A Design Tool for

Timber Gridshells
The Development of a Grid Generation Tool

MSc Thesis by M.H. Toussaint

May 2007

Delft University of Technology

Faculty of Civil Engineering and Geosciences
Section of Structural and Building Engineering
Structural Design Lab
 Master’s thesis

A Design Tool for

Timber Gridshells
The Development of a Grid Generation Tool

Submitted in partial fulfilment of the

requirements for the degree of

MASTER OF SCIENCE

in

CIVIL ENGINEERING

by

M.H. Toussaint

born 02-03-1980
in Naaldwijk, The Netherlands

Delft University of Technology

Faculty of Civil Engineering and Geosciences
Section of Structural and Building Engineering
Structural Design Lab
 Preface

Preface
This report describes the result of my Master's thesis project. This project is the completion of
my MSc study in Building Engineering at Delft Technical University, faculty Civil Engineering
and Geosciences. The subject of the Master's thesis is the development of a (conceptual)
design tool to determine the geometry of a gridshell structure. The research was performed
at the Structural Design Lab from January 2006 to April 2007. One of the research topics of
the SDL is innovative and accessible use of ICT in design. This thesis is part of this.

I would like to thank all members of my graduation committee for their contribution to this
report, and for their comments and advices during our meetings. I would also like to thank Mr
R. Harris (Buro Happold) for his quick reaction to questions that I put to him trough email.
Next, I would like to thank my fellow students in room 0.72 for their support and all the
pleasant coffee breaks. Furthermore I would like to thank my family and friends who have
been supportive and contributed to this report, especial Riny Toussaint who checked parts of
this report on spelling and grammar errors. I also want to thank my parents for supporting
me throughout my study period. Finally special thanks go to Azahara van Bergen for her daily
support and encouragements.

Matthijs Toussaint
's-Gravenhage, May 2007

Graduation committee:
Prof. Ir. L.A.G. Wagemans
Ir. J.L. Coenders
Dr. Ir. J.W.G van de Kuilen
Dr. Ir. P.C.J Hoogenboom

i
A Design Tool for Timber Gridshells

ii
 Table of contents

Table of contents
Preface ...........................................................................................................................i

Abstract ....................................................................................................................... vii

1 Introduction...............................................................................................................1

1.1 Introduction to the problem .................................................................................1

1.2 Problem definition and thesis goal.........................................................................3
1.3 Research questions..............................................................................................4

2 Timber shell structures................................................................................................7

2.1 Introduction........................................................................................................7
2.2 Use of timber through history...............................................................................7
2.3 Shells in theory ................................................................................................. 16
2.4 Examples of timber shells................................................................................... 21
2.5 Timber shells in practice: field research ............................................................... 29

3 Timber gridshells ...................................................................................................... 35

3.1 Introduction...................................................................................................... 35
3.2 Structural principles of the gridshell .................................................................... 36
3.3 Gridshell example projects ................................................................................. 50
3.4 Comparison of the gridshells .............................................................................. 66

4 Form finding ............................................................................................................ 69

4.1 Introduction...................................................................................................... 69
4.2 Form finding techniques..................................................................................... 69

5 Grid generation tool for arbitrary surfaces................................................................... 81

5.1 Introduction...................................................................................................... 81
5.2 Tool development set-up ................................................................................... 82
5.3 Shape analysis .................................................................................................. 83

6 Development of the grid generation tool..................................................................... 87

6.1 Introduction...................................................................................................... 87
6.2 Proposed method .............................................................................................. 89
6.3 Assumptions and starting points ......................................................................... 92
6.4 The gridshell design tool .................................................................................... 93
6.5 Results compared with reality by physical modelling ........................................... 117

iii
A Design Tool for Timber Gridshells

7 Members in bending ............................................................................................... 121

7.1 Introduction.................................................................................................... 121

7.2 Stresses induced by the formation process ........................................................ 122
7.3 Interaction between the laths........................................................................... 143
7.4 Curvature and RD-forces.................................................................................. 150
7.5 Conclusions .................................................................................................... 170

8 Conclusions and recommendations........................................................................... 173

8.1 Introduction.................................................................................................... 173

8.2 Conclusions .................................................................................................... 173
8.3 Recommendations ........................................................................................... 174
8.4 Evaluation of the gridshell design tool ............................................................... 175

References ................................................................................................................. 179

List of symbols............................................................................................................ 183

Appendices................................................................................................................. 185

Appendix 1: Determination of the maximum bending radius................................ 187

Appendix 2: GenerativeComponents.................................................................. 192
Appendix 3: Proof of R >> mesh size................................................................ 197
Appendix 4: Problems encountered in developing the grid generation tool............ 199
Appendix 5: Physical modelling......................................................................... 203
Appendix 6: Contact information....................................................................... 223

iv
Table of contents

v
 Abstract

Abstract
In the last few years the timber gridshell has gained popularity. Recently two gridshells were
constructed: the Weald and Downland gridshell in 2002 and the Savill Garden gridshell in
2006. These structures are examples from which the benefits of timber gridshells become
apparent. A gridshell can display elegance and style, with its slender ribs curved into shape. It
is also a sustainable structure, as the use of material is small, due to the shell behaviour. Also
the timber can source from sustainable resources. Despite the advantages of the timber
gridshell and the interest in sustainable engineering and free form architecture, the gridshell
is not used very often. What can be seen as an important reason for this is the fact that the
design process for a free form gridshell is rather complicated. An iterative design process is
used to determine the grid geometry, which is only known to a few people. Main goal of this
Master’s thesis is a study into the application of a design tool based on the geometrical
properties of the grid of a grid shell, i.e. equal distance between the nodes on the quadrangle
grid.

The gridshell structure is a structure built with long slender laths. The laths are positioned in
a flat quadrangle mat with one or more layers in two directions. This mat is then pushed and
pulled into the desired shape by bending the laths and deforming the quadrangle meshes into
rhombic shapes. When the desired shape is achieved, the laths are attached to edge supports
and the structure is stiffened by diagonal bracing or applying a continuous layer on top of the
laths. Timber is outstandingly suitable for this kind of building method. It is lightweight
compared to its strength and can be bent and twisted relatively easy.

Timber has always been used in structures by mankind. It was not until the twentieth century
for timber to be used in large scale shell structures. In 1975 the first large scale timber
gridshell was finished. This structure is the Multihalle in Mannheim. The structure can be seen
as true pioneers work. The geometry of the structure was determined by physical form
finding and it was constructed by pushing up the flat mat of laths by aid of scaffolding towers
and fork lifts.
More recently the Weald and Downland Gridshell and the Savill Garden gridshell were
constructed. The former was constructed by lowering down the flat mat into shape on a
special movable scaffolding. The latter was constructed by simply laying out the grid on a
pre-shaped formwork. The gridshells were designed by aid of a computer form finding
technique.

The gridshell design tool has been set up to generate a gridshell grid on an arbitrary surface.
The method used to generate the gridshell geometry uses two spheres to determine the
intersection points of the gridshell laths. If the two spheres are positioned in such a way that
their midpoints are located on the surface and that the two spheres are intersecting, there
will be two intersection points between the two spheres and the surface. Together with the
sphere midpoints, these four points form a mesh in the gridshell grid. A script has been
created to execute this determination of points in a sequence which locates all possible
intersection points on the surface.
The sequence starts from start-off sections, interpolated on the surface. This implies that the
correctness of the grid is dependent on the correctness of these start-off sections. Although
the results of the grid generation tool look promising, further testing is advised to prove this.

The design tool has been set up having the possibility to check the curve angles of the
generated grid. From these angles the bending stresses can be calculated and checked if the
bending or torsion stress criteria are exceeded. If this is the case, the checked element is
given a colour. The generated structure can be checked visually for stress levels exceeding
the stress criteria.

vii
A Design Tool for Timber Gridshells

The design tool has been tested on a surface consisting of two semi-spherical parts
connected by an anti-clastic part. The resulting grid was used to construct a physical model
to check the computer model on reality. Deviations were found in shape and geometry
between the two models, but also similar effects were found. Therefore the conclusion must
be that the design tool creates a grid which is correct. Form finding is needed to adjust the
shape to a surface that complies with the shape that will be formed by the grid in reality. This
form finding can be performed manually using the results of the physical model. The results
also show that (semi-) spherical surfaces are hard to create by means of a gridshell structure.
The laths have to bend and scissor too much to comply with the curvature. These kinds of
surfaces should be avoided.

In the construction of a gridshell the laths are bent into the desired shape on internal
supports. If this shape is not equal to the equilibrium bending position of the lath, the lath
will deflect toward this equilibrium position when the internal supports are removed. This
results in an undesired change of geometry and stress level. This behaviour was analysed
with a single lath. The structural analysis software GSA was used for this. A maximum
deviation of 23% was found in the stress level. This can lead to breakages if this is
unaccounted for.

The formation process of a gridshell results in bending and torsion stresses in the members.
After relaxation of the timber, a residual stress level remains in the structure. This stress has
to be accounted for in structural analysis. The stress levels can be derived from the curve
angles in the structure. These angles are part of the output of the gridshell design tool and
can therefore easily be utilised. First it has been tested if the formation bending stresses can
be implemented as a load case in GSA. For a single lath this method gets accurate results but
in a 3D structure the results are less usable. The bending stresses had better be used as a
superposition load with other load cases that are analysed in GSA, like wind and snow loads.
The stress levels can be added to the stress levels resulting from the GSA-analysis, after
applying a reduction factor which takes the timber relaxation into account.
It has been found that a complex stress distribution is present in a 3D structure, while the
stress levels resulting from the curve angles show a more continuous stress distribution. The
laths interact with each other when bent into shape, resulting in a combination of bending,
torsion and axial force. When edge disturbances are neglected, a deviation of approximately
10% is found between the stress levels resulting from analysis and the stresses calculated by
using the curve angles. The conclusion of this analysis is that the calculated stress levels can
be used for analysis. However, to get safe results, the reduction factor should not be taken
too low. More research is desired to verify the use of the curve angles to determine stress
levels and to determine a safe reduction factor.

viii
Abstract

ix
A Design Tool for Timber Gridshells

x
 Introduction

1
Introduction

This chapter is the introduction to the problem analysed in this Master’s thesis. First the
scope of the subject of the thesis will be introduced and an introduction to the problem will
be displayed in Section 1.1. In Section 1.2 the research problem will be defined, together with
the main goal of the thesis. Finally, in Section 1.3 the main goal of the thesis will be
translated into research questions, which will be guidelines in this thesis.

1.1 Introduction to the problem

One main characteristic of a shell is its large span to thickness ratio. By nature a shell uses a
very efficient way of diverting forces to a support structure, which is called shell action. This
makes it possible to create large spans with little material. This principle is also used by
nature itself, shown in examples like soap bubbles, sea shells and bird eggs. A shell structure
can show elegance and efficiency when designed correctly.
Since ancient times shell structures are used by mankind. One of the first types of shells used
by mankind is the dome. Before domes, large spans were hard to create and columns were
needed to support a roof. Stone domes structures were first seen in Roman civilisation, which
constructed semi spherical domes. One example is the Pantheon in Rome, built in 125 AD.
Parallel to this, in Persia domes also developed. Pendentives were first used here, which
enabled the dome to be supported by four columns. One of the largest dome achievements
by the Byzantine Empire is the Hagia Sofia (537 AD) in Constantinople, the modern Istanbul1.
Until the twentieth century, domes can be characterized as weighty structures. Thick walls
were needed to resist the horizontal forces resulting from the heavy stone dome. This
changed with the rapid development of computer technology, after World War II. Shell theory
was already known, but with the computer it became possible to derive and verify solutions
for very slender shells with large spans. This enabled engineers to create very thin shells with
large spans. Especially in the fifties and sixties, quite a lot of large span shells were built. The
concrete shells by Heinz Isler are excellent examples of efficient shell behaviour (Figure 1.2).
Also timber proved itself to be very suitable for shell structures, especially in hypar shells
(Figure 1.3). Timber shells lack the need for an expensive casing, which is needed for
concrete shells.
Timber also proved to be very useful for free form architecture. With a system of long
continuous timber laths, a free form lattice shell can be built. Such structure is known as a
gridshell. The shape of gridshell is obtained by bending and deforming a flat mat of timber
laths. When the quadrangle mat of laths is deformed in the desired shape, the laths can be
pinned to an edge construction and stiffened with diagonal bracing. This way, a 3D shell
structure is created out of 2D base, with only the natural behaviour of the timber. This
construction method was first used on a large scale in 1975 at the Multihalle Mannheim
gridshell.

1
http://litestraboen.blogspot.com/2007/01/domes.html accessed 15-03-2007

1
A Design Tool for Timber Gridshells

Figure 1.1: Pantheon in Rome2 Figure 1.2: Shell structure by Heinz Isler3

Figure 1.3: Timber hypar shell in Figure 1.4: Weald & Downland gridshell4
Dortmund (Holzbau Konstruktionen)

Unfortunately, increasing labour costs made the construction of shells too expensive
compared to conventional concepts in the last few decades. This caused the interest in shell
structures to diminish. However, the rise of free form architecture and the interest in
sustainable buildings has renewed the interest in shells in timber. This was acknowledged by
the completion of the Weald and Downland gridshell in 2002, and the Savill Garden gridshell
in 2006. These structures display style and elegance and were built with timber from
sustainable sources. They combine architecture and sustainable engineering in well thought-
out designs.
The structures mentioned above are the only large-scale timber gridshells existing today.
Despite the advantages in appearance and sustainability, the gridshell is not used very often.
The fact that the design process is quite complicated, probably is an important factor in this.
A gridshell is designed by means of an iterative design process which is little transparent and
the existing gridshells are all designed by only a few people. To increase the use of the
gridshell system, the knowledge on gridshell design should be spread and the design process
made more transparent. This thesis is meant to be part of this.

2
http://www.nazionaleroma.it/english/where_We_Are/surroundings/rome_pantheon.html accessed 13-02-2006
3
http://staff.bath.ac.uk/abscjkw/OrganicForms/HistoryPictures/HeinzIsler.jpg accessed 19-03-2007
4
http://www.wealddown.co.uk/downland-gridshell.htm accessed 18-12-2005

2
 Introduction

1.2 Problem definition and thesis goal

Timber gridshell structures are not often used, despite the advantages. The complicated
design process is considered the main problem in gridshell design. The largest problem in this
design process is the determination of the geometry of the structure. During construction the
gridshell lattice is bent and deformed, approximating the desired shape. The geometry of the
shape, which is dependent on the bending behaviour of the material, is not known in
advance. To be able to predict the structural behaviour of the gridshell, the design model
should be an accurate approximation of the outcome of the construction sequence. An
iterative process which takes the bending behaviour into account is needed to do this. This
design process is little transparent and without knowledge of the subject this is a huge
obstacle.

The problem in the design process can possibly be solved by introducing a different method
to determine the grid geometry. This method should be more accessible and transparent. The
proposed method in this thesis is based on the geometrical properties of a gridshell grid,
which is the equal distance between nodes. The bending behaviour of the material can be
implemented as a boundary condition to which the geometry has to comply.
This method can be implemented in a design tool. By basing this design tool on commercially
available software, the tool is kept accessible. The tool should be usable as a plug-in for
anybody who purchases this software.

A goal of this thesis is to implement the proposed method in a conceptual design tool. To
verify the results of this tool, a study is needed into the bending behaviour of the grid
members. This can be summarised into the main goal of this thesis, which is:

A study into the applicability of a geometrical design tool to the design process of
a timber gridshell

3
A Design Tool for Timber Gridshells

1.3 Research questions

The main goal stated in the previous section can be split into two subjects: the
implementation of the proposed method into a conceptual design tool and the study into the
bending behaviour of the members of a gridshell. The first part can be researched by the
following research questions and sub-questions:

1. Is the proposed geometrical model suitable for determination of the gridshell

geometry for an arbitrary surface?

a) Which free form gridshells exist today and what are their characteristics?

This question is researched in Chapter 2, 3 and 4. Chapter 2 will give an insight into
the development of timber structures trough history and an overview of timber shell
structures. In Chapter 3 existing gridshells will be studied and compared. Chapter 4
will give an overview of form finding techniques, by which the geometry of free
form structures is determined.

b) How can the proposed geometrical method be used to create the gridshell
structure?

This question leads to the development of the actual design tool. The proposed
method will be implemented in a tool which can generate the grid of a gridshell.
This process will be displayed in Chapter 5 and 6. First, in Chapter 5 the proposed
grid generation method will be further specified. In Chapter 6 the set-up of the
actual design tool will be explained and the resulting geometry will be displayed.

c) To what extend is reality approximated by the proposed method?

To answer this question, the results of the design tool should be verified in a real
structure. This is attempted by creating a physical model of a gridshell, which
models the physical behaviour of the structure. The physical modelling will be
compared with the computer model. This can be found in Section 6.5.

4
 Introduction

To be able to analyse and assess the results of the design tool, more knowledge is needed of
the bending behaviour of the gridshell members during construction. This leads to the second
research question and sub-questions:

2. What stresses occur during the process of bending the gridshell into shape?

a) What stresses occur in a slender member which is bent into shape over internal
supports?

The members of a gridshell are bent into shape while supported by internal
supports. The bending results in bending stresses, but what happens when the
internal supports are removed after construction? This process will be analysed by
modelling the process step by step in Chapter 7.

b) Can the formation stresses be deduced from the curvatures which are determined
by the geometrical design tool?

To structurally analyse the formation process, the stresses resulting from the
formation process should be known. If these can be deduced from the generated
grid geometry, it should be possible to implement these as a load case in the
structural analysis of a grid shell. This will be researched in Chapter 7.

5
A Design Tool for Timber Gridshells

6
 Timber shell structures

2
Timber shell structures

2.1 Introduction

In this chapter an overview is displayed of the use of timber in shell structures. As an

introduction to this, the use of timber trough history is investigated in Section 2.2. To be able
to assess structures as shells the different definitions of a shell and a general overview of
shell theory will be displayed in Section 2.3. After this, in Section 2.4 some illustrative
examples of existing timber shell structures will be presented to give an overview of
possibilities in timber construction. The chapter will be finalized with the display of the field
research executed in different timber shell subjects by the author.

2.2 Use of timber through history

Wood is one of the oldest building materials known to mankind. The oldest known wooden
artefacts date back some 14000 years, and probably wood is used in structures since the
ancestors of modern mankind started to build shelters5. Because of the perishability of wood,
not much is known about those ancient times. Only archaeological reconstruction can provide
some information about what might have been. Remains of dwellings in Central Europe from
around 3000 BC show us that round wood was used as the main construction material in
those days (Kuklik, cited in Thelandersson & Larsen, 2003, p.1). A lot of timber structures
also disappeared when the use of steel became common. Either they were replaced by steel,
or simply rotted away because maintenance wasn’t considered important anymore (Yeomans,
1999).
The history of the use of timber can be studied by looking at the remaining examples. The
oldest examples of timber structures still remaining date from the middle ages. In Western
Europe some 14th century structures still remain. Also in Scandinavia, where wood has always
been a resource widely available, some medieval buildings still remain. The oldest
Scandinavian building is the Borgund church in Norway, which was built in the twelfth
century. Some parts of Asia also have a long tradition of timber construction. In Japan, some
seventh century structures still remain (Thelandersson & Larsen, 2003). The largest all wood
historical building in the world also stands in Japan: the Diabutsu-Den at the Todai-ji temple
in Nara (Figure 2.1). The current building, measuring 57x50m and 47m high, was built in
1709 and houses world’s largest bronze statue of Buddha. The original, even greater building
dated from 749, but it was destroyed by fire6.

5
http://www.arplus.com/broch/articles/araug05pdfs/araug05reviewsP103.pdf accessed 26-01-2006
6
http://web-japan.org/atlas/historical/his13.html accessed 15-01-2006

7
A Design Tool for Timber Gridshells

7
Figure 2.1: Diabutsu-Den, Nara, Japan

Studying the development of timber as structural material in Western Europe, four periods
can be distinguished. In the Middle Ages wood was the main construction material and some
fine examples of all wood buildings still remain today. In the 16th to 18th century, brick came
into use and timber was mainly used in roof structures. With the development of iron in the
19th century, timber connections were improved with iron elements and laminated timber was
introduced. In the 20th century, mass produced connectors and engineered timber such as
Glulam, made the timber construction to what it is today (Ross, 2002).

2.2.1 The middle ages

In the middle ages, timber was the main building material. Lacking tension capacity, stone
was only applicable in compressive structures, like arches and domes. Therefore it was only
used in prestigious buildings. Timber was universal for roofs and framed structures. The most
common frame type was the cruck structure, used in houses and barns. One of the oldest
remaining examples is the Leigh Court barn near Worcester, UK and was built around 1325
(Horn, 1973) (Figure 2.2 and Figure 2.3). Loads are mainly transmitted trough contact
pressure in the joints, with triangulation for stability. The joints were obviously critical
elements. Joints could become quite complicated and remaining examples show carpentry
was a true skill.

Figure 2.2: Leigh Court barn (Horn 1973) Figure 2.3: Inside Leigh Court barn (Horn 1973)

7
http://www.taleofgenji.org/todaiji.html accessed 15-01-2006

8
 Timber shell structures

Figure 2.4: Schematic drawing of the Leigh Court barn (Horn 1973)

9
A Design Tool for Timber Gridshells

The most common joints in the middle ages were:

• The tennon joint, which is in compression only (Figure 2.5a)
• The lap joint, which has small tension capacity (Figure 2.5b-c)
• The scarf joint, to extend members. The joint was placed in a non critical section
in the length of the beam. A variety of locking methods was used. (Figure 2.5d-g)
• The post head and tie beam joint, which deals with the critical point where the
roof truss sits on a post. Tension capacity was needed, to support the reaction
force of the tie beam. (Figure 2.5h)

Figure 2.5: Common joints (Ross 2002)

10
 Timber shell structures

The best known Western-Europe example is probably Westminster Hall in London (Figure
2.6). It is considered a true masterpiece in timber engineering from those times. The original
building was built in the 11th century and was reconstructed in 1395. The roof spans 20,5m
and is 72m long and is of hammer-beam type. The main rafters are supported by Crown,
Queen and Hammer posts. The Hammer post rests on the braced Hammer beam, which is in
tension and resists the outward trust of the main rafter (Figure 2.7). Of course the flow of
forces in such a complicated structure is dependent on a lot of factors like support deflections
and connection stiffness. There was an ongoing discussion on whether the loads are brought
down by the great arch, or directly by the main rafters to the wall head. This was ended by
tests on a scale model and numerical models. It proved that almost all of the vertical dead
weight is supported by the corbels, and the load is brought down by combined action of the
hammer post and the great arch. The major horizontal trust is resisted by the walls, halfway
between the corbel and the wall head. Also the Hammer beam was proved to be in tension
and to relieve the wall head of horizontal force (Courtenay & Mark, 1987).

Figure 2.6: Westminster Hall (Courteny &

Mark 1987)

Figure 2.7: Drawing of Westminster Hall by Violett-le-Duc (Courteny & Mark

1987)

11
A Design Tool for Timber Gridshells

2.2.2 Early modern period

The Renaissance brought a different architectural style, based on Greek and Roman buildings.
In this new architecture, timber frames had no place. The use of brick became common and
trusses came into use for the roof structure. The structure was hidden behind a plaster ceiling
for architectural needs. Most common was the king post truss (Figure 2.8). Iron straps were
introduced to reinforce the tension connection with the tie beam. Alternative roof shapes and
truss configurations appeared, such as the queen post and multiple-bay trusses, as longer
spans were attempted. (Ross, 2002)

Figure 2.8: King post truss with iron strap connection (Ross 2002)

12
 Timber shell structures

Quality and efficiency of the truss depended on the skills and knowledge of the carpenter,
which he acquired from his master. Like in the Middle Ages this knowledge was transferred
from master to apprentice only. In the 16th century the first carpenters’ manuals were
published. The earliest were mainly on measuring and face estimation though. It was not
until the 18th century, for the first manual which dealt with geometry and construction
properly to be written by Francis Price.
In the profession of carpentry, a transformation can be seen from carpentry as a craft activity
to a production profession. If desired an architect could be hired to provide the designs,
although knowledge on construction was still provided by the carpenter. If a more modest
building was required, a tradesman capable of building in the required style could be hired.
As new architectural forms were introduced and planning, construction and decorative forms
became more sophisticated, it became more common to hire an architect. (Yeomans, 1986)
Another development took place in the theory of structures and material properties. The first
significant effort in the theory of elasticity was undertaken by Galileo (1564-1642). Robbert
Hooke (1635-1703) formulated his famous law and Petrus von Musschenbroek (1692-1761)
performed the first major series of tests on various species of timber, to determine the
strength properties of the material (Booth, 1964).

The first structural method for large span arches in timber was invented by the French
architect Philibert de l’Orme (1515-1577). In 1561, he announced his invention of a
composite timber member, composed of two or three planks on its side and sawn off radially,
then joined together with wooden pegs at several points. The longitudinal sides of the planks
were cut to an arch shape. This method used considerably less material than conventional
methods and large spans were possible. De l’Orme made designs for domes with spans up to
60m and he believed spans of 200-400m would be feasible. Despite the time-consuming
production and poor stiffness due to the large amount of parts and joints, his methods were
used until well in the nineteenth century. The largest dome using it was the dome roof of the
Halle au Blé in Paris, spanning 41m and built in 1783 (Müller, 2000).

Figure 2.9: De l’Orme’s composite member (Müller 2000)

Figure 2.10: De l’Orme arch designs (Müller 2000)

13
A Design Tool for Timber Gridshells

2.2.3 Nineteenth century

In the nineteenth century, the industrial revolution brought radical changes in society and
technology. Large industrial buildings were needed. Floors had to bear heavy loads, so large
primary beams were used. Also the roof structures needed to span larger distances. Iron was
increasingly used in strap connections to improve the tension connections. The industrial
revolution brought machine driven saws and mass produced nails and bolts, which reduced
costs dramatically and made assembly of trusses a lot easier. With the development of
wrought iron, a construction material with high tension capacity was introduced. As the
principles of statics became more clearly understood, the tie member was replaced by an iron
rod more often (Ross, 2002).
To compete with steel, new methods for timber were searched. This led to the invention of
laminated timber. Instead of using short pieces of wood like De l’Orme’s method, the long
length of the material was taken advantage of. Thin planks were bent into shape and then
jointed together with clamping bolts and collars. At first, laminated timber was mainly used in
bridges. Spans over 60m were already possible at the start of the century. Convinced of the
advantages on De l’Orme’s method, Armand Rose Emy (1771-1851) was one of the first
engineers who used laminated timber in his designs for arched structures. His methods were
widely adopted for military and factory buildings in France, with spans over 40m. Confident of
his own method he even made designs for spans over 100m (Booth, 1971; Müller, 2000).

Figure 2.11: Designs for large halls by Emy (Müller 2000)

14
 Timber shell structures

2.2.4 Twentieth century

In the first half of the last century, a wide range of mass produced connectors and shear
connectors became available. The connections in trussed structures were further improved
with these and possible spans became larger again.
At the start of the century, probably the first attempt to construct a lattice shell with a three
dimensional load transfer was made. Around 1904, Fritz Zollinger (1880-1945) transformed
De l’Orme’s composite members in a three dimensional frame (Figure 2.12). By opening up
the two parts of the member, he created a diamond shaped lattice. In this way a three
dimensional curved structure was made. At each joint one plank was going trough. The
connection was made with bolts (Figure 2.13). The system was no success. Lacking moment
capacity in these joints, the structures showed large deflections. Later, the jointing system
was improved and successfully used in Germany, see Section 2.4.1.

Figure 2.12: The Zollinger system (Müller 2000)

Figure 2.13: Node of the Zollinger system (Müller 2000)

Laminated timber was further developed in the twentieth century. Production techniques
were enhanced and new glues were developed. In early glued laminated timber, organic
glues were used. These perform well in dry conditions, but moisture degrades the glue. In
the 1930s moisture resistant Phenol-formaldehyde glues were developed. While this glue
needs heat for curing, it was difficult to produce large cross sections. Urea-formaldehyde
glues, which were developed in the 1940s, cured with normal temperatures. From the 1950’s,
glued laminated beams became in general use in construction, with only transportation
considerations limiting its dimensions (Ross, 2002).
The last few decades, production techniques improved by aid of computer technology. It is
now possible to create complex curved shapes, with double curved Glulam beams. Timber is
graded at high speed by automatic grading machines, and timber of high quality can be
produced by extracting wood deficiencies and joining the pieces with advanced joining
techniques. To illustrate the vast possibilities of modern timber construction some examples
are shown in Section 2.4.

15
A Design Tool for Timber Gridshells

2.3 Shells in theory

2.3.1 Shell surfaces

The diversity of shell surfaces is vast. Any surface which is curved in one or more directions
can be seen as a shell surface. One way of defining shell surfaces is by Gaussian curvature.
Another is the way the surface is generated. Both are displayed here.

Gaussian Curvature
A shell can be described by curves. When looking at a point on a shell, different curves can
be drawn on the surface through this point, which all have a different radius of curvature.
The curves which have the minimum and maximum value of curvature are the principle
curvatures κ1 and κ 2. The Gaussian curvature is the product of these two: κ g = κ1 ⋅ κ 2
(Hoefakker & Blauwendraad 2005). Three different types of Gaussian curvature are defined,
which are shown in Figure 2.14. A shell is typed by its type of Gaussian curvature. These
types are:

• κ g < 0: Principle curvatures are opposite. This is called an anti-clastic surface.

• κ g > 0: Principle curvatures are of the same sign. This is called a clastic surface.
• κ g = 0: At least one of the principle curvatures is zero. This results in a cylindrical
surface or a plane when both κ1 and κ2 are zero.

(a) (b) (c)

Figure 2.14: negative (a), positive (b) and zero Gaussian curvature (c) surfaces

16
 Timber shell structures

Surface generation
Surface generation is the way the surface is created. Some of the different possibilities to do
this will be displayed here:

• Surface of revolution: these surfaces are generated by revolving a curve around the
axis of revolution. This curve is called a meridian curve. Examples are the cone, the
dome and the hyperboloid, but also a cylinder is a surface of revolution.

(a) (B)

(c) (d)

Figure 2.15: surfaces of revolution. Spherical shell (a), cone (b), hyperboloid (c), cylinder (d) (Pestman)

• Surface of translation: these surfaces are created by translating one plane curve
along another, while keeping the sliding curve’s orientation constant. The curve along
which the other one slides is called the generator. When the generator is a straight
line, the translation of a curve results in a cylindrical surface. Figure 2.16 shows
examples of the surface of translation:

(a) (b) (c)

Figure 2.16: surfaces of translation. Elliptical paraboloid (a), cylindrical paraboloid (b), Hyperbolic paraboloid (c)
(Hoefakker & Blaauwendraad 2005)

17
A Design Tool for Timber Gridshells

• Ruled surfaces: a ruled surface is generated by sliding the ends of a straight line
along their own generating curve, keeping the straight line parallel to a prescribed
direction. Examples are the conoid and the hypar. The hypar is a special case in this,
because it can also be cut out of a hyperbolic paraboloid by four straight lines along
the surface.

Figure 2.17: Hypar shell as a ruled surface (left) (Hoefakker & Blaauwendraad 2005) and as a part of a hyperbolic
paraboloid (right) (Pestman)

Figure 2.18: Conoid shell (Pestman)

18
 Timber shell structures

2.3.2 General principles of shell theory

Shell structures have a few unique properties, which makes them interesting for designers
and structural engineers. Shells can display elegance and lightness if designed correctly. With
a minimum of material, large spans can be made. A shell can be recognized by its small
thickness to span ratio. What makes this possible is the principle of membrane action, which
is unique for shell structures.
The basic assumption of membrane theory is that in a distributed loaded thin shell only pure
membrane stress fields are developed. In this stress field, only normal and in-plane shear
stresses are developed, which are uniformly distributed over the cross section. Bending
stresses are negligible small compared to the in-plane stresses. Due to the initial curvature a
shell can resist in-plane forces as well as out-of-plane loads by membrane action. (Hoefakker
& Blaauwendraad 2005).

Figure 2.19: Stress resultants and load components on a shell element

However, in some cases membrane theory does not satisfy equilibrium and/or the
displacement requirements anymore and bending theory is needed. Disturbance of
membrane behaviour occurs when:

• boundary conditions and deformation constraints are not compatible with the
requirements of a pure membrane stress field (Figure 2.20 b&c)
• the shell is loaded by a concentrated load (Figure 2.20 d)
• a change in shell geometry occurs (Figure 2.20 e)

(a) (b) (c)

(d) (e)

Figure 2.20: Membrane disturbances (Hoefakker & Blaauwendraad 2005)

19
A Design Tool for Timber Gridshells

To resist the forces that disturb the membrane behaviour of the shell, additional structural
elements are needed. In a lot of shell structures ribs and/or edge beams are added, resulting
in a structure where membrane action and bending behaviour is combined to resist load.
True shells where the loads are resisted by membrane behaviour only are not seen very
often. So if the definition of a shell depends on whether loads are transferred through
membrane action or not only, probably a lot of structures which are considered to be shells
are in fact not.
In timber, true shell behaviour is not seen very often. A lot of double curved timber structures
consist of a lattice of ribs combined with one or more continuous layers of timber ply. This
continuous layer provides interaction between the ribs and membrane action in a certain
extend. The question if a timber shell is in fact a shell depends on in what extend a
membrane stress field can develop in its surface.

In the next section examples of timber shells are given. If possible an answer will be provided
to the question whether membrane action is present or not.

20
 Timber shell structures

2.4 Examples of timber shells

To give an overview of the possibilities of modern timber construction for shell structures,
some representative examples are presented in this section. Looking at the shell surfaces
defined in Section 2.3, the next surfaces will be used in the examples:

• cylindrical shells: lattice barrel vaults

• spherical shells and domes: radial rib dome

lattice dome

• Hypar shell

• Other: Suspended shells

Gridshells

As stated in the previous section, it is uncertain if all these structures are in fact shells from a
structural system point of view. When only the shape is concerned, the examples are all
curved structures and therefore shells. Therefore these structures will be designated as shells
and it is tried to answer the question whether shell action is present or not in the overview.

The gridshell, which is the structure concerned in this thesis, is also defined as a type of shell.
Here another definition problem appears, as the words grid and lattice have the same
meaning. In literacy, the terms point to the same kind of structures, although the term lattice
shell points more often to structures with triangulated grids and gridshell to quadrangle grids.
In this thesis a difference is added. The term lattice shell will be used to refer to ribbed shells,
where rib members are connected to each other in the joints. This lattice can be combined
with a structural continuous layer to provide membrane action. A gridshell is a different kind
of structure. In a gridshell the ribs are continuous from support to support and connected to
each other at the intersections. Stiffness is generally provided by triangulation. In Chapter 3,
gridshells will be investigated.

