 Cable
4 0.1743 S–V 9 0.3261 Cable–girder
5 0.2272 S–V 10 0.3297 S–T

Note: Cable—the mode shape of suspension cable.

Fig. 14. Hybrid FE model of Humen Bridge for full-order flutter analysis.

0.01 0.35

0.00 0.30
Imaginary part

-0.01 0.25
Real part

Complex mode 1
-0.02 Complex mode 2
0.20
Complex mode 3
Complex mode 4 0.15
-0.03 Complex mode 5
Complex mode 6
-0.04 Complex mode 7 0.10
Complex mode 9
Complex mode 10
-0.05 0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Wind velocity (m/s) Wind velocity (m/s)

Fig. 15. Variation of complex eigenvalues versus wind velocity obtained using theoretical flutter derivatives: (a) real part and (b) imaginary part.

Table 5 the bridge and only three flutter derivatives A∗2 , A∗3 and H1∗
Three possible flutter states for Humen Bridge have been obtained under wind attack angle of both 0◦ and
+3◦ [26,27]. Fig. 12 shows the measured flutter derivatives.
Flutter state Ũf f (Hz) Uf (m/s) Mainly contributing modes
With the advancement of experimental and system identifica-
1 9.180 0.2766 90.37 4, 5, 10, 16 tion techniques, all the 18 flutter derivatives can be identified
2 11.754 0.2973 124.37 2, 3, 22 from wind tunnel testing presently [28–30].
3 11.820 0.4531 190.67 4, 5, 30
The structural model, as shown in Fig. 13, is composed of
736 elements. The first 30 natural modes are extracted using
Lanczos method in ANSYS, and the natural frequencies and
of 888 m [25]. The bridge deck is a streamlined steel box girder mode shapes of the first 10 modes are summarized in Table 4.
of 36.1 m wide and 3.0 m high. The two main cables are 33 m
apart and the bridge deck is suspended by hangers with inter- 4.2.1. Flutter analysis using theoretical flutter derivates
vals of 12 m. The two bridge towers are 150.5 m high reinforced In recognizing the only three measured flutter derivatives
concrete structures. Wind tunnel tests have been conducted on and the closeness of the bridge deck to an airfoil, flutter is first
a sectional model of bridge deck and an aeroelastic model of carried out using theoretical flutter derivatives of the airfoil-like
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 547

Amplitude 1

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
400
Angle (°C)

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
0.4
Energy ratio

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Mode number

1
Amplitude

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
400
Angle (°C)

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
0.4
Energy ratio

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Mode number

1
Amplitude

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
400
Angle (°C)

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
0.5
Energy ratio

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Mode number

Fig. 16. Modal characteristics of the participating modes in the three critical flutter states: (a) the first; (b) the second; and (c) the third.
548 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

0.02 0.35

0.01 0.30

Imaginary part
0.00 0.25
Real part

-0.01
Complex mode 1 0.20
-0.02 Complex mode 2
Complex mode 3
Complex mode 4 0.15
-0.03 Complex mode 5
Complex mode 6
-0.04 Complex mode 7 0.10
Complex mode 9
Complex mode 10
-0.05 0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Wind velocity (m/s) Wind velocity (m/s)

Fig. 17. Variation of complex eigenvalues versus wind velocity obtained using measured flutter derivatives at 0◦ attack angle: (a) real part and (b) imaginary part.

0.04 Complex mode 1 0.35

Complex mode 2
0.03 Complex mode 3
Complex mode 4
Complex mode 5 0.30
0.02 Complex mode 6
Imaginary part
Complex mode 7
0.01 Complex mode 9 0.25
Real part

Complex mode 10
0.00
0.20
-0.01
-0.02 0.15
-0.03
0.10
-0.04
-0.05 0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Wind velocity (m/s) Wind velocity (m/s)

Fig. 18. Variation of complex eigenvalues versus wind velocity obtained using measured flutter derivatives at 3◦ attack angle: (a) real part and (b) imaginary part.

