Engineering Structures 120 (2016) 82–91

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Aerodynamic stability of long span suspension bridges with low

torsional natural frequencies
Michael Styrk Andersen a,⇑, Jens Johansson a, Anders Brandt a, Svend Ole Hansen a,b
a
Dept. of Technology and Innovation, University of Southern Denmark, Campusvej 55, Odense, Denmark
b
Svend Ole Hansen ApS, Skt. Jørgens Allé 5c, 1615 København V, Denmark

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

Article history: Classical flutter of suspended bridge decks can be avoided if the torsional frequencies are lower than the
Received 5 July 2015 vertical. Wind tunnel tests of single boxes and twin box section models with torsional natural frequencies
Revised 7 April 2016 above and below the vertical frequency has been conducted. Flutter was avoided in all tests where the
Accepted 11 April 2016
torsional frequency was lower than the vertical. But too low torsional stiffness caused large static
Available online 3 May 2016
displacements of the girder at medium–high wind speeds and steady state oscillations driven by a
combination of torsional divergence and stalling behavior at the critical wind seed. In order to design
Keywords:
aerodynamically stable suspension bridges with low torsional natural frequencies it is suggested to
Long-span bridge
Flutter
increase the mass moment of inertia and provide adequate torsional stiffness by the main cables spacing.
Torsional divergence Ó 2016 Elsevier Ltd. All rights reserved.
Aerodynamics
Wind–structure interaction
Aeroelasticity
Wind tunnel experiments
Non-flutter design principle

1. Introduction In traditional bridge design where the torsional-to-vertical

The aerodynamic stability of the bridge deck has been a major
issue in the design of long span bridges since the disastrous
collapse of the first Tacoma Narrows in 1940. During the last two
decades, 3 world records in main span lengths have been broken
each using different approaches to obtain aerodynamic stability.
The Akashi Kaikyo Bridge in Japan has a main span of 1991 m
where a slotted deck is supported on a truss girder. The slots
improve the aerodynamic stability.
The Xihoumen Bridge in China spans 1650 m and has a twin box
girder with a central gap of 6 m. The central gap was introduced to
increase the critical flutter wind speed. The Great Belt Bridge in
Denmark has a main span of 1624 m and is constructed with a
single box shaped girder with fairings at the windward and lee-
ward edges.
The possibility of longer spans is related to the implementation
of stronger and lighter materials. At the moment, new world record
breaking spans are being considered for e.g. fjord crossings along
the Coastal Highway Route 39 in Norway from Kristiansand to
Trondheim.
⇑ Corresponding author.
E-mail addresses: mian@iti.sdu.dk (M.S. Andersen),
(J. Johansson), abra@iti.sdu.dk (A. Brandt).
frequency ratio, cx = xa/xh, is above 1 both mass, stiffness and
frequency ratio must be increased in order to raise the critical flut-
ter wind speed. A pure increase in mass seems less favorable than
an increase in stiffness. However, increased stiffness tends to
require an increase in mass. Furthermore, for long span bridges,
the torsional rigidity of the girder declines and the mass moment
of inertia in the main cables increases with increased span length.
This naturally leads to a reduction of the frequency ratio cx.
Thin airfoil theory [1] was previously used to estimate an upper
limit for the flutter wind speed for single box shaped girder
suspension bridges. But after the introduction of the twin box
girder, critical flutter wind speeds may exceed the thin airfoil
flutter speed by as much as 200% [2]. This may however not be
sufficient for all super long span bridges.
There is no theoretical solution to the critical flutter wind speed
for airfoils having cx < 1. Therefore, a lighter and more economical
suspension bridge design may thus be developed based on the
principle that the frequency of the fundamental torsional mode
of the bridge deck deliberately should be lower than the frequency
of the corresponding vertical mode. Similarly this can be obtained
by either an increase in the mass moment of inertia in the torsional
motion or a decrease in the torsional stiffness. A decrease in
torsional stiffness does however lead to a reduction of the critical
jensj@iti.sdu.dk wind speed for torsional divergence.