21
A Design Tool for Timber Gridshells

2.4.1 Cylindrical lattice shell

After some less successful attempts to build a cylindrical lattice shell using the Zollinger
system mentioned in Section 2.2.4, it was successfully used in 1989, in the roof of a sports
hall in Berlin-Charlottenburg (Figure 2.21 and Figure 2.22). The diamond lattice shell is
stiffened by diagonal sheeting, which provides in membrane action in a certain extend. The
spacing between the ribs is 2m and the connections are made rigid using steel plates and pin
joints. The horizontal support reaction is resisted by raised ties. Despite of this stiff system,
four steel stiffening beams were added to be sure.

2.4.2 Radial rib dome

The radial rib dome is one of the earliest structural forms. Some nomadic tribes have used
this system in their tents for centuries. The primary structural members are three pinned arch
ribs (Figure 2.23). Ring purlins resist tangential membrane forces. Because of this geometry,
the span of the purlins varies considerably, which is the main disadvantage of the radial rib
dome. Also sheeting causes problems near the top because of this. Although not a dome, a
good example of a radial ribbed construction is the ice rink in Davos, Switzerland, built in
1979-1980 (Figure 2.24 to Figure 2.26). Loads are carried by a heavy weight structure with
members with a depth up to 1950mm, to be able to resist high snow loads. The sheeting
does not provide membrane action.

22
 Timber shell structures

Figure 2.21: Sports hall Berlin-Charlottenburg (Müller Figure 2.22: Node of the barrel vault lattice (Müller 2000)
2000)

Figure 2.23: Radial rib dome (Müller 2000) Figure 2.24: Davos ice rink (Müller 2000)

Figure 2.25: Davos ice rink under construction (Müller Figure 2.26 Davos ice rink under construction (Müller
2000) 2000)

23
A Design Tool for Timber Gridshells

2.4.3 Lattice domes

In search of a structural form with identical
members, the geodesic dome was developed.
This dome is generated by projecting
icosahedrons (20 sided three dimensional figure,
composed of equilateral triangles) onto the
enclosing sphere surface (Figure 2.27). The
network can be separated into ten large
triangles. This results in a system of hexagons
with pentagons at the nodes of the large
triangles.

The main problem with the geodesic dome is the

irregular edge lengths. This causes design Figure 2.27: Geodesic sphere8
problems at the supports, when building a
geodesic dome that is not hemispherical.
To overcome this edge problem, the ensphere dome was developed. This is a combination of
the hexagonal and triangular dome. The outer ring of the triangular dome is used, to avoid
irregular edge lengths. The other rings are formed as a hexagonal dome, with ribs parallel in
three axes. One of the largest ensphere domes is the Tacoma dome in Washington with a
span of 160m (Figure 2.28-Figure 2.31). Primary members are glued Douglas fir and measure
170-220mm wide and 750mm deep. The sheeting is made of Douglas fir planks of 50mm
depth. The dome is supported on a pre-stressed concrete beam. Only two months were
needed to erect the dome.

Another fine example of the lattice shell is the roof of the thermal baths in Bad Sulza (1990)
(Figure 2.32 to Figure 2.34). It consists of two intersecting domes on an irregular plan. It is
supported by concrete columns and edge arches, which are curved in two directions. The ribs
are continuous over two bays and are connected with hard wood dowels. Steel plates were
placed on top of the joints to secure the members during construction. The sheeting is
arranged diagonally in two directions and is nailed and glued to provide membrane action and
interaction between the members.

8
http://www.skymind.com/~ocrow/dome/dometop.gif accessed 07-03-2007

24
 Timber shell structures

Figure 2.28: Tacoma dome (Müller 2000) Figure 2.29: Tacoma dome under construction (Müller
2000)

Figure 2.30: Tacoma dome under construction (Müller Figure 2.31: Tacoma dome under construction (Müller
2000)l 2000)

Figure 2.32: Bad Sulza inside (Müller 2000)

Figure 2.33: Bad Sulza vertical section (Müller 2000) Figure 2.34: Bad Sulza roof structure (Müller 2000)

25
A Design Tool for Timber Gridshells

2.4.4 Suspended shells

A different kind of structure is the suspended shell. Here, loads are transferred to the
supports through tension forces instead of compression. To provide stiffness against wind
suction and to achieve the necessary pre stress, the form of the shell has to be curved in two
directions. Tents are a good example of suspended structures. In timber it is the suspended
lattice shell.
The largest suspended shell built in timber, is the roof over a waste plant in Vienna (Figure
2.35). The plant was built in 1982 and consists of 48 radial ribs suspended from a reinforced
concrete tower of 67,35m high. The overall diameter is 170,6 m. A double layer of boards has
been nailed in two directions to provide membrane action and extra stability. It was built by
assembling the ribs in pairs with purlins and one layer of cladding on the ground (Figure
2.36). After hoisting the ribs into place, the remaining cladding was installed.

Figure 2.35: Waste plant in Vienna (Müller 2000) Figure 2.36: Waste plant under construction (Müller 2000)

A well known suspended lattice shell is the roof of brine baths at Bad Dürrheim (Figure 2.37).
It was designed by Geier and Geier in collaboration with the engineers Wenzel, Frense and
Barthel, and built in 1987. It is designed with the computer program EASY, which is a
program used for form finding membrane structures. The roof is hanged from five tension
rings, supported by tree-like columns between 9.1 and 11.5m high ( Figure 2.39). The lattice
has been constructed of double curved and sometimes twisted Glulam ribs with two layers of
diagonal sheeting to link the ribs. The meridian ribs, measuring 200x205mm, are suspended
from ring to ring, or ring to perimeter arch and follow the catenary line. Following the primary
stress trajectories, these are primary loaded in tension. The annular ribs are 80x80 or
120x140mm with 800mm spacing.
The tension rings and perimeter arches were designed as box section in such way that the
ribs can be pinned between the box panels (Figure 2.38). The corners of the arches are
supported by large cast steel bearings (Figure 2.41), mainly to resist horizontal forces. The
deadweight of the arches is supported by the façade.
Of course for such a structure, a price has to be paid. The extra work during planning and
assembly was considerable. Now, finished and famous, the extra expense weighs out the
extra costs, as it mostly does in any one-off special structure with high aesthetics.

26
 Timber shell structures

Figure 2.37: Bad Dürrheim (Müller 2000) Figure 2.38: Box section under construction (Müller 2000)

Figure 2.40: Bad Dürrheim under construction9

Figure 2.39: Tree columns (Müller 2000) Figure 2.41: Cast steel bearing (Müller 2000)

9
http://www.burgbacher.de accessed 20-01-2006

27
A Design Tool for Timber Gridshells

2.4.5 Hypar shells

Quite successful as a timber shell is the hypar. A hypar is double curved surface and it is a
part of a hyperbolic parabolic shell, described by hyperbolas and parabolas. It can also be
defined as a ruled surface, according to Section 2.3 and generated by straight lines on the
surface.
Big advantage of the hypar shell is the constant stress in the material along the surface. This
is because a (distributed) vertical load on a hypar shell is transmitted trough shear forces
only. Consider a small element of a hypar shell (Figure 2.42). Because of the torsion of the
surface, the shear forces along the edges of the element result in a vertical component. This
resultant is in equilibrium with the vertical load. The shear forces are constant along the
descriptive straight lines. At the edges the forces are absorbed by the edge beam and
transmitted to the supports as a normal force (Pestman).

Figure 2.42: A small element with shear forces along the

edges. The edge beam transmits the forces to the
supports (Pestman)

In construction the timber hypar is mostly built up out of several layers of timber in different
directions to provide membrane action, and edge beams of laminated timber. By combining
several hypars, larger roofs can be created. A beautiful example is the expo roof in Hannover,
built for the EXPO 2000 (Figure 2.43 and Figure 2.44). The roof consists of ten canopies of
39x39m carried by 18 m. high towers. Each canopy consists of four prefabricated shell
segment supported by cantilever arms. A two layered sheeting of boards is fixed at an angle
to the ribs and provides bracing of the shell. The boards are attached at a spacing of 100 mm
for transparency and ventilation of the timber. Also the synthetic roof cover is attached 50
mm clear of the lattice for ventilation.

Figure 2.43: Exporoof Hannover (D), (source Figure 2.44: Perspective view of one element of 39x39m. in
unknown) the Exporoof canopy(Müller, 2000)

28
 Timber shell structures

2.5 Timber shells in practice: field research

Not all knowledge can be found in books. Practical experience can be an important source of
information in any field of science. To get an insight in the field of practice in timber
engineering, three companies that could provide relevant information on timber shells were
visited. First Luning adviesburo, a consulting firm specialized in technical timber structures
was visited. Second, Heko Spanten was visited, who produces laminated timber structures.
Third visit was to Van Drenth Buighout. This company does not produce structural timber, but
their production techniques for curved timber elements could be interesting for curved timber
shells. The following sections report on these visits.

2.5.1 Luning adviesburo voor technische houtconstructies B.V.

Luning adviesburo in Doetinchem (NL) is a consulting agency specialized in technical timber
structures. Geodesic domes is one of their specialties, as they have developed their own node
system. Figure 2.45 shows a project designed by Luning, a geodesic dome under construction
for the planetarium in Artis zoo in Amsterdam.

Figure 2.45: Geodesic dome by Luning (GeoDomeDesign)

In a conversation with Mr. Luning, some problems with timber shell structures were
discussed. The first subject discussed was that true shells should be made of isotropic
material to generate true membrane action. Timber is not an isotropic material. When
designing a timber shell, this should be taken into account. This problem rises for instance
when designing edge connections. As an example the hypar shell which was built for the
Bundesgartenschau of 1970 in Dortmund was discussed (Figure 2.46). When using timber as
membrane skin, different layers will be applied in different directions. The stresses will pass
different layers of wood in different directions, so stresses parallel to the grain in one layer
will be rolling stresses perpendicular to the grain in the next, which is a much weaker
direction. Such detailing problems are typical for shells with a stressed timber skin. Attention
has to be paid which stress is transmitted to which layer of wood and in which direction to
the grain.

29
A Design Tool for Timber Gridshells

Figure 2.46: Hypar shell, Bundesgartenschau Dortmund 1970 (Holzbau Konstruktionen)

When timber is used as a membrane skin another problem arises. The swelling and shrinkage
of the timber due to moisture content will cause the stressed skin to change in length. As a
shell such as this hypar is statically undetermined, these changes will cause undesired
stresses in the structure. This should be taken into account in the design. According to Mr.
Luning, with the Dortmund hypar this movement is compensated by the flexibility of the
anchor cables.

With respect to glued joints, attention has to be paid to the production process. The quality
of the glued joint is dependent on this, because errors can not be corrected. Especially when
gluing on site, moist, dirt and temperature are factors that should be watched closely.
Another point of attention should be moist in the service state. Glued joints are not allowed
to be degradable by moist so the right type of glue has to be picked for each application.

When discussing the Weald and Downland gridshell, the question rose whether the shrinkage
of the green oak after construction wasn’t a problem. The bolts in the nodes probably had to
be re-tightened after a while.

A “new” type of wood was discussed: acetylated wood. The treatment method of acetylation
has been studied as early as the 1930’s. In the past twenty years this research has been
intensified and only recently, the process has become economically available.
Most traditional treatment methods use toxics such as oil, arsenics, ammonia or metal
compounds to impregnate the cell walls of the wood. Non-toxic methods like thermal
modification change the appearance of the wood and weaken it, making the method
unsuitable for most applications.
Acetylating treatment lacks all these disadvantages. The process uses acetyl, which is derived
from vinegar, to physically alter the molecular structure of the wood. No toxic chemicals are
used for this. The acetyl, being only made out of carbon, hydrogen en oxygen, is bonded to
the free hydroxyls in the wood, which are naturally present in its structure. The research
demonstrated that the physical alteration of the wood improves various material properties
considerably10:

• durability: class 5 durability softwood can be improved to class 1

• dimensional stability: swelling and shrinkage is reduced to 70-80% compared to
untreated wood
• Decay resistance: acetylated wood is largely fungi and insect repellent. Wood-eating
insects are unable to digest acetylated wood.
• Hardness: an increase of 30% in hardness can be reached
• Retention of colour: acetylation improves the stability of wood colours when exposed
to day light, ensuring consistent aesthetics

10
http://www.titanwood.com/ accessed 06-04-2006

30
 Timber shell structures

Furthermore the treatment has no negative impact on strength properties of the material and
on the appearance. Acetylated timber can easily compete with tropical hard wood, as the
durability is the same or even better. At the same time, fast growing wood such as beech can
be used, which is much cheaper than tropical hard wood.

2.5.2 Heko Spanten, Ede

The company Heko Spanten in Ede (NL) is specialized in fabricating laminated timber frames,
girders and columns. They are capable of creating cross sections of any desired size, straight
or curved, with only transportation dimensions as limitation. Figure 2.47 shows an example
project. The parts for complete projects can be pre fabricated and can be delivered ready for
assembly, e.g. making slots and drilling holes can be done in the factory (Figure 2.48).
Straight beams are produced in a large straight press (Figure 2.49). Curved beams are
produced in a press which can be adjusted to fit the desired curvature ( Figure 2.50). Also
curvature perpendicular to the main direction is possible, although limited to the height of the
press.

Figure 2.47: Tree centre, Baarn (NL) 11 Figure 2.48: Heko Spanten factory hall.

Figure 2.49: Straight laminated beam in press. Figure 2.50: Adjustable curved press.

11
http://www.bomencentrumnederland.nl/?page=hetpaviljoen_fotos accessed 06-04-2006

31
A Design Tool for Timber Gridshells

2.5.3 Van Drenth Buighout:

Van Drenth Buighout in Culemborg (NL) is part of Van Drenth Groep. This company fabricates
curved laminated elements, mainly for the furniture industry (Figure 2.51 and Figure 2.52).
Also rotary die shells are made for the packaging industry. These half circle shells are used
for stanching package shapes out of large sheets of paper or cardboard.
In its factory, half fabricates are made such as back- and armrests, seats and complete chair
seating. Either these can be delivered finished or unfinished for further processing such as
applying furnishing.
The timber mainly used is beech which has good properties for the production process. Also
other species can be used as the client desires, or different outer layers can be applied to
only give a different look. The standard size of the laminates used is 2m wide and 0.7 to
4mm thick. Standard thickness is 2mm, which gives a bending radius appropriate for most
applications. The maximum width is 2.6m, which is limited by the suppliers and the presses
used in the factory.
The products are made by pressing stacks of laminates into moulds. First the laminates are
stacked in the right order and glue is applied on every laminate (Figure 2.53). The stack is
pressed into shape in its mould and heated (Figure 2.54). The result is a curved timber shell
(Figure 2.55). It is possible to create a cross section of every desired thickness. This is limited
by cost-effectiveness, as for large cross sections production time becomes considerable large.
For every mm of thickness 1 minute of heating time is needed.

The glue used is Kaurit 325, which is urea formaldehyde resin glue, with a hardener. During
pressing, heat is needed for proper bonding of the glue and laminates. Also high frequency
heating is used.
For every different shape a different mould is needed. As every client wants his own different
shape, this results in a lot of different moulds in stock (Figure 2.56). The moulds are made of
timber mostly. Also aluminium is used for products produced in very high numbers. The
production of a simple timber mould with single curvature costs approximately 2000 to 3000
euro. This is why this production technique is only cost-effective for mass production.
After pressing, the product is further processed. The desired shape is milled out of the raw
product and holes are drilled by a 5 axial CNC machine, which makes it possible to approach
the product from every side and under every angle. For smaller number of products, sawing,
drilling and sanding is performed by hand.

There is little known about the structural value of these products. Should this technique be
applied for structural elements, first the behaviour of the adhesive used should be known. If
this behaviour is unsatisfactory, a different adhesive should be applied. Secondly, there is a
problem of cracking. With double curved elements, cracking can occur in the layers ( Figure
2.57). As the stack of laminates is pressed into its mould, the layers are not only bent, but
also deformed into shape. If it is a non-structural element which is furnished later on this is
not a problem. When it is a structural element, it is a problem as the cracking degrades the
strength of the element. Especially for the outer layers this is a problem, as stresses due to
bending are the highest in these layers. Because of the cracks, these outer layers will not act
as part of the structural section anymore.

32
 Timber shell structures

Figure 2.51: Armrests Figure 2.52: Curved elements in stock

Figure 2.53: Glue is applied on laminates

12 Figure 2.55: End result (Van Drenth Groep)

Figure 2.54: stack of laminates

Figure 2.56: Mould in store

Figure 2.57: Cracks in upper layer.

12
http://www.vandrenthgroep.nl/leaflet accessed 04-04-2006

33
A Design Tool for Timber Gridshells

34
 Timber gridshells

3
Timber gridshells

3.1 Introduction

As the interest in free form architecture has seen a mayor increase in the last decades, also
the interest in timber gridshells increased. With a timber grid, a double curved shape can be
created fairly easy, as will be explained later, and a few elegant examples of this exist today.
The first large scale gridshell structure was built in 1975 in Mannheim, Germany, which is
described in Section 3.3. Some 30 years later the building is still in use, an acknowledgement
of the success of the structure as it was designed for only one winter and one summer
(Burkhardt et al 1978). Despite the success of the Mannheim gridshell, gridshells were not
used very often anymore. At universities experimental gridshells were built by students, but
not on a large scale. An increase of labour costs made gridshells only less popular, as the
construction process is rather labour intensive, and so was the design process.
The process of designing and form finding a gridshell is not straight forward (Harris & Kelly,
2002). An iterative process is needed to find a smooth surface which is possible to create
from a flat mat of laths. In this design process, mayor developments took place in the past
few decades. As computer aided design methods improved dramatically it became possible to
use the computer in this iterative design process. 25 years after the construction of the
Mannheim gridshell, computer technology helped developing the gridshell at the Weald and
Downland open air Museum in Sussex, UK (see Section 3.3.2). Also the latest construction
and gluing techniques were used in the construction process. The Weald and Downland
gridshell is a perfect example of what is possible in timber gridshells with today’s modern
technologies. Also from the point of view of sustainability the structure can be seen as an
example. Use of material is minimized by using a shell and timber is a sustainable material
which can be obtained from environmentally sustainable sources.

The benefit of timber gridshells becomes apparent in the construction stage. Complex forms
can be shaped relatively easy (Harris & Kelly, 2002). This is achieved by laying out a flat mat
of continuous timber laths in two directions. After connecting the laths at the intersections
using a pin connection, the grid can be deformed by bending the laths and deforming the
quadrangles of the maze into rhombic shapes. If the required shape is reached, the laths are
fixed to the edge boundaries and the nodes are tightened. To keep construction easy, this
method can only be used with a material which is light, can be bent without too much effort
and has enough capacity to resist the loads after construction. Here, the properties of timber
are taken full advantage of, as timber is a light weight material and can be bent relatively
easy, with enough strength to resist loads and bending moments.

With use of modern technologies, the gridshell has become an efficient and environmentally
sustainable structure. Despite this, there seems to be a reluctance to use it more often.
Possible reason could be that the design process is considered complicated. The gridshells
that have been built were designed on basis of experience and with time consuming design
processes. Other possible reason could be that difficulties are encountered in forming the
double curved shape from an initial flat mat of laths (Kelly et al, 2001). If the possibilities and
advantages of a gridshell are unknown, a more conventional structure is quickly chosen.

35
A Design Tool for Timber Gridshells

In this chapter the principles of the gridshell will be investigated. First the structural principles
of the gridshell will be explained in Section 3.2. The topics shell behaviour, stiffness and
strength will be analysed and the bending behaviour of a gridshell element will be
investigated by determining the method to calculate the maximum bending and torsion
stresses. To learn from existing buildings, Section 3.3 displays examples of existing gridshells.
In these examples, the characteristics of each gridshell will be displayed, as well as design
and construction methods. The gridshells will be compared by their characteristics in Section
3.4.

3.2 Structural principles of the gridshell

The gridshell structural system is based on the use of continuous laths which are pinned at
their intersections. From an initial flat mat of laths, the structure is shaped by bending the
laths and deforming the mat by deforming the quadrangles of the mat to rhombic shapes
(Figure 3.1). After the shape is formed, the nodes are tightened and the structure is stiffened
by diagonal bracing.

(a) (b)

Figure 3.1: A flat mat of laths (a) is deformed to a spherical structure (b), by bending the laths and deforming the
quadrangles of the mat to rhombic shapes.

This building method creates a rather complex structural system of bent laths working
together to resist loads. The general behaviour of the system is analysed in this section. First
shell behaviour is analysed. In Section 3.2.2 the structural strength and stiffness is analysed.
After this the moment capacity of the laths is reviewed in Section 3.2.3. In Section 3.2.4 it is
determined how the stresses in an element can be determined by using the curve angle of a
member, and how these stresses should be checked on ultimate stress criteria.

36
 Timber gridshells

3.2.1 Shell action

The definition gridshell suggests that is a gridshell structure is in fact a shell. This is incorrect
as will be explained here. For a continuous shell a distributed load results in shear and normal
stresses (Figure 3.2), as explained in Section 2.3.2. This creates a rather rigid system as
every element of the continuous surface is locked in by the internal stresses and transfers
these to the neighbouring elements.
With a gridshell, one could say shell behaviour is imitated by the system of continuous
members. For a gridshell, the continuous layer is discretizised by transferring all material of
the shell element into the edges. The result is a system of four laths joined in the nodes,
which can only transmit forces in the direction of the laths and can resist out of plane
bending. The (distributed) normal stress in the continuous shell element is transferred to the
edges too, which results in normal forces on the laths (Figure 3.3). The normal stresses that
are present in the shell element are now accounted for.

Figure 3.2: Continuous and gridshell elements Figure 3.3: Gridshell element with normal forces

Without additional measures, a gridshell

structure can be seen as a series of slender
parallel arches, which work together to resist
the applied loads. When shell action is desired
the shear forces that were present in the shell
element should be accounted for by the
gridshell element. By linking the laths
diagonally, diagonal stiffness is introduced in
the gridshell and the shear forces can be
transmitted from one edge of the gridshell
element to the opposite one. The laths will
work together and the gridshell will perform Figure 3.4: Gridshell element with diagonal bracing
more as a continuous shell.

Diagonal stiffness can be provided in several ways:

• Rigid joints
• Cross ties
• Rigid cross bracing
• A continuous layer

37
A Design Tool for Timber Gridshells

Triangulation of the grid, either by applying cross ties or bracings is realized quite easy. By
applying rigid bracings, the structural behaviour of the grid would be comparable with a
continuous shell. It is also possible to create diagonal stiffness by applying a continuous layer
on top of the laths of the structure. This provides structural stiffness and cladding of the
structure in the meantime. Bracing with cross ties leaves the option to vary the diagonal
stiffness by altering the pre stress, thickness or the material of the ties (Burkhardt et al
1978).
Diagonal stiffness provided by rigid joints is less easy to realize. Rigid connections transfer
shear forces through bending moment to the supports. This can be achieved either with
connectors or gluing of the joints. Timber connectors such as dowel type fasteners or
connector plates always have a certain rotation capacity which decreases the moment
resistance and thus the stiffness of the structure. Gluing can provide good moment
connection, but complicates the construction process as gluing conditions have to be
optimized to guarantee the quality of the joint.

3.2.2 Stiffness and strength of the gridshell

As stated in the previous section, diagonal bracing is needed to provide diagonal stiffness to
the structure. If diagonal bracing is omitted, the gridshell is a series of slender arches,
resisting a load together. When this is compared with a continuous shell, a load on such
structure can be transmitted to the supports in a direct line to the supports, which keeps
deflections small. With a gridshell, the load cannot be transmitted directly but activates the
laths, which deflect to a position in which there is equilibrium of forces (Happold & Liddell
1975). The load is transmitted trough normal forces and bending moments.

When a gridshell is left unbraced the stiffness of the structure depends on the ability of the
laths to deflect to an equilibrium position and thus on the stiffness of the laths. The ultimate
deflection capacity is dependent on the ultimate moment capacity of the material used.
This ultimate moment capacity is dependent on ultimate stress level fu of the material. When
elastic behaviour is assumed, the material collapses as fu is exceeded. In a pin supported
curved element an asymmetric load results in a combination of normal and bending stresses.
The combination of these stresses determines the actual stress level. Due to normal force FN
the normal stress level is σN. The bending stresses σM due to moment M can increase until fu
is reached (Figure 3.5). This increase is the moment capacity. As the normal load increases,
less stress capacity is left and the moment capacity of the cross section decreases.

Cross section Increase in

moment

Fn

σn σm σtot

Figure 3.5: Combination of normal and bending stress; σn + σm ≤ fu

When the laths in a gridshell are already loaded by a high compressive load, little stress
capacity is left for the structure to resist moment stress, i.e. deflect to equilibrium. This
behaviour can be seen as a decrease of structural stiffness and is typical for compression
structures. In contrast with this, tension structures only stiffen as loads are increased
(Burkhardt et al 1978). Figure 3.6 shows the load-deflection curve of a continuous shell, a

38
 Timber gridshells

gridshell and a tension net under disturbance load. The continuous shell has a much higher
collapse load than the gridshell. As stated before, in a continuous shell the normal forces can
be transmitted trough the entire surface in stead of just a few laths. An increase of load has
less effect on the stress distribution and therefore the ultimate load capacity is higher.
Load

continuous shell tension net

grid shell

Deflection

Figure 3.6: load-deflection diagram (Burkhardt et al 1978)

3.2.3 Moment capacity of the cross section

In Section 3.2.2 it was found that the load resistance of a gridshell is dependent on the
moment capacity of the cross section of the grid. The out-of-plane moment capacity can be
improved greatly by increasing the moment of inertia (I ) of the members. This can be
established by increasing the height of the structural members. Applying one or more
additional layers of laths in each direction is an effective method to do this. Installing shear
blocks between the layers will provide composite action of the layers, which will further
increase I.

The disadvantage of applying a double layer of laths is that it complicates the construction
process. On top of the scissoring of the laths, the laths of the outer layer must be able to slip
relative to the laths of the inner layer when the laths are bent into shape. Figure 3.7 shows
this slip for two laths of equal length. Of course this movement must also be possible in the
joints. To be able to tolerate these two movements, the joints need to be very loosely
connected during construction. If tightened too much, twisting and bending of the laths can
prevent the layers from slipping and scissoring, which could result in breakage of the laths.
With the Mannheim gridshell and the Weald & Downland gridshell different solutions for this
problem are used, which will be discussed in Section 3.3.1.4 and 3.3.2.5 respectively.

Figure 3.7: Slip of the laths. Two laths of equal length are bent to the
same radius. Fixed in the middle, the outer edges move relatively to
each other.

39
A Design Tool for Timber Gridshells

3.2.4 Bending behaviour of an element

The shape of a gridshell is created by applying a large deflection to the system of laths. The
laths are bent and twisted to shape the mat of laths to the desired shape. The possibility to
bend and twist makes it possible to create this shape, but also puts restrictions on the shape.
The laths can only bend to a curvature in which the ultimate stress level is reached.
In this section, it is explained how the maximum curvature and torsion angle can be
determined. First the maximum curvature is calculated using the ultimate bending stress of a
material in Section 3.2.4.1. Next, in Section 3.2.4.2 the design bending stress which should
be used for design purposes is modified using Eurocode5. Section 3.2.4.3 reviews the
maximum torsion angle. Finally, the combinations of stresses occurring in a bent and twisted
member are investigated.

3.2.4.1 Maximum curvature

The maximum curvature of a timber member depends on the maximum bending strength fm
and the modulus of elasticity E0. An increasing fm leads to a decreasing bending radius. A
stronger piece of timber can be bent further prior to failure than a weaker piece, but a larger
bending strength also implies a larger modulus of elasticity. This means the bending radius
does not decrease proportionally to the increase of moment, e.g. a larger moment or force is
needed to bend a strong piece of timber to the same radius of a weaker one. This can be
reviewed using simple mechanical analysis.

To analyse bending behaviour, a segment of a member shown in Figure 3.8 is considered.

The member is subjected to a bending moment, which results in internal stresses and a
bending curvature.

Δx

M Δw

θ
Figure 3.8: Bending member

40
 Timber gridshells

The bending moment due to the curvature can be calculated as:

M = EI κ ( 3.1 )

Where

I = 1 bh3 ( 3.2 )
12

and

1
κ= ( 3.3 )
R

With:
M = bending moment;
E = modulus of elasticity;
I = moment of inertia;
κ = curvature of the beam;
b = width of the member;
h = height of the member;
R = radius of curvature.

The bending stress in the outer fibres of a member can be calculated with:

M M
σm = = ( 3.4 )
W 1
6 bh
2

With:
σm = bending stress;
W = moment of resistance.

The formula above can now be rewritten as:

1 Ebh3
12
= 1
6 bh 2σ m ( 3.5 )
R

Replacing σm by the maximum bending stress of the material fb, this leads to:

Eh
Rmax = ( 3.6 )
2 fm

The maximum bending radius of a timber member is thus dependent on the maximum
bending stress of the material and its moment of elasticity. When a high curvature is desired,
e.g. a small bending radius, timber with a small E / fm ratio should be selected. This can be
achieved by selecting the timber with the desired properties by strength grading. For
example, in the Weald and Downland gridshell oak was used, which has a low E / fm ratio by
nature. Also a high timber grade was achieved by selecting timber by strict requirements such
as limiting grain slope and avoiding knots (Harris & Kelly 2002) (see Section 3.3.2.4)

41
A Design Tool for Timber Gridshells

The derivation made above can be found more extensively in Appendix 1: Determination of
the maximum bending radius. Also an analysis of the maximum bending radius related to the
strength class material properties can be found here.

For design purposes, the design value of the ultimate stress capacity should be used. The
design bending strength value can be determined according to Eurocode5 with:

kmod f m ,k
f m,d = ( 3.7 )
γM

Where:
fm,k = the characteristic bending strength
kmod = the modification factor which takes into account the influence of load duration,
service class and material type
γ m. = the partial factor for material properties

Values for kmod and γm can be found in Eurocode5. The bending stress due to the construction
process can be seen as a medium term load (1 week to 6 months). For medium term loads
kmod =0.8 can be used. For the partial factor for material properties, γm =1,3 is recommended
for solid timber.

42
 Timber gridshells

3.2.4.2 Curved timber and moment capacity

Curving a beam induces bending stresses in the material. The distribution of these stresses in
a curved or tapered beam is non-linear (Blass et al, 1995). This has an effect on the
maximum bending stress and should be taken into account when designing such a beam.
Additionally the bending stress causes radial stresses perpendicular to the grain.

For design purposes, the maximum bending stress of a curved beam can be calculated
approximately with simple bending theory, modifying M/W with a shape factor kl, which
depends on the ratio between the height of the cross section and the radius of curvature.
This factor takes into account the strength reduction due to bending of the laminates during
production (Blumer, 1975, 1979 referenced in Blass et al, 1995). This also can be found in
Eurocode5 part 1-1, Section 6.4.3. According to the Eurocode this theory applies for glued
laminated timber and LVL only. The theory is based on the theory of thin anisotropic plates,
taking into account the influence of stresses perpendicular to the grain. It is not known if
taking this influence into account gives correct results when members that are not built up
out of thin layers are concerned, but are strongly curved. However with a gridshell the effect
of a non-linear stress distribution is present due to the strong curvature in combination of
bending stresses. Therefore the factor kl is still applied to take into account this effect.

The non-linear bending stress distribution can be illustrated by a small section in bending.
Based on Navier's theory of elasticity, the strain of the outer fibres of a curved beam is
smaller than the strain of the inner fibres. To regain equilibrium of forces the neutral line has
to shift down and according to Hooke's law and the maximum bending stress |σo| in the outer
fibres is smaller than the maximum bending stress in the inner fibres |σi|. (Blass, 1995).

σo Δlo

lo

h M

li
M

b σi Δli

Figure 3.9: Distribution of bending stresses in a curved beam (Blass et al, 1995)

43
A Design Tool for Timber Gridshells

According to Eurocode5 the design bending stress in the apex zone of the curved beam
(Figure 3.10) should be calculated by:

6M ap , d
σ m , d = kl ( 3.8 )
bhap2

With:

2 3
⎛ hap ⎞ ⎛ hap ⎞ ⎛ hap ⎞
kl = k1 + k2 ⎜ ⎟ + k3 ⎜ ⎟ + k4 ⎜ ⎟ ( 3.9 )
⎝ R ⎠ ⎝ R ⎠ ⎝ R ⎠
σm,d = the design bending stress;
Map,d = the design moment in the apex zone
hap = height of the beam in the apex zone
b = width of the beam

Figure 3.10: Curved beam with its apex zone (Eurocode5)

Factors k1 to k4 depend on αap, which is the angle of taper in the beam. This angle is zero in
case of a curved beam. The factors become:

k1 = 1
k2 = 0.35
k3 = 0.6
k4 = 6

44
 Timber gridshells

For a beam of 50x50mm bent to the maximum curvature, the factor kl is 1.0013. When hap
<< r the influence of the curvature on the design bending strength becomes small. The
effect is therefore neglected in this thesis. Also the additional stress perpendicular to the
grain is related to the angle of curvature and can be neglected due to the small factor hap /r.

For verification of the failure condition of bending, Eurocode5 also applies a reduction factor
kr to take into account the bending stresses in the laminates of curved glued laminated timber
due to production. A curved glued laminated beam is produced by bending the laminates and
then gluing them together. The stresses due to this bending reduce in time due to plastic
deformation and relaxation, but still have to be taken into account when verifying the failure
condition. By applying the reduction factor it is recognized that the laminates are in a
stressed condition, but these stresses are not equal to the full stress induced in the
production process (Harris, R., pers. comm., 22 dec. 2006). This effect also counts for the
laths of a gridshell, where relaxation reduces the bending stresses which were caused by
bending in the formation process. By applying factor kr these reduced bending stresses can
be taken into account.

The failure condition due to bending can now be verified with:

σ m , d ≤ kr f m , d ( 3.10 )

With:
⎧ R
⎪⎪1 for in ≥ 240
t
kr = ⎨ ( 3.11 )
⎪0.76 + 0.001 Rin R
for in < 240
⎪⎩ t t
σm,d = the design bending stress
fm,d = the design bending strength
t = the design bending stress
Rin = the inside radius of a beam

When a lath of 10m in length and a cross section of 50x50mm and timber grade D30 is
considered and this lath is bent to its minimum bending radius kr becomes:

R = c * h = 167 *50 = 8350

(for values of constant c, see Appendix 1: Determination of the maximum bending radius)

Rin 8350 − 25
= = 166.5 < 240
t 50

rin
kr = 0.76 + 0.001 = 0.9265
t

45
A Design Tool for Timber Gridshells

3.2.4.3 Torsion
The maximum torsion stresses occur in the middle of the edges of a member. According to
Eurocode5 the following expression needs to be satisfied:

τ tor , d ≤ kshape f v , d ( 3.12 )

With
⎧ h
⎪1 + 0.15
k shape = min ⎨ b for rectangular cross sections
⎪⎩2.0

τtor,d = the design torsion stress;

fv,d = the design shear strength;
kshape = factor depending on the shape of the cross section;
h = the larger cross-sectional dimension;
b = the smaller cross-sectional dimension.

