Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.

com, ISSN 1743-3509

Analysis and design of cable net structures

through optimization techniques
P.M. Sufian, A.B. Templeman
Department of Civil Engineering, University of
Liverpool, PO Box 147, Liverpool L69 3BX, UK

ABS TRACT
This paper shows that numerical optimization methods provide a com-
prehensive and rigorous basis not only for design but also for a variety
of very practical analysis problems associated with grossly non-linear
cable structures. The static analysis is achieved through unconstrained
optimization of the total potential energy stored in the structure. The
pretension design is set in a multicnteria optimization context. A
minimax solution is found by means of an entropy-based optimization
algorithm. Illustrative examples are solved.

INTRODUCTION
The behaviour of cable net structures is characterised by a combination
of geometrical and physical non-linearities. Geometric non-linearities
arise from the fact that such structures equilibrate applied loading by
large changes of shape but small strains. Physical non-linearities are
caused by the fact that cables are able to carry only tension forces and
become slack under compressive loads. Further non-linearity may arise
from possible yielding of the tension cables.
The analysis of these non-linear structures involves two iterative
calculations: one to determine the zero configuration, which results from
the application of pretensioning to an untensioned net, and one to de-
termine the final configuration, resulting from the application of service
loading upon the prestressed structure. The design of cable net structures
consists of finding the best distribution of prestress to achieve satisfac-
tory performance in terms of cable stress and joint displacement levels,
and is even more complex than the analysis problem. The overall theme
of this paper is to show that computer-based numerical optimization
methods provide a comprehensive and rigorous basis for practical analy-
sis and design problems associated with grossly nonlinear cable net
structures.
The static analysis of cable net structures is achieved through direct
unconstrained optimization of the total potential energy stored in the
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

492 Optimization of Structural Systems

structure. This approach to analysing cable net structures has been de-
scribed by Buchholdt [1] and later developed by Sufian [2] and Sufian
and Tempieman [3] to include pretensioning analysis through specifica-
tion of cable element shortnesses representing externally applied preten-
sioning forces. The zero configuration is found by introducing desired
pretension forces via end cable elements in specified locations. These
end elements are assumed to be slightly shorter than geometrically nec-
essary by specified shortnesses over which they are stretched into posi-
tion during the process of pretensioning. This analysis approach is
described in detail in Refs. [2,3] and is used in the present work. A brief
outline of this method is given in the next Section of this paper.
Nodal displacements can usually be reduced by increasing the levels
of pretensioning forces. However, this requires the use of larger diameter
cables, more robust clamps and anchors and much stiffer supporting
structures, hence, a more expensive structure. For this reason, prestress-
ing is one of the most important features to be accounted for in designing
cable nets and is an ideal candidate for optimization.
As a design goal, it is desirable to find as low a level of prestressing
as possible which satisfies the performance requirements for cable net
structures. This paper develops a numerical method which seeks the
minimum level and optimum distribution of prestress whilst satisfying
prescribed limits upon cable stresses and nodal deflections under any
given load condition which the structure encounters during its service
life. This is achieved by posing an optimization problem in terms of
shortness variables for which values are sought, design is posed in a
vector (multicriteria) optimization format in which normalised constraint
goals, governing the target behavioural restrictions, are assumed as ob-
jective functions to be minimized. The formulated minimax design prob-
lem is conveniently converted into a convex scalar optimization function
by means of an entropy-based technique [4,5] which can be solved using
any unconstrained optimization algorithm.

STATIC ANALYSIS
The analysis approach is to consider the determination of equilibrium
as a minimization process of the total potential energy. The total po-
tential energy, TT, in any loaded structure is the summation of the strain
energy, S, stored in the structure and the potential energy, [/, of the
external loads and may be expressed as
7r = S-hC7 (1)
The condition for equilibrium can be mathematically expressed as
dn = 0 (2)

Cable nets subjected to pretensioning

The zero configuration geometry and pretensioning forces can be calcu-
lated using the minimum energy method under the following conditions
1. The unstressed configuration geometry is known and is pretensioned
by stretching any of the end cable elements.
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 493

2. These end cable elements are assumed to be initially shorter than is

geometrically necessary to stretch the cables on to the supports.
Consider a shortened member, k , connected to adjoining inner members,
i, at node j in space as shown in Figure 1. The end member, k, is of
initial length Lr* which is shorter by an unknown amount SLr* than the
physical distance between joints j and m. Applying some pretension force
P* to member k stretches this member on to support m so that the cable
system pretension forces are then TL in members i and T<* in member k.
The free joints n andy will displace from their initial coordinates
(xm, y™, 2m) and (xrj, Vrj, Zrj) to a new equilibrium position (.r^, yon, 2<>n) and
(*«>, )'„, z,y) respectively.