http://dx.doi.org/10.1016/j.engstruct.2016.04.025
0141-0296/Ó 2016 Elsevier Ltd. All rights reserved.
 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91
The slope of the time mean aerodynamic moment dCm/da acting
on airfoils at different angles of attack is considerable larger than
corresponding values of streamlined bridge deck girders and
especially twin box girders. Therefore, the critical wind speed for
torsional divergence is larger for bridge decks than for thin airfoils.
Adequate torsional stiffness may be provided by the cable plane
spacing alone and might thus not need to be obtained through
the deck girder as in traditional design.
The self-excited forces on the bridge deck occurring from the
motion of the bridge deck itself is expressed in terms of Aerody-
namic Derivatives (AD’s) introduced for bridge decks by Scanlan
and Tomko in [3].
The idea of designing a suspension bridge with cx < 1 was
originally proposed by [4] with the introduction of the twin bridge.
Walshe and Wyatt [5] mentioned the idea and stated that the
needed gap width between the twin boxes also depends on the
static rotations under eccentric traffic load.
Bartoli et al. [6,7] tested section models of a twin box with a
large gap width having cx < 1 and reported the sections to be aero-
dynamically stable. Torsional divergence did occur, but only at
wind tunnel speeds corresponding to full scale values above the
required critical wind speed. Wind tunnel results of aerodynami-
cally stable flat plate section models with cx < 1 were reported
by [8]. However, torsional flutter of bluff bridge sections occurred.
Larsen and Larose [9] wrote that classical flutter may be prevented
if cx < 1, but that issues with e.g. eccentricities due to traffic loads,
buffeting response, vortex shedding and torsional divergence need
to be resolved before this design principle has matured.
The present paper aims to explain why the critical flutter wind
speed tends to decline as the span width increases in traditional
bridge design. Wind tunnel tests of single boxes and twin boxes
with different width-to-depth ratios, B/D, and gap-to-width ratios,
Z/B, has been tested at torsional-to-vertical frequency ratios, cx,
above and below unity. The observed critical wind speeds are
presented in Section 4 and compared with the corresponding thin
airfoil critical wind speeds.
2. Background
Cable supported bridges are flexible structures that respond to
the wind in a dynamic manner. Therefore structural dynamics
plays a vital role within bridge aerodynamics. It is important to
Table 1
Torsional girder constant Kt and mass moment of inertia Id.
Case 1 2 3 4
Kt m4 5 0.5 0 0.15
Id 103 kg m2/m 250 250 250
Fig. 1. Two degrees of freedom section model of a thin plate with a central gap.
83
correctly estimate the vertical and torsional still air natural
frequencies during the design phase because their ratio are crucial
for the determination of the flutter wind speed. An ‘‘exact solution”
to the fundamental symmetric frequencies is given in [10,11].
2.1. Declining torsional-to-vertical frequency ratio with increasing
span length
The change in torsional-to-vertical frequency ratio with span-
length will here be illustrated with four simplified cases. Table 1
presents the four cases with different torsional girder constants
Kt and mass moment of inertia Id, each of which is considered at
different main span lengths. No suspended side spans are included.
The cable sag is equal to 10% of the span. The distance from the
shear center of the bridge deck to the main cable planes are
a = 7.5 m and the Young modulus of the cables is 205 GPa. The
bridge deck width is a constant B = 18.3 m and has a shear modulus
of 66.7 GPa. Its mass is md = 12.5
103 kg/m.
The area of the main cables obviously depends on the span
length. In the present the necessary cable area is estimated follow-
ing [12]. Assuming a distributed liveload on the bridge deck
p = 46.58 kN/m and a concentrated liveload representing special
vehicles P = 675 kN the cable area Ac is calculated by the equation
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
½ðg d þ pÞlm þ 2P lm þ 16km
Ac ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1Þ
8f cbd km  ccb lm lm þ 16km
2 2
The design stress and the equivalent density of the main cable
are fcbd = 800 MPa and ccb = 84 kN/m3, respectively. The span
length and the sag is denoted lm and km respectively. The deadload
of the bridge deck per unit length is gd = md ⁄ 9.81 m/s2. The cable
mass per unit length depends on the cable area and is given by
l = 8560 kg/m3 Ac. The mass and mass moment of inertia per unit
length of the deck and the cables are m = (md + 2l) and I = (Id +
2la2) respectively. The weight of hangers has been neglected.
Fig. 2 shows the relationship between the span-to-width ratios
and cx, as well as the critical flutter wind speed in m/s. The calcu-
lations of the critical flutter wind speed have been simplified by
assuming that the flutter derivatives corresponds to thin airfoil
derivatives and that the vertical and torsional modes are shape
wise similar. The calculation routine is given in Section 2.2. Case
1 has a high torsional stiffness and thus a high frequency ratio at
low span-to-width ratios. Increase in span-to-width ratio causes
a relatively rapid decrease in frequency ratio. This rate of change
decreases when the torsional stiffness of the deck is reduced, as
illustrated with cases 2 and 3. Hence, the effect of the torsional
stiffness in the girder on cx is seen to decline as the span length
1250
increases. No critical flutter wind speeds were found for
84 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91

Fig. 2. Torsional-to-vertical frequency ratio for four cases of torsional moment of inertia and mass moments of inertia of the bridge deck at different span-to-width ratios (a)
and the corresponding thin airfoil flutter wind speed (b) assuming a structural damping ratio of f = 0.3% and 50% mass participation.

