Sei sulla pagina 1di 8

Aerodynamic flutter

analysis of suspension
bridges by a modal
technique
T. J. A. Agar

Civil Engineering Department, Glasgow University, UK


(Received November 1987; revised January 1989)

Suspension bridges are long, slender flexible structures which have the
potential to be susceptible to a variety of types of wind-induced
instabilities, the most serious of which are divergence (due to stationary
wind forces) and flutter (due to aerodynamic forces). Flutter occurs at
certain wind speeds where aerodynamic forces acting on the deck feed
energy into an oscillating structure, so increasing the vibration ampli-
tudes. If this situation is approached the basic safety of the bridge is
threatened. This paper describes a computational method for predicting
flutter speed based on a modal technique. A selection of the lowest
vertical and torsional natural mode shapes is included with the aero-
dynamic forces in an interaction analysis, which yields an unsymmetric
matrix eigenvalue problem. Flutter instability is indicated when, at some
wind speed, one of the complex eigenvalue pairs resulting from the
solution of the eigenvalue problem has a zero real part and a non-zero
imaginary part.

Keywords: bridges, aerodynamics, instability

In the early part of this century, developments in the that, at certain critical wind speeds, the aerodynamic
aesthetics of bridge building and improvements in forces act to feed energy into the oscillating structure and
materials led to the construction of progressively longer, increase the magnitude of vibration, sometimes to catas-
structurally more efficient and slender bridges. It was only trophic levels. The critical flutter speed of a bridge
after the Tacoma Narrows suspension bridge collapsed depends on its vertical and torsional natural vibration
shortly after its completion in 1940 that the potentially characteristics and also on the cross-sectional shape of
unstable behaviour of this type of bridge under wind its deck (since this affects the aerodynamic forces that act
action began to be investigated. It subsequently became during pitch and heave oscillations). Current UK Code
appreciated that such bridges could be liable to violent guidance on stability 1 is available for structures with
aerodynamic oscillations that could not be diagnosed individual spans not exceeding 200 m. However, for spans
from static analyses taking account of steady wind forces, greater than this, advice is that the stability should be
even those for maximum design wind speeds. verified by wind-tunnel tests, which are both time-
One of the essential requirements of modern suspension consuming and expensive.
bridge design is to avoid significant levels of wind excited As it was useful at a design stage to be able to examine
oscillations, of which there are two main types: limited the effect of different structural configurations of a major
amplitude (non-divergent) oscillations produced by vor- suspension bridge being designed by Mott, Hay &
tex shedding; and divergent oscillations produced by both Anderson, London, UK, for a Far East client in relation
galloping and classical flutter types of instability. to, among other things, the flutter problem, a computa-
The former class of behaviour may, in limit state tional package ANSUSP 2 was developed to analyse
terminology, be considered primarily as a 'serviceability' suspension bridge three-dimensional dynamic behaviour.
problem responsible mainly for excessive levels of vibra- Initially, the program predicted flutter behaviour in a
tion and having a potential for serious fatigue damage module using a time integration approach and which
in the long term. In contrast, the latter class, in particular included the effects of geometrical nonlinearity. The
flutter, may be considered to be an 'ultimate' condition. aerodynamic deck excitation forces used were the analytic
Flutter is a self-excited oscillatory instability of a body Theodorsen functions appropriate to an oscillating flat
suspended in an air stream. It involves the interaction of plate. Successive modules (mostly based o n the time
aerodynamic, inertial and elastic structural forces such integration scheme) were added to predict different types

0141-0296/89/02075-08/$03.00
© 1989 Butterworth & Co (Publishers) Ltd Eng. Struct. 1 989, Vol. 1 1, April 75
Modal flutter analysis. T. d. A. Agar
of bridge response as follows:
t~ B LI

(a) loads that can be considered as static; !£ Z]