Reference configuration
Zero configuration

Figure 1. Shortened end cable element connected to

adjoining elements at joint j in space

Total potential energy function The total potential energy function, TT, is
given by

w (3)
,=1 6=1
where / = (nm - ne), nm is the total number of cable elements in the net
structure and ne is the total number of end cable elements stretched,
K( = EAjL) is the axial rigidity, ?„, and e^ are the elongations of cable
elements i and k and can be expressed in terms of the nodal coordinates
thus
f., = L,,-Lr, W
(5)
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

494 Optimization of Structural Systems

where £/„ and Z/& are untensioned and tensioned lengths of member i re-
spectively and are given by
L,, = {(*„, - *,/ + CXrn - y/ + (*rn - *r/} "* (6)

*•« = {(*„, - *<>/ + (y« - */ + (*», ~ *„•)¥* (7)

Similarly

4. = {(*„ - *r/ + (Xm - *•/ + (*™ ~ *r/} "* ~ ^rk (8)

4. = (K, - *<,/ + (Xm - >•</ + (*m - VV" 0)

where (.r^y™, z*,) are the coordinates of the fixed joint m. Equation (3)
is a function of two sets of unknown variables: the first set contains the
three free joint coordinates (*„, 7,, a,) and the second set contains the end
member shortnesses (5 1/,*. The derivatives of % with respect to these two
sets of unknowns with respect to the unknowns are given [2,3] by

(*°J~*°Ji ,...
Vo,— -— (ID

with similar derivatives of n with respect to y,, and z<,/. nj is the number
of cable elements meeting at joint j and x<>n, yon and z<>n are the coordi-
nates of the joint connecting the remote end of cable element i. If an end
member, k, meets joint j then **,, )',„, £<,„ in Equations (11) become
*m, Jm, 2m respectively.
At the stationary point of TT, setting Equations (10) and (11) to zero
gives the equilibrium condition of the pretensioned cable net structure.
If P* are specified, Equations (10) and (11) are solved to find the required
initial shortnesses dL* and the free joint coordinates (%„, y,, z<,). However,
in practice it is more practical for the initial shortnesses to be specified
than the pretension forces P*. Therefore, if 5Lrk are specified, Equations
(10) and (11) can be solved to give values for coordinates (*<,, y, , z*) and
the pretension forces P* necessary to satisfy equilibrium.
In this work Equation (3) was minimized directly to determine equi-
librium, rather than using Equations (10) and (11) which are non-linear
and hard to solve. Examining Equation (3) further, it can be seen that
if the initial shortnesses dLrk of end members are specified the first two
terms become functions of the unknown coordinates only and the third
term is effectively constant. Consequently the minimization process of
Equation (3) can be considerably simplified by minimizing

,) 4- ) (12)
i=l 6=1
over the free joint coordinates only.
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 495

Response to static loading

The applied load on a cable net structure is assumed to act at the nodes
only. Once the zero configuration and the pretension forces in the cables
have been found, the final configuration and member forces can be
worked out in the same way as for prestressing, using the zero config-
uration as datum.
The expression for the total potential energy function is given by

where e is the elongation of a cable element due to static service loads,

To is the initial pretension forces in the cable element, s, is the strain
energy in the cable element due to pretension force and (F& F%, F,/) and
(<5*,, ;%,-, dv) denote the components of the applied load at joint j and "he
displacements of joint j respectively, nf, is the total number of loaded
joints.
The elongations and nodal deflections can be expressed in terms of
the variable joints coordinates thus for element i

where L« and Z/p are the prestressed and loaded lengths which can be
related to the global coordinates as
= {(* - *•)* + (y«, - y, + (*<,„ - )-}"' (15)

Finally, the nodal deflections can also be expressed in terms of the global
coordinates in the following way

Hence Equation (13) is expressed in terms of the known coordinates of

the zero configuration and the unknown joint coordinate variables over
which the equation is minimized for equilibrium to find the final con-
figuration.

