EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

19,45 -62 (1990)

NON-LINEAR EARTHQUAKE-RESPONSE ANALYSIS OF

LONG-SPAN CABLE-STAYED BRIDGES: THEORY

ALY S. NAZMY*
Civil Engineering Department. Polytechnic University, 333 Jay Street. Brooklyn, New York 11201, U.S.A.

AND

AHMED M. ABDEL-GHAFFAR'
Civil Engineering Department. University of Southern California. Los Angeles. California 90089-0242, U.S.A.

SUMMARY
The non-linear dynamic analysis of three-dimensional long-span cable-stayed bridges when subjected to seismic loading is
formulated. All possible sources of non-linearity, such as cable sag, axial force-bending moment interaction in bridge
towers and girders and change of geometry of the whole bridge due to large displacements are considered in the analysis.
Both cases of uniform and multiple-support seismic excitations are considered in the non-linear formulation of the
problem. A tangent stiffness, iterative procedure is utilized to capture the the non-linear seismic response. The non-linear
equations of motion are solved using a step-by-step integration technique in the real displacement coordinate space as
well as in the modal coordinate space to save computational time.

INTRODUCTION
The future trend in the design of cable-stayed bridges to longer centre or effective spans makes non-linear
analysis inevitable.' - 4 This is essential not only for evaluating the stresses and deformations induced by
environmental loads, such as vehicular traffic, wind and earthquakes, but also for assuring safety during
construction. Since the non-linearity in the behaviour of this special type of flexible, long-span bridge is of
geometric type, and mainly due to large deformations, it is essential to point out that when the centre-span
length increases, which will result in a considerable increase in the displacement and deformations of the
bridge under strong shaking, a pronounced non-linearity in the response may be expected. Therefore, the non-
linear behaviour of cable-stayed bridges, keeping the future range of spans in mind, should be investigated,
and the validity of the linear, dynamic response assumption (see Figure I), which prevailed for several year^^,^
must be re-examined.
In this paper, a general step-by-step integration technique is presented for the evaluation of the dynamic
response of geometrically-non-linear cable-stayed bridges subjected to strong ground shaking. The structure
is discretized in space into finite elements, mainly beam-column elements and cable elements, and the Wilson-
6 method.',* is used for the time discretization, to assure numerical stability of the algorithm for all time
increments. Direct integration in the real displacement coordinate space, and, alternatively, integration in the
modal coordinate space, using the normal mode shapes as an orthogonal basis for the purpose of coordinate
transformation, are used to integrate the incremental equations of motion. The latter approach takes less
computation time than the former, when the modes of vibration used for the coordinate transformation do
not need to be recomputed at the beginning of each time step, which is the case for systems with mild non-
linearity.

*Assistant Professor.
'Professor.

0098-8847/90/0l0045-18$09.00 Received 13 October I988

0 1990 by John Wiley & Sons, Ltd. Revised 20 June 1989
46 A. S. NAZMY AND A. M. ABDEL-GHAFFAR

NON-LINEARITIES IN CABLE-STAYED BRIDGES

In spite of the fact that the behaviour of the material of the structural elements in a cable-stayed bridge is, in
general, linearly elastic, the overall-load-displacement relationship for the structure is non-linear under
normal design loads.'- l 1 This overall non-linear behaviour originates from three primary sources. Previous
investigators5v6* 9*l o * ' * have proposed that these three sources are:

(i) the non-linear axial force-elongation relationship for the inclined cable stays due to the sag caused by
their own weight;
(ii) the non-linear axial force and bending moment interaction for the tower and longitudinal girder
elements; and
(iii) the geometry change caused by the large displacements which can occur in this type of structure under
service as well as environmental design loads (Figure 1).

In the present study, all the above mentioned sources of non-linearity are taken into consideration.
The non-linear behaviour of inclined cables due to the sag effect is accounted for by using the concept of an
equivalent modulus of elasticity for an equivalent straight chord member. This modulus combines the effects
of material and geometric deformations such that the axial stiffness of the equivalent chord member becomes
equal to the apparent axial stiffness of the actual curved cable. This equivalent modulus is a function of the
cable tension and the horizontal projected length of the cable which are time-dependent and should be
recomputed at the end of each iteration cycle during the non-linear dynamic analysis.
The non-linear behaviour of the tower and girder elements due to axial force-bending moment interaction
is accounted for by using the stability functions to modify both the bending and axial stiffnesses of the element
at the end of each iteration cycle. A detailed description of the non-linear stiffness formulation of cable and
beam-column elements is given by the authors in References 1 and 13.
The third source of non-linearity in cable -stayed bridges, namely the overall geometry change due to large
deformations, is accounted for by updating the bridge geometry by adding the incremental nodal
displacements to the previous nodal coordinates at the end of each iteration cycle before recomputing the
stiffness of the bridge in the deformed state.

NON-LINEAR DYNAMIC ANALYSIS PROCEDURE

Among the many techniques available for non-linear dynamic analysis, probably the most efficient is the step-
by-step integration procedure. 5 - 2 0 In this approach, the response is evaluated at successive increments of
time At, usually taken to be of equal length for computational convenience. At the beginning of each time

CABLE -STAYED BRIDGES

Generalized
Force Linear

Dynamic

--- - - - - .- - - -
5.1 STATIC DEFORMATION
DUE TO D . L .

