Вы находитесь на странице: 1из 38

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2004; 60:10111048 (DOI: 10.1002/nme.987)

Aeroelastic forces and dynamic response of long-span bridges

Massimiliano Lazzari1 , Renato V. Vitaliani1, , , and Anna V. Saetta2


1 Department of Construction and Transportation, via Marzolo 9, University of Padova, 35131 Italy
2 Department of Architectural Construction, Tolentini 191, IUAV - Venice, 30135 Italy

SUMMARY
In this paper a time domain approach for predicting the non-linear dynamic response of long-span
bridges is presented. In particular the method that leads to the formulation of aeroelastic and buffet-
ing forces in the time domain is illustrated in detail, where a recursive algorithm for the memory
terms integration is properly developed. Moreover in such an approach the forces expressions, usu-
ally formulated according to quasi-static theory, have been substituted by expressions including the
frequency-dependent characteristics. Such expressions of aeroelastic and buffeting forces are made ex-
plicit in the time domain by means of the convolution integral that involves the impulse functions and
the structural motion or the fluctuating velocities. A finite element model (FEM) has been developed
within the framework of geometrically non linear analysis, by using 3-d degenerated finite element. The
proposed procedure can be used to analyze both the flutter instability phenomenon and buffeting re-
sponse. Moreover, working in the geometrically non-linearity range, it verifies the possibility of strongly
flexible structures of actively resisting the wind loading. Copyright ! 2004 John Wiley & Sons, Ltd.

KEY WORDS: long-span bridges; aeroelastic forces, flutter analysis; buffeting analysis; geometrical
non-linearity; time-domain approach

1. INTRODUCTION

The long-span suspension bridge shows special characteristics such as high flexibility, low
structural damping and light weight, being thus very vulnerable to wind excitations. Generally,
the wind loads as buffeting (loads caused by the fluctuating part of the wind velocity) and self-
excited forces (loads caused by interaction between wind and structures motion) are functions
of the geometric configuration of the bridge sections, of the incoming wind fluctuations and
the reduced frequency. The wind forces with frequency-dependent aerodynamic characteristics
are generally described in terms of experimentally quantified flutter derivatives for the self-
excited forces and in terms of admittance and span-wise coherence functions for the buffeting

Correspondence to: R. V. Vitaliani, Department of Construction and Transportation, via Marzolo 9, University of
Padova, 35131 Italy.
E-mail: rvit@caronte.dic.unipd.it

Received 24 January 2003


Revised 9 June 2003
Copyright ! 2004 John Wiley & Sons, Ltd. Accepted 24 July 2003
1012 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

forces. Taking into account these unsteady characteristics of aerodynamic forces is essential for
their accurate evaluation and for the evaluation of the bridge response. There are two main
approaches to the problem of simulating structural response under constant or turbulent wind:
the analysis in frequency domain and the analysis in time domain.
In the first one, the input used for the simulation includes the statistical characteristics of
the wind, the wind density spectral matrix and the correlation functions of fluctuation wind,
while its output comes in the form of statistical quantities describing the structures motion.
Since the main hypothesis of the frequency analysis is the linearity behaviour analysis, such
an approach results inadequate for the final assessment of the bridge, due to the nonlinear
response of the wide-span structures.
The second approach performs the numerical integration of the structural motion equation
in the time domain, obtaining the bridge motion as output. Many papers consider quasi-
steady formulation of the action and consequently ignore the dependence of the aeroelastic and
buffeting terms on frequency (e.g. References [13]). A quasi-steady approach cannot always
be considered formally correct, however, especially when the reduced velocity is small: the
experimentally obtained aeroelastic derivatives are not used in the formulation. More recently,
it has been developed an analysis that implements the expression of the aeroelastic forces
by means of rational functions into the time domain (e.g. References [4, 5]). In this paper
an explicit analysis for the formulation of an aeroelastic finite element in time domain that
enables the analysis of both buffeting and flutter is presented. Such an analysis offers the great
advantage of considering the dependence of flutter and buffeting coefficients on frequency and
of enabling the geometrical non-linear behaviour of the structure to be taken into account. The
final expressions are in time domain, exploiting the representation of the forces by means of
the convolution integral.
It is worth noting that a non-linear aerodynamic force model should be used to take into
account the non-linear dependence of the aeroelastic forces on the effective angle of incidence.
In the proposed formulation the turbulence effects on the aerodynamic forces are included
according to the conventional linear load approaches (e.g. Reference [6]), by linearizing the
aerodynamic forces around the mean displacement position. Such a formulation agrees with
the forces model hypothesis that allows the representation of the forces by means of the
convolution integral. Within such an approach, it is assumed that there is no interaction between
the aeroelastic and buffeting forces. This circumstance is partially compensated by measuring
the forces coefficients under condition of turbulence, e.g. Reference [7].
However the accuracy of such an approach need further examination, especially in the
cases for which the nonlinear aerodynamics may become increasingly critical, e.g. when the
aerodynamic characteristics of innovative bridge deck designs, with attractive aerodynamic
performance, exhibit significant sensitivity with respect to the effective angle of incidence and
with the increases in the bridge span. (e.g. Reference [8]).
In summary this paper contains a review of the more common methods developed both in
frequency and time domain to model aerodynamic forces on bluff bridge decks. Moreover, it
discusses the relationship among the force descriptors for static, self-excited and buffeting force
components.
It is worth noting that, in order to provide a complete theory concerning both terms of
the wind load, i.e. aeroelastic and buffeting forces, in the paper the proposed formulation
transforms the expressions of the general wind load (aeroelastic and buffeting load) from the
frequency-time domain to time domain.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1013

The main aim of the proposed procedure, and consequently of proposed example, is to
highlight the capability and the potential difficulties of the aerodynamic problems solution
obtained by using a recursive formulation for the memory term. For that reason the aeroelastic
analysis of the Akashi Kaikyo bridge has been developed to find the flutter velocity out (in
this case is basic the coupling, described by aeroelastic derivatives, between wind load and
structural motion), for which the flutter derivatives are the only necessary data for the response
simulation.
Moreover, a comparison of the proposed approach with expressions fine-adjusted by other
authors in terms of theory and implementation algorithms are developed.

2. AERODYNAMIC FORCES

In the case of a bridge deck along-wind section with three degrees of freedom (h, vertical dis-
placement, p, lateral displacement and !, twist) it has become conventional (e.g. Reference [7])
to separate the associated time-dependent sectional force vectors into the static (s), aeroelastic
(ae) and buffeting (b) contributions per unit length of the deck, Figure 1 [4]:

Lift : L(t) = Ls (t) + Lae (t) + Lb (t) (9)

Drag : D(t) = Ds (t) + Dae (t) + Db (t) (10)

Moment : M(t) = Ms (t) + Mae (t) + Mb (t) (11)

The mean static components per unit length are expressed as:

2Ls (t)
= BCL (!s ) (12)
"U 2
2Ds (t)
= BCD (!s ) (13)
"U 2
2Ms (t)
= B 2 CM (!s ) (14)
"U 2

Figure 1. Aerodynamic forces on bridge section [4].

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1014 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

where " is the air density, B the width of the deck, U the mean wind speed and CL , CD ,
and CM are mean lift, drag and pitching moment coefficients; !s is the mean static angle
of attack of the bridge section and 21 "U 2 is the wind dynamic pressure.
Adopting the non-stationary theory, assuming a harmonic motion of the profile h(t) =
h 0 exp(i#t), p(t) = p 0 exp(i#t) and !(t) = ! 0 exp(i#t), the aeroelastic forces per unit of deck
can be expressed in terms of aerodynamic derivatives [9], Table I (1)(2), where k = #B/U =
2$nB/U = 2$/Ur is the reduced frequency, # is the circular frequency of vibration of the
bridge, (2) are aeroelastic derivatives depending on the frequency of the system (these aeroe-
lastic derivatives can be theoretically obtained for slender profiles, but can only be evaluated
experimentally for the bluff profile of bridges). The over-dot denotes the derivatives with re-
spect to time t. For slender profiles, the aeroelastic derivatives are related to the Theodorsen
function C(k) = F (k) + iG(k) (e.g. References [10, 11]).
The (1) represent the self-exciting forces expressed into frequency-time domain and - using
the Fourier transform (F [ ] is the Fourier transform operator)the (1) for the lift force can
be expressed as

! "
2F [Lse (t)] kH1 kH2 B k 2 H4 kH5 k 2 H6
=
F [h]+ 2
F [!]+k H3 F [!] + F [h]+ F [p]+
F [p]
"U 2 B U U B U B
(15)

= i#F [h], and similarly for drag and moment, can be written in
Equation (15), using F [h]
the compact form as follows:

F [Lse (t)] k 2 (H4 + iH1 ) k 2 (H6 + iH5 ) k 2 B(H3 + iH2 ) F [h(t)]

F [Dse (t)] = 1 "U 2
k 2 (P6 + iP5 ) k 2 (P4 + iP1 )

k 2 B(P3 + iP2 ) F [p(t)]
2
F [Mse (t)] k 2 B(A + iA ) k 2 B(A6 + iA5 ) 2 2
k B (A + iA ) F [!(t)]
4 1 3 2

(16)

In Reference [12] it is pointed out that there are redundancies in the classic analysis of aeroe-
lastic forces. Starting from the hypotheses that self-excited loads derive from an overlapping of
linear mechanisms, some authors (e.g. References [1214]) propose another straightforward nu-
merical model for self-excited forces in terms of the convolution integral between the structural
motion and the impulse response function; for the lift force it is
)
1 t * +
Lse (t) = "U 2 ILseh (t %)h(%) + ILsep (t %)p(%) + ILse! (t %)!(%) d% (17)
2

where the general expression IWsej (t %) is the impulse functions of the self-exciting forces W
(W = lift, drag and moment) relative to the displacement component j (j = h(t), p(t) and !(t)).
Equation (17) describes the self-excited lift force exclusively into time-domain, unlike (1)-(2).
The assessment of unsteady aerodynamic forces in time domain (17) requires identification of
aerodynamic impulse response functions, whose determination (for bluff bridge sections) shows

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
Copyright ! 2004 John Wiley & Sons, Ltd.