R lm
span-to-width ratios above 109.3 where cx = 1.20 in Case 3. Case 4 hðxÞaðxÞdx
has a low torsional stiffness in the girder but a large mass moment
C h ¼ R0 lm 2
hðxÞ dx
of inertia. The torsional-to-vertical frequency ratio declines from R lm
0
ð4Þ
hðxÞaðxÞdx
cx = 1.05 at the lowest span to cx = 0.95 at the longest. No critical C a ¼ R0 lm 2
flutter speeds were found at any spans in this case. aðxÞ dx
0

2.2. Flutter
The following complex eigenvalue (CEV) analysis of a bridge
deck in wind assess the critical flutter wind speed by evaluating
the real part of the complex eigenvalues of the equation of motion
of a bridge deck in wind. The modal mass, linear viscous structural
damping and structural stiffness per unit length in still air are
given by the matrices M, Cstr and Kstr respectively. The modal
coordinates are defined by r = [h, a]T where h and a are the
generalized vertical and torsional degrees of freedom respectively.
The self-excited wind loads Fm ¼ Kae r þ Cae r_ represents the
aeroelastic forces acting on the bridge deck per unit length. It is
divided into aerodynamic stiffness Kae and aerodynamic damping
Cae defined by 8AD’s, Ai ; Hi where i = [1, 2, 3, 4]. The AD’s are
expressed as functions of the reduced non-dimensional frequency
K ¼ bUx and are given in e.g. [13]. The full scale speed is denoted U,
the bridge deck width B and x is the angular frequency
of oscillation of the bridge deck at the wind speed U. The density
of air is denoted q. With these definitions, the equation of
equilibrium becomes
where h(x) and a(x) is the mode shapes along the bridge deck axis.
Subtracting the aerodynamic damping and stiffness matrices
from the structural gives the effective damping and stiffness, i.e.
Ceff = Cstr  Cae and Keff = Kstr  Kae. The effective equation of
motion in wind is then given by
M€r þ Ceff r_ þ Keff r ¼ 0 ð5Þ
Aerodynamic damping of bridge decks can be modeled as non-
proportional viscous damping which means that complex mode
shapes of the bridge deck is expected in wind. Flutter is, by defini-
tion, a dynamic instability which is recognized when one of the
complex eigenvalues of the system has a positive real part, i.e. neg-
ative damping. The critical reduced flutter wind speed is found at
the point where the eigenvalues have zero real part, i.e. zero effec-
tive damping. At larger wind speeds negative effective damping
causes oscillations to increase until either the wind is reduced or
the bridge deck collapses.
2.2.1. State-space equation of motion
To evaluate the eigenvalues, the effective equation of motion in
wind is transferred to the state-space equation of motion

M€r þ Cstr r_ þ Kstr r ¼ Kae r þ Cae r_ ð2Þ z_ ¼ Az ð6Þ

The aerodynamic matrices with the self-excited forces are given _ and the state-space system matrix
where the state vector z ¼ ½r; r
T

by A is
2 3  
H4 0 I
24
C h H3 A¼ ð7Þ
Kae ¼ 12 qU BK 2 B 5 1 1
M Keff M Ceff
C a A4 A3 B
" # ð3Þ The eigenvalues are found by searching for the non-trivial solu-
H1 C h H2 B tions kr to the characteristic equation given by
Cae ¼ 12 qUBK
C a A1 B A2 B2 detðA  kIÞ ¼ 0 ð8Þ

