Engineering Tripos Part IIB

Module 4A10 Flow Instability

Flow-Structure Interaction

1 Introduction
In the first eight lectures of this course all solid boundaries were
considered to be rigid and stationary. This is not always accurate.
In reality, flow causes structural motion, which in turn alters the
flow.
Fluid flow

Structural force on fluid Flow force on structure

Structural
motion

This can give rise to devastating results. For example:

• in 1965, three cooling towers collapsed in a high wind (figure
1). The full report is in the library (shelf mark TI.11). We
shall explore why this occurred in section 3;
• for a different reason the Tacoma Narrows suspension bridge
collapsed in 1940 (figure 2). Although there is still debate
as to what exactly caused this, it was one of the effects
explained in sections 5 and 6;
• pipes in chemical plants and in the home frequently rever-
berate. In section ?? we shall examine why a flow within a
pipe often makes the pipe unstable;
• turbulent buffeting, while random, can excite natural fre-
quencies of structures. This will be covered in section ??.

2 Non-dimensional parameters for flow-

structure interaction
Engineers and physicists frequently use non-dimensional parame-
ters because they contain no information about man-made scales
of length, time, mass etc. Hence they provide the most convenient
way of describing scale-invariant phenomena. These parameters
are the ratios of two influential quantities (see box).

1
Figure 1: Aerial photo of Ferrybridge power station after a
storm. The down-wind cooling towers have collapsed.

Figure 2: Tacoma Narrows suspension bridge. Before and

after failure.

2
 D U/f
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaa
y
U

Figure 3: The reduced velocity, Ur is defined as the ratio of

the path length per cycle, U/f to the diameter of the object,
D.

Non-dimensional parameters and order of

magnitude estimates

When working out the tendencies of a physical situ-

ation the ‘∼’ sign is often used. This sign means ‘is
of the order of’ or ‘scales with’ or ‘is similar to’. It is
generally used to drop constants without losing im-
portant physics. For instance, the inertial force per
unit length from a fluid on a cylinder is given by:
1 2
Inertial force = ρU CD D ∼ ρU 2 D
2
The viscous shear force is given by:
Z
Viscous shear force = τ · nt ds
S

where the tangential shear, τ · nt , is integrated

around the surface πD. We make an estimate for
τ and obtain:
Z  
∂u U
τ · nt ds ∼ µ πD ∼ µ D = µU
S ∂r D

Putting this together we obtain:

Inertial forces ρU D
Reynolds number = =
Viscous forces µ

2.1 Reduced Velocity

The mass in figure 3, oscillating at frequency f , traces out a sinu-
soid in the fluid moving past it at velocity U . The wavelength is
U/f . The reduced velocity, Ur is the ratio of this wavelength to
a characteristic dimension of the mass, in this case its diameter,
D:
Path length per cycle U
Ur ≡ = (1)
characteristic dimension of model fD

3
 Why should this be called the reduced velocity? What
does the product f D represent physically?

Definitions vs. constraints

The sign ‘≡’ means ‘is equivalent to’ or ‘is defined

as’. It is used when you are defining a new variable.
It is different to ‘=’, which is usually used to specify
a constraint (e.g. one of the boundary conditions or
the equation of motion).

2.2 Mass Ratio

The mass ratio is a measure of the relative independence of a mass
in a fluid.
mass of model m
Mass ratio ≡ = (2)
mass of fluid displaced ρV

V is volume, not velocity. If the mass ratio is high (e.g. a grand-

father clock’s pendulum in air), inertial forces from the fluid will
have little effect on the motion. If it is low (e.g. a pingpong ball
in water), buoyancy and inertial forces from the fluid dominate
the behaviour.

2.3 Reduced Damping (Scruton Number)

energy dissipated per cycle 4πmζ
Sc ≡ Mass Ratio × = (3)
total energy of structure ρV
where the damping factor, ζ is defined as:
1 energy dissipated per cycle
ζ≡ (4)
4π total energy of structure

A car’s suspension is as close to critical damping (ζ = 1) as

possible. If you sit on the bonnet and jump off, it stops moving
after roughly one cycle. On the other hand, a tuning fork has
ζ  1 and vibrates through several thousand cycles before coming
to rest. At the other extreme, a squidgy executive toy, which
returns to its original shape a day after being deformed has ζ  1.
What is the physical significance of the reduced damp-
ing. Hint: Think about the instant when all the struc-
ture’s energy is kinetic.

4
 Q.1 Non-dimensional parameters at Ferry-
bridge

The Ferrybridge Cooling

179 ft
Towers had a wall thick-
ness of 5 inches, a density 60 ft
165 ft
of 2400 kgm−3 and a res-
onant frequency measured 375 ft
at 0.6Hz for the side-to-
side motion. On the night
of the storm which de-
stroyed them, the wind
averaged around 70 mph.
Estimate the reduced ve- 210 ft
locity and the mass ratio.

3 Vortex Shedding
A cylinder in steady flow sheds vortices alternately from each
side at a well-defined frequency. Interesting things happen when
the natural vortex shedding frequency is similar to the natural
frequency of the cylinder.
In this section, we shall:
• describe vortex shedding qualitatively, assuming a station-
ary cylinder;
• describe the qualitative behaviour when the cylinder is vi-
brated at an externally-imposed frequency;
• describe the qualitative behaviour when the cylinder is free
to vibrate at any frequency it chooses as the flow velocity
is increased;
• evaluate a quantitative expression for the response of a
cylinder to vortex-induced forcing at a certain frequency;
• discuss mechanisms of avoiding dangerous oscillations due
to vortex shedding.

Bluff Bodies

An object is described as a bluff body if the boundary

layers separate as the flow passes over the body:

aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaa
U a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaa
aaaaaaaaaaaa
aaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
a aaaaaaaaaaaaaaa a aaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa a aaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
a a a a a a a a a aaaaaaaaaaaa
a aaaaaaaaaaaaaaa a aaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa a aaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaa a aaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaa

5
 Fast Fluid
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaa
Slow Fluid Parallel shear layers
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaa

Fast Fluid

a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
Sinuous instability of
aaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
the double shear layer
a a a a a a a a a

aaa a a a a a a a
aaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
Non-linear development
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaa
a a a a a a a a a
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
of the sinuous instability
aaa a a a a a a a
aaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a

Figure 4: Vortex shedding behind a cylinder (bottom) is the

non-linear development of a sinuous instability of the wake
flow (middle). The wake flow can be modelled as a double
shear layer (top).