Derivatives of the total energy function with respect to the variable coor-
dinates Xf,yr and Zf at joint j can be expressed as:
/v / \
(r, + Ke)i *g"*""' - f (18)

with similar derivatives of n with respect to y^ and ZQ. where %/%, y/>, and
2/h are the coordinates of the joint connecting the remote end of cable
element i.
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

496 Optimization of Structural Systems

Minimization of the total potential energy

The minimization of the total energy functions is treated as an uncon-
strained optimization problem and is expressed in the following form
minimize n (X) (19)
x
where n(X) is the total energy function expressed in terms of, X, the
joints coordinate variable vector. Checks on the cable strains during the
optimization process are incorporated and the strain energy terms corre-
sponding to slack cable elements are removed when necessary.
Problem (19) is a straightforward unconstrained non-linear opti-
mization problem for which a variety of solution methods are available.
The one used in the present work was the conjugate gradient method [6].

PRETENSION DESIGN
As was noted earlier, the magnitude of the nodal displacements may be
reduced by increasing the levels of prestressing forces in the cable net.
As a design goal, it is desirable to find as low a level of prestressing as
possible which satisfies the performance requirements for a cable net
structure.
One important performance requirement for a cable net is that all
of the cable elements remain stressed under the various loading cases.
The prestress should be neither too large nor too small but sufficient to
maintain a tensile force under all conditions. The other important re-
quirement for a pretensioned cable net is that no excessive deflection of
the joints should occur under any of the loading cases that the structure
may encounter during its service life.

Mini max problem formulation

Assume that an initially untensioned (reference) configuration of a cable
net structure is given and the cross sectional areas of the cable elements
are known. Pretensioning is carried out by specifying shortnesses in any
of the end cable elements as described earlier. A combination of shortness
magnitudes and shortened cable elements positions which give the lowest
possible prestressing forces is desirable. If the total number of end cable
elements chosen to be stretched is denoted by ne and the shortnesses in
these elements denoted by (5L, then, <5L is a vector of end cable element
shortnesses, <5L,, i = 1, , ne . These are design variables for the opti-
mization.
One set of goals arises from the requirement that under any dead
or static service loading, the displacement of the joints should all be as
small as possible. A question then arises as to which configuration the
displacements should be measured relative to: the initial reference con-
figuration or the zero configuration. In this work it was decided that a
maximum desirable value of the displacements dm** of any joint under
applied, dead or static service loads from the zero configuration should
be imposed. If &/, 5yj and 6,, are the displacement components of joint j
from the zero configuration due to applied load, then the displacement
goals are
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 497

<5.v^mn.x; <5,,^<W M*m« (20)

The transverse displacements in the direction of the applied loads are
large in magnitude compared to those in the longitudinal directions, and
therefore are more critical. It is, therefore, only necessary to constrain
the displacements in the direction in which the the loads are applied. If
a sign convention is chosen such that the y axis represents the transverse
direction in which the load is applied, then out of constraints (20) only
displacement goals (21) for joint j need be considered.

It is only necessary to limit the displacements of those joints which are

prone to undergo large deflections.
Imposing a lower limit on the total forces in a cable ensures that
cable elements do not become slack. If TLn is the value of a minimum
desirable tensile force in any cable element, a minimum tension force
goal can be written in the form
7% > T"min (22)
where 7% is the final force in a particular cable element i . Similarly, the
maximum force in any cable element may be constrained not to exceed
a desirable maximum value 7L*x by means of the goal
TfiZT^ (23)
The optimization method used in this work requires that all these goals
should be cast in a normalized form. The goals (21), (22) and (23) may
be written in the normalized form as follows:
&.
G:(5L) = -j^- - 1 < 0 j = 1,......., J (24)
' <5nHuc

-7^-l<0 *=c/+l,......., J + nm + l (25)

