Computers and Structures 212 (2019) 145–161

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

Integrated finite strip flutter analysis of bridges

Hamidreza Naderian a,b,⇑, Moe M.S. Cheung b,c, Majid Mohammadian d, Elena Dragomirescu d
a
Department of Civil and Mineral Engineering, University of Toronto, Canada
b
Department of Civil and Environmental Engineering, Hong Kong University of Science and Technology, Hong Kong
c
Western China Earthquake and Hazards Mitigation Research Centre, Sichuan University, Chengdu, China
d
Department of Civil and Environmental Engineering, University of Ottawa, Canada

a r t i c l e i n f o a b s t r a c t

Article history: The recently developed integrated finite strip method (IFSM) is extended to the aerodynamic flutter anal-
Received 14 April 2018 ysis of bridges under wind effects. The methodology is capable of performing a three-dimensional (3D)
Accepted 10 October 2018 flutter analysis in the IFSM environment, and the solution falls into the category of both multi-mode
Available online 3 November 2018
and full-mode flutter analysis. Aerodynamic stiffness and damping matrices, as well as structural prop-
erty matrices, are derived by IFSM. In addition, an optimal scheme is proposed for solving the flutter
Keywords: eigenvalue problem. Furthermore, a simple technique has been developed in order to handle different
Flutter
end boundary conditions. The proposed finite strip solution is very straightforward in terms of amount
Long-span cable-stayed bridge
Aerodynamic
of input data, boundary conditions, modeling, and the flutter analysis process. Moreover, the convergence
Aeroelasticity rate of the method is very high due to the semi-analytical and localization nature of the IFSM. Benchmark
B3 spline numerical investigations are presented, including the study of a simply supported thin flat shell and a
Integrated finite strip method model of the Kap Shui Mun Bridge, an existing long-span cable-stayed bridge. The numerical results show
that IFSM significantly improves the convergence of the critical flutter frequencies, and therefore leads to
smaller storage requirements and faster flutter eigenvalue extraction.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction solid-mechanics, dynamics, and fluid-dynamics can investigate the

aeroelasticity of an airfoil or a long-span bridge.
One of the main challenges in structures surrounded by wind Modern long-span bridges are usually constructed with very
flow is aerodynamic instability due to the effects of self-excited slender deck structures, which results in an increased risk of insta-
forces. If the wind-structure interaction leads to an increase in bility under extreme dynamic loads like wind and earthquakes.
the magnitude of oscillatory motions, then aeroelastic instability With the advent of new advanced materials such as composite
occurs. This is caused by aerodynamic forces induced on the struc- fiber-reinforced polymers (FRP), longer-span bridges are being
ture as a consequence of its motion. Such aerodynamic forces are built throughout the world. On the other hand, these types of
called self-excited forces, and this self-excited oscillatory instabil- materials are very lightweight, and consequently, the stability of
ity phenomenon is called flutter. Some examples of structures crit- bridge structures against wind effects has become more critical.
ical in countering flutter are airfoils in the aerospace industry like The nature of wind load is very complex and turbulent, which
aircraft wings, and long-span cable-stayed bridges in civil engi- means it is not easy to understand the structural behavior of
neering. In flutter, at a critical wind speed, the self-excited aerody- bridges under wind flow. Flutter is one of the most important
namic forces acting on an oscillatory system start feeding the criteria in the analysis and design of suspension and long-span
energy of the system instead of dissipating it, which leads to cable-stayed bridges. The determination of the dynamic and
destructive divergent vibrations in the structure. The aerodynamic aerodynamic characteristics of a particular bridge is a crucial step
behavior of a structure depends on the interaction between the in solving the flutter problem. Since the 1960s, a number of publi-
aerodynamic forces of inertia and damping, and the structure’s cations have dealt with both analytical and experimental studies
stiffness properties. Therefore, a multi-disciplinary study between on flutter in long-span bridges. Tanaka et al. [1], and Ge et al. [2]
investigated flutter in plates and bridges by multi-mode and full-
mode approaches. Huang et al. [3] studied the aerodynamic behav-
⇑ Corresponding author at: 35 Saint George Street, Dept. of Civil and Mineral ior of bridges under skew wind by performing a series of section
Engineering University of Toronto, Toronto, ON, M5S 1A4, Canada. model tests. Dowell et al. [4] investigated the flutter of rectangular
E-mail address: [email protected] (H. Naderian). plates under a three-dimensional axial flow through numerical and

https://doi.org/10.1016/j.compstruc.2018.10.003
0045-7949/Ó 2018 Elsevier Ltd. All rights reserved.
146 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

experimental tests, and Wu et al. [5] presented a framework for formulation of the self-excited flutter forces is based on applying
linear and nonlinear aeroelastic analyses of cable-supported the self-excited forces on the center of elasticity of the deck [19].
bridges. The topic of bridge flutter is still highly important, and According to the present research, however, the spatial distri-
researchers have tried to develop different types of experimental bution of self-excited aerodynamic forces is not only inefficient,
and analytical techniques for a better understanding of the flutter but also it might be against general assumptions of the flutter for-
phenomenon [6–12]. mulation, which might lead to less accurate results. Although the
The finite strip method (FSM), pioneered by Y. K. Cheung and M. idea behind the uniform distribution of wind loads is to consider
S. Cheung [13,14], is one of the most efficient and accurate numer- the effects of the spatial distribution of aerodynamic forces over
ical techniques for modelling long-span structures with highly har- the cross-section of the bridge deck, our recent studies indicate
monic behaviour along the longitudinal direction. Among these, that the aerodynamic performance is not sensitive with regard to
one can mention buckling and free vibration of thin-walled struc- this factor. Therefore, spatial discretization of aerodynamic loads
tures, as well as the structural investigation of long-span bridges. on the entire deck system decreases the efficiency of the solution
Despite the merits of the finite strip method, its application to procedure. The latter will be discussed more in detail later in this
bridge engineering has been limited to modeling only the bridge paper when providing the numerical examples. Last but not least,
deck, while other structural components of the structure, including Cheung et al.’s [18] proposed method might not perfectly and effi-
piers, towers, cables, etc., have been modeled as special boundary ciently consider the structural interactions between decks, piers,
conditions of the deck [13]. For that reason, and with the advance- towers, and cables because the compatibility between the dis-
ment of other numerical techniques, the popularity of the finite placements of the interaction joints is obtained through an itera-
strip method among bridge engineers has declined in recent dec- tive process. However, this technique could be effective for
ades. In order to overcome this obstacle, Naderian et al. [15,16] simple types of structures under static/quasi-static forces. In the
developed the so-called Integrated Finite Strip Method (IFSM), case of complex structures like long-span cable-stayed bridges or
which is able to model an entire cable-stayed bridge system. IFSM structures under dynamic and aerodynamic excitations such as
is, in fact, an evolved version of the Spline Finite Strip Method non-uniform seismic excitations or aerodynamic self-excited
(SFSM) [17], which is itself a more comprehensive format of the forces, such simplification can no longer be adopted.
conventional Finite Strip Method (FSM), at the expense of incorpo- All the above-mentioned concerns have been eliminated in the
rating more degrees of freedom. However, as a tool, this is more present study by introducing the Integrated Finite Strip Flutter for-
powerful than the ordinary finite strip method in terms of dealing mulation, as follows: the shell spline finite strip models the bridge
with structures with complex geometry, boundary, and loading deck, the one-dimensional column strip models the towers and
conditions. Moreover, SFSM is also capable of modeling multi- piers, and the cable strip models the bridge cables. Transition sec-
span bridges and structures subjected to not only uniformly dis- tion elements in each of these strips provide the connectivity and
tributed forces but also concentrated forces. In IFSM, the bridge boundary conditions between different structural components of
deck is divided into a number of flat shell spline finite strips in a cable-stayed bridge.
the longitudinal direction, and B3 spline functions define the longi- In this research, first the IFSM is briefly reviewed, and then the
tudinal shape function of the displacement function. Using B3 general formulation of the flutter equation of motion is presented.
splines, the continuity over the second derivative of the displace- The integrated finite strip flutter methodology along with the
ment function is achieved. Furthermore, the splines have localiza- derivation of the aerodynamic stiffness and damping matrices, as
tion properties through which the bandwidth of the global well as the structural property matrices, are fully explained. In
property matrices becomes very narrow. As a result, the computa- addition, a simplified method has been developed to handle the
tional time can be dramatically decreased. When it comes to com- boundary conditions corresponding to the flutter problem. Thirdly,
plicated analyses, for instance, vibration-based real-time structural multi-mode and full-mode aeroelastic approaches are integrated
health monitoring of huge bridge structures under earthquake, within the finite strip treatment. Finally, a simple eigenvalue
typhoon, or high traffic loads, where numerous iterative dynamic analysis based on linearization of the flutter equation of motion
analyses must be performed consistently, even a small percentage is proposed through which the critical flutter frequencies and con-
of computational savings is highly appreciated. Therefore, IFSM can sequently critical wind speeds can be obtained by solving a stan-
be a suitable platform for the vibration investigation of modern dard eigenvalue analysis. The proposed finite strip flutter
bridge structures. Naderian et al. [16] recently extended the finite solution is applied to two numerical examples where the flutter
strip method to the seismic analysis of long-span cable-stayed behavior of a thin-walled flat plate, as well as the long-span
bridges. In this paper, the IFSM is further extended to aerodynamic cable-stayed Kap Shui Mun Bridge, are investigated. The accuracy
flutter analysis. of the method in predicting critical flutter speeds and frequencies,
It is worth mentioning that M. S. Cheung et al. [18] used a com- as well as the convergence of the currently developed finite strip
bination of the spline finite strip method and the finite element method for flutter analysis, is examined. The results show that
method for flutter analysis of long-span cable-stayed bridges. IFSM and the proposed flutter analysis scheme is a powerful and
Despite the merits of their method, there are still considerable con- efficient tool for the aerodynamic analysis of a plate in general
cerns about the efficiency and accuracy of the solution, which are and for bridge structures as a system. Rapid convergence and small
addressed in the following: (I) the aerodynamic stiffness and storage requirements are highlights of the proposed method.
damping matrices are not derived by the spline finite strip method
but are based on lumped aerodynamic forces. The distribution of 2. Review on integrated finite strip method
the aerodynamic forces along the nodal lines of the strip is based
on trapezoidal discretization, while in the spline finite strip The Integrated Finite Strip Method (IFSM) [15,16] is the latest
method it must be done by B3 spline discretization; (II) Only the version of the finite strip method and is able to fully simulate
deck is modeled by spline finite strips, while other components like and analyze the bending and vibration of long-span cable-stayed
piers, towers, and cables are simulated by equivalent beam finite bridges. Not only the bridge deck but also all other structural parts
elements. In other words, their numerical solution is a partial finite of a long-span cable-stayed bridge including towers, piers, linked
strip solution; and (III) The aerodynamic self-excited forces are beams, and cables, as well as bearings, can be modeled in the envi-
spatially distributed on the deck of the bridge, while the general ronment of the spline finite strip method by using different types
 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 147