Static
I
I
I

Yo.,. Generalized
Displacement
-
Figure 1. Non-linear force-displacementrelationshipsfor cable-stayed bridges
 LONG-SPAN CABLE STAYED BRIDGES THEORY 47

interval, the condition of dynamic equilibrium is established. The response for an incremental time step At is
then evaluated approximately, on the basis that the structural properties remain constant during the interval
At. The non-linear nature of the system is accounted for by re-evaluating the structural properties at the end of
the time step, to be appropriate to the current deformed state at that time. Iterations are then made to achieve
the state of dynamic equilibrium at the end of this time step and the computed velocities and displacements
are then used as initial conditions for the next interval; thus the process may be continued step by step from
the initiation of loading to any desired time.
There are many procedures available for pertorming the step-by-step integration of the incremental
equations of motion.5'7'8,12~15-17~21 The method used herein is the Wilson-8 m e t h ~ d ~ , ' * ' in~ *which
'~
numerical stability of the solution process is assured regardless of the magnitude selected for the time step; for
this reason, this method is said to be unconditionally stable.
It should be mentioned at this point that in the earthquake-response analysis, it is important to use an
unconditionally stable method, because the time step size At can then be selected giving regard only to the
required accuracy in the low-mode response; i.e., the time step size need not be small enough to satisfy a
stability criterion. It is for this reason that the Wilson-6 method was used for the present non-linear
earthquake-response analysis.
The basic assumption of the Wilson-6 method is that the acceleration varies linearly over the time interval
from ti to ti + $At, where 8 2 1. Wilson et ai.' have shown that, for 6 2 1.37, the method becomes
unconditionally stable, and the method is most accurate when 6 is about 1.4.'

INCREMENTAL FORM OF THE EQUATIONS O F MOTION

Assuming constant mass and damping matrices, while the stiffness of the finite-element model changes as a
function of nodal displacements, the equations of dynamic equilibrium of the bridge at the time ti are given, in
a matrix form, by
[MI { i;(ri)} + [CI {Y(ti)) + CK(yi)I {y(ti)>= { p ( t i ) > (1)

in which [MI and [C] are the mass and damping matrices of the finite element model, [K(yi)] is its stiffness
matrix, which is a function of the nodal displacements at time ti, { y(ti)}and { P(ti)}are the vectors of dynamic
nodal displacement and externally applied dynamic nodal forces at time t i , respectively.
At a later time T , where T =$At, the equations of dynamic equilibrium are given by
[ M I { j ; ( t i + ~ ) )+ C c I { j ( t i + ~ ) +
} CK(y(ti+T))I{~(ti+t)}={P(ti+T)} (2)
where [ K ( y ( t i + T))] is the stiffness matrix at time ti + t,which is a function of the nodal displacements at that
time, and {y(ti+ T ) } is the vector of dynamic nodal displacements at time ti T . +
Subtracting equation (1) from equation (2) results in the incremental equations of motion, namely:
[MI {aj;i> + CCI { a j i )+ CK,(Y~)I
{aYi>={aPi> (3)
In the above equation, the hat over A indicates that the increments are associated with the extended time step
t = 8At; thus,

It should be noted that [K,(yi)] is the tangent stiffness matrix of the finite element model at time t i , which
is a function of the nodal displacements at time t i , and is assumed to remain constant during the increment of
time T.
48 A. S. NAZMY AND A. M. ABDEL-GHAFFAR

Equation (3) involves some approximation, since the incremental elastic force vector { 8 F r } is computed
from
{ aF;} = C K T ( y i ) l { a y i > (5)
where the tangent stiffness [KT] is used instead of the secant stiffness (see Figure 2). This is done because the
secant stiffness cannot be determined until the displacement at the end of the time step is computed. Therefore,
depending upon the size of the time step and the non-linearities present in the system, there will be a residual
force vector at the end of the extended time step, to be computed as follows:
{ p'(ti -k T)} = { p(ti T)} - [ M] { y ( t i -k T)} - [c]{ j(ti 4- T)} - { F'(ti -k T)} (6)
The exact evaluation of { F'(ti + T)} of the above equation is explained later.
The value of the residual force vector given above may not be negligible, in which case iteration should be
carried out in each time step until convergence to some acceptable tolerance is achieved.
To avoid accumulation of small errors in the numerical solution, it is usually preferable to add the last
residual force vector computed at the end of the previous time step, after convergence is satisfied, to the
incremental dynamic load vector when solving for the incremental displacements. Thus, the incremental
equations of motion [equations (3)] become
[ M ] { ~ Y i } + [ c l { a ~ i }+ [ K T ( y i ) l { a y i ) ={api} +{ P i } ~ ~ M ~ { ~ i } ~ ~ C ~ { ~ i (7)
} ~ { F ~ }

EVALUATION OF THE DYNAMIC AND ELASTIC FORCE VECTORS FOR NON-LINEAR

SYSTEMS
The dynamic force vector { P ( t ) }
The equations of motion of the three-dimensional vibration of the bridge when subjected to multiple-
support seismic excitations at the two anchor piers and the two tower bases (Figure 3) can be expressed in a
partitioned matrix form as1*22*23

where the subscript 'g' designates the degrees of freedom corresponding to the points of application and
directions of ground motions (with total number q degrees of freedom), and the subscript 's' corresponds to all
other structural degrees of freedom of the bridge model (with total number N degrees of freedom). [M,,],
[ C,,] and [ K , , ] are rectangular mass, damping and stiffness matrices, respectively of order N x q. These
rectangular matrices represent the coupling between the structural nodes not connected to the ground, and
the support displacements due to seismic motion.

