NHRE Masters Course Project Wind Engineering Portfolio

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

NHRE masters course Project Wind Engineering portfolio

Wind Induced Vibration of Long-span Bridges:


Flutter stability analysis of the Lillebælt suspension bridge

Samuel T., Simon K. and Mark M.

Supervisors: Prof. Dr. Guido Morgenthal and Dr. Abbas Tajammal

August 2015
Abstract

Long-span bridges are prone to wind induced vibration. Flutter is one of these wind induced aeroelastic instability where the bridge deck vibrate at
high divergent-amplitude, in a combined heave and pitch modes, and collapse. After the 1940 Tacoma incident, flutter analysis have been one of the
critical design concern of long-span bridges. The flutter stability of these structures is usually checked by calculating the wind speed at which flutter
occurs. This study discus the CFD (Computational Fluid Dynamics) based flutter stability analysis of long-span bridges. Taking the study case to
be the Lillebælt suspension bridge, the static divergent and the dynamic flutter stability limits are predicted using the vortex particle based CFD
flow solver VXflow [1]. The dynamic (fully coupled) aeroelastic simulation is performed for the case of single slice (1m strip bridge deck section) and
multi-slice (Quasi-3D) models. Theoretical back ground of flutter analysis is presented, results from different approaches are compared, conclusions
are drawn and recommendations for applications and future research offered.

Introduction

If history teaches us anything, it will be the severity of wind to long-span bridges, specifically to Tacoma narrow bridge. The 1940 Tacoma bridge
dramatic failure can be taken as an eye opening event towards the extreme wind induced vibration of flexible structures. Before the failure of Tacoma,
there seemed to be almost no awareness of the dynamic fluid structure interaction i.e. bridge aerodynamic.

Wind load analysis of flexible structures is a complex process and it relies on several interconnected parts. The pioneer in the field of wind engineering,
A. Davenport, breaks down this complex process in to wind loading chain [4]. In this approach, the wind load on a structure is determined by the
combined effect of the local wind climate, exposure of the structure to the local wind climate, the aerodynamic and dynamic characteristics of the
structure, and the criteria considered in evaluating the wind load. For pre-selected site, the wind load analysis boils down to the wind-structure
interaction i.e. aerodynamic modeling.

In general, the aerodynamic load vector F , on a unit segment of bridge deck, is a nonlinear function of the bridge deck dynamic response parameters

1
and the incoming flow.
F = Fi (α, α̈, α̇, ry , r˙y , r¨y , rz , r˙z , r¨z , U∞ , u, w) (1)

Where u(t) and w(t) are the longitudinal and vertical turbulence components of the incoming flow respectively. Other parameters are defined in
figure 1. Over-bars represent static deformation and over-dot represent derivative of the variables with respect to time.

Fz
r θ + rθ
FL

FD
α
Fx
β Fθ

β rz (c)
Urelt
w − r_z
U1 + u − r_y rθ
Z rz
ry
(b)
D X

B
ry
(a)

Figure 1: Instantaneous bridge deck wind interaction

Different analytical/numerical schemes aim on simplifying this nonlinear load-response relationship. The conventional semi-empirical methods base
their formulation on a linearized load-response scheme. In this approach, the aerodynamic loads are dealt as linear combination of the self excited
load referred as flutter (aeroelastic) force and the turbulence induced load referred as buffeting (aerodynamic) force.

2
   
Fae Fb
 x  x
Fae (α, α̈, α̇, ry , r˙y , r¨y , rz , r˙z , r¨z , U∞ ) = 
Faez  ,
 Fb (U∞ , u, w) = 
 F bz 
 (2)
Faeθ F bθ

The flutter force vector, Fae , is function of the dynamic structural response parameters, while the buffeting force vector, Fb , is function of the
fluctuating component of the incoming flow. The total aerodynamic load is determined by linear superposition of the buffeting and the flutter loads.
Considering the two momentous degree of freedoms, the dynamic equation of motion is formulated as

mr̈z + 2mξz ωz ṙz + mωz2 rz = Fz = Faez + Fbz (3a)

I r̈α + 2Iξα ωα ṙα + Iωα2 rα = Fθ = Faeθ + Fbθ (3b)

For the case of flutter, due to the co-dependency of the response and excitation force, there exist the response terms in the excitation part of the
equation. In order to solve the equation, dependency criteria must be established. For small amplitude structural response, it is observed that the
principle of potential flow theory holds, i.e. no separation of flow. This means, the self-excited aerodynamic loads are linear and in phase to the
structural displacement, velocity and acceleration; where the coefficients of these expressions referred to aerodynamic derivatives.