of strips, including shell spline strips, columns strips, and cable 4. Integrated finite strip flutter formulation
strips. The key parameter of the originality of the IFSM is the so-
called transition section, which is based on unequal spline strips. 4.1. General
Transition sections provide compatibility of displacements and
vibrations in the intersection joints of different components, and Both the structural and aerodynamic properties of different
they also handle the boundary conditions of bearings. The compat- components of a bridge can be derived by the integrated finite strip
ibility can be achieved by selecting an extremely small length for a method. In the case of a long-span cable-stayed bridge, only the
spline transition section [15]. Another important feature of the self-excited flutter forces acting on the bridge deck are considered
IFSM is that the structural interactions between different compo- in the analysis. In the following sub-section, the integrated finite
nents, for example between towers and cables, can be modelled. strip discretization as well as the derivation of the structural and
This capability is very important, especially for dynamic analyses aerodynamic matrices in the environment of IFSM are presented.
of long-span bridges like seismic analysis and flutter analysis, in
which there is considerable interaction and load effect transmis- 4.2. Integrated finite strip discretization
sion between different structural elements. More details on the
development of the IFSM have been addressed by Naderian et al. Considering the bridge deck as a shell structure, both in-plane
[15]. and out-of-plane degrees of freedom need to be considered. A flat
shell spline strip is shown in Fig. 1, where each knot of a nodal line
has four degrees of freedom, three translational (u, v, w), and one
3. Aerodynamic flutter equation of motion
rotation ðhÞ.
The vector of the displacement parameters of a shell spline strip
The self-excited aeroelastic forces acting on a plate or a bridge
centered at ym is given as:
structure are caused by wind effects. This means that flutter-
resultant forces can be considered as external forces. Similar to fdgm ¼ ½uim ; mim ; wim ; him ; ujm ; wjm ; hjm T ð5Þ
general dynamic systems, the equilibrium of the effective forces,
including the inertia force, the damping force, the elastic force, It must be noted that the parameters of Eq. (5) are in fact spline
and the applied flutter force, will form the flutter equation of the coefficients and not the physical degrees of freedom. The inte-
motion. For a structure for which displacements are defined by a grated finite strip method is based on the use of unequally spaced
vector fdg including n degrees of freedom, matrix analysis can be B3-spline functions, through which the transition section is devel-
used to formulate the flutter equation of motion as: oped for connecting different structural components together.
Moreover, it allows the locations of the supports and the concen-
n o n o
½Ms  d€ þ ½C s  d_ þ ½K s fdg ¼ fF aer g ð1Þ trated load to coincide with the knots on the nodal lines, thus lead-
ing to more accurate results. Moreover, the introduction of
where ½Ms , ½C s , and ½K s  are the global structural mass, damping, unequally spaced interior knots allows one to describe accurate
and stiffness matrices respectively with the order of n  n, while responses in the region of high stress gradients, or at the locations
n o of abrupt geometric changes, by spacing the knots more closely.
fF aer g is the global self-excited aerodynamic force vector, and d_ In a flutter analysis problem of a flat shell or a bridge deck, the
n o
and € d are the vectors of the first and second derivatives of dis- unequally spaced B3-spline function centered at ym may define the
longitudinal shape of function as:
placement respectively. The self-excited forces fF aer g are dependent 8
on the vertical (w), lateral (u), and rotation ðhÞ displacements of the >
> 0 y < ym2
>
>
structure and their first derivatives, and it can be expressed as: >
> Am ðy  ym2 Þ3 ym2 6 y < ym1
>
>
n o >
>
< Am ðy  y Þ3 þ C m ðy  y Þ3 ym1 6 y < ym
fF aer g ¼ ½C aer  d_ þ ½K aer fdg ð2Þ Um ðyÞ ¼
m2 m1
ð6Þ
> Bm ðymþ2  yÞ3 þ Dm ðymþ1  yÞ3
> ym 6 y < ymþ1
>
>
in which ½C aer  and ½K aer  are the n  n global aerodynamic damping >
> B ðy
>
> m mþ2  yÞ
3
ymþ1 6 y < ymþ2
>
>
and stiffness matrices respectively. The self-excited aerodynamic :
0 ymþ2 6 y
force associated with the second derivative of displacement, called
aeroelastic inertial forces, is omitted, as the air density is signifi- in which
cantly lower than that of the structural materials. Substituting Eq.  1
(2) into Eq. (1) yields the flutter equation of motion as: Am ¼ ðymþ1  ym2 Þðym  ym2 Þðym1  ym2 Þ
 1
n o n o n o Bm ¼ ðymþ2  ym1 Þðymþ2  ym Þðymþ2  ymþ1 Þ
½Ms  d€ þ ½C s  d_ þ ½K s fdg ¼ ½C aer  d_ þ ½K aer fdg ð3Þ 
C m ¼ ðymþ2  ym2 Þ ðymþ2  ym1 Þðymþ1  ym1 Þðym  ym1 Þðym1  ym2 Þ
1
 1
Dm ¼ ðymþ2  ym2 Þ ðymþ1  ym2 Þðymþ1  ym1 Þðymþ1  ym Þðymþ2  ymþ1 Þ
or
ð7Þ
n o n o
½M €d þ ½C  d_ þ ½K fdg ¼ 0 ð4Þ The section lengths of the B3 spline function can be chosen to be
equal for flutter analysis of simple cases like a long-span flat shell
in which ½M ¼ ½Ms , ½C  ¼ ½C s   ½C aer , and ½K  ¼ ½K s   ½K aer , where where there is no need to employ unequally spaced splines. The
½M , ½C , ½K  are the system mass, damping, and stiffness matrices length of the plate strip L is divided into m sections of equal length
respectively. The aerodynamic damping ½C aer  and stiffness matrices h. A typical local B3-spline function of equal spaced is defined as:
½K aer  are not necessarily symmetrical because of the non- 8
>
> ðy  ym2 Þ3 ym2 6 y 6 y  ym1
conservative nature of the aerodynamic forces. Therefore, the sys- >
>
>
> 3 2 2 3
> h þ 3h ðy  ym1 Þ þ 3hðy  ym1 Þ  3ðy  ym1 Þ ym1 6 y 6 ym
tem stiffness ½K  and system damping ½C  matrices are asymmetric, 1 <    
Um ðyÞ ¼ 3 h3 þ 3h2 ymþ1  y þ 3hðymþ1  yÞ2  3 ymþ1  y 3 ym 6 y 6 ymþ1
which results in coupling effects between the flutter response fre- 6h >
>
>
> ðy
> 3
quencies. In other words, the asymmetry of the aerodynamic prop- >
> mþ2  yÞ ymþ1 6 y 6 ymþ2
:
erties enables the flutter response of the structures to be coupled 0 otherwise
between multiple mode shapes. ð8Þ
148 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

Fig. 1. Shell spline finite strip.

