Design of An Active Controller For Quincy Bayview Bridge, Against Seismic Excitation - Model Updatingdutta2008 PDF

STRUCTURAL CONTROL AND HEALTH MONITORING
Struct. Control Health Monit. 2008; 15:10571077

Published online 2 January 2008 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/stc.232
Design of an active controller for Quincy Bayview Bridge,

Illinois, U.S.A., against seismic excitation}Part I:
Model updating
Atanu Kumar Dutta, Anjan Dutta*,y and Sajal Kanti Deb

Department of Civil Engineering, Indian Institute of Technology, Guwahati 781039, India
SUMMARY
This paper deals with the updating of a 3-D nite element model of an existing cable-stayed bridge. The
objective of the updating is to arrive at a model that captures experimentally observed modes to the closest
extent. Simultaneously, it has been attempted to keep the model as compact as possible as well as
numerically stable such that the model may be used for the implementation of active control strategy
against seismic excitation. The updating has been attempted primarily by using catenary element for
modeling of cables in place of equivalent truss element, constraint equations in place of bearing elements in
the deckpylon interface as well as for rigid links connecting the spine with cables. Further, due to the
existence of large axial force, the PD eect for deck and pylon members has also been considered. The
conguration of the updated model for linear dynamic analysis has been adopted from the dead-load-
deformed shape, which has been arrived at by carrying out nonlinear static analysis. The modal
characteristics of the updated model have been validated with experimentally observed values. Moreover,
the performance of the updated model has also been compared with alternative models formed by a
combination of truss and catenary element in modeling of stay cables, with or without PD eect on deck
and pylon. Copyright # 2008 John Wiley & Sons, Ltd.
KEY WORDS: catenary element; kinematic constraint; PD eect; dead-load-deformed structure; non-
linear static
1. INTRODUCTION
Cable-stayed bridges have gained popularity in the category of long-span bridges over the past
three decades due to improved structural performance and aesthetic appeal in comparison with
suspension bridges. It is appreciated that the development of a numerical model, which
simulates natural frequencies and mode shapes of the structure, is an important step for the
computation of dynamic responses of a cable-stayed bridge. Moreover, the model needs to be
*Correspondence to: Anjan Dutta, Department of Civil Engineering, Indian Institute of Technology, Guwahati 781039,
India.
y
E-mail: adutta@iitg.ernet.in, anjan civil@yahoo.com
Received 21 May 2007

