You are on page 1of 196

DEPARTMENT OF STRUCTURAL ENGINEERING

ROYAL INSTITUTE OF TECHNOLOGY

SE-100 44 STOCKHOLM

Numerical Analyses of Cable


Roof Structures
Gunnar Tibert

KUNGL TEKNISKA HÖGSKOLAN


INSTITUTIONEN FÖR BYGGKONSTRUKTION

TRITA-BKN. Bulletin 46, 1999


ISSN 1103-4270
ISRN KTH/BKN/EX--46--SE

Licentiate Thesis
Numerical Analyses
of
Cable Roof Structures

Gunnar Tibert

Department of Structural Engineering


Royal Institute of Technology
SE-100 44 Stockholm, Sweden

TRITA-BKN. Bulletin 46, 1999


ISSN 1103-4270
ISRN KTH/BKN/B--46--SE

Licentiate Thesis
Gunnar
c Tibert 1999
KTH, TS–Högskoletryckeriet, Stockholm 1999
Abstract

This thesis deals with the techniques used in the numerical analysis of cable roof
structures. These structures are usually very light and flexible and require analysis
methods, which take their non-linear behaviour into account.
An extensive literature survey, concerned with both practical and theoretical aspects
of cable roofs, is presented. Some aspects included are: structural systems, roof
erection procedures, different cable types and their properties, structural details,
roof loads and analysis methods.
As the initial shape of a cable roof depends on the internal force distribution, it
cannot be described by simple geometrical models. Special iterative methods, usu-
ally not familiar to the structural engineer, have to be utilised in order to find the
pretensioned configuration of the roof. The simple force density method is presented
in detail and applied to a number of different types of cable roof structures. The
method worked well for structures composed of only cables, but not for structures
with compression members.
Three analytical finite cable elements are presented. Two elements are mathemat-
ically exact and can accurately model both taut and slack cables using only one
element per cable. It is shown that the analytical elements are advantageous in
modelling cable behaviour.
A static analysis of the Scandinavium Arena in Gothenburg has been performed.
The results from this analysis were compared with results from the original design
of the same object. It was found that the bending moments in the supporting
structure—the concrete ring beam—were very sensitive to its shape. This explained
the large discrepancy in the bending moment distribution between the analyses.
Results from a simplified method, used for preliminary calculations, agreed well
with those of the more accurate finite element calculations, for a studied symmetric
load case.
Failure stage analysis of the class of self-stressed cable structures called tensegrity
structures has been identified as an area of further research.

Keywords: cable roof structures, loads, form-finding, force density method, finite
cable elements, static analysis, the Scandinavium Arena.

iii
Preface

The research work in this thesis was carried out at the Department of Structural
Engineering, Structural Mechanics Group, at the Royal Institute of Technology in
Stockholm, under the supervision of Professor Anders Eriksson. The work reported
in this thesis was financed through a personal grant from KTH.
First of all, I express my gratitude to my supervisor Professor Anders Eriksson for
his scientific guidance and valuable advice.
I also thank Docent Costin Pacoste for help with the selection of a suitable beam
element for the static analyses.
I would also like to thank Professor Emeritus Alf Samuelsson at Chalmers Uni-
versity of Technology in Gothenburg and Mr. Nils Dahlstedt, Technical Manager
at the Scandinavium Arena in Gothenburg, for the valuable information about the
Scandinavium Arena.
Finally, I am grateful to all people at the Department of Structural Engineering that
have helped me in the work with this thesis.

Stockholm, April 1999

Gunnar Tibert

v
Contents

Abstract iii

Preface v

List of symbols xi

1 Introduction 1
1.1 Aims and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 General structure of thesis . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Literature review 5
2.1 Historical review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Structural systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Simply suspended cable structures . . . . . . . . . . . . . . . 10
2.2.2 Pretensioned cable trusses . . . . . . . . . . . . . . . . . . . . 10
2.2.3 Pretensioned cable net structures . . . . . . . . . . . . . . . . 12
2.2.4 Tensegrity systems . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Roof erection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.2 Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.3 Axial stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.4 Corrosion protection . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Cladding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5.1 Fabrics and foils . . . . . . . . . . . . . . . . . . . . . . . . . . 21

vii
2.5.2 Metal sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.3 Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Structural details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.1 End fittings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.2 Intermediate fittings . . . . . . . . . . . . . . . . . . . . . . . 25
2.6.3 Saddles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.4 Anchorages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Roof loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7.1 Wind load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.2 Snow load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7.3 Earthquake load . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.7.4 Other loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.8 Analysis methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3 The initial equilibrium problem 41


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.1 Physical modelling . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Literature review of initial equilibrium solution methods . . . . . . . 42
3.2.1 The non-linear displacement method . . . . . . . . . . . . . . 44
3.2.2 The grid method . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.3 The force density method . . . . . . . . . . . . . . . . . . . . 49
3.2.4 Least squares stress determination methods . . . . . . . . . . 51
3.2.5 A combined approach . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.6 Initial equilibrium of tensegrity structures . . . . . . . . . . . 53
3.3 The force density method . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.1 The linear force density method . . . . . . . . . . . . . . . . . 56
3.3.2 The non-linear force density method . . . . . . . . . . . . . . 61
3.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4.1 Smaller cable nets . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4.2 A large cable net . . . . . . . . . . . . . . . . . . . . . . . . . 73

viii
3.4.3 Cooling towers . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.4 A structure composed of both cables and struts . . . . . . . . 79
3.4.5 Cable dome . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4.6 Tensegrity structures . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4 Finite cable elements 85


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2 Analytical cable solutions . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2.1 The inextensible catenary . . . . . . . . . . . . . . . . . . . . 87
4.2.2 The elastic catenary . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.3 Effect of cable bending stiffness . . . . . . . . . . . . . . . . . 92
4.3 Literature review of cable elements . . . . . . . . . . . . . . . . . . . 95
4.3.1 Elements based on polynomial interpolation functions . . . . . 95
4.3.2 Elements based on analytical functions . . . . . . . . . . . . . 97
4.4 Straight and parabolic elements . . . . . . . . . . . . . . . . . . . . . 99
4.4.1 Straight bar element . . . . . . . . . . . . . . . . . . . . . . . 99
4.4.2 Elastic parabolic element . . . . . . . . . . . . . . . . . . . . . 101
4.5 Catenary elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.5.1 Elastic catenary element . . . . . . . . . . . . . . . . . . . . . 105
4.5.2 Associate catenary element . . . . . . . . . . . . . . . . . . . . 107
4.5.3 Convergence of solution . . . . . . . . . . . . . . . . . . . . . 111
4.6 Comparison of elements . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.6.1 Comparison example 1 . . . . . . . . . . . . . . . . . . . . . . 113
4.6.2 Comparison example 2 . . . . . . . . . . . . . . . . . . . . . . 114
4.6.3 Comparison example 3 . . . . . . . . . . . . . . . . . . . . . . 115
4.6.4 Conclusions from the comparisons . . . . . . . . . . . . . . . . 117

5 Static analysis 123


5.1 Static analysis of the Scandinavium Arena . . . . . . . . . . . . . . . 123

ix
5.1.1 The Scandinavium Arena—background . . . . . . . . . . . . . 123
5.1.2 Prestressing forces . . . . . . . . . . . . . . . . . . . . . . . . 126
5.1.3 Finite element model . . . . . . . . . . . . . . . . . . . . . . . 130
5.1.4 Calculation results . . . . . . . . . . . . . . . . . . . . . . . . 135
5.1.5 Calculation results from 1972 . . . . . . . . . . . . . . . . . . 138
5.1.6 Comparison of the results . . . . . . . . . . . . . . . . . . . . 142
5.2 Sensitivity of bending moment to the shape of the ring beam . . . . . 142
5.2.1 Description of the structure . . . . . . . . . . . . . . . . . . . 142
5.2.2 Different shapes of the ring beam . . . . . . . . . . . . . . . . 144
5.2.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . 145
5.3 Comparison with a simplified method . . . . . . . . . . . . . . . . . . 154
5.3.1 Results and discussion . . . . . . . . . . . . . . . . . . . . . . 154

6 Conclusions and further research 157


6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.1.1 The initial equilibrium problem . . . . . . . . . . . . . . . . . 157
6.1.2 Finite cable elements . . . . . . . . . . . . . . . . . . . . . . . 158
6.1.3 Static analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.2 Further research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.2.1 Failure analysis—background . . . . . . . . . . . . . . . . . . 159
6.2.2 Failure analysis—further research . . . . . . . . . . . . . . . . 163

Bibliography 165

A Numerical data for the Scandinavium Arena 175

x
List of symbols

The following is a list of the most important symbols that appear in the chapters
of the thesis. Symbols not included in this list are defined when they first appear.
The number refer to the page where the symbol first appear.

A cross-sectional area, 45
A0 cross-sectional area of core wire, 19
Ai cross-sectional area of a wire in layer i, 19
A equilibrium matrix, 52
B compatibility matrix, 54
C length of cable chord, 102
Cp pressure coefficient, 30
C connectivity matrix for free nodes, 56
Cf connectivity matrix for fixed nodes, 56
Cs connectivity matrix for all nodes, 56
d vector of nodal displacements, 54
E Young’s modulus, 45
E0 Young’s modulus of core wire, 19
Ei Young’s modulus of wires in layer i, 19
e vector of bar elongations, 54
F force in global coordinate system, 99
F component of cable force in local coordinate system, 87
f vector of nodal loads, 52
H horizontal component of the cable force T , 88
h projection of cable profile on z  -axis, 87
I moment of inertia, 87
Ii moment of inertia of wire in layer (around its own centerline), 92
K tangent stiffness matrix in global coordinate system, 101
K tangent stiffness matrix in local coordinate system, 102
KE elastic stiffness matrix in local coordinate system, 100
KG geometric stiffness matrix in global coordinate system, 50
L length, 43
L0 unstrained length, 45
l projection of cable profile on x -axis, 87
mi number of wires in layer i, 19
n number of wire layers, 19
p wind pressure at time t, 30
q0 intensity of distributed load on cable, 87

xi
q vector of force densities, 58
Ri wire radius in layer i, 19
ri radius of wire centerline helix in layer i, 19
s arc length (elastic cable), 87
s0 arc length (inextensible cable), 88
T cable force, 43
Tb cable force at the base (s0 = 0), 91
T transformation matrix, 100
t vector of bar axial forces, 52
U mean wind velocity, 29
u turbulence component of the wind field in the x-direction, 29
u vector of free x-coordinate differences, 58
V total wind velocity, 30
v turbulence component of the wind field in the y-direction, 29
v vector of free y-coordinate differences, 58
w turbulence component of the wind field in the z-direction, 29
w vector of free z-coordinate differences, 58
xg vector of nodal coordinates, 50
x vector of free x-coordinates, 58
xf vector of fixed x-coordinates, 58
y vector of free y-coordinates, 58
yf vector of fixed y-coordinates, 58
z vector of free z-coordinates, 58
zf vector of fixed z-coordinates, 58
αi angle of wire centerline helix in layer i, 19
β angle between the cable chord and horizontal, 102
ν Poisson’s ratio, 92
φ angle between x - and x-axis, 103
ρ air density, 30
θ angle between tangent to cable profile and x -axis, 87
θm mean wind direction, 30
θv azimuth angle of turbulent wind component v, 30
θw elevation angle of turbulent wind component w, 30

xii
Chapter 1

Introduction

Tensile architecture represents the new trend in design: construction with the mini-
mum amount of material. As is well-known, the primary advantage of tensile mem-
bers over compression members is that they can be as light as the tensile strength
permits. With new materials, such as high strength steel cables and silicone-coated
glass fibre membranes, larger distances can be spanned using the same amount of
material as before.
Tensile structures have always fascinated architects and engineers, mainly because
of the aesthetic shapes they produce. Despite this, very few tensile structures have
been built. Why are they not more common, if they are both economic and beauti-
ful? One answers might be that tent-like structures have always been thought of as
temporary. Although, a probably more correct answer is that they are more difficult
to analyse and construct than traditional buildings. From a structural viewpoint,
tension structures have several special features, such as light weight and flexibility.
These features require special care in the design; for example, an error in the distri-
bution of the pretensioning forces may lead to damage of the cladding under large
loads.
If the numerical analysis of building structures is concerned, the finite element
method is the dominating tool. In this method, the structural characteristics and
external loads are described by matrices and vectors. The sought parameters, e.g.
displacements and internal forces, are found by matrix operations.
The first step in the analysis process is the definition of the geometry of the struc-
ture, which generally is known a priori. However, this is not the case for tensile
structures. Due to the negligible flexural stiffness of cables and membranes, the
initial configuration of these structures must be stressed, even if the self-weight is
disregarded. Thus, before the analysis of the behaviour of the structure to external
loads can be performed, the initial equilibrium configuration must be found. The
shape of a tensile structure, which very much depends on the internal forces, also
governs the load-bearing capacity of the structure. Therefore, the process of deter-
mining the initial equilibrium configuration calls for the designer’s ability to find
an optimum compromise between shape, load capacity and constructional require-
ments. Several numerical methods, applicable to the initial equilibrium problem,

1
CHAPTER 1. INTRODUCTION

can be found in literature. Most of these methods are not included in general finite
element programs, e.g. ABAQUS 1 and are not familiar to the practising structural
engineer.
After the initial reference configuration has been determined, the structural members
have to be described by stiffness matrices and force vectors. Special elements for
cables or chains are often not available in commercial finite element programs. The
single cable is instead modelled by one or several other elements depending on
the sag-to-span ratio. Nevertheless, this approach has problems such as numerical
instability of the solution algorithms. To avoid these problems it is desirable to have
at hand a robust element which accurately describes the behaviour of both taut and
slack cables. Since cable structures in general are very flexible, a geometrically non-
linear solution method has to be used. The most common is the Newton-Raphson
algorithm, embedded in more or less sophisticated load incrementation techniques.
The final step in the analysis process is to define the external loads on the structure.
For civil engineering structures there are a number of loads that must be considered:
self-weight, vehicles, wind, rain, snow, ice, earthquakes, temperature, etc. The
magnitude and distribution of these loads is a constant source of research. The
present knowledge in the area is found in the national building codes, which aid
the engineers in their decisions. Tensile structures often have irregular shapes and
low self-weights which may give rise to unforeseen effects such as very high snow
loads and flutter instability due to wind. To ensure the safety of the structure,
experimental tests have to be undertaken together with statistical analyses to find
the magnitudes of the snow and wind loads.
Even with the right tools, the design of tensile structures will not be straightforward.
Each new roof type has its own features. It is no surprise that experience and
good engineering judgement are frequent characteristics among famous designers
of tensile structures: Fritz Leonhardt, Jörg Schlaich, Frei Otto, Horst Berger and
David Geiger, to mention a few.

1.1 Aims and scope


The aim of this work is to study the mechanical aspects of cable supported shell type
structures (roofs, cooling towers, etc.). The first part of the work is concerned with
the basic aspects of these specific types of structures. These include: the principal
arrangements of the cables, pretensioning schemes necessary to obtain a prescribed
shape, and the practical aspects of connections between and supports for the cables.
Further, basic computational models are to be studied. These include, but are not
limited to, the methods for analysis and force distribution. The second part is
concerned with the formulation of suitable finite elements, which take into account
the non-linear behaviour of a cable. Basic analyses are performed, and verified.
Theoretical and numerical studies are included in this thesis, but no experimental
1
ABAQUS is a registered trademark of Hibbitt, Karlsson & Sorensen, Inc., 1080 Main Street,
Pawtucket, RI 02860-4847, U.S.A. Internet: http://www.abaqus.com.

2
1.2. GENERAL STRUCTURE OF THESIS

work is conducted. In the discussed methods, only elastic structures and static loads
are considered.
All of the numerical calculations in this thesis has been coded in the Matlab 2 lan-
guage. Some expressions have been derived using the computer algebra package
Maple 3

1.2 General structure of thesis


To get an overview of the structure of this thesis, the contents of the chapters are
presented below.
In Chapter 2, an extensive literature study on cable roof structures is presented.
The study includes both practical and theoretical aspects of cable roofs. Among
the practical aspects are: different structural systems, roof erection processes, dif-
ferent types of cables and their properties, roofing materials, and structural details.
Different types of loads and their effect on cable roofs are also presented. Finally,
methods used to analyse the behaviour of cable roofs under loads are reviewed.
In Chapter 3, a review of the numerical methods used to find the initial equilibrium
configuration of cable structures and structures of mixed type (cables and stiff struc-
tural members) are presented. One of the methods—the force density method—is
further described in detail and coded. A variety of examples are analysed to illus-
trate both the advantages and the drawbacks of the force density method.
In Chapter 4, the difficulties of modelling cable behaviour using finite elements based
on the conventional approach, i.e. using shape functions, are discussed. Further, four
finite cable elements are presented: the straight bar, the parabolic cable, the elastic
catenary and the associate catenary. The internal force vectors and tangent stiffness
matrices are presented and the elements are compared by some simple examples.
In Chapter 5, an existing cable roof structure is analysed by a finite element program
written by the author. The structure is the Scandinavium Arena in Gothenburg,
which consists of a pretensioned cable net anchored in a nearly circular concrete
ring beam. The results of the calculations are compared to the results from the
initial design process and the reasons for discrepancies in the results are discussed.
In addition, results from a simplified method, mainly used in preliminary design,
are compared to the results from the finite element calculations.
In Chapter 6, the conclusions of this study are stated and directions for further
research are suggested.
In Appendix A, data used in the analysis of the Scandinavium Arena are presented.

2
Matlab is a registered trademark of The MathWorks Inc., 24 Prime Park Way, Natick, MA
01760-1500, U.S.A. Internet: http://www.mathworks.com.
3
Maple is registered trademark of Waterloo Maple Inc., 57 Erb Street W., Waterloo, Ontario,
Canada N2L 5J2. Internet: http://www.maplesoft.com.

3
Chapter 2

Literature review

2.1 Historical review

The first structures regarded as cable roofs are four pavilions with hanging roofs
built by the Russian engineer V. G. Shookhov at an exhibition in Nizjny-Novgorod
in 1896. During the 1930’s a small number of roof structures of moderate sizes were
built in the U.S.A. and Europe, but none of major importance [88].
A big step in the development of suspended roofs came in 1950 when Matthew
Nowicki designed the State Fair Arena, Figure 2.1, at Raleigh, North Carolina,
USA. Sadly, Nowicki died that same year in a plane crash, but his work continued
through the architect William Henry Deitrick and civil engineer Fred Severud and
in 1953 the arena was completed [88].

(a) (b)

Figure 2.1: The State Fair Arena at Raleigh, North Carolina, U.S.A., (a) Repro-
duced from [10], (b) Structural system, reproduced from [16].

On an exchange visit to the U.S.A. in 1950 a German student in architecture, named


Frei Otto, previewed the drawings for the Raleigh Arena in the New York office of
Fred Severud. Otto saw that the project embodied many of his own ideas about how

5
CHAPTER 2. LITERATURE REVIEW

to construct with minimal amount of material. After graduation in 1952 Otto began
a systematic investigation of suspended roofs. The investigation was presented in
the doctoral thesis Das Hängende Dach (The Suspended Roof), which was the first
comprehensive documentation on the subject [31].
The thesis caught the attention of Peter Stromeyer of Stromeyer Company, one of
the largest tent manufacturers in the world. Stromeyer contacted Otto and they
began a fruitful cooperation. In 1957 Otto formed the Development Centre for
Lightweight Construction in Berlin in order to further increase the research about
tensile architecture. In 1964 he incorporated the centre into the Institute of Light
Surface Structures at the University of Stuttgart. A massive research work was un-
dertaken at the two institutes during 1957–1965 and published in Tensile Structures
(two volumes) [31, 124].
Frei Otto is considered by many to be responsible for the development of modern
tensile architecture. He was involved in the construction of many of the large tensile
structures during the mid 1960’s to early 1970’s. Among these was the first large
cable net structure with fabric cladding, the German pavilion at the World’s fair in
Montreal 1967 [10], Figure 2.2.

Figure 2.2: The German pavilion at the World’s fair in Montreal 1967. Reproduced
from [10].

Another pioneering structure at this time was the large low-profile super elliptic
air-supported roof, Figure 2.3, with a membrane attached to a diagonal cable net.
This structure was designed by David Geiger for the United States pavilion at the
World’s fair in Osaka 1970 [10].

6
2.1. HISTORICAL REVIEW

Figure 2.3: David Geiger’s air-supported roof at the World’s fair in Osaka 1970.
Reproduced from [10].

Following the success of the cable net in Montreal, Frei Otto produced a very elegant
development of the Montreal design for the Olympic Stadium in Munich 1972 [87],
Figure 2.4.

Figure 2.4: The Olympic Stadium in Munich. Reproduced from [31].

After the Osaka dome, several air-supported domes were built around the world,
because they provided the economically best alternative to span large distances.
However, several of them deflated due to heavy snow loads or compressor failure.
To overcome the deflation problems, David Geiger invented another structure 1986—
the cable dome. The cable dome concept was inspired by the tensegrity principle by

7
CHAPTER 2. LITERATURE REVIEW

Kenneth Snelson and Richard Buckminster Fuller. The first two domes were built
for the 1988 Seoul Olympics. The latest and biggest, the Georgia Dome, was built
in Atlanta 1994, Figure 2.5.

Figure 2.5: The Georgia Dome in Atlanta, U.S.A., during construction. Reproduced
from [10].

In the year 2000, the Millennium Experience will be held in Greenwich, London,
close to the Greenwich meridian. This exhibition will be held inside the largest
dome ever. The diameter of the dome is 364 m and the height is 50 m [70].

Figure 2.6: The Millennium Dome in London, during construction. Reproduced


from the cover of Bautechnik, Vol. 75, No. 11, 1998.

8
2.2. STRUCTURAL SYSTEMS

2.2 Structural systems

In this section, the traditional cable roof systems are presented together with a fairly
new one. Each roof type is presented very briefly, but references to more information
are given.
Cable roofs can be divided into different categories depending upon the criterion
used for classification. In accordance with how the cables are used, they can be
classified as [57]:

1. cable supported roofs, and

2. cable suspended roofs.

Cable supported roofs are, in principle, similar to cable-stayed bridges. In these


roofs, the cables only provide additional support for elements which themselves
carry a major part of the load. In cable suspended roofs the load is carried directly
by the cable system [57]. The cable supported roofs, for which the cables only have
an auxiliary function, will not be considered in this thesis.
The cable suspended roofs may be divided into the following categories [16]:

1. simply suspended cables,

2. pretensioned cable trusses, and

3. pretensioned cable nets.

Further, the pretensioned cable structures may be either self-balancing or non-self-


balancing. In a self-balancing structure, the forces in the cables are balanced inter-
nally in the supporting structure, e.g. a ring beam. In a non-self-balancing structure,
the cable forces are resisted by ground anchors [16].
In general, the stiffness of a pretensioned cable structure depends on [16]:

• the curvature of the cable,

• the cross-sectional areas of the cables,

• the level of pretension, and

• the stiffness of the supporting structure.

The cladding will not, unless it is in the form of a concrete shell, significantly increase
the stiffness of a roof. In the following, the traditional types of cable suspended roofs
will be described. In each category, the structural systems are illustrated by a limited
number of figures. More examples can be found in the references 16 and 57.

9
CHAPTER 2. LITERATURE REVIEW

2.2.1 Simply suspended cable structures

The first type is the simply suspended roof. These roofs have a single curvature or
a positive double curvature (like a bowl). Systems of this type have no stiffness. To
reduce the displacements caused by any form of applied loading, the roof cladding
must either be very heavy or stiff. Concrete is perhaps, therefore, the most suitable
roofing material; both prefabricated slabs and in situ cast concrete are used [16].
One can compare this roof type to a suspension bridge which is stiffened by the
bridge deck.

Figure 2.7: Simply suspended roof.

The simply suspended roofs, which are stiffened by the cladding material, will not
be considered in this thesis; only systems, which can be pretensioned before the
cladding is applied, will be analysed.

2.2.2 Pretensioned cable trusses

Lighter and stiffer systems than the simply suspended systems can be achieved if a
second set of cables with reverse curvature is connected to the hanging cables. A ca-
ble truss is quite stiff if it is tensioned to a level which ensures that both the hanging
and the bracing cables remain in tension under any load case. The basic cable truss
configurations with vertical connecting elements are shown in Figure 2.8. Another
system is the cable truss with diagonal ties, Figure 2.9, developed by the Swedish
engineer David Jawerth. Generally, the cable trusses with the vertical connecting
elements are structural mechanisms if they are considered as pin-jointed trusses.
However, the cable truss with diagonal connecting elements is statically indetermi-
nate [76]. Therefore, the Jawerth truss is stiffer than the other trusses [51]. The
cable trusses may be arranged in parallel planes, Figure 2.8, or radially, Figure 2.10.
A parallel Jawerth system was used in the Johanneshov Ice Stadium in Stockholm,
Sweden. An extensive study of cable trusses is presented in reference 76.

10
2.2. STRUCTURAL SYSTEMS

(a) Convex cable truss structure with corrugated metal roof decking

(b) Concave cable truss structure with corrugated metal roof decking

(c) Convex-concave cable truss structure with corrugated metal roof


decking

Figure 2.8: Cable trusses. Redrawn from [16].

11
CHAPTER 2. LITERATURE REVIEW

Figure 2.9: Cable truss system developed by the Swedish engineer David Jawerth.
Redrawn from [16].

Figure 2.10: Radial cable truss structure—Lev Zetlin’s cable roof over the audito-
rium in the city of Utica, U.S.A. Reproduced from [10].

2.2.3 Pretensioned cable net structures

The third type of cable roof structures is that in which the hanging and bracing
(pretensioning) cables all lie in one surface and form a net. To be pretensioned, this
surface must be anticlastic (saddle-shaped) at every point [16].
The stiffness of a cable net depends mainly on: the curvature of the net surface
and the level of pretension. In order to minimise the material in both the net and
the supporting structure it is advantageous to have a surface with a relatively small

12
2.2. STRUCTURAL SYSTEMS

radius [31]. The prestressing force must not be exceeded by any type of loading, or
the cables become slack. Areas of slack cables may damage the cladding or give rise
to the destructive phenomenon of flutter [16].
Cable nets can be designed with masts and edge cables or with stiff boundaries such
as beams, arches and rings, Figure 2.11. The first type is generally less stiff and
more complicated to construct than the latter ones. The cladding is often placed
directly on the cable network [16, 57].

Figure 2.11: A cable net structure—the Scandinavium Arena in Gothenburg, Swe-


den.

More information on different types of cable nets and their properties can be found
in references 16 and 76.

2.2.4 Tensegrity systems

A pure tensegrity structure is a structure composed of a relatively few non-touching,


straight compression members which are suspended in a net of tension members. The
key feature of such structures is that they are self-stressed; no external devices to
equilibrate the cable forces are needed. Tensegrity structures can be said to have
been invented by Kenneth Snelson and Richard Buckminster Fuller [99].
Several new systems, based on the tensegrity principle, have been developed in recent
years. The most well-known of these new systems is the cable dome concept by
David Geiger. The cable dome, Figure 2.12, is not a pure tensegrity structure since
a curved ring beam is used to balance the cable forces. The cable dome concept was
developed as an economically equal alternative to air-supported structures, which
several times have deflated due to mechanical failure or excessive snow loads. Today,
at least eight cable domes exist, but more will surely be build.

13
CHAPTER 2. LITERATURE REVIEW

Figure 2.12: The cable dome by David Geiger. Redrawn from [99].

More about tensegrity structures and some other fairly new structural concepts can
be found in reference 99. Tensegrity structures are further discussed in Chapter 3.

2.3 Roof erection

Theoretically, cable and membrane structures can be given infinitely many different
shapes. In practice, the number of configurations is restricted, as shown in the
previous section. Of course, with the use of scaffolding, more shapes would be
possible, but this eliminates some of the benefits with cable structures. Generally,
cable roof construction has two advantages over other forms of roof construction:
very little or no scaffolding is required, and quite rapid erection process. However,
these advantages do not indicate that the erection of a cable structure is an easy
task. Every step of the erection process must be computer controlled to avoid over-
stressing of the supporting structures. It is important that the contractor responsible
for the roof erection fully understands and exactly follows the erection plan specified
by the designer [57].
Cable trusses have the easiest erection process among cable roof structures and may
be assembled in the air or on the ground, of which the latter is to prefer. After
being assembled on the ground the truss is hoisted into position and prestressed
by applying tension at both ends simultaneously. Depending upon whether the
centre of the truss needs to be lifted or lowered, tension is applied to the suspension
or prestressing cable. Since the cable trusses usually do not interact with each
other before the cladding is applied several trusses can be erected and prestressed
simultaneously to reduce overall construction time [16]. Double layer grids with
radial symmetry can be erected in the same fashion as trusses, but care has to be
taken to not over-stress the compression ring as bending moments are introduced
when just a few trusses are tensioned. An erection scheme for radial double layer
grids is given in [57].

14
2.4. CABLES

Cable nets may be preassembled on the ground and hoisted into position or as-
sembled in the air. Nets with flexible boundaries (i.e. edge cables) are usually pre-
assembled on the ground, but for nets with stiff boundaries either of the methods
can be used [16]. With use of computational methods (see Chapter 3) the shape
of a net and the corresponding cable forces can be very accurately determined. To
obtain the computed shape of the real net a high dimensional accuracy in fabrication
is required. Small errors in unstrained length may cause large errors in force. One
method to achieve a high accuracy at a minimal cost is to specify a net with a square
unstrained mesh and uniform cable stresses. In this way, the same cable dimension
can be used for the whole net (not the edges) and the equidistant cable-to-cable
connections can be factory-assembled. But, even with a high accuracy some ad-
justment can be necessary after the net has been lifted into its final position. This
adjustment is possible if tensioning devices (turnbuckles) are incorporated at the
ends of the cables [63].
For tensegrity structures, suitable methods for prestressing large tensegrity frame-
works have not yet been developed. This is probably the main reason for the very
few large tensegrity structures today. Nonetheless, one exception is the cable domes
by David Geiger. These domes were developed as an economically equal alternative
to air-supported structures, but without the risk for deflation. From the economic
point of view, it was necessary that the domes could be constructed without any
scaffolding. Figure 2.13 shows the steps of erection of a cable dome [99]. For a
complicated structure, the best way to plan the erection steps is to build a physical
model of the structure [99].

2.4 Cables

The main load carrying element in the structures considered in this thesis is the
cable. In structural applications, the term ‘cable’ means a flexible tension member.
However, a cable can have different configurations. In this section, the different
types of cables and their characteristics will be examined.

2.4.1 Products

The smallest single tension element in a cable is the steel wire. It is usually circular
in cross section, with a diameter between 3 and 8 mm, but may be non-circular in
locked coil strands. The wire has a high tensile strength that is obtained by cold
drawing or cold rolling [35].
A spiral strand, Figure 2.15(a), is an assembly of wires laid helically around a central
straight wire. An assembly of a small number of wires is called a spiral strand and
if there are more than three layers it is called a spiral bridge strand. The successive
layers are usually wound in opposite directions to get equal torsional stiffness in
both directions [35].

15
CHAPTER 2. LITERATURE REVIEW

(a)

(b)

(c)

(d)

(e)

Figure 2.13: Cable dome erection steps: (a) The upper cables are hung, then (b) a
hoop and struts are hung, raising the inverted ridge cables. More hoops
and struts, (c)–(e), further raise and tension the ridge cables. Drawn
from data given in [39].

16
2.4. CABLES

Figure 2.14: A wire rope and its parts. Reproduced from [29].

Locked coil strands, Figure 2.15(b), are similar to spiral strands but are composed
of two types of helically laid wires: the core is a spiral strand with helically laid
circular wires, and at least the two outer layers have wires with a special Z-shape
that interlock with each other. The special shaped wires together with the self-
compacting effect of the helical arrangement result in a tight surface and a low void
ratio in the outer layers [42].
A wire rope, Figure 2.15(c), is an assembly of spiral strands that are laid helically
around a central core that can be a strand or another independent wire rope. The
spiral strands are usually laid in the opposite direction to the wires in the spiral
strands (ordinary lay) but can be laid in the other direction (Lang’s lay) [35].
The helical lay of wires increases the flexibility of the cable, but reduces the strength
and stiffness. In some applications, particularly suspended bridges, a high strength
and stiffness are more important than flexibility and therefore products with parallel
wires and strands have become popular. Other benefits with parallel strands and
wires are easier handling and transportation. In the last decade, parallel strands
have also found use in roof construction. Parallel strand systems were used as the
hoop and ridge cables in the cable domes by David Geiger [100]. The development
of parallel products over recent years is reviewed by Walton [125].

17
CHAPTER 2. LITERATURE REVIEW

(a) Bridge strand (b) Locked coil bridge (c) Wire rope
strand

Figure 2.15: Cable cross sections. Reproduced from [16].

2.4.2 Strength

For the wires commonly used in cables the guaranteed minimum tensile strength
is 1570 MPa and the guaranteed 0.2 % proof stress is 1180 MPa. The limit of
proportionality (0.01 % proof stress), which is the absolute upper limit for the
stresses in the service condition, has a value of 65–70 % of the tensile strength.
When deciding the allowable stress level, the effect of relaxation must also be taken
into account. Tests on steel wires show that the relaxation accelerates when the wire
is held under a permanent stress larger than 50 % of the tensile strength. Therefore,
the stresses from permanent loads should not exceed 45 % of the tensile strength [42].

2.4.3 Axial stiffness

For structural applications, the perhaps most important property of the cable, be-
sides the tensile strength, is the axial stiffness. As mentioned above, a cable with
helical wires has a lower stiffness than a cable with straight wires. In the design of
cable structures, it is of cardinal importance to know the axial stiffness of the cables
since the force distribution in, for example, a cable net is very sensitive to small
errors in the cable properties (modulus and length). Several methods have been
developed to calculate the axial stiffness of a helically wound cable, see for example
reference 22. Most of these methods are based on contact theories and are, thus,
very complex. Nevertheless, two simple and accurate methods have been found and
will be presented in this section. For explanation of the notations see Figure 2.16.

18
2.4. CABLES

Figure 2.16: Geometry of a helically wound cable. Reproduced from [58].

In reference 58, Kumar and Cochran linearised the equations from Costello [29] and
arrived at the following closed-form expression for the axial stiffness:

n
 
(AE)eq = A0 E0 + mi Ai Ei sin αi 1 − (1 + ν)pi cos2 αi , (2.1)
i=1

where
Ai = πRi2 , (2.2)
and     
Ri Ri2 ν
pi = 1 − ν cos αi 1 − 2 1 −
2 2
cos 2αi cos αi . (2.3)
ri 4ri 1+ν
Kumar and Cochran [58] also provide an even simpler expression for the equivalent
axial stiffness
n

(AE)eq = A0 E0 + mi Ai Ei sin3 αi 1 − ν cot2 αi . (2.4)
i=1

Another method, in which the wire layers are modelled as orthotropic sheets, has
been developed by Raoof [98]. The method is quite cumbersome and not suitable
for practical design work. Therefore, Raoof derived a simplified procedure, by para-
metric studies of different cable dimensions. In that, Hruska’s1 parameter is first
computed as:
n
mi Ai
κ= cos4 αi , (2.5)
i=1
A T

in which

n
AT = mi Ai . (2.6)
i=1
1
From F. H. Hruska, Calculation of stresses in wire ropes, Wire, Vol. 26, No. 9, 1951.

19
CHAPTER 2. LITERATURE REVIEW

The relation between the full-slip modulus (no friction between wires) and steel
modulus is computed as:
Efull-slip
= −0.26442 − 2.004046κ + 6.5735κ 2 − 3.3068κ 3 , (2.7)
Es
Denoting Efull−slip /Es = ϑ, the no-slip modulus is found from
Eno-slip
= 3.998 − 7.916ϑ + 7.238ϑ2 − 2.321ϑ3 . (2.8)
Efull-slip
The full-slip and no-slip axial stiffnesses are obtained by multiplying Efull-slip and
Eno-slip , respectively, with AT . The method by Raoof, equations (2.5)–(2.8), are
included in Eurocode 3 [35]. Raoof’s method has been checked against experimental
results in [47]. It was found that the experimental moduli of newly manufactured
cables agreed well with the theoretical full-slip modulus.
The methods presented above have also been compared to other analytical methods
and it is concluded that the overall elastic behaviour of helical cables under axial
loading is well represented by the available mechanical models. Which model one
should use is dependent on the size of the cable [22]. The expressions by Kumar
and Cochran is expected to yield higher accuracy for cables with few layers of wires,
while the opposite can be said about the method by Raoof, [22].
Although any of the methods presented above gives an accurate value for the axial
stiffness, a newly assembled cable does not have a linear stress-strain relationship.
The reason is that a cable consists of moving parts which need a run-in period.
In order to obtain a more linear behaviour the cable is, after the assembly, loaded
repetitively to a load well within the elastic limit of the wire material. The purpose
of this procedure is to remove the constructional stress and, thereby, obtain an
almost linear stress-strain curve [16]. However, despite this linearising process, the
cable stiffness will vary; it is lower when the cable is new and becomes higher during
the useful life of cable [97].