3.1 The causes of vortex shedding

To a first approximation, the flow behind a bluff body can be
modelled as two parallel shear layers, as shown in figure 4 (top).
Each shear layer is unstable to the Kelvin-Helmholtz instability
(section 4.1), which will tend to amplify any disturbances. How-
ever, the double shear layer has an additional mode of instability,
where the two layers snake up and down in phase (figure 4 mid-
dle). In technical terms, each single shear layer is convectively
unstable and acts as an amplifier, whereas the double shear layer
is absolutely unstable and acts as an oscillator. This oscillator
beats at its own frequency, which in this case is determined by
the thickness of the wake. The non-linear development of this
sinuous instability is vortex shedding.
The wake is absolutely unstable over a range of frequencies.
When the body is rigid and stationary, the natural frequency of
vortex shedding, fs , is that with the highest growth rate. This
can be expressed as a non-dimensional parameter, the Strouhal
number:
fs D
S≡ (5)
U
where U is the free stream velocity and D is the distance between
the shear layers. D is usually taken to be a characteristic trans-
verse dimension of the body. Note that the Strouhal number is
similar to the inverse of the reduced velocity, equation (1), but
with a different frequency.

6
Figure 5: Strouhal number S ≡ fs D/U as a function of
Reynolds number for circular cylinders. (From Blevins p48)

Figure 6: Strouhal number S ≡ fs D/U as a function of

Reynolds number for non-circular sections. D is measured
over the largest cross-section presented to the flow.

The Strouhal number is a function of the Reynolds number

and the bluff body’s roughness, which affect the wake’s thickness
(figure 5). The Strouhal number for different bluff body shapes
can be seen in figure 6. In this figure, D is the maximum
width of the body. What would figure 6 look like if D
were taken as the wake thickness instead? (The discussion
on p51 of Blevins is particularly interesting on this point).

7
 0.5

0.4 Will lock in

0.3
Ay / D

0.2

0.1
Won’t lock in Won’t lock in
0
0.4 0.6 0.8 1 1.2 1.4 1.6
ωf / ωs

Figure 7: A cylinder of diameter D is vibrated with fre-

quency ωf and amplitude Ay . This figure shows the small-
est amplitude at which the vortex shedding frequency locks
into the forcing frequency. The natural vortex shedding fre-
quency of the stationary cylinder is ωs . The experimental
results are assembled from Blevins p55.

Q.2 Vortex Shedding frequencies

Using figure 5 calculate the vortex shedding fre-

quency from:
• a telephone pole 20cm in diameter in a 10m/s
wind;
• a marine pipeline 1m in diameter in a current
of 3m/s;
• the cooling towers at Ferrybridge.
The density of air is 1.3 kgm−3 and its dynamic vis-
cosity at 15◦ C is 1.8 × 10−5 Nsm−2 . The dynamic
viscosity of sea water is 1.0 × 10−3 Nsm−2 and its
density is 1025 kgm−3 .

3.2 Effect of forced bluff body motion on vortex

shedding
If a bluff body is vibrated, it can be made to shed vortices away
from its wake’s natural frequency, fs (figure 7). When the vibra-
tion frequency, f , is near to the natural vortex shedding frequency,
fs , it only takes a small amplitude for the vortex shedding to ‘lock
in’ to this new frequency. As f moves further away from fs it re-
quires larger amplitude vibrations to achieve lock-in.
This is consistent with the view that a range of frequencies
are absolutely unstable, with amplification rates highest at the
natural frequency fs . Other frequencies can dominate but only if
the forcing is sufficiently strong that the natural frequency, which
amplifies faster, does not overtake.
Correlation of the vortex shedding along the length of the

8
cylinder is also enhanced by forced vibration. In addition, you
can get lock-in when the cylinder is excited at a sub-harmonic of
the natural frequency of vortex shedding.

3.3 Feedback between bluff body motion and

vortex shedding
Bluff structures have a natural vibration frequency and a natural
vortex shedding frequency. The latter varies with flow velocity, so
‘lock-in’ can occur over a range of free stream velocities.
In figure 8, the reduced velocity is plotted on the horizontal
axis and fs /f is on the vertical axis. A stationary body sheds
vortices at a fairly constant Strouhal number so its shedding fre-
quency is proportional to U/f D, which is the diagonal line. The
shedding frequency from a bluff body which can vibrate follows the
diagonal line until it approaches the natural vibration frequency.
Then the two frequencies lock and the oscillations have a much
higher amplitude, shown here by Ay /D. At higher velocities they
lock out again. At Re ∼ 104 lock in occurs at a reduced ve-
locity of between 5 and 6. How could you have predicted
this from figure 5?

Q.3 Vortex lock-in at Ferrybridge?

Given the reduced velocity you calculated for

the Ferrybridge cooling tower in Question 1, would
you expect lock-in to occur for a single cooling tower?

At Ferrybridge, the design wind speed was not ex-

ceeded during the storm. It’s clear that something
unexpected happened. What might it have been?

3.4 Harmonic Model

There are many different models for the interaction between a
structure’s vibration and vortex-shedding. Details can be found
in Blevins (p60). We shall examine the simplest: a forced mass
on a spring (figure 9). It neglects the effect of the structure on
the flow. The force due to vortex shedding, FL , is assumed to be
harmonic at an imposed frequency ωs . We model its amplitude
with a lift coefficient CL :
 
1 2
FL = ρU D CL sin(ωs t) (6)
2

The harmonic model cannot model the feedback mechanism

by which lock-in is achieved. However, it does provide useful
intuition. The equation of motion of the cylinder is:
1
mÿ + 2mζωy ẏ + ky = ( ρU 2 D)CL sin(ωs t) (7)
2

9
Figure 8: Top: Experimentally-determined frequency of vor-
tex shedding fs for a spring-supported cylinder as flow ve-
locity is increased. The natural frequency of the system is
f . Bottom: Amplitude Ay divided by the diameter of the
cylinder D. (From Blevins p60)
FL
aaa a a a a a a a
aaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
U aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
D
a aaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
y aaaaaaaaaaaaaaaaa

Figure 9: The forced mass on a spring shown here is a rudi-

mentary model for the response of a structure to vortex shed-
ding.