(b) foundation motion to simulate seismic action;
(c) time-dependent loads, e.g. to determine response to
wind gusting;
(d) natural frequency analysis. V m+o
A further module was developed to enable problems
to be analysed by a modal technique and a facility was
included that allowed experimentally measured bridge
deck flutter derivatives to be used, thus giving more
realistic aerodynamic interaction forces.
This paper describes some features of the modal
technique used in the program and gives examples of
flutter analyses that have been performed. This module
has been developed largely at Glasgow University (the Figure 1 Aerodynamic forces on oscillating flat plate
main body of the ANSUSP package having been written
by the author while working with Mott, Hay &
Anderson). Consider a flat-plate deck as shown in Figure I subject
to a smooth horizontal air flow. The section is assumed
to have two degrees of freedom, h, and = with uncoupled
Dynamic analysis vertical and torsional stiffnesses represented by kh and
k=. The forces Lh and M= are aerodynam',c lift and moment
Methods of solution acting on the section which is oscillating in the air flow
The equation of motion in terms of a nodal displacement at a circular frequency to in both degrees of freedom.
vector {U} for a structural system can be expressed in For the case of a thin aerofoil in incompressible flow,
conventional matrix notation as Theodorsen s used potential flow theory to derive expres-
[M]{O} + [c]{¢2} + [K]{U} = {P} (1) sions for Lh and M= per unit length of section:

where the matrices [M], [C], [K] and {P} have their usual L h = - ~pbE2FVh - bV(1 + F) +
meaning. For a linear elastic structural system, a com-
monly used technique for solving equation (1) is that of - 2VtoGh
modal synthesis. This has the computational advantage
that the response of a structure discretized as a large - {2V2 F - bVtoG}a] (2)
order n degrees of freedom configuration can be effectively
reduced to one with m degrees of freedom (m < < n), M , = - n p b [ b V F h + { b2v(1-F)2 bV__~G}~
where m is the number of modes included.
In general, there is a nonlinear relationship between - b VtoGh
forces and displacements due to significant changes in
the geometry of a suspension bridge as it deflects. Hence + { b 2 ~ toG bV2F}a] (3)
the stiffness matrix [K] is displacement dependent and the
modal synthesis technique is not strictly appropriate. An
alternative for solving equation (1) is numerically to where F[k) and G(k) are the real and imaginary parts of
integrate the equation of motion directly to give a the Theodorsen circulation function:
time-history response, and various methods 3-6 have C(k) = F(k) + iG(k) (4)
been available to perform this. One of these methods, the
Newmark fl method 6, vas initially used in modules of F and G are functions of the non-dimensional parameter
ANSUSP to predict general structural response including k = tob/V, which is the reduced frequency of vibration of
flutter. Details of this approach have been given else- the system. Added mass terms have been n e e d .
where 7. It has not been possible to develop expressions for
aerodynamic coefficients associated with bluff bodies
Nature of aerodynamic forces from basic fluid-flow principles. However, Scanlan and
Classical flutter is an aeroelastic phenomenon in which Tomko 9 have shown experimentally that for small dis-
two degrees of freedom of a structure, a rotation and a placements the lift and moment may be treated as linear
translation, couple together in a flow-driven unstable functions of h, = and their first two derivatives. These
oscillation. Coupling of the two degrees of freedom, have been expressed as
indispensable for flat-plate thin-aerofoil flutter has come
to be the identifying sign for classical flutter. Single degree L~ =½pV2(2B(KH * ~~+ KH* -~b~+K2H~a] (5)
of freedom flutter can be associated with bluff, unfaired
bodies where the flow is strongly separated. Some of the
older suspension bridge deck sections could give rise to
single degree of freedom flutter, but modern designs tend
to make the aerodynamics of the deck closely resemble
M==½pV2(2B2 KA* + KA*--~+ K2 A~ o~ (6)

where the flutter derivatives H~'. Af are functions of/~


1
that of a flat plate. an alternatively defined reduced frequency ~toB/V.