2.4.4 Corrosion protection

Cables made of high strength steel wires are extremely vulnerable to phenomena such
as stress and fretting corrosion. Add to this that most of the wires will be inaccessible
for inspection and maintenance in the completed cable and that numerous of cavities
are present between wires, and one understands that it is essential to ensure that
the corrosion protection is of highest quality, particularly in the regions of end or
intermediate fittings [35, 42].
It is nowadays normal practice to protect the wires in a cable by galvanization.
Both electrolytic and hot-dip techniques can be used, although the hot-dip technique
has become the preferred method. There are different classes of coating thickness
dependent on the severity of the exposure conditions. The coating is usually of pure
zinc but zinc-aluminium alloys are also used. Hydrogen embrittlement of galvanized
steel is not recognised as a real problem with wire ropes and strands [125].

20
2.5. CLADDING

It is today generally agreed that cables should have two barriers against corrosion.
For spiral strands, wire ropes and locked coil strands the second barrier consists of
filling the interstices between the wires with a blocking material and coating the
outer surface. The primary purpose of the blocking material is to prevent ingress of
moisture [42]. Suitable blocking materials are synthetic waxes and compounds based
on petrolatum (petroleum jelly), which are hydrophobic and have good adherence.
The final coating can be ordinary paint or, if necessary, a more displacement resistant
compound [125].
If the cable is exposed to an aggressive environment it is normally sheathed with a
tube made of steel or polyethylene. The space between the tube and the cable is
filled with a suitable compound such as polymer cement grout or petroleum wax [42].
Sheathing is the most effective method for corrosion protection and it is considered
as impermeable. Materials used for sheathing must be ductile and if polyethylene is
used it must be resistant to ultraviolet radiation. An alternative sheathing method
is to extrude polyethylene directly onto the cables (no filling) [35].

2.5 Cladding

In analysis of a prestressed cable structure the cladding is usually assumed not to add
any contribution to the structural stiffness. Some contribution will in any case be
added to the performance of the building, which cannot be neglected. Especially the
damping properties of the roof will be enhanced, which have significant importance
for the dynamic behaviour of the structure.
There are two main categories of cladding: continuous membranes and unit cov-
erings. Membranes can be made of fabric, foil or metal sheet. Unit coverings are
panels of metal, wood or plastic [23]. The choice of cladding material depends on
the type of structure (e.g. its shape), the expected lifetime, static and dynamic be-
haviour, security and maintenance. What type of cladding to be used should be
decided upon at an early stage in the design process in order to avoid large changes,
which might effect the cable spacing and the design of structural details [16].

2.5.1 Fabrics and foils

Fabric is today the most common cladding material used for lightweight tension
structures. As a structural element, the fabric must have the strength to span
between supporting elements, carry wind and snow loads, and be safe to walk on.
To comply with these requirements, the fabric must be prestressed, since it has
a negligible bending stiffness. The amount of prestress and the patterning of the
membrane, i.e. how the membrane should be cut and assembled, is given by the
structural analysis of the roof. Besides the structural requirements, the fabric must
meet the requirements which affect the environment inside the building; these are
air tightness, water protection, fire resistance, heat insulation, light transmission,
acoustic properties, maintenance and durability [10].

21
CHAPTER 2. LITERATURE REVIEW

Fabric membranes are composite materials. Inside the membrane there are filament
fibre yarn, designed to resist tensile forces, woven in different directions forming
an anisotropic surface. For permanent buildings with expected long lifetimes only
two types of fibres can be used: glass and aramid (Kevlar2 ) fibres, of which glass
fibre is the most common. The mechanical properties of these fibres compared to
the properties of a steel wire are shown in Table 2.1. To protect the fibres from
environmental degradation, they are coated with some resin. Resin used are PTFE3
(Teflon2 ), silicone and PVC4 [99].

Table 2.1: Comparison of filament yarn characteristics [42, 130].


Property Glass Aramid Steel
(E-HTS glass) (Kevlar 49) (Cold drawn wire)
Density (g/cm3 ) 2.55 1.44 7.86
Young’s modulus (GPa) 69 124 205
Tensile strength (MPa) 2410 2760 1570
Max. elongation (%) 3.5 2.5 4.0
Temp. resistance (◦ C) 350 250 500

Fibreglass coated with PTFE has found the broadest use for permanent buildings.
PTFE is a clear material which is chemically inert, so all dirt washes off without
damaging the coating. It is also resistant to abrasion and highly reflective, absorbing
little light as well as heat. The fact that Teflon comes in two forms, PTFE and
FEP5 , with different melting points makes it possible to heat weld seams, which
enables a fast installation of the roof cladding. In addition to its high initial cost,
PTFE-coated fibreglass has two disadvantages: the material is brittle and requires
considerable care in the packing, shipping and installation of panels, and it has little
elastic forgiveness and must therefore be accurately patterned [99].
Fibreglass coated with silicone is more flexible than PTFE-coated fibreglass, so it is
less likely to be damaged during shipment and installation. With a silicone coating,
the fabric can be made more translucent than with PTFE and the need for artificial
lightning during daytime can be almost eliminated. Fabric joints are chemically
bonded or glued. The self-cleaning properties of silicone rubber are not yet as good
as those of PTFE; it is recommended to clean the membrane once a year [99].
Fabrics of Kevlar have high tensile strength, high stiffness and very low weight.
These properties make it possible to span large distances with Kevlar fabrics without
a supporting cable net. One major disadvantage with fibres of Kevlar is that they
are highly susceptible to ultraviolet radiation and cannot be coated with translucent
resin. The fibres must be shielded with an opaque carbon black coat. Due to the
sensitivity to ultraviolet radiation the joints of Kevlar fabrics cannot be heat welded
with clear Teflon. The seams must instead be sewed, but it is impossible to develop
2
Kevlar and Teflon are registered trademarks of E. I. du Pont de Nemours and Company
3
Abbreviation for Polytetrafluoroethylene
4
Abbreviation for Polyvinyl Chloride
5
Abbreviation for Fluorinated Ethylene Propylene

22
2.5. CLADDING

the full strength of the fabric through the joints due to the high strength of the base
material [40].
The newest membrane material is EFTE6 foil, which is not a woven fabric but a
polymer film sheet. From a structural viewpoint, EFTE foil is interesting because
of its high tear resistance. In addition to the structural properties, the foil has
many properties that make it work well as enclosure material. For example, it
can be considered as incombustible, impervious to ultraviolet radiation and most
chemicals, and it can be manufactured with a translucency of over 90 % [99].

2.5.2 Metal sheets

Instead of a fabric membrane a metal membrane can be chosen. Sheets of aluminium


or steel sheets with thicknesses of 1 to 5 mm are found to be suitable for this appli-
cation. Due to the low bending stiffness of the sheets, it is necessary to prestress the
membrane to prevent buckling. Prestressing is achieved by applying the membrane
before the roof is fully erected. When the roof is raised to the final position the
membrane is pretensioned. The metal membrane is composed of small accurately
cut sections jointed by welding, gluing or bolting. Metal sheet membrane is a fea-
sible choice for long-life structures and can be designed with openings covered with
glass to provide natural lightning. Heat loss is prevented by attaching insulation
material internally [23]. In [131], Yeremeyv and Kiselev describe the manufacturing
and erection of a number of large projects in Russia where metal sheets are used as
covering.

2.5.3 Panels

A cable net with cable spacing of around half a meter is ideal for small elements
(panels). The elements are either shape-cutted or jointed in such a way so that they
will conform to the shape of the structure. The panel system is most economical if it
is made of light material, not to impose extra weight on the cable structure. Panels
of fibreboard, aluminium and plastic are appropriate to use for covering roofs [23].
For the Olympic Stadium in Munich, a system with translucent plastic panels (Plex-
iglas7 ) with thickness of 4 mm and size of 2.90 m × 2.90 m was used. The panels
were fastened to the supporting cable net with shock absorbing flexible connections
to prevent cracking of the panels under roof movements. The joints between the
panels were sealed with continuous neoprene profiles, as seen in Figure 2.17 [63].
However, it should be mentioned that many architects, e.g. Philip Drew [31], find
the Plexiglas cladding of the Olympic Stadium ugly. Therefore, it will probably not
be used again.
6
Abbreviation for Tetrafluoroethylene
7
Plexiglas is a registered trademark of AtoHaas Americas Inc.

23
CHAPTER 2. LITERATURE REVIEW

Figure 2.17: Acrylic panels for the Olympic stadium in Munich. Reproduced
from [57].

2.6 Structural details

Already at early stages of the design process the designer has to pay attention
to the design of the structural details. Structural details are fittings, saddles and
anchorages. Fittings are attachments used to grip the cable at the ends or along its
length. They can be classified, in accordance with the type of application, as the
friction or clamp type, the pressed or swaged type, and the socketed type. Saddles
are used when the cable has to run continuously over masts and other supports.
In self-supporting systems, cables are anchored into structural members, such as a
concrete ring or an arch. In other systems the cable forces are resisted by anchors
in the ground [57]. A comprehensive survey of structural details is given by Chaplin
et al. [23].

2.6.1 End fittings

An end fitting (terminal) is an attachment, which transmits the cable force to the
supporting system. To be totally effective, the end fitting must withstand the full
breaking force of the cable without significant yielding, endure dynamic loading
without risk of fatigue failure and not induce fatigue failure of the cable. For ap-
plications where large forces are to be transmitted to the supporting structure two
different end fittings are accepted [125]: the socketed type and the swaged type,
Figure 2.18.

24
2.6. STRUCTURAL DETAILS

(a) Socketed type (b) Swaged type

Figure 2.18: Cable end fittings with pin connectors. Reproduced from [16].

The most reliable, but also the most expensive, of the end fittings is the socketed
type. It is manufactured by splaying the end of the cable a prescribed length and
cleaning the individual wires. When the wires are cleaned and dried the conical
socket of machined or casted steel is positioned on the splayed cable section. Then
molten socketing material is poured into the socket, hardens and forms a cone, Fig-
ure 2.18(a). As tension is applied to the cable the cone is drawn into the socket
and wedging forces are developed which grip the wires. As socketing material either
of zinc or resin is used. Pure zinc has been used for over a century and it offers
a cathodic protection for the cable, but it is sometimes criticised for impairing the
fatigue resistance of the cable in this region. Another, more important, disadvan-
tage with sockets filled with pure zinc is that they are prone to creep effects under
high stresses. Therefore zinc alloy, with improved creep resistance, is often used.
Polyester or epoxy resin has better creep resistance. As the resin is casted at low
temperature the fatigue resistance of the cable will not be impaired. Socketed end
fittings can be used for all cable sizes but cables of smaller diameter, approximately
less than 38 mm, can be terminated by means of hydraulically compacted fittings
called swaged end fittings. Swaged end fittings are cheaper than socketed types
but they are only guaranteed to resist 95 % of minimum breaking load of the ca-
ble. All end fittings are manufactured, installed and rigorously tested by the cable
manufacturer [16, 125].

2.6.2 Intermediate fittings

Intermediate fittings are used to connect cables to other cables. These fittings are
usually not standard appliances and their behaviour depend on the frictional force
between the cable and the clamp. To prevent sliding of the clamp, the clamping force
must be large and thereby high radial stresses are induced. Cables are more prone
to fatigue when the pressure between adjacent wires is high and it is, therefore,
important to use fittings where the clamping force is evenly distributed over the
cable. The resistance of a spiral strand and a locked coil strand to clamping forces,
where the latter has the higher resistance, can be found in Eurocode 3 [35]. When
the cable is tensioned the diameter will decrease and consequently the clamping
force. It can therefore be necessary to retension the clamp bolts to prevent sliding.

25
CHAPTER 2. LITERATURE REVIEW

To avoid abrasion between the clamp and cable under cable movements, which can
result in fatigue failure, the ends of the fittings must be radiused. Different types of
intermediate fittings are shown in Figures 2.19–2.20.

(a) Clamp connection (b) Swaged clamp connection

Figure 2.19: Cable connections for dual-strand cable nets. Reproduced from [16].

(a) Single U bolt connection (b) Double U bolt connection

Figure 2.20: Cable connections for two-way cable nets. Reproduced from [16].

In the search for the best economical solution one key is to use few types of structural
details, as the number of fittings in, for example, a cable net can be quite large. A
way to achieve this is to use a fitting which can be adjusted for different angles
between cables. The fitting shown in Figure 2.19(b) can be mounted in a factory
and thereby it is possible to reach a high accuracy. As mentioned above, accurate
assembly of the fittings is necessary in order to obtain the desired internal force
distribution in a cable net.

2.6.3 Saddles

When the cables have to run continuously over supports like columns and masts,
they have to be supported by saddles, Figure 2.21. When designing a saddle one
has to take the bending stiffness of the cable into account. Two factors have to be
checked:

• the tensile stress in the outer wires, and


• the pressure between the cable and the saddle.

26
2.7. ROOF LOADS

If the pressure between the cable and the saddles is too high the fatigue resistance
of the cable will be affected. The common rule is that the diameter of the saddle
should not be less than 30d, where d is the diameter of the cable [16, 35].

Figure 2.21: Saddle. Reproduced from [16].

2.6.4 Anchorages

In self-supporting systems, the cables are anchored into the boundary structures,
which resist the cable forces due to either geometry or self-weight. These structures
are usually rings, arches and masts made of concrete or steel. In open systems
the cable forces are resisted by tension anchors in the ground. A survey of exist-
ing tension anchors and methods for estimating their capacities for various ground
conditions can be found in [16]. Which of the two anchorage alternatives that will
be most economical, if both are architecturally accepted, depends upon the ground
conditions, cost of material, and availability of expertise and labour skill.

2.7 Roof loads

Today, structural analyses are performed using commercial finite element programs,
which contain elements for almost every application. New elements are constantly
being developed and older refined in an attempt to obtain more accurate results.
Nevertheless, the accuracy of the results will mainly depend on the errors in the
prescribed loads acting on the structure. Since most loads are environmental loads
with random distributions, durations and magnitudes, the ‘exact’ values will never
be known. In an attempt to achieve higher accuracy in the results from a structural
analysis more reliable data on the extreme loads acting on buildings are needed.
Apart from the prestress, the loads acting on cable roofs are the same as any other
type of loads acting on more conventional buildings. However, it is well known that
non-uniformly distributed loads are more dangerous to cable structures than uniform
loads. Therefore, it is important to determine the ‘true’ load distribution on the
structure. Nonetheless, the unusual shape of these structures, together with their
low weight and large scale, make this a difficult task. A further complication is that
practically no guidance is available from codes of practice. This implies additional

27
CHAPTER 2. LITERATURE REVIEW

costs to the project, because of the need for expertise. The latest methods for
determining the loads on roofs of general shapes involve very sophisticated physical
and computational modelling techniques, which require expensive equipment and
powerful computers. In this section, these methods are reviewed. The loads are
viewed in order of their importance on the structural behaviour of tension structures.

2.7.1 Wind load

Due to the low weight of cable roofs with membrane cladding, wind pressure is one of
the most important forms of loading. The variability and large number parameters
involved in the determination of wind effects on structures make it a very complex
problem. Some undesirable effects and partial collapses have been caused by wind
on tension structures [16]. Among these can be mentioned the vibrations due to
wind on the roof of the Raleigh Arena, U.S.A., which made it necessary to insert
supplementary internal cables.

The nature of wind

Wind is initiated by pressure differences between points of equal elevation, caused


by variable solar heating of the atmosphere of the earth. The motion of the air mass
is modified by the rotation of the earth and close to the ground the velocity of the
moving air is reduced due to friction. At a certain height above the surface of the
earth the effect of the surface friction becomes negligible. Above this boundary layer
a frictionless wind balance is established, and the wind flows with the gradient speed
along lines of equal barometric pressure. The height of the atmospheric boundary
layer normally ranges from a few hundred meters to several kilometres, depending
upon wind intensity, roughness of terrain, and angle of latitude [32].
Physically, the wind is composed of two different velocity components [16]. The first
component is the velocity of a steady flow determined by the long-term pressure
variations (approximately four day periods). This component is called the mean
wind velocity. The second velocity component, which is superimposed on the steady
flow, is due to a turbulent fluctuating system with high frequency components,
which is caused by the friction between the air and the surface of the earth. The
two velocity components are clearly seen when the wind velocity is plotted in a van
der Hoven power spectrum, Figure 2.22. This spectrum shows the variations of the
mean square of the amplitudes of the fluctuating components against the frequencies
of these components.
Hence, the analysis of linear structures can be divided into to two parts: the cal-
culation of the quasi-static response due to the steady velocity component and the
response caused by the turbulence components. As cable structures have a non-
linear behaviour this division is generally not valid. Instead, the total wind load
must be used in the dynamic analysis of cable structures. In the sequel to this
section the common expressions for description of the wind load on buildings and

28
2.7. ROOF LOADS

ways to obtain the pressure distribution will be described in brief.

Figure 2.22: Spectrum of horizontal wind speed after van der Hoven. Reproduced
from [26].

Mathematical description of natural wind

To describe the wind velocity mathematically a Cartesian coordinate system is ap-


plied, with the x-axis in the direction of the mean wind velocity, the y-axis hori-
zontal and the z-axis vertical, positive upwards. The total wind velocity at time t,
V (x, y, z, t), is formulated as:

V (x, y, z, t) = U (z) + u(x, y, z, t) + v(x, y, z, t) + w(x, y, z, t), (2.9)

where U (z) is the mean wind velocity in the mean direction θm , u, v and w, are
turbulence components of the wind field in the x, y and z directions, respectively.
It can be noted that the mean wind velocity U (z) only depends on the height above
the ground. The turbulence components are treated mathematically as stationary,
stochastic processes with a zero mean value. The mean wind velocity U (z) and the
turbulence component u in the wind direction are often most important, as they
usually give the main contributions to the wind forces on a structure [32].
Three laws have been proposed to describe the way in which the mean velocity U
varies with height [32]. The first law is the power law, which has been adopted in
many codes. The second law is the logarithmic law, which is derived not only from
empirical data, but also from theoretical considerations. The Deaves and Harris
model, which is the third law, is the most exact one since it is fitted to experimental
data [16,26]. In urban areas, where stadiums and other large roofs usually are built,
the terrain roughness might change if buildings are erected or demolished [32]. This
directly affects the mean wind velocity and has to be considered at the design stage.
The wind in the boundary layer is always turbulent, which means that the flow is
chaotic, with random periods varying from fractions of a second to several minutes,
Figure 2.22. In order to describe a turbulent flow, statistical methods must be
applied [32].

29
CHAPTER 2. LITERATURE REVIEW

Wind load on a structure

The earliest method for the assessment of the action of turbulent wind is the quasi-
steady vector model [27]. It makes the simple, but inaccurate, assumption that the
pressure fluctuations correspond exactly with the variations of the wind velocity.
Other methods may be found in [27], but wind loads on buildings are determined
using the quasi-steady model in many codes [64]. Therefore, a detailed description
of the quasi-steady theory will be given here. For a point on a surface, (x,y,z), the
instantaneous pressure, p, is given by [27, 64]
1
p = ρV 2 C p (θm + θv , θw ), (2.10)
2
where V is the wind velocity given by equation (2.9). C p (θm + θv , θw ) is the mean,
with respect to time, pressure coefficient for the instantaneous azimuth angle, θv
and the elevation angle θw , of the wind velocity vector measured from the mean
wind direction θm . The magnitude of the wind velocity is given by

V 2 = (U + u)2 + v 2 + w2 . (2.11)

The instantaneous azimuth angle θv is given by


v
θv = tan−1 . (2.12)
U +u
In the same way the vertical component θw can be expressed as
w
θw = tan−1 . (2.13)
U +u
By removing small second order terms, the full quasi-steady model is linearised and
the velocity magnitude reduces to

V 2 ≈ U 2 + 2U u. (2.14)

The fluctuating wind directions are assumed linear for small v and w, which gives

v ∂C (θ )
w ∂C (θ )
p m p m
C p (θm + θv , θw ) ≈ C p (θm ) + + . (2.15)
U ∂θv U ∂θw
Substituting (2.14) and (2.15) into equation (2.10) yields

v ∂C (θ )
w ∂C (θ ) 
1  2 p m p m
p(t) ≈ ρ U + 2U u C p (θm ) + + . (2.16)
2 U ∂θv U ∂θw

Dividing both sides of equation (2.16) by the mean dynamic pressure 12 ρU 2 , ex-
panding and discarding small turbulent cross terms gives the instantaneous pressure
coefficient

u
v ∂C (θ )
w ∂C (θ )
p m p m
Cp (θm + θv , θw , t) ≈ C p (θm ) + 2 C p (θm ) + + .
U U ∂θv U ∂θw
(2.17)

30
2.7. ROOF LOADS

Taking the time average of equation (2.17) leaves the expected result

C p (θm + θv , θw , t) = C p (θm ). (2.18)

To evaluate the performance of the quasi-steady theory, full-scale wind velocity and
pressure measurements were recently done on an full scale experimental building at
the Texas Tech Field Research Laboratory [64]. The results from that study showed
that the area-averaged pressures over a substantial area of the roof as well as root
mean square (rms) and peak pressure coefficients can be well predicted using the
quasi-steady theory. However, the spectra of the pressure coefficients cannot be
predicted at high frequencies using the quasi-steady model.

Wind tunnel testing

The value of the wind pressure coefficient Cp is a function of shape, scale, surface
condition, surroundings, wind velocity and wind direction [32]. Because of the com-
plexity of these factors, the pressure coefficients must be determined by full-scale
measurements or wind tunnel tests. Full-scale measurements are the most accurate,
but not possible in practice and is therefore only carried out to verify the wind
tunnel tests. Hence, the most appropriate method for determining the wind load
is to test a model of the structure in a wind tunnel. Surface pressure coefficients,
based on such tests, for traditional building shapes can be found in different codes.
However, as mentioned above, the shapes of tension structures are not covered by
the codes [40].
To interpret the results from a model test, the model must satisfy several laws [32].
These model laws are formulated by introducing a number of non-dimensional pa-
rameters. In wind engineering, the number of parameters is so large that it is
impossible to satisfy all the conditions simultaneously. Therefore, some parameters
that are of minor importance have to be disregarded. Besides the laws for the model
itself, the wind tunnel must also be able to simulate the wind climate in the atmo-
spheric boundary layer for the site considered. A general description of model laws
and boundary-layer wind tunnels can be found in e.g. [32].
Earlier, rigid models were used in wind tunnel studies to determine the pressure
distribution on the exterior and interior surfaces of a building, for a variety of wind
directions. The rigid model studies have been developed over many years of test-
ing conventional buildings and are relatively straightforward. However, high wind
speeds can change some of the pressure coefficients when aeroelastic models are
used [49]. These changes are attributed to roof deflections and to the non-linear
stiffness of the roof. The conclusion is that the use of rigid pressure-tapped models
can underestimate the pressure coefficients for flexible structures undergoing large
displacements. Another reason, maybe the main one, for conducting wind tunnel
experiments with aeroelastic models is to search for unforeseen aerodynamic insta-
bilities, i.e. large amplitude vibrations [49]. In references 33 and 49 wind tunnel tests
of tension roofs with aeroelastic models are performed, but no types of aerodynamic

31
CHAPTER 2. LITERATURE REVIEW

instabilities were found. Certain aspects that have to be taken into account when
modelling cable and membrane structures in a wind tunnel are found in [121].

Methods of analysis

In wind tunnel testing the pressure distribution is measured using electronically


scanned multi-channel pressure systems, with up to 512 channels and sampling fre-
quencies of up to 100 Hz [11]. Hence, the amount of data from one series of testing
is enormous and in its raw state very difficult to use for analysis. A method to de-
scribe the wind velocity profile and wind pressure pattern was recently rediscovered
in the field of wind engineering. This method, often used for stochastic problems,
is the proper orthogonal decomposition (POD), known also as the Karhunen-Loeve
expansion8 . The POD method resembles the modal analysis used in structural dy-
namics [11, 12].
The main objective of the POD method is to find a deterministic function Φ(x, y)
which is best correlated with all the elements of a random field [11]. The deter-
ministic function Φ(x, y) is found through a maximisation of the projection of the
random pressure field p(x, y, t) on Φ(x, y)

p(x, y, t)Φ(x, y)dxdy
= max . (2.19)
Φ2 (x, y)dxdy

If the maximisation of equation (2.19) is performed in the mean-square sense for a


discrete pressure field, it leads to the following eigenvalue problem

Rp Φ = λΦ, (2.20)

where Rp is the covariance matrix of the pressure space, and Φ and λ are, respec-
tively, a vector and a value, both to be determined. The eigenvectors Φn (xi , yj ) are
base functions in a series expansion of the pressure field

p(xi , yj , tk ) = an (tk )Φn (xi , yj ), (2.21)
n

where the expansion coefficients, i.e. modal amplitudes, an (tk ) are easily computed
due to the orthogonality of the eigenfunctions Φn (xi , yj )

j p(xi , yj , tk )Φn (xi , yj )
an (tk ) =
i
2 . (2.22)
i j Φn (xi , yj )

The eigenvalue λk is the measure of the contribution of each eigenmode to the


pressure mean squares [30]. Depending on the number of terms included in the ex-
pansion, equation (2.21), different levels of accuracy are reached. In [12] about 30 %
8
The expansion was derived independently by a number of investigators; Karhunen in
1947, Loeve in 1948, and Kac and Siegert in 1947 (according to “Stochastic Finite Ele-
ments: A Spectral Approach” by Ghanem, R. G. and Spanos, P. D., which can be found at
http://venus.ce.jhu.edu/book/)

32
2.7. ROOF LOADS

of the eigenvectors were required in the expansion to represent the peak pressure
with an error of approximately 10 %. However, only one term, the first eigenvector
was needed to represent the mean point and area-average roof pressure with an error
of approximately 1 %. A physical interpretation of the first three eigenvectors can
be provided by the quasi-steady theory [122]. It was shown in reference 122 that
the first eigenvector was closely related to the mean pressure distribution C p (θm ),
the second and third eigenvectors were related to ∂C p (θm )/∂θv , and ∂C p (θm )/∂θw ,
respectively. These results, together with the full scale measurements by Letchford
et al. [64] indicate that the pressure field over certain roof types can be described
by the quasi-steady theory and that the POD method and the quasi-steady theory
in some sense are related to each other.
A simplified method to calculate wind loads on tension structures with irregular
shapes has been presented by Tabarrok and Qin [115]. The method simplifies input
data, because it does not need experimentally measured wind pressure coefficients.
In their method a membrane structure was discretized with constant strain shell
elements. To calculate the wind load on each element, the designer specifies a
magnitude of the pressure coefficient Cp that defines the wind pressure on a vertical
surface normal to the wind direction, and a direction that defines the source of the
wind. The wind pressure normal to each element is then computed by scaling the
wind velocity by the cosine of the angle between the wind direction and the outward
normal to the element. This means that the model gives zero pressure for surfaces
parallel to the wind and suction on leeward surfaces.

Computational wind engineering

Wind tunnel testing, including model making, is expensive, tedious and in some cases
inaccurate due to limitations associated with the boundary layer wind tunnel [109].
These limitations might be overcome if the pressure values could be derived by
numerical computer simulations. Savings, in both time and money, would also be
possible.
Application of Computational Fluid Dynamics (CFD) to wind engineering problems
means large computer memory and CPU time, because very fine computational grids
are needed to deal with the modelling of turbulence, complex building configurations
and the large area of model domain [109]. Today, only smaller structures with coarse
grids can be analysed. However, it is anticipated that current limitations due to
long CPU time and large memory requirements will be overcome in the near future
through new computational, parallel-processing based architectures, faster computer
hardware and more efficient computational algorithms [11].
The current state-of-the-art of Computational Wind Engineering (CWE) is reviewed
by Stathopoulos [109]. Results from CFD simulations are compared with those
obtained from wind tunnel or full scale experiments for buildings of different shape
and various wind directions. According to Stathopoulos: “Disagreements proved to
be higher than what is tolerable, particularly for cases that require complex building
shapes, surroundings and for results other than mean pressure coefficients.” He

33
CHAPTER 2. LITERATURE REVIEW

further concludes that “at present time CWE may be used only for the assessment of
the wind environment around buildings. For cases which involve mean values of the
wind speed and pressure coefficient the numerical results may be used for preliminary
design purposes.” Several areas have to be improved before CFD can be used in
design, which include: numerical accuracy, description of boundary conditions and
refinement of turbulence models [109].
The most optimal method today seems to be a hybrid analysis, where experimental
wind tunnel data are combined with numerical simulations [11]. For engineering
problems, a hybrid analysis could also involve combining the experimental data
with those from available CFD software packages.
To summarise wind loads on cable roof structures:

• The mean and turbulent parts of the wind cannot be separated in analysis due
to the flexibility (geometric non-linearity) of the roofs.

• Pressure coefficients have to be determined by wind tunnel tests. The pressure


distribution from the tests can be described mathematically using the POD
method.

• To be able to recognise any unforeseeable wind related instability, the wind


tunnel model should be of the aeroelastic type.

Although several methods are at hand for the wind engineer, the assessment of
wind effects on structures is certainly not easy. As mentioned before, this area
is very complicated and for a more in-depth analysis one can refer to the many
references given in this section, especially [26,27,32]. As the design recommendations
concerning tension roof structures are non-existent, good engineering judgement and
experience are important characteristics of designers of tensile structures.

2.7.2 Snow load

Apart from wind loads, snow loads play an important part in the design of struc-
tures. Many modern buildings have moved away from traditional shapes and their
behaviour with respect to snow accumulation is not known well enough [41]. When
constructing a tension roof, the load corresponding to the expected intensity of snow
has to be considered. As for the wind loads this is more or less straightforward for
ordinary types of roofs, and is found in national building codes. For cable and
membrane roofs this is considerably more difficult.

Snow distribution

The snow intensity is measured at meteorological weather stations as the ground


snow depth. Prior to 1970, many buildings were designed and built assuming uni-
formly distributed snow loads [118]. After a number of failures, attention was given

34
2.7. ROOF LOADS

to unbalanced loads, due to snow drift. Therefore, surveys of actual snow loads
were started to determine the difference between ground and roof snow load. The
results from the surveys showed that, in cold and windy areas, the roof snow load
was considerably lower than the ground snow load. Nonetheless, on certain parts of
some roofs the load was significantly higher.
Today, building codes make provision for drifting of snow by specifying a number
of snow load cases for the type of roof considered. Some shapes of roofs tend to
accumulate more unbalanced loads than others, and the load cases try to cover the
possible snow distributions over the roof. Unfortunately, the roof types covered by
the codes are usually traditional. Tension structures, such as cable and membrane
roofs, with sculptural forms are not covered by the codes. Due to the flexibility of
tension roofs, ponding of snow can occur in flat areas or under heavy snow loads.
This requires consideration in design and can only be analysed with the aid of wind
tunnel or water flume experiments.

Wind tunnel and water flume testing

Like wind tunnel experiments, some model laws has to be followed when modelling
the snow in air or water. In wind tunnels, granular materials, such as tea, glass, and
nut shells, are used to simulate dry snow, while sand is used in water flume experi-
ments, Figure 2.23. One limitation in modelling, which cannot easily be overcome,
concerns the great variation in snow properties. Common simulation materials can-
not model sticky snow. For example, sand will not stay on steep surfaces which
makes it difficult to simulate snow accumulations on steep slopes where snow will
accumulate before eventually sliding off. Another limitation in model studies is that
only one wind direction is considered at a time but the overall seasonal environment
consists of a sequence of snow storms and high winds from different directions. In
reality, the final snow accumulation depends on the chronological order and dura-
tion of the storms and on temperature, sunshine, humidity, etc. Surrounding terrain
may also affect total snow accumulation and drift patterns on structures. Whether
air or water is the medium, a model can provide a good simulation of the flow
around structures. However, the state-of-the-art of snow drift modelling prevents
the measurement of quantitative results [50, 118].
Recently a water flume test was used to determine the snow loads on a large tension
roof at Denver International Airport, U.S.A. [10]. Denver is known for its heavy snow
falls, and the shape of the roof leads to high snow load intensities being expected.
The tests also showed that in the valleys of the roof the design snow intensity was
very high, 3.8 kN/m2 , Figure 2.23. The predicted snow pattern from the model
test agreed well to that seen on the roof after the first snow falls, confirming the
reliability of the test.

35
CHAPTER 2. LITERATURE REVIEW

Figure 2.23: Investigation of snow drift with the help of a model test, where water
replaces air and sand represents snow. Reproduced from [10].

Computer simulations

In the design of the tension roof of Denver International Airport, the water flume
experiments were supplemented by a computer program based on the Finite Area
Element (FAE) method (not to be confused with the Finite Element Method) [38].
This is a so-called hybrid method, which means that the wind velocity field is ob-
tained from wind tunnel experiments. A brief description of the FAE method and
its properties is given below. First, the roof is divided into many area elements by
a grid. The wind velocities are measured at grid intersection points. Time histories
of meteorological data concerning the wind direction and speed are used as input
for the computations. Snow drift is computed using empirical relationships for snow
flux versus wind velocity. By computing the mass fluxes into and out of each ele-
ment, the rate of build up or depletion of snow mass in the element due to drifting
is determined. The mass balance computations at each time step include the addi-
tional mass from snow fall and the depletion due to melting. The method also takes
into consideration the less significant drift of snow that has been rained upon, or
that has experienced a melting episode. Some surfaces, which are rough or ribbed,
have high snow storage capacities and can trap snow permanently (at least until it
melts). Therefore, the area elements are assigned with a certain storage capacity for
snow depending on surface roughness. Also included in the FAE method is a heat
balance used to calculate the melting rate of the snow pack inside each element, and
the ability of snow to store liquid water and thereby increasing the snow density.
The FAE method has proved to be a good tool to supplement the model studies,
and overcome the limitations associated with them. With this method quantitative
results can be obtained with higher accuracy [38].
A purely computational method for predicting snow accumulation, called SNOW-
SIM, has been developed under a research project at Narvik Institute of Technology
in Norway [7]. The method includes a commercial CFD program, combined with a
simplified drift-flux model to simulate snow drift. A computer simulation of snow
drift has the advantage over wind tunnel or water flume experiments that it can
be more available and less expensive. Simulations can be done with snow drifts

36
2.7. ROOF LOADS

from different directions and with variations in velocity and intensity. Simulations
in three dimensions were presented, but due to limitations in computer power only
small buildings of regular shape with a coarse mesh resolution, and simulation times
of up to 50 seconds could be studied. Therefore, the simulations could only be re-
garded as an indication of where the snow will deposit, not as an exact quantity
calculation. Compared with real measurements the simulations gave similar snow
drift patterns. Bang et al. [7] gave a number of problems that have to be resolved
before quantitative results can be available. These included the proper treatment
of turbulence in the CFD program, evaluation of different drift-flux models, and
modelling of structures with their local terrain. Thus, a complete computer simula-
tion of snow drift magnitude is today not available even for buildings of traditional
shape [7].
As in the case with wind loads on tension structures with complex shapes, Tabarrok
and Qin [115] have proposed a simplified method to calculate the snow load distri-
bution. In their method, vertical snow loads are generated based on the horizontal
projection of each elemental area and a snow load magnitude per unit horizontal
area specified by the designer. This means that there is full snow load on a hor-
izontal surface and zero load on a vertical surface. This method is of course very
approximative as it cannot handle snow drift.
It has been seen that the determination of snow load magnitude and distribution is a
task of equal difficulty as that for wind load. For a roof with a complex shape the only
way to find the sought quantities, i.e. magnitude and distribution, is through model
tests. This procedure is expensive, time consuming and requires special knowledge
and experience.

2.7.3 Earthquake load

Another important form of loading, which has to be considered in certain parts of


the world, is earthquake ground motion. Even smaller earthquakes may lead to
collapse of stiff structures. Many studies have been concerned with the earthquake
response of building structures but, like the wind and snow load studies, very few
have included cable roof structures. Two works on the topic have been found and
are briefly presented in the following.
In reference 78, a cable truss with diagonal ties (system Jawerth) is subjected to
vertical and horizontal earthquake loadings. Both a linear and a non-linear analysis
was performed. The maximum displacements did not differ significantly between
the analyses. It was also found that under a horizontal earthquake, all the diagonals
became slack at many instances. As far as the diagonal forces are concerned, the
response was, according to Mote and Chu, “very erratic” [78].
An elastic earthquake response analysis of a type of cable dome—the suspen-dome—
is presented in reference 117. The suspen-dome is a single-layer truss dome stiffened
with a tensegrity system. Tatemichi et al. conclude that the analysis “confirmed
effectiveness of the suspen-dome against earthquake motions, particularly vertical

37
CHAPTER 2. LITERATURE REVIEW

motions.”
In general, the response of structures to dynamic loading is determined by a finite
element analyses. Such analyses of cable roofs have shown that these structures
usually have a long period of vibration. In addition, the supporting structures are
relatively much stiffer and heavier than the cable system. Therefore, high-frequency
contents of the earthquake ground motion will be amplified by the supporting struc-
tures. Conversely, low-frequency components will be reduced considerably by the
time they reach the cable system. Hence, the response of the structure is dependent
on the low-frequency content in the ground motion [57].