10
 2 2
ρU CL / 2k = 0.1 ζ = 0.0316
1.5 ζ = 0.1
ζ = 0.316
y/D

0.5

0
0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4

0
φ

−1

−2
0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
ωs / ωy

Figure 10: Response of the cylinder on a spring, with res-

onant frequency ωy to forcing from vortex shedding at fre-
quency ωf . The damping factor, ζ, varies.

where ωy is the resonant frequency of the cylinder. The steady

state response has amplitude Ay and phase difference φ from the
imposed force:
y = Ay sin(ωs t + φ) (8)
The solution to the non-homogenous ordinary differential equation
(7) is given in Thompson 1993 (page 51, shelf mark RT.103). One
obtains:
1
y ρU 2 CL sin(ωs t + φ)
= ( 2 )1 (9)
D   2 2  2 2
k 1 − ωωys + 2ζ ωωys

Where the phase angle φ is given by:

2ζωs ωy
φ = tan−1 (10)
(ωs2 − ωy2 )

This is shown graphically in figure 10. Unsurprisingly, the re-

sponse is highest when the vortex shedding frequency is roughly
the same as the resonant frequency of the cylinder. The amplitude
when ωs = ωy is approximately:

Ay ρU 2 CL
= (11)
D 4kζ

11
 Q.4 Maximum response of a damped oscillator

A close inspection of figure 10 reveals that the maxi-

mum amplitude occurs when the steady state vibra-
tion frequency ωs is slightly off-set from the resonant
frequency ωy . Use equation (9) to show that this off-
set is given by:
ωs p
= 1 − 2ζ 2 (12)
ωy

What would happen if you built two identical bridges

side by side?

Q.5 Reduced amplitude at lock-in

Show that equation (11) can be expressed in terms

of the non-dimensional parameters introduced in sec-
tion 2:
Ay CL
= 2 2 (13)
D π S Sc
Given that S is the only parameter to contain U
and that S is a constant 0.2 for natural vortex
shedding, what does equation (13) tell us about the
velocity-dependence of the amplitude at lock-in?
What is going on here?

Hint: use the fact that ωs2 = ωn2 ≈ k/m.

3.5 Reduction of vortex-induced vibration

3.5.1 Modifying the flow field
Vortex shedding arises from a wake instability behind the bluff
body. If the body is streamlined so that there is no wake, there
will be no vortex shedding. Obviously this is only practical when
the object can rotate or when the flow always comes from the
same direction.
If separation is inevitable, one can add a suppression device.
These disrupt or prevent the formation of a two-dimensional vor-
tex sheet. Some devices are shown in figure 11. Design rules for
such devices are given in Blevins, pp 79-81. More devices can be
found in Naudascher p161.

3.5.2 Modifying the structure

Vortex-induced vibration is most damaging when the shedding
frequency fs locks in to the resonant frequency of the structure

12
Figure 11: Vortex suppression devices. (a) helical strake;
(b) shroud; (c) axial slats; (d) streamlined fairing; (e) split-
ter; (f) ribboned cable; (g) pivoted guiding vane; (h) spoiler
plates. (From Blevins, p78)

f . Lock-in can be avoided by stiffening the structure. To avoid

lock-in, one should keep the reduced velocity below approximately
1.

3.5.3 Damping
If all else fails, one can increase the Scruton number; equation
(3). As a rough design rule, when the Scruton number exceeds
60, peak amplitudes at resonance are less than 1% of diameter.
The Scruton number can be increased by increasing either the
mass, the damping of the structure, or both.
How do the helical strakes in figure 11 (a) suppress
vibration of a chimney?

13
4 Modelling fluid forces on a structure
In some situations, a fluid’s effect on a structure can be modelled
by assuming that the structure is heavier or more damped and
then calculating its motion in a vacuum. In this section we shall:
• examine the pressure force that a stationary fluid exerts on
an accelerating structure;
• introduce the concept of ‘added mass’;
• examine the pressure field that an accelerating fluid exerts
on a stationary structure;
• derive Morison’s equation, which expresses the force on an
accelerating structure within an accelerating fluid;
• introduce the concepts of ‘added damping’ and ‘added stiff-
ness’;
• examine the effect of fluid viscosity and wave radiation on
oscillating structures, in terms of ‘added mass’ and ‘added
damping’.

4.1 Added Mass

If a ping pong ball is released under water, how quickly does it
accelerate? One could look at the buoyancy force on the ball,
F = (ρl − ρg )V g ≈ 0.3N divided by the mass of the ball, m ≈ 5g.
This gives a ≈ 60 ms−2 . After one second at that acceleration,
the ball would have travelled 30m and be going at 135 mph. What
is wrong?
The problem is that the mass ratio (Eq. 2) is small (∼ 0.1),
so the buoyancy force on the ball mainly accelerates the water
around the ball, rather than the ball itself. We shall now see how
the mass of the water can be modelled by assuming that the ball
is heavier than it actually is.
For simplicity we will look at the quasi-2D case. The kinetic
energy of the fluid around a moving cylinder is given by:
1
K.E.L = ρL U 2 πa2 (14)
2
per unit length. The derivation of this is in Derivation 1. The
total kinetic energy is:
1
m + ρL πa2 U 2


K.E. = (15)
2
where m is the mass per unit length of the cylinder. The work
done on the cylinder in time δt is f U δt, where f is the force per
unit length on the cylinder. This equals the change in total kinetic
energy:

f U δt = δ(K.E.) (16)
 
d 1 2

2
⇒ fU = m + ρL πa U (17)
dt 2

dU dU
⇒ f = m + ρL πa2 = m0 (18)
dT dT

14
 So to examine the equation of motion of an object in a sta-
tionary fluid, you can simply add some virtual mass (in this case
ρL πa2 ) to its mass and perform the calculation as if it were in a
vacuum. The added masses for other shapes are given in Blevins
(page 25).
In an unsteady flow, where the fluid is accelerating and there-
fore contains a pressure gradient, there is a further force on the
object. This is just like the buoyancy force that one obtains due
to the pressure gradient caused by gravity. You will meet this in
question 6.
As an aside, we derived the added mass expression for an
infinitely large volume of inviscid irrotational incompressible fluid.
The fluid is in phase with the object everywhere and consequently
there can be no travelling waves. If travelling waves could
exist (e.g. in an inviscid compressible fluid) would it be
easier or harder to accelerate the mass?