Table I. Self-exciting and buffeting loads into frequency and time domain.
Frequencytime
Wind loads domain Time domain

Self-excited load

a0,11 h(t) + a1,11 (B/U )h(t)
2Lseh (t) h h

AEROELASTIC FORCES AND DYNAMIC RESPONSE


Lift BkH1 + Bk 2 H4 H1 , H4 ,
m+2
a0,11 , a1,11 , a2,11 , ag,11 , g = 3, m + 2
"U 2 U B +
+a2,11 (B/U )2 h(t) ! g,11 (t)
g=3

a0,13 B!(t) + a1,13 B(B/U )!(t)


2Lse! (t) B !
BkH2 + Bk 2 H3 ! H2 , H3 ,
m+2
a0,13 , a1,13 , a2,13 , ag,13 , g = 3, m + 2
"U 2 U +a2,13 B(B/U )2 ! (t) + ! g,13 (t)
g=3

a0,12 p(t) + a1,12 (B/U )p(t)



2Lsep (t) p p
BkH5 + Bk 2 H6 H5 , H6 ,
m+2
a0,12 , a1,12 , a2,12 , ag,12 , g = 3, m + 2
"U 2 U B +a2,12 (B/U )2 p(t)
+ ! g,12 (t)
g=3


a0,21 h(t) + a1,21 (B/U )h(t)
2Dseh (t) h h
Drag BkP5 + Bk 2 P6 P5 , P6 ,
m+2
a0,21 , a1,21 , a2,21 , ag,21 , g = 3, m + 2
+
+a2,21 (B/U )2 h(t) ! g,21 (t)
Int. J. Numer. Meth. Engng 2004; 60:10111048

"U 2 U B
g=3

a0,23 B!(t) + a1,23 B(B/U )!(t)


2Dse! (t) B !
(1) BkP2 + Bk 2 P3 ! (2) P2 , P3 (3) ,
m+2
(4) a0,23 , a1,23 , a2,23 , ag,23 , g = 3, m + 2
"U 2 U +a2,23 B(B/U )2 ! (t) + ! g,23 (t)
g=3

a0,22 p(t) + a1,22 (B/U )p(t)



2Dsep (t) p p
BkP1 + Bk 2 P4 P1 , P4 ,
m+2
a0,22 , a1,22 , a2,22 , ag,22 , g = 3, m + 2
"U 2 U b
2
+a2,22 (B/U ) p(t) + ! g,22 (t)
g=3


a0,21 Bh(t) + a1,21 B(B/U )h(t)
2Mseh (t) h h
Moment B 2 kA1 + B 2 k 2 A4 A1 , A4 ,
m+2
a0,31 , a1,31 , a2,31 , ag,31 , g = 3, m + 2
"U 2 U B
2
+a2,21 B(B/U ) h(t) + ! g,21 (t)
g=3

1015
1016

Table I. Continued.
Frequencytime
Wind loads domain Time domain
B
U
a0,23 B 2 !(t) + a1,23 B 2 ! (t)
2Mse! (t)
m+2
B !
B 2 kA2 A2 , A3 a0,32 , a1,32 , a2,32 , ag,32 , g=3, m+2
U
+ B 2 k 2 A3 !
"U 2 +a2,23 B 2 (B/U )2 ! (t) + ! g,23 (t)
g=3
,

B
p(t)
2Msep (t) p U
a0,22 Bp(t) + a1,22 B
m+2
p
B 2 kA5 A5 , A6 a0,33 , a1,33 , a2,33 , ag,33 , g=3, m+2
U B
+ B 2 k 2 A6
"U 2 +a2,22 B(B/U )2 p(t)
+ ! g,22 (t)
g=3

Copyright ! 2004 John Wiley & Sons, Ltd.


,

Buffeting loads
2Lbu (t) u(t) m
Lift &Lbu (2CL B) g=1 " g,11 (t)
dg U
U B
(2CL B)&Lbu e0,11 + Ag,11 m e0,11 , eg,11 , g = 1, m
"U 2 g=1
! ! " "
, ,

w(t) w(t) m m d U
2Lbw (t) g
&Lbw " g,12 (t)
U U
(CL B + CD B)&Lbw (CL B + CD B) e0,12 + Ag,12 e0,12 , eg,12 , g = 1, m
"U 2 g=1 g=1 B
! ! " "
, ,

u(t) u(t) m m d U
2Dbu (t) g
Drag (2CD B)&Dbu &Dbu (2CD B) "g,21 (t)

U U
e0,21 + Ag,21 e0,21 , eg,21 , g = 1, m
"U 2 g=1 g=1 B
! ! " "
, ,

w(t) w(t) m m d U
2Dbw (t) g
(5) (CD B)&Dbw (6) &Dbw
(7) (CD B) "g,22 (t)
(8)
U U
e0,22 + Ag,22 e0,22 , eg,22 , g = 1, m
"U 2 g=1 g=1 B
! ! " "
M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

, ,

u(t) u(t) m m d U
2Mbu (t) 2 g
Moment (2CM B 2 )&Mbu &Mbu (2CM B ) "g,21 (t)

U U
e0,31 + Ag,31 e0,31 , eg,31 , g = 1, m
"U 2 g=1 g=1 B
! ! " "
, ,

w(t) m m d U
2Mbw (t) g
B 2 )&
(CM Mbw &Mbw (CM B 2 ) w(t) e
0,32 A g,32 " g,22 (t)
U U
+ e0,32 , eg,32 , g = 1, m
"U 2 g=1 g=1 B
! ! " "
, ,

Int. J. Numer. Meth. Engng 2004; 60:10111048


AEROELASTIC FORCES AND DYNAMIC RESPONSE 1017

still unsolved difficulties. On the contrary, the technique for identifying the frequency domain
force parameters (flutter derivatives and admittance functions) has been fully established, and
a large data set is available now for a host of geometric configurations of bridge sections (e.g.
References [1416]).
In the same way as in (1)(15), the Fourier transform of (17) is

2F [Lse (t)]
= F [ILseh h + ILsep p + ILse! !] = F [ILseh ]F [h] + F [ILsep ]F [p] + F [ILse! ]F [!]
"U 2
(18)

The compact form of (18) for lift, drag e moment is



F [Lse (t)] F [ILseh ] F [ILsep ] F [ILse! ] F [h(t)]

F [Dse (t)] = 1 "U 2 F [IDseh ] F [IDsep ] F [IDse! ]
2 F [p(t)] (19)
F [Mse (t)] F [IMseh ] F [IMsep ] F [IMse! ] F [!(t)]

Comparing the two expressions (16) and (19) the following equalities are given, which represent
the relationship between the aerodynamic impulse function and flutter derivatives:

F [ILseh ] = k 2 (H4 + iH1 ) = ILseh

F [ILsep ] = k 2 (H6 + iH5 ) = ILsep (20)

F [ILse! ] = Bk 2 (H3 + iH2 ) = B ILse!

F [IDseh ] = k 2 (P6 + iP5 ) = ID seh

F [IDsep ] = k 2 (P4 + iP1 ) = ID sep (21)

F [IDse! ] = Bk 2 (P3 + iP2 ) = B ID se!

F [IMseh ] = Bk 2 (A4 + iA1 ) = B IM



seh

F [IMsep ] = Bk 2 (A6 + iA5 ) = B IM



sep
(22)

F [IMse! ] = B 2 k 2 (A3 + iA2 ) = B 2 IM



se!

The flutter derivatives are usually known only at discrete values of reduced frequency k.
Since the inverse Laplace transform can hardly be used to quantify the impulse function, it
is necessary to have approximate continuous functions of the reduced frequency to describe
frequency-dependent force parameters (e.g. References [17, 18]). For the self-excited forces,
the most common form for approximating the coefficients of aerodynamic force (22) is the

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1018 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

rational functions approximation knows as Rogers approximation [19, 20]:

,
m+2 ik
ILseh (i#) = a0,11 + a1,11 (ik) + a2,11 (ik)2 + ag,11
g=3 ik + dg
- . - .2
i#B i#B ,
m+2 i#
= a0,11 + a1,11 + a2,11 + ag,11 dg U (23)
U U g=3 i# + B
- . - .2 ! "
i#B i#B ,
m+2 ag,11 4$2 ag,11 2$dg Ur
= a0,11 + a1,11 + a2,11 + 2 2 2
+i 2 2
U U g=3 dg Ur + 4$ dg Ur + 4$2

where aa1,a2a3 is coefficient approximation in which a1 is linked to the term order of the
approximation function (e.g. a1 = 0, 1, . . . , 6), a2 is the index of the aerodynamic forces
(e.g. a2 = 1 = lift, 2 = drag, 3 = moment) and a3 are index of the degree of freedom (e.g.
a3 = 1 = h(t), 2 = p(t), 3 = !(t)). The coefficient aa1,a2a3 (a1 = 0 6, a2 = 1 3 and
a3 = 1 3) and dg (g = 3 m + 2) are frequency-independent coefficients; the first and second
terms of (23) are respectively the non-circulatory static-aerodynamic and the aerodynamic
damping; the third term is the additional aerodynamic mass which is normally negligible; and
the rational terms are the unsteady components which lag the velocity of body motion and
permit an approximation of time delays through positive values of parameter dg . The value of
m determines the level of accuracy of this approximation and the size of additional equations
representing the aerodynamic state.
The two real and the two imaginary parts of (23) and (20) are compared for h d.o.f.:

4$2 ,
m+2 4$2
k 2 H4 = a0,11 2
a2,11 + a1,11 2 2
Ur g=3 dg Ur + 4$2
Real: (24)
U2 ,
m+2 U2
H4 = a0,11 r2 a2,11 + a1,11 2 2 r
4$ g=3 dg Ur + 4$2

2$ m+2
, 2$d1 Ur
k 2 H1 = a1,11 + a1,11 2 2
Ur l=3 d1 Ur + 4$2
Imaginary: (25)
,
Ur m+2 dg Ur3
H1 = a1,11 + a1,11
2$ g=3 2$(dg2 Ur2 + 4$2 )

In the same way, the approximation is applied to the other flutter derivatives and for Lift force
it results:

Ur2 ,
m+2 Ur2
Lift-! d.o.f.: H3 = a0,13 a2,13 + ag,13 (26)
4$2 g=3 dg2 Ur2 + 4$2

,
Ur m+2 dg Ur3
H2 = a1,13 + a1,13 (27)
2$ g=3 2$(dg2 Ur2 + 4$2 )

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1019

Ur2 ,
m+2 Ur2
Lift-p d.o.f.: H6 = a0,12 a2,12 + ag,12 (28)
4$2 g=3 dg2 Ur2 + 4$2

,
Ur m+2 dg Ur3
H5 = a1,12 + ag,12 (29)
2$ g=3 2$(dg Ur2 + 4$2 )
2

All the components, which are approximated by rational functions, are written in the following
compact form:
2

k (H4 + iH1 ) (a0,11 a1,11 a2,11 a3,11 a4,11 a5,11 a6,11 )


Lift: Lse (t) k 2 (H6 + iH5 ) (a0,12 a1,12 a2,12 a3,12 a4,12 a5,12 a6,12 )



2
Bk (H3 + iH2 ) B(a0,13 a1,13 a2,13 a3,13 a4,13 a5,13 a6,13 )
2

k (P6 + iP5 ) (a0,21 a1,21 a2,21 a3,21 a4,21 a5,21 a6,21 )


Drag: Dse (t) k 2 (P4 + iP1 ) (a0,22 a1,22 a2,22 a3,22 a4,22 a5,22 a6,22 ) dg , g = 3 6



2 (30)
Bk (P3 + iP2 ) B(a0,23 a1,23 a2,23 a3,23 a4,23 a5,23 a6,23 )
2

Bk (A4 + iA1 ) B(a0,31 a1,31 a2,31 a3,31 a4,31 a5,31 a6,31 )


Moment: Mse (t) Bk 2 (A6 + iA5 ) B(a0,32 a1,32 a2,32 a3,32 a4,32 a5,32 a6,32 )



2 2
B k (A3 + iA2 ) B 2 (a0,33 a1,33 a2,33 a3,33 a4,33 a5,33 a6,33 )