2.7.4 Other loads

For the majority of civil engineering structures, e.g. bridges, the dead load is a
large part of the total load. This is not the case for prestressed cable roofs. For
these roofs the dead load consists of the weight of cladding, insulation, cables and
fittings, etc., [57]. The magnitude of the dead load for a prestressed roof with fabric
cladding is very low, values as low as 0.1 kN/m2 are common [99]. Hence, the dead
load cannot be considered as important in ensuring the safety of correctly designed
prestressed cable roofs. Nevertheless, wind suction may cause large deflections or,
even worse, flutter instability of a flexible structure with a low dead load. Undoubtly,
for suspended roofs the weight and stiffness of the cladding is more important, as it
governs the stiffness of the roof.
Of the permanent loads, the prestress is in many cases the most important one.
The magnitude of the pretensioning force varies from structure to structure, but
must, due to stress relaxation, not be greater than 45 % of the breaking force of the
cable [42], section 2.4.2. Apart from relaxation, loss of cable tension also occurs as a
result from creep in the supporting structure, slippage of cables at anchorage points
and increase in temperature [57].
Live loads are usually taken into account by specifying the intensity of a uniformly
distributed load. Usually, cable roofs have curved shapes and may therefore be
considered as inaccessible to people except for maintenance purposes. This justifies
the use of a lighter design live load for the cable system and supporting structure,
but for the cladding a normal design live load should be used [57].
Of course, all the different loads (wind, snow, dead, live, temperature) presented in
this section are not considered separately. Design load cases consist of combinations
of the different loads. The load combinations forming these cases are given in the
national building codes and there is no reason to expect that the cases will be
different for cable roof structures.

38
2.8. ANALYSIS METHODS

2.8 Analysis methods

In this section, some techniques used for analysis of cable structures will be men-
tioned. None of the numerical techniques described here are applicable to only cable
structures. Therefore, only a short historical review is considered necessary.
Early analyses were done by applying membrane shell theory to cable nets. Appli-
cation of the membrane shell theory results in a set of differential equations. Except
for special cases, these equations are difficult to solve in closed form. In most
cases, the equations are solved by numerical techniques, such as the finite difference
method [112]. Shore and Bathish [107] used double Fourier series to transform the
differential equations to a system of algebraic equations. One flat and one hyperbolic
paraboloid prestressed cable net, both square in plan, were analysed numerically and
experimentally. The agreement between the results was acceptable. Recently, Tärno
performed a parametric study of saddle-shaped networks with elliptic plane layouts
and stiff contours [116]. Such a study serves as an aid in choosing the dimensions
of the roof and the structural elements. In general, membrane shell theory is less
accurate if the cable mesh in a net is coarse; it is inadequate for complicated roof
shapes.
Since the introduction of computers in the 1960’s, several numerical methods have
been developed for the general analysis of structures. Among these methods, the
stiffness technique (finite element method) have been widely adopted. Originally,
the method was developed to analyse structures with small displacements. Under
the action of external loads cable nets undergo large displacements and it became
evident that the stiffness technique was not applicable to such structures in its
original form. Therefore, the original method was modified and applicable structures
with geometrically non-linear characteristics. Several iterative methods have been
applied to the non-linear stiffness method. The most popular is the Newton-Raphson
technique, which has proven to be accurate, efficient and applicable to the majority
of cable structures [1]. A comprehensive description of the finite element method and
the Newton-Raphson technique can be found in, for example, [28]. Other authors,
e.g. [16, 111] have used a method based on the minimisation of the total potential
energy of the structure. The minimisation was done using the the conjugate gradient
method. The dynamic relaxation technique has been used by several authors for
both form-finding [8, 66] and load analysis [68, 69].
Approximate methods for the preliminary design of cable trusses and simple cable
nets, can be found in references 16, 57 and 76. For elliptical cable nets the method
described by Tärno [116] is recommended.
The following chapters will investigate some earlier reported analysis methods and
some new variants of them, aiming at accurate analyses of the form-finding and
normal usage stages of some cable roof structures. Failure stage analysis will be
identified as a topic for further research.

39
Chapter 3

The initial equilibrium problem

3.1 Introduction

For structural analysis, the equilibrium configuration of a structure is generally


known in advance. This is not the case for tension structures, i.e. cable and mem-
brane structures. Due to the low flexural stiffness of the cables and the fabric these
structures have to be constructed so that they will experience a significant prestress
at all times. Thus, there is no compatible unstressed configuration for a tension
structure, even if no external loads are applied and its self-weight is neglected.
Therefore, the designer must specify a reference configuration for the structure that
is stressed. The shape of the reference configuration depends upon the internal
stresses and forces. Hence, the load bearing behaviour and the shape of the struc-
ture cannot be separated and cannot be described by simple geometric models. In
addition to satisfying the equilibrium conditions, the initial configuration must ac-
commodate both architectural, structural and constructional requirements [44,114].
Finding the stressed initial configuration is an inverse structural problem, in which
the specified force distribution is the driving parameter in the process. This is
inverse to standard problems where the forces are the structural response to the
deformations of the structure [13].
The problem of finding a configuration that satisfies the laws of equilibrium is usu-
ally called form-finding or shape-finding. Haber and Abel [44] thought that this
nomenclature was inappropriate to use when describing methods in which variables
besides the shape were adjusted to satisfy equilibrium. Therefore, they used the
term initial equilibrium problem instead. Throughout the present chapter and the
rest of the thesis this term will mainly be used.
The objectives of this chapter are to describe the initial equilibrium problem and
review the existing computer methods for solving it. All the methods that are to
be described are applicable to mainly cable structures, membrane structures and
bar frameworks. Among the methods, one is especially interesting, namely the force
density method. This method will further on be described in detail and applied to
a number of different problems. First, a brief description of the methods that were

41
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

used before computer methods were available will be given.

3.1.1 Physical modelling

Due to the mathematical difficulty and spatial complexity of the structural forms of
tension structures, physical modelling was the primary method to solve the initial
equilibrium problem of tensile structures until 1969 [44], when the cable roofs for the
1972 Munich Olympic Games were about to be built [62]. A pioneer in the physical
modelling field was Frei Otto, who performed extensive experiments using a variety
of different media, including soap films, fabric, and wire models.
The most simple modelling tools are soap film models, which are obtained by dipping
wire frames into soap water. These models are used as a first check if the curvature of
the roof surface is appropriate. As is well known, a soap film always contracts to the
minimal surface. The minimal surface may be the most aesthetic shape, but it is not
always the best structural form, as the minimal surface approach tends to produce
very flat areas, which may induce flutter (see also section 3.2.1). After the soap film
models, working models of larger scale and of other materials were built for further
processing within the design process [87]. The Institute for Lightweight Structures
in Stuttgart has worked with physical modelling techniques for several decades and
a large number of structures with complex structural shapes of different scales have
been realised during the years, e.g. [31] or [87]. The Institute was involved in the
construction of the some of the largest and most complicated cable nets built in the
world: the German pavilion at the 1967 World’s fair in Montreal and the Olympic
cable roofs in Munich.
Models give many useful insights into the the behaviour of tension structures. A
large number of configurations can in a short time be studied if for example soap film
models are used. For practical design work, however, they do not provide sufficient
accuracy and have to be replaced by computational methods [31, 44]. Nevertheless,
physical models of tension structures will always be made, because of the excellent
structural visualisation they provide [77].

3.2 Literature review of initial equilibrium solu-


tion methods

In a numerical method a solution to the initial equilibrium problem consists of a


combination of parameters describing an equilibrium configuration of the structure.
One method of distinguishing the different solution methods is to indicate which
of the parameters are specified by the designer and which are treated as problem
unknowns [44]. The parameters involved in the initial equilibrium problem are:

• Structural topology
The structural topology defines the connectivity of the material of the struc-

42
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

ture. This is done via the stiffness matrix in the finite element method (sec-
tion 3.2.1) or the connectivity matrix in the force density method (section 3.3).

• External loads
Two types of loadings may act on the structure: body forces and surface trac-
tions. Inclusion of these loads often complicate the initial equilibrium problem
as the direction and magnitude of the loads may depend on the unknown ref-
erence configuration.

• Structural geometry
The actual shape or surface geometry of the structure is one of two key parame-
ters in the initial equilibrium problem. It plays a major role in determining the
stresses that will act in the structure at various times. For a tension structure,
curvature is the parameter that mostly affects the structural behaviour.

• Boundary conditions
In methods where the geometry is treated as an unknown, it is necessary to
introduce some boundary conditions to ensure a unique solution.

• Internal force distribution


The internal force distribution is the second key parameter. In order to obtain
a safe and economical design, it is crucial to find an appropriate force pattern.

The initial equilibrium problem is a pure statics problem. Therefore, it is not neces-
sary to introduce kinematic equations. However, some methods, e.g. the non-linear
displacement method, are using kinematic equations to solve the initial equilibrium
problem. This method requires material properties to be specified, although these
need not be the actual properties. Fictitious material properties may be used to
control the solution of the reference configuration, [44].
As mentioned above, external loads may complicate the initial equilibrium problem.
Therefore, it is assumed that the structural members are weight-less and that no
loads act at the nodes. However, for completeness the external forces will still be
present in many of the equations presented in this chapter, but the usual approach
is to set them to zero.
Initially, the only requirement put on the reference configuration is that it should
be in equilibrium. Consider a node i in a cable net where four cables meet, Fig-
ure 3.1. The equilibrium equations in the x-, y-, and z-directions at that node can
be expressed as:
x j − xi xk − xi x l − xi x m − xi
Tij + Tik + Til + Tim + Fxi = 0, (3.1)
Lij Lik Lil Lim
yj − yi y k − yi yl − yi ym − yi
Tij + Tik + Til + Tim + Fyi = 0, (3.2)
Lij Lik Lil Lim
zj − zi zk − zi zl − zi zm − zi
Tij + Tik + Til + Tim + Fzi = 0. (3.3)
Lij Lik Lil Lim

43
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

k
i

l m

y x

Figure 3.1: A connection in a cable net

Since the initial equilibrium is a static problem any configuration where equa-
tions (3.1)–(3.3) are satisfied at each node is a solution to the present problem.
Nevertheless, some solutions are better than others. The different methods used to
obtain these solutions will be described below. Their respective merits and draw-
backs will be brought to light. At the end of each section, the features of each
method are summarised.

3.2.1 The non-linear displacement method

Among the first computer methods applied to the solution of the initial equilibrium
problem was the non-linear displacement method, which is based on the large dis-
placement finite element technique used for analysis of structural behaviour under
external loads. As the same program can be used for both the initial equilibrium
problem and the load analysis, this approach is quite common. Nonetheless, there
are some serious disadvantages associated with this technique. This section will be
divided into two subsections: the first one dealing with cable nets and the second
one with membranes.
The non-linear displacement method may be summarised as follows. First, an ele-
ment mesh in equilibrium with a prescribed force distribution is established in the
horizontal plane. A three-dimensional form of the mesh is created by displacing the
support points almost vertically until they attain their prescribed positions, Fig-
ure 3.2. An iterative algorithm, e.g. the Newton-Raphson method, is used to obtain
the equilibrium configuration of the deformed structure.

44
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

Cable nets

Argyris et al. [4] were among the first to use the non-linear displacement method to
solve the initial equilibrium problem for cable nets. Their method was developed in
order to find the form of the cable roofs at the 1972 Olympics in Munich. Straight
bar elements were used to represent the cables.
A method similar to the non-linear displacement method has been used by Barnes [8].
This method is an application of the dynamic relaxation method, where an initially
out-of-balance structure is allowed to undergo damped vibrations until a steady
equilibrium shape is obtained.
The displacements of the fixed nodes may give rise to an unfavourable force dis-
tribution in the net, when actual material properties are used. Therefore, when
the fixed nodes have reached their final positions, a force adjustment procedure is
applied to the net. In this procedure the original unstrained lengths of the ele-
ments are recomputed in such a manner that the desired force values are obtained.
For a straight cable element satisfying Hooke’s law this is straightforward as the
total lengths of the elements before and after adjustment must be the same, i.e.
L0 + ∆L0 = L0 + ∆L0 . It leads to the following relation:

L0 + ∆L0 L0 + ∆L0
L0 = = . (3.4)
1+ 1 + T /AE

After this adjustment step the structure is no longer in equilibrium. Therefore,


some more iterations are needed to re-impose equilibrium. But, these iterations will
not change the final force distribution very much, so it will be close to the desired
one. Another way to keep control over the forces is to use a very small modulus of
elasticity for the cables, but then the control over the cable lengths is lost. With the
procedure outlined above control of both the forces and cable lengths is possible.

Figure 3.2: The principles of the non-linear displacement method. Reproduced


from [16]

45
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Membrane structures

In principle, the application of the non-linear displacement method to membrane


structures does not differ very much from the case of cable nets. Some general ap-
proaches used for finding the reference configuration of membranes will be discussed.
First, a finite element suitable for membrane representation has to be selected. Ow-
ing to the great geometric non-linearity of membrane structures, it is preferable to
use a dense mesh of primitive elements rather than a coarse mesh made up of higher
order elements [77, 115]. Then, a stress distribution has to be chosen. Generally,
two different approaches are used:

• the minimal surface approach, or

• the nonuniform stress approach.

The simplest choice for an initial equilibrium shape is the minimum surface config-
uration, which is characterised by a state of isotropic tensile stress. In order to find
the minimal surface, it is assumed that the flat membrane has a very small modulus
of elasticity and is in the isotropic prestressed state [115]. For other members in the
structure, such as beams or bars, actual material properties should be used [106].
Due to the small modulus used, the specified stresses in the membrane will only
slightly change, even though large deformations occur during the displacements of
the fixed nodes [115].
The advantages of the minimum surface are its aesthetically pleasing shape and
the associated uniform tensile stress. However, in some cases the minimal surface
configuration cannot satisfy all the architectural and structural requirements [115].
Since the mean curvature for minimum surfaces is zero, such surfaces are rather flat
and these have poor load bearing capacities. A nonuniform stress approach has to
be used. In this, a very small modulus of elasticity is still used for the membrane,
but as the name of the approach implies, the initial stresses are no longer specified
uniformly. Following the same procedures as for the minimal surface approach, the
final configuration should be in equilibrium with the nonuniform prestress. Several
trial calculations are usually needed to find a satisfactory equilibrium shape. Hence,
it is not obvious how to choose the nonuniform stresses [114].
For structures, where it is difficult to specify nonuniform initial stresses, an alterna-
tive approach can be used. This approach, which is based on elastic deformations,
is similar to that of Argyris et al. [4] for cable nets. In this approach, the actual
modulus of elasticity of the membrane is used. As for the cable net, the deformed
equilibrium configuration will have nonuniform and possibly large stresses. At this
stage, the stresses due to deformation are removed by a stress adjustment algorithm
and only the initial stresses are retained [115].
It was explained above that the way to keep the stresses within the elements con-
stant during displacement is to assign a very small modulus of elasticity to the
elements. However, in many cases the small elasticity has undesirable consequences,
such as numerical instability and divergence of the solution [65]. These problems

46
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

stem from the assumption of small strains made in the derivation of the membrane
elements. For an ill-chosen initial surface, i.e. in most cases a horizontal surface,
this assumption is often violated because gross changes in the element geometry
take place during the displacements of the fixed nodes. A way to avoid numerical
instability is to choose a mathematically defined initial surface close to the final
shape. Similar convergence problems were also reported in [36].
Another problem is that for surfaces which exhibit high curvatures, elements gather
in certain regions of the surface and leave the remaining regions represented to a
lesser accuracy. It is suggested that a suitable element arrangement in complex cases
should be chosen with the aid of a physical model [66].
Recently, Bletzinger [13] used a method called the updated reference strategy, which
is a numerical continuation method, to solve the initial equilibrium problem of mem-
branes with minimal surfaces. This technique had to be used because of the occur-
rence of a singular stiffness matrix, which excludes the use of the ordinary Newton-
Raphson algorithm. The stiffness matrix is singular when the nodal displacements
are tangential to the membrane surface. To understand that, consider a plane, which
obviously is a minimal surface. The surface area of that plane does not change if
the geometry of discretization is changed, e.g. by small tangential displacements
within the plane. Hence, the area variation of in-plane displacements is zero. A
special case of the updated reference strategy is the force density method. As the
updated reference strategy is claimed to be absolutely robust, it is perhaps the best
non-linear displacement method available for the initial equilibrium problem.
There are some drawbacks of the non-linear displacement method applied to the
initial equilibrium problem. Both the final shape and the stresses in the structure
are difficult for the designer to control. It is not an easy task to specify a desirable
force distribution [44]. If actual material values are used it is possible for some el-
ements in the structure to end up in compression [4]. The specification of material
properties (fictitious or real) represents unnecessary additional decision making for
the designer. In addition, the computations involved in this method are time con-
suming for large structures [44]. An advantage is that the program used to solve
the initial equilibrium problem can also be used for further load analysis.
The non-linear displacement method may be summarised as follows [44]. The vari-
ables specified by the designer are:

• structural topology,

• boundary conditions, and

• material properties.

The problem unknowns are:

• structural geometry, and

• internal force distribution.

47
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

The following additional constraint is placed on the solution:

• an initial force distribution may be specified.

3.2.2 The grid method

Other methods for solving the initial equilibrium problem have been developed to
overcome the problems associated with the non-linear displacement method. In
many of these methods, a variety of limitations are imposed on the solution to
transform the general non-linear problem into a linear one. The earliest method
of this type was developed for orthogonal cable nets by Siev and Eidelmann in
1962 [44].
Their method use the equations (3.1)–(3.3). By placing restrictions on the cable net
regarding the geometry, boundary conditions, and the internal stress distribution
the remaining vertical equilibrium problem becomes linear [108]. How this is ac-
complished will be shown below. Siev and Eidelmann assumed that the horizontal

∆l

∆l

∆l ∆l
∆l
∆l

Figure 3.3: Cable net with orthogonal horizontal projection

projection of the cable net is orthogonal, i.e. xi = xk = xm and yi = yj = yl , with


a grid size equal to ∆l, Figure 3.3. This gave the following modified equilibrium
equations in the x- and y-directions (with zero external loads):
∆l ∆l
Tij + Til =0 (3.5)
Lij Lil
∆l ∆l
Tik + Tim =0 (3.6)
Lik Lim

48
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

Noting that Tij ∆l/Lij and Til ∆l/Lil are the horizontal components of the cable
forces in the x-direction, and Tik ∆l/Lik and Tim ∆l/Lim the horizontal components
in the y-direction, one can see that the horizontal force is constant in those directions.
If the horizontal forces in the x- and y-directions at node i are denoted by Hix and
Hiy , respectively, equation (3.3) can be written as:

Hix (zj − 2zi + zl ) + Hiy (zk − 2zi + zm ) + Fiz = 0 (3.7)

If the horizontal components of the forces in the cables are specified, equation (3.7)
becomes linear, and the only unknowns are the z-coordinates of the free nodes.
Equation (3.7) is the discrete version of the vertical equilibrium equation of a shear-
free membrane [119]:
2 2
Hx ∂ z + H y ∂ z + Fz = 0, (3.8)
∂x2 ∂y 2
where H x, H
 y are the horizontal components of the prestressing force distribution
(N/m) in x- and y-directions, respectively. Fz is the vertical (z-direction) load
intensity (N/m2 ).
The grid method may be summarised as follows [44]. The variables specified by the
designer are:

• structural topology, and

• boundary conditions.

The problem unknowns are:

• structural geometry, and

• internal force distribution.

The following additional constraints are placed on the solution:

• limited to line cable elements,

• constant horizontal force along cables, and

• limited to cable nets with straight-line plan projections.

3.2.3 The force density method

A linear solution to the initial equilibrium problem was derived in section 3.2.2 for
orthogonal cable nets. However, because of the restrictions placed on the structure
in that method, the resulting shapes are few. In this section, a more general method,
called the force density method, will be presented.

49
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

The force density method is a strategy to solve the equations of equilibrium for a
cable net without requiring any initial coordinates for the structure, just by taking
advantage of a mathematical trick, [43]. Consider the equilibrium equations (3.1)–
(3.3). These equations are non-linear since the element length L is a function of
the node coordinates. If, instead of the element forces, the force-to-length ratios
(denoted by q) for each element are specified (3.1)–(3.3) can be written as:

qij (xj − xi ) + qik (xk − xi ) + qil (xl − xi ) + qim (xm − xi ) = 0, (3.9)


qij (yj − yi ) + qik (yk − yi ) + qil (yl − yi ) + qim (ym − yi ) = 0, (3.10)
qij (zj − zi ) + qik (zk − zi ) + qil (zl − zi ) + qim (zm − zi ) = 0. (3.11)

It is obvious that the main advantage of using the force densities as description
parameters for a cable net is that any state of equilibrium can be obtained by
the solution of one system of linear equations. The equilibrium state so obtained
has the prescribed force density in each element. No other conditions, such as
equidistant meshes or constant element forces, are fulfilled. For a first impression
of the shape of the structure it is not necessary to consider auxiliary constraints,
but in a detailed analysis these have to taken into account. For this purpose a non-
linear displacement method or the extended non-linear force density method can be
used. The difference between these methods is that for the non-linear displacement
method described in section 3.2.1 the number of equations is equal to the number of
degrees of freedom, while for the non-linear force density it is equal to the number
of additional constraints. In most cases the number of constraints is much smaller
than the number of degrees of freedom [105].
Mollaert [75] applied the force density method to structures composed of both cables
and compression members. To obtain a solution out of the plane of the fixed nodes
the tensile and compression parts of the structure were separated. At the common
nodes the removed part was replaced by external forces. Both parts were then
designed separately.
In [77] the force density method was used together with a least squares minimisation
approach presented in [43] to generate the cutting pattern for membrane structures.
Although the problem of determining the membrane cutting pattern is outside the
scope of this thesis it should be mentioned that the force density method can be
used to solve also this problem. Due to the simple formulations of the force density
method and the least squares minimisation technique, a solution can be obtained
in short time although a very fine mesh is used. These properties make the force
density method a better choice than other methods, such as the dynamic relaxation
method, at the patterning stage in the design of fabric structures [77].
As presented in [105] the force density method is limited to line cable elements. In
reference 44 an extended version of the force density method, with curved cable
and membrane elements, is presented. The method is based on assumed geometric
stiffness matrices
KG xg = 0, (3.12)
where KG is the geometric stiffness matrix of the structure and xg the vector of
nodal coordinates (x-, y- and z-coordinates). Equation (3.12) can be applied to any

50
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

finite element structural model. Although (3.12) has the form of a standard stiffness
equation, the unknowns are the nodal coordinates rather than nodal displacements.
For structures composed of only straight bar elements, the set of equations in (3.12)
are identical to the corresponding equations in the force density method. Neverthe-
less, if the choice of suitable force densities was quite easy, it is much more difficult
to choose the geometric stiffness matrices. For simple elements closed form expres-
sions can be established, but for many elements the matrices have to be found by
numerical integration. Even after the geometry has been solved, the determination
of stresses for complex elements can be a problem [44].
Christou [24] implemented an elastic catenary element in the force density method
to be able to take into account loads distributed along the cables. After the shape is
found the force in each cable has to be found by iteration since the horizontal force is
described by a non-linear equation. However, refinement of the force density method
to take into account distributed loads has less importance at the ‘form-finding’ stage
since the loads are often neglected to simplify the problem.
More recently Lai et al. [59] used the force density method to find the form of a
deployable reflector for space applications. They transformed the original membrane
into an equivalent cable network and could, therefore, use the original equations of
the force density method. This work shows that although the force density method
was developed back in 1971 it finds new areas of application.
The force density method may be summarised as follows [44]. The variables specified
by the designer are:

• structural topology, and

• boundary conditions.

The problem unknowns are:

• structural geometry, and

• internal stress distribution.

The following additional constraints are placed on the solution:

• limited to line cable elements, and

• force density prescribed for each element.

3.2.4 Least squares stress determination methods

In all of the methods above the structural geometry is one of the problem unknowns.
For structures where the geometry for some reasons is known the cable forces to

51
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

satisfy equilibrium has to be determined. In this section two methods that are
appropriate for such cases will be presented. Both methods are derived from

At = f , (3.13)

which is the matrix form of (3.1)–(3.3). Three cases can occur for (3.13) of which
the following two are especially interesting: the over- and the under-determined
cases [44]. The third case corresponds to a square matrix A.
In the over-determined case there are more equilibrium equations than unknown
cable forces. Because there is no exact solution to such a set of equations, it is
necessary to seek an optimal set of forces which will only approximately equilibrate
the loads on the structure. The optimal set of cable forces is selected by the least
squares method
AT At = AT f . (3.14)
It should be emphasised that this method satisfies equilibrium in a least squares
sense only. A disadvantage of this method is that the designer does not have much
control over the force pattern. There is no restriction against compressive forces
and the distribution of forces may be highly irregular. Some force control may be
obtained by prescribing some of the forces. A major advantage of the method is
that the solution is obtained by solving a set of symmetric linear equations [44].
The under-determined case occurs when there are more unknowns than equilibrium
equations (membrane structures often fall into this category). For this case, there
exists an infinite number of exact equilibrium solutions for the forces. Therefore, an
ideal force distribution t∗ has to be defined and solved for. Generally, these ideal
forces will not satisfy equilibrium. The actual forces are expressed in terms of the
ideal forces and a set of deviations from the ideal force values,

t = t∗ + ∆t. (3.15)

Since the ideal forces are specified directly, the force deviations ∆t becomes the
problem unknown. Equation (3.13) can now be written as:

A∆t = f − At∗ . (3.16)

The optimal solution to (3.16) is defined by the set of force deviations that have the
smallest Euclidean norm. This optimal solution is found by solving the following
minimisation problem with Lagrange multipliers:

∆tT ∆t − 2kT [A∆t − (f − At∗ )] → min . (3.17)

The solution to (3.17) is


 −1
∆t = AT AAT (f − At∗ ) . (3.18)

The actual forces, obtained from (3.15), should satisfy equilibrium exactly. Since the
force distribution has a minimal variation from the specified ideal forces it should be
fairly smooth. However, large force deviations may occur if the structural geometry

52
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

and the prescribed force distribution are incompatible. An advantage of the under-
determined least squares method is that the designer is given some control over
the force distribution while the geometry may be specified exactly. As in the over-
determined case, the solution procedure only involves linear symmetric matrices [44].
The least squares stress determination methods may be summarised as follows, [44]:
the variables specified directly by the designer are:

• structural topology,

• boundary conditions, and

• structural geometry.

The problem unknown is

• internal force distribution.

3.2.5 A combined approach

No single solution method is optimal for all problems. It is possible to use several
of the better solution methods in combinations to create more flexible design tools.
A combined approach lets the designer experiment with various methods to find the
optimal solution. Approximate results from one solution method may be used as
input data to another method to obtain an improved solution [44].
For cable structures it seems that the best strategy is to first use the force density
method, which uses linear cable elements, and then take this solution to a non-linear
finite element program, which uses more refined cable formulations (see Chapter 4).
Since the solution obtained by the force density method is approximate, but still
very good, only a few iterations should be needed to find the ‘true’ equilibrium
configuration.

3.2.6 Initial equilibrium of tensegrity structures

The most interesting class of space structures is that of the self-stressed systems,
called tensegrity systems. Basically, tensegrity systems are composed of two sets of
elements, a continuous set of cables, and a discontinuous set of rectilinear struts [81].
The self-stress makes these structures rigid without requiring any support to bal-
ance the stresses [80]. This property makes them very interesting from an economical
point of view. Concerning initial equilibrium configurations of tensegrity structures,
much research has been done on geometrical basis [81]. However, geometrical meth-
ods do not guarantee mechanical equilibrium and the solutions have to be checked
by numerical techniques.

53
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Pellegrino and Calladine have in a number of publications, [18–21, 92, 93], analysed
statically and kinematically indeterminate frameworks with known geometries. In
their analyses they used the equilibrium equation (3.13), repeated here,

At = f , (3.19)

and the compatibility equation


Bd = e. (3.20)
By considering the four subspaces of the equilibrium matrix A (= BT ) useful infor-
mation concerning the structure is obtained. The essence of their method is given
in Figure 3.4 and Table 3.1. From a structural view, the perhaps most important
result from their studies is that only some kinematically indeterminate frameworks
with more than one independent state of self-stress (s > 1) can be stiffened to the
first-order by a single state of prestress. This is crucial, because tensegrity systems
and cable nets must have first-order stiffness in all possible modes [21] to prevent
excessive displacements, which may lead to collapse.
In most cases, the geometry of a tensegrity structure is unknown. To find the self-
stressed configuration of a general tensegrity structure has proven to be much more
difficult than it is for cable and membrane structures, according to the few works
dealing with this problem.

Table 3.1: Four different types of structural assemblies. From [91].


Assembly type dim N (A) Static and kinematic features
and
dim N (AT )
I Statically determinate and s=0 Both (3.19) and (3.20) have a
kinematically determinate m=0 unique solution for any right
hand side (r.h.s.).

II Statically determinate and s=0 (3.19) has a unique solution for


kinematically indeterminate m>0 some particular r.h.s., but other-
wise no solution. (3.20) has an
infinite number of solutions for
any r.h.s.

III Statically indeterminate and s>0 (3.19) has a infinite number of


kinematically determinate m=0 solutions for any r.h.s. (3.20)
has a unique solution for some
particular r.h.s., but otherwise.
no solution.

IV Statically indeterminate and s>0 Both (3.19) and (3.20) have an


kinematically indeterminate m>0 infinite number of solutions for
some particular r.h.s, but other-
wise no solution.

54
3.2. LITERATURE REVIEW OF INITIAL EQUILIBRIUM SOLUTION METHODS

Dim. Equilibrium A Compatibility B


Row space R(AT ): Column space R(B):
bar tensions in compatibility bar
r equilibrium with the = elongations
loads in the column
space.
Bar space
Rb ⊥ ⊥

Null-space N (A): Left null-space N (BT ):


states of self-stress. incompatible bar
s (Solutions of At = 0) = elongations.

Column space R(A): Row space R(BT ):


loads which can be extensional
r equilibrated in the = displacements.
initial configuration.

Joint space
⊥ ⊥
R3j−c
Left null-space N (AT ): Null-space N (B):
loads which cannot be inextensional
m equilibrated in the = displacements.
initial configuration (Solutions of Bd = 0)

Figure 3.4: The four fundamental subspaces associated with the equilibrium matrix
A and the compatibility matrix B(= AT ). The sign ‘=’ indicates that
the two subspaces coincide, while ‘⊥’ indicates that they are orthogonal
complements of one another (note that s = b − r and m = 3j − c − r).
Redrawn from [91].

Hanaor [45] used a non-linear displacement method based on Newton-Raphson iter-


ations to find the form of double layer tensegrity dome. He also stated, contrary to
what is given in this chapter and in [82], that the initial equilibrium problem is kine-
matic [45]: “The assumed geometry is, in general a mechanism, when constraints
on strut and tendon elongations are considered. Shape finding consists essentially
of activating the mechanisms, until a state is reached when only elastic deforma-
tions are possible. This is the prestressable geometry.” But, an algorithm for the
kinematic formulation was not presented in reference 45.

55
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Motro et al. [82] used both the dynamic relaxation method and the force density
method to find the form of simple tensegrity systems. The results were compared to
analytical solutions. The dynamic relaxation technique worked well for systems with
few nodes but suffered from convergence problems when the number of nodes was
increased. Although the force density method was applied only to small tensegrity
structures, Motro et al. anticipated that the force density method could be conve-
nient for the initial equilibrium solution of more complicated tensegrity systems [82].

3.3 The force density method

As shown in section 3.2.3, the force density method is popular among space structure
researchers. The method was developed by Linkwitz and Schek [71] for the initial
equilibrium problem of the cable roofs at the 1972 Olympic Games in Munich. It
was first published in [71] and later extended in [72] and [105]. The procedure
in this section will mainly follow that in [105] but some additional comments and
explanations will be given. Throughout this section the following notation will be
used: V = diag (v), where v is a vector.

3.3.1 The linear force density method

In the force density method it is assumed that the cables are straight and pin-jointed
to each other or to the supporting structure [105]. First, a graph of a network is
drawn and all nodes are numbered from 1 to ns , all elements from 1 to m. The nf
nodes which are to be fixed points are taken at the end of the sequence. All the
other n nodes are free. Thus, the total number of nodes is ns = n + nf . Then the
connectivity matrix Cs is constructed with the aid of the graph. Each element j
has the node numbers k and l (from k to l). The connectivity matrix Cs for the
structure is define by (i = 1, 2, ..., ns ):


+1 for i = k,
cs (j, i) = −1 for i = l, (3.21)


0 in the other cases.
This connectivity matrix can be divided into two matrices
 
Cs = C Cf , (3.22)
where C and Cf contains the free and fixed nodes, respectively. Denoting the vectors
containing the coordinates of the n free nodes x, y, z, and similarly for the nf fixed
nodes xf , yf , zf , the coordinate differences for each element can be written as:
u = Cs xs = Cx + Cf xf , (3.23)
v = Cs ys = Cy + Cf yf , (3.24)
w = Cs zs = Cz + Cf zf . (3.25)

56
Table 3.2: Summary of initial equilibrium solution methods (‘∗’ designer specified, ‘?’ problem unknown, ‘−’ not included). From [44].
Structural External Boundary Structural Internal Material
Description Notes
topology loads conditions geometry forces properties
The non-linear ∗ ∗ ∗ ? ? ∗ A–G
displacement method
The grid method ∗ ∗ ∗ ? ? − H–K
The force density ∗ ∗ ∗ ? ? − H, L
method
The over-determined ∗ ∗ ∗ ∗ ? − F, M–O
least squares stress
method
The under-determined ∗ ∗ ∗ ∗ ? − B, N
least squares stress
method

57
Decription of notes (related to zero external loads)
A: Requires trial surface geometry.
B: Requires trial stress distribution.
C: Requires material properties.
D: Zero elastic modulus gives a configuration in equilibrium with precribed forces.
E: Large elastic modulus gives control over element length.
F: Local compression may occur.
G: Expensive nonlinear solution.
H: Limited to line cable elements.
I: Constant horizontal force along cables.
J: Limited to structures with orthogonal straight-line horisontal projections.
K: Edges must be fixed (or rigid).
L: A force density must be prescribed for each element.
M: Equilibrium solution is approximate.
N: Stresses may be constrained.
O: The force distribution may be irregular.
3.3. THE FORCE DENSITY METHOD
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

The equilibrium equations for the free nodes for the x-, y- and z-directions are
written as:

CT UL−1 t = fx , (3.26)
CT VL−1 t = fy , (3.27)
CT WL−1 t = fz . (3.28)

By using the force-to-length ratios for the elements, i.e. the force densities, as de-
scription parameters, (3.26)–(3.28) are written as:

CT Uq = fx , (3.29)
CT Vq = fy , (3.30)
T
C Wq = fz , (3.31)

where the vector q, of length m, is described as:

q = L−1 t. (3.32)

Using (3.23)–(3.25) and the following identities:

Uq = Qu, (3.33)
Vq = Qv, (3.34)
Wq = Qw, (3.35)

equations (3.29)–(3.31) are written as:

CT QCx + CT QCf xf = fx , (3.36)


CT QCy + CT QCf yf = fy , (3.37)
CT QCz + CT QCf zf = fz . (3.38)

By setting D = CT QC and Df = CT QCf , equations (3.36)–(3.38) can be written


as:

Dx = fx − Df xf , (3.39)
Dy = fy − Df yf , (3.40)
Dz = fz − Df zf . (3.41)

Equations (3.39)–(3.41) are solved using elementary algebra:

x = D−1 (fx − Df xf ) , (3.42)


y = D−1 (fy − Df yf ) , (3.43)
z = D−1 (fz − Df zf ) . (3.44)

Two cases for matrix D can occur [82, 105]:

1. Determinant of D = 0
The matrix D has full rank and the form of the structure is governed by the

58
3.3. THE FORCE DENSITY METHOD

values chosen for the force densities. In the case of a prestressed cable net with
a given connectivity (i.e. fixed C and Cf ) the number of equilibrium shapes
is identical to the number of vectors q. This justifies the use of the force
densities as description parameters for a cable net. Attention must be paid to
the specific case which occurs when all fixed nodes are coplanar, because then
the solution will also be planar and without interest (see section 3.4.4).

2. Determinant of D = 0
The system can be solved for x only when vectors Df xf lie is the space spanned
by the linearly independent vectors of matrix D (if the external loads are zero).
Similarly for y and z.

To illustrate the properties of the linear force densities a simple example will now
be given. Consider the structure in Figure 3.5 with all fixed nodes in the x–y plane.

6
1 2

1 2

3 4 5 6

3
7 8
8 9
7 10
4 5
y
11 12

9
x
z

Figure 3.5: A simple cable structure with zero external loads. The arrows indicate
the directions of the elements.

