Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE.

For personal use only; all rights reserved.

A Stochastic Finite Element Approach to Determine the Safety of Suspension

Bridge Cables

A. Montoya1, R. Betti2,3, G. Deodatis2,4, and H. Waisman2,5

1
University of Texas at San Antonio, Department of Civil and Environmental
Engineering, One UTSA Circle, San Antonio, TX 78249; PH (210) 458-7516; FAX
(210) 458-6475; email: arturo.montoya@utsa.edu
2
Columbia University, Department of Civil Engineering and Engineering Mechanics,
500 West 120th Street #610, New York, NY 10027; PH (212) 854-3143; FAX (212)
854-6267; email: 3betti@.civil.columbia.edu, 4 deodatis@civil.columbia.edu,
5
waisman@civil.columbia.edu

ABSTRACT

A new methodology to determine the safety of suspension bridge main cables is

proposed and illustrated on a corrosion-deteriorated cable composed of 9061 wires.
The approach is the first one incorporating a finite element (FE) model to predict the
cables failure load, accounting for load recovery due to friction in broken wires and
simulating the reduced cable's strength as a three dimensional random field. In order
to obtain the breaking load of a cable, the load is increased gradually (quasi-static
loading) in a cables FE model, having wires break a few at a time according to their
residual strength. Because of the load transfer to surrounding wires, the breakage of
an individual wire affects the stress state inside the surrounding wires. This local
damage eventually causes a global reduction in the load carrying capacity of the
cable, up to a complete failure. The safety of the cable is determined through a Monte
Carlo simulation, in which the reduced strength of the cable is generated for every
realization through the Spectral Representation Method (SRM) and is input as a
material parameter in the FE model. The statistics of the load that will drive a
suspension bridge cable to failure under a hypothetical deterioration state are obtained
at the end of the simulation.

INTRODUCTION
The structural function of the main cables in suspension bridges is to transfer the
tension load, derived by supporting the roadway, to the towers. The main cables are
composed of thousands of high strength parallel steel wires with a diameter of
approximately 5 mm bundled together in strands either built in situ or prefabricated.
These strands are then compacted and tightened together and eventually the cross
section of the cable becomes semi-circular. The wires in pristine conditions have a
strength ranging from 1570 MPa to 1800 MPa. However with aging, fatigue loading,
and harsh environmental conditions, the wires strength reduces significantly (Shi et
al., 2007). Field observations of aging suspension bridges indicate serious distress of

Computing in Civil Engineering (2013)

 2 COMPUTING IN CIVIL ENGINEERING
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

cable wires. Many wires are corroded and even broken, jeopardizing the safety of the
entire bridge. Moreover field surveys performed by Suzumura and Nakumura (2004)
have indicated that the corrosion level varies according to the wire positions inside
the cable. Current inspection procedures rely on visual observations which are
considered unreliable and expensive. Today the biggest challenge suspension bridge
owner authorities face is to estimate the current and remaining safety of a main cable
in order to help them decide whether it is necessary to provide immediate
maintenance or rehabilitation (or even replacement) of such cables.
In order to provide an accurate assessment of the cable's remaining strength, it
is critical to model the load transfer mechanism between wires and the corrosion
uncertainty within the cable. One of the most important structural characteristics of
suspension bridges is the high internal redundancy of its main cables. If a wire breaks,
the load carried by that wire is redistributed to the unbroken wires: Montoya et al.
(2012) showed through some numerical examples that, in the vicinity of the break,
the neighboring wires take most of the load burden while, at some distance away
from the break, the broken wire recovers fully or partially its load carrying capacity
due to friction. The overall carrying capacity of the cable decreases as individual
wires start to break and, as the number of broken wires increases, it might reach a
point that corresponds to the complete failure of the cable.
The objective of this paper is to introduce a methodology that determines the
safety of an existing cable by estimating the failure load of a suspension bridge main
cable. This is accomplished through the Monte Carlo Simulation by generating a
number of realizations of the cables strength within a FE model.