Revised 7 August 2007
Copyright # 2008 John Wiley & Sons, Ltd. Accepted 29 August 2007
1058 A. K. DUTTA, A. DUTTA AND S. K. DEB
compact and free from numerical ill-conditioning so that it may be used for the implementation
of control strategy.
However, building a numerical model to represent dynamic characteristics of a cable-stayed
bridge is rather dicult as this exible structure exhibits complex behavior with the exural,
lateral and torsional motions being very often coupled. As a consequence, structural
characteristics and responses predicted by idealized nite element (FE) models have
discrepancies and errors compared with those obtained from the actual measurements. Lack
of proper idealization of the complex structure is regarded as one of the main causes of such
discrepancies in a numerical model for a cable-stayed bridge. Modal identication through full-
scale testing is the most reliable method to determine the true dynamics properties (e.g. natural
frequencies, damping ratios and mode shapes) of a structure. This serves as a basis for validating
and/or updating an analytical model of a structure so that the model represents the actual
structural properties and boundary conditions.
FE model updating method has emerged in the 1990s as a subject of great importance for
mechanical and aerospace engineering. The main approaches to model updatings [1] are (a)
optimal matrix updating, (b) sensitivity-based parameter estimation, and (c) neural-networks
updating. However, model updating has been found to be dicult to apply to civil engineering
structures, because of the diculties in prototype testing and experimental data analysis arising
from the nature, size, location and usage of these structures. Only recently, the civil engineering
community has begun to adopt this technique. Aktan et al. [2] conducted a parameter grouping
and sensitivity study on a bridge by considering material properties, boundary and continuity
conditions and geometric properties as the model parameters. However, there is very little
literature available on model updating of cable structures. Research on applications of model
updating method to dynamic assessment of a cable-stayed bridge was investigated by
Brownjohn and Xia [3]. Zhang et al. [1] used optimization approach to update the FE model
of a suspension bridge. In another study, Zhang et al. [4] updated the model of a cable-stayed
bridge using sensitivity-based parameter estimation approach.
The cable-stayed bridge model sought to be updated here is the FE model proposed by
Wilson and Gravelle [5] for Quincy Bayview Bridge in IL, U.S.A. The model was innovative in
its representation of deck as a spine with translational and rotational lumped masses connected
to the spine by rigid links. The cables were also connected to the spine through rigid links. Other
components of this model were (1) cable as truss element using Ernsts equivalent Youngs
modulus and (2) bearing links to model the deck pylon connection. The conguration due to
dead load was not arrived at using any nonlinear static analysis. The stiness of the bridge
structure needed for eigenvalue analysis was instead evaluated based on the design
conguration. The model was validated with the ambient vibration results presented by Wilson
and Liu [6]. It was found that the numerical model was fairly successful in representing the
exural modes albeit with relatively lower frequencies. The observed lower frequencies were
perhaps due to the adoption of a linear model for eigenanalysis. For a highly nonlinear structure
such as the cable-stayed bridge, it is always appropriate to use the dead-load-deformed
conguration arrived at through nonlinear static analysis as the basis for the model to be used
for eigenanalysis. However, more importantly the model was not successful in capturing the
early transverse experimental modes of the structure. In this study, it is attempted to update the
numerical model with an objective to capture the experimental modes more accurately.
Researchers [7, 8] have pointed out the discrepancies of using truss element for modeling the
cable in a cable-stayed bridge and have preferred to adopt the exact elastic catenary element.
Copyright # 2008 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2008; 15:10571077
DOI: 10.1002/stc
DESIGN OF AN ACTIVE CONTROLLER 1059
Jayaraman and Knudson [7] presented the formulation for a small strain elastic catenary
element and demonstrated its utility through static and dynamic problems. Karoumi [8]
simplied the formulation and used the same for 2-D analysis of a long-span cable. The results
were compared with single truss element with equivalent Youngs modulus and it was observed
that catenary element resulted in less deection than truss element for high tension/initial
tension ratio of the cable. It was pointed out that equivalent modulus approach accounted only
for slackening eect of the cable due to sag and not for its stiening eect due to large
displacements. Wang and Yang [9] made a parametric study on cable-stayed bridges for
individual eects of dierent sources of nonlinearity. It was concluded that for the dead-load
deformation analysis, the large deection is found to contribute the most to the nonlinearity,
with cable sag contributing the least. It may be recalled from the observations of Karoumi [8]
that the formulation for equivalent modulus for cable modeled as truss element accounts only
for sag, whereas the formulation for cable as elastic catenary considers the geometry of the
cables. Thus, it can be concluded from the observations by Wang and Yang [9], in conjunction
with Karoumi [8], that catenary formulation for cable is more appropriate when dead-load-
deformed structure is used for modal analysis. Hence, the primary update contemplated here is
to use the catenary element to represent the cables. However, optimization scheme could not be
adopted as the identication data were not available with the authors.
The earlier numerical model adopted by Wilson and Gravelle [5] was fairly successful in
representing the vertical as well as the torsional modes in experimental results. Hence it is
concluded that the spine model is a good representation of the deck. More involved modeling
approach for deck as suggested by Ren and Peng [10] has not been adopted so as to avoid
numerous additional degrees of freedom (DOFs), which will render the active control strategy to
be presented in a companion paper to be an impractical one. In the earlier model, bearing
elements were idealized as vertical and horizontal links with ctitious material and sectional
properties. However, the introduction of elements with ctitious sectional and material
properties is found to cause numerical ill-conditioning in nonlinear static analysis. In the
updated model, the bearings are represented by constraint equations.
Further unlike the earlier numerical model, the updated model for modal analysis is based on
the dead-load-deformed structure arrived at through nonlinear static analysis considering PD
eect on deck and pylon. As the deformed structure is stier, it is expected to raise the
frequencies of the model to the level of experimentally observed frequencies. In representing
boundary conditions, the more conventional penalty approach is avoided as it introduces
numerical ill-conditioning in the nonlinear static analysis.
Four analytical models have been planned and the best updated model is to be selected after
examining their performance in representing the experimentally observed modal characteristics
of the same bridge. Dierent models have been arrived at with dierent combinations of
modeling strategy by adopting truss or catenary element for cable and by considering or
ignoring the PD eect for representing nonlinearity in deck and pylon.
2. THE BRIDGE
The Quincy Bayview Bridge in IL, U.S.A., has two H-shaped concrete towers, double-plane fan-
type cables and a composite concrete-steel girder bridge deck. The main span is 274 m and there
are two equal side spans of 134 m for a total length of 542 m. The material and sectional
DOI: 10.1002/stc
properties of deck and pylons are described in detail in Wilson and Gravelle [5] and are not
repeated. There are a total of 56 high-strength steel cables, 28 supporting the main span and 14
supporting each of the side spans. The cables are of four dierent cross-sectional areas: (1)
89:67 cm2 ; (2) 67:74 cm2 ; (3) 53:67 cm2 and (4) 34:83 cm2 as shown in Figure 1.
3. FE MODELING
3.1. Modeling of deck

The modeling of deck is followed from the earlier model adopted by Wilson and Gravelle [5].
The deck cross-section has been assumed to be of equivalent steel thin-walled open C-section as
shown in Figure 2. Pure torsional constant for the open section has been modied to include the
warping constant of the section. The sectional properties that have been adopted for equivalent
C-section are as follows: area, 0:8268 m2 ; transverse moment of inertia, 19:75 m4 ; exural
Figure 1. Line diagram of the bridge with cable type.
Figure 2. The equivalent steel C-section for deck.
DOI: 10.1002/stc
moment of inertia, 0:3409 m4 and torsional moment of inertia, 0:0271 m4 : The equivalent
sectional and material properties of the C-section have been ascribed to the spine. The spine is
located along the shear center of the C-section and is discretized using beam elements.
The connectivity of the cable ends to the spine is done through rigid link as per the model
proposed by Wilson and Gravelle [5]. The translational masses, arising out of discretization of
the deck as a spine, are lumped at the end of the rigid links as shown in Figure 3. The calculated
translational mass for each deck segment for the main span is 113 857:75 kg; whereas the same
for the side span is 108 401.58 kg.
Certain corrections are to be made to the mass moment of inertia of the spine node as the
distributed masses are lumped and placed at the end of the rigid links. The applied corrections
are in accordance with the following equation:
Dx Imx 2Mr2 1
where Dx is the corrective mass moment of inertia about the x axis (x is for X; Y; Z), M is the
lumped mass (kg) and r (m) is the radial distance of lumped mass from shear center. The
corrections for the side span and main span are shown in Table I.
3.2. Modeling of pylon