Fe

Fe(t ,+z)
m
r
F:
U
-we
.-

Generalized Displacement

Figure 2. Tangent and secant stiffness of the non-linear elastic force-displacement

 LONG-SPAN CABLE STAYED BRIDGES: THEORY 49

3-D CABLE - STAYED BRIDGE

MULTIPLE EARTHQUAKE
INPUTS

Figure 3. A cable-stayed bridge model subjected to three-component, multiple earthquake inputs

Assuming that the displacements can be decomposed into vibrational and pseudo-static displacements,
then

and substituting from equation (9) into equation (8), one obtains
r ~ s s l ( { w+ {JPS})+Wsg1{jpg}+ [ c s s l ( { 3 v s ~+ ~ 3 p , ~ ) + C C s g l { 3 p n ~
+C ~ S S l ( ~ Y +
V S(YPS>)
~ + tK,,l {Ypg) = P I (10)
In the above equations, { y,,} is the vector of ground displacements at the supports of the structure. The elastic
force computed in the above equation as [K,,] ( { y,,} + {y,,}) can be viewed as the sum of two parts,
vibrational elastic force (due to vibrational displacement), and pseudo-static force due to support displace-
ment (see Figure 4). This yields:
CK,,(Y", + Yps)l({Yvs} + {Y,,)) = CK,,lI { Y V J + CK,,21 (Y,J (1 1)
Substitution from equation (1 1) into equation (10) gives
+
CMSSI {jivs> CCSSl {YVJ + CKSSII {YW) + C ~ s s z {lY p s l + CK,,I {Yp,}

=- ~ ~ s s l ~ ~ p s ~ - ~ ~ s g l { C~,,1{3p,}-C~,,1{3,,}
Y p g } - (12)
For an unloaded structure with a static condition of support displacements the equilibrium is expressed by
CKssZ1 ( Y p J + [K,,l {Ypg} = (01 (13)
Substitution from equation (13) into equation (12) gives
+ CC,,I {3,J + CKSs13{YVJ = {w>
CM,,I {jVJ (14)
where
( p ( t ) > =- r ~ s s l { -~[K,I{Y,,)
p s ~ - ~ ~ s s l ~ -3 CCS,1{3,,~
ps} (15)
It should be mentioned at this point that equation (14) is the same as equation (l), which means that the
dynamic displacement that appeared in equation (1) is actually the vibrational part of the total displacement;
50 A. S. NAZMY AND A. M. ABDEL-GHAFFAR

=e/A 3 - D CABLE STAYED BRIDGE

~

% LATERAL GROUND MOTION .' y

d

Figure qa). A qualitative example of the pseudo- or quasi-static displacement

t Generalized
Force

Generalized
k Y0.L.
T
1
yv,
'1
(YV,+Yd
Yp.

d
Displacement

Figure qb). Decomposition of the dynamic elastic force into vibrational part and pseudo-static part in the non-linear dynamic analysis
of cable-stayed bridges

it also means that the stiffness matrix (whether the secant stiffness used in the equation of dynamic
equilibrium, or the tangent stiffness used in the incremental equations ofmotion) is a function of the dead-load
displacement plus the vibrational displacement (see Figure 5). Thus, the pseudo-static displacement is not
included in computing the stiffness.
To simplify the expression of the dynamic force given in equation (15), the expression of the structural
displacement due to support motions can be utilized.
Substitution from equation (1 3) into equation (1 5 ) yields
{ P ( t ) }= ~ CK,,I - [Msgl){Ypg}+(Ec,slCKssZ1-l [K,,I-
~ ~ s s 1 ~ ~ s s 2 1 - 1 [Csg1){jpg} (16)
Generally, the contribution due to the damping term in the above equation is neglected; also, if a lumped-mass
idealization is used, [M,,] becomes a null matrix, in which case equation (16) reduces to
 LONG-SPAN CABLE STAYED BRIDGES: THEORY 51

(D.L. + VlBR \ DISPL

Figure 5(a). Evaluation of the elastic force vector

( NONLINEAR ANALYSIS )

Figure 5(b). Static deformation due to dead-load equilibrium position

It can be noted from equation (13) that the expression for {yPs) contains
- CKss21r1CKS,1 = CCpsl (18)
where [G 3 is the matrix of pseudo-static functions, whose columns represent the static displacement
ps .
corresponding to unit displacement of the supporting points [Figure qa)].
It can be assumed, at this point, that these pseudo-static functions given in equation (18) remain constant
during the earthquake-response excitation, and that they are not affected by the change in the stiffness of the
structure. Such an assumption is possible since the changes in [Kss2] - and [ K , , ] can cancel each other when
they are multiplied, and since these pseudo-static functions are normalized functions, whose maximum values
equal unity.
The above assumption makes it possible to avoid inverting the stiffness matrix [K,,2] at the beginning of
each iteration cycle, and thus saves a considerable amount of computational time. Based on this assumption,
substitution from equation (18) into equation (17) gives
{W)>
=-c~ssl~Gpsl~Yp,~ (19)
Equation (19) gives the forcing function for the multiple-support seismic excitation case. For the uniform
earthquake input case, this equation can be further simplified to become
ow=-CMssIW (20)
where { g } is the vector of the input ground acceleration (it is of order N ) , whose elements are given by the x-,
y- and z-components of ground acceleration.
It should be mentioned at this point that there is no approximation involved in the case of uniform
earthquake input, since the pseudo-static displacement in this case is represented by a rigid body motion of the
52 A. S. NAZMY A N D A. M. ABDEL-GHAFFAR

structure. Accordingly, neither the stiffness nor the relative displacement will be affected by this kind of
motion (in this case, K,, and Kssl in Figure 4(b) are identical).
There is yet an alternative approach, which involves no approximation, for solving equation (8) without
decomposing the displacements into vibrational and pseudo-static parts.' Since, it is important to separate
the vibrational response from the pseudo-static response in order to examine the dynamic or inertial as well as
the kinematic effects, and in order to compare the results with those of the linear dynamic analysis, the
approach described herein is utilized.