Taking this for granted, the flutter force in frequency domain, Fae , for flat plate type of cross section where the longitudinal displacement is negligible,
is formulated in terms of the aerodynamic derivatives as shown below.
 
1 2 ∗ r˙z ∗ B α̇ 2 ∗ 2 ∗ rz
Faez = ρU∞ B KH1 (K) + KH2 (K) + K H3 (K)α + K H4 (K) (4a)
2 U∞ U∞ B
 
1 2 2 ∗ r˙z ∗ B α̇ 2 ∗ 2 ∗ rz
Faeθ = ρU∞ B KA1 (K) + KA2 (K) + K A3 (K)α + K A4 (K) (4b)
2 U∞ U∞ B

K= (4c)
U∞
where ρ is density of the incoming flow, A∗i and Hi∗ are flutter derivatives which represent the ratio between the deck motion and the aerodynamic
forces at different reduced frequency K and ω is the vibration frequency of the bridge. Except for stream line sections, where analytic solution known
as the Theodorsen’s circulation function is available, in general the flutter derivatives are retrieved either from wind tunnel measurements or CFD
simulations.

3
Flutter Instability

The dynamic and static structural responses are directly related to the mean wind velocity. In fact, for a given mean wind velocity above some
threshold value, the response may develop towards what is perceived as unstable state. The case of flutter instability is visible when the dynamic
motion of the structure is formulated for non-turbulent flow (Fb = 0) and single mode (pure heave and pitch)scenario as shown in equation 5 below.
 
2 1 2 ∗ r˙z 2 ∗ rz
mr̈z + 2mξz ωz ṙz + mωz rz = Faez = ρU∞ B KH1 (K) + K H4 (K) (5a)
2 U∞ B
 
2 1 2 2 ∗ B α̇ 2 ∗
I r̈α + 2Iξα ωα ṙα + Iωα rα = Faeθ = ρU∞ B KA2 (K) + K A3 (K)α (5b)
2 U∞
Re-arranging the terms
1 1
mr̈z + 2mξz ωz − ρU∞ BKH1∗ r˙z + mωz2 − ρU∞ KH4∗
 
rz = 0 (6a)
|2 {z } |2 {z }
Aerodynamic damping Aerodynamic stif f ness
| {z } | {z }
ξef f Kef f

For ξef f < 0, i.e. 4mξz < ρB 2 H1∗ =⇒ Galloping instability. For Kef f < 0 =⇒ Static galloping instability (divergence)
1 1
I r̈α + 2Iξα ωα − ρU∞ B 3 KA∗2 r˙α + Iωα2 − ρU∞ KA∗3
 
rα = 0 (6b)
|2 {z } |2 {z }
Aerodynamic damping Aerodynamic stif f ness
| {z } | {z }
ξef f Kef f

For ξef f < 0, i.e. 4Iξα < ρB 4 A∗2 =⇒ Torsional Flutter instability. For Kef f < 0 =⇒ Static torsion instability (divergence)

As can be seen from equation 6; the aerodynamic force supply energy to the oscillatory motion by altering the stiffness and damping properties of
the system and leads to a progressive increase of the structural motion. In fact above a specific critical wind velocity, U∞ > Ucrt , a negative damping
will be introduced and leads to instability. This unique feature differentiates the flutter phenomenon from the resonance problems.

Flutter is usually coupled motion of at least two degree of freedoms, namely the vertical and torsion motions. This combination is common when the
torsion mode natural frequency is only slightly larger than the vertical mode natural frequency, which is often the case of slender long-span bridge
decks. The target of flutter instability analysis is to trace the critical condition of flutter occurrence i.e. determine the critical flutter velocity. After
identifying the critical flutter velocity different measures can be taken to fit the desired safety condition. This includes enhancing the dynamic and/or
aerodynamic behaviors of the bridge deck.