15
Q.6 Added Mass in an accelerating flow (slow
way)

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

The net pressure force per unit length is:

Z
f = − p ndS (19)
S

where p is given by Bernoulli:

∂φ 1
p = −ρL − ρL u2 (20)
∂t 2
and the velocity potential for a cylinder is:

a2
 
φ = U (t) r + cos θ (21)
r

Also,
 2  2
∂φ 1 ∂φ
u2 = u2r + u2θ = + (22)
∂r r ∂θ

Derive expressions for ∂φ/∂t and u2 in terms of r,

θ, a and dU/dt. Where is u2 a maximum and where
is it a minimum? What are the values of u2 at these
points?

Using the facts that ds = adθ and that n = ex cos θ+

ey sin θ, substitute relevant expressions into equation
(19) and show that the force exerted by the acceler-
ating fluid on the cylinder is:
dU
f = 2ρL πa2 (23)
dt
Compare this with the force exerted by a stationary
fluid on a cylinder which is accelerating at dU/dt.
(This is given in the notes). What is the magnitude
of the extra force? Physically, how does this extra
force arise? Could you have used physical intuition
to write down this result without going through the
derivation?

Hint: For the last part, evaluate the gradient of

the pressure field in a fluid accelerating in the x-
direction? What is the volume of the cylinder per
unit length?

16
 Q.7 Added Mass in accelerating flow (quick
way)

A beam has square cross-section of side length 2a

and is accelerated in a stationary fluid. The added
mass is given by 1.51ρL πa2 .

What would the added mass be if the beam were

stationary and the fluid were accelerating?

4.2 Morison’s Equation

The inviscid analysis in the previous section is limited to potential
flows. In practical situations, there is also a drag force, which can
be modelled by the classic expression FD = 1/2ρU 2 CD . The
total inertial force per unit length for a stationary object in an
accelerating flow is the sum of the buoyancy force and the added
mass force. If the object has displacement y, the inertial force FI
and the fluid dynamic drag force FD are given by:

FI = ρAU̇ + Ca ρA(U̇ − ÿ) (24)

1
FD = ρ|U − ẏ|(U − ẏ)DCD (25)
2
where A is the cross-sectional area and Ca is the added mass
coefficient (this is 1 for a cylinder in potential flow). Adding these
together we obtain Morison’s equation:
1
F = ρAU̇ + Ca ρA(U̇ − ÿ) + ρ|U − ẏ|(U − ẏ)DCD (26)
2
Physically we have merged two contradictory models. To cal-
culate the added mass, we assumed a potential flow in which there
can be no flow separation and no drag. Then we added drag,
which is entirely caused by flow separation. It’s ugly, but it is a
useful expression because it includes the three important physical
effects: buoyancy, added mass and fluid dynamic drag. One is
free to adjust Ca and CD to fit experimental data, which is quite
successful over a variety of regular and irregular oscillating flows.
Morison’s equation is almost exact for 1) acceleration
in an inviscid fluid and 2) Steady velocity at Re > 1000.
Why is this?

17
 Q.8 Morison’s Equation

A pier stands in shallow coastal water, lapped by

the waves. The waves have amplitude H and the
pier’s legs have diameter D. In shallow water,
the horizontal component of velocity is given by
U = (H/2)(g/d)1/2 cos ωt, where H is the trough-
to-crest wave height, d is the mean water depth and
g is gravitational acceleration. The wavelength is
given by λ = 2π(gd)1/2 /ω. Take CD ≈ 1 and Ca = 1.

Under what conditions of H/D will the horizontal

force on the pier’s legs be dominated by drag forces
rather than buoyancy and inertial forces?

4.3 Added Stiffness

When you jump onto a boat in a still lake, it bobs up and down.
The buoyancy force is equal and opposite to the displacement, so
it behaves like an added stiffness. Note that this added stiffness is
not the same as the buoyancy term in the added mass expression
evaluated in question 6, which arose due to the acceleration of a
fluid which completely surrounded the body.

18
 Q.9 Added stiffness

Y0

A buoy is made from a hollow cylindrical drum with

diameter D, height H and mass mD attached to a
hemi-spherical mass of concrete, mC . The depth of
the bottom of the drum below sea-level is Y0 .

The buoy rests in an absolutely calm sea. Derive an

expression for Y0 . What is the minimum height of
the drum in order for the buoy not to sink if D = 1m,
mD = 10kg, mC = 500kg, ρl = 1025kgm−3 ?

The buoy is pushed downwards by a distance Y (to

Y0 + Y ). What is the buoyancy force upwards? Ex-
press this as a spring constant k.

4.4 Added Damping

With the exception of the drag force in Morison’s equation, all the
forces encountered so far have been proportional to acceleration,
since they are inertial forces. (In an oscillating system, one would
say that the forces are in phase with the acceleration).
You are already familiar with forces which are proportional
to (or in phase with) velocity, such as viscous drag. These forces,
which allow the dissipation of energy, give rise to an ‘added damp-
ing’ term in an equation of motion.
We shall shortly come across another set of forces which are
in phase with the velocity of the object but which have the oppo-
site effect to damping. They pump energy from the fluid to the
structure. These can be thought of as ‘negative damping’ forces.
It is these which give rise to the most destructive flow-structure
interactions.

4.5 Harmonic motion of objects in fluids & added

coefficients
The equation of motion of an oscillating object is:
mÿ + 2mζωy ẏ + ky = F (t) (27)
If y and F (t) are both harmonic and at the same frequency, the
force F (t) can be split into components that are in phase with ÿ,

19
Figure 12: Effect of viscosity on (a) added mass and (b)
added damping coefficients for a cylinder vibrating with
small amplitudes within a rigid, concentric cylindrical shell
[From Naudascher p43]. ωR2 /ν is a Reynolds number. As
it decreases, the added mass and added damping increase.

ẏ and y:
F (t) = −A0 ÿ + −B 0 ẏ + −C 0 y (28)
By substitution into (27), the equation of motion becomes ho-
mogenous:

(m + A0 )ÿ + (2mζωy + B 0 )ẏ + (k + C 0 )y = 0 (29)

A0 is the added mass, B 0 is the added damping and C 0 is the

added stiffness. In practice, it is difficult to distinguish between
added mass and added stiffness, since they are both in phase, and
these terms are often lumped together. Now if you work out the
added mass and added damping due to the fluid, you can treat
the oscillating system as if it were in a vacuum.

4.5.1 Effect of viscosity

Increasing the fluid viscosity increases both the added mass and
the added damping, as shown in figure 12.