The elastic force vector { Ff }

It is possible to calculate the elastic force vector { F r } by simply adding up the incremental force changes
{AFe}.However, because the stiffness matrix [KT] is, in general, only an approximation, significant errors
can accumulate in this procedure.
Therefore, it is preferable to compute the elastic force vector {Ff} from the dead-load (D.L.) displacements
plus the vibrational displacements at time ti as follows [see Figure 5(a)]:
{Fi'}= CK~(YD.L. + Y ~ ) I ( Y D L +Yi)-
. {FL.,.) (21)
in which [K,(y,,,. + yi)] is the secant stiffness based on the joint displacement and member forces due to both
D.L. and vibrational displacements.
This process is usually carried out on the element level, where the member-end actions are computed for
each element based on the D.L. and vibrational displacements using the secant stiffness of the element at this
particular deformed state. Then, for each node, these internal forces are assembled and projected on the
system global coordinates (X, Y, Z), for all the elements joining at this paiticular node. By inverting the sign
of these assembled forces for all the nodes, and subtracting from them the externally-applied dead-load forces
at the joints, the elastic force vector due to dynamic displacements, {Ff}, is obtained.

SOLUTION OF THE EQUATIONS OF MOTION USING STEP-BY-STEP INTEGRATION

The numerical integration of the non-linear equations of motion is executed utilizing the Wilson-%method.'
In this method, the acceleration may be represented by a linear function during the time step T = %At,This
yields

in which {8ji}is given by equation (4c).

Integration of equation (22) twice between the limits ti and t yields

and

Evaluation of equations (23) and (24) at the end of the extended interval t = ti + T gives
{ A Y i )= { j i } T + T {
1 A-Yi}T (25)

in which { h y , } and { A j i }are defined by equations (4a) and (4b), respectively. If equation (26) is solved for the
incremental acceleration { a j i } and substituted in equation (25), one obtains
6 6
{&J= - p { A y i } - - { j i ] - 3 { y i ] (27)
T
 LONG-SPAN CABLE STAYED BRIDGES: THEORY 53

and
3 t
@ji} = - - T{ & y i } - 3 { j i } - - {2j i }

Linear system
The tangent stiffness matrix of the structure at the dead-load deformed state [ K T ] [see Figure 5(b)] is
assumed to be constant throughout the linear dynamic analysis: hence, the residual force vector given by
equation (6) equals zero, and equation (7) reduces to equation (3), which involves no approximation.
Substitution from equations (27) and (28) into equation (3) gives

which can also be written as

[ R ]{ A y , } = { & P i }
In the above equation,
6 3
= C K T l + f [ M 1 +;[Cl
= effective stiffness
and

= effective incremental load vector

Equation (30) is then solved for { & y j } ;no iteration is required since [KT] is constant.

Non-linear system
The stiffnessmatrix [KT] for the non-linear dynamic analysis of cable-stayed bridges is computed once at
the beginning of each time step and used during iterations in this time step until convergence is achieved to
some acceptable tolerance. The modified Newton-Raphson technique is used in the iterative procedure to
achieve equilibrium (to some tolerance) at the end of each time step. Substitution for equations (27)and (28)
into equation (7) gives

and

{ api}= O({P ( t i + A t ) } - { P ( t i ) } ) +{ P i } + [ M I
1
2{ j i } + { j i } - { Ff} (34b)
54 A. S.NAZMY A N D A. M. ABDEL-GHAFFAR

Equation (33) is then solved for { Ayi}. Due to the fact that this equation involves an approximation (since
[KT] is assumed to be constant during the entire time step), the residual force vector computed at the end of
the extended time step, using equation (6), may not be negligible. To compute this residual force, the
displacement, velocity and acceleration at time ( t i + z) must be computed first, as follows:
equation (4a) * { y ( t i + T I } = { y ( t i ) }+ { A y i } (354
3 5
equations (4b) and (28) 3 { j ( t i + T)} = - { A y i } - 2 { j i }- { j i } (33-4
5

6 6
equations (4c) and (27) * { y i ( t i + T)} = T"{ 8 y i } - { j i } - 2 { yi} (354

Then the residual force vector is computed from equation (6):

{Pr(ti+ T)} = {pi} + e({Pi+ I } {Pi})- [ M I { j ; ( t i +TI}- [CI { j ( t i + t > }- { F e ( t i + T)}
- (36)
In order to compute the correlation to the displacement increment {Aayi}, the above residual force vector is
applied to the right hand side of equation (3) as a dynamic force vector; then equation (3) gives
[MI { ~ A j i i }+ CCI { AAji} + CK,(Y~)I {AAyi} = {Pr(ti+ z)} (37)
where { A A y i } and { A a j i } are the corrections to the incremental acceleration and velocity due to the
correction in the incremental displacement { AAyi}. Knowing that { ji}and {yi}have no corrections, one can
obtain the corrections for the incremental acceleration and velocity from equations (27) and (28), respectively,
as follows:
6
equation (27)* { A8jii} = t'{ A\Byi} (384

