Structural Engineering and Mechanics, Vol. 62, No.

1 (2017) 85-95
DOI: https://doi.org/10.12989/sem.2017.62.1.085 85

Form-finding analysis of suspension bridges

using an explicit Iterative approach
Hongyou Cao1a, Yun-Lai Zhou1, Zhijun Chen2b and Magd Abdel Wahab3,4,5
1
Department of Civil and Environmental Engineering, National University of Singapore, 117576, Singapore
2
School of Civil Engineering & Mechanics, Huazhong University of Science & Technology, Wuhan, 4300743, China
3
Division of Computational Mechanics, Ton Duc Thang University, Ho Chi Minh City, Vietnam
4
Faculty of Civil Engineering, Ton Duc Thang University, Ho Chi Minh City, Vietnam
5
Soete Laboratory, Faculty of Engineering and Architecture, Ghent University,
Technologiepark Zwijnaarde 903, B-9052 Zwijnaarde, Belgium

(Received September 19, 2016, Revised December 16, 2016, Accepted January 23, 2017)

Abstract. This paper presents an explicit analytical iteration method for form-finding analysis of suspension bridges. By
extending the conventional analytical form-finding method predicated on the elastic catenary theory, two nonlinear governing
equations are derived for calculating the accurate unstrained lengths of the entire cable systems both the main cable and the
hangers. And for the gradient-based iteration method, the derivation of explicit calculation for the Jacobian matrix while solving
the nonlinear governing equation enhances the computational efficiency. The results from sensitivity analysis show well
performance of the explicit Jacobian matrix compared with the traditional finite difference method. According to two numerical
examples of long span suspension bridges studied, the proposed method is also compared with those reported approaches or the
fundamental criterions in suspension bridge structural analysis, which eventually confirms the accuracy and efficiency of the
proposed approach.
Keywords: form-finding analysis; suspension bridge; analytical method; elastic catenary cable; finite element analysis

1. Introduction cable sustains both self-weight along its length and

concentrated forces from the hangers. The FE-based method
Suspension bridges analysis requires a form-finding identifies the target cable shape by updating the nodal
analysis prior to the finite element (FE) model to determine positions and/or initial force of the cable elements
the geometric parameters and the initial force of the cable according to the nonlinear structural analysis results based
system since the configuration of a cable is arbitrary in on an assumed initial suspension bridge model. Wang and
stress-free state and its stiffness is derived from the applied Yang (1996) proposed a successive substitution approach,
external loads in contrast to the frame structures (Irvine which predicates on the Newton-Raphson method and
1981). In the past years, many numerical and analytical modifies the initial tension of the cable elements, to
form-finding approaches have evolved for the tensegrity minimize vertical deformation and/or bending moment of
structures (Zhang et al. 2014, Koohestani and Guest 2013, the girder under dead load. Karoumi (1999) utilized the
Faroughi et al. 2014) and the suspension bridges (Sun et al. trial-and-error procedure for the form-finding analysis of
2014, 2016, Kim et al. 2014, Jung et al. 2015) as well. suspension bridges. Like the simplified method, these
The form-finding analysis methods for suspension approaches also assume that the configuration of main cable
bridges can be summarized into three categories: 1) the follows a parabolic shape and fix the nodal position during
simplified approach; 2) the FE-based method; and 3) the the iteration process. However, the final configuration of
analytical method. The simplified method assumes the main the main cable is not a parabola rigorously. To overcome
cable to be a parabola under uniform load along its span these drawbacks, several methods, such as the target
direction (Chen et al. 2014, Gimsing and Georgakis 2011, configuration under dead load (TCUD) method (Kim and
Lonetti and Pascuzzo 2014). However, the accuracy of the Lee 2001) , the initial force method (Kim et al. 2002), the
simplified approximation is insufficient since the main improved TCUD (I.TCUD) method (Kim and Kim 2012),
the G.TCUD method (Jung et al. 2013) and the coordinate
iteration method (Sun et al. 2014), have been developed
Corresponding author, Ph.D. successively. These methods find out the target profile of
E-mail: zhouyunlai168168@gmail.com the bridge by treating the nodal positions of the main cable
Corresponding author, Professor and/or the unstrained length of the bridge’s members, e.g.,
E-mail: magd.abdelwahab@tdt.edu.vn, the main cable, the pylon and the girder as variables and
magd.abdelwahab@ugent.be solve them in NFEM iteration formula with additional
a
Research Fellow constraints derived from bridge’s geometric requirements
b
Professor under dead load. However, the convergence of these FE-
Copyright © 2017 Techno-Press, Ltd.
http://www.techno-press.org/?journal=sem&subpage=8 ISSN: 1225-4568 (Print), 1598-6217 (Online)
86 Hongyou Cao, Yun-Lai Zhou, Zhijun Chen and Magd Abdel Wahab

lx methods for form-finding analysis of suspension bridge.

This study extends the elastic catenary theory based
O
x analytical form-finding method to be capable of calculating
Xj Fj
the accurate unstrained length of the entire cable system,
j including both main cable and hangers. The Jacobian matrix
Fx for the iteration process is derived in explicit form in
Fi
ly

i accordance to the differential form of elastic catenary

Fy equations and then sensitivity analysis is carried out to
Xi
validate the reliability of this explicit method in comparison
y w with FDM. Finally, two long span suspension bridges are
numerically studied, and the results are also compared with
those existing approaches.
Fig. 1 Two-dimensional catenary cable