Pylons are discretized with beam elements with nodes introduced at locations of sectional
changes. A comparatively coarser mesh has been considered adequate as the pylon is much
stier than the deck. The beam formulations with or without stability factors accounting for P
D eect have been adopted in order to evaluate the inuence of beam column eect on the model
as the structure deforms under dead load. This approach has also been adopted to model the
deck.
Figure 3. Finite element model for deck.
Table I. Mass moment of inertia kg m2 correction at the spine node.

Correction component Side span Main span
DX 3 437 505.38 3 273 011.26
DZ 3 041 032.37 2 315 571.11
DY 7 971 638.25 700 154.47
DOI: 10.1002/stc
In the modeling of deck and pylon, beam formulations with or without consideration for PD
eect have been adopted for a comparative study of its eects on the static nonlinear behavior of
the structure. The FE formulation for beam element is well established and is not presented
here. The formulation to account for the PD eect in beam element through stability functions
has been enumerated in Nazmy and Abdel-Ghaar [11].
3.3. Modeling of cable

Two primary approaches for modeling of cables have been discussed in this section. The cable
has been modeled as both truss element with equivalent modulus and as catenary element.
3.3.1. Stiness of cable as an equivalent truss element. Modeling the cable as a truss element is
done with Ernsts equivalent Youngs modulus and is expressed as
E
Eeq 2
wLx 2 AE
1
12T 3
where Eeq is the equivalent modulus, E is Youngs modulus of cable material, w is the unit
weight of cable per meter, Lx is the horizontal projected length of the cable, A is the cross-
sectional area of the cable, and T is the prevalent cable tension. Koroumi [8] observed that the
formulation in Equation (2) can account only for the slackening of the cable due to sag eect
and due to self-weight; the stiening eect on the cable due to large displacement, which is
generally the case with exible structures such as the cable-stayed bridge, cannot be properly
accounted for in this formulation.
3.3.2. Stiness of cable as an elastic catenary element. The stiness formulation of cable as a
catenary element has been developed following the procedure enumerated in Jayaraman and
Knudson [7]. The typical cable prole with associated forces is illustrated in Figure 4. The
owchart in Figure 5 explains the iterative procedure for arriving at the nodal forces of the cable
element. After the evaluation of the nodal forces, the element stiness matrix for elastic catenary
is expressed as [7]
8 9 2 38 I 9
>
> dFxI >
> a1 a2 0 a1 a1 0 > du >
>
> >
> >
> >
>
>
> > 6 7 > >
>
> dFyI >
>
> 6 a 0 a a 0 7>>
> dv I >
>
>
>
> >
> 6 6 4 3 4 7 >
> >
>
>
> >
> 7 >
> >
I >
< dF I = 6 F =H 0 0 F =H 7 < dw =
z 6 1 1 7
6 7 3
>
> J> 6 7> J >
> dFx >
> > 6
>
a1 a2 0 7>> du >
> >
>
>
> >
> 6 7> >
>
> dFyJ >
> 6 Symmetric a4 0 7>>
> dvJ >
>
>
>
> >
> 4 5 >
> >
>
>
> >
> >
: >
: J; J;
dFz F1 =H dw
where the coecients are in terms of the nodal forces. It may be noted that the stiness in the
out of plane (YZ plane) is extrapolated from the inplane stiness (XY) by dividing the nodal
force by the horizontal intercept of the cable (H). Since the value of H is quite large, it is clear
that the out-of-plane stiness of the cable will be quite low compared with the inplane stiness.
When this catenary element is used to model the cables for the bridge, it is expected that the
DOI: 10.1002/stc
Figure 4. Catenary element.
above-mentioned low out-of-plane stiness should help in the representation of the transverse
mode of the deck as observed in the experimental results.
The stiness matrix is evaluated in the local XY plane (in the plane of the element) and is
transformed to the global XYZ coordinate system for assemblage into the structure stiness
matrix by the usual coordinate transformation technique. The transformed stiness matrix is as
given in the following equation:
" #
C C
S 4
C C
where
2 3
F1 =Hl 2 a1 m2 a2 m F1 =Hlm a1 ml
6 7
C 6
4 a2 l a4 a2 l 7
5
F1 =Hlm a1 ml a2 m F1 =Hm2 a1 l 2
Here, m and l are the cosine and sine of the angle between the local XY plane and the global XY
plane of the cable.
It has been observed that the formulation of elastic catenary element accounts for large
displacement of the nodes in addition to sag eect of the cable due to self-weight. Thus, catenary
element represents the true nature of the cables as structural elements, which are exible in
nature and undergo large displacement in eld condition. Most importantly, this formulation
represents the low out of stiness of the cable to a closer extent.
3.3.3. Element mass matrix. Mass discretization is done by lumping of the element mass at both
ends of the cable. Lumped masses at the nodes are given by 0:5rALu ; where Lu is the unstressed
length of the cable, A is the area of cross-section of the cable and r is the mass density of the
cable material.
DOI: 10.1002/stc
Figure 5. Flowchart for catenary element analysis.
3.4. Rigid links

