Documente Academic
Documente Profesional
Documente Cultură
_
_
_
Y (3)
Fig. 2. Cross-section of passive deckaps control system.
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 68
where m is the total mass of the system, S
a
is the rst-order moment of inertia of the
system about the deck elastic center, S
b
and S
g
are the rst-order moments of inertia
of the leading and trailing aps about their hinges, respectively, I
a
is the second-
order moment of inertia of the system about the deck elastic center, and I
b
and I
g
are
the second order moments of inertia of the leading and trailing aps about their
hinges, respectively. The damping matrix is assumed to be diagonal, i.e.,
C
s
= diag c
h
b
2
Y c
a
Y c
b
Y c
g
_ _
, with the entries corresponding to damping coecients
of respective degrees of freedom. The stiness matrix, K
s
, consists of two matrices,
K
s
= K
s1
K
s2
X (4)
K
s1
describes stiness of the deck-aps system without control cables, i.e.,
K
s1
= diag k
h
b
2
Y k
a
Y k
b
Y k
g
_ _
, where k
h
Y k
a
Y k
b
and k
g
are stiness coecients of
heaving and pitching motion, and leading and trailing ap connection, respectively.
Matrix K
s2
represents the stiness of the cable connections and supporting beam and
is found as
K
s2
= T
/
p
K
b
K
b
K
cc
( )
1
K
cc
T
p
(5)
where T
p
describes a transformation of displacements of the deck and aps, x, into
total vertical elastic displacements of the supporting beam and control cables,
T
p
=
0 r
cb
x
cb
r
cb
r
b
_ _
0
0 r
cg
x
cg
0 r
cg
r
g
_ _
X (6)
The stiness matrix of the supporting beam is
K
b
=
k
bbb
k
bbg
k
bbg
k
bgg
_ _
(7)
and K
cc
is the stiness matrix of the additional cables, which takes into account their
nonlinear properties, i.e., their lack of stiness in compression. K
cc
takes the
following values
K
cc
=
k
cb
0
0 k
cg
_ _
Y (8a)
K
cc
=
k
cb
0
0 0
_ _
Y (8b)
K
cc
=
0 0
0 k
cg
_ _
Y (8c)
K
cc
=
0 0
0 0
_ _
X (8d)
Stiness matrix (8a) is assumed when both cables are in tension, that of Eqs. (8b) or
(8c) when only the leading or trailing ap cable is in tension, and a matrix with zero
coecients (8d) when none of the cables is in tension. The condition for existence of
tensile force is checked by computing the sign of the summation of elastic
displacements of the supporting system, i.e., beam and additional cables, due to
vibration and prestressing.
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 69
The forces due to prestressing moments, M
b0
and M
g0
, are represented by vector
P
ps
. The forces P
ps
depend on the existence of tensile forces in the control cables. The
initial moments can be expressed in terms of initial ap displacements, b
0
and g
0
, and
stiness of the connections
M
b0
= k
b
b
0
Y (9a)
M
g0
= k
g
g
0
X (9b)
3. Modeling of aerodynamic forces
3.1. Formulation
A theoretical description of the self-excited aerodynamic forces of an oscillating
airplane wing was derived from potential ow theory by Theodorsen [9]. Theodorsen
and Garrick [10] extended this solution to characterize the nonstationary ow about
a wingailerontab combination. Both solutions describe the unsteady aerodynamic
forces due to steady-state oscillations in terms of the frequency-dependent
Theodorsen circulatory function. An extension of Theodorsen's theory to arbitrary
motions was presented by Edwards [11]. He introduced the generalized Theodorsen
function to describe the self-excited aerodynamic forces caused by arbitrary motion
of the wing.
Roger [12] proposed a modeling method which can transform the aeroelastic
equation of motion of an airplane into the frequency-independent time domain
equation. This method approximates aerodynamic force coecients by rational
functions of the Laplace variable. The size of the equation after approximation is
extended, but the overall analysis is greatly simplied. The augmented system, in
Roger's formulation, has a relatively large number of newly added aerodynamic
states, and modications of his method were proposed by Dunn [13] and Karpel [14].