Speaking about the coefficients, the first (a1 = 0) and third (a3 = 0) terms are function only of
the real part of the pair of aeroelastic derivatives (e.g. for lift force H4 , H3 and H6 ), whereas
the second term (a2 = 0) is function of the imaginary components (e.g. for lift force H1 , H2
and H6 ). These three terms are strongly linked with the physical phenomena. In particular,
as above-described, the second term represents the aerodynamic damping and it is function of
H1 which, into the Scanlan formulation, links the lift force with the vertical velocity of the

bridge (h(t)). The last rational terms of the (23), as its shown by (24) and (25), approximates
at same time the real and the imaginary part of the pair of the aeroelastic derivatives.
The coefficients a0,nq , a1,nq , a2,nq , ag,nq , g = 3, m + 2 can be evaluated by approximation
to the linear or nonlinear least-squares fit methods using the experimentally obtained flutter
derivatives and (24)-(25). In the following simulation the four coefficients dg (g = 3 6) are
assumed constant for all the terms, but such coefficients could be easily considered as variable
values.
The approximation is done only for the complex pairs IW
sej
(W = lift, drag and moment;
j = p(t), h(t) and !(t)) so the multiplier term 1, B, B 2 , B 3 and B 4 must be considered in
the final relations.
After the approximation, the coefficients can be described by continuity function of reduced
velocity so its possible to come back in time domain by an inverse transform of (20)(22).
Changing the base s = i#, the former approximating rational function of the aeroelastic
transfer function (23) can be extended into the Laplace domain. The inverse of the Laplace
transform of (23), with the mathematics change s/(s +r) = 1r/(s +r), yields the aerodynamic

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1020 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

impulse function:

ILseh (t)
3 - . - .2 m+2 - .4
sB sB , dg U/B
= L1 [ILseh (s)] = L1 a0,11 +a1,11 +a2,11 + ag,11 1
U U g=3 s+(dg U/B)
- .2
B B
= a0,11 '(t) + a1,11 '(t) + a2,11 '(t)
U U
- .
,
m+2 dg U ((dg U/B)t)
+ ag,11 '(t) e (31)
g=3 B

where '(t) is the Dirac function and the over-dots indicates the derivatives with respect of
the time. Substituting (31) in (17), considering the last term in the expression (31) as !f/!%
and proceeding with a integration by parts, the following expression 2Lseh (t)/"U 2 can be
obtained:

- .2
2Lseh (t) B B
= a0,11 h(t) + a1,11 h(t) + a2,11 h(t)
"U 2 U U
- .
) t dg U
,
m+2 dg U B (t%)
+ ag,11 h(t) ag e h(%) d%
g=3 B

- .2
B B
= a0,11 h(t) + a1,11 h(t) + a2,11 h(t)
U U
- .
) t dg U
,
m+2 B (t%)
+ ag,11 e d%
h(%) (32)
g=3

In compact form, (32) can be re-written as (3) and (4), where it has been assumed that:
) t dg U
!L,gh (t) = ag,11 e B (t%) h(%)
d% (33)

Equation (33) represents the memory of the system which introduces the non-conservative loads
(the principal aspect of the aeroelastic forces) into the time domain analysis. In this case the
forces applied to bridge deck section are functions of the previous story of the motions. So

the !L,gh (t) needs the integration of the h(%) between [ t] for all time steps of simulation
analysis.
Similarly, the previous equations could be written including the contributions of p and ! to
the lift, drag and moment forces. In such a way, the lift, drag and moment forces (1), usually

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1021

given in terms of time and aeroelastic derivatives (2) (i.e. in terms of reduced frequency) are
expressed using approximation coefficients exclusively in time domain (3)(4).
The buffeting forces per unit span (5)-(6) are expressed in terms of steady average lift,
drag and moment force coefficients CL = !CL /!!, CD = !CD /!!, CM = !C /!!. A is the
M n
projected area of the deck, normal to wind direction; u(t) and w(t) are along-wind and vertical
gust velocity components of the wind, and (6) is the aerodynamic admittance transfer function
between fluctuation wind velocity and buffeting forces (which are dependent on frequency and
deck configuration) (e.g. References [7, 8, 14]).
In the same way as in (2)(1) the buffeting forces (6)(5) can be carried into the time
domain by the convolution integral between the aerodynamic impulse response function and
the fluctuating wind velocity; for lift, the force is
) - .
1 t u(%) w(%)
Lb (t) = "U 2 ILbu (t %) + ILbw (t %) d% (34)
2 U U

where IWbj (t %) are the aerodynamic impulse functions of the buffeting forces W (W = lift,
drag and moment) relative to the turbulent wind velocity component j (j = u(t) and w(t)).
Equation (34) describes the buffeting force exclusively into the time-domain unlike (5)(6).
Comparing the two Fourier transform expressions of the (5) and (34), it results the following
equalities that represent the relationships between the impulsive function and the aerodynamic
transfer function:

F [ILbu ] = 2CL B&Lbu = 2CL B ILbu (i#)


Lift: (35)
F [ILbw ] = (CL + CD )B&Lbw = (CL B + CD B)ILbw (i#)

F [IDbu ] = 2CD B&Dbu = 2CD B ID bu (i#)


Drag: (36)
F [IDbw ] = CD B&Dbw = CD B ID bw (i#)

F [IMbu ] = 2CM B 2 &Mbu = 2CM B 2 IM



bu
(i#)
Moment: (37)

F [IMbw ] = CM
B 2 &Mbw = CM B 2 IM

bw
(i#)

The admittance functions as flutter derivatives are normally known only at discrete values of
reduced frequency k, so a continuous approximation function of ILbu (i#), ILbw (i#), ID bu (i#),
ID bw (i#), IM
(i#) and I (i#) is looked for into (35)(37) by rational functions approxima-
bu Mbw
tion. Equation (38) is the rational approximation for lift buffeting force about the w component
of the turbulence wind, whose coefficients e0 , eg with g = 1, m can be determined by linear
or non-linear least square fit of the coefficient &Lbw (6):

! "
,
m i# ,m eg,12 4$2 eg,12 2$dg Ur
ILbw (i#) = e0,12 + eg,12 = e0,12 + +i 2 2
g=1 i# + (dg U/B) g=1 dg2 Ur2 + 4$2 dg Ur + 4$2
(38)

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1022 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

All the components are written in the following compact form:


&Lbu = ILbu (i#) (e0,11 e1,11 e2,11 e3,11 e4,11 )


Lift: Lse (t)
&Lbw = ILbw (i#) (e0,12 e1,12 e2,12 e3,12 e4,12 )

9
&Dbu = ID bu (i#) (e0,21 e1,21 e2,21 e3,21 e4,21 )
Drag: Dse (t) dg with g = 3 6
&Dbw = ID bw (i#) (e0,22 e1,22 e2,22 e3,22 e4,22 )
9
&Mbu = IM

bu
(i#) (e0,31 e1,31 e2,31 e3,31 e4,31 )
Moment: Mse (t)
&Mbw = IM

bw
(i#) (e0,32 e1,32 e2,32 e3,32 e4,32 )
(39)

The coefficients dg , g = 1, m are constant coefficients and they can be the same for the
coefficients of (23). Changing the base s = i# the (38) is brought into the Laplace domain,
and developing the inverse of the Laplace transform it comes out:

ILbw (s) = (CL B + CD B)ILbw (s)


! - - .."
,
m d U d U
= (CL B + CD B) e0,12 '(t) +
g g
eg,12 '(t) exp t (40)
g=1 B B

Substituting (40) in (34), the term 2Lbw (t)/"U 2 is obtained. Considering the last term in
expression (40) as !f/!% and proceeding with a integration by parts, the relation becomes as
(7)-(8); here the memory term is
) t dg U w(%)
"bw (t) = eg,12 e B (t%) d% (41)
U

In this term the integral involves the time history of the wind turbulence velocity w(%)
and the mean wind velocity U . Equation (41) is not connected with the motions of the
system and this relation clearly shows the difference with the aeroelastic memory
term (33).
The variation of the turbulence parameters (e.g. integral scales, cross-spectral function, cross-
coherence function) has a consequence on the wind velocity simulation and in particular in the
expressions of buffeting forces develop in time domain by Equations (7), (8) and (41). It is
worth noting that the wind turbulence velocities, u and w components, can be obtained by a
simulations algorithms (e.g. wave superimposition or ARauto regressive and ARMAauto
regressive moving average) discussed in a previous paper by the same authors [21]. The above
observation is congruent with forces model hypothesis that permits to write the integrals (17)
and (34).

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1023

3. THE INDICIAL FUNCTION

In classic wing profile theory, the aeroelastic forces can be written as two indicial functions,
as suggested by Reference [22] for lift force:
) )
1 2 t 1 2 t d%
Lseh (t) = "U #Lseh (t %)h(%) d% + "U #Lseh (t %)h(%) (42)
2 2

Equation (42) was developed from wing profiles and applied to bridges [12]. In Reference [13]
Equation (42) is described as a redundant relation, given that it incorporates the information
provided by h(%) and h(%) in the interval [, t]. That is why representation (17), based
on the impulses rather than on the indicial function, is taken as the starting point to describe
aeroelastic forces in the time domain. Equation (42) can be connected to (4) and (33), developing
the latter relations of (42). In particular, considering a rational approximation that involves
a0,11 , a1,11 , ag,11 , g = 3, m + 2 and not a2 , (4) can be written as

)
2Lseh (t) B ,
m+2 t dg U
= a1,11 h(t) + (a0,11 + ag,11 e d%
B (t%) )h(%)
"U 2 U g=3
) t
B d%
= a1,11 h(t) + #Lseh (t %)h(%) (43)
U

where the indicial functionsassuming a rational approximation of the complex pairs of trans-
, dg U
B (t%)
formed aeroelastic derivatives (23)becomes #Lseh (t) = a0,11 + m+2 g=3 ag,11 e and
:t

the term a1,11 (B/U )h(t) represents the integral of the second indicial function #Lseh (t
d% introduced by (42):
%)h(%)
) t
2Lseh (t) B d%
= a0,11 h(t) + a1,11 h(t) + (#Lseh (t %) a0,11 )h(%) (44)
"U 2 U

Similarly, all the other contributions can be expressed starting from the indicial functions
#Wsej (t) (W = lift, drag e moment and j = h(t), p(t) and !(t)). #IseJ (t) can be described by
means of different indicial functions whose coefficients are generally estimated by nonlinear
fitting with the aeroelastic derivatives (23). One way to illustrate the indicial functions is the
wing profile theory:

,
m+2 2
#Lseh (t) = a0,11 + ag,11 edg ( s
(45)
g=3

where ( = Prandtl-Glauert = 1 M 2 and M is the Mach number, where ( is usually assumed
equal to 1; in this case (45) coincides with (43) to approximate the Wagner functions (e.g.
References [23, 6, 24]).
Other time domain expressionswhich include only the dependence on the displacement,
acceleration and rotation velocityare based on indicial functions; thus, for lift and drag forces

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1024 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

and pitched moment they are [8]:


) t - .
2Lse (t) h(%)
p(%)
= B (C L + CD )#Lseh (t %) 2C #
L Lsep (t %) d%
"U 2 U U
) t
B ((CL + CD )#Lse! (t %)!(%)) d%

) - .
2Dse (t) t h(%)
p(%)
=B (CD CL )#Dseh (t %) 2CD #Dsep (t %) d%
"U 2 U U
) t (46)
+B ((CD CL )#Dse! (t %)!(%)) d%

2Mse (t)
) t -
h(%)
p(%)
.
= B2
CM #Mseh (t %) 2CM #Msep (t %) d%
"U 2 U U
) t
+ B2
(CM #Mse! (t %)!(%)) d%

This relation (46) can be linked with the impulsive function (17). Proceeding with a integration

by parts and with the definition of '(t) ('(t)) and neglecting initial conditions of motion, for
lift force we can write
B +# Lseh (0)'(t) + # Lseh (t))
ILseh (t) = (C + CD )(#Lseh (0)'(t)
U L
B +# Lsep (0)'(t) + # Lsep (t))
ILsep (t) = (2CL )(#Lsep (0)'(t) (47)
U
ILse! (t) = B(CL + CD )(#Lse! (0)'(t) + # Lse! (t))

whereas developing the Fourier transform of (46) its possible to join the Fourier transform of
the impulsive functions, the indicial function and the pair of flutter derivatives.
Impulsive func. Indicial func. Flutter derivatives
B
F [ILseh ] (C + CD )(i#)2 F [#Lseh ] k 2 (H4 + iH1 )
U L (48)
B
F [ILsep ] 2 CL (i#)2 F [#Lsep ] k 2 (H6 + iH5 )
U
F [ILse! ] B(CL + CD )(i#)F [#Lse! ] Bk 2 (H3 + iH2 )
In the same way as it is done in (46)(48), its possible to write the Buffeting forces by indicial
functions which are linked with impulsive and admittance functions by the following relation:

) t - .
1 2
u(%)
w(%)
Lb (t) = "U B (2CL )#Lbu (t %) + (CL + CD )#Lbw (t %) d%
2 U U
) t - .
1 2
u(%)
w(%)
Db (t) = "U B (2CD )#Dbu (t %) + (CD )#Dbw (t %) d% (49)
2 U U

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1025

) - .
1 t
u(%)
w(%)
Mb (t) = "U 2 B 2 (2CM )#Mbu (t %)
+ (CM )#Mbw (t %) d%
2 U U
B Lbu (t))
ILbu (t) = (2CL )(#Lbu (0)'(t) + #
U
B Lbw (t))
ILbw (t) = (C + CD )(#Lbw (0)'(t)w + # (50)
U L

Impulsive func. Indicial func. Admittance func.


F [ILbu ] (2CL )BF [#Lbu ] (2CL )B&Lbu = (2CL )B ILbu (i#) (51)
F [ILbw ] (CL + CD )B[#Lbw ] (CL + CD )B&Lbw = (CL B + CD B)ILbw (i#)

4. MEMORY TERMS OF AEROELASTIC FORCES

The memory terms (33) and (41) represent the memory of the phenomenon and their evaluation

at the instant t depends on the temporal history of h(%) and w(%) for aeroelastic (non-stationary
force) and buffeting forces. A strict calculation of the integral (33) and (41) calls for the
use of all the previously-calculated variables and thus involves a great deal of memory and
processing, that considerably increases the computational effort of the analysis [18]. Considering
the Theorem of derivation under the integral sign and applying it to (33) and (41), the
result is

dg U
aeroelastic forces: ! L,gh (t) =
!L,gh (t) + ag,11 h(t) (52)
B

bw (t) = dg U "bw (t) + eg,12 w(t)


buffeting forces: " (53)
B U

The relations in Table I (3) and (52) for aeroelastic forces and in Table I (7) and (53) for
buffeting forces are consistent with the expressions obtained by Reference [14]. Equations (52)
and (53) define the !L,gh (t) and "bw (t) as augmented aerodynamic state vector (additional
variables).
The following part of this paragraph develops the same consideration about the memory
term (33) of the aeroelastic forces because it is the most important term. The importance of
the argument of integral (33) or (52) rapidly declines with the damping due to the exponential
term; the damping rate depends on the power of the exponential term. This finding enables
the integration limits to be changed from [, t] to [tlimit , t], where tlimit can be defined as
follows [25, 26]:
dg U dg U
% [, t] and t % > tlimit e B (t+$t%) e B (t%) 0 (54)
In a numerical analysis, in time domain, the force increment of the term relating to the sys-
tems memory is fundamental. The variation of (52) can consequently be written by means

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1026 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

of (54) as

) t - dg U dg U .
d!L,gh = !L,gh (t + $t) !L,gh (t)
= ag,11 e B (t+$t%)
e B (t%)
h(%) d%
tlimit
) t+$t dg U
+ ag,11 e B (t+$t%) h(%)
d% (55)
t

In the evaluation of (55) for each instant t analysed, it must be calculated the memory integral
in the interval [tlimit , t]. The burden of calculations is considerable, since tlimit
= 20% of t for
a good-quality integration.
In order to obtain a recursive relation for calculating the memory term of the phenomenon
(e.g. References [18, 5]), (33) is developed in the following way t = tj , tj = tj 1 + $t,
[, tj ] = [, tj 1 ] + [tj 1 , tj ]:
) tj 1
!L,gh (tj ) = e(dg U/B)$t d%
ag,11 edg U/B(tj 1 %) h(%)

) tj 1 +$t dg U
+ ag,11 e B (tj 1 +$t%) h(%)
d%
tj 1
- .
(56)
dg U dg U (tj 1 +$t+tj 1 )
B tj 1 +$t
(t t )
= e B j j 1 !L,gh (tj 1 ) + ag,11 e
2

- .
tj 1 + $t + tj 1
h $t
2

dg U dg U - .
$t
!L,gh (tj )
= ! L,gh (tj 1 ) = e
B (tj tj 1 ) !L,gh (tj 1 ) + ag,11 e 2B (tj tj 1 ) h
tj 1 + $t
2
(57)

From Equations (56)(57), it can be seen that only the quantities involving ! L,gh (tj 1 ) and
j 1 + $t/2)$t at time tj 1 + $t/2 are needed to be stored for evaluating ! L,gh (tj ). In this
h(t
way !L,gh (tj ), Equation (33), is approximated by ! L,gh (tj ), Equation (57) and if the analyses
are carried out in time domain with tj tj 1 = $t = cos (56) can be expressed as
- .
$t
! L,gh (tj ) = bg ! L,gh (tj 1 ) + ag,11 bg h tj 1 + $t (58)
2
dg U dg U ;
where the terms bg = e B (tj tj 1 ) and bg = e 2B (tj tj 1 ) = bg are two constants of the
simulation time step.
In Equation (58) the evaluation of the term h(t j 1 + $t/2)$t is the most delicate aspect
because it has a big influence on the following simulation step. For this reason some approxi-
mations have been developed for the determination of h(t j 1 + $t/2)$t so that the algorithm
implementation would be efficient. In the next four schemes will be briefly described:

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1027

Scheme AStep with initial time reference t1 ; the approximation of h(tj 1 + $t/2)$t is
j 1 + $t/2)$t is rewritten as
carried out by the instantaneous value of the velocity tj 1 so h(t
- .
$t
h tj 1 + $t j 1 )$t = $h(tj 1 )
= h(t (59)
2
j 1 + $t/2)$t is very simple and it does not require the inter-
Approximation (59) of the h(t
j 1 + ($t/2)).
mediate evaluation of the term h(t
Scheme BNumerical derivation; developing h(t i1 + $t/2) by Taylors series up to the
third order, the result is

- . j 1 ) - $t .2
$t D 2 h(t
h tj 1 + j 1 ) $t +
j 1 ) + D h(t
= h(t
2 2 2 2
j 1 ) - $t .n
D n h(t
+ + +
n! 2
- . j 1 ) - $t .2
$t D 2 h(t

= h tj 1 + j 1 ) $t +
j 1 ) + D h(t
= h(t (60)
2 2 2 2

Equation (60) becomes (59) considering only the first term of the Taylors series. For the
derivatives operator D, considering the approximation of the derivatives till the third order, the
j 1 ), D h(t
term h(t j 1 ) and D 2 h(t
j 1 ) can be written as

- . - 2 3 .
h(tj 1 ) = Dh(tj 1 ) = 1 ln 1 h(tj 1 ) = 1 ln + + j 1 )
+ h(t
$t 1 $t 2 3
3 1
- .
, 1 11 3 1
= g h(tj 1 ) = h(tj 1 ) 3h(tj 2 ) + h(tj 3 ) h(tj 4 ) (61)
g=1 g $t 6 2 3

3 1
- .
, g 1 2 1 3
D h(tj 1 ) = h(tj 1 ) = + + h(tj 1 ) (62)
g=1 g 2 3
! "2 - .
,3 1 1 4 1 6 2 4 1 5
2
D h(tj 1 ) = g 2 3
h(tj 1 ) = + + + + + h(tj 1 )
g=1 g 4 9 3 3
(63)

where h(t) = h(t) h(t $t) and g h(t) = g1 h(t) g1 h(t $t) is the back operator.
Scheme CCubic Spline; the approximation of h(t) into the last known interval [tj 2 tj 1 ]
j 2
(t) and its second derivatives by h
j 2
is defined by h3 3(t):
j 2 tj 1 t t tj 2
h 3 (t) = cj 2 + ci1 (64)
tj 1 tj 2 tj 1 tj 2

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1028 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

where t [tj 2 tj 1 ] and the two coefficients cj 2 and cj 1 are the values of the second
derivatives of spline function approximation at time tj 2 and tj 1 . Equation (64) can be written
over all time intervals where the dates are observed (e.g. n time intervals). Integrating Equation
j 2
(64) with its continuity boundary conditions for the last interval [tj 2 tj 1 ] (h 3 (tj 1 ) =
j 2 2 2 j 2
h3 (tj 1 ) and h 3 (tj 2 ) = h3 (tj 2 )) and with time step $t = cos, for h 3 (t) and
j j
j 2
h 3 (t) at time t = t + ($t/2) the expression will be
i1

j 2 cj 2 ci1
h 3 (t) = (tj 1 t)3 + (t tj 2 )3
6$t 6$t
- . - .
hj 2 $t hj 1 $t
+ cj 2 (tj 1 t) + cj 1 (t tj 2 ) (65)
$t 6 $t 6
- . - . - .
j 2 $t 1 9 hj 2 $t hj 1 $t
h 3 ti1 + = cj 2 $t + cj 1 $t cj 2 + cj 1
2 8 8 $t 6 $t 6
(66)
j 3
The coefficients cm are determined by the n + 1 first order continuity conditions h 3 (tj 2 ) =
j 2
h 3 (ti2 ). These conditions carry out the three-dimensional diagonal system (n 1) (n 1):
cj 1n , cj 1 = basic assumptions
, i = 1, 2, . . . , n (67)
cj i + 4cj 1i + cj 2i = 6(hj i 2hj 1i + hj 2i )/$t 2

In the recursive scheme, the value h(t) is required to be out of the last interval time in the
point time ti1 + $t/2 which is into the interval [tj 1 tj ]. Over the boundary point it occurs
the most important free condition of the spline formulation, c0 and cn , which can be valued
by polynomial function of IIIth order h(ti ) = a + bh(ti ) + ch2 (ti ) + dh3 (ti ) fitting for the point
h(tj i ), i = 1 4.
Scheme DNumerical integration: the last term of (56) is carried out by direct integration
j 2
of the spline function approximation h 3 (t) by C point over the last interval [tj 1 tj ]:
- .
) tj dg U 6
dg U $t $t - .
$t , 2 2 xs j 2 $t
ag,11 e B (tj 1 $t%) h(%)
d% Bs ag,11 e B
h 3 tj 1 + (1 + xs )
tj 1 2 s=1 2
(68)

where Bs and xs with s = 1, 6 are the weights and the co-ordinates of the GaussLegendre
points into the interval normalized [0 1] = 1/$t[tj 1 tj ].