where B is the bridge deck width, as illustrated in Fig. 1, and q is the

density of air. 2.2.2. Complex eigenvalue analysis
The similarity between the vertical mode and torsional mode is The complex eigenvalues of the system matrix A are defined by
defined by the non-dimensional coupling coefficients kr ¼ fr xr  ixdr where fr is the damping ratio and xr is the
 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91
circular frequency while xdr is the damped circular frequency.
Flutter is identified when fr becomes negative, i.e. when
Reðkr Þ P 0. At the critical wind speed the vertical and torsional
frequency are equal, i.e. x = xa = xh and therefore K = Ka = Kh.
The critical flutter wind speed is assessed using an iterative
approach in the MATLAB environment [14].
2.3. Torsional divergence
The aerodynamic moment and lift forces acting on a twin box
section are illustrated in Fig. 1. The torsional aerodynamic moment
FM rotates the bridge deck in order to establish equilibrium
between the aerodynamic moment and the torsional stiffness of
the deck-cable structure. The rotation a changes the angle of attack
of the wind. At increased angles of attack the aerodynamic
moment increases which forces the bridge deck to rotate even
more.
A simplified engineering approximation is to assume that FM is
linear with respect to the bridge deck rotation a and fully corre-
lated along the span. The critical torsional divergence wind speed
Udiv is then calculated according to e.g. [15]:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ka
U div ¼
qB2 dCdaM
where ka is the structural torsional stiffness.At this wind speed, the
bridge deck will twist in a divergent static rotation. With this
approach, a maximum allowable value of dCM/da is given by
2ka
dC M =da <
qB2 U 2div
Examples of dcm/da values from the literature are given in
Table 2.
Collapse due to divergence is only possible if the bridge deck is
not able to twist more than a critical angle of attack, acr, where
flow separation will cause the bridge deck to lose its lift force. This
phenomenon is quite similar to a stall of an airfoil. Instead of
collapse, flexible structures like cable-supported bridges may
experience large steady oscillations with a non-zero static mean
value and peak values occurring at acr.
2.4. Calculation of critical wind speeds
The critical wind speeds for a bridge having different values of
cx are plotted in Fig. 3. The bridge has an equivalent modal mass
and modal mass moment of inertia per unit length of me = 8750
kg/m and Ie = 481
103 kg m2/m respectively. The deck width is
B = 18.3 m, the vertical natural frequency is fh = 0.1 Hz and the
torsional stiffness is proportional to cx.
It has been assumed that the self-excited forces are given by the
thin airfoil AD’s and that the structural damping ratios are fh,str =
fa,str = 0.3%. No critical wind speeds for flutter were found for
cx 6 1:032.
Fig. 3 illustrates the critical wind speed for coupled flutter at
different frequency ratios using the mechanical properties of the
Bridge and the AD’s of an airfoil using the above described CEV
analysis. Critical flutter wind speeds according to Selberg’s
Formula [18] and Theodorsen’s method [1] with the AD’s given
Table 2
Aerodynamic moment coefficient slope references.
Section type
Thin airfoil [16]
The Great Belt Bridge deck conf. H9.1, Scale 1:80 [11]
The Stonecutters bridge [17]
A Simplified Twin Box Girder, Z:B  0.62 [6]
85
Fig. 3. Critical wind speeds for thin airfoil flutter and torsional divergence at
different torsional-to-vertical frequency ratios cx.
in [13] are shown as a reference. Furthermore, the critical torsional
divergence speed of a Joukowski wing having dCM/da = p/2 [16] is
ð9Þ
plotted using Eq. (10). These curves of critical wind speeds are
often found in the literature on bridge aerodynamics, e.g. in [19].
But as seen in Fig. 3 larger critical wind speeds for torsional
divergence is obtainable if dCm/da is reduced.
2.5. Coupled flutter between modes of different order
ð10Þ
Flutter involving coupling of higher order torsional modes with
vertical modes of lower order should be analyzed if cx < 1. How-
ever, due to the orthogonality of normal modes, the coupling
coefficients given in Eq. (4) will be zero if the stiffness of the bridge
deck girder is zero. This means that coupled flutter cannot occur
unless the vertical and torsional mode are of the same order.
Multimodal flutter effects has been carefully studied in [20] where
it was concluded that the shapewise similarity between vertical
and torsional modes are the most governing criteria for multi-
modal effects.
2.6. Torsional flutter
Torsional flutter is a known issue for bluff bodies and was
reported already by [3]. According to [21,22] it has been observed
that sections with B=D 6 10 are sensitive to torsional flutter in a
1-dof torsional system due to positive A2 -values. If A2 > 0 Torsional
Flutter can occur even though cx < 1. But if the values of A2 are
negative, zero, or small compared to the structural damping and
the mass moment of inertia, torsional flutter will not occur.