The application of the rational function approximation for utter analysis of bridges
of various cross-sections was reported by Wilde et al. [15].
The vector of the self-excited forces, P, is selected as
P
/
= L
h
b M
a
M
b
M
g
_
Y (10)
where L
h
is a lift force, and M
a
M
b
and M
g
are torsional moments acting on
respective degrees of freedom.
The unsteady aerodynamic forces acting on a wingailerontab combination
(Fig. 3), according to Theodorsen and Garrick [10], may be represented as
F =M
a
q
U
b
_ _
C
a
q
U
b
_ _
2
K
a
q C s ( )
U
b
_ _
RS
2
q C s ( )
U
b
_ _
2
RS
1
qY (11)
where
q
/
= q
h
ab q
a
q
b
q
g
_
X (12)
Heaving and pitching motion with respect to the center of rotation of the wing are
denoted by q
h
and q
a
, whereas q
b
and q
g
are relative angles of rotation of the aileron
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 70
and tab, respectively. The corresponding vector of generalized forces is
F
/
= L
q
h
b M
q
a
M
q
b
M
q
g
_
X (13)
U is the mean velocity of the oncoming wind. The matrices M
a
, C
a
, K
a
, R, S
1
and S
2
depend on system geometry, namely the location of the wing rotation center, ap size
and location of hinges. The exact formulas can be derived from Theodorsen and
Garrick [10] and are not shown here due to their length and complexity.
Vectors describing the motion of the bridge deckaps system and wingaileron
tab combination are related through linear transformation
q = VxX (14)
Consistently, the relationship between the self-excited aerodynamic forces is given as
P = V
/
FX (15)
Thus, the formula for the unsteady aerodynamic forces of a bridge deck-aps system
may be written as
P = V
/
M
a
V x
U
b
_ _
V
/
C
a
V x
U
b
_ _
2
V
/
K
a
Vx C s ( )
U
b
_ _
V
/
RS
2
V x
C s ( )
U
b
_ _
2
V
/
RS
1
VxX (16)
The function C s ( ) appearing in Eqs. (11) and (16) is the generalized Theodorsen
function [11] which is expressed in terms of the Ha nkel functions, H
2 ( )
n
(n = 0Y 1), of
nondimensional Laplace variable, s, as
C s ( ) =
H
2 ( )
1
s ( )
H
2 ( )
1
s ( ) iH
2 ( )
0
s ( )
Y s = sbaUX (17)
Fig. 3. Aerodynamic forces on a wingailerontab system due to Theodorsen.
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 71
The rst three terms of the formula (11) represent lift and moment of a
noncirculatory origin, while the last two terms describe the forces due to vorticity in
the wake generated at the trailing edge.
The vector of the bueting forces, P
buf
, is
P
/
buf
= L
bufY h
b M
bufY a
M
bufY b
M
bufY g
_
Y (18)
where L
bufY h
is the lift force, and M
bufY a
, M
bufY b
, and M
bufY g
are torsional moments
acting on respective degrees of freedom. For a at plate, bueting forces due to
uctuations in longitudinal wind direction, u, may be neglected. Bueting forces due
to uctuations in the vertical direction, w, may be expressed as [16]
P
buf
=
U
b
_ _
2
V
/
R
_
t
+
0
w t ( )
U
dc t
+
t ( )
dt
+
dtY (19)
where t
+
= tUab is nondimensionalized time, and c is the Ku ssner function.