76 Eng. Struct. 1989, Vol. 11, April


Modal flutter analysis: T. J. A. Agar

These derivatives have also been determined experimen-


tally in Japan 1°'11 and in France 12. Hay Is has reviewed
current theories on bridge aerodynamics.
o,~"
Modal flutter analysis
For a suspension bridge subject to a transverse wind flow,
the right-hand side of equation (1) becomes:
{Vaero} = [ A ] { U } -{- [ B ] { ~ } (7)
where the coefficients of [A] and [B] arise from equations
(2) and (3) or equations (5) and (6). Applying the modal
synthesis technique leads to an m th order system in terms Figure 2 Suspensionbridge idealization in 3-D space
of generalized coordinates {~}:
+ [c*]{$} + = {0} (8) )
"~i+2
where
[C*] = [X]T([C] -- [B])[X];
[K*] = ([A] - [X]r[A][X])

and [X] is a matrix containing m natural modes and [A] x


is a diagonal matrix containing the corresponding cir-
cular frequencies squared.
Looking for a solution to equation (8) of the form
{~6} = {O}e ;aleads to the standard matrix eigenvalue form
of order 2m
Figure 3 Program internal nodal numbering scheme used in modal
[a]{z}: (9) analysis
where
bridge vibrations are considered to consist essentially of:

(i) x, y, 0x displacements of cable nodes;


and (ii) y, 0=, 0x displacements of deck nodes;
(iii) x, 0y displacements of tower top nodes;

M a t r i x organization
Solution of equation (9) yields a set of ,l and {O} l
With methods of predicting dynamic response that do
Omitting vector and matrix brackets hereafter, for a
not require explicit formation of a structural stiffness
complex conjugate pair of eigenvalues 2 = # +/to and
matrix, nodal numbering is consequently largely irrele-
,~ =/a - io~ and corresponding eigenvalues • = p + iq and
vant to computational efficiency. However, with the
• = p - iq the characteristic motion may be expressed as
modal approach the stiffness matrix [K] is formed
~b= ~'[(p + q)sin tot + (q - p)cos cot] (10) explicitly and it is advantageous to renumber the nodes
from the scheme used for the direct integration approach 7
To ensure dynamic stability in a system, the real parts # to that shown in Figure 3, where the half-bandwidth of
of all the eigenvalues should be negative. The onset of the stiffness matrix is limited to a maximum of 12 since
flutter instability occurs at the lowest wind speed, for there are three degrees of freedom per node. Renumbering
which an eigenvalue is complex and has a zero real part. is performed automatically without user involvement.
Divergence instability is indicated by a real eigenvalue The global stiffness matrix is assembled from contribu-
becoming zero. tions from the deck, hanger, tower and cable element
stiffness matrices, the last of these including a geometric
component is that accounts for the gravity stiffness of the
Structural ideafization and computer implementation structure. The inertial characteristics of the structure are
modelled by lumping of the member element mass at the
nodes.
Structural model Having formed the stiffness and mass matrices, the
The program ANSUSP idealizes a suspension bridge resulting free vibration eigenvalue problem is solved by
as a three-dimensional framework in a similar manner a simultaneous iteration algorithm 16 for a specified
to Iwegbue and Brotton t4. It is a two-cable idealization number of the lower vibration modes [X]. ANSUSP is
and comprises the types of structural elements shown in structured such that natural mode shape, frequencies and
Figure 2. Since the flutter phenomenon is normally all other pertinent information can be stored in user-
considered to involve motion in the x y plane, unaccom- defined binary files and later read back if required for
panied by an transverse motion, vertical and torsional subsequent analysis.

