Composites in Construction 2005 Third International Conference

Lyon, France, July 11 13, 2005

CABLE STRUCTURE WITH LOAD-ADAPTING GEOMETRY

A.S. Jlich, J.F. Caron and O. Baverel
Institut Navier - Lami, ENPC-LCPC
6 et 8, avenue Blaise Pascal Cit Descartes Champs-sur-Marne 77455 Marne-la-Valle
Cedex 2, France
[email protected], [email protected], [email protected]
ABSTRACT: For an application in a composite cable-stayed footbridge a system is developed,
which is to ensure an equal distribution of static or quasi-static life loads in the composites
retainers. Thus the allowed life load can be maximized for this kind of structures while maintaining
the necessary wide safety margin. The structures optimal geometry for the control procedure is
determined by means of an algorithmic formfinding process, based on the method of force density.
The results of the shape optimization seem to match with a mechanical device to be developed.
1.

INTRODUCTION

The fragile behaviour of composite materials (carbon and glass fibres) and the lack of experience
require high safety factors, which are prejudicial to the introduction of composites in civil
engineering. The studys context is an all-composite cable-stayed footbridge for which a shape
control system is developed. The shape of the footbridge will adapt itself to external life-loads to
equalize as much as possible the stresses within the different types of elements. As a result the
allowable life loads can be maximized even if high safety factors are to be kept.
This paper presents briefly the method of force density. Then this method is applied in an
algorithmic form to a simplified 2D structure inspired by the final cable-stayed footbridge. Finally
the optimisation results are interpreted for an application to a mechanical control device.
2.

FORCE DENSITY METHOD

With the force density method, introduced by Linkwitz in 1971 [1,2], equilibrium shapes of
prestressed cable-structures can easily be found. As a highly non-linear system of equations is
linearised, this method is proved to be highly efficient when determining consistent equilibrium
forms for tension nets.
The method of force densities is briefly introduced means a simplified structure whose geometry is
chosen to be close to the footbridges geometry we plan to study. Fig. 1 shows the Finite Element
model created with ANSYS for the cable-stayed footbridge. Fig. 2 shows the simplified model
chosen for the shape optimisation with Scilab and the notations used.

Composites in Construction 2005 Third International Conference

Lyon, France, July 11 13, 2005

y
10

S2

S3

S1

S4
C2

C1
0

1
0

C3

Fy

10

15

Figure 2 Simplified model with notations

Figure 1 FE-Model in 3D.

The simplified structure consists of four stays (S1, S2, S3 and S4) and a cable made of three parts
(C1, C2 and C3). Both element types work in tension only. Points 1, 2, 3, 4, 5 and 6 are
constrained in both x- and y-direction, thus their coordinates are known. We will determine then the
coordinates of points 7 and 8 according to the equilibrium shape for a load set. The structure loads
we consider are prestress and eventually a force on nodes 7 and/or 8. In the following equations
we chose to apply a vertical load Fy on node 7.
2.1

The equations of the method of force density

Considering the structures equilibrium on nodes 7 and 8, we can write down the following
equations:
(x7

x1 )

Node 7
(y7

y1 )

(x8

f(17 )
l(17 )
f(17 )
l(17 )

x7 )

Node 8
(y8

with:

xk

y7 )

+ (x7
+ (y7

f('78 )
l( 78 )
f( 78 )
l(78 )

+ (x8
+ (y8

x2 )
y2 )

f( 27 )
l( 27 )
f( 27 )
l( 27 )

x3 )
y3 )

x -coordinate of node k

y k y-coordinate of node k

+ (x7

x4 )

+ (y7

f( 38 )
l( 38 )
f( 38 )
l( 38 )

y4 )

+ (x8
+ (y8

f( 47 )
l( 47 )
f( 47 )
l( 47 )

x5 )
y5 )

+ (x7
+ (y7

f( 58 )
l( 58 )
f( 58 )
l( 58 )