4
Theoretical back-ground: Flutter Instability Analysis

Unstable behavior is caused by the effects of the aerodynamic stiffness and damping. This introduces changes in the modal frequency of the structure
relative to the modal frequencies at wind velocity equal to zero. The stability limit can be determined by investigating the structural motion or
technically from the eigenvalue solution of the impedance matrix. For a bridge deck immersed in a turbulent flow, the modal time domain equilibrium
equation is given by
M0 η̈(t) + C0 η̇(t) + K0 η(t) = Fb (t) + Fae (t) (7)

where M0 , C0 and K0 are the modal mass, damping and stiffness matrix respectively, the subscript zero indicates that the values are determined
for the case of U∞ = 0. η is the time dependent response function modal vector. Fb and Fae are the modal flow induced (buffeting) and motion
induced (flutter) load vector respectively. Taking the Fourier transform of the above equation,

− M0 ω 2 + C0 ω + K0

aη (ω) = aF b (ω) + aF ae (ω) (8)

where aη , aF b and aF ae are the Fourier amplitudes of η, Fb and Fae respectively. Assuming that the amplitude of the modal motion induced load
to be proportional and in phase to the structural motion,

aF ae = (−Mae ω 2 + Cae ω + Kae ) aη (ω) (9a)

Substituting in to equation 8
aF b = − (M0 − Mae ) ω 2 + (C0 − Cae ) ω + (K0 − Kae ) aη (ω)
 
(9b)

Mae , Cae and Kae are Nmode by Nmode matrices that contain the aeroelastic cross sectional terms whose elements are shown below. The contribution
of the aerodynamic mass Mae is minimum and often neglected.

ρB 2 h i2 Z ρB 2
Z
Kaeij = ωi (U∞ ) Φi T K̂ae Φj , Caeij = ωi (U∞ ) Φi T Ĉae Φj (10)
2 2
Lexp Lexp

K̂ae and Ĉae are 3 by 3 matrices whose element are the flutter derivatives which are functions of the frequency of motion, the mean wind velocity
and the type of cross section. Φi is the ith mode eigen vector considering only the 3 primary degree of freedoms. ωi (U∞ ) is the ith mode mean wind

5
velocity dependent resonance frequency. Lexp is the length of the bridge exposed to the wind. Pk∗ , k=1-6 is the aerodynamic derivative associated
with the horizontal direction which is often neglected for streamlined section.
     
∗ ∗ ∗ ∗ ∗ ∗
φx (x) P P5 BP2 P P6 BP3
   1   4 
∗ ∗ ∗ ∗
Φi = φz (x) , K̂ae =  H5
   H1 BH2  , Ĉae =  H6
  H4 BH3∗ 

 (11)
∗ ∗ 2 ∗ ∗ ∗ 2 ∗
φθ (x) BA5 BA1 B A2 BA6 BA4 B A3
i

From equation 9b
aη (ω) = Hη (ω). aF b (ω) (12a)

Where the impedance matrix Hη (ω), which is Nmod by Nmod matrix that incorporate the stiffness and damping properties of the structure as a
function of the mean wind velocity dependent resonance frequencies, is given by
n h 1 i h1i o
2
H (ω, U∞ ) = I − kae − ω.diag + 2iω.diag .(ξ − ξae ) (12b)
ωi ωi
I is identity matrix, kae and ξae are Nmod by Nmod normalized aerodynamic stiffness and damping matrix respectively as shown below, ξ is the
structural damping ratio matrix.

Φi T K̂ae Φj Φi T Ĉae Φj
R R
" #2
Kaeij ρB 2 ωi (U∞ ) Lexp Caeij ρB 2 ωi (U∞ ) Lexp
kaeij = 2 = T
, ξaeij = 2 = (13)
Φi T .Φj
R R
ωi M0i 2m0i ωi Φi .Φj ωi M0i 4m0i ωi
L L

ωi and M0i are the ith mode resonance frequency and modal mass calculated in vacuum (U∞ = 0). While m0i is the modal equivalent and evenly
distributed mass associated with M0i . L is the length of the bridge segment.