59
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

The connectivity matrix Cs for this structure with, first, n = 5 free nodes and, then,
nf = 4 fixed nodes, giving ns = 9 is written as (elements left out are zero):
 
1 −1
 
 1 −1 
 
 
 1 −1 
 
 1 −1 
 
 
 −1 
 1 
 
 1 −1 
 
Cs =  .
 1 −1 
  (3.45)
 
 1 −1 
 
 
 1 −1 
 
 
 1 −1 
 
 −1 
 1 
1 −1
    
C Cf

The matrices D and Df describe the equilibrium at the free and fixed nodes, re-
spectively, and are written as:
 
q1 + q3 + q4 0 −q4 0 0
 
 0 q2 + q 5 + q 6 −q5 0 0 
 
 
D= −q4 −q5 q4 + q5 + q8 + q9 −q8 −q9 
 
 −q8 
 0 0 q7 + q8 + q11 0 
0 0 −q9 0 q9 + q10 + q12
(3.46)
and  
−q1 −q3 0 0
 
 −q2 0 −q6 0 
 
 
Df =  0 0 0 0 . (3.47)
 
 0 −q −q11 
 7 0 
0 0 −q10 −q12
Figure 3.6 shows the resulting shapes of this structure for different force density
values. It is seen that for a single element an increase in the force density relative to
the others results in a contraction of that element. The opposite holds for a decrease
in the force density, even more emphasised with negative values.

60
3.3. THE FORCE DENSITY METHOD

(a) Elements 1–12 have q = 1 (b) Elements 1–3, 5–12 have q = 1 and
element 4 has q = 10

(c) Elements 1–3, 5–12 have q = 1 and (d) Interior elements have q = 1 and edge
element 4 has q = −0.1 elements have q = 5

Figure 3.6: Different equilibrium configurations for the plane structure in Figure 3.5.

3.3.2 The non-linear force density method

Multiple equilibrium shapes can be obtained with the linear force density method.
However, these shapes may be unsatisfactory from a structural point of view. For
example, the mesh may be irregular and the force distribution unsmooth. There-
fore, it is necessary to find a configuration which is in equilibrium and which also
satisfies additional conditions. These conditions are generally non-linear and so is
the extended force density method. It is preferred to start the non-linear computa-
tions using the shape found with the linear method. In contrast to the non-linear

61
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

displacement method, the number of non-linear equations here is identical to the


number of additional conditions and independent of the number of nodes. Thus,
this non-linear approach is more efficient than the non-linear displacement method.
The additional constraints are put in the following form [105]:

g (x, y, z, q) = 0, (3.48)

where it is assumed that the constraints are functions of the coordinates and force
densities. Since the coordinates also are functions of the force densities, (3.48) is
written as:
g∗ (q) = g(x(q), y(q), z(q), q) = 0. (3.49)
Equation (3.49) is generally non-linear and has to be linearised to be solvable

∂g∗ (q0 )
g∗ (q0 ) + ∆q = 0. (3.50)
∂q

Equation (3.50) can be written as:

GT ∆q = r (3.51)

where
∂g∗ (q0 )
GT = (3.52)
∂q
and
r = −q∗ (q0 ). (3.53)
Equation (3.51) is similar to the non-linear Newton-Raphson equation used in finite
element analysis. In many cases the number of constraints is less than the number
of elements, i.e. m > r. This means that (3.51) is under-determined and has m − r
linearly independent solutions. From these solutions a single solution can be chosen
by considering the following variational principle [105]:

∆qT ∆q → min . (3.54)

A Lagrange multiplier method is used to solve for the force density correction ∆q.
The variations of the following functional:

M = ∆qT ∆q − 2kT GT ∆q − r (3.55)

give the equations


1 ∂M
= ∆q − Gk = 0, (3.56)
2 ∂∆q
1 ∂M
= GT ∆q − r = 0. (3.57)
2 ∂k
The solution to (3.56) and (3.57) is given by (3.62)–(3.64). A damped version
of (3.51) is
GT ∆q = r + ω. (3.58)

62
3.3. THE FORCE DENSITY METHOD

The minimum principle associated with (3.58) is


∆qT ∆q + ω T Pω → min . (3.59)
Another approach, based upon modified damping, is
GT ∆q = Ωr. (3.60)
The associated minimum principle of (3.60) is
∆qT ∆q + (i − ω)T P (i − ω) → min . (3.61)
where iT = (1, 1, ..., 1). According to Schek, [105], the approach given by (3.61) is
useful if there are large changes in the force densities because the damped iterations
converge without oscillation. Another advantage of this approach is that for con-
straints which cannot be satisfied the iterations may stop at a shape which fulfils
the constraints as closely as possible. The Lagrange factors are given by
k = T−1 r, (3.62)
where 
 T
G G in case (3.51), (3.54),
T = GT G + P−1 in case (3.58), (3.59), (3.63)

 T
G G + P−1 R2 in case (3.60), (3.61).
The solution to the minimisation problem is
∆q = Gk. (3.64)
For the subsequent iterations, a new q is computed as:
q1 = q0 + ∆q, (3.65)
until convergence within a given tolerance.
In [105] the following additional constraints are considered:

• node distance,
• element force, and
• unstrained length.

The Jacobian matrix GT for these constraints will now be given.

Node distance constraints

If elements with very large axial stiffnesses are used this kind of constraint may
arise. This condition, with r prescribed node distances ls , is written as:
gd = l − ls . (3.66)
The Jacobian matrix for the node distance constraint is written as:
−1 
GTd = −L U C D−1 CT U + V C D−1 CT V + W C D−1 CT W . (3.67)
This matrix is of dimension r × m.

63
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Force constraints

In many cases it is desirable to specify the forces for some elements. The force
condition, with r prescribed forces tv , is written as:
gf = L q − tv = Q l − tv (3.68)
The Jacobian matrix for this case is
−1 
GTf = L − Q L U C D−1 CT U + V C D−1 CT V + W C D−1 CT W . (3.69)
In [105] no further comment to this equation is given. However, inspection of (3.69)
reveals that the subtraction cannot be performed as the matrix L is of dimension
r × r and the dimension of the resulting matrix after the minus sign is r × m. This
problem can be solved if matrix L is expanded with m − r zero columns at the
positions corresponding to the numbers of the unconstrained elements. Then the
subtraction can be performed.

Unstrained length constraints

For fabrication purposes it may be necessary to prescribe the unstrained lengths of


some elements (see also section 3.4.1). This condition, with r unstrained lengths
luv , is written as:
gu = lu − luv . (3.70)
For a stressed straight cable element i, which satisfies Hooke’s law, the unstrained
length is computed as:
AEi
lui = li . (3.71)
AEi + Ti
The Jacobian matrix for this unstrained length constraints is
2 −1 2 −3 
GTu = −Lu K −Lu L U C D−1 CT U
(3.72)
+ V C D−1 CT V + W C D−1 CT W .
As for the force constraints, the subtraction to obtain GTu cannot be performed
due to different dimension of the matrices involved. The remedy is the same as
2 −1
above; after −Lu K is computed the resulting matrix is expanded with m − r zero
columns. After considering the following limit case:
lim GTu = GTd (3.73)
AE→∞

a misprint in the expression for GTu was detected in [105].

Mixed constraints

In many cases one may want to take into account more than one of the constraints
given above. This can be accomplished by writing the total Jacobian matrix as:
 
G = Gd Gf Gu . (3.74)

64
3.4. EXAMPLES

Practical experience has shown that convergence is very slow if the magnitude of
the numerical force values are much larger than the numerical values for the lengths.
This occurs when an element is assigned with both a force and a length constraint.
Nevertheless, the convergence rate can be significantly improved if the force values
are scaled down (divided by a scale parameter) so that they become of the same
order as the lengths. No theoretical analysis has been done to justify this scaling
procedure, but it has worked very well for the examples given in the next section.

3.4 Examples

Until now mainly theoretical issues concerning form-finding have been discussed.
The force density method will be described by some additional examples in this
section. Some of the examples that will be presented below are chosen to highlight
problems that need special techniques and some just to show the versatility of the
force density method in finding the shape of different cable net structures. Finally,
the applicability of the method for structures including both cables and struts will
be checked. It should be emphasised that the aim of the following examples is
to investigate the behaviour of the force density method when applied to different
types of cable structures. Numerical stability and convergence rate are in particular
studied. No load bearing or architectural aspects are taken into account. In many of
the examples the dimension of the structure and internal forces are chosen arbitrarily.

3.4.1 Smaller cable nets

To construct a cable net one must know how the cables are best arranged. Generally,
two arrangements are worth considering:

• Geodesic mesh, in which the cables run along the geodesic lines in the surface.
A geodesic line is the shortest way between two points on a surface. This
approach minimises the use of material but the manufacturing can be quite
complicated.

• Uniform mesh in the unstrained state. From a constructional point of view


this approach is the best one. As, an error in length of 0.1 % can give rise to an
error in force of about 50 %, accurate placing of the connections is crucial [63].
An equidistant mesh enables the mounting to be done in a factory. At the
building site, the net can be assembled on the ground and hoisted into position.
Figure 3.7 shows an example from [72], where certain elements are assigned
with a constant unstrained length.

65
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

4 5

Figure 3.7: The uniform mesh approach. Elements on the sides of non-shaded areas
should have equal unstrained lengths. Redrawn from [72].

Table 3.3: Coordinates in metres for the fixed nodes of the structure in Figure 3.7.
Coordinates
Node
x y z
1 0.0000 0.0000 0.0000
2 −6.4897 2.2285 3.0000
3 −6.9932 9.3859 0.0000
4 −2.4917 11.7810 4.0000
5 3.8300 11.5901 0.0000
6 7.0874 4.2153 5.0000

Considering the magnitude of the prestressing force, the usual rule of thumb is that
the magnitude should be such that no element goes slack under any load condition.
However, according to Leonhardt and Schlaich [63] this rule “causes unacceptably
high forces.” They further conclude that “it will never be possible to establish
general rules for finding the shape of cable net structures.” This is due to the
many factors which affect the shape and load-bearing characteristics of a cable net.
Among the factors are: prestress magnitude and distribution, mesh geometry of the
net, edge rigidity (concrete ring or edge cables), angle between net cables and edge
cables and stiffness ratios of the members [63].
In general, the magnitude of the prestressing force is determined by the allowable
deformations and fatigue strength of the cables. However, an increase in prestressing
force is not as effective in reducing the deformations as an increase in curvature of the
net [63]. Of course, the distribution of the prestressing force must be quite uniform.

66
3.4. EXAMPLES

Several configurations have to checked before finding one which best satisfies all
requirements put on the structure.
A common approach for prestressed cable nets is given in reference 72. In this
approach, the interior cables are assigned with unstrained length constraints and
the elements in contact with the edges are assigned with force constraints. In this
way a good compromise between load-bearing behaviour and construction ease is
reached. A special problem which may arise when using the outlined technique is
that the elements connected to the edge cables may have too large angle changes,
Figure 3.8. Since the real cable is continuous and has a finite bending stiffness,
such angles cannot occur in practice. A technique to avoid these angle changes is to
assign force constraints also to the interior cables. For a three-dimensional structure
a configuration that satisfies all the constraints exactly may not be found. But if
the modified damped version of the non-linear force density method is used, the
iterations stop at a satisfactory shape.
To find the shape of the structure in Figure 3.7, the following procedure was used:

1. All interior cable were assigned with a force density equal to 200 kN/m. The
force density for the edge cables was 1200 kN/m. The interior cables had the
following constraints: cable force equal to 200 kN and unstrained length of 1
m. No constraints were assigned to the edge cables. The heights of all the
fixed nodes are changed to z = 0. Thus, the cable net lies in the x–y plane.
The stiffness of the cables was AE = 100 MN.

2. With the prescribed force densities, the linear force density method gave the
shape shown in Figure 3.8. The net has a nearly square mesh, but some large
distortions occur near the edges.

3. With all the fixed nodes still in the same plane, 20 iterations with the non-
linear force density method gave a fairly smooth layout without large angle
changes near the edges. Note that this configuration does not satisfy the
constraints. This step is an intermediate step to get rid of the irregularities in
the edge areas and get a nearly equidistant interior net with a uniform force
distribution.

4. In this, the last step, all the fixed nodes have their original positions given by
Table 3.3. The force constraints for all interior cables assigned with unstrained
length constraints are removed. Only interior cables connected to the edge ca-
bles still have force constraints. The reason for this modification is that for a
three-dimensional cable net both force and unstrained length constraints can-
not generally be satisfied within the net. However, for the plane configuration
in step 3 it is possible to satisfy both constraints. To obtain the final shape
shown in Figure 3.9 required 19 iterations.

The reason for the somewhat slow convergence of the non-linear force density method,
if one compares with the Newton-Raphson technique used in finite element analy-
ses, is probably due to the highly distorted meshes in some parts of the net. If

67
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

large changes in the force densities of only a few elements are needed to satisfy the
constraints, slow convergence follows.

Figure 3.8: Too large angles of the cables occur near the edges if only unstrained
length constraints are used for interior cables. Dashed line = desired
position of the cable. Redrawn from [72].

Figure 3.9: Three dimensional view of the equidistant cable net with smooth force
distribution.

68
3.4. EXAMPLES

The next example will show that different final configurations will be obtained de-
pending on the initial force density values of the edge cables. In the first example the
target shape is an interior mesh width of 1 metre and a prestressing force of 200 kN
for each of the cables in the edge area. The stiffness of the cables was AE = 1000
MN. For both nets the starting value of the force density for each interior cable
was qinterior = 200 kN/m. For the first net, Figure 3.10, the force density in each
of the edge cables was qedge = 5qinterior and for the second net, Figure 3.11, it was
qedge = 50qinterior . The same procedure as for the previous example was used. To
obtain the plane net, with force constraints for all interior cables, 8 iterations were
required for net 1 and 7 iterations for net 2. This difference is probably due to the
fact that the starting shape of net 2 is closer to the final shape, see Figure 3.11(b).
As above, the final three-dimensional configuration is obtained by removing the force
constraints from interior cables not connected to the edge and fixing the support
nodes in their original positions. For both nets the final shape was obtained with
only 4 iterations. The solutions show that, depending on the starting values of the
force densities in the edge cables, which are unconstrained, different configurations
and force values are obtained.

69
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

(a) Initial configuration. qinterior = 200 (b) Final configuration.


kN/m and qedge = 1000 kN/m.

( 6,6,4 )

3.976

4.025
3.858

( 0,12,0 ) 3.944
3.789
(−6,6,4 )
3.916
3.789
3.944
3.858
4.025

3.976

( 0,0,0 )

(c) Final configuration. Coordinates of fixed points (m) and edge


cable forces (MN).

Figure 3.10: Hyperbolic paraboloid net 1.

70
3.4. EXAMPLES

(a) Initial configuration. qinterior = 200 (b) Final configuration.


kN/m and qedge = 10000 kN/m.

( 6,6,4 )

6.314
6.378
6.217

6.317
( 0,12,0 )
6.165
(−6,6,4 )
6.297
6.165
6.317

6.217
6.378

6.314

( 0,0,0 )

(c) Final configuration. Coordinates of fixed points (m) and edge


cable forces (MN).

Figure 3.11: Hyperbolic paraboloid net 2.

71
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

One of Frei Otto’s most famous membrane structures is the star-shaped pavilion over
the dance fountain in Cologne. It was built for the 1957 Federal Garden Exhibition.
The tent is still standing, although it was planned for a single summer [87]. This
structure inspired the author in the next example—a star-shaped cable net. The net
vertices lie on two circles with radii 14.421 m and 8.165 m, respectively. The height
difference between the vertices is 4 m. All cables have the axial stiffness AE = 100
kN. The starting values of the force densities are: 200 N/m for the interior cables,
2000 N/m for the valley cables, 6000 N/m for the ridge cables and 1000 N/m for
the edge cables. An unstrained length of 1.8 m are assigned to all net cables that
are perpendicular to the ridge cables. Force constraints are assigned to all interior
cables (not ridge or valley cables) and ridge cables. The force values are 200 N and
8000 N, respectively. To obtain the final shape, shown in Figure 3.13, 10 iterations
were required.

Figure 3.12: The pavilion over the Cologne Dance Fountain. Reproduced from Ar-
chitectural Design, No. 117, “Tensile Structures”, 1995.

72
3.4. EXAMPLES

(a)

(b)

Figure 3.13: A star-shaped cable net structure inspired by Frei Otto’s pavilion over
the Cologne Dance Fountain.

3.4.2 A large cable net

The previous examples, which had simple, symmetric configurations, show that the
outlined form-finding procedure gave the desired shapes. In this section, a more
complicated cable net will be analysed. The cable net of the German pavilion at
the 1967 World’s fair in Montreal, Figure 3.14, inspired the author in the layout of
the present example. As in the previous examples, a plane connectivity mesh with
equidistant width is first drawn, Figure 3.15. Then the elements are assigned with
force density values of different magnitudes so that the initial plane shape resembles
that of the drawn mesh. The initial values of the force densities are: 67 kN/m for
net cables, and 667 kN/m for edge cables (including the large cables within the

73
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

net, illustrated in Figure 3.15). The target shape is an unstrained mesh width of
3 m and a force of 200 kN in each of the net cables connected to the edges. The
stiffness of each cable is 100 MN. Since the initial shape, shown in Figure 3.16(a),
is highly distorted in some parts, more than 100 iterations were required to get a
good plane initial shape. Nonetheless, a three-dimensional shape, which satisfies
the constraints, could not be obtained. The reason for this is that the heights of
the fixed nodes are large relative to the other dimensions of roofs. Thus, an initial
equidistant mesh in plane is incompatible with an equidistant mesh of the three-
dimensional structure. In some cases it might be possible to draw a slightly finer
mesh width than the target mesh width in space. However, this problem is not
easily solved since quite detailed knowledge of the final shape of the net is needed.
One way to overcome this problem is to use a physical model to calculate the correct
number of cables. For cable nets with smaller height-to-length ratios an equidistant
initial mesh will work.
Even though the desired shape could not be found, a three-dimensional shape was
obtained by using the force densities of the plane structure in Figure 3.16(b) and the
original coordinates for the fixed points. As shown in Figure 3.17 the ‘final’ shape
has a smooth cable arrangement.

Figure 3.14: Aerial view of the German pavilion. Reproduced from Architectural
Design, No. 117, “Tensile Structures”, 1995.

74
3.4. EXAMPLES

12
11

10 13
1

4
2

14
9 3

5
8
6

Figure 3.15: Mesh for a large cable net.

Table 3.4: Coordinates in metres for the fixed nodes of the large cable net in Fig-
ure 3.15.
Coordinates
Node
x y z
1 0.000 52.500 10.000
2 10.893 42.188 0.000
3 0.000 22.500 15.000
4 40.893 42.188 20.000
5 44.126 12.362 0.000
6 15.910 6.590 0.000
7 0.000 0.000 0.000
8 −18.106 9.142 0.000
9 −22.500 22.500 0.000
10 −14.317 56.974 0.000
11 −1.874 67.382 0.000
12 45.028 71.901 0.000
13 67.218 56.574 0.000
14 66.015 25.790 0.000

75
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

(a) Initial plane configuration

(b) ‘Final’ plane configuration

Figure 3.16: Equilibrium configurations in the horizontal plane.

76
3.4. EXAMPLES

Figure 3.17: Three-dimensional views of the ‘final’ configuration of the large cable
net.

3.4.3 Cooling towers

Cooling towers enable a nuclear power plant to be built at any location independent
of natural water supplies. The power station capacity is dependent on the cooling
capacity. A high cooling capacity can be generated either by a pair of large or a
number of small cooling towers. Due to influences concerning the air intake and
wind forces when several towers are used, it is better to built only one or two [110].
Usually, the cooling towers are made of concrete.
After the Munich Olympics in 1972, the structural engineers in Germany saw the
potential of using a prestressed cable net shell for cooling towers instead of the ordi-
nary concrete shell. The main advantage with the cable net tower is its high safety.
Due to its flexibility and lightness it is insensitive to earthquakes and settlements
on bad ground, which would damage a concrete tower. If the pylon is made of steel

77
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

the safety is further enhanced [110].


The tower with the highest throughput of air per hour is that at Schmehausen-
Uentrop. The height of its pylon is 180 m. Nevertheless, besides the economical and
structural questions the one concerning the harmony of the landscape is a strong
issue for not building cooling towers at all. The cable net cooling towers do not, like
most other tension structures, fit in the landscape [110].
The force density method can be used to find the shape also for cooling towers
with different mesh configurations. For both the towers shown in Figure 3.18, the
radius of the base is 5 m and the height of the pylon is 22 m. In the analyses,
it is assumed that the pylon top is a fixed node. The stiffness of the elements is
AE = 100 MN. The diamond-shaped towers had the following starting values for the
force densities: 200 kN/m for the net cables, 400 kN/m for the hangers and −2700
kN/m for the segments of the upper ring. For the rectangular net the values were:
1200 kN/m, 400 kN/m and −3300 kN/m, respectively. These values were chosen by
studying the shapes of the nets for different force density values (trial and error).
Each tower consists of 16 hangers and 16 ring segments. The target shape was, for
both towers, determined by the radius of the upper compression ring and the length
of the hangers. Node distance constraints were assigned to the segments of the
compression ring and the hangers. The strained lengths of each ring segment and
hanger are 1.5607 m and 5.3852 m, respectively. This corresponds to a compression
ring with radius of 4 m. For both nets, 5 iterations were required to obtain the
desired shape. It should be noted that, in the present examples no attention has
been paid to load bearing capacity or other requirements such as equidistant interior
net.

78
3.4. EXAMPLES

(a) Diamond net (b) Rectangular net

Figure 3.18: Two types of cooling towers.

3.4.4 A structure composed of both cables and struts

Until now structures with mainly cables has been studied in detail. Therefore, it
is necessary to check the applicability of the force density method to structures
composed of both cables and struts. Consider the simple structure in Figure 3.19.

(x0 0 z0 )
z

x
y

Figure 3.19: A simple structure composed of both cables and a strut.

79
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

The governing matrices for this can be written as:


 
2qc + qs −qs
D= , (3.75)
−qs 2qc + qs
 
−qc −qc
Df = . (3.76)
−qc −qc
If det(D) = 0 then the solutions, with the fixed points given in Figure 3.19, are

xT = x0 + l/2 x0 + l/2 , (3.77)
T

y = 0 0 , (3.78)

zT = z0 z0 . (3.79)

Hence, the length of the strut is zero irrespective of the values chosen for the force
densities as long as they do not give a singular matrix D. The elements for the
structure do not have their specified force densities. Solutions ‘out-of-plane’ can be
obtained by fixing nodes that are not coplanar to the other nodes. However, then the
equilibrium of that node is disturbed since any unbalanced load can be resolved by
the fixed support. One way to avoid this problem is to introduce more elements and
connect one of the ends of the new elements to the free nodes. The other end is fixed
at the positions which the free nodes are supposed to have, Figure 3.20. Then, the
new elements are assigned with constraints saying that the lengths of these elements
should approach zero. If the force densities of the constrained elements do not change
during the iterations, the element forces will also approach zero. This results in the
desired shape with almost the prescribed element forces. Nevertheless, since the
geometry for a prestressed structure, composed of both cables and struts, in most
cases is specified the suggested approach is useless. Better methods are at hand,
e.g. the subspace method by Pellegrino and Calladine [91]. For more complicated
structures the approach suggested by Mollaert [75], where the compression members
are replaced by external forces, may be used. No calculations have, however, been
performed to check this approach.

x
y

Figure 3.20: The structure in Figure 3.19 with additional elements.

80
3.4. EXAMPLES

3.4.5 Cable dome

The cable domes by David Geiger represent the most advanced type of very large
space structures. Pellegrino [92] has analysed a cut-down version of a larger dome
(Figure 3.22). For this dome the degree of kinematic indeterminacy m, i.e. the
number of independent mechanisms, is 13 and the degree of static indeterminacy s,
i.e. the number of independent states of self-stress, is one. The mechanisms must
be first-order infinitesimal, which they are if and only if there exists a state of self-
stress which can impart positive first-order stiffness to every mechanism. Structures
with higher-order mechanisms are not stiff enough to be used as real structures.
Two of the thirteen inextensional mechanisms are shown in Figure 3.23. The single
state of self-stress (s = 1) is obtained by computing the basis for the null-space of
A. Whether this self-stress imparts the structure with first-order stiffness can be
checked by a method given in [20].

Figure 3.21: David Geiger’s cable dome

81
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Figure 3.22: The cut-down version of a cable dome analysed by Pellegrino [92].

With the approach by Calladine and Pellegrino [93], properties of the initial config-
uration of a framework with known geometry can be obtained. However, for more
complicated structures, where the geometry is unknown, this approach cannot be
used.

(a) Mechanism 1 (b) Mechanism 2

Figure 3.23: Two of the thirteen inextensional mechanisms of the cable dome. Re-
drawn from [92].

82
3.4. EXAMPLES

3.4.6 Tensegrity structures

As mentioned in section 3.2.6 very little work has been done on form-finding of
tensegrity structures. It was shown in section 3.4.4 that to avoid a coplanar solution
of a structure composed of cables and struts additional nodes must be fixed. This
approach was used by Motro et al. [82] to find the configuration of the structure in
Figure 3.24(a). Four of the eight nodes had to be fixed to get a satisfactory solution.
But is this solution in a state of self-stress? Remember that the equilibrium at a
fixed node (support) in a framework always can be resolved. Some improvements
could perhaps be provided by the approach suggested by Mollaert [75]. Neverthe-
less, it seems that a reliable method for the initial equilibrium problem of tensegrity
structures is not available. Examples of some simple structures are shown in Fig-
ure 3.24

(a) Skew 4-prismatic system (b) Truncated tetrahedron

Figure 3.24: Some small tensegrity structures.

Larger tensegrity structures can be constructed, Figure 3.25, by assembling elemen-


tary self-stressed modules [79], Figure 3.24(a). This example shows the potential
of tensegrity structures as economical solutions for spanning large space. However,
these ideas have yet to be realised at a larger scale.

83
CHAPTER 3. THE INITIAL EQUILIBRIUM PROBLEM

Figure 3.25: A double layer single curvature tensegrity system. Redrawn from [79].

3.4.7 Conclusions

The examples have shown that the force density method is well suited to find the ini-
tial equilibrium configuration of different kinds of cable nets. However, for complex
nets, physical models might be needed to construct the initial mesh.
The convergence rate of the non-linear force density method was relatively high
when only one constraint was assigned to specific elements. When several elements
were assigned with two constraints, the convergence rate was much lower. For some
nets it is impossible to satisfy two constraints, e.g. prescribed force and unstrained
length for interior cables, and for them the iterations converged to a configuration
close to the final shape. It was also found that the starting shape had a considerable
effect on the number of iterations needed for convergence.
For structures composed of both cables and struts it is more difficult to control
the final solution. For these structures, the most common approach is to specify
the geometry and then solve for the cable forces by analysing the subspaces of the
equilibrium matrix A.

84
Chapter 4

Finite cable elements

4.1 Introduction

Today, in many technological fields, the finite element method is the dominating
analysis tool. In structural analysis all members of a structure, such as plates,
beams, bars, cables etc., need to be represented by suitable finite elements. These
elements are formulated to accurately correspond to the behaviour of the real mem-
ber. For beams, a number of finite elements are available for different applications.
However, for cable problems special elements are rarely found in commercial finite
element software, e.g. ABAQUS, and the common approach is to use straight bar
elements to model the cables. The lack of suitable cable elements stems from the
highly non-linear behaviour of a cable, which hardly can be modelled using the stan-
dard Galerkin technique. This non-linearity, which is of the geometric rather than
material type, arises due to the very low bending stiffness of a cable, Figure 4.1.
For very taut cables, the straight bar element is a good representation of the cable,
but if the cable is subjected to a compressive force it will easily buckle and lose
its stiffness. The steepness of the stiffness transition, shown in Figure 4.2, depends
on the relation between the physical properties of the cable. Hence, it is the low
flexural rigidity that makes cable modelling difficult.
In this chapter, a number of finite elements which are applicable to general cable
problems will be presented, but before starting with the elements it is necessary to
first present the analytical solutions for both the inextensible and extensible cables
subjected to uniformly distributed loads.

4.2 Analytical cable solutions

A perfectly flexible cable supported at its ends and acted upon by a uniform grav-
itational force assumes a curve called the catenary 1 . The equation of the catenary
was first obtained by Leibniz, Huygens and Johann Bernoulli in 1691. They were
1
The Latin word for chain

85
CHAPTER 4. FINITE CABLE ELEMENTS

−1

L0 = 13 m

−3
z (m)

−5

−7

−9
0 2 4 6 8 10 12
x (m)

Figure 4.1: Various configurations of an inextensible cable (x = 2, 4, 6, 8, 10, 11.9,


11.99, 11.999 m; z = −5 m).

700

600

500
Horizontal stiffness (kN/m)

q0 = 800 N/m

400

300
L0 = 80 m
AE = 51 MN
200

100
q0 = 80 N/m
0
75 76 77 78 79 80 81 82 83 84 85

Distance between supports (m)

Figure 4.2: Comparison of the stiffness of an extensible cable and a straight bar with
same properties.

86
4.2. ANALYTICAL CABLE SOLUTIONS

responding to a challenge put out by Johann’s brother Jacob to find the equation
of the chain curve [48]. In this section, the analytical solutions to the inextensi-
ble and extensible cables will be derived. These solutions can be found elsewhere,
e.g. [48, 57, 61], but since different notations have been used in these references they
cannot easily be compared. For consistency and clarity, the derivation of the solu-
tions will be repeated in this section. Finally, the validity of the assumption of zero
bending stiffness of the cable will be discussed.
z

F6

θ2
j
F4

h
E, A

q0

s θ
F1
x
i
θ1

F3
l

Figure 4.3: Extensible catenary element in x–z plane (x  x , z  z  ).

4.2.1 The inextensible catenary

To derive the equation of the inextensible catenary, certain assumptions have to be


made about the properties of the cable. It is assumed that the cable is perfectly
flexible (EI ≡ 0), inextensible (AE → ∞), free of torsional rigidity and able to
sustain only tensile forces. As a result of these assumptions, the cable force is
tangential to the cable at every point. Consider a small segment of the extensible
cable in Figure 4.3. Let AE → ∞ and the segment becomes inextensible, Figure 4.4.
It may seem inconsistent to define the self-weight positive downwards, Figure 4.4, but

87
CHAPTER 4. FINITE CABLE ELEMENTS

since only loads acting downwards are considered in this chapter this convention is
chosen (another argument is that this convention was chosen for the cable elements,
which will be used later, cf. [1, 52]). Thus, in all the expressions below only the
magnitude of the gravitational force, not its direction, needs to be inserted.
 
dz d dz
T + T ∆s0
ds0 ds0 ds0

∆s 0
 
dx d dx
T + T ∆s0
ds0 ds0 ds0
dx
T
ds0
limAE→∞ s = s0
dz q0 ∆s0
T
ds0
Figure 4.4: Segment of an inextensible cable

Returning to Figure 4.4, vertical equilibrium of the cable segment yields


 
d dz
T = q0 , (4.1)
ds0 ds0
where
dz
= sin θ. (4.2)
ds0
Horizontal equilibrium of the cable segment gives
 
d dx
T = 0, (4.3)
ds0 ds0
where
dx
= cos θ. (4.4)
ds0
Integrating (4.3) yields
dx
T = H, (4.5)
ds0
where H is the horizontal component of cable tension, which is constant along the ca-
ble since no horizontal loads are acting. Substituting equation (4.5) to equation (4.1)
gives
d2 z ds0
H 2 = q0 . (4.6)
dx dx
Using the following geometric constraint:
 2  2
dx dz
+ = 1, (4.7)
ds0 ds0

88
4.2. ANALYTICAL CABLE SOLUTIONS

equation (4.6) can be written as:


  2 1/2
2
dz q0 dz
2
= 1+ . (4.8)
dx H dx

The solution to equation (4.8) with the boundary conditions x = 0, y = 0 and x = l,


y = h, is
H
q x
0

z= cosh + ζ − cosh ζ , (4.9)
q0 H
in which  
−1 q0 h
ζ = sinh − η, (4.10)
2H sinh η
and
q0 l
η= . (4.11)
2H
Equation (4.9) is the inextensible catenary equation. Differentiating (4.9) gives the
slope as:
dz
q x
0
= sinh +ζ . (4.12)
dx H
The length of the inextensible catenary can be computed with the following equa-
tion [84]:
l2
L2 = 2 sinh2 η + h2 . (4.13)
η
An approximate equation for the inextensible catenary can be obtained for a shallow
cable, that is a cable with a small sag-to-span ratio. In this case ds0 /dx ≈ 1 and
equation (4.6) simplifies to
d2 z
H 2 = q0 . (4.14)
dx
Using the same boundary conditions, the solution to (4.14) is the following parabolic
equation: 

x 2
x
x
z = ηl − +h . (4.15)
l l l
Equation (4.15) is much easier to work with than (4.9). Therefore, the parabola
has been used for many years as the cable equation in the development of approx-
imate formulae for preliminary design of cable structures, see for example [42, 57].
As mentioned above, for a cable where only the self-weight is acting, the parabolic
equation (4.15) is approximate and the error increases as the sag-to-span ratio in-
creases. To obtain an estimation of the difference between the parabola and the
catenary, we consider a cable with level supports, i.e. h = 0. For sag-to-span ra-
tios larger than 0.2 the parabola is a rather crude representation of the catenary,
Figure 4.5. Nevertheless, the use of the parabolic equation may in some cases be
more correct, e.g. for a suspension bridge, where the cables sustain a load which is
uniformly distributed along their span and much larger than the self-weights of the
cables, that is the bridge deck [57].

89
CHAPTER 4. FINITE CABLE ELEMENTS

11

10
zcat (l/2) = 0.3l
9

8
zcat (l/2) = 0.3l
zcat (x)−zpar (x)
7
Percent error

zcat (l/2)

6 Tcat (x)−Tpar (x)


Tcat (x)
5
zcat (l/2) = 0.2l
4

2
zcat (l/2) = 0.2l zcat (l/2) = 0.1l
1
zcat (l/2) = 0.1l
0
0 0. 05 0. 1 0. 15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x/l

Figure 4.5: Difference between sags and cable forces for the inextensible catenary
and parabola with same spans l and horizontal forces H

4.2.2 The elastic catenary

A real cable is not inextensible as it has a finite axial flexibility. The differential
equations of equilibrium of a stretched chain, satisfying Hooke’s law, were derived by
Jacob and Johann Bernoulli, but the solution was first given in 1891 by Routh [48].
To derive the equations for the elastic catenary, the extensible cable segment in
Figure 4.6 is considered. Horizontal and vertical equilibrium of the segment yields
dx
T = H, (4.16)
ds
dz
= q0 s0 − F3 . T (4.17)
ds
It can be noted that due to the conservation of mass the total weight of the cable
segment is unaffected by the elongation of the cable. For the elastic catenary, the
geometric constraint to be satisfied is
 2  2
dx dz
+ = 1. (4.18)
ds ds
It is assumed that the cable material satisfies Hooke’s law
 
ds
T = AE −1 , (4.19)
ds0
where A is the uniform cross-sectional area in the unstrained profile.

90
4.2. ANALYTICAL CABLE SOLUTIONS

T dz
ds

T dx
ds

s
H
q0 s 0

F3

Figure 4.6: Segment of an extensible cable.

Then, (4.16) and (4.17) are squared, added and substituted into equation (4.18),
which gives the cable force at any point s0 as [48]:
 1/2
T (s0 ) = H 2 + (q0 s0 − F3 )
2
. (4.20)

By noting that dx/ds0 = (dx/ds)(ds/ds0 ) and dz/ds0 = (dz/ds)(ds/ds0 ) the para-


metric solutions x(s0 ) and z(s0 ) are derived, since dx/ds, dz/ds and ds/ds0 are given
by the equations (4.16), (4.17) and (4.19), respectively. The solutions are
    
Hs0 H −1 F3 −1 q0 s0 − F3
x(s0 ) = + sinh + sinh , (4.21)
AE q0 H H
! 

    "1/2 !
 2
  2 "1/2
F s0 q0 s0 H q0 s0 − F3 F3
z(s0 ) = 3 
−1 +  1+ − 1+ .
AE 2F3 q0 H H
(4.22)
It is obvious that the cable force increases with height above the lowest point of the
cable. If the equations (4.20)–(4.22) are combined we find the following interesting
result [102]:
 1/2
T (z(s0 )) = (AE + Tb )2 + 2AEq0 z(s0 ) − AE, (4.23)
where Tb is the cable force at the base, i.e. at s0 = 0. Hence, the cable force at
any value of z(s0 ) is independent of the span l. Practically, this result is important
because the cable force at the base Tb is an input value in the design of some cable
structures, for example guyed masts [102].
Recently, Russell and Lardner [102] did some experiments with a scale model of a
guy cable, for which the elasticity is important. The purpose with their experiments

91
CHAPTER 4. FINITE CABLE ELEMENTS

was to compare the numerical predictions from the equations (4.20)–(4.22) of the
elastic catenary with the measured values. The results show very good agreement
between the experimental and theoretical cable force at the base, x = 0, for a number
of horizontal spans l. An average error of 2.5 % below the theoretical values was
reported.

