DETERMINATION OF WIND Until now experimental identification of model

LOADS CAUSING FLUTTER parameters were confined only to model

EFFECTS
Ladis1.allsLwambllka 1
Email: liiuambuka@yahoo.de
ABSTRACT
A new method on the determination of wind forces by
experimental means in a wind tunnel is discussed. The
method reported herein is based on the identification
methodology of model parameters under time response
(i.e. time domain identification).
As a test model, a thin plate of cypress wood, elastically
suspended in three degrees of freedom to represent the
idealized form of a classical airfoil, for which theoretical
wind load parameters exist, has been employed. Wind
load parameters, experimentally obtained on the test
model under the new method, are compared with the
theoretical values of classical airfoils.
Advantages of the new method against the conventional
methods are discussed.
Key words:
Flutter = elastic vibrations due to
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)
vibrations in still air (natural model). In this
paper the identification technique is being
introduced to determine vibration parameters
of a model under wind flow in a wind tunnel
(aerodynamic model). The parameters obtained
from testing the aerodynamic model are non-
constant. They are dependent on wind speed
and vibration frequency.
The primary objective of wind tunnel tests
reported herein is to extract, from the
aerodynamic model parameters, the influence
of wind in the form of wind load derivatives.
For this, the constant natural model parameters
constitute reference. values.
The previous method used to determine wind
load parameters under free response was
developed by Scanlan (1971 and 1975) and is
widely used to predict flutter instability for
long span bridges. With the method of Scanlan,
wind load parameters were obtained under
several test procedures depending on the
number of vibration modes to be excited.

wind loads Technical constraints as well as the adopted

Flutter instability = magnified elastic approximation techniques which form the
vibrations signifying the critical wind speed essential part of the method had adverse effects
System parameters = properties of a structure in on the accuracy of results obtained.
natural modes
Wind Load Parameters = properties ofa structure
signifying the influence of wind load The method presented in this paper has the
Flat plate = idealized model with basic advantage in that the limitations inherent
structural behaviour as a streamlined body in the conventional method are eliminated. In
Time domain = responses on basis of time this approach only one test procedure is
interval required because the method is based on
Frequency domain =responses on basis of coupled modes of vibrations. The application
frequency interval of the method presented herein, however, is
possible only by means of high speed digital
INTRODUCTION computers and appropriate software.

Experimental identification of system 2. BASIC ANALYTICAL PRINCIPLE

parameters from time response was developed
by Ibrahim (1973 and 1976) and its applicability The motion of a structural object under wind
proved to be very effective. ·Since then action and assuming a steady flow is described
subsequent improvements of the method was through the following set of equations (1):
made by Badenhausen (1985) and Ibrahim
(1985) in terms of enhanced accuracy,
resolution and anti-disturbance.

I
Dr. Ing., Senior Lecturer, Department of Structural Engineering,
University of Dar es Salaam,Tanzania.
 L. Lwambuka

U+ K U~~
/1';",':' I
M s U + Ds U + K s U =M ·".'{Y + D w w (1)
with
M s , D s ; K s = System Parameter Matrices
M w' D w, K w = Wind Load Parameter Matrices

U = The Displacement Vector

The fundamental similarities.of both sides of eq.(l) allow one to re-write eq.(l) in the form of:

(2)

which can be simplified to read:

MU+DU+KU=O (3)

with
M=M
- - s -M
_w

D=D
- - s -D
-w (4)
K=K
- - s -K
-w
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Eq.(3) is homogenous and defines a free 3. SYSTEM PARAMETER

response motion, the solution of which leads to
an eigenvalue problem.
In the albescence of wind flow (i.e. zero wind
speed) it has been shown by Lwambuka (1988)
that the identification yields system parameter
matrices M s, D s, K s (natural model). Model
parameters identified under the influence of
wind (aerodynamic model) are found to differ
considerably from those obtained in still air.
The influence of wind (i.e wind load
parameters) can be extracted by use of eq. (4) as
functions of wind speed.
IDENTIFICATION TECHNIQUE
3.1 Test Model
The identification of parameters is based on the
measurement of acceleration response signals
by use of accelerometers which are mounted on
the test model at locations which correspond to
the degrees of freedom. The test model selected
for this purpose was a thin plate, elastically
suspended to allow three degrees of freedom as
shown in fig.(l). The model was selected to
comply with the classical airfoil idealization
(=thin plate) from which theoretical wind loads
be derived using the methodology as described
by Theodorsen (1935) and Lwambuka (1988).