3
equation (28)=. { A A j i ) = (38b)

Substitution from equations (38a) and (38b) into equation (37) results in
[ K i ] {AAyi} = { P r ( t i+ T)} (39)
where [K,]is given by equation (34a).
Solving equation (39) for {AAyi}, one can correct the incremental displacement as follows:
{ I)} = {Ayy)}+ { AAyy)} (40)
in which (n)represents the number of the iteration cycle. One should repeat the computation of {AAyy)} until
convergence to some acceptable tolerance is achieved; at this point,
{AYi} = {Ayi"+')}

Linear and non-linear systems

Having computed the incremental displacement { a y i } at the end of the extended time interval, one can use
equation (27) to compute { h j i } .

The incremental acceleration { A j i } for the normal time interval At is then obtained by a simple linear
interpolation. Hence,
1
{ A j i } = { Ajii}
~
(42)
0
 LONG-SPAN CABLE STAYED BRIDGES THEORY 55

To calculate the incremental velocity { A j i } and the incremental displacement {Ayi} corresponding to the
normal interval Ar, use is made ofequations (25) and (26) with the extended time interval parameter T replaced
by At; that is,

1
equation (25)a{ A j i } = {yi}At+-{Aji}Af (43)
2

equation ( 2 6 ) a { Ay,} = { j i }At + -21 { j i }(At)2 + 61 { Ayi} (At)2

- (44)
Finally, the displacement, velocity and acceleration at the end of the normal time interval are calculated by
{yi+il = {Yi} + (AYiI (453)
{ j i + 1 1 = { j i >+ {Aji) (43.4
{ j i i + l } = { j i i } + {Aji} (454
Once the displacement, velocity and acceleration vectors have been determined at the time t i + ( = t i + At), the
above procedure is repeated to calculate these quantities for the next time step a t time = t i + l + At) and
the process is continued to any desired final time.

CONVERGENCE CRITERIA
The criterion used to check the convergence of the numerical solution is the ratio of the final residual force
vector to the incremental force vector computed in the beginning of the time step, where both vectors are
measured in some specific vector norm, i.e.,

~ ~ ~ ' ( " ' ( ~ i <+ Tol,

~ ) ~ ~ ~ / ~ ~ ~ ~ i ~ ~ ~ (46)

where Tol, is the tolerance for force convergence, usually taken to be less than or equal to 0.01.
At the same time, the ratio of the final correction for the incremental displacement vector to the final value
of the incremental displacement vector, both measured in a vector norm, should also be checked; i.e.

11A15,yl"'II"IJdyl"'''llE < Tol, (47)

where Tol, is the tolerance for displacement convergence, usually taken also to be less than or equal to 001.
In equations (46) and (47), llallEis the Euclidean norm given by

IlallE = J.. j= 1

THE USE OF MODE SHAPES IN SOLVING THE NON-LINEAR DYNAMIC PROBLEM

The solution of equation (7) for non-linear systems, using a direct step-by-step integration procedure, requires
the iterative solution of a non-linear system of algebraic equations [equation (33)], equal in order to the
llurnber of degrees of freedom (DOF) in the finite element model. This may require the solution of a very large
order system. Therefore, the modal response method is utilized to replace the original system of equations for
the structure nodal displacements by a smaller system of equations for a system of generalized
coordinates. 5 . t 2.2 1 , 2 4 - 2 6
In principle, the use of mode shapes simply involves a coordinate transformation from the finite element
displacement coordinates to the modal coordinates. This change of basis is particularly effective if only a few
modal coordinates need to be employed.
56 A. S. NAZMY A N D A. M. ABDEL-GHAFFAR

Mathematical formulation
Consider again the equation of dynamic equilibrium at time t i in its matrix form, equation (l), and define
the modal transformation
{Yi> = [@I{qi} (49)
where { qi} is the vector of generalized coordinates at the time ti; and [O] is the matrix of mode shapes,
obtained from the solution of the eigenvalue problem
K71- wz CMll{41= (0) (50)
In the above, [KT] is the tangent stiffness matrix of the structure in the dead-load deformed state (Figure 5).
If N is the order of { y,}, and [0] is an N x p matrix whose columns are the lowest p mode shapes of the
structure, then p is the number of generalized coordinates. The method works well if p can be taken to be much
less than N without a great loss of accuracy in the solution.
The substitution of equation (49) into equation (l),and premultiplication of the resulting equation by [@ITT,
reduces equation (1) to the form
[MI ( i i i ) + CCl {4i} + [Eil {qi)= {Qi) (51)
where