cross section. To implement full-order flutter analysis, a series where is the mode shape obtained at the previous wind ve-
of Matrix27 elements are attached to the nodes at the central locity step Ui−1 and  is the mode shape obtained at the cur-
girder. A total of 148 Matrix27 elements are used, half of them rent wind velocity step Ui . More the value of C is close to 1,
modeling the aeroelastic stiffness while the remaining modeling more likely the two mode shapes are similar. From Fig. 15,
the aeroelastic damping. Fig. 14 illustrates the FE model of flutter instability is originated from the model branch 10, and
the bridge incorporating Matrix27 elements for flutter analysis, flutter critical wind velocity and flutter frequency are estimated
where the boundary conditions for the constrained nodes are to 85.90 m/s and 0.2781 Hz, respectively. Then the complex
not displayed for clarity. eigenvalue analysis is conducted again by assuming that the
By disregarding structural damping and sweeping wind ve- damping ratios of all modes are 0.5%. In this case the critical
locity from 0 to 120 m/s, the first 10 conjugate pairs of com- flutter wind velocity and flutter frequency are computed to be
plex eigenvalues and complex eigenvectors are extracted and 92.23 m/s and 0.2752 Hz.
the variation of 9 of the 10 complex eigenvalues versus wind Multimode flutter analysis is also carried out to obtain the
velocity is shown in Fig. 15 (the mode 8 corresponding to the flutter critical wind velocity and flutter frequency. In the mul-
vibration of suspension cable is not shown). The frequency lines timode flutter analysis, the first 30 natural modes for the orig-
may intercross one another at varying wind velocities due to the inal bridge are selected for participating modes; the struc-
closely spaced eigenvalues. Thus, the closely spaced eigenval- tural damping ratios for all modes are assumed to 0.5%; and
ues of the suspension bridges make it quite difficult to correctly the range of reduced wind velocity is from 1 to 15. Table 5 lists
trace a specified mode during the sweep of wind velocity and the results of multimode flutter method and Fig. 16 illustrates
frequency iteration course. To ensure a correct mode trace, the the modal characteristics of participating modes in terms of
following correlation coefficient of mode shapes is examined modal phase angle , modal amplitude r and modal energy ratio
for each increment step of wind velocity: e in the three possible flutter critical states. For the first flutter
state, the modes 4, 5, 10 and 16 (symmetrical flexural and tor-
( T
)2 sional modes) are the mainly contributing modes; for the second
C= , (24)
( T
)(T ) flutter state, the modes 2, 3 and 22 (asymmetrical flexural and
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 549

1
Amplitude

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

400
Angle (°C)

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
0.4
Energy ratio

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Mode number

Fig. 19. Modal characteristics of the participating modes in the critical flutter state of the lowest flutter wind velocity obtained using measured flutter derivatives
0◦ attack angle.

1
Amplitude

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
400
Angle (°C)

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
0.5
Energy ratio

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Mode number

Fig. 20. Modal characteristics of the participating modes in the critical flutter state of the lowest flutter wind velocity obtained using measured flutter derivatives
3◦ attack angle.

torsional modes) are the mainly contributing modes; and for Then the measured flutter derivatives together with the theoret-
the last flutter state, the mainly participating modes are 4, 5 and ical flutter derivatives are used to predict bridge flutter under
30 (symmetrical flexural and high-order torsional modes). The the wind attack angle of both 0◦ and +3◦ , and the modal damp-
lowest flutter critical wind speed and the corresponding flutter ing ratios for all the concerned modes are set as 0.5% which
frequency are 90.37 m/s and 0.2766 Hz, which are in a close was also adopted and validated in the aeroelastic model test of
agreement with the full-order method. the bridge [26].
The flutter analyses are repeated using the full-order method
4.2.2. Flutter analysis using measured flutter derivates and the multimode method, respectively. Figs. 17 and 18 il-
As mentioned before, only three flutter derivatives were mea- lustrate the variation of the predicted complex eigenvalues
sured for the Humen Bridge. The remaining flutter derivative versus wind velocity obtained using the full-order method un-
parameters are taken from their theoretical values of an airfoil. der the wind attack angle of 0◦ and +3◦ , respectively. From
550 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551

Table 6
Comparison of flutter analysis results of Humen Bridge

Method Flutter velocity (m/s) Flutter frequency (Hz)

0◦ attack angle 3◦ attack angle 0◦ attack angle 3◦ attack angle

Full-order method 89.36 75.20 0.2973 0.3050

Multimode method 90.22 76.10 0.2998 0.3078
NACS 89.55 71.68 0.3009 0.3089
Wind tunnel tests 88 72 – –