3.2. Rational function approximation (RFA)
The most common approximation of unsteady aerodynamic forces, used in
aeronautics, is based on rational functions [12,14]. Although these methods require a
rather complicated optimization procedure [17], they have proved to be very useful
for direct approximation of the tabular data of unsteady aerodynamics obtained by
experiment [15]. For the deckaps system, a theoretical formulation of unsteady
forces is available and the general approach results in unnecessary computational
burden. It will be shown that the minimum state RFA formulation of the deckaps
aerodynamics can be obtained from approximation of the generalized Theodorsen
function.
The generalized Theodorsen function is approximated by rational functions of the
nondimensionalized Laplace variable as
~
C s ( ) = a
0
n
l
i=1
a
i
s b
i
Y (20)
where
~
C s ( ) denotes approximation. The partial fractions, a
i
a s b
i
( ), are commonly
called lag terms, because each represents a transfer function in which the output lags
behind the input and permits an approximation of the time delays inherent in
unsteady aerodynamics. The coecients of the partial fractions, b
i
, are referred to as
lag coecients. Addition of each partial fraction introduces into the resulting state-
space realization new states referred to as aerodynamic states. The number of partial
fractions is denoted by n
l
, and is found as a compromise between the precision of the
approximation and the size of the state-space realization.
The Ha nkel functions, which are used to determine the generalized Theodorsen
function, have analytical continuation in the whole s-plane except the branch cut-o
along the negative real axis. The generalized Theodorsen function is in fact dened
over a two-dimensional domain and its approximation is a fairly dicult task.
However, if our interest is restricted only to purely imaginary values of the
nondimensionalized Laplace variable, s = ik, where k = obaU, the approximation
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 72
problem is greatly simplied. Since utter phenomena occur for points in the s-plane
which lie along the imaginary axis, this restriction may be justied. The
approximation formula for the generalized Theodorsen function with the argument
s = ik also has analytical continuation in almost all points in the s-plane. Thus, it is
supposed that slightly decaying or growing motions would be well approximated.
However, the approximation may fail in prediction of strongly damped motions.
The Laplace transformation of Eq. (16) with the approximated formula for the
Theodorsen function (20) and zero initial condition yields
L P ( ) = V
/
M
a
s
2
U
b
_ _
C
a
a
0
RS
2
( )s
U
b
_ _
2
K
a
a
0
RS
1
( )
_
U
b
_ _
2
R
n
l
i=1
a
i
s Uab ( )b
i
S
2
s
U
b
_ _
S
1
_ _
_
VL x ( )X (21)
The aerodynamic states are dened in Laplace domain as
L x
a
( ) = sI
U
b
_ _
R
b
_ _
1
U
b
_ _
EL x ( )X (22)
The matrices appearing in Eq. (22) have the forms
E = [b
1
Y F F F Y b
n
l
[
/
S
2
V [1Y F F F Y 1[
/
n
l
times
S
1
VX (23a)
and
R
b
= diag b
1
Y F F F Y b
n
l
( )X (23b)
As the parameter k varies in the range from 0 to , C k ( ) and
~
C k ( ) trace two
curves in the complex plane denoted by c k ( ) and
~
c k ( ), respectively. The objective of
the approximation procedure is to minimize the error of tting the exact curve with
the approximated one. This error can be evaluated as an integral
e =
_
B
A
D c k ( )Y
~
c k ( )
_ _
dc k ( ) [ [Y (24)
where D c k ( )Y
~
c k ( )
_ _
denotes distance between points of curves
~
c k ( ) and
~
c k ( )
corresponding to the same value of parameter k, and d
~
c k ( )
is the length of an
innitesimal arch of c k ( ). A and B are the terminal points of this part of the exact
curve which is approximated. Integral in Eq. (24) can be parameterized with respect
to k as
e =
_
k
up
k
low
C k ( ) C k ( ) [ [
dC k ( )
dk
dkY (25a)
where
dC k ( )
dk
= i
H
2 ( )
0
k ( )
_ _
2
H
2 ( )
1
k ( )H
2 ( )
0
k ( )ak H
2 ( )
1
k ( )
_ _
2
H
2 ( )
1
k ( ) iH
2 ( )
0
k ( )
_ _
2
X (25b)
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 73
The integration limits in Eq. (25a), k
low
and k
up
, are the lower and upper bounds of
the reduced frequency interval on which tting is performed. In this study
optimization is performed by the nonlinear technique proposed by Nelder and Mead.