+ (x 8
+ (y 8

x8 )
y8 )

f( 78 )
l( 78 )
f( 78 )
l( 78 )

x6 )
y6 )

=0

(1)
= Fy

f( 68 )
l( 68 )
f( 68 )
l( 68 )

=0

(2)
=0

f( jk ) force within the element between node j and k

l( jk ) length of the stress-extended element between
node j and k

The distance between node j and k, i.e. the length of the stress-extended element, is described
using Pythagoras law l( jk ) = ( x j

x k )2 + ( y j

y k )2 . The system of equations given by (1) and (2)

is then obviously highly non-linear regarding the geometrical variables (x7, y7, x8 and y8).
Introducing force densities defined as q(jk) = f(jk) / l(jk) the equations (1) and (2) can be linearized as
shown in equations (3) and (4).
Node 7
Node 8

x 7 (q(17 ) + q( 27 ) + q( 47 ) + q( 78 ) ) x 8 q( 78 ) = x1q(17 ) + x 2 q( 27 ) + x 4 q( 47 )
y 7 (q(17 ) + q( 27 ) + q( 47 ) + q( 78 ) ) y 8 q( 78 ) = y1q(17 ) + y 2 q( 27 ) + y 4 q( 47 ) + Fy
x 7 q( 78 ) + x 8 (q( 78 ) + q( 38 ) + q( 58 ) + q( 68 ) ) = x 3 q( 38 ) + x 5 q( 58 ) + x 6 q( 68 )
y 7 q( 78 ) + y 8 (q( 78 ) + q( 38 ) + q( 58 ) + q( 68 ) ) = y 3 q( 38 ) + y 5 q( 58 ) + y 6 q( 68 )

(3)
(4)

with:

q( jk ) the force density of the element between nodes j and k, i.e. f( jk ) l( jk ) .

We obtain a linear system of equations with 4 equations and 4 unknowns (x7, y7, x8 and y8). A
particular set of force densities q(jk) relates to a unique equilibrium shape.
2.2

Application to the prestressed structure

As Linkwitz noticed [3], investigations and practical experiments justify the choice of very simple
types of force densities to create equilibrium shapes as an initial approach to formfinding: a force
density q = c is assigned to elements of equal length while a force density of q = c / l is used for
structures with elements of irregular length. In both cases c is a constant value, but while the
constant c means a constant force density when using the first type of force densities, the
constant c stands for a constant force when applying force densities of the second type. The
structure we study has elements of irregular length. Hence for the shape optimization the force
densities q = c / l of the second type are chosen.
Two different sets of force densities qs for the stays and qc for the cable elements are chosen, i.e. a
constant force cs = fs for all the stays and a constant force cc = fc for the cable. These force
densities state a prestress distribution within the structure. Remembering the force density
definition q = f / l, it becomes obvious that implementing a non-zero force density to an element
results in an equilibrium form with a non-zero force for this element. Thus all elements are
subjected to a force. The two different sets of force densities qs / qc form two different states of
prestress distribution. No force Fy is applied.
The equilibrium forms for the initial shape 1 are given in Fig 3 and Fig 4. The force densities
chosen for Fig. 3 are qs = 1 / l and qc = 2 / l, with l relating to the length of the stress-extended
element in the initial geometry. The equilibrium shape of Fig. 4 results from choice of the force
densities qs = 1 / l and qc = 10 / l. In Fig. 5 the ratio of qs / qc = 1 / 10 is preserved but the
equilibrium is calculated for an initial shape 2. The initial geometry is drawn in a dotted line and the
equilibrium form is represented as an unbroken line.
10

10

10

0
0

10

Figure 3 Initial shape 1 and

equilibrium for
qs = 1 l and qc = 2 l .

15

10

Figure 4 Initial shape 1 and

equilibrium for
qs = 1 l and qc = 10 l .

15

0
0

10

Figure 5 Initial shape 2 and

equilibrium for

qs = 1 l and qc = 10 l .