5. FINITE ELEMENT FORMULATION

The geometrical non-linear response of the structures has been analysed using a finite element
code named Loki (geometrically non-linear finite element analysis of cable, membrane and

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1029

skew beam elements), worked out according to the total Lagrangian formulation developed in
Padua University (e.g. References [18, 21, 27]).
The equations of motion in terms of a nodal displacement vector {X} for a bridge structural
system under wind loads can be expressed in conventional matrix notation (finite element
method) as
+ [C]{X}
[M]{X} + [K]{X} = {F } + {Fb } + {Fse } (69)

where {F } is static force (e.g. dead loads, pre-stress), {Fb } is buffeting forces (7)-(8) and (41),
{Fse } is aeroelastic forces (3)-(4) and (33) and [M], [C] and [K] = [K({X(t)})] are mass,
damping and structural stiffness matrices.
Considering the skew beam element, [27, 28], the aeroelastic element can be obtained. In
three-dimensional analysis a skew beam element has four nodes and six degrees of freedom,
whereas the section model of the deck has three degrees of freedom; in this way the corre-
spondence between cinematic (displacements, velocities and accelerations) and forces quantities
(lift, drag and moment) can be written as (Figure 2):

i i j
j u x (t)
ux (t) 0 0
i j i j
uy (t) p(t)
u y (t)

p(t)


i j
h(t)
i j
u z (t) h(t)


u (t)
{i Xj (t)} = z
i j =
, {i X j (t)} = j = ,
)x (t) !(t)
i ) (t) ! (t)
x
i j 0

j 0
i ) (t)
)x (t) x
i j
) (t) 0 i j 0
x )x (t)
i j

u x (t)
i j 0
u y (t)

p(t)
i j
u z (t) h(t)
i j
{ X (t)} = j = (70)
i ) (t) ! (t)
x
j 0
i ) (t)
x
i j 0
)x (t)


0 0
i j
Fy (t) Dse (t)

i j L (t)
Fz (t) se
{ Fse (t)} =
i j
i j = $x (71)
Mx (t) Mse (t)

0 0

0 0

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1030 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

Figure 2. Skew beam model section.

where i is ith-element and j is j th-node (with j = 1 4 for skew beam element implemented
into Loki code). Considering ith-element and j th-node of the finite element model with its
i $x j influence length of the bridge. The load vector {i F j (t)} can be written by (3)-(4) as
se

! - .2 "
1 B B j
{i Fsej (t)} = "U 2 B [i A0 ]{i Xj (t)} + [i A1 ]{i X j (t)} + [i A2 ]{i X j (t)} + {i ! }
j j j
2 U U
(72)

where


0 0 0 0 0 0

0 a0,22 a0,21 Ba0,23 0 0


0 a0,12 a0,11 Ba0,13 0 0
i j

j

i
[ A0 ] = $x

0 Ba0,32 Ba0,31 B 2 a0,33 0 0


0 0 0 0 0 0

0 0 0 0 0 0

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1031


0 0 0 0 0 0

0 a1,22 a1,21 Ba1,23 0 0


0 a1,12 a1,11 Ba1,13 0 0
i j

j

i
[ A1 ] = $x

0 Ba1,32 Ba1,31 B 2 a1,33 0 0


0 0 0 0 0 0

0 0 0 0 0 0

0
0 0 0 0 0 0
i j
0 a2,22 a2,21 Ba2,23 0 0 ! (t)
Fy

0 a2,12 a2,11 Ba2,13 0 0 ij
i j !Fz (t)
i j i j
[ A2 ] = $x [ ! ] = (73)

0 Ba2,32 Ba2,31 B 2 a2,33 0 0 i j
!Mx (t)

0 0 0 0 0 0
0

0 0 0 0 0 0
0

This formulation gets to a lumped expression of the aeroelastic forces, but it can be developed
towards a consistent matrix expression using the shape function of the finite element [29]. By
means of (73) the expressions of whole ith-element, which has 4 nodes, can be written as:

1
i 1 i 1 i 1
{ u (t)} { u (t)} { u (t)} {i ! (t)}


{i u2 (t)} {i u 2 (t)} {i u 2 (t)} {i ! 2 (t)}
1 i i i
{i Fse (t)} = "U 2 B
[ A0 ]

+[ A1 ]

+[ A2 ]

+



2 i 3
{ u (t)} i 3
{ u (t)} i 3 3
{ u (t)} {i ! (t)}

i 4 i 4 i 4

{ u (t)} { u (t)} { u (t)} i 4
{ ! (t)}
(74)

[ i
A1g ] 0 0 0 0 (=1




0 [i A2g ] 0 0
B
i 1 (=
where [ Ag ] = ( , with g = U (75)
0 0 [i A3g ] 0
- .2



B
2 ( =
0 0 0 [i A4g ] U

In this formulation, {i X(t)} is the aerodynamic displacement due to wind load actions and
it doesnt include the displacement given into the state 0 (e.g. dead load and pre-stress) by

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1032 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

{i Xstate 0 (t)}. Consequently, for the aeroelastic problem, the aerodynamic displacement, velocity
and acceleration are rewritten as


{iXse (t)} = {iX(t)} {iXstate 0 (t)}, {iX se (t)} = {iX(t)},
{iX se (t)} = {i X(t)}
(76)

Assembling all the elements, the motion equation by finite element method can be written for
the wind engineering problem as

+ [C]{X}
[M]{X} + [K]{X} = {F } + {Fb } [Mse ]{X}
[Cse ]{X}

[Kse ]{X} + [Kse ]{Xstate 0 } + {Fse,m } (77)

where [Mse ], [Cse ] and [Kse ] are aeroelastic mass, aeroelastic damping and aeroelastic
stiffness matrices (which are obtained by assembling the approximation coefficients (4) by
matrices (75) for all finite elements of the model); by means of (76) [Kse ]{Xstate 0 } introduces
in (69) the static deformation (which is not important for the aerodynamic and aeroelastic
phenomena), while {Fse,m } is the memory force.
Using (77), it can be rewritten (69) as


([M]+[Mse ]){X}+([C]+[C
se ]){X}+([K]+[Kse ]){X}={F }+{Fb }+[Kse ]{Xstate 0 }+{Fse,m }
(78)
X}
[M]{ + [C]{
X} + [K]{X}
= {F } + {Fb } + [Kse ]{Xstate 0 } + {Fse,m }

The mass matrix, [M] = [M] + [Mse ], damping matrix, [C] = [C] + [Cse ] and stiffness matrix,

[K] = [K] + [Kse ], with memory term, {Fse,m }, represent the basic matrices of the new
aeroelastic finite element developed into time domain for bridges non linear geometric analysis
starting from skew beam formulation.
Due to the coupling between structural behaviour and aereoelastic effect, the aereoelastic
forces could induce self-excitation effects. In particular the development of an aeroelastic
response (e.g. flutter analysis), involves the coupling of two structural modes (e.g. flexural and
torsional) which affects the whole bridge.
The system of equations governing the aeroelastic response, Equation (78), is a coupled
system, where the total mass, damping and stiffness matrices, as well as the memory term,
include the contribution of both structural behaviour (i.e. eigenforms) and aeroelastic effect.
Such matrices have been obtained by assembling the rational coefficient (e.g. [A2 ], [A1 ],
[A0 ], Equation (73), if considering ith-element and j th-node of the finite element model).
Therefore the solution of the total coupled system implicates that the oscillations accord-
ing to eigenforms of the structures are included in the dynamic response of the bridge. In
such a way structural motion and wind load are correlated each other by a non conservative
system.
Moreover the system (78) is in general non-linear and non-symmetric because the
matrices [Mse ], [Cse ] and [Kse ] are non-linear and non-symmetric. The non-symmetry of
matrices is a consequence of assuming aeroelastic wind load as non conservative
load [27].

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1033

The solution of the equations of motion was carried out by Reference [30]:
time domain integration by Newmark step-by-step integration methods:

{X(t
+ $t)} = a0 ({X(t + $t)} {t X(t)}) a2 {X(t)}
a3 {t X(t)}
(79)
+ $t)} = a1 ({X(t + $t)} {X(t)}) a4 {X(t)}
{X(t
a5 {X(t)}

where the coefficients aj with j = 0 7 are the Newmark constants defined by

1 ' 1 1 'N
a0 = 2
, a1 = , a2 = , a3 = 1, a4 = 1
!N $t !N $t !N $t 2!N !N
- . (80)
$t 'N
a5 = 2 , a6 = $t (1 'N ), a7 = 'N $t
2 !N
and !N , 'N are two parameters which describe the kind of approximation.
non-linear geometric response by NewtonRaphson algorithmic. The new displacement
{X(K) (t + $t)} and the tangent stiffness matrix [KT(K1) (t + $t)] at time t + $t and K
non-linear iterations are
{XK (t + $t)} = {XK1 (t + $t)} + {$XK }
<
!([K(t + $t)]{u(t + $t)}) << (81)
[KT (K1) (t + $t)] = <
!{u(t + $t)} {u(K1) (t+$t)}

System (78) can be rewritten in compact form as

[K(t + $t)]{u(t + $t)} = {F (t + $t)} (82)

where the two terms [K(t + $t)] and {F (t + $t)} are equivalent stiffness matrix and equivalent
load vector:

+ a1 [C]){$u
[K(t + $t)] = ([KT (K1) (t + $t)] + [Kse ] + a0 [M] (K) }
! "
[KT (K1) (t + $t)] + [Kse ] + a0 [M] + a0 [Mse ] + a1 !R [M]
= {$u(K) } (83)
+a1 (R [KT (t)] + a1 [Cse ]

{F (t + $t)} = {F (t + $t) } ([K (K1) (t + $t)] + a0 [M]


+ a1 [C]){u
(K1) (t + $t)}

= F (t + $t) + Fb (t + $t) + [Kse ]{ustatico (t)} + {Fse,mem (t)}

([K(K1) (t + $t)] + [Kse ]){u(K1) (t + $t)}


! "
a0 ({u(K1) (t + $t)} {u(0)} {u(t)} + {u(0)})
+ ([M] + [Mse ])
+a2 {u(t)}
+ a3 {u(t)}

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1034 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

+ (!R [M] + (R [KT (t)] + [Cse ])


! "
a1 ({u(K1) (t + $t)} {u(0)} {u(t)} + {u(0)})
(84)
+ a4 {u(t)}
+ a5 {u(t)}

The last expression of the two sides of (82) is obtained setting damping matrix as Reyeighs
damping [C] = !R [M] + (R [KT (t)] where !R and (R are Reyeighs damping coefficients.
System (82), with (83) and (84), represents the final system for aeroelastic response of bridge
with non-linear geometric behaviour.