3. Model tests
The experimental validation of the aerodynamic stability of
sections having frequency ratios below unity are presented in this
section. Four cross sectional solutions have been investigated.
Fig. 4 shows their cross sectional shapes and their geometric
dimensions are given in Table 3. Some of the series have a rather
small length-to-width ratio as a consequence of the physical
dimensions of the tunnel. The effect of this remains untested.
dcm/da
p/2 3.1. Wind tunnel and experimental setup
1.17
0.54
The experiments have been performed in the wind tunnel at
0.20
Svend Ole Hansen ApS in Copenhagen. The wind tunnel is a
86 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91
Fig. 4. Cross sectional shapes of section models.
Table 3
Geometric dimensions of section models.
Series B/D Z/B B (m)
1 10 0 0.24
2 24 0 0.576
3 27.8 0.14 0.667
4 34.8 0.31 0.835
boundary layer tunnel of the open return flow type having a
1.7 m
1.5 m cross section and a maximum wind speed approxi-
mately equal to 12 m/s. The longitudinal turbulence intensity Iu
measured was between 1% and 2% in the tests conducted.
Fig. 5 shows the experimental model-rig system. Outside the
wind tunnel, two horizontal bars connected to the model via cen-
tral rods, is suspended from springs at configurable positions. The
transient vibration tests were executed with combined vertical and
torsional initial displacements. The initial displacements of the
section model were carefully adjusted prior to the execution in
order to avoid initiating rolling motion. An electromagnetic release
mechanism was used to release the model simultaneously at the
same initial conditions on both sides of the tunnel through all tests.
Only the torsional and vertical degrees of freedom were of interest
in the present study. Therefore drag wires were installed to avoid
lateral motion.
The section model was suspended by parallel connected
springs. The spring eccentricity can be adjusted to the desired
torsional rigidity. Dummy masses at the horizontal bar allows
configuration of the mass moment of inertia through their
Fig. 5. Dynamic test rig in the wind tunnel. (1) Windward drag wire. (2) Horizontal
extension rod. (3) Load cell. (4) Leeward drag wire with horizontal spring. (5)
Section model. (6) Vertical spring. (7) Horizontal arm used to connect the springs
with the section model.
eccentricity to the center of gravity. Hence, the spring and dummy
mass configuration determine the torsional natural frequency, and
the torsional-to-vertical frequency ratio.
L (m)
1.7
3.2. Test procedure and signal processing
1.7
1.7
1.7 The wind speeds were generally varied with approximately
1 m/s until flutter occurred or the maximum tunnel wind speed
was reached. The spring constants ki were determined by single
degree of freedom tests. The results are given in Table 4. The modal
vertical stiffness and the corresponding equivalent modal torsional
P
stiffness was estimated to be kh = ki and ka ¼ kh e2k respectively.
The vertical load was measured through HBM Load Cells at Chan-
nel 0 (Ch0) to Channel 3 (Ch3) at a sampling rate of fs = 500 Hz.
The forces were only measured at the top of the rig. Therefore, only
the upper springs i = {5, 6, 7, 8}, is used for kh and ka, when the
vertical and torsional response is estimated using Eqs. (11) and
(12) respectively. When the modal mass is estimated all springs
are taken into account. Test Series 1A to 1H and 4B to 4E used only
the upper springs.
1
h¼ ðCh0 þ Ch1 þ Ch2 þ Ch3 Þ ð11Þ
kh
1 1 Ch1  Ch0 þ Ch3  Ch2
a¼ sin ð12Þ
ka 2ek
Coupled free vibration tests were carried out in still air. An
implementation of Prony’s method [23] in the Abravibe Toolbox
for Noise & Vibration Analysis [24], including stabilization
diagram, was used to determine the natural frequencies and damp-
ing ratios. The modal mass and mass moment of inertia was calcu-
lated based on the still air natural frequencies determined.
4. Results and discussion
Table 5 summarizes the critical wind speeds observed for Series
1–4 at different torsional-to-vertical frequency ratios. The accuracy
of the obtained critical wind speeds depend, among other things,
on the wind speed increments in the tunnel. Hence, in the follow-
ing the highest stable wind speed observed is denoted U1 while the
lowest instable speed observed is denoted U2. The critical wind
speed Ucr was in between U1 and U2. All wind speeds are
non-dimensional and defined by Ur = U/(fhB) where fh is the still
air vertical frequency. The nature of instability at U2 is explained
by abbreviations where CF is coupled flutter, LSD is large static
displacements defined as mean vertical displacements lh/D > 1
and NR is used if no instability was observed at the highest obtain-
able wind speed. The thin airfoil reference critical wind speeds for
coupled flutter, UCF, and torsional divergence Udiv is calculated with
the CEV method described and Eq. (9) respectively. The constant /
expresses the ratio of U1/min(UCF, Udiv). Still air damping ratios
 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91 87