The membrane displacement functions u and v and the flexural ½Uim  ¼ ½ U1 U0 U1 U2 .. .. . . Um2 Um1 Um Umþ1  ð14Þ
displacement function w are expressed as the product of transverse
½U1m , ½U2m , ½U5m  and ½U6m  are related to the displacements u and v
polynomials and longitudinal B3-splines:
of nodal lines i and j respectively, while ½U3m , ½U4m , ½U7m  and ½U8m 
X
rþ1 are related to the displacement w.
u¼ ðN 1 U1m ðyÞuim þ N2 U5m ðyÞujm Þ ð9Þ In the short form, Eq. (12) is expressed as:
m¼1
ff g ¼ ½Nt ½Ufdg ð15Þ
X
rþ1
where ½N t  is a 3  8 matrix and ½U is an 8  8 (m + 3) matrix, while
v¼ ðN1 U2m ðyÞv im þ N2 U6m ðyÞv jm Þ ð10Þ
m¼1
fdg is an 8 (m + 3)1. Thus, fdg is an 8 (m + 3)1 matrix.

X
rþ1 4.2.1. Column and cable strips
w¼ ðN3 U3m ðyÞwim þ N 4 U4m ðyÞhim þ N 5 U7m ðyÞwjm The cantilever behavior of the piers and towers of a long-span
m¼1 cable-stayed bridge can be simulated by the so-called three-
þ N6 U8m ðyÞhjm Þ ð11Þ dimensional (3D) Column Strip (CS3), which is a vertical shell
spline strip fixed at one end, for providing the support boundary
where r is the total number of longitudinal sections on a nodal line. conditions, and free at the other end, as shown in Fig. 2.
In the matrix form, Eqs. (9)–(11) can be rewritten as: In some cases, a one-dimensional (1D) column strip (CS1) is
8 9 2 3 more applicable and is actually easier to be employed for mod-
<u>
> = N1 0 0 0 N2 0 0 0
elling piers and towers. In a CS1, each knot belonging to the nodal
6 7
ff g ¼ v ¼ 4 0 N1 0 0 0 N2 0 0 5 line has three translational degrees of freedom. Consequently, the
>
: > ;
w 0 0 N3 N4 0 0 N5 N6 displacement function can be defined as:
2 38 9
½U1m  > uim >
> > X
rþ1
6 >
7> >
> u¼ um Um ðzÞ ð16Þ
> v im >
7>
6 ½U2m  > >
6 7> >
>
6 7>
> >
>
m¼1
6 ½U3m  7>
> wim >
> >
>
6 7> > X
>
7< him >
rþ1
6 ½U4m  =
6 7 w¼ wm Um ðzÞ ð17Þ
 6 7
6 ½U5m  7> >
7>
> ujm >
m¼1
6 >
6 7>
> >
>
6 ½U6m  7>
> v
> jm >
>
>
6 7>
> >
6 7>w >
> >
>
4 ½U7m  5>
> jm >
>
>
: >
;
½U8m  hjm
ð12Þ
The transverse shape functions adopted in Eq. (12) are cubic
Hermite polynomial functions for vertical displacement variation
and linear interpolation for in-plane displacements as:

N 1 ¼ 1  X; N2 ¼ X; N3 ¼ 1  3X 2 þ 2X 3 ; N4 ¼ xð1  2X þ X 2 Þ;
N 5 ¼ ð3X 2  2X 3 Þ
N 6 ¼ xðX 2  XÞ
ð13Þ
In Eq. (12), ½U1m  to ½U8m  are the row matrices, and each of them
includes (m + 3) local B3-splines as: Fig. 2. Column Strip (CS3).
 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 149

X
rþ1 For example, in order to model a fixed bearing, which allows rota-
v¼ v m Um ðzÞ ð18Þ tion but restricts translations, the displacements of knots 3 and 8
m¼1
must be equal in order to achieve compatibility, which can be
For the displacement-strain relationships, only the bending in implemented through the transition section as detailed above.
the vertical and transverse directions and the axial stress are con-
sidered, while the shear stress and torsional moment are assumed 4.3. Finite strip structural property matrices
to be negligible, as the amounts of these forces are very low in CS1.
In order to model the bridge cables in the finite strip environ- After the discretization of the entire bridge system is performed
ment, the cable strip is employed. The cable strip is a simplified ver- using the IFSM, the principle of minimum potential energy can be
sion of the CS1. Generally, cables can resist only axial tension applied to derive the mass matrix [m] and stiffness matrix [k] of
stresses and not compression. For this reason, in the cable strip the shell strip, column strip, and cable strip, as follows:
model, only the axial stress defines the strain-displacement Z
relationship. ½m ¼ ½NTi q½Ni dV ð19Þ

Z
4.2.2. Connectivity provisions
½k ¼ ½BTi ½D½Bi dV ð20Þ
In contrast with the finite element method in which auxiliary
bridge structural elements like piers and cables can be easily con-
in which q is the density of the strip, [D] and [B] are the elastic
nected to the bridge deck, the concept ‘‘element” in the longitudi-
matrix and the strain matrix respectively, and [N] is the shape func-
nal direction is not defined in the conventional finite strip
tion matrix, which is the multiplication of the transverse functions
methods. In other words, it is impossible to insert an extra compo-
and the longitudinal spline functions as ½Nt ½U. It must be noted
nent at an intermediate point within a spline strip. In order to
that in the case of a one-dimensional column strip (CS1) and a cable
eliminate this impediment, a special transition section has been
strip, the integral is performed over the length of the strip. Similar
developed within the IFSM [15]. The methodology has been
to the FEM, the IFSM strip properties are converted to the knots of
applied to connect different components of complex structures
the nodal lines along the strip during the model formulation pro-
such as long-span cable-stayed bridges [15]. Therefore, deck-pier
cess. However, the number of knots required is significantly
and deck-tower connectivity as well as cable-deck and cable-
reduced due to the semi-analytical nature of the IFSM. The struc-
tower connectivity are modelled by employing these newly devel-
tural matrices of the strips can be assembled using the standard
oped transition section elements.
assembly techniques, and the global structural stiffness matrix
The transition section is created by using unequally spaced B3-
½K s  and global structural mass matrix ½Ms  of the entire bridge sys-
spline functions. A typical transition section connecting two differ-
tem are formed after assembling all structural components.
ent components is shown in Fig. 3. Without losing generality, it is
The structural damping matrix ½C s  is usually a function of the
assumed that the length of the normal and the transition sections
structural stiffness and mass matrices, and it can be obtained from
are H and Ht respectively. In Fig. 3, the vertical line is a nodal line
the following formula, referred to as ‘‘classic Rayleigh damping”:
on the column strip, while the horizontal line is a nodal line on the
shell spline strip of the deck. This schematic can also be applied to ½C s  ¼ a½Ms  þ b½K s  ð21Þ
the cable strip and shell spline strip joints in order to ensure con-
where a and b are the Rayleigh damping factors, which can be
nectivity. The vertical and horizontal lines overlap at knots 3 and 8
determined by having two structural damping ratios n associated
of the shell and column strips, respectively, and in order to achieve
with two specific frequencies. The damping ratio n of the nth free
identical displacements at knots 3 and 8, the ratio of Ht =H should
vibration mode of a structure is given by:
be infinitely small. Depending on the complexity of the structure,
different values of h can meet the required accuracy. Numerical 1 x
nn ¼ aþ nb ð22Þ
studies have shown that under normal circumstances good accu- 2xn 2
racy can be obtained with the ratio of Ht /H = 0.001 [15]. Using
where xn is the angular frequency of the nth mode. Assuming the
the developed transition section in the spline finite strip procedure,
same damping ratio n for two different modes, a and b can be
compatibility for the displacements of different components of the
calculated.
structure is satisfied. It must be noted that the bearings can be
modeled as special boundary conditions for the transition section.
4.4. Finite strip aerodynamic property matrices