6. APPROXIMATION BY MEANS OF A RATIONAL FUNCTION

This section illustrates the approximation by means of a rational function through nonlinear
least-squared of the complex pair (22) by means of Equations (24) and (25). The flutter
derivatives of the Akashi Kaikyo bridge of Japan which, with a main span of 1990 m and two
side spans of 960 m, is the worlds longest suspension bridge is investigate. The stiffening truss
girder has a 35.5 m width, 14 m depth, 6.74 m2 projected area normal to the wind direction and
an offset distance between the shear and gravity centres of the 3 m. The static aerodynamic
coefficients of the truss girder are CD = 1.987, CL = 0.0182, CM = 0.0046, CD = 0.217,
CL = 1.612 and CM = 0.381.

In Figure 3 the rational approximation functions are compared with the values of the aerody-
namic derivatives of the stiffening truss girder (modified section) measured in the wind tunnel
by test on section models. It worth noting that H5 , H6 , A5 , A6 and P4 are not significant and
are thus neglected in this study, while P1 = 2/kCD An /B is based on the quasi-steady for-
mulation (e.g. References [4, 31, 32]). The experimental data are given by the HonshuShikoku
Bridge Autority (1995). This comparison is drawn so as to emphasize how a complex rational
function can be used to follow up the experimental trend of complex pairs of flutter derivatives.
In the next part of the paper some values of the coefficients dg , smaller than those assumed in
Reference [4], will be used by allowing the integration of the memory term for the application
of a recursive formula. The approximation will be developed considering four rational terms; it
is evident that the quality of the approximation improves as the number of terms increases, but
the contribution of such an improvement is minimal. Moreover the dg coefficients are assumed
to be constant and equal for all the complex pairs, Table II. The approximation by rational
function is very close to experimental values of the derivatives and only the pair (P6 + iP5 )
gives the same imprecision.

6.1. Numerical integration of the memory of the system-time step variation


The integration of the memory of Equation (57) for the different force contributions (lift,
drag and moment) is numerically complex because of the damping of the signal h(t),
p(t)
and ! (t) induced by the exponential function and because the complesity of exact integration
with the recursive expression (58). To emphasize these drawbacks, lets start by considering a

velocity signal h(t) and the relative sinusoidal displacements h(t) with a period corresponding
to T = 6.5 s. These data correspond to the second symmetrical vertical frequency of Akashi
Kaikyo bridge.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1035

2 30
H*3
1
25 H*2
0 H*3_RF
20
-1 H*2_RF

-2 15

-3 10
-4
H*4 5
-5 H*1
H*4_RF 0
-6
H*1_RF
-7 -5
0 5 10 15 20 25 0 5 10 15 20 25
(a) Ur=nU/B (b) Ur=nU/B

1 0.5

0
0.5
-0.5

0 -1

-1.5
-0.5
-2
P*5 P*4
-2.5
-1 P*6 P*1
P*5_RF -3 P*4_RF
P*6_RF P*1_RF
-1.5 -3.5
0 5 10 15 20 25 0 5 10 15 20 25
(c) Ur=nU/B (d) Ur=nU/B

-0.5

-1

-1.5

-2

P*3
-2.5
P*2
-3 P*3_RF
P*2_RF
-3.5
0 5 10 15 20 25
(e) Ur=nU/B

0.2 5

0 A*3
4 A*2
-0.2 A*3_RF
3 A*2_RF
-0.4

-0.6 2

-0.8
A*4 1
-1 A*1
A*4_RF 0
-1.2
A*1_RF
-1.4 -1
0 5 10 15 20 25 0 5 10 15 20 25
(f ) U r=nU /B (g) Ur=nU/B

Figure 3. Aeroelastic derivatives (Akashi Kaikyo bridge [4]): lift force (a) h and (b) !; drag force
(c) h, (d) p and (e) !; pitch moment (f) h and (g) !.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1036 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

Table II. Rational functions approximation.


Stiffness Damping Mass Memory term
3 4 5 6
dg 0 1 2 0.5 0.75 1.0 1.25

ag Drag P6 P5 0.03005 5.12068 0.00785 10.66392 100.244 260.8759 191.791


P4 P1 0.00341 0.72464 0.00022 0.26906 1.74866 3.61093 2.28922
P2 P3 0.10937 3.95222 0.00349 1.9387 2.70848 24.09315 35.6992

ag Lift H4 H1 0.08202 3.49127 0.00195 4.092 24.281 48.702 32.615


H3 H2 1.50349 12.35023 0.00995 37.83506 305.223 715.6156 491.441

ag Moment A4 A1 0.01012 0.3791 0.00026 0.51397 2.92577 5.36878 3.17246


A3 A2 0.26626 1.63735 0.00192 5.81679 45.647 105.1135 71.5452

Table III. Rational functions approximation Types A and B.


Type A analysis[4] Type B analysiscurrent analysis
3 4 5 6 3 4 5 6

dg 0.5 1.5 5.0 8.0 0.5 0.75 1.0 1.25


ag 0.0339 0.4148 4.9585 6.1810 5.81679 45.647 105.1135 71.5452

The memory terms are calculated with the coefficients ag,23 and dg given in Table III,
which are those reported in literature (Type A), and those (Type B) obtained directly from the
approximation of the aeroelastic derivatives by means of nonlinear least-square method, Table
II. Its worth noting that the global force terms given by (3) must theoretically remain constant
in relation to variations in the rational approximation terms, but each memory component
(58) varies with the coefficients ag , dg . In this test the memory term is not coupled with the
systems motion, i.e. a pure integration test is performed with a recursive type of algorithm.
In real dynamic, memory and structural motion are totally coupled, therefore the mistakes on
memory evaluation repeat themselves in the subsequent simulation steps. In this simulation the
recursive expression (58) and step velocity approximation (59) are applied, obtaining thus a
manifest variation of the result with the time step. Figures 4 and 5 show the results relating
to the memory ! g,23 (t) of the system assessed using the two previously-described sets of
coefficients and various signal integration steps: $t = 0.2, 0.1, 0.05, 0.01, 0.005 s where the
reference parameter is:
! "= ! "
,4 ,4
R = max
!g,23 (t), $t = 0.2 s max
!g,23 (t), $t = 0.005 s (85)
g=1 g=1

In Figures 4(a)(d) it should be noticed that, as the integration step diminishes, the signals
amplitude tends towards a constant value, very different from the one estimated using higher

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1037

3.E-01

2.E-01
2.E-01

Mmem/0.5(UB)2
1.E-01
5.E-02

0.E+00
-5.E-02
-1.E-01
-2.E-01
-2.E-01
-3.E-01
(a) 0 20 40 60 80 100 120 140 160 180 200

8.E-02

6.E-02
Mmem/0.5(UB)2

4.E-02

2.E-02

0.E+00

-2.E-02

-4.E-02

-6.E-02

-8.E-02
(b) 0 20 40 60 80 100 120 140 160 180 200

3.E-02

2.E-02
2.E-02
Mmem/0.5(UB)2

1.E-02
5.E-03

0.E+00
-5.E-03
-1.E-02
-2.E-02
-2.E-02
-3.E-02
0 5 10 15 20 25 30 35 40 45
(c)
3.E-02

2.E-02
2.E-02
Mmem/0.5(UB)2

1.E-02
5.E-03

0.E+00

-5.E-03
-1.E-02

-2.E-02
-2.E-02

-3.E-02
0 5 10 15 20 25

(d) Time [s]

Figure 4. Memory term ! g,23 (t): Type A [4] analysis $t=0.1 s (a); 0.05 s (b); 0.01 s (c); and 0.005 s (d).

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1038 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

6.E-01

4.E-01

Mmem/0.5(UB)2
2.E-01

0.E+00

-2.E-01

-4.E-01

-6.E-01
(a) 0 20 40 60 80 100 120 140 160 180 200

8.E-01

6.E-01
Mmem/0.5(UB)2

4.E-01

2.E-01

0.E+00

-2.E-01

-4.E-01

-6.E-01

-8.E-01
(b) 0 20 40 60 80 100 120 140 160 180 200

8.E-01

6.E-01

4.E-01
M mem /0.5(UB)2

2.E-01

0.E+00

-2.E-01

-4.E-01

-6.E-01

-8.E-01
(c) 0 5 10 15 20 25 30 35 40 45

8.E-01

6.E-01
Mmem/0.5(UB)2

4.E-01

2.E-01

0.E+00

-2.E-01

-4.E-01

-6.E-01

-8.E-01
0 5 10 15 20 25

(d) Time [s]

Figure 5. Memory term ! g,23 (t): Type B analysis $t=0.1 s (a); 0.05 s (b); 0.01 s (c); and 0.005 s (d).

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1039

30.000

25.000

20.000

R
15.000

10.000

5.000

0.000
0 0.05 0.1 0.15 0.2 0.25
(a) t step integration [s]

1.02
1

0.98

0.96

0.94

0.92
R

0.9

0.88

0.86

0.84

0.82
0 0.05 0.1 0.15 0.2 0.25

(b) t step integration [s]

Figure 6. Memory term error ! g,23 (t): Type A [4] analysis (a); and Type B analysis (b).

integration steps. Similarly, with smaller integration steps, the initial transient is strongly influ-
enced by $t and by the initial conditions.
The graphics Figures 5(a)(d) show that the signal remains substantially constant in amplitude
and shape as the integration step varies, and there is none of the effects of the starting mode of
motion evidenced in Figures 4(a)4(d). The reason for this different response of the recursive
algorithm lies in the damping rate of the memory term. The choice of suitable exponential
terms that reduce the damping rate (dg : 0.5, 1.5, 5.0, and 8.0 dg : 0.5, 0.75, 1.0 and
1.25) facilitates the recursive integration and limits its dependence on the integration step. If
the integration of the memory term is carried out into Laplace domain (hypothesis of linear
structural behaviour), for example by modal co-ordinated state space approach, the valuation
of the memory term by (31) is easier then time domain, so it is possible to use high dg
coefficient.
In Figure 6 the integration-step-related variation in the memory term is further emphasized
with reference to the Table III. Clearly, the integration Type B induces minimal variations in
the memory term in relation to variations in the integration step, while the integration Type
A is totally dependent on the step selected in the dynamic analysis. This means that the
simulation step can no longer be selected merely with reference to the structure-and to the

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1040 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

fundamental period ($t T0 /20)but must ensure an accurate integration of the bridges
aeroelastic dynamics.