4.5.2 Effect of wave radiation

In a compressible fluid, waves will carry energy away from a vi-
brating object. To the object, this appears to be a dissipation
of energy, just like damping. This will give rise to a phase shift
between the force and acceleration. For example, if the displace-
ment is y = sin(ωt), the force will be F = F0 sin(ωt + φ). In terms
of ÿ and ẏ this is:
F (t) = −A0 ÿ − B 0 ẏ (30)
Hence wave radiation affects the added mass term A0 and creates
an added damping term B 0 . These terms depend on the Froude

20
Figure 13: Added mass and added damping coefficients for
a flat piston of radius R vibrating at frequency f [From
Naudascher p41]

number f R/c, where c is the wave speed. The coefficients A0 and

B 0 for a flat piston vibrating in a wall are shown in figure 13 . Note
that as the wavespeed tends to infinity, one recovers the potential
flow situation, where there is no added damping.

Q.10 Radiation damping

The buoy in question 9 is displaced by a distance Y

and left to bob up and down on the absolutely calm
sea. Ignoring damping, write down the equation of
motion.

Take the added mass of the buoy to be ρπD3 /24.

What is the natural bobbing frequency? Describe
qualitatively what will happen over time.

The buoy is in shallow water, depth 2m, with wave

speed c = (gd)1/2 . From figure 13, estimate the
added damping due to wave radiation and re-write
the equation of motion containing the damping term.

Hence estimate the damping factor ζ, equation (4),

the mass ratio, equation (2) and the Scruton number,
equation 3.

21
 Derivation 1 Kinetic energy of the fluid around a
moving cylinder

Consider a cylinder of radius a moving at constant velocity U

through a still fluid. We will calculate the kinetic energy of the
fluid around the cylinder, assuming potential flow.
The velocity potential for flow around a cylinder is:

a2
 
φ = U0 r + cos θ
r

where θ is the angle with the free stream flow. The kinetic energy
of the fluid around the cylinder is:
Z Z ∞ Z 2π
1 2 1 2
K.E. = ρu dV = ρu rdθdr
V 2 r=a θ=0 2

1 2 1  2
ρ u + u2θ

but ρu =
2 2 " r
 2  2 #
1 ∂φ 1 ∂φ
= ρ +
2 ∂r r ∂θ
1 2 1 2 a 4 1 2  a 2

= ρU + ρU − ρU 2 cos 2θ
2 2 r 2 r
The first term is the kinetic energy of a uniform flow with no
obstacle, summed over all space. We shall subtract this in a mo-
ment. The second term is a rotationally symmetric average veloc-
ity increase around the obstacle. The third term is θ-dependent.
It accounts for the fact that the flow goes faster around the top
and bottom of the cylinder while coming to a stop along the left
and right edges. When integrated around the cylinder it disap-
pears.
The difference between the kinetic energy of the flow around
the cylinder and the plug flow is:

1 2 ∞ 2π  a 4
Z Z   a 2 
∆K.E. = ρU −2 cos 2θ rdrdθ
2 r=a θ=0 r r
1 2 πD2
 
= ρU
2 4

22
5 Galloping
A flow can cause a structure to vibrate if the structure’s lift co-
efficient changes with angle of attack. This is called galloping
and is independent of vortex shedding. Most structures with non-
circular cross-sections can gallop.
As an example, let us consider the object with square cross-
section shape in figure 14. Figure 14(a) shows the object vibrating
in the laboratory frame. Figure 14(b) shows the object in a frame
of reference following the object and oriented with the apparent
flow direction. Figure 14(c) shows how the boundary layers and
the body force change with flow orientation. As the mass moves
downwards, the apparent angle of attack increases. The pressure
on the lower side of the body decreases, causing a downwards force
on the body. This continues until the boundary layer re-attaches.
The resultant plot of force coefficient Cy ≡ Fy /(1/2ρU 2 ) vs. α is
shown in figure 15(a)i.

aaa a a a a a a a aaa a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
D aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa
a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
.
a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa
y a a a a a a a a a
aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaa

(a) (b)

aaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa

aaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a

(c)

Figure 14: (a) A mass on a spring in a cross-flow. (b) When

the mass moves up and down, the apparent angle of attack
α changes. (c) This changes the vertical force on the mass.
Under certain conditions this can lead to positive feedback
and galloping

23
 Q.11 Calculating Cy from CL and CD

For the galloping beam in figure 14, whose velocity

is ẏ (defined as positive in the downwards direction)
in a free stream velocity U , write down expressions
for the relative velocity Urel and the apparent angle
of attack α in terms of U and ẏ.

In the frame of reference moving with the mass

(figure 14b), the lift force is FL (positive upwards)
and the drag force is FD (positive to the right).
Write down an expression for the vertical force in
the stationary frame, Fy (positive downwards), in
terms of FL , FD , and α.

The coefficients CL , CD and Cy are defined such that

2 2
FL = ρUrel DCL /2, FD = ρUrel DCD /2 and Fy =
2
ρU DCy /2 (note that Cy is defined in terms of U ,
not Urel ). Show that Cy is related to CL and CD by
the following exact expression:
CL cos α + CD sin α
Cy = − (31)
cos2 α
Show that, for small angles around α = 0:
  
∂Cy  ∂CL
= − + C D (32)
∂α α=0 ∂α α=0

5.1 Soft Excitation

Under some conditions, the mass will start to gallop from rest.
This is known as soft excitation. As before, the equation of motion
of the forced mass-spring-damper system is:

mÿ + 2mζωy ẏ + ky = F (t) (33)

And this time the force F (t) is given by:

1 2
F (t) = ρU DCy (34)
2

24
 The quasi-steady assumption

We have assumed that we can work out the force (Eq.

34) on the mass by using a coefficient Cy which was
calculated for a steady flow. This assumes that any
unsteady flow structures have been swept far enough
downstream that their influence on the force is neg-
ligible. This is the quasi-steady assumption and it is
only valid when the flow velocity Urel is much greater
than the body velocity (∼ f D). In other words the
reduced velocity Ur ≡ U/f D is much greater than
unity (in practice it needs to be above 20). This ap-
proximation is valid for aeroplane wings but is less
valid for bridges. We shall meet the methods used
to analyse gallop on bridges in section 6.3.