The concept of kinematic constraint is used for modeling the rigid links, where constraint
equations relate the displacements of two joints. The procedure is enumerated in References
[12, 13]. One of the joints is termed as master to which the displacements of the other joint,
termed as slave, can be related. Equations (5) relate the translations (u1; u2; u3), rotations
(r1; r2; r3) and coordinates (x; y; z) of the master and the slave, all taken in the constraint local
coordinate system:
u1s u1m r2m Lz r3m Ly

DOI: 10.1002/stc
r1s r1m
r2s r2m
r3s r3m 5
where Lx xs xm ; Ly ys ym and Lz zs zm : Here, the subscript s stands for slave

and m represents master node. The displacements of the slave node can thus be transformed
from the master node through the following transformation matrix:
8 9 2 38 9
>
> u1s >> 1 0 0 0 Lz > u1m >
Ly > >
>
> >
> 6 7>
> >
>
>
> u2 >
> 60 1 0 Lz 0 7>
>
Lx 7> u2 >
>
>
> s >
> 6 > m >
>
>
> >
> 6 7>
> >
>
< u3s >
> = 60 < u3m >
> =
6 0 1 Ly Lx 0 7
7
6 7 6
>
> >
r1s > 60 0 7> r1ms >
>
> >
> 6 0 0 1 0 7>
> >
>
>
> >
> 6 7>
>
>
>
>
>
>
> r2 >
> 60 0 0 0 1 0 7>
> r2 >
m >
>
> s>
> 4 5>
> >
>
>
: >
; >
: >
;
r3s 0 0 0 0 0 1 r3m
In the context of this study, the spine nodes are considered as the master nodes and the nodes
connecting the cables to the spine via rigid links are considered as the slave nodes for the
evaluation of stiness. For transformation of mass, the lumped mass nodes on either side of the
spine are considered as salve nodes. Thus, corresponding to each master node along the spine,
there exist two slave nodes on either side of the spine. With these transformation matrices, the
mass and stinesses of the slave nodes are transformed to the spine master nodes. The
representation of rigid links in terms of kinematic constraints can thus result in substantial
reduction in the degrees of freedom.
3.5. Modeling of bearing