The eigenvalue solution of the impedance matrix gives Nmode roots and each value represent a stability limit associated with all the considered
relevant mode shapes. Hη (ω) is a complex valued matrix, therefore while solving the eigenvalue problem, both the real and the complex matrix
determinant must be zero simultaneously. Once the instability mean velocity dependent frequency ωi (U∞ ) is determined, the associated mean wind
velocity U∞ is taken as critical flutter wind speed. In fact of all the eigenvalues that may be extracted, the lowest mean wind velocity is taken to be
the critical flutter velocity.

As discussed above, close to a stability limit the vertical and torsion modes dominate the response. Considering bridges; the static torsion instability
(divergence in torsion), dynamic instability in vertical direction (Galloping), dynamic instability in pure torsion and dynamic instability in combined

6
vertical and torsion modes (flutter) are common. Considering the first torsion and vertical modes Φθ and ΦZ (that have three components as shown
in equation 11) with corresponding ωθ and ωZ , the impedance matrix reduced to;
( " # " #  ωr 2   
ωr
   )
1 0 kae11 kae12 ω
0 ω
0 ξ1 − ξae11 −ξae12
H (ωr , Ucrt ) = − − θ  2  − 2i  θ   .    (14)
ωr
0 1 kae21 kae22 0 ωr 0 ωZ
−ξae21 ξ2 − ξae22
ωZ

where ωr = ωθ (Ucrt ) = ωZ (Ucrt ) is the mean wind velocity dependent resonance frequency. As can be seen from equation 10, the calculation of the
aerodynamic stiffness and damping matrix is vastly reduced if pure modes ΦZ and Φθ with corresponding ωz and ωθ are considered, i.e.
h iT h iT
Φθ (x) ≈ 0 0 φθ (x) , ΦZ (x) ≈ 0 φz (x) 0 (15)

In the case of flutter instability, the coupling of the torsion and vertical modes occur via the off-diagonal terms of the stiffness and damping matrix
of equation 14. The characteristic equation of the eigenvalue problem constitute a complex quadratic equation with unknown flutter frequency ωr .
The solution obtained will in general be complex, ωr = ω1 + i ω2 , whose real and imaginary parts can be interpreted as representing both decay
and oscillatory components respectively in the natural frequencies. This implies for ω1 > 0 the response will be divergent and for ω1 < 0 decaying
response. When ω2 ≈ 0 the oscillatory part vanishes and the phenomenon of static divergence, loss of either vertical and/or rotational stiffness,
observed. In this case, the instability effect is caused by negative stiffness and it is static not dynamic.

Solution Strategy

The aerodynamic derivatives are function of the reduced frequency K, which in turn is function of frequency of the motion and the mean wind
velocity as shown in equation 4. Since the aerodynamic stiffness and damping are function of the aerodynamic derivatives, in order to solve the eigen
problem of equation 14, iteration is required.

In CFD based studies what is done is that, first the aerodynamic derivatives for range of K (from 0 to 16) are simulated using forced vibration, for
vertical and torsion motions, by only varying the frequency of motion. For any desired K value the, aerodynamic derivatives can be derived by using
either linear or higher order interpolation. To get the static divergence critical frequency, a preliminary value of K is chosen and the procedure is
Bωr
repeated until ω2 vanish or approximately equals to zero. To that solution ωr = ω1 and Ucrt = K
.

The solution strategy is that, a dynamic coupled analysis (in which the real modal frequencies are used) is simulated at the vicinity (often at the
range of ± 5) of the static divergence critical wind velocity. Then decision is made by evaluating the displacement time history.

7
Study subject: The Lillebælt suspension bridge

The Lillebælt suspension bridge is part of Denmark’s motor express system. Inaugurated in 1970 and became the first suspension bridge in Denmark.
It spans 1.44 km and consists two towers that rise 120 meters above sea level. It is solidly anchored in 33.000 tons of armored concrete on each side
and accommodate ships as tall as 42 meters.