Since we shall be dealing with small angles of attack α, we

can make the following approximations around the equilibrium
position at α = 0:

∂Cy 
Cy (α) = Cy |α=0 + α + O(α2 ) (35)
∂α α=0
   2
ẏ ẏ ẏ
α = tan−1 ≈ + O( ) (36)
U U U
Urel = U (1 + α2 )1/2 ≈ U + O(α2 ) (37)
If we off-set the steady force due to Cy |(α=0) against a steady
force from the spring at equilibrium, this term drops to zero and,
dropping the O(α2 ) terms one obtains:

ẏ ∂Cy 
Cy = (38)
U ∂α α=0

The equation of motion (33) becomes:


1 2 ∂Cy  ẏ
mÿ + 2mζωy ẏ + ky = ρU D (39)
2 ∂α α=0 U

  
1 ∂Cy 
⇒ mÿ + 2mζωy − ρU D ẏ + ky = 0
2 ∂α α=0
(40)
So the force can be expressed as an added damping term. How-
ever, note that if ∂Cy /∂α|α=0 > 0 there is negative damping. By
this mechanism the fluid pumps energy into the harmonic motion.
The oscillator will be unstable to soft excitation when:

1 ∂Cy 
ρU D > 2mζωy (41)
2 ∂α α=0

Since the volume per unit length, V , is equal to D2 , this can be

written as: 
∂Cy 
Ur > 2Sc (42)
∂α α=0

25
 steady−state response
Cy

2
(∂ Cy / ∂α)0

α Ur / Sc

(a)i (a)ii

steady−state response
Cy

2
(∂ Cy / ∂α)0

α Ur / Sc

(b)i (b)ii

Figure 15: The velocity at which galloping starts (through

soft excitation) depends on ∂Cy /∂α around α = 0 (Eq. 42).
Plots of Cy against α for two galloping oscillators are shown
on the left and the steady state responses are shown on the
right as a function of the reduced velocity Ur and reduced
damping Sc. The bottom oscillator starts galloping at lower
Ur than the top oscillator because ∂Cy /∂α|α=0 is larger.

where Ur is the reduced velocity (1) and Sc is the reduced damping

(3). For a given Scruton number, the critical velocity decreases as
the gradient ∂Cy /∂α|(α = 0) increases (figure 15).

26
 Q.12 Frozen raindrops on a power line

Power lines normally have circular cross-sections, which means that CL and CD don’t
change with apparent angle of attack. Consequently ∂Cy /∂α is negative (Eq. 32) and the
lines are always stable to galloping.

However, a ridge of frozen raindrops on the line can give it lift. This will make the line
gallop, depending on where the ridge is. The aim of this question is to find out the range of
angles α (defined below) which can give rise to galloping and to estimate the critical velocity.

The drag and lift coefficients for a power line with a ridge of frozen raindrops is shown
in the above figure. The angle α is defined on the left. Note that it is in the opposite
direction to the apparent angle of attack α, as defined in figure 14. We are concerned with
soft excitation: i.e. small angles around α.

If CD were zero at all angles, between what values of α would the power line be potentially
unstable to galloping? Hint: use equation (32).

Estimate the magnitude of ∂Cy /∂α at α = 0◦ . Note that α needs to be converted to radians.

A power line of diameter 5cm and average density 3000 kgm−3 has a damping factor ζ of
0.001 and a vibration frequency of 4 Hz. Calculate the Scruton number. If droplets freeze
on it at α = 0◦ , calculate the reduced velocity at which it will start to gallop. What wind
velocity does this correspond to? Don’t forget to take CD into account.

5.2 Hard Excitation

If ∂Cy /∂α is positive at some angle of attack, but not at zero angle
of attack, galloping is still possible. The critical flow velocity must
be exceeded and an impulse large enough to reach point 1 on figure
16(a) is required.
A soft excitation response can develop into a hard excitation
response if the gradient ∂Cy /∂α increases with α. Let us consider
an object with the Cy curve given in figure 16(b)i. The object will
start to gallop when the critical velocity is reached, at point 2.
Then, because ∂Cy /∂α increases as the apparent angle of attack
increases, the galloping gets harder and harder and will reach the

27
 steady−state response
Always Potentially unstable
stable to hard excitation
Cy

1
1
0

α Ur / Sc

(a)i (a)ii

steady−state response
3
Cy

2
0
α Ur / Sc

(b)i (b)ii

Figure 16:

steady-state response at point 3. It seems that vigorous gallop-

ing has suddenly started from nothing! Futhermore, the critical
velocity required to sustain galloping reduces. The flow velocity
could reduce all the way to point 1 before the object would come
to rest again. More details can be found in Naudascher, p204.

Q.13 Hard Excitation of an iced power line

The ridge of ice on the power line in question 12 has

formed at α = 80◦ . At what values of α might the
power line be prone to galloping through hard exci-
tation? Galloping will occur when the wind velocity
is high enough and when the initial velocity of the
power line is high enough. Estimate these values.

28
 S-shaped curves

The plots on the right of figure 16 are steady state so-

lutions to the equation of motion (33). The solid lines
represent stable solutions. The dotted lines repre-
sent unstable solutions. In fluid mechanics and com-
bustion, one often encounters response curves which
double back on themselves, as in figure 16(b)ii, with
an unstable region sandwiched between two stable
regions. In a flame, point 2 on the figure would rep-
resent sudden ignition of two reactants as their tem-
perature is increased. Calculation of these plots is
beyond the scope of this course, but details can be
found in Naudascher p 187.

5.3 Torsional Galloping & Torsional divergence

Galloping systems have a single degree of freedom. In the previous
case, this was translational but it can also be rotational. The
equation of motion is:

I θ̈ + 2ζωθ I θ̇ + kθ θ = Fθ (43)

where Fθ is the moment, which varies with both θ and with θ̇. The
concept is the same. Further details can be found in Blevins p111.
Another type of instability occurs if the aerodynamic moment
increases with θ faster than the restoring force. This happens to
the bonnet of your car if you’re driving fast and it hasn’t been
shut properly. This is called torsional divergence.

29
 Q.14 Torsional Divergence

A car engine cooler consists of a spring-loaded flap

which opens up wider as the car goes faster:

The aerodynamic moment on the flap is

1 2 2
Fθ = ρU D Cm (θ)
2
Write down the force balance at equilibrium.