*fl*
rp
Gi(5L) = -=£• -- 1 < 0 i = J + nm + 2,......., J + 2nm (26)
max
where J is the total number of joint displacement restrictions and nm is
the number of cable elements.
It is desirable to change the values of the variable vector 5L in such
a way that the values of all the elements of the goals are made as small
as possible. This can be achieved by minimizing the maximum of the goal
element values over variables <5L thus casting the problem in a minimax
format which may be defined as
Minimize Maximum < G,-(<5L) > (27)
6L i
in which G,, i = 1,.........., J + 2nm, are the goal functions (24) to (26).
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

498 Optimization of Structural Systems

Minimax optimization
Problem (27) is discontinuous and non-differentiable which makes its
numerical solution by direct means difficult. In this work the solution to
the minimax problem (27) is found indirectly by the unconstrained min-
imization of a scalar function which is continuous, and thus considerably
easier to solve. The following theorem [4,5] can be proved.

Theorem
The vector x which solves the vector minimax problem
Minimize Maximum <Gj(x)> (28)
xeX (=1....../
where G is a vector of dimensionless goal functions, is generated by
solving the scalar optimization problem

Minimize ( ) l n expCpG^)] <29)

with positive values of the parameter p increasing towards infinity.

The minimax optimization problem (27) for the cable net structure
with goals defined by Equations (24) to (26) was solved by the scalar
minimization problem:

Minimize Fp = (-j-) In (30)

6L *

over an increasing positive sequence of/?, where M = J + 2nm.

Scalar optimization
Problem (30) may, in theory, be solved by a variety of conventional scalar
optimization methods. Derivatives of Fp with respect to all variables 6L
may be formed but contain the first derivatives of the goal functions
Gj(dL) . Some of these goals are implicit functions of AL and their de-
rivatives are consequently very hard to obtain by algebraic means. A
possible alternative to analytical derivatives is the use of numerical de-
rivatives. However, in this cable net problem this would require ne + 1
complete analyses of the cable net under all loading cases in order to
calculate numerical first derivatives for use in a first order solution
method for problem (30). This would be enormously time consuming.
The solution to problem (30) may be achieved by other methods
which do not require derivative values and thus the complication of their
evaluation can be avoided. The choice of unconstrained optimization
methods that can be used to solve problem (30) is, therefore, restricted
to those in which derivative evaluations are not required. Optimization
methods which do not require derivative values are less efficient than
those which require derivatives. However, the increased efficiency which
could be achieved by first order optimization methods is outweighed by
the complication involved in evaluating the first derivatives. In the
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 499

present work, therefore, a zeroth order optimization method, namely the

nonlinear Simplex method [7], was used to solve problem (30).

Optimization Design algorithm

The main steps in the algorithm used for the optimization of cable net
structures can be summarised as follows
• Step 1: Input data: reference configuration, material properties, all
loading cases and an initial set of shortnesses as the starting point
# Step 2: Choose a value of p for solving problem (28)
• Step 3: Analyse the the net structure for pretensioning plus each ap-
plied loading system using the minimum energy method
• Step 4: Find the numerical values of all goal functions for use in
problem (28)
• Step 5: Solve problem (28) by any zeroth optimization method (such
as the simplex method). This gives new estimates for all end cable
element shortnesses.
• Step 6: If iteration have converged stop iteration and go to the next
step otherwise return to step 3
• Step 7: Retain the new design and use it as a new starting point and
increase the value of p then repeat the optimization by returning to
Step 3. If the difference between the starting point and the new de-
sign is small stop the algorithm and use the new design as the opti-
mum.
The minimax optimization algorithm requires a sequence of positive
values of p increasing towards infinity. In the present work a value in
the range 10< to 10* was used for the solution of problem (28). Further
algorithmic details and computational aspects can be found in Ref.[2].

DESIGN EXAMPLES
The above optimization procedure has been used to find the optimal
shortness set and the corresponding pretension state for two cable net
structures. The first net model examined is simple and has been solved
in order to illustrate the proposed design algorithm and to validate and
confirm the accuracy of the results. The method is then used to find the
optimum distribution of prestressing forces in a more realistic cable net
structure.