Figs. 15, 18 and 19, it is found that the modal damping of com- Two case studies are provided to validate the capability of
plex mode 10, which is a torsional mode at zero wind velocity both methods in determining the flutter critical wind velocity
as shown in Table 4, turns from the negative value to the posi- and flutter frequency. In these examples, the flutter analysis
tive value with increasing wind velocity for all the cases studied results obtained by the full-order and multimode methods are
with the different sets of flutter derivatives. It is thus sufficient compared with those obtained by an analytical solution, com-
to trace the variation of the low-order torsional modes versus puter code NACS or wind tunnel testing, and a favorable agree-
wind velocity to identify the lowest critical flutter wind veloc- ment is observed. It is also found that only torsional modes
ity. Figs. 19 and 20 show the modal characteristics of the par- of a structure may become dynamic instability with increas-
ticipating modes in the critical flutter state of the lowest flutter ing wind velocity for the cases studied in the full-order flutter
wind velocity obtained with the multimode method under the analysis. Therefore, it may be sufficient to track the variation
wind attack angle of 0◦ and +3◦ , respectively. From Figs. 16, 19 of complex eigenvalues of the torsional modes versus wind ve-
and 20 it is clear that the low-order mode combination (namely locity for the purpose of reducing computational costs in the
modes 4, 5, 10 and 16) rather than the high-order model combi- full-order flutter analysis.
nations leads to the lowest critical flutter wind velocity, and the
mode combination is of the same mode shape characteristics Acknowledgments
in terms of symmetry and asymmetry. This is helpful to select
the relevant participating modes for multimode flutter analysis. The work described in this paper was supported by Nat-
Table 6 shows the flutter wind velocity and flutter frequency ural Science Foundation of China under (NSFC) Grant nos.
predicted by the full-order method and multimode method 50738002 and 50478051. These supports are gratefully ac-
and using the measured flutter derivatives at wind attack an- knowledged.
gle of both 0◦ and 3◦ . A computer code NACS, which was
originally developed for geometrical nonlinear analysis of
cable-supported bridges [31] and later expanded to include References
wind-induced flutter and buffeting analysis based on the M–S
[1] E. Simiu, R.H. Scanlan, Wind Effects on Structures: Fundamentals and
method [14], is also applied to flutter analysis of Humen Bridge
Applications to Design, third ed., Wiley, New York, 1996.
by assuming the same participating modes, modal damping ra- [2] F. Bleich, Dynamic instability of truss-stiffened suspension bridges under
tios and flutter derivatives and the results are given in Table 6 wind action, Trans. Am. Soc. Civ. Eng. 114 (1949) 1177–1222.
for comparison and validation. As mentioned previously, an [3] R.H. Scanlan, J.J. Tomko, Airfoil and bridge deck flutter derivatives,
aeroelastic full model of the whole bridge was tested in wind ASCE J. Eng. Mech. Div. 91 (1971) 1117–1137.
tunnel to obtain the critical flutter wind velocity as well and the [4] R.H. Scanlan, Action of flexible bridges under wind, 1: flutter theory,
J. Sound Vib. 60 (1978) 187–199.
experimental results are also provided in Table 6 [27]. A fa-
[5] R.H. Scanlan, N.P. Jones, Aeroelastic analysis of cable-stayed bridges,
vorable agreement between the results obtained by the present ASCE J. Struct. Eng. 116 (1990) 270–297.
procedures, NACS and wind tunnel testing are observed. [6] T. Miyata, H. Yamada, Coupled flutter estimate of a suspension bridge,
J. Wind Eng. Ind. Aerodyn. 33 (1990) 341–348.
5. Summaries and conclusions [7] N.N. Dung, T. Miyata, H. Yamada, N.N. Minh, Flutter responses in
long span bridges with wind induced displacement by the mode tracing
method, J. Wind Eng. Ind. Aerodyn. 77–78 (1998) 367–379.
In this study, the full-order and multimode methods of flutter
[8] Y.J. Ge, H. Tanaka, Aerodynamic analysis of cable-supported bridge by
analysis using ANSYS are presented. The full-order method uti- multi-mode and full-mode approaches, J. Wind Eng. Ind. Aerodyn. 86
lizes a novel FE model to represent the coupled wind-structure (2000) 123–153.
system and it is straightforward to implement in ANSYS. How- [9] Q.S. Ding, A.R. Chen, H.F. Xiang, Coupled flutter analysis of long-
ever, it is more computationally expensive. The multimode span bridges by multimode and full-order approaches, J. Wind Eng. Ind.
method takes advantage of modal superposition techniques to Aerodyn. 90 (2002) 1981–1993.
[10] J.M. Xie, H.F. Xiang, State-space method for 3-D flutter analysis of
reduce the computational efforts and can point out the mainly
bridge structures, in: Proceedings of the 1st Asia Pacific Symposium on
contributing modes to flutter instability; however, the computer Wind Engineering, India, 1985, pp. 269–276.
implementation in ANSYS is much more difficult than the full- [11] T.J.A. Agar, Aerodynamic flutter analysis of suspension bridges by a
order method. modal technique, Eng. Struct. 11 (1989) 75–82.
 X.G. Hua, Z.Q. Chen / Finite Elements in Analysis and Design 44 (2008) 537 – 551 551