A general formula for the maximum torsion stress and torsion angle for a circular rod is:

Tz
τ max = ( 3.13 )
Ip

And

Tl
θt = ( 3.14 )
KG

With
T = the torsion moment
Ip = the polar moment of inertia
G = the shear modulus
l = the length of the member
θt = the angle of torsion

According to Young and Budynas (2002) these formula can be modified to calculate the
maximum torsion moment and torsion angle in a member with a solid rectangular cross
section (Figure 3.11). This maximum stress occurs at the midpoint of each side. With the
modifications τmax and θt can be calculated with:

46
 Timber gridshells

3T ⎡ ⎛b⎞ ⎤
2 3 4
⎛b⎞ ⎛b⎞ ⎛b⎞
τ max = ⎢1 + 0.6095 ⎜ ⎟ + 0.8865 ⎜ ⎟ − 1.8023 ⎜ ⎟ + 0.9100 ⎜ ⎟ ⎥ ( 3.15 )
8ab 2 ⎣⎢ ⎝a⎠ ⎝a⎠ ⎝a⎠ ⎝ a ⎠ ⎦⎥
for a ≥ b

Tl
θt = ( 3.16 )
KG 2b
With:
⎡16 b⎛ b4 ⎞⎤
K = ab3 ⎢ − 3.36 ⎜1 − ⎟⎥ 2a
⎣3 a ⎝ 12a 4 ⎠⎦
for a ≥ b Figure 3.11: Rectangular cross section

The maximum shear strength fv can be used in equation 3.15 to calculate a maximum torsion
moment T. This can then be used to calculate the maximum torsion angle θt,max, with
equation 3.16.

47
A Design Tool for Timber Gridshells

3.2.4.4 Combination of stresses

When a gridshell is bent in shape, the laths are subjected to bending moment in two
directions, torsion and shear forces in two directions. This deformation can be seen as an
ultimate limit state deformation, as these stresses are present in the shaping process as a
maximum. After construction, relaxation of the timber takes place and part of the formation
stresses will disappear. It is estimated that the formation stresses will decrease to 0,5 of the
initial stress level. This half will remain present in the material as a residual stress level. It is
needed to take this residual stresses into account in structural analysis.

To check the stress combination in the construction phase for failure conditions, a criterion
which combines all stress combinations is desired. Guidelines exist for only for certain
combinations of stresses. A criterion for the combination of stresses in a gridshell is set-up
here using existing combination criteria.

First, the available stress combination criteria available are stated.

According to Eurocode5 a member bent in two directions in ultimate limit state should satisfy
the following condition:

σ m, y ,d σ m, z ,d
+ km ≤1 ( 3.17 )
f m, y ,d f m, z ,d

σ m, y ,d σ m, z ,d
km + ≤1 ( 3.18 )
f m, y ,d f m, z ,d

With:
km = factor which makes allowance for re-distribution of stresses and the effect of
inhomogeneities of the material in a cross section. For solid timber, glued
laminated timber and LVL, km is 0,7 for rectangular sections.
σm,y/z,d = the design bending stress in the y or z direction;
fm,y/z,d = the design bending strength in the y or z direction.

These criteria are valid for cases without buckling or tilting. It is assumed that no buckling or
tilting occurs, as elements in a grid shell are relatively short (0,5-1m) The criteria for the
combination of bending and tension or compression can be stated as:

σ t / c ,0, d σ m, y ,d σ m, z ,d
+ + km ≤1 ( 3.19 )
f t / c ,0, d f m, y ,d f m, z ,d

σ t / c ,0, d σ m, y ,d σ m, z ,d
+ km + ≤1 ( 3.20 )
f t / c ,0, d f m, y ,d f m, z ,d

With:
σc/t,0,d = the design compression/tension stress parallel to the grain;
fc/t,0,d = the design compression/tension strength parallel to the grain.

48
 Timber gridshells

Eurocode5 does not provide guidance for the combination of shear and torsion stresses. The
following criterion is given by Möhler and Hemmer (1977, cit. Blass et al (1995)):

2
τ tor , d ⎛ τ xy , d ⎞
+⎜ ⎟⎟ ≤ 1 ( 3.21 )
f tor , d ⎜ f
⎝ v,d ⎠

With:
τxy,d = the design shear stress;

Furthermore, timber can be checked on a combination of stresses with the stress criterion of
Norris (1964, cit. Van de Kuilen & De Vries, 2005)

2 2 2
⎛ σ m,d ⎞ ⎛ τ v,d ⎞ ⎛ σ c / t ,90, d ⎞
⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ ≤ 1 ( 3.22 )
⎝ f m,d ⎠ ⎝ fv,d ⎠ ⎝ f c / t ,90, d ⎠

Where:
σc/t,90,d = the design compression/tension stress perpendicular to the grain;
fc/t,90,d = the design compression/tension strength perpendicular to the grain.

The last part of this formula takes into account the stress perpendicular to the grain for
timber sawn angular to the grain. This is not the case in a gridshell and can be left out.

Based on the criterion on Norris, the following criterion for the combination of bending,
shear/torsion and tension/compression can be stated:

2 2 2
⎛ σ m,d ⎞ ⎛ τ v,d ⎞ ⎛ σ c / t ,0, d ⎞
⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ ≤ 1 ( 3.23 )
⎝ f m,d ⎠ ⎝ fv,d ⎠ ⎝ f c / t ,0, d ⎠

in which:
σ m, d = σ m, y , d + σ m, z , d
τ v , d = τ xy , d + τ tor , d

This criterion is quite conservative, because reduction factors have been left out. However,
the timber needs to have enough stress capacity for resisting different loads during the
lifetime of the building, on top of the (diminished) stresses resulting from the formation
process. For a first analysis, the criterion gives a satisfying approximation. Further
investigation is desired to determine if a more accurate criterion can be used.

49
A Design Tool for Timber Gridshells

3.3 Gridshell example projects

3.3.1 Multihalle Mannheim

3.3.1.1 Background
Every two years a garden exhibition is held in one of the major cities in the Federal Republic
of Germany. In 1970 it was decided that this Bundesgartenschau of 1975 was to be held in
Mannheim (Happold & Liddell 1975). A master plan was developed for the area of
Herzogenried park, where the festivities would take place. The plan included a large covered
space, where a variety of activities could take place. The winning design for this space was
the gridshell of Frei Otto and Ove Arup & Partners as structural engineers. Otto, famous for
his structures of tension nets, uses hanging models for his designs. Also the Multihalle of
Mannheim was designed with hanging models.
The complex consists of a multi-purpose hall, where a range of activities can take place, such
as exhibitions, flower shows, entertainment, concerts, theatre, sports activities, etc. In a
second smaller hall a restaurant is situated. The halls are connected by a covered link. Figure
3.12 shows an aerial view of the complex. Figure 3.13 shows an inside picture (Burkhardt et
al 1978)

Figure 3.12: Arial view of the Multihalle (Burkhardt et al Figure 3.13: Inside view of the Multihalle (Burkhardt et
1978) al 1978)

3.3.1.2 The roof

The Multihalle lies as an artificial hill in its surroundings. The gridshell is designed in such way
that the shape continues the hilly landscape of the garden. The enclosed building area of
3600 m2 is air conditioned and it is covered with PVC coated fabric. The grid has a maximum
longitudinal span of 85m. It is built up out of a double layered mat of laths of Hemlock Pine.
This performed best in test with respect to shrinkage and creep. It was also selected for its
straight grain and availability in long lengths. The laths have a cross section of 50x50mm and
are spaced on 500mm. Approximately 72000m of lath was used to construct the shell. The
grid is supported by four different edge supports: concrete foundations, cables, laminated
timber beams and arches. Diagonal stability is improved by applying cross ties. Pairs of 6mm
cables are spaced at 4,5m in both directions.
After completion the roof was tested by loading it to 1,7 times the design load. This was
applied by hanging dust bins filled with 90 l. of water on every ninth node. Deflections stayed
well under the calculated deflection, proving the gridshell a safe structure.

50
 Timber gridshells

3.3.1.3 Structural modelling and analysis

Physical modelling
The initial form finding of the gridshell was entirely performed with physical modelling. At
first, a wire model was made of the preliminary design. A second hanging model was made
using the system line of the structure, to determine the initial data on node coordinates
(Figure 3.14). This model had to be very accurate as an error will be enlarged when
transferring the data on a full scale construction. Other methods of form finding, i.e. drawing
and computing the coordinates were considered, but these appeared not to be better. To
correctly compute the coordinates, input data was needed on form behaviour of the grid, but
this was not available. Also the number of iterative steps to calculate the coordinates would
be numerous and time consuming.
The model was made of rings and links (Figure 3.15). Although these were machine
manufactured, the sizes of the elements were not exact, due to tolerances in the
manufacturing process and wearing of the tools, so it was impossible to rule out all
imperfections. One of every 3 laths was modelled. In the model a mesh was used of 15mm,
which represented 1,5m in the full scale structure. The intermediate nodes would be
interpolated afterwards. The model was set up on a marble plate, so inaccuracies due to
shrinkage or distortion of the base of the model was ruled out.

Figure 3.14: Hanging model (Burkhardt et al 1978) Figure 3.15: Detail of the model (Burkhardt et al 1978)

Computer form finding

The coordinates of the nodes were determined by taking stereo photographs of the model.
With these coordinated, the structure could be analysed by computer calculations. Because of
inaccuracies of the model, not all members of the hanging model were in tension. To correct
this, the correct geometry was calculated using the force density method. The intermediate
results of the iterative steps were analysed. Deviations of a medium force smaller than 15%
were considered non critical. When larger deviations occurred, adjustments were made in
geometry. From this calculated suspended net, the data needed for production and erection
was derived (Burkhardt et al 1978).

51
A Design Tool for Timber Gridshells

Structural analysis
When Ove Arup & Partners started with their designs, very little reference material was
available. Only three much smaller gridshells were built before. Initial studies were performed
to determine the design load and hand calculations on shell buckling were made. Structural
design was started before the final geometry was finished. To gain knowledge on behaviour
of gridshells, tests on a working model of the Essen gridshell ware executed. The results
showed that lath size had to be increased to enhance buckling resistance, to 100x100mm.
This would give problems however with forming the shell, as more force is needed to bend
the laths with a bigger cross section. Also the contract was already let and a lath size of
50x50mm was agreed. Decided was to apply a double layered mat, so bending flexibility was
maintained during construction. After applying shear blocks between the layers, sufficient out
of plane bending strength will be provided.
Design loading was determined by using wind and snowfall records in the area and by wind
tunnel testing on a 1:200 scale model. This way, the design could be fully optimized, instead
of just using normative average loading values. Also tests were carried out on the nodes and
to investigate stress relaxation of the timber.

To determine a collapse load, tests were executed on a model of the gridshell. To correctly
model the full size structure, dimensional analysis was used, which means scaling the factors
that govern the behaviour of the physical system. Perspex members were used to model the
grid. One model member represented six double layer members on full scale (Figure 3.16).
The model was tested by hanging 100mm nails on the nodes and dial gauges were used to
measure the deflections (Figure 3.17).
The test results were compared with the results of computer calculations. The collapse load
determined by the tests was 2,8 kN/m2. The calculations gave a value slightly over 1 kN/m2.
This difference occurs because the model does not scale the shear deformation of the full
scale structure. This shear resistance is largely controlled by the individual slip per unit of
force of the joints. The Perspex model corresponds to a full size structure with a very high
value of joint stiffness, thus resulting in a high collapse load.

Figure 3.16: Test model (Burkhardt et al 1978) Figure 3.17: Tests on the model (Burkhardt et al 1978)

52
 Timber gridshells

3.3.1.4 Connection details

Typical node joint
The laths are bolted together in the nodes. To provide slipping of the outer layers during
erection, these layers have slotted holes. After erection, shear resistance is needed, so the
bolts are tightened to provide sufficient friction. Testing indicated that tension in the bolts
would decrease in time, due to shrinkage of the timber. To prevent this, spring washers are
applied (Figure 3.18 and Figure 3.19).

Figure 3.18: Typical node joint (Burkhardt et al 1978) Figure 3.19: Photo of a typical node joint13

Joints in the laths

The laths are prefabricated into laths up to 40m by finger jointing. The joints used were
20mm with a 6mm root, to suit the machines of the factory. Quite a lot of joints broke during
site handling and erection due to this short connection length. The laths were repaired by
nailing 50x25mm lapping pieces to each side. This was also used to lengthen the laths into
the required length.

Boundary connections
Four types of edge connections are used in the Multihalle (Figure 3.20). Originally, Frei Otto
designed all boundaries on columns as cable edges. Cable edge supports are possible where
boundary forces are more or less constant and where the change in angle of the boundary
system line at the column is not so large that excessive support reactions are caused. These
conditions were only met at parts of the restaurant. To connect the gridshell to the cable, it is
first connected to a plywood board. This board also helps to cope with the differences in
distributed lath forces and the cable reaction force. At the columns the cables are brought
together (Figure 3.21).
Where cable supports could not be applied, edge beams are used. 60x500mm laminated
timber beams are connected on either side of the grid. The laths are bolted to these beams.
Also where the grid is connected to the concrete support, it is first connected to timber
beams.
The timber beams are connected to the steel columns with steel plates. These plates are
bolted to the beams and welded to the columns (Figure 3.23). The connection itself was
simple, but the geometry was not as for every connection the angles were different. The
cutting profile for the plates had to be determined accurately to provide production drawings.
A special computer program was written for this task. Also the production drawings for the
arches were produced by a specially written computer program.
At the valley between the Banana and the Multihalle a laminated timber beam is applied with
a circular section. Steel connecters connect the grid to the valley beam (Figure 3.22).

13
http://www.kunst.uni-stuttgart.de/wendland/progetti/mannheim/mannheim_img accessed 20-01-2006

53
A Design Tool for Timber Gridshells

Figure 3.20: Edge layout (Burkhardt et al 1978)

Figure 3.21: Cable edge connection (Burkhardt et al Figure 3.22: Valley beam connection (Burkhardt et al
1978) 1978)

Figure 3.23: Edge column connection (Burkhardt et al

1978)

54
 Timber gridshells

3.3.1.5 Assembly
As with the Essen gridshell, the Mannheim grid was supposed to be lifted into shape by
cranes. Calculations however, showed that four 200 tonnes cranes were needed over a period
of three weeks. The high costs of this forced the contractors to think of other options. Finally,
the gridshell was erected by pushing up the lattice from underneath. Fork lifts were used to
lift the scaffolding towers (Figure 3.25). By using these, the horizontal movement of the
scaffolds as the shape of the lattice changed could be followed easily. To spread the forces
on the grid, H-shaped timber spreaders were used. A ball joint between the scaffold and
spreader provided rotation to fit the curve of the shell. To reduce costs, as few scaffolds as
possible were used. This resulted in quiet long spans between the scaffolds. To eliminate low
areas between the scaffolds, flying struts were used (Figure 3.26).
The PVC coated fabric is applied and fitted to the structure on site (Figure 3.27). It is made of
sheets of the fabric, hot welded together, and attached to the grid with over 400.000 staples.

Figure 3.24: Under construction (Burkhardt et al 1978) Figure 3.25: Scaffolding towers (Burkhardt et al 1978)

Figure 3.26: Intermediate strut (Burkhardt et al 1978) Figure 3.27: Applying the roof skin (Burkhardt et al
1978)

55
A Design Tool for Timber Gridshells

Figure 3.28: Interior (Wilhelm 1985)

56
 Timber gridshells

3.3.2 Weald and Downland gridshell

3.3.2.1 Background
The Weald and Downland gridshell, shown in Figure 3.29 and Figure 3.30 is built at the
Weald and Downland Open Air Museum in Singleton in the UK, and was finished in 2002. This
museum has over 45 historical buildings from South East England, which have been rescued
and rebuilt there. The museum needed a new building for study and practice of building
conversation, especially the timber framing tradition in England. Also a new conservation
store for collection items was needed. To extend the collection of timber structures into the
21st century, the new building should be an example for modern rural buildings. The result of
the combination of skills of the architect Edward Cullinan Architects, the engineer Buro
Happold and the carpenter, Green Oak Carpentry Company, truly is a display of modern
craftsmanship (Harris & Kelly, 2002). The basement of the building is sunken into the hillside
and houses the conservation store. The workshop is situated on the ground floor and is
roofed by the gridshell. The roof has the shape of a triple-bulb hourglass, to mirror the rolling
shapes of the West Sussex Downs.

Figure 3.29: The Weald & Downland gridshell14 Figure 3.30: Inside view of the gridshell14

3.3.2.2 The roof

The triple-bulb hourglass roof is 48m long and between 11-16m wide. It has an internal
height of 7-10m. The roof is clad with Red Cedar boards and polycarbonate glazing. The grid
is built up out of 50x35mm oak laths in four layers, like the gridshell in Mannheim, to provide
good out of plane resistance. The laths have a spacing of 500mm in areas with high load, and
1000mm in other areas. Original designs were made with 500mm spacing for the whole
structure. Careful examination of the forces and stresses by computer analysis lead to
increasing of the spacing, which saved construction time and reduced costs significantly.
Shear blocks are screwed between the different layers to provide composite action between
the layers. To increase stability, diagonal bracing is applied. In the lower parts the bracing
laths run in longitudinal direction, at the top in transverse direction. These also provide
support for the cladding boards.
The cladding consists of polycarbonate glazing, which covers the upper part of the roof, and
Western Red Cedar boards. This closed part is insulated with a multi-layered composite
insulation material14

14
http://www.wealddown.co.uk/downland-gridshell.htm accessed 18-12-2005

57
A Design Tool for Timber Gridshells

3.3.2.3 Structural modelling and analysis

Physical modelling
Physical modelling involved an important part of the modelling process. It provided a lot of
information on form, structure and construction of the shell. Scheme models of wire mesh
were made to research the form. After this a larger 1:30 model using wood strips was built.
This proved to be very instructive. The geometry of the model was used to determine the
boundary conditions for the computer model. It also served as presentation model (Harris &
Kelly 2002).
After this an accurate wire mesh model was made, to explore the formation of the shape and
to determine self weight bending. Dimensional analysis was used to correctly model the full
scale structure (Jensen 2000). A boundary template was used to determine if the correct
shape was reached. Also internal scaffolding was modelled to approximate the formation
procedure as good as possible. The wire mesh was loaded by hanging large steel nail on the
nodes.
Main conclusion of the experiment was that the saddles would not form themselves under
dead weight. External forces are needed to reach the final shape. As the lattice becomes
more curved, larger forces are needed to stretch the lattice, so the initial lay out of the lattice
should be already stretched instead of with square angles. In was also concluded that
formation of the waists costs a lot of force when first a barrel shape is adopted. Formation of
the valleys and tops should be formed simultaneously to prevent breakage of the laths.

Computer form finding

The geometry of the grid was obtained using a form finding technique called dynamic
relaxation (see 4.2.2.2). This is an iterative process which modifies the desired shape by
minimizing kinetic energy of the model as it is made to oscillate around an initial
approximation of the shape. Numerous iterations are needed to create a smooth shape
(Harris & Kelly 2002). The method is generally used to examine oscillations of a pure
catenary shape to generate a final shape. The W&D gridshell however is not a purely
catenary shape. It is not possible to create the saddles with a hanging chain model. The
method had to be modified by including the bending stiffness of the laths, to correctly model
the shape.

Structural analysis
The structure is analysed and designed in accordance with the Eurocode5, using timber grade
D30 with a characteristic bending strength of 30N/mm2. Structural analysis was performed
using the elastic analysis software STAAD Pro. Two methods were used. Dynamic relaxation
was used for second order analysis of buckling instability. Using the STAAD model a
deflection analysis to compare the deflected shape with the non-linear analysis under the
same load. This proved that under working load the behaviour is elastic, with adequate factor
of safety against buckling. Detailed stress checks were made using the information provided
by the STAAD model.
It was also concluded that the shape of the grid enhanced the load bearing capacity. The
waisting along the building improved the strength and stiffness against asymmetric loads
(Harris & Kelly 2002).

58
 Timber gridshells

3.3.2.4 Timber
Selection
A number of species of timber were considered for construction of the gridshell. Based on
their properties and the results of a series of tests, oak was selected for the grid. This species
performed best in the structural tests carried out at Bath University, with respect to bending
behaviour. It proved to be stiffer than other species and has a considerably higher bending
strength. Although it needs more force than other species to be bent it can achieve a smaller
bending radius prior to failing. Also it showed a somewhat plastic failure mode, compared to
more brittle timber. Additionally, oak has high natural durability, so no treatment would be
necessary. Possible leakages will not lead to decay of the timber (R. Harris, pers. email
comm. 14 March 2006).
Another reason is that oak is one of the most common used materials in the museum’s
collection of buildings and the species was readily available from sustainable sources in UK.
Strangely enough, eventually the timber was sourced in Normandy, because better timber
with a lower rejection rate was available there.
The main disadvantage of oak is that the direction of grain varies significantly, due to the
growth characteristics. This was overcome by cutting out the defects and joining the pieces
together to create laths of the required length. Selection was made on the following
requirements (Harris & Kelly 2002):
• a maximum slope of grain of 1:10
• No dead knots or live knots. Only small pin knots were allowed, provided that they
formed no more than 20% of the width of any face.
• No shakes or splits
• No sapwood (sapwood is not naturally durable and not resistant to infestations)

Lath production
Although tests indicated that the shell could be formed using dry oak, green oak was used.
Green timber is easier to bend, thus making the forming process of the shell easier. One
disadvantage of oak is its acidity, making it difficult to joint with adhesives. Using green oak
only make this worse, also because of the moisture content of the green timber. After an
adhesive was found which is not affected by this acidity and the moisture content, the use of
it was no problem anymore.
In total, approximately 6000 linear meters of lath were needed. The average length of
individual pieces was 0.6m so 10.000 finger joints were used in the structure. Laths of 6m
length were made off site, using a special machine, to maximize the quality with a minimum
wastage. The finger joints are hardly visible, so despite the amount the joints have minimal
visual impact (Figure 3.31).
On site the laths were jointed into laths up to 37m long. Here, a scarf joint with a slope of 1:7
is used (Figure 3.32). The joints were made in a tunnel tent to avoid weather influences.
There is an interesting contrast between the two jointing techniques used. Finger jointing was
the latest technology, while scarf jointing has been used for centuries.
The joints performed well during construction. There were approximately 145 breakages
during forming. Almost all broken joints were finger joints. Main causes were (Harris & Kelly
2002):

• Pinching of the lattice on scaffold supports

• Tight curvature
• Tension build-up because restriction of the relative slipping between the layers
• Dry joints

59
A Design Tool for Timber Gridshells

Figure 3.31: Finger joint14

Figure 3.32: Scarf joint14

3.3.2.5 Connection details

Typical node joint
For the connection between the laths a special connector was designed. It consists of three
plates, connected with four bolts (Figure 3.33). The middle plate has a pin in the centre,
keeping the connection into place. The outer layers can slide freely in their direction during
shaping of the structure. Two of the four bolts can be used to connect the diagonal bracing.
The connecter proved to be very successful and has been patented.

Figure 3.33: Assembly of a typical node joint14

Edge connection
At the edges the gridshell is connected to the floor of the structure. The laths are bolted
between two layers of plywood and connected to the floor beams. The floor and floor beams
are cut into shape (Figure 3.34) and the first layer of plywood boarding is connected to the
Glulam floor beams with angle brackets prior to the erection of the roof. Holes are drilled in
the boarding aligned with the holes in the brackets. Figure 3.35 shows one of the positions of
the brackets. After the grid is lowered, blocks are installed on this location to fix the sandwich
structure firmly to the brackets on the inside (Figure 3.36). Also the gaps between the layers
are filled up with timber where the grid overlaps the boarding to create a solid section four
times the depth of a lath. The second layer of plywood is attached and the laths and plywood
layers are bolted together (Figure 3.37). The whole sandwich is bolted to the brackets on the
inside to create a rigid edge connection.

60
 Timber gridshells

Figure 3.34: Floor and beams are cut into shape14 Figure 3.35: Location of the brackets14

Figure 3.37: Edge detail14

Figure 3.36: Connection to the edge14

Figure 3.38: Finished connection14

61
A Design Tool for Timber Gridshells

Figure 3.39: Flat mat of laths14 Figure 3.40: Start of lowering14

Figure 3.41: Adjustable jack14 Figure 3.42: Angled jacks14

Figure 3.43: Halfway down14 Figure 3.44: Completed form14

62
 Timber gridshells

3.3.2.6 Assembly
Instead of pushing or lifting up the grid against gravity, the Weald and Downland gridshell
was lowered into position. The flat grid of laths was laid out on a special scaffolding system
at the level of the valleys of the shape. This scaffolding system used adjustable jacks to
accurately alter the heights to form the shape of the gridshell.
The mat was laid out at a height of 7m. As concluded from the experiences of physical
modelling, the mat was not laid out with 90° angles between the laths. 96° and 84° angles
were used and the resulting mat was 47x25m (Kelly et al, 2001).
The process of lowering the grid was carefully monitored visually and with the information
provided by scaffolding jacks system. The longitudinal centre line was used as a reference
line, as this line was not to move transversely. The nodes on this line were painted white, to
be able to visually check if the nodes remained on a straight line.
The scissoring and sliding of the laths was influenced with straps in plane of the grid. By
tensioning the straps in the desired direction, the scissoring was stimulated. The strapping
arrangement was continuously monitored. Failing of the lattice to scissor or of the laths to
slide relatively to each other would lead to breakages so the process was observed carefully.
After formation was finished, the valleys resembled the designs very well. The domes
however appeared to be too low. Also the perimeter nodes around the domes were 300mm
too low. Adjustments were made by pushing up the perimeter nodes using small jacks.
The formation process was very successful. Observation was seen as the key control of the
formation process. Potential problems could be isolated and dealt with continuously by
observing the behaviour of the lattice.

63
A Design Tool for Timber Gridshells

3.3.3 Savill Garden gridshell

3.3.3.1 Background
Opened in June 2006, the Savill Garden visitors centre is the latest timber gridshell. The Savill
Garden gridshell is the roof of the new visitor centre of The Royal Landscape. This park,
which is part of Windsor Great Park near Windsor Castle in the UK needed a new visitors
centre. Existing facilities were obsolete and scattered around the park. The Savill Garden
visitors centre now forms the central gate to the Royal Landscape.
As a giant leaf the gridshell lies between the sloping surroundings. The gridshell itself forms
the roof of the public part of the visitors centre, housing an information desk, shop, bar,
restaurant and a banqueting hall. The slopes adjacent to the roof are actually the vegetated
roof of different facilities like offices, stock, kitchen and toilets.
The building is designed by Glenn Howells Architects. For the engineering and realisation, the
same companies as the Weald and Downland gridshell were contracted: Buro Happold and
The Green Oak Carpentry. After years of preparation, the structure was built in three years,
costing 8,3 million Euros. The careful preparation has resulted in a well thought design of the
complex structure. The result is well recognized for its excellence in design and sustainability.

3.3.3.2 The roof

The roof is a symmetrical surface, formed by a central dome with a smaller dome on either
side and ends pointing slightly upward. The roof is 98m in length, has a maximum of 28m in
width and is 4,5 to 8,50m high15. The shape can be described by a sinusoidal wave in
longitudinal direction and a parabola in transverse direction. The structure is a double layered
grid of laths with a mesh size of 1m. The laths are 80mm wide and 50mm high. Shear blocks
of 80 *120*300mm connect the two layers. This way a structural cross section with a height
of 190mm is created.
To save costs, it was chosen to apply a continuous layer on top of the laths for diagonal
stiffness, in stead of diagonal bracing. Birch plywood is used for this, which also serves as a
roof cover together with isolation and an aluminium rain cover.
Support construction is a circumferential waving steel tube, which is supported by steel
angled columns. Timber laths are bolted to the edge tube, trough laminated veneer lumber
elements, provides strong connection with less support points.

3.3.3.3 Timber and assembly

Weeks of computing were needed to structurally model the structure. To find the right timber
for the structure and to verify the results of the structural modelling, prototypes of different
timbers species were tested. It was found that Larch was the most suitable for the structure.
Larch is known for its natural strength and its natural durability16, which adds to the
structure’s environmental value. All modelling work paid off, as breakage only occurred twice
during construction.
All timber used was FSC approved. 400 larch trees were carefully selected and processed into
a total of 35km of laths. For this 11000 finger joints were used.
The laths are finger joined with water based PUR glue into 6m pieces. On site 35m laths were
created. These laths were laid out onto a temporary frame and manoeuvred into exact
position by hand, using laser measuring equipment and simple drainage tubes. In place, the
laths are screwed together. The construction method is fairly simple, but it took a full year of
labour with 20 craftsmen to construct the timber roof.

15
http://www.e-architect.co.uk/oxford/savill_building.htm accessed 10-04-2007
16
http://www.scottishwood.co.uk/hardwoods.htm#Larch accessed 10-04-2007

64
 Timber gridshells

Figure 3.45: the Savill Garden gridshell (De Groot, 2007)

Figure 3.46: inside view of the roof structure (De Groot, Figure 3.47: steel ring tube on angled columns (De Groot,
2007) 2007)

Figure 3.48: gridshell under construction (De Groot, 2007) Figure 3.49: the gridshell under construction17

17
http://www.greenoakcarpentry.co.uk/g-body-savill-garden-gridshell.htm accessed 24-03-2007

65
A Design Tool for Timber Gridshells

3.4 Comparison of the gridshells

The almost 30 years that are between the constructions of first and latest two gridshells
makes them differ in a lot of ways. Especially the aid of advanced computer techniques made
the architectural and structural design and design process of the Weald and Downland
gridshell and the Savill Garden gridshell a lot different. Advanced production and construction
techniques made it possible to create a high quality building. The quality of the Mannheim
gridshell lies maybe not in elegant construction solutions, but more in the fact that building it
was true pioneers work and in the architectural design itself.

To compare the structures, the differences are displayed in Table 3.1. Also the strengths and
weaknesses of these differences are stated. Apart from design methods, one of the main
differences is found in the construction method. The mat of laths of Mannheim was pushed
up, with Weald and Downland it was lowered, which proved to be more controllable and
accurate than the method used with Mannheim. With the Savill Garden gridshell, construction
was performed without deforming a mat of laths, but by manoeuvring every single lath into
position.

The Savill Garden gridshell is probably built with the simplest methods of the three. On site,
the laths were placed into position and simply screwed together. Also the problem of sliding
layers was avoided by adding the different layers one by one. If this method is used, accurate
data is needed on the positions of the laths and nodes. Because of the framework, not much
room is left for on site adjustments.
The temporary framework is another downside of the building method, because this is
probably partially once-off. But the method probably saves a lot of money, compared to the
expensive adjustable scaffolding used with the Weald and Downland gridshell. Also
manoeuvring every single lath into place and screwing them together are downsides, as it
takes a lot of manual labour. On the other hand, the bolts and connectors used in the other
gridshells also had to be placed by hand, and probably also had to be retightened due to
shrinkage of the timber.
The simple building method probably also contributed to the low amount of breakages, which
is two. In the Weald and Downland this was 145. During construction of the Mannheim
gridshell there were numerous breakages, but no figure could be found.

The structures themselves differ most in means of stability. With Mannheim, stability is
provided by steel cable bracing, which makes the stability dependent on the tension in the
cables. Weald and Downland uses timber laths as rigid bracing. With the Savill Garden
gridshell, again a different system was used: continuous sheeting providing shell action.

Construction costs can also be compared. The Weald and Downland has costed 1611€/m2
gross internal floor area (Harris et al. 2003), with a total of 2,35 million Euro18. For the Savill
Garden gridshell this is 3009€/m2 gross internal floor area19, with a total of 7,05 million
Euro20, costing approximately twice as much as the Weald and Downland gridshell. The
difference between the two structures can probably be explained by the 'luxurious' features,
like the steel edge structure and a glass façade of 8m. high. The Savill Garden gridshell itself
costed only 34,6% of this amount (1038€/m2). The Weald and Downland gridshell costed
only 16% of its total amount of building costs per square meter (260€/m2). It is uncertain
what causes this difference because no specification of the building costs could be found. It
probably can be explained by the fact that the Savill Garden is approximately four times
larger than the one in Weald and Downland, taking more time to construct and thus more
labour costs. The costs for the Weald and Downland gridshell can be compared with the

18
http://vs2.i-dat.org/unstructured02/eco2.html accessed 09-04-2007
19
http://www.cnplus.co.uk/qsweek/images/Savill_Gardens.pdf accessed 09-04-2007
20
http://www.qsweek.com/nav?page=qsweek.contentspage&view_resource=5941138 accessed 09-04-2007

66
 21
Mannheim strength/weakness (+/-) Weald&Downland strength/weakness (+/-) Savill Garden strength/weakness (+/-)
Design Mainly physical form - labour intensive Combination of physical + high accuracy possible unknown (assumably
finding - less accurate and computational form computational form finding)
finding
Broken laths Pushed up with fork lift + cheap Lowered down with special + controllable positioned lath by lath on + simple
trucks scaffolding formwork
- hard to control - expensive + controlable
- labour intensive
Stability Cable bracing + easy to construct Ridged braces - high accuracy needed continuous layer + stiff system
- stiffness depends on tension + same type of connection as
in ties the grid

Node Slotted holes - holes weaken the cross Special connector + works very good screwed + simple
connection section - development takes time and + cheap
costs money

Timber used Air dry + strength and shape stays Green + bending is easy unknown
constant - distortion and shrinkage
possible due to drying
building could be built economically.

- special glue needed

Species Hemlock Pine + straight grain Oak + high bending strength Larch + strong
+ durable + durable
- irregular grain direction
Lath joint Finger joint - short weld length caused Finger joint by jointing + high quality Finger joint with water
breakages machine based pur glue
that the Savill Garden is a quite expensive building.