Example 1: Cable structure

A simple five member cable structure is shown in Figure 2. Before per-
forming the design calculation it is appropriate to examine the behaviour
of the loaded cable structure under different prestressing states. The
structure is to carry a point load of 60kN in the vertical direction at the
free joint, 2, as shown in Figure 2. If the prestressing procedure is re-
stricted to the stretching of cable elements 4 and 5 only by equal
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

500 Optimization of Structural Systems

shortness then, it would be desirable to know the final force distribution

and nodal deflection under varying shortness magnitudes.

E
CN

E
CM

Figure 2. Cable structure example

To this end the loaded net was analysed for increasing shortness values
specified in elements 4 and 5. The resulting nodal deflection and force
distribution are depicted graphically by Figure 3a, curve 1, and Figure
3b respectively. Curve 1, in Figure 3a reveals that no significant re-
duction in the joint displacement can be achieved for shortnesses greater
than 0.004m . The minimum displacement possible is about 0.0021m.
Examination of the final force distribution results depicted by Figure 3b,
reveals that for shortness values less than 0.004m cable elements 1 and
2 become slack and for values greater than 0.004m element 3 remains
slack in the final configuration. This means that no shortness solution
exists which satisfies the non-slackening condition. For this criterion to
be satisfied other members need to be stretched together with members
4 and 5. The effect of stretching member 3 along with members 4 and 5
has been examined by re-analysing the cable structure subjected to
varying shortness magnitudes specified in elements 4 and 5 combined
with shortnesses of 5L* = 0.004m and 0.008m at element 3. Displacement
results are shown in Figure 3a, curves 2 and 3, and the final force dis-
tribution in the members are shown in Figure 3c and 3d.
If limits upon the joint displacement and maximum and minimum
member forces are imposed such that J,max < 0.00155m , 7%™ <i 25£iY and
Tfmm > IQkN then, a close examination of Figure 3 reveals that an opti-
mum shortness solution which simultaneously satisfies all of these goals
does not exist. However, the best possible solution whereby the degree
to which these goals are violated is a minimum may be predicted by the
use of Figure 3. The optimum solution lies in the region marked A in
Figure 3c, in which dt^, <5Lr4, and 6L* have values of approximately
0.004m
Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509
Optimization of Structural Systems 501

Curves:
1 A <5Lp3= 0.00
2 4- <5Lrj= 0.004 m
3 ^r 6L^j= 0.008 m

0.0

0.004 m

4 6 '8 10

0.008 m

A Cable element 1
Shortness in elements 4 <k 5 (mm) V Cable element 3
-|- Cable element 4
(d)

Figure 3. Force and displacement variation

with changes in shortness magnitudes
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

502 Optimization of Structural Systems

For the above simple example with only two design variables, it has
been possible to predict the optimum solution with reasonable accuracy.
This provides a means of checking the design result obtained using the
present optimization method.
Design calculations have been carried out for the above example.
A starting point dL° — 0.001 was used. An initial value of p — 10 was
subsequently increased until no further improvement to the solution was
obtained. The solutions obtained for the different p values are given in
Table 1. It can be seen from Table 1 that the solution is fairly insensitive
to changes in the p value. It is noted that the function Fp monotonically
decreases as p increases. Significant, improvement to the solution ceases
at p — 10* and the optimum design is in agreement with that predicted
from the above analysis.

7}m,x(AA") S,m>*(kN) F,
10' 4.0078 4.1594 9.910 28.149 0.15356 0.123024 0.255929
102 3.8491 4.0503 9.212 27.418 0.15356 0.097291 0.109929
103 3.8365 4.0366 9.140 27.380 0.15356 0.095194 0.096482
10< 3.8268 4.0379 9.126 27.310 0. 15356 0.096250 0.096332
10* 3.8262 _ 4.0377 9.122 27.298 0. 15356 0.096270 0.096283
106 3.8262 4.0377 9.122 27.298 0.15356 0.096270 0.096271
10? 3.8262 4.0377 9.122 27.298 0.15356 0.096270 0.096270

