2016 Advanced Timber Structures

Advanced Timber Structures
Yves Weinand (Ed.)
Advanced
Timber
Structures
Architectural Designs and
Digital Dimensioning
Birkhäuser
Basel
Table of contents
Introduction 6
1 Folded plate structures 13
2 Advanced architectural geometry 49
3 Active bending 91
4 Form-finding and mechanical investigations

of active bended systems 125
5 Customized construction 189
The Research Laboratory IBOIS at the EPFL Lausanne 236
Picture credits 237
5
Introduction
How can a schedule and a technologically

innovative process shift the perspective of the
construction industry toward sustainability?
Yves Weinand
Timber construction has a promising future—especially ners ought to meet this demand by initiating and guid-
in relation to climate change and our need to find sus- ing innovative processes like the ones addressed here.
tainable solutions for the construction industry and to Furthermore, solutions for so-called “non-standardized”
implement the use of appropriate building materials. architectural forms that are sustainable and economic
Timber’s comparatively low energy consumption has been should be found.
a known fact for years. Challenges around sustainability
in the construction industry also touch on the question of
architectural form. At the Department of Timber Struc- Sustainability
tures, IBOIS, EPFL Lausanne, we posed the question of
how a formal and technologically innovative process With the onset of climate change, the concept of sus-
might be developed from a sustainable perspective. tainability has finally become a central issue for our
The renewal of construction technologies and technical twenty-first-century society. In this context, the research
procedures in timber, taking into account the innate presented here focuses on the following question: by ex-
qualities of the material, could lead to the increased use panding the applications of wood as a construction mate-
of timber in contemporary construction. This is not about rial, could its use in the construction of public buildings
the principle of longevity or permanence, on which mo- be boosted?
dernity is based, “modern” implying that which lasts, Regardless of its typology or function, a building
or that which remains. Timber is generally regarded as always consists of a myriad of small elements. Wood
being a traditional building material. This preconceived and wood-based materials are made up of the assembly
image is advantageous to socially legitimate research of smaller parts. Solid wood, timber beams, plywood,
that is about finding complex shapes, or creating free- laminated timber, and laminated veneer timber panels
form structures, when it is done in timber. are all produced from the amalgamation of smaller-scale
Choosing to approach complex geometries from parts. For this reason, the technology of the junction
the perspective of the designer, rather than merely connections should also be considered in the synthesis
in terms of generating forms, can be seen as stepping of these materials in a building. The variety of existing
away from fashionable trends in architecture, especial- timber materials and the considerable versatility in
ly the free, amorphous forms of so-called “Blob” archi- their application should determine the manufacturing
tecture. Many such free-form buildings completely ig- and prefabrication methods.
nore the problem of sustainability, partly due to their The aim of the research is to find solutions to
choice of materials and partly due to their energy a number of questions. We are interested in discovering
consumption and the cost of maintenance. In contrast, new construction solutions that can be easily incorpor-
the savings in overall energy consumption by the use ated into hardware stores’ offerings, to enable the af-
of wood as a building material, in terms of life cycle fordable realization of unconventional architecture. One
ana lysis and demolition, are undeniable. As a natural of the most important ways to reduce construction costs
material, timber requires less energy to produce, to is to use a digital design tool. The development of specif-
transform, to assemble, and to supply sustainably than ic, application-oriented digital tools would thus appear
other build ing materials. The public has come to expect to be imperative. Our tools will help at the interface of
this technologically innovative process. Today, public architecture/civil engineering, mechanical/geometric de-
and private clients alike demand novel solutions in sign, and form-finding/parametric digital prefabrication,
terms of both sustainable, high-quality architectural in specific, project-related steps.
design and construction methods. Architects and plan-
6 Introduction
The chair of wooden structures, IBOIS/EPFL
Within the framework of the Department of Timber

Structures, IBOIS has initiated various research areas
that explore the relationship between engineering
and architectural design. IBOIS is part of the Civil Engi-
neering Institute ENAC/EPFL, but it is also affiliated
with the Department of Architecture, where an architec-
Fig. 1 Three-layer panels
ture studio is made available to engineering master’s
students. Thus, collaboration between architects and
engineers is encouraged, providing the environment for
a wider scientific community within architecture schools current ability to use structural planes plays a vital
Europe-wide. The research results presented here focus role—for example, large-scale multilayered boards, ply-
on construction and the challenges of realizing complex wood, or laminated veneer lumber panels.
shapes and free forms. What is the relationship between
basic research and applied research? What is the con-
nection between pure research and applied research? Innate material qualities of timber
Or between curiosity-driven research and problem-
oriented research? And finally: how can the scientific re- Even today, one can see timber construction as a perpetu-
search in architecture be reconciled with the artistic di- ation of traditional forms and methodologies. The major-
mension of research, in order to bring them into harmony ity of buildings continue to utilize traditional structures
with one another?1 or building methods, such as timber-framing or truss
IBOIS provides a place to innovate, where the systems. However, due to the availability of new tim-
fascinating inductive-experimental approach is com- ber-based materials, there is a need to invent new con-
bined with the clarity of deductive-scientific methods. struction methodologies. As opposed to steel or reinforced
This is undertaken with the aim of creating new forms concrete, building with timber relies to a greater degree
and types of structures—particularly timber structures. on the development of cutting types and junctions,
In addition to its sustainable qualities, timber also has and the understanding of the relationship between them.
exceptional mechanical properties, which can be utilized Thus, for example, when designing a timber building, it is
in specific structural forms. imperative to consider the junctions, which are integral
For centuries, timber construction has been to the structure, and to specify the entire structure, in-
governed by the use of linear elements connected to cluding its joints.
truss systems. However, in contrast to steel and re- It is essential for engineers who specialize in tim-
inforced concrete—the dominant building materials of ber to provide comprehensive details that can be integrat-
the nineteenth and twentieth centuries—engineers have ed into the overall structure when formulating their
done little to develop the use of timber as a building structural models. In this way, these engineers take on
material. Now, thanks to the availability of digital tools, the role of designers. It becomes clear that timber con-
applications of this material can be expanded signifi- struction calls for interdisciplinary collaboration between
cantly; new geometries can be created; and innovative engineers, architects, and contractors right from the be-
construction materials and methods can be developed. ginning of the planning process. In particular, the choice
In short, we can undertake an innovative exploration of of panel types should play a central role in determining
structural engineering with regard to timber. Here, the the form and typology of the selected support system.
7
Characteristic Small Constructional Difference Car tesian axial system. This corresponds to historically
properties Specimen Element to the small applied geometries that define the principles of material
specimen
strength. A Cartesian axis system describes an isotropic
Bending Strength 68 37 46 % material in an efficient manner, but this is less true for
(mean value N/mm2)
an anisotropic material, such as timber. The longitudinal
Tensile Strength 80 30 63 % direction of the fibers, for example, is assumed to be
parallel to fibers
(mean value N/mm2) perfectly rectilinear. Though this is partially true, in re-
ality the natural longitudinal orientation of tree growth
Compression Strength 40 32 20 %
parallel to fibers (i.e., tree fiber) tends to converge conically toward the
(mean value N/mm2) top of the trunk. More accurate modeling techniques
should therefore take into account these specific proper-
Table 1
ties. It would be interesting to develop “tree-specific”
mechanical models, such as the scanning of the exact
fiber configuration of a particular tree trunk in order to
Tensile strength N/mm²
10 000
mechanically evaluate a specific application.
The problem of timber anisotropy has already
1000
been mentioned, such as in the invention of plywood,
where the fibers are confined. If several layers are glued
100
across each other, this results in a homogeneous, or quasi-
isotropic, structure.
10
Systems
1 A chain breaks as soon as its weakest link fails.
In contrast, a system will continue to function even with
Cellulose
Microfibres
Traceïdes
products
‘Clear Wood’
timber
Paper
Derived timber
Constructional
a broken weakest link. Engineered wood—a dual system

of laminated beams, or veneer laminated lumber pan-
els—that consists of several elements glued together can
Fig. 2 be regarded as a system. These are known as “multilay-
ered systems.” The principle can be applied to a specific
material as well as to a structural system. A common ex-
Table 1 Resistance values for small test elements from spruce
ample is a laminated beam, which consists of several
Development of resistance values for cellulose
Fig. 2
super imposed layers. The resistance value can be adjust-
(f y = 100 N/mm²) to wood (f y = 10 N/mm²)
ed, depending on the number of layers. If the number of
layers does not exceed four, then the beam is regarded
as a conventional support. If the number of layers ex-
ceeds four, then it can be regarded as a system, thereby
Specifically, the following observations about timber can increasing the resistance of the beam. From the view-
be listed: point of probability theory, the likelihood of failure of
a beam decreases with the increase in the number
The question of scale of its layers.
The size of the timber member used determines Another pertinent example is a reciprocal sys-
its strength and hence the range of its applications. Com- tem, such as the Zollinger system. If a member of the
paratively, a small “clear wood” test block performs far diamond-shaped configuration of a Zollinger mesh net-
better than a standard-sized beam. work fails (for example, due to a particularly strong
The surprisingly high values of small test blocks wind load), then the system does not fail as a whole.
(for example, spruce) were categorized by Peter Niemz2 One also speaks of a “social” support system, where the
in table 1. Therefore, great losses in the assessment pro- weakest link is supported by the adjacent members.
cess can be determined, since in this case considerably
lower values were reported. These observations lead to the following conclusions:
The anatomy of wood – It makes sense to produce timber materials

Figure 3 shows the three principal axes of wood where the “system effect” strengthens the total
fibers using the example of a cut tree trunk: longitudi- resistance.
nal, tangential, and radial alignment of the fibers. In – It makes sense to develop structural systems
practice, radial and tangential alignments are hardly where the interdependency of the elements is
distinguishable from one another, and an average value maximized.
is generally taken. The three axes are positioned in a
8 Introduction
If these considerations are applied to wood-based ma- On the basis of his research, de l’Orme foresaw the use
terials such as laminated beams, it is foreseeable that, in of small-scale wood elements that could be used as an
future applications, these structural systems will also alternative material. He combined this principle with
be able to benefit from the system factor. geometric innovations, allowing him to achieve greater
Traditional carpentry would rarely benefit from spans. Unfortunately, de l’Orme’s innovative ideas did
a system effect; most of the time, the failure of a roof not succeed, as they were too labor-intensive: every
beam or joist will lead to the collapse of the roof or attic. individual piece and each connection had to be cut in-
The same is true for traditional timber connections, dividually by hand. These past obstacles to building
where local failure of a connection results in the collapse de l’Orme’s structures could be overcome today with in-
of the element that is held in place by the connection. dustrial fabrication and CNC milling, and the construc-
The structures presented in the following section consist tion could thereby be made affordable. The architec-
of a large number of small elements. The importance tural expression of a networked system can certainly be
of developing such support systems, which amplify the classified as contemporary.
mutual dependence of the elements, is reflected in all De l’Orme’s findings influenced the French ar-
the structural systems demonstrated below. my’s military structures. General Armand Rose Emy
advocated the use of a large number of small-scale tim-
ber elements to cover their arena structures. In this in-
A new generation of structures stance, however, the boards were installed horizontally,
or rather in a horizontally curved position. Thus the
Can wood perform better than it currently does, or his- total curvature of the arched structure was achieved
torically has? And could the architectural image of tim- by the local bending of each board. This avoided waste;
ber buildings be given a more contemporary expression? a second advantage was that, as the orientation of
When one takes a close look at the buildings of Philibert the longitudinal fibers of each board coincided with
de l’Orme, already envisaged back in the early 1600s, the the line of force of the arc, they functioned far better
answers to these questions tend toward the positive. structurally.
Longitudinal
Transverse section
Fiber direction
on
cti
d ire
ia l ia l
Ta
d d
Ra
n
Ta
Ra
ge
n
nt
ge
ia
nt
ld
ia
Radial section
ire
l
ct
io
Fig. 3 Tangential section

n
Direction of fibers
0°
Glued connection
Elasticity modulus E – normal deflection:
z 45°
y z local
90°
y local
x
x local
longitudinal (EL) tangential (E T )

radial (ER)
Shear modulus G – shear deflection: Layer
Fig. 5
Elasticity modulus E (in tangential, Shear modulus G (in tangential,

longitudinal, and radial direction) longitudinal, and radial direction) The three strands of wood fibers are inserted
Fig. 3
E T / ER / EI = 1 / 1.7 / 20 (soft wood) GLR / GLT = 1 / 1 (soft wood) into a Cartesian axial system.
E T / ER / EL = 1 / 1.7 / 13 (hard wood) GLR / GLT = 1.7 / 1 (hard wood) Fig. 4 Elastic and tangential modules vary greatly.
Fig. 4 Fig. 5 The principle of cross-laminated timber
9
The much-celebrated Mannheim Multihalle by Otto
Mutsch ler (1975) is an extraordinary example of a spatial
structure composed of networked elements. The double-
layered network structure consists of actively bent tim-
ber slats with a square cross-section, which follow the
thrust line and absorb the normal forces well.
The third and final example of an upgraded struc-
tural system is the design for an industrial hall in
Botrange, Belgium. The supporting structure consists of
simple boards, which are multi-layered and crossed over
one another to form a mesh network. The boards pass
through the nodal points and are connected only with
vertical pins inserted laterally. A spatial structure is thus
Fig. 6 Philibert de l’Orme:
created out of a complex combination of conjoined small
The panels define a part of
the curve and are installed parts. The local rigidity of the nodes can be increased by
vertically. Thus, the edge inserting an additional bolt, thereby increasing the over-
of each panel describes a all rigidity of the system. Timber-frame construction and
section of the arc.
post-and-beam structures were, and remain, widely used
A significant proportion of a
panel’s wood fiber therefore systems in timber. With these composite systems, junc-
runs into a void, or into the tions are added individually on-site. In addition, semi-
edge of the panel, as it prefabricated floor elements and wall structures are
does not follow the primary
now available, which can also be incorporated.
direction of the curve, as this
would require the structural As a result, an attempt should be made to create
behavior of an arc. made-to-measure prefabricated systems incorporating
Fig. 6
connection technologies and precise prefabricated ele-
ments. Due to their specific shape or geometry, these
custom-made fixtures could only be installed in a specif-
ic location and in a unique position within the overall
system. Errors that often occur on–site could thereby
be avoided.
The manner in which building sites are organ-
ized today corresponds, in many ways, to nineteenth-
century models. The necessity for a foreman who reads
and understands construction plans, and then connects
them with the delivered components, needs to be replaced
by a stronger and different kind of planning. Access
connectivity systems, predetermined assembly sequences,
and integral mechanical connection techniques should
determine the site-work schedule. For this reason, we are
interested in geometric algorithms, subdivision processes,
Fig. 7
planarization processes, connective sequences, automat-
ed milling technologies, tool development, mechanical
test trials, and the execution of manual as well as robotic
joining processes. IBOIS’s areas of interactive research
have been summarized in the diagram below. Folding
systems, discretized free-form structures, woven and ac-
tively bent structures, and mechanically induced struc-
tural systems will be presented. Special attention will be
paid to the connections.
Notes
Fig. 8
1Thanks to Professor Pierre-Alain Croset for his critical
notes in this area.
Fig. 7 Multihalle Mannheim: General view of the shell structure with
2 Keunecke, D. and P. Niemz. “Axial stiffness and
a span of 60 m. Four rib layers are connected in one node. selected structural properties of yew and spruce micro-
Fig. 8 Competition project for an industrial hall in Botrange, Belgium tensile specimens.” Wood Research, 53, 1–14, 2008.
10 Introduction
Material Wood
• Eco
• Lightweight Surface
• Strong Structures
• Carbon-neutral
Discretization
Assembly Planning • Shape Optimization
• Discretization (with
• Interlocking Planar or Singly
Assemblies Curved Plates)
• Sequences
Prefabrication
• New Technologies
Assembly • Optimization
• Manual Assembly • Computerized Tools
• Robotic Assembly for Fabrication
• Robot Vision
Joint
Mechanics
• Analysis
• Testing
• Tools for
Dimensioning
Fig. 9 Intercations between

IBOIS’s areas of research
11
1 Folded plate structures
1. 1 Folded plates made from timber panels 14

Yves Weinand
1. 2 Structural optimization of timber folded-plate

structures 32
Andrea Stitic and Yves Weinand
1. 3 “It’s important to progress

from research to a product” 40
Interview between Ueli Brauen and Yves Weinand
1. 4 The Chapel of the Deaconesses of St-Loup

at Pompaples, Switzerland 44
Marielle Savoyat
13
1. 1
Folded plates made from timber panels

Yves Weinand
Folded plate structures are fascinating Thus far, structures made of folded wood-
constructions, where form and structure en panels could not be implemented.
are inextricably linked to one another. Formwork structures made of reinforced
Folded panels are also able to assume concrete were already being designed and
many different forms. By folding timber, executed in the 1930s. At that time, the
these structures acquire a high level of construction of identical, prefabricated
rigidity, despite the fact that the material prismatic supports was significant, as it
strength remains reduced in relation minimized the cost of timber shuttering.
to the span. Furthermore, this material Structures made completely from folded
allows filigree structures that are also timber panels were being developed in
economical. the 1970s. However, this concept was not
pursued any further.
In this essay, we will look at possible ways

that folded timber panel structures might
establish themselves in the marketplace.
To illustrate the concept, three projects
will be presented to highlight the possible
development of these structures. In each
project, form-related, mechanical, and
production-related aspects are highlight-
ed. As already mentioned, folding struc-
tures have the ability to torsion form and
structure. Experimenting with relatively
soft, large surfaces or panels is made
possible by combining the panels along
their edges, in “wrinkles” or folds. The geo -
metric manipulation of the folds and the
“freezing” of a certain arbitrarily chosen
Fig. 1

Fig. 2 Fig. 3 a Fig. 3 b
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Fig. 4
4550
Fig. 1 Folding principle: rigidity is achieved by folding.

Fig. 2 Experimental folded plane
Fig. 3 a and b Axonometric representation of a folded
structure: elevation and plan
Fig. 4 Axonometric representation of the basic
elements of the original geometric configuration
Fig. 5 Floor plan of the base element of the original
geometric configuration
6300
Fig. 5
15
diagonal member
section 24 × 24
principal beam
A‘ h: 24
4550
A B
“Kerto“ panel
57 mm
diagonal member
section 24 × 24 “Kerto“ panel
principal beam diagonal member
h: 24 principal beam
30
55
Fig. 6 Solution of the support system: the panel is transfor med

into a truss, clad in far thinner panels.
geometric position during the folding lumber panels—then this results in a

process opens up the potential for the required panel thickness of around 14 cm.
structural optimization of such structures. Panels of this thickness are, however,
rather expensive.
The first example to be analyzed is the
roof of the sports hall “des Îles” in the The challenge of this type of construction
Swiss lakeside town of Yverdon. is not in the cutting or the installation
This building was initially planned as a of the panels, but rather in the execution
folded-panel structure. In preliminary of their connections. For this reason,
studies, the geometric consequence of alternative construction options were
having a folding roof with a span of over proposed that aimed to reduce the panel
45 m was investigated. If one calculates thickness and find optimal locations for
the scenario as illustrated in fig. 3 a— the connections. One of these alternative
the geometry as sole self-supporting solutions is illustrated in fig. 6.
membrane made of laminated veneer

Laminated beams form a triangular truss, – Due to this reduction, a strengthen-
which is covered with panels. Thus, the ing of the slenderer panels is neces-
truss is visible from within the hall. sary. The superposition of the origi-
On the exterior, the panels provide the nal “natural” folding geometry over
necessary insulation and roofing surface. that of the edge support and its
The connection is made possible here by stiffeners diminishes the conceptu-
cutting the edge supports. The peripheral al strength of the folding geometry.
laminated beams running along the edges – The folded structure is effectively
have to be angled, cut, and then bolted changed from a structural support-
together. The overall geometry of the fold ing surface to a frame structure by
creates a geometric challenge along its shifting the main components of
“folded joint,” which in this case is solved the shear forces along the edge slats.
by the cut. The structural consequence of
this is that the initial structural strength Fortunately, our subsequent attempt to
gained from surface areas is, thanks to build a folded structure succeeded, with
the edge rigidity, effectively altered into our execution of the Chapel of Saint Loup.
a truss support structure. Due to the However, it is important to note that
superimposition of linear elements and the span of only 9.5 m was far less than in
their triangular stiffeners onto the folded Yverdon, where the span was approxi-
geometry, the unified appearance of mately 45 m. For the chapel, 60-mm-thick,
the folded structure is unfortunately lost. three-layer panels span the chapel space,
achieving a reduction of 1/158. The folded
Even for this altered folded structure for geometry relates to the roof but also to
the hall in Yverdon, the high tenders the walls, as the static efficiency of
given by contractors were an indication folding in the roof area is employed in
of the skepticism regarding its economic a similar form in the wall surfaces.
viability. Ultimately, the structural
efficiency of the folded panel was rejected The tool developed by Hans Ulrich Buri at
in favor of a frame system. However, IBOIS generates the overall geometry, as
additional rigidity still does occur spa- well as determining the cutting and bend-
tially above the folded roof structure ing of the panels. The montage images
and the wall truss. display a self-supporting framework that
functions efficiently as a single fold. Our
In summary, the following is ascertained: idea was to leave the supporting struc-
– With cost saving in mind and the ture—i.e., the panels—visible, at least on
aim of reducing the thickness of the the inner side. The panels are therefore
panels, a panel thickness of 14 cm connected to one another on the exterior
was determined. by means of a 2-mm-thick metal plate.
17
The transfer of forces from one plate to the – The connection method suits the
next can thus take place by linear supports. surface supporting nature of the
As a consequence of the cut sizes, five structure.
detail types were identified. The nail
plates remain identical, but the number It should be noted that the generation of
of nails per plate varies from one type to the Chapel of Saint Loup’s overall geo-
the next. The torsion peaks that occur metry is rather arbitrary, as the lateral
at the frame corners can be avoided. surface of the chapel corresponds to an
These bending moments can be distrib- unfolded sheet of paper. In theory, there
uted along the edges and do not all have are no offcuts. An incline for rainwater
to be absorbed at the frame corner. runoff was generated by the folds.
In the most extreme structural state that The obvious weak point of the overall
might occur in the case of the chapel, structure is the open edge—the first fold
bending moments at the frame corners of located above the entrance, where the
about 10 kNm/m can be reduced by half. largest span occurs.
The proposed detail thus corresponds to The history of architecture reveals how
the surface structural nature of this the understanding of structural folding
supporting folded structure. Unlike tradi- has led to a wide variety of effective
tional timber construction, where the solutions.
prismatic sections are connected by pins
of some sort (nails, screws, bolts), that The understanding of the mechanical ri-
weaken the cross-section, a folded gidity of the folds in the overall geometry
structure, such as the one at Saint Loup, described here does not result in only a
allows an optimal exploitation of the single feasible geometry. Rather, numer-
panel thickness as structural support. ous geometric options were demonstrated
by Hans Ulrich Buri’s work.
In summary, the following can be
ascertained: But ultimately, transverse and longitudi-
– 40-mm-thick panels for the walls nally generated profiles remain purely
and 60-mm-thick panels for the geometric elements. In this respect,
roof were found to be optimal sizes the tool proposed by Hans Ulrich Buri
for an efficient and material-saving remains a formal, architectural, and
system in this category range. constructive tool. It also allows for a me-
– With the selected connection chanical optimization that, however, re-
method, weakening of the panel mains intuitive. In the case study (Fig. 24)
cross-section is avoided. it is shown how a given folded geometry
can be optimized from a mechanical

Fig. 7 Finite-element model of a single-
folded, three-layer panel structure:
support conditions and edge connections
Fig. 8 Finite-element model of a single-
folded, three-layer panel structure:
load cases; a) permanent weight,
b) dead load, c) snow, d) wind
Fig. 9 Finite-element model of a

single-folded, three-layer panel struc-
ture: deformation and membrane forces
Fig. 7
b)
a)
c)
d)
Fig. 8
deformation d x (mm) deformation d y (mm)

max = 2.5159, min = 0.0354 max = 0.2145, min = –9.5408
2.5 9.5
2.0 7.5
1.5 6
1.0 4
0.5 2
0 0
–0.5 –2
–1.0 –4
–1.5 –6
–2.0 –7.5
–2.5 –9.5
Membrane effort N x (kN) Membrane effort Nz (kN)

max = 101.4, min = –24.8 max = 15.7, min = –158.9
100 159
80 127
60 95
40 64
20 32
0 0
–20 –32
–40 –64
–60 –95
–80 –127
–100 –159
Fig. 9
19
Fig. 10 a
150 nail perforations per rm, 150 nail perforations per rm,
2 mm , 56 m length in total 2 mm , 40 m length in total
160
170
320
340
160
170
150 nail perforations per rm, 175 nail perforations per rm,
2 mm , 81 m length in total 3 mm , 48 m length in total
120
120
240
300
120
180
150 nail perforations per rm,

3 mm , 32 m length in total
120
240
120
Fig. 10 b Fig. 10 c
Fig. 10 a, b, and c Detail of a panel connection: 2-mm Fig. 11 Detail of the plate connection after Fig. 14 Fabrication drawings for each panel
nail plates are divided into five detail types, where completion
Fig. 15 Axonometric view of the chapel and the same
the number of nails varies. The plates themselves
Fig. 12: Plate detail at the base geometry unfolded
were previously folded.
Fig. 13 Detailed view of the roof edge during
construction
Fig. 11 Fig. 12 Fig. 13

12864 8601
6849
2969
2809
2250
8698 10419 10119
2906
2715
2250
9606 9366
11030
2789
2713
2250
10409 8442
11744
2904
2778
2250
10860 8397 12134
2927
2818
2250
8166
11087 12312
2961
2828
2250
12181 11725
13343
2967
2753
1736
Fig. 14
Fig. 15
21
Fig. 16
Fig. 17
Fig. 18
Fig. 16 Installation of the panels.

Each section is self-supporting.
Fig. 17 The chapel during construction
Mounting of the roof membrane
Fig. 18
and the battens
Fig. 19 a and b Completed chapel
Fig. 19 a Fig. 19 b

perspective. By increasing the height of
the folds, or by increasing the frequency
of the pleats at the edge of the shell,
structural performance can be enhanced.
In summary, the geometric adjustments

that contribute to overall rigidity:
– By increasing the fold depth, the
moment of inertia is increased.
– If the frequency of the pleats is
increased toward the edges of the
shell, then additional horizontal
rigidity can be generated in these Fig. 20
edge regions by simply adapting

the geometry. This adaptation is
convincing as, unlike conventional
shell construction solutions, where
the edge has to be reinforced, here
it is possible to enhance the struc-
tural capacity by simply taking
advantage of the folded geometry, Fig. 21
thereby avoiding the need for

an extra element in a different,
non-timber material. Fig. 20 The trapezoidal cross-section also corresponds to a fold.
The scale of the folds can be adapted to the specific
Fig. 21
building—here, the roof of Graz city hall.
Fig. 22 a and b Generation of possible folding geometries
Fig. 22 a Fig. 22 b
23
Verbiers Concert Hall: Transverse and longitudinal
Fig. 23
profiles enable the creation of a variety of folding geometries.
– The danger of local buckling of the the lower the price per square meter,
panels can be mitigated by reduc- and the lower the transport costs.
ing the surface of the panels and
increasing the fold frequency. Thus far, the presented tool has been
(Since the fold edges are substan- aimed specifically at the engineering
tially more rigid than the panels, community. We hope to arouse their
they would buckle before the fold interest in the overall form and the
edge fails due to instability.) mechanical optimization. In addition,
– The reduction of the panel surfaces the above-mentioned geometric manipu-
also offers an economic advantage, lations improve the rigidity of the shell,
as the price of wood panels depends to accommodate both horizontal and
on their size: the smaller the panel, vertical forces.

Fig. 24 Geometric manipulation of folds
Static improvements achieved: de-
Fig. 25
formations can be reduced by one-third.
Fig. 26 Prototype
Typical reverse-fold geometry.
Fig. 27
Transverse and longitudinally generated
profiles enable the creation of various
folding geometries.
Fig. 24
Internal forces Deformations
1. max. vectorial
displacement
3.3 mm
2. max. vectorial
displacement
2.2 mm
3. max. vectorial
displacement
1.9 mm
Fig. 25
Fig. 26 Fig. 27
25
Fig. 28 The folding geometry of a tennis court hall is optimized. The folds are positioned close to the edges of the hall.

The design of a prototype was generated The distances stipulated by regulations
with the following geometric could not initially be complied with, as
characteristics: the idea was to design with relatively
thin panels. As a result, various different
A given area integrates the following options for positioning of the panels
surfaces: and panel thicknesses were explored.
– A continuous fold with a All test configurations are feasible and
V-shaped cross-section have the same advantages provided
– A complete fold with the by the panel connections as detailed in
longitudinal profile of a the chapel described above.
triangle (inverse fold)
– Diamond fold Nevertheless, these connections remain
– Herringbone fold costly. Furthermore, the traditional allo-
cation of cost groups for timber volume,
The comparative parameters are as connections, and assembly costs are un-
follows: altered. Therefore, the critical considera-
– Vertical deflection in the center tion lies not only in the design of the
– Vertical deflection at the edge overall geometry from a formal, topologi-
– Absolute tension under its own cal, and mechanical perspective (as is
weight shown in Andrea Stitic’s work) but also in
– Structural behavior under the integration of the connection details.
asymmetrical wind loads In a third project, the ideas developed
in the hall in Yverdon and the Chapel of
Here, a potential mechanical optimization Saint Loup, along with the integration
of a factor of 5 can be determined. of a new connection, are unified and exe-
cuted. A folded system with small sections
As with previous projects, careful con- was chosen for Vidy’s new theater.
sideration was given to the means of con- By doubling the number of folds, two
nection. Although the connections chosen 40-mm-thick panels are positioned
for the Chapel of Saint Loup led to a 25 cm apart, thus forming two shells.
coherent solution, the chosen panels did
not form an integral part of the geometry. The principle behind the connection of
Bolted connections that do not comply the two shells is illustrated in fig. 34.
with regulations were used for initial Both surfaces of the inner shells of the
trials. folds project through to the outer shell.
This results in a stiffening of the fold,
a kind of rigid corner. Thus, conventional
purlins or timber beams can be omitted.
27
Fig. 29
Fig. 30 a Fig. 30 b
Fig. 30 c Fig. 30 d
Fig. 29 Finite-element model of a folded structure Variations of folds: a sketch design for a
Fig. 31 Testing a prototype. The double-shell
Fig. 33
concert hall for the Verbier Festival structure consists of 8-mm-thick panels situated
Fig. 30 a, b, c, and d Execution
of prototypes, which
10 cm apart.
also led to technical test experiments. Fig. 32 A double-folded structure is generated.
Visualization of the virtual parametric model

An innovative timber connection is
created, derived directly from the
existing geometry.
The strengthening of the panels along the

folds, or the reinforcement of the edges,
also decreases the risk of the panels
bending. The trapezoidal surface is re-
duced by the reinforced edges. In addition,
all the panel connections were designed
as timber–timber connections.
Fig. 31 Fig. 32
Fig. 33
29
Fig. 34 Fig. 35
Fig. 36 Fig. 37 a
North facade
A–A
West facade
B–B
East facade
Fig. 37 b Fig. 37 c C–C

Fig. 34 Principle of the corner junction in each fold.
Both layers of shell project to the outer shell and are
connected.
Fig. 35 Woven patterns are created along the edges,
as the panels are connected by pins throughout.
Axonometric drawing and plan of folded geometry,
Fig. 36
Vidy Theater
Fig. 37 a, b, and c Ground-floor plan, elevations, cross-sections,
Vidy Theater
Fig. 38 Vidy Theater, Lausanne: presentation images
Fig. 38
31
1. 2
Structural optimization of timber

folded-plate structures
Andrea Stitic and Yves Weinand
This
In diesem
paper Beitrag
analyzeswird the analysiert,
potential ofwelchesdifferent Poten-
pos- 1 Introduction
zial
sibleverschiedene
folded-form gefaltete
topologiesTopologien
for generatingfür die folded-
Her-
surface
stellung structures
von Faltwerkenmade from aus Holzplatten
timber. The bergen.advan- Architectural and technical applications of origami-
Der
tage größte
of such Vorteil
structuressolcher
lies primarily
Konstruktionen
in the realmliegt inspired structures (the resulting forms need not neces-
of
in ecology
ihrem ökologischen
and sustainability.und nachhaltigen
By utilizing anProfil. inte- sarily be developable) employ the structural potential
grated
Durch die method
integrale
of construction
Bauweise, beithat der fulfills
Stützfunktion
both a of the folding principle to use less material and increase
supporting
und Verkleidungsfunktion
and a covering function,
vereint sind, extremely
lassenlight-
sich structural efficiency. By placing the material farther
sehr
weight leichte
structures
Bauwerke
can berealisieren.
achieved. Also, Zudem a greater
ist ein away from the axis of flexure, i.e. folding, the moment
degree
hoher Vorfertigungsgrad
of prefabrication ismöglich,possiblewodurch
leading to diea Ge-
re- of inertia is increased, inherently leading to greater
duction
samtkosten in thereduziert
overall werden.
cost. TimberHolzfaltwerke
folded-surface
beste- structural stiffness. This paper focuses on folded struc-
structures
hen aus einer consist
Vielzahlof einzelner,
a large numberdünner Platten,
of discrete, die tures of engineered panels made from timber. Common
thin
miteinander
plane elezu ments,
einerconnected
gefaltetentogether
Fläche verbunden
to form an ter minology for describing these structures, which utilize
overall
werden.folded Dazusurface.
sind entsprechende
Proper edgewise Verbindungs-
connection the benefits of folding, include: “folded plates,” “folded
details
elemente areanneeded
den Kanten
in order erforderlich,
to ensure an die efficient
ein effi- slabs,” and “corrugated structures.” However, the struc-
zientes
load-bearing
Tragsystemsystem. entstehen
For structures
lassen. Für made Holztrag-
from tures being considered are composed of multiple inclined-
werke
timber ist products,
dies eine
this besondere
presents anHerausforderung,
enormous chal- plane structural surfaces, joined together to form a glo-
die
lenge,modernste
requiring the Verbindungstechniken
use of state-of-the art erfordert.
joining bally folded form, while their load-bearing behavior
Aus
techniques.
diesem Grund Thus, the
kamenuse Holzfaltwerke
of folded timber im Ingeni-
plates combines both a slab and a plate mechanism (Fig. 1). In this
eurbau
in civil engineering
bisher auchapplications
nur sehr selten has been zumvery Einsatz.
limit- instance, the terminology “folded-surface structures” is
ed
Dochto indate.
jüngster
However,
Zeit kam
new es technical
zur Entwicklung
solutionsneuerhave preferred.1
technischer
recently been Methoden,
proposed mit fordenenefficient
sich Holzplatten
edgewise Several structures made from timber engineered
an
joining
den of Kanten
thin timber
effektivpanels.
verbinden
This paper
lassen.focuses
Der vor-
on panels have been realized, as described by Hans Ulrich
integrated
liegende Beitrag mechanical
behandelt
attachment
integrale techniques
mechanische that Buri 2 and Regine Schineis3. In these examples, the width
Verbindungsmethoden,
utilize digital prefabrication bei denen
to integrate
die Verbindungs-
connectors of the structure was spanned with a single element.
through
mittel währendpanel geometry.
der digitalenTakingVorfertigung
into considerationin die However, due to manufacturing and transportation con-
Platten
material, geometrie
fabrication,
eingefügt
and connection
werden. Unter detail Beach-
constraints, timber elements are only available in limited
straints,
tung von Werkstoff,
various topologies
Fertigungare undexamined
Anschlussfor details
the sizes. Consequently, in order to cover longer distances, ef-
considered
werden verschiedene
application.Topologien
Furthermore, für the
diestructural
jeweilige ficient connection details between adjacent plane
Anwendung
behavior of folded geprüft.systems
Des Weiteren
is studied,wird anddas three
Trag- elements are needed along the span of the structure. A
verhalten
feasible forms der Faltwerke
are compareduntersuchtby meansund drei of mach-
finite folded-surface structure made from prefabricated cross-
element
bare Formen analysis.
mittelsFinally,
einer Finite-Elemente-Berech-
observations are made laminated timber panels and assembled with screwed
on
nunga case
miteinander
study of verglichen.
a built prototype
Schließlich
structure,
wird eineand miter joints was proposed. 4 The folded-form geometry
Fallstudie
the structural an einem
potential
realenof Prototypen
the proposed durchgeführt
systems is chosen was based on folded rhombus elements. After ex-
outlined.
und das konstruktive Potenzial des gewählten Sys- amining its load-bearing performance, it was concluded
tems dargestellt. that the screwed miter joint connections were not suffi-
ciently resistant to withstand the resulting transverse
Keywords folded-surface structures, folded form bending moments, and that the joints for large panel
topology, integral
gefaltete mechanical attachments,
Flächentragwerke, assemblies such as these would have to be improved.
foldedgefaltete
plates structural behavior
Topologie, The issue of designing adequate thin panel
integrale mechanische Verbindung, edgewise joining details presents a major challenge in
Tragverhalten gefalteter Platten timber engineering, as it is difficult to address by using
standard timber panel joining techniques. However, the

performance of folded-surface structures depends on digital prefabrication to integrate connectors through
these linear edge-to-edge panel connections, and they panel geometry. The main advantage to using these form-
are considered a key design component. Since on-site fitting joints is that, in addition to their load-bearing
gluing is not possible due to a lack of uniform, stable function (connector feature), they also integrate features
conditions for curing of the adhesive, joints are usual- for fast and precise positioning of thin elements (locator
ly achieved by using metal fasteners. Nevertheless, ac- feature). Moreover, these joints do not impose any con-
cording to current regulations, the minimum distance straints on the panel thickness.
from the screw to the panel edge is set to 4d (d = dia- In this paper we present various known folded
meter of the screw) which restricts the minimum value form topologies and examine their structural potential
of the panel thickness, depending on the size of the for application in timber folded-surface structures. Sub-
fastener. 5 Consequently, the final panel thickness is usual- sequently, feasible forms are derived, taking into ac-
ly not dimensioned according to the structural require- count the material and the chosen connection detail
ments regarding the load-bearing capacity, but more fabrication, as well as assembly-related constraints. The
in line with the minimum requirements imposed by structural behavior of timber folded plates is studied
the connection detail. The recent rediscovery of integral and the chosen forms are compared by means of finite
mechanical attachments has provided an innovative element analysis. Finally, a case study of a built proto-
method for edgewise jointing of timber panels.6,7 In this type structure is presented, along with important obser-
paper, we focus on an integrated mechanical attachment vations regarding the material, fabrication, and element
technique developed by Robeller et al. 8 that utilizes assembly.
fold slab plate
Fig. 1 Load-bearing behavior of a single fold, combination of

the slab and plate effect
33
Bidirectional folded forms
a) Simple b) Spot or facet c) Rhombus-based d) Anti-prism-based folded form

corrugated surfaces folding folded form and its quadrilateral variation
Fig. 2 Classification of folded-surface structures;

perspective and transverse cross-section view
2 Topology of the folded form surface: equilateral triangles, squares, and regular hexa-
gons. Others either require the use of semi-regular tessel-
Fig. 2 shows the various folded-form topologies that are lations or irregular polygon geometry. In the classifica-
classified by their ability to discretize flat and singly- tion presented, only regular tessellations are shown, as
curved surfaces. In order to describe these forms, we will irregular tessellations offer a myriad of different topolo-
use the terminology derived from computer graphics gies. The second group (Fig. 2c) consists of rhombus-based
where a surface is represented by a polygon mesh. 9 The folded surface forms. In origami literature, this form is
topology of such a mesh is defined using a set of vertices also known as the Miura Ori or herringbone pattern. The
and additional information on how they are connected. third and final bidirectional folded form considered is
This connectivity further determines the bounding edges the form based on the anti-prism. This is composed of
of the form’s surfaces. The spatial arrangement of verti- isosceles triangle faces and is also known as the Yoshi-
ces, as well as their connectivity, can be regular or irreg- mura or diamond pattern. A variation of this form where
ular, creating a form that is composed of either identical quadrilateral faces are used is also included (Fig. 2d) . This is
or diverse base polygon geometry. obtained by duplicating the anti-prism form vertices and
The first group (Fig. 2a) contains simple corrugated introducing an additional connection line between them,
surfaces. They are composed of quadrilateral faces with resulting in trapezoid faces. Neither of these forms can
bounding edges that form a succession of either parallel or discretize flat surfaces, and they are limited to curved
oblique lines. Folded forms such as these are commonly cross-sections.11
used in practice. However, due to size constraints on tim- The bidirectional folded forms classified above
ber panel elements, the covered surface area of such struc- are subsequently compared according to their folding
tures is limited. In order to overcome this constraint, principles, making them more or less feasible for struc-
topologies where vertices and their connections form a tural application in timber folded-surface structures. All
spatial grid are used. By doing so, bidirectional folded structures are considered to have pinned supports at the
forms are obtained (Fig. 2b, c, d). Such forms consist of multi- sides, while connections between the plates are regarded
ple elements in two distinct directions of the structure. as line hinges, allowing rotations about the face edge
The first bidirectional folded form considered direction. A cutting plane is positioned perpendicular
(Fig. 2b) includes spot or facet folded forms, consisting of to the longitudinal axis and the obtained transverse
vertices where several faces converge together in one cross-section profile is observed (Fig. 2). It should be noted
single spot.10 Forms such as these are obtained by taking that for quadrilateral-based facet folded forms, both
a basic polygon and vertically raising its centroid point. flat and singly curved, a continuous longitudinal hinge
This vertex is then connected to the vertices of the ini- line is formed in every second vertex of the transverse
tial polygon to form triangular faces. For such forms, only cross-section. Even with a rise sufficient to span the
three types of polygons allow regular tessellations of a ratio of the structure, the number of such hinges should

not exceed three (including the ones at side supports).12 a)
When more than three hinges are used, the system trans-
forms into a mechanism and is no longer stable for more <20°
b>70°
than two quadrilateral polygons per span. Also, the static
height of the structure at this section point is very low
b)
and equal only to the panel thickness. With regard to oth-
er polygonal facet folded forms, the system remains sta- <20° <20°
ble irrespective of the number of hinges due to the offset b>140°
of each transverse string of polygon elements in relation
to the previous one. Each individual string is still a c)
mechanism in its own right, but is kept stable by the
neighboring string of elements. Furthermore, as the <20° <20°
number of polygon sides in facet-folded forms increases, b >70°
the dihedral angle between the faces surrounding the
raised centroid point also increases. It is considered that Minimum value of the apex angle, β, and the maximum
Fig. 3
value for the assembly direction deviation from the plane
angles close to 90° function particularly well for folded normal to the assembled edges for the a) rhombus-based,
surface systems, whereas the structure will lack rigidity b) anti-prism-based and c) trapezoid-based folded forms
when the angle is incrementally increased.13 Moreover,
the dihedral angle values in integrally attached timber
structures are restricted to angles between 50° and 130° of panel edges that can be connected simultaneously. The
by the connection detail choice (see section 3.1 below). simultaneous assembly of multiple edges requires that
Consequently, as the dihedral angles become very ob- the panel assembly direction be chosen within the inter-
tuse for five-sided polygons, such forms are regarded section of what are called the “individual edge rotation
as unsuitable for this study. windows.” These “windows” represent all possible assem-
As a result, only the rhombus-based folded form bly directions for each edge. The range of this intersec-
and the anti-prism-based form, along with its trapezoid tion depends on the number of assembled edges, tool in-
variation, were considered worthy of further investiga- clination limitations, and the dihedral angle between the
tion. These three forms offer a particular structural ad- plates.14 In order to ensure that assembly is possible, i.e.
vantage by having adjacent folds both in a longitudinal that respective rotation windows intersect, the geometry
and a transverse direction, together providing the neces- of the panels must be chosen in such a way that the final
sary structural height along the entire cross-section. assembly direction does not deviate from the plane nor-
mal to each assembled edge by more than 20° (Fig. 3).
3 Structural considerations 3.2 Geometry of the structures

For each folded form, the base element geometry
The chosen bidirectional forms were further examined was kept constant throughout the whole system so that
by comparing their ability to discretize a hollow cylin- the form remained regular. Apex angles, β, were kept
drical segment with a set of predefined parameters. Its within the limits imposed by assembly constraints and
outer radius was set to Rext = 2.5 m and the cord length, i.e. held to a minimum of 70° for the rhombus- and trapezoid-
span, to S = 3 m in order to obtain an optimal structure based folded forms, and a minimum of 140° for the anti-
height of h = 0.5 m, as well as a favorable height-to-span prism-based form (Fig. 3). The panel thickness t, widths of
ratio of 1 : 6. Additionally, for restricting the maximum elements w, and the dihedral angle φ, were also taken
static height of the folded system, h s, the inner radius was as invariables for all forms. The number of individual
set to Rint = 2.3 m. elements along a transverse cross-section, which can be
used to approximate a set curvature with a defined max-
3.1 Joint constraints imum static height and structure span, was determined
The integral attachment technique chosen for according to the following formulae:
connecting the panels imposes a certain number of con-
straints which need to be taken into consideration when } tot
n= #cos
designing the folded surface geometry. One of the main –1
S R ext – h s X &
constraint considered is robotic accessibility where, due R ext
to the limited tool inclination, the dihedral angle be- } tot
ñ=
tween the connecting panels, φ, is restricted to angles be- #cos –1
T
R ext – h s &
Y
eq. 1
tween 50° and 130°. Furthermore, due to the local joint R ext – Qh s /2 V
geometry, the orthogonal as well as the non-orthogonal
assembly of individual elements also presents specific re- ñ if ñ is even
n$G
strictions concerning the geometry and the number ñ + 1 if ñ is odd eq. 2
35
Rhombus-based Anti-prism-based Trapezoid-based Where ψtot is the central angle of the circular segment
folded form folded form folded form
observed, Rext is the outer radius of approximated hollow
1,15 h
cylinder, and hs is the static height. For anti-prism- and
h
h
0,86 h s hs hs
0,94 S S S trapezoid-based forms, there is a unique result for a
required number of whole, uncut triangular elements
(equ. 1) . On the other hand, for rhombus-based folded
{ {
{
forms, the result is defined as the minimum one neces-
w
Fig. 4
w w sary. Additionally, only an even number of elements
is considered, as any odd number results in end seg-
ments with different orientations, leading to a height
difference between side supports. Therefore, the result
is rounded up to the next even number (equ. 2). For the
set parameters, three segments for the anti-prism- and
trapezoid-based form, and a minimum of six segments
0 0.25 0.5 0.75 1
for the rhombus-based form were obtained (Fig. 4). It can be
noted that approximating the same curvature with a de-
fined height by using the rhombus-based folded form
resulted in a slightly smaller span, as well as reduced
maximum static height of the system. In the case of the
trapezoid-based form, the overall height of the system
Rhombus-based had to be increased in order to keep the side cross-
folded form
sectional segments uncut and to avoid unnecessary cor-
Anti-prism-based
folded form rugations at the supports.
Trapezoid-based
folded form
3.3 Structural analysis
Bidirectional folded-surface structures carry
the load via a combination of extensional and flexural
action. The relative proportions of extensional and flex-
ural effects depend on several factors: the overall form of
the structure, connection detail stiffness, support condi-
tions, and the loading configuration. Extensional or shell
action resists the external transverse loading through
the action of in-plane forces. This is a result of the overall
shell-like form of the structures, but also the individual
position of the plates in relation to the acting force.
Fig. 5
The structural behavior of an inclined plate such as this
is divided into two different mechanisms of resistance
and their combinations. Firstly, plate mechanisms, where
the component of the acting force directed parallel to the
surface is resisted by in-plane forces, and secondly, slab
h Funicular curve mechanisms, where the component of the acting force
Anti-prism-based folded form
Trapezoid-based folded form
directed at right angles to the surface is resisted by bend-
Rhombus-based folded form ing and twisting moments. The latter mechanism of local
plate bending forms the flexural part of the overall struc-
ture’s load resisting action.
The conversion of out-of-plane external load into
in-plane extensional action is explained below. The load
S/2
normal to the plate is transferred by bending to the edges,
S
Fig. 6
where it is resolved into components situated in the
planes of the joining plates. When these in-plane forces
are not in equilibrium, the resulting force is further
Fig. 4 Transverse cross-section and perspective view
of the three examined structures transferred to the supporting plates by in-plane shear
along certain edges. The size of this in-plane shear de-
Fig. 5 Vertical displacements along a normalized span (above)
and an overlay plot of the deformed and undeformed state, pends on a number of factors, such as the uniformity
uniformly scaled for all structures (below) of the dihedral angle between the observed plate and
Fig. 6 Deviation of folded form central line from the equivalent its supporting plates and stiffness of the plates, as well
smooth shell funicular curve as the edge connection detail and the nature of the load.

The local plate bending behavior also depends on similar As a result of the analysis presented here, it is concluded
factors. However, the values of the edge bending mo- that the choice of the folded form has a significant influ-
ments that cause edge rotation depend primarily on the ence on the edge connection load-bearing requirements
rotational stiffness of the connection detail. and that the semi-rigidity of the joint will certainly play
The three obtained structures were modeled in an important role in the behavior of such systems in
finite element analysis software ( ABAQUS) and their reality. Furthermore, the best structural performance of
performance was observed under structural self-weight. the folded-surface structure is found when the external
It was concluded that the local structural behavior of load is primarily resisted by shell-like extensional action.
the system and the inherent global structural stiffness This can be enabled by choosing a form which best ap-
depend to a great degree on the connection details’ capac- proximates a corresponding hollow cylinder segment, i.e.
ity to transfer the occurring shear forces and bending a smooth shell solution. To that end, a comparison is made
moments between the adjacent plates. It was decided between each folded transverse cross-section central line
to model the actual semi-rigid behavior of the joints as and a funicular curve of the equivalent smooth shell. It
being completely rigid. This simplification enabled us can be seen that the central line of the anti-prism form is
to obtain maximum values of the occurring edge forces the one closest to the funicular solution (Fig. 6). The amount
for each folded form and to compare them with each other. of flexural action generated in order to resist the load
Furthermore, the boundary conditions were modeled depends on how much the folded form deviates from
as if pinned, allowing rotation, but no movement in all the funicular. A substantial difference observed between
three directions. The plates were modeled as convention- the rhombus-based form and the other two structures is
al shell elements and their geometry was defined at a result of their fundamental topological distinction. The
the reference surface that coincides with each plate’s rhombus-based form consists of quadrilateral faces with
mid surface. In this way, a three-dimensional continuum vertices alternately lying on the outer and inner surface
could be approximated with a two-dimensional theory of the hollow cylinder, i.e. the distance between these two
due to the fact that the plate thickness is small compared surfaces is equal to the maximum static height of the sys-
to its remaining dimensions. Material properties of tem. On the other hand, the anti-prism-based and the
simplified linear elastic, orthotropic, 21-mm-thick, struc- trapezoid-based forms consist of faces where all vertices
tural LVL timber panels were fed into the model with are situated on the outer surface. This topological differ-
values according to the VTT No. 184.15 ence makes the structural height of the latter two forms
Fig. 5 shows vertical displacements along a more uniformly distributed along the transverse cross-
normalized span of the mid transverse cross-section of section, consequently providing a higher overall rigidity.
the observed structures. The overlay plot of the deformed
and underformed structure shape, uniformly scaled for
all three forms, is shown below. The lowest vertical dis- 4 Observations on a prototype structure
placements under a dead load were found for the anti-
prism-based structure. The ratios between the values ob- A prototype structure was constructed and tested in or-
tained for the other two observed structures and the der to explore the behavior of timber folded-surface struc-
anti-prism-based form were equal to about 1.5 for the tures under load. Initial findings concerning the fabri-
trapezoid-based structure and up to 3.5 for the rhombus- cation and assembly of its constituting elements were
based structure. Even higher discrepancies were found collated. Integral mechanical attachments were used as
for the values of the bending moments about the indi- connection details, and their geometry was designed ac-
vidual edges. The ratio between the rhombus-based and cording to certain extant examples.15,16,17 Regarding the
the anti-prism-based structure amounted to 5. Values for material, two types of engineered wood products were
shear forces were highest for the rhombus-based struc- considered: Kerto-Q structural grade laminated veneer
ture, with a ratio of 3 : 1 when compared to the value for lumber (LVL) and cross-laminated timber (CLT) panels.
the anti-prism-based structure. It was noticed that the more homogeneous and mech-
The ratio between the trapezoid-based and anically sturdy peeled-veneer laminates offer particular
the anti-prism-based structure was equal to about 1 : 3. advantages over CLT panels. First, considerably thinner
A distinct distribution of shear forces was found when cross-sections are possible, making lighter-weight struc-
comparing the rhombus-based structure with the other tures possible. Second, an important shortcoming of
two structures. The shear forces appear on all edges of CLT was recognized while milling the panel edges. As
the rhombus-based structure, whereas in the anti-prism- CLT panels consist of several layers of longitudinal
based as well as the trapezoid-based structure they timber planks that are not glued to each other on the
are only present at the edges that are not parallel to the sides, depending on the angle of the joint with respect
main axis of the structure. The edges are shorter in the to the individual layer plank orientation, there is a risk
trapezoid-based structure than in the anti-prism-based that considerable pieces of the joint could simply chip off
structure. This accounts for the aforementioned increase after manufacture. Another issue when using integral
in shear. mechanical attachments was recognized early on in the
37
lmin
Fig. 8 a Fig. 8 b
leff
l
Fig. 7
prototype design process. Due to specific joint geometry, recognized. In such a form, the edge-to-edge connectivity
the effective connecting length of each edge is always is realized between every element along the span and the
shorter than its total length. This can pose a problem neighboring transverse string element (Fig. 8b). In contrast
when trying to achieve an efficient connection of a num- to this is the isosceles triangle solution where a connec-
ber of plane elements which converge at one vertex. In tion is only achieved in every second element (Fig. 8a). Ad-
order to reduce the vertex valence number, the prototype ditionally, by using isosceles trapezoids, a wider span can
was designed using the isosceles trapezoid-based folded be realized with the same number of elements along a
form (vertex valence = 4) rather than the anti-prism- transverse cross-section.
based one (vertex valence = 6). Using the RhinoPython application program-
The minimum edge length of the trapezoid ele- ming interface, a computational tool was developed that
ment was restricted in order to secure a minimum con- instantly generates both the geometry of the individ-
nection between the panels and to provide at least one ual components and the machine G-code required for
pin on each adjoining plate edge (Fig. 7). As a result, an ad- fabrication. Exploiting this geometric freedom, we have
ditional benefit of the trapezoid-based folded form was tested our computational tool by designing a double-
curved folded-surface prototype with alternating convex-
concave transverse curvature (Fig. 9). The built structure
spanned three meters and was constructed with 21-mm-
Fig. 7 Connection detail in a thick Kerto-Q structural grade LVL panels (7-layer,
two-panel assembly I-III-I). The structure’s total weight amounted to only
Fig. 8Edge-to-edge connectivity between the faces of 192 kg. Boundary conditions that restrain displacements
a) an anti-prism-based folded form and of the supports in the transverse direction were applied
b) the isosceles trapezoid version
on both sides. This was achieved by fabricating form-
Fig. 9 Timber folded-surface prototype fitting lateral timber supports that made it possible to
Fig. 10 Load-displacement curve of the load test straighten the edges in order to position two large
50
40
Force (kN)
30
20
10
0
0 20 40 60 80 100
Displacement (mm)
Fig. 9 Fig. 10

concrete blocks on each side of the structure. A longi- References
tudinal line load was introduced gradually by applying 1 Sedlak, V. “Folded surface structures.” Architectural science
pressure to a leveled steel beam placed on top of the struc- review, 1978, 21(3), 58–59.
ture. Vertical displacements were measured at the center 2 Buri, H.U. Origami-Folded Plate Structures, PhD Thesis,
point of the transverse cross-section. Fig. 10 shows the EPF Lausanne, 2010.
load-displacement curve obtained from the test. The 3 Schineis, R. “Gefalteter Klangkörper Musikprobensaal
dashed line represents the first applied load cycle where Thannhausen (Thannhausen Rehearsal Room),” in:
the maximum load reached up to the approximated 10. Internationales Holzbau Forum, Forum-Holzbau, CH-Biel,
Garmisch-Partenkirchen, 2004.
proportional limit of the curve.
It can be seen that the load of 20 kN corresponds 4 See note 2 above.
to a vertical displacement of 18 mm. Next, a load up to 5 EN 1995-1-1. Eurocode 5: design of timber structures—part
1-1: general-common rules and rules for buildings. European
the failure of the structure was applied. This displayed a
Committee for Standardization, Brussels, 2004.
high structural efficiency (ratio of the maximum load
6 La Magna, R. et al. “From nature to fabrication: Biomimetic
over the dead weight of the structure) of the tested proto- design principles for the production of complex spatial
type, which reached to 23.44 when loaded with 45 kN. structures,” International Journal of Space Structures, 2013,
The failure of the structure occurred in the con- 28(1), 27–40.
nection detail when the out-of-plane rotation of the edges 7 Robeller, C., A. Stitic, P. Mayencourt, and Y. Weinand.
closest to the applied line load exceeded the joint capacity. “Interlocking folded plate, integrated mechanical attachment
This kind of failure mode indicated that the connections for structural timber panels,” in: Block, P., J. Knippers, N. J.
Mitra, and W. Wang (eds.) Advances in architectural geometry
used can provide a certain ductile capacity for the folded 2014: Proceedings of the AAG’14 Conference, Springer, London,
surface system made from brittle timber members. In or- 2014, 281-294.
der to assess the overall ductility of the structure and the 8 Ibid.
influence of the joint semi-rigidity, the prototype was 9 Botsch, M. et al. Polygon Mesh Processing. AK Peters, 2010.
modeled in the FE analysis software with completely
10 Trautz, M. and R. Herkrath, “The application of folded plate
rigid joints, in the same way as previously explained principles on spatial structures with regular, irregular and
(see section 3.3). The load was applied in the same manner free-form geometries,” in: Domingo, A. and C. Lazaro (eds.)
as in the test, to the limit of 20 kN, which resulted in a International Association for Shell and Spatial Structures:
2 mm vertical displacement. Such results suggest that the Proceedings of the IASS Symposium on Evolution and
Trends in Design, Analysis and Construction of Shell and Spatial
introduction of semi-rigid connections has a significant
Structures, Valencia, 2009.
influence on the overall structural behavior and can po-
11Frostick, P. “Antiprism Based Form Possibilities for Folded
tentially provide a beneficial ductile behavior for timber Surface Structures.” Architectural science review, 1978, 21(3),
folded-surface structures under load. However, further 59–67.
studies are needed to precisely quantify their effects. 12 Huybers, P. See-through structuring: A method of construc-
tion for large span plastic roofs. PhD Thesis, TU Delft, 1972.
13 Bechthold, M. Innovative Surface Structures: Technology and
5 Conclusion Applications. Taylor and Francis, 2008.
14 Mattoni, G. Folded Plate Structure—Design and Analysis of
The prototype presented here demonstrated the realiza- Woodworking Joints for Structural Timber Panels. MA Thesis,
tion of a very lightweight structure with a weight-to- ENCP and EPFL, 2014.
surface-area ratio of only 11.5 kg/m 2 . It successfully com- 15“VTT certificate for structural laminated veneer lumber,”
No. 184–03, 24 March 2009, Certification body (S017,
bined the structural advantages of timber panels with
EN 45011) accredited by FINAS.
the efficiency of folded plates, while conforming to the
16 See note 6 above.
constraints of integrated mechanical attachments. Fur-
thermore, it demonstrated the high structural capacity
Roche, S., C. Robeller, L. Humbert, and Y.Weinand, “On the
of the chosen folded form and validated the fabrication 18
semi-rigidity of dovetail joint for the joinery of LVL panels.”
and assembly methods used for its construction. Howev- European Journal of Wood and Wood Products, 2015.
er, it should be noted that further research is required
regarding the physical connections used. Their mechan-
Reprinted from International Journal of Space Structures,
ical behavior and load-bearing capacity still need to be Vol. 30, Num. 2, 2015.
confirmed. Nevertheless, a great structural potential is
recognized in the proposed timber panel structural sys-
tems, and the forms discussed within this paper are
considered to be of significant influence in establishing
these structures on a building scale.
39
1. 3
“It’s important to progress from research

to a product”
Interview between Ueli Brauen (Brauen & Wälchli Architectes, Lausanne) and Yves Weinand
April 21, 2015
Yves Weinand: Let’s go back to the history of Ueli Brauen: I would like to start with your research, and
IBOIS, which you are familiar with. When I how you came to work on folded structures. You have an
became the director of the laboratory, we initial- engineering and architecture background. Hans Ulrich
ly began by placing an emphasis on architecture Buri undertook this first research on folded structures
and form creation. The first theses of Hans Ulrich as an architect. When you take an element like the
Buri1 and Ivo Stotz2 dealt with the creation of fold—which is familiar to everybody—as an object of
overall forms. Hans Ulrich Buri describes folded research, it’s very interesting from an architectural
structures in his “Origami-Folded plate struc- point of view, because it opens up new horizons in the
tures” thesis; he created a tool that allows them field. At a given time, the engineer intervenes and
to be generated while respecting a certain undertakes research at a structural and mathematical
number of geometric constraints. It’s intuitive level, to grasp the forms and determine their structural
work, inspired by the Japanese art of folding. potential.
This very graphic work illustrates the wide It does not surprise me that people consider you to be
range of geometries that can be used to create “unusual,” because there are plenty of specialists, who
pleated structures. But the approach remains are very competent in their domains, but the great chal-
purely geometric, as mechanical or structural lenge resides in the capacity to see links with other
optimization aspects are not addressed. domains. What is really interesting in your research is
Then with the thesis of the architect, Christopher that it combines the culture of architecture with that of
Robeller, 3 we were interested in wood-wood engineering. The drive for architecture is very strong in
assemblies. The folded structures we had al- your work. When you work with a form, it must be pure.
ready studied were very rigid. However, during This is also true for assemblages. You are interested in
tests on rupture, it is the assembly or connec- wood-wood assemblages that don’t need any other
tions between panels that are the first to give binding material. That is your sense of aesthetics and
way. So long as we seek to work with thin panels, elegance speaking. Your research interest would be
we are not able to respect existing norms, since different if you were trained solely as an engineer.
minimum distances between screw axis and For me, it’s clear that it will take some time before you
panel’s edge need to be respected. 4 Therefore, are able to bring together static and functional require-
the question of joints is of great interest to us. ments, with the requirements of form (whether it’s for
In the context of her thesis, civil engineer folds or assemblages). You will either manage to find
Andrea Stitic was interested in the topology of a structural form for a broad function that is convincing
form. Her research involves the integration thanks to its form, or you will find a form that is struc-
of joints or connections, while simultaneously turally reliable. For the latter, price will be a deter-
considering manufacturing constraints. She was mining factor. A system can be interesting at one scale,
able to develop calculation models, that take but no longer at another. Each structure has its optimal
into account the spring rigidity of wood-wood scale. You dream of large folded structures, such as the
assemblages. She’s interested in the topology of Saint-Loup Chapel. This increase in scale, aligned with a
form, which is unusual for an engineer. It’s truly need for rigidity, will be a difficult challenge to address.
an architecture/civil engineering undertaking.
In terms of innovation, the engineer can contrib- Y. W.: We are not talking about mega-structures
ute a lot to the appreciation of a global form. here, but structures with a range of 40 to
This brings me to my first question: What do you 45 meters. A folded structure would be strong
consider to be the potential of folded structures? enough for the scale of a sports arena, such
Do you think they could be used for types of as the one in Yverdon.5 We have noted—and this
buildings other than pavilions? was also the case in the context of the Isles Sports

Center in Yverdon—that when the thickness of Y. W.: I’m fascinated by vernacular architecture.
the wood becomes too significant, the costs sky- I’m participating in a drone airport project
rocket. We therefore reconsidered the project and in Africa, which will include two systems: one
designed a hybrid structure: a lattice beam rein- medical, the other commercial. The constructed
forced by a continuous panel. It’s for this reason surface on stilts represents approximately 15 by
that we would like to resolve the problem of 50 meters. It will consist of covering areas for
assemblages and, in doing so, that of cost. Even drones, which will be stored beneath these
in 2006/07, at the time of the Yverdon project, our structures.
research was not sufficiently advanced for that. I suggested providing a range of tool data to
people on location, who could then use products
U. B.: And today, do you have a less expensive solution derived from wood for an in situ production.
that could work? I have always liked very sharp angles, patterns
reproduced in a very repetitive fashion, which
Y. W.: It’s a nuanced answer. We are still unable to can often be found in African carpets. By trans-
convince enterprises to open up entirely to mitting our technology, adaptable on location,
production automation and to thereby reduce for cutting in Africa, the architectural expression
the costs of production for folded structures. could find a local reference. It’s not about impos-
However, as long as wood is increasingly applied ing a modernist attitude here, but transmitting a
to wooden constructions, more and more enter- tool that allows for adapting or specifying archi-
prises are becoming specialized in wood manu- tectural solutions based on their cultural context,
facturing and the situation could change. while taking into account the essence of the avail-
More commonly, companies charge for the cut able wood, and the local architectural expression.
depending on the number of panels, rather than We could create sections of columns with acute
accounting for “machine-hours.” In considering angles, of about 15 meters in height, by using
the cost of the panel and the cutting time, then connections that would be simply clipped.
yes, I think production costs could be reduced. This would imply the use of an automated cut-
So, to answer your question, I have a less ex- ting machine (CNC) on location.
pensive solution, so long as you are producing Today, from an architectural point of view, the
in-house. Unfortunately, to date we have not elaboration of construction detailing has become
yet found an enterprise that would be willing difficult because of the multiplicity of construc-
to produce at a lower cost. tion layers. The connection is no longer magni-
fied, as it was in the past. I have the impression
U. B.: Yes, because they are protecting their market. that wood-wood joint development, like what we
They have no need to go any further, since demand is are doing, could be a return to basics.
there in any case. Glued-laminated is efficient and
automated. I think we need to gain the interest of a U. B.: Attention to detail still exists today, but in specific
company that would develop the folding and assem- buildings, where we invest money for cultural or sym-
blages with you, and would have a return on investment bolic reasons. But in most architectural applications, the
in the future. But I also think that wood-wood assem- technical detail must be effective, in that it must be
blages could be very interesting for residential build- economical and sustainable. In those cases, the aesthetic
ings, for both economic and ecological reasons (no glue, cost is of little importance. We have much to learn from
no additional elements or materials). By putting together artisans today. Attention to detail is important. In terms
prefabricated elements, assembly and dismantling of performance, costs, and sustainability, it must be
would be facilitated. There’s tremendous potential. optimal. I understand that, as an architect, you have a
41
desire for form, which comes through in the elegance of Y. W.: We recently presented our work in Frank-
assembly. But in Africa, using a drone and making it furt, 6 where we were approached by DETAIL
work is far more important than the building. In that magazine. They wanted us to focus on the theme
precise case, I would not talk about the idea of the form of construction detail, all the while showcasing
as a cultural value, because it adds something excessive and demonstrating the structure in its entirety.
that leads to the impression of a cost overrun. In any By the way, we currently have a book in prepa-
case, the form is a question of adaptation over time. The ration, in which we present the detail in relation
Roman aqueduct is a 100% utilitarian building with the to overall form.
purpose of transporting water. Today, most people have
an emotional reaction to these forms, as the arch has U. B.: I admire the overall work you have been doing for
become something familiar to us. The mast of a high- years, with the means at your disposal, the engineers’
tension line has the same structural aesthetic value to mathematics and tools. I’m under the impression that at
me, but nobody is moved by it, because we have yet IBOIS you are in a sort of playground, where you enjoy
to adopt its form as a cultural value. Architecture yourselves, and have the tools to research the properties
must seem self-evident to a user. To arrive at this self- and potential of your “toys.”
evidence and touch on the collective culture of the user
is important. Y. W.: We always work on very specific programs,
but I’m not sure how we could go towards more
Y. W.: Can construction detail once again lead generalist applications.
to cultural value? In certain cases, the architec-
tural detail holds particular importance. But U. B.: One must recognize that you are involved in
generally, in ordinary architecture—as you put fundamental research, not for profit. It’s important to
it so well—it’s rarely the case. For me, the con- progress from fundamental research to arrive one day
structive detail has much to contribute to archi- at a product, the result of which was not distorted by
tectural expression. Its elaboration should be a economic or political constraints. You are free, and so is
priority concern. This was once the case and your research, and it’s normal that such things take
has been the case for architecture in Japan, or time. I like collaborating and dreaming with you.
for architects like Mies van der Rohe. I appreciate your knowledge and your desire to create
a pure and elegant architecture, which we see less and
U. B.: Yes indeed, that would be wonderful. But there are less nowadays.
other examples, such as Apple, where everything was
derived from one person who had an aesthetic sensibili- Y. W.: In the end, my intuition is that we should
ty, an attention to detail and business. It was the first evolve towards the scale of larger buildings
time that we considered adapting machines to humans. in order to be convincing. I’m working actively
Which is to say that the common culture of humans towards building on a larger scale but, for that,
allowed for the instantaneous use of a complicated ma- we have to create partnerships with industry.
chine. For me, architecture must have this same sense of We have the opportunity to perpetuate our
certainty; we must find the most obvious entry point research in the context of a NCCR7. This allows
but, once we’re there, we must immediately understand us to undertake applied research, with the
which direction to follow. I hope that attention to detail support of the SNF8, and to collaborate with
will be developed with this same sense of evident industrial partners.
functionality.

References
1Buri, H. U. “Origami —Folded Plate Structures,
thesis n° 4714.” EPFL, Lausanne, 2010.
2Stotz, I. “Iterative geometric design for architecture,
thesis n° 4572.” EPFL, Lausanne, 2009.
3 Robeller, C. “Integral Mechanical Attachment
for Timber Folded Plate Structures, thesis n° 6564.” EPFL,
Lausanne, 2015.
4EN 1995-1-1:2004+A1, Eurocode 5 Design of Timber
Structures.
5Isles Sports Center, Yverdon-les-Bains, Architects: Brauen
& Wälchli et Yves Weinand; Engineers: Bureau d’études
Weinand; Client: Town of Yverdon-les-Bains, Competition:
1991, Project: 2006–09, Implementation dates: 2010–11.
6Advancing Wood of Architecture, Deutsches Architektur-
museum, Frankfurt, 27.03.2015.
7 National Centre of Competence in Research Digital Fabrica-
tion (NCCRs). The National Centres of Competence in Re-
search –NCCRs – are a research instrument of the Swiss
National Science Foundation. They promote long-term
interdisciplinary research networks in areas of strategic
importance for the future of Swiss science, economy and
society (www.dfab.ch)
8 FNS/SNF Swiss National Science Foundation (www.snf.ch)
43
1. 4
The Chapel of the Deaconesses of St-Loup

at Pompaples, Switzerland
Marielle Savoyat
Design Local Architecture/Danilo

Localarchitecture, Danilo Mondada
Mondada Following a competition awarded in the summer of 2007
sowie SHEL Architecture,
architecture firm and ShelEngineering
Architecture,
and to renovate the buildings occupied by the St-Loup Dea-
Production Design
Engineering, and Production Design conesses in Pompaples, a wooden chapel with an abun-
dance of folds was realized in 2008. It was intended to be
Client St-Loup Deaconnesses Institute
Diakonissengemeinschaft Saint Loup,
a temporary structure, as renovations to the original
1318 Pompaples, Schweiz
Switzerland.
chapel in one of the buildings of the deaconess institution
Contact:
Kontakt: Sister Marianne
Schwester MorelMorel
Marianne
would only be accessed again in early 2010. However, the
Completion 2008 sisters fully appropriated this new innovative space and
wanted to keep their temporary timber chapel, even after
Location Pompaples,
Pompaples, Switzerland
Schweiz the renovation period. The idea of building this ephemer-
al, yet warm and welcoming place of worship—using high-
quality, inexpensive material—became an obvious and
preferred solution. In collaboration with the Laboratory
for Timber Constructions (IBOIS) at the EPFL, Swiss Feder-
al Institute of Technology, Lausanne, an innovative struc-
Fig. 1 Detail ture based on panels of cross-laminated timber was creat-
Fig. 2 Structural panels seen from the inside ed, based on research undertaken by Hans Ulrich Buri
1
2
2 1 24 mm pine eaves board
3 2 20 mm three-ply lam. Sheeting
4 prematurely greyed, screw fixed
4 to 30/60 mm pine distance
pieces
3 bituminous sealing layer
5
4 2 mm sheet steel bent to shape
2
5 60 mm lam. cross-boarded
3 sheeting, inner face untreated
6
7 8
6 fabric covering on 20/20 mm
wood framing
7 60/120 mm pine column
8 10 mm polycarbonate cellular
slab in 10 mm alum. channels
9 40 mm lam. cross-boarded
9 sheeting
10 30/30 mm aluminium angle
9
10
9 11 100/60 mm pine plate
11
12 timber bearing plate
12
13 Ø 8 mm copper lightning
13 14 conductor
14 200/150 mm concrete blocks
Fig. 1

Fig. 2
45
Fig. 3
Fig. 3 Sun and shade expression of lateral facade

Fig. 4 View of the entrance with its translucent enclosure
Fig. 5 The building in its countryside setting
Fig. 4 Fig. 5

and Professor Yves Weinand on folded wooden structures,
inspired by origami, the Japanese art of folded paper. The
audacity of the sisters was mitigated: The cost of the new
space was equal to that of renting a provisional structure
on the construction site.
IBOIS researcher Hans Ulrich Buri designed a
digital tool that generates folds inspired by the art of
folded paper. Shel analyzed the structural behavior and
established construction plans. Thanks to this invention,
the chapel was delivered in the record time of two months.
Prior to this, only small-scale prototypes had been pro-
duced at the IBOIS.
The fact that a thin surface becomes more rigid
when it is folded was applied. The principle made it
possible to construct the envelope and chapel structure
with large wooden panels, rather than with a traditional
post-and-beam frame. This technique was feasible thanks
to advances in the wood industry. It is now possible to
produce massive surfaces with irregular and differing
cutouts for each piece at no additional cost. The panels are
cut with a numerically-guided saw, based on 3D drawings.
This allows the construction of designs with a high level
of complexity. In the case of the St-Loup Deaconesses‘
chapel, the massive 45-mm-thick plywood vertical pan- Fig. 7
els and 60-mm-thick horizontal panels in spruce, are all

different sizes. The panels were simply assembled by
metal plates and fixed in place with screws, before being
Fig. 6 View on the altar
covered on the exterior with a layer of bituminous water-
Elevations
proofing and clad in three-ply wood panels. It is impor- Fig. 7
tant to note that every second ridge is vertical, thus facil- Fig. 8 Plan and section
itating on-site assembly. The inverted roof surfaces
distribute the water runoff on both sides of the long ele-
vations. The two gabled facades, composed of a textile
membrane and skeletal reliefs with diagonal drawings,
recall the composition of a stained glass window and al-
low natural light to enter the space.
The coherence between structure and architecture
and the choice of using wood, creates an extraordinary
sense of serenity and a profound architectural quality.
a a
Fig. 6 Fig. 8
47
2 Advanced architectural geometry
2. 1 Regular and fractal-shaped,

discretized tensile structures 50
Yves Weinand
2. 2 Iterative geometric design for architecture 58

Ivo Stotz, Gilles Gouaty, and Yves Weinand
2. 3 Geometrical description and structural analysis

of a modular timber structure 70
Sina Nabaei, Yves Weinand, and Olivier Baverel
2. 4 “Discover processes where the parameters

are insufficient” 78
Interview between Mark Pauly and Yves Weinand
2. 5 Modular pavilion:
a structure for the Paléo Festival 88
Marielle Savoyat
49
2. 1
Regular and fractal-shaped, discretized

tensile structures
Yves Weinand
In the previous chapter, it was shown ple, by wind or earthquakes—produce

how the overall stiffness of a structure bending in domes that cannot be ab-
can be increased by creating folds. sorbed well. Historically, civil engineers
The origami tool undertakes this within have mainly specified standard areas for
the following constraint: the geometry shell structures. Well-known examples of
and folds generated can always be re- these are buildings by Félix Candela and
duced to a flat surface. The following Eduardo Torroja. Pier Luigi Nervi also
project envisages embedding folds into envisaged the use of folds in these stand-
open, free areas. ardized areas as, thanks to the folds,
asymmetrical and horizontal forces can
Shell structures are known to perform be better absorbed. Furthermore, light can
well under their own weight and vertical penetrate into the domes.
loads, because these forces are dissipated
as membrane forces, thereby avoiding The aim of the work presented here was
any bending in the shell (except at the to develop a tool that could generate folds
support points). It is also known, however, in regulated as well as free-form areas.
that horizontal loads—caused, for exam- In addition, our aim was to integrate
Comparison of standard and folded areas.

Fig. 1
Identical dimensions and support conditions are evaluated
for a standard surface and an equivalent surface with folds
(under snow load).
Using the developed tool to create free forms,
Fig. 2
standard surfaces and folded or fractal surfaces
can be generated. The division pattern is also visible.
d Zmax = 220 mm d Zmax = 45 mm

Fig. 1

Fig. 2
51
the subdivision of the surfaces into the first chapter, can also be solved by the
form-finding process. use of folds. The user is free to suggest
and generate changes in form, depending
The starting point and the basis of our on their architectural or engineering-
work are iterative functions that have oriented considerations. Thus, the intro-
the ability to retain a selected basic topo- duction of foreign design elements is
logical form regardless of the trans- rendered unnecessary.
formations applied to the geometry.
Thus, it is possible to introduce standard- Fig. 1 shows how a standard surface is
ized or fractal curves as transformations. improved by the subdivision and folding
The surface-generated fractals have the of the overall geometry. The mechanical
advantage of introducing folds into the behavior of both geometries—folded
overall geometry. With the help of weight and unfolded—is tested. The rigidity of
points, the folds can be strengthened. the folded geometry shows an improve-
The subdivision process can be carried ment factor of 5, their deformations
out either homogeneously, over the entire thus amounting to a mere 20 percent of
surface, or locally, thus supporting the those found in a standard, unfolded shell.
structural requirements. The partition function also defines the
local geometry of the edges or borders of
The developed form-finding tool can thus each element, which is also exportable.
also be used as a mechanical optimization
tool: folds can be moved or increased. The prototype described at the close of
The issues regarding the necessary edge this chapter was initiated by Ivo Stotz
reinforcements of domes, raised in the and Gilles Gouaty, who strove to develop
Fig. 3 The tool describes the overall

geometry. After subdivision
into separate parts, the precise
geometric description of each
facet is specified.
Fig. 4 The data is then transcribed
into machine code, which controls
the milling head.
Fig. 3 Fig. 4

3517.91
2351.51
1887.10
351
7.9
2124.70
1
2369.38
35
24
19
.09
76
. 00
.70
2124
1976.00
Fig. 5 Fig. 6
Fig. 5 Floor plan of the generated shell construction

Fig. 6 Axonometric of the generated shell construction
Fig. 7 A prototype of the generated shell construction
53
existing structures by the use of discrete Observations of the behavior of the overall
elements. The mold was generated by us- geometry have shown that even freely
ing the tool they described. It remains up rotatable edges are sufficient for the
to the user to control the criteria for the overall rigidity of the system, though
shape or the parameters that fit the mold. connections remain necessary. However,
if one considers the connections to the
Which type of connection should be used overall geometry and their complexity, the
to join the panels? The digital tool con- connections shown here do not solve the
tains the description of the panels’ geo- problem.
metry. This description can also be used
for the assembly of the connections. For this reason, a further experiment
The structure must be able to withstand was undertaken with a different kind of
lateral forces, regular forces, and bending shell structure, this one composed of
moments. The formation of these connec- facets. The connections were directly
tions in timber is significant, as this has a considered in the initial design phase.
considerable effect on construction costs. If the connections developed in the Stotz/
Gouaty prototype were to be added to
Experiments were carried out on four the example here, then incisions would be
types of connections and mechanically made into the panels in order to secure
tested. In the initial attempts, only self- the connections by means of contact
tapping, diagonally mounted screws zones.
were used. Here, torsion along the edges
is hardly prevented. A folded sheet is The advantage of the recommended integ-
inserted into the second row in the axis ral connections between the panels is
of the plates. Here, a lever arm at half- that they do not require any additional
plate thickness is used. measures or connections. The connection
is an integral part of the plate and is
In the third test series, two folded metal produced simultaneously with the cut
sheets divide the thickness of the panel panel. Even if one can assume that this
into thirds, thereby increasing bending structure looks promising, it is also
rigidity. The fourth series of tests com- clear that mechanical improvements are
bined plates with screws, thus achieving not only possible but are even on the
the highest bending rigidity possible. verge of being realized.
The previous example shows a facet-like Two assessment steps took place with
structure, the execution of which requires students from the Civil Engineering
specialized connection technology. Department of the ENAC/EPFL. First, the
nodes of the structure were analyzed

Concave element
Convex element
Fig. 8
U-joint
Convex element
Concave element
Fig. 9
Fig. 10
Fig. 8 Prototype of a test body in which two 10-cm-thick laminated

timber panels are connected with oblique, self-sinking screws.
In the lower image, four 10-cm-thick laminated timber panels are
connected using such screws. Furthermore, the panels are also
screwed together vertically.
Fig. 9 Prototype of a test body. Additional sheets are inserted
between the plates.
Fig. 10 Axonometric view of the contact zones
Fig. 11 Detailed view of the shell construction. The panels are
interwoven or wedged together.
Fig. 11
55
Fig. 12 Completed arch pavilion
and mechanically calibrated, and the relates to precisely this point: how can
frictional resistances were modeled, one initiate an optimization process that
measured, and calibrated. As shown in takes both the global issues of the overall
figs. 8 to 12, geometric manipulations of geometry and the local requirements
the overall structure were proposed, mod- of the connection details equally into
eled, and recorded in a secondary step. account? For this, geometrical, mechani-
A satisfactory solution for the connections cal, and manufacturing-related criteria
can only be found after the overall geo- should be considered.
metry has been revised. The structural
optimization described by Sina Nabaei is As the following discussion with
also affected by a modification of the local Mark Pauly highlights, it is of interest
connection. The latter is, however, only to expand the interaction of edge con-
made possible by the manipulation of the straints. If solely geometric constraints
entire geometry. One of the key issues play a role with the form-finding tool
that has concerned us for a long time developed by Stotz and Gouaty, then

mechanical form-finding remains
intuitive. Thus, the generation and distri-
bution of folds remains intuitive and a
quantitative control is not incorporated.
The pavilion shown in fig. 12 introduces
a connection technology that will be
discussed in the fourth chapter.
In fact, it is already possible to expand the
digital form-finding tool, and thereby
expand the overall geometric conditions
respecting material-related and mechani-
cal constraints. An example of this is
presented in the work on mechanical
form-finding shown in Chapter 4.
57
2. 2
Iterative geometric design for architecture

Ivo Stotz, Gilles Gouaty, and Yves Weinand
This
Dieses
interdisciplinary
interdisziplinäre research
Forschungsprojekt
project is presented
wird von 1 Introduction
einer
by a group
Gruppe of aus
architects,
Architekten,
mathematicians,
Mathematikern and com-
und
Informatikern
puter scientists vorgestellt,
who research die nach
newneuen
methods Methoden
for the In order to present the geometric design method studied
efficient
für die effiziente
realization Realisierung
of complex komplexer
architectural architek-
forms. here, the mathematical background must first be clari-
The
tonischer
present Formen
work investigates
suchen. Im vorliegenden
methods of iterative
Beitrag fied. Before explaining the principles of transforma-
geometric
werden Methodendesign inspired
der iterativen
by theFlächengestaltung
work of Michael tion-driven geometric design, a series of historic exam-
Fielding
präsentiert,Barnsley.
die auf Several
der Arbeititeratively
von Michaelconstructed
Fielding ples will be examined. This will introduce the reader to
geometric
Barnsley basieren.
figures will
Es werden
be discussed
mehrere in iterativ
order tokon-
in- the methods of iterative geometric design. The relation-
struierte
troduce the geometrische
notion of Formen
transformation-driven
erläutert, um damit geo- ship between the mathematical method of geometric
das
metric
Konzept
design.der Theaufdesign
Transformationen
method studied basierten
allows surface design and the physically constructed building
Flächengestaltung
interaction with thevorzustellen.
design, forming Die affine
untersuchte
trans- will be shown by examples in the second part of this
Bemessungsmethode
formations and generating ermöglicht
discrete diegeometries.
interaktive presentation.
Veränderung
Furthermore, der Formthe durch
handlingaffineofTransformatio-
specific con-
straints
nen, wodurch
is discussed.
diskrete Geometrien
Geometrical entstehen.
and topological
constraints
Außerdem aim towird
facilitate
der Umgang
the production
mit bestimmten
of archi- 2 Mathematical background
Randbedingungen
tectural free-form objects.
erläutert. A surface
Bestimmtemethod geometri-
based
on
schevector
und topologische
sums is studied,Randbedingungen
allowing the sollen
designdieof 2.1 Monster curves
Herstellung
free-form surfacesarchitektonischer
that are entirely
Freiformobjekte
composed er- of The Cantor set (Fig. 1), also called Cantor dust, is
planar
leichtern.
quadrilateral
Es wird eine elements.
auf Vektorsummen
The combination
basieren-
of named after the German mathematician Georg Cantor. It
de
theMethode
proposed untersucht,
surface method
mit der sich
and Freiformflächen
transformation describes a set of points that lie on a straight line. At the
gestalten
driven iterative
lassen,design
die ausschließlich
provides new ausform-finding
ebenen Vier- end of the 19th century, this figure attracted the atten-
ecken
possibilities
bestehen. whileDiefulfilling
Kombination
a number ausofgewählter
material tion of mathematicians because of its apparently contra-
Methode
and construction
und iterativer
constraints.
Flächengestaltung
Finally, the findings
mittels dictory properties. Cantor himself described it as a per-
Transformationen
are tested on a series eröffnet
of applications.
ganz neue Möglichkeiten
The studied fect set, which is not dense anywhere.1 Further properties,
für
testdie
scenarios
Formfindung aim to undevaluate
erfüllt zudem
the advantages
zahlreiche such as self-similarity, compactness, and discontinuity,
of
Randbedingungen,
discrete geometric diedesign
Werkstoffin terms
und of
Konstruktion
efficient in- were studied years later.
stellen.
tegratedSchließlich
productionwerden
of free-form
die Ergebnisse
architecture.
an diver- The geometric construction of the Cantor set can
sen Anwendungen getestet. In den Tests sollen die be explained as follows: Take a straight line segment, di-
Vorteile der diskreten Flächengestaltung in Bezug vide it into three parts of equal length and remove its
auf einearchitecture,
Keywords integrierte applied
Herstellung architektonischer
discrete geometry, IFS, middle third; divide again each of the resulting line
Freiformen untersucht
timber werden.
construction segments and keep removing their middle thirds. If you
repeat this for each of the new line segments, you will
end up with the Cantor set.
Architektur, angewandte diskrete The von Koch curve is one of the best-known
Geometrie, IFS, Holzbau fractal objects and among the first found. In 1904, the
Swedish mathematician Helge von Koch described it for
the first time. 2 The curve is constructed step by step.
Beginning from a straight line, a meandering curve with
strange properties is created:
– It does not possess a tangent, which means that
it cannot be differentiated.
– The length of any of its sections is always infinite.

T1
T2
T3
T4
Fig. 1 Fig. 2
Fig. 1 Cantor Set

Von Koch Curve
The geometric construction of the von Koch curve is iter- Fig. 2
ative, where each of the construction steps consists of Example of a free-form object designed
Fig. 3
using the presaged surface method
four affine geometric transformations. The primitive is
a section of a straight line, which is scaled, rotated, and
displaced by each of the transformations {T1…T4}. Four
duplicates are generated per construction step and each applied iteratively. In our case, a function is an affine ge-
of these will, in turn, produce four more duplicates in the ometric transformation. Iterative means that the con-
following construction step (Fig. 2). struction is done step by step. The input of a construction
step is the result of the previous step.
2.2 Iterative geometric figures What is really new in Barnsley’s work is that
The peculiar properties of the aforementioned the resulting geometric figures are not defined by the
objects led mathematicians to name them “monster curves.” primitive used, but rather by its transformations. As
In 1981, Barnsley defined a formalism based on Hutchin- shown in fig. 4, the construction of a Sierpinski triangle
son’s operator 3 that was able to describe objects such as may use a fish as a primitive. Analogous to this, the
these in a deterministic way. 4 His IFS-method (see section von Koch curve might be constructed on the basis of the
2.3) consists of a set of contracting functions that are letter “A.” The end result remains exactly the same.
Fig. 3
59
of transformation {T i} is applied an infinite number of
times. The attractor A is the unique non-empty compact,
such that
A = F Ti A
i
Fig. 4 The attractor is a theoretical object, which will not be

analyzed any further here. We will work on the inter-
mediate construction steps that are given by the follow-
ing sequence:
K n+1 = F Ti Kn
i
Where K represents the initial figure, the so-called “germ.”

Fig. 5 K might be any arbitrary object, for example a fish, as
shown in fig. 4, or a straight line segment, as used
for the construction of a von Koch curve or a Bézier curve
(Figs. 3 and 5) . Finally, there is the equation relating to
the geometric figure K in the construction step n→∞ to the
attractor A:
lim (Kn ) = A
n→∞
Fig. 6
Within the following, the modeled objects are defined as
the projection of an IFS via a projection operator P.
Sierpinski triangle and von Koch curve according to
Fig. 4
Barnsley’s IFS-formalism
Fig. 5 Iterative construction of a Bézier curve 3 Discrete iterative geometric design
Fig. 6 First construction steps of a Bézier curve
The strange properties of the geometric figures discussed
in section 2.1 and 2.2 concern the limit state; the attractor.
For practical applications, the theoretical object of the at-
The conclusion is that it is theoretically possible to use tractor is far less relevant than its intermediate construc-
any form of primitive for the construction of geometric tion state K. In the scope of this work, the construction
figures such as these. This led to the hypothesis that it is state K has the following properties that are beneficial for
basically possible to use construction elements as primi- the application of free-form geometries in architecture:
tives. Instead of using fishes like Barnsley did, we would – K is computational point by point
rather choose to use construction elements such as beams – The resulting geometry is always expressed by a
or panels. finite number of elements
In order to complete this series of introductory
examples, we would like to briefly address the Bézier 3.1 Transformation-driven geometric design
curve. In 1959, de Casteljau discovered a method for the As stated in section 2.3, the final aspect of the it-
construction of the curve known today as the Bézier eratively constructed geometry is defined by transforma-
curve. De CasteIjau’s method 5 is based on iterative con- tions. In order to control the resulting figure, the geomet-
struction (Fig. 5), which is extremely similar to the con- ric design method has to provide solutions that will allow
struction of a von Koch curve. The actual Bézier curve the manipulation of the transformations used for the
was analytically described by Bézier in 1961 as a poly- construction of the figure.
nomial function. His pioneering work forms the corner- Let us analyze an example of an iteratively
stone of today’s CAD software. constructed Bézier curve as shown in fig. 6. The figure
shows the first construction steps of a Bézier curve with
2.3 Iterated function systems (IFS) three control points P shown in red.
The geometric figures of the examples shown at
the beginning of this section are all defined by a set of p1,x p2 ,x p3,x
transformations {T i}. As Barnsley teaches us, the result is P = (p1, p2 , p3 ) = ( p1,y p2 ,y p3,y )
p1,z p2 ,z p3,z
indifferent to the initial object on which the set of trans-
formations is applied. This is true for the limit state, also
called attractor A, which is the resulting figure if the set

The transformation matrices shown above are fairly
standard. These matrices work with fixed values that
always produce the same result: a smooth Bézier curve
with three control points.
A closer look at the transformation matrices of the Bézier

curve shows that they have certain values in common.
For example: The column c1 of T1 as well as the last
column c6 of T2 is fixed. Furthermore, the last column of
Fig. 7
T1 is identical to the first column of T2 . These dependen-
cies guarantee that the resulting curve will be continu-
ous. The rest of the values are free and can be modified
At the beginning (n = 0), the initial figure K consists of according to the designer’s wishes.
the line connecting the end points of the control polygon lf we generalize the above-mentioned findings,
[p1, p3]. Per construction step, two transformations {T1, we can say that certain values of the transformation
T2} are applied to each element. The first three steps of matrices are constrained, whereas others are free. Based
the construction sequence are shown below. on the example of an iteratively constructed continuous
curve with two transformations and three control points,
PK0 = PK the general scheme of the transformation matrices is as
PK1 = PT1K0 FPT2 K0 = PT1K FPT2 K follows:
PK2 = PT1K1 FPT2 K1 = PT1T1K FPT1T2 K FPT2T1K FPT2T2 K
1 c2,1 c3,1 c3,1 c5,1 0
PK3 = PT1K2 FPT2 K2 = PT1T1T1K F7 FPT2T2T2 K
T1 = ( 0 c2,2 c3,2 ) , T2 = ( c3,2 c5,2 0 )
0 c2,3 c3,3 c3,3 c5,3 1
Each element of PKi can be computed by following the
construction tree shown in fig. 7. In order to act on the free values of the transformation
We use transformation matrices to describe af- matrices, we defined a graphic data input method. We
fine geometric transformations. We work with barycent- introduce the set of points si which we call “subdivision
ric coordinates, which means that each computed point is points” (Fig. 8). Each point si corresponds to the projection
based on a combination of the entry points. In the present via P of columns ci.
example of a Bézier curve with three control points, each
transformation can be expressed by a single 3 × 3 matrix: Si = Pci
1 0.5 0.25 By this equation, the values of the columns ci can be ob-
T1 = (c1, c2 , c3 ) = ( 0 0.5 0.5 ) tained by the position of the subdivision points s, relative
0 0 0.25 to the position of the control points P (Fig. 9) using the
equation below:
0.25 0 0
T2 = (c4 , c5, c6 ) = ( 0.5 0.5 0) ci = P –1si
0.25 0.5 1
Fig. 7 Construction sequence: Tree representation

Fig. 8 Representation of control and subdivision points
Acting on the transformation matrices by manipulating
Fig. 9
the subdivision points (blue)
Fig. 8 Fig. 9
61
ological and geometrical constraints. The constraints are
mainly dictated by physical and production conditions
from the field of construction. Within the following, a
few examples of different constraints are presented.
For example, an important point might be that
the free-form object will be built out of planar timber
panels. According to this, the geometric constraint de-
mands that the virtual 3D-model ought to be completely
comprised of planar panels. We will work on this con-
straint in sections 3.4 and 4.2.
Fig. 10
In section 2, we have presented a series of itera-
tively constructed objects. Not all of them are suitable
for physical realization. The Cantor set, for instance, is
simply a set of discontinuous line fractions. Since we
generally need material continuity (unless designing
ornaments or similar forms), we have to verify that the
created elements building up the geometric figure are
connected with one another. Continuity represents a
topological constraint.
In order to avoid complex detailing of the nodes
of a wire frame structure, it is advantageous to know
the number of bars coinciding in one node. To keep the
number of bars per node at 6, we might work with sur-
faces composed entirely of regular triangular faces. This
is a topological constraint.
On the one hand, constraints will make the phys-
ical realization of free-form objects easier. On the other
hand, they may limit the design possibilities and there-
Fig. 11
fore restrict the form-finding process, and this ought to
be avoided as far as possible.
Fig. 10 IFS-Curve curve design: adjustment of roughness /
smoothness 3.4 A constrained surface model
In order to create iterative surfaces that are
Fig. 11 Surface design by vector sum
entirely composed of planar elements, we will work on so-
called “vector sums.” Generally, classical CAD-software
computes NURBS-surfaces by tensor products, which have
3.2 Potential of transformation-driven the unsuitable property of being composed locally of
geometric design double-curved faces. Great effort is required for their
Whether a figure is smooth or rough only depends production. The principle of using vector sums, more
on the affine geometric transformations. The same curve precisely Minkowski sums 6, for the generation of free-
might be smooth or rough. By changing the subdivision form surfaces has already been studied by Schlaich7 and
parameters, the smoothness and the roughness respec- Glymph. 8 Surfaces such as these are combinations of two
tively can be adjusted, as shown in fig. 10. The input of curves. Fig. 11 shows the curves a and b. The vector sum of
the subdivision parameters is given by the position of any two segments of the curves (a, b) creates a parallelo-
the subdivision points (as we referred to them). Alongside gram, which is part of the entire surface. The surface is
the control points, which are widely known in classical completely composed of parallelograms and therefore it
CAD-software, subdivision points augment a variety of meets the geometrical constraint which requires that all
design possibilities. They provide a graphic way to its parts have to be planar.
manipulate the affine geometric transformations, which The discrete curves a and b used for the con-
are expressed in the user-unfriendly form of n-dimen- struction of the surface are represented by two lists of
sional matrices. These points work below the surface of points A and B. The resulting vector sum surface is repre-
the graphic user interface. sented by the quad mesh M:
⟶ ⟶
3.3 Constrained geometric design Mi,j = M0,0 + A0 Ai + B 0 Bi
The goal is to develop design strategies that make
Note: M 0,0 is the origin of the quad Mesh.
the design and production of free-form surfaces easier. It may be any arbitrary point in the design space.
Therefore, the geometric design should meet certain top-

division points allow the manipulation of the local aspect
of the surface: smoothing / roughening.
This surface method offers great design potential.
It unifies in one formalism the hitherto separate para-
digms of “smooth” and “rough.” Furthermore, it verifies
a certain number of geometric constraints, allowing the
optimization of the production of free-form architecture.
This will be shown by examples in the next section.
4 Applications
In order to realize physical buildings out of discrete vir-

tual geometries, the elements that constitute the 3D
models are replaced by construction elements. For an iter-
atively designed curve, the line sections will be replaced
by linear construction elements, such as planks or beams.
Fig. 12
In the case of a discrete surface, we replace its faces
by planar construction elements (panels, plates, etc.).
The replacement of geometric elements by construction
The design possibilities of vector sums are limited com- elements poses a certain number of questions, as the geo-
pared to NURBS surfaces. The quad mesh generated by vec- metric figures do not have physical dimensions like
tor sums is composed entirely of parallelograms. In order thickness. We will first discuss the more demonstrative
to augment the design capabilities, Schlaich7 and Glymph 8 case of a two-dimensional figure, namely the Bézier curve.
extended the method such that the resulting quad mesh
is composed of trapezoids as well as parallelograms.
In order to extend the design capabilities, we em-
ploy methods of projective geometry. The IFS-formalism
will be extended by the possibility of assigning different
weights to its control and subdivision points. Each point
will be defined by four coordinates (w, x, y, z). Assigning
different weights w to the points allows deforming of
the quad mesh more or less locally. Thus, the resulting
quad mesh is no longer only composed of parallelograms,
but largely of convex planar quadrilaterals. To illustrate
the effect of point weight editing on vector sums, fig. 12
shows the simple case of a regular quad mesh where two
point weights have been edited. Fig. 13 shows the same
principle applied to more complex meshes.
Within the scope of projective geometry, the co-
ordinates (w, x, y, z) are called homogeneous coordinates.
A point (x, y, z) in the model space R 3 with a weight w has
the corresponding homogeneous coordinates (w, xw, yw,
Fig. 13
zw). Reciprocally, each point of homogeneous coordinates
(w, x, y, z) has corresponding R 3 coordinates (x/w, y/w, z/w).
The path from homogeneous coordinates to R 3 is called
projection.
In order to work with IFS-subdivision in homo-
geneous coordinates, the control points P must be defined
in homogeneous coordinates. Here, the IFS-subdivision Fig. 14
and the computation of the quad mesh M are performed
using homogeneous coordinates. Finally, the result ob-
tained is projected back to the model space R 3. Fig. 12 Deformation of vector sum meshes by editing the
The proposed surface method functions with any weight of certain points
pair of discrete curves. However, the use of IFS-curves Point weight editing of smooth and rough vector sum
Fig. 13
creates the ability to control the global shape of the sur- surfaces
face via its control points. In addition to that, the sub- Fig. 14 Iterative Bézier curve
63
Fig. 15
Fig. 15 Shape study
Fig. 16 Reduced scale model
4.1 Discrete Bézier vault structure Once the shape has been defined, the curve will be subdi-
In the following example, we build a vault struc- vided into its constituent parts until we obtain a suitable
ture based on an iteratively constructed Bézier curve with length for the construction elements. On the one hand,
four control points (Fig. 14). The straight-line sections that the lengths of the elements should not exceed the length
build up the curve will be replaced by untreated timber of the most common sizes of planks existing on the
planks. The vault is composed of a series of arched curves, market. On the other hand, the subdivision should be
placed jointlessly alongside one another. The planks are small enough to obtain a smooth rendering of the curve.
then screwed together in order to create a massive timber The relevant dimensions needed for the pro duc-
vault structure. The shape of the vault’s section can be tion of the construction elements are directly produced
controlled via the control points of the Bézier curve. by the geometric figure. The lengths of the planks cor-
Fig. 15 shows a form study where the curve’s control points respond to the lengths of the curve’s line sections. The
have been deformed such that the resulting shape is a chamfer angle can also be deduced from the geometric
meandering element with inflection points. The line model (bisector angle of two adjacent line segments). The
segments of the underlying discrete Bézier curve have design is therefore limited to two steps:
been replaced by construction elements. – Shape control, via the control points
– Subdivision control by choosing the adequate
level of iteration.
Fig. 16 shows a reduced scale model of such a massive

timber vault structure. The different planks have been cut
by a numerically-controlled computerized routing table.
The question of how to subdivide a free-form
object into a coherent set of construction elements be-
comes obsolete as it is directly given by the iterative
geometric construction method. A direct link from design
to production has been established, which is an impor-
tant cost and time factor for the production of free-form
architecture.
Fig. 16

4.2 Shell structure—feasibility test
In this section we will discuss the application of
an iteratively constructed free-form surface as a panel
construction. The surface method used for the design of
the free-form objects has been described in section 3.4
The design used for the realization of the larger proto-
type (Fig. 21) was mainly driven by following para-
meters: On the one hand, the aim was to create a small,
dome-like structure, presenting a smooth arc in its longi-
tudinal section; on the other hand, we designed a rough
curve, providing folds to the transverse section of the
structure.
In the present example, the faces that compose
the surface are replaced by planar timber panels. The
choice of the thickness of the timber panel is important,
as the virtual 3D surface does not consider any thickness.
A volumetric model has to be derived from the surface
model. The extrusion process is illustrated by fig. 18. Fig. 17
Firstly we generate a parallel offset surface that main-

tains a constant distance from the initial surface. That
distance corresponds to the thickness of the timber panel. 4.3 Integrated manufacturing
Secondly, the bisector planes are calculated, which will In order to test the established digital production
be used at a later stage for the chamfer cut of the chain, an extract of an iteratively designed free-form
panels. In this manner, we design free-form objects that surface was first produced by a 5-axis CNC-machine. The
are entirely built up of planar constructional elements. procedure to advance from the geometry data of the con-
Fig. 19 shows an example of an IFS-surface with faces that struction elements to the machine code has mainly been
have been entirely replaced by extruded construction automated. To create the elements of such complex shapes,
elements. the following work steps are necessary:
Fig. 17 Integrated manufacturing of

the constructional elements
Fig. 18 Parallel offset mesh generation
Fig. 19 Extruded IFS-surface
Fig. 18
Fig. 19
65
A unique address for each construction element is logisti-
cally necessary, in order for the different elements to
be assembled in the right place. Each element has to be
oriented according to the coordinate system of the CNC-
machine, the dimensions of the raw material, and the fiber
direction of the plywood panel.
Automatic generation of the machine code for
each element—the material properties, the type of ma-
chine, and the nature of the cutting tools—is of vital im-
portance for the integrated production of the elements,
which all have different sizes and shapes.
Fig. 17 shows a sequence of the machining process. The
production of each multi-part plate has been divided into
three working steps:
– Piercing of the fixation holes. Each part is

screwed on the machining table.
– Engraving of the addresses of each element.
– Contour cut: machining of the actual element.
The assembled manufactured elements give an accurate

rendering of the surface designed on the computer screen.
This shows that practical realization of iteratively con-
structed surfaces becomes possible.
After the realization of an early partial proto-
type (Fig. 20), we continued testing the method on a more
Fig. 20 a
complex structure comprising 256 construction elements.
The shell structure shown in fig. 21 presents a small vault
spanning over 4.5 meters, built from 10-mm-thick spruce
plywood panels. Although the realized objects remain
relatively small in scale thus far, they allowed us to veri-
fy the validity of the proposed design method, since the
employed manufacturing techniques are also applicable
for 1:1 scale construction elements.
5 Discussion and outlook
The applications presented here show that the design

and construction of free-form surfaces using our method
requires a relatively small planning effort. Several prob-
lems appeared during the manufacturing process due
to the extremely low tolerances permitted by the perfectly
fitting pieces. Large-scale free-form buildings will proba-
bly have higher tolerances, but the logistics and the as-
sembly will probably become more complicated. The effi-
ciency of the method presented is only proved insofar as
the processing of the data, design to production, takes
only a few moments.
We have shown that the design method which al-
lows the generation of rough and smooth objects could
be employed for the design of bearing plate structures
using the rigidity of the fold to improve the structural
Fig. 20 b
strength of the free-form surfaces.
Fig. 22 illustrates a preliminary investigation
Fig. 20 a and b Assembling of the partial prototype into the bearing potential of such folded structures.
Fig. 21 a and b Prototype of the generated shell structure On its left side, a smooth symmetrical plate structure is

Fig. 21 a
Fig. 21 b
67
Acknowledgements
shown. The same model is shown on the right side
with additional folds. In order to get a rough idea of the This research has been supported by the Swiss National
Fund (200021–112103) and (200020– 120037/1). We would
global structural behavior, we imagined a plate struc-
also like to thank our project partner, Dr. Eric Tosan from
ture spanning over 12 meters, built of 20-mm plywood the LIRIS, Université Lyon I, France, for his support on the
panels (spruce). First, we designed a smooth, dome-like development of the mathematical model. Our special thanks
shell structure, where we applied an asymmetrical load goes to Johannes Natterer for the preliminary investigation
of Fz = 1500 N/m 2—a typical design value for snow-load on the load-bearing capacity of IFS surfaces.
in central Switzerland. The maximum deflection occurred
along the z-axis, which was at about 266 mm. After add- References
ing folds, the FEM-analysis showed a maximum deflec- 1Cantor, G. “De la puissance des ensembles parfaits de
tion value of 15 mm. points.” Acta Mathernatica 1884, p. 381–392.
Note: This analysis was assuming hinge joints 2 von Koch, H. “Une courbe continué sans tangente, obtenue
between the individual construction elements. In reality, par une construction géometrique élémentaire.” Arkiv för
the joints may present a certain bending rigidity. Since Matematik I, 1904, p. 681–704.
the construction elements are fixed along all four sides 3 Hutchinson, J. E. “Fractals and self-similarity.” Indiana
(except the border panels), any possible rotation is greatly University Mathematics Journal 30, 1981, p. 713–747.
limited by the system. For real-scale applications, we 4 Barnsley, M. F. Fractals Everywhere, Academic Press. 1988.
believe that adequate detailing will be imperative. Fur- 5 De Casteljau, P. “Courbes à poles.” INPI, 1959.
ther information about the joints of folded timber plate 6 Minkowski, H. Geometrie der Zahlen. Leipzig, Teubner, 1896.
structures can be found in Buri9 and Haasis10. Buri et al.9 7 Schlaich, J., H. Schober. “Filigrane Kuppeln. Beispiele,
used 2-mm-thick folded steel plates for joining massive Tendenzen und Entwicklungen.” TEC21, 2002, vol. 128, no. 12,
40-mm-thick timber panels. Haasis et al.10 studied the p. 21–27.
8 Glymph, J., D. Shelden, C. Ceccato, J. Mussel and H. Schober.
bending rigidity of screwed connections.
“A parametric strategy for freeform glass structures using
The above-mentioned considerations about de-
quadrilateral planar facets.” Automation in Construction, 2004.
tailing and structural properties of the presented work
9 Buri, H., Y. Weinand. “Gefaltet. Holztragwerke.” TEC21,
are possible subjects for future research. 2009, no. 8, p.18–22.
In the future, we will continue the development of 10 Haasis, M., Y. Weinand. “Origami -folded plate structures.”
larger and more complex objects. The potential of the new Engineering. 10th World Conference on Timber Engineering. 2008,
design method for free-form surfaces is far from being ex- Miyazaki, Japan.
hausted. We hope to have further opportunities to test our
method on applications such as suspended ceilings, free- Reprinted from Journal of the International Association for Shell
form facades, climbing walls, halls, and shell structures. and Spatial Structures, vol. 50, num. 1, 2009, pp. 11–20.
Comparison
Fig. 22
of the a smooth and
rough IFS-surface
under asymmetrical
d Zmax = 220 mm d Zmax = 45 mm load

69
2. 3
Geometrical description and

structural analysis of a modular timber structure
Sina Nabaei, Yves Weinand, and Olivier Baverel
The
Das ambitious
ehrgeizigegoal Zielofdertheaktuellen
ongoing research
Forschungsarbei-
at IBOIS, 1 Introduction
the
ten laboratory
am IBOIS ist of timber
es, neuartige
constructions
großmaßstäbliche
at the École
Polytechnique
KonstruktionenFédérale
aus innovativen
de Lausanne Holzerzeugnissen
(EPFL), is to 1.1 IBOIS, the re-interpretation of timber
develop
unter Verwendung
the next generation
textiler Fügungsprinzipien
of timber construc- zu construction
entwickeln.
tions made from Vorgestellt
innovative
wird timber-derived
hier eine modulare prod- In recent years, the necessity of using renewable
Tragstruktur
ucts, by the aus application
gefalteten of Holzplatten,
textile principles
wobeionins-a and sustainable resources in the building sector has be-
building
besondere scale.
die geometrischen
The structure presentedMethodenisfür a modular
die Her- come obvious, and interest in timber as a building mate-
composition
stellung modularer
of timber Muster
foldedundpanels,
Gitter
notably
für Holzbau-
demon- rial has revived.1, 2, 3 Novel timber-derived products, such
strating
werke beispielhaft
an exampleerläutert
of the application
werden. Esof wird
geometric
gezeigt, as massive block panels, have emerged and the use of
dass
techniques
komplexe usedräumliche
to produce Tragwerke
modular mitpatterns
einfacher
and such products is increasing. 4, 5 The ambitious aim of the
Verbindungstechnologie
lattices to timber structures. zwischen
Effectively,
den Elementen
it is shown ongoing research at IBOIS, the laboratory of timber con-
geplant
that complex
werdenspatial
können. structures
Darüber hinaus
can belassen
designed
sich structions at the École Polytechnique Fédérale de Laus-
dank
using moderner
simple connection
CAM-Verfahren technologykomplexe
between Holzele-
ele- anne is to develop the next generation of timber construc-
ments.
mente in Moreover,
großem Maßstab
by takingzuschneiden
advantage of undadvanced
präzise tions made from innovative timber-derived products by
zusammenfügen,
CAM process, complex so auch planar
der timber
Prototyp,elements
der in die-
are the application of textile principles on a building scale. 6, 7
cut
seminBeitrag
large scale
vorgestellt
and assembled
wird. Daswith Falthigh
prinzipprecision,
beruht The unprecedented exploration and study of timber-
as
aufis
einem
the case
ebenen forStabflechtwerk.
the prototype Die of the
Grundmodule
structure related structures and their structural analysis is sought
presented
bestehen aus in this
zweipaper.
sich gegen
The folding
seitig stützenden
concept corre- ge- within a framework integrating the mechanical and struc-
sponds
falteten to
Holzplatten,
a planar reciprocal
die ineinandergesteckt
frame structure. werden
The tural principles of textiles. Since timber can be viewed
und
basicschließlich