Eng. Struct. 1989, Vol. 11, April 77


Modal flutter analysis: 7. J. A. Agar
Modal flutter analysis until (Otrial, the circular frequency used to evaluate the
Having the natural modes available, the calculation of aerodynamic terms, agrees with the response of interest.
matrix [,4] of order 2m in equation (9) is straightforward, In practice, since the eigensolution of equation (9) yields
only the two upper sub-matrices C* and K* requiring m complex conjugate pairs of eigenvalues of the form
significant computation. In calculating C*, normally C, 2 i = #i + ie9i, there is a potential problem in knowing
the structural damping matrix, is conservatively taken as which one of the response o9~ the next trial frequency,
being zero. The eigensolution of equation (9) is affected ~ot,iaishould be aligned with. Initially, with very low wind
as follows: speeds, the response frequencies are almost equal to the
(1) the real unsymmetric matrix A is reduced to a matrix natural frequencies and the real parts of the eigenvalues
.4 of upper Hessenberg form by similarity transforma- are negative, indicating decaying motions. If the response
tions. The eigenvalues of .4 are the same as those of .4. frequencies are plotted on an Argand diagram they all
(2) The eigenvalues 2i of A (and hence A) are evaluated lie just to the left of the imaginary axis. When analyses
using the Double QR method, which is a computationally are carried out at successively higher wind speeds, all
more efficient extension of Francis's QR method 17. eigenvalues start to migrate along paths further away
(3) If required, any eigenvector {zi} corresponding to from the imaginary axis, corresponding to increasing
eigenvalue 2~ is then computed by a back substitution damping, while oscillation frequencies may increase or
process. decrease from the natural frequency values. Eventually,
at some higher wind speed, one of these complex eigen-
value paths turns and heads back to intersect the axis.
Characteristics of the solution This corresponds to an oscillation with zero damping
Since the method predicts eigenvalues and vectors de- and the onset of flutter instability.
fining the response for given wind speeds, investigation
of flutter stability involves a sweep through a range of
Severn Bridge analysis
increasing windspeeds until instability is indicated. Fur-
thermore, the solution of the eigenproblem as in (1)-(3) Figure 4 shows a structural idealization of the Severn
above at any given wind speed must be iterative in itself. Bridge where the left side, main and right side span
This is because the aerodynamic terms are functions of decking is represented by 5, 16 and 5 beam elements,
response frequency, and re-analysis has to be performed respectively. This idealization leads to a structural system
of order 215. With the properties as shown in Figure 4,
a natural frequency analysis in ANSUSP yields the first
13 natural frequencies and modal characteristics as given
Oiston~=between ¢
in Table 1. Table 2 shows the CPU times required to
f !llO'2m compute the lowest 13 and 30 frequencies and mode
shapes for a range of Severn Bridge idealizations. All
computation has been performed on a DEC microVAX
5 elements 8 elements
305m 488rn II computer.
Figure 4 Severn Bridge flutter analysis idealization and properties. If a modal eigenvalue analysis is performed, first
Structural properties as follows. Cables: CSA = 0.158 m2; mass = including only modes 2 and 8, i.e. the flutter response is
1338 kg/m. Hangers: CSA = 0.0146 m2; mass = 124 kglm. Towers: considered to be a combination of the fundamental
/ (longitudinal) = 2.13 m4; J (torsion) = 8.52 m4; E(steel) = 207 x
10s kN/m 2. Deck: I (vertical) = 1.079 m4; J (torsion) = 3.995 m4; symmetric flexural and torsional modes, then the Argand
mass = 10350 kglm; mass moment of inertia = 696000 kg-m2/m; diagram of Figure 5 is produced, where the vibration
overall deck width = 31.85 m. frequency at flutter (# = 0) follows a path starting at the

Table 1 Natural frequencies of 5-16-5 idealization of the Severn Bridge

Mode shape

Mode Natural Flexural Symmetric Displacements Designation a


no. frequency (F)/torsional (T) (S)/anti(A) in
(rad/s) Centre Side
span spans

1 0.801 F A ~, 1-A-F(C)
2 0.878 F S ~ ~/ 1-S-F(C,S)
3 1.251 F S 2-S-F(C,S)
4 1.372 F A 1-A-F(S)
5 1.698 F A ~ ~ 2-A-F(C)
6 1.718 F S 3-S-F(C,S)
7 2.221 F S ~, 3--S-F(C)
8 2.272 T S ~ -,~// 1-S-T(C,S)
9 2.755 F A 3-A-F(C)
10 3.039 F S - ~. 1-S-F(S)
11 3.039 F A - ~/ 2-A-F(S)
12 3.361 F S ~, - 4--S'-F(C)
13 3.580 T A ~/ 1-A-T(C)

aMode designation: e.g. 3-S-F(C,S) implies the 3rd Symmetric Flexural mode involving displacements in the Centre and Side spans