4.2.3 Effect of cable bending stiffness

Some authors, e.g. [1, 52], have termed the elastic catenary exact, which under the
stated assumptions is correct. However, the elastic catenary is never exact in reality,
because the assumptions are violated in many cases. One assumption which is more
important than the others, and needs further studies, is that of the zero bending
stiffness. For a real cable, as that in a suspension bridge, the magnitude of the
flexural rigidity can be very large. Therefore, the effects of a non-zero flexural
rigidity on cable geometry and forces will be analysed in this section.
First we need to know how large the bending stiffness of a cable actually can be.
Calculating the bending stiffness is more complex than finding the axial stiffness,
but if we think of how the cable is assembled, see section 2.4, two extreme cases can
be distinguished:

1. All wires are stuck together. The cable has one neutral axis and the rigidity
of the cable is similar to that of a beam.

2. The friction between the wires is zero, making them bend around their own
neutral axis.

The first case represents the upper bound of the bending stiffness while the second
represents the lower bound. Several methods to calculate the flexural rigidity for
helically wound cables have been suggested in literature [22]. Following the recom-
mendations made by Cardou and Jolicoeur [22], a good estimate of the upper bound
of the cable bending stiffness is obtained using the approach by Lanteigne [60]:


n
(Ri2 /2 + ri2 ) 3
(EI)max = E0 I0 + mi Ai Ei sin αi . (4.24)
i=1
2

For the lower bound of the bending stiffness the frictionless model by Costello [29]
should be used [22]:


n
sin αi
(EI)min = E0 I0 + mi Ei Ii . (4.25)
i=1
(2 + ν cos2 αi )

The true bending stiffness is somewhere between the two extremes. Which of the
extreme values that is closest to the real weakness depend from case to case. To
further complicate things, experiments have shown that the bending stiffness is not
constant; it varies along the length of the cable [22].

92
4.2. ANALYTICAL CABLE SOLUTIONS

The effects of a non-zero bending stiffness on the cable geometry and forces has been
analysed by Wang and Watson [127]. In their studies, they assumed that the cable
was inextensible and used the elastica equation,

d2 θ
EI = H sin θ − q0 s0 cos θ, (4.26)
ds20

first formulated by Euler, to represent the cable, Figure 4.7. By dividing all forces by
EI/S 2 and all lengths by S, where S = L/2, equation (4.26) is made dimensionless.
Equation (4.26) then becomes

d2 θ
2 = H sin θ − K s0 cos θ,
3
(4.27)
d s0

where K = Sq0 (EI)−1/3 . K can be seen to represent the relative importance of


1/3

density and length to flexural rigidity or the ratio of half the cable length to the
bending length (EI/q0 )1/3 [127]. Since no closed form solution to equation (4.27)
exists, it has to be integrated numerically. However, due to the extreme numerical
stiffness of (4.27) for large values of K, all classic numerical techniques, such as
Newton’s method with the shooting algorithm or a finite difference approximation,
fail to find the correct solution [128]. Solutions for values of K in the whole range
can only be provided by a sophisticated continuation method [127, 128].
Some of the results from Wang and Watson’s studies are presented in Figures 4.8
and 4.9. For a > 0.8 the effect of K on the shape can be neglected, Figure 4.8. For
the maximum curvature shown in Figure 4.9(a), the effect of K is considerable for
a < 0.8. The horizontal force is very sensitive to K for all values of a, Figure 4.9(b).
The following conclusions can be drawn from the studies by Wang and Watson:

• The effect of bending stiffness on the cable shape can be ignored for taut
cables.

• Ignoring the bending stiffness leads to a softer model, which in dynamic anal-
yses gives lower eigenfrequencies.

Still, the assumption of a constant bending stiffness along the length of the cable is
not correct according to the experimental results reported in [22]. As a concluding
remark one can say that for the quasi-static behaviour of a taut cable the bending
stiffness is of less significance.

93
CHAPTER 4. FINITE CABLE ELEMENTS

q0 S T
z
a
H

s0 θ
x

Figure 4.7: Coordinate system for an inextensible cable with non-zero flexural rigid-
ity. Redrawn from [127].

Although the effects of neglecting the bending stiffness of the cable have been anal-
ysed one assumption still remains: the zero torsional rigidity. To check the effects of
this assumption the cable model would have to be extended with rotational degrees
of freedom, but this will not be done here.
In their present form, the analytical solutions are of little use. For a general analysis
of cable structures the cable must be represented by a finite element.

Figure 4.8: The height b as a function of the span a for an inextensible cable with
bending stiffness. Redrawn and slightly modified from [127].

94
4.3. LITERATURE REVIEW OF CABLE ELEMENTS

H

ds0
(0)

a a

(a) The maximum curvature (b) The horizontal force

Figure 4.9: Some results for an inextensible cable with bending stiffness. Redrawn
and slightly modified from [127].

4.3 Literature review of cable elements

The objective of this section has not been to give a historical review of all the
developed cable elements. Instead, it is focused on different formulations used to
solve the cable problem. The literature review has indicated that generally two
approaches for the development of finite cable elements are used. The first approach
is the use of polynomials to describe the shape and displacement field, which is the
ordinary approach in the development of finite element. In the second approach
analytical expressions are used, which in a mathematical sense exactly describes the
cable under certain load conditions.

4.3.1 Elements based on polynomial interpolation functions

In this class basically three types of elements are found in literature:

• straight bar element,

• curved isoparametric bar elements, and

95
CHAPTER 4. FINITE CABLE ELEMENTS

• curved elements with rotational degrees of freedom.

Two-node straight bar element

The straight bar element is the most common element used in the modelling of
cables. A number of element formulations have been presented for geometrically
non-linear analysis, see for example [5, 9, 14, 16]. However, in the end they should
give similar results. Straight bar elements, which possess only axial stiffness, often
provide suitable representation of highly pretensioned cables, such as those in cable
nets and trusses. For slack cables with large curvature, the standard approach is to
represent the cable by a large number of bar elements. This technique is inefficient
as the number of degrees of freedom drastically increases. Another drawback is the
spurious slope discontinuities occurring at nodes where no concentrated loads act.
These discontinuities are due to the straight element assumption and may lead to
convergence problems in the analysis [37, 61].
One method often used to model slack cables is the equivalent modulus or stiffness
approach. In this method, the cable is assigned a certain stiffness that take the cable
sag into account. This equivalent stiffness was derived by Ernst [34]. By equating
the stiffness of a straight bar element to that of a parabolic element, he obtained a
simple expression for the equivalent axial stiffness of a slack cable. The equivalent
stiffness is a function of the cable force, the self-weight of the cable, the length of
the element and the axial stiffness of a straight cable. However, for cables with large
sag a large number of elements is still needed.

Multi-node isoparametric finite elements

Instead of using many bar elements with linear interpolation functions one can use
fewer elements with interpolation functions of higher order. By adding more nodes
to the finite element, higher order polynomials for the shape and displacements of
the element can be defined [61]. Most common are the three and four node el-
ements, which use parabolic and cubic interpolation functions, respectively. The
tangent stiffness matrix and equivalent nodal forces are obtained using the isopara-
metric formulation. Due to the complex expressions involved in this formulation,
the tangent stiffness matrix and equivalent nodal forces have to be found by numer-
ical integration. These curved elements give accurate results for cables with small
sag. For larger sag more element must be used. Between element nodes only the
displacement continuity is enforced [89].

Curved elements with rotational degrees of freedom

The continuity of the slopes can be enforced by adding rotational degrees of freedom
to the nodes. Such an element was developed by Gambhir and Batchelor [37]. Cubic
polynomials described the displacement field and shape of the cable. Due to the

96
4.3. LITERATURE REVIEW OF CABLE ELEMENTS

simplifications made for this element it is applicable only for cables with small sag-
to-span ratios. To model a cable that has a large curvature more elements have to be
used, but, contrary to the other elements presented above, no slope discontinuities
will occur.
It may be argued that a cable can be modelled using beam elements with a low
bending stiffness. For cables where the bending stiffness is known or assumed this
alternative can provide more reliable results. However, for very flexible cables, a
large number of beam elements with a very low bending stiffness are needed to model
the cable [54]. Since rotational degrees of freedom are present in the beam elements,
the total number of degrees of freedom will, therefore, substantially increase.
Some advantages associated with the polynomial based cable elements are:

• The formulation with polynomials is almost universal.

• The out-of-plane response of the cable can be captured if the multi-node


isoparametric formulation is used.

• For dynamic analysis the mass matrices are consistent.

The disadvantages are:

• Without rotational degrees of freedom slope discontinuities occur, which may


induce numerical convergence problems.

• Many elements are needed to model slack cables with large sag-to-span ratios.

4.3.2 Elements based on analytical functions

The second class contains the analytical elements, which are based on analytical
formulae to take into account the effect of loading applied along the length of the
cable. Three elements, each associated with a certain type of uniformly distributed
loading, have been presented in literature. They are:

• The parabolic element for which the load is uniformly distributed along the
horizontal span of the cable.

• The elastic catenary element for which the load is uniformly distributed along
the unstrained length of the cable.

• The associate catenary element for which the load is uniformly distributed
along the strained length of the cable.

97
CHAPTER 4. FINITE CABLE ELEMENTS

Parabolic element

As mentioned earlier the parabola has, because of its simpler form compared to
the catenary, been more frequently used in the analysis of cable structures. One
type of parabolic element, the three-node isoparametric element, has already been
presented. Two other formulations, which have much in common, are presented in [1,
76]. Both formulations use shallow cable assumptions, which make them applicable
only for cables with small sag-to-span ratios. For both elements, the cable force is
obtained by solving a cubic equation.

Elastic catenary element

For a perfectly flexible cable subjected to self-weight load only this element is exact.
Basically, two elastic catenary elements have been developed. The first element was
presented by Peyrot and Goulois [94, 95]. They used the expressions by O’Brien et
al. [53, 83, 84] to obtain the flexibility matrix. The tangent stiffness matrix was de-
rived by taking the inverse of the flexibility matrix. In reference 94 the plane version
of the element was used in the analysis of transmission lines, while in reference 95
three-dimensional structures were analysed. The tangent stiffness given in [95] is be-
lieved to be incorrect, because the out-of-plane stiffness, i.e. in the y  -direction, was
set to zero. This was later corrected by Jayaraman and Knudson [52] who demon-
strated the capability of the element for a number of examples. Irvine [48] derived
a similar element, but reported that the stiffness matrix should be unsymmetric
due to the geometric non-linearity. However, it is observed that the stiffness matrix
in reference 48 in fact is symmetric, which it ought to be due to the conservative
nature of the self-weight [3]. The second elastic catenary element was derived by
Ahmadi-Kashani [1]. This element will be presented later in this chapter together
with the last element in the class, the associate catenary element.

Associate catenary element

The last element in this section is a special one. For this element the load is dis-
tributed along the strained length of the element. A load of this type is the snow
load. Under the action of the snow load the element stretches and thereby increases
the available length for the snow flakes to land on; it can be said that the element
becomes heavier as it elongates. Thus, the total load is dependent on the displace-
ments. In this case the loading is non-conservative and therefore the tangent stiffness
matrix is unsymmetric. This element was derived in reference 1 and the equations
expressing the nodal forces and the tangent stiffness matrix are utterly complex.
The main advantage with the analytical elements is that only one element is needed
to model a cable with a high degree of accuracy. More elements may be used in,
for example, dynamic analysis and thanks to the exact analytical expressions no
discontinuities occur across the element boundaries.
Although these elements work very well there are some disadvantages:

98
4.4. STRAIGHT AND PARABOLIC ELEMENTS

• The equivalent nodal forces and the tangent stiffness matrix have to be found
by iteration.

• The use of trigonometric functions, such as the tangent function, in the for-
mulations make them undefined for certain angles or load cases.

• Consistent mass matrices are not available.

Finally, should be mentioned the general cable element in reference 1 which can
be used for any type of loading along the length of the cable. For this element, a
finite segment approach is used to calculate the equivalent nodal forces and tangent
stiffness matrix. Using this general element only one element is needed to model a
cable, irrespective of the variation of the load along the length of the cable. Thus,
this element is very useful for the analysis of long guy cables or underwater cables,
which exhibit varying wind and fluid loads.
From the elements presented in this review, four elements are selected and will be
described in the following sections. These include: the straight bar, the elastic
parabolic element, the elastic catenary element, and the associate catenary element.
They are chosen because they all have six degrees of freedom and closed-form ex-
pressions for the equivalent nodal force vectors and tangent stiffness matrices, which
means that no numerical integration is necessary to obtain them.

4.4 Straight and parabolic elements

These two elements are chosen to be presented in the same section because they are
somewhat linked to each other by the equivalent stiffness by Ernst [34]. It will be
shown that their tangent stiffness matrices have some similarities.

4.4.1 Straight bar element

This element, shown in Figure 4.10, is the most simple finite element for structural
analysis. It has appeared in many forms, e.g. [5, 14, 16], but in the end practically
all of them give the same results. The formulation used here is from [9], with no
detailed discussion considered necessary. The vector of equivalent nodal forces in
the local coordinate system is written as:

F 1 = −T 0 0 T 0 0 .
T
(4.28)

The elastic stiffness sub-matrix in the local coordinate system is given as:
 
1 0 0
AE 

kE = 0 0 0 . (4.29)
L
0 0 0

99
CHAPTER 4. FINITE CABLE ELEMENTS

F6 x

z
j F5
z y 
F4
L

F1
y
i
F2
F3

Figure 4.10: Straight bar element in space

Since the analysis of cable structures is non-linear the following geometric stiffness
sub-matrix is needed:  
1 0 0
T
kG =  0 1 0  . (4.30)
L
0 0 1
The total elastic matrix in local coordinate system tangent stiffness is written as:
 
 kE −kE
KE = , (4.31)
−kE kE
while the geometric stiffness is given as:
 
kG −kG
KG = . (4.32)
−kG kG
The transformation matrix from the local to the global system is
 
t1 0
T1 = , (4.33)
0 t1
in which
 x j − xi y j − yi zj − zi 
 L L L 
 yj − yi x j − xi 
 
 − 0 
t1 =  Lxy Lxy , (4.34)
 
 (x − x )(z − z ) (yj − yi )(zj − zi ) Lxy 
 − j i j i
− 
LLxy LLxy L

100
4.4. STRAIGHT AND PARABOLIC ELEMENTS

where  1/2
Lxy = (xj − xi )2 + (yj − yi )2 . (4.35)
Thus, the equivalent nodal forces and tangent stiffness matrix in global coordinates
can now be written as [9]:
F1 = TT1 F1 , (4.36)
and
K1 = TT1 KE T1 + KG . (4.37)
The straight bar element can be used in the modelling of slack cables if the axial
stiffness AE/L is substituted by the equivalent axial stiffness by Ernst [34], which
is given as:  
AE AE
=  . (4.38)
L eq q02 l2
L 1+ AE
12T 3
For cables where the chord length is longer than the unstrained length the cable
tension can be computed with the following well-known equation:
L − L0
T = AE . (4.39)
L0
When L ≤ L0 equation (4.39) yields a zero or compressive force. How to compute
the cable force in those cases will be described in the next section.

4.4.2 Elastic parabolic element

As mentioned in section 4.3, for this type of element the load q0 is distributed along
the horizontal span of the cable. Some authors [1, 76] have derived the nodal forces
and element stiffness matrix for parabolic elements with small sag-to-span ratios. An
elastic parabolic element for large sags has not yet been developed. The formulation
for the elastic parabolic element presented in this section was given in [1].
The vector of equivalent nodal forces for a taut parabolic element is given as:

F 2 = F1 0 F3 F4 0 F6 ,
T
(4.40)

where

F1 = −H, (4.41)


 
 h q0 l
F3 = −H − , (4.42)
l 2H
F4 = −H, (4.43)
 
 h q0 l
F6 = H + , (4.44)
l 2H

101
CHAPTER 4. FINITE CABLE ELEMENTS

from which we can obtain the end angles


h q0 l
tan θ1 = − , (4.45)
l 2H
h q0 l
tan θ2 = + . (4.46)
l 2H
The expressions for the equivalent nodal forces and end angles are applicable for
elastic parabolic elements with large sag-to-span ratios. However, to obtain the tan-
gent stiffness matrix some simplifications have to be introduced in order to solve the
complex equations that arise in the formulations [1]. By introducing the assumption
of small a sag-to-span ratio, these equations were simplified and solved in reference 1
resulting in the following local tangent stiffness matrix:
 
k1 0 k2
 H 
k2 = α1  0 0 , (4.47)
α1 l
k2 0 k3

in which
 
L0 C − L0 q 2 l2
k1 = 2
cos β + sin2 β + 0 2 cos2 β sin2 β, (4.48)
C C 8T
 2 2

2L0 q l
k2 = − 1 − 0 2 cos2 β sin β cos β, (4.49)
C 8T
 
L0 C − L0 q 2 l2
k3 = 2
sin β + cos2 β + 0 2 cos4 β, (4.50)
C C 8T
AE
α1 =  2 2
. (4.51)
q0 l 2
C 1+ AE cos β
12T 3

The element matrix for the elastic parabolic element, in the local coordinate system,
is written as:  
 k2 −k2
K2 = . (4.52)
−k2 k2

The transformation matrix for the parabolic element is not the same as T1 . Since the
x -axis is not oriented along the chord of the element, T1 cannot be used. However,
since the z  -axis is vertical, which in the majority of cases is also the direction of the
z-axis, the transformation matrix for the parabolic element is easily derived; for a
cable loaded only in the z-direction the projection of the element is straight in the
x–y plane. Thus, the rotation matrix is the same as the well-known rotation matrix
for a plane beam element (if the rotational degrees of freedom are substituted by
the translations in the z-direction)
 
t2 0
T2 = , (4.53)
0 t2

102
4.4. STRAIGHT AND PARABOLIC ELEMENTS

z

F6
F5
j
h y
F4

F1
i q0
F2
F3
l

x

Figure 4.11: Catenary element in x –z  plane

where  
cos φ − sin φ 0
t2 =  sin φ cos φ 0 . (4.54)
0 0 1
This transformation matrix is also compatible with the elastic and associate cate-
naries presented in section 4.5.
The global nodal forces and tangent stiffness matrix for the parabolic element can
now be written as:
F2 = TT2 F 2 , (4.55)
and
K2 = TT2 K2 T2 . (4.56)
The tangent stiffness matrix is a function of the unknown cable force T . This force
can be found from the following cubic equation:
   22 
1 3 C − L0 q0 l q 2 l2
F (T ) = T − 2
T + cos β T − 0 cos2 β ≡ 0. (4.57)
4
AE C 12AE 24

It is assumed that the contribution of the third term to the equation is negligible, [1],
and the equation is simplified to
 
1 3 C − L0 q02 l2
F (T ) = T − T −
2
cos2 β ≡ 0. (4.58)
AE C 24

103
CHAPTER 4. FINITE CABLE ELEMENTS

z
z

F6

F5

y
F4

F1 q0

F2
φ
F3 x

Figure 4.12: Rotation of catenary in the x–y plane

From Descartes’ rule of signs [96] it follows that there is only one positive root of
equation (4.58). Equation (4.58) is in [1] solved with a Newton-Raphson technique.
If the stiffness sub-matrix of the elastic parabolic element (4.47) is compared with
that of the straight bar element, equation (4.37), it is noted that, apart from the
cos2 β-term, α1 is identical to the equivalent axial stiffness given by Ernst, equa-
tion (4.38). It is claimed in [1] that the formulation presented here should provide a
better approximation for the taut parabolic element than the bar element with an
equivalent axial stiffness. However, with the load intensity q0 and angle β equal to
zero, equation (4.47) becomes
   
L0 AE
 C 0 0 
 C 
 T 
k2 =  0 0 , (4.59)
 C   
 (C − L0 ) AE 
0 0
C C
where the following relation for the taut parabola is used:
H
T = . (4.60)
cos β
Comparison of (4.59) and the sub-matrix of (4.37) indicates that the geometric
stiffness terms in the x- and z-direction are missing. This results in a softer behaviour

104
4.5. CATENARY ELEMENTS

of this element compared to a bar element. In addition to that, a small fraction,


i.e. (C − L0 )/C, of the elastic stiffness AE/C also contributes to the stiffness in
z-direction, which is peculiar. Hence, the elastic parabolic element is not identical
to the bar element when the load q0 along the span is removed.

4.5 Catenary elements

In the preceding section two different approaches were presented for a parabolic cable
element, where one was said to be slightly more accurate than the other [1]. However,
the accuracy of the results using a parabolic element to model a cable subjected to
load uniformly distributed along its length is acceptable only for cables with small
sag-to-span ratios. In addition to this, the ability of the straight bar element to take
compressive forces may induce numerical instability in the analysis of, for example,
cable nets. Therefore, a routine which eliminates the element from the model if a
compressive force is detected, has to be used in the computer program [14, 16]. The
element is eliminated by setting the axial stiffness to a very low value, say 10−20 .
In this section elements are studied, which in a mathematical sense are exact for
cables subjected to uniformly distributed loads along their lengths. These elements
can be used for any sag-to-span ratio; both very slack and taut cables can be analysed
with a high degree of accuracy.

4.5.1 Elastic catenary element

For this element the load is uniformly distributed along the unstrained length of the
element
q(s0 ) = q0 . (4.61)
The self weight of the cable is a load of this type. Thus, this element gives the same
results as the elastic catenary equations (4.20)–(4.22) and is therefore called elastic
catenary element. In a mathematical sense this element is exact for a cable loaded by
its self weight only. As mentioned in section 4.3, two different formulations have been
developed for this element. The first formulation, which was given in [52, 95], was
based on a flexibility method by O’Brien et al. [53, 83, 84]. The second formulation,
which will be presented here, was derived by Ahmadi-Kashani [1]. The equations
that express the equivalent nodal forces and tangent stiffness matrix are written in
terms of H, θ1 and θ2 , where θ1 and θ2 are the angles between the horizontal and
and the tangent to the cable at the ends, Figure 4.3. The vector of nodal forces for
the elastic catenary shown in Figure 4.3 can be written as:

F 3 = F1 0 F3 F4 0 F6 ,
T
(4.62)

105
CHAPTER 4. FINITE CABLE ELEMENTS

where

F1 = −H, (4.63)


   
 −1 q 0 L0
F3 = −H sinh cosh −µ , (4.64)
2H sinh µ
F4 = H, (4.65)
   
 −1 q 0 L0
F6 = H sinh cosh +µ , (4.66)
2H sinh µ

where  
q0 l L0
µ= − . (4.67)
2 H AE
The element stiffness matrix, in the local coordinate system, for the elastic catenary
element is written as:  
 k3 −k3
K3 = , (4.68)
−k3 k3
where
 
q 0 L0
 AE + sin θ2 − sin θ1 0 cos θ1 − cos θ2 
 
 H 
 
k3 = α2  0 0 , (4.69)
 α2 l 
 
 q0 l 
cos θ1 − cos θ2 0 − sin θ2 + sin θ1
H

in which
q0
α2 =   , (4.70)
θ 2 − θ1 θ1 + θ2 θ 2 − θ1 q02 L0 l
4 sin µ cos − sin +
2 2 2 AEH
where
   
−1 q 0 L0
tan θ1 = sinh cosh −µ , (4.71)
2H sinh µ
   
−1 q 0 L0
tan θ2 = sinh cosh +µ . (4.72)
2H sinh µ

It should be emphasised that the expressions for the nodal forces and tangent stiff-
ness matrix are applicable only if node j is above or at the same level as node i, i.e.
zj ≥ zi . However, this does not cause any serious problem for the implementation;
it is just a matter of switching the directions of some of the stiffness coefficients
and the horizontal forces at the nodes and then change the rows and columns of the
tangent stiffness matrix and the force vector.
One particular case, which deserves attention, concerns an elastic catenary element
where the intensity of the loading q0 approaches zero. For this case the tangent

106
4.5. CATENARY ELEMENTS

stiffness matrix (4.69) becomes


 
AE 2 T AE
 L cos θ1 + L 0
L
sin θ1 cos θ1 
 
 T 
  
lim k3 =  0 0 . (4.73)
q0 →0  L 
 AE 
 sin θ1 cos θ1 0
AE 2
sin θ1 +
T 
L L L

where the following relation between the unstrained length L0 and the strained
length L for a straight elastic element is used:
 
T
L = L0 1 + . (4.74)
AE
It is noted, that the stiffness sub-matrix in equation (4.73) is identical to that of a
straight bar element rotated an angle θ1 . Hence, the elastic catenary element can be
used to represent weight-less cables if, to avoid numerical instabilities, a very small
value for q0 is specified [2].
It is seen that the equivalent nodal forces and the tangent stiffness matrix are func-
tions of only one variable, i.e. the horizontal force H. For a given geometric config-
uration, H is found by the following equation:
4H 2 h2
2 − L0 ≡ 0.
2 2
F (H) = sinh µ +    (4.75)
q02 q 0 L0
1+ coth µ
2AE
Equation (4.75) is solved using the well-known Newton-Raphson algorithm for non-
linear equations of one variable, where the derivative of (4.75) is given as:
 2
dF 8H 2l lL0 hq0
= 2 sinh µ − sinh 2µ −
2
   3 . (4.76)
dH q0 q0 q 0 L0 H sinh µ
2AE 1 + coth µ
2AE
In the case of an inextensible cable element, where AE → ∞, (4.75) simplifies to

4H 2 q0 l
L20 = 2
sinh2 + h2 , (4.77)
q0 2H
which is identical to equation (4.13).

4.5.2 Associate catenary element

To understand the formulation of this element one can think of a long elastic cable
which is supported at its ends and placed outdoors. One example of such a cable
is the conductor in a transmission structure. Assume that outdoors there is a light
wind and after a while it starts to snow. Every snow flake that lands on the cable

107
CHAPTER 4. FINITE CABLE ELEMENTS

stretches it. The snow will be uniformly distributed along the strained length of
the cable. Thus, the total load on the cable is dependent on the deformation of the
cable. In this case the displacement dependent load is also non-conservative.
In other words, the load along this element, in terms of the unstrained length, is not
constant. Instead, the load along the unstrained length is expressed as:
ds
q(s0 ) = q0 , (4.78)
ds0
where q0 is the constant load per unit strained length of the cable. An element of
this type was developed by Ahmadi-Kashani [1], who gave it the name associate
catenary element. The vector of nodal forces for this element is

F 4 = F1 0 F3 F4 0 F6 ,
T
(4.79)

where

F1 = −H, (4.80)


   
q0 h
F3 = −H sinh sinh−1 −η , (4.81)
2H sinh η
F4 = H, (4.82)
   
 −1 q0 h
F6 = H sinh sinh +η , (4.83)
2H sinh η
(4.84)

where η is given by (4.11). From the expressions for the equivalent nodal forces, the
end slopes are found as:
   
−1 q0 h
tan θ1 = sinh sinh −η , (4.85)
2H sinh η
   
−1 q0 h
tan θ2 = sinh sinh +η . (4.86)
2H sinh η

It can be seen that these end slopes are identical to those of the inextensible catenary,
equation (4.12). Hence, the shape of the associate catenary is exactly the same
as that of the inextensible catenary. Considering the properties of the associate
catenary element, this result is not a surprise. Despite the elongation of the element
the load intensity per actual length unit is always q0 , which is also the case for
the inextensible catenary. For a suspended elastic catenary the load intensity per
strained length unit is always less than q0 . The tangent stiffness matrix of the
associate catenary element in the local coordinate system, as shown in Figure 4.11,

108
4.5. CATENARY ELEMENTS

is written as:
 
a 0 b −a 0 −b
 
 H H 
 0 0 0 − 0 
 α3 l α3 l 
 
 
 λ1 a − λ2 c 0 λ1 b + λ2 d −λ1 a + λ2 c 0 −λ1 b − λ2 d 
 
K4 = α3  ,
 −a 0 −b a 0 b 
 
 
 −
H H 
 0 0 0 0 
 α3 l α3 l 
 
−λ3 a + λ4 c 0 −λ3 b − λ4 d − e λ3 a − λ 4 c 0 λ3 b + λ 4 d + e
(4.87)
where
ψ
α3 = − , (4.88)

H
γ= , (4.89)
AE
q0 L
a = sin θ2 − sin θ1 + , (4.90)
AE
b = cos θ1 − cos θ2 , (4.91)
q0 h
c= + ξ tan θ2 , (4.92)
H
q0 l
d= + ξ, (4.93)
H
q0
e= , (4.94)
α3 sin θ2
λ1 = tan θ1 , (4.95)
cos θ1 + γ
λ2 = , (4.96)
cos θ1
q0 h
λ3 = tan θ2 − , (4.97)
H sin θ2
sin θ1
λ4 = λ 2 , (4.98)
sin θ2
H
ψ= , (4.99)
q0 cos θ1 (cos θ1 + γ)
ψ
∆ = − [ξL (cos θ1 + γ) + al + bh] , (4.100)
H   
 q 0 l q 0 L0
ξ = 1 − γ (cos θ2 + γ) γ
2

H AE
  (4.101)
sin θ2 sin θ1
− − .
cos θ2 + γ cos θ1 + γ
It is seen that the tangent stiffness matrix in (4.87) is not symmetric. This is due
to the non-conservative nature of the load acting on the element.
As in the case of the elastic catenary element, the equivalent nodal forces and the
tangent stiffness matrix of the associate catenary element are functions of the sole

109
CHAPTER 4. FINITE CABLE ELEMENTS

variable H. To obtain H, the following non-linear equation has to be solved, using


the Newton-Raphson technique outlined in section 4.5.3,
# #
AE ## (m + p) (m − n) ## AEl
F (H) = L0 − ln − ≡ 0, (4.102)
mq0 # (m − p) (m + n) # H

where the derivative of (4.102) is



dF H 2AE sgn(κ) (m − n) [(m2 − 1) + r(p − 1)]
=− ln |κ| −
dH AEq0 m3 Hq0 m |κ| (m − p)(m + n)
(m + p) [(m − 1) − t(n − 1)] (m + p)(m − n) [(m2 − 1) − r(p − 1)]
2
+ −
(m − p)(m + n) (m − p)2 (m + n)

(m + p)(m − n) [(m2 − 1) + t(n − 1)] AEl
− + 2
(m − p)(m + n) 2 H
(4.103)

in which
(m + p)(m − n)
κ= , (4.104)
(m − p)(m + n)
  2 1/2
H
m= 1− , (4.105)
AE
H a1
n=1+ e , (4.106)
AE
H b1
p=1+ e , (4.107)
AE 
−1 q0 h
a1 = sinh − η, (4.108)
2H sinh η
 
−1 q0 h
b1 = sinh + η, (4.109)
2H sinh η
h
t = 1 + (η coth η − 1) + η, (4.110)
L
h
r = 1 + (η coth η − 1) − η. (4.111)
L
The strained length L of the associate cable element is given by (4.13), but can be
calculated by
H
L= (tan θ2 − tan θ1 ) . (4.112)
q0
In references 1 and 3 the derivative of (4.102) is not given as in (4.103). When the ex-
pression for dF/dH given in these references was used in the Newton-Raphson algo-
rithm the iterations did not converge. Therefore, equation (4.102) was checked with
the mathematical symbolic software Maple. The differentiation of (4.102) thereby
yielded (4.103).
It should be mentioned that in practice this element may not be very useful. To
explain the reason for this one can think of a load case where both wind and snow

110
4.5. CATENARY ELEMENTS

loads are acting. In the load analysis the design snow load is already on the cable
when the wind load is applied. If we apply the wind load at the nodes, the cable will
stretch and the total load on it will grow although the snowing has stopped. This
behaviour is due to the formulation of the element, and from the discussion here it
has less physical justification.

4.5.3 Convergence of solution

In order to obtain the equivalent nodal forces and tangent stiffness matrix for the
elastic and associate catenary elements, either equation (4.75) or (4.102) is solved
by the classic Newton-Raphson method [3]. Due to the nature of this iterative tech-
nique, convergence to the correct solution cannot be guaranteed unless the original
algorithm is modified. How this modification should be formulated is determined
by the behaviour of the function considered.
The relationship between the horizontal force H and the unstrained length L0 , given
by equation (4.102), is shown in Figure 4.13. It can be seen that for an element
where the unstrained length L0 is shorter than the chord length C, there is a unique
relation between H and L0 . However, for a slack element where L0 > C, three
different values for H can be found, each representing a certain equilibrium state.
These three values are indicated by H1 , H2 and H3 in Figure 4.13. Only one value,
that is H = H1 , corresponds to the correct solution. For the other two extraneous
solutions H is negative and requires the element to be in compression, Figure 4.14.
Thus, for a taut element convergence is always to the correct solution, while for a
slack element the solution may converge to any of the three solutions of which only
one is correct [1].
By studying the relation between H and L0 shown in Figure 4.13, Ahmadi-Kashani
suggests the following modified Newton-Raphson algorithm
 $  dF 
H − F (H ) for Hi+1 > 0,
i i
Hi+1 = dH i (4.113)

Hi /2 for Hi+1 < 0,

in order to avoid the unwanted solutions. The objective of this modification is


to change an initial overestimate of H, which may cause convergence to a wrong
solution, to an initial underestimate of H. This is due to the fact that an initial
underestimate of H will always converge to the correct solution [1].
Although this algorithm will converge to the correct solution, the number of itera-
tions needed depends on the starting value for H. Ahmadi-Kashani [1, 3] presents
two initial estimates depending on the element being slack or taut:
Case 1. In this case it is assumed that the unstrained length of the cable is longer
than its chord length. If the small effect of elasticity is ignored, the length of the
cable is described by
l2
L20 = 2 sinh2 η + h2 . (4.114)
η

111
CHAPTER 4. FINITE CABLE ELEMENTS

L0 (m)

90
COMPRESSION 80 TENSION
L0 > C
70
60
50 L0 = C
40
30
H3 H2 20 H1
−105 −104 −103 −102 −101 −100 −10−1 10−1 100 101 102 103 104 105
H (kN)

Figure 4.13: Three possible solutions for catenary elements. Redrawn from [1].

H3
H1 H2

H1 H2 H3

(a) (b) (c)

Figure 4.14: Element configurations for the three different solutions: (a) H = H1 ,
(b) H = H2 , and (c) H = H3 . Redrawn from [1].

A non-dimensional geometric parameter δ is defined as:


 2 1/2
L0 − h2
δ= . (4.115)
l2
Using this parameter it is shown in [1] that the following expressions provide a good
estimate for the horizontal force H

1/2

 (120δ − 20)1/2 − 10 for 1 < δ ≤ 3.67,
η≈

2.337 + 1.095 ln δ − 0.00473 (7.909 − ln δ)2.46 for 3.67 < δ < 4.5 · 105 .
(4.116)
After η has been obtained from (4.116), H can be calculated by (4.11).
Case 2. In this case it is assumed that the unstrained length of the cable is shorter

112
4.6. COMPARISON OF ELEMENTS

than its chord length. The effect of element elasticity cannot be neglected and,
therefore, equation (4.116) is not applicable. An initial estimate for the cable force
T is in this case provided by the following equation [1, 3]:
%
1/3 1/3
b2 + a2 /3 for b2 > a2 ,
T ≈ 1/3 (4.117)
a2 + b2 /2a22 for b2 ≤ a2 ,

where  
C − L0
a2 = AE (4.118)
C
and
q02 L20
b2 = AE cos2 β. (4.119)
24
Once T has been obtained a starting value for H is found from the following relation:
Tl
H= . (4.120)
C
It should be noted that for the particular case δ = 1, which, for example, occurs
when C = l = L0 and h = 0, an initial estimate cannot be found using the suggested
expressions. In that case a good initial value for H can be found by putting δ = 1.001
in (4.116). The use of the modified Newton-Raphson algorithm and the initial
estimates given above ensure a fast convergence to the correct solution.

4.6 Comparison of elements

Most of the elements presented in this chapter are developed by one author only.
Therefore, to confirm the reliability of the finite cable elements and the present im-
plementation of those it is necessary to analyse some simple structures using the
presented elements and compare the results to those by other authors. Due to the
complexity of the expressions for the nodal forces and tangent stiffness matrices,
given in the preceding sections, differences between the equations are difficult to
distinguish. Therefore, a numerical comparison between the nodal forces and some
selected stiffness components will be presented for a single cable suspended at dif-
ferent heights, Figure 4.15. Included in this comparison are: all elements in this
chapter and the elastic catenary element by Jayaraman and Knudson [52]. In order
to get a verification of the present computer implementation of the elements, the
cable in Figure 4.15, which was presented in [3], will be used in the force and stiffness
comparison studies. Furthermore, two simple examples, which have been studied by
many authors [52, 74, 84, 103, 129] are analysed.

4.6.1 Comparison example 1

In this example, the nodal forces and stiffness components of the tangent stiffness
matrices will be compared. For the comparison, the single suspended cable shown

113
CHAPTER 4. FINITE CABLE ELEMENTS

in Figure 4.15 will be used. The following values are used: AE = 51 MN, q0 =
0.04 kN/m and L0 = 80.0 m. Following reference 3, the nodal forces and stiffness
components will be presented for four different chord lengths C and six angles θ. A
comparison of nodal forces was not made in [3], but for the stiffness components the
values obtained here will be compared with the values obtained by Ahmadi-Kashani.
The results are presented in the Tables 4.5–4.8 on the pages 118–121.

z
j

β
i x

Figure 4.15: Configuration for the cable used in the comparison of nodal forces and
stiffness components.