The formulation of the bueting forces (19) in a form of dierential equations is
obtained through the approximation of the Ku ssner function after Dowell et al. [16]:
c t
+
( ) = 1 0X5e
0X13t+
0X5e
t
+
X (26)
Introducing the above approximation into Eq. (19) and taking Laplace transforma-
tion with zero initial condition yields
L P
buf
( ) =
U
b
_ _
2
V
/
R
0X065
s 0X13
0X5
s 1
_ _
L
w
U
_ _
X (27)
Gust states in the Laplace domain are dened as
L x
g
_ _
= sI R
g
_ _
1
E
g
L
w
U
_ _
Y (28)
where R
g
= diag 0X13Y 1 ( ) is the matrix of gust lag terms and E
/
g
= 1 1 [ [. Hence,
the bueting forces are given by
P
buf
= Q
buf
x
g
Y (29a)
where
Q
buf
=
U
b
_ _
2
V
/
R[0X065Y 0X5[ (29b)
and x
g
is a time domain counterpart of (28).
Taking the inverse Laplace transformation of Eqs. (21) and (22), and inserting
them together with Eqs. (29a) and (29b) into the equation of motion (1) yields a
state-space equation of motion
x
x
x
a
x
g
_
_
_
_
=
0 I 0 0
M
1
K M
1
C M
1
Q M
1
Q
buf
Uab ( )E 0 Uab ( )R
b
0
0 0 0 R
g
_
_
_
_
x
x
x
a
x
g
_
_
_
0
0
0
1aUE
g
_
_
_
_
w
0
M
1
0
0
_
_
_
_
P
ps
Y (30a)
where
M = M
s
V
/
M
a
VY (30b)
C = C
s
U
b
_ _
V
/
C
a
a
0
RS
2
( )VY (30c)
K = K
s
U
b
_ _
2
V
/
K
a
a
0
RS
1
n
l
i=1
a
i
RS
2
_ _
VY (30d)
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 74
Q =
U
b
_ _
2
V
/
R[a
1
Y F F F Y a
n
l
[X (30e)
The augmented state vector contains aerodynamic states, x
a
, and gust states, x
g
. In
this RFA formulation, addition of one lag term results in addition of only one new
aerodynamic state. The state-space equation of motion is nonlinear due to the
variable stiness of the cables supporting the aps.
3.3. Interpretation of aerodynamic states
For a better understanding of the rational model of unsteady aerodynamic forces,
the physical interpretation of aerodynamic states is provided herein. For the sake of
simplicity only two degrees of freedom of the deck, h and a, and one aerodynamic
state, x
a1
, are considered. Furthermore, the equation of motion is decomposed into
symmetric and skew-symmetric parts. The graphical representation of the symmetric
part of the equation of motion is shown in Fig. 4. The rotational degree of freedom a
is represented as a translation. The aerodynamic forces introduce coupling between
the heaving and pitching motion through the aerodynamic stiness and damping and
the aerodynamic state. The aerodynamic forces also modify the stiness and
damping that connect the heaving and pitching displacement to the supports. Since
the added aerodynamic stiness can be negative, the divergent type of the instability
can also be modeled. Notice that the aerodynamic state in this formulation
represents a displacement.
4. Simplied equation of motion
Due to vibrations caused by wind gust, the control cables can become slack. In
such a case not only will undesirable motion of the additional cables occur, but aps
Fig. 4. Representation of bridge model with RFA of aerodynamic forces.