6.2. Numerical integration of the memory of the systemkind of velocity approximation


The following part describes a comparison between the different ways to evaluate the memory
term (33) using the medium and maximum residual value of the memory term as reference
parameters:

! g,31 (t) !g,31 (t)


Maximum residual: Rg,max = maxT (86)
maxT !g,31 (t)
)
! g,31 (t) !g,31 (t)
Medium residual: Rmax = dt (87)
T maxT !g,31 (t)

The term !g,31 (t) is the close form of the memory integration term obtained by sinusoidal
hypothesis for displacement and velocity motion of the bridge deck. In the next part the
comparison is developed also with reference to sinusoidal signal with a random superposition
signal.
Figures 7 and 8 show the comparison between the maximum (86) and medium (87) residual
for the memory term integration by (57), whose velocity is approximated by the above-described
model (59), (60), (66) and (68) with and without the random component added to sinusoidal
velocity. The sensibility analysis of recursive time integration is carried out with several inte-
gration time steps in order to relate the scheme type and the time-step integration.
About Figure 7 its worth noting that in general the quality of time integration improves as
the scheme for velocity approximation goes from scheme A to scheme D and when the time
step $t reduces itself. The quality increases rapidly when the time step comes down and the
difference between the different scheme decreases as well. The maximum residual reduces from
13.0% (scheme A) to 0.3% (scheme D) and the difference is more evident for the medium
residual which reduces from 8.5% (scheme A) to 0.2% (scheme D) for the highest time
step 0.2 s. If the time step is set to 0.05 s the maximum residual changes from 3.5%
(scheme A) to 0.5% (scheme C) and the difference is more evident for the medium residual
which goes on from 2.0% (scheme A) to 0.3% (scheme C). The difference between the
scheme A to scheme B, C and D is very high and not comparable when the time step is within
the range 0.2 to 0.05 s. Such a difference becomes less important than above when the time
integration is very small. Instead, the schemes B, C and D give similar result when the $t is
about 0.1 s, while there is the same difference with bigger time step. The highest quality is
obtained when the recursive integration of memory uses the scheme D where the integration
uses the spline (scheme C) approximation of the velocity.
Figure 8 shows how the difference among the schemes A, B, C and D is smaller when
the signal of velocity isnt sinusoidal but there is a random component added to it. Assuming
the maximum residual as reference parameter, Figure 8(a), scheme B gives a similar time
integration quality of the scheme A and for the time step 0.2 scheme B is better than scheme
A; at the same time, scheme D is a worse algorithm than scheme C for all time steps. Increasing
the time step, the difference between the schemes decreases and for time step about 0.05 s

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1041

14.00%

12.00%

Maximum residual
10.00%

8.00%

6.00%

4.00%

2.00%

0.00%
0.20000 0.10000 0.05000 0.01000 0.00500
scheme A 1.32E-01 6.93E-02 3.56E-02 7.27E-03 3.64E-03
scheme B 2.47E-02 6.52E-03 1.66E-03 6.75E-05 1.69E-05
scheme C 1.58E-02 6.12E-03 3.64E-03 8.73E-04 4.47E-04
scheme D 3.41E-03 4.24E-04 5.71E-05 9.62E-06 9.61E-06
(a) Time step [s]

9.00%

8.00%

7.00%
Medium residual

6.00%

5.00%

4.00%

3.00%

2.00%

1.00%

0.00%
0.20000 0.10000 0.05000 0.01000 0.00500
scheme A 8.55E-02 4.49E-02 2.28E-02 4.64E-03 2.32E-03
scheme B 1.58E-02 4.17E-03 1.06E-03 4.29E-05 1.07E-05
scheme C 1.03E-02 4.00E-03 2.33E-03 5.58E-04 2.85E-04
scheme D 2.15E-03 2.68E-04 3.28E-05 6.10E-06 6.10E-06

(b) Time step [s]

Figure 7. Maximum (a); and medium (b) residual of the memory of the system ! g,23 (t) when the

signal is a sinusoidal displacement h(t) and velocity h(t).

the schemes can be divided into two groups (scheme AB and scheme CD). The maximum
residual changes from 18.0/4.57 (scheme A) to 7.5/1.5% (scheme C) for the times step
0.2/0.05 s.
The case of integration with scheme D gives a bigger residual than scheme C for high time
step when the medium residual is considered. With this parameter (as shown by Figure 8(b)),
the scheme A offers a very great difference from the other three models and this difference
grows as the random component increases. Regarding schemes B, C and D the difference is
distributed only when the time step is high, whereas the diagram is close with low time step.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1042 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

20.00%
18.00%
16.00%

Maximum residual
14.00%
12.00%
10.00%
8.00%
6.00%
4.00%
2.00%
0.00%
0.20000 0.10000 0.05000 0.01000 0.00500
scheme A 1.77E-01 1.11E-01 4.57E-02 8.48E-03 4.93E-03
scheme B 1.83E-01 1.07E-01 4.05E-02 6.55E-03 3.94E-03
scheme C 7.37E-02 6.18E-02 1.65E-02 2.47E-03 1.63E-03
scheme D 1.32E-01 7.89E-02 2.06E-02 3.14E-03 1.80E-03
(a) Time step [s]

8.00%

7.00%
Medium residual

6.00%

5.00%

4.00%

3.00%

2.00%

1.00%

0.00%
0.20000 0.10000 0.05000 0.01000 0.00500
scheme A 7.08E-02 3.81E-02 1.95E-02 4.95E-03 2.09E-03
scheme B 4.58E-02 1.13E-02 2.03E-03 1.49E-04 3.76E-05
scheme C 2.04E-02 6.21E-03 2.25E-03 6.18E-04 2.61E-04
scheme D 2.59E-02 4.96E-03 5.70E-04 4.42E-05 1.52E-05

(b) Time step [s]

Figure 8. Maximum (a); and medium (b) residual of the memory of the system ! g,23 (t) when the
signal is a linear superposition of a sinusoidal and random (! 10% of the mean and active each 1 s)

signal for h(t) and velocity h(t).

7. CROSS SECTION MODEL OF BRIDGE DECK

This section outlines the first results of the simulation carried out by using a cross section
model of Akashi Kaikyo bridge with two main degrees of freedom, with the aim of comparing
the results obtained with others reported in the literature. Moreover, this study only considers
the effect of aeroelastic forces to evaluate the critical flutter velocity but the problem can be
generalized for buffeting analysis . The cross section model for mass, stiffness and damping
was calibrated so that the two own modes of vibration correspond to the second symmetrical

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1043

vertical mode (#f = 0.1211 Hz, T = 8.26 s) and the first symmetrical torsional mode (#T =
0.1467 Hz, T = 6.81 s) [4]. The frequencies were selected in this way because in flutter
conditions the structure induces the coupling of such two modes. The tendency of coupling
of these two modes of vibration is demonstrated by several studies in literature dealing with
Akashi Kaykio bridge. The analyses are carried out with a gradually increasing incident wind
velocity, studying the response in the field of both large and small displacements. By observing
the diagrams of the equivalent lift and moment forces due to the contributions of mass,
damping, stiffness and memory, calculated explicitly on the basis of (33), it can be deduced
that the memory diagrams corresponding to the acceleration terms offer limited contributions
by comparison with the other terms. The greatest contribution coincides with the damping
term, which introduces energy in the structural system and determines an amplification of the
oscillations with increasing velocities. Conversely, the structural stiffness terms cause variations
in the own vibration frequency of the system.
In particular, with an incident wind velocity equating to U = 80 m/s, the structure has a
damping action and the amplitude of oscillations diminishes progressively as shown in Figure
9(a)(c) for central displacement, rotation and phase diagram where the trajectory converges to
the center. From this diagram its evident how the motion of bridge is influenced by the initial
condition necessary to set off of the process.
With a greater velocity, Figure 10 U = 86.5 m/s, the amplitude of oscillations remains
constant and the dissipated energy due to damping compensates for the one introduced by the
flow. In this case, amplitude of displacement and rotation remains constant, Figure 10(a)10(c),
and the two vibration modes are characterized by the same frequency of oscillation Figure
10(d), (f = 0.136 HZ, T = 7.35 s), but are out of phase each other. A situation in which
displacements and rotations are of constant amplitude and equal frequency but out of phase by
approximately $/2 identifies the critical condition and defines the flutter velocity, i.e. 86.5 m/s.
The flutter frequency is into the frequency range set from vertical and torsion mode (#f =
0.1211 Hz ! #Flutter = 0.136 Hz ! #T = 0.1467 Hz). The flutter velocity and frequency are in
good agreement with analytical result in literature (U = 79.1 m/s, #Flutter = 0.146 Hz [32])
and with measurement (U = 84.0 m/s, #Flutter = 0.135 Hz [32]).
With a greater incident velocity, the amplitude of displacements and rotations increases,
since the structure is unable to damp the energy introduced by the forces self-excited by the
aeroelastic contributions and this is clear in Figure 11(a)(d). Figure 11 shows the analyses
with U = 110 m/s for a period long enough to bring out the effect of the geometrical
non-linearity. Clearly, the more the amplitude of the displacements and rotations increases, the
more the geometrical non-linearity influences the structural response, to the point of limiting the
maximum amplitudes due to motion. The motion is consequently confined within the maximum
amplitude limits of displacement and rotation defined by the structural damping action. At these
amplitudes this phenomenon occurs because the energy introduced by the forces is dissipated
by the system.
This effect, Figure 11(d), should be more deeply investigated in future development of the
research, in order to understand if it describes a correct response of the structures or if it
comes from a correct interpretation of non-linear geometric response of the system or if it is
a limitation of the wind model force introduced into the analysis (Figure 12).
Its important to understand that this aeroelastic finite element model is obtained by the
combination of non-linear geometric response of the structures and linear model of aeroelastic
and buffeting forces into time domain. The linear behaviour hypothesis for the wind load is

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1044 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

4.00E-02

3.00E-02

2.00E-02

Displacement [m]
1.00E-02

0.00E+00

-1.00E-02

-2.00E-02

-3.00E-02

-4.00E-02
0 20 40 60 80 100 120 140 160 180 200

(a) Time [s]


3.00E-04

2.00E-04

1.00E-04
Rotation [rad]

0.00E+00

-1.00E-04

-2.00E-04

-3.00E-04

-4.00E-04

-5.00E-04
0 20 40 60 80 100 120 140 160 180 200

(b) Time [s]


3.00E-04

2.00E-04

1.00E-04
Rotation [rad]

0.00E+00

-1.00E-04

-2.00E-04

-3.00E-04

-4.00E-04

-5.00E-04
-3.00E-02 -2.00E-02 -1.00E-02 0.00E+00 1.00E-02 2.00E-02 3.00E-02

(c) Displacement [m]

Figure 9. Wind velocity: 80 m/s (a)(b) displacement and rotation time history; and (c) phase diagram.

intrinsic into the expression written into frequency domain by (1)-(2) for aeroelastic force and
(5)(6) for buffeting force and into time domain by the two convolution integral expressions
(17) and (34). Advanced modeling of aerodynamic forces, which take into account the nonlinear

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1045

4.00E-02 3.00E-04

3.00E-02 2.00E-04

1.00E-04
2.00E+02
Displacement [m]

0.00E-00

Rotation [rad]
1.00E-02
-1.00E-04
0.00E-00
-2.00E+04
-1.00E-02
-3.00E-04
-2.00E-02
-4.00E-04

-3.00E-02 -5.00E-04

-4.00E-02 -6.00E-04
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200

(a) Time [s] (b) Time [s]

3.00E-04 4.00E-02 3.00E-04

2.00E-04 3.00E-02 2.00E-04

1.00E-04 2.00E-02 1.00E-04

-- Rotation [rad]
- Displacement [m]
0.00E-00 0.00E-00
Rotation [rad]

1.00E-02
-1.00E-04 -1.00E-04
0.00E-00
-2.00E+04 -2.00E+04
-1.00E-02
-3.00E-04 -3.00E-04
-2.00E-02
-4.00E-04 -4.00E-04

-5.00E-04 -3.00E-02 -5.00E-04

-6.00E-04 -4.00E-02 -6.00E-04


-3.00E-02 -2.00E-02 -1.00E-02 0.00E-02 1.00E-02 2.00E-02 3.00E-02 0 10 20 30 40 50 60 70

(c) Displacement [m] (d) Time [s]

Figure 10. Wind velocity 86.5 m/s: f = 0.131 Hz, T = 7.63 s; (a)(b) displacement and rotation time
history; and (c)(d) phase diagram and time history.

dependence of the aeroelastic forces on the effective angle of incidence, should be introduced
to improve the proposed approach.