Tanzania Journal 0/ Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 1
Ladislaus Lwambuka

y
>-

c c.
'1

Fig.I: Elastically Restrained Flat Plate Model

With The author adopted the vertical coordinate

M a , Mb= flat plate masses system which he found to be adequate for
8 a ,8b = flat pate moment of inertia response measurements by use of
k = elastic stiffness accelerometers. In fig.2 is a plan view of the test
c = reference geometrical length model showing positioning of accelerometers,
v = wind speed whereas the model suspension in wind tunnel
Ul, U2, U 3 = vertical coordinates is shown in fig . 3
h, a, p = classical coordinates
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Fig.2: Test Model

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1. (No. 2), March, 2007
 Determination of Wind Loads Causing Flutter Effects

Fig.3: Model Suspended in Wind Tunnel

3.2 Response Data Processing

The accumulated response signals are first k=~ (6)
m,
interpolated to fonn Langrange's polynomial
Whereby m, the polynomial interpolation
function of 7th order. In order to minimize the
order, equals 7.
noise modes nonnally associated with the
measurements, the polynomial function is
Eq.(5) as it stands is a product of three
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

integrated numerically. Henceforth, the

integration processes and therefore contains 3
terminology "integrated acceleration" is
unknown constants, i.e. the initial response
adopted. But since the application of eq.(3)
values . Determination of the 3 integration
requires also the knowledge of the velocity and
constants, which represent the initial
displacement functions, two additional
conditions, can be avoided by employing a
successive integrations are necessary.
differentiation concept, which is prescribed by
Therefore, after three numerical integration
Badenhausen (1985). The differentiation
processes of the acceleration response, eq.(3)
process reduces the number of identification
leads to eq.(5) which is known as "the
equations by 2 to form eq.(7):
identification equation":

M sV c + DsV e + K sV e = 0, e = l,k (5 ) M sU e + D sU c + K sU e = 0, e = 1,p (7)

Where the response vectors are designated:
u integrated acceleration
u integrated velocity and
u = integrated displacement.
If for each measuring station on the test modet
5 response signals have been recorded, then the
total number of identification equations k
obtained and represented by eq.(5) will be
Tanzania Journal of Engineering and Technology, (TJET) Vol. 1 (No. 2), Morch, 2007
with
p =k-2
3.3 Response Data Quantity
In order to facilitate the datum process as
prescribed above, the collection of a sufficient
quantity of data constitutes the first objective of
the test. The accuracy of identification depends
on the degree of indetenninacy of eq.(7), which
is given by the number of identification
equations, p. One must therefore respond .to the
basic question: How many response. s~gnals
will be required to produce a mmunum
number of p equations?
 L~dislaus Lwambuka J

The quantity of data procured at each Experience gained through the investigations
measuring station is given by of Badenhausen (1985) and Lwambuka (1988),
recommends that the rate of signal recording
should exceed at least 4 times the highest
(8) natural frequency of the vibrating system.
whereby From the model specifications of Fig.(l), the
mass and stiffness matrices can be derived to
I'1t = recording time interval (in seconds) obtain the form:
Tm = test duration (in seconds)

Mo
eo Mo eo
--+-
2
- --- 0
4 c 4 c2
M=
s
Mo eo
- - - - -1 ( M +M +
4 c2 4
1
b c 2 (e +e b )
0
O
J Mb
---
4
e
b
c2
Mb _ e b Mb e b
0 -+-
4 c2 4 c2

0 0]0
Ks =
l k
0
o
k
0 k

On setting up the experiment the following data were established:

(10)
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Mo =Mb = 1.376. 10-4 [kNms- 2 ]

c =0.093[m]
eo =e b =1.570.10- 7 [kNms- 2 ]
k = 0.288[kNm- 1 ]

for which the vibration modes are characterized by the three natural frequencies:

It = 7.53, 12 = 11.57, 13 = 14.48[HZ] (11)

with

(12)