PROPOSED METHOD TO DETERMINE THE CABLES SAFETY

The Finite Element Method can be used as the means to incorporate the interaction
mechanisms between wires into the assessment of the remaining strength of a cable.
A FE model can serves as an invaluable tool for evaluating bridges that exhibit
distress signs. In addition, because of the uncertainties associated with corrosion, a
stochastic analysis is required to determine the effects of corrosion uncertainty on the
cable's failure load. The drawbacks associated with generating a model for a large
number of wires and integrating it into a stochastic analysis is the computational
effort associated with it. However, nowadays parallel computing provides a feasible
way to solve stochastic mechanics problems.
The approach proposed in this paper estimates the failure load of a suspension
bridge main cable by performing a finite element analysis coupled with a Monte
Carlo Simulation. Each realization of the simulation considers different deterioration
conditions in each wire. The cable considered for this study has a diameter of 24.37
cm, a prescribed length of 18.3 m, and is composed of 9061 wires with a diameter of
0.4826 cm as shown in Figure 1. The length of the model is associated with the
recovery length of a broken wire within a cable. Assuming the cable has been well
compacted, the overall strength of the cable can be determined from a segment of
limited length, since good compaction of the wires guarantees load recovery in
broken wires (Mateo, 1994). Three clamps having a width of 20 cm are placed over
the length, spaced at 6.096 m from each other (see Figure 1b). The cable is pulled at
both ends (in the length direction: z-axis,).

Computing in Civil Engineering (2013)

 COMPUTING IN CIVIL ENGINEERING 3
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)
Figure 1. Cable's model: (a) cross sectional and (b) longitudinal views.

Finite Element Model for Main Cables

In order to facilitate a full stochastic analysis of the problem and reduce the resulting
computational effort to manageable levels, truss elements were selected to model the
wires. Figure 2(a) illustrates a segment of two wires in a three dimensional space,
exemplifying the behavior between a broken and an unbroken wire. The left wire
(simulating an unbroken wire) carries an axial traction Tz at the end. The wires
transfer stresses through friction related shear forces that are generated along their
contact line of interaction. This interaction allows the broken wire to recover load and
alleviates the load burden on the unbroken wire. An equivalent system is presented
in Figure 2(b) when modeling wires with truss elements. Truss elements having a
stiffness equivalent to the wire are placed at the center of each wire. The external
force resulting from the axial traction is placed at node 1 of the unbroken wire. The
frictional interaction between wires can be modeled using spring elements connecting
both wires. The shear stress at the contact line is translated into a shear force, (),
acting at the nodes of the spring elements.
Coulombs Friction Law (e.g. Wriggers (2006)), specifies that the maximum
shear force, , that can be transferred between two surfaces is a function of a normal
force:

where is the friction coefficient (an empirical property of the contacting
materials), is the normal applied force, and is the resulting tangential force due
to friction in a direction that is opposite to the motion of the object. For main cables,
the normal force is provided by the compaction action of the steel bands. The
pressure generated due to a concentrated force decays to zero exponentially when
moving in the longitudinal z-direction as specified by the Boussinesqs solution to a
point load applied on a half space of a semi-infinite solid (Boussinesq, 1885). Thus,
the clamping action and frictional interaction between wires takes place in very close
proximity to the clamps.

Computing in Civil Engineering (2013)

 4 COMPUTING IN CIVIL ENGINEERING
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)
Figure 2. (a) Load transfer mechanism between a broken and an unbroken wire and
(b) an equivalent finite element model

An equivalent user friction model to capture the Coulombs Law consists on

providing an elasto-plastic behavior to the spring elements between the wires in
contact. This concept was proposed by Waisman et al. (2011). The stick behavior is
characterized by the linear elastic stiffness and the slip part by the perfect plasticity,
where the yielding force FT represents the maximum shear force that can be
transferred at that point. The yielding force decays when moving away from the
clamp as observed in Figure 3(a) for four different locations from the clamp. For
illustration purposes, the mesh for 19 parallel wires is shown in Figure 3(b). Note that
the spring elements connect wires in the x-y plane and provide frictional interaction
in the longitudinal direction.
In order to obtain the breaking load of a cable, the applied load is increased
incrementally (quasi-static loading), having only a few wires break at a time. For
breakage to occur, the stress has to exceed the Reduced Tensile Strength (RTS) of the
element. The RTS is input as a material property of the element before loading begins
and is obtained according to the Spectral Representation Method (see next section).
At the initial step of the analysis, all wires are assumed to be unbroken; thus, the
spring elements are inactive. As the cable is loaded, the sections of the wires that
reach their RTS will break. This break is reflected in the model by eliminating the
stiffness of the individual wire element, the so called element deletion technique. As
the element breaks, friction on the surfaces between the broken wire and the
surrounding one transfers load between adjacent wires. Numerically, this is simulated
by activating user friction model elements (springs) connecting the sliding wires.
Montoya et al. (2012) emphasize that when such a break happens in one wire, the
load in other wires (depending on their proximity to the broken wire) will
automatically be adjusted and hence may enter a critical stress state which will lead to
their failure, i.e. the domino effect.