4.4.1. Aerodynamic self-excited forces

In order to derive the aerodynamic damping and stiffness
matrices of a flat shell or a bridge deck using the finite strip
method, first, the self-excited flutter forces need to be recognized.
According to the linear aerodynamics, three types of uniformly dis-
tributed line loads acting along the center of elasticity of a moving
structure, including lift force Lf , drag force Df , and pitching
moments Mf , as illustrated in Fig. 4, define the self-excited flutter
forces per unit length of the structure as:
 
1 w0 Bh0 w Bu0 u
Lf ¼ qair U 2 B K w H1 þ K w H2 þ K 2w H3 h þ K 2w H4 þ K w H5 þ K 2w H6
2 U U B U B
ð23Þ
 
1 u0 Bh0 u w0 w
Df ¼ qair U 2 B K w P 1 þ K w P2 þ K 2w P 3 h þ K 2w P 4 þ K w P 5 þ K 2w P 6
2 U U B U B
Fig. 3. Typical transition section element. ð24Þ
150 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

 0 0 0

1 w  Bh  w u u
M f ¼ qair U 2 B2 K w A1 þ K w A2 þ K 2w A3 h þ K 2w A4 þ K w A5 þ K 2w A6
2 U U B U B
ð25Þ

where u, w, and h are the vertical bending, lateral bending, and tor-
sional displacements, as shown in Fig. 5, while u0 , w0 ; and h0 are the
first derivatives of the corresponding displacements. K w is the non-
dimensional reduced frequency given by K w ¼ Bxf =U, where xf is
the flutter angular frequency xf , B is the deck width, and U is the
mean wind velocity. The terms Hi , P i , and Ai (i = 1, 2, 3, . . . , 6) are
the aerodynamic derivatives associated with the self-excited lift
Lf , drag Df , and moment Mf aerodynamic forces respectively. The
flutter derivatives are all functions of the reduced frequency K w
and are usually measured from experimental tests performed on
the cross-section of the structure. Also, qair refers to the air density.
When the cross-section of the deck is a slab girder, the center of
elasticity is equal to the geometric shear center, while in the case Fig. 5. Aerodynamic displacement components.
of a box girder bridge, the aerodynamic forces need to be mapped
on the nodal lines of the strip. This procedure, however, is out of Substituting Eq. (27) into (26) yields:
the scope of the current research, as the deck of the bridge is Z Z
assumed to be a slab girder. Nevertheless, this can be achieved sim- ff aer g ¼ ½/T ½Nt T Lf dA þ ½/T ½Nt T Df dA
ply by adjusting the displacement functions. Also, it must be noted
Z
that the wind velocity acting on the cross-section of the plate (deck)
þ ½/T ½Nt T M f dA ð28Þ
might come from different angles of attack. It is the responsibility of
the aerodynamic design engineer to investigate which wind attack
angle(s) can cause the maximum aerodynamic force and subse- where ½/T is the longitudinal B3 spline function of Eqs. (6) or (8).
quently instability of the structure. ½N t  is the transverse shape function of Eq. (13), where its integra-
Rb
tion 0 ½Nt T dx over the width of strip b corresponding to the aerody-
4.4.2. Spline discretization of aerodynamic forces
namic forces Lf , Df , and Mf is given by the following matrices
In conventional analytical methods and the finite element
method, the uniformly self-excited forces are linearly lumped respectively:
along the structure, whereby one-half of each element load acts 2 3
upon each element-end. Herein, the B3-spline interpolation is 0 0 0 0 0 0 0 0
6 7
employed to distribute the aerodynamic forces along the plate. 40 0 0 0 0 0 0 05 ð29Þ
The spline discretization of the aerodynamic forces leads to a more 0 0 1 0 0 0 0 0
accurate distribution because of the higher continuity degree (C2).
In the finite strip formulation, it is more convenient to locate a 2 3
nodal line passing through the aerodynamic line loads, acting on 1 0 0 0 0 0 0 0
6 7
the centre of elasticity of the deck. Applying the principle of min- 40 1 0 0 0 0 0 05 ð30Þ
imum potential energy in the IFSM environment, the self-excited 0 0 0 0 0 0 0 0
aerodynamic forces originating from the wind velocity as an exter-
nal wind-induced force can be distributed on the knots along the 2 3
nodal line of a spline strip. Integrating the forces over the elements, 0 0 0 0 0 0 0 0
6 7
one obtains: 40 0 0 0 0 0 0 05 ð31Þ
Z
T 0 0 0 1 0 0 0 0
ff aer g ¼ ½N qdA ð26Þ
The integration of the spline representations of Eqs. (6) or (8)
where q represents the self-excited aerodynamic forces given by: over the length of the strip L for each degree of freedom (displace-
ment parameter) results in m + 3 values corresponding to the knots
q ¼ Lf þ Df þ M f ð27Þ 
of a nodal line. This integral can be called /f . Assuming the

Fig. 4. Self-excited aerodynamic forces [1].