If 1][ is defined to represent the ratio between the rate of recording, fA, and the highest natural
frequency of the system, 13' then the recommendations stated above implies that
. fA 1
17f mm=-= ~4 (13)
, f3 m.l1t.f
3

Tanzania Journal 0/ Engineering and Technology, (TJET) Vol. 1. (No.2), March, 2007
 Determination of Wind Loads Causing Flutter Effects
-- "" -- -_.. _--_...._......_-_._.. _..._....._- - --

A.recording time intervalqf say e)}ceed abo1,lt)O tiIf1.es; that o~ the pedod ofJhe
I1t = 0.002 sec lowest natural frequency. In our case, the
is arbitrarily selected. Hence with 13 and m =7 duration of the test is therefore obtained from
we obtain J;. as Tm = 1.328sec.
'7[ = 4.93.
For practical purposes, the author recommends
Observing that the vibrating system of fig.1 has
that the actual test duration T should be at least
three recording stations as portrayed in fig.(2),
double the analytical duration T",: :
I'::.t specifies a recording capacity requirement
of IA = 500 signals/sec on all the three T = 2Tm '.;. (14)
channels simultaneously.
The recommendation in eq.(14) allows for a
The duration of the test Tm can be obtained
free selection of the amplitude U 0 as reference
from recommendations on experiences set forth
for the analysis fig.4.
by Badenhausen (1985). In general, it should

I
I
A-
,I
I..
" IUo

~ ,
I
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

r,."
T

Fig.4: The desirable duration Tm and the actual duration T of the vibration test.

4. WIND LOAD PARAMETERS

After numerically processing the procured
response signals to form eq.(7), the
identifica tion of system parameters M, 12, K
follows directly by use of the time domain
methodology as described by Badenhausen in
1985 and Lwambuka in 1988. With eq.(4), wind
load parameter matrices M w, !2 w , Kw can be
extracted and eq.(l) becomes definitive. For
evaluation purposes, we seek a classical
presentation of the wind load parameters by
transforming eq.(l) into classical coordinates
U. A transformation matrix is defined as
follows:

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 1
Ladislaus Lwambuka

U
1

h.o 1 0'
1 1
o (15)
c c
fJ 1 2 1
c c c
so that
M =I- T MT- 1
D =T-T DT- 1 (16)
- - --

U=TU
- -- (17)

Introducing an arbitrary, dimensional factor matrix, Fy of the form

LPV 2C 0
Fy = 0 LpV 2 C 2 (18)
[ o 0
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Eq.(l) can be re-written in dimensionless wind load matrices M:,Dw,K: as well as dimensionless

response vectors * * [l * given

~,~, by eq.(19) below

(19)

The dimensionless response vectors of eq.(19) assume the form:

ch
v2
h h
2 •• - -
c a v c
'" v2 '" '"
U* -- U* -- ca
v
U* -- a (20)
2 ••
c fJ
v2
cfJ fJ
v

Tanzania Journal 0/ Engineering and Technalogy, (TJET) Vol. 1. (No.2), March, 2007
 Determination Wind Loads Causing Flutter Effects

wher,eas'Jhe dimenionless wind load parameters are presented as follows:

* "" ""
'"
Mwll M*w12 M*w13

;*=[: 0
1
0 n '"
M*w21

'"
M*w3\
""
M*w22

M*w32
""
M*w23

M*w33
(21)

[.~ '" '" ""

n
Dwll D* w12 D*w13
~ 0
"" '"
* (22)
D*== llJ D*w21 D* w22 D* w23
0
"" '" '"
D* w31 D* w32 D* w33
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

* '" '"

[ :.]
'"
Kwll K*w12 K*w13

- --
K*==
*
~ llJ
0

0
*
'"
K*w21 K* w22
'"
K*w23
(23)

"" '" ""

K*w31 K* w32 K* w33

The matrix coefficients in Equations (21), (22) and (23) contain a total number of 27 elements. These
are presented in this paper in Figures 5, 6 and 7 as real functions of the reduced frequency.