Table 1. Effect of p on the solution for example 1

A solution which satisfies all the target goals does not exist for this
particular example. A design which satisfies any two limit conditions
greatly violates the third. The optimum solution given in Table 1 is the
best compromise in which all the critical goals are exceeded by similar
proportions.
Example 2: Hyperbolic paraboloid
The hyperbolic paraboloid cable net, shown in Figure 4 and used by
Krishna [8] for analysis, has been used here as a design example for
which the optimum distribution of prestressing forces was determined.
The same model has been previously used as a design example by
Cinquini and Contro [9] and Simoes and Templeman [10]. However, the
extent to which comparison can be made is limited due to differences in
the specification of the design problems by the respective authors. The
extensional stiffnesses (EA) of the sagging and hogging cables are 293.6
MN and 197.5 MN respectively. Three loading cases were designed for
and are detailed in Refs. [2,8,9,10]. Load case l i s a vertical loading of
IQQkN at every joint in the net. Load cases II and III are non-symmetric
but turn out to be non-critical.
Symmetry of the cable layout implies that this net can have eight
different end cable element shortnesses 6L, (i = 1, ,8) as design variables
as shown in Figure 5a. Imposing a symmetrical shortness distribution in
Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 503

n 17

12 22
6
13 19 23 25 A
C
14 24

15 21

Plan

3.66m
3.66m

Elevation

Figure 4. Hyperbolic paraboloid

 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

504 Optimization of Structural Systems

this manner has the advantage of a reduced number of design variables

and consequently less computational effort. However, for problems in
which geometrical symmetry of the structure is not accompanied by
symmetrical layout of applied loading, imposing symmetrical shortness
distributions may not result in an optimum design. Since, for the present
problem, load cases II and III are not symmetric, the most logical number
of design variables to have is eleven, as shown in Figure 5b. Design
calculations have been carried out using both eight and eleven shortness
variables and the results from both calculations were found to be similar.
This is due to Load case I being the dominant load in the sense that it
produced all the active goals. Results of calculations with eight variables
only have been quoted for the purpose of the following discussion.

//Ol_R \
% »
/%@ 6k; rtiV
<5L2 6L0
/due 6L2 a&
<5U <5U (5Ur\
yffl-a A\A\ auX
<5U <3U 5U <5L? /
\
6L2
\&4(
<5L2 6L, siyr
<5 La/
\^L*
%
^
xy.y

a. 8 shortness variables b. 11 shortness variables

Figure 5. Shortness variables for example 2

To appraise the practical benefit arising from the proposed method,
it is appropriate to evaluate the final optimized design with respect to
an initial design. To this end an initial design in which a uniform pre-
stress distribution of about 800&N in all the outer cable elements, as
given in Krishna [8], is considered. The maximum and minimum cable
element forces and maximum nodal deflection resulting from this design
are tabulated in Table 2, column 1. The values in bracket are the
equivalent end cable prestress forces corresponding to the given
shortnesses in cm.
It would be desirable to improve this design in order to reduce the
maximum final forces in the cable elements and the maximum nodal
displacement, but ensuring at the same time that the forces do not fall
below a specified bound as a safety measure against slackening. In an
attempt to achieve this, three design problems, 1, 2, 3, corresponding to
different combination of permissible force and displacement target limits
were examined.
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 505

Problem 1
In this problem, the extreme force and displacement values,
7)max = 17406N, 7>nun = 318/eAT and J,max = 0.85m, resulting from the non-
linear analysis of the net structure when subjected to uniform prestress-
ing forces of SOOkN , were used as target constraints. Bounds on the
vertical displacements of joints 7, 12, 13, 14, and 19 were considered to
control the global deformability of the structure. The shortness vector
(<5L?) corresponding to the uniform prestressing force of SQQkN was con-
sidered to be the most appropriate starting point for the design iteration.
The final design shown in Table 2, column 2, was obtained after two
iterations of the design problem, with p = 10*. Further iterations with
updated starting points produced no further significant improvements.
It can be seen that the optimized shortnesses give a solution in which the
final maximum force and nodal displacement dre smaller and the mini-
mum force is larger than those of the uniform prestress design. Changes
in these values from the specified targets are all in the same proportion,
as indicated by the calculated goal values shovn in brackets.