The deck-tower bearings have been modeled using the constraint equations (5). In the FE
model, the spine node (slave node) is constrained with the pylon nodes at lower strut (master
nodes) as illustrated in Figure 6. The constraint equations are applied with respect to
translational DOFs in the longitudinal direction and rotational DOFs about the vertical axis.
These equations represent the boundary conditions that permit the end of the deck to rotate
freely about the vertical Y and transverse Z axes. Rotation about the longitudinal axis X
and all three translational DOFs are restrained in the two abutments.
The discretized model of the cable-stayed bridge as per the model proposed by Wilson and
Gravelle [5] is shown in Figure 7. The model has 82 beam elements, and 56 cable elements. In the
updated model 10 constraint equations have been used to represent the bearings between deck
and pylon. The 128 rigid links of the model connecting the spine nodes with the cable ends and
the translational masses are also represented by constraint equations. The updated model has a
total DOF of 458, whereas the original 3-D FE model in Wilson and Gravelle [5] contains a total
DOF of 1248. The pylon bases have been considered as xed with respect to all the displacement
components.
DOI: 10.1002/stc
Figure 6. The master and slave nodes in pylon-bearing connection.
Figure 7. Finite element model.
4. NONLINEAR STATIC ANALYSIS
The mixed incremental-iterative procedure has been adopted for nonlinear static analysis of the
structure. As the nonlinearity in cable-stayed bridges is predominantly geometric in nature, the
stiness has been updated for the change in geometry alone, considering materials in the elastic
range. The tangent stiness at the beginning of each load step is considered for the step-by-step
analysis. A large number of load steps are used in order to account for the huge dead load of the
structure. Four numerical models have been considered for nonlinear static analysis for the
bridge. The objective is to evaluate the inuence of dierent modeling strategies on the nal
conguration of the bridge under the dead load. The descriptions of each of the models are as
follows:
Model I: Cable modeled as truss element with equivalent modulus; deck and pylon modeled
with beam element without considering PD eect.
Model II: Cable modeled as truss element with equivalent modulus; deck and pylon modeled
with beam element considering PD eect.
DOI: 10.1002/stc
Model III: Cable modeled with elastic catenary element; deck and pylon modeled with beam
element without considering PD eect.
Model IV: Cable modeled with elastic catenary element; deck and pylon modeled with beam
element considering PD eect.
It may be mentioned that the proposed updated model for the numerical model considered by
Wilson and Gravelle [5] corresponds to the Model IV. Other models (Models IIII) have been
considered to compare their relative performance and observe whether Model IV qualies as the
most suitable update. A computer program has been developed in Matlab1 to perform the
nonlinear static analysis. The initial deck prole has been assumed to be horizontal. In the
absence of data on pretension, an iterative way has been adopted to nd out the nal tension of
the cable at dead-load-deformed conguration. For this a pretension of 30% of ultimate cable
tensions of the respective cable types has been considered as initial tension. Member forces
including the cable tensions have been updated after each load step. It may be mentioned that
the stability factors in beam element formulation will remain as unity in the case of Models II
and IV, where PD eect for deck and pylon is not considered. The incremental-iterative
procedure is continued till the unbalanced joint loads converge. The nal tensions serve as the
initial tension for the next iteration. An outer iterative loop, as illustrated in the owchart
(Figure 8), has been adopted to arrive at the convergence of cable tensions. The outer
convergence corresponds to the dead-load-deformed structure with nal cable tensions and nal
member forces. It may be mentioned that irrespective of the initial pretension, the nal tension
conguration attained is the same if the iterations are carried out.
It may be mentioned that the proposed updated model for the numerical model in Wilson and
Gravelle [5] corresponds to the Model IV. Other models (Models IIII) have been adopted to
compare their relative performance and to see whether Model IV qualies as the most suitable
update. A computer program has been developed in Matlab1 to perform the nonlinear static
analysis. The horizontal prole has been adopted as the initial deck prole. A pretension of 30%
of ultimate cable tensions of the respective cable types has been considered as the initial tension
in the analysis. Member forces including the cable tensions have been updated after each load
step. It may be mentioned that the stability factors in beam element formulation will remain as
unity in the case of Models II and IV, where PD eect for deck and pylon is not considered.
The incremental iterative procedure is continued till the unbalanced joint loads converge. The
nal tensions serve as the initial tension for the next iteration. An external iterative loop, as
illustrated in the owchart (Figure 8), has been carried out to arrive at the convergence of cable
tensions. Thus, the outer convergence corresponds to the dead-load-deformed structure with
nal cable tensions and nal member forces.
5. MODAL ANALYSIS
The stiness matrices used in the modal analysis of the four numerical models correspond to the
nal geometry, member forces, as well as the cable tensions of the dead-load-deformed
structure. The frequencies and mode shapes obtained from modal analysis have been compared
with those from experimental results reported by Wilson and Liu [6]. As modal characteristics
are the best indicator of the appropriateness of a model, this comparative study would indicate
whether the proposed updated model is the best among the four alternatives. The eect of using
PD eect of beam element on the modal characteristics is also explored in this section.
DOI: 10.1002/stc
Figure 8. Flowchart for the nonlinear static analysis.
6. RESULTS AND DISCUSSION
The results of nonlinear static analyses and the corresponding dynamic characteristics of all the
four analytical models have been presented in this section.
6.1. Nonlinear static analysis: dead-load-deformed shape