On site joint Lap pieces - holes weaken the cross Scarf joint + continuous laths possible unknown
section

- stiffness is reduced - on site gluing needed

Broken laths unknown 145 (mainly finger joints) 2
Cladding PVC coated fabric + translucent Western Red Cedar boards + insulation possible Continuous layer of Birch - no daylight possible
plywood
Table 3.1: comparison of the Mannheim and the Weald and Downland gridshells

- on site welding of seams and polycarbonate glazing + daylight trough the roof + insulation possible

http://www.worldtrading.net/currency/old_converter.html accessed 10-04-2007

needed

- requires accurate assembly

for aestetics
Building costs of 435 Euro per m^2 of 260 Euro per m^2 of gross 1038 Euro per m^2 of
the roof only usable floor area floor area gross floor area
this is 175€/m2. The total building costs of the building per square meter were 435€/m2

increased labour costs makes the figures differ. According to Burkhardt et al. (1978) the
(DM850/m2). It is hard to compare the Mannheim gridshell with the latest two. Especially
building costs of a typical building in medium costs range (Harris et al. 2003). This implies

exchange rate of 200121 this should be 1,83 million Euros. With 10500m2 usable floor area
The Mannheim gridshell roof costed DM 3586000 (Burkhardt et al, 1978). Converted with the

67
Timber gridshells
A Design Tool for Timber Gridshells

68
 Form finding

4
Form finding

4.1 Introduction

In free form architecture the search for the desired and optimal shape is always present. In
general, the search for the optimal shape is called form finding. In architecture, the search is
for an optimal shape with respect to the desired appearance or functionality of the enclosed
space. Architectural from finding can be seen as shape finding with the optimum being rather
subjective to the opinion of the architect. In structural design, the shape itself is not
optimized, but the optimum is found in an optimal structural behaviour. The architectural and
structural optima do not often have the same shape, which conflicts the interests of the
architect and structural engineer. The art in designing is joining these interests in an early
stage of the design process, combining them to create an overall optimum.
For finding the optimal structural shape, different techniques can be used. In Section 4.2
these techniques will be elaborated.

4.2 Form finding techniques

Form finding can be applied on many subjects. In general, it is used to find an optimized
solution to a problem, with respect to one or more criteria. To be able to solve it, the problem
has to be defined in such way that these criteria can be optimized. Mathematically, this
optimization can be performed in various ways. ‘Classical’ methods such as the LaGrange
function, linear and non-linear programming, but also intelligent and adaptive models, which
are part of Artificial Intelligence, can be seen as optimization methods. Coenders (2004)
made an extensive overview of available methods.
Focussing on the optimization of structures, form finding is closely related to structural
optimization. For instance, the optimization problem can be the structural behaviour of the
structure, with a minimal use of material as optimization criteria. There are different
techniques for finding this optimum. Physical modelling is the easiest method. With physical
models, a quick insight in structural behaviour and form can be gained. High accuracy
however is difficult to gain. More advanced computational methods are available to provide
the accuracy necessary for further design and construction.

69
A Design Tool for Timber Gridshells

4.2.1 Physical modelling

With physical modelling the principles of nature are used to determine shapes. By evolution,
nature has developed ways to transmit forces with minimal use energy and material. A
spider’s web, the shape of soap bubbles and radiolarians (oceanic organisms) are examples
of structures by nature, highly efficient with respect to use of energy.

Figure 4.1: Structures by nature. A spider’s web22, soap bubbles (Otto 1982) and a radiolarian23

Back to structures of mankind, cable structures and fabric membranes are good examples of
physical form finding. When loaded tension structures adapt to a unique shape which is not
known in advance (Lewis 2003). The structure adopts itself to resist the load in an optimal
way, minimizing the energy needed to resist the load. This behaviour can be modelled quite
easy with flexible materials such as pantyhose fabric. Soap film is also very suitable to model
tension fabric structures, as the soap film adopts the minimal energy surface between its rigid
supports. Well known for using and exploring these techniques is Frei Otto and his Institut für
Leichte Flächentragwerke (ILEK), which conducted a lot of research in structures in nature
and form finding techniques. Otto used physical form finding is his designs, for instance the
net roof of the Olympic stadium in Munich.

Compression structures can also be modelled physically. Soap bubbles and pneumatic
membranes can be seen as the inverse of a compression shell. Arch structures can be
modelled by threads or cables. Under a uniform load a cable structure adopts to the catenary
line. Lacking moment capacity, there will be only pure tension in that cable. Made rigid and
turned upside down this would become an arch structure with only pure compression. Antoni
Gaudi used this principle in his famous designs, such as the Sagrada Familia in Barcelona
(Figure 4.2). Gaudi used threads to model the columns and arches of his structures. Little
weights hanged on the cables modelled the weight of building elements. Also with the
Mannheim gridshell Otto used hanging models to form find the shape of the structure (see
Section 3.3.1.3).

22
http://gallery.hd.org/_c/natural-science/spider-web-closeup-backlit-against-dark-sky-in-corner-of-door-frame-
mono-1-DHD.jpg.html accessed 16-04-2007
23
http://www.cs.brown.edu/courses/cs024/imagesTechnology.html accessed 16-04-2007

70
 Form finding

Figure 4.2: Hanging model by Gaudi and its inverse24

The design methods used by for instance Gaudi and Otto can be seen as the oldest form of
form finding. The geometry of a design is completely determined by physical models. The
building of these models is very time consuming however. A change in design can not be
passed through on the model easily. Nowadays, physical models are mainly used as physical
representations of the design, a presentation model to other parties. The design is made with
the aid of drawing and engineering software, in which changes can be made quickly.
However, using physical modelling in the design stage can still play an important role in
gaining better understanding of their design, especially with complicated designs with double
curved surfaces.

24
http://staff.bath.ac.uk/abscjkw/OrganicForms/SlideShow/ accessed 4-5-2006

71
A Design Tool for Timber Gridshells

4.2.2 Computational form finding

Computational form finding is a numerical optimization process. With numerical optimization,
an iterative calculation sequence is used to determine the optimum of a problem. In
structural optimization, starting point of the optimization process is the initial shape of the
structure, which does not comply with the desired optimization conditions. In the iterative
form finding process, numerical algorithms describing mechanics of structures perform
geometric adjustments of the structure, until static equilibrium is reached. The outcome of
the form finding process is a computational model of the structure, describing its shape,
stresses and deformations under load conditions by a set of numerical and graphical data
(Lewis 2003). To reach the desired optimum, iterations steps can become quite numerous,
especially when designs are complicated. Present-day computer technology makes it possible
to make these calculations in a matter of minutes without much difficulty.

What makes numerical optimization especially useful for form finding is the possibility of
solving non-linear problems in a relative easy way. When deflections get large, structures can
display geometrically non-linear behaviour, which means the deflection is not linear with
respect to its loadings. Tents and membrane structures are known for this behaviour. Non-
linear problems are hard to solve with normal mechanics formulae, but it can be solved
relatively easy with an iterative calculation process. Numerical optimization is an iterative
process, which makes it suitable for form finding structures which display non-linear
behaviour.

The two computational form finding techniques mainly used and developed over the last
decades, are the force density method and the dynamic relaxation method. These methods
are explained in the following sections.

72
 Form finding

4.2.2.1 Force density

The force density (FD) method allows you to generate shapes of tension structures that are
in static equilibrium. It is a method that uses a linear system of equations to model static
equilibrium of pre-tensioned cable net under prescribed force/length ratios. By assuming a
constant ratio of force to length, non-linear system of equations becomes linear and can be
solved. The next example is used to illustrate the principles of the FD method, which is
reproduced from Lewis, 2003.

4
3
4
3

1
1 y 2

x
Figure 4.3: Structure with 4 members

Principles of the force density method

Figure 4.3 shows a structure of 5 nodes and 4 members, which can be seen as part of a
larger structure such as a cable net. The equilibrium of forces at node 5 can be determined
by resolving the tension forces Tm into x, y and z components (m = member number). This is
achieved by multiplying Tm by the direction cosines, defined as the ratios of the member
lengths projected on the direction axes, to the actual lengths Lm. The directional force
components can be defined as:

Tm ( xi − xk )
Tm , x =
Lm
Tm ( yi − yk )
Tm , y = ( 4.1 )
Lm
Tm ( zi − zk )
Tm , z =
Lm

73
A Design Tool for Timber Gridshells

Equilibrium of forces can be calculated by adding up the directional components, which

should be equal to the external directional load vectors Px, Py, and Pz acting on node 5.

T1 ( x1 − x5 ) T2 ( x2 − x5 ) T3 ( x3 − x5 ) T4 ( x4 − x5 )
+ + + = Px
L1 L2 L3 L4
T1 ( y1 − y5 ) T2 ( y2 − y5 ) T3 ( y3 − y5 ) T4 ( y4 − y5 )
+ + + = Py ( 4.2)
L1 L2 L3 L4
T1 ( z1 − z5 ) T2 ( z2 − z5 ) T3 ( z3 − z5 ) T4 ( z4 − z5 )
+ + + = Pz
L1 L2 L3 L4

In the above system of equations, the member lengths Lm are non linear. By introducing a
constant tension coefficient qm, the force density, which is defined as the ratio Tm/Lm, the
system becomes linear:

q1 ( x1 − x5 ) + q2 ( x2 − x5 ) + q3 ( x3 − x5 ) + q4 ( x4 − x5 ) = Px
( 4.3)
q1 ( y1 − y5 ) + q2 ( y2 − y5 ) + q3 ( y3 − y5 ) + q4 ( y4 − y5 ) = Py
q1 ( z1 − z5 ) + q2 ( z2 − z5 ) + q3 ( z3 − z5 ) + q4 ( z4 − z5 ) = Pz

This linear system of three equations has three unknowns, the coordinates x5, y5 and z5, and
can therefore be solved. Notice that if the force density qm has a constant value for all
members, the calculated coordinate system is independent of qm. If a change in configuration
of the network is desired, different values for qm can be defined for each member.

Using of the force density method

With the FD method the initial shape of cable nets and membranes can generated with only
the boundary coordinates and the force densities to be specified. In the example only four
members were used. The system of equations can be expanded for larger structures, but of
course the system of equations becomes larger with an increasing number of nodes and
members. Matrix notation and calculation are used to keep the system manageable.
The method has developed into a non-linear FD method, which makes it possible to
implement other requirements than only the boundary conditions, such as constant mesh
width or constant pre-stress, making it a versatile design method for tension structures
(Lewis 2003). One software package using FD in which this is possible is the programme
EASY, in which complete tension structures can be designed and calculated. The package is
not limited to fabric membranes only. The package also includes an application for calculating
element with bending resistance, EasyBeam. The roof of Bad Dürrheim (Section 2.4.4) was
designed and engineered using this application.

74
 Form finding

4.2.2.2 Dynamic relaxation

In dynamic relaxation, form finding is performed by a pseudo-dynamic process, which can be
explained as follows. The mass of the structure is lumped in the nodes and oscillate around
the initial position under influence of the out-of-balance forces. Due to artificial damping, the
masses come to rest in an equilibrium position (Lewis 2003). In its original form, the iterative
process uses viscous damping, where the movement of the nodes is damped by damping
coefficients in its formulae. The alternative process with ‘kinetic damping’ proved to be more
stable and to converge more rapidly when dealing with large displacements. This procedure
traces the motion of the structure. When a peak in total kinetic energy is detected, all
velocities are set to zero and calculations are restarted with the current geometry. The
energy peaks generally decrease and the process is repeated until all vibrations have been
dissipated and static equilibrium is reached (Barnes 1999).

Computation procedure
The computation procedure includes the following steps (Barnes 1999):

First, the resultant force for node i is calculated with Newton’s second law for every direction.
For simplicity, only the x-direction is stated here:

Rixt = M iVixt ( 4.4)

where
Rixt = resultant force for node i
M i = lumped mass at node i
Vixt = acceleration of node i

The acceleration at time t can be approximated with 4.5:

Vixt +Δt / 2 − Vixt −Δt / 2 ( 4.5 )

Vixt =
Δt

where
Vixt = acceleration at node i
Vixt ± Δt / 2 = speed of node i at (t ± Δt / 2)

Substituted equation 4.5 into 4.4 the velocity at (t + Δt / 2) can be found:

Δt
Vixt +Δt / 2 = Vixt −Δt / 2 + ⋅ Rixt
Mi ( 4.6 )

The geometry can then be updated to time (t + Δt ) :

xit + Δt = xit + Δt ⋅ Vixt + Δt / 2 ( 4.7 )

75
A Design Tool for Timber Gridshells

Now the geometry is completely updated and the kinetic energy (KE) at (t + Δt / 2) can be
determined. If the current KE is larger than the previous KE (at t − Δt / 2) the system returns
to 4.6 after calculating the new link forces and its resultant by:

t +Δt
⎛T ⎞
⋅ ( x j − xi )
t +Δt
R t +Δt
ix =P t +Δt
ix + ∑⎜ ⎟ ( 4.8 )
⎝ l ⎠m

where
Pixt + Δt = updated link force acting on node i
Tm = tension in link m connecting nodes i and j
lm = length of link m
t + Δt
⎛T ⎞
∑ ⎜⎝ l ⎟⎠ = summation over all links m connecting to i
m

If the current KE is less than the previous KE a peak in KE has occurred somewhere between
(t − 3Δt / 2) and (t + Δt / 2) . The process is then restarted at 4.4 with all velocities set to zero,
after correcting the geometry. Because the peak occurred some time before (t + Δt / 2) the
current geometry, which is set to xit + Δt , has to be corrected to a more exact position
corresponding with the KE peak time t* (Figure 4.4).

Kinetic energy

t∗ t
t
t − 3Δt / 2 t + Δt / 2 t + Δt / 2

Figure 4.4: Graph of peak in kinetic energy

76
 Form finding

t +Δt
⎛T ⎞
In equation 4.8 the term ∑ ⎜⎝ l ⎟⎠ can be seen as the nodal stiffness Si. Si can be
m

adjusted to the appropriate stiffness for different structures. In the example above, Si is the
geodetic stiffness of a tension structure. For cable net structures the elastic stiffness can be
added and S becomes:

⎛ EA T⎞
Si = ∑ ⎜ +g ⎟ ( 4.9 )
⎝ l l ⎠m

where g is a factor for allowing the geometric stiffness to increase due to large change of
length of the links during analysis.

For form finding structures where bending stiffness is present, Si should include the bending
stiffness of connecting beam links:

2 EI
Si = ∑ ( 4.10 )
l3

What is easily forgotten is the nodal forces resulting from the bending of the laths. The
continuous laths are initially straight and bent into shape. This results in moment in the laths
and shear forces si acting on the nodes, normal to the chord between the nodes. In next
figure, two subsequent links in a gridshell are considered:

Figure 4.5: Moment and shear forces in laths

M
j α

sa sb sa sb
la lb

i lc k

Figure 4.5: Moment and shear forces in laths

If the bending radius R and moment M are calculated with:

lc EI
R= M =
2sin α R

with E assumed constant.

The shear forces sa and sb at j thus are:

2 EI gsin α 2 EI gsin α
sa = sb =
la glc lb glc

For further elaboration on DR we refer to Barnes, 1999.

77
A Design Tool for Timber Gridshells

Figure 4.7: British Museum Great Court Image by C.J.K. Williams24

Figure 4.6: Computer image of the

British Museum Great Court roof Image
by C.J.K. Williams24

Evaluation of the dynamic relaxation method

Big advantage of DR over matrix methods like force density, is that no matrix manipulations
are needed, therefore saving computer CPU time. Although more iterations are needed to
reach equilibrium, the iterations take little CPU time, still resulting in less total CPU time
compared to matrix methods. Furthermore dynamic relaxation shows high numerical stability
and is capable of providing solutions to highly non-linear problems. This makes dynamic
relaxation widely recognized as very successful for modelling pre-stressed cable nets and
membranes (Lewis 2003).
Good example of application of DR is the British Museum Great Court roof (Figure 4.6 and
Figure 4.7), designed by Chris Williams. Williams is known for writing his own small
mathematical software applications for his designs. Figure 4.8 shows the results of one of his
(simpler) DR programs.
Besides from writing your own software, one software application in which it is possible to
apply DR is Oasys GSA, which is a frame and finite element programme. Also a DR form
finding module called GsRelax is implemented.

Figure 4.8: Dynamic relaxation of a basketball net. Left: initial net. Right: loaded net adjusted by dynamic relaxation.
Image and software by C.J.K. Williams24

78
Form finding

79
A Design Tool for Timber Gridshells

80
 Grid generation tool for arbitrary surfaces

5
Grid generation tool for
arbitrary surfaces

5.1 Introduction

As stated in Section 3.1, the gridshell structure is not used very often. One of the reasons is
the difficult design process. When designing a gridshell, the major problem is developing the
grid geometry. If any structural analysis is to be performed on a structure, data is needed on
position of the joints and length of the elements. The distance between two nodes in a
gridshell is constant, so only the locations of the joints have to be calculated. The problem is
that there is no standard design tool which can generate such a grid on an arbitrary surface.
With former gridshell projects, the grid itself is was determined either by physical form
finding, used to design the Mannheim gridshell (Burkhardt et al. (eds) 1978), or by an
iterative form finding process, which involves numerous iterations to develop a smooth shape.
This was performed for the Weald & Downland gridshell (Harris & Kelly, 2002). The process is
based on dynamic relaxation techniques and mathematical expressions describing the starting
grid of the geometry. One can understand that setting up such a process can become quite
complicated. First, the form finding has to be implemented, which takes into account the
physical properties of the material. Second, the surface has to be described mathematically,
which can be quite difficult when it concerns a free form surface.

Making the gridshell design method better accessible would stimulate the use of the gridshell.
It would make research into the feasibility of these structures for arbitrary free form surface
designs easier and it would contribute to the usability of the method for a larger number of
engineers.
One way of increasing the method's accessibility is providing a grid generation tool based on
commercially available software. With this tool it should possible to generate a grid on an
arbitrary surface, so creating the geometry of the grid would no longer be a problem. For this
Master's thesis research it is attempted to create a first prototype of such a design tool. The
results will be analysed with structural analysis and physical modelling.

In this chapter, the set-up for the tool development will be displayed in Section 5.2.
Furthermore, the shape that will be used to test the design tool will be described and
analysed in Section 5.3.

81
A Design Tool for Timber Gridshells

5.2 Tool development set-up

The development of the grid generation tool consists of the following parts: First the tool
itself is developed. The tool development consists of the following parts:

• To develop and test the grid generation tool a shape is needed. The shape which is
used is analysed in Section 5.3.
• In Section 6.2 the grid generation technique which is used will be explained.
• The grid generation technique is implemented in a usable grid generation tool, based
on commercially available software. Section 6.4 describes the structure of the tool
that has been developed.
• The output of the tool should be usable in third party software for structural analysis.
The tool is set up to generate a text output with all relevant data, which can be
processed further.
• The tool is tested by generating a grid on test case surfaces. The main test case
surface is described in Section 5.3. Also an eggoid shaped surface is used. The result
are be used to review the generation method. The result can be found in Section
6.4.2.
• The generation tool is set up to perform a curvature check on the generated grid.
This curvature check is based on the maximum curve and torsion angles determined
in 3.2.4.1. This addition to the tool is described in Section 6.4.1.5. Results with the
curvature check can be found in Section 6.4.5.

The grid generation tool is reviewed by using the results in a few studies:

• A physical model is constructed with the case study shape to globally verify results. If
the geometry of the generated grid is correct, it should be possible to construct a
physical model with corresponding geometry. Also a conceptual model of the eggoid
is built Section 6.5 discusses the results of the physical modelling.
• A study is performed into the bending stresses that occur in a structure that is bent
into shape. First a single lath under different loads is analysed. Second, a 3D
structure is analysed. In these studies, it is tried to relate the bending curvature to
the stress levels that occur in the structure. The structural analysis software GSA25
will be used for this. This study is displayed in Chapter 7.

After the studies have been performed and the result analysed, the gridshell design tool will
be reviewed and conclusions can be drawn. In in Chapter 8, conclusions and
recommendations for enhancement of the tool will be presented and the Gridshell design tool
will be reviewed.

25
Oasys GSA© by Oasys Limited

82
 Grid generation tool for arbitrary surfaces

5.3 Shape analysis

The architectural design that will be used to test the design tool is shown in Figure 5.1 and
Figure 5.2. This model is a design for a large hall26. The dimensions of the building are
approximately 40x60 m. large enough for a variety of functions, such as a wellness centre, an
exhibition hall or a tennis court. The surface of the design consists of two clastic parts: a
(partial) sphere and a ruled surface, a hyperbolic entrance and two connecting anti-clastic
parts. The model is a poly surface, i.e. a surface built out of different elements, and can be
viewed and modified in the program Rhinoceros27. It is not a requirement to keep this shape
as it is. Modifications are allowed when delaying difficulties in the study process are expected
or experienced, or to optimize the shape structurally. However the concept of the shape is to
be maintained.

The shape used is not very challenging from a contemporary architectural point of view. The
shape is comparable with the shape of the Mannheim gridshell. However, from a structural
point of view it has interesting aspects such as the double curvature and the anti-clastic area,
which is rather flat. Analyzing the surface curvature, a plot of the Gaussian curvature can be
made. This plot shows in which extend the surface is smooth with continuous curvature.
Looking at Figure 5.3 some remarks can be made. As can be seen, there are some sudden
changes of colour, indicating rapid change of curvature or discontinuities in the surface. At
the edges of the different parts of the surface, disturbances are clearly visible. These
probably were created by the software’s routines when joining the different parts of the
surface together. In the right part of the structure there is also an unexpected change of
curvature takes place, indicated by the sharp change of colour. The discontinuities become
better visible when curvature plots of section AA’ and BB’ are made (Figure 5.4 and Figure
5.5). The discontinuities show themselves as sharp jumps in the graph.

The tubular entrance shows a colourful pallet, indicating a rapid change of curvature. The
rapid change to a much stronger curvature is likely to give problems from a structural point of
view. The laths of the grid have to be able to follow the curvature. Possibly the entrance can
be better implemented with an edge structure between the main structure and entrance.
Therefore this part of the surface is simplified for further analysis. The entrance is replaced
by a smooth surface as can be seen in Figure 5.6.

Figure 5.1: Rhino model perspective view Figure 5.2 Rhino model perspective view back

26
Model provided by Prof. Ir. L.A.G. Wagemans
27
Rhinoceros© NURBS modeller for Windows by R. McNeal & Associates, version 3.0 SR4

83
A Design Tool for Timber Gridshells

B’
Discontinuities

A A’

Figure 5.3: Gaussian curvature plotted on the surface

Figure 5.4: Curvature plot of section AA’ Figure 5.5: Curvature plot of section BB’

To avoid problems, the surface should be optimized to a smoother surface. Now the tubular
entrance is deleted from the surface, this becomes relatively easy to do in Rhino. First a
number of longitudinal sections have to be extracted from the surface. The surface can then
be recreated by lofting a surface trough the sections. The result is a more continuous NURBS
surface (Figure 5.8). This surface can be further smoothened by editing its control points.

An interesting question rises whether the gridshell method is also applicable on a more
modern design. In contemporary free form designs a development can be seen from dome
like shape like Mannheim, to a shape where the curve bends inward at the lower part of the
structure, to touch the ground at an angle larger than 90o (Figure 5.9 and Figure 5.10). The
shape bending inwards is a challenging shape from a structural point of view because it
results in large bending moments in the elbow of the structure. Apart from the structural
feasibility of a gridshell with such a shape, the question raises whether it is geometrically
possible to create such shape with the gridshell method. The laths have to be able to bend
and scissor enough to create the surface. In Section 6 it is attempted to physically model an
eggoid shape with a gridshell.

84
 Grid generation tool for arbitrary surfaces

Figure 5.6: Model without entrance Figure 5.7: Longitudinal sections

Figure 5.8: Lofted surface with Gaussian curvature analysis plot

Figure 5.9: Example of free form design: The BMW Figure 5.10: Example of free form design: WEB van
Bubble by Bernhard Franken, Frankfurt28 Noord Holland by Kas Oosterhuis (Boer & Oosterhuis,
2002)

28
http://www.e-architekt.cz/index.php?PId=1232&KatId=90 accessed 17-10-2006

85
A Design Tool for Timber Gridshells

86
 Development of the grid generation tool

6
Development of the grid
generation tool

6.1 Introduction

As stated before, there is no software was found which can generate a grid on a surface with
elements of equal length. Some standard grid generation methods available in software tools
like Rhinoceros are:

• Points by UV-coordinates (Figure 6.1)

• Point projection in z-direction (Figure 6.2)
• Line projection (Figure 6.3)
• Surface meshing into polygons (Figure 6.4)

As can be seen, none of these methods create a grid with equal mesh edge length. Projection
methods disregard any distance between points and simply create the projection of a point or
line by shifting them in z-direction and locating the intersection position with the surface.
Meshing techniques create the best approximation of the surface with a maximum allowed
polygon size, shape and/or number. Also UV projection does not take length into account, but
uses a parameter along the surface. As standard grid generation methods do not provide the
desired grid, a different approach is needed. One approach is creating the grid with a
graphical method by hand. This option is investigated here.

The chapter starts with displaying and explaining the proposed grid generation method on
which the design tool will be based in Section 6.2. In Section 6.3, the assumptions and
starting points for the development of the tool will be displayed. Next, the actual design tool
will be displayed and reviewed in Section 6.4. In this section the set-up of the scripts that are
created for the design tool will be explained. Also results are presented for the grid generated
on the test surface. In Section 6.5 these results will be compared with a physical model of the
structure.

87
A Design Tool for Timber Gridshells

Figure 6.1: Line grid by UV coordinates Figure 6.2: Points projected in Z-direction

Figure 6.3: Lines projected on a surface Figure 6.4: Meshed surface

88
 Development of the grid generation tool

6.2 Proposed method

When point 1, 2 and 3 are located on an arbitrary surface (Figure 6.5,) with equal distance x
to each other, a fourth point on the surface can be defined by locating the intersection point
between the surface and two spheres with radius x and their centre points at point 2 and 3.
With an intermediate step of creating section curves of the spheres and the surface, point 4 is
located on the intersection point of these curves. The distance from point 2 and 3 to point 4
is now equal to x.

Spheres

3
1 x
x
x 4

2 x

Section curves

Figure 6.5: Locating a point with equal distance to the two outer
points by aid of two spheres

By repeating this point location sequence until all points on the surface are located, a grid
suitable for a gridshell is created. It should be possible to perform this repetition of steps in a
3D modelling tool. By using a third party software package, it is possible to use the already
made features of the package to do this and there is no need to program the 3D model
yourself.

Two software applications are selected in which running the series of steps should be
possible. The package in which it was first attempted to create a grid-generation tool is
GenerativeComponents29 (GC). This is a parametric associative design system, where
standard features can be used to setup the series of steps. The parametric associative
character of the tool has high potential, allowing the user to adjust his design by changing
just a few parameters. However difficulties were encountered in an early stage of developing
the tool. The software was still a beta version at the time of use and no user’s guide was
available. Either a flawed set-up of the model or bugs in the software made it unable to
determine the desired surfaces and intersections. A different method might have been more
successful. The experiences with the development of the grid generation tool with GC and the
problems encountered are described in Appendix 1: Determination of the maximum bending
radius.

29
GenerativeComponents© by Bentley Systems, version 8.9.0.0 Beta

89
A Design Tool for Timber Gridshells

Second application used is Rhinoceros (Rhino). Rhino is a 3D NURBS modeller in which

complex 3D models can be created. Almost any 3D surface can be modelled. It is also
possible to run command scripts to execute series of commands automatically. Rhino
provides a plug-in for creating scripts, known as RhinoScript30. RhinoScript exposes the
internal working of Rhino, enabling the user to implement specific Rhino commands and it
provides support for VBScript. VBScript, which is short for Microsoft Visual Basic Scripting
Edition, is a variation on the Visual Basic programming language31.

For grid generation with the sphere method described above, a script will be used to execute
the process of locating the intersection points on the surface. The following steps will be
processed in Rhino by the script:

• Creating sections (Figure 6.6): the user is asked to create sections by picking points
in the top view. These sections are used as the base of the grid generation. The use
of these sections will be further elaborated in Section 6.3.
• Split sections: The grid generation will start at the intersection of the two sections.
Therefore these are split at the intersection point (point 1).
• Divide by length (Figure 6.7): The sections are divided by the specified length. This
length is equal to the desired mesh size.
• Create two spheres with radius equal to the mesh size at the first two points (point 2
and 3, Figure 6.8).
• Create intersection curves of the spheres with the surface (Figure 6.9).
• Create intersection point (point 4) of these two curves. This point will be one of the
next points to create the next intersection point (Figure 6.9).

By repeating this sequence of creating spheres and intersections at point 5 and 4, point 6 can
be located (Figure 6.10 and Figure 6.11). This sequence is repeated until all possible
intersection points are located on the surface. After this, the points can be connected, to
create the structure.

One difficulty to overcome in the scripting tool, is creating continuous laths. A gridshell
structure is built with continuous laths in two directions and the laths are connected together
at each intersection with free rotation around the node’s z-axis. When the structure is to be
implemented in a structural analysis program, the data on the geometry of the structure
should be correct, i.e. it must be known which element is connected to which node and this
must be correct for the entire structure.
The fact that rotation around the node’s z-axis must be accommodated in the model means
that the elements in different directions cannot be connected to the same node in the
computer model.
The separation of the elements and associated nodes is created by using the sections as the
main directions. First, all created nodes are copied. Two different sets of nodes are created
this way: the originals and the copies. The separation between the directions is created by
connecting the elements in one direction to one set of nodes and the elements in the other
direction to the other set of nodes. This is further elaborated in Section 6.4.1.2.

30
RhinoScript plugin help file
31
http://en.wikipedia.org/wiki/Vbscript accessed 19-02-2007

90
 Development of the grid generation tool

Figure 6.6: Creating sections Figure 6.7: Sections are divided by length

1
3
3

2
2 4

Figure 6.8: Spheres and intersections created Figure 6.9: intersection point located

3
2
4 4
5
5

Figure 6.10: Spheres are created at point 5 and 4 Figure 6.11: Point 6 is created at the intersection of the
intersection curves

91
A Design Tool for Timber Gridshells

6.3 Assumptions and starting points

For the grid generation technique to work properly, a few assumptions have to be made:

• Line elements between the points are created as straight lines. These lines should
follow the surface, as in real the timber is curved as well. The difference in length
between the curved element and the straight line is neglected, based on the fact that
radius of curvature >> element length (see Appendix 3: Proof of R >> mesh size).
• The sections from which the grid generation takes off have such a shape that it can
be approximated by curving and twisting a lath.
• Rhino is able to create and locate sections and section points with the necessary
accuracy.

The sections from which the generation takes off are necessary as basis for the generation.
The sections are used as basic directions for the grid and the first points will be located on
this curve. The correctness of the grid depends on the correctness of the sections. The curve
of the section should be curved in such manner that the lath is can follow this curve by
bending and twisting.
Rhino has the function "InterpCrvOnSrf", which interpolates a curve on a surface between or
through the desired points. It creates a curve running smoothly over the surface. It is
assumed that these curves approximate the needed section curves and are usable for the
grid generation.

Furthermore, the global and local axes of the members are shown in Figure 6.12. The local x-
axis of a member points in the direction of the member. Its z-axis is pointing upward, in the
positive direction of the global z-axis. The local y-axis is orthogonal to the local z- and x-axes

Figure 6.12: global and local axes of the members32

32
OASYS GSA helpfile

92
 Development of the grid generation tool

6.4 The gridshell design tool

In this section the gridshell design tool is described, along with the results. The gridshell
design tool is set up in such way that different sub-tools are used for different tasks. The
main tool is the grid generation tool itself, which generates the grid on the surface pointed by
the user. The script which is executed when running this tool is described in Section 6.4.1.
The results are presented in Section 6.4.2. Some of the problems encountered while
developing this script can be found in Appendix 4: Problems encountered in developing the
grid generation tool.
A second tool can be used to trim the structure to a desired height. The script for this tool is
described in Section 6.4.3. Finally, an output has to be generated when the generated grid is
to be used for further structural analysis. A third tool is created for this, which extracts all
information needed from the model. This can be found in Section 6.4.4.

Rhino provides the possibility to create a custom toolbars and buttons. A gridshell toolbar as
shown in Figure 6.13 can be created as a graphical representation of the gridshell design tool.
The three buttons represent the different tools.

Figure 6.13: Toolbar with buttons for the CreateGrid,

TrimStructure and ExportStructure tools

93
A Design Tool for Timber Gridshells

6.4.1 Grid generation script

In this section the process in the script is elaborated. Flow charts are used for this.
Figure 6.14 shows a flow chart of the script's main structure. A subroutine is indicated with a
diamond shape. The rectangles with rounded corners represent one or more commands that
are executed. A Boolean operation is represented by a six cornered polygon.
The main structure of the tool starts with an input subroutine, in which the user is asked for
the required input, after which the grid generation takes place. This consists of the
generation of the grid of points, connection of the points with lines and if desired the check of
the curvature and torsion angles. The different subroutines are discussed more detailed in the
next sections.
The result of the script is a series of points and elements. When the geometry is found
correct, a second script described in Section 6.4.4 can be executed to generate text output
which can be used in a structural analysis package.

Input

Grid
Generation

Create Grid Create Points Subroutine

Draw Lines Commands

Check Angles Boolean

Figure 6.14: Flow chart of the script's main structure

The script is created with a main loop, which enables the user to exit without error messages.
This is possible every time a message box appears. This happens when user input is required.
The user has the option to click "Cancel", which exits the main loop. By default it is also
possible to exit a script in Rhino by pressing the escape key. With this script this is needed
several times before the script exits entirely.

Figure 6.15: text box with options OK and CANCEL

94
 Development of the grid generation tool

6.4.1.1 Input
The grid generation tool starts with statements which ask the user for input. This sequence is
shown in the flow chart below:

Input

Select Surface

Create Sections

MeshSize

Check
Curvatures?
Yes No

Timber properties
No
Use
the entered
properties?

Grid
Yes
Generation

Figure 6.16: Flow chart input section

The input routine starts with asking the user to select the surface which will be used for grid
generation. After this the user is asked to create the sections. This Create Section subroutine
is shown in Figure 6.17. It exists of:

• Creating two sections

• Asking the user if the sections should be used. If NO is selected, new sections can be
created
• The sections are split at their intersection points. There is no RhinoScript command
for this, so the user is prompted to do this manually by following the command line
instructions. There are now four section parts.
• If a section's start point is below its end point the section is flipped. If this is omitted,
it can occur that the generation starts at the bottom of the structure. This will lead to
failure of the grid generation.

95
A Design Tool for Timber Gridshells

Create Sections

Create Section 1

Create Section 2

Use these
No
sections?
Yes

Split Sections

Sections
Z-end > Z-start?

Yes

Flip Sections

Figure 6.17: Create Sections subroutine

After creating the sections, the user is asked for the desired mesh size and if it is needed to
check the curvatures in text boxes (Figure 6.18). The latter is asked because checking the
curvatures consumes quite a lot of time. If YES is clicked, input is required on the material
properties of the timber. The following input is required:

• Height of the timber laths htimber;

• Width of the timber laths wtimber;
• E-modulus of the timber;
• Shear modulus of the timber;
• Bending strength of the timber;
• Shear strength of the timber;
• Modification factors kmod and γm.

For every input a text box appears, such as shown in Figure 6.19. When all material
properties are known, the maximum angles of curvature and torsion are calculated. The
angles of curvature determined in the curvature check have to be equal or smaller than these
values. The maximum angles are calculated according to Section 3.2.4.