78 Eng. Struct. 1989, Vol. 11, April


Modal flutter analysis: T. J. A. Agar
Table 2 CPU times for natural frequency analyses for various

I
structural idealizations of the Severn Bridge
2.5
CPU time (s)
Number Mode8
of degrees
Idealization of freedom 13 modes 30 modes

3-8-3 107 55 184


5-16-5 215 112 344
8-26-8 359 183 627 vf = 76,9 m/s
10-32-10 449 240 763 {.0f = t . 4 9 r o d / s (O3triol following
12-36-12 521 276 887 mode6) •-I.o
.4- 0
17-54-17 773 414 1321 e-- ~" e~e,---e~ Mode 2

'0.5

2.5 -23 -o'.2 -o'., o "-


..,.4p---o~ Mode 8 Figure 6 Argand plot for Severn Bridge analysis with modes 2, 6
and 8. ( 0 ) represent plotted points for wind speeds V (m/s): 10,
f .2.0 20, 30, 40, 50, 50, 65, 70, 72.5, 75.0, 77.5, respectively

,.--e
I~ = 71.0m/s J Mode 12
~)f = 1.545 rod/s
3
Mode I0
.I.O 3.0
-in--- Q • O~O------O~
Mode2

2.5
Mode 8
Mode 7

-2.0
-o.3 -c .2 p
-o'., o
Mode6
Figure 5 Argand plot for Severn Bridge analysis with modes 2 and ,e 3
8. (O) represent plotted points for wind speeds V (m/s): 10, 20, Yf = 7 6 . 9 m / s Lof= 1 . 4 9 5 m / s ~
30, 40, 50, 60, 65, 70, 72.5, respectively
Mode :5

e ~ O • • •
'1.0
Mode 2
fundamental torsional frequency while the path of the
bending frequency is directly away from the o~ axis. Here
oJ,,l., has been set to follow the reduction in torsional -0.5
frequency. Including only these two modes, a flutter wind-
speed Vf of ~ 71.0 m/s and flutter frequency o~f of 1.545
rad/s is predicted. -0'.3 -o.2
' :~., o ---
Consider next the case shown in Figure 6, where the
Figure 7 Argand plot for Severn Bridge analysis with all symmetric
additional mode 6 has also been included. Here, the modes from Table 1. (Q) represent plotted points for wind speeds
correct solution corresponds to co, le, following the path V (m/s): 10, 20, 30, 40, 50, 60, 65, 70, 72.5, 75, 77.5, respectively
of mode 6 and not that of the torsional mode. Flutteya~ ~
predicted at Vf = 76.9 m/s and frequency o~f= 1.49 rad/s.
The prediction obtained by following the frequency of path does reverse direction sharply for V > 60 m/s and
mode 8 (broken line) can be seen to give a similar path subsequently crosses the axis.
to that of the correct solution: however, the intersection Now consider an analysis including modes 2, 3, 6, 7,
on the o~ axis is at Vf = 76.0 m/s and 1.53 rad/s. This is 8, 10, 12, i.e. all symmetric modes from Table I. For the
obtained with wt,~,l = 1.75, compared with 1.53 predicted; idealized (symmetric) Severn Bridge, the lowest flutter
consequently, the solution in this cause is relatively speed corresponds to a symmetric displacement pattern
insensitive to the value of o~memused to evaluate the with antisymmetric natural modes not contributing to
aerodynamic force coefficients. However, there is no the response. The Argand diagram is shown in Figure 7,
guarantee that this will be the case in general. Even for where again mode 6 frequency has been followed, and
this three-mode analysis full automation of a numerical the flutter prediction is again at Vf = 76.9 m/s and
procedure to produce the correct solution is not a trivial o~f --- 1.495 rad/s. Table 3 shows the flutter speed predic-
matter. For example, between wind speeds 40 and 50 m/s, tions and modal involvement in the flutter response
the mode path 8 a ~ e a r s to start to return to intersect obtained by including different combinations of modes
the o~ axis, but actually turns away from the axis for in an analysis. The conclusion to be drawn from a range
V> 60 m/s. In fact, O~tr~omshould follow mode 6, whose of the above types of analysis is that, in this case, the