Table 4
Spring constants.

i 1 2 3 4 5 6 7 8
k (N/m) 246.78 246.22 246.91 246.62 257.66 259.74 256.61 258.16

Table 5
Section model tests results.

Test series cx me Ie fh U1 U2 UCF Udiv Nature of instability /

xa kg/m kg m2/m Hz Ur Ur Ur Ur –
xh

1A 0.71 11.23 0.50 1.17 39.64 – 1 56.89 NR >0.70

1B 0.72 11.06 0.48 1.18 38.02 – 1 56.52 NR >0.67
1C 0.86 11.06 0.34 1.18 38.52 – 1 56.82 NR >0.68
1D 0.88 11.23 0.33 1.17 38.91 – 1 57.28 NR >0.68
1E 1.02 11.06 0.24 1.18 4.13 8.64 6.79 56.62 CF 0.61
1F 1.09 11.06 0.21 1.18 15.24 16.11 17.27 56.60 CF 0.88
1G 1.19 11.23 0.18 1.17 18.14 21.79 22.87 57.21 CF 0.79
1H 2.10 8.96 0.05 1.31 26.01 29.30 42.13 53.21 CF 0.62
2A 1.20 13.19 0.22 1.52 8.09 8.10 7.51 11.07 CF 1.08
2B 1.33 13.19 0.23 1.51 9.08 9.77 9.1 12.55 CF 1.00
3A 0.92 13.07 0.25 1.52 10.67 – 1 6.75 NR >1.58
3B 0.99 15.39 0.63 1.40 11.68 – 1 11.53 NR >1.01
3C 1.22 13.02 0.35 1.52 8.84 8.90 7.06 10.59 CF 1.25
4A 0.71 12.85 0.40 1.53 8.59 – 1 4.20 NR >2.05
4B 0.72 12.48 0.38 1.11 8.58 9.24 1 4.15 LSD 2.07
4C 0.82 12.94 0.46 1.09 9.52 10.83 1 5.21 LSD 1.83
4D 0.93 12.94 0.54 1.09 10.85 11.52 1 6.40 LSD 1.70
4E 1.43 12.94 0.73 1.09 12.19 – 8.75 11.44 NR >1.39