Figure 3: Schematic elevation and plan view of the bridge [3]

Figure 2: Perispective view of the Lillebælt bridge [2]

Figure 5: Simplified analysis bridge deck section [2]

Figure 4: Pre-fabricated bridge deck segment: 24m erection unit [2]

8
Fundamental Study

Forced vibration simulation

In order to determine the flutter derivatives, the bridge deck is oscillated harmonically (forced motion) in the vertical (heave) and rotational (pitch)
degree of freedoms separately. The procedure is repeated for K values between 0 and 12. To get the desired K values; only the bridge deck vibration
frequency is manipulated. The constant inputs parameters are shown in table 1 below,

B [m] U∞ [ ms ] Heave amplitude [m] Pitch amplitude [rad] Motion type


33 10 0.5 0.5 Harmonic - Sine function

Table 1: Forced vibration input parameters

Figure 6: Forced heave harmonic oscillation Figure 7: Forced pitch harmonic oscillation

The aerodynamic derivatives are determined by comparing the measured force of the forced oscillations with the force at zero motion. As shown
in figure 8 and 9, for comparison the Theodorsen equivalent flutter derivatives (flat plate with the same width) are presented together. Linear
interpolation is used to derive the intermediate values.

9
0 3
Lillebaelt Lillebaelt
-2 Theodorsen Theodorsen
2
-4

-6 1
H$1

H$2
-8 0
-10
-1
-12

-14 -2
1 2 4 6 8 10 12 14 16 1 2 4 6 8 10 12 14 16
U1 U1
B#! [-] B#! [-]

10 2
Lillebaelt Lillebaelt
Theodorsen Theodorsen
0 1

-10 0
H$3

H$4
-20 -1

-30 -2

-40 -3
1 2 4 6 8 10 12 14 16 1 2 4 6 8 10 12 14 16
U1 U1
B#! [-] B#! [-]

Figure 8: Heave motion flutter derivatives

10
3 0

2.5
-0.5

2
-1
Lillebaelt
A$1

A$2
1.5
Theodorsen
-1.5
1

-2
0.5
Lillebaelt
Theodorsen
0 -2.5
1 2 4 6 8 10 12 14 16 1 2 4 6 8 10 12 14 16
U1 U1
B#! [-] B#! [-]

8 0.8
Lillebaelt
Theodorsen

6 0.6
A$3

A$4
4 0.4

2 0.2
Lillebaelt
Theodorsen
0 0
1 2 4 6 8 10 12 14 16 1 2 4 6 8 10 12 14 16
U1 U1
B#! [-] B#! [-]

Figure 9: Pitch motion flutter derivatives

11
Static divergence flutter velocity

Using the simulated and Theodorsen aerodynamic derivatives the static divergent flutter velocity is determined by applying the iteration procedure.
The result is summarized in table 2 below.

Aerodynamic derivatives K [-] ω [ rad


s
] Ucrt [ ms ]
Lillebælt 8.16 2.20 94.2
Theodorsen 8.62 2.08 93.8

Table 2: Static divergent flutter parameters

Single slice dynamic (fully coupled) simulation

Here the response of a 1m strip bridge section located at the middle of the central span is used to evaluate the global flutter response of the 3D
system. Ten simulations, at the vicinity of the static divergent critical velocity i.e. 94.2 m/s, are carried out. Specifically from 90 to 100 m/s with 1
m/s increment. The minimum wind speed where the structure starts to flutter is taken as the critical wind speed. The bridge is set to vibrate with
first heave and pitch modal frequencies, identified from the 3D FEM analysis, simultaneously. Details of the considered modes is shown in figure 11.
As shown in figure 10, for both modes a normalized modal displacement of (0 1 0) (edge center edge) is considered.
m
1

1
B
0

Modal deformation

Figure 10: Considered normalized modal displacement

12
After running different simulations, the structure starts to flutter at wind speed of 96 m/s. For comparison the vertical displacement time history of
the bridge at 95 m/s is presented.

Figure 11: First heave and pitch modes

Figure 12: Vertical displacement time history: Left 95 m/s , Right 96 m/s, ( ) lead and ( ) trail node

13
The flow field at 96 m/s is shown below. The bridge deck vibrates in combined heave and pitch modes with high amplitude and collapse.

140

120

100

80

60

l.edge
t.edge
Lift
0.5 Moment
d/dmax, L/Lmax, M/Mmax

−0.5

−1
0 40 80 120 160 200 240 280 320 360 400
tU/B

Figure 13: Flow field visualization at 96 m/s

14
Multi-slice (Quasi-3D) dynamic simulation

Multi-slice dynamic analysis is a fully coupled simulation the same as the single dynamic analysis, except in this case different locations along the
longitudinal axis of the bridge are considered. Moreover, instead of taking only the fundamental heave and pitch modes, additional higher pitch and
heave modes are also incorporated in the analysis. Seven slices (5 in the central span and 1 each in the approach span) are considered.