Modeling StrengthVariation in Main Cables

Field inspections have indicated that in small regions within the cables cross section,
neighboring wires will be at a similar corrosion stage. However, current
methodologies treat the wire strengths as uncorrelated (Mateo et al., 1994) or only
account for the correlation along its length (Shi et al., 2007). The weakest points of
the simulated strength of two adjacent wires can be considerably far from each other.

Computing in Civil Engineering (2013)

 COMPUTING IN CIVIL ENGINEERING 5
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

This implies that the one dimensional, decoupled random field approach cannot
simulate homogeneous-like corrosion in small regions within the cross section of the
cable.

(a) (b)
Figure 3. (a) Elasto-plastic behavior of the spring (green) elements to account for
friction induced by the clamps (b) and cables FE model.

The strength of a corrosion-deteriorated cable is simulated as a three-

dimensional stochastic field by the Spectral Representation Method (SRM)
(Shinozuka and Jan, 1972). This approach considers the spatial correlation of the
cable strength both within the cross-section (over different wires) and along the
length of the cable. Shinozuka and Deodatis (1996) have provided the theoretical
background for simulating a homogeneous stochastic field in three dimensions. The
prescribed three dimensional Spectral Density Function (SDF) required by the SRM
formulation can be obtained by combining a SDF that captures the behavior in the
cross section of the cable (x-y plane) and a SDF that accounts for the spatial
correlation along the length (z-direction). The correlation is obtained from field data
obtained from an actual main cable. Sample wires extracted from the cable are cut
into small specimens of appropriate length (usually 25 30 cm) for tensile testing. A
standard tensile strength test is performed on each sample to define the average
strength and average elongation and their statistical deviations.
The pool of data for this study came from a thorough inspection performed on
the Williamsburg Bridge cable in 1988 (Steinman,1988). The data indicated a strong
dependency of the strength with the y-coordinate (cables top to bottom). The
computational time increases considerably when generating sample functions of a
stochastic field in three dimensions according to the SRM. However, the Fast Fourier
Technique (FFT) decreases the computational time dramatically, facilitating a three-
dimensional stochastic analysis of the cable's strength (Brigham, 1988). Figure 4
shows the strength generated for three pairs of adjacent wires according to the FFT
technique for three-dimensional stochastic fields (Shinozuka and Deodatis, 1996).
The wire pairs are taken from the bottom, center, and top sectors of the cables cross

Computing in Civil Engineering (2013)

 6 COMPUTING IN CIVIL ENGINEERING
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

section. The distribution of the wire strength corresponds well with the conditions
observed in the visual inspections of the cable.

Figure 4. Generated strength of wires by considering the spatial correlation of the

cable strength in space. The weakest point along the length is represented by o.

PARALLEL COMPUTING
Due to the overall number of equations of the model, parallelization of the problem is
essential to facilitate the stochastic analysis that determines the effect of corrosion-
related uncertainties on the cable's failure load. The mesh, composed of truss and
spring elements, is independent of the realization; while the material property that
determines the wire breaks (RTS) in each wire changes for each run according to the
stochastic representation. The simulation methodology is illustrated in the flow chart
of Algorithm 1.

Algorithm 1
1. Read Input Files. Store general finite element information.
2. Loop1 over each Monte Carlo realization.
A. Set the displacement vector and force vector to zero. u=0, F=0.
B. Generate the wire corroded strengths
a. Loop2 over each displacement increment (20 steps through the
displacement controlled method).
i. Loop3 for convergence of Nonlinear Problem
1. Solve Linear Equations
ii. End Loop3
iii. Increase displacement of the cable , +1 = +
b.End Loop2
3. End Loop1

RESULTS
The failure load of the cable is obtained through a Monte Carlo Simulation
(MCS) with 1000 realizations. The cable is loaded using the displacement controlled
approach for twenty steps at increments of u=0.2298 cm. The clamps are assumed

Computing in Civil Engineering (2013)

 COMPUTING IN CIVIL ENGINEERING 7
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

to be sufficiently tightened to provide full recovery after one cable band. Figures 5
shows a sample step for a run of the model. The force-displacement plot (Figure 5a)
compares the response of the cable to the response of a cable in perfect conditions.
The cross-sectional plot (Figure 5b) is at the critical section along the cables length
where the greatest number of broken wires is found for this run. A cluster of broken
wires (dark blue wires indicating zero stress) with random pattern is formed at the
bottom sector. A highly stressed (orange section) region surrounds this cluster of
broken wires.