Fig. 6. The aerodynamic damping factors fa,ae and fh,ae as a function of reduced wind speed where fh and fa is the vertical and torsional frequency in wind.

affects UCF. These were estimated by a covariance driven subspace
identification method [25] using the unbiased correlation func-
tions of slightly cropped transient signals down sampled from
fs = 500 Hz to fs = 20 Hz. The realized systems were in the order
of n = {4, 6, 8, . . ., 50} while the block Toeplitz matrices consisted
of 35 time lags. The poles having natural frequencies within
±0.5% of the vertical and torsional frequencies and positive damp-
ing ratios below 3% of critical were considered physically stable
and analyzed. From the set of stable poles, the mean value of the
damping ratios, rejecting the 10 highest and lowest outliers, was
used as the still air damping ratios relative to critical in the calcu-
lation of UCF of the corresponding Test Series.
4.1. Series 1: Single rectangular box B/D = 10
The torsional stiffness was identical in Series 1A to 1G. Dummy
masses were used to tune cx. These were removed in Series 1H and
the spring eccentricity ek was doubled.
Dominant torsional motion was seen at the critical wind speeds
in Series 1F to 1H. At wind speeds Ur > U2, coupled flutter would
eventually occur. In Series 1E coupled flutter occurred at U2.
The aerodynamic torsional damping ratio is denoted fa,ae =
fa,eff  fa,str where fa,str and fa,eff is the still air and effective tor-
sional damping ratio respectively estimated by Prony’s method
selecting the physically most stable modes. In Fig. 6 it is seen that
88 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91

Fig. 7. Test Series 1H and 1A.

Series 3B U/(f h B)=11.68 Series 3C U/(f h B)=8.9

0.5 1

0.5
h/D

h/D

0
0

-0.5 -0.5
0 15 30 0 15 30

2 4
torsion (deg)

torsion (deg)

1 2

0 0

-1 -2
0 15 30 0 15 30
Time (s) Time (s)

Fig. 8. Test Series 3B and 3C.

fa,ae is not only related to the cross sectional shape, but also
depends on the mechanical properties of the section, e.g. the mass
moment of inertia and cx. A single outlier in Series 1H and Series
1G was removed before polynomials of order 3 were fitted to the
damping ratios by the least squares method.
It is indicated that rectangular B/D = 10 boxes are not always
prone to torsional flutter. Also, the mass moment of inertia is a
significant parameter, because the structural viscous damping,
i.e. the structural damping ‘‘buffer” against negative aerodynamic
damping, is proportional to the inertia.
In Fig. 7 the responses at cx < 1 and cx > 1 are shown. The
highest stable wind speed (U1) is significantly larger for the setup
having cx < 1. Furthermore the response is clearly stable contrary
the setup with a frequency ratio of cx = 2.10.
4.2. Series 2: Faired flat plate B/D = 24
Coupled flutter occurred close to the theoretical thin airfoil
wind speeds. A 8% higher wind speed was observed in Series 2A,
which is a measure of the uncertainty involved in assessing deter-
ministic values for critical flutter wind speeds. It has previously
been reported in the literature [26] that the stability boundary is
stretched over a considerable range of mean wind speeds. The lack
of synchronization of motion induced and fluctuating pressures
along the span may decrease the span-wise correlation of wind
forces and thus increase the critical wind speed.
4.3. Series 3: Twin boxes B/D = 27.8
Series 3A was conducted with 8 springs and a spring eccentric-
ity ek = 0.125 m. Small static displacements lh/D =  0.49 and l(a)
= 3.38° were seen at the maximum wind speed reached. In Series
3B the spring eccentricity was raised to ek = 0.2 m and dummy
masses were placed to reduce the torsional frequency and obtain
cx < 1. Very small static displacements were found at the maxi-
mum wind speed. Series 3C was identical to 3B, but without the
dummy masses giving a lower mass moment of inertia and
frequency ratio cx = 1.22. Coupled flutter occurred at a wind speed
1.25 times the corresponding thin airfoil flutter speed. Fig. 8 shows
the response of Series 3B at the highest obtained stable wind speed
U1 and the lowest instable wind speed U2 of Series 3C.
 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91 89

Fig. 9. Test Series 4C.

