Engineering Structures 31 (2009) 2162–2170

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

An approach based on the catenary equation to deal with static analysis of three
dimensional cable structures
Miguel Such a,∗ , Jesus R. Jimenez-Octavio a , Alberto Carnicero b , Oscar Lopez-Garcia c
a
Analysis and Design Department, Institute for Research in Technology, Universidad Pontificia Comillas de Madrid, Alberto Aguilera, 23, 28015 Madrid, Spain
b
Escuela Superior de Ingeniería-ICAI, Universidad Pontificia Comillas de Madrid, Alberto Aguilera, 23, 28015 Madrid, Spain
c
Instituto Ignacio Da Riva, Universidad Politécnica de Madrid, Plaza Cardenal Cisneros 3, 28040, Madrid, Spain

article info abstract

Article history: In this paper a novel method to solve three dimensional cable structures based on the catenary equation
Received 29 May 2008 is proposed. The method is a generalization of a previous engineering application to compute the initial
Received in revised form equilibrium of railway overheads. The major contributions of this paper are: the extension of the previous
26 February 2009
engineering application to simulate arbitrary three dimensional cable structures; cable elasticity is
Accepted 17 March 2009
Available online 9 April 2009
incorporated into the formulation; and due to the fact that the method relies on the analytical catenary
equations, high numerical efficiency is exhibited. In order to show the validity of the method, comparisons
Keywords:
with several well reported cable structure problems are presented. The agreement between the proposed
Cable structures method and published results is excellent.
Catenary © 2009 Elsevier Ltd. All rights reserved.
Non-linear analysis

1. Introduction
Due to their efficiency and aesthetics, cable structures became
quite popular from the 1950s onwards. From the mid 60s to the
end of the 70s a significant number of articles dealing with cable
structures were published, see for instance [1–7] among others.
Nowadays, cable structures are widely used in many applications
as, for example, power transmission lines, railway overheads, cable
transportation systems, cable roof structures etc.
Cable structures pose well known challenging problems, and
the modelling of such structures has always been a subject of
research and innovation. Cable members are light, very flexible and
do not experience bending and compression stiffness. Therefore,
cable structures exhibit a high non-linear behaviour. Another
important problem of cable structures is the determination of the
initial equilibrium configuration. That is, the computation of the
stressed reference configuration which is an inverse structural
problem. Reference [8], is one of the pioneering works dealing
with the classification of the methods to solve initial equilibrium
problems. Local behaviour of particular types of cables is another
quite difficult problem that modelling of cable structures should
∗
Corresponding author. Tel.: +34 91 542 28 00; fax: +34 91 542 31 76.
E-mail addresses: Miguel.Such@iit.upcomillas.es (M. Such),
Jesus.Jimenez@iit.upcomillas.es (J.R. Jimenez-Octavio),
carnicero@dim.icai.upcomillas.es (A. Carnicero), oscar.lopez.garcia@upm.es
(O. Lopez-Garcia).
0141-0296/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2009.03.018
deal with. Helically wound cables present interwire friction which
influences axial stiffness [9]. Cables can show hockling or kinking
phenomena as a result of torsional stability of single and double
rope systems, [10]. For instance, the validity domain assessment
of the mechanical behaviour of simple straight strands which
are layers of helical wires wound around a central straight
wire core has appeared in [11]. This paper focuses on macro-
scale modelling of complex cable structures, that is, the initial
equilibrium configuration computation and the cable structure
response to external load equilibrium under general loading.
Therefore, the modelling of the local behaviour of wire cables is
beyond the scope of the paper.
Broadly speaking, the methods used to model cable structures
can be classified into two main groups. Following the nomencla-
ture proposed in [8] these approaches are called: the non-linear
displacement method and the force density method.
The method of non-linear displacement is based on an iterative
process that modifies step by step the geometry from one
configuration to another fulfilling the equilibrium equations.
Argyris’s pioneering work, [12], applies this method to the
design of the cable roof of the Olympic Stadium of Munich,
replacing real cables with truss elements. The dynamic relaxation
method with kinetic damping is used in [13] to determine the
initial equilibrium configuration and analyse prestressed nets and
membranes. Based on the method of non-linear displacements,
some authors, [14] and [15], modelled the cable as a series of
straight linear trusses developing specific formulations to improve
the method performance. Trying to improve these formulations,
 M. Such et al. / Engineering Structures 31 (2009) 2162–2170 2163

Karoumi [16] developed a curved finite element, Jayaraman and 2.1. Global formulation of a single three dimensional cable
Knudson [17] formulated a two node finite element based on the
exact equation of the elastic catenary and, more recently, Andreu As is well-known the catenary is the equilibrium shape of a
et al. in [18] implemented a deformable catenary element into cable that hangs between fixed points under its own weight. For
a finite element method. The tangent stiffness of a cable using a comprehensive review of cable mechanics see [28] and for the
the catenary equation is provided by [19]. Finally, it can be said local catenary formulation followed here see [29]. Let us consider
that most of the current non-linear displacement methods are the single three dimensional
based on the finite element method. An alternative approach to the
D E cable shown in Fig. 1. Using the local
reference system Õξ1 ξ2 ξ3 , the catenary equation can be written
finite element method is presented in [20] where an engineering
application based on the catenary equation is used to analyse the as:
initial equilibrium and stiffness computation of railway overheads.
ξ1i
 
The force density method was initially introduced by ξ = c cosh
i
2 (1)
Scheck, [21]. The non-linear nature of the problem is considered c
by means of the projections of the forces at every node and their
where c = t1i /w is the catenary shape parameter, t1i the horizontal
non-linear dependency with the nodal coordinates, that are the un-
component of the cable tension (direction ξ1 ), and w is the weight
knowns of the problem. However, other authors tried to obtain an
per unit of length of the cable. The length of the cable between the
equivalent linear problem taking into account certain assumptions
lowest point of the catenary, l, and a general point, i, is denoted by
about the final solution. Among them, Siev and Eidelmann suggest
sli and is defined as
in [1] imposing perpendicularity condition on the horizontal pro-
jection of the equilibrium configuration of a cable net. This restric- ξ1i
 
tion allows the initial equilibrium to be calculated simply by means
li
s = c sinh . (2)
c
of solving an equivalent system of linear equations. However, this
assumption only made the method applicable to a reduced range Finally, the tension at point i, t i , can be expressed by
of problems. Indeed, this method was revisited and applied in [22]
to solve the initial equilibrium problem of structures composed ξ1i
 
of mixed cables and rigid elements under compression loads. The t i = c w cosh . (3)
c
force density method was enhanced by Haber and Abel in [8].
Instead of straight beam elements they added curved and shell el- The three previous equations, i.e. (1)–(3), summarise the
ements to the force density formulation. The density force method behaviour of a cable under its own weight into a local reference
was also combined with least squares optimization techniques system. Now, the positioning of the single cable should be referred
to generate the cutting patterns of the membranes that compose to as a three dimensional frame. The following notation will be
a shell structure in [23]. This method is used in [24] to design used: subscript letters refer to directions and superscript letters
the shape of deployable membrane reflectors within aerospace refer to a catenary point. Let us consider a general point, i, of the
applications. catenary represented in Fig. 1. The spatial position of an arbitrary
This article deals with the static behavior of three dimensional point i of the catenary, X i , is described by the coordinates with
cable structures. The method proposed in this paper is based on the respect to a global cartesian reference system hOX1 X2 X3 i. This
exact solution of the catenary. The use of the analytic expression position vector can be expressed in the form
of the catenary to solve or formulate complex problems has been
used previously by other authors, among others see [25,26,19, X i = X Õ + R · ξ
i
(4)
18,20,27]. One of the most important differences between the
t
where ξ ≡ ξ , ξ , 0
i i i
aforementioned models and the herein proposed one is that while 1 2 t is the transpose, and the rotation matrix
previous models use either catenary tension or node positions R is defined by
that are unknown, the herein proposed method uses all possible
sin ϕ ab − cos ϕ ab
 
unknowns in a cable structure problem, that is, the catenary 0
variables such as tension, length and the catenary parameter as R=  0 1 0  (5)
well as node positions. In this way, both the initial equilibrium cos ϕ ab 0 sin ϕ ab

configuration and the cable structure response to external loads

with ϕ ab = atan X1b − X1a / X3b − X3a . The rotation matrix R




can be calculated using the same strategy. The method has
been implemented in a general purpose computer code toolbox
D E
transforms the vectors from the local system, Õξ1 ξ2 ξ3 , to the
called CALESCA. Moreover, from this perspective, the proposed
formulation should be regarded as a mix between a non-linear global system, hOX1 X2 X3 i. Considering Eqs. (4) and (1) it is possible
displacement method and a force density method. to express the vertical coordinate of the point i as
The remainder of the paper is organised as follows. Section 2
describes the numerical foundations of the method. Section 3 ξ2i = X2i − X2Õ = c cosh λi (6)
provides the application of the method to several well documented
cable structures. The comparison between the method proposed where λi = ξ1i /c. Applying Eq. (6) at the two end nodes, a and b,
and the published results are also presented. Finally, in Section 4, of the catenary arch, see Fig. 1, the following relationship can be
the main conclusions of this study are summarised. found:

ξ2b − ξ2a = X2b − X2a = c cosh λb − cosh λa

(7)
2. Mathematical formulation
where λa and λb should be written as functions of the unknowns
of the problem, that is λa (X a , X b , c ) = α ab − c asinh β ab and


In what follows the mathematical formulation of the method
proposed in this paper is presented. First, the global formulation of λb (X a , X b , c ) = α ab + c asinh β ab where


the catenary equations into a three dimensional reference system
is shown. Next, the global equilibrium of a cable structure is 1
defined. α ab = dab (8)
2
2164 M. Such et al. / Engineering Structures 31 (2009) 2162–2170

Fig. 1. Local and global coordinate systems.

and where α is the coefficient of thermal expansion and ϑ(s) the

temperature along the cable. It is assumed that the temperature
X2b − X2a
β ab =   (9) distribution along the cable should be known or computed by
dab other means. However, a realistic assumption is to consider both
2 c sinh 2c
q the coefficient of thermal expansion and the temperature constant

2
2 along the cable. Under these circumstances, the increase of the
and dab = X1b − X1a + X3b − X3a .
length due to temperature is
In this paper the cable is described by the catenary parameter, c ,
the catenary arch length, sab , and the cable tension, t a . Considering 1sϑ = αϑ sab
Eq. (2) at the two end nodes, a and b, see Fig. 1, it is possible to
where sab is the cable length given by Eq. (11).
establish a functional dependence between the coordinates of the
Finally, applying Eq. (3) at, for instance, the end node a, it is
end nodes, X a , X b , the catenary parameter, c, and the catenary arch
possible to write a relationship among the coordinates of the end
length, sab , that is:
nodes, X a , X b , the catenary parameter, c, and the tension of the
E1 sab , c , X a , X b = 0 node a, t a , that is an equation of the form


(10)
E2 t a , c , X a , X b = 0


where (15)
ab
− c sinh λ − sinh λ . b a


E1 = s (11) where
The catenary length sab is computed including the cable ta
E2 = − cosh λa .
elasticity as: cw
Eqs. (10) and (15), completely define the behaviour of a single
0 + 1se + 1sϑ
sab = sab (12)
three dimensional cable. These two equations provide a system of
0 is the unstressed cable length, 1se the increase of the
where sab seven degrees of freedom, that is, given seven known parameters
length due to elastic deformation, and 1sϑ the length increase from the set sab , c , X a , X b and t a the cable behaviour is completely
due to temperature. The increase of the length due to elastic determined. However, although these equations are useful when
deformation can be expressed by studying a simple cable, these results are not enough to tackle the
modelling of complex cable structures.
T (s)
Z
1se = ds (13)
sab EA
2.2. Three dimensional cable structure
where T (s) is the tension at each point of the cable which can be
computed according to Eq. (3), and EA is the axial stiffness. After Let us consider a general three dimensional cable structure
some algebra, Eq. (13) can be rewritten as subject to gravity and point loads, see Fig. 2. In such a complex cable
w c hc i net, there are many cables with different known and unknown
1se = sinh 2λb + sinh 2λa + dab .




(14) parameters which determine the individual behaviour of each
2EA 2
cable. However, the cables are linked together due to the fact that
It should be remarked that in some applications the strain on on one hand they should satisfy force equilibrium and on the other
the cables can be neglected. However, in certain problems the hand the cable structure is subjected to some constraints that
variation of length due to elasticity should be considered. should be fulfilled.
The length increase due to the temperature is written as Let a be a node of a cable structure with an arbitrary distribution
Z
of forces defined in the global reference system hOX1 X2 X3 i.
1sϑ = αϑ(s)ds Furthermore, let us consider that ja cables are joined at node a,
sab
 M. Such et al. / Engineering Structures 31 (2009) 2162–2170 2165

Fig. 2. Three dimensional cable structure description.

see Fig. 2. Abusing the notation, the subindex now indicates the to given external loads F a . Eqs. (10) and (15) are applied to every
cable number. The force equilibrium at node a can be expressed by cable providing 2nc equations. The force equilibrium at every node
(17) gives 3nn equations. The number of equations of the set of
ja
constraints is nt . Then, summing up, the following system of non
X
Tja + F a = 0 (16)
linear algebraic equations is obtained:
j =1

with Tja = T1a , T2a , T3a

t
the components of the tension of cable G (η) = 0 (19)
j
a t
j at node a, and F a = F1a , F2a , F3 the external force applied at


where
node a. The tension of cable j, Tja , is obtained by transforming the
j , cj , X , X
= 0,
E1 sab 1 . . . nc
a b


tension of the catenary j at node a, tja from the local reference 
E2 tja , cj , X a , X b = 0, 1 . . . nc


D E
system Õξ1 ξ2 ξ3 to the global system hOX1 X2 X3 i. The components G=
E3 cj , tja , X a , X b , F a = 0, 1 . . . nn




of the local tension vector tja can be written in terms of the variables E4 cj , sab
j , tj , X , X , F
a a b a
= 0, 1 . . . nt


of the problem as t1a = wj cj and t2a = wj cj sinh λÕa . Therefore,
the force equilibrium at the global coordinate system leads to the and η = (η1 , η2 , . . . , ηn ) are the unknowns of the problem and
following algebraic equation: n = 3 nn + 2 nc + nt .
The solution of the non-linear system of algebraic equations
E3 cj , tja , X a , X b , F = 0


(17) (19) has been carried out using a general non-linear equation
where solver.1 As is well know most common non-linear solvers require
∂G
ja the computation of the Jacobian matrix, Jij = ∂ηi . One of the
X j
E3 = Rj · tja + F a advantages of the proposed method is that the Jacobian can be
j =1 computed analytically due to the fact that the catenary equations
and their derivatives are obtained in closed form. The size of the
and Rj being the rotation matrix corresponding to cable j defined
by Eq. (5). equation system is the lowest possible because the formulation
Finally, to correctly describe the cable structure a set of is completely independent of spatial discretisations, as occurs, for
algebraic constraints should be incorporated into the formulation. instance, when the finite element method is used. The size of the
This set of constraints is made up of fixed node positions or system of equations is exclusively determined by the geometry of
even mathematical models of different devices, such as structure the structure: the number of cables and nodes. All these aspects
supports, pulleys with or without friction, springs, counterweight provide a fast, robust and very flexible method from a numerical
masses, tensioning devices etc . . . The set of constraints can be point of view. Depending on the kind of problem, the unknown
described by the following algebraic equation variables can be node location or catenary parameters. The toolbox
CALESCA manages the equation formation supplying the solver
E4 cj , sab
j , tj , X , X , F
a a b a
= 0.


(18) with the non-linear system of algebraic equations in which these
Generally speaking the parameters of the problem can be of two unknown variables are mixed altogether.
types: node positions or cable parameters. Node positions are the

t
equilibrium coordinates of the nodes, X a = X1a , X2a , X3a . Cable
parameters consist of the following variables: catenary parameter, 1 A trust region dogleg algorithm included in the Optimization toolbox 2.1.1 of
c, cable length, s, the tension, t a , at a node a. Let us assume that Matlab 7.0 has been used in all the examples carried out in this paper, performing
the cable structure is made of nc cables and nn nodes subjected robustly and precise in all the verification examples.
2166 M. Such et al. / Engineering Structures 31 (2009) 2162–2170

Fig. 3. Comparison between equilibrium configurations of cable system together with pulley.

3. Verification examples Thus, 15 equations will be needed in order to find the equilibrium
configuration. These 15 unknowns are the following: reactions,
In this section four examples of cable structures are presented R1 and R2 , at supports P1 and P2 , the horizontal position, x3 , of
in order to validate the previously proposed formulation. The cable the pulley P3 , the reactions, R3y and R3z , at the pulley and the
structures have been selected to show not only the versatility of the characteristics of cables 1 and 2, catenary constant, c1 and c2 , cable
herein proposed formulation but also the accuracy of the method. lengths, l1 and l2 and tension of the cable at the pulley, t1 and t2 .
The first example shows explicitly the formulation explained in the The information provided is the horizontal reaction at P3 which
Section 2. A pulley between a planar cable including cable elasticity must be zero as the friction is not considered, the position of the
is presented as a detailed example in order to clarify the theoretical fixed points P1 and P2 and the vertical position of the pulley, P3 at
concepts. The second example shows a three dimensional cable z3 = 100.
structure which models a skyline system. The third one is the As has been discussed in Section 2, in order to solve the value
computation of the stiffness of a railway overhead. The last one for the unknowns, the system (19) must be applied.
is another three dimensional cable structure with a spring and
The explicit form of equation Eq. (10) applied to cables 1 and 2
considers temperature and cable elasticity.
is
w c1 h c1  
P
 
P

P P
i
3.1. Transport pulley system E11 ≡ L1 + sinh 2λ13 + sinh 2λ11 + d13 1
2EA 2
 
Bruno and Leonardi analysed the non-linear static behaviour − c1 sinh λ
P3 P
− sinh λ11 = 0 (20)
1
of cable systems supported by sheaves, see [30], in order to be
applied to ski lifts, electrical transmission lines and cable systems w c2 h c2  
P
 
P

P P
i
E12 ≡ L2 + sinh 2λ23 + sinh 2λ22 + d23 1
in erection procedures of long-span bridges. One of the tests 2EA 2
carried out by these authors has been reproduced and compared 
P P

to the CALESCA results. Specifically, the case study consists of a − c2 sinh λ23 − sinh λ21 = 0. (21)
cable with an initial length of L0 = 500 m long, Young’s modulus
E = 16 GPa, cross section A = 8.05 × 10−4 m2 and linear density Using the same procedure, the explicit form of (15) is
w = 6.327 kg/m. Moreover, the cable is fixed at P1 = (0, 0, 0) and P
P2 = (300, 0, 50), and a pulley is located at z3 = 100 m which is t1 3 P
E21 ≡ − cosh λ13 = 0 (22)
free to move in the horizontal direction, see Fig. 3. As in the original c1 w
paper the elasticity of the cable is taken into account. As mentioned P
above, the treatment of pulleys can be easily incorporated into the t2 3 P
E22 ≡ − cosh λ23 = 0. (23)
formulation as a constraint. In this particular case the pulley has c2 w
been modelled as a boundary node with the constraint that the
The system of Eqs. (17) is the sum of forces at each point.
reaction in the horizontal plane should be ignored. The original
The sum at point 1 is
reference [30] neglects pulley friction, thus, for this particular test
case friction has not been considered. E31x ≡ w c1 − R1x = 0 (24)
The configuration of the problem is composed of 3 nodes, nn ,
E31y ≡ R1y = 0 (25)
2 cables, nc and 2 constrains to model the pulley, nt . Therefore,
the number of unknowns will be n = 3 nn + 2 nc + nt = 15. E31z ≡ w c1 sinh λÕ1 P1 − R1z = 0 (26)
 M. Such et al. / Engineering Structures 31 (2009) 2162–2170 2167

Table 1 is composed of three cables. Each cable hangs between a node at

Differences between equilibrium configurations of cable system together with the top of a tower and a common end in which a load can be hung.
pulley.
The three cables are linked together at the common node and the
Bruno and Leonardi CALESCA ∆ (%) load can be transported inside the triangle whose vertexes are the
x3 (m) 282.81 282.2 0.21569 towers. Fig. 4 shows a sketch of this cable structure.
L1 (m) 446.37 446.92 0.12355 In what follows the comparison between the particular
L2 (m) 52.86 53.078 0.41335 triangular skyline presented in [31] and the results obtained using
T (kg) 1830 1831.4 0.07626
CALESCA are shown. The three cables are fixed to the following
x3 (m) 47.67 47.33 0.71302 points: P1 = (260, 210, 786), P2 = (320, 685, 790) and P3 =
L1 (m) 111.07 110.96 0.10145
L2 (m) 388.18 389.04 0.22224
(15, 680, 771). The linear density of the cables is w = 0.1 N/m
T (kg) 1478 1481.23 0.22052 and the weight of the hanging load is W = 100 N. Due to the
fact that the considered stresses are relatively low, the elastic
deformations are neglected. For further information Ref. [31]
at point 2, provides the original data. Two different cases are considered.
E32x ≡ w c1 − w c2 = 0 (27) The first one, Case I, cable tensions, T1 , T2 and T3 are known and
the unknown parameters are load position, x, y, and z, and the
E32y ≡ R3y = 0 (28)
length of cables, s1 , s2 and s3 . The second one, Case II, the known
E32z ≡ w c1 sinh λÕ1 P3 + w c2 sinh λÕ2P3 − R3z = 0 (29) parameters are the load position at the horizontal plane, x, y, and
the vertical tension, T3 . The results obtained using CALESCA are
and at 3 results shown in Table 2 together with those previously published in [31].
E33x ≡ w c2 − R2 x = 0 (30) The differences are all less than 0.1% which shows not only the
versatility of the tool but also the accuracy of the formulation.
E33y ≡ R2 y = 0 (31)
E33z ≡ w c2 sinh λÕ2 P2 − R2 z = 0. (32) 3.3. Railway overhead stiffness
Last the constraints equations model the pulley behaviour. The
constraints applied must maintain the total unstretched length In railway transport systems, the catenary structure or railway
overhead consists of a complex distribution of cables that provides
of the cable and match the tensions at both sides of the pulley
the electric energy supply to the train. The catenary structure
assuming that there is no friction in it.
presents low bending stiffness and the two major cables, the
E41 ≡ L1 + L2 = L0 (33) contact and the messenger wires are tensioned with quite a high
P P mechanical tension.
E42 ≡ t1 3 − t2 3 = 0. (34) The verification example presented in this subsection shows
Therefore a non-linear system of 14 equations is obtained that can the computation of the static stiffness of a railway overhead.
be used to solve the 14 unknowns. The static stiffness of a railway overhead is usually defined
Fig. 3 shows the equilibrium configuration of the cable system as the ratio between a punctual force applied to the contact
obtained both with CALESCA, continuous lines, and by Bruno and wire and the corresponding displacement. The computation of
Leonardi, dashed lines. Furthermore, Table 1 collects the numerical the stiffness distribution of a railway overhead using the Finite
results where: x3 is the horizontal equilibrium position of the Element Method (FEM) is reported by Wu and Brennan in [32].
pulley; L1 is the length of the cable from its left end, P1 , to the The main characteristics of the particular railway overhead are
pulley; L2 is the length from its right end, P2 , to the pulley; T is the summarised in Table 3. The results of the reference [32] have been
compared with the ones obtained using CALESCA in Fig. 5. Both
cable tension at the pulley and; ∆ the relative difference between
results together with the relative difference are shown in Table 4.
both methods. It can be seen that the differences between both
It can be seen that the maximum difference appears at the position
results are less than 1%. The comparison between both methods
of the first dropper and they are about 6%. The results obtained
is quite a remarkable match.
using the FEM computation and the CALESCA formulation shows
It is quite remarkable that the trust region dogleg included
very good agreement.
in the version of Matlab 7.0 converges in around 25 iterations
depending on the initial values for a tolerance of 10−14 , spending
in all cases less than a quarter of second for an Intel(R) Core(TM)2 3.4. Three dimensional cable system with spring
CPU - T7200 with 2 GHz and 2 GB RAM.
In order to show further the versatility of the formulation a
As it was previously pointed out, the convergence can be
three dimensional cable structure together with a spring is also
slightly affected by the election of the initial values. This fact can
simulated. The chosen cable structure was proposed by Peyrot
become a key aspect in case of bifurcations, i.e. the equilibrium
and Goulois in [7]. This cable system is composed of three cables
of the system can reach more than one configuration of the
joined at a point which hangs from a vertical spring with stiffness
cables. For this example, Bruno and Leonardi discuss in [30] the
k = 1000 N/m. The point at which the cables are joined up
influence of the initial values, that evidences the non-uniqueness
is horizontally loaded with a load P = 1000 N. Fig. 6 shows
of equilibrium configurations for a given cable structure. In this
a sketch of this structure, where cables 1, 2 and 3 are linked at
example each configuration can be reached selecting a different
the equilibrium point A0 , with the initial point denoted by A. The
set of initial values: length and tension of the cables and the initial cables of the structure have a coefficient of thermal expansion of
location of the moving constraint. α = 6.5 10−6 C−1 , axial stiffness of EA = 2.9 105 N and they are
subjected to a temperature increment of ϑ = 100 C. For further
3.2. Triangular skyline system information, see reference [7].
Table 5 shows the displacements u0 , v0 and w0 from the initial
The cable structure presented in this subsection is a triangular point A to the equilibrium point A0 together with the relative
skyline system. A triangular skyline system allows the transport of difference ∆ between the results. Once again, it can be said
any kind of load through an operational zone without interfering that there is good agreement between the results. It can also be
or damaging anything in its way. This particular transport system remarked that the differences are less than 0.5%.
2168 M. Such et al. / Engineering Structures 31 (2009) 2162–2170

Fig. 4. Configuration of a triangular skyline system.

Table 2
Comparison of results: Case I and Case II.
Case I Case II
Kanzaki CALESCA ∆ (%) Kanzaki CALESCA ∆ (%)
T1 3200 3200 – 3200 3199.9 0.003
Stress (N) T2 2994 2994 – 2994 2994.1 0.002
T3 4069 4069 – 4069 4069 –
x 145.5 145.5 0 145.5 145.5 –
Location (m) y 610.3 610.29 0.013 610.3 610.3 –
z 751.6 751.58 0.003 751.6 751.58 0.003
s1 418 418.06 0.06 418 418.06 0.060
Length (m) s2 193.7 193.7 0.0048 193.7 193.7 0.005
s3 149.3 149.23 0.069 149.3 149.23 0.069

Table 3
Mechanical and geometrical parameters of the railway overhead used in [32].
Mechanical data Messenger wire Contact wire Droppers

Mass per unit length (kg/m) 0.697 0.988 0.1

Tension (kN) 9.8 9.8 0
Geometrical data
Length between masts (m) 1.4
Mast to first dropper length (m) 2.5
Distance between droppers (m) 5
Number of droppers 10

railway overheads [20]. In order to show the validity of the method

Fig. 5. Comparison between computations of the railway overhead stiffness
distribution.
4. Concluding remarks
In this paper a novel method to deal with three dimensional
cable structures has been proposed. The method is based on the
analytical catenary equations and is a generalisation of a previous
engineering application to compute the initial equilibrium of
comparisons with several well-known cable structure problems
have been presented. The agreement between results is very good
and the differences between them are always less than seven per
cent in the worst case and on average the differences are about
one per cent. The major improvements of the present method
compared to the Ref. [20] can be conveniently summarised as
follows:
• The previous method considered only two dimensional vertical
cable structures. Now taking advantage of the fact that cable
catenaries are located on a plane, a full three dimensional
treatment of cable structures has been incorporated into the
formulation.
• Cable elasticity is taken into account. Although cable elasticity
can usually be neglected, some applications involve important
elastic deformation of the cable structure.
 M. Such et al. / Engineering Structures 31 (2009) 2162–2170 2169

Table 4 • Due to the fact that the method is based on the exact catenary
Differences between computations of the railway overhead stiffness distribution. equations, it inherits most of the benefits of the previous
Distance (m) Stiffness (N/m) ∆ (%) one, [20]. From a numerical point of view, the method exhibits
Wu and Brennan CALESCA high numerical efficiency, i.e. does not depend on spatial
0 4094.26 3960.50 3.37748
discretisation and the problem size is determined by the
1.13636 4299.18 4160.64 3.32975 topology of the cable structure.
2.43506 5672.13 5379.85 5.43284 • Moreover, most of the solvers of systems of non-linear algebraic
3.24675 4299.18 4028.27 6.72522 equations require the computation of the Jacobian. Despite the
3.8961 3459.02 3342.47 3.4869 fact that this computation is quite lengthy and cumbersome the
5.19481 2905.74 2726.63 6.56891
method implements the analytical expressions of the Jacobian
6.33117 2659.84 2608.29 1.97621
7.46753 2659.84 2615.20 1.70676 providing for its fast and accurate evaluation.
9.09091 2106.56 2074.07 1.56659 • The initial equilibrium problem and analysis of static equilib-
10.7143 1881.15 1831.49 2.71125 rium under the loading of cable structures are treated using
12.3377 1881.15 1881.88 0.0387361 the same algorithm strategy. Due to the management of known
15.4221 1532.79 1522.65 0.66552
and unknown parameters both kinds of problems can be easily
17.5325 1614.75 1585.57 1.8405
20.4545 1409.84 1374.33 2.58364
solved.
22.5649 1491.80 1470.02 1.48188 The most important limitation of the method is that when
25 1368.85 1335.49 2.49843
it is used for extremely high tensioned cable structures, the
resulting non-linear system of algebraic equations can become ill-
Table 5 conditioned because the catenary parameter can attain too high
Displacement to the equilibrium point. a value. To overcome this deficiency the solution is twofold. On
Peyrot and Goulois CALESCA ∆ (%) the one hand, the solver of the non-linear system of algebraic
u0 26.47 26.527 0.2153
equations needs to be robust. On the other hand, instead of using
v0 41.13 41.105 0.0608 catenary equations to model cables, a simple straight bar would be
w0 −2.87 −2.8833 0.4613 helpful. From the authors’ experience, this problem has appeared
with very extremely high tensioned cable structures which cannot
be considered representative of usual cable structures. Another
minor limitation of the method presented herein is that the
cable’s own weight, uniform and punctual loads are the only loads
considered. Thus arbitrary non-uniform distributed loads along the
cable length are not considered. However, the practical situation
treated in this paper is perfectly valid for the most common
engineering applications, such as railway overheads, cable-based
energy transport systems, metropolitan subways, cableways, etc.

Acknowledgements

The authors would like to thank an anonymous reviewer of

Journal of Engineering Structures for encouraging us to provide
a more general framework of the original work on which this
article is based. This work has been partially funded by the Spanish
Ministerio de Fomento under the project ‘‘Cálculo del desgaste
en el sistema de captación de energía de trenes de alta velocidad.
TRAN2004-06889/TREN’’ (Wear evaluation in high speed electric
overhead systems). This support is gratefully acknowledged.

Appendix

In order to explicit some of the equations defined in Section 2,

Fig. 6. Initial and final structure configurations of Huang and Lan cable structure.
• Constraints on the variables of cable structures are taken into
account in a natural way. Moreover, the constraints can be used
to include in the cable structure models of mechanical devices
that are commonly used, such as pulleys, cable tensioning
systems, cable supports, springs, etc.
Regarding other well-known methods to simulate cable struc-
tures, the advantages of the herein proposed method can be
described as follows:
herein some of the most common special members and constraints
are presented. These are just a few cases which enhanced the
generality of the method providing trusses and pulleys. However,
other special constraints could be added to reproduce any
structural problem.
The equations to model a pulley, as it is applied in Example 3.1,
must apply two additional constraints. The cable which passes
through the pulley is divided in two cables. The sum of lengths
of both cables must be equal to the length of the original cable.
Moreover, assuming no friction in the pulley, the tension of the
cable at both sides of the loop must be the same. Therefore, the
constraint equations, E4 , will be
L1 + L2 = L0 (A.1)
t1 = t2 . (A.2)
If trusses were to be added inside a cable structure, it will
be necessary to include point forces equations, E3 , and additional
constraints, E4 .
2170 M. Such et al. / Engineering Structures 31 (2009) 2162–2170

Since the tension must lie on the axis of the truss and the self [6] Ozdemir H. A finite element approach for cable problems. Int J Solids Struct
weight is distributed on the end nodes, the forces which will be 1979;15:427–37.
[7] Peyrot A, Goulois A. Analysis of cable structures. Comput & Structures 1979;
included in the system of equations E3 must be 10(5):805–13.
[8] Haber R, Abel J. Initial equilibrium solution methods for cable reinforced
x2 − x1
Tx1 = t q (A.3) membranes. Part I—Formulations. Comput Methods Appl Mech Eng 1982;
30(3):263–84.
x221 + y221 + z21
2
[9] Raoof M, Davies TJ. Influence of variations in the axial stiffness of steel cables
on vertical deflection of cable trusses. J Constr Steel Res 2004;60(3–5):411–20.
y2 − y1 [10] Ermolaeva N, Regelink J, Krutzen M. Hockling behaviour of single- and
Ty1 = t q (A.4)
multiple-rope systems. Eng Failure Anal 2008;15:142–53.
x221 + y221 + z21
2
[11] Ghoreishi S, Messager T, Cartraud P, Davies P. Validity and limitations of linear
analytical models for steel wire strands under axial loading using a 3D FE
z2 − z1 l model. Int J Mech Sci 2007;49(11):1251–61.
Tz1 = t q +w (A.5) [12] Argyris JH, Angelopoulus T, Bichat B. A general method for the shape finding
2
x221 + y221 + 2
z21 of lightweight tension structures. Comput Methods Appl Mech Eng 1974;3(1):
135–49.
x2 − x1 [13] Barnes M. Form-finding and analysis of prestressed nets and membranes.
Tx2 = −t q (A.6) Comput & Structures 1988;30(3):685–95.
x221 + y221 + z21
2 [14] Vilnay O. Static and dynamical response of cable nets. Int J Solids Struct 1990;
26(3):299–312.
y2 − y1 [15] Pellegrino S. Structural computational with singular value decomposition of
Ty2 = −t q (A.7) the equilibrium matrix. Int J Solids Struct 1993;30(21):3025–35.
[16] Karoumi R. Some modeling aspects in the nonlinear finite element analysis of
x221 + y221 + z21
2
cable supported bridges. Comput & Structures 1999;71(4):397–412.
[17] Jayaraman H, Knudson W. A curved element for the analysis of cable
z2 − z1 l
Tz2 = −t q +w (A.8) structures. Comput & Structures 1981;14(3):325–33.
2 [18] Andreu A, Gil L, Roca P. A new deformable catenary element for the analysis of
x221 + y221 + z21
2
cable net structures. Comput & Structures 2006;84(29):1882–90.
[19] Der kiureghian A, Sackman J. Tangent geometric stiffness of inclinated cables
where (xp , yp , zp ) are the coordinates of point p, w is the self weight under self-weight. J Struct Eng 2005;131(6):941–5.
of the truss, l is the length of the truss, t is the axial tension inside [20] Lopez-Garcia O, Carnicero A, Torres V. Computation of the initial equilibrium
of railway overheads based on the catenary equation. Eng Struct 2006;28(10):
the truss and (xba , yba , zba ) are the relative coordinates of point b
1387–94.
to a. [21] Schek H. The force density method for form finding and computation of general
The compatibility between tension and length must be networks. Comput Methods Appl Mech Eng 1974;3(1):115–34.
conserved. Therefore [22] Mollaert M. Form finding of mixed structures. In: Guildford, editor. Third
  conference on space structures. 1984. p. 180–5.
l [23] Moncrieff E, Topping B. Computer methods for the generation of membrane
q
x221 + y221 + 2
z21 − l+t =0 (A.9) cutting patterns. Comput & Structures 1990;37(4):441–50.
EA [24] Lai C, You Z, Pellegrino S. Shape of deployable membrane reflectors. J Aerosp
Eng 1998;11(3):73–80.
must apply. [25] Pipkin AC. Catenary deformations of inextensible networks. J Eng Math 1994;
In order to use a spring with

stiffness k as in Example 3.4 instead 28(5):401–6.
of a truss, the term l + t EAl should be substituted by l + kt .


[26] Suzuki Y, Miyata E, Iverson S. Static analyses of the triangular running skyline
system: A three-dimensionally movable logging cable system. Comput &
Structures 1996;60(4):579–92.
References [27] Huang Y, Lan W. Static analysis of cable structure. Appl Math Mech 2006;
27(10):1425–30.
[1] Siev A, Eidelman J. Stress analysis of prestressed suspended roofs. J Struct Div, [28] Irvine H. Cable structures. Cambridge (MA): MIT Press; 1981.
ASCE 1964;90:103–21. [29] Cella P. Methodology for exact solution of catenary. J Struct Eng 1999;125(12):
[2] O’brien T. General solution of suspended cable problems. J Struct Div, ASCE 1451–3.
1967;93:1–26. [30] Bruno D, Leonardi A. Nonlinear structural models in cableway transport
[3] Poskitt T. Numerical solution of nonlinear structures. J Struct Div, ASCE 1967; systems. Simul Pract Theory 1999;7:207–18.
4(August):69–94. [31] Kanzaki K, Sakai T. Studies on the cable crane hung at three supports (I)(1)
[4] Hitchings D, Ward P. The non linear steady-state response of cable networks. A way of calculation supposing static balanced state. J Jpn For Soc 1972;54:
Comput Methods Appl Mech Eng 1976;9(2):191–201. 103–12.
[5] Irvine HM, Sinclair GB. The suspended elastic cable under the action of [32] Brennan TWu. Basic analytical study of pantograph-catenary system dynam-
concentrated vertical loads. Int J Solids Struct 1976;12(4):309–17. ics. Veh Syst Dyn 1998;30(6):443–56.