Uniform prcst. Problem 1 Problem 2 Problem 3

Start
oLf 6 Li
point

6L,(cm) 3.70 (800) 3.916 (842) 3.480 (748) 6.877 (1474) 4.037 (903)
Jiz 7.45 (800) 6.469 (709) 0.663 (723) 11.088 (1183) 8.248 (923)
6Ls 11.14 (800) 10.870 (769) 10.010 (751) 14.563 (1071) 13.385 (978)
6L. 14.60 (800) 1280 (829) 13.760 (789) 18.374 (1074) 19.066 (1058)
6U 2.40 (800) 0.376 (434) 2.770 (896) 5.236 (1682) 3.348 (1119)
6L> 4.93 (800) 4.044 (705) 6.070 (924) 7.658 (1200) 5.203 (865)
6Li 7.43 (800) 7.510 (810) 7.440 (769) 9.786 (1049) 9.883 (1028)
6L* 9.76 (800) 9.329 (769) 8.640 (701) 13.990 (1160) 14.133 (1090)

1740 1705 (-0.0190) 1665(0.11) 1968 (-0.016) 1945 (-0.027)

318 324 (-0.0185) 305 (0.10) 594 (0.77) 583 (-0.403)
0.85 0.834 (-0.0191) 0.88(0.10) 0.787 (-0.016) 0.778 (-0.027)

-0.015869 -0.02735

Table 2. Hyperbolic net design results

Problem 2
For this problem, target values for the maximum and minimum forces
were 15006N and 335kN respectively, and for the maximum nodal dis-
placement was 0.8m. These targets are more stringent than those before,
i.e. cable elements may not be as highly or lightly loaded as in problem
1 and the joints may not displace to the same degree.
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

506 Optimization of Structural Systems

Table 2, column 3, gives the final design results obtained after two
iterations of the design problem initiated from starting point JL? (see
Table 2, column 1) again with p — 10* . It can be seen from these results
that the target bounds have not been achieved and have all been pro-
portionately violated as indicated by the goal values. However, the
overall objective, of reducing the maximum force and displacement and
increasing the minimum force, with respect to the uniform design, has
been partly met by the reduced final maximum force. However, this is
accompanied by an increase in nodal displacement and a reduction in
minimum cable force.
It is apparent from the above results that excessive reduction in the
cable forces lead to larger deflections. Therefore, to achieve a reduced
degree of displacement, it may be necessary to impose a more liberal
maximum force constraint. To this end the following problem was solved.

Problem 3
This problem had the least stringent maximum force limit among the
problems studied, T,nax was 20006 JV. T,,un and i),,^ were as for problem 2.
In addition to solving this problem using the usual starting point i)L?, the
problem was also solved using a second initial point c>L?. The shortness
vector values of JZ/% are shown in Table 3. Both calculation results are
tabulated in Table 2 for comparison purposes. As anticipated, two differ-
ent design solutions resulted corresponding to the two different starting
sets of shortnesses, as can be seen from Table 2. This reinforces the sub-
jective conclusion that the final design is dependent upon the position
of the initial trial point.

Starting
point <5Li(cm) <5L,2 6^3 6L+ 6Lt> 6L^ 5Li dLg
c5L? 3.70 7.45 11.14 14.60 2.40 4.93 7.43 9.76
<5L2 1.00 5.00 10.00 15.60 5.00 3.00 7.00 10.00

Table 3. Starting shortnesses for hyperbolic net optimization

In the context of the original aim of attempting to improve the results
with respect to the uniform prestress design, it can be seen that this ob-
jective has been met in reducing the displacement and increasing the
minimum force. This, however, results in an increased maximum cable
force. Viewing the results in the context of achieving the targets set for
this particular problem, it is clearly seen that both solutions are within
the specified bounds.
There are several points of interest to be noted from these solutions.
c5L? gives a better solution than dL\ as indicated by the lower Fp value.
Moreover, a relatively lower degree of prestress in the design with c5L?
achieves a smaller nodal displacement. This is contrary to what might
intuitively be expected to occur and gives some insight into the complex
and highly non-linear nature of cable net structures. The last point worth
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

Optimization of Structural Systems 507

noting is that the extreme force and displacement values fall within the
set bounds to disproportionate extents, the maximum force and joint dis-
placement constraints being the active goals, as indicated by the calcu-
lated goal values in Table 2. The solutions given in Table 2 for problems
1, 2, and 3 all correspond to different target values. They can all be in-
terpreted, therefore, as compromise (or Pareto optimal) solutions of the
design problem.