 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 151

location of the aerodynamic loads on the first nodal line of the strip conventional assembly techniques. Finally, the aerodynamic equa-
i, and after mathematical rearrangement, Eq. (28) is expanded to tion of motion for the entire structural system in the IFSM environ-
the following matrix formulation: ment is formulated in the general form of Eq. (2).
2 38 9
P 4 ½/fdiag  ½0 P 6 ½/fdiag  BP3 ½/fdiag  > u
6 > i >
7> >
> 4.5. Boundary conditions
6 ½0
6 
½0 ½0 ½0 7>
>
7>
>
v i >
>
>
6 H ½/fdiag  > >
7> >
6 6 ½0 H4 ½/fdiag  BH3 ½/fdiag  7>
> wi > >
> The strip in the FSM must be defined with preset boundary con-
6  >
7< >
> >
1 6
2 2 6 BA6 ½/fdiag  ½0 BA4 ½/fdiag  B2 A3 ½/fdiag  7 hi = ditions. There are several techniques that can be applied to model
ff aer g ¼ qair U K w 6 7
2 7> u >
6 7> j > the boundary conditions of the strip [13]. Despite the advantages
6 7>
> > >
6 7>
> vj > >
6 7>
>
> >
>
>
of the spline finite strip method over the finite element method
6 7> >
4 5>
>
> >w j
>
> in terms of computational efficiency, handling of a complex
: ;
hj amended scheme of local splines for considering the end and inter-
2 38 0 9
P 1 ½/fdiag  ½0 P 5 ½/fdiag  BP 2 ½/fdiag  > ui >
> >
nal boundary conditions, keeps this solution untidy. In other
6 7>
> v0 > > words, current amendment schemes for boundary conditions are
6 ½0 ½0 ½0 ½0 7>
> i> >
6  7>
> 0>
> >
> unable to be generalized, and dealing with boundary conditions
6 H ½/fdiag  ½0 H1 ½/fdiag  BH2 ½/fdiag  7>
> w >
i>
6 5 7>
> >
>
6  7> > using standard techniques like penalty functions is complicated.
7< hi =
0
1 6 BA5 ½/fdiag  ½0 BA1 ½/fdiag  B2 A2 ½/fdiag 
þ qair UBK w 6
6
7
7> u0 > Herein, a straightforward method for modelling boundary condi-
2 7>
6
6 > j >
7> >
> tions based on replacing the spline displacement parameters with
6 > v 0j >
7> >
6 7>
> >
> >
6 7>
> 0>
>
>
physical degrees of freedom is proposed. This will result in a gen-
4 5> w j>
>
> 0 > > eral unified formulation of otherwise very complex and tedious
: ;
hj
amended schemes of local splines in the vicinity of the boundary
ð32Þ supports and at any internal support. This makes the method more
   versatile and adjustable with other numerical techniques.
in which /fdiag is the diagonal form of /f with the order of (m
For an equally spaced B3-spline function, one can rewrite the
+ 3)  (m + 3), and is given by: physical displacements uphy ; v phy ; wphy and the rotation hphy at the
2 3 end knots of a nodal line i of a spline strip as:
/f 1
  6 .. 7 1
/fdiag ¼ 6
4 .
7
5 ð33Þ uphy0 ¼ ðu1 þ 4u0 þ u1 Þ
6
/fmþ1 1
v phy0 ¼ ðv 1 þ 4v 0 þ v 1 Þ
6
1
4.4.3. Finite strip aerodynamic stiffness and damping matrices wphy0 ¼ ðw1 þ 4w0 þ w1 Þ
6
The first and second matrices in Eq. (32) are the aerodynamic 1
stiffness and damping matrices of a spline strip respectively. As hphy0 ¼ ðh1 þ h1 Þ
2h ð36Þ
there are m + 3 knots on each nodal line, the number of each dis- 1
placement parameter in Eq. (32) will be m + 3. Applying the spline uphym ¼ ðum1 þ 4um þ umþ1 Þ
6
finite strip derivation, the aerodynamic stiffness and damping 1
matrix of a shell spline strip can be obtained by: v phym ¼ ðv m1 þ 4v m þ v mþ1 Þ
6
2 3
P4 ½/fdiag  ½0 P 6 ½/fdiag  BP3 ½/fdiag  1
wphym ¼ ðwm1 þ 4wm þ wmþ1 Þ
6 7 6
6 ½0 ½0 ½0 ½0 7
6  7 1
6 H ½/  ½0 H ½/  BH ½/  7 hphym ¼ ðhm1 þ hmþ1 Þ
6 6 fdiag 4 fdiag 3 fdiag 7 2h
6 7
6 BA ½/  ½0 BA ½/  B2 A ½/  7
1 6 7 where h is the spline section length. Dividing a spline strip into m
½kaer  ¼ qair U 2 K 2w 6 6 fdiag 4 fdiag 3 fdiag
7 ð34Þ
2 6 7 equal sections will give 8(m + 3) spline parameters which define
6 7
6 7 the displacement function of the shell spline strip. Similarly, the
6 7
6 7 displacements and rotations of the intermediate knots are also pre-
6 7
4 5 sented by spline parameters. The transformation matrix corre-
sponding to each degree of freedom ui v i wi hi uj v j wj hj is
expressed by:
2 3 2 1 3
P 1 ½/fdiag  ½0 P 5 ½/fdiag  BP 2 ½/fdiag  0 1
6 7 2h 2h
6 ½0 ½0 ½0 ½0 7 6 7
6  7 61 4 1 7
6 H ½/fdiag  ½0 H1 ½/fdiag  BH2 ½/fdiag  7 66 7
6 5 7 6 6 6
7
6  7 6 7
1 6 BA5 ½/fdiag   
½0 BA1 ½/fdiag  B2 A2 ½/fdiag  7 6 1 4 1 7
½caer  ¼ qair UBK w 6
6
7
7 6 6 6 6 7
2 6 7 6 7
6 7 6 7
6 7 6 7
6 7 6 7
6 7 6 7
4 5 6 .. 7
T¼6 . 7 ð37Þ
6 7
6 7
6 7
ð35Þ 6 7
6 7
6 7
Except for the spline strip with self-excited forces, the aerody- 6 1 4 1 7
6 6 6 6 7
namic stiffness and damping matrices of the rest of the strips in 6 7
6 7
the entire IFSM model is zero, as there is no self-excited force act- 6 1 4 1
7
4 6 6 6 5
ing on other elements of the model. The global aerodynamic stiff- 1 1
0
ness ½K aer  and damping ½C aer  matrices are formed using the 2h 2h
152 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

8n o9
where T is a 8  8 (m + 3) matrix. The global transformation matrix ( ) >
n o _
zn < €d >
=  kfwg
o
of a flat shell spline strip is presented as: X_ ¼ ¼ n o ¼ kekt ð47Þ
2 3 d_ : d_ >
> ; fwg
T
6 T 7
6 7
6 7
6 T 7 5.1. Full-mode finite strip aeroelastic analysis
6 7
6 T 7
6 7
Tt ¼ 6 7 ð38Þ
6 T 7 In the full-mode flutter analysis, the flutter eigenvalues of an
6 7
6 T 7 original n degrees of freedom system, rather than an m order sys-
6 7
6 7 tem modified by m natural modes, is calculated. The full-mode
4 T 5
aeroelastic analysis employs the vector iteration method to solve
T partial eigenvalues of Eq. (43). The iterative analysis must converge
Therefore, the transformed structural matrices can be obtained to the most dominant value of k, which is equal to the highest fre-
by: quency xf . For a complicated structural system like a long-span
cable-stayed bridge with a massive number of degrees of freedom,
½K st  ¼ ½T t T ½K s ½T t  ð39Þ full-mode eigenvalue flutter analysis requires a large amount of
time, which is computationally very inefficient. Research has
½Mst  ¼ ½T t T ½M s ½T t  ð40Þ shown that difference between the results of full-mode flutter
analysis and multi-mode analysis is slight [1]. The flutter phe-
After transferring all degrees of freedom to the physical coordi-
nomenon usually occurs under the effect of the first few torsional
nate system, for each restrained degree of freedom a corresponding
and/or heave modes of the structure. For that reason, full-mode
zero value is imposed on the physical displacement vector, which
analysis might not be considered as efficient for the eigenvalue
means the corresponding rows and columns in the structural and
flutter analysis.
aerodynamic properties matrices are eliminated. The proposed
amended scheme is more applicable to the equal section splines,
5.2. Multi-mode finite strip aeroelastic analysis
while for the unequal section splines the penalty function is used.
A globally accepted algorithm for solving the eigenvalue flutter
5. Flutter eigenvalue analysis problem, especially for linear elastic structural systems, is the flut-
ter modal analysis in which the aerodynamic response of the struc-
Eq. (4) is an eigenvalue problem which can be solved using the ture is discretized into m degrees of freedom instead of high-order
techniques of either frequency domain or time domain analyses. In n degrees of freedom. Multi-mode flutter analysis is an approxi-
the present research, the flutter eigenvalue problem is solved in mate approach in which the system flutter oscillation mode is
the environment of the frequency domain method. The flutter assumed to be the combination of a few natural modes of the tar-
response frequencies of the structure are primarily dependent on get structure. Herein, m is the number of natural modes involved in
the natural frequencies of the system, and different schemes can the flutter oscillation. When multi-mode flutter analysis needs to
be used for solving the flutter equation like PK, PK-F method, Jacobi be performed, the flutter equation of motion (Eq. (4)) must be pre-
diagonalization, QL or QR transformation, subspace iteration, etc. sented in the modal space, in which the structural and aerody-
Herein, a simple method is proposed which is based on lineariz- namic matrices are generalized as:
ing the quadratic flutter Eq. (4). The philosophy of linearization of
the flutter eigenvalue problem is to allow one to use the eigenvalue ½M   ¼ ½wT ½M ½w ð48Þ
n o
solution of an undamped system. Assuming z ¼ d_ , one can refor-
½C   ¼ ½wT ½C ½w ð49Þ
mulate Eq. (4) as:
8n o9 or

> € >
( n _ o )
½M  ½0 < d = ½C  ½K  d
n o ¼ ð41Þ ½C   ¼ ½wT ð½C s   ½C aer Þ½w ¼ ½wT ½C s ½w  ½wT ½C aer ½w ð50Þ
½0 ½I >: d_ >
; ½I ½0 fdg
and
or ½K   ¼ ½wT ½K ½w ð51Þ

( z_ )

½M  ½0 n o ½C  ½K  z or
¼ ð42Þ
½0 ½I d_ ½I  ½0 fdg
½K   ¼ ½wT ð½C s   ½C aer Þ½w ¼ ½wT ½C s ½w  ½wT ½C aer ½w ð52Þ
In short form, Eq. (42) is presented by:
n o
½A X_ ¼ ½BfX g ð43Þ 5.3. Solution procedure

where In the case of full-mode flutter analysis, generalization of the


½M ½0 property matrices is not necessary because all natural frequencies
½ A ¼ ð44Þ
½0 ½I  of the system must be considered, and the natural mode shapes are
not required to be decoupled. For both full-mode and multi-mode