*
OJ = -
me (24)
v

Where

(i)
= frequency of the vibrating system
c = reference geometrical length
v = wind speed

Tanzania Jodrnal 0/ Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 1
Ladislaus Lw'!mbuka
------.-------- --_..._... _-_... __._---_._----_.-.-- ._ .. _ - - - - - - - - - - - - _... __. __._._-_._--_. _....... __ •._----_._----_.._..__ .....__. _ - - _ .__ ._.._ - - - - - - - - -

~ r~"
....•

/
30 /: _
20 -10 20

10 - 10 10
._~ __________<1. ____0.,-
o 10 20 30
0~====1~0~:==2=O:===;30; o 10 10

I~' /
r" >0
':'J ~ r"
_10
30

20

10
/
__________ 0 __ :::1
j
.. I'

10 20 30 10 20 lO
0
10 20
----=2..7
30 ..

'0
20

---- .- .
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

20 -20

to
~
10
f •
-10
_~_. ___ -_______G_ __0.
o 10 20 10 10 20 10
0
10 20 30 . '

experimental values
theoretical values

Fig.S: The Wind "Mass" Coefficients, M * wij

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1. (No.2), March, 2007
 Determination of Wind Loads Causing Flutter Effects

r~'>
-20

°l
-20

~
2

,, ; ........... ,
I
, " ,,,
, -
-10 /- ~_.
___ _
'.. ... .. ... ... ,
h
'0 20 10 10 20 30 o I~ 20 JO ,,'

0':" 0;.
,
,
,,
10 10

,, ,0 . I

, ,, ,"
I ·6

,,
. ,I
., , , I

-2

10 20 20
0 .' 10 20 30

_10 10 ••0

-8
,, "
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

-, I
-6
I
_L .L

-2
--- --- --- --- .2

Ol-~~~'O~----~20~--~3~0 30
o ~ ":"'-=---'~O------:2'::O---1:t-h ::

experimental values
theoretical va.lues

Fig.6: The wind II damping" coefficients, D * wij

Tanzania Journal of Engineering and Technology, (TJET) Val. 1 (No.2), March, 2007
 1
Ladislaus Lwambuka
.. _ - - - _. . . - - _ . _ . . .....-.... .- ......................... -

'L

EY
I '·
LO /./

/
10

10
,-
L-....::....~'O--~20--3....
0 ,,'

so
LO
-20

_IS lO

.. ~
20
-10
to
~----~-
10 20 ]0 to 30

10
_20
,t'
' :0
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

10

.
. •

,,
.
~

/ .. ___ ". Z..,

10 10

experimental values
theoretical values

Fig.7: The wind "stiffness" Coefficients, K * wi)

In Figures 5, 6 and 7, experimentally obtained
functions of wind load coefficients under this
method are being compared with real
theoretical values, given in APPENDIX I,
which are derived from the Theodorsen's
complex functions.
5. RESULTS
Good agreement between experimental and
theoretical results can be observed in Figs. 5, 6
and 7 for most of the coefficients. The
deviations stem from the discrepancy of the
test model from an ideal flat plate as well as the
approximated overlapping of elastic supports
and vertical coordinates (Figs. 1 and 2).
The accuracy of the method can further be
verified by treating the model suspension as a
real structure and solving the eigenvalue
problem using experimentally obtained wind

Tanzania Journal 0/ Engineering and Technology, (TJET) Vol. 1. (No.2), March, 2007
 Determination 0/ WindLoads Causing Flutter Effects
----------; --

load parameters. The solution presented in Vf = 9.8ml s,

fig.8 provides the critical wind speed of
which defines the onset of flutter instability.

60

, '~
30 , -
28 -------~

-,-,ZOql-:-S----T.-:,o::-o---~ --:-:15:-------:;2~O-----2~5~----v;;vlm/sl
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Fig.S: Determination of critical wind speed using experimentally

obtained wind load parameters.

Repeating the analysis with theoretical wind load parameters leads to solutions presented in Fig. (9)
with a critical wind speed of Vf = 11.7ml s,

Tanzania Journal 0/ Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 J
Ladislaus Llivambuka

J ______
to"l
eo

10

63 - •.• - ••. - - ....