Figure 14: Considered slice locations Figure 15: VXflow Quasi-3D Model

In addition to the first heave and pitch modes shown in figure 11, 2 heave and 2 pitch modes, shown below, are considered. Furthermore, the actual
normalized modal displacement of each mode is feed in to the simulation.

Figure 16: Second heave and pitch modes

15
Figure 17: Third heave and pitch modes

In this case, the quasi-3D bridge model is set to vibrate with the selected modes for mean wind velocity in the range of 85 m/s to 95 m/s. The same
as the single slice dynamic analysis, decision is made based on the displacement response of the system. Following that, the critical flutter velocity
in this case becomes 89 m/s. The vertical displacement time history of the central section of the bridge at 88 m/s and 89 m/s is presented below.

Figure 18: Vertical displacement time history for the central section (SL4 ): Left 88 m/s , Right 89 m/s, ( ) lead and ( ) trail node

16
The vertical displacement of the other sections (slices) for the case of 89 m/s is shown below.

Figure 19: Vertical displacement time history at 89 m/s : Left SL6 , Right SL7 , ( ) lead and ( ) trail node

Unlike the single slice dynamic flutter vertical displacement, in the quasi-3D dynamic analysis the axis of vibration shifts. This shows that there is
a static deformation.

New axis of vibration

Original axis of vibration

Figure 20: Static deformation: SL3 vertical displacement time history at 89m/s, ( ) lead and ( ) trail node

17
Figure 21: Quasi-3D VXflow flow visualization at 89 m/s

18
Summary

The critical flutter velocity determined using the different methods is summarized below.

Approach Description Ucrt [ ms ] Remark


Theodorsen Uses Theodorsen equivalent aerodynamic derivatives 93.8 Pure analytical
Static divergence Uses forced simulation aerodynamic derivatives 94.2 Actual bridge dynamic behavior - not considered
Single slice dynamic simulation Fully coupled single slice simulation 96.0 Considers the 1st heave and pitch modes
Quasi-3D dynamic simulation Fully coupled multi-slice simulation 89.0 Considers the first three heave and pitch modes

Table 3: Summary of analysis result

Conclusion

• The flutter stability limit (critical flutter velocity) is highly influenced by the method of analysis. In all cases the flutter vertical displacement
exceeds the ultimate structural limits. In reality the structure will deteriorate and local damages propagate and lead to collapse before reaching
this abrupt amplitude as it is witnessed in Tacoma bridge.

• Since the cross section of the bridge deck is close to its flat plate equivalent, the static flutter velocity determined using the Theodorsen
(pure-analytical) and the simulated (numerical) aerodynamic derivative are very close. Even though it is expected the single slice dynamic
and the quasi-3D approach critical velocity to be close, the difference is noteworthy. This might be due to the fact that the contribution of
higher modes and/or the static deformation noticed in the quasi-3D approach is significant. Considering this case, for preliminary design the
Theodrosen approach gives satisfactory result with minimum computation cost.

• From the considered approaches, the quasi-3D dynamic simulation is a better representative of the real system. And it happens to be the
critical velocity from this approach is the minimum one. Therefore, for final design it is vital to study the 3D wind-structure interaction.

• The static deformation noticed in the quasi-3D dynamic analysis requires further investigation. Its source need to be identified and its
contribution to the overall stability of the system have to be quantified.

19
Bibliography

[1] G. Morgenthal. Aerodynamic analysis of structures using high-resolution vortex particle methods. PhD Thesis, 2002.

[2] C. Ostenfeld, A. G. Frandsen, J. J. Jessen, and G. Hass. Design and construction of the bridge.

[3] C. Ostenfeld, A. G. Frandsen, J. J. Jessen, and G. Hass. Model tests for superstructure of the suspension bridge.

[4] S. Roberts. Wind wizard: Alan G. Davenport and the art of wind engineering. Princeton university press, 2012.

20

You might also like