½C  ½K  flutter analysis, the eigenvalue analysis of Eq. (4) yields a set
½B ¼ ð45Þ of complex conjugate pairs of eigenvalues ki ¼ li  ixi
½I  ½0
(i ¼ 1; 2; . . . ; m=2Þ. The real part li is in fact the logarithmic decre-
 (n o)  ment of the structural system, while the imaginary part xi is the
z d_ kfwg kt
fX g ¼ ¼ ¼ e ð46Þ flutter frequency response of the structure at a certain wind speed.
fdg fdg fwg The real part li of at least one complex conjugate pair must be pos-
itive in order to identify the occurrence of flutter instability. In
 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 153

other words, when the real part of the eigenvalue becomes zero for including plate length L = 300 m, plate width B = 40 m, plate thick-
the first time at the lowest wind speed, the critical flutter condition ness t = 1 m, modulus of elasticity E ¼ 3:5  103 MPa, plate mass
occurs. In performing the flutter eigenvalue analysis, the critical density q ¼ 2:5  103 kg=m3 , Poisson’s ratio t ¼ 1=6, and the air
flutter speed U f and critical flutter frequency xf can be solved
density qair ¼ 1:225 kg=m3 : The plate is divided into two strips of
simultaneously from the real and imaginary values of the resultant equal width in the transverse direction, and 10 equal-length sec-
eigenvalue. The range of wind speeds U min to U max in increments of tions in the longitudinal direction. The finite-strip discretization
DU, as well as the iteration of the oscillation frequency from the of the plate is illustrated in Fig. 6. The angle of the wind speed is
initial value of the corresponding natural mode, are important fac- selected to be zero degrees.
tors in the investigation of flutter susceptibility. In the case of full- First, a free-vibration spline finite-strip analysis is performed in
mode analysis, the number of eigenvalues k is equal to 2n, while for order to obtain the natural frequencies and mode shapes of the
multi-mode analysis this number is equal to 2 m. The eigenvalue plate. The developed finite strip program is also capable of 3D-
equation of Eq. (4) (or Eq. (43)) is solved iteratively with the incre- plotting the deformed shape of the plate in different mode shapes.
mental value of the wind speed U. The aerodynamic derivatives are In Table 1 and Fig. 7, the first ten natural frequencies as well as the
obtained from the flutter angular frequency xf , and the critical mode shapes of the simply supported flat shell are presented. From
wind speed and flutter frequency are traced when the aeroelastic the deformed shape of the plate, it can be concluded that the sec-
logarithmic decrement of the system becomes zero. The intersec- ond and fifth modes correspond to the first symmetric heave and
tion of the flutter frequency curves of two or multiple modes deter- torsional modes with natural frequencies of 0.0189 Hz and
mine the critical condition. 0.2050 Hz respectively. As can be seen in Table 1, the targeted nat-
ural frequencies are in excellent agreement with the frequencies of
6. Numerical examples the same plate, as presented in Ref. [18], with a difference of less
than 5%.
In order to illustrate the effectiveness of the proposed finite Theodorsen’s function [20] is used to calculate the aerodynamic
strip solution in the flutter analysis of long-span plates and bridge flutter derivatives of a plate with a uniform thin cross-section as:
structures, in this section two examples of frequency-domain flut- C ðK w Þ ¼ F ðK w Þ þ iGðK w Þ ð53Þ
ter analysis are presented. First, classical thin airfoil aerodynamics
in the category of both multi-mode and full-mode analysis where
approaches is employed for the numerical flutter investigation of 0:165 0:335
a thin flat shell. In the second case study, the aerodynamic perfor- F ðK w Þ ¼ 1   2   2 ð54Þ
mance of the Kap Shui Mun Bridge is evaluated. 1 þ kw0:0455
1 þ 0:3
kw

   
6.1. Flutter of long-span flat shell 0:165 0:0455
kw
0:335 0:3
kw
GðK Þ ¼   2   2 ð55Þ
A long-span plate with simply supported ends is selected as a 1 þ kw
0:0455
1 þ kw
0:3

case study to investigate the accuracy of the proposed finite strip

in which kw ¼ 0:5K w . The values of Theodorsen’s functions F ðK w Þ
solution. For verification purposes, the material and geometric
properties are the same as those addressed in reference [18], and GðK w Þ against the kw are illustrated in Fig. 8.
Using Theodorsen’s functions F ðK w Þ and GðK w Þ, the flutter
derivatives Hi and Ai (i = 1, 2, 3, 4) for a plate section can be ana-
lytically presented as follows:
F
H1 ¼ 2p
Kw
 
p 1 G F
H2 ¼  þ þ
Kw 2 Kw 2
  ð56Þ
p KwG
H3 ¼  2 2F 
K 2
w 
p 4G
H4 ¼ 1þ
2 Kw
Fig. 6. Simply supported long-span flat shell [18].

Table 1
Natural frequencies of the long-span thin flat shell.

Mode number Frequency f (Hz) Angular frequency x (rad/s) Mode shape

FSM Ref. [18]
1 0 – 0 –
2 0.0189 0.019 0.1185 Symmetric heave 1
3 0.0755 – 0.4744 Antisymmetric heave 1
4 0.1702 – 1.0694 Symmetric heave 2
5 0.2050 0.194 1.2879 Torsional 1
6 0.3035 – 1.9069 Antisymmetric heave 2
7 0.4150 – 2.6076 Torsional 2
8 0.4768 – 2.9961 Symmetric heave 3
9 0.6351 – 3.9902 Torsional 3
10 0.6939 – 4.3600 Antisymmetric heave 3
154 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

Mode 6
Mode 1

Mode 7

Mode 2

Mode 8
Mode 3

Mode 9

Mode 4

Mode 5 Mode 10

Fig. 7. Finite strip mode shapes of the simply supported thin flat shell.

Fig. 8. Real and imaginary parts of Theodorsen’s circularity function C ðK w Þ ¼ F ðK w Þ þ iGðK w Þ.

 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 155

Fig. 9. Aerodynamic flutter derivatives of the thin flat shell.

Fig. 10. Torsional flutter response frequencies of the simply supported long-span plate.

pF In the environment of the spline finite strip method and using

A1 ¼
2K w an iterative frequency domain analysis, the response flutter fre-
 
p 1 G F quencies of the plate for different wind speeds are obtained and
A2 ¼ þ þ
Kw 8 2K w 8 expressed in Fig. 10. In addition, the variation of the logarithmic
! ð57Þ decrement against the wind speed is shown in Fig. 11. Only the tor-
2
p Kw F KwG
A3 ¼ 2 þ  sional mode has been considered in this study using both multi-
K w 64 2 8 mode and full mode flutter analysis. Also, the structural damping
pG is neglected in this example.
A4 ¼ 
2K w According to Fig. 11, the logarithmic decrement becomes zero at
a wind speed of 113 (m/sec), and shows the occurrence of flutter
The variations of the theoretically calculated flutter derivatives
instability. The corresponding response frequency for the critical
Hi and Ai of the plate section against the reduced velocity are illus-
wind speed of 113 (m/sec) is 0.1055 (Hz), which represents the
trated in Fig. 9.
critical flutter frequency. In Table 2, the critical flutter wind speed
156 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

and frequency of the present plate are compared with the results aerodynamic forces allocated to each knot in a spline nodal line
reported by other researchers. The comparison between the results is slightly less than that of the present finite strip formulation.
shows a maximum 2% and 11% difference between the critical flut- Therefore, less wind energy is input into the structure, and this
ter frequencies and critical flutter wind speeds respectively, which initiates the occurrence of flutter at a higher speed. The other
confirms the accuracy of the present finite strip formulation and reason behind the difference between the critical wind speed
the corresponding flutter analysis scheme. Also, there is almost of the present formulation and that obtained by Chang et al.
no difference between the numerical finite strip results of the [18] originates in the spline distribution of the aerodynamic
single-mode and full-mode analyses, while the computational forces in the current flutter finite strip method versus the trape-
time for these analyses was 23 s and 1620 s respectively, which zoidal distribution of the same forces employed in the work of
shows a dramatic difference. It can be concluded that there is no Cheng et al. [18]. In the spline distribution, more forces are allo-
need to include all the vibration modes in the flutter analysis. cated to the middle knots than to the knots close to the two
The reason for the higher critical wind speed in the case of ends of the plate, while in the trapezoidal distribution, except
the PK-F finite strip flutter analysis [18] could be the spatial dis- for the two ends of the nodal lines, the same amount of forces
tribution of the self-excited loads, in which the magnitude of the are allocated to the rest of the knots.