The initial and nal cable tensions along with the ultimate cable tension for all the cables for the
proposed updated Model IV have been shown in Table II.The cable numbers are as indicated in
DOI: 10.1002/stc
Table II. Initial and nal cable tensions for the Model IV.
Ultimate tension Initial tension Final tension (MN) Percentage of ultimate
Cable no. (MN) (MN) (Model IV) tension (%)
1, 28, 29, 56 14.3 4.29 7.01 49
2, 27, 30, 55 14.3 4.29 7.12 50
3, 26, 31, 54 10.8 3.24 5.23 48
4, 25, 32, 53 10.8 3.24 4.94 46
5, 24, 33, 52 5.6 1.68 2.82 50
6, 23, 34, 51 5.6 1.68 2.54 45
7, 22, 35, 50 5.6 1.68 2.35 42
8, 21, 36, 49 5.6 1.68 2.38 42
9, 20, 37, 48 5.6 1.68 2.42 43
10, 19, 38, 47 5.6 1.68 2.85 50
11, 18, 39, 46 8.6 2.58 4.15 48
12, 17, 40, 45 8.6 2.58 3.96 46
13, 16, 41, 44 10.8 4.29 5.32 49
14, 15, 42, 43 10.8 4.29 5.35 50
Figure 1. The nal cable tensions in the individual cables have been found to be in the vicinity of
50% of their ultimate tensions. This would result in an equivalent Youngs modulus that is
essentially equal to the original modulus, which is in agreement with the observations in Wilson
and Gravelle [5].
The deected shapes of the bridge, corresponding to Models I and IV, have been shown in
Figures 9 and 10. The vertical displacements of the spine and horizontal displacements of the
pylons have been exaggerated to appreciate the deected shape for the scale of the structure. The
presented shapes in Figures 9 and 10 correspond to the two extreme modeling schemes. The
deected shapes corresponding to the intermediate models have not been shown as the
dierence between them is not appreciable. The computed deections of the dead-load-
deformed structure for the four numerical models have been presented for selected spine nodes
as shown in Table III.
It is evident from the tabulated data that the midspan deection for the Model I is down-
ward, while the same has an upward trend from Model I to Model IV. The value ranges
from 0:02 m (downwards) to 0:97 m (upwards) indicating that the models are
gradually stiening from Model I to Model IV. The other spine point deections are also
observed to be compatible with the midspan deection. The deections for the select pylon
nodes (Table IV) are also showing close correspondence with the spine deections.
While a rightward deection of 0.008 m has been observed at the top of the left pylon node
for Model I, a leftward deection of 0.208 m has been observed for the node at the same location
for Model IV.
Thus, it is observed from the results in Tables III and IV that the models with the cables
idealized as catenary element are stier compared with the ones with cables modeled as
equivalent truss element. Further, it has been observed that the models taking into account the
PD eect in beam element formulation are stier than their counterparts without the same.
Hence, it is concluded that the model involving the PD eect as well as the catenary cable
element is the stiest. The modal characteristics of the four models have been presented in the
DOI: 10.1002/stc
Figure 9. Deected shape for Model I.
Figure 10. Deected shape for Model IV.
Table III. Displacements (m) of representative spine nodes for Models IIV.
Node locations Model I Model II Model III Model IV
Left side span, center 2:25e 03 8:38e 03 1:75e 02 7:04e 02
1
4 center span 8:10e 03 3:46e 02 8:46e 02 3:80e 01
Midspan 2:06e 02 8:82e 02 2:16e 01 9:70e 01
3
4 center span 8:09e 03 3:46e 02 8:45e 02 3:80e 01
Right side span, center 2:55e 03 8:14e 03 1:90e 02 7:05e 02
Table IV. Displacements (m) of representative pylon nodes for Models IIV.
Model I Model II Model III Model IV
Node location Left Right Left Right Left Right Left Right
Lower strut joint 7:68e 5 7:6e 5 2:38e 4 2:3e 4 7:05e 4 7:05e 4 2:66e 3 2:65e 3
Deck level joint 8:83e 5 8:8e 5 3:76e 4 3:7e 4 1:05e 3 1:05e 3 3:97e 3 3:97e 3
Upper strut joint 2:13e 3 2:1e 3 9:07e 3 9:1e 3 2:05e 3 2.05e3 9:58e 2 9:57e 2
Lower cable joint 2:72e 3 2:7e 3 1:16e 2 1:2e 2 2:43e 2 2:43e 2 1:22e 1 1:22e 1
Upper cable joint 8:08e 3 8:1e 3 2:02e 2 2:0e 2 4:83e 2 4:83e 2 2:08e 1 2:08e 1
DOI: 10.1002/stc
following section. Comparison of these characteristics with those from the numerical model of
Wilson and Gravelle [5] and experimental observations [6] would help to establish whether the
proposed updated model is the best candidate for model updating.
6.2. Modal analysis}frequencies and mode shapes

The modal characteristics of the four models have been evaluated in this section and have been
compared with established results. The observations from visual inspection are presented in
Table V. Asterisks indicate missing modes in the numerical models. It has been observed that the
exural modes (1, 2 and 7) for all the four numerical models correspond well with the respective
modes from the numerical model by Wilson and Gravelle [5] as well as with those from
experimental observations [5]. However, a stiening trend exhibits from Model I to Model IV, as
indicated by progressively increasing frequencies. This stiening eect may be attributed to the
consideration of PD eect in nonlinear analysis as well as due to the adoption of catenary
element to model the cable in place of the truss element. It has been observed that these mode
shapes and frequencies from Model IV resemble the experimental observations very closely.
The most signicant observation in terms of mode shapes is that the rst two coupled
torsional-transverse/transverse-torsional modes of Models I and II are strikingly dierent from
the corresponding modes of Models III and IV. A more precise understanding of this interesting
observation may be attained by studying the numerical dierence among the modes through the
modal assurance criterion (MAC).
Table V. Comparison of modal characteristics.