Eng. Struct. 1989, Vol. 11, April 79


Modal flutter analysis." T. J. A. Agar
Table 3 Flutter speed predictions and modal involvement in C~_ ((
response for analyses with different combinations of symmetric
mode shapes

Mode shape Modal involvement as % flutter response

Flexural
2 100.0 85.4 65.7 57.3 59.3
3 - 14.6 - 9.8 9.1
6 34.3 30.7 31.6
7 - - - 2.1
10 - --0 --
12 - 0.1

Torsional I I I
0 l ~2 5 0 .~ 5 0 .~ 7 5 0 5
8 100.0 100.0 100.0 100.0 98.6
18 - - 1.4 Figure 9 Deck vertical response at flutter from modes 2, 3, 6 for
Severn Bridge analysis. Curves: A, t + (2/'/8); B, t + (3T/8); C,
~of(rad/s) 1.545 1.551 1.495 1.495 1.495 t + (T/8); D, t + (4T/8); E, t, t + T; F, t + (5T/8); G, t + (7T/8); H,
Vf (m/s) 71.0 70.9 76.9 76.9 76.9 t + (6T/8). T = oscillation Period = 2n/oh

Note -, indicates mode not included in analysis. Mode 18 is the


2nd symmetric torsional mode fundamental symmetric torsional (mode 8) and flexural
(mode 2) modes with Vf = 77 m/s obtained by including
flutter response consists of a combination of modes 2, 6 at least modes 2, 6 and 8, there may appear to be some
and 8, i.e. the first and third symmetric flexural and inconsistency. However, on examination the inference
fundamental torsional modes. These mode shapes are from the former analysis is that if the bridge were
shown in Figure 8, together with the second symmetric artificially constrained to allow flexural motion propor-
flexural mode, while Figure 9 shows the resulting flexural tional to the fundamental symmetric mode only, flutter
deck displacements predicted during an oscillation at could appear as predicted at 71 m/s. However, in reality
flutter. the response induced by the aerodynamic forces will be
the combination of all mode-shapes that the aero-
Discussion and comparison with alternative methods dynamics drive. The flexural response at flutter shown in
Figure 9 is consistent with that of Bleich I s, who reported
Initially, on comparing the flutter prediction of Vf = 71 the effect of incomplete matching between flexural and
m/s obtained by including only the similarly shaped torsional modes as increasing flutter speed as the initially
half-sine wave fundamental flexural mode develops a
reverse curvature.
Comparison of the flutter wind speed Vr = 76.9 m/s
obtained above with predictions by other methods are
I
shown in Table 4. Use of the numerical integration
time-history module 7 in ANSUSP (which uses exactly
Mode 2 ( l - S - F) the same formulation for the aerodynamic force coef-
ficients) produces a value in the region of 77.5 m/s with
a flexural response showing a reverse curvature very
similar to that shown in Figure 9. Indeed, it is pertinent
/
to note that although the time integration method was
./ started with initial deck velocities proportional to the
r-..

MOde 3( 2 - S - F ) Table 4 Comparison of flutter windspaed predictions for the


Severn Bridge by various methods

Flutter speed
fl .~, Method Reference (m/s)

Multi degree-of-freedom, modal This method 76.9


I
Multi degree-of-freedom, Agar 7 77.5
Moas 6 (a-s-~) ¢- time-history
Two degrees-of-freedom, - 65.6
time-history (by author)

Mode8 ( I - S - T )
f T Two degrees-of-freedom, modal
(by author)
Two degrees-of-freedom,
experimental and calculation
-

Smith la
65.6

65.7

(Smith-NPL)
Figure 8 Natural modes 2, 3, 6 and 8 for Severn Bridge analysis
(tower displacements shown at magnification of 5 compared with Selberg semi-empirical Selberg 2° 68
vertical deck displacements)