Fig. 11. Logarithmic decrement of the simply supported long-span plate.

Table 2
Flutter analysis results of the simply supported long-span thin flat shell.

Method Single-mode Full-mode FEM SFSM PK-F van der Put formula
Reference This paper This paper [21] [18] [18]
Critical flutter frequency f (Hz) 0.1055 0.1055 0.1034 0.1045 –
Frequency error% – 0 +2 +0.9 –
Critical wind speed Uf (m/sec) 113.1 113.2 125.4 123.5 121.9
Wind speed error% – 0 11 9 7

Table 3
Convergence study for critical flutter wind speed and frequency.

Number of strips Number of sections

4 8 12 16 20
f (Hz) Uf (m/sec) f (Hz) Uf (m/sec) f (Hz) Uf (m/sec) f (Hz) Uf (m/sec) f (Hz) Uf (m/sec)
1 0.1071 102.3 0.1060 110.1 0.1058 112.8 0.1055 113.1 0.1055 113.1
2 0.1069 102.5 0.1059 110.4 0.1055 113.1 0.1055 113.1 0.1055 113.1
3 0.1069 102.6 0.1059 111.4 0.1055 113.1 0.1055 113.1 0.1055 113.1
4 0.1069 102.9 0.1058 111.6 0.1056 113.1 0.1055 113.1 0.1055 113.1

Fig. 12. Kap Shui Mun Bridge.

 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 157

Fig. 13. Geometrical properties of the deck: (a) top view; (b) front view.

It is worth noting that a number of parametric studies were per- the natural frequencies, are obtained by the IFSM and are pre-
formed in the present research, as summarized in Table 3, in order sented in Table 5. The finite strip results are compared with those
to investigate the convergence of the flutter frequency and flutter obtained from the finite element (SAP 2000) analysis [15] as well
wind speed results against the spline finite strip mesh size. The as with the field measurements [22]. The very good agreement
numerical results of Table 3 show that the number of sections in between the results validates the accuracy of the IFSM in frequency
the longitudinal direction of the strip directly influences the con- analysis. In Figs. 16–19, the first symmetric and antisymmetric
vergence of the results, while the number of strips has a lower heave and torsional mode shapes of the Kap Shui Mun Bridge are
impact on the accuracy and convergence rate of the solution. The illustrated. The corresponding natural frequency of the first sym-
reason for this is that the greater the number of knots in the longi- metric torsional and heave modes is 0.7526 (Hz) and 0.4250 (Hz)
tudinal direction of the spline strip, the more uniform distribution
of the aerodynamic self-excited forces along the center of elasticity
of the deck. This results in faster and better convergence of the
results.

6.2. Aerodynamic performance of the Kap Shui Mun bridge

For the second case study, the integrated finite strip method is
employed to investigate the flutter performance of the Kap Shui
Mun Bridge, as illustrated in Fig. 12. The geometric and material
properties of the deck are expressed in Fig. 13 and Table 4 respec-
tively. The finite strip modeling of the west and east towers, as well
as the 3D integrated finite strip modeling of the entire cable-stayed
bridge system, are displayed in Figs. 14 and 15 respectively.
Detailed information regarding the integrated finite-strip modeling
of the Kap Shui Mun Bridge system can be found in Ref. [15].

6.2.1. Free vibration analysis

The flutter frequency analysis is performed by selecting the nat-
ural frequencies of one or a number of specific mode shapes as the
start point of the response frequency. Therefore, free vibration
analysis is a necessary step before flutter analysis. In this regard, Fig. 14. Towers models of the Kap Shui Mun Bridge.
a free vibration analysis is performed to obtain the natural fre-
quencies of the Kap Shui Mun Bridge and their corresponding
mode shapes. The first ten mode shapes of the bridge, including

Table 4
Material properties of the deck.

Properties Main span Side spans

Modulus of elasticity (kPa) 2.00  108 3.00  107
Mass density (kg/m3) 3880 3630
Poisson’s ratio 0.3 0.2
Moment of inertia (vertical) (m4) 191 363
Moment of inertia (vertical) (m4) 2530 5560
Fig. 15. Three-dimensional model of the Kap Shui Mun Bridge.
158 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

Table 5
Modal characteristics of first ten modes of the Kap Shui Mun Bridge.

Mode number Natural frequency f (Hz) Nature of mode shape

IFSM (1) FEM [15] (2) Field tests [22] (3)
1 0.2113 0.2061 – Tower
2 0.2409 0.2338 – Tower
3 0.4250 0.4226 0.39 Symmetric heave mode of the deck
4 0.5217 0.5160 0.49 Lateral bending of the deck
5 0.7526 0.7179 0.83 Symmetric torsional mode of the deck
6 0.8523 0.8500 0.66 Antisymmetric heave mode of the deck
7 0.9305 0.9257 0.90 Lateral bending of the deck
8 1.0032 1.0023 1.07 Heave mode of the deck
9 1.1381 1.1048 – Tower
10 1.1391 1.1058 – Tower

Fig. 16. First symmetric heave mode of the deck (0.4250 Hz).
Fig. 18. First symmetric torsional mode of the deck (0.7526 Hz).

Fig. 17. First antisymmetric heave mode of the deck (0.8523 Hz).

respectively, while this value is 1.3419 (Hz) and 0.8523 (Hz)

respectively for the first antisymmetric torsional and heave modes.
Fig. 19. First antisymmetric torsional mode of the deck (1.3419 Hz).

6.2.2. Kap Shui Mun bridge flutter derivatives

In the design of Kap Shui Mun Bridge, because of the critical ration both with and without traffic and in which the road curb is
location of the bridge, typhoon conditions have been considered also considered. The vertical incidence angle of the wind, some-
in which the wind loading is much stronger (might need to con- times called the angle of the wind attack, in the present study is
sider of wind speed to up 95 m/s) [23]. The Kap Shui Mun Bridge -2.5 degrees.
was tested for aerodynamic derivatives in order to determine the
critical wind velocities for the bridge deck [24]. The aerodynamic 6.2.3. Flutter analysis of the Kap Shui Mun bridge
flutter performance of Kap Shui Mun Bridge has been investigated After obtaining all structural and aerodynamic properties of the
through the experimental flutter derivatives of the bridge as the Kap Shui Mun Bridge by using the integrated finite strip
reported by Jon Raggett at the West Wind Laboratory at California method, the flutter analysis can be performed. The damping ratio
in collaboration with Robert Scanlan at Johns Hopkins University of the bridge system is chosen to be 0.02. In the previous example,
[25]. The static aerodynamic coefficients and the aerodynamic flut- it was shown that full-mode flutter analysis takes much more stor-
ter derivatives have been derived through wind tunnel tests. The age and time comparing with the multi-mode flutter analysis,
flutter derivatives, obtained experimentally, are reproduced in while the results of both techniques are almost the same. The
Figs. 20 and 21. Herein, the deck width B is 35.7 m. All the flutter full-mode flutter analysis for a massive structure like the Kap Shui
derivatives are related to the final main span bridge deck configu- Mun Bridge with thousands of degrees of freedom will take a mas-
 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 159

Fig. 20. Flutter derivatives A1 to A3 of the Kap Shui Mun Bridge against reduced velocity.

sive amount of storage and time. Therefore, multi-mode flutter 6.3. Efficiency analysis
analysis is applied in this example for aeroelastic analysis of the
Kap Shui Mun Bridge. The first symmetric torsional mode of the Using a personal computer with an Intel Core2 Duo CPU
deck is taken as the critical mode shape for the flutter. The inte- (1.66 GHz) and 3 GB physical memory, the computer times
grated finite-strip flutter eigenvalue analysis results are plotted required by IFSM and FEM for two test cases with similar grid con-
in Figs. 22 and 23, where the flutter frequencies as well as the log- ditions were compared. For the first case study, the computational
arithmic decrements of the Kap Shui Mun Bridge for different wind time in IFSM and FEM was 4000 s and 4400 s, respectively, and in
speeds are calculated and displayed. The results show that both the the second case study it was 12,300 s and 13,600 s, respectively,
flutter frequency and logarithmic decrement decrease when the which shows superiority of IFSM in terms of computational cost.
wind speed increases, which could be a sign of flutter instability. In other words, around 10% percent decrease in computation time
However, the logarithmic decrement never becomes zero, which can be witnessed in finite strip method compared to finite element
means that the flutter instability does not happen within the stan- analysis. Due to the semi-analytical properties in the longitudinal
dard values for wind speeds and even in typhoon conditions, for direction, the number of sections for each strip in the IFSM model
which the critical wind speed could be 95 m/sec. Therefore, this could be significantly reduced without losing accuracy. Conse-
proves that the aerodynamic flutter design of the Kap Shui Mun quently, the time for computation could also be reduced, and
Bridge meets the requirements for aeroelastic stability. greater efficiency is achieved. It must be noted that this case study
160 H. Naderian et al. / Computers and Structures 212 (2019) 145–161

Fig. 21. Flutter derivatives H1 to H3 of the Kap Shui Mun Bridge against reduced velocity.