(a) (b)
Figure 5. Sample Step of a Realization: (a) Force-Displacement Curve of the cable
and (b) Longitudinal Stresses (MPa) in the 9061 wires.

Load redistribution allows the cable to carry the new load level with an
uneven distribution among wires. Eventually, the cable fails in a domino like pattern
in which the failure of weak wires causes other less weak wires to fail at the same
external load. The load is kept constant for each of the 1000 realizations, but the cable
fails at a different load as shown in Figure 6(a). The response of the system is
nonlinear for each realization; the stiffness of the overall cable significantly decreases
after the breaks. The corrosion-related uncertainties play a major effect on the
breaking load of the cable, as significant scattering is observed in the failure load. The
histogram for the breaking load is illustrated in Figure 6(b).

CONCLUSIONS
Reliable assessment of the cable's remaining capacity is needed by bridge
authorities in order to make important decisions regarding maintenance and
rehabilitation of main cables in suspension bridges. The proposed methodology
provides an estimation of the distribution of the failure load, which may help
preventing an unexpected failure. This approach is the first one incorporating a finite
element analysis within a Monte Carlo simulation to evaluate the performance of
main cables. This method accounts for the correlation of the cable's strength along its
length and within the cross section. The proposed formulation is capable of capturing
the nonlinear behavior of the cables strength as wires break, starting from the most
corroded areas in the cables cross-section (usually the bottom sector).

Computing in Civil Engineering (2013)

 8 COMPUTING IN CIVIL ENGINEERING
Downloaded from ascelibrary.org by UFSC - Universidade Federal de Santa Catarina on 08/16/16. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)
Figure 6. 1000 realizations: (a) Force-displacement curve and (b) histogram

REFERENCES
Boussinesq, J. (1885). Appications des potentials a l'tude de l'quilibre et du
mouvement des solides lastiques.Gauthier-Villars.
Brigham, E.O. (1988). The Fast Fourier Transform and its Applications. Prentice
Hall, New Jersey.
Gronquist Bridsall Steinman, Boynton and Columbia University, 1998. Williamsburg
Bridge cable investigation program: Final report. Submitted to the New York
State Dept of Transp. and New York City Dept. of Transportation, New York.
Mateo, J., Deodatis, G. and Billington, D. (1994) . Safety analysis of suspension
bridge cables:Williamsburg bridge. ASCE-Journal of Structural Engineering,
120(11):3197-3211.
Montoya, A., Waisman, H. and Betti, R. (2012). A simplified contact-friction
methodology for modeling wire breaks in parallel wire strands. Computers
and Structures, 100-101:39-53.
Shi, Y., Deodatis, G. and Betti, R. (2007). Random field-based approach for
strength evaluation of suspension bridge cables. ASCE-Journal of Structural
Engineering, 133(12):1690-1699.
Shinozuka, M. and Deodatis, G. (1996). Simulation of multi-dimensional gaussian
stochastic fields by spectral representation. Applied Mechanics Rev.,
49(1):29-52.
Shinozuka, M. and Jan, C. M. (1972) . Digital simulation of random processes and
its applications. J. Sound and Vibration, 25, 111128.
Suzumura, K. and Nakamura, S. (2004). Environmental factors affecting corrosion
of galvanized steel wires. ASCE-Journal of Materials in Civil Engineering,
16(1):1-7.
Waisman, H., Montoya, A., Betti, R. and Noyan, I.C. (2011). Load transfer and
recovery length in parallel wires of suspension bridge cables. ASCE-Journal
of Engineering Mechanics, 137(227).
Wriggers, P. (2006). Computational contact mechanics. 2nd edition, Springer,
Heidelberg.

Computing in Civil Engineering (2013)