80 Eng. Struct. 1989, Vol. 11, April


Modal flutter analysis: T. J. A. Agar
fundamental flexural and torsional mode shapes, the stability of the 5-16-5 idealization was assessed at 15
flutter condition was not excitable at 70-71 m/s and not wind speeds, including seven natural modes, the total
with a resulting response comprising essentially only the CPU time involved (including the initial computation of
fundamental flexural and torsional modes. the lowest 30 modes) is ,-, 420 s. Alternatively, including
The other predictions in Table 4, based on modelling 13 and 20 natural modes in the flutter analyses would
flutter behaviour with only two degrees of freedom (one require ~ 600 and 1030 s respectively.
vertical and one torsional), yield consistently lower flutter By comparison, the multi-degree-of-freedom time-
speeds. The two of these predictions carried out by the history integration method with the same 5-16-5 struc-
author again used exactly the same Theodorsen coef- tural idealization 7 required ~ 1500 CPU s per wind speed
ficients as the multi-degree-of-freedom modal and time considered. This amounts to 22,500 CPU s for 15 such
integration methods. This represents further substantia- wind speed investigations. Consequently, use of the modal
tion for the significant difference in the flutter predictions flutter analysis method described in the paper represents
obtained from multi-degree of freedom and two-degree- a saving in computation by a factor of between 20 and
of-freedom modelling. 50 over the time integration approach for comparable
It should be noted that the multi-degree-of-freedom accuracy.
modal analysis described here predicts a divergence wind
speed of 74.7 m/s, which agrees with that obtained from Conclusions
two-degree-of-freedom idealizations. The divergence
solution was obtained by using COtri=, = 0 in the aero- A method of computing the flutter speeds of suspension
dynamic coefficient evaluation. Consequently, the full bridges has been presented. The method uses a multi-
16% increase in flutter speed from -~ 66 m/s from two- degree-of-freedom modal technique and includes the
degree-of-freedom to -~ 77 m/s from multi-degree-of- aerodynamic force interaction in the governing equations
freedom analyses would not, in practice, be completely of motion to produce an unsymmetric eigenvalue prob-
realized since the static divergence instability becomes lem. The characteristic roots of the matrix involved are
the governing phenomenon. indicative of the nature of possible motions that can exist
at a given wind speed. This has been used to identify the
flutter condition, which involves a divergent oscillation
Computational e~ciency involving both vertical and torsional deck motion. Com-
Table 5 indicates the CPU times required to perform parison with two-degrees-of-freedom methods indicates
modal flutter analyses (having previously computed and that there may well be a significant conservatism in
stored the natural modes, see Table 1) with various methods that assume the flutter response to be dependent
numbers of natural modes included for the 5-16-5 on only the fundamental flexural and torsion natural
idealization of the Severn Bridge. Normally, on average, modes. The modal method described is demonstrated to
approximately three analyses need to be carried out for be computationally much more efficient than an alterna-
each wind speed considered to allow convergence of O~triam tive time-history integration method, but yields com-
and the response frequency of interest. Very modest CPU parable flutter speed prediction accuracy.
times are required for analyses including up to ~ 12
natural modes. Although, in retrospect, it can be recog- Acknowledgements
nized that in this case flutter speed prediction accuracy
is not significantly improved by including any more than The author acknowledges that most of the initial program
modes 2, 6 and 8, it would be prudent in practice to development of ANSUSP was carried out at Mott, Hay
perform analyses with the first 12 symmetric modes for the & Anderson, Consulting Engineers, London, UK.
symmetric Severn Bridge idealization considered, or more
if an unsymmetric bridge was being investigated. If the References
1 Bridge Aerodynamics. Proposed British design rules, Thomas
Telford Ltd, London. 1981
Table 5 CPU times for modal flutter analyses on a 5-16-5 2 Agar, T. J. A. Analysis of suspension bridges--programme ANSUSP
idealization with various numbers of natural modes included user guide, Mott Hay & Anderson Computer Department, 1980
3 Wilson, E. L. and Clough, R. W. 'Dynamic response by a step-by-
step analysis'. Prec. Symp. on the Use of Computers in Civil
CPU time (s) Engineering, Lisbon, 1982
4 Bathe, K. J. and Wilson, E. L. 'Stability and accuracy analysis of
No. of Size of direct integration methods', Earthquake Eng. Struct. Dynamic,
modes Mode nos eigenvalue for one for one 1973, 1, 283-291
included problem eigensolution wind speeda 5 Hilbcr, H. M., Hughes, T. J. R. and Taylor, R. L. 'Improved
numerical dissipation for time integration algorithms in structural
2 2,8 4 x 4 0.60 1.8
dynamics', Earthquake Eng. Struct. Dynamic, 1977, 5, 283-292
6 Newmark, N. M. 'A method of computation for structural dy-
3 2,6,8 6 x 6 0.75 2.2
namics'. Prec. ASCE J. 1959, 68, (EM3), 67-94
5 2,3,6,8, 10 x 10 1.30 3.9
10
7 Agar, T. J. A. 'The analysis of aerodynamics flutter of suspension
bridges', Computers and Structures 1988, 30, 593-600
7 2,3,6,7, 14 x 14 1.85 5.6
8,10,12
8 Theodorscn, T., 'General theory of aerodynamic instability and the
13 1 . . . . . 13 26 x 26 5.70 17
mechanism of flutter', NACA Technical Report, No. 496 1935
20 1. . . . . 20 40 x 40 15.4 46
9 Scanian, R. H. and Torako, J. J. 'Airfoil and bridge deck flutter
30 1 . . . . . 30 60 x 60 44.3 133
derivatives'. Prec. ASCE J. 1971, 97, (EM6), 1717-1737
10 Okubo, T. and Narita, N. 'A comparative study of aerodynamic
forces acting on cable-stayed bridge girders'. Prec. 2nd. US-Japan
mConvergence of o~a aand response frequency requires, on average, Seminar on Wind Effects on Structures, Kyoto, University of Tokyo
approximately three eigensolutions at each wind speed Press, 1976, pp 271-283