It can be seen that:

the force density ratio controls the radius of curvature of the cable (elements C1, C2 and
C3).
the initial shape has an influence on the equilibrium shape. The information of the initial
shape is not included through the coordinates of the 4 unknown (x7, y7, x8 and y8) in
equations (3) and (4) but through the lengths of each element in the force densities q.
Actually, to determine the equilibrium shape, the force densities are calculated depending
on the initial extended length of the elements and evidently different initial geometries
match with different initial element lengths.

15

3.

ITERATIVE STRESS-EQUALIZING METHOD

The chosen algorithm is very simple. Other types have been proposed by [4,5] but as the aim is
not to develop a highly efficient and fast algorithm but a stress control device, this approach seems
justifiable.
The aim of the presented algorithm is to determine geometries providing a (quasi-)equal stress
distribution within each type of element. As each type of element has a determined cross-section,
to equalize stresses means to equalize forces within the elements no matter what their length.
Fig. 6 shows the iteration process chosen; troughout all the iterations each c (cs and cc) is kept
constant and thus corresponds to the equalized forces fs and fc to be reached (if possible) within
the elements.

INITIAL SHAPE
x (k,)initial , y (k,)initial
Calculation of the stressextended lengths

l (jk)

Calculation of the force densities q(jk) =c/l (jk)

Solve force density equations

new x (k) and y (k)

convergence?

or
maximum iteration number?

yes

no

OPTIMIZED SHAPE
~c
x (k,)opti , y (k,)opti , f (jk,)opti =
Figure 6 Iteration process to generate equilibrium forms equalizing stresses within
elements.
However, it should be noticed that it is not always feasible to determine a shape that leads to the
imposed forces in each element. Nevertheless the iterative process can determine a shape as
close as possible to the requirements; this will be the optimized geometry. This is why two classical
criteria are tested to stop the iteration procedure. On the one hand we have a convergence test
and on the other hand a maximum iteration step.
3.1

Application 1: prestressed structure

For this evaluation the initial shape 2 is chosen because of the greater variation of the elements
length. This allows an easier evaluation of the influence of the initial geometry. Fig. 7 shows the
improvement of the equilibrium shape for qs = 1 / l and qc = 10 / l throughout the iterations, i.e.
iterations i = 1, 2, 10. Table 2 gives the maximum element length difference of all elements for
these iterations. This maximum difference dlmax
is the maximum difference between the distance
i
of nodes ji and ki (or the stress extended length of the element between nodes ji and ki) of
iteration i and the distance of nodes ji - 1 and ki - 1 of the previous iteration step i - 1, i.e.
dlmax
= max l( jk ,)i l( jk ,)i 1 . Studying the maximum length difference, the improvement of the
i
equilibrium geometry can be observed. When the new geometry does not particularly differ from
the previous one, i.e. the maximum length difference dlmax
gets close to zero, the force densities,
i
depending on the lengths, do not vary much either. The algorithm converges. This can easily be
noticed in Table 2 on the fact that the average forces of the stays fs,i c s = 1 and of the cable

elements fc,i c c = 10 quite reach the required forces c in iteration step i = 10. Additionally the
standard deviations ss,i and sc,i. indicate that a good equalization of the forces is archived for i = 10.
Table 2

Trends of dlmax
, fs,i and fc,i ,
i

10
initial shape

s s,i and sc,i

dl imax

1 1.896
2 0.136
10 0.013

fs,i

s s,i

fc,i

s c,i

1.621
1.074
1.000

0.859
0.048
0.002

7.127
9.803
10.024

0.723
0.157
0.093

shape i=1

5
shape i=2
optimized shape

0
0

10

15

Figure 7 Improved equilibrium shape for

iteration steps i = 1, 2, 10.