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 75
will not be properly guided as well. Therefore, for practical reasons the control
cables should always remain taut. Numerical simulations of the response of the
bridge under wind action, conducted in Part II of this paper, showed that for
suciently large prestressing moments, M
b0
and M
g0
, and initial angles of rotation,
b
0
and g
0
, the supporting cables may always remain in tension. Moreover, if the
supporting system has adequately large stiness, the rotation of the aps can be
assumed to be proportional to the rotation of the deck:
b = t
b
aY (31a)
g = t
g
aY (31b)
where control gains, t
b
and t
g
, are determined from the geometry of the control
system as
t
b
=
r
cb
x
cb
r
cb
r
b
Y (32a)
t
g
=
r
cg
x
cg
r
cg
r
g
X (32b)
In order to validate the assumptions on ap motions (31), the relative errors
between the actual ap motion and the assumed one are evaluated. The errors are
dened as follows:
e
b
=
rms b t
b
a
_ _
rms t
b
a
_ _
1007Y (33a)
e
g
=
rms g t
g
a
_ _
rms t
g
a
_ _
1007Y (33b)
where rms() denotes the root-mean-square value, and a, b and g are time histories of
the solution of the nonlinear equation of motion (30).
The simplied governing equation of motion derived from Eq. (30a) becomes
z
z
z
a
z
g
_
_
_
_
=
0 I 0 0
M
1
c
K
c
M
1
c
C
c
M
1
c
Q
c
M
1
c
Q
bufY c
Uab ( )E
c
0 Uab ( )R
b
0
0 0 0 R
g
_
_
_
_
z
z
z
a
z
g
_
_
_
0
0
0
1aUE
g
_
_
_
_
wY (34a)
where
M
c
= M
dd
M
df
T T
/
M
fd
T
/
M
ff
TY (34b)
C
c
= C
dd
C
df
T T
/
C
fd
T
/
C
ff
TY (34c)
K
c
= K
dd
K
df
T T
/
K
fd
T
/
K
ff
TY (34d)
P. Omenzetter et al. / J. Wind Eng. Ind. Aerodyn. 87 (2000) 6179 76
Q
c
= Q
d
T
/
Q
f
Y (34e)
Q
bufY c
= Q
bufY d
T
/
Q
bufY f
Y (34f)
E
c
= E
d
E
f
TX (34g)
The mass, damping and stiness matrices, as well as Q
d
, Q
f
, Q
bufY d
, Q
bufY f
, E
d
and
E
f
are all obtained from the partition of the global system matrices with respect to
the degrees of freedom corresponding to the deck, x
/
d
= [hab a[, and aps, x
/
f
= [b g[.
The matrix T in Eqs. (34bg) denes the relationship between controlled system
displacement vector z
/
= [hab a[ and the vector of displacement of aps x
f
:
x
f
= Tz (35)
and is given by
T =
0 t
b
0 t
g
_ _
X (36)
5. Passive control system synthesis
In this paper, the considered control law given by Eq. (31), relates rotation of the
aps proportionally to the pitching motion of the deck. The objective of the
optimization of the passive system is to nd the optimal control gains. The primary
requirement for a passive control system is to provide the highest possible critical
wind speed, U
cr
, of the system. Thus, the rst proposed performance index evaluates
relative improvement of the critical wind speed.
Let the critical wind speeds of an uncontrolled and a controlled bridge deck be U
0
cr
and U
c
cr
, respectively. The control system performance index is dened as
I
1
=
U
c
cr
U
0
cr
U
0
cr
1007X (37)
This criterion proved to be adequate for systems with symmetric cable connections,
which in general provide limited improvement in critical wind speed. However,
systems with asymmetric cable connections turned out to be capable of improving
critical wind speed far beyond the practical requirements. Thus, for these systems,
comparison of control laws based merely on the maximum critical wind speed
criteria does not provide meaningful information. Therefore, the second of the
suggested performance indices assesses the system's degree of stability over the
selected wind velocity range.
The control system is modeled by an nn homogenous state-space equation of
motion given by