Fig. 10. Test Series 4D.

4.4. Series 4: Twin boxes B/D = 34.8
The number of springs and the spring eccentricity in Series 4A
were similar to Series 3A. At the highest wind speed minor static
displacements were seen. In Series 4B the same test was conducted
with only 4 springs, i.e. approximately half the stiffness, resulting
in large static displacements at U2. Wind speeds higher than U2
resulted in vertical oscillations around a non-zero torsional and
vertical mean value. In Series 4C the spring eccentricity was raised
to ek = 0.155 m. At Ur = 12.10 steady torsional and vertical displace-
ment occurred immediately after the release from the initial condi-
tions (See Fig. 9, right). At a rotation of a = 15° the section model
stalled until it gained lift again at a = 3.5°. The section model
stalled a second time, but this time it continued stalling until
a = 5° where it stopped and started to oscillate around a large
static mean displacement lh/D = 1.7 and la = 1.17° (nose
down). The absolute peak displacements was h/D = 3.31 and
a = 6.11°. Despite large amplitudes the oscillations were not
increasing. The oscillations were most likely driven by vortices
shed in the wake of the windward section impinging on the
leeward section.
The spring eccentricity was raised to ek = 0.19 m in Series 4D
which raised the critical wind speed compared to Series 4C.
Torsional Divergence did not occur in any of the tests which are
explained by the fact that the section model is able to rotate more
than the critical angle of attack when the section model stalls. In
Fig. 10 the torsional and vertical response of Series 4D is shown.
Series 4E had a frequency ratio of cx = 1.43 that was reached by
increasing the spring eccentricity to ek = 0.34 m. No instability was
observed. But the torsional and vertical frequencies were identical
at the highest wind speed which indicates that classical flutter
might occur at marginally higher wind speeds. Thus the critical
flutter wind speed for this section is at least 1.39 times higher than
the corresponding thin airfoil flutter wind speed.
90 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91
Stable
Non-Zero Mean
Zero Mean
Stable
Zero Mean
Fig. 11. Response depending on the torsional-to-vertical frequency ratio cx.
In summary it is seen that flutter is avoided if cx < 1. Instead
stable static displacements will occur as the wind speed increases.
At higher wind speeds large steady state oscillations will occur. As
illustrated in Fig. 11 these wind speeds are possibly larger than the
critical flutter speeds for sections having cx > 1.
5. Conclusions
The present paper demonstrates that the torsional-to-vertical
frequency ratio cx of suspension bridges decreases with an
increase in span length. This leads to low critical flutter wind
speeds and thus challenges the construction of super long span
bridges. Possible solutions are to increase the flutter speed using
twin box cross sections with central gaps. Wider gap widths seem
to be more beneficial than narrow gap width, which is still better
than no gap.
A more radical solution is to suppress coupled flutter by setting
cx < 1. This can either be done by decreasing the torsional stiffness
or by increasing the mass moment of inertia. If the torsional
stiffness is low, large static displacements will occur.
A better approach is to increase the mass moment of inertia.
This leads to aerodynamically stable behavior at the highest wind
speeds reached as shown for the rectangular B/D = 10 boxes having
cx < 1.
At frequency ratios cx < 1 flutter did not occur for any of the
tested twin box section models. Torsional Divergence did not occur
either because the section model was able to twist more than
a  15° where the bridge deck stalls. In the cases with low
torsional stiffness large static displacements occurred followed
by steady state oscillations.
The tests conducted with the wide gapped twin boxes reached
stable wind speeds up to approximately 2 times the corresponding
critical torsional divergence wind speed for a thin airfoil. The
higher the torsional stiffness, the higher the stable wind speeds
were clearly observed for the wide gapped twin boxes at cx < 1.
Instead of designing for critical torsional divergence wind speeds
it is more reasonable to use limit state values for the maximum
and mean static displacements allowed.
In summary, super long span bridges having cx < 1 might be
constructed by
(A) Optimizing the torsional stiffness.
(B) Increasing the mass moment of inertia.
(C) Decreasing the angle derivatives of the aerostatic force
coefficients.
The introduction of a large gap between twin boxes comply well
with the above mentioned criteria and may thus facilitate the
construction of long and super long span bridges. In practice this
could possibly be obtained by having the twin boxes external to
the cable planes ensuring that the radius of gyration is larger than
half the cable spacing. This would also reduce the bending moment
of the secondary span, i.e. the stresses in the cross beams between
the twin box girders.
Stable Steady state oscillations
Non-Zero Mean U
Aerodynamic Instability (Flutter)
U
Zero Mean
Acknowledgments
We want to extend a special thanks to the entire staff at Svend
Ole Hansen ApS for their hospitality and generous help during the
periods of wind tunnel tests. Fruitful discussions with bridge con-
sultancies in Denmark and with the Public Road Administration in
Norway as well as the great work and contributions from students
at the department are greatly acknowledged. A special acknowl-
edgement to Michael Lenius and Jesper R. Læsø for their valuable
work in the wind tunnel on Test Series 2 and 3 as well as prepara-
tion of Figs. 1 and 5 in the present work. Also the anonymous
reviewers are gratefully acknowledged. Their comments signifi-
cantly improved the paper.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.engstruct.2016.
04.025.
References
[1] Theodorsen T. General theory of aerodynamic instability and the mechanism of
flutter. NACA Report 496; 1934. p. 291–311.
[2] Sato H, Kusuhara S, Ogi K, Matsufuji H. Aerodynamic characteristics of super
long-span bridges with slotted box girder. J Wind Eng Ind Aerodyn
2000;88:297–306.
[3] Scanlan RH, Tomko JJ. Airfoil and bridge deck flutter derivatives. J Eng Mech
Div 1971:1717–37.
[4] Richardson JR. The development of the concept of the twin suspension bridge.
NMI R125; 1981.
[5] Walshe DE, Wyatt TA. Bridge aerodynamic 50 years after Tacoma Narrows –
Part 2: a new discipline world-wide. J Wind Eng Ind Aerodyn 1992.
[6] Bartoli G, D’Asdia P, Febo S, Mannini C, Noe S, Procino L. Innovative
configurations for long-span suspension bridges. In: Proceedings of the 5th
European and African conference on wind engineering, Florence, Italy; 2009.
[7] Bartoli G, D’Asdia P, Febo S, Mannini C, Pastò S, Procino L. Innovative solutions
for long-span suspension bridges. In: BBAA VI international colloquium on:
bluff bodies aerodynamics & applications; 2008.
[8] Nowicki T, Flaga A. Relation between shape and the phenomenon of flutter for
bridge deck-like bluff bodies. Arch Mech 2011;63:201–20.
[9] Larsen A, Larose G. Dynamic wind effects on suspension and cable-stayed
bridges. J Sound Vib 2015.
[10] Richardson JR. Advances in technique for determining the aeroelastic
characteristics of suspension bridges; 1975.
[11] Larsen A, Gimsing NJ. Wind engineering aspects of the east bridge tender
project. J Wind Eng Ind Aerodyn 1992;41–44:1405–16.
[12] Gimsing NJ, Georgakis CT. Cable supported bridges. 3rd ed. Wiley; 2012.
[13] Dyrbye C, Hansen SO. Wind loads on Structures. Wiley; 1997.
[14] MATLAB. MATLAB R2015A. Natick (MA, United States): The MathWorks Inc;
2015.
[15] Simiu E, Scanlan RH. Wind effects on structures: fundamentals and
applications to design. 3rd ed. New York, Chichester: Wiley; 1996.
[16] Joukowski N. Aérodynamique, Paris; 1916.
[17] Xu Y-L. Wind effects on cable-supported bridges. Wiley; 2013.
[18] Selberg A. Oscillation and aerodynamic stability of suspension bridges,
Trondheim; 1961.
[19] Selberg A, Hjorth-Hansen E. The fate of flat plate aerodynamics in the world of
bridge decks; 1976.
[20] Øiseth OA. Dynamic behaviour of cable supported bridges subjected to strong
natural wind. Trondheim: Department of Structural Engineering, Norwegian
University of Science and Technology; 2011.
[21] Matsumoto M. Aerodynamic damping of prisms. J Wind Eng Ind Aerodyn
1996;59:17.
 M.S. Andersen et al. / Engineering Structures 120 (2016) 82–91 91

[22] Matsumoto M, Mizuno K, Okubo K, Ito Y. Torsional flutter and branch
characteristics for 2-D rectangular cylinders. J Fluids Struct 2005;21:597–608.
[23] de Prony GR. Essai éxperimental et analytique: sur les lois de la dilatabilité de
fluides élastique et sur celles de la force expansive de la vapeur de l’alkool, à
différentes températures. J de l’école Polytechnique 1795.
[24] Brandt A. ABRAVIBE – a MATLAB toolbox for noise and vibration analysis and
teaching <http://www.abravibe.com>; 2011.
[25] Rainieri C, Fabbrocino G. Operational modal analysis of civil engineering
structures – an introduction and guide for applications. New York: Springer;
2014.
[26] Jakobsen J. Fluctuating wind load response of line-like engineering structure
with emphasis on motion-induced wind forces. Department of Structural
Engineering, University of Trondheim, Norway; 1993.