8. CONCLUSIONS

This paper illustrates the procedure that leads to the formulation of aeroelastic and buffeting
forces in the time domain. The classical expressions of self-excited aeroelastic and buffeting
forces formulated according to quasi-static theory and generally used in time-domain studies,
have been replaced by forces expressions which takes into account the effect of the frequency.
The aerodynamic instability forces are made explicit in the time domain by means of the
convolution integral that involves the aeroelastic impulses and the structural motion. Similarly,
the expressions of buffeting that include the aerodynamic transfer function are described by
means of the convolution integral between the impulse functions and the fluctuating velocities.
In both cases, the aerodynamic impulse functions are identified by means of a comparison in
the frequency domain with the couples of aeroelastic derivatives or with the transfer functions.
The continuity needed for the explicit formulation in the time domain is obtained through a
rational approximation of the complex pairs of aeroelastic derivatives. The final expressions
of the aeroelastic forces in time show that the elimination of the frequency dependence takes

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1046 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

1.00E-01 4.00E-04

8.00E-02 2.00E-04

0.00E+00
6.00E-02
Displacement [m]

-2.00E-04

Rotation [rad]
4.00E-02
-4.00E-04
2.00E-02
-6.00E-04
0.00E+00
-8.00E-04
-2.00E-02
-1.00E-03

-4.00E-02 -1.20E-03

-6.00E-02 -1.40E-03
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200

(a) Time [s] (b) Time [s]

1.00E-01 4.00E-04
4.00E-04

8.00E-02 2.00E-04
2.00E-04

0.00E+00 0.00E+00
6.00E-02

- Displacement [m]

- - Rotation [rad]
-2.00E-04 -2.00E-04
Rotation [rad]

4.00E-02
-4.00E-04 -4.00E-04
2.00E-02
-6.00E-04 -6.00E-04
0.00E+00
-8.00E-04 -8.00E-04
-2.00E-02
-1.00E-03 -1.00E-03

-1.20E-03 -4.00E-02 -1.20E-03

-1.40E-03 -6.00E-02 -1.40E-03


-6.00E-02 -4.00E-02 -2.00E-02 0.00E-02 2.00E-02 4.00E-02 6.00E-02 8.00E-02 1.00E-01 0 10 20 30 40 50 60 70 80 90 100

(c) Displacement [m] (d) Time [s]

Figure 11. Wind velocity 110 m/s; (a)(b) displacement and rotation time history; and (c)(d)
phase diagram and time history.

Figure 12. Wind velocity: 110 m/s non linear geometrical response of the structures.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
AEROELASTIC FORCES AND DYNAMIC RESPONSE 1047

place through:
explicit terms, functions of the components of the structures displacement, velocity and
acceleration. These terms express the fluidstructure interaction during the evolution of the
phenomenon. Within the framework of finite element approach they allow the definition
of a new aeroelastic element;
memory term that introduces the story of the structural dynamic evolution. The memory
term is the direct consequence of the non-stationary formulation of the forces taken as
the basis for the procedure; it represents the known term in the structural dynamics and
describes the influence of past motion on the development of present motion.
The formulation is developed with the purpose of obtaining a direct integration in the time
domain, which enables the effect of non-linear geometric behaviour in the structural response
to be considered without any limitation (similarly the material non linear behaviour could be
included in a further development of the model). Thus it enables the analysis to be extended
to a thoroughly study of the flutter phenomenon and buffeting. In particular, it enables the
starting and evolution of flutter to be studied in a single procedure in combination with the
effect of turbulence even at low velocities. The present paper does away with the hypothesis
of a structure with a linear behaviour, but introduces the need to assess the memory term
accurately. In particular, a recursive expression is formulated that contains the calculation effort
and the memory requirements, by relying exclusively on what was calculated at the previous
step.

REFERENCES
1. Diana G, Cheli F, Zasso A, Collina A, Bruni S. Aerodynamic design of very long-span suspension bridges.
Long-span and high-rise structures. IABSE Symposium, Kobe, 1998; 115128.
2. Lazzari M, Saetta A, Vitaliani R. Analisi dinamica non lineare di un ponte strallato soggetto allazione del
vento. Proceedings of the 6th National Congress of Wind Engineering, ANIV 2000, Italy, Genova, 2000;
1821.
3. Mendes PA, Branco FA, Saraiva JG. Numerical simulation of the aeroelastic behaviour of Vasco Da Gama
Bridge. Proceedings of New Technologies in Structural Engineering, Portugal Lisbon, 1998.
4. Boonyapinyo V, Miyata T, Yamada H. Advanced aerodynamics analysis of suspension bridges by statespace
approach. Journal of Structures Engineering (ASCE) 1999; 125(12):13571366.
5. Ding Q, Lee PKK. Computer simulation of buffeting actions of suspension bridges under turbulent wind.
Journal of Computers and Structures 2000; 76:787797.
6. Borri C, Hoffer R. Aeroelastic wind forces on flexible bridge girders. Meccanica 2000; 35(1):115.
7. Simiu E, Scanlan R. Wind effects on structures. An Introduction to Wind Engineering. Wiley & Sons: New
York, 1996.
8. Chen X, Kareem A. Advanced in modeling of aerodynamic forces on bridge decks. Journal of Engineering
Mechanics (ASCE) 2002; 128(11):11931205.
9. Sarkar PP, Jones NP, Scanlan RH. Identification of aeroelastic parameter of flexible bridge. Journal of
Engineering Mechanics (ASCE) 1994; 120(8):17181742.
10. Theodorsen T. General theory of aerodynamic instability and the mechanism of flutter. NACA Technical
Report 496, 1935.
11. Scanlan RH. The action of flexible bridges under wind. 2:Buffeting theory. Journal of Sound and Vibration
1978; 60(2):201211.
12. Lin YK, Yang JN. Multimode bridge response to wind excitations. Journal of Engineering Mechanics (ASCE)
1983; 109(2):586603.
13. Bucher CG, Lin YK. Stochastic stability of bridges considering coupled modes. Journal of Engineering
Mechanics (ASCE) 1988; 114(12):20552071.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048
1048 M. LAZZARI, R. V. VITALIANI AND A. V. SAETTA

14. Chen X, Matsumoto M, Kareem A. Time domain flutter and buffeting response analysis of bridges. Journal
of Engineering Mechanics (ASCE) 2000; 126(1):716.
15. Cobo del Arco D, Aparicio Bengoechea AC. An analysis of wind stabilityImprovements to the response
of suspension bridges. Monograph CIMNE (International Center for Numerical Methods in Engineering)
1999; 49.
16. Walshe DE, Wyatt TA. Measurement an application of the aerodynamic admittance function for the a
box-girder bridge. Journal of Wind Engineering and Industrial Aerodynamics 1999; 14:211222.
17. Lazzari M, Saetta A, Vitaliani R. Non linear dynamic response of long-span bridges. Towards a better built
environmentinnovation, sustainability, information technology, IABSE Symposium, Melbourne, 2002.
18. Lazzari M. Geometrically non-linear structures subjected to wind actions. Ph.D. Dissertation, University of
Padua, Italy, 2002.
19. Tiffany SH, Adams WM. Nonlinear programming extensions to rational function approximation methods for
unsteady aerodynamic forces. NASA Technical Paper 2776, 1988.
20. Roger KL. Airplane math modeling methods for active control design: structural aspects of active control.
AGARD-CP-8, Advisory Group for Aerospace Research and Development, Neuilly-sur-Seine, France, 1977.
21. Lazzari M, Saetta A, Vitaliani R. Non-linear dynamics analysis of cable-suspend structures subjected to wind
actions. Journal of Computers and Structures 2001; 79(9):953969.
22. Bisplinghoff RL, Ashley H, Halfman RL. Aeroelasticity. Dover Publications: New York, 1955.
23. Fung YC. An Introduction to the Theory of Aeroelasticity. Dover Publications: New York, 1993.
24. Hffer R. Stationre und instationre modelle zur zeitbereichssimulation vor windkrften an linienfrmigen
bauwerken. Ph.D. Dissertation, University of Bochum, Germany, 1997.
25. Borri C, Costa C, Zahlten W. A computational model of nonstationary flow forces for the simulation of
aeroelastic instability of bridge decks. ECCM-2001, Cracow, Poland, 2000.
26. Borri C, Costa C, Zahlten W. Non-stationary flow forces for the numerical simulation of aeroelastic instability
of bridge decks. AIMETA-2001, Taormina, Italy, 2001.
27. Vitaliani RV, Gasparini AM, Saetta AV. Finite element solution of the stability problem for nonlinear
undamped and damped systems under nonconservative loading. International Journal of Solids and Structures
1997; 34(19):24972516.
28. Martini L, Vitaliani R. On the polynomial convergent formulation of a C isoparametric skew beam element.
Journal of Computers and Structures 1988; 29(3):437449.
29. Bontempi F, Malerba PG, Giudici M. La formulazione matriciale dei problemi aeroelastici di ponti sospesi
e strillati. Studi e RicerchePolitecnico di Milano 2000; 21.
30. Bathe KJ. Finite Element Procedures. Prentice-Hall: New Jersey, 1996.
31. Miyata T, Yokoyama K, Yasuda M, Hikami Y. Akashi Kaikyo Bridge: Wind effects and full model wind
tunnel tests. Aerodynamics of Large Bridges. Larsen A: Belkema Rotterdam, 1992.
32. Katsushi H, Jones NP, Scanlan RH, Akiyama H. Multi-mode flutter and buffeting analysis of the Akashi-
Kaikyo bridge. Journal of Wind Engineering and Industrial Aerodynamics 1998; 7778:431441.
33. Larose GL. Mann J. Gust loading on streamlined bridge decks. Journal of Fluids and Structures 1998;
12:511536.

Copyright ! 2004 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2004; 60:10111048

Вам также может понравиться