2. Iterative approach for form-finding analysis

based approaches highly depend on the assumed initial FE
model (Wriggers 2008). This section introduces the elastic catenary theory based
The analytical approaches, which separates the bridge analytical form-finding analysis approach for suspension
into two substructures: the cable system and the pylon- bridges (Chen, Cao, and Zhu 2013). This approach includes
girder, couples them via the hanger forces and deals with the entire cable system information (main cable and
the cable system independently using analytical method, hangers) into the governing equations.
possess a much higher computational efficiency and faster
convergence in contrast to the simplified and the FE-based 2.1 Elastic catenary equations
methods. Based on the elastic catenary equations, O’Brien
and his co-workers (1964, 1967) proposed a successive The elastic catenary theory assumes a perfectly flexible
calculation procedure that divides the cable into segments at cable with self-weight uniformly distributed along its length
the positions of the concentrated loads and solves the static as shown in Fig. 1, where xi and xj denote the coordinates of
equilibrium to determine the cable configuration. This node i and j, Fi and Fj refer to the nodal forces at both ends,
method is mainly used to predict the accurate cable respectively. The elastic catenary equations follow
configuration of the suspension bridges during the hanger
installation process (Chen et al. 2015, Fan et al. 1999, Luo x j  xi  [lx , l y ]T  φ  Fi , L0  (1)
2004) since it requires the unstrained length of each cable
element known as prior. Chen et al. (2013) improved this F j  Fi  wL0 (2)
method to the form-finding analysis of the suspension under
bridge finished stage through considering the unstrained where L0 represents the unstrained length of element,
length instead of the horizontal distance of the cable w=[0,w]T, w means the self-weight per length unit of the
element (the space between the two adjacent hangers) as cable. The two components of φ are given in Eqs. (3) and
variables. Wang et al. (2015) extended this approach to the (4) as follows
preliminary analysis of self-anchored suspension bridge
 
Fx L0 Fx  
 
2
while neglecting the self-weight of the hangers in their main x    ln  Fx  wL0  Fy
2
 wL0  Fy 
cable analysis model. Similar to above approaches, Jung et EA w  
(3)
al. (2015), proposed a simplified analytical method for the
optimization of the initial shape analysis in self-anchored
 ln  Fx2  Fy2  Fy 
suspension bridges. This approach assumes the main cable
segments between hangers to be parabolic under self-weight wL20 1  2 
 
Fy L0 2
instead of the elastic catenary element. In essence, the y      Fx  wL0  Fy  Fx2  Fy2  (4)
EA 2 EA w  
analytical form-finding procedure becomes solving a series
of implicit nonlinear governing equations derived from the where E and A denote the elastic modulus and the cross-
cable system. sectional area of the cable, respectively.
The Netwon-Raphson iteration method is the most
widely used and efficient approach to solve the nonlinear 2.2 Governing equations
equations. However, the numerical methods, e.g., finite
difference method (FDM), has to be adopted to approximate Form-finding analysis is an inverse problem
the Jacobian matrix during the governing equations determining the nodal positions and cable tensions or
resolving process due to the implicit expression of the unstrained lengths according to the predetermined
equations (Chen et al. 2013, Luo 2004). Although the FDM geometric parameters of the bridge under the dead load
is simple in application, both truncation and round-off state. The following steps detail the analytical form-finding
errors that it often suffers will heavily affect the Jacobian analysis procedure,
matrix approximation quality (Lund 1994). An explicit (1) Dividing the bridge into two sub-structures coupled
Jacobian matrix estimation method will not only overcome by the hanger forces. As shown in Fig. 2(a), ls and lm denote
these defects but also improve the current analytical the span of side-span and main-span,
 Form-finding analysis of suspension bridges using an explicit Iterative approach 87

2n
 d x  lm
i
ls i 1 ls

f
d xi

a. A suspension bridge

Ri
Side span 2n-1 supports and the reactions denoted by Ri Side span
b. The hangers are replaced by supports

Ni
2n-1 node forces Ni acting on the cable and Ni=Ri
Side span Side span
c. The hangers are replaced by node forces

Fig. 2 Structural decomposition of the girder-tower system and cable system

y (x2n+1,y2n+1)

o x 2n
h

(x2n,y2n)
1 1
F x
F i
(x2n-1,y2n-1)
Fx1 2n-1
f

(x2,y2) A
(x1,y1) 1 (x3,y3) (xn+1,yn+1)
2
3
Ni  Ghi

d 1x d x2 d x3 d xn d xn 1 d x2 n 1 d x2n

(a) The main cable system

Upper anchorage on the

main cable

Ni  Ghi Tii 1
hi

Lower anchorage on
Ni the girder T ji Ni  Ghi

The girder
(b) The free body diagram of (c) The free body diagram at the detail
No.n+1hanger of A

Fig. 3 The main cable system and free body diagrams of the connecting points between girder, hanger and main cable

respectively. f refers to the sag at the mid-span of main self-weight of the hanger. The structural parameters such as
i
cable and d means the horizontal distance between the i th spans, the layout of the hangers and the sag-span ratio are
x
th usually determined prior.
and the (i-1) hangers. (3) Calculating the hanger forces at the lower anchored
(2) Replacing the hangers by vertical supports and points, Ni, using the pylon-girder substructure. The cable
calculating the support forces Ri (i denotes the ith support) system thus can be considered as an independent system as
according to the target shape of the stiffening girder under shown in Fig. 2(c).
dead load as shown in Fig. 2(b). By this way, the hanger Fig. 3 illustrates the mechanical sketch of the cable
forces acting on the stiffening girder Ni (i refers to the ith T
hanger) are eventually determined by Ri+Ni=0. The force system. Fi1   Fx1, Fy1  (the subscript and superscript of F
 
acting on the main cable equals the summation of Ni and the denote the left/right node of the cable element and the
88 Hongyou Cao, Yun-Lai Zhou, Zhijun Chen and Magd Abdel Wahab

element number, respectively) denotes the nodal forces of mid-span and the end are known and can be expressed as
node i of the first cable element. Recalling the elastic
x n 1  x1  l / 2, h / 2  f 
T
catenary element equations Eqs. (1)-(4), the coordinate and (15)
nodal forces of node j for the first cable element become
x 2n 1  x1  l , h
T T
x2  x1  d1x , d1y  (5) (16)
 
where l denotes the span of the bridge, f refers to the mid-
F j  Fi  wL0 (6) span sag of the main cable and h represents the relative
height between neighboring pylons as shown in Fig. 3(a).
 
where, L10   x1 Fx1, Fy1, d1x , d1y   y Fx1, Fy1, L10 .   With Eqs. (13)-(16) and substituting the expression of
As shown in Fig. 3(b), hi (i represents the ith hanger, Li0 into the equations, the governing equations of form-
i=1~2n-1) denotes the length of the hanger with elongation. finding method for a plane-shape cable becomes
Assuming h0i defines the unstrained length of the ith T
     
n i i 1
2n
i i 1 
hanger, and the following equations describe the    y Fx , Fy ,  x Fx , Fy , d x ,   y Fx , Fy ,  x Fx , Fy , d x
i i i i i i
i 1 i 1 (17)
relationship between hi and h0i
T
  d yn 1, d y2n 1    h / 2  f , h 
T
hi Ni  wh x  
0 0 dx  h0i  hi (7)
Eh Ah
2.3 Nonlinear iteration procedure
  Ni  Eh Ah    Ni  Eh Ah  2
 2 Eh wh Ah h i
h0i  (6) The following gradient-based iteration method is
wh
employed to solve Eq. (17),
where wh refers to the weight of the hanger per length unit, Step 1: Assuming the main cable shape follows a
Eh and Ah represent the modulus of elasticity and the cross parabola under dead load, the initial iteration forces,
sectional area of the hangers, respectively. The self-weight Fi1 = [Fx1, Fy1 ]T , can be estimated by the following equations
of ith hanger Ghi becomes:
qd l 2  8 f 2  q m l 2
Ghi  whh0i (9) Fx1   1  2  (18)
8f  3l  8 f

According to the equilibrium condition at ith node as
shown in Fig. 3(c), the tension at node j of ith element and h  4 f 
the tension at i node of (i+1)th element lead to Fy1  Fx1 tan   Fx1 (19)
l

 Fxi    Fyi  wLi0 

2 2
T ji  (10) where, qd and qm are the dead load and self-weight per unit
length of stiffening girder and main cable, respectively.
Step 2: Calculating the unstrained length of the first
 Fxi    Fyi  wLi0  Ni  Ghi 
2 2
Tii 1  (11) element, L10 , and the nodal coordinate, (x2, y2), and the
The force relationship between jth node of ith element force, F1j , of node j according to Eqs. (1)-(2).
and ith node of (i+1)th element expresses as Step 3: Calculating the node i forces, Fi2  [ Fx2 , Fy2 ]T , of
Fii  Fii 1  wLi01  Ni 1 (12) the second element according to Eq. (12), and calculating
where Ni 1  [0, Ni 1  Ghi 1 ]T , i 
i=2~2n.
2 2n the unstrained length of the second element, L20 , the nodal
If the nodal force at node i of the first cable element, coordinate, (x3, y3), and the forces, F j2 according to Eqs.
T
Fi1   Fx1, Fy1  , is known, calculating the coordinate and (1)-(2). Repeating this procedure until all the cable
  elements’ nodal coordinates and forces are calculated.
nodal force of each element from 1 to 2n successively with Step 4: calculating d yn+1and d y2n+1 according to the y-
Eqs. (5)-(6) and Eq. (12), the coordinate of mid-node and
the end node of the main cable can be obtained as direction coordinate of node 1, n+1 and 2n+1, and checking
the convergence of the iteration procedure using the
 
n T n T
xn 1  x1    d xi , d iy   x1   d xi , y Fxi , Fyi , Li0  (13) following convergence criteria
i 1
  i 1  
2 2
 d n 1  d n 1   d 2 n 1  d 2 n 1 
R   x n 1 x , m    x 2 n 1 x , m    (20)
 
2n T 2n T
x2n 1  x1    d xi , d iy   x1    d xi , y Fxi , Fyi , Li0  (14)  dx   dx 
i 1
  i 1  
where, d xn,m1  h / 2  f , d x2,nm1  h , and  is the tolerance
According to the structural parameters of the bridge,
such as span, sag and the relative height between the defined by users. If the convergence criterion is satisfied,
neighbouring pylons, the coordinates of the main cable at terminate the iteration, or else go to the Step 5.
 Form-finding analysis of suspension bridges using an explicit Iterative approach 89

Step 1: small. Therefore, a compromise between both types of

Estimate the iteration initial values errors is required during application of FDM.
Fig. 4 gives the flowchart for the flexible iteration
Step 2:
procedure. The TCUD of the main cable in the main span is
Calculate No. n element’s unstrained length, obtained under the hypothesis that the tower is in zero-
internal force and node coordinates bending moment state under dead load which means the
n=n+1

t=t+1
horizontal component of the side main cable tension equals
Step 3:
Calculate the node force of the i node of No. to that of the main cable. When calculating the TCUD of
(n+1) element the main cable in the side span, the first term in Eq. (17) can
be removed, and the governing equation of the form-finding
Step 5: analysis for side span is reduced as a single variable
Step 4:
No Calculate the Flexible
The convergence criteria
matrix and calculate the nonlinear equation. The calculation procedure for side span
is satisfied?
new iteration values is also the same as the flowchart shown in Fig. 4 with the
Yes horizontal component of Fi1 as known.

Iteration is terminated

3. Jacobian matrix derivation and sensitivity analysis

Fig. 4 The flow chart of the flexible iteration procedure
This section derives an explicit gradient matrix formula
from the recursion relation in Eq. (12) and the differential
Step 5: If the convergence criteria in Eq. (20) is not form of elastic catenary equations to overcome the
achieved, the following iteration formula is adopted shortcomings of FDM. A sensitivity analysis is also
 d xn,m1   d xn 1,t  conducted to verify the feasibility and efficiency.
 2 n 1    2 n 1, t    J 2 2 Fi
1, t
(21)
 d x , m   d x 
3.1 Jacobian matrix derivation
where, t denotes the iteration step. Fi1,t can be expressed
as According to the flexible iteration procedure,
substituting Eq. (12) into Eq. (17) leads to a form only
1  d  d xn 1,t 
n 1
T
F1,t   J 2 2  2xn,m1 2 n 1, t  (22) respecting to variables Fi1   Fx1, Fy1  .
 d x , m  d x   
T
     
n 1 1 1 1 1 i 
and Fi1,t 1 for the next iteration step t+1 is
2n
1 1 1 1 1 i
   y Fx , Fy ,  x Fx , Fy , d x ,   y Fx , Fy ,  x Fx , Fy , d x 
i 1 i 1  (25)
Fi1,t 1  Fi1,t  Fi1,t (23) T
  d yn 1, d y2n 1    h / 2  f , h 
T
 
Finally, go back to Step 2 for the next iteration step.
The matrix shown in Eq. (21) is called flexible matrix Assuming function U as
indicating the relationship between the vector function,  d yn 1  h / 2  f 
T U 
φ  d yn1, d y2n1  , and the vector variable, F 1,t
from  d 2 n 1  h 
(26)
  i
 y 
mathematical viewpoint. Because function φ expresses in an
Linearizing the nonlinear function U by first-order
implicit form, the gradient matrix, [J], usually is
approximated by FDM. Taylor expansion at Fm1 , 
, Fm2n , and then
The FDM approximates the derivative of the response
 
  F 
based on the response function values 1
C1 Cn
0  m 
 
0

i i  x  x j e j   i  x 
U  U Fm
1
, Fm2n
 
,

(27)
x j

x j
(24) C1 Cn   Fm2n 
 C2n 

Δxj is a small perturbation applied to the jth design where,

variable xj, and ej is a vector unit in the jth direction. One
additional function evaluation is need for every design Ci  
y x y 0  ,
y x y 0
  F i , F i , Li  F i , F i , Li   
variable xj to compute the full gradient. Even though, this  Fx i
Fyi 
 
method is straightforward and can be easily implemented
T
for any response function, the approximation quality Fi   Fxi , Fyi  .
strongly depends on the step size Δxj, which is an obvious  
inherent drawback of this method. Both truncation and Recalling Fig. 3, the length of the hangers can be
round-off errors are present in FDM. Truncation errors are calculated by
due to the terms disregarded in the Taylor expansion in Eq. hi  yi 1  yid1  yi   iy  yid1 (28)
(24). A smaller perturbation Δxj will reduce this error. In
contrast, round-off errors increase strongly when Δxj is very Substituting Eq. (8) and Eq. (28) into Eq. (9) leads to the
90 Hongyou Cao, Yun-Lai Zhou, Zhijun Chen
and Magd
L Abdel
1 T  Fy  wL0
Wahab
x
  0  ln 2
Fx EA w T1  Fy
differential equation with respect to  iy
Fx2  1 1 
    (34a)
Ghi w  T1 T1  Fy  T2 T2  Fy  wL0  

Eh Ah
wh   wh  
 yi (29)
 Eh Ah  N i 1   2 Eh Ah wh h i 1
2

 x F F
 x  x (34b)
Eh Ah L0 EA T2
where   .
 Eh Ah  Ni 1   2 Eh Ah wh hi 1
2
When φx=dx is a constant, considering the implicit
The relationship between ΔFi and ΔFi-1 becomes function and Eq. (3), the following relationships can be
according to Eq. (12) obtained
1 0  i 1  Li 1 Li 1  L0  
Fi   F  w  0 , 0  Fi 1  x / x (35a)
 Fx Fx L0.
0 1   Fxi 1 Fyi 1 
 
 0 0 
L0  
 i 1   x / x
  wh   y  iy1  Fi 1 Fy Fy L0.
(35b)
 i 1  (30)
 Fx Fyi 1 
Substituting Eqs. (33)-(34) into Eq. (30) can achieve
 1 0 
 i 1 i 1  d y  y  y  x  x
 i
L0 1  y Li  1  y  Fi 1    / (36a)
  w i 1   wh i 1
1  w 0   wh
i 1  Fx Fx L0 Fx L.
 Fx Fx Fy Fyi 1 

If neglecting the effect of the hanger elongation on the d y  y  y  x  x

   / (36b)
self-weight of the hanger, Eq. (9) can be simplified as Fy Fy L0 Fy L.
Ghi  whh0i  whhi (31) The matrix C2×2n can be directly calculated from Eq.
and thus the coefficient γ in Eq. (29) equals 1. (36) and the matrix F2n ×2 can be calculated from Eq. (30)
With the recursion relation expressed in Eq. (30), Eq. and Eqs. (35)-(36).
(26) can be simplified as
3.2 Sensitivity analysis
U  U Fm
1
 
  J 22 Fm

1
 (32)
Fig. 5 shows a simplified single-span suspension bridge
 Fm
1  with a span of 120 m and 11 hangers, the sag-span ratio for
  the main cable is 1/10, and the area, density and elastic
where, J22  C22nF2n2   F2n2Fm .
1
and  modulus of the cable are 0.08 m2, 8.005×103 kg/m3 and
 2n 
 Fm  2.00×1011 Pa, respectively. To validate the explicit flexible
According to Eq. (4), the following differential equation matrix derived in section 3.1, three loading cases are
can be obtained considered for calculating the Jacobian matrix.
Case 1: the cable only acts by its self-weight with
 y  y  y neglecting the effect of the hangers and deck girder;
d y  dFx  dFy  dL0 (33)
Fx Fy L0 Case 2: the cable acts by its self-weight with
concentrated load 5000 kN at each hanger shown in Fig. 5
where while ignoring the effect of the hangers and stiffening
 x  y Fx  1 1  girder;
     Case 3: Only considering the load at each lower anchor
Fy Fx w  T1 T2 
of the hangers as 5000 kN with 0.01 m2 for each hanger
 y L0 1  Fy Fy  wL0  area and the same material as the main cable.
     The obtained results are also compared with those
Fy EA w  T1 T2 
calculated by FDM with different perturbation. Eq. (18)
 y Fy wL0 Fy  wL0 determines the initial load for calculating the flexible
  
L0 EA EA T2 matrix. Fm1 =[-966.51, 386.61]T kN for case 1, and Fm1 =[-
73154.01, 29261.61]T kN for case 2 and case 3.
T1  Fx2  Fy2 Table 1 lists the sensitivities computed by FDM with
different perturbations and by the proposed explicit method.
 
2
T2  Fx2  Fy  wL0 The perturbation Δxi is defined as a percentage between the
variables in the FDM during the computations due to the
Differentiating Eq. (3) respecting to Fx and L0, value of the two variables are not at same magnitude. From
respectively, lead to Table 1, the sensitivities calculated by FDM vary with the
perturbations and these sensitivities are not converging to
 x L 1 T2  Fy  wL0
  0  ln those calculated by the proposed analytical method as
Fx EA w T1  Fy perturbation Δxi decreases. The optimal perturbation for this
Fx2  1 1 
   
w  T1 T1  Fy  T2 T2  Fy  wL0  
 
 Form-finding analysis of suspension bridges using an explicit Iterative approach 91

Table 1 Sensitivities computed with the FDM and the l=120m

proposed explicit method (Unit: m/N)

f=12m
Cases Jij J11 J12 J21 J22

3m
Δxi =1% -1.13×10-5 -5.91×10-5 2.29×10-6 -1.18×10-4
Δxi Fig. 5 A simplified cable structures
=1×10-4 -1.14×10-5 -5.91×10-5 2.35×10-6 -1.18×10-4
Finite %
Δx
Table 2 Material and cross-sectional properties of the bridge
difference i
Case 1 method =1×10-10 -1.14×10-5 -5.91×10-5 2.32×10-6 -1.18×10-4 Structural member E (Gpa) A (m2) I (m4) w (kN/m)
%
Δxi Main span cable 210 0.4 - 32.9
=1×10-13 -2.11×10-5 -9.42×10-5 -2.02×10-5 -9.65×10-4 Side span cable 210 0.41 - 33.8
%
Hanger 210 0.025 - -
Present work -1.14×10-5 -5.91×10-5 2.35×10-6 -1.18×10-4
Deck 210 0.5 1.66 72.4
Δxi =1% -1.84×10-7 -8.20×10-7 -3.00×10-8 -1.64×10-6
Δxi
=1×10-4 -1.86×10-7 -8.20×10-7 -3.03×10-8 -1.64×10-6
% 4. Numerical examples
Finite
difference Δxi -10
Case 2 method =1×10 -1.30×10-7 -8.96×10-7 -6.49×10-8 -1.55×10-6 Great Belt suspension bridge (Kim and Lee 2001) and
% Yingwuzhou Yangtze River Bridge et al. 2013) are
Δxi investigated in order to show the efficiency of the improved
=1×10-13 -3.90×10-8 -8.10×10-5 7.30×10-8 -8.12×10-5
analytical form-finding method in determining the initial
%
shape of suspension bridges.
Present work -1.86×10-7 -8.20×10-7 -3.03×10-8 -1.64×10-6
Δxi =1% 1.84×10-7 8.20×10-7 3.04×10-8 1.64×10-6 4.1 The Great Belt suspension bridge
Δxi
=1×10-4 1.86×10-7 8.20×10-7 3.07×10-8 1.64×10-6 For Great Belt suspension bridge, form-finding analysis
Finite %
has been performed by (Karoumi 1999), Kim and Lee
difference Δxi -10 (2001), Kim and Kim (2012). Fig. 6 shows the TCUD
Case 3 method =1×10 1.95×10-7 6.53×10-7 3.24×10-8 1.63×10-6
% parameters and node number of the simplified Great Belt
Δxi suspension bridge where the superstructure is supported by
=1×10-13 -3.25×10-5 8.11×10-5 -3.25×10-8 1.62×10-4 a roller on the cross beam of the pylons. The main cable is
% connected to the stiffening girder at the center of the bridge.
Present γ 1.85×10-7 8.20×10-7 2.99×10-8 1.64×10-6 In Fig. 6, the nodal points 1, 9, and 21 represent the position
work γ=1 1.85×10-7 8.20×10-7 2.99×10-8 1.64×10-6 of the left spray saddle, tower saddle, and sag point at the
center span with the predetermined y-coordinate of 0.00 m,
180.00 m and 0.001 m, respectively. Besides these, prior to
numerical example is Δxi=1×10-4% where results calculated a form-finding analysis, nodal coordinates for several
by the proposed method agrees well with almost all the structural points are also pre-determined for a target
sensitivities calculated by FDM under the three loading configuration as follows: 1) The hangers are vertically
cases. By comparing with the analytical values, truncation arranged and the camber of the deck girder was not
errors still exist for 1% perturbation and the round-off considered; 2) The x-coordinates of nodal points at which
errors are clearly visible when Δxi smaller than 1×10-4%. the main cable and hangers interconnect are known; 3) The
And when Δxi=1×10-13%, the computed gradient matrix is y-coordinates of nodal points at which the deck girder and
incorrect. The effectiveness of the FDM depends on the hangers interconnect are 0.00 m. The other nodal
perturbation Δxi. For instance, the optimal perturbation is coordinates are unknown parameters, for example, y-
Δxi=1×10-4% in this numerical example, however, it will coordinates of main cable for the node no. 2-8 and 10-20,
change and it is difficult to select for other functions and which shall be determined by a form-finding analysis.
calculation point. On the other hand, the proposed method Table 2 summarizes the material and cross-sectional
can compute the analytical values directly, which gives a properties of the bridge. As this bridge is an earth-anchored
better choice compared with FDM. bridge, the axial forces and longitudinal displacement of the
In addition, the gradient matrix calculated with γ=1 is deck girder are not expected and the analytical method
the same as that calculated with the actual γ determined by based form-finding analysis considers cable-only system,
Eq. (30) from Case 3, this means that the self-weight of the therefore the stiffness of the tower does not affect the form-
hangers can be approximated by their deformed lengths finding analysis result and the parameters about the tower
during the gradient matrix derivation as for almost all were not listed.
suspension bridges the elongation of the hangers is small Table 3 summarizes the differences between the
and the hangers should be in elastic state during operating converged nodal coordinates of the main cable calculated
period. The gradient matrix calculation procedure thus can by the proposed method, TCUD and I.TCUD methods. In
be greatly simplified with γ equals 1. Table 3, the third and the fourth columns were obtained
92 Hongyou Cao, Yun-Lai Zhou, Zhijun Chen and Magd Abdel Wahab

535m 1624m 535m

180.00 9 180.00
10
8

3 4
0.00 2 20 0.001 0.00
1 21

Fig. 6 Geometric parameters of the simplified suspension bridge

Table 3 Calculated nodal coordinates of the main cable

Node y (m) (3)-(1) (3)-(2)
x (m) Remarks
No. (1) By TCUD (2) By I.TCUD (3) Present work (m) (m)
1 0.000 0.000 0.000 0.000 0.000 0.000 main cable anchor
2 66.875 13.637 13.640 13.640 0.003 0.000
3 133.750 29.967 29.973 29.974 0.007 0.001
4 200.625 48.714 48.724 48.725 0.011 0.001
5 267.500 69.961 69.975 69.976 0.015 0.001
6 334.375 93.696 93.715 93.716 0.020 0.001
7 401.250 119.935 119.959 119.961 0.026 0.002
8 468.125 148.687 148.716 148.719 0.032 0.003
9 535.000 179.964 180.000 180.000 0.036 0.000 tower saddle
10 602.667 151.101 151.132 151.129 0.028 -0.003
11 670.333 124.787 124.812 124.811 0.024 -0.001
12 738.000 101.011 101.032 101.031 0.020 -0.001
13 805.667 79.764 79.781 79.780 0.016 -0.001
14 873.333 61.038 61.050 61.050 0.012 0.000
15 941.000 44.824 44.833 44.833 0.009 0.000
16 1008.667 31.116 31.122 31.123 0.007 0.001
17 1076.333 19.908 19.912 19.913 0.005 0.001
18 1144.000 11.196 11.198 11.198 0.002 0.000
19 1211.667 4.976 4.977 4.977 0.001 0.000
20 1279.333 1.244 1.245 1.245 0.001 0.000
21 1347.000 0.001 0.001 0.001 0.000 0.000 sag point

Table 4 Horizontal tension of the main cable (×106N)

Parameter TCUD I.TCUD Present work calculated by TCUD illustrate that the vertical deformation
of the tower leading to the actual sag of the main cable is
Horizontal tension of the main cable 194.00 193.75 193.69
3.6 cm smaller than that of the other two methods. I.TCUD
method can eliminate the deformation of the tower during
the form-finding analysis by introducing initial force into
from Kim and Kim (2012). The results from these three the tower. The results calculated by the proposed method
methods agree well with each other. However, the TCUD and I.TCUD agree well with each another. The maximum
and I.TCUD methods require proper initial configuration difference occurs at the two nodal points near the tower
estimation for the main cable ahead to ensure the saddle with 3 mm. This difference may be resulted from
convergence of the implementation and lead to challenge in that the proposed method is derived from the complete form
engineering application. The convergence of the proposed of elastic catenary equations while the cable element in
method is independent of the initial configuration of the I.TCUD is derived from the linearized elastic catenary
main cable. Therefore, the proposed analytical method has equations.
obvious advantages. Since the proposed method divides the Table 4 shows that the three methods provide consistent
bridge into two substructures and the form-finding analysis results for the horizontal tension of the main cable. The
only refers to the cable system, the shortening of the main maximum horizontal tension of the main cable occurs in the
tower induced by the cable force does not affect the vertical TCUD method. As discussed above, TCUD cannot
coordinates of the main cable. However, the results eliminate the vertical deformation of the tower, which
 Form-finding analysis of suspension bridges using an explicit Iterative approach 93

Fig. 7 Yingwuzhou Yangtze River Bridge

40+11@15+20=225 20+54@15+20=850 20+54@15+20=850 20+11@15+40=225

1 12 13 67 68 122 123 134

Fig. 8 Elevation view of Yingwuzhou Yangtze River Bridge (units in meters)

Fig. 9 The FE model of Yingwuzhou Yangtze River Bridge

implies the final mid-span sag will be smaller than the

predetermined value as the tower deforms in vertical
direction and leads to the increase in the horizontal force of
the main cable.

4.2 Yingwuzhou Yangtze River Bridge

Yingwuzhou Yangtze River Bridge is employed to

verify the applicability of the proposed method in large-
scale real engineering application. As shown in Fig. 7,
Yingwuzhou Yangtze River Bridge is the 8th bridge across
the Yangtze River located in Wuhan with a span of Fig. 10 Calculated displacement of the stiffening girder and
(225+850+850+255) m. The total number of the hangers on main cable
each cable plane is 132, with 55 hangers in the main span
and 11 hangers in each side span. Figure 8 shows the layout
and number of the hangers. The cross sectional area, the Fig. 9 shows the FE model of the Yingwuzhou Yangtze
cross sectional moment of transverse and vertical inertia of River Bridge developed by Midas/Civil. Midas/Civil
the stiffening girder are 1.563 m2, 207.598 m4, 1.105 m4, supplies elastic catenary cable element recognized as the
respectively. The cross sectional area and elastic modulus most accurate element type to simulate the nonlinear
of the main cable are 0.296 m2, 2.05×1011 Pa, and the sag- properties of the cable elements. However, unlike beam or
span ratio of the main cable in main span is 1/9. The truss element, users need to define the initial tension or
coordinates of the saddles on the three towers are (0, 0, unstrained length for each cable element to calculate its
162.5), (±850, 0, 144.5) and that of the two anchors are stiffness matrix, which means the accuracy of the FE model
(±1075, 0, 39.0). The dead load is 342.82 kN/m including highly depends on these initial tensions or unstrained
the secondary dead load of 59.02 kN/m. lengths. As shown in Fig. 9, the FE model of Yingwuzhou
Generally speaking, an ideal suspension bridge FE Yangtze River Bridge contains 1619 nodes and 1601
model should satisfy the following three requirements: (1) elements, including 548 elastic catenary cable elements in
the bridge’s displacement under dead load equals to zero; total. The stiffness girder and tower are modeled by beam
(2) the calculated tension of the hangers should equal to that element while the main cable and the hangers utilize the
of the predetermined tension; (3) the calculated bending catenary element. The nodal coordinate of the main cable
moment of the stiffness girder equals to that of calculated and unstrained length of all catenary elements were
by the multi-support continuous beam model. In this calculated by the proposed form-finding approach. To
analysis, these three points are considered as criterions to eliminate the vertical deformation of the towers, proper
investigate the effectiveness of the proposed method. initial forces equal to the internal forces induced by the
94 Hongyou Cao, Yun-Lai Zhou, Zhijun Chen and Magd Abdel Wahab

Fig. 12 Bending moment diagrams of the multi-support continuous beam and suspension bridge (kN.m)

Fig. 11 Comparison of the calculated hanger tensions and the predetermined hanger tensions

main cable’s compression to the towers have been obtained from elastic catenary equations with no
introduced to the corresponding elements. simplification in the deviation process. Meanwhile, the
Fig. 10 demonstrates the displacement curves of the elastic catenary cable element supplied by Midas/Civil is
stiffening girder and main cable under dead load calculated derived from the linearized elastic catenary equations with
from the FE model developed utilizing the results obtained neglecting the high-order terms of the equations that shall
by the form-finding analysis. The maximum displacement lead to errors between analytical method and FE method.
of the stiffening girder (smaller than 3 mm) occurs at the Besides, the unavoidable error of the nonlinear solver of the
mid-span of each main span, and the maximum FE software from round-off errors will also amplify these
displacement of the main cable does not exceed 1 mm, errors between the two different methods. In summary, the
which is much smaller compared with that of the stiffening errors in the FE model developed by the proposed method
girder. Fig. 11 shows the predetermined tensions of the odd are very small and acceptable and the form-finding method
numbered hangers from 1 to 67 calculated by the multi- proposed in this paper can provide an accurate form-finding
support continuous model and that calculated from the FE analysis result for suspension bridges.
model under dead load. The vertical label on the left side in
Fig. 11 denotes the tension of the hanger and that on the
right side shows the ratio of the calculated tension to 5. Conclusions
predetermined tension. The line shows the variation of the
tension ratio varies with different hangers. It shows that the This study addressed an explicit analytical iterative
maximum error of calculated hanger tensions is smaller method for form-finding analysis in suspension bridge
than 0.5% of the predetermined tensions. Fig. 12 compares based on the gradient matrix derived from the differential
the bending moment of the stiffening girder calculated by form of the elastic catenary equations, where a sensitivity
the FE model and those yielded by the multi-support analysis is conducted for verification. Afterwards, two
continuous beam model and the results agree well with each suspension bridges are investigated numerically to illustrate
other, and the maximum error between them are less than the accuracy and efficiency of the proposed form-finding
1.5%. analysis method. To conclude, following remarks can be
To summarize, the proposed method performs well in obtained:
real bridge model and agrees well with the three criterions. (1) An improved form-finding method is developed for
The proposed method is an analytical method, where all the accurate unstrained lengths calculation for the hangers
cable members’ geometric parameters including nodal based on the conventional analytical form-finding
coordinates and unstrained lengths, were rigorously method. The Jacobian matrix for gradient-based
 Form-finding analysis of suspension bridges using an explicit Iterative approach 95

iteration algorithms is derived in an explicit form to Faroughi, S., Kamran, M.A. and Lee, J. (2014), “A Genetic
overcome the inherent defects of FDM. algorithm approach for 2-D tensegrity form finding”, Adv.
(2) The sensitivity analysis demonstrates that the Struct. Eng., 17(11), 1669-1679.
gradient matrices calculated by the proposed method Gimsing, N.J. and Georgakis, C.T. (2011), Cable Supported
Bridges: Concept and Design, John Wiley & Sons.
and by FDM with proper perturbations are consistent. Irvine, H.M. (1981), Cable Structures, The MIT Press, Cambridge.
The proposed method directly derives the accurate Jung, M.R., Min, D.J. and Kim, M.Y. (2013), “Nonlinear analysis
Jacobian matrix while FDM suffers both truncations methods based on the unstrained element length for determining
errors and round-off errors. initial shaping of suspension bridges under dead loads”,
(3) In the first example, the cable coordinates and Comput. Struct., 128, 272-285.
horizontal tensions comparison between those obtained Jung, M.R., Min, D.J. and Kim, M.Y. (2015), “Simplified
by the proposed analytical method and those derived by analytical method for optimized initial shape analysis of self-
two NFEM-based form-finding methods confirms the anchored suspension bridges and its verification”, Math. Prob.
accuracy and efficiency of the proposed method. The Eng., 2015, Article ID 923508, 14.
Karoumi, R. (1999), “Some modeling aspects in the nonlinear
proposed method avoids the unfavorable influence finite element analysis of cable supported bridges”, Comput.
induced by the shortening of towers compared with Struct., 71(4), 397-412.
TCUD while the final results evaluated by it agrees well Kim, H.K. and Kim, M.Y. (2012), “Efficient combination of a
with those derived by I.TCUD. TCUD method and an initial force method for determining
(4) The proposed methodology shows capacity in the initial shapes of cable-supported bridges”, Int. J. Steel Struct.,
application of large-scale practical engineering 12(2), 157-174.
structures from Yingwuzhou Yangtze River Bridge Kim, H.K., Lee, M.J. and Chang, S.P. (2002), “Non-linear shape-
nonlinear FE analysis. In comparison with the finding analysis of a self-anchored suspension bridge”, Eng.
fundamental criterions for suspension bridge, the Struct., 24(12), 1547-1559.
Kim, K.S. and Lee, H.S. (2001), “Analysis of target configurations
maximum displacement error is smaller than 3 mm and under dead loads for cable-supported bridges”, Comput. Struct.,
the maximum relative error of hanger tensions and 79(29), 2681-2692.
bending moment for the stiffening girder do not exceed Kim, M.Y., Kim, D.Y., Jung, M.R. and Attard, M.M. (2014),
0.5% and 1.5%, respectively, which implies that the “Improved methods for determining the 3 dimensional initial
proposed method satisfies the requirements in shapes of cable-supported bridges”, Int. J. Steel Struct., 14(1),
engineering practice. 83-102.
(5) The proposed method might be easily extended to Koohestani, K, and Guest, S.D. (2013), “A new approach to the
spatial-shape cable from plane-shape cable and self- analytical and numerical form-finding of tensegrity structures”,
anchor suspension bridges that gives a promising future Int. J. Solid. Struct., 50(19), 2995-3007.
Lonetti, P, and Pascuzzo, A. (2014), “Optimum design analysis of
in real engineering application. hybrid cable-stayed suspension bridges”, Adv. Eng. Softw., 73,
53-66.
Lund, E. (1994), “Finite element based design sensitivity analysis
Acknowledgements and optimization”, Institute of Mechanical Engineering, Aalborg
University, Denmark.
This research was supported by the Fundamental Luo, X.H. (2004), “Numerical analysis method for cable system of
Research Funds for the Central Universities (WUT: suspension bridges”, J. Tongji Univ., 4, 5.
2015IVA015) and the National Natural Science Foundation O’Brien, T. (1967), “General solution of suspended cable
of China (Grant No. 51408249). The authors would like to problems”, J. Struct. Div., 93(1), 1-26.
O’Brien, W.T. and Francis, A.J. (1964), “Cable movements under
express their appreciation to A/Prof. Xudong Qian from two-dimensional loads”, ASCE J. Struct. Div., 90, 89-123.
National university of Singapore for discussions. Sun, Y., Zhu, H.P. and Xu, D. (2014), “New method for shape
finding of self-anchored suspension bridges with three-
dimensionally curved cables”, J. Bridge Eng., 20(2), 04014063.
Reference Sun, Y., Zhu, H.P. and Xu, D. (2016), “A specific rod model based
efficient analysis and design of hanger installation for self-
Cao, H., Chen, Z., Wu, Q., Zhu, H.P. and Kang, J. (2016), “A anchored suspension bridges with 3D curved cables”, Eng.
simplified model for multi-span suspension bridges based on Struct., 110, 184-208.
single cable theory”, China J. Highw. Tran., 29(4), 77-84. Wang, P.H. and Yang, C.G. (1996), “Parametric studies on cable-
Chen, Z., Cao, H., Ye, K., Zhu, H. and Li, S. (2013), “Improved stayed bridges”, Comput. Struct., 60(2), 243-260.
particle swarm optimization-based form-finding method for Wang, S., Zhou, Z., Gao, Y. and Huang, Y. (2015), “Analytical
suspension bridge installation analysis”, J. Comput. Civil Eng., calculation method for the preliminary analysis of self-anchored
29(3), 04014047. suspension bridges”, Math. Prob. Eng., 2015, Article ID
Chen, Z., Cao, H. and Zhu, H. (2015), “An iterative calculation 918649, 12.
method for suspension bridge’s cable system based on exact Wriggers, P. (2008), Nonlinear Finite Element Methods, Springer
catenary theory”, Baltic J. Road Bridge Eng., 8(3), 196-204. Science & Business Media.
Chen, Z., Cao, H., Zhu, H., Hu, J. and Li, S. (2014), “A simplified Zhang, L.Y., Li, Y., Cao, Y.P. and Feng, X.Q. (2014), “Stiffness
structural mechanics model for cable-truss footbridges and its matrix based form-finding method of tensegrity structures”,
implications for preliminary design”, Eng. Struct., 68, 121-133. Eng. Struct., 58, 36-48.
Fan, L., Pan, Y. and Du, G. (1999), “Study on the fine method of
calculating the erection-parameters of long-span suspension
bridges”, Chin Civil Eng. J., 32, 20-25. CC