Figure 6.18: Text box Check Curvatures? Figure 6.19: String input message box

96
 Development of the grid generation tool

6.4.1.2 Grid Generation

Figure 6.20 shows the flow chart of the grid generation sequence. The actual grid generation
is split into the subroutine "Create Grid" which generates the node points and the part which
connections of the nodes in different directions.

Grid
Generation

Create Grid

Copy nodes for

different directions

Connect nodes,
direction x:
(x,y) and (x+1,y) No
Check
curvature Check angles
= True? Yes

Connect copied
nodes, direction y:
For Section j=0 to 3 (x,y) and (x,y+1) No
Check
curvature Check angles
= True? Yes

For all copied nodes

Figure 6.20: Flow chart Grid Generation sequence

The global directions in which the elements of the grid are drawn are indicated by the
sections (Figure 6.21). The sections also divide the surface into quadrants. The grid
generation is executed quadrant for quadrant between two subsequent sections, starting with
section 0 and 1. First the subroutine "Create Grid" creates an array of points (Figure 6.22).
The subroutine "Create Grid" is described in the next section. When all nodes are created,
they are copied to make the separation of the lath directions possible. The local directions of
the grid are direction x and y. These are the directions of the sections used, which are section
0 and 1 in case of the first quadrant. Connecting the original nodes together, laths in
direction x are created (Figure 6.23). This corresponds with the global direction 1. After this
the laths in direction y are created (Figure 6.24), corresponding with global direction 2. These
are connected to a different set of nodes by connecting the copied nodes. All elements are
named after the nodes between which they are created to be able to reproduce this
information. If the curvature check is desired, which was asked in the input subroutine, the
subroutine "Check Angles" is executed for every created element.

97
A Design Tool for Timber Gridshells

2 Direction 2 1

Direction 1
3 0
Figure 6.21: Sections 0 to 3 and global directions Figure 6.22: First quadrant filled with points

1 1
Direction y

Direction x
0 0

Figure 6.23: Lines in direction x (1) are drawn between Figure 6.24: Lines in direction y (2) are drawn between
the original nodes the copied nodes

2 1 2 1

Direction x

Figure 6.25: Nodes between section 1 and 2 are created Figure 6.26: Lines in direction x (2) are drawn between
the copied nodes

Direction y

3 0

Figure 6.27: Lines in direction y (1) are drawn between Direction x Direction y
the original nodes
Figure 6.28: The final quadrant between section 3 and 0
is filled.

98
 Development of the grid generation tool

After the first quadrant is filled with nodes and elements, the quadrant between section 1 and
2 is processed. First the array of points is created (Figure 6.25). Because the process has
shifted to the next quadrant, the x-direction is determined by section 1 and the y-direction by
section 2. The quadrants have to connect to each other in the global directions in order to
create continuous laths over the entire structure. In the first quadrant the original nodes were
used to create the lines in direction x (global direction 1). The lines in direction y were
connected to the copied nodes for global direction 2. For the current quadrant connecting the
original nodes will create lines in the x-direction, which corresponds to global direction 2
(Figure 6.26). Creating lines with the copied nodes will create lines in the y-direction,
corresponding with global direction 1 (Figure 6.27).
To connect the first and second quadrant, the points located on the adjacent section (section
1 in this case) need to be the same points for both quadrants. To connect the laths in global
direction 2 (local direction x for the current quadrant), the points at the array location (x,0),
encircled in Figure 6.26, need to be the same points as the copied points created at the array
location (0,y) in the first quadrant, which are already connected to the lines in direction 2 in
that quadrant. Therefore these points are copied into the point array of the current quadrant.
An equal process is performed for the lines in direction 1. The original nodes at location (0,y)
in the first quadrant have to be copied to the copied nodes array of the current quadrant
(encircled in Figure 6.27).

The process of creating points, copying them and connecting them in the correct directions is
repeated until the entire structure is processed. In the last quadrant an extra copy operation
is performed. Not only the points located at the section adjacent to the previous quadrant
need to be copied, but also the points located at the section between the last and the first
quadrant.

When the process finishes, the lines in different directions are connected to different point
sets, enabling the user to reproduce the laths as continuous in the different directions for
structural analysis.

99
A Design Tool for Timber Gridshells

6.4.1.3 Create Grid

With the subroutine "Create Grid" the generation of the point grid is created. The process
starts with checking if it is the first time a quadrant is processed. If this is true, the first
section is divided by length MeshSize. If it is not the first quadrant the points located on the
sections are copied to connect the adjacent quadrants, which has been explained in the
previous section.

The generation of a point is processed by the subroutine "Create Points". This subroutine is
repeated until the final points on the sections are used. This results in an array of points with
the dimension xmax,ymax (Figure 6.29). Because this process does not fill the entire surface with
points, it has to be checked if there is another point possible at the position (xmax+1,i) or at
(j,ymax+1). This is checked by the following process.

Let's consider the check in direction y. A plane is created through points (j,ymax), (j,ymax-1) and
(j,ymax-2) (Figure 6.30 and Figure 6.31). This plane is scaled to be able to locate the
intersection point(s) with the sphere intersection curve at location (j,ymax) (Figure 6.32). If
there are two of these intersection points, the outer one is another grid point. If there is only
one intersection point, no next grid point is possible at this location and the check is
performed at location (j+1,ymax).
If a next grid point is located, the "Create Points" subroutine creates another row of points.
When all points at the edge in y direction are checked and no additional grid points are
located, the checking process is performed at the edge in x direction.
By using a plane trough the three edge points to locate the next point, a next element can be
created in the same plane as the preceding two elements. The new element continues the
curvature of those elements.

ymax

Direction y
(j,ymax)

(j,ymax-1)

(j,ymax-2)
xmax
Direction x

Figure 6.29: An array of dimension xmax,ymax Figure 6.30: Points for creating the plane

Intersection
points

Figure 6.31: A plane trough points (j,ymax), (j,ymax-1) and

Figure 6.32: Intersection points. The bottom one is point
(j,ymax-2) is created.
(j,ymax+1)

100
 Development of the grid generation tool

Create Grid

Yes
First quadrant?

No
Copy points (0,y)
Devide Section j by
of previous
MeshSize
quadrant to (x,0)

Devide Section i by Create intersection

MeshSize curves at (x,0)

CreatePoints

Create Plane at
edge y-direction

Intersect plane
with intersection Create Points
curve

Another point
Yes
possible?

3x No

Create plane at
edge x-direction

Intersect plane
with intersection Create Points
curve

Another point
Yes
possible?

No

Figure 6.33: Flow chart of the "Create Grid" subroutine

101
A Design Tool for Timber Gridshells

6.4.1.4 Create Points

The "Create Points" subroutine is the part which locates the points on the surface with equal
mesh distance. It executes the routine described in Section 6.2 in series, until all points on
the sections are used and a grid array of the dimension xmax,ymax is created. When a sphere
intersection curve is created, it is checked if the result of the intersection procedure is a
single object and if it is a curve. It is possible that the intersection procedure creates more
than one object (see Section 6.4.4). If the intersection is created with more than one curve,
the parts are joined.

Create Points

Create Sphere at
(x,y)

Intersect sphere
with surface
y = y+1
x=0 Is
No
Intersection a
single object?

Yes Join curve parts

does
No intersection
(x+1,y-1)
exits?
Yes

Intersect
intersection (x,y)
and (x+1,y-1)

No two points?

Yes
x = x+1
select correct point
as point (x+1,y)

Figure 6.34: Flow chart of the "Create Point" subroutine

102
 Development of the grid generation tool

The points are located row for row until the outer nodes on the section is processed. Before
two intersection curves are intersected to locate the grid points, it is checked if the next curve
on the previous row exists (location j+1,i-1). If this curve does not exist, the outer node is
reached and the process starts with the next row. If this curve exists, curve (j,i) and (j+1,i-1)
are intersected. If two intersection points are located, one of these points is the next grid
point. If there is only one point, the intersection curve (j+1,i-1) is located at an edge of the
surface, or something went wrong with the intersection process. In either case, the process
starts with a new row.

(j,i)

(j,i-1)
Intersection
curve (j,i)

Intersection
curve (j,i-1)
Figure 6.35: The grid is created row by row Figure 6.36: Intersection curve (j+1,i-1) does not exist.
The process starts with the next row

103
A Design Tool for Timber Gridshells

6.4.1.5 Check Angles

In the input section, the user was asked if the curvatures should be checked. If YES was
answered the angle of curvature in y and z direction and the torsion angle in x direction are
checked every time an element is added. The flow chart of the checking procedure is shown
in Figure 6.41 and explained in this section.

The angles are determined graphically in the Rhino model. To illustrate how, let's consider
two elements between the arbitrary points j-1, j and j+1 on a surface (Figure 6.37). Angle αy
of the curvature around the y-axis between element 1 and 2 is determined with the following
steps:

• At point j, the normal to the surface is created (Figure 6.38).

• A plane trough this normal and element 1 is created. This is directional plane of the
element. The tangent to the surface is created by creating a normal to the directional
plane (Figure 6.39)
• This tangent is copied to point j+1 and the intersection of his copy and the directional
plane is be located (Figure 6.40).
• The angle of curvature around the y-axis is now defined as the angle between
element 1 and the line between node j and the intersection node. This angle is called
α y.

Normal to surface
j-1
j-1
1

j
j
2
z
y
x z
j+1 y
j+1
x
Figure 6.37: Two elements under an angle with each Figure 6.38: Normal to the surface created
other

Directional surface

j-1 j-1

j
j

Tangent Tangent copy

z j+1
z
y j+1 y αy
x
x Intersection point

Figure 6.39: A directional plane trough element 1 and Figure 6.40: The angle αy is determined
normal created. The normal to this plane is tangent to the
surface

104
 Development of the grid generation tool

Check angles

Create normal at determine angle

point (j,i) alfa_z

Create directional Copy normal to

plane (j+1,i)

Create tangent at Intersect normal

point (j,i) with tangent plane

Create tangential determine angle

plane alfa_y

move normal of
Create normal
point to (j-1,i)
plane
to (j,i)

Copy tangent to determine angle

(j+1,i) alfa_x

Intersect tangent Do stresses

with dir plane exceed
citerium?

Yes

Assign colour
Figure 6.41: Flow chart of the "Check Angles" subroutine

Angle αz of the curvature around the z-axis is determined in a similar way (Figure 6.42):

• The tangential plane is created trough the tangent and element 1.

• The normal is copied to point j+1.
• The intersection point of the tangent plane and the normal copy is determined.
• The angle αz is determined as the angle between element 1 and the line between
point j and the intersection point.

j-1 Tangent plane

Intersection point
j

z j+1
αz
y
x Normal copy

Figure 6.42: Tangent plane and αz

105
A Design Tool for Timber Gridshells

The determination of the angle of torsion angle αx is determined by using the element created
previous to element 1: element -1, and the normal to the surface in point j-1:

• The normal -1 is copied to point j.

• Then the element -1 is copied to the end of the normal -1 copy.
• The plane trough the normal and tangent of point j is created. This is called the
perpendicular plane.
• The intersection of this copied element and the perpendicular plane is located
• The torsion angle is angle between the normal of point j and the line between this
intersection point and point j.

Element -1 copy αx
Normal -1

Normal -1 copy
j-1 j-1
Element -1

j j

Perpendicular
plane
z z
y y
j+1 j+1
x x
Figure 6.43: Element -1 and normal point j-1 are copied Figure 6.44: Determination of αx

Now all angles are known, it can be checked if the stresses due to the curvatures exceed the
stress criterion. The implemented criterion in the tool for the bending angles is the criterion
for bending in two directions described by equations 3.17 and 3.18 in Section 3.2.4.4:

σ m, y ,d σ m, z ,d
+ km ≤1 ( 3.17 )
f m, y ,d f m, z ,d

σ m, y ,d σ m, z ,d
km + ≤1 ( 3.18 )
f m, y ,d f m, z ,d

For this the bending stresses in different directions need to be calculated from the angles of
curvature. The following equations are used for this:

The bending radius R can be calculated with equation 3.6 according to Section 3.2.4.4.

Eh
R= ( 3.6 )
2 fm

When R is know, the current bending stress can be calculated by rewriting equation 3.6 as:

Eh
σm = ( 6.1 )
2R

106
 Development of the grid generation tool

R can also be calculated with:

1
R= ( 6.2 )
κ

In which curvature κ can be written as (see Appendix 1: Determination of the maximum

bending radius):

dθ
κ= ( 6.3 )
dx

When for dθ the angle of curvature α is used and for dx the mesh size L is used, σm can be
calculated with:

Ehα
σm = ( 6.4 )
2L

For torsion the check is performed by comparing the determined torsion angle with the
maximum torsion angle, which is calculated according to Section 3.2.4.3, equation 3.14:

Tmax L
α t ,max = ( 3.14 )
KG

If one of criteria in equation 3.17 and 3.18 is exceeded, or then the torsion angle between
two members is larger than the maximum torsion angle, a colour is assigned to the reviewed
element. This way, the curvatures of the grid are checked visually. The colours that are used
are:

• red when one of the bending criteria is exceeded;

• blue when the torsion criterion is exceeded;
• green when both the bending and torsion criteria are exceeded.

107
A Design Tool for Timber Gridshells

108
 Development of the grid generation tool

6.4.2 Grid generation results

The grid generation tool creates a graphical representation of the grid structure in Rhino. All
information needed for output to a structural analysis program is stored in the names of the
nodes and elements and the coordinates of the nodes, as described in Section 6.4.4. Figure
6.45 and Figure 6.46 show the surface and the used sections used in this generation. Figure
6.47 and Figure 6.48 show the result of a grid generation with mesh size 2,5m in top and
perspective view. With a mesh size 0,5m, which is a more realistic mesh size, the structure
shows a smooth surface.

Figure 6.45: Used section curves Figure 6.46: Perspective view

Figure 6.47: Result of the grid generation with mesh size Figure 6.48: Perspective view of the result of the grid
2,5m generation with mesh size 2,5m

Figure 6.49: Structure with a grid with mesh size 0,5m

109
A Design Tool for Timber Gridshells

In Section 5.3 it was stated that an eggoid shape will be analysed to test if this shape is
possible with a gridshell structure. The shape used for this analysis with the sections used is
shown in Figure 6.50 and Figure 6.51. Figure 6.52 to Figure 6.54 show the result of a grid
generation.

Figure 6.50: Eggoid shape top view

Figure 6.51: Eggoid shape perspective view

Figure 6.52: Results of grid generation with mesh size 1,5m

Figure 6.53: Results of grid generation with mesh size Figure 6.54: Perspective view of the eggoid
1,5m.

110
 Development of the grid generation tool

6.4.3 TrimStructure
The generation process does not create a grid which ends at the edge as it is now (Figure
6.55). The edge of the grid stays jagged because the script can not create an element if it is
not between two nodes with a distance of the mesh size to each other. Of course this can be
corrected by hand before exporting the structure for structural analysis, but this is not
desired. This will also erase information on grid geometry that is stored in the names of the
nodes and elements that will be trimmed or deleted. A solution to this problem is trimming
the generated grid by a horizontal plane. This can be simply performed by hand, but when
this operation is performed by a simple script, the elements and points can be named
automatically to the desired format to create a correct output file.
If trimming of the shape is not desired, the surface could be extended prior to grid generation
so that the grid can be trimmed back to the original shape. Figure 6.56 to Figure 6.59 show
trimmed versions of the case shape and the eggoid.

Figure 6.55: A rather jagged edge is created.

Figure 6.56: Trimmed structure top view Figure 6.57: Trimmed structure perspective view

Figure 6.58: Trimmed eggoid Figure 6.59: Trimmed eggoid perspective view

111
A Design Tool for Timber Gridshells

The trimming tool set-up is displayed in the flowchart below. The script starts with
determining which elements are intersecting with a horizontal plane at the desired trimming
height. When an element and the plane intersect, an intersection point is located and a new
line is created between the old start point and the intersection point.
The script is set up in such a way that the naming of the elements and points stays
unchanged. The newly created line and point are named after the old element and point. The
output script would give an erroneous output if this was skipped. After all elements are
checked, elements and points that are situated below the intersection plane are deleted.

TrimStructure

add object to
Select all objects
element array
No

Point object?

Yes
add object to point
Enter trimming heigth
array

create horizontal
section plane

move plane to
trimming heigth

Determine which
elements intersect
with section plane

Determine which
endpoint is below
section plane

Save point's and

element's name

intersect element
with section plane

add new line

rename intersection
point and new line to
old names

Delete objects below

section plane

Figure 6.60: flow chart of the TrimStructure script

112
 Development of the grid generation tool

6.4.4 Output Script

The output of the grid generation script is the graphical representation of the points and
elements, created in Rhinoceros. If it is desired to use this model in a structural analysis
program, probably a compatibility problem arises. It is not possible to load the file format of
Rhino is a different program. Rhinoceros provides an export feature, by which the model can
be exported to the file formats of other drawing software such as AutoCad and 3Dstudio. It is
also possible to create text files of the coordinates of points.
These export features exports the geometrical data of the model. The information that is
needed for structural analysis is however the geometry setup data of the model, i.e. between
which nodes which element is situated. This information is stored in the names of the
elements by the grid generation script. All points are named by the generation script as:

"point number; base x-coordinate in the flat mat; base y-coordinate in the flat mat"

The elements are named:

"begin point; end point"

To extract this information another script is developed. This script asks the user to select all
points and elements which have to be exported. The script sorts the selected objects in an
element array and a point array. The user is then asked to enter a save location for the file.
After this the file is opened and filled with the points names and coordinates and the element
names.

The values are separated by a semicolon. The saved file is a *.txt file and has the following
lay-out:

Point no. Base X Base Y X Y Z

… … … … … …
… … … … … …
Element Start End point
no. point
… … …
… … …

The file can be opened with a spread sheet program to further process the data.
Unfortunately the data has to be ordered in the spread sheet. The nodes are not in ascending
order and there are nodes without a number. These nodes are not part of the structure and
can be neglected.
Also the node numbering needs to be restructured. The nodes are numbered by the original
numbering of the script. When the model is edited, e.g. trimmed, some nodes have been
deleted and new node created. When the node numbering is not restructured and copied into
GSA, this software will mix up the structure because the program renumbers the nodes, but
the start and end nodes of the elements are copied in the analysis program unchanged. For
example when the structure considered in Figure 6.62 is trimmed, four new nodes were
created. From the trimmed structure an output is created by the output script. A list of nodes
1 to 6 is produced which are numbered {1,2,6,7,8,9} and a list of element start and end
nodes. When this list is copied into GSA, this program will just number the nodes as
{1,2,3,4,5,6} but not renumbers the elements’ start and end nodes. The result is that only
the element between node 1 and 6 is recognized.

113
A Design Tool for Timber Gridshells

ExportGrid

add object to
Select all objects
element array
No

Point object?

Yes
Enter Save
add object to point
Location and file
array
name

Open File

Write to file: name

+ point coordinate
of point i

For all points

Write to file: name

of element i

For all elements

Close file

Figure 6.61: Flow chart of the Export script

1 2 1 2 1 2

6 7 8 9 3 4 5 6

Interpretation by structural
3 4 5 Trimmed structure analysis program
Original structure
Figure 6.62: Trimmed structure and the interpretation by GSA

Nodes Elements
original numbering Renumbered as start node End node
1 1 1 6
2 2 1 7
6 3 2 8
7 4 2 9
8 5
9 6

114
 Development of the grid generation tool

6.4.5 Grid generation with visual curve analysis

The curvature check can be executed as an option during grid generation. As stated in
Section 6.4.1.5, the maximum bending stress criteria found in Section 3.2.4.4 are
implemented in the grid generation tool to check the curvature of the grid members. When
the maximum bending stress criteria or torsion angle criterion is exceeded, the member will
change colour. The result is a generated grid with visual information on the local curvature of
the laths. If the criteria are exceeded anywhere, it is necessary to adjust the structure.
This adjustment can be made in a few ways. First, a smaller element height or width of the
timber elements can be chosen, making a smaller radius of curvature possible. Second,
timber with a smaller E / fm ratio can be selected (see Section 3.2.4.1). Another option is to
adjust the shape of the surface to a larger curvature, if this is allowed by the architect. One
could say this process of adjusting the shape to the results of the grid generation is a form of
backward form finding.

With the case shape, the design tool is tested with curvature check implemented. Figure 6.63
shows a grid of mesh size 2.5m, which is generated on the case shape. An element cross
section of 50x50mm is used. For the modification factors kmod = 0,8 and γm = 1,3 is used. The
heavy printed elements are elements in which one of the bending criteria is exceeded. The
dotted elements are elements where both torsion and bending criteria are exceeded. As can
be seen, the radius curvature is exceeded in a large part of the surface. A modification of the
structure or surface is thus needed to fulfil the demands of curvature.
First modification to the structure is reduction of the element’s height. A grid is generated
with cross section 35x50mm, which is shown in Figure 6.64. Still, in quite some elements, the
maximum bending criteria are exceeded. Another option is to change the shape of the
structure to a larger bending radius. For the grid in Figure 6.65, the curvature in the waist of
the shape is lessened, resulting in a better fitting grid. One could also use different start-off
sections. The sections used in Figure 6.66 have less (change of) curvature than the previous
ones, resulting in a grid in which the elements are less curved.

There are still areas in the structure in Figure 6.66 where the laths of the grid cannot comply
with the curvature created by the design tool. Main problem is found in the left dome, of
which the curvature is the larges. To overcome this problem, the curvature of this part of the
surface is decreased by scaling that part of the structure (Figure 6.67). One could also flatten
the part of the surface to decrease the overall curvature in this area, but in this case the
shape of the left and right part of the surface would become alike and less interesting in an
architectural point of view.

The grid that is created by the design tool after this modification still shows areas in which
the bending criteria are exceeded. The problem is concentrated at the ends of the structure.
It seems to be problematic to create a gridshell structure for surfaces like this; that is a
surface which is (semi-)spherical. The laths of the structure need to scissor and bend largely
to create such surface. Additional measures such as reducing the elements’ cross section in
this area are needed to be able accommodate the curvature.

115
A Design Tool for Timber Gridshells

Figure 6.63: Visual curve analysis. The dark members Figure 6.64: Grid generated with a smaller element
indicate the maximum curvature is exceeded height, 35mm

Figure 6.65: Shape with less curved waist Figure 6.66: Different start-off sections

Figure 6.67: Change of the shape of the smaller dome Figure 6.68: Grid generated on the shape with smaller
(wire frame view of the shapes) curvatures

116
 Development of the grid generation tool

6.5 Results compared with reality by physical modelling

To compare the results of the grid generation tool with reality, a physical model has been
constructed. This model has been created from an initially flat mat of bending members to
model the shaping process during construction. The process of creating this model and an
extensive comparison with the computer model can be found in Appendix 5: Physical . The
result of the physical modelling is the structure shown in Figure 6.69 to Figure 6.72. Although
the physical model does not resemble the computer model exactly, both models give an equal
image of the structure. Similar effects in bending and scissoring of the laths have been found.

Figure 6.69: 3D view of the created model

Figure 6.70: Physical model in top view.

Figure 6.71: Left dome of the physical model in side Figure 6.72: Right dome of the physical model in side
view view

117
A Design Tool for Timber Gridshells

With the grid generation tool it was found that bending curvatures become too large when
the laths of the grid are bent to a (semi-)spherical surface. This problem was also found in
the physical modelling process. Especially the right dome of the structure shows less
scissoring and curvature in the laths as intended by the computer model (Figure 6.73).
Without additional force, the structure would not achieve a position equal to the computer
model. Large force was needed to pull the structure into a position, creating tension in parts
of the structure (Figure 6.74). This effect can be compared with the tensioning of the fabric
of a hyperbolic tent structure. Together with the results of the grid generation tool, this
results show that it is hard to create a (semi-)spherical surface with a gridshell structure. It is
therefore advisable to avoid such shapes when a gridshell structure is applied.

Figure 6.73: Comparison of the right dome in the two models. Moving the members in the
pointed directions by force (arrows in top right) will result in a better approximation of the
computer model. Some laths are printed bold for comparison. The dark and bold lines
indicate the physical model. Red lines indicate the computer model.

Figure 6.74: Structure, with the nodes at top right side moved in the direction of the arrow.
The models resemble better, but tension exists in the encircled area.

118
 Development of the grid generation tool

Deviations between the two models have been found in the shape of the structure. First, the
two models differ in height. The physical model has larger height than the computer model.
This can be explained by the fact that the scissoring of the laths at the ends of the structure
is less than in the computer model. The laths in the physical model have the same length as
in the computer model however. This results in a larger height, as the laths are less curved
than intended.
Second, the cross sections of the physical model show a parabolic chape, whereas the cross
sections of the computer model show a more circular shape. These deviations between the
two models can be well explained. The parabolic shape is generally considered as a more
natural shape to divert axial forces trough an arched structure. The fact that the models
deviate does show that the computer model does not represent a shape that is optimized for
structural behaviour. The model should be of a more parabolic shape.
Building the physical model is a type of physical form finding. Therefore the results of the
physical model can be used to modify the computer model to a more optimized shape. One
could say this is a reversed form finding process. In stead of first creating a shape by form
finding and then creating a structure, here the structure is created first, using the grid
generation script. The shape is then adjusted to a shape corresponding with the physical
model. A new grid can then be generated on the adjusted, optimized shape for structural
analysis.

B C

Section AA'

D
D'

Section BB' B'

A'
C'

Section CC' Section DD'

= Physical model = Computer model

Figure 6.75: Sections of the two models

119
A Design Tool for Timber Gridshells

120
 Members in bending

7
Members in bending

7.1 Introduction

The shape of a gridshell is created by applying a large deflection on the members. The laths
are bent and twisted to create the desired shape with the initial flat mat of laths. The
properties of timber make it possible to bend and twist the laths into shape, but also put
restrictions on the shape possibilities. The material will fail when the ultimate stress capacity
is exceeded. By bending and twisting of the lath, stresses are induced in the material. After
construction, the formation stresses will diminish due to relaxation and approximately half of
the initial stress level will remain present. On top of these residual stresses, the structure has
to withstand loads like wind and snow, which desires a certain load capacity of the laths after
construction. To be able to understand the behaviour of the structure and to be able to
structurally analyse the gridshell structure, it is important to fully understand the behaviour of
members. This is investigated in this section.

First, the question rises which stress distributions occur in a member during construction. In
the form shaping sequence, the member is first bent into shape on internal supports. After
pinning the structure down at the ends, the internal supports are removed. The stress
distribution in the laths will probably change and thus will the shape of the structure. This
behaviour will be investigated in Section 7.2 by analyzing a simple beam on two supports.
This investigation will be extended to a 3D structure in Section 7.3.

For structural analysis, it is desired to take into account the formation stresses. In Section 7.4
a method to do this will be tested on a simple beam and a 3D structure. This method is based
on the assumption that the bending stresses can be deduced from the angles of curvature in
the members of the structure.

121
A Design Tool for Timber Gridshells

7.2 Stresses induced by the formation process

When a gridshell structure is erected, the members are deflected into shape by self weight
and applied load. The members are supported at several points along the span. Due to the
applied displacement, stresses are induced. When the shape forming process is ended, the
structure is pinned or clamped at the edges and the temporary supports are removed. If the
displacements applied are equal to the natural bending shape, the removal of these supports
will not cause any changes. This natural bending shape is the equilibrium position which is
adopted by a member by moving the ends to each other. When this is not the case, the
internal stress distribution is not in equilibrium after the internal supports are removed. The
beam will deflect to this equilibrium position and redistribution of the internal stresses will
occur. This can be seen as a relaxation reaction of the beam. Therefore, this deflection is
called the relaxation deflection. This change of shape is undesired, because it can cause
unexpected and undesired change of geometry and internal stress distribution.

To understand this change of system and to be able to predict the behaviour of the structure,
the different steps of the construction process are analysed. These steps are displayed in a
flow chart, shown in Figure 7.1. The following steps can be distinguished:

• The first step is to lay out the laths in a flat mat on the internal supports.
• When the supports are moved, the mat will first deflect under self weight.
• The deflection by self weight only is probably not enough to create the desired
shape, so force is needed to push or pull the mat into shape.
• During the formation process, the shape of the structure should be reviewed. When
the resulting shape complies with the desired shape within acceptable boundaries,
the structure can be pinned or clamped to the final supports. If not, additional forces
are needed.
• After the formation process is ended, the internal stresses induced by the
deformation should be estimated.
• It should be determined if the internal supports are removed immediately, or if these
are left in place for a period of time. In case of the former, the internal stresses are
probably not in equilibrium. The structure will deflect to an equilibrium position and
redistribution of stresses takes place. In case of the latter, the internal stresses due
to the formation process diminish in a certain extend due to relaxation of the timber,
prior to the removal of the internal supports. The relaxation deflection will be smaller
in this case.
• The internal supports are removed
• The relaxation deflection and corresponding stresses can be estimated.
• The structure should be re-analysed in equilibrium position.

To be able to predict the final result of the formation process, the amount of relaxation
during construction should be estimated. Also the internal stress distribution after completion
of the formation process and in the different steps should be estimated. To be able to make
this estimation, the different steps are analysed in the next section.

122
 Members in bending

laths are placed in flat

mat

mat deflects under self

weight

calculate force needed

to bend structure into
desired shape

deflection by force

resulting shape
complies to desired No
shape?

Yes

Pin structure to
supports

Estimate internal
stresses

direct removal of No
relaxation takes place
internal supports?

estimate time before

Yes internal supports are
removed

remove internal
estimate relaxation
supports

estimate deflection and

stresses

re-analyse the structure

in equilibrium position

Figure 7.1: Flow chart for analysis of the construction phase

123
A Design Tool for Timber Gridshells

The different steps of the construction phase are analysed by studying the bending behaviour
of a simple beam. This beam is a timber lath is used, 10m length with a cross section of
50x50mm and timber properties of strength class D30, supported by two pendulum columns
which act as internal supports.

The formation process is modeled in the following steps (see Figure 7.2):
• The lath is first supported by two pendulum bars at 2,5 and 7,5m. For stability the
lath is supported by a vertical roll at mid span.
• The lath is deformed by a vertical and/or a horizontal force at the end nodes (step 1)
• The end points are pinned at the deformed position (step 2)
• The pendulums are deleted. The lath will deflect to an equilibrium position (step 3)

These steps will be analysed in the following load cases:

• Deflection induced by vertical force at the end nodes (load A)
• Deflection induced by horizontal force at the end nodes (load B)
• Deflection induced by a combination of horizontal and vertical forces in different
ratios (load C)
• Deflection induced by horizontal and/or vertical force at the end nodes, combined
with self weight (load SW)

The different cases are named with a step letter and a load number, e.g. case A3 is the case
with vertical load and the pendulums removed.

The analysis performed is non-linear to take into account the geometrical non-linear
behaviour of the structure. A linear analysis is based on the assumption that deflections stay
small. Only deflection in z direction is accounted for and displacements in x-direction stay
zero. When large deflections occur, which is the case in this analysis, this assumption is no
longer valid. Therefore non-linear analysis is needed.

The analysis solver for non-linear analysis in GSA is based on Dynamic Relaxation (see
Section 4.2.2.2). The non-linear analysis performed by GSA is based on two different
effects33:

• Geometric non-linearity, where the load causes large deflections which must be taken
into account in order to get an accurate solution.
• Material non-linearity, where the load causes material to behave in a non-linear
manner, typically through yielding.

The latter is not considered here, as linear behaviour of the material itself is assumed.

33
Help file of OASYS GSA

124
 Members in bending

Structure

F F

Step 1

Step 2

Step 3

Figure 7.2: The structure is deformed in different steps

125
A Design Tool for Timber Gridshells

7.2.1 Load case A1

For the first load case, the slender beam is bent into shape by applying a vertical force of 100
N on both ends of the beam (Figure 7.3). Due to this load, the beam deflects to the shape
shown in Figure 7.4, with a total deflection of 541mm in vertical direction. The corresponding
moment, shear and axial force diagrams are shown in Figure 7.5 to Figure 7.7.

100 N 100 N

Figure 7.3: Case A1, applied force

uz = 148 mm

uz = -393 mm Uz = -393 mm
ux = 3,73 mm Ux = -3,73 mm

Figure 7.4: Case A1, deflection

0.247 kNm

Figure 7.5: Case A1, moment line

-98.4 N
-0.01 N

98.4 N

Figure 7.6: Case A1, shear force

17.4 N 17.4 N

-0.6 N

Figure 7.7: Case A1, axial force

126
 Members in bending

7.2.2 Load case A2

When the structure has reached its desired shape, it is pinned to the edge supports. The
deflection of the beam in case A1 is considered as this desired shape. Thus, the beam of A1
can be pinned down at this deformed position. In GSA this is implemented as an applied
displacement of the end nodes, equal to the deflection of the end nodes in case A1. It is
found that when the structure is pinned at the deformed position, the deflection stays equal
to the case where the structure is forced down. Because all boundary conditions are equal,
also the internal force distribution does not change.

uz = 148 mm

uz = -393 mm uz = -393 mm
ux = 3,73 mm ux = -3,73 mm

Figure 7.8: Case A2, deformation

0.247 kNm

Figure 7.9: Case A2, moment line

-98.2 N

-0.1 N

98.2 N

Figure 7.10: Case A2, shear force

17.1 N 17.1 N

-1.2 N

Figure 7.11: Case A2, axial force

127
A Design Tool for Timber Gridshells

7.2.3 Load case A3

The following step in the formation process is removing the internal supports. With this
structure, this implies removing the pendulum supports. The deflection of the end nodes is
kept equal and the beam can be re-analysed. This can be seen as the structure relaxing to an
equilibrium position. It is found that the nodes where the pendulums were attached deflect
down and the top deflects up. The bent member shifts to a more parabolic shape. This
change is also seen in the moment line. The discontinuities in the moment line are
disappeared, which results in a more energy efficient stress distribution. The axial force
increases from approximately zero to approximately -515N (compression).

uz = 155 mm

uz = -393 mm uz = -393 mm
ux = 3,73 mm ux = -3,73 mm

Figure 7.12: Case A3, deformation

0.284 kNm

Figure 7.13: Case A3, moment Line

-87.4 N

87.4 N

Figure 7.14: Case A3, shear force

-507 N -515 N -507 N

Figure 7.15: Case A3, axial force

128
 Members in bending

7.2.4 Load case B1

To analyse the effect of a horizontal deformation force, this is analysed in this case. In the
former load case an axial force of 515 N was found after removing the pendulums. This load
is taken as the deformation force in this case. The first step in the formation process is again
deforming the structure, supported by the two pendulums. A horizontal load of 515N is
applied on both ends of the member. To induce deflection in the right direction, a vertical
force of 1 N is applied on the middle. It is found that the horizontal force gives a deformation
approximately equal to the final step of load case A, e.g. the natural bending deflection of the
beam.