[12] A. Namini, P. Albrecht, H. Bosch, Finite element-based flutter analysis [22] R.W. Clough, J. Penzien, Dynamics of Structures, second ed., McGraw-
of cable-suspended bridges, ASCE J. Struct. Eng. 118 (1992) 1509– Hill, New York, 1993.
1526. [23] Z.Q. Chen, T.J.A. Agar, Finite element-based flutter analysis of cable-
[13] H. Tanaka, N. Yamamura, M. Tatsumi, Coupled mode flutter analysis suspended bridges—discussions, ASCE J. Struct. Eng. 120 (1994)
using flutter derivatives, J. Wind Eng. Ind. Aerodyn. 42 (1992) 1044–1045.
1279–1290. [24] T. Theodorsen, General theory of aerodynamic instability and the
[14] Z.Q. Chen, The three dimensional analysis and behaviors investigation mechanism of flutter, NACA Report No. 496, 1935.
on the critical flutter state of bridges, in: Proceedings of the International [25] M.Z. Zheng, G.Z. Yang, The Humen Pearl River Bridge, Struct. Eng.
Symposium on Cable-Stayed Bridges, Shanghai, China, 1994, pp. 10–13. Int. 2 (1998) 93–94.
[15] A. Jain, N.P. Jones, R.H. Scanlan, Coupled flutter and buffeting analysis [26] State-key Laboratory for Disaster Reduction in Civil Engineering
of long-span bridges, ASCE J. Struct. Eng. 122 (1996) 716–725. (SLDRCE), Investigations on wind-resistant behavior of Humen
[16] H. Katsuchi, N.P. Jones, R.H. Scanlan, Multimode coupled flutter and suspension bridge, Research Report, Bulletin of Laboratory of Wind
buffeting analysis of the Akashi-Kaikyo Bridge, ASCE J. Struct. Eng. Tunnel, Tongji University, Shanghai, China, 1995 (in Chinese).
125 (1999) 60–70. [27] H.F. Xiang, A.R. Chen, J.Z. Song, On wind resistant properties of Tiger
[17] X. Chen, M. Matsumoto, A. Kareem, Aerodynamic coupling effects on Gate suspension bridge, Wind Struct. 1 (1998) 67–75.
flutter and buffeting of bridges, ASCE J. Eng. Mech. 126 (2000) 17–26. [28] A.R. Chen, X.F. He, H.F. Xiang, Identification of 18 flutter derivatives
[18] X.J. Zhang, H.F. Xiang, B.N. Sun, Nonlinear aerostatic and aerodynamic of bridge decks, J. Wind Eng. Ind. Aerodyn. 90 (2002) 2007–2022.
analysis of long-span suspension bridges considering wind-structure [29] A.G. Chowdhury, P.P. Sarka, Identification of eighteen flutter derivatives
interaction, J. Wind Eng. Ind. Aerodyn. 90 (2002) 1065–1080. of an airfoil and a bridge deck, Wind Struct. 7 (2004) 187–202.
[19] X. Chen, Improved understanding of bimodal coupled bridge flutter [30] H.W. Niu, Z.Q. Chen, X.G. Hua, A novel 3-DOF forced vibration system
based on closed-form solution, ASCE J. Struct. Eng. 133 (2007) 22–31. for identification of eighteen flutter derivatives, in: Proceedings of the
[20] A.H. Namini, Analytical modeling of flutter derivatives as finite elements, 12th International Conference on Wind Engineering, Australian Wind
Comput. Struct. 41 (1991) 1055–1064. Engineering Society, 2007, pp. 2247–2254.
[21] ANSYS User’s Manual, Version 8.0, Swanson Analysis Systems Inc. [31] Z.Q. Chen, T.J.A. Agar, Geometric nonlinear analysis of flexible spatial
(SASI), Houston, PA, 2004. beam structures, Comput. Struct. 49 (1993) 1083–1094.