I
60

50

s Ie IS 1Q 25 vl..v,1

r
"'(~l
.')
s

,.
!
------

'1/ , 2 :
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

10 20 lS v ",'.1

.J
Fig.9: Detennination of critical wind speed using theoretical
wind load parameters (Appendix I)

The actual wind speed observed in the wind
tunnel at which the model structure began to
flutter was
Vf :::; lO.3m/ s,
Since the critical wind speed obtained by use of
experimentally obtained wind loads is fairly
close to reality, the accuracy of the new method
would appear to be well justified.
experimental detennination of wind load
parameters.
2) In comparison with the conventional
method., the new method not only
reduces the technical requirement and
experimental work on wind tunnel tests,
but also increases the accuracy of results.

6. CONCLUSIONS 3) Although the method of approach

presented in this paper bears the
On the basis of experimental results presented advantages over the conventional one, as
in this paper the following conclusions can be pointed out under item 2 above, the data
drawn. processing effort is extremely high. It
requires the use of a high speed digital
1) The time domain identification technique computer and a set of special software for
. an advanced for d~ta processing.

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1. (No.2), March, 2007
 Determination of Wind Loads Causing Flutter Effects .

7. REFERENCES Lwambuka, L., (1988). Berechnung

Winderregter Schwingungen auf der
Badenhausen, K., (1985). Identification der Grundlage Experimenteller
Modellparameter Elastomechanischer P arametridentifikation, Disserta tion,
System aus Schwingungsversuchen. University of Kassel.
Dissertation, University of Kassel.
Scanlan, RH.; and Tomko, J.J., (1971).
Ibrahim, S. R, (1973). A time Domain Airfoil and Bridge Deck Flutter
Model Model Test Technique Shock and Derivatives Proc. of the American Society
Vibration Bulletin 43, part 4 of Civil Engineers, Journal of the
Engineering Mechanics Division.
Ibrahim, S. R, (1985). Advances in Time
Domain Identification and Modeling of Scanlan, R H., (1975). Theory of the Wind
Structures Second International Analysis of Long-Span Bridges Based on
Symposium on Aeroelasticity and Data obtained from Section Model Tests
Structural Dynamics Technical University Proc. 4th Intern. Conf. on Wind Effects on
Aachen. Buildings and Structures.

Ibrahim, S. R; and Mikulcik, E.C., (1976) Theodorsen, T., (1935). General Theory of
The Experimental Determination of Aerodynamic Instability and the
Vibration Parameters from Time Response Mechanism of Flutter National Advisory
Shock and Vibration Bulletin 46, P. 187 - Committee of Aeronautics Report No. 496
196. (Original).
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 1
Ladislaus Lwambuka

Appendix I: Theodorsen's Real Functions of Wind Load Parameters

The wind "mass" coefficients tum out to constant values of the form:

M*wll =-1(';"
.

MWI3
• =--
2
3
M: 21 = 0
* 1(
Mw22 =--
8
M' =-~
w23 16

•
M w31 =-32
M' =-~
w32 16
• 1
M w33 = - 1((- + - -12)
32 21(

The wind "damping" coefficients will read

Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

2 •
1(W 2mv' 2

Tanzania Jaurnal of Engineering and TectJnology, (TJfT) Vol. 1. (No.2), March, 2007
 Determination of Wind Loads Causing Flutter Effects

Whereas the wind "stiffness" coefficients take the form:

• nG
KwZl = --.
W

•
Kw22 = 7r{ W'F z - 2w' G}
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

with
F = The real part of Theodorsen - Function
G = The imaginary part of Theodorsen - Function

Appendix II: List of Symbols

* Notation for dimensionless parameter

c Reference geometrical length
k Spring constant
L Profile length
co Vibration frequency
F Real part of Theodorsen - Function
G Imaginary part of Theodorsen - Function

M* .. D* .. K* .. Wind Load Coefficients

WI)' WI)' WI)

U Displacement vector in classical coordinates h, a f3

U Displacement vector in vertical coordinates u 1, u z ,U 3
I Transformation matrix

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1 (No.2), March, 2007
 1
Ladislaus Lwambuka

MDK
-s,-s,-S System matrices based on U
MDK
_S,_S,_5 System matrices based on U
Fv ,.'.: Dimensional factor matrix

M D
-w,-w,-W K Wind load matrices based on U
Wind load matrices based on U
Reproduced by Sabinet Gateway under licence granted by the Publisher (dated 2011.)

Tanzania Journal of Engineering and Technology, (TJET) Vol. 1. (No.2), March, 2007