DISCUSSION AND CONCLUSION

The design examples studied in this paper and others presented in Ref.[2]
give rise to several discussion points about the present design method
and the behaviour of cable net structures. It has been shown that the
optimization of cable net structures is a particularly difficult problem.
A completely automatic optimum design program is currently not an
achievable goal. The use of optimization in a user-interactive mode has
been shown to result in significant prestress reductions.
The non-linear analysis of pretensioned cable net structures involves
considerable computation because the shape-finding problem must be
done iteratively. At least two shape-finding problems must be solved in
each analysis: first, to find the shape of the zero configuration under
pretensioning forces only; second, to find the shape for the final config-
uration for each applied load case. Because the non-linear analysis can
only be done numerically, it is not possible to get closed form algebraic
expressions for any of the functions needed in the optimization model.
Consequently the optimization must be based upon models which use
numerical function values and do not employ first derivative values.
The minimax formulation adopted in this work was found to be sat-
isfactory in that it allowed the simultaneous optimization and control of
the different engineering goals. The entropy-based approach to solving
the minimax optimization formulations proved to be very successful in
transforming the problem to a scalar optimization problem. The solution
of the scalar problem was found to be insensitive to changes in values
of the parameter p. A sufficiently high p value of about 10* was quite
adequate to ensure a smooth convergence to an optimum solution.
The examples solved in this paper provided considerable insight into
the behaviour of prestressed cable net structures. Joint displacements and
individual cable element forces were sometimes very sensitive to com-
paratively small changes in the initial pretensioning forces. It was diffi-
cult to predict which cable element might become slack or which joint
would have the largest displacement. Very different pretensioning oper-
ations produce very similar prestress force distribution. This is disturbing
from a design point of view. It suggests that many local optima may exist
in the design optimization problem. A good design solution depends on
a good choice of an initial starting point. This presents a major disad-
vantage in that time must be devoted to choosing the initial shortness
set. However, viewed in the context of an overall strategy which aims
at taking an initial design for a cable net structure and making succes-
sive improvements to that design until the rate of improvement becomes
 Transactions on the Built Environment vol 2, © 1993 WIT Press, www.witpress.com, ISSN 1743-3509

508 Optimization of Structural Systems

too small to warrant further computational effort, the present algorithm

was very successful.

REFERENCES

1. Buchholdt, H. A. Introduction to cable roof structures,

Cambridge University Press, London, 1985.

2. Sufian, F. M. Analysis and design methods for pretensioned cable

net structures, PhD Thesis, University of Liverpool, 1992.
3. Sufian, F. M. and Templeraan, A. B. "On the Non-linear analysis
of Cable Net Structures." Structural Engineering Revieu, Vol. 4,
No. 2, pp. 147-158, 1992.
4. Templeman, A. B. "Entropy-bared optimization methods for engi-
neering design." Advanced Techniques for Structural Optimization,
ed. S. Hernandez, Computational Mechanics, Southampton, 1993.
5. Li, X. "An entropy-based aggregate method for minimax opti-
mization." Engineering Optimization, Vol. 18, pp. 277-285, 1992.

6. Fletcher, R. and Reeves, C. M. "Function minimization by con-

jugate gradients." Computer J., Vol. 7, No. 2, pp. 149-154, 1964.
7. Nelder, J. A. and Mead, R. "A simplex method for function min-
imization." Computer J., Vol. 8, pp. 155-162, 1965.
8. Krishna, P. Cable suspended roofs, McGraw-Hill, New York, 1978.
9. Cinquini, C. and Contro, R. "Prestressing design method for
cable net structures." Engineering Structures, Vol. 7, pp.
183-139, 1985.
10. Simoes, L. M. and Templeman, A. B. "Entropy-based synthesis
of pretensioned cable net structures." Engineering Optimization,
Vol. 15, No. 2, pp. 121-140, 1980.