and
{Qi1 = [@IT{Pi1 (524
In the above, [A?] is a diagonal matrix of generalized masses M , ; [C] is a diagonal matrix of generalized
damping coefficients 2(,w,M,; and [Ki] is the generalized stiffness matrix, which is not diagonal for non-
linear systems.
Equation (51) represents the equation of motion in the modal coordinate space, and it is in a form similar to
equation (l), but it has a lower dimension.
It should be noted that, although the above described analysis procedure is based on a modal solution, it is
appropriate to view the method simply as a coordinate transformation, where the normal mode shapes serve
as an orthogonal basis for the purpose of transformation from the real displacement coordinate space to the
modal coordinate space. The accuracy of the predicted response depends entirely on the quality of the
transformation vector used. For this reason, the mode shapes obtained from solving the eigenvalue problem
of equation (50) can be used for coordinate transformation without a great loss of accuracy only if the system
has a mild non-linearity or for systems with only local non-linearities. Otherwise, the transformation will
have to be performed in each time step using the mode shapes corresponding to the structures properties at
the beginning of this time step, which is a very costly process and its efficiency is questionable.
In general, the use of a modal solution (Galerkins method) in non-linear dynamic analysis is very effective in
earthquake-response analysis, since, for this type of analysis, only a relatively few mode shapes need to be
considered in the analysis, and the system is usually only locally non-linear.
By following the same procedure mentioned before for writing the incremental equations of motion in the
real displacement coordinate space, an equation similar to equation (7) can be obtained in the modal
coordinate space,
[ G I { a q i } + CCI{&,) + ~ R T ( y i ~ ~ { a ~ i } = { s Q i } + { Q i } - ~ M I { ~ i > - ~ c I { ~ i } -CIT{FP}
(53)
In the above equation, [R,(yi)] is the transformed tangent stiffnessmatrix that is recomputed at the beginning of
each new time step, and remains constant during iterations (modified Newton-Raphsom iteration technique).
 LONG-SPAN CABLE STAYED BRIDGES: THEORY 57

For solving equation (53) using the Wilson-8 method, equations similar to equations (33) and (34) can be
derived:

where

and

(55b)
Equation (54) is then solved for { aqi}. But there will be a residual force vector, owing to the assumption that
[KT(yi)] remains constant during the time step.
To compute this residual force, the vectors of {q(t,+ T)}, {q(ti, T)} and { q ( t i , r ) } should be first computed,
similar to equations (35), with qi instead of yi. Then, the residual force vector can be obtained from
{Q'(ti + z ) } = { Q i } + 8 ( { Q i + 1 } - {Qi})- [MI {ii(ti + .I}- CCI { 4 ( t i + TI} - C@I'{F'(ti + TI} (56)
The correction to the displacement is obtained from
CRil { A a q i } = {Q'(ti + (57)
After obtaining { Ahqi}, the updating of the process is given by
{ a q y + l ) }= { a q y ) }+ {A@)}

By repeating the computation, a new correction vector {A&?)} is obtained until convergence to some
acceptable tolerance is achieved.
Then, using the final value of {asi},the vectors of {&} and {Aiji} can be computed (similar to those of
equations (41) and (42)) and the vectors of {Aqi} and {Aq} (similar to those of equations (43) and (44)).
Finally, one obtains
{ qi+ 11 = qi} + { M i } * { Y i + 1) = [@I {qi+ 1 1 (594

The procedure is repeated for the next time step.

APPLICATION AND VALIDATION

As mentioned earlier,the accuracy of the non-linear response computed by the modal superposition method
depends on the 'quality' of the transformation vectors, i.e. the number and accuracy of the mode shapes used
for this purpose. Since the modes used here are those obtained using the tangent stiffness matrix of the bridge
in the deal-load deformed state, and are assumed to remain constant for the purpose of coordinate
transformation during the entire earthquake analysis, the results obtained by this method need to be verified.
For examining the accuracy of this method when used with cable-stayed bridges, a simple 2-D model (as
shown in Figure 6 ) is analysed once using the direct time integration method in the real displacement
coordinate space, and another time using Galerkin's method with 20 modes to compute its non-linear
response to earthquake excitation. In both analyses, ground motion records taken from the Imperial Valley,
California, earthquake, M, = 6.6, of 15 October 1979 are utilized in the present study to define representative
58 A. S. NAZMY A N D A. M. ABDEL-GHAFFAR

2 - D CABLE. STAYED BRIDGE

LOCATIONS OF CALCULATED RESPONSE QUANTITIES

Figure 6. The two-dimensional finite element model

2-D CABLE-STAYED BRIDGE - SPAN 1100 FT

UNIFORM EARTHQUAKE INPUT

X-DISPL OF JOINT 10
v.00
3.85
VIBRATIONAL A
3.00 f t

NONLINEAR (Wilson - 8 method)

-3.00
- ----__NONLINEAR (Wilson - 0 with 20 modes ) " . -2.81

0.00 1.00 2.00 3 00 Y.00 5.00 6.00 1.00 6.00 9.00

TIME IN SECONDS

Y - DlSPL OF JOINT 11
i I
- VIBRATIONAL
0.60--

-
I..
0
0.UO .-
x
-
0.20.-

~ -0.UO--

NONLINEAR (Wilson 8 method ) -

-0.80.- -
NONLINEAR (Wilson 8 with 20 modes.: - 0 . 7 7
i
0.00 1.00 2.00 3.00 Y.00 5.00 6.00 7.00 8.00 9.00