Numerical Experimental
Mode no. Model I Model II Model III Model IV model [5] observation [6]
Modal frequencies (Hz) and nature of modes
1 0.371 0.373 0.373 0.375 0.371 0.375
(exural) (exural) (exural) (exural) (exural) (exural)
2 0.48 0.48 0.49 0.5 0.5 0.5
3 0.57 0.58 0.57 0.58 0.577 0.56
(tor-trans) (tor-trans) (trans-tor) (trans-tor) (tor-trans) (trans-tor)
4 * * * * * 0.63
(transverse)
5 * * * * * 0.70
(tor-trans)
6 0.63 0.68 0.70 0.73 0.633 0.74
(trans-tor) (trans-tor) (tor-trans) (tor-trans) (tor-trans) (tor-trans)
7 0.77 0.78 0.78 0.8 0.77 0.8
8 * * * * * 0.80
(torsional)
9 0.73 0.78 0.81 0.86 (mixed-tor) 0.733 0.89
(mixed-tor) (mixed-tor) (mixed-tor) (mixed-tor) (mixed-tor)
10 0.85 0.87 0.88 0.88 0.854 0.89
The * indicates missing mode in numerical models.
DOI: 10.1002/stc
MAC is a measure for examining the numerical similarity or dissimilarity of two modes,
which may not be perceptible by visual examination. The MAC number is dened as a matrix of
normalized inner product of mode shape vectors [1] as expressed in the following equation:
P
j pl1 jal;i jbl;j j2
MAC Pp P 7
l1 jal;i jal;j pl1 jbl;i jbl;j
where jal;i is the modal amplitude at the location l of the ith mode of model a, jbl;i is the modal
amplitude at the location l of the ith mode of model b and p is the order of mode shape vector.
MAC equal to unity indicates good agreement between the two mode shapes, whereas the same
equal to zero indicates no agreement between the same.
The comparative MAC study corresponding to dierent modes among the numerical models
is presented in Table VI. Visual observations have also been tabulated to supplement the MAC
values. MAC is examined for a particular mode between Models I and III, i.e. to observe the
dierence between the modal characteristics of model with cable modeled as truss element and
that with cable modeled as catenary element. These two classes of models do not consider PD
eect for deck and pylon in nonlinear static analysis. Similarly, the MAC is examined for
Models II and IV, which take into account the PD eect for deck and pylon in nonlinear static
analysis.
It has been observed that the MAC numbers are close to unity corresponding to the numerical
mode numbers 1, 2, 5, 6 and 7. The MAC numbers for higher modes are also close to unity and
have not been reported. On the other hand, it is observed from Table VI that the MAC numbers
corresponding to 3rd and 4th modes are close to zero. It implies that the 3rd and 4th modes of
the models with cables modeled as truss elements (Models I and II) are entirely dierent from
the respective modes of the models with cables modeled as catenary element (Models II and IV).
By visual inspection, the 3rd mode is transverse-torsional in Models III and IV, whereas it is
torsional-transverse in Models I and II. Similarly, the 4th mode is torsional-transverse in
Models III and IV, whereas it is transverse-torsional in Models I and II.
6.3. Selection of updated model

In this section, the best updated model is selected from the four numerical models. It has already
been observed from Table V that Model IV simulates the experimentally observed modal
Table VI. Comparison of MAC values.

Nature of mode MAC values
Numerical Models I Models II
mode number Models I and II Models III and IV and III and IV
1 Flexural Flexural 0.99 0.98
3 Torsional-transverse Transverse-torsional 2:3e 08 3:5e 08
4 Transverse-torsional Torsional-transverse 2:6e 13 2:7e 13
6 Mixed-torsional Mixed-torsional 0.99 0.98
DOI: 10.1002/stc
characteristics to the closest extent. As such, this model is the most suitable update of the
numerical model of Wilson and Gravelle [5]. However, a more comprehensive comparison of
this model with the earlier numerical model [5] vis-a-vis the experimental result presented by
Wilson and Liu [6] has been attempted here.
The comparison of frequencies of the rst few exural modes has been presented in Figure 11.
It has been observed that the proposed numerical Model IV conclusively updates the
earlier numerical model [5] as far as capturing of identied exural frequencies is
concerned. Most importantly, the models with catenary element (Models III and IV) are
successful in simulating the observed transverse-torsional behavior of the structure. Thus,
the most signicant observation of this study is the representation of the transverse-
torsional mode by Model IV (as well as by Model III). It is worth recalling that this
distinctive behavior of the model is due to catenary element being weaker in the plane
orthogonal to the plane of the cable (for the present case, the global transverse plane). This
conclusively establishes the importance of catenary element in modeling the cable in a cable-
stayed bridge. It has also been observed that among Models III and IV, the latter represents the
transverse behavior better and the frequencies captured by it are closest to the experimental
results.
Comparison of torsional frequencies obtained from the numerical model proposed by Wilson
and Gravelle [5] and that obtained from the updated model (Model IV) against those of the
experimental observation is shown in Figure 12. It is evident from the plot that Model IV
suitably updates the earlier numerical model with respect to the torsional modes as well.
Another noteworthy observation is that there is a softening eect in all the numerical models as
Figure 11. Comparison of exural modes for dierent models.
DOI: 10.1002/stc
Figure 12. Comparison of torsional modes for dierent models.
Figure 13. First (exural) mode.
Figure 14. Second (exural) mode.
the number of torsional modes increases. This is in agreement with Wilson and Gravelle [5]. It
has been explained that as the section becomes more and more twisted in the case of higher
torsional modes, the actual structure becomes stier due to warping resistance. In contrast, the
torsional stiness of the FE model does not change as the equivalent torsional constant Jeq
remains constant. This softening eect is more pronounced in Models I and II as well as in the
numerical model proposed by Wilson and Gravelle [5]. However, it is least conspicuous in
Model IV with its higher torsional frequencies being fairly close to those of the experimentally
determined values. This is another important observation in favor of Model IV to be regarded
as the best updated model. Important mode shapes corresponding to the updated model (Model
IV) have been shown in Figures 1319.
DOI: 10.1002/stc
Figure 15. Third (transverse-torsional) mode.
Figure 16. Fourth (torsional-transverse) mode.
Figure 17. Fifth (exural) mode.
7. CONCLUSIONS
This study has successfully updated the FE model of a cable-stayed bridge to represent the
experimentally observed modes to the closest extent. The most signicant observation of
the study is that the updated model with the cables modeled as catenary elements reproduces the
transverse behavior of the bridge, which the earlier numerical model with cable as the truss
element failed to represent. Simultaneously, this model also reproduces the exural modes more
closely. The model further succeeds in the representation of higher torsional modes with
frequencies closer to the experimentally determined values. Thus, the proposed model is an
DOI: 10.1002/stc
Figure 18. Sixth (mixed torsional) mode.
Figure 19. Seventh (exural) mode.
appropriate update of the earlier model. The model is also numerically compact as it uses
constraint equations for the representation of bearing elements as well as rigid links. The study
employs a mixed-incremental-iterative nonlinear static analysis scheme accounting for PD
eect for deck and pylon to arrive at the dead-load-deformed conguration. This model, by the
virtue of its compactness, may serve as a base for linear dynamic analysis and also for active
control study.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the help of Dr J. M. Caicedo, University of South Carolina, U.S.A., in
allowing the rst author to use the FEM toolbox in Matlab1 developed by him.
REFERENCES
1. Zhang QW, Chang CC, Chang TYP. Finite element model updating for structures with parametric constraints.
Earthquake Engineering Structural Dynamics 2000; 29:927944.
2. Aktan E, Catbas N, Turer A, Zongfen Z. Structural identication: analytical aspects. Journal of Structural
Engineering (ASCE) 1998; 124(7):817829.
3. Brownjohn JMW, Xia P-Q. Dynamic assessment of curved cable-stayed bridge by model updating. Journal of
Structural Engineering 2000; 126(2):252260.
4. Zhang QW, Chang TYP, Chang CC. Finite-element model updating for the Kap Shui Mun cable-stayed bridge.
Journal of Bridge Engineering 2001; 6(4):285293.
5. Wilson JC, Gravelle W. Modelling of a cable-stayed bridge for dynamic analysis. Earthquake Engineering and
Structural Dynamics 1991; 20:707721.
6. Wilson JC, Liu T. Ambient vibration measurement on a cable-stayed bridge. Earthquake Engineering and Structural
Dynamics 1991; 20:723747.
DOI: 10.1002/stc
7. Jayaraman HB, Knudson WC. A curved element for analysis of cable structures. Computers and Structures 1981;
14:325333.
8. Karoumi R. Some modeling aspects in the nonlinear nite element analysis of cable supported bridge. Computers
and Structures 1999; 71:397412.
9. Wang P-H, Yang C-G. Parametric studies on cable-stayed bridges. Computers and Structures 1996; 60(2):243260.
10. Ren W-X, Peng X-L. Baseline nite element modeling of a large span cable-stayed bridge through eld ambient
vibration tests. Computers and Structures 2005; 83:536.
11. Nazmy AS, Abdel-Ghaar AM. Three dimensional nonlinear static analysis of cable-stayed bridges. Computers and
Structures 1990; 34(2):257271.
12. SAP2000, Analysis Reference Manual. Computers and Structures Inc.: Berkley, CA, U.S.A.
13. Cheng FY. Matrix Analysis of Structural Dynamics}Application and Earthquake Analysis. Marcel Dekker Inc.: New
York, 2000.
DOI: 10.1002/stc