Eng. Struct. 1989, Vol. 11, April 81


Modal flutter analysis. T. J, A. Agar
11 Okubo, T. and Yokoyama, K. 'Some approaches for improving Hill Book Co., New York, 1968
wind stability of cable-stayed bridges', Proc. 4th Int. Conf. on Wind 16 Corr, R. B. and Jennings, A. 'A simultaneous iteration algorithm for
Effects on Buildings and Structures, London. 1985, Cambridge symmetric eigenvalue problems', Int. J. Num. Methods Eng., 1976,
University Press, 1986, pp 241-249 10, 647-663
12 Loiseau, H. and Szechenyi, E. 'Etude du comportment aero~lasti- 17 Francis, J. G. F. 'The QR transformation, parts I and II', Computer
quedu tablier d'un pont d haubans', T. P. 1975-75, Office National J., 1961 4, 265-271, 332-345
d'Etudes et de Recherches Aerospatiales, Chatillon, France 18 Bleich, F. 'The flutter theory' The mathematical theory of vibration
13 Hay, J. S. 'An introduction to some current theories on the in suspension bridges (Bleich, F., McCullough, C. B., Rosecrans, R.
aerodynamic behaviour of bridges', Transport and Road Research and Vincent, G. S., eds), US Government Printing Office,
Laboratory, TRRL Supplementary Report 542, 1980 Washington, DC, 1950, Ch. 7, pp 241-281
14 Iwegbue, I. B. and Brotton, D. M. 'A numerical integration method 19 Smith, I. P. 'The aeroelastic stability of the severn suspension
for computing the flutter speeds of suspension bridges in erection bridge', NPL Aero Report 1105. 1964
conditions', Proc. ICE (Part 2), 1977, 63, 785-802 20 Selberg, A. 'Oscillation and aerodynamic stability of suspension
15 Pr-zemieniecki, J. S. Theory of matrix structural analysis, McGraw- bridges', .4cta Polytech. Scand., Civil Eng. 1961, No. 13,

82 Eng. Struct. 1989, Vol. 11, April

Potrebbero piacerti anche