The aerodynamic moment coefficient Cm (θ) has the

simple form:

Cm (θ) = Cm0 + Cmθ θ

where Cm0 > 0. Sketch the magnitudes of Fθ and

kθ on the same graph. The two lines intersect at
the equilibrium value of θ. What happens as U
increases? Evaluate an algebraic expression for the
angle θ at equilibrium.

Sketch a graph of θ against U . Hence or otherwise,

evaluate an upper bound for the critical divergence
velocity, where the flap suddenly snaps open.

5.4 Prevention of Galloping

Galloping occurs when the component of the fluid force in phase
with velocity exceeds the damping force. This is expressed (Eq 41)
as: 
1 ∂Cy 
ρU D > 2ωn ζm (44)
2 ∂α α=0
By inspection of this equation, a number of techniques to avoid
galloping become obvious:

5.4.1 Modifying the flow field

One method is to ensure that ∂Cy /∂α|α=0 is negative. This can
be achieved through smooth aerodynamic contours which prevent
flow separation. For example, an aerofoil will be stable since Cy ≈

30
Figure 17: The Severn road bridge has a broadly aerody-
namic cross-section, which is intended to reduce the ten-
dency of this suspension bridge to gallop.

Figure 18: Examples of passive dampers on structures

−CL at small α (Eq. 31) and ∂CL /∂α|α=0 is positive until stall.
This philosophy was used in the design of the Severn Road Bridge
(figure 17). Other examples can be found in Naudascher, p260
and Blevins, p140.
Note that this technique will not necessarily prevent gallop
due to hard excitation. If ∂Cy /∂α becomes positive at higher
angles of attack, as in figure 16a, the object will gallop when
given a sufficiently large starting impulse or displacement.
How might this starting impulse be provided? Could
an aerofoil shape be susceptible to hard excitation?

5.4.2 Modifying the structural components

Modifying the flow field alone is often not enough. Indeed, most
suspension bridges are theoretically prone to galloping in very high
winds. However, one can design the critical velocity to exceed the
highest expected wind.
√ By expressing the term ωn m on the RHS
of equation (44) as km, one sees that increasing the structural
stiffness or mass (or both) increases the critical galloping velocity.
This technique worked for the second Tachoma Narrows bridge,
which was 40% more massive than the first.

5.4.3 Damping
If all else fails, one can always increase the damping (as for the
Millenium Bridge across the Thames). Passive dampers can be
used, as in figure 18, or the rather more intelligent Tuned Mass
Damper, shown in figure 19. Details of these and how they work
can be found in Naudascher p261 and Simiu p357.

31
Figure 19: Tuned Mass Damper system installed in the Citi-
corp Center, New York City

32
Figure 20: The Glasgow Wing Tower. According to the de-
signers, it is “actively rotated in order to control the wind
direction relative to the tower. The main staircase of the
tower has been designed to have an airfoil shaped cross sec-
tion to reduce the wind loads in line with the incident wind
and to reduce the potential for vortex shedding. In addition
there are a number of peripheral airfoils to promote attached
flow over the main body and hence reduce the drag.”

6 Flutter
6.1 General Description
If an object has both translational and torsional degrees of free-
dom (e.g. the Glasgow Wing Tower in figure 20) it is possible
to get coupled mode flutter. The remarkable feature of coupled
mode flutter is that negative damping is only required in one of
the modes for the coupled mode to become unstable.
For example, in a non-stalled aerofoil the translational mode
is always stable. This is because ∂Cy /∂α|α=0 is always negative
(section 5.1). The translational mode is shown in the left hand
line of the graph in figure 21, where the natural frequency and the
added damping are shown as Ur increases. The added damping,
B 0 , is always positive for the translational mode. The torsional
mode, on the other hand, does become unstable (B 0 < 0) at a
critical velocity. Above this reduced velocity, the modes lock into
each other and become unstable.
A qualitative description of steady state flutter motion is also
shown in figure 21. As the aerofoil starts to twist, it catches the
wind, which pushes it up. The aerofoil’s inherent stiffness acts to
counteract the twist and at the top of the cycle it is horizontal
again. However, its angular momentum carries it through the
horizontal and then it catches the wind again, this time forcing it
down. So the cycle continues, with the phase of torsion leading
that of translation by 1/4 cycle.

33
 aaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaa

aaaaaaaaaaaaaaaa a a a a a a a a a
aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa aaa a a a a a a
aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaa

aaa a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaa

Figure 21: Classic flutter is a combination of a translation

mode and a torsion mode, which are shown in the box at
the top left. At a critical reduced velocity Ur , one of the
modes becomes unstable (i.e. has negative added damping
B 0 ), as shown at the top right. This renders the combined
flutter mode unstable at a frequency ωn . The torsion mode
leads the translation mode by 1/4 of a cycle, leading to the
motion shown in the bottom of the figure.

34
 aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

Figure 22: Diagram of a spring-supported aerofoil with

translational and torsional degrees of freedom. The aero-
dynamic force, which acts on the aerodynamic centre, is re-
solved to Fy and Fθ around the location of the springs (the
elastic axis). The chord length is C.

6.2 Analysis at high reduced velocity

In order to analyse flutter, we need an equation of motion for each
degree of freedom. These equations are cross-coupled and have to
be solved together. We solve the equations by writing them in
matrix form and setting the determinant to zero. This type of
solution crops up frequently in flow instability analysis. You will
have met the basic concepts in A-level maths.
Referring to figure 22, one can write down the translational
and torsional equations of motion. These are derived in Blevins
Appendix B:

mÿ + Sx θ̈ + (2mζy ωy )ẏ + ky y = Fy (45)

Iθ θ̈ + Sx ÿ + (2Iθ ζθ ωθ )θ̇ + kθ θ = Fθ (46)

The moment, Fθ (positive clockwise) and the vertical force, Fy

(positive downwards) are taken about the elastic axis. These are
modelled by the expressions:
 ( )
1 2 ∂Cy  ẏ ct θ̇
Fy = ρU c θ+ − (47)
2 ∂α α=0 U U
 ( )
1 2 ∂Cy  ẏ ct θ̇
Fθ = ρU cca θ+ − (48)
2 ∂α α=0 U U

These are derived in the same way as the galloping force (Eq.
38), noting that the aerofoil has actual angle of attack θ as well as
apparent angle of attack (ẏ/U − ct θ̇/U ). In practical applications,
one can also use the fact that Cy ≈ −CL .
To solve these two equations, we use the same approach as
that in section 2 of the flow instability course, where we con-
sider small harmonic perturbations around y = θ = 0 of the form
y = Y0 est . The exponential allows amplification (Real{s}) and

35
simple harmonic motion (Imag{s}). One substitutes the follow-
ing expressions into equations (45) and (46):

y = Y0 est θ = θ0 est
ẏ = sY0 est θ̇ = sθ0 est
ÿ = s2 Y0 est θ̈ = s2 θ0 est

and then one expresses the equations as a matrix:

(ms2 + 2mζy ωy s − qs (Sx s2 − q + qcUt s )

  
U + ky ) Y0
=0
(Sx s2 − qcUa s ) (Iθ s2 + 2Iθ ζθ ωθ s + kθ − qca + qcaUct s θ0

where q ≡ 1/2ρU 2 c(∂Cy /∂α|α=0 ).