Design of An Active Controller For Quincy Bayview Bridge, Against Seismic Excitation - Model Updatingdutta2008 PDF

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Design of An Active Controller For Quincy Bayview Bridge, Against Seismic Excitation - Model Updatingdutta2008 PDF

Încărcat de

Drepturi de autor:

Formate disponibile

STRUCTURAL CONTROL AND HEALTH MONITORING

Struct. Control Health Monit. 2008; 15:10571077

Design of an active controller for Quincy Bayview Bridge,

Atanu Kumar Dutta, Anjan Dutta*,y and Sajal Kanti Deb

Received 21 May 2007

3.1. Modeling of deck

Figure 1. Line diagram of the bridge with cable type.

Figure 2. The equivalent steel C-section for deck.

3.2. Modeling of pylon

Figure 3. Finite element model for deck.

Table I. Mass moment of inertia kg m2 correction at the spine node.

3.3. Modeling of cable

Figure 4. Catenary element.

Figure 5. Flowchart for catenary element analysis.

3.4. Rigid links

u1s u1m r2m Lz r3m Ly

where Lx xs xm ; Ly ys ym and Lz zs zm : Here, the subscript s stands for slave

3.5. Modeling of bearing

Figure 6. The master and slave nodes in pylon-bearing connection.

Figure 7. Finite element model.

4. NONLINEAR STATIC ANALYSIS

Figure 8. Flowchart for the nonlinear static analysis.

6. RESULTS AND DISCUSSION

6.1. Nonlinear static analysis: dead-load-deformed shape

Figure 9. Deected shape for Model I.

Figure 10. Deected shape for Model IV.

6.2. Modal analysis}frequencies and mode shapes

Table V. Comparison of modal characteristics.

6.3. Selection of updated model

Table VI. Comparison of MAC values.

Figure 11. Comparison of exural modes for dierent models.

Figure 12. Comparison of torsional modes for dierent models.

Figure 13. First (exural) mode.

Figure 14. Second (exural) mode.

Figure 15. Third (transverse-torsional) mode.

Figure 16. Fourth (torsional-transverse) mode.

Figure 17. Fifth (exural) mode.

Figure 18. Sixth (mixed torsional) mode.

Figure 19. Seventh (exural) mode.

S-ar putea să vă placă și