Fig. 22. Torsional flutter response frequencies of the Kap Shui Mun Bridge.

was performed on a simplified model. In the case a complicated in comparison with FE analysis [15]. For instance, when a continu-
long-span cable-stayed bridge model, the efficiency of the pro- ous real time analysis is required in which a lot of iterative analy-
posed integrated finite strip solution will be more appreciated ses are performed, the advantages of the proposed methodology in
due to the need for fewer iterations and less degrees of freedom time saving will be more illustrated.
 H. Naderian et al. / Computers and Structures 212 (2019) 145–161 161

Fig. 23. Logarithmic decrement of torsional flutter of the Kap Shui Mun Bridge.

7. Conclusions [3] Huang Ming-Hui, Lin Yuh-Yi, Weng Ming-Xi. Flutter and buffeting analysis of
bridges subjected to skew wind. J Appl Sci Eng 2012;15(4):401E413. 401.
[4] Chad Gibbs S, Wang Ivan, Dowell Earl. Theory and experiment for flutter of a
The integrated finite strip method was extended to the area of rectangular plate with a fixed leading edge in three-dimensional axial flow. J
wind engineering, where aerodynamic flutter analysis can be per- Fluids Struct 2012;34(2012):68–83.
[5] Wu Teng, Kareem Ahsan, Ge Yaojun. Linear and nonlinear aeroelastic analysis
formed in a minimal computational time, providing high accuracy
frameworks for cable-supported bridges. Nonlinear Dyn 2013;74(3):487–516.
and efficiency. Due to the semi-analytical nature and narrow band- [6] Kreis ES, Andr´e JC. A numerical inquiry into the flutter phenomenon in long-
width of the dynamic properties of the system, the results con- span bridges. Latin Am J Solids Struct 2005;2(2005):321–37.
[7] Matsumoto Masaru, Matsumiya Hisato, Fujiwara Shinya, Ito Yasuaki. New
verged rapidly, with a few degrees of freedom. The aeroelastic
Consideration on flutter properties basing on SBS -Fundamental flutter mode,
damping and stiffness matrices were derived by using the inte- similar Selberg’s formula, torsional divergence instability, and new coupled
grated finite strip method, and the total potential energy of a shell flutter phenomena affected by structural coupling. BBAA VI international
strip was obtained from the algebraic summation of the in-plane colloquium on: bluff bodies aerodynamics & applications Milano, Italy, July,
20-24 2008, 2008.
and out-of-plane deformations. In the current finite strip solution, [8] Phan Duc-Huynh, Nguyen Ngoc-Trung. Flutter and buffeting control of long-
the boundary conditions were modelled by using a simple tech- span suspension bridge by passive flaps: experiment and numerical
nique based on a physical coordinate system. Also, the flutter prob- simulation. Int’l J Aeronaut Space Sci 2013;14(1):46–57. https://doi.org/
10.5139/IJASS.2013.14.1.46.
lem was solved through a very straightforward eigenvalue [9] Canor Thomas, Caracoglia Luca, Denoël Vincent. Application of random
frequency analysis. Both multi-mode and full-mode flutter analy- eigenvalue analysis to assess bridge flutter probability. J Wind Eng Ind
ses approaches have been outlined in the IFSM environment in Aerodyn 2015;140:79–86.
[10] Xie Gui-hua, Yin Jie, Liu Rong-gui, Chen Bei, Cai Dong-sheng. Experimental and
order to evaluate the critical flutter speed as well as the critical numerical investigation on the static and dynamic behaviors of cable-stayed
flutter frequency of the structural system. The numerical results bridges with CFRP cables. Compos Part B: Eng 2017;111:235–42.
show that performing the full-mode flutter analysis is much more [11] Yang Yongxin, Zhou Rui, Ge Yaojun, Zou Xiaojie, Zhang Lihai. Flutter
characteristics of twin-box girder bridges with vertical central stabilizers.
time-consuming than the multi-mode analysis, while the resultant
Eng Struct 2017;133:33–48.
critical flutter frequencies and corresponding flutter wind speeds [12] Lee N, Lee H, Baek C, Lee S. Aeroelastic analysis of bridge deck flutter with
do not change significantly. Therefore, it is suggested to use the modified implicit coupling method. J Wind Eng Ind Aerodyn 2016;155:11–22.
[13] Cheung MS, Li W, Chidiac SE. Finite strip analysis of bridges. 1st
multi-mode technique for flutter analysis. Also, the results show
ed. London: Spon; 1996.
that the uniform distribution of self-excited flutter forces on the [14] Cheung YK, Tham LG. The finite strip method. USA: CRC Press; 1998.
centre of elasticity of a cross-section is reasonably good when com- [15] Naderian Hamidreza, Cheung Moe MS, Shen Zhenyuan, Dragomirescu Elena.
pared with the spatial distribution of aerodynamic forces on the Integrated Finite Strip Analysis of long-span cable-stayed bridges. Comput
Struct 2015;158:82–97.
entire surface of the deck. Last but not least, the method is capable [16] Naderian Hamidreza, Cheung Moe MS, Shen Zhenyuan, Dragomirescu Elena.
of performing a three-dimensional flutter analysis. The successful Seismic analysis of long-span cable-stayed bridges by an integrated finite strip
extension of the finite strip method to aerodynamic flutter analysis method. ASCE J Bridge Eng 2015. ISSN 1084-0702.
[17] Cheung YK, Fan SC, Wu CQ. Spline finite strip in Structure Analysis. In:
can open a new door for further development of this methodology Proceedings, the international conference on finite element method, Shanghai.
in different areas of wind engineering, such as buffeting analysis of p. 704–9.
a bridge, wind-vehicle-structure (WVS) interactions, and the aero- [18] Lau DT, Cheung MS, Cheng SH. 3D flutter analysis of bridges by spline finite
strip method. J Struct Eng 2000:1246–54. https://doi.org/10.1061/(ASCE)
dynamics of composite hybrid FRP bridge structures. 0733-9445(2000)126:10(1246).
[19] Scanlan RH, Lin WH. Effects of turbulence on bridge flutter derivatives. J Engrg
Acknowledgements Mech, ASCE 1978;104(4):719–33.
[20] Simiu E, Scanlan RH. Wind effects on structures. 3rd ed. New York: Wiley;
1996.
The authors wish to gratefully acknowledge the financial sup- [21] Cheng SH. ‘‘The 3D finite element flutter analysis of long-span bridge”, Master
port of the Canada NSERC Industrial Postgraduate Scholarship. thesis. Shanghai, China: Tongji University; 1994.
[22] Lau CK, Mak WP, Wong KY, Man KL, Chan WY, Wong KF. Structural
performance measurement and design parameter validation for Kap Shui
References Mun Cable-Stayed Bridge. Adv Steel Struct 1999;2(1).
[23] Yung Benedict Kai Kwan. A critical analysis of the Kap Shui Mun Bridge.
[1] Ge YJ, Tanaka H. Aerodynamic flutter analysis of cable-supported bridges by Proceedings of bridge engineering conference, 23 April 2008, England, 2008.
multi-mode and full-mode approaches. J Wind Eng Ind Aerodyn [24] Scanlan RH, Stroh SL, Raggett JD. Methods of wind response investigation
2000;86:123–53. employed for the Kap Shui Mun Bridge. J Wind Eng Ind Aerodyn 1995:1–11.
[2] Ding QS, Chen AR, Xiang HF. Coupled flutter analysis of long-span bridges by [25] Raggett Jon, Scanlan Robert. Flutter derivatives of Kap Shui Mun Bridges using
multimode and full order approaches. J Wind Eng Ind Aerodyn wind tunnel tests. West Wind Laboratory, California & Department of Civil
2002;90:1981–93. Engineering, Johns Hopkins University; 1997.