4.6.2 Comparison example 2

This problem, which consists of a suspended cable subjected to uniform and concen-
trated loads, was first considered by Michalos and Birnstiel [74] and later analysed
by O’Brien and Francis [84], Saafan [103], and Jayaraman and Knudson [52]. The
initial configuration and the data for this structure are found in Figure 4.16 and
Table 4.1, respectively. The results from the computations are shown in Table 4.2
together with the results obtained by other authors.

1 x 3
8 kip 100 ft
400 ft 2
500 ft
1000 ft

Figure 4.16: Prestressed cable under self-weight and concentrated load. Redrawn
from [52].

114
4.6. COMPARISON OF ELEMENTS

Table 4.1: Initial data for the structure in Figure 4.16.


Description Magnitude
Cross-sectional area of cable 0.85 in2
Equivalent modulus of elasticity 19000 kips/in2
Self weight of cable 3.16 lb/ft
Segment 1–2 412.8837 ft
Unstrained length
Segment 2–3 613.0422 ft

Table 4.2: Displacement at load point for the structure in Figure 4.16.
Displacements (ft)
Investigator Element type
Vertical Horisontal
Saafan [103] Straight bar −17.954 −2.774
O’Brien & Francis [84] Elastic catenary −18.460 −2.820
Michalos & Birnstiel [74] Straight bar −17.953 −2.773
Jayaraman & Knudson [52] Straight bar −17.951 −2.772
Jayaraman & Knudson [52] Elastic catenary −18.458 −2.819
Present Elastic parabola −18.377 −2.842
Present Elastic catenary −18.457 −2.819
Present Associate catenary −18.555 −2.820
Present Elastic catenarya −18.457 −2.819
a
Elastic catenary by Jayaraman and Knudson [52]

4.6.3 Comparison example 3

Here a slightly more complex structure will be studied. The structure considered is
the prestressed cable net shown in Figure 4.17. This structure was first studied by
Saafan [103] and subsequently analysed by West and Kar [129], and Jayaraman and
Knudson [52]. Initial data are given in Table 4.3. It seems that these values were
chosen arbitrarily and therefore the structure is not in equilibrium in the assumed,
prestressed configuration. The results for this example are shown in Table 4.4 to-
gether with the results by other authors.

115
CHAPTER 4. FINITE CABLE ELEMENTS

z y

100 ft x
1 2
f
100 ft f
f
3 4 f

100 ft
Assumed initial configuration
f = 30 ft
100 ft 100 ft 100 ft

Figure 4.17: Prestressed cable net under vertical loads. Redrawn from [52].

Table 4.3: Data for the assumed configuration structure shown in Figure 4.17.

Description Magnitude
Cross-sectional area of cables 0.227 in2
Equivalent modulus of elasticity 12000 kips/in2
Self weight of cablea 0.0001 kip/ft
Horizontal members 5.459 kips
Prestressing force
Inclined members 5.325 kips
Load acting vertically downward at nodes 1, 2, 3 and 4 8.0 kips
a
A small self weight is assumed for the analytical cable elements

116
4.6. COMPARISON OF ELEMENTS

Table 4.4: Displacement for node 4 under concentrated loads for the structure in
Figure 4.17.
Displacements of node 1 (ft)
Investigator Element type
x-dir. y-dir. z-dir.
Saafan [103] Straight bar −0.1324 −0.1324 −1.4707
West & Kar [129] Straight bar −0.1325 −0.1324 −1.4698
Jayaraman & Knudson [52] Elastic catenary −0.1300 −0.1319 −1.4643
Jayaraman & Knudson [52] Straight bar −0.1322 −0.1322 −1.4707
Present Straight bar −0.1322 −0.1322 −1.4707
Present Elastic parabola −0.1338 −0.1338 −1.4873
Present Elastic catenary −0.1328 −0.1328 −1.4764
Present Associate catenary −0.1338 −0.1338 −1.4874
Present Elastic catenarya −0.1328 −0.1325 −1.4765
a
Elastic catenary by Jayaraman and Knudson [52]

4.6.4 Conclusions from the comparisons

From the element comparison the following observations are made:

• The difference between horizontal and maximum cable forces for the associate
and elastic catenary elements is limited when the cable is slack. H and T for
the two different formulations for the elastic catenaries are exactly the same.

• Like the forces, the stiffness components for the two formulations for the elastic
catenaries give identical results in almost all cases. Small differences occur for
C = 80 and 80.7 m. For the associate catenary the differences are more
frequent but still very small. The reason for this might be the errors in some
of the equations in [2], which were reported in section 4.5.2 or the chosen
tolerance for which the computations were discontinued.

• The numerical results above demonstrate that the elastic catenary can simulate
a weight-less cable very well.

117
Table 4.5: Horizontal force H and maximum cable force T (kN)
β (deg) Element type H(60.0)a H(70.0) H(80.0) H(80.7) T (60.0)b T (70.0) T (80.0) T (80.7)
Straight bar, Eeq – – – 446.250000 – – – 446.250000
Parabolic 0.848507 1.512016 27.918128 442.492270 1.469682 2.060629 27.963939 442.495213
0 Elastic catenary 0.888008 1.543304 27.904380 446.359216 1.829907 2.223013 27.950214 446.362084
Associate catenary 0.888027 1.543356 28.012831 442.434778 1.829950 2.223089 28.058537 442.437722
Elastic catenaryc 0.888008 1.543304 27.904380 446.359216 1.829907 2.223013 27.950214 446.362084
Straight bar, Eeq – – – 444.551884 – – – 446.250000
Parabolic 0.838857 1.494822 27.670871 440.806749 1.521051 2.135766 27.960610 442.633591
5 Elastic catenary 0.881930 1.532164 27.727619 444.659844 1.932461 2.340404 28.018741 446.500682
Associate catenary 0.881949 1.532215 27.741897 440.750759 1.932508 2.340485 28.033176 442.577961
Elastic catenary 0.881930 1.532164 27.727619 444.659844 1.932461 2.340404 28.018741 446.500682
Straight bar, Eeq – – – 404.439850 – – – 446.250000
Parabolic 0.631662 1.125619 22.191992 401.000859 1.519621 2.117649 25.133374 443.075675
25 Elastic catenary 0.744692 1.281188 23.684242 404.520855 2.289398 2.703852 26.854539 447.018261
Associate catenary 0.744706 1.281227 23.680705 400.973085 2.289457 2.703945 26.851040 443.109723
CHAPTER 4. FINITE CABLE ELEMENTS

Elastic catenary 0.744692 1.281188 23.684242 404.520855 2.289398 2.703852 26.854539 447.018261
– – – 315.546401 – – – 446.250000

118
Straight bar, Eeq
Parabolic 0.299996 0.534606 12.436112 312.829318 1.187058 1.615572 18.404714 443.215200
45 Elastic catenary 0.483408 0.807754 15.659681 315.584355 2.547063 2.891783 23.325135 447.437426
Associate catenary 0.483415 0.807772 15.667895 312.828976 2.547130 2.891876 23.337268 443.550704
Elastic catenary 0.483408 0.807754 15.659681 315.584355 2.547063 2.891783 23.325135 447.437426
Straight bar, Eeq – – – 188.593399 – – – 446.250000
Parabolic 0.064049 0.114140 3.741958 186.959046 0.647669 0.844192 9.471374 443.001083
65 Elastic catenary 0.209004 0.325844 6.638973 188.600989 2.712710 2.955816 17.217899 447.720119
Associate catenary 0.209007 0.325849 6.641730 186.961717 2.712782 2.955904 17.224896 443.854031
Elastic catenary 0.209004 0.325844 6.638973 188.600989 2.712710 2.955816 17.217899 447.720119
Straight bar, Eeq – – – 38.893250 – – – 446.250000
Parabolic 0.000562 0.001001 0.094025 38.555887 0.111009 0.133465 1.217796 442.519323
85 Elastic catenary 0.023131 0.031125 0.475646 38.893159 2.795775 2.995888 7.207200 447.844763
Associate catenary 0.023133 0.031126 0.474635 38.555923 2.795714 2.995938 7.196160 443.989433
Elastic catenary 0.023131 0.031125 0.475646 38.893159 2.795775 2.995888 7.207200 447.844763
a
Horizontal force H for chord length C
b
Maximum cable force T for chord length C
c
Elastic catenary by Jayaraman and Knudson [52]
Table 4.6: Stiffness component K44 (kN/m)
a
β (deg) Element type K(60.0) ∆K(60.0)b K(70.0) ∆K(70.0) K(80.0) ∆K(80.0) K(80.7) ∆K(80.7)
Straight bar, Eeq – – – – – – 631.655484 0.000231
Parabolic 0.028281 – 0.086374 – 212.500000 – 626.168427 –
0 Elastic catenary 0.041931 0.000000 0.106723 0.000000 212.709289 0.000000 637.188184 0.000000
Associate catenary 0.041932 −0.000002 0.106730 −0.000009 214.204762 −1.650363 626.267917 −0.038286
Elastic catenaryc 0.041931 – 0.106723 – 212.709265 – 637.188184 –
Straight bar, Eeq – – – – – – 626.901728 −0.041780
Parabolic 0.027960 – 0.085230 – 210.888461 – 621.458426 –
5 Elastic catenary 0.041654 0.000000 0.105809 0.000000 211.096344 0.000000 632.392428 −0.000001
Associate catenary 0.041656 −0.000002 0.105816 −0.000009 211.160614 −0.217011 621.634540 −0.037004
Elastic catenary 0.041654 – 0.105809 – 211.096325 – 632.392428 –
Straight bar, Eeq – – – – – – 519.871708 −0.987519
Parabolic 0.021156 – 0.061449 – 174.600851 – 515.395564 –
25 Elastic catenary 0.035543 0.000000 0.085890 0.000000 174.781096 0.000000 524.416773 0.000000
Associate catenary 0.035544 −0.000001 0.085895 −0.000007 174.620453 0.052123 516.995976 −0.010611
Elastic catenary 0.035543 – 0.085890 – 174.777391 – 524.413889 –
– – – – – – 318.671286

119
Straight bar, Eeq −2.764841
Parabolic 0.010606 – 0.026998 – 106.359921 – 315.945260 –
45 Elastic catenary 0.024590 0.000000 0.052044 0.000000 106.508412 0.000000 321.438571 0.000000
Associate catenary 0.024591 −0.000001 0.052046 −0.000002 106.573848 −0.111539 318.624227 0.025888
Elastic catenary 0.024590 – 0.052044 – 106.507574 – 321.438570 –
Straight bar, Eeq – – – – – – 117.405852 −4.542093
Parabolic 0.002977 – 0.005925 – 38.044727 – 116.395631 –
65 Elastic catenary 0.013870 0.000000 0.023058 0.000000 38.179209 0.000000 118.394574 0.000000
Associate catenary 0.013870 0.000000 0.023058 0.000000 38.201664 −0.031623 117.980625 0.044986
Elastic catenary 0.013870 – 0.023058 – 38.179207 – 118.394531 –
Straight bar, Eeq – – – – – – 10.288241 −5.487735
Parabolic 0.000108 – 0.000168 – 1.627559 – 10.199018 –
85 Elastic catenary 0.005679 0.000000 0.006847 0.000000 1.700314 0.000000 10.330274 0.000000
Associate catenary 0.005680 −0.000001 0.006847 0.000000 1.693119 0.007082 10.280973 0.047929
Elastic catenary 0.005679 – 0.006847 – 1.700310 – 10.330293 –
a
Stiffness component K for chord length C
b
Difference in stiffness components defined as: K from [3] subtracted by K in this study
4.6. COMPARISON OF ELEMENTS

c
Elastic catenary by Jayaraman and Knudson [52]
Table 4.7: Stiffness component K66 (kN/m)
a
β (deg) Element type K(60.0) ∆K(60.0)b K(70.0) ∆K(70.0) K(80.0) ∆K(80.0) K(80.7) ∆K(80.7)
Straight bar, Eeq – – – – – – 5.529740 −5.529740
Parabolic 0.014142 – 0.021600 – 0.348977 – 5.483176 –
0 Elastic catenary 0.022873 0.019058 0.027786 0.078937 0.349186 0.000000 5.531117 0.000000
Associate catenary 0.022874 0.019056 0.027788 0.078932 0.350541 −0.001354 5.482487 0.048630
Elastic catenaryc 0.022873 – 0.027786 – 0.349186 – 5.531117 –
Straight bar, Eeq – – – – – – 10.285886 −5.491003
Parabolic 0.014141 – 0.021924 – 1.958746 – 10.197993 –
5 Elastic catenary 0.022912 0.000000 0.028103 0.000000 1.957877 0.000000 10.329228 −3.145456
Associate catenary 0.022913 0.000000 0.028105 −0.000001 1.959130 -0.001839 10.201084 −3.051004
Elastic catenary 0.022912 – 0.028103 – 1.957877 – 10.329228 –
Straight bar, Eeq – – – – – – 117.369714 −4.510313
Parabolic 0.013690 – 0.027246 – 38.205226 – 116.359613 –
25 Elastic catenary 0.023531 −0.000100 0.033981 0.000000 38.166263 0.000000 118.357913 −3.113978
Associate catenary 0.023532 0.000000 0.033984 −0.000001 38.134140 0.011373 116.731535 −2.170844
CHAPTER 4. FINITE CABLE ELEMENTS

Elastic catenary 0.023531 – 0.033981 – 38.165457 – 118.357310 –


– – – – – – 318.671286

120
Straight bar, Eeq −2.099717
Parabolic 0.010606 – 0.026998 – 106.359921 – 315.945260 –
45 Elastic catenary 0.023252 0.000000 0.037513 0.000000 106.139936 0.000000 321.435840 −2.484440
Associate catenary 0.023253 0.000000 0.037516 −0.000002 106.210122 −0.111544 318.682722 −0.901773
Elastic catenary 0.023252 – 0.037513 – 106.139098 – 321.435840 –
Straight bar, Eeq – – – – – – 520.037904 −0.987640
Parabolic 0.004601 – 0.013365 – 174.565951 – 515.564451 –
65 Elastic catenary 0.021216 0.000000 0.028604 0.000000 173.673255 0.000000 524.579643 0.000000
Associate catenary 0.021218 0.000000 0.028606 −0.000002 173.782197 −0.145141 522.944803 0.035348
Elastic catenary 0.021216 – 0.028604 – 173.673245 – 524.579484 –
Straight bar, Eeq – – – – – – 627.209367 −0.042005
Parabolic 0.000214 – 0.000653 – 210.885926 – 621.771224 –
85 Elastic catenary 0.020019 0.000000 0.020149 0.000000 201.455127 0.000000 632.697081 0.000000
Associate catenary 0.020021 0.000000 0.020150 0.000000 200.541388 0.904867 632.586334 0.047290
Elastic catenary 0.020019 – 0.020149 – 201.454638 – 632.697508 –
a
Stiffness component K for chord length C
b
Difference in stiffness components defined as: K from [3] subtracted by K in this study
c
Elastic catenary by Jayaraman and Knudson [52]
Table 4.8: Stiffness component K46 (kN/m)
a
β (deg) Element type K(60.0) ∆K(60.0)b K(70.0) ∆K(70.0) K(80.0) ∆K(80.0) K(80.7) ∆K(80.7)
Straight bar, Eeq – – – – – – 0.000000 0.000000
Parabolic 0.000000 – 0.000000 – 0.000000 – 0.000000 –
0 Elastic catenary 0.000000 0.000000 0.000000 0.000000 0.000004 0.000949 0.000000 0.001250
As. cat. K46 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
As. cat. K64 0.000001 −0.000001 0.000003 −0.000003 0.006724 −0.000052 0.019990 0.000000
Elastic catenaryc 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
Straight bar, Eeq – – – – – – 54.363005 0.480134
Parabolic 0.001218 – 0.005581 – 18.419973 – 53.890853 –
5 Elastic catenary 0.001319 0.000000 0.005605 0.000000 18.417802 0.000000 54.843022 0.000000
As. cat. K46 0.001319 0.000000 0.005606 −0.000001 18.423413 −0.018971 53.914194 −0.007344
As. cat. K64 0.001320 0.000000 0.005609 −0.000001 18.430077 −0.018977 53.925997 −0.007487
Elastic catenary 0.001319 – 0.005605 – 18.417800 – 54.843022 –
Straight bar, Eeq – – – – – – 239.841599 2.118072
Parabolic 0.004449 – 0.020381 – 81.274988 – 237.776264 –
25 Elastic catenary 0.005233 0.000000 0.022315 0.000000 81.247867 −0.000001 241.959426 0.000000
As. cat. K46 0.005233 −0.000001 0.022316 −0.000002 81.172982 0.024283 238.557555 −0.026916
As. cat. K64 0.005234 −0.000001 0.022319 −0.000001 81.179087 0.024284 238.539786 −0.027548
Elastic catenary 0.005233 – 0.022315 – 81.246142 – 241.958136 –
Straight bar, Eeq – – – – – – 313.141547 3.430022

121
Parabolic 0.003535 – 0.016198 – 106.140079 – 310.463135 –
45 Elastic catenary 0.005163 0.000000 0.022156 0.000000 106.046946 0.000000 315.906789 0.666610
As. cat. K46 0.005163 0.000000 0.022157 −0.000002 106.112197 −0.111394 313.187830 2.200997
As. cat. K64 0.005164 0.000000 0.022159 −0.000002 106.117036 −0.060671 313.154858 2.239225
Elastic catenary 0.005163 – 0.022156 – 106.046109 – 315.906789 –
Straight bar, Eeq – – – – – – 239.919097 2.118016
Parabolic 0.000967 – 0.004433 – 81.349830 – 237.855438 –
65 Elastic catenary 0.002254 0.000000 0.009515 0.000000 81.173395 0.000000 242.037885 0.000000
As. cat. K46 0.002254 −0.000001 0.009515 −0.000001 81.221285 −0.067641 241.289523 −0.005948
As. cat. K64 0.002254 0.000000 0.009516 −0.000001 81.224258 −0.067644 241.261681 −0.006585
Elastic catenary 0.002254 – 0.009515 – 81.173390 – 242.037804 –
Straight bar, Eeq – – – – – – 54.389920 0.480114
Parabolic 0.000009 – 0.000043 – 18.448948 – 53.918340 –
85 Elastic catenary 0.000139 0.000000 0.000483 0.000000 18.129526 0.000000 54.870271 0.000000
As. cat. K46 0.000139 0.000000 0.000483 0.000000 18.048903 0.079253 54.863038 −0.000062
As. cat. K64 0.000139 0.000000 0.000483 0.000000 18.049581 0.079256 54.856529 −0.000207
Elastic catenary 0.000139 – 0.000483 – 18.129481 – 54.870397 –
4.6. COMPARISON OF ELEMENTS

a
Stiffness component K for chord length C
b
Difference in stiffness components defined as: K from [3] subtracted by K in this study
c
Elastic catenary by Jayaraman and Knudson [52]
Chapter 5

Static analysis

In this chapter, it will be demonstrated how a large prestressed cable roof structure
is analysed using the finite element method. The structure chosen for the analysis is
the Scandinavium Arena in Gothenburg. It was intentioned to also analyse another
large Swedish cable roof structure—the Johanneshov Ice Stadium in Stockholm.
However, it was found that this pioneering cable truss structure by David Jawerth
had already been extensively analysed under different loadings, see for example
references 1, 76 and 78.

5.1 Static analysis of the Scandinavium Arena

The analysis of the Scandinavium Arena was performed using a geometrically non-
linear finite element method. Data about the Scandinavium Arena were obtained
from construction drawings, articles [55, 56], a Master’s thesis [86] and unpublished
material [104]. Further, the results from the calculations were compared to the
results from previous calculations [86], and results from a simplified method recently
presented at the Department of Structural Engineering at the Royal Institute of
Technology in Stockholm, Sweden.

5.1.1 The Scandinavium Arena—background

Before the analysis begins some background information on the Scandinavium Arena
will be given. An extensive description can be found in reference 55.
In 1948 an architect competition concerning an indoor sports building in central
Gothenburg was announced. It was won by a working group led by the architect
Poul Hultberg. In 1962 the preliminary design works started and a final decision
concerning the realisation of the structure was taken in June 1969. In May 1971
the Scandinavium Arena was completed. With space for 14000 spectators it was
at the time the largest covered arena in northern Europe [56]. The arena has been
and is still used for activities such as concerts, theatre shows, ice-hockey, soccer,

123
CHAPTER 5. STATIC ANALYSIS

Figure 5.1: The Scandinavium Arena after completion. A pylon is seen almost in
the middle of the view. Reproduced from [55].

swimming, etc.
Roof structure. The roof consists of a prestressed cable net cladded with thermal
and water insulated corrugated steel plates. All cables are anchored in a space-
curved reinforced concrete ring. The concrete ring is supported by four stiff pylons
and 40 circular columns. The surface of the roof conforms nearly with a hyperbolic
paraboloid. From the centre point of the roof the hanging cables rise 10 m to the
top and the bracing cables fall 4 m to the valley of the ring. The cable spacing is
nearly constant and equals to 4 m in both directions. Foundation. The building
is supported partly by rock and partly by concrete piles. Two of the pylons are
supported by concrete foundations that rest on 115 piles. The large number of piles
needed is due to the horizontal forces that occur at the connections between the ring
beam and the pylons.
Ring beam. The ring beam has a rectangular cross-section with a width of 3.5 m and
a height of 1.2 m. An alternative solution with a ring beam made as a hollow steel
box was investigated during the design work, but it was found to be too expensive.
Columns and pylons. The circular columns are cast in place and designed to carry
mainly axial forces. The pylons consist of radially oriented concrete walls, with a
side length of 3.5 m connected by beams, Figure 5.2. The space between the walls
is approximately 3.5 m wide and filled with ventilation equipment. The pylons are
relatively stiff and can take large horizontal forces. Therefore, the ring beam is
discontinuous at the top of the pylons which affects the prestressing forces in the
cable in the areas between the pylons and the top of the ring beam. The forces
in the bracing cables are there significantly smaller than in other parts of the roof.
Tension rods. The colour telecasting (remember that the arena was built in 1971)
required that the light and sound systems had to be stable. Therefore, it was not
considered suitable to attach the systems directly to the roof. Instead, the light

124
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

Figure 5.2: The arena after completed erection of the cables. Reproduced from [55].

and sound systems are suspended in a cable system supported by the pylons. This
radially oriented cable system also serves as tension rods for the ring beam, which
is discontinuous at the pylons, Figure 5.5.
Original calculations. Preliminary dimensions of the ring beam and the cables were
estimated by a simplified method in which the cable net was represented by a shear-
free membrane. The stiffness of the ring beam was taken as the stiffness of a plane
ring beam with the same dimensions as the real one and supported at four stiff
pylons [56]. The membrane stresses were approximated by sectionally constant
values in each direction and the deflection of the roof by polynomials [56, 104]. The
unknowns were determined from equations expressing the vertical equilibrium of the
membrane and the compatibility between the membrane and the ring beam. Axial
forces and bending moments in the ring beam due to snow and wind loads were
modified with respect to the inclination of the ring beam. Accurate values of the
twisting moment could not be obtained by the simplified method.
Thereafter, a more accurate analysis was performed using a mixed finite element
method. In that method, the structure was divided into two substructures, the
cable net and the ring beam on columns. The two substructures were analysed sep-
arately by the stiffness method and then connected by compatibility and equilibrium
expressions. Since the analysis was non-linear, the substructures had to be itera-
tively connected. The maximum deflection of the roof surface under full snow load
was found to be 64 cm. Comparison with the simplified method showed a difference
of at most 10 % in bending moments in the ring [56]. More results from these finite
element calculations will be presented in section 5.1.5.

125
CHAPTER 5. STATIC ANALYSIS

Figure 5.3: Erecting the sheet roofing. Reproduced from [55].

5.1.2 Prestressing forces

A cable net with a fine mesh can be assumed to behave like a membrane free of shear
stresses. In that case, it is possible to obtain an analytical solution by introducing a
number of assumptions concerning the load distribution and behaviour of the cables
and ring beam, see for example [116]. In this section, a simplified approach will be
used to determine the magnitude of the initial prestressing forces in the cables.
According to reference 86, the roof surface of the Scandinavium Arena is described
by the following equation:

x 2
y 2

fx − fy for x ≤ xp ,

y 2 3  x − x 2
R R
z=
x 2 (5.1)

fx − fy − fy
p
for x ≥ xp ,
R R 4 R

where xp is the x-coordinate of the pylon, and fx , fy are defining height measures.
R is the radius of the horizontal projection of the ring beam.

126
127
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

Figure 5.4: Construction drawing K27:1 for the Scandinavium Arena, showing primarily the dimensions of the concrete ring beam.
CHAPTER 5. STATIC ANALYSIS

128
Figure 5.5: Construction drawing K27:3 for the Scandinavium Arena, showing primarily the configuration of the cable net (horizontal
projection).
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

z
R

fx
y
x
fy

Figure 5.6: A hyperbolic paraboloid with a circular plane projection.

In the original preliminary calculations, it is assumed that the shape of the roof
surface is described by the first expression in (5.1), i.e. the equation of the hyperbolic
paraboloid [104]:

x 2
y 2
z = fx − fy . (5.2)
R R
Using polar coordinates the equation of the ring beam can be described as:


xring = R cos θ,
yring = R sin θ, (5.3)


zring = fx cos θ − fy sin θ.
2 2

The equilibrium equations in x-, y- and z-directions for a membrane free of shear
forces, can be stated as [119]:

∂H x
+ Fx = 0, (5.4)
∂x
∂Hy
+ Fy = 0, (5.5)
∂y
   
∂  ∂z ∂  ∂z
Hx + Hy + Fz = 0, (5.6)
∂x ∂x ∂y ∂y

where Hx, H y are the horizontal components of the prestressing force distribution
(N/m) in x- and y-directions, respectively. Fx , Fy , Fz are the load intensities (N/m2 )
is the x-, y- and z-directions, respectively. With only vertical loads, (5.4)–(5.6)

129
CHAPTER 5. STATIC ANALYSIS

simplify to:
∂Hx
= 0, (5.7)
∂x
∂Hy
= 0, (5.8)
∂y
2 2
x ∂ z + H
H  y ∂ z + Fz = 0. (5.9)
∂x2 ∂y 2
x = H
Inserting (5.2) into (5.9) and prescribing H y = H
 0 yields [104]:

0 = − Fz R2
H (5.10)
2(fx − fy )
Equation (5.10) shows that H  0 increases as fy increases. For the Scandinavium
Arena Fz = −0.6 kN/m (the combined weight of the cladding and cables), fx = 10
2

m, fy = 4 m, and R = 54 m. These values gives H  0 = 145.8 kN/m. The horizontal


component of the force in a single cable is obtained by multiplying H  0 with the

cable spacing. For non-equidistant cable spacing, H0 is multiplied with the sum of
half the distance between adjacent cables, e.g. for a cable i in the y-direction it is
(xi+1 − xi−1 )/2.

5.1.3 Finite element model

The Scandinavium Arena will in this section be analysed with finite elements, as
a comparison to the original calculations. As a demonstration, the comparison
will be performed for only one load case: uniformly distributed dead load of −0.6
kN/m2 and snow load of −0.75 kN/m2 on the whole roof. Due to symmetry in both
structure and load case only a quarter of the structure had to be modelled for this
case, Figure 5.7.

x
y

Figure 5.7: One quarter of the Scandinavium Arena.

The finite element model of a quarter of the structure is shown in Figure 5.8. The
beam nodes on the symmetry lines x = 0 and y = 0 have the following boundary

130
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

conditions: θy = 0, θz = 0, and ux = 0. Cable nodes on x = 0 are prevented to


move in the y-direction and vice versa for the other symmetry line. All columns are
pin-jointed to the ground and to the ring beam, while the pylon is rigidly connected
at both ends.

Figure 5.8: Element model.

All cables are modelled using the elastic catenary element presented in Chapter 4.
This element was chosen instead of the bar element to avoid the problems with local
mechanisms and numerical instability due to slackening cables. The ring beam and
the pylon were modelled by non-linear three-dimensional beam elements based on
cubic shape functions. The co-rotational approach is used to compute the tangent
stiffness matrix and internal force vector [90]. The Matlab routines for the beam
element have been developed by Costin Pacoste and have been used to analyse
other structures [85]. Since the ring beam is relatively stiff, the results using the
non-linear beam elements were compared with results using linear three-dimensional
beam elements. The differences in the results were very small. Despite the small
difference, the non-linear beam element was used for the analyses in this chapter.
It should be mentioned that some of the beam elements are very short and stiff,
and therefore do not fit into the beam assumption. Still, the beam model is used
for these elements. A more accurate analysis, which avoids the short elements but
keeps the cable spacing would require the use of solid elements. However, some
difficulties in connecting the solid elements and the cable elements may arise since
the solids cannot cope with high concentrated load in the same way as the beam
elements. The columns were modelled with straight bar elements. The magnitudes
of the prestressing forces are computed according to subsection 5.1.2. The radially
oriented tension rod, shown in Figure 5.5, was not included in the finite element
model of the structure.
As mentioned earlier, the main difference in analysis between cable structures and
other structures, such as frames and trusses, is that the initial configuration is
generally unknown for cable structures. According to Møllmann [76], the following
iterative procedure is used for a cable structure with an elastic boundary structure
(arches or beams):

131
CHAPTER 5. STATIC ANALYSIS

1. Assuming that the boundary joints are fixed in the positions corresponding to the un-
stressed state of the arch, the shape of the cable net is determined corresponding to
cables in vertical planes.

2. The cable forces at the boundary joints obtained from the previous stage are now re-
garded as external loads acting on the arch. The arch is then analysed separately for
these forces and for the weight of the arch members.

3. Keeping the boundary joints fixed in the positions obtained from stage 2, the shape of
the net is now recalculated (cables in vertical planes or geodesic net).

4. Return to 2.

With this procedure, the boundary structure and the cable net are calculated sep-
arately until the displacement changes of the joint coordinates of the cable net and
boundary are sufficiently small. A somewhat similar procedure was used in the
analysis of the Scandinavium Arena. The only difference is that at step 2 the pre-
tensioned cable net is numerically attached to the unstressed ring beam. This means
that the stiffness of the whole structure is used to compute the displacements of the
ring beam. Some cables will be unloaded during step 2, but since the ring beam is
quite stiff only 3–4 iterations are needed to get a deviation in the horizontal com-
ponent of the cable forces of a most 0.5 %. What this error corresponds to in the
unstrained length of a cable will now be checked. For a bar the following equation
holds:
∆L L − L0
T = AE = AE . (5.11)
L0 L0
This equation can be written as:
L
L0 = . (5.12)
T /AE + 1

The cables for the Scandinavium Arena have AE = 343 MN, T ≈ 145.8·4 = 583 kN.
Assuming a cable length of 108 m, the error in unstrained length is 0.92 mm. This
tolerance can not be reached in practice. However, one should bear in mind that a
small error in unstrained length may give large errors in force; for example, an error
in unstrained length of 0.05 % gives an error in force of 30 % with the cable data
given here. For the form-finding of the cable net the grid method (section 3.2.2)
was used. Nevertheless, equation (3.7) cannot be used in this case due to a non-
equidistant mesh. Instead, the following equation expressing the vertical equilibrium
is used:
   
zj − zi zi − zl zk − zi zi − zm
Hix − + Hiy − + Fiz = 0. (5.13)
xj − xi xi − xl y k − yi y i − ym

Equation (3.7) is a special case of (5.13). Note that yl = yi = yj and xm = xi = xk


in this case, Figure 3.1.
Material and cross-sectional properties for pylons, the ring beam, columns and cables
are given in Figure 5.10. More data, including coordinates for beam elements, node
loads, prestressing forces, etc., can be found in Appendix A.

132
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

(a) Initial equilibrium procedure—step 1

(b) Initial equilibrium procedure—step 2

(c) Initial equilibrium procedure—step 3

Figure 5.9: Numerical procedure to find the initial equilibrium configuration sug-
gested by Møllmann [76]: (a) the shape of the cable net is obtained by
assuming fixed nodes, (b) the cable forces are regarded as external loads
on the ring beam (step 1 and 2 are repeated until convergence), and (c)
the two structures are connected and the whole structure should now be
in equilibrium.

133
CHAPTER 5. STATIC ANALYSIS

z
650

Pylon
E = 32 GPa
G = 12.8 GPa
3500

y A = 4.55 m2

x
Ix = 0.566 m4
Iy = 4.645 m4
Iz = 17.021 m4
4500

z

Ring beam
E = 32 GPa
G = 12.8 GPa
1200


y A = 4.2 m2
x
Ix = 1.581 m4
3500 Iy = 0.504 m4
Iz = 4.288 m4

Column
800 E = 32 GPa
G = 12.8 GPa
A = 0.503 m2
Ix = 0.040 m4
Iy = 0.020 m4
Iz = 0.020 m4

Cable
E = 162 GPa
A = 2.12 · 10−3 m2

Figure 5.10: Cross-sectional and material properties for finite elements. Drawn di-
mensions in millimetres.

134
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

5.1.4 Calculation results

Some results from the present analysis of the Scandinavium Arena are presented
in Figures 5.11–5.16. The maximum deflection of the net due to the snow load
is 54 cm. Figure 5.11 shows that the contour curves of the net displacement are
almost circular. All cable forces increase from the pretension values due to the
snow load. The forces in the bracing cables increase as a result of the outwards
movement of the valley of the ring beam. The axial force diagram indicates that
the pylon significantly reduces the axial force in the lower part of the ring beam
between the pylons. Luckily, the bending moments in these parts are not that
large. A discontinuity in the bending moment diagram occurs due to effects from
the pylon. The ‘height’ of this discontinuity depends on the relation between the
bending stiffness of the ring beam and the torsional stiffness of the pylon. The
twisting moments are much lower than the bending moments. As for the bending
moment diagram, the pylon affects the twisting moment distribution. Some more
comments concerning the results are given in section 5.1.6, where the present results
are compared with previous results. Note that in all bending moment diagrams in
this chapter, a positive moment corresponds to tension at the outer side of the ring
beam.

(0 , 5 2 )
55 (− 1 ,5 1 )
(− 2 ,4 8 )
50 (− 4 ,4 3 )

45 (− 1 1 ,3 4 )

40
(− 2 3 ,2 4 )
−200
35
(− 3 6 ,1 4 )
y (m)

30
−300
−100
(− 5 0 ,6 )
25

−400
20
(− 6 2 ,2 )

15 −500
(− 7 1 ,0 )
10

5
(− 7 5 ,0 )

0
0 5 10 15 20 25 30 35 40 45 50 55
x (m)

Figure 5.11: Contour lines of net displacement and ring beam displacements in x–y
plane due to snow load (mm).

135
CHAPTER 5. STATIC ANALYSIS

400
380 379 378 376 373 373
371 368 362
351
350
322

300
Increase in force (kN)

248
250

200

150

100

50

0
0 5 10 15 20 25 30 35 40 45 50
y (m)

Figure 5.12: Increase in the horizontal component of the cable forces for cables in
the x-direction (hanging cables) due to snow load.

80
75
71
70
67
65

60 58
54
Increase in force (kN)

50 49
48
45
43
41 41
40

30

20

10

0
0 5 10 15 20 25 30 35 40 45 50
x (m)

Figure 5.13: Increase in the horizontal component of the cable forces for cables in
the y-direction (bracing cables) due to snow load.

136
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

Column
Pylon
0
Axial force (MN)

−5

−7.8 −8.1 −8.6


−7.7 −9.3 −8.9
−10

−13.9 −13.1 −13 −13 −12−11.6 −11.2


−14.3 −14.1 −10.8
−15
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.14: Axial force in ring beam under snow load.

Column
Pylon
15000 12988
Bending moment (kNm)

10204 10793
10000
7148
5312
5000 5037
3897
2927
0
−729

−5000
−4535

−10000 −7066
−8208
−15000 −8762 −9881
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.15: Bending moment around the local z-axis (stiff direction) for beam ele-
ments under snow load.

137
CHAPTER 5. STATIC ANALYSIS

Column
Pylon
2000
Twisting moment (kNm)

1515
1500

1000 856

472
500
209
0
0

−500 −595 −585


−687 0
−716 −647 −533 −384
−1000
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.16: Twisting moment around local x-axis for beam elements under snow
load.