There is a trivial solution to this matrix equation, where the
vector [Y0 , θ0 ] = [0, 0]. For non-trivial solutions, the determinant
of the matrix must be zero. This gives us an equation of the form:

C0 s4 + C1 s3 + C2 s2 + C3 s + C4 = 0 (49)

which can be solved for s. If s has a positive real component,

small disturbances will tend to amplify and the coupled mode is
unstable. This occurs when:

C1 C2 C3 < C0 C32 + C4 C12 (50)

by the Routh-Hurwitz stability criteria (Mechanics Data Book).

This produces an expression for the exact speed at which flutter
will start. In question 15 you will determine the significance of
the positions of the aerodynamic center and the centre of mass.
As an aside, there will only be four values of s which satisfy
equation 49. These are known as the eigenvalues. This type of
problem is an eigenvalue problem.

36
Q.15 Simplified Flutter Analysis
The flutter analysis can be simplified if one ignores
damping and models the forces as:

1 2 ∂Cy 
Fy = ρU c θ
2 ∂α α=0

1 2 ∂Cy 
Fθ = ρU cca θ
2 ∂α α=0

Show that the coefficients C0 ...C4 in equation (49)

are:

C0 = mIθ − Sx2
C1 = 0
C2 = mkθ + Iθ ky − qca m + qSx
C3 = 0
C4 = ky kθ − qca ky

Solve equation (49) for s, in terms of C0 , C2 and C4 .

Examine the sign of the real component of s under

these two situations:
• 4C0 C4 > C22
• C0 C4 < 0
Hence write down the stability criteria for this
system. What do these two situations represent
physically?

Now consider and comment on the following situa-

tions:
• The aerodynamic centre, the centre of mass
and the pivot point all coincide;
• Only the centre of mass and the pivot point
coincide
Note that this question concerns a very special case.
All the damping terms, both positive and negative,
are eliminated. Consequently the instability mech-
anism is different from that in figure 21, which was
due to energy being pumped into the object via a
negative damping (B 0 ) term.

37
6.3 Analysis at low reduced velocity (e.g. for
bridge decks)
So far we have used the quasi-steady approximation, which as-
sumes that the lift and drag coefficients are constant. This re-
quires that the fluid force doesn’t depend on the previous mo-
tion of the object. In other words any flow structures which
have formed must be swept downstream very quickly relative to
the body’s motion. This is only the case for reduced velocities
Ur ≡ U/f D greater than around 20, which means that the analy-
sis will not work for bridges, where Ur ∼ 1. Two ways round this
problem are presented here.

6.3.1 Theodorsen’s and Scanlan’s approach

When a bridge deck is stable, oscillations die away exponentially.
When it is unstable, they grow exponentially. At the stability
boundary they remain harmonic. Bridge designers are only really
interested in the stability boundary, in order to find the wind
velocity where flutter starts. At this boundary the bridge’s motion
is harmonic, so the forces will also be harmonic:

Fy = c1 y + c2 ẏ + c3 θ + c4 θ̇ (51)
Fθ = c5 θ + c6 θ̇ + c7 y + c8 ẏ (52)

These are substituted into the equations of motion (45 & 46).
When the quasi-steady approximation is not valid, the force co-
efficients ci depend on the incident wind speed and the frequency
of oscillation. They are worked out from wind tunnel tests. Typ-
ically, for various wind speeds, a section model is moved through
prescribed harmonic motions in the wind tunnel and the resulting
pressures measured. These are integrated to find the net aerody-
namic force and moment, allowing the ‘flutter derivatives’ ci , to
be determined.
Designing a typical long-span bridge may involve hundreds of
tests on each potential cross-section before a profile with a suitable
set of flutter derivatives is found. Until recently, this was the most
common approach.

6.3.2 Discrete Vortex approach

Theodorsen and Scanlan’s method is firmly rooted in a structural
engineering approach to the problem: find a way of modelling the
fluid force and treat the bridge as a mass on a spring. Recently,
however, structural engineers have adopted the discrete vortex
method from fluid mechanics. You met this in 3A1 Incompressible
Flow when you modelled the flow over an aerofoil by distributing
vortices on its surface.
A large number of vortices are placed around the section to
be modelled (figure 23). The vortex strengths are determined by
the condition that there must be no flow through the surface and
no separation at the trailing edge.
Then the vortices are released from the surface of the section,
modelling diffusion and convection of vorticity into the flow. New
vortices are instantly created on the surface. The spilled vortices

38
U aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a

Figure 23: Discrete Vortex approach. The flow around

an aerofoil can be calculated numerically by distributing vor-
tices on the aerofoil surface. The strengths of the n vortices
are determined by the (n − 1) constraints that there is no
flow across the surface of the aerofoil, together with the final
constraint, which is the Kutta condition. (This is another
example of an eigenvalue problem).

then roll up forming the familiar vortex street. The fluid force
on the section can then be calculated. This gives us the coupling
between body motion, fluid motion and fluid force that we were
looking for.
The discrete vortex method is now applied to the design of all
large suspension bridges (e.g the 3.3km span across the Straights
of Messina between Italy and Sicily). From a numerical point of
view it is much more efficient than finite volume or finite element
techniques.

7 Prevention of flutter
Like galloping, flutter can be prevented by increasing the mass,
stiffness and damping of a structure. If the situation corresponds
to that in figure 22, the aerodynamic centre can also be moved for-
wards or the centre of mass moved back. Reducing ∂Cy /∂α|(α=0)
will increase the critical velocity at which flutter starts. How-
ever, changing its sign will not necessarily prevent flutter. These
answers drop out of the solution to question 15.

39