TIME IN SECONDS

Figure 7. Comparison between the direct integration method and the modal superposition method in computing the non-linear joint
displacement

and appropriately correlated input support motions. The 1979 earthquake, which is considered to be the
largest in California in the last quarter century, generated the most comprehensive set of data on ground
shaking yet recorded from a damaging earthquake anywhere in the world (and can be appropriately used for
studies of multiple-supported and extended structures). Some of the data were recovered from stations that
 LONG-SPAN CABLE STAYED BRIDGES THEORY 59

were very close to each other in the vicinity of the causative fault and that were aligned approximately along
and transverse to the fault (see References 14 and 27). Array No. 6 from the earthquake records of 1979 was
used for the uniform ground excitation input.' The time increment, At, is 0.01 sec, and the solution is carried
out for the first 10 sec of the earthquake.
Figure 7 shows the comparison between the results of the two analysis for the longitudinal displacement of
the left tower top (Joint lo), and the vertical displacement of the deck (Joint 1l), while Figure 8 shows this
comparison for the vertical shear force in Member 16 and the bending moment in Member 19.
It is evident that a good agreement between these response quantities is obtained from the two methods of
analysis. This indicates that the use of modal superposition for the non-linear dynamic analysis of cable-
stayed bridges gives good and reasonably accurate results when, of course, a sufficient number of modes is
used.
It should be mentioned at this point that poor agreement was obtained between the results of this method
and those of the direct integration method, when only 10 modes were used. The response curves show that, as
the number of modes used is increased, the closer the obtained response becomes to the non-linear response

2-D CABLE-STAYED BRIDGE - SPAN 1100 FT

UNIFORM EARTHQUAKE INPUT

Y-SHEAR IN MEM 16 AT JOINT 2

1 , 1"

-2.00
-NONLINEAR ( Wilson - 0 method )
-2.440

0 00 I 00 2.00 3.00 Y.00

i
- - - - - -. NONLINEAR (Wilson - 0 with 20 modes)
5 00 6 00 1.00 8.00
,
9.00
-1.344

TIME IN SECONDS

Z-MOMENT IN MEM 19 AT JOINT 11

- 1.50
VIBRATIONAL A
I.UO

-I.ZOy
- - _ - _ _NONLINEAR (Wilson - 0 with 20 modes) .1.11
T
000 1.00 2.00 1.00 U.00 1.00 6.00 1.00 8.00 9.00

TIME IN SECONDS

Figure 8. Comparison between the direct integration method and the modal superposition method in computing the non-linear element
forces
60 A. S. NAZMY AND A. M. ABDEL-GHAFFAR

2-D CABLE-STAYED BRIDGE - SPAN 1100 FT

UNIFORM EARTHQUAKE INPUT

Y-SHEAR IN MEM 16 AT JOINT 2

1 I

G, it
o.uo t
-00.0

j/'
\I \ I
-LINEAR
. ~ . y D ~ - - - - - - NONL.INEAR -2.33
4 I
o 00 1.00 2.00 3.00 q.00 s.oo 6.00 1.00 8.00 4.00

TIME IN SECONDS

Z-MOMENT IN MEM 19 AT JOINT 11

VIBRATIONAL

-0.90

-1.13
0.00 1.00 2.00 Y.00 4.00 5.00 6.00 1.00 8.00 9.00

TIME IN SECONDS

Figure 9. Comparison between the linear and non-linear dynamic responses

computed by the direct integration method. This agrees with Bathe's conclusion in Reference 21. Finally,
Figure 9 shows the comparison between the linear and non-linear responses of this relatively short span
bridge.
Based on the above discussion, Galerkin's method was used for the non-linear earthquake analysis of the
3-D models, described in the companion paper,28 using 30 modes which cover a broad band of the bridge
vibration frequencies. No attempt was made to perform non-linear direct integration for the 3-D models since
the finite element model has over 550 degrees of freedom and a very large computation time is required to
perform such an analysis.

SUMMARY OF THE ANALYSIS

The non-linear dynamic analysis of cable-stayed bridges when subjected to synchronous and non-
synchronous seismic loadings is formulated. The initial calculations of the direct step-by-step integration
algorithm involve the following steps.
 LONG-SPAN CABLE STAYED BRIDGES THEORY 61

1. Form the tangent stiffness matrix at the dead-load deformea state; for the modal analysis solve the
eigenvalue problem for the lowest 30 modes.
2. Initialize the displacement and velocity by zeros, and { j o }= [M]-{P,}.
3. Select At (usually equal to the time step in earthquake input records) and set O = 1.4.
4. Form the effective linear stiffness matrix [equation (31a)l; in the linear analysis, triangularize this
matrix.
5. In the linear analysis, form the effective incremental load vector [equations (31a, b)]; then solve for the
incremental displacement vector from equation (30).
6. In non-linear analysis (i) update the tangent stiffness matrix using the nodal displacement and member
forces due to dead-load displacement plus vibrational displacement; (ii)form the effective stiffness matrix
[equation (34a)l and then triangularize it; (iii) compute the elastic force vector (equation (21)] at the
beginning of each time step; (iv) form the effective incremental load vector [equation (34b)3; (v) solve for
the incremental displacement vector [equation (33)]; (vi) iterate for dynamic equilibrium.
7. For both linear and non-linear analyses calculate the incremental acceleration, velocity and displace-
ment for the real time step At [equations (41), (43), (44)]; and then finally, compute the displacement,
velocity and acceleration at the end of the real time step [equations (45a, b, c)].
8. In the non-linear modal algorithm the generalized matrices have to be computed using the modal
transformation and the generalized displacement vector has to be computed.

ACKNOWLEDGEMENTS
This paper is based on the Ph.D. dissertation of the first author that was supervised by the second author and
presented to the faculty of the Department of Civil Engineering of Princeton University in May, 1987. The
research was supported partially by a grant (No. ECE-8501067) from the National Science Foundation with
Dr S. C . Liu as the Program Director, and partially by the Egyptian Ministry of Higher Education via a
Government Mission; this support is greatly appreciated.