5.1.5 Calculation results from 1972

In reference 86, the Scandinavium Arena is analysed with the finite element method.
Like the analysis above, the calculations were done on one quarter of the structure.
The ring beam was modelled with linear three-dimensional beam element and the
cables with straight bar elements. As mentioned above, the ring beam and the cable
net were analysed separately with the stiffness method and connected by a flexibility
approach. Two load cases were considered: snow load (−0.75 kN/m2 ) on the whole
roof and wind load (0.4 kN/m2 ) on the whole roof. The maximum deflection under
full snow load was 64 cm.
The finite element model in [86] differs from the model analysed in the previous
section. The most important differences between the two structural models are:

• Cable spacing. While the cable spacing according to the drawings was used
in the present analysis of the Scandinavium Arena, a non-equidistant mesh
of cables was used in [86]. The cables are connected to the joints between
the beam elements, to which also the columns and the pylon are connected,
Figure 5.17. It appears that this mesh has been chosen to reduce the number
of unknowns.
• Contour shape. No information on the x- and y-coordinates of the ring could
be obtained from any of the references 56, 86 and 104. Nonetheless, the ring

138
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

appears to be circular in the figures in [86]. In the present analysis the co-
ordinates of the ring beam were obtained from the construction drawing in
Figure 5.4. It is important to know the ‘exact’shape of the ring beam as it
considerably affects the distribution of the bending moment, see section 5.2.
• Cable forces. As mentioned above, the cable forces in the upper bracing cables
are lower than those in the lower bracing cables. In addition, the five lowest
bracing cables were also post-tensioned by introducing a gap between the ring
beam and the cable net. The final cable forces after the post-tensioning were
not given. Therefore, a uniform cable force distribution was used in the present
analysis.
• Young’s modulus. No information concerning the modulus of elasticity for
the concrete is given in [86]. According to the drawings, concrete K400 (old
notation) was used for the ring beam. A characteristic value of 32 GPa for the
modulus of elasticity for this concrete class is found in the Swedish building
codes. This value was used in the present analysis. The shear modulus is
0.4E, which corresponds to Poisson’s ratio of 0.2.

Some of the results from reference 86 are given in Figures 5.17–5.21 to facilitate
comparison.
y

(−1,85 )
(−3,79 )
−100 (−6,68 )

(−13,54 )
−200
−300 (−24,39 )

−400 (−39,27 )

(−54,17 )
−500

(−72,7 )

(−87,2 )
−600

(−99,1 )

(−105,0 )

x
−600 −500 −400 −300 −200 −100

Figure 5.17: Contour lines of net displacement and ring beam displacements in x–y
plane due to snow load (mm). Redrawn from [86].

139
CHAPTER 5. STATIC ANALYSIS

900
845
815
800 769
750

700
659
Increase in force (kN)

600
560

500

400
335
300 277

200

100
45
7
0
0 5 10 15 20 25 30 35 40 45 50 55 60
y (m)

Figure 5.18: Increase in the horizontal component of the force for cables in the x-
direction (hanging cables) due to snow load. Redrawn from [86].

700

612
600
558

500 490
472
Increase in force (kN)

426
400

300

200

100

0
−25 −14 −10 −9 −1

−100
0 5 10 15 20 25 30 35 40 45 50 55 60
x (m)

Figure 5.19: Increase in the horizontal component of the force for cables in the y-
direction (bracing cables) due to snow load. Redrawn from [86].

140
5.1. STATIC ANALYSIS OF THE SCANDINAVIUM ARENA

Column
Pylon
15000
Bending moment (kNm)

10000 10787 10003


8649 7659
6649
5000 2962
2452 1648

0
−2844
−3344
−5000
−6374
−10000
60 −9238 −7639
50 −9807 60
40 50
30 40
y (m) 20 30
20 x (m)
10 10
0 0

Figure 5.20: Bending moment around the local z-axis (stiff direction) for beam ele-
ments under snow load. Redrawn from [86].

Column
Pylon
1393
1500
1265
Twisting moment (kNm)

1000 943

500 504

0 121

−500 −57
−618 −633 −625 −248
−611 −485
−1000
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.21: Twisting moment around local x-axis for beam elements under snow
load. Redrawn from [86].

141
CHAPTER 5. STATIC ANALYSIS

5.1.6 Comparison of the results

The differences in midpoint deflection and in-plane displacements are probably due
to a higher Young’s modulus of the concrete in the present analysis and different
initial cable forces in the analyses. For both analyses, both the cable forces in the
x- and y-directions increase. This is due to the outwards displacement of the valley
of the ring beam. A detailed comparison of the increase in the cable forces can not
be done since the intial forces differ much in some parts of the roof. No axial forces
were presented in [86]. However, Figure 5.14 clearly shows that the pylon takes a
large horizontal force. Concerning the twisting moments, they agree very well; the
distributions of the twisting moments are similar and the values do not differ very
much. In the case with the bending moments, their distributions differs qualitatively
very much. The maximum magnitude of the bending moments is about the same
for the two analyses.
Thus, it can be concluded that all the results except the bending moment distri-
bution have a satisfactory agreement. The disagreement in the bending moment
distribution will be explained by another example in the next section.

5.2 Sensitivity of bending moment to the shape


of the ring beam

In order to explain the discrepancies in the bending moment distributions for the
calculations of the Scandinavium Arena and to verify the finite element program
another structure will now be analysed. The structure shown in Figure 5.22 has
been analysed by Møllmann [76] and was chosen because of its similarities with the
Scandinavium Arena.

5.2.1 Description of the structure

The system used for the calculation consists of nine hanging and nine bracing cables
in the net, and 28 straight space beam elements which form the ring beam. The joints
between the beam elements coincide with the joints where the cables are attached.
The ring beam is supported by vertical columns at all joints and it is assumed that
vertical displacement is prevented at these joints. There are three main supports:
joints 8, 15 and 22, Figure 5.22. In addition to the vertical constraints due to the
columns, joints 8 and 15 are prevented from moving in the y-direction and joint 15
cannot move in the x-direction.

142
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

y
z
1

26 4 8.60
5.40 x
25 5

52.00 52.00

22 8 x Dimensions in m
z

11 8.60
19 y
5.40
18 12

15 52.00 52.00

Figure 5.22: Geometry of the cable structure by Møllmann. Redrawn from [76].

All cables have the cross-sectional area A = 5.2 · 10−3 m2 and Young’s modulus
E = 160 GPa. The concrete ring beam is assumed to be elastic with the material
and cross-sectional properties given in Figure 5.23. The cross-section of the ring
beam twists along the perimeter. The longer side of the rectangle is parallel to
the tangent of the roof surface in the direction normal to the boundary. In the
initial state, the cables are in vertical planes and the cable joints are very nearly
located on a hyperbolic paraboloid, Figure 5.22. The initial state is the equilibrium
configuration where the cables are pretensioned and the combined weight of the
cables and the cladding acts on the net. The initial equilibrium configuration is
determined according to the procedure described in section 5.1.3. The horizontal
components of the cable forces in the initial state are 2600 kN for both the hanging
and the bracing cables.

z

Ring beam
E = 20 GPa
G = 10 GPa (ν = 0)
A = 2.6 m2
y
1300

Ix = 0.877 m4
x Iy = 0.366 m4
Iz = 0.867 m4

2000

Figure 5.23: Cross-sectional and material properties. Drawn dimensions in millime-


tres.

143
CHAPTER 5. STATIC ANALYSIS

In reference 76, the following four load cases were analysed:

1. Uniformly distributed dead load plus snow load over the whole roof.

2. Uniformly distributed dead load on the whole roof plus snow load on half the
roof (x > 0).

3. Uniformly distributed dead load on the whole roof plus snow load on half the
roof (y > 0).

4. Uniformly distributed dead load plus wind load over the whole roof.

Only the first load case will be considered in this section. The vertical loads on
the cable net (measured per unit horizontal area in the initial state) are: dead load
(weight of cables and roof cladding)= −0.6 kN/m2 and snow load= −0.75 kN/m2 .
In the present analysis, the whole structure (375 degrees of freedom) was modelled
using the same program and finite elements as for the analysis of the Scandinavium
Arena.

5.2.2 Different shapes of the ring beam

In Figure 5.22 the ring beam is drawn as a circle. However, in [76] the boundary arcs
26–1–4, 5–8–11, 12–15–18 and 19–22–25 were replaced by parabolas with the same
rise as the corresponding circular arcs. This was done in order to make the projected
boundary curve conincide with the line of compression for the projected cable forces
in the initial state. As is seen in Figure 5.22, along these arcs the ring beam is in the
plane loaded only in one direction (either the x- or the y-direction). To investigate
the effects of a change of the shape of the ring beam arcs, the structure studied by
Møllmann will be analysed for three different shapes of the arcs: circular, parabolic,
and cosine shape. These shapes are described by the following equations for the arc
26–1–4 (−3R/5 ≤ x ≤ 3R/5):

ycircle = (R2 − x2 )1/2 , (5.14)


5x2
yparabola = − + R, (5.15)
9R   
5x −1 4
ycosine = R cos cos . (5.16)
3R 5

The y-coordinates for the different shapes are shown in Figure 5.24.

144
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

1
2

y (m)
Node x (m) z (m)
Circle Parabola Cosine
2 10.400 50.949 50.844 50.808 −4.840
3 20.800 47.659 47.378 47.288 −3.160

Figure 5.24: Difference between some ring beam shapes: circle and parabola—
visually, circle, parabola and cosine—quantitatively.

5.2.3 Results and discussion

Some of the results from the calculations are given in Table 5.1, but the most
interesting results are shown in Figures 5.26–5.38 on pages 147–153.

Table 5.1: Midpoint and ring beam displacements under full snow load.
Shape of ring beam arcs
Description Ref. 76
Circle Parabola Cosine
Midpoint displacement (m) −1.174 −1.176 −1.177 −1.171
Ring beam displacement in 0.189 0.190 0.190 0.188
y-direction at (x = 0, y = R) (m)
Ring beam displacement in −0.222 −0.222 −0.222 −0.221
x-direction at (x = R, y = 0) (m)

It is shown in Table 5.1, Figures 5.26–5.29 that the shape of the ring beam has very
small effects on the displacements, the cable forces and the axial force in the ring
beam. In addition, for these quantities the current results agree very well with the
results given in reference 76. However, regarding the bending moment, large changes
occur when the shape is varied, see Figures 5.30–5.35. Also the twisting moment

145
CHAPTER 5. STATIC ANALYSIS

changes, but not as much as the bending moment. Nevertheless, the twisting mo-
ment is not so important in this case since its maximum value is less than 10 % of
the maximum value of the bending moment. The reason for the large difference in
the bending moment distribution will be discussed below.
Structures with circular or parabolic shape are optimised for a certain load case (cf. a
bicycle wheel or a stone arch). If the load case changes, the structure may undergo
large changes in displacements and force distribution. Anyone who has broken a
spoke in a well-built bicycle wheel will agree that the structure adjust itself to the
new load case and for some structures the adjustment might be large. The reverse
must also hold: change the shape of the structure but keep the load distribution
and possibly large changes in some quantities will follow. For the present example,
consider the two statically equivalent systems in Figure 5.25. The eccentricity e =
M/N can be considered as a measure of the importance of the bending moment
compared to the normal force. Remember, in the present example, the axial force
N changed very little when the shape of the ring beam was varied. This means that
if the shape of the ring beam is changed, the eccentricities of the axial forces are
changed.

N1

M1 N2

(a) M2

e1
N1

N2 e2
(b)

Figure 5.25: A segment of the ring beam. The systems (a) and (b) are statically
equivalent if e1 = M1 /N1 and e2 = M2 /N2 .

For the parabolic arcs, the eccentricities in the initial state for nodes 2 and 3 are
e2 = 227/12431 = 0.018 m and e3 = 189/13700 = 0.014 m. The coordinate changes
for these nodes are ∆y2 = 0.105 m and ∆y3 = 0.281 m. Hence, if the coordinate
changes are larger than the eccentricities one can expect significant changes of the
bending moment diagram, Figures 5.30 and 5.31.
If the axial force is much larger than the bending moment, i.e. a small eccentricity e,

146
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

the moment is considered as ‘small’. This means that the whole cross-section of the
ring beam will be subjected to compressive stresses only. It is possible to calculate
the eccentricity e0 that, for a rectangular section of height h and width b, gives zero
stress at the outermost fiber:
N 6M N 6N e0
σ=− + 2 =− + =0 (5.17)
bh bh bh bh2
Equation 5.17 yields e0 = h/6. If the eccentricity e is smaller than e0 no tensile
stresses will occur. However, the compressive stresses on the other side of the cross-
section can be very high and may crush the concrete. It is therefore desirable to
have a small eccentricity in the whole ring beam.
It is concluded that the bending moment distribution is strongly dependent on the
shape of the ring beam. The shape of the ring beam in the present analysis of the
Scandinavium Arena and in the previous analysis was different. Thus, the bending
moment distributions from the two studies cannot be compared. Also for the exam-
ple by Møllmann, the present analysis gave bending moments values that differed
quite much from those given in [76], see Figures 5.31 and 5.34. As no coordinate
values for the ring beam were given in reference [76], a detailed comparison of the
bending moments is not possible.

3500

3343 3384 3357 3339 3357 3384 3343


(3346) (3382) (3346) (3331) (3346) (3382)
3171 (3346) 3171
3000 (3190) [3347] [3382] [3342] [3327] [3342] [3382] [3347] (3190)
[3195] {3370} [3195]
{3370}

2500
Cable force (kN)

2000

1500

1000

500

0

50 −
40 −
30 −
20 −
10 0 10 20 30 40 50
y (m)

Figure 5.26: Forces in the hanging cables under full snow load. The following nota-
tion is used: ·=Circle, (·)=Parabola, [·]=Cosine, {·}=Møllmann [76].

147
CHAPTER 5. STATIC ANALYSIS

3500

3329 3380 3347 3347 3380


3322 3329
(3351) (3354) (3342) (3342) (3354)
(3336) (3351)
3000 3108 [3344] 3108
[3356] [3340] [3340] [3340] [3344] [3356]
(3119) (3119)
[3123] [3123]
{3340} {3340}
2500
Cable force (kN)

2000

1500

1000

500

0

50 −
40 −
30 −
20 −
10 0 10 20 30 40 50
x (m)

Figure 5.27: Forces in the bracing cables under full snow load. The following nota-
tion is used: ·=Circle, (·)=Parabola, [·]=Cosine, {·}=Møllmann [76].

−2000

−4000
Axial force (kN)

−6000
{−11770}
{−13860}
−8000 −11740 −12429 −13685 −13271 −13674 −12421
(−11746) (−12431) (−13700) (−13270) (−13696) (−12427)
−10000 −11729
[−11747] [−12431] [−13701] [−13269] [−13700]
[−12428] (−11741)
−12000 [−11743]

−14000
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.28: Axial force in beam elements in the initial state. The following notation
is used: ·=Circle, (·)=Parabola, [·]=Cosine, {·}=Møllmann [76].

148
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

0
−2000
−4000
−6000
Axial force (kN)

−8000 {−14930}
{−17590}
−10000
−15027 −15842
−15921 −17567 −17176 −17500
−12000 (−15033)
(−15918) (−17587) (−17170) (−17505) (−15844) −14939
[−15032]
−14000 [−15916] [−17586] [−17164] [−17500] [−15840] (−14949)
[−14947]
−16000
−18000
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.29: Axial force in beam elements under full snow load. The following no-
tation is used: ·=Circle, (·)=Parabola, [·]=Cosine, {·}=Møllmann [76].

2500 2288
Bending moment (kNm)

2000
1913
1500
1000
500
139
0
−500
−1000 −391
−1074
−1225
−1500 1109
−2000
60
50 −1697
60
40 50
30 40
30
y (m) 20
20
10 10 x (m)
0 0

Figure 5.30: Bending moment around local z-axis for beam elements in the initial
state. Circular shape of ring segments.

149
CHAPTER 5. STATIC ANALYSIS

300
{770}
Bending moment (kNm)

245 227 189


200
109
100

0
−49
−100

−200
−181
−300
−283
−400 {−1150}
60 −310
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.31: Bending moment around local z-axis for beam elements in the initial
state. Parabolic shape of ring segments. {·}=Møllmann [76].

800
Bending moment (kNm)

600 693 497


400 340
200 240

0
137
−200
−400
−472 −270
−600
−800
−1000 −843
60
50
60
40 50
30 40
y (m) 20 30
20 x (m)
10 10
0 0

Figure 5.32: Bending moment around local z-axis for beam elements in the initial
state. Cosine shape of ring segments.

150
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

6000 5038
Bending moment (kNm)

4000
2468
2000
648
980
0 −114

−2000 −1323

−4000

−6000 −4468

−8000
60 −6547
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.33: Bending moment around local z-axis for beam elements under snow
load. Circular shape of ring segments.

3000
{3030} 2258
2482 2020
Bending moment (kNm)

2000 2555

1000
219
0

−1000

−2000

−3000 −2648
−4000

−5000 −4107
60 {−5570}
50 −4543
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.34: Bending moment around local z-axis for beam elements under snow
load. Parabolic shape of ring segments. {·}=Møllmann [76].

151
CHAPTER 5. STATIC ANALYSIS

4000
Bending moment (kNm)

3000 2504
3109 2499
2000 1434
1000 706
0
−1000
−2000
−3000
−4000 −3483

−5000 −4064
60 −3928
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.35: Bending moment around local z-axis for beam elements under snow
load. Cosine shape of ring segments.

200 169
Twisting moment (kNm)

100

21 10
0

−30
−100
−133
−200
−153
−300
60 −285
50
60
40 50
30 40
20 30
y (m) 10
20
0
10 x (m)
0

Figure 5.36: Twisting moment around local x-axis for beam elements under snow
load. Circular shape of ring segments.

152
5.2. SENSITIVITY OF BENDING MOMENT TO THE SHAPE OF THE RING BEAM

200
Twisting moment (kNm)

109
100
21
0 0
−51
−100

−200 −83

−300
−255
−330
−400
60
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.37: Twisting moment around local x-axis for beam elements under snow
load. Parabolic shape of ring segments.

100 80
Twisting moment (kNm)

34
0 14

−100 −74

−54
−200

−300 −230

−400
60 −390
50
60
40 50
30 40
y (m) 20 30
20
10 10 x (m)
0 0

Figure 5.38: Twisting moment around local x-axis for beam elements under snow
load. Cosine shape of ring segments.

153
CHAPTER 5. STATIC ANALYSIS

5.3 Comparison with a simplified method

In reference 116, a simplified method to analyse a prestressed cable net anchored


in an elliptic contour beam is presented. The simplified method is well suited for
preliminary design, where the dimensions of the structure and its structural members
are to be decided. However, in further design work a more accurate finite element
method is often needed. In this section the reliability and limitations of the simplified
method will be investigated.
In the simplified method, a number of assumptions concerning the behaviour of the
structure and load cases are imposed to obtain closed-form solutions. The most
important assumptions are:

• The cable net is substituted by a continuous shear-free membrane. This can


be considered to be valid for a net with a dense mesh.

• The projection of the ring beam in the horizontal plane is an ellipse.

• The ring beam is assumed to deform like a linear plane beam. Thus, the
space-curved shape of the ring beam is not taken into consideration.

• The ring beam is supported by a continuous wall which permits the ring to
move freely in the horizontal plane.

• In the initial state, the roof is uncladded. This is in contrast to the initial
states of the Scandinavium Arena and Møllmann’s structure.

The complete list, which includes 16 assumptions is given on the pages 34–36 in
reference 116.
First, the example by Møllmann will be investigated. The geometry of the roof will
differ from that shown in Figure 5.22 due to the last assumption given above. For
 0 in both directions and Fz = 0, equation (5.9) can
the same prestressing intensity H
be written as:
2 2
0 ∂ z + H
H  0 ∂ z = 0. (5.18)
∂x2 ∂y 2
Solving (5.18) yields fx = fy as expected. The heights fx and fy will be different also
for the Scandinavium Arena as the self-weight is zero in the initial configuration.

5.3.1 Results and discussion

The results for Møllmann’s example and the Scandinavium Arena are presented in
Tables 5.2 and 5.3. Møllmann’s example were calculated with both circular and
parabolic shapes of the contour arcs.

154
5.3. COMPARISON WITH A SIMPLIFIED METHOD

Table 5.2: Comparison between the simplified method and the finite element method
for Møllmann’s example. Load case: uniformly distributed load equal to
−0.75 kN/m2 on the whole roof (N.B. zero load in the initial state).
Method
Description Finite element
Simplified
Circular Parabolic
fx (m) 7.002 7.002 7.000
fy (m) 6.998 6.998 7.000
Midpoint displacement (m) −1.689 −1.692 −1.634
Force in the mid cable 3009 2996 3196
in the x-direction (kN)
Force in the mid cable 2974 2991 3107
in the y-direction (kN)
Axial force at (x = 0, y = R) (kN) −13538 −13549 −15430
Axial force at (x = R, y = 0) (kN) −13464 −13471 −15001
Bending moment at (x = 0, y = R) (kN) 2877 4238 5577
Bending moment at (x = R, y = 0) (kN) −7154 −5237 −5577

For Møllmann’s example, the midpoint displacement agree very well for both the cir-
cular and parabolic shapes of the contour arcs. The same holds for the cable forces.
For the axial forces and bending moments, however, the differences are larger. It
should be mentioned that axial forces given for the finite element calculation are not
the maximum axial forces, cf. Figure 5.29. The maximum axial force for the circular
and parabolic shapes are −15785 kN and −15816 kN respectively. Nevertheless, the
maximum positive and negative moments occur at the bottom and the top of the
ring beam. The bending moments for the ring beam with a parabolic shape agree
best with the simplified method. The difference in the positive bending moment
may be a result of the assumption of a plane ring beam in the simplified method.
The difference in axial force in the ring beam follows from the differences in the
bending moments since the structure must be in equilibrium.

155
CHAPTER 5. STATIC ANALYSIS

Table 5.3: Comparison between the simplified method and the finite element method
for the Scandinavium Arena (without the pylon). Load case: uniformly
distributed load equal to −0.75 kN/m2 on the whole roof (N.B. zero load
in the initial state).
Method
Description
Finite element Simplified
fx (m) 6.611 6.703
fy (m) 6.795 6.703
Midpoint displacement (m) −1.999 −1.821
Force in the mid cable 891 932
in the x-direction (kN)
Force in the mid cable 720 707
in the y-direction (kN)
Axial force at (x = 0, y = R) (kN) −10423 −12197
Axial force at (x = R, y = 0) (kN) −8797 −9250
Bending moment at (x = 0, y = R) (kN) 24985 39729
Bending moment at (x = R, y = 0) (kN) −40350 −39729

All comments on Møllmann’s example also hold for the Scandinavium Arena: mid-
point deflection and cable forces show good agreement but the differences in axial
forces and bending moments are larger.
It should be pointed out that the simplified method assumes an elliptic contour
beam. As the cable net is assumed to behave like a continuous membrane, the
forces at every point on the ring beam act in both the x- and y-direction. In practice,
some sections of the ring beam are subjected to forces in only one direction. In such
cases these sections should have a parabolic shape otherwise the bending moment
distribution will not be smooth as assumed in the simplified method. Nevertheless,
it can be concluded that the simplified method is well suited for preliminary design
works. Since the bending moment is very sensitive to the shape of the ring beam,
analysis with a finite element program must be used in detailed design work.

156
Chapter 6

Conclusions and further research

6.1 Conclusions

The conclusions from the study are divided into three subsections, each having the
same name as the chapter from which the respective conclusions are drawn.

6.1.1 The initial equilibrium problem

The conclusions from Chapter 3 are:

• In the form-finding of cable structures, the optimal strategy seems to be a


combined approach. A simple method is first used to obtain a starting shape
that is close to the final shape. Then, a more accurate, iterative method is
used to obtain the final shape.

• The linear version of the force density method is very simple to use, but it is
not simple to specify force densities that give the desired force distribution or
shape. Therefore, the linear force density method is used to obtain a starting
shape for the non-linear force density method.

• The linear force density method is not suitable for a cable net with an orthog-
onal projection in the horizontal plane. A better method for this case is the
grid method.

• For a structure composed of only cables, the force density stiffness matrix D is
positive definite and, thus, the solution is unique. However, this matrix can be
singular for structures composed of both tension and compression members.
Therefore, the force density method, as presented in this thesis, is not suitable
for such structures.

• No suitable method to find the initial equilibrium configuration of pure tenseg-


rity structures has been found. For a tensegrity structure with known geome-
try, the independent states of prestress and the number of inextensional mech-

157
CHAPTER 6. CONCLUSIONS AND FURTHER RESEARCH

anisms are given by the dimensions of the subspaces of the equilibrium matrix
A.

6.1.2 Finite cable elements

The conclusions from Chapter 4 are:

• The elastic and associate catenary elements are mathematically exact under
the assumption that the cable is perfectly flexible. Only one element per
cable is necessary to model the static behaviour of both slack and taut cables
subjected to uniformly distributed loads.

• In the case when the self-weight of the cable approaches zero the tangent
stiffness matrix of the elastic cable element approaches that of the straight
bar element. Hence, both light and heavy cables can be modelled using the
elastic catenary element.

• The parabolic element is not mathematically exact but yields extremely accu-
rate results for cables with small sag-to-span ratios.

• The tangent stiffness matrices of the analytical cable elements are functions of
the horizontal force and have to be obtained by iteration. Good starting values
and a modified Newton-Raphson method ensure that the correct horizontal
force is found with only 3–4 iterations.

• Comparisons between the finite cable elements show, as expected, similar re-
sults for taut elements. For slack elements, however, the discrepancies are
larger and are not negligible.

• A real cable has a bending stiffness, but this can be neglected for very taut
cables. However, for span-to-length ratios less than 0.8 the bending stiffness
cannot be disregarded.

6.1.3 Static analysis

The conclusions from Chapter 5 are:

• In the comparison between the new and old results for the Scandinavium
Arena, the maximum vertical net displacement, in-plane ring beam displace-
ments, cable forces and twisting moments show good agreement. The differ-
ences can be explained by differences in cable spacing, initial cable forces and
modulus of elasticity for the concrete. However, the large discrepancies in the
bending moment distribution cannot be explained in this way.

• The bending moment distribution is found to be very sensitive to the shape of


the elastic ring beam. A circular ring beam yields an irregular bending moment

158
6.2. FURTHER RESEARCH

distribution in both the initial and loaded states. A ring beam consisting of
parabolic in-plane arcs is shown to provide a smoother moment distribution
with lower values. This can be explained in that the parabolic shape is closer
to the line of compression for the projected cable forces in the initial state.

• The twisting moments in the ring are much lower than the bending moments
and do not seem to be a problem in design.

• The pylons stiffen the ring beam significantly; much larger bending moments
and net displacements are obtained if the pylons are not present. The pylons
also considerably reduce the axial compressive force in the lower parts of the
ring beam. With a low compressive force in the concrete ring beam, the
importance of the bending moment increases as large tension stresses may
arise.

• Most of the results from a simplified method, which assumes that the cable net
behaves like a shear-free membrane, agree well with the results from the finite
element calculations. However, due to the many assumptions, the simplified
method can only be recommended for preliminary calculations. Since the
bending moments of the accurate and simplified methods differ more than
the other quantities, the most important of the assumptions in the simplified
method seems to be that of a plane ring beam.

6.2 Further research

Static analyses of elastic cable nets and cable trusses are ubiquitous in literature
which are shown by the published monographs, e.g. [16, 48, 57, 61, 113], related to
this topic. The author, therefore, does not consider it necessary to do more work
on elastic static analyses of these two types. Also their dynamic behaviour in the
elastic range is extensively described in several of these references.

6.2.1 Failure analysis—background

The current trend in structural design is to optimise the load-bearing to weight ratio
of structures, and cable structures are very much involved in this trend. Structural
optimisation leads to non-linear phenomena and parameter sensitivity. For safety
reasons, it is very important to know the behaviour of a structure under large loads.
Generally, a structure has different modes of failure: elastic or plastic instability, and
material failure at ultimate strength. The first failure type—the elastic instability
phenomenon—has been extensively analysed for bar, beam and shell structures at
the Department of Structural Engineering at the Royal Institute of Technology,
Stockholm. An analysis tool has been developed for these structures, and this tool
can also be used to analyse the stability of cable structures. In the following sections
the other aspects related to failure analysis of cable structures are reviewed. In the
last section, a structural concept is discussed.

159
CHAPTER 6. CONCLUSIONS AND FURTHER RESEARCH

Plasticity

The high strength steel used in modern cables exhibits linear stress-strain character-
istics over only a portion of its usable strength. Therefore, in ultimate load analyses
of general cable structures, the resulting formulations must consider material non-
linearity. Under large external loads, some cables will go into the plastic range while
some will stay in the elastic zone and other will lose their pretension. Hence, the
cable elements must handle all three cases. In their present form, the elements in
this thesis are only valid for the elastic and the slack cases and therefore have to
be extended to consider plasticity. In reference 1 the general cable element was
extended to include material non-linearity. Since the general cable element forms
the basis in the derivation of the analytical cable elements, it is anticipated that it
is possible to include material non-linearity also in the analytical elements.
Some works have been concerned with the plastic analysis of cable structures. For
example, Ma et al. [73] used four-node isoparametric cable elements in the plastic
dynamic analysis of a saddle-shaped cable net. More recently, Atai and Miodu-
chowski [6] derived conditions for the stability of cable structures in the plastic
state and addressed important issues, such as load history, path-dependency of so-
lution and unloading of a cable from a plastic to a slack state. From these works
some important conclusions have been drawn and they ought to provide a good
starting point for further improvement of the plastic analysis of cable structures.

Parameter and imperfection sensitivity

The load-bearing capacity of general cable structures depends on several parameters,


of which the level of prestressing is the most important. For ultimate load analysis it
is necessary to determine the sensitivity of the structure to changes in some of these
parameters. Also different types of imperfections, e.g. non-straight compression
elements or misplaced cables, may have a serious impact on the maximum allowable
loads on the structure. Two works that can be referred to in this context are
presented below.
Lewis et al. [67] analysed the cladding stiffening effects in prestressed cable roofs.
A parametric study with different cladding-to-net stiffness ratios showed that for
ratios found in practice the cladding significantly contributes to the net stiffness.
As a result of the cladding, the prestressing forces could be lowered. However, in
this case the composite action between the cladding and cable net must be assured
throughout the lifetime of the structure. The connection between the cladding and
net must be able to transmit large shear forces, which may give rise to higher cost
and thereby decreases the benefits of the cladding-net interaction.
In general, space trusses are regarded as highly redundant structures with the abil-
ity to survive the loss of several members without losing overall stability. These
structures also have the property to be very sensitive to imperfections. Wada and
Wang [123] have investigated the effect that different types of imperfections have on
the load-bearing capacity of a double layer space truss. Their investigation included

160
6.2. FURTHER RESEARCH

variation of member strength, initial imperfection of member length, and errors in


the assembly process. The conclusions were that the fabrication errors have minor
influence on the capacity of the truss, but the human errors, like assembly errors,
have enormous influence on the mechanical behaviour of the structure. For the space
truss analysed, if two or more members out of 288 members had errors the structure
had a large probability of collapsing. A reduction of the load-bearing capacity may
also occur in cable structures if cable connections are not assembled in their right
positions.

Analysis of tensegrity structures

During the last decade new structural principles have been developed, the struc-
tural behaviour of which are not yet fully understood. The most interesting of the
new structural principles is that of self-stressed systems, also known as tensegrity
systems. Self-stressing is interesting as it allows cable-strut structures to be built
without the need for supporting structures to equilibrate the stresses in the initial
configuration. In addition to this advantage, tensegrity structures have other bene-
fits that make them interesting for research; for example, they are lightweight and
earthquake resistant. There is also a strong desire from architects to implement the
tensegrity principle in buildings because of the pure shapes it produces. Although
the research activity in this area is high, see for example references 46, 80, 81, 126,
there are still several questions that need to be answered before full scale applica-
tion can be a reality. In this section, analysis aspects of tensegrity structures will
be discussed.
Morphology studies. From the invention of the tensegrity concept in the late 1940’s
until the beginning of the 1990’s most research projects have dealt with the geo-
metrical shapes of tensegrity networks. According to Hanaor [46] “It appears that
the morphological study of tensegrity networks has reached a degree of saturation,
whereby the range of conceivable patterns exceeds by far the likely range of appli-
cations.” Nevertheless, double-layer tensegrity grids have been developed by joining
tensegrity simplexes [79]. Wang [126] concludes that “the future work will be con-
centrated on applying other simplexes in space structures.”
Initial equilibrium configurations. Several different shapes of tensegrity structures
are available from the morphology studies. But, form-finding with geometrical meth-
ods does not guarantee mechanical equilibrium and solutions must be checked with
a numerical method [81]. Motro et al. [82] applied both the dynamic relaxation
method and the force density method to solve the initial equilibrium problem. They
anticipated that the latter method is more suitable for large systems, but that more
work has to be done to check its efficiency. One possible way to modify the force den-
sity method to apply to tensegrity systems might be to adopt Mollaert’s approach
given in reference 75. In that approach, the compression and tension members
are separated to ensure a solution out of the plane. Recently, Bruno [15] used a
method based on the minimisation of the potential energy to find the equilibrium
configuration of simple two-dimensional tensegrity systems. A more mathematical

161
CHAPTER 6. CONCLUSIONS AND FURTHER RESEARCH

approach was developed by Roth and Whiteley [101] and extended by Connelly and
Whiteley [25]. This approach has been used by Burkhardt [17] to analyse quite
complicated tensegrity domes.
Mechanism elimination. Finding a configuration which satisfies equilibrium is not
enough to solve the initial equilibrium problem for a tensegrity system; the internal
mechanisms for that configuration must be identified, classified and, if possible, elim-
inated by prestressing [80]. The method by Calladine and Pellegrino [20, 93] can be
used to find the mechanisms and determine the stability of the initial configuration.
Recently, Tomka [120] introduced a technique called the method of stabilising force
to analyse the stability of cable structures. The advantage of this method is that
in addition to the qualitative result (stable or unstable), obtained by the method
by Calladine and Pellegrino, quantitative conclusions can also be drawn concerning
the measure of stability.
After an acceptable solution has been found, the behaviour of the tensegrity system
has to be studied under the effect of external loads. In particular, the stability of
the self-stressing configurations should be studied [80]. It is of cardinal importance
to know if mechanisms can reappear as a result of external loadings. According to
Motro [80] this subject “still remains a fairly open matter.”
Construction. Besides the theoretical aspects mentioned above, the analysed struc-
tures must be possible to build. In the present state, there has not been much
application of the tensegrity principle in the construction field. The reason for this
is that several fundamental technical problems still need to be solved [46, 80]. The
main problems are:

• suitable prestressing procedures,

• efficient node systems, and

• incorporation of cladding.

Finding a suitable construction and prestressing procedure is quite difficult. These


procedures tend to be cumbersome and uneconomic because general tensegrity sys-
tems are geometrically complex and lack rigidity prior to prestressing [80]. The
prestressing methods must be reliable and assure the level and permanence of the
tension that has been put in. The efficiency of the node systems is very much related
to the construction procedure and the prime objective is to have compact connec-
tions. Concerning the roof covering, a flexible membrane is preferable because of the
flexibility of the tensegrity frameworks. It is important that the membrane forms
an integral part of the design, as it is not a trivial matter to obtain a correct stress
distribution in the membrane [46]. According to Hanaor [46] “none of the studies
carried out to date, consider the surface membrane.”

162
6.2. FURTHER RESEARCH

6.2.2 Failure analysis—further research

Because the tensegrity structures are made stiff by prestressing, the effects from loss
of the prestress are more severe in these structures than in other cable structures.
Cable structures relying on supporting structures or foundations to equilibrate the
unbalanced loads have generally a higher safety to failure (if the supporting struc-
tures themselves are stable). Tensegrity structures are in many aspects very inter-
esting, but also very complex. From the review in the previous section, the following
research directions are suggested:

• Analysis of the elastic stability of tensegrity structures under external loads.


Most tensegrity structures are kinematically indeterminate, but in many cases
it is possible to stabilise the mechanisms to the first order by prestressing.
However, it is of great importance to know if any of the mechanisms can
reappear under external loads due to the loss of the prestress. First, the
stability of an ideal configuration composed of only cables and struts should
be analysed. Then, it would be of interest to study what additional effects the
membrane cover has upon the overall stability.

• The effects of imperfections on the elastic stability of tensegrity structures.


The stability of the ideal configuration represents the theoretical upper bound
of loading. Real structures have imperfections of different kinds and in many
cases, these greatly affect the stability of the structures. A sensitivity anal-
ysis identifies those parameters that significantly affect the behaviour of the
structure. More or less automatic procedures for these analyses are highly
desirable.

• Analysis of the plastic stability of tensegrity structures under external loads.


A further step would be to include also material non-linearities in the analysis.
These non-linearities introduce several new problems such as load history and
path-dependency of the solution. To simplify the analyses certain assumptions
may be introduced to avoid some of these problems, for example that only
linearly increasing loads are considered.

163
Bibliography

[1] Ahmadi-Kashani, K. Development of cable elements and their applications


in the analysis of cable structures. PhD thesis, University of Manchester In-
stitute of Science and Technology (UMIST), 1983.

[2] Ahmadi-Kashani, K. Representation of cables in space subjected to uni-


formly distributed loads. International Journal of Space Structures Vol. 3, No.
4 (1988), pp. 221–230.