515 N 515 N

1N

Figure 7.16: Case B1, applied forces

uz = 164 mm

uz = -393 mm uz = -393 mm
ux = 3,85 mm ux = -3,85 mm

Figure 7.17: Case B1, deformation

0.288 kNm

Figure 7.18: Case B1, moment line

-89.5 N

89.5 N

Figure 7.19: Case B1, shear force

-507 N -515 N -507 N

Figure 7.20: Case B1, axial force

129
A Design Tool for Timber Gridshells

7.2.5 Load case B2

The structure of case B1 is again pinned down at its deformed position. The deformation and
moment line are displayed in the figures below. It is found that the deformation and internal
force distribution is again equal to step 1, where the structure is pushed down by force.

uz = 164 mm

uz = -393 mm uz = -393 mm
ux = 3,85 mm ux = -3,85 mm

Figure 7.21: Case B2, deformation

0.288 kNm

Figure 7.22: Case B2, moment line

-89.4 N

89.4 N

Figure 7.23: Case B2, shear force

-512 N -520 N -512 N

Figure 7.24: Case B2, axial force

130
 Members in bending

7.2.6 Load case B3

After the structure has been pinned down, the pendulums are removed. The deformation and
moment line are displayed in the figures below. It is found that the deformation and internal
force distribution is equal to the begin situation, where the structure is deformed by a
horizontal force and supported by pendulums. In this first step, the member already deflects
to the natural equilibrium bending shape. Pushing a member into shape with a horizontal
force thus approximates the natural bending shape of the pinned member

uz = 164 mm

uz = -393 mm uz = -393 mm
ux = 3,85 mm ux = -3,85 mm

Figure 7.25: Case B3, deformation

288 kNm

Figure 7.26: Case B3, moment line

-89.7 N

89.7 N

Figure 7.27: Case B3, shear force

-507 N -515 N -507 N

Figure 7.28: Case B3, axial force

131
A Design Tool for Timber Gridshells

To compare load cases A and B, the differences in deflection and stress distribution are
investigated. First the different steps of analysis are reviewed. Figure 7.29 shows the
difference in deflection between case A1, A2 and A3. This difference is the relaxation
deflection. As can be seen pinning down the structure after pushing down (A2 - A1) has
almost no effect on the deformation. Removing the pendulums does have an effect (A3 –
A2). The nodes between node 8 and 14 shift up, the other nodes shift down. The maximum
difference in deformation occurs at nodes 5, 11 and 17. This difference is -2,1% for node 5
and 17. In the middle this is 1,33%.

10
Δu (mm)

2 A2 - A1

0 A3 - A2
0 5 10 15 20
-2

-4

-6

-8
Node number

Figure 7.29: Relaxation deflection for load case A

As an example, if the structure should be 8m high, the difference after removing the internal
supports is 168 mm down at ¼ and ¾ of the span and approximately 106 mm up in the
middle. This is not a very large difference but it should be accounted for, e.g. when designing
the facade.

When we compare case B1, B2 and B3, it is found that the difference between deformations
is approximately zero. This shows that when a horizontal force is used, the deformation
approximates the natural deflection. However, a much larger force is needed to induce the
deformation.

It was fount that when a vertical force is used, the geometry changes after removing the
internal supports, although the change is not large. When the internal stress distribution is
investigated, a more important effect is found. Figure 7.30 shows the moment stress
distribution of the upper edge of the beams of the different cases. From case A1 to A3, the
internal moment stress relaxes from a discontinuous moment to a continuous distribution.
Going from B2 to B3 no difference can be seen. Figure 7.31 shows the change as a
percentage of the deformation. A change of 23,8% is found for nodes 6 and 16, which are
the nodes where the pendulum columns were attached. The change is 12,9% for the middle
node, node 11. To avoid breakage of the laths these changes should be taken into account,
especially when the laths are bent close to their ultimate bending strength.

132
 Members in bending

σm (N/mm2) 16

14

12

10 A1
A3
8
B1
6 B3

0
0 5 10 15 20
Node number

Figure 7.30: Bending stresses in the upper edge

15
Δσ (%)

10

0
0 5 10 15 20
-5 A3 - A1

-10 B3 - B1

-15

-20

-25

-30
Node number
Figure 7.31: Bending stress difference in the upper edge(%)

It should also be noted that the axial stress increases going from A1 to A3, from
approximately zero for case A1 to 0,206 for A3. Going from B1 to B3 however, has
approximately no effect on the axial stress. In these cases, the axial stress of 0,206 N/mm2 is
already present due to the horizontal force.

As shown in case A2 and B2, approximately no difference occurs between step 1 and 2
(pushing or pinning the structure down). Therefore, this step is skipped in the following
cases.

133
A Design Tool for Timber Gridshells

7.2.7 Load cases C1 and C3:

In load case A and B it is found that a horizontal force gives a much better approximation of
the equilibrium bending shape of a member, than when a vertical load is applied. This
suggests only horizontal force should be used to deform a grid shell. However, a much larger
force is needed to reach the same deformation as with vertical forces only, which complicates
the formation process. When a horizontal force is combined with a vertical force, it might be
possible to reach a better approximation of the equilibrium bending shape than with a vertical
force only and with less force than with a horizontal force only. This is tested by applying a
load of 100 N vertical and 100 N horizontal on the beam ends (Figure 7.32).

First, the member supported by pendulums is deflected into shape. This gives the deflection
and internal force distribution shown in Figure 7.33 to Figure 7.36.

Fz = -100N Fz = -100N
Fx = 100N Fx = -100N

Figure 7.32: Case C1, deformation

uz = 185 mm

uz = -482 mm uz = -482 mm
ux = 5,65 mm ux = -5,65 mm

Figure 7.33: Case C1, deformation

0.294 kNm 0.312 kNm 0.294 kNm

Figure 7.34: Case C1, moment line

-119 N

-13.3 N

13.3 N

119 N

Figure 7.35: Case C1, shear force

134
 Members in bending

-76.1 N -101 N -76.1 N

Figure 7.36: Case C1, axial force

After the deformation by force, the structure is pinned down (step 2, which is omitted here)
and the pendulums are removed (step 3). The deformation and internal force distribution
become:

uz = 192 mm

uz = 482 mm uz = 482 mm
ux = 5,65 mm ux = 5,65 mm

Figure 7.37: Case C3, deformation

0.349 kNm

Figure 7.38: Case C3, moment line

-107 N

107 N

Figure 7.39: Case C3, shear force

-503 N -515 N -503 N

Figure 7.40: Case C3, axial force

Before the pendulums are removed, the total deflection between end points and the middle
of the beam was 667mm. When the pendulums are removed, the middle of the beam deflects
7mm upward, 1,10% of the total deflection. In load case A (vertical force only) the relaxation
deflection was 1,29% of the total deflection. For bending stress, the change when the
pendulums are removed is also smaller: 19,1% in stead of 23,8% with vertical force only.

135
A Design Tool for Timber Gridshells

In load case C the ratio vertical force over horizontal force (v/h) is 1. To investigate the
influence of the vertical force on the end nodes, the following cases are investigated, with
and without pendulums (step 1 and 3):

• C1a and C3a: 150 N vertical, 200 horizontal force (v/h = 0,75)
• C1b and C3b: 100 N vertical, 200 N horizontal (v/h = 0,5)
• C1c and C3c: 75 N vertical, 300 N horizontal (v/h = 0,25)

To give an overview, the results of the deformation are put together in Figure 7.41

400
u (mm)

200

0 C1a
0 5 10 15 20 C1b
C1c
-200
C3a
C3b
-400
C3c

-600

-800 Node number

Figure 7.41: Deformation of the different cases

In the figure above the deflection of the cases with and without pendulums is plotted. The
differences between step 1 and 3 are hard to see in this figure. To analyse the spring-back
deflection better this difference plotted in Figure 7.42.

10
Δu (mm)

2
C3a-C1a
0 C3b-C1b
0 5 10 15 20 C3c-C1c
-2

-4

-6

-8

-10
Node number

Figure 7.42: relaxation deflection load case C

136
 Members in bending

7.2.8 Discussion increasing diagonal force

In the table below, the deflection results of the analysis with increasing v/h ratio are put
together. This is data on deflection step 3, with the pendulums removed. To indicate the
difference between the different load cases, the differences between case 3 and 1 are
displayed in the same table. The change of deflection Δu is indicated for the nodes with
maximum change: node 6, 11 and 16.

To analyse the relaxation deflection, this difference between the case with and without
pendulums is plotted in Figure 7.43 as a percentage of the total deflection. The percentage is
used because the total deflection is not equal for all cases. As can be seen, the relaxation
deflection becomes relatively smaller when the horizontal force applied is made larger, going
from C3 to C3a, C3b and C3c . The maximum change in percentage occurs at the nodes close
to the supports. Because the absolute change is small for these nodes, this maximum is
disregarded.

The ratio utot/F shown in Table 7.1 indicates the deflection in millimetres that is caused by 1 N
of applied force. This ratio is the larges when only a vertical force is applied (case A3).
Downside is the fact that also the relaxation deflection and change in bending stress is the
larges (see Figure 7.43). Ratio utot/F is the smallest for case B3 with horizontal force only. In
this case the relaxation deflection is almost zero, but for approximately every mm of
deformation 1 N of force is needed.

Table 7.1: Differences in deformation between cases with and without pendulums
Case F (N) v/h utot utot/F Δu (mm) node Δu (mm) node 6
(mm) (mm/N) 11 and 16
A3 100 v. - 548 5,48 7,1 (1,29%) -6,1 (1,22%)
C3 100 v 1 674 3,37 7,4 (1,10 %) -5,8 (1,22%)
100 h
C3a 120 v 0,75 923 3,30 8,8 (1,0%) -6,8 (1,04%)
160 h.
C3b 100 v. 0,5 871 2,90 7,5 (0,86%) -5,7 (0,9%)
200 h.
C3c 75 v. 0,25 942 2,51 5,6 (0,59%) -4,4 (0,66%)
300 h.
B3 515 h. 0 557 1,08 ~0 ~0

137
A Design Tool for Timber Gridshells

1.5
Δu (%)

0.5

0
C3 - C1
0 5 10 15 20
C3a - C1a
-0.5
C3b - C1b

-1 C3c - C1c

-1.5

-2

-2.5
Node number

Figure 7.43: Percentage difference in deflection between pendulum cases and relaxed cases

The same effect can be seen with the bending stress. The bending stress in the upper edge is
displayed in Table 7.2 for the different cases. Increasing the ratio v/h causes an increase in
bending stress. The difference between the cases with and without pendulums (step 1 and 3)
are displayed in Figure 7.44.

Table 7.2: Differences in bending stress between cases with and

without pendulums
Case F (N) v/h Δσm (%) Δσm (%)
Node 11 Node 6 and 16
A3 100 v. - 12,9 -23,8
C3 100 v 1 10,5 -19,1
100 h
C3a 120 v 0,75 9,09 -16,6
160 h.
C3b 100 v. 0,5 8,14 -14,8
200 h.
C3c 75 v. 0,25 5,75 -10,3
300 h.
B3 515 h. 0 0,24 0,25

138
 Members in bending

15
Δσm (%)

10

0 A3-A1
0 5 10 15 20
-5 C3-C1
C3a-C1a
-10 C3b-C1b

-15 C3c-C1C

-20

-25

-30

Figure 7.44: Difference in bending stresses in the upper edge (%)

The increase of the bending stress is the most important to take into account. Especially
when the laths are bent close to their ultimate bending strength, the sudden increase can
cause breakage of the laths.
When a choice has to be made on which forces should be applied to induce deformation, it is
advisable to use a combination of horizontal and vertical force. When for example a ratio v/h
of 0,5 is used, the ratio utot/F is still profitable and the spring-back deflection is already small.

The change in bending stress has to be taken into account in the design to avoid breakage.
However in the test cases only two supports are used. In real, the number of supports will be
larger and the shaping will be performed in a controlled manner. Therefore the effect of the
increase of stresses will be less for the maxima found in this analysis.

139
A Design Tool for Timber Gridshells

7.2.9 Analyzing the effect of self weight

To analyse the effect of self weight on the deflection, a few cases are analysed with self
weight included. First the deflection of the beam by self weigh only and the moment, shear
and axial force diagrams are shown in Figure 7.46 to Figure 7.49.

qz = 13,25*10-3 kN/m

Figure 7.45: Case SW, applied load

uz = 4.2 mm

uz = -28,9 mm uz = -28,9 mm
ux = 0,18 mm ux = -0,18 mm

Figure 7.46: Case SW, deformation (x25)

0.041 kNm 0.041 kNm

Figure 7.47: Case SW, moment line

-32 N -32 N

32 N 32 N

Figure 7.48: Case SW, shear force

0.24 N 0.24 N

-0.15N -0.15N

Figure 7.49: Case SW, axial force

140
 Members in bending

Load cases A, B and C (vertical, horizontal and diagonal load) are analysed with self weight
included to study the effect of the self weight on deflection and stress distribution. The
stresses that result from the deformation by self weight are added to the stresses induced by
the applied loads. Effect of the self weight is found the most prominent in stress distribution.
This will be analysed in Table 7.3. The bending stress diagrams of the different cases can be
found in Figure 7.50 to Figure 7.52. In these figures the bending stress in the upper edge is
displayed for both the first step (supported by pendulums) and step 3 (pendulums are
deleted). The deformation in the different cases increases due to the self weight, although
the change is small. The change of deformation is analysed in Table 7.3 by comparing the
percentage of change with and without self weight.

20
σM (N/mm2)

15

A1sw
10
A3sw

0
0 5 10 15 20
Node number

Figure 7.50: Case A1sw and A3sw (deformation by a combination of vertical force and SW), bending stress in the
upper edge
σM (N/mm2)

20

15

B1sw
10
B3sw

0
0 5 10 15 20
Node number
Figure 7.51: Case B1sw and B3sw (deformation by a combination of horizontal force and SW), bending stress in the
upper edge
σM (N/mm2)

20

15

C1sw
10
C3sw

0
0 5 10 15 20
Node number

Figure 7.52: Case A3sw and C3sw (deformation by a combination of diagonal force and SW), bending stress in the
upper edge

141
A Design Tool for Timber Gridshells

The differences between the cases with and without pendulums are presented in the table
below. The differences of the cases without self weight are included to compare the different
cases.

Table 7.3: Differences between cases with and without pendulums

Δu (%) node 11 Δσm (%) Node 11 Δσm (%)Node 6 and 16
Without SW With SW Without SW With SW Without SW With SW
A3 – A1 1,29 2,09 12,9 19,7 -23,8 -37,5
B3 – B1 0 0,86 0,24 8,4 -0,25 -15,5
C3 – C1 1,10 1,58 10,5 14,9 -19,1 -28,2

It can be seen that including self weight has an increasing effect on the differences between
the cases with and without pendulums. This can be explained by two things. First the
deflection by self weight and deflection by applied force have to be superimposed to acquire
the total deformation and stress distribution. Second, including self weight adds to the effect
of non-linear behaviour, especially for the cases with horizontal force. By the deflection due
to self weight the moment arm for the horizontal force gets larger, which increases the
bending moment at the supports at node 6 and 16.

As stated in Section 7.2.8, the increase in deflection and bending stresses has to be taken
into account when designing a gridshell, although the effect will be less prominent when
more supports are used. This is also true for the effect of self weight. When more supports
are used, the weight is divided and also its influence will be less.

142
 Members in bending

7.3 Interaction between the laths

To investigate the interaction between the laths, a 3D case is analysed. For this case, four
equal laths are used with the same properties of the 2D case. The 3D structure, which is
shown in Figure 7.53, has a middle beam of 10m in length which is supported by three similar
beams at 2,5, 5,0 and 7,5 m. The middle beam is roll supported in the x direction. The
transverse beams are pinned at their ends and restricted in the x direction to avoid buckling
during analysis. The beams are linked together at their intersection nodes.
The interaction will be investigated by deflecting the supporting beams and analyzing the
effect of their deflection on the middle beam. Two situations will be analysed: first the
situation where all three supporting beams are deformed with equal deflection; second the
supporting beams will be deformed by unequal deflection.
In the former cases the analysis started with a structure supported by pendulum columns,
pushing the member down and then pinning it down. These first two steps will be skipped in
this analysis, as the effect of these steps is already known.

Beam 1

Beam 2

Beam 3

Beam 4

Figure 7.53: 3D structure in initial flat position

143
A Design Tool for Timber Gridshells

7.3.1 Supporting beams with equal deformation

In situation 1 the structure is deformed by deflecting the middle beams with deflection equal
to the deflection of case C1, which was deflected by 100N vertical and 100N horizontal force.
This results in the following deflection and force diagrams (Figure 7.54 to Figure 7.57):

ux = 6,89 mm uz = -482 mm
uy = 56,5 mm

uz = 178 mm

uz = 203 mm

uz = 178 mm

ux = -6,89 mm
uz = -482 mm
uy = 56,5 mm

Figure 7.54: Case 3D1, deformation

Mxx = 0,319 kNm Myy = 0,200 kNm

Mxx = 0,251 kNm

Mxx = 0,319 kNm

Myy = 0,037 kNm

Mxx = -0,438 kNm

Figure 7.55: Case 3D1, moment line

144
 Members in bending

-141 N

-83,5 N
-74,5 N
-141 N
93,9 N

80,4 N -94,8 N

-74,5 N
74,5 N

94,8 N -80,4 N

141 N 93,9 N
74,5 N

83,5 N

141 N
Figure 7.56: Case 3D1, shear force

-1030 N
66,2 N

-1040 N

-1030 N
49,7 N

-1040 N
-1030 N

66,2 N

-1030 N
Figure 7.57: Case 3D1, axial force

145
A Design Tool for Timber Gridshells

7.3.2 Supporting beams with unequal deformation

In situation 2, beams 2 and 4 are deflected by a deflection equal to the deflection of case
C1a, which was deflected by a horizontal force of 200N and a vertical force of 100N. The
horizontal deflection of Beam 3 is kept the same as the previous situation. However, the end
nodes are shifted down to the same vertical displacement of beam 2 and 4. This means that
the end nodes will be at the same vertical position, which would be the case in real structure
on horizontal foundations.

uz = -620 mm
uy = -94,6 mm
ux = 10,4 mm uz = -620 mm
uy = -56,5 mm

uz = 184 mm
uz = -620 mm
uy = -94,6 mm
uz =8,5 mm

uz = 184 mm

ux = -10,4 mm

uz = -620 mm
uy = 56,5 mm

Figure 7.58: Case 3D2, deformation

Mxx = 0,422 kNm

Myy = 0,476 kNm
Mxx = 0,073 kNm

Mxx = 0,422 kNm

Myy = 0,653 kNm

Mxx = -0,486 kNm

Figure 7.59: Case 3D2, moment line

146
 Members in bending

-287 N

-277 N
-41 N

190 N -335 N
-287 N

-385 N

-277 N
277 N

385 N
-190 N
287 N
335 N
277 N
41 N

287 N

Figure 7.60: Case 3D2, shear force

-1850 N

1817 N

-1870 N

1787 N
-1850 N
-1850 N

1817 N

-1870 N

-1850 N

Figure 7.61: Case 3D2, axial force

147
A Design Tool for Timber Gridshells

7.3.3 Analyzing the interaction

By deflecting the three transverse beams, the middle beam is also deflected. The three
beams interact with beam 1 and deflection of beams 2,3 and 4 cause deflection and stresses
in beam 1. This interaction occurs in the links between the beams and can be translated into
link forces. For situation 1 these link forces are:

• Link between beams 1-2 and 1-4: 176 N

• Link between beam 1-3: 190 N

For situation 2 the link forces are:

• Link between beams 1-2 and 1-4: 577 N

• Link between beam 1-3: 771 N

The internal stress distribution of each beam can be seen as the superposition of two load
cases: the applied displacements and the link force acting on it. Due to the interaction
between the beams, the internal stress distribution changes from what we have found in the
analysis cases in Section 7.2 to the stress distribution found in this section.
In situation 2 the link forces are of a larger magnitude than in situation 1. This difference is
the result of the difference of deflections of the transverse beams. The middle beam is pulled
down by the middle transverse beam much stronger than in situation 1. This results in large
compression forces in the outer transverse beams and a large tension force in the middle
transverse beam (Figure 7.61). Also the moments change. From approximately zero at mid
span for beam 1 in situation 1 to 0,653 kNm in situation 2. The maximum moment in beam 3
stays approximately equal, but the moment diagram does change in shape. For beam 2 and
4, the moment at the link decreases to approximately zero due to the downward force. The
field moment however increases from 0,3 to 0,4 kNm. In beam 1 the moment becomes more
than twice as large at the links with beam 2 and 4 and increases from 0,2 to 0,47 kNm

The link forces are shown in Figure 7.62 for situation 2. The link forces between the outer
transverse beams and beam 1 are compression forces. This results in axial compression in
beam 2 and 4. The compression force acts in upward direction on beam 1. The curvature of
the middle beam and the outer beams are both in the same direction and when a surface is
imagined over the beams, this would be a clastic surface.
The link force between beam 3 and beam 1 is a tension force, which results in an axial
tension force in beam 3 and a force pulling beam 1 down at mid span. This results in an
opposite curvature in beam 1 and 3. When a surface is imagined over the beams again, the
surface would be anti-clastic in this area

In a larger structure, a lot more members interact with each other. The effects that members
have on each other will be of the same principle with compression and tension link forces.
When a clastic surface is formed, the members affect each other with compression link
forces. When the shape pushed from clastic to anti-clastic, tension link forces are to be
expected.

148
 Members in bending

Beam 1

577 N

771 N

577 N

Beam 2

Beam 3

Beam 4

Figure 7.62: Exploded view of the structure (situation 2)

149
A Design Tool for Timber Gridshells

7.4 Curvature and RD-forces

As found in the previous sections, curving a lath results in a bending moment in the lath. It
was also found that when a lath is curved in a shape on a series of intermediate supports, the
deflected shape is probably not equal to the natural bending shape. The lath will therefore
deflect back to a natural equilibrium position when the internal supports are removed: the
relaxation deflection (RD).
After relaxation of the timber, approximately half of the stress level remains as residual
stresses. For structural analysis, it is needed to take into account the residual formation
stresses and it is therefore desired to determine these stresses. When the final geometry of
the gridshell is known, the bending and torsion stresses can be calculated, using the curve
angles as described in Section 3.2.4. With the gridshell design tool this becomes easy,
because the angles are part of the output of the tool. By using a spreadsheet the curve
stresses are quickly calculated for all nodes.

When this method is used to calculate the formation stresses, the final geometry of the
structure must be known or at least a proper approximation is needed. The problem is that it
is uncertain in what extend the design tool approximates the final geometry. If the generated
geometry is not equal to a natural bending shape and the mat of laths is forces in this
geometry during construction, a relaxation deflection will occur when the internal supports
are removed, as seen in Section 7.2. To estimate the relaxation deflection, a (geometrical)
non-linear analysis can be used as used in previous sections. For a simple beam this is no
problem, but when a complete gridshell with a few thousand nodes and elements is to be
analysed, non-linear analysis becomes time consuming. A simpler method to determine the
relaxation deflection and RD-stresses would greatly help the design process of a gridshell. In
this section it is researched if the curvatures of a structure can be used for this, based on the
assumptions described below.

Let’s consider a structure that is bent in a non-natural bending shape on internal supports.
What keeps the structure in its shape prior to removing the internal supports is the reaction
forces at the internal support. When the internal supports are removed, but these support
reactions are kept forcing on the structure as an applied force, the structure would still keep
the same shape. When equal but reversed forces are superimposed on the structure, the
support reactions will be equalled out and the structure will deflect back to its natural
equilibrium position. One could say that these reversed forces are the RD-forces for an
asymmetric curved lath.
This theory is applied a lath curved in a non-natural way. One can simply calculate the
moments that are the result of the curvature, according to Section 3.2.4.1. This curvature
with its moment can be seen as the result of forces pushing the structure in a shape (Figure
7.63a and b). When the opposite of these forces is applied on the structure the structure
would deflect back (Figure 7.63 c and d). These opposite forces can be calculated by
translating the moment in shear forces, by taking the gradient of the moment line
(Hartsuijker, 1999). When the lath is divided into discrete elements, for every element a
shear force can be calculated. The difference in shear force at a node between two elements
would be the RD-forces acting on this node.

150
 Members in bending

F
F (b)
(a)

-F

-F

(d) -F (c)

Figure 7.63: The forces F in on a natural curved beam (a) result in the moment line (b). The opposite of the forces F
(c) should result in a deflection back to the natural curve (d). Deflections, forces and moment line are indicative.

Applied on a large structure which is not curved in a natural way, it would be simpler to
calculate the RD-stress distribution with this method than with non-linear analysis. When all
curvatures at all nodes are known, the moments due to the curvatures can be calculated and
therefore also the resulting RD-forces are known. By applying these forces, a linear load case
"relaxation deflection" is created. The result of this load case would be a structure deflected
to the natural curved shape. One could say form-finding is applied in a very basic way.

The theory described above is tested on a simple structure as follows:

• Situation 1: a lath (lath 1) is asymmetrically curved over internal supports. The

support reactions in the supports would be the inverse of the RD-forces.
• From the initially asymmetric curved lath the curve angles between segments are
determined.
• These angles are translated into moments, shear forces and RD-forces;
• Situation 2: the lath is implemented in GSA with its curved geometry. This is the
strainless situation.
• Situation 3: the RD-forces are applied on the structure, which results in deflections.
• Situation 4: an initial straight lath (lath 2) with a length equal to the length of lath 1
is bent up in a GSA analysis. The deflection is the natural bending curve of lath 1.
• Situation 5: the inverses of the RD-forces are applied on lath 2. This load case
should be equal to situation 1 and so should be the deflections. Also the resulting
moments should be equal to the moments that are derived from the curve angles.
• The different situations can be compared. The deflected shape of situation 3 should
be equal to the shape in situation 4. The deflected shape of situation 5 should be
equal to the shape in situation 2.

151
A Design Tool for Timber Gridshells

7.4.1 Initially curved beam

The theory described in the previous section is tested on a simple structure. Let’s consider a
lath that is bent on internal supports in a non-natural bending shape (situation 1). This lath is
inputted strainless in GSA (situation 2). The lath shown in Figure 7.64 is divided into 9
elements, each 2,5m long. The angles at each node between the different elements are
determined. From these angles the moments can be calculated which would have occurred
when the lath would have been bent into this shape from an initial straight position. The
moments can be translated into shear forces. This data is displayed in Table 7.4.

The moments at every node can be calculated using the formulas given in Appendix 1:
Determination of the maximum bending radius. The next formula can be stated, which
calculate the stress due to a certain bending angle:

hEα b
σm = ( 7.1 )
2L

Where:
αb = the bending angle
L = the element’s length

Together with

M
σm = ( 7.2 )
W

the bending moment can be calculated.

The shear forces at each node can be calculated with:

M j − M j −1
Fs , j = ( 7.3 )
L

Where:
Fs,j = the shear force at node j
Mj = the moment at node j

Figure 7.64: Curved beam

152
 Members in bending

Table 7.4: Angles between the nodes with corresponding moments and shear forces
Between elements α M Fshear ΔFshear
Node (rad) (Nmm) (N) (N)
1 - 1 - 0 0 -
2 1 2 0,068 1,42*105 56,716 -46,040
3 2 3 0,081 1,68*105 10,676 6,230
4 3 4 0,101 2,11*105 16,906 9,572
5
5 4 5 0,133 2,77*10 26,478 13,730
6 5 6 0,181 3,77*105 40,208 12,781
7 6 7 0,245 5,10*105 52,988 -9,455
8 7 8 0,297 6,19*105 43,533 -47,469
9 8 9 0,292 6,09*105 -3,936 -239,632
10 9 - 0 -243,568

This data can also be displayed in a moment and shear-force diagram. The moments on the
structure due to curvature are:

0,510
0,377 0,619

0,277

0,211 0,609

0,168

0,142

Figure 7.65: Moment due to curvature (x106 Nmm)

53,0
40,2 43,5
26,5

16,9

10,7 -3,9
56,7

-243,5

Figure 7.66: Shear forces due to curvature (N)

153
A Design Tool for Timber Gridshells

The difference in shear at each node results in the RD forces which are applied on the
structure (situation 3). This should result in a bending shape which approximates the natural
bending shape. The deflection caused by this load is shown in Figure 7.67. Corresponding
moment, shear and axial force diagrams are shown in Figure 7.68 to Figure 7.70.

-47,5 -239,6
-9,5

-46,0 13,7 12,8

9,6
6,2

Figure 7.67: Deformation due to curvature forces (N)

0,205

0,450

Figure 7.68: Moment line due to curvature forces (kNm)

84,3

181
46,9

Figure 7.69: Shear forces due to curvature forces (N)

-113,8

-108,5 -87,3

Figure 7.70: Axial forces due to curvature forces (N)

154
 Members in bending

7.4.2 Comparison with a straight lath

The deformation of this asymmetric curved lath by the RD-forces (situation 3) is compared
with the deformation of a straight lath which is bent up (situation 4). The deflection of this
lath is the natural bending shape. The lath is deflected by moving the end nodes until the
distance between the end nodes is equal to the distance of the end nodes of the curved
beam. The straight lath has the same length as the curved lath of 22,5 m. the end node is
moved 2,32 m. to the left. This results in the deflection shown in Figure 7.71.

2,32 m

Figure 7.71: Deformation of a straight lath

To compare the deformation of the initially curved lath with the deformation of the initially
straight lath, the deflections are put together in a graph (Figure 7.72). To give a complete
picture, the deformations of the initial curved lath are calculated by linear and non-linear
analysis. The deformation of the initial straight lath is calculated by non-linear analysis.

5
Deflection (m)

4.5

3.5

3 Initial curved lath

(linear analysis)
2.5
Initial curved lath
2 (non-linear analysis)

1.5 Initial straight lath

0.5

0
0 2 4 6 8 10
Node number

Figure 7.72: Deformation of the initial straight and initial curved lath

155
A Design Tool for Timber Gridshells

In the graph it can be seen that there is a small difference between the different cases. The
initial straight lath bends to the natural deflection shape. Difference for the middle nodes
(node 5 and 6) between the natural deflection and the deflection calculated by non-linear
analysis for the initial curved lath is 60 mm downward (1,4%). With linear analysis the
difference is 57mm upward for node 4 and the deformation shape keeps a small eccentricity
to the right.

When this lath is loaded with the inverted RD-forces (situation 5), the deflection should
become equal to the deflection of situation 2. The following results for deflection and moment
are found:

Figure 7.73: Deformation of the initial straight lath, by applied displacements and inverted RD-forces

Figure 7.74: moments as result of the applied displacements and inverted RD-forces

When these results are compared with the coordinates of the initial situation (situation 2) and
the moments that are calculated from the curve angles, it is found that the two situations do
not differ much (Table 7.5). For the deflection a maximum difference of 2% is found. In the
moments the maximum difference is 4,4%.

Table 7.5: Difference in deflection and moment between situation 2 and 5

z-coordinate (m) difference Moment (Nmm) difference
Node situation 5 situation 2 (m) % situation 5 calculated (Nmm) %
1 0 0 0 - 0 0 0-
2 1.236 1.228 0.008 0.64 141200 141789 -589 -0.42
3 2.343 2.305 0.038 1.62 169100 168479 621 0.37
4 3.261 3.196 0.065 1.98 214800 210743 4057 1.89
5 3.931 3.847 0.084 2.14 286900 276938 9962 3.47
6 4.256 4.172 0.084 1.97 394900 377457 17443 4.42
7 4.102 4.045 0.057 1.38 533100 509927 23173 4.35
8 3.327 3.320 0.007 0.22 638300 618759 19541 3.06
9 1.892 1.929 -0.037 -1.95 611800 608920 2880 0.47
10 0 0 0 - 0 0 0-

156
 Members in bending

Comparing the results of situation 3 and situation 4, it can be concluded that the deflection
due to the RD-forces gives an accurate result for the equilibrium position of the lath. Only a
small difference is found between the results calculated with non-linear analysis and the
results of the calculation with RD-forces. A non-linear analysis gives the most accurate result,
so a non-linear analysis cannot be ruled out if accuracy is desired. This non-linear analysis is
however much simpler than the non-linear analysis that is needed to calculate an equilibrium
position for an initial flat mat of laths. In stead of calculating the deflection from initially flat
to deformed, a series of forces is applied on an already curved structure and only the
relaxation deflection has to be calculated.
Comparing situation 5 and situation 2, it is shown that the bending moments that can be
derived from the curve angles are correct. In the analysis the derived moments only differ
4,4% with the moments calculated with the non-linear analysis.

157
A Design Tool for Timber Gridshells

7.4.3 RD-forces in a three dimensional grid

An equal analysis as in the previous section is performed with a 3D structure. As a test case,
a shape is needed that is not symmetric. The tip of a rotated ellipsoid is chosen (Figure 7.75).
Using the grid generation tool, a grid is generated with a mesh size of 2,5m, shown in Figure
7.76. This can be seen as the structure deformed on internal supports (situation 1).

Figure 7.75: Tip of a rotated ellipsoid in side view and 3D

Figure 7.76: grid generated on the ellipsoid tip

The following steps are needed to perform the comparison:

• Import the structure generated by the gridshell design tool in GSA. This is the
unstrained situation of the structure (situation 2).
• Calculate RD forces out of the curvatures. Curvature around the z-axis results in
horizontal forces. Curvatures around the element’s local y-axis result in vertical
forces.
• Apply all forces on the unstrained structure. This results in the deformation of the
unstrained structure (situation 3)
• Determine the natural bending shape of the grid with a non-linear analysis, by
applying generated coordinates as applied displacements (situation 4)
• Compare the geometry of this strained situation with the coordinates after
deformation of the unstrained structure.

158
 Members in bending

The grid is generated with a lath cross section of 35mm high and 50mm wide. With the
output tool the model data is exported. This data can be used to model the structure in GSA,
by copying the node coordinates and the element begin- and endpoints into the program.
This model is the strainless situation of the structure (situation 2). The data in the output file
is ordered in such way that the laths can be implemented in GSA as continuous laths in two
directions. The intersections between the laths can be implemented as link nodes, with
unrestrained rotation around the z-axis.

In the output file, also the angles of curvature in x-, y- and z- direction can be found. From
these angles, bending and torsion stresses can be derived for every node, in the same way as
described in Section 7.4.1. These stresses can be translated in moments, and the moments in
shear forces acting on every element. The difference between the shear forces at every node
are the RD forces. These can be applied on the strainless structure (situation 3). If the
assumption made in this section is correct, this should result in a shape similar to the natural
bending shape.

The deflected shape of the strainless structure should be compared with the natural bending
shape. This natural bending shape can be determined with a non-linear analysis in GSA. With
a relatively small structure as this, a non-linear analysis is quickly performed. This analysis
starts with an initially flat mat of laths. As with the cases in Section 7.2, applied
displacements are enforced on all nodes of the structure to deform the structure to a curved
shape. The applied displacements in x, y and z direction are equal to the x, y and z-
coordinates of the nodes that are generated by the design tool. It should be noticed that a
permanent applied displacement can only be enforced on a node which has a support. The
applied displacements that are enforced on the nodes that do not have an edge support are
used to push the structure in an initial shape. In the non-linear analysis iterates to an
equilibrium position and what is left is the applied displacements of the supported edge nodes
and the equilibrium displacements of the other nodes.

Because a non-linear analysis iterates to a situation in which the structure is in equilibrium,

this shape should be the natural bending shape of the grid (situation 4), complying with the
applied displacements of the supports. Now this shape is known, the deflections of the
unstrained structure can be compared with the natural bending shape.

First, the unstrained structure is implemented in GSA (Figure 7.77). This structure is then
loaded by the RD-forces in horizontal and vertical direction (Figure 7.78 and Figure 7.79).
Results will be displayed on the points indicated by their node number in Figure 7.77, the
loads are displayed. The resulting deflection is shown in Figure 7.80 with a magnification of
20.

159
A Design Tool for Timber Gridshells

20

16

3 5 7
1/2
22

34 29
37

43

46

Figure 7.77: Unstrained structure in GSA (with node numbers for indicative sections)

Table 7.6: Angles of curvature at nodes, with corresponding moments, shear forces and RD-forces
Node αyy αzz σyy σzz Myy Mzz τxz τxy RD forces RD forces
number (rad) (rad) (N/mm2) (N/mm2) (Nmm) (Nmm) (N) (N) (N) (N)
34 0 0 0 0 0 0 - - - -
29 0.138 0.026 9.687 1.799 98891 53528 39.556 21.411 -39.223 -41.933
22 0.140 0.001 9.769 0.075 99724 2225 0.333 -20.521 2.431 30.359
1 0.149 0.013 10.446 0.901 106635 26819 2.765 9.838 3.280 -5.142
3 0.170 0.019 11.926 1.296 121746 38558 6.044 4.695 5.745 -3.072
5 0.212 0.020 14.813 1.432 151217 42617 11.789 1.624 -72.275 -18.671
7 0 0 0 0 0 0 - - - -

20 0 0 0 0 0 0 - - - -
16 0.141 0.016 9.864 1.133 33723 5584 40.278 13.489 -40.349 -16.128
9 0.141 0.013 9.847 0.911 27125 5283 -0.071 -2.639 2.528 -6.250
2 0.149 0.002 10.448 0.165 4904 5192 2.456 -8.889 3.428 10.625
37 0.170 0.004 11.889 0.311 9245 5221 5.884 1.736 5.713 29.471
43 0.210 0.042 14.729 2.932 87262 5330 11.597 31.207 -71.741 -66.112
46 0 0 0 0 0 0 - - - -

45,4 N
11,3 N

13,1 N
6,8 N
4,4 N 2,1 N

3,6 N 5,2 N 2,9 N

7,4 N
17,3 N

29,3 N 21,3 N 20,6 N

2,7 N
16,1 N

6,5 N 46,3 N

Figure 7.78: Horizontal RD-force, which result from bending around the z-axis (N) (figures are the sum of two nodes
in an intersection and therefore not equal to the figures in Table 7.6)

160
 Members in bending

86,7 N 66,2 N

6,5 N

10,1 N

78,7 N
6,7 N

11,6 N
81,4 N
13,6 N

Figure 7.79: Vertical RD-forces, resulting from bending around the (local) y-axis (N) (figures are the sum of two nodes
in an intersection and therefore not equal to the figures in Table 7.6)

uz = -18,2 mm

uz = 18,2 mm

uz = -23,1 mm
uz = 12,8 mm

Figure 7.80: Deformation by the RD forces (magnified 20x) (mm)

161
A Design Tool for Timber Gridshells

For the non-linear analysis, the structure is first implemented as a flat mat (Figure 7.81). The
structure is then deformed by applied displacements. The results of the non-linear analysis
are displayed in Figure 7.82.

0,25 mm 0,30 mm 0,33 mm

0,36 mm
0,16 mm

0,16 mm 0,40 mm

0,27 mm

0,42 mm
0,32 mm

0,48 mm
0,43 mm

0,46 mm 0,46 mm

0,49 mm
0,62 mm 0,61 mm
0,54 mm
Figure 7.81: Undeformed mat of laths with applied displacements of the support nodes (m)

uz = 1,115 m

Figure 7.82: Deformed structure, by non-linear analysis

162
 Members in bending

Now the results are known, the structures can be compared. Two sections are reviewed,
displayed in Figure 7.83. The coordinates of the nodes on these sections are displayed in the
graphs in Figure 7.84 and Figure 7.85.

A A'

B'

Figure 7.83: Sections AA' and BB'

1.2
u (m)

deformed by
0.8 displacements
unstrained

0.6 unstrained,
deformed by force

0.4

0.2

0
0 2 4 6 8 10 12 14
x (m)
Figure 7.84: Section AA' over the different structures

163
A Design Tool for Timber Gridshells

1.2
u (m)

0.8

deformed by
0.6 displacements
unstrained

unstrained,
0.4
deformed by force

0.2

0
0 1 2 3 4 5 6 7 8 9 10
x (m)
Figure 7.85: Section BB' over the different structures

As can be seen, the shape that is determined by non-linear analysis by applied displacements
shows a different shape than the unstrained structure, which is equal to the original surface.
This deformed shows a more parabolic shape. The difference between the form found
structure and the unstrained structure is 30,3mm in vertical direction, which is 2,8% of the
total deflection. When the RD forces are applied, this is reduced to 23,5mm (2,1%)
When the RD forces are applied on the unstrained structure, the structure does deflect
toward the form found shape, but not much. With the single lath, applying the RD-forces
gives a deflection back to the natural bending shape of the lath. With this 3D structure, less
resemblance is found. This is probably due to the fact that a 3D structure reacts more stiff to
applied loads than a single element.

From this analysis it cannot be concluded that the RD-forces can be used as a load case in
structural analysis to implement the stresses due to the formation process of the shape. With
a single lath the results look promising, so more research with 3D structures could prove the
method usable.

164
 Members in bending

7.4.4 Stress level derived from curve angles

Another option to implement the formation stresses is to use the stresses calculated from the
curvatures of the generated grid as a superposition load case. Load cases like snow and wind
can be analysed in a structural analysis program. By adding the formation stress manually,
the structure can be checked for the combination of stresses.
To analyse the usability of these stresses, they are compared with the stresses that are found
in GSA in the form found structure.

For the comparison, the two laths are reviewed that are indicated by node numbers in Figure
7.77. The bending stress resulting from bending around the y-axis can be calculated from the
curve angles and is displayed in the Table 7.7. The results from the non-linear analysis are
less straight forward. Due to the interaction between laths, there is not one bending stress
for every node, but it can differ for each element attached to the node. In Figure 7.86 the
moment lines of the moment around the y-axis is displayed. This clearly shows jumps in the
moment at certain nodes.
In Table 7.7, the bending stress for the elements left and right to the node (element 1 and
element 2) is displayed, as well as the average of the two. These stresses are compared with
the calculated stress. The difference is displayed as percentage of the calculated stress.

20
16
9
3 5 7
1/2

22

37
29
34

43

46
Figure 7.86: Moment lines on the form-found structure (numbers indicate the reviewed nodes)

Table 7.7: Bending stress in the upper edge of the elements

element1 element2 avarage calculated % diff 1 % diff 2 % diff av.
Node (N/mm^2) (N/mm^2) (N/mm^2) (N/mm^2) (%) (%) (%)
34 - 0 0 0 0 0 0
29 20.07 17.63 18.85 9.687 107.18 81.99 94.58
22 8.936 8.948 8.942 9.769 -8.53 -8.40 -8.46
1 10.2 11.86 11.03 10.446 -2.35 13.54 5.59
3 6.76 8.52 7.64 11.926 -43.32 -28.56 -35.94
5 23.32 25.59 24.455 14.813 57.43 72.75 65.09
7 0 0 0 0 0 0 0

20 - 0 0 0 0 0 0
16 20.49 18.33 19.41 9.864 107.72 85.83 96.78
9 5.808 5.813 5.8105 9.847 -41.01 -40.96 -40.99
2 17.22 6.34 11.78 10.448 64.82 -39.32 12.75
37 7.803 7.165 7.484 11.889 -34.37 -39.73 -37.05
43 22.59 24.42 23.505 14.729 53.37 65.79 59.58
46 0 - 0 0 0 0 0

165
A Design Tool for Timber Gridshells

When the difference between the average of the bending stress from the GSA analysis and
the bending stress calculated from the curve angles is reviewed, it is noticed that the
difference in the outer nodes differ largely. Moving to the middle of the structure, the
difference becomes smaller. For the middle nodes, node 1 and 2, the difference is 5,6 and
12,7 respectively.
The large difference between the analysed stress and the calculated stress can be explained
by the fact that in GSA the supports are modelled as pin supports. The bending stress
therefore has to increase from zero in the support to a bending moment corresponding with
the bending radius. The stresses calculated from the bending angles provided by the design
tool are based on a continuous curve and therefore a more uniform stress distribution.
The stresses in the middle nodes differ less with each other than the outer nodes. The edge
disturbance has less influence on the stress level when the distance to the edge becomes
larger. Also this analysis is performed with a mesh size of 2,5m. In reality the mesh size will
be smaller, e.g. 0,5m or 1,0m. The peaks in moment found in the form-found structure will
be also smaller, because the edge disturbance is spread over more elements.
It can therefore be assumed that in a larger structure, a better resemblance can be found
between the stresses provided by GSA and the calculated stresses. In a larger structure the
edge moments will settle to a more uniform stress distribution within the first few elements.

To give an indicative view on the level of stress that is caused by the formation process, the
stress level in the two reviewed laths is checked with the stress criterion stated in 3.2.4.4.
The combination of stresses in the formation process should comply with:

2 2 2
⎛ σ m,d ⎞ ⎛ τ v,d ⎞ ⎛ σ c / t ,0, d ⎞
⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ ≤ 1 ( 7.4 )
⎝ f m,d ⎠ ⎝ fv,d ⎠ ⎝ f c / t ,0, d ⎠

in which:
σ m,d = σ m, y ,d + σ m, z ,d
τ v , d = τ xy , d + τ tor , d

This is first checked for the calculated stress level. Using the curve angles, stress levels for
moment in two directions, shear and torsion can be determined. Moment and torsion is
determined according to Sections 3.2.4.1 and 3.2.4.3 respectively. Shear stress can be
calculated by deriving shear forces from the bending moments as described in Section 7.4.1.
The shear stress can be calculated with:

3 Vd
τ xy = ( 7.5 )
2 A

In which:
Vd = the shear force
A = the cross section of the element

It is not possible to determine levels for axial stress, so this is left out. This results in the
stresses displayed in Table 7.8.

166
 Members in bending

For the structure analysed with GSA, the stress levels can be collected from the output tables.
First the results that are found by GSA for bending and shear stress in two directions and
axial stress is displayed in the figures below:

Figure 7.87: Bending around the elements' y-axis, Myy

Figure 7.88: Bending around the elements' z-axis, Mzz

167
A Design Tool for Timber Gridshells

Figure 7.89: Shear stress in the elements' y-direction, τxy

Figure 7.90: Shear stress in the elements' z-direction, τxz

Figure 7.91: Axial stress in the elements, σxx

168
 Members in bending

The stress levels from the both cases can be displayed in tables with the result for the stress
combination criterion. For the stresses calculated from the curvature angles, this is displayed
in Table 7.8. The stress levels that are collected from the GSA output for the analysed
structure are displayed in Table 7.9. If stress levels at both sides of a node differ, the
maximum stress level is used.
When the calculated stresses are used to check the stress level, all elements comply with the
criterion. The stress level is 0,32 to 0,97 times the maximum allowed stress combination.
When the stresses found by GSA are checked, it is found that in the nodes attached to the
edge elements, the stress combination criterion is exceeded. This resembles with the findings
that in these nodes a much larger bending moment exists. The other nodes comply with the
criterion within a range of 0,31 to 0,99, which resembles the results of the calculated
stresses. If the peaks in moment are neglected, based on the assumptions made on page
166, it can be stated that stress levels found by using the curve angles give a usable
approximation for superposition in structural analysis.

Table 7.8: Stress levels calculated from curve angles

σyy σzz τxx τxz τxy
Node (N/mm2) (N/mm2) (N/mm2) (N/mm2) (N/mm2) Criterion
34 0 0 0 - - -
29 9.69 1.80 0.40 0.03 0.02 0.45
22 9.77 0.07 0.38 0.00 -0.02 0.32
1 10.45 0.90 0.38 0.00 0.01 0.42
3 11.93 1.30 0.37 0.01 0.00 0.56
5 14.81 1.43 0.38 0.01 0.00 0.82
7 0 0 0 - - -

20 0 0 0 - - -
16 9.86 1.13 0.40 0.03 0.01 0.41
9 9.85 0.91 0.38 0.00 0.00 0.38
2 10.45 0.16 0.37 0.00 -0.01 0.37
37 11.89 0.31 0.37 0.01 0.00 0.48
43 14.73 2.93 0.38 0.01 0.03 0.97
46 0 0 0 - - -

Table 7.9: Stress levels found by GSA in the structure

σyy σzz τxx τxy τxz σxx
Node (N/mm2) (N/mm2) (N/mm2) (N/mm2) (N/mm2) (N/mm2) Criterion
34 0.00 0.00 0.00 -0.15 0.28 0.01 0.01
29 20.07 7.45 0.00 -0.15 0.28 0.01 2.23
22 8.94 2.82 -0.09 0.04 -0.02 -0.07 0.41
1 11.86 5.84 0.40 -0.01 -0.01 -0.06 0.96
3 8.52 6.58 -0.05 0.05 0.04 -0.08 0.67
5 25.59 5.81 0.00 -0.05 -0.15 -0.04 2.90
7 0.00 0.00 0.00 -0.05 -0.15 -0.04 0.01

20 0.00 0.00 0.00 0.10 -0.24 -0.14 0.01

16 20.49 5.84 0.00 0.10 -0.24 -0.14 2.04
9 5.81 4.42 0.07 -0.04 0.04 -0.22 0.31
2 17.01 1.38 0.03 0.01 -0.03 -0.19 0.99
37 7.80 4.67 0.08 0.02 0.00 -0.19 0.46
43 24.42 10.17 0.00 0.10 0.16 -0.22 3.53
46 0.00 0.00 0.00 0.10 0.16 -0.22 0.02

169
A Design Tool for Timber Gridshells

7.5 Conclusions

From the analysis performed on members in bending a few conclusions can be drawn. On the
subject of maximum curvature the analysis showed that it is beneficial to select timber with a
low E /fm ratio. This will result in a small maximum bending radius and thus large curvatures
can be reached prior to failure. This is also true for the torsion angle. It would be interesting
to look at E /fm of green timber. By applying green timber, a larger curvature should be
possible. Proper property information however is scarcely available. With the Weald and
Downland gridshell green timber was applied. Extensive bending testing proved the timber to
have sufficiently moment and bending capacity.

On the subject of the formation stresses, the analysis showed that care should be taken when
deforming the laths over internal supports. When the deflection does not approximate the
natural equilibrium deflection, the structure will deflect to its equilibrium position after
removal of the internal supports. This results in a change of geometry and a change in stress
distribution. When this is left unaccounted for, breakage of the laths can occur. The analysis
showed that the deflection by applying a horizontal force approximates the equilibrium shape
best. A large force is needed however to reach the desired deflection. A combination of
horizontal and vertical forces gives a larger deflection with less applied force and an
acceptable difference from the equilibrium shape.
In the 3D analysis, it was found that the laths in a 3D structure influence each other by
interaction in the nodes. Link forces act on the laths in the nodes and can be seen as forces
superimposed on the applied deflection.

The assumption that from the curvature of a curved member relaxation deflection forces can
be derived is also tested and the results look promising. When the RD-forces are applied on
the curved member, the member deflects to a shape that approximates the equilibrium
position of a bent lath. This can be seen as a form finding process, as the structure's
equilibrium position is approximated. However, for the 3D structure analysed, the results are
less accurate. This is probably due to the fact that a 3D structure reacts much stiffer to a load
due to the 3D force flow. More research is needed to validate the use of the RD-forces to
determine the equilibrium position.

When the bending stress from the formation process is calculated by using the curve angles,
the results are accurate when the single lath is reviewed. This method thus is correct.
However, in the 3D structure, the interaction between the laths disturb the moments that
would be the result of curvature only. The difference that is found between the stress in the
form found structure and the generated structure is 6% for the top node to 60-97% for the
edge nodes. It is assumed that edge disturbances are the cause for the large difference at
the edges. In the middle of the structure, the edge disturbances have less influence on the
stress distribution. In a larger structure with smaller mesh size this will result in a more
continuous stress flow. It is therefore assumed that the curve angles can be used for
structural analysis for large part of the structure, to give an indicative stress level. Because of
the possible differences one should be conservative in using the calculated bending stress
level. When for example a relaxation factor of 0,7 is applied in stead of 0,5, the calculation
will give a safe solution for the bending stress level after relaxation of the timber.

170
Members in bending

171
A Design Tool for Timber Gridshells

172
 Conclusions and recommendations

8
Conclusions and
recommendations

8.1 Introduction

In this Master’s thesis, a study has been performed into the applicability of a design tool
based on the main geometrical property of the gridshell. This is the equal distance between
the mesh nodes in a quadrangle grid. This method has been implemented in a design tool.
With the design tool, a gridshell grid can be generated on an arbitrary surface. The resulting
model contains all geometrical information which is needed for exporting the geometry to a
third party structural analysis software package. The output data also contains angles of
curvatures of the members in different directions. These can be used to determine the
bending stress levels resulting from the form shaping process.
The findings of this Master’s thesis research will be stated here as conclusions and
recommendations for further research.

8.2 Conclusions

• The grid generation method proposed in this thesis can be used to determine the
geometry for the grid of a gridshell. The method has been implemented in a design tool.
This design tool can be used for determining the grid geometry. An output can be
created, which can be processed further for structural analysis. The created geometry is
not necessarily correct. A large amount of experience is still required to review and
analyse the created geometry.
• The use of start-off sections provides a method for determining a grid with user
determined directions of the laths. This makes testing of different grid directions
possible.
• The resemblance between the physical model and the computer model show that the
grid generation technique creates geometry that approximates reality.
• It is uncertain whether the start-off sections as created in Rhino represent the directions
that will occur in reality when shaping a gridshell.
• The script can be set-up with a sequence that checks stress levels based on the curve
angles in the grid members that will occur when the laths are bent to the curvatures
needed.
• The curve check in the design tool shows that a (semi-)spherical surface is hard to
create from an initial straight mat of laths. The laths need to curve and scissor largely to
comply with the surface. This was also shown in the physical form finding of the
gridshell.
• An iterative design process is needed to adjust the initial surface to a surface which is
optimized for a gridshell grid. The graphical setup of the design tool provides information
on whether the maximum bending stress is exceeded in the members. This gives quick
insight in the fitness of the shape for application to a gridshell.
• The use of internal supports to create the desired shape can result in undesired changes
in shape and stress levels in the elements when the internal supports are removed. If
the intended shape is not equal to the natural equilibrium shape of the grid, the
structure will deflect to this equilibrium position when the internal supports are removed.

173
A Design Tool for Timber Gridshells

Changes in stress level up to 23% have been found in this research. Change of shape is
less prominent. Changes of 1,3% or less have been found.
• Stress levels that can be calculated using the angles of curvatures can be used to
estimate the residual stress levels due to the bending of the laths during the formation
of the shape. These stress levels can be used in structural analysis by superposition to
other load cases.
• It is hard to estimate the relaxation deflection by using the difference in shear force that
can be determined by using the bending stresses corresponding to the curve angles. For
a single lath applying these forces as relaxation deflection forces gives a deflection
corresponding to the natural bending shape of a lath. For a 3D structure, less
correspondence was found.

8.3 Recommendations

• It would be largely beneficial if the script could approximate the input surface, complying
with the maximum angles. The way the design tool has been set up now, the grid
generation process does not take account of the properties of the timber when creating
the grid points. The curvature has been checked after the structure has been created.
The shape of the surface has to be adjusted after the grid generation will have been
completed, if maximum curvature angles are exceeded. Form finding has to be
performed by hand. Automatic form finding would enhance the design tool.
• Further testing is advisable to verify the use of the correctness of the start-off sections.
The correctness of the generated grid depends on the correct input of these start-off
sections.
• It is advised to avoid surfaces with an angle of attack on the horizontal larger than
approximately 30 degrees. The results of the design tool and the physical modelling
show that such surfaces are difficult to create using a gridshell structure.
• The output of the design tool should be enhanced. In the current output the nodes and
elements are not ordered well. Especially when the structure is trimmed to remove the
jagged edge, the output becomes discontinuous. The output file then needs processing
before it can be used in a structural analysis program.
• The design tool should be optimized to save computer time. Especially the graphical
determination of the curve angles takes relatively much time. An analytical determination
might save time.
• Further research should be performed to determine a correct load factor on the initial
bending stress resulting from the formation process, by checking the stress level after
relaxation of the timber.
• Further research should be performed on the applicability of the stresses that can be
derived from the angles of curvature, provided by the design tool. Especially the
differences in stress level at the edge nodes that were found in this thesis should be
analysed.

174
 Conclusions and recommendations

8.4 Evaluation of the gridshell design tool

From the results of the grid generation tool, the conclusion has been drawn that the
proposed grid generation method can be used for the determination of the geometry of a
gridshell grid. The gridshell design tool as it is created for this Master’s thesis thus complies
with the demands stated in this thesis The results of the grid generation look promising. The
created grid is not necessarily correct however. This is dependent on the correctness of the
start-off sections. When the created grid was reviewed, it can be stated that a structure was
created that flowed smoothly over the surface. When a physical model was created based on
the generated geometry, similar results were found.
One result found both by physical modelling and the design tool was the difficulties in
forming a (semi-)spherical shape by means of a grid shell structure. The laths could not
bend and twist enough to comply with the curvature needed to form the shape.
Because the two models show similar results, the assumption can be that the use of the
start-off sections is legitimate. However the method is used depends on the correctness of
these sections, as the generated grid is based on this. Further testing is needed to verify the
usability of the sections created by Rhino.
Also deviations have been found between the two models. The shapes of the two models
differed in height and cross sectional shape. The physical model assumed a more parabolic
shape in cross section than the computer model. This can be explained by the fact that a
parabolic shape is a more natural bending shape for a bending member to take on. Similar
effect can also be found with form finding techniques. The parabolic shape can be compared
with the catenary line which is assumed by a chain hanging between supports. In the design
process of a gridshell, the design tool can contribute to the conceptual design stage. First the
tool is used to check if a gridshell is possible in the initially desired shape. When the tool
shows that the maximum stress criteria are exceeded, the shape needs to be adjusted. Also
by adjusting the section curves, an optimal direction for the laths can be searched. When
adjusting the surface, one should also take account of physical form finding principles. It is
therefore advised to determine the shape with a shape optimization software tool.

When a gridshell is deformed into shape, bending and torsion stresses occur in the members.
Due to relaxation of the timber the stress levels will diminish approximately by factor 0,5.
When structurally analysing a gridshell, these stresses should be taken into account. The
bending stresses in the members can be derived from the curve angles in structure. These
angles are part of the design tool output and can therefore be utilized easily. The use of the
bending stresses as an implemented load case has been analysed. This analysis shows that
this is hard to achieve. The theory of deriving relaxation deflection forces from the bending
stresses gives accurate results for a single member, but when a 3D structure is analysed the
results are less accurate. This is probably due to the fact that a 3D structure reacts much
stiffer to an applied load.
It is easier to use the formation stresses as a superposition load case after analysis in GSA.
The stress levels (reduced for relaxation) can be added to the stress levels that will be found
found by analysis of load cases like wind and snow loads. A problem is that in a 3D structure
a complex member interaction takes place. When an element rotates around its y-axis, this
results in a torsion moment in an element perpendicular to it, and vice versa. This results in
jumps in the stress levels. This behaviour is hard to implement when using the curve angles
to determine the stress levels. In the elements close to the edge supports, large deviations
between the stress levels analysed by GSA and the stress levels calculated with the curve
angles have been found. For nodes closer to the middle of the structure, a better
resemblance has been found. Therefore the assumption is that the use of the curve angles to
determine the stress levels can be applied in structural analysis. To take account of deviations
from reality, a reduction factor which gives conservative results is advised. Further research
should estimate this reduction factor.

What also should be taken into account are the relaxation deflection and the related change
in stress distribution. When the gridshell is shaped on internal supports and the shape is not

175
A Design Tool for Timber Gridshells

equal to the natural equilibrium position of the deformed structure, the structure will deflect
when the internal supports are removed. In the analysis a maximum change of 23% in stress
level was found. If this is neglected, breakages can occur during construction. It was tried to
estimate the relaxation deflection using the relaxation deflection forces, derived from the
curve angles, but this result was not found accurate enough.

If the general use of gridshell structures is reviewed, it can be stated that this type of
structure has high potential. Recently built structures prove the gridshell to be a worthy
addition to prestigious structures. The waving shapes that are possible add architectonical
value to the building. It is also possible to design the structure as a sustainable building.
Using timber from sustainable sources make sure the building has highly sustainable value, as
timber is renewable material. Also the fact that the structure is a shell, which has efficient use
of material related to the span, adds to the minimization of material use.
The power of the gridshell is the result of the construction method. From an initial flat mat of
laths a continuous shell structure can be created. A 2D surface can be transformed to 3D,
without disconnecting the elements. In the different grid shell examples, different
construction methods were used. The simplest one was used in the Savill Garden gridshell.
Here, the laths were just simply placed into position lath by lath on formwork. If this is also
the most economic one can be questioned, because this method is labour intensive. On the
other hand, in the other gridshells all laths had to be placed into position in the flat mat and
connected to each other too. An advantage is that this can be performed on a flat working
area. The building method also depends on the size of the structure. The Savill Garden
gridshell is four times as large as the Weald and Downland Gridshell. For large surfaces it is
probably a lot more difficult to deform the entire surface into the desired shape at once.

It is shown in this thesis that not every shape is possible using a gridshell structure.
Especially (semi-)spherical surfaces are problematic. The laths that are connected to each
other in the nodes need to bend and scissor beyond their minimum bending radius to create
such a surface. The structure is constrained in shaping by the timber properties and by the
fact that the elements are connected to each other. When the constraint of the attached
nodes is released, it might be possible to create different surfaces. If a mat of laths is
imagined where the laths are only connected to each other in parts of the structure, e.g. in
the parts which would be the tops of anti-clastic parts, different geometry might be created.
If the laths at the edges are free to move in plane of the surface, this might make other
geometries possible. This would be possible by using a modified version of the connector
used in the Weald and Downland gridshell. In the current structure, a pin restrains the middle
laths from moving. If this pin is omitted, the constraint of equal distance between the nodes
is released and the nodes are free to move within certain boundaries.

176
Conclusions and recommendations

177
A Design Tool for Timber Gridshells

178
 References

References
Barnes, M.R. 1999, ‘Form Finding and Analysis of Tension Structures by Dynamic Relaxation’,
International Journal of Space Structures, Vol. 14, No. 2, 1999

Blass, H.J. Aune, P., Choo, B.S., Görlacher, R., Griffiths, D.R., Hilson, B.O., Steck, G. (eds)
1995, Timber engineering STEP 1: basis of design, material properties, structural
components and joints, Centrum Hout, Almere.

Boer, S. & Oosterhuis, K. 2002, Architectural Parametric Design and Mass Customization
available at
http://www.oosterhuis.nl/quickstart/fileadmin/Projects/129%20the%20web%20of%20
north%20holland/02_Papers/000-040603-ECPPM.pdf

Booth, L.G. 1964, ‘The strength testing of timber during the 17th and 18th centuries’ in
YEOMANS, D (ed.) 1999, The Development of Timber as a Structural Material, Ashgate
Publishing Limited, Hampshire, pp. 211-236.

Booth, L.G. 1971, ‘The development of laminated timber arch structures in Bavaria, France
and England in the early nineteenth century’ in Yeomans, D (ed.) 1999, The
Development of Timber as a Structural Material, Ashgate Publishing Limited,
Hampshire, pp. 291-304.

Burkhardt, B. et al. (eds) 1978, IL13: Multihalle Mannheim, Institut für leichte
Flächentragwerke, Stuttgart.

Coenders, J.L. 2004, ‘Structural Form Finding and Structural Optimization Techniques’, in
Form Finding and Structural Optimization, MSc thesis, Delft University of Technology

Courtenay, L. & Mark, R. 1987, ‘The Westminster Hall roof: a historiographic and structural
study’ in Yeomans, D. (ed.) 1999, The Development of Timber as a Structural Material,
Ashgate Publishing Limited, Hampshire, pp. 127-150.

GeoDomeDesign, 'Planetarium Artis', [brochure]

Groot, de, H. 2007, ‘Rechthoekig Gebogen, Savill Garden Visitor Centre, Windsor Great Park
London’ in Het Houtblad, jaargang 19, nummer 2, Het Houtblad BV, Almere

Holzbau Konstruktionen, Hölzernes Hängedach Bundesgartenschau in Dortmund, [Brochure],

Sonderdruck für die Arbeitsgemeinschaft Holz e.V. aus DETAIL, Zeitschrift für
Architectur + Baudetail.

Happold, E & Liddell, W.I. 1975, Timber lattice roof for the Mannheim Bundesgartenschau,
The Structural Engineer, No. 3, Volume 53, March 1975

Harris, R. & Kelly, O., ‘The structural engineering of the Downland gridshell’ in Parke, G.A.R.
& Disney, P. (eds) 2002, Space Structures 5 volume 1, Thomas Telford, London

Harris, R., Romer, O.,Kelly, O. & Johnson, S. 2003, ‘Design and Construction of the Downland
Gridshell’ in Building Research and Information, Volume 31, Number 6, pp. 427-454,
Routledge, part of the Taylor & Francis Group

Hartsuijker, C. 1996, Mechanica van Constructies 1b, deel 2, Reader TU Delft

179
A Design Tool for Timber Gridshells

Hartsuijker, C. 1999, Toegepaste Mechanica deel 1, Evenwicht, Acedemic Service,

Schoonhoven

Heymans, J. 1967, ‘Westminster Hall roof’, in Yeomans, D. (ed.) 1999, The Development of
Timber as a Structural Material, Ashgate Publishing Limited, Hampshire, pp. 151-176.

Hoefakker, J.H. & Blaauwendraad, J. 2005, Theory of Shells, Course CT5143, Delft University
of Technology, Delft

Horn, C. & W. 1973, ’The cruck-built barn of Leigh Court, Worcestershire, England’ in
Yeomans, D. (ed.) 1999, The Development of Timber as a Structural Material, Ashgate
Publishing Limited, Hampshire, pp. 1-25.

Jensen, F. 2000, Erection Procedure for the Downland Gridshell, MSc thesis, University of
Bath

Kelly, O.J. 2003, Harris, R.J.L., Dickson, M.G.T. & Rowe, J.A. 2001, ’Construction of the
Downland Gridshell’, The Structural Engineer, No. 17, Volume 79, September 2001

Kuilen, van de, J.W.G & Vries, de, P.A. 2005, CT3051A Constructieleer 3A, Reader, Delft
Technical University

Leupi, J. 2002, ‘Parametric design for the structural elements of timber rib shells’, Space
Structures 5 volume 1, Thomas Telford, London

Lewis, W.J., Tension Structures, form and behaviour, Thomas Telford, London.

Müller, C. 2000, Holzleimbau, Laminated Timber Construction, Birkhäuser, Basel; Berlin;

Boston.

Otto, F. et al. 1982, Natürliche Konstruktionen, Deutsche Verlags-Anstalt, Stütgart

Pestman, J.H., Vormgeving in hout, C.A. Spin en Zoon N.V., Amsterdam

Ross, P. 2002, Appraisal and Repair of Timber Structures, Thomas Telford Publishing,
London.

Thelandersson, S. & Larsen, H.J. (eds) 2003, Timber Engineering, John Wiley & Sons, LTD,
Chichester.

Van Drenth Groep, Uit het juiste hout gesnede [Brochure]

Wilhelm, K. 1985, Architekten heute, Frei Otto, Quadriga-Verlag Severin, Berlin.

Yeomans, D (ed.) 1999, The Development of Timber as a Structural Material, Ashgate

Publishing Limited, Hampshire.

Young, W.C. & Budynas, R.G. 2002, Roark’s Formulas for Stress and Strain, 7th edn, McGraw-
Hill, New York

180
References

181
A Design Tool for Timber Gridshells

182
 List of symbols

List of symbols
κ = curvature of the beam;
κ1 = first principle curvature;
κ2 = second principle curvature;
κg = Gaussian curvature;
σn = normal stress;
σm = bending stress;
fu = ultimate stress strength;
σm,d = the design bending stress
fm,d = the design bending strength
σm,y/z,d = the design bending stress in the y or z direction;
fm,y/z,d = the design bending strength in the y or z direction;
σc/t,90,d = the design compression/tension stress perpendicular to the grain;
fc/t,90,d = the design compression/tension strength perpendicular to the grain.
σc/t,0,d = the design compression/tension stress parallel to the grain;
fc/t,0,d = the design compression/tension strength parallel to the grain;
σo = stress level at the outer edge of a curved beam;
σi = stress level at the inner edge of a curved beam;
M = bending moment;
E = modulus of elasticity;
I = moment of inertia;
W = moment of resistance;
R = radius of curvature;
Rin = the inside radius of a beam;
lo = outer length of a curved beam;
Δlo = change of outer length of a curved beam;
li = inner length of a curved beam;
Map,d = the design moment in the apex zone;
b = width of the member;
h = height of the member;
hap = height of the beam in the apex zone;
αap = angle of taper;
kl = reduction factor for design bending stress in curved beams;
kr = a reduction factor which takes into account the bending stresses in the laminates of
curved glued laminated timber due to production;
km = factor which makes allowance for re-distribution of stresses and the effect of
inhomogeneities of the material in a cross section;
t = thickness of a laminate;
c = constant by which the minimum radius can be estimated;
Vd = the shear force;
A = the cross section of the element;

183
A Design Tool for Timber Gridshells

184
 Appendices

Appendices

185
A Design Tool for Timber Gridshells

186
 Appendices

Appendix 1: Determination of the maximum bending

radius
To analyse bending behaviour, a segment of a member is considered shown in Figure A.1.
The member is subjected to a bending moment, which results in internal stresses,
deformation and deflection.

First let's consider the kinematics relations, linking deformations and displacement
(Hartsuijker, 1996)

Δx

M Δw

θ
Figure A.1: Bending member

According to the hypothesis of Bernoulli, if deflections are considered to be small sections

perpendicular to the member's axis stay perpendicular when the member deforms. For the
angle of rotation of the section we can thus write:

dw
θ =− ( A.1 )
dx
where:
θ = rotation of the section about the y-axis
w = displacement in the z-direction
x = distance along the x-axis

The angle θ is also the angle of the strain diagram. The strain in the fibres due to
deformation can be written as:

ε ( z) = κ z z ( A.2 )

with:

dθ
κ= ( A.3 )
dx
where:
κ = gradient of the strain diagram
ε(z) = strain over the height of the beam
z = distance from the member axis to the outer fibres

κ can also be defined as the curvature of a curve. Mathematically this is defined as the
change of direction angle of the tangent per curve length:

187
A Design Tool for Timber Gridshells

Δθ dθ
κ = lim = ( A.4 )
Δs → 0 Δs ds

where:
κ = curvature
s = curve length

The curve length s of a fibre in a beam is dependent on the strain of that fibre (equation
A.4). Therefore also s is dependent on z, which is the distance from the referred fibre to the
member axis. When the strains are considered small (ε(z)<<1), it can be neglected that the
strain that the strain differs over the height of the beam and s can be considered equal over
the height of the member. Equation A.4 can then be written as:

Δθ Δθ d θ
κ z = lim = lim = ( A.5 )
Δs → 0 Δs Δs →0 Δx dx

This shows that the curvature of the beam is equal to the gradient of the strain diagram. The
absolute value of the curvature reciprocal is the curve radius R:

1
R= ( A.6 )
κ

Next, constitutive relations are considered which link the section forces to the deformation.
Together with Hook's law and equation A.2 the curvature can be calculated as a function of
the bending stress:

σm
σ m = Eε = Eκ z z → κ= ( A.7 )
Ez

When equation A.7 is substituted in equation A.6, the bending radius of the member can be
calculated as a function of the bending stress. By using the maximum bending stress of the
material fm, the minimum bending radius of the member can be calculated:

zE
Rm = ( A.8 )
fm

Combining equation A.1, A.6 and A.8, the maximum angle of curvature becomes:

xf m
θ max = ( A.9 )
zE

When this is re-written to Rm=c*d, Rm can be calculated by multiplying the height of the beam
d (=2*z) by a constant c:

E
c= ( A.10 )
2 fM

For timber class C and D these values can be calculated:

C24: R0=229d D30: R0=167d

C27: R0=213d D35: R0=143d
C30: R0=200d D40: R0=138d
C35: R0=186d D50: R0=140d
C40: R0=175d D60: R0=142d

188
 Appendices

As a rule-of-thumb, the average of c can be used. For C-class timber this is 200. For D-class
this is 145. This average value can be used for checking the preliminary design. The final
design has to be checked with the corresponding strength value of the timber used.

For use in a gridshell, a timber grade is desired which is able to bend to the smallest
curvature needed to accomplish the desired shape without breaking. This can be translated in
a ratio E/fm as low as possible, i.e. a low bending stiffness combined with a high bending
strength.

To analyse the influence of the E/fm ratio, this ratio can be compared with the maximum
possible bending radius. For this analysis, the timber strength classes with corresponding
material properties are used according to the code EN338. This calculation is made for timber
of class C24 to D60. The results are plotted to the corresponding class ultimate bending
strength in Figure A.2.
The use of properties according to timber strength classes can be questioned, as the
properties of a timber beam do not have to be equal to values of its assigned strength class.
A timber population is assigned to a strength class, when its characteristic values of bending
strength, modulus of elasticity and density are equal to or greater than the corresponding
values of the related strength class. These characteristic values are determined as the
weighted means of the lower 5-percentiles for strength properties and density, and as the
weighted mean for modulus of Elasticity (Blass et al, 1995).
Therefore, the values of the properties of a beam are likely to be different than the values of
its assigned strength class. However, in this analysis the values of the different grades are
used to give an insight in the bending behaviour related to the material properties.

1.4

1.3
Indexed value

1.2

1.1 (E/fm)/( E/fm)0 D-class

(E/fm)/(E/fm)0 C-class
R/R
R/R00 D-class
D-class
(E/fm)/( E/fm)0 C-class
(E/fm)/(E/fm)0 C-class
1 R/R
R/R00 C-class
C-class

0.9

0.8

0.7
20 25 30 35 40 45 50 55 60 65
Bending strength (N/mm2)

Figure A.2: E /fm ratio and R, indexed to the values of timber class C24 for C-class timber and to the values of D30 for
D-class timber, plotted to the corresponding grade ultimate bending stress.

To compare the moment needed to accomplish the maximum bending radius for different
timber grades, the maximum radius R and the maximum moment M for timber of class C24
to D60 are plotted to the corresponding class ultimate bending strength in Figure A.3. The
values are indexed again to the values of C24 for the C-class timber and to the values of D30
for D-class timber.

189
 A Design Tool for Timber Gridshells

2.2

2.0
Indexed value

1.8

1.6
R/R
R/R00 D-class
D-class
M/M
M/M0 0 D-class
D-class
1.4
R/R
R/R00 C-class
C-class
R/R
M/M00 C-class
C-class
1.2

1.0

0.8

0.6
20 25 30 35 40 45 50 55 60 65
Bending strength (N/mm2)

Figure A.3: R0 and Mu, indexed to the values of timber class C24 for C-class timber and to the values of D30 for D-
class timber, plotted to the corresponding grade ultimate bending stress.

From these figures it can be concluded that for timber of coniferous species (C-class), it is
beneficial to choose a timber quality with high bending strength with respect to the
curvature. Figure A.3 shows a moment M increasing proportionally to the bending strength fm
and a maximum curvature radius R decreasing disproportionably to fm. A higher bending
strength corresponds with a lower modulus of elasticity and therefore a lower maximum
radius is possible. This allows a stronger curved surface.

Reviewing deciduous species (D-class), Figure A.2 shows an optimum of the smallest bending
radius, coinciding with the optimum of the highest fm/E ratio. This optimum corresponds with
a timber strength class D40. For timber of deciduous species it can be concluded that
choosing timber with material properties corresponding to a higher bending strength than
class D40 is not beneficial. A higher strength class will not lead to a decreasing bending
radius, due to the fact that the fm/E ratio does not increase. This can also be seen in Figure
A.3. A higher bending strength implies a higher maximum bending moment, but the minimum
bending radius does not decrease above class D40.

190
 Appendices

The maximum torsion angle can be calculated with the formula in 3.2.4.3. This is performed
with the timber properties corresponding with timber class C24 to D60. These are plotted to
the corresponding strength class in Figure A.4. The figure shows that an increasing timber
class does not necessarily imply a larger torsion capacity. With D-class timber, there is a
maximum at strength class D40 of 0.16 rad. This coincides with the optimum of the fm/E ratio
(Figure A.2).

0.45
Torsion angle (rad)

0.4

0.35

0.3
C-class
D-class
0.25

0.2

0.15

0.1
20 25 30 35 40 45 50 55 60 65
Bending strength (N/mm2)

Figure A.4: Maximum torsion angle plotted to the bending strength

191
A Design Tool for Timber Gridshells

Appendix 2: GenerativeComponents
GenerativeComponents is a parametric associative design system, which combines direct
interactive manipulation of design methods with visual and traditional programming
techniques34. GC is a plug-in for Bentley System’s Microstation and it was still a Beta version
at the time of use for this Master’s thesis. The consequence of this is that there are bugs in
the program to be expected. A big disadvantage is the lack of a user guide, which means
everything has to be learnt from other users’ experiences and trial and error. The high
potential of the software makes it attractive to use however.

With GC, it is possible to create your design based on adjustable parameters by using built in
features. For example, when designing an office building, Storey height could be assigned as
a parameter. By simply adjusting this parameter, your design is adjusted to the new storey
height. It is also possible to build your own features. For example, it is possible to create a
‘Storey’ feature consisting of an entire storey including columns, façade, ceiling etc., and
attach a parameter ‘number of floors’ to it. By simply changing this parameter, more storeys
will be added to your design.

For this thesis, using a parametric setup for the structural grid is a great advantage. The
mesh size of the grid is a key parameter for this model. The idea is to set up the model with
a few other parameters as well. By using the dome height and radius, the design can be
adjusted in size and height. The angle between the horizontal plane and the tangent to the
bottom of the surface can be a parameter, so more complex shapes can be reviewed.
Furthermore, the base angle between the grid elements can be variable as well, to be able to
experiment with different starting angles of the grid.
The declared parameters can be adjusted by moving sliders between user defined intervals
(Figure A.5). The graphics update themselves in real time. Figure A.6 to Figure A.9 show the
changing graphics when parameters are changed.

Figure A.5: Graph variables user interface. Parameters can be adjusted by moving sliders.

34
http://www.smartgeometry.com/tech.htm accessed 10-10-06

192
 Appendices

Figure A.6: Start position. Figure A.7: changing ground tangent.

Figure A.8: Changing dome height. Figure A.9: Changing dome radius.

Figure A.10: Imported points. Figure A.11: Curves trough points.

Figure A.12: Surface trough curves.

193
A Design Tool for Timber Gridshells

After exploring the possibilities of GC, the aimed design should be imported into the
application for processing. A compatibility problem with Rhinoceros raises, as it is not possible
to import the Rhino model into GC. To be able to do this, the following steps were necessary.
First, in Rhino sections have to be made of the model. Then, a series of points has to be
created on each section and these have to be exported to a text file. By importing these text
files, the sections can be redrawn GC. The surface can be recreated by lofting a surface
through these sections. Figure A.10 to Figure A.12 show the created surface in GC.

The way the sections are created is also of importance. The proceeding points on each
section have to be in the same direction; otherwise the lofting process will produce a
distorted surface. Also discontinuity in a section is not wanted, as the curve trough the
imported points needs to be continuous for the lofting process. The tube-like entrance in the
design is such discontinuity, as it is not possible to create continuous sections trough the
shape. As stated in Section 5.3, the entrance is left out for analysis.

Next step is to project the grid points on the surface. The variable grid angle is to be
implemented in this step. A first try-out is projecting the intersection points of a horizontal
grid on the surface. This horizontal grid is changeable in angles by moving one of the two
pivot points (Figure A.13 & Figure A.14).
Because this projection method does not create the wanted grid with uniform element length,
the sphere method is tested in GC. Two spheres with radius x are created with their centre
on the surface (Figure A.15). The intersection curves of the spheres with the surface are
needed to locate the two wanted intersection points of the spheres and the surface. The
software was unable to do this however, creating a problem for this grid generation
technique. A different way of creating the surface is tested. By using standard elements, like
parts of spheres, the surface should be easier to process (Figure A.16). At the same time, by
using standard elements implementing the variables like dome height and tangent angle
should possible. However, using a simpler surface does not solve the problem.

194
 Appendices

Figure A.13: Changeable grid. Figure A.14: Points projected on surface.

Figure A.15: Spheres on surface. Figure A.16: Parts of domes.

Review GC

GenerativeComponents is a promising software application with a lot of possibilities. The

parametric set up makes it possible to change your design easily, even when it is in a
developed state. When changing a parameter, the 3D view is updated in real-time so the
changes are visible immediately. Also the possibility of creating your own feature gives the
designer a powerful tool in CAD.
However, the problems with surface modelling and surface processing which the author ran
into during the stage of study and try-out are a major obstacle for the grid development
process. The Beta status of the application is probably the cause of the problems found, as
there are probably still bugs in the software which makes it hard to process complex
surfaces. Also the lack of a users guide does not ease the use of the application. Because of
these problems, it is decided to stop using GC and to switch to Rhinoceros.

195
A Design Tool for Timber Gridshells

196
 Appendices

Appendix 3: Proof of R >> mesh size

The assumption that the difference in length between the curved element and the straight
line can be neglected is proved by the following calculation:

When a timber lath of 50*50mm with timber grade D30 is assumed, the minimum bending
radius R is 167*50= 8,35 m, according to Section 3.2.4.1. The mesh size is assumed to be
500mm.

The length s of a curve can be calculated with:

s
⎛l ⎞
θ = sin ⎜ ⎟
−1

⎝R⎠ l

s =θR
R
Where:
θ
s = curve length
l = mesh size
R = Curve radius
θ = angle of curvature

With l = 500mm and R = 8,35m, s becomes:

⎛ 500 ⎞
s = sin −1 ⎜ ⎟ *8350 = 500, 299mm
⎝ 8350 ⎠

The difference between mesh size l and curve length s is 0,299mm. This difference is 0,06%
and is small enough to be neglected.

197
A Design Tool for Timber Gridshells

198
 Appendices

Appendix 4: Problems encountered in developing the grid

generation tool
Different problems are encountered developing the scripts. The problems encountered are on
one hand related to the scripting itself and on the other hand related to the way Rhino
handles the surfaces creates intersections. The script related problems are about finding the
right VBScript or RhinoScript command. A lot of registered commands that can be used in
Rhino at the command prompt are implemented in RhinoScript as standard features, but not
all. Commands that are not available as a feature can be executed by using the
"Rhino.Command" command. This runs a registered Rhino command as if it has been typed
in the command prompt35. The example below uses a standard feature and the
Rhino.Command to create an intersection between a sphere and a surface. The sphere and
the surface have already been identified by the script as "arrSphere(j,i)", which is the sphere
at array position (j,i), and "Surface". "Rhino.SelectObject" selects the sphere and the surface.
The intersection is created by adding the Rhino command "Intersect" to the RhinoScript
command "Rhino.Command". "Intersect" is now executed with the two selected objects.

Rhino.SelectObject arrSphere(j,i)
Rhino.SelectObject Surface
Rhino.Command ("Intersect")

The problem of this method is what is returned in the script as result. Rhino.Command
returns a Boolean, true or false, indicating success or failure of the command. What is
needed however to be able to use the intersection created in Rhino, is the identifier of the
created object. In most cases it is sufficient to select the last created object, but here a
problem arises in the way Rhino handles surfaces and creates intersections.

When the intersection curve between a sphere and the surface is made, Rhino creates one
closed curve in most cases. The single object can be identified and used by the script. In
some cases however, this process creates more than one object. If the last created object is
selected for further processing, the wrong object is probably used. The next problems are
encountered with the intersection procedure:

• When a sphere is intersected by a discontinuity in the surface, like the edge of two
different surface parts, Rhino encounters problems with the intersection process
(Figure A.17). The created intersection is not continuous and can consist of several
open curves. These curves need to be joined to create a closed curve (Figure A.18).
• Close to a discontinuity, the intersection is not created as a closed curve.
• Close to a discontinuity a curve is created with a loop. Until this problem was met,
the curves were selected by open or closed status. The loop creates two closed
curves, one of them being very small (Figure A.19). This creates a selection problem,
as both curves are closed.

35
RhinoScript Help file

199
A Design Tool for Timber Gridshells

extra curve

Discontinuity

Figure A.17: Intersection problems occur at Figure A.18: At the left two intersection parts are
discontinuities in the surface created: a closed curve and an extra curve right to the
discontinuity.

Loop

Discontinuity

Discontinuity
Figure A.20: An open curve is created caused by the
Figure A.19: Zooming in on the right intersection curve discontinuity in the surface
of Figure A.18, near the discontinuity. A loop is created
to close the curve.

Also problems were encountered creating the intersections between two curves. The result
of this operation should be two intersection points. This is not always the case. The following
problems were found:

• One or more small curves are created in stead of a point. This is a problem related to
the discontinuity of the surface.
• More than two points are created. During one test runs, more than 10 points were
created at one intersection. This is probably an error in the intersection process.

These unexpected intersection problems lead to a trial and error approach to filter out all
options and slow down the process of developing the script. For every newly found error, the
script has to be modified to rule out all incorrect options. This also makes the selection
procedure difficult. If more than one object is created, it is not possible to simply select the
object created last. Every object created in one command has to be checked.

The main reason for a lot of problems is discontinuities in the surface. The shape which has
been used to test the script contains quite some discontinuities when it is used as a poly
surface (see Section 5.3). When the surface is simplified and recreated as a single NURBS
surface, a lot of intersection problems can be obviated.

200
Appendices

201
A Design Tool for Timber Gridshells

202
 Appendices

Appendix 5: Physical modeling

1. Introduction
Building a physical model in an early stage of design provides insight in shape, possibilities
and structural behaviour. The later is most important when it concerns a structure with
complicated structural behaviour such as a gridshell structure. With the gridshell, also the
construction phases can be modelled and reviewed, as the same building method is used for
a real size gridshell as for the model.
This model described here is intended as a geometrical model. It will be used to compare the
results of the computer model with reality. Deviations are to be expected between the
computer model and the physical model. These will be caused by inaccuracies of the physical
model and, probably more important, because the shape of the physical model will be
dictated by laws of nature and not by the constraints of the computer model. The shape of
the physical model is therefore unlikely to coincide precisely with the computer model, which
a 3D representation of the shape desired by the architect. Building the model is therefore also
physical form finding. The results of the physical model can be used to adjust the computer
model to a more natural shape. This will result in a more efficient structural behaviour.
In this chapter the process of creating a physical model is described. The geometry of the
adjusted structure of Section 6.4.5 is used to create the model and the physical model will be
compared with the geometry of this model.
The materials which are used for the model are discussed first after which the building
process is described in Section 3. Third the observations made are described. After the
conclusions are presented in Section 5, the chapter is ended with the description of the
conceptual model for the eggoid shaped gridshell.

2. Materials
To reach a desired accuracy and correct modelling of the structure, materials have to be
selected with care. The properties that have to be modelled correct:

• scissoring of the laths: flexible nodes are needed

• mat layout: continuous members are to be used
• building method: initial mat has to be flat
• bending: the bending behaviour of the modelled members should be approximately
the same as the members in real. This implies bending resistance of equal
magnitude in the two directions perpendicular to the member axis and similar
torsion behaviour
• bending strength: model members need to be able to bend to the desired scaled
curvature without breakage or plastic deformation

For a geometrical model it is less important to scale all the characteristics of the structure in a
correct way, since there is no intention to measure deflection or forces.

203
A Design Tool for Timber Gridshells

The following materials have been used to build the model (Figure A.22):

• For the members a steel welding rod of 1mm diameter is selected. This material has
adequate bending behaviour. The rods can bend without breakage, but care has to
be taken for yielding. A simple bending test proved that the desired curvature can be
reached without yielding, so no problems are to be expected. Downside of this
material is that the circular cross section has equal bending stiffness in every
direction in stead of just y and z direction. The difference between Iyy of a square
cross section and I of a circular cross section of equal size is 5,5% (see Figure A.21).
It is assumed that the bending behaviour of a square cross section with torsion
capacity is approximated sufficiently by the bending behaviour of a circular cross
section.

• A flexible node can be created by rubber bands. A simple double knot, firmly
tightened, provides the desired flexibility of the mat of rods. The materials roughness
provides enough resistance against slipping if tightened sufficiently.

• As a base, MDF of 12 mm thick is selected. This material provides a smooth and flat
surface and is very suitable for pinning the structure down with nails or pins.

• Nails and pins, for pinning down the structure at the edges

h R

y
y

z
z

I yy = 1 h4 I zz = 1
4 π R4
12

A = h2 A = π R2

Acircle = Asquare R =1 → h= π

I yy , square = 1 π 2 = 0.8225 I yy ,circle = 1

4 π = 0.7854 = 0.955*Iyy,square
12

Figure A.21: difference in Iyy between a square and circular rod of equal cross section in size

Figure A.22: Materials used for geometrical model

204
 Appendices

3. Formation process
In this section the formation process is illustrated in steps.

• A mat of steel rods is tied with rubber bands. The dimensions of the mat and lengths
of the rods are determined by using a template print-out of the flat structure (Figure
A.23). This template is created by using data generated by the generation script.
Using the base coordinates which are in the output text file, the lengths and layout of
the flat mat can be plotted. By placing the rods on the template print-out, the rods
can be tied at the correct position and equal distance from each other (Figure A.25).

• The rods are cut to the correct length.

• The ends are bent into eyes for ease moving and positioning of the members.

• Rubber bands are tightened and cut short.

• The mat is placed on the MDF base and on a top view template (Figure A.24 and
Figure A.25) of the structure in final position. By using the template, the end position
of the member ends is determined. Also a visual check of the curve of the members
can be performed after the formation process.

• Rods are placed through the end eyes, divided in sections of 10 to 20 cm. By doing
this, not every node itself has to be moved and pinned (Figure A.26).

• On the long side, the mat is pinned at two points which are into place already (Figure
A.26).

• The structure is tensioned by pinning down the edges on the short sides and then
pushing inward from the free long edge (Figure A.27). A relatively large force is
needed to create an initial curvature. This is due to the fact that the members have
to buckle into shape.

Figure A.23: Template of the flat mat Figure A.24: Template of the deformed structure

205
A Design Tool for Timber Gridshells

Figure A.25: Flat mat of rods is lied out on a template print-out

Figure A.26: Edge rods are added. The structure is pinned at two points which are already on their final location..

Figure A.27: The structure is tensioned by pushing on the long edge and preventing movement at the short edges.

206
 Appendices

Figure A.28: A vaulted shape is formed. Figure A.29: By pushing from the sides, domes are
shaped.

Figure A.30: Pushing from the sides to form domes

207
A Design Tool for Timber Gridshells

• The structure is further tensioned by pushing the long edge inward. A shallow
vaulted shape is created (Figure A.28).

• The structure is now properly tensioned for the further forming process. First the
domes are formed by pushing in diagonal direction (Figure A.29 and Figure A.30).

• By going in circles around the structure, pushing inward bit by bit, the curvature of
the structure gets larger in a controlled manner (Figure A.31, Figure A.32 and Figure
A.33). It is found that pushing inward perpendicular to the length axis of the
structure is relatively easy. The formation of the domes requires larger force.

• On the edge which was pinned in the beginning, all the edge sections reach their
final position quite soon in the formation process. The anti-clastic part can now begin
to form properly (Figure A.31).

• An intermediate assessment shows that the left dome is almost finished in terms of
position of the edges. The right dome needs to move inward further (Figure A.34).
Also a larger curvature is needed going down to the edge supports (Figure A.36). The
saddle between the two domes also needs a larger curvature and is already in too
high position (Figure A.35 and Figure A.36).

• Slowly pushing the edges of the domes to their intended end position the curvature
gets larger. The left dome reaches its final position without much trouble. For the
right dome it takes quite a lot more force however to reach the desired curvature and
to push the edges to their intended final position. The final shape is shown in Figure
A.37 to Figure A.39.

Figure A.31: Pushing the edges inward going around in circles shapes the structure. At one edge the final position is
reached quite soon (dashed line). The curved S-shape of the members becomes visible.

208
 Appendices

Figure A.32: Domes and saddle are shaped Figure A.33: Further shaping of the structure

Figure A.34: Shallow dome (left) formation is almost finished. The formation of the stronger curved dome (right)
needs to move upward to reach the desired height.

Figure A.35: Left dome approaches final shape, although Figure A.36: Right dome must become higher and
the saddle is to high curved stronger. The saddle is already too high.

209
A Design Tool for Timber Gridshells

Figure A.37: After further pushing the edges inward, the edges reach final position.

Figure A.38: Left dome with edges near final position Figure A.39: Right dome with edges near final position

210
 Appendices

4. Observations
Shape
First, the computer model and the physical model are
compared in terms of shape. The height and position
of nodes in the physical model are measured at
normative sections shown in Figure A.41. The height
of the nodes' position is determined by using a
plummet to determine the length of the plumb line.
The horizontal location along the section line is
measured by the distance of the base of the plumb
line to the edge of the structure (Figure A.40). From
these measurements and the coordinates of the
corresponding nodes from the computer model, the
sections shown in Figure A.41 are created. This
shows that the physical model has too much height
compared to the computer model. Section AA' over
the shallow left dome shows that in this section the
two models coincide quite well. The physical model is
slightly higher and shows a more parabolic curve,
Figure A.40: Measuring the plummet line
whereas the computer model shows a circular arched
cross section.

Looking at section CC' a larger deviation is shown. The computer model shows a circular
arched cross section of a smaller radius than cross section AA'. The physical model shows a
parabolic shape again, but the deviation from the computer model is larger. It should also be
noticed that there is also a horizontal deviation. The top of the parabola is situated more to
the left than the top of the arc and the right side of parabola is more flattened than the left.
This is also shown in the longitudinal section DD', where the right end side has a more
flattened shape than the section of the computer model. Figure A.42 shows the difference
between the two models.

As a result of the parabolic shape and larger height of the structure, also section BB’ is higher
than the computer model. Section BB’ also shows a different shape. At the right side the
section approaches the support with a larger curvature and touches down at the support with
an almost vertical angle.

211
A Design Tool for Timber Gridshells

B C

Section AA'

D
D'

Section BB' B'

A'
C'

Section CC' Section DD'

= Physical model = Computer model

Figure A.41: Sections of the two models

Figure A.42: Difference in curvature between the physical model and the computer model in the right dome.

212
 Appendices

Curvature
Second, the curvature of the individual members is reviewed. For the two domes different
observations are made. First the shallow dome is reviewed (Figure A.43). The computer
model of this part of the structure coincides with the physical model within acceptable
accuracy, corresponding with the observations of the shape. Seen in top view, the members
of the physical model have only a small deviation from the computer model. The moderate S
shape of the members which is seen in the computer model also occurs in the physical
model. Also the end points of the member coincide with their position in the computer model.
Moving to the right of the structure, the deviation between the computer model (grey line)
and the physical model (black line) becomes larger. This also corresponds with the shape
observations, as the deviation in shape from the computer model was already found to differ
more as we move to the right part of the structure. In these figures an inaccuracy must be
taken into account, because of the difference between the perspective view of the photo and
the isometric computer plot.

When the smaller dome is reviewed, a larger deviation is noticed (Figure A.44). This
corresponds with the shape analysis of the models. Especially in the transition zone between
anti-clastic and clastic shape, the members of the physical model show less curvature in
horizontal direction than the members of the computer model. This relates especially for the
members which run from top left to bottom right in Figure A.44. More to the right in the
structure the curvature of the members in the model coincides better again.
This difference in curvature and also the difference in height and shape of the dome can be
explained as follows. First, the member ends do not coincide with the support positions of the
computer model. This is relates especially for the members ending in the top right of Figure
A.44. As the length of the rods is bent with a smaller distance between the end nodes, the
arcs become higher. Second, the cross section in the anti-clastic part has a much steeper
angle with the base, as the computer model does (see Figure A.41). This has the effect that
the nodes in the transition zone are shifted in horizontal direction.

What is also found in the grid generation using the curvature check, is the difficulties in the
ends of the structure. According to the script, the members cannot meet with the curvature
needed to create the semi-spherical shape. This is also shown in the deviations between the
computer and the physical model. Especially in the smaller dome, the laths have difficulties
bending and scissoring to the desired end position.
It is possible to approximate the shape of the computer model with a better fit. The physical
model can be modified by moving the ends of the members closer to their desired end
position by force (see Figure A.45). All nodes then shift to a position fitting better with the
computer model. The force needed to pull the nodes into position results in a tension force
which extends over the structure in diagonal direction. This is the encircled region in Figure
A.45. This results in a lower shape and a member curvature corresponding better with the
computer models height and shape. Especially in the anti-clastic zone, the structure is pulled
down by the tension in the rods. However, by doing this the members are forced to a position
which exceeds the maximum curvature.

213
A Design Tool for Timber Gridshells

Figure A.43: Comparison of the left dome in the two models. The grey lines indicate the computer model

Figure A.44: Comparison of the right dome in the two models. Moving the members in the pointed directions (arrows
in top right) will result in a better approximation of the computer model.

214
 Appendices

Figure A.45: Structure, with the nodes at top right side moved in the direction of the arrow. Tension exists in the
encircled area.

215
A Design Tool for Timber Gridshells

5. Conclusions
The deviations found between the computer and physical model can be well explained. The
cross sections deviate from circular arc to more parabolic shape. This shape is generally
considered as a more natural shape to divert axial forces trough an arched structure. Because
of this tendency to parabolic shapes, the horizontal S-shape in the members is less prominent
and as a result of this the physical model has larger height as intended.

Although the physical model deviates from the computer model, this does not mean that the
computer script creates a grid which is incorrect. By moving the members ends to their
computed position the shape and curvature of the computer model is approximated better.
However to do this force is needed to pull the members in position and tension is created in
the members which cross the anti-clastic part of the structure. This can be compared with the
tensioning of the fabric of a hyperbolic tent structure.

The fact that the models deviate does show that the computer model does not represent a
shape that is optimized in structural behaviour. The model should be of a more parabolic
shape. As stated in the introduction of this chapter, building the physical model is a type of
physical form finding. Therefore the results of the physical model can be used to modify the
computer model to a more optimized shape. One could say this is a reversed form finding
process. In stead of first creating a shape by form finding and then creating a structure, here
the structure is created first by the grid generation script. The shape is then adjusted to a
shape corresponding with the physical model. A new grid can then be generated on the
adjusted, optimized shape for structural analysis.

With the grid generation tool it was found that bending curvatures become too large when
the laths of the grid are bent to a (semi-)spherical surface. This problem is also found in the
physical modelling process. It is therefore advisable to avoid such shapes or to take additional
measures, such as applying members with a smaller cross section in the problematic area.

216
 Appendices

6. Conceptual model of the Eggoid

As stated in section 5.3 it is interesting to investigate the possibilities of an eggoid shaped
gridshell. To do this a conceptual model is built. The materials used for this model are:

• For the grid members strips of VIVAK are cut. VIVAK is a plastic material which has a
high bending capacity around its weak axis before failure. Therefore it is very suitable
for testing different geometries. Downside is that the difference between bending
capacity in different directions is large, due to the rectangular cross section.

• To create flexible nodes, binding wire is tied through holes in the VIVAK (Figure
A.46). With a hot needle these holes are easily pierced in the strips.

• As a base a rigid cardboard box is used. An ellipse is cut out of the base, to act as an
edge boundary template.

• Needles are used to pin down the structure.

Some interesting observations are made creating the model. First the bending behaviour of
the grid is observed with respect to the direction of the members. The mat is supported in
the middle and deflects under self weight. When the members intersect at a small angle, the
stiffness of the mat is generally directed in one direction and the weight of the members is
carried in that direction (Figure A.47). When the members are moved apart, the number of
members per unit of surface becomes smaller in one direction and larger in the other
direction. The stiffness of the structure is now divides itself between the two principle
directions (Figure A.48). When the members are moved apart further, the stiffness in the first
direction becomes too small and the structure snaps through its balance point (Figure A.49).

By doing this simple experiment, it is physically experienced what the influence of the
direction of the member has on the structural behaviour of the mat. Obviously, there is an
optimum in angle between the members. An angle too small makes the structure too stiff and
all load transfer only takes place in one direction. An angle too large gives an inefficient
structural behaviour. To use the structures in an optimal way the optimal direction should be
estimated to fully use the bi-directional character of the structure.

Figure A.46: The grid is tied with binding wire Figure A.47: Load carried in one direction

217
A Design Tool for Timber Gridshells

Figure A.48: Load carried in two directions at an angle Figure A.49: The structure as snapped though its
close to the structures snap-through point ultimate point of balance.

The experiences of previous experiment are used when creating the model. First the mat is
orientated at an angle slightly orientated in one direction, at an angle of 60 degrees between
the members. The mat is carefully pushed down in the boundary template. Soon the
curvature becomes too large. The members cannot bend to the curvature needed and buckle
inward (Figure A.51 and Figure A.52)

Figure A.50: Flat mat orientated with an angle of Figure A.51: The mat is pushed into shape, but buckles
approximately. 60 degrees quite soon at the long edge.

Figure A.52: Edge buckling

218
 Appendices

Figure A.53: Mat orientated at approximately 140 Figure A.54: curvature of the structure with a member
degrees. orientation 140 degrees

Figure A.55: Mat orientated at approximately 80 Figure A.56: A large curvature is possible, but at the
degrees narrow sides buckling occurs

Figure A.57: Buckling of the structure at the small end.

219
A Design Tool for Timber Gridshells

When the mat is orientated mainly in the other direction (Figure A.53), the members come
very close to each other at the narrow sides when the mat is pushed into shape. This effect is
already seen in the free form shape in the small dome with large curvature, but here the
effect becomes extreme. The curvature cannot be made very large before the mat starts
buckling and resisting at these edges (Figure A.54).

At an orientation of approximately 80 degrees the mat lets itself being pushed into shape
more easily and to a larger curvature (Figure A.55 and Figure A.56). However, again at the
narrow sides of the ellipse the members start buckling. When the mat is being pushed further
into the desired eggoid shape, the members are buckling and twisting (Figure A.57). This is
due to the fact that the members cannot bend enough due to the relatively large bending
stiffness in the stiff direction, in combination with a large curvature of the shape.

The structure is finished with an angle between the members of approximately 80 degrees
and a moderate curvature (Figure A.58 and Figure A.59). This results in a smooth surface
without buckling. To compare the model with a computer model a grid is generated on a
similar eggoid shape (Figure A.60 and Figure A.61). This also shows the members scissoring
to each other at the narrow ends. When the generated grid is trimmed at the same height as
the physical model, the two models coincide.

The most important observation made while creating this model is the member behaviour
with large curvatures. A large curvature forces the members to scissor to each other and to a
large curvature of the individual members. If the bending and rotation capacity of the
members is insufficient, the members will buckle and twist. It is therefore advisable to limit
the curvature if possible.

Figure A.58: Final structure with moderate curvature Figure A.59: Final structure with an angle of
approximately 80 degrees

Figure A.60: Computer model of the eggoid (untrimmed).

Scissoring is clearly visible.
Figure A.61: top view of the computer model
(untrimmed)

220
Appendices

221
A Design Tool for Timber Gridshells

222
 Appendices

Appendix 6: Contact information

Contact information student:

Name: Matthijs Toussaint

Student no.: 9840294
Address: Vlierboomstraat 445, 2564JD DEN HAAG, The Netherlands
Phone: 06-14348691
E-mail: [email protected]

Contact information graduation committee:

Prof. Ir. L.A.G. Wagemans

Organisation: Delft University of Technology
Faculty: Civil Engineering and Geosciences
Section: Structural and Building Engineering
Address: Stevinweg 1, 2628 CN Delft
Room no.: ST II 1.59
Telephone: 015 - 27 84752
E-mail: [email protected]

Ir. J.L. Coenders

Organisation: Delft University of Technology
Faculty: Civil Engineering and Geosciences
Section: Structural and Building Engineering
Address: Stevinweg 1, 2628 CN Delft
Room no.: ST II 1.58
Telephone: 015 - 27 85711
E-mail: [email protected]

Dr.ir. P.C.J. Hoogenboom

Organisation: Delft University of Technology
Faculty: Civil Engineering and Geosciences
Section: Structural Mechanics
Adress: Stevinweg 1, 2628 CN Delft
Room no.: OG 6.48
Telephone: 015-2788081
E-mail: [email protected]

Dr.ir. J.W.G. van de Kuilen

Organisation: Delft University of Technology
Faculty: Civil Engineering and Geosciences
Section: Structural and Building Engineering
Address: Stevinweg 1, 2628 CN Delft
Room no.: ST II 2.56
Telephone: 015-2782322
E-mail: [email protected]

223