REFERENCES
1. Aly S. Nazmy and Ahmed M. Abdel-Ghaffar, Seismic responses analysis of cable-stayed bridges subjected to uniform and multiple-
support excitations, Report No. 87-SM-I, Department of Civil Engneering, Princeton Univeristy, 1987.
2. Ahmed M. Abdel-Ghaffar and Aly S. Nazmy, Effects of three-dimensionality and nonlinearity on the dynamic and seismic behavior
of cable-stayed bridges, Proc. 4th struct. congr. 1987, Bridges transmission line struct. ASCE, Orlando, Florida (1987).
3. Ahmed M. Abdel-Ghaffar and Aly S. Nazmy, Dynamic and seismic-analysisconsiderations for the design of long-span cable-stayed
bridges, Proc. 3rd joint U S - J a p a n workshop performance strengthening bridge struct. research needs Tokyo, Japan (1987).
4. Ahmed M. Abdel-Ghaffar and Aly S. Nazmy, Response evaluation and research needs for seismic design of cable-stayed bridges,
Proc. 4th joint U.S.-Japanworkshop bridge eng. performance strengthening innovation San Diego, CA (1988).
5. N. F. Morris, Dynamic analysis of cable-stiffened structures, J. struct. div. ASCE 100, 971-981 (1974).
6. J. F. Fleming and E. A. Egeseli, Dynamic behaviour of a cable-stayed bridge, Earthquake eng. struct. dyn. 8, 1-16 (1980).
7. E. L. Wilson, I. Farhoomand and K. J. Bathe, Nonlinear dynamic analysis of complex structures, Earthquake eng. struct. dyn. 1,
24 1 -252 (1 973).
8. K. J. Bathe and E. L. Wilson, Linear and nonlinear earthquake analysis of complex structures, Proc. 5th world conf. earthquake eng.
Rome, Italy, 2, 17961805 (1973).
9. J. F. Fleming, Linear versus nonlinear behavior of cable-stayed bridges, Wind seismic effects, Proc. 14th joint panel conf. US-Japan
cooperative program natural resources 343-360 (1983).
10. J. F. Fleming, Nonlinear static analysis of cable-stayed bridge structures, Comput. struct. 10, 621-635 (1979).
11. The Task Committee on Cable-Suspended Structures of the Committee on Special Structures of the Committee on Metals of the
Structural Division, Tentative recommendations for cable-stayed bridge structures and Commentary on the tentative recommen-
dations for cable-stayed bridge structures, J . struct. diu. ASCE 103, 929-959 (1977).
12. N. F. Morris, The use of modal superposition in nonlinear dynamics, Comput. struct. 7 , 65-72 (1977).
13. Aly S. Nazmy, Nonlinear earthquake-response analysis of cable-stayed bridges subjected to multiple-support excitations, Ph.D.
Dissertation, Department of Civil Engineering, Princeton University, 1987.
14. A. G. Brady, V. Perez and P. N. Mork, The Imperial Valley earthquake, October 15,1979: digitization and processing of acceleration
records, Open-File Report 80-703, U.S. Geological Survey, Seismic Engineering Branch, Menlo Park, CA, 1980.
15. R. W. Clough and J. Penzien Dynamics ofStructures, McGraw-Hill, New York, 1975.
16. M. Paz, Structural Dynamics, Theory & Computation, 2nd edn, Van Nostrand Reinhold, New York, 1985.
17. K. J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1982.
18. J. H. Argyris, P. C. Dunne, and T. Angelopoulos, Nonlinear oscillations using the finite element technique, Comput. methods appl.
mech. eng. 2, 203-250 (1973).
19. H. Adeli, J. M. Gere and W. Weaver, Algorithms for nonlinear structural dynamics, J. struct. div. ASCE 104, 263-280 (1978).
62 A. S. NAZMY AND A. M. ABDEL-GHAFFAR

20. C. A. Felippa, and K. C. Park, Direct time integration methods in nonlinear structural dynamics, Report No. LMSC-D013361,
Applied Mechanics Laboratory, Lockheed Palo Alto Research Laboratory, Palo Alto, CA 94304, 1978.
21. K. J. Bathe and S. Gracewski, S. On nonlinear dynamic analysis using substructuring and mode superposition, Comput. struct. 13,
699-707 (1 98 1).
22. A. M. Abdel-Chaffer, R. H. Scanlan and L. I. Rubin, Earthquake response of long-span suspension bridges, Report No. 83-SM-13,
Department of Civil Engineering, Princeton University, 1983.
23. F. Baron, M. Arikan and E. Hamati, The effects of seismic disturbances on the Golden Gate Bridge, Report No. E E R L 76-31,
University of California, Berkeley, CA, 1976.
24. R. E. Nickell, Nonlinear dynamics by mode superposition, Comput. methods appl. me&. eng. 7 , 107-129 (1976).
25. A. K. Noor, Recent advances in reduction methods for nonlinear problems, Comput. struct. 13, 3 1 4 (1980).
26. R. W. Clough and E. L. Wilson, Dynamics analysis of large structural systems with local nonlinearities, Cornput. methods appl. mech.
eng. 17/18, 107-129 (1979).
27. C. Rojahn, Selected papers on the Imperial Valley, California, earthquake of October 15, 1979; Open-File Report 80-1094, U.S.
Geological Survey, Seismic Engineering Branch, Menlo Park, CA, 1980.
28. Aly S. Nazmy and Ahmed M. Abdel-Ghaffar, Non-linear earthquake-response analysis of long-span cable-stayed bridges:
Applications, Earthquake eng. struct. dyn., 19, 63-76 (19%).