[3] Ahmadi-Kashani, K., and Bell, A. J. The analysis of cables subject


to uniformly distributed loads. Engineering Structures Vol. 10, No. 3 (July
1988), pp. 174–184.

[4] Argyris, J. H., Angelopoulos, T., and Bichat, B. A general method


for the shape finding of lightweight tension structures. Computer Methods in
Applied Mechanics and Engineering Vol. 3 (1974), pp. 135–149.

[5] Argyris, J. H., and Scharpf, D. W. Large deflection analysis of pre-


stressed networks. Journal of the Structural Division, ASCE Vol. 98, No. ST3
(March 1972), pp. 633–654.

[6] Atai, A. A., and Mioduchowski, A. Equilibrium analysis of elasto-plastic


cable nets. Computers & Structures Vol. 66, No. 2–3 (1998), pp. 163–171.

[7] Bang, B., Nielsen, A., Sundsbø, P. A., and Wiik, T. Computer simu-
lation of wind speed, wind pressure and snow accumulation around buildings
(SNOW-SIM). Energy and Buildings Vol. 21 (1994), pp. 235–243.

[8] Barnes, M. R. Form-finding and analysis of prestressed nets and membranes.


Computers & Structures Vol. 30, No. 3 (1988), pp. 685–695.

[9] Bathe, K. J. Finite Element Procedures. Prentice-Hall, New Jersey, 1996.

[10] Berger, H. Light structures—structures of light: the art and engineering of


tensile architecture. Birkhäuser, Basel, 1996.

[11] Bienkiewicz, B. New tools in wind engineering. Journal of Wind Engineer-


ing and Industrial Aerodynamics Vol. 65 (1996), pp. 279–300.

[12] Bienkiewicz, B., Tamura, Y., Ham, H. J., Ueda, H., and Hibi, K.
Proper orthogonal decomposition and reconstruction of multi-channel roof

165
BIBLIOGRAPHY

pressure. Journal of Wind Engineering and Industrial Aerodynamics Vol.


54/55 (1995), pp. 369–381.

[13] Bletzinger, K.-U. Form finding of tensile structures by the updated refer-
ence strategy. In IASS International Colloquium on “Structural Morphology—
Towards the New Millenium” (Nottingham, U.K., 1997), J. Chilton, B. Choo,
W. Lewis, and O. Popovic, Eds., pp. 68–75.

[14] Broughton, P., and Ndumbaro, P. The analysis of cable and catenary
structures. Tomas Telford, London, 1994.

[15] Bruno, E. A. Design tools for tensegrity structures. Bachelor’s thesis, The
Pennsylvania State University, 1997. Online. Internet. 19 January 1999. Avail-
able HTTP: www-scf.usc.edu/˜ebruno/tensegrity/thesis2.html.

[16] Buchholdt, H. A. An introduction to cable roof structures. Cambridge


University Press, Cambridge, 1985.

[17] Burkhardt, R. A technology for designing tensegrity domes and spheres.


Online. Internet. 19 January 1999. Available HTTP:
www.channel1.com/users/bobwb/prospect/prospect.html.

[18] Calladine, C. R. Buckminster Fuller’s “tensegrity” structures and Clerk


Maxwell’s rules for the construction of stiff frames. International Journal of
Solids and Structures Vol. 14, No. 2 (1978), pp. 161–172.

[19] Calladine, C. R. Modal stiffnesses of a pretensioned cable net.


International Journal of Solids and Structures Vol. 18, No. 10 (1982), pp.
829–846.

[20] Calladine, C. R., and Pellegrino, S. First-order infinitesimal


mechanisms. International Journal of Solids and Structures Vol. 27, No. 4
(1991), pp. 505–515.

[21] Calladine, C. R., and Pellegrino, S. Further remarks on first-order


infinitesimal mechanisms. International Journal of Solids and Structures Vol.
29, No. 17 (1992), pp. 2119–2122.

[22] Cardou, A., and Jolicoeur, C. Mechanical models of helical strands.


Applied Mechanics Reviews, ASME Vol. 50, No. 1 (January 1997), pp. 1–14.

[23] Chaplin, F., Calderbank, G., and Howes, J. The technology of


suspended cable net structures. Construction Press, London, 1984.

[24] Christou, P. M. An integrated technique for the analysis of frame and


cable structures. PhD thesis, University of Florida, 1996.

[25] Connelly, R., and Whiteley, W. Second-order rigidity and prestress


stability for tensegrity frameworks. SIAM Journal of Discrete Mathematics
Vol. 9, No. 3 (1996), pp. 453–491.

166
BIBLIOGRAPHY

[26] Cook, N. J. The designer’s guide to wind loading of building structures,


Part 1: background, damage survey, wind data and structural classification.
Butterworths, London, 1985, ch. 7.

[27] Cook, N. J. The designer’s guide to wind loading of building structures,


Part 2: static structures. Butterworths, London, 1990, ch. 12.

[28] Cook, R. D., Malkus, D. S., and Pleshna, M. E. Concepts and


applications of finite element analysis, third ed. John Wiley & Sons,
Chichester, 1989.

[29] Costello, G. A. Theory of wire rope, second ed. Springer-Verlag, New


York, 1997.

[30] Davenport, A. G. How can we simplify and generalize wind loads?


Journal of Wind Engineering and Industrial Aerodynamics Vol. 54/55
(1995), pp. 657–669.

[31] Drew, P. Frei Otto: form and structure. Verlag Gerd Hatje, Stuttgart,
1976.

[32] Dyrbye, C., and Hansen, S. O. Wind loads on structures. John Wiley &
Sons, Chichester, 1997.

[33] El Ashkar, I., and Novak, M. Wind tunnel studies of cable roofs.
Journal of Wind Engineering and Industrial Aerodynamics Vol. 13 (1983),
pp. 407–419.

[34] Ernst, H.-J. Der E-modul von Seilen unter Berücksichtigung des
Durchhanges. Der Bauingenieur Vol. 40, No. 2 (1965), pp. 52–55. (In
German).

[35] European Committee for Standardisation. Part 2: steel bridges,


Annex A: high strength cables. pr ENV 1993-2 Eurocode 3: design of steel
structures (February 1997), pp. 74–98.

[36] Fujikake, M., Kojima, O., and Fukushima, S. Analysis of fabric


tension structures. Computers & Structures Vol. 32, No. 3/4 (1989), pp.
537–547.

[37] Gambhir, M. L., and Batchelor, B. D. A finite element for 3D


prestressed cablenets. International Journal for Numerical Methods in
Engineering Vol. 111 (1977), pp. 1699–1718.

[38] Gamble, S. L., Kochanski, W. W., and Irwin, P. A. Finite area


element snow loading prediction—Applications and advancements. Journal
of Wind Engineering and Industrial Aerodynamics Vol. 41-44 (1992), pp.
1537–1548.

167
BIBLIOGRAPHY

[39] Geiger, D., Stefaniuk, A., and Chen, D. The design and construction
of two cable domes for the Korean Olympics. In Shells, Membranes and
Space Frames, Proceedings IASS Symposium, Vol. 2 (Osaka, Japan, 1986),
pp. 265–272.

[40] Geiger, D. H. Membrane structures. In Encyclopedia of Architecture:


design, engineering and construction, J. A. Wilkes, Ed., vol. 3. John Wiley &
Sons, 1989, pp. 398–437.

[41] Ghiocel, D., and Lungu, D. Wind, snow and temperature effects on
structures based on probability. Abacus Press, Tunbridge Wells, 1975, ch. 14.

[42] Gimsing, N. J. Cable supported bridges: concept and design. John Wiley &
Sons, Chichester, 1997.

[43] Gründig, L., and Bahndorf, J. The design of wide-span roof structures
using micro-computers. Computers & Structures Vol. 30, No. 3 (1988), pp.
495–501.

[44] Haber, R. B., and Abel, J. F. Initial equilibrium solution methods for
cable reinforced membranes: Part I—formulations. Computer Methods in
Applied Mechanics and Engineering Vol. 30 (1982), pp. 263–284.

[45] Hanaor, A. Aspects of design of double-layer tensegrity domes.


International Journal of Space Structures Vol. 7, No. 2 (1992), pp. 101–113.

[46] Hanaor, A. Developments in tensegrity systems: an overview. In Fourth


International Conference on Space Structures (Surrey, U.K., September
1993), pp. 987–997.

[47] Hobbs, R. E., and Raoof, M. Hysteresis in bridge strand. Proceedings of


the Institution of Civil Engineers, Part II Vol. 77 (December 1984), pp.
445–464.

[48] Irvine, H. M. Cable structures. Dover Publications, New York, 1992.

[49] Irwin, H. P. A. H., and Wardlaw, R. L. A wind tunnel investigation of


a retractable fabric roof for the Montreal Olympic Stadium. In Wind
Engineering, Proceedings of the Fifth International Conference, Fort Collins,
Colorado, July 1979 (New York, 1980), J. E. Cermak, Ed., Pergamon Press,
pp. 925–938.

[50] Iversen, J. D. Drifting snow similitude – Drift deposit rate correlation. In


Wind Engineering, Proceedings of the Fifth International Conference, Fort
Collins, Colorado, July 1979 (New York, 1980), J. E. Cermak, Ed.,
Pergamon Press, pp. 1035–1047.

[51] Jawerth, D. Förspänd hängkonstruktion med mot varandra spända linor.


Byggmästaren Vol. 38, No. 10 (1959), pp. 223–236. (In Swedish).

168
BIBLIOGRAPHY

[52] Jayaraman, H. B., and Knudson, W. C. A curved element for the


analysis of cable structures. Computers & Structures Vol. 14, No. 3/4 (1981),
pp. 325–333.

[53] Jennings, A. Discussion of “Cable movements under two-dimensional


loading” by W. T. O’Brien and A. J. Francis. Journal of the Structural
Division, ASCE Vol. 91, No. ST1 (February 1965), pp. 307–311.

[54] Karoumi, R. Nonlinear analysis of cable supported bridges. Tech. Rep.


1997:21, Department of Structural Engineering, Royal Institute of
Technology, Stockholm, 1997.

[55] Kärrholm, G., and Karlsson, S. Inomhusarena med dubbelkrökt


hängtak. Byggmästaren Vol. 50, No. 9 (September 1971), pp. 20–24. (In
Swedish).

[56] Kärrholm, G., and Samuelsson, A. Analysis of a prestressed cable-roof


anchored in a space-curved ring beam. In Ninth IABSE Congress
(Amsterdam, May 1972).

[57] Krishna, P. Cable-suspended roofs. McGraw-Hill, New York, 1978.

[58] Kumar, K., and Cochran, Jr., J. E. Closed-form analysis for elastic
deformations of multilayered strands. Journal of Applied Mechanics, ASME
Vol. 54 (December 1987), pp. 898–903.

[59] Lai, C.-Y., You, Z., and Pellegrino, S. Shape of deployable deflectors.
Journal of Aerospace Engineering Vol.11, No. 3 (July 1998), pp. 73–80.

[60] Lanteigne, J. Theoretical estimation of the response of helically armored


cables to tension, torsion and bending. Journal of Applied Mechanics, ASME
Vol. 52 (June 1985), pp. 423–432.

[61] Leonard, J. W. Tension structures. McGraw-Hill, New York, 1988.

[62] Leonhardt, F., and Schlaich, J. Cable-suspended roof for Munich


Olympics. Civil Engineering, ASCE Vol. 42, No. 7 (July 1972), pp. 41–44.

[63] Leonhardt, F., and Schlaich, J. Structural design of roofs over the
sports arenas for the 1972 Olympic Games: some problems of prestressed
cable net structures. The Structural Engineer Vol. 50, No. 3 (March 1972),
pp. 113–119.

[64] Letchford, C. W., Iverson, R. E., and McDonald, J. R. The


application of the quasi-steady theory to full scale measurements on the
Texas Tech building. Journal of Wind Engineering and Industrial
Aerodynamics Vol. 48 (1993), pp. 111–132.

[65] Lewis, W. J. Mathematical formulae as a medium for shaping membrane


forms. In Fourth International Conference on Space Structures (Surrey,
U.K., September 1993), pp. 907–915.

169
BIBLIOGRAPHY

[66] Lewis, W. J., and Gosling, P. D. Stable minimal surfaces in


form-finding of lightweight tension structures. International Journal of Space
Structures Vol. 8, No. 3 (1993), pp. 149–166.

[67] Lewis, W. J., Jones, M. S., Lewis, G., and Rushton, K. R.


Cladding-network interaction in pretensioned cable roofs, studied by dynamic
relaxation. Computers & Structures Vol. 19, No. 5/6 (1984), pp. 885–897.

[68] Lewis, W. J., Jones, M. S., and Rushton, K. R. Dynamic relaxation


analysis of the non-linear response of pretensioned cable roofs. Computers &
Structures Vol. 18, No. 6 (1984), pp. 989–997.

[69] Lewis, W. J., and Shan, J. Numerical modelling of the nonlinear static
response of clad cable net structures. Computers & Structures Vol. 35, No. 1
(1990), pp. 15–22.

[70] Liddell, I. The roof of the Millennium Dome. New Steel Construction Vol.
6, No. 4 (August/September 1998), pp. 30–32.

[71] Linkwitz, K., and Schek, H.-J. Einige Bemerkungen zur Berechnung
von vorgespannten Seilnetzkonstruktionen. Ingenieur-Archiv Vol. 40 (1971),
pp. 145–158. (In German).

[72] Linkwitz, K., Schek, H.-J., and Gründig, L. Die


Gleichgewichtsberechnung von Seilnetzen unter Zusatzbedingungen.
Ingenieur-Archiv Vol. 43 (1974), pp. 183–192. (In German).

[73] Ma, D., Leonard, J. W., and Chu, K. H. Slack-elasto-plastic dynamics


of cable systems. Journal of the Engineering Mechanics Division, ASCE Vol.
105, No. EM2 (April 1979), pp. 207–222.

[74] Michalos, J., and Birnstiel, C. Movements of a cable due to change in


loading. Journal of the Structural Division, ASCE Vol. 86, No. ST12
(December 1960), pp. 23–38.

[75] Mollaert, M. Formfinding of “mixed structures”. In Third International


Conference on Space Structures (Guildford, 1984), Elsevier, pp. 180–185.

[76] Møllmann, H. Analysis of hanging roofs by means of the displacement


method. PhD thesis, Technical University of Denmark, 1974.

[77] Moncrieff, E., and Topping, B. H. V. Computer methods for the


generation of membrane cutting patterns. Computers & Structures Vol. 37,
No. 4 (1990), pp. 441–450.

[78] Mote, S. H., and Chu, K.-H. Cable trusses subjected to earthquakes.
Journal of the Structural Division, ASCE Vol. 104, No. ST4 (April 1978),
pp. 667–680.

[79] Motro, R. Tensegrity systems and geodesic domes. International Journal


of Space Structures Vol. 5, No. 3&4 (1990), pp. 341–351.

170
BIBLIOGRAPHY

[80] Motro, R. Tensegrity systems: the state of the art. International Journal
of Space Structures Vol. 7, No. 2 (1992), pp. 75–83.
[81] Motro, R. Structural morphology of tensegrity systems. International
Journal of Space Structures Vol. 11, No. 1&2 (1996), pp. 233–240.
[82] Motro, R., Belkacem, S., and Vassart, N. Form finding numerical
methods for tensegrity systems. In Spatial, Lattice and tension structures,
Proceedings of the IASS-ASCE International Symposium 1994 (Atlanta,
U.S.A., 1994), J. F. Abel, J. W. Leonard, and C. U. Penalba, Eds., ASCE,
pp. 704–713.
[83] O’Brien, W. T. General solution of suspended cable problems. Journal of
the Structural Division, ASCE Vol. 93, No. ST1 (February 1967), pp. 1–26.
[84] O’Brien, W. T., and Francis, A. J. Cable movements under
two-dimensional loads. Journal of the Structural Division, ASCE Vol. 90,
No. ST3 (June 1964), pp. 89–123.
[85] Olsson, A. Object-oriented finite element algorithms. Licentiate thesis,
Royal Institute of Technology, Stockholm, 1997.
[86] Olsson, N., and Wennerström, H. Jämförelse mellan olika
approximativa beräkningsmetoder vid förspända lintak. Master’s thesis,
Chalmers University of Technology, Gothenburg, 1972. (In Swedish).
[87] Otto, F., and Rasch, B. Finding Form. Edition Axel Menges, Stuttgart,
1995.
[88] Otto, F., and Schleyer, K. Tensile structures, Vol 2: cable structures.
MIT Press, Cambridge, 1969.
[89] Ozdemir, H. A finite element approach for cable problems. International
Journal of Solids and Structures Vol. 15 (1979), pp. 427–437.
[90] Pacoste, C., and Eriksson, A. Beam elements in instability problems.
Computer Methods in Applied Mechanics and Engineering Vol. 144, No. 1–2
(1997), pp. 163–197.
[91] Pellegrino, S. Analysis of prestressed mechanisms. International Journal
of Solids and Structures Vol. 26, No. 12 (1990), pp. 1329–1350.
[92] Pellegrino, S. A class of tensegrity domes. International Journal of Space
Structures Vol. 7, No. 2 (1992), pp. 127–142.
[93] Pellegrino, S., and Calladine, C. R. Matrix analysis of statically and
kinematically indeterminate frameworks. International Journal of Solids and
Structures Vol. 22, No. 4 (1986), pp. 409–428.
[94] Peyrot, A. H., and Goulois, A. M. Analysis of flexible transmission
lines. Journal of the Structural Division, ASCE Vol. 104, No. ST5 (May
1978), pp. 763–779.

171
BIBLIOGRAPHY

[95] Peyrot, A. H., and Goulois, A. M. Analysis of cable structures.


Computers & Structures Vol. 10, No. 5 (1979), pp. 805–813.

[96] Råde, L., and Westergren, B. Beta β Mathematics Handbook,


second ed. Studentlitteratur, Lund, Sweden, 1990.

[97] Raoof, M. Comparison between the performance of newly manufactured


and well-used spiral strands. Proceedings of the Institution of Civil
Engineers, Part II Vol. 89 (March 1990), pp. 103–120.

[98] Raoof, M. Design recommendations for steel cables. Structural


Engineering Review Vol. 4, No. 3 (1992), pp. 223–233.

[99] Robbin, T. Engineering a new architecture. Yale University Press, New


Haven, 1996.

[100] Robison, R. Fabric meets cable. Civil Engineering, ASCE Vol. 59, No. 2
(February 1989), pp. 56–59.

[101] Roth, B., and Whiteley, W. Tensegrity frameworks. Transactions of


the American Mathematical Society Vol. 265, No. 2 (1981), pp. 419–446.

[102] Russell, J. C., and Lardner, T. J. Statics experiments on an elastic


catenary. Journal of the Engineering Mechanics, ASCE Vol. 123, No. 12
(December 1997), pp. 1322–1324.

[103] Saafan, S. A. Theoretical analysis of suspension roofs. Journal of the


Structural Division, ASCE Vol. 96, No. ST2 (February 1970), pp. 393–404.

[104] Samuelsson, A. Kommentarer till G. Kärrholm, A. Samuelsson: Analysis


of a prestressed cable-roof anchored in a space-curved beam. 15 pages, (In
Swedish), April 1972.

[105] Schek, H.-J. The force density method for form finding and computation
of general networks. Computational Methods in Applied Mechanics and
Engineering Vol. 3 (1974), pp. 115–134.

[106] Shan, W., Yamamoto, C., and Oda, K. Analysis of frame-cable


structures. Computers & Structures Vol. 47, No. 4/5 (1993), pp. 673–682.

[107] Shore, S., and Bathish, G. N. Membrane analysis of cable roofs. In


Space structure: a study of methods and developments in three-dimensional
construction resulting from the International Conference on Space
Structures, University of Surrey, September, 1966 (Guildford, U.K., 1967),
R. M. Davies, Ed., pp. 890–906.

[108] Siev, A., and Eidelman, J. Stress analysis of prestressed suspended


roofs. Journal of the Structural Division, ASCE Vol. 90, No. ST4 (August
1964), pp. 103–121.

172
BIBLIOGRAPHY

[109] Stathopoulos, T. Computational wind engineering: past achievements


and future challenges. Journal of Wind Engineering and Industrial
Aerodynamics Vol. 67&68 (1997), pp. 509–532.

[110] Steel net brings new principle to cooling tower design. Energy International
Vol. 16, No. 2 (February 1979), pp. 13–14.

[111] Stefanou, G. D., and Nejad, S. E. M. A general method for the


analysis of cable assemblies with fixed and flexible elastic boundaries.
Computers & Structures Vol. 55, No. 5 (1995), pp. 897–905.

[112] Subcommitee on cable-suspended structures of the task


committee on special structures, of the committe on metals,
of the structural division. Cable-suspended roof construction
state-of-the-art. Journal of the Structural Division, ASCE Vol. 97, No. ST6
(1971), pp. 1715–1761.

[113] Szabo, J., and Kollar, L. Structural design of cable suspended roofs.
Ellis Horwood, Chichester, 1984.

[114] Tabarrok, B., and Qin, Z. A finite element procedure for form finding of
tension structures. Transactions of the Canadian Society for Mechanical
Engineering Vol. 16, No. 3/4 (1992), pp. 235–250.

[115] Tabarrok, B., and Qin, Z. Nonlinear analysis of tension structures.


Computers & Structures Vol. 45, No. 5/6 (1992), pp. 973–984.

[116] Tärno, I. Effects of contour ellipticity upon structural behaviour of


hyparform suspended roofs. Licentiate thesis, Royal Institute of Technology,
Stockholm, 1998.

[117] Tatemichi, I., Hatato, T., Anma, Y., and Fujiwara, S. Vibration
tests on a full-size suspen-dome structure. International Journal of Space
Structures Vol. 12, No. 3&4 (1997), pp. 217–224.

[118] Templin, J. T., and Schriever, W. R. Loads due to drifted snow.


Journal of the Structural Division, ASCE Vol. 108, No. ST8 (August 1982),
pp. 1916–1925.

[119] Timoshenko, S. P., and Woinowsky-Krieger, S. Theory of plates and


shells. McGraw-Hill, New York, 1959.

[120] Tomka, P. Lateral stability of cable structures. International Journal of


Space Structures Vol. 12, No. 1 (1997), pp. 19–30.

[121] Tryggvason, B. V. Aeroelastic modelling of pneumatic and tensioned


fabric structures. In Wind Engineering, Proceedings of the Fifth
International Conference, Fort Collins, Colorado, July 1979 (New York,
1980), J. E. Cermak, Ed., Pergamon Press, pp. 1061–1072.

173
BIBLIOGRAPHY

[122] Uematsu, Y., Yamada, M., Inoue, A., and Hongo, T. Wind loads and
wind-induced dynamic behaviour of a single-layer lattice dome. Journal of
Wind Engineering and Industrial Aerodynamics Vol. 66 (1997), pp. 227–248.

[123] Wada, A., and Wang, Z. Influences of uncertainties on mechanical


behaviour of a double-layer space truss. International Journal of Space
Structures Vol. 7, No. 3 (1992), pp. 223–235.

[124] Waldon, R. S. Otto, Frei. In Encyclopedia of Architecture: design,


engineering and construction, J. A. Wilkes, Ed., vol. 3. John Wiley & Sons,
1989, pp. 647–649.

[125] Walton, J. M. Developments in steel cables. Journal of Constructional


Steel Research Vol. 39, No. 1 (1996), pp. 3–29.

[126] Wang, B. Definitions and feasibility studies of tensegrity systems.


International Journal of Space Structures Vol. 13, No. 1 (1998), pp. 41–47.

[127] Wang, C. Y., and Watson, L. T. The elastic catenary. International


Journal of Mechanical Science Vol. 24, No. 6 (1982), pp. 349–357.

[128] Watson, L. T., and Wang, C. Y. A homotopy method applied to


elastica problems. International Journal of Solids and Structures Vol. 17,
No. 1 (1981), pp. 29–37.

[129] West, H. H., and Kar, A. K. Dicretized initial-value analysis of cable


nets. International Journal of Solids and Structures Vol. 9 (1973), pp.
1403–1420.

[130] Yang, H. H. Kevlar aramid fiber. John Wiley & Sons, Chichester, 1993.

[131] Yeremeyv, P. G., and Kiselev, D. B. Thin sheet metal (membrane)


suspended roof structures. International Journal of Space Structures Vol. 10,
No. 4 (1995), pp. 237–241.

174
Appendix A

Numerical data for the


Scandinavium Arena

150 152
155
275
280 157

159

285
163

290
167
120
125

112

170
103 Zoom
295

93

174
83

72

300
61 176

49

37
178

25 36

13 24
305
181
1 12
307
183

Figure A.1: Computational model for the Scandinavium Arena. A circle indicates
a node. Unfilled circles in the cable net are nodes which are loaded.

175
APPENDIX A. NUMERICAL DATA FOR THE SCANDINAVIUM ARENA

174 178
170 176
167 181
163

159

157
314
313 315
155 312 316
311 317
152
310
318
309

308 188 189


187 190
191
186 192
185 193
184 194

Figure A.2: Element model for the ring beam and columns. Node and element
numbers.

z

z
x
y
2

y 1

Figure A.3: Beam element in space with local and global coordinate systems.

176
Table A.1: Initial coordinates for beam elements (unstressed configuration). Node
numbers according to Figures A.1 and A.2.
Node Node coordinates (m)
number x y z
150 0.000 53.927 −3.948
151 2.000 53.927 −3.948
152 3.550 53.927 −3.948
153 6.000 53.629 −3.785
154 10.000 53.143 −3.520
155 10.671 53.062 −3.475
156 14.000 52.237 −3.031
157 18.000 51.229 −2.490
158 21.300 49.944 −1.827
159 25.202 48.424 −1.044
160 26.000 47.942 −0.826
161 29.180 46.000 0.048
162 29.500 45.811 0.133
163 32.436 44.017 0.941
164 33.900 42.570 1.452
165 34.462 42.000 1.650
166 37.194 39.294 2.605
167 38.504 38.000 3.064
168 41.585 34.000 4.257
169 42.500 32.822 4.610
170 43.222 31.883 4.892
171 44.376 30.000 5.367
172 46.600 26.377 6.284
173 46.834 26.000 6.379
174 47.437 25.015 6.629
175 48.817 22.000 7.225
176 50.612 18.000 8.001
177 51.835 14.000 8.547
178 52.733 11.058 8.947
179 52.894 10.000 9.020
180 53.506 6.000 9.299
181 53.855 3.710 9.458
182 53.855 2.000 9.458
183 53.855 0.000 9.458
184 3.550 53.927 −19.500
185 10.671 53.062 −19.500
186 18.000 51.229 −19.500
187 25.202 48.424 −19.500
188 32.436 44.017 −19.500
189 38.504 38.000 −19.500
190 43.222 31.883 −19.500
191 47.437 25.015 −19.500
192 50.612 18.000 −19.500
193 52.733 11.058 −19.500
194 53.855 3.710 −19.500

177
APPENDIX A. NUMERICAL DATA FOR THE SCANDINAVIUM ARENA

Table A.2: Third point in x –y  plane for beam and bar elements. Node and element
numbers according to Figures A.1 and A.2.
Node number Coordinates for third point (m)
Element
1 2 x y z
275 150 151 0.000 55.659 −4.202
276 151 152 3.653 55.659 −4.202
277 152 153 6.223 55.347 −4.026
278 153 154 10.223 54.863 −3.753
279 154 155 11.006 54.769 −3.699
280 155 156 14.440 53.920 −3.228
281 156 157 18.527 52.893 −2.660
282 157 158 21.951 51.562 −1.951
283 158 159 25.947 50.010 −1.126
284 159 160 26.908 49.425 −0.854
285 160 161 30.085 47.491 0.045
286 161 161 30.403 47.297 0.135
287 162 163 33.473 45.429 1.003
288 163 164 35.095 43.819 1.584
289 164 165 35.662 43.257 1.787
290 165 166 38.390 40.548 2.764
291 166 167 39.826 39.124 3.279
292 167 168 42.918 35.106 4.510
293 168 169 43.828 33.924 4.872
294 169 170 44.642 32.868 5.196
295 170 171 45.807 30.961 5.689
296 171 172 48.024 27.334 6.628
297 172 173 48.255 26.955 6.725
298 173 174 48.949 25.820 7.019
299 174 175 50.334 22.769 7.634
300 175 176 52.196 18.601 8.462
301 176 177 53.425 14.547 9.026
302 177 178 54.370 11.431 9.460
303 178 179 54.544 10.286 9.541
304 179 180 55.150 6.285 9.825
305 180 181 55.521 3.837 9.999
306 181 182 55.521 2.000 9.999
307 182 183 55.521 0.000 9.999
308 152 184 4.550 53.927 0.000
309 155 185 11.671 53.062 0.000
310 157 186 19.000 51.229 0.000
311 159 187 26.202 48.424 0.000
312 163 188 30.851 44.235 0.941
313 167 189 39.504 38.000 0.000
314 170 190 44.222 31.883 0.000
315 174 191 48.437 25.015 0.000
316 176 192 51.612 18.000 0.000
317 178 193 53.733 11.058 0.000
318 181 194 54.855 3.710 0.000

178
Table A.3: Load area for the loaded nodes according to Figure A.1. Nodal load =
(load area)×(load intensity).
Nodes Load area (m2 )
1–48 16.0
49–60 14.6
61–71 16.0
72–82 16.4
83–92 15.8
93–102 15.4
103–111 17.2
112–119 18.8
120–125 16.2

Table A.4: Horizontal component of the initial cable force. Cables in y-direction are
numbered from left to right, and in x-direction from down to up.
Cable Force (kN)
number x-dir. y-dir.
1 583.20 583.20
2 583.20 583.20
3 583.20 583.20
4 583.20 583.20
5 583.20 532.17
6 583.20 583.20
7 583.20 597.78
8 583.20 575.91
9 583.20 561.33
10 583.20 626.94
11 583.20 685.26
12 583.20 590.49

179
List of Bulletins from the Department of Structural Engineering, The Royal Institute of
Technology, Stockholm

TRITA-BKN. Bulletin
Pacoste, C., On the Application of Catastrophe Theory to Stability Analyses of Elastic Structures.
Doctoral Thesis, 1993. Bulletin 1.

Stenmark, A-K., Dämpning av 13 m lång stålbalk—“Ullevibalken”. Utprovning av dämpmassor


och fastsättning av motbalk samt experimentell bestämning av modformer och förlustfaktorer.
Vibration tests of full-scale steel girder to determine optimum passive control. Licentiatavhandling,
1993. Bulletin 2.

Silfwerbrand, J., Renovering av asfaltgolv med cementbundna plastmodifierade avjämningsmassor.


1993. Bulletin 3.

Norlin, B., Two-Layered Composite Beams with Nonlinear Connectors and Geometry—Tests and
Theory. Doctoral Thesis, 1993. Bulletin 4.

Habtezion, T., On the Behaviour of Equilibrium Near Critical States. Licentiate Thesis, 1993.
Bulletin 5.

Krus, J., Hållfasthet hos frostnedbruten betong. Licentiatavhandling, 1993. Bulletin 6.

Wiberg, U., Material Characterization and Defect Detection by Quantitative Ultrasonics. Doctoral
Thesis, 1993. Bulletin 7.

Lidström, T., Finite Element Modelling Supported by Object Oriented Methods. Licentiate Thesis,
1993. Bulletin 8.

Hallgren, M., Flexural and Shear Capacity of Reinforced High Strength Concrete Beams without
Stirrups. Licentiate Thesis, 1994. Bulletin 9.

Krus, J., Betongbalkars lastkapacitet efter miljöbelastning. 1994. Bulletin 10.

Sandahl, P., Analysis Sensitivity for Wind-related Fatigue in Lattice Structures. Licentiate Thesis,
1994. Bulletin 11.

Sanne, L., Information Transfer Analysis and Modelling of the Structural Steel Construction Pro-
cess. Licentiate Thesis, 1994. Bulletin 12.

Zhitao, H., Influence of Web Buckling on Fatigue Life of Thin-Walled Columns. Doctoral Thesis,
1994. Bulletin 13.

Kjörling, M., Dynamic response of railway track components. Measurements during train passage
and dynamic laboratory loading. Licentiate Thesis, 1995. Bulletin 14.

Yang, L., On Analysis Methods for Reinforced Concrete Structures. Doctoral Thesis, 1995.
Bulletin 15.

Petersson, Ö., Svensk metod för dimensionering av betongvägar. Licentiatavhandling, 1996.


Bulletin 16.

Lidström, T., Computational Methods for Finite Element Instability Analyses. Doctoral Thesis,
1996. Bulletin 17.

Krus, J., Environment- and Function-induced Degradation of Concrete Structures. Doctoral


Thesis, 1996. Bulletin 18.

Editor, Silfwerbrand, J., Structural Loadings in the 21st Century. Sven Sahlin Workshop, June
1996. Proceedings. Bulletin 19.
Ansell, A., Frequency Dependent Matrices for Dynamic Analysis of Frame Type Structures.
Licentiate Thesis, 1996. Bulletin 20.

Troive, S., Optimering av åtgärder för ökad livslängd hos infrastrukturkonstruktioner.


Licentiatavhandling, 1996. Bulletin 21.

Karoumi, R., Dynamic Response of Cable-Stayed Bridges Subjected to Moving Vehicles. Licentiate
Thesis, 1996. Bulletin 22.

Hallgren, M., Punching Shear Capacity of Reinforced High Strength Concrete Slabs. Doctoral
Thesis, 1996. Bulletin 23.

Hellgren, M., Strength of Bolt-Channel and Screw-Groove Joints in Aluminium Extrusions.


Licentiate Thesis, 1996. Bulletin 24.

Yagi, T., Wind-induced Instabilities of Structures. Doctoral Thesis, 1997. Bulletin 25.

Eriksson, A., and Sandberg, G., (editors), Engineering Structures and Extreme Events—proceed-
ings from a symposium, May 1997. Bulletin 26.

Paulsson, J., Effects of Repairs on the Remaining Life of Concrete Bridge Decks. Licentiate Thesis,
1997. Bulletin 27.

Olsson, A., Object-oriented finite element algorithms. Licentiate Thesis, 1997. Bulletin 28.

Yunhua, L., On Shear Locking in Finite Elements. Licentiate Thesis, 1997. Bulletin 29.

Ekman, M., Sprickor i betongkonstruktioner och dess inverkan på beständigheten. Licentiate
Thesis, 1997. Bulletin 30.

Karawajczyk, E., Finite Element Approach to the Mechanics of Track-Deck Systems. Licentiate
Thesis, 1997. Bulletin 31.

Fransson, H., Rotation Capacity of Reinforced High Strength Concrete Beams. Licentiate Thesis,
1997. Bulletin 32.

Edlund, S., Arbitrary Thin-Walled Cross Sections. Theory and Computer Implementation.
Licentiate Thesis, 1997. Bulletin 33.

Forsell, K., Dynamic analyses of static instability phenomena. Licentiate Thesis, 1997. Bulletin
34.

Ikäheimonen, J., Construction Loads on Shores and Stability of Horizontal Formworks. Doctoral
Thesis, 1997. Bulletin 35.

Racutanu, G., Konstbyggnaders reella livslängd. Licentiatavhandling, 1997. Bulletin 36.

Appelqvist, I., Sammanbyggnad. Datastrukturer och utveckling av ett IT-stöd för byggprocessen.
Licentiatavhandling, 1997. Bulletin 37.

Alavizadeh-Farhang, A., Plain and Steel Fibre Reinforced Concrete Beams Subjected to Combined
Mechanical and Thermal Loading. Licentiate Thesis, 1998. Bulletin 38.

Eriksson, A. and Pacoste, C., (editors), Proceedings of the NSCM-11: Nordic Seminar on Compu-
tational Mechanics, October 1998. Bulletin 39.

Luo, Y., On some Finite Element Formulations in Structural Mechanics. Doctoral Thesis, 1998.
Bulletin 40.

Troive, S., Structural LCC Design of Concrete Bridges. Doctoral Thesis, 1998. Bulletin 41.
Tärno, I., Effects of Contour Ellipticity upon Structural Behaviour of Hyparform Suspended Roofs.
Licentiate Thesis, 1998. Bulletin 42.

Hassanzadeh, G., Betongplattor på pelare. Förstärkningsmetoder och dimensioneringsmetoder för


plattor med icke vidhäftande spännarmering. Licentiatavhandling, 1998. Bulletin 43.

Karoumi, R., Response of Cable-Stayed and Suspension Bridges to Moving Vehicles. Analysis
methods and practical modeling techniques. Doctoral Thesis, 1998. Bulletin 44.

Johnson, R., Progression of the Dynamic Properties of Large Suspension Bridges during Construc-
tion—A Case Study of the Höga Kusten Bridge. Licentiate Thesis, 1999. Bulletin 45.

Tibert, G., Numerical Analyses of Cable Roof Structures. Licentiate Thesis, 1999. Bulletin 46.

The bulletins enumerated above, with the exception for those which are out of print, may be
purchased from the Department of Structural Engineering, The Royal Institute of Technology,
S-100 44 Stockholm, Sweden.

The department also publishes other series. For full information see our homepage
http://www.struct.kth.se

You might also like