Considering the evolution of the equilibrium geometry throughout iteration steps, it can be fixed
that:
the greatest variation of shape occurs between initial geometry and equilibrium geometry of
the first iteration step i = 1.
the optimized geometry is determined by the initial shape. This can be noticed considering
that even though a symmetrical prestress load is applied the lengths of elements C1 and
C3 are not equal.
3.2

Application 2: loaded and prestressed structure

The influence of external load on the optimized shape of the prestressed structure is now studied.
As noticed in the previous chapter, the initial geometry is significant when determining new
optimized shapes. This is why the initial prestress shape of the structure has to be chosen
carefully. The selection procedure for an optimized initial shape will not further be discussed in this
paper, but it should be kept in mind that this detail will lead to additional investigation.
In this paper, the selected optimized prestressed shape is the equilibrium form shown in Fig. 2.
The force density ratio that leads to this optimized shape is close to the ratio resulting from the
previous Finite Element study of the footbridge. The structure is loaded in two different manners.
First load case (LC1): The structure is loaded symmetrically with a load Fy = 10 applied on
nodes 7 and 8. These forces model the pedestrian load. To take into account the new internal
stress distribution resulting from external loading the adapted force densities allotted are qs = 10 / l
and qc = 25 / l. These force densities are also determined based on the Finite Element computation
results.
Second load case (LC2): The structure is loaded asymmetrically with Fy = 10 applied only on
node 7. The matching force densities are qs = 5 / l and qc = 20 / l.
The Fig. 8 and 9 illustrate the equilibrium shapes for both load cases at iteration step i = 10. As for
the previous figures, the dotted line shows the initial geometry and the unbroken line the
equilibrium form.
10

10

10

15

10

15

Figure 8 Equilibrium for LC1 with qs = 10 l

Figure 9 Equilibrium for LC2 with qs = 5 l

and qc = 25 l .

and qc = 20 l

The initial stress-extended length lprestress and the stress-extended additional element length lLC
each element requires to create the equilibrium geometry are displayed in Table 3 for both load
cases. The additional stress-extended element length is lLC = lLC lprestress . The additional element
length

Table 3

Initial stress-extended length and required additional stress-extended element length to

create optimized shape.

Element

S1
6.371

S2
10.650

S3
10.650

S4
6.371

C1
4.795

C2
5.525

C3
4.795

Cable
15.115

lLC1

-0.708

0.304

0.304

-0.708

-0.789

1.667

-0.789

0.089

lLC 2

-0.176

0.946

-0.013

-0.620

-0.927

1.464

-0.453

0.085

lprestress

4.

lLC(,C ) = lLC(,C1) + lLC(,C2 ) + lLC(,C3 ) .

lLC(,C ) of the entire cable is calculated by

EVALUATION OF A CONTROL DEVICE MOCK-UP

All the lengths given in the previous chapters are lengths of stress-extended elements. The
optimization approach takes into account neither the geometry of the elements nor the elements
material properties. In fact with the force density method only the structures geometry is adapted
to create a force flow matching with a determined force distribution.
The next step is the construction of a mock-up of the simplified structure to validate the numerical
results and to start developing a control device. When determining the cutting pattern of the mockup, the slack lengths of the elements are needed. It becomes necessary to consider the elements
geometry and material. In order to make the notations clear, the Fig. 10 shows graphically all
notations used for the lengths.

f LC(,jk)
l LC(,jk)

l LC(,jk)

slack

l LC(,jk)

slack

l prestress(,jk)
slack
l LC(,jk)

l LC(,jk)
slack
l LC(,jk)

l prestress(,jk)
l LC(,jk)

l prestress(,jk)

l prestress(,jk)
slack
l prestress(,jk)

f prestress(,jk)

k
f prestress(,jk)

Prestress

f LC(,jk)

Load Case

Figure 10 Notations of lengths l( jk ) and

l(slack
jk ) ,

additional element lengths

elongations l( jk ) and additional lengths

l( jk ) and

l(slack
jk ) .for prestress and load cases.

Considering Hookes linear law to be applicable for the elements material of this structure, the
slack lengths l(slack
jk ) of the elements can easily be determined by the following equations:
l( jk ) = l(slack
jk ) + l( jk )
l( jk ) =

l(slack
jk )
EE A E

f( jk ) =

(5)
l(slack
jk )
EE A E

cE

(6)

l(slack
jk )

with:

= l( jk )

cE
1+
EE A E

(7)

l( jk )

length of the stress-extended element between node j and k

l(slack
jk )

slack length of the element between node j and k

l( jk )

elastic elongation of the element between node j and k

f( jk )

force within the element between node j and k

EE
AE
cE

stiffness of the material of the element type E

area of the cross section of the element type E
constant matching with the imposed force for element E

slack
Table 4 shows the slack length for each element in the initial state of prestress called lprestress
, and

slack
. The additional slack element length
for both load cases called l LC

each load case (LC) is also calculated with

slack
slack
lLC
= lLC

slack
lLC
of each element for

slack
lprestress
. According to the FE-model of the

footbridge the stiffness of the stays and of the cable is ES = EC = 40 000 MPa, the cross-section of
the stays is AS = 0.0001 m2 and the cross-section of the cable is AC m2. To match with the
dimensions, all lengths in Table 4 are given in meters.
Table 4

Slack lengths lslack and required slack additional length

Element

slack
lLC
to create optimized shape.

S1

S2

S3

S4

C1

C2

C3

Cable

slack
lprestress

6.369

10.647

10.647

6.369

4.795

5.525

4.795

15.114

slack
l LC
1

5.649

10.927

5.649

10.927

4.006

7.192

4.006

15.204

slack
l LC
1

-0.720

0.280

-0.720

0.280

-0.789

1.667

-0.789

0.090

slack
lLC
2

6.187

11.281

10.624

5.744

3.868

6.989

4.342

15.199

slack
lLC
2

-0.182

0.634

-0.023

-0.625

-0.927

1.464

-0.453

0.085

Because of the symmetry of LC1, the slack length and additional length of the element S3 are
identical to those of element S2, and similarly element S4 corresponds to element S1, and C3 to
C1.
The greatest stricture needed is the one to form the equilibrium shape of LC1. The element S1
slack
(and the element S4) has to be shortend by
lLC
1(,S1) = 0.720m slack length.
The most important additional length is required for the equilibrium geometry of LC2. Here the
slack
element S2 needs lLC
2(,S 2 ) = 0.634m extra slack length.
The cable remains almost the same length: the maximum additional length is

slack
lLC
1(,C ) = 0.090m.

Considering the application on the 3D-footbridge, the necessary additional slack length will
probably be smaller because of a greater number of retainers and cable elements. The results of
this first evaluation allow further reflections; it should be possible to design a control device able to
adapt the elements length within these margins.
5.

CONCLUSIONS

The final aim is to control the stresses within elements of a loaded structure by adapting the shape.
In the first part of this paper a short synopsis of the force density method was given. Then a basic
7

iteration process was presented. The forces in the different element types are to be equalized
adjusting the lengths of the elements to the load situation. In the third part a preliminairy approach
allowed an evaluation of the element elongations required for this adaptive shape structure. The
order of magnitude of the calculated elongations seemed suitable to the whole geometry and
applying such displacements on the structure should be possible with a mechanical device to be
developed.
[1] Schek H.J. The force density method for form-finding and computation of general networks.
Computer Methods in Applied Mechanics and Engineering 3; 1974; p.115-134.
[2] Lewis W.J. Tension structures, Form and behaviour. 1st ed. London: Thomas Telford; 2003;
p. 51-57.
[3] Linkwitz K. About formfinding of double-curved structures. Engineering Structures; 1999; 21;
p. 709-718.
[4] Linkwitz K. Formfinding by the direct approach and pertinent strategies for the conceptual
design of prestressed and hanging structures. International Journal of Space Structures; 1999; 14;
No. 2; p. 73-87.
[5] Singer P. Analogies between minimal surfaces and membrane constructions (numerical part).
Natrliche Konstruktionen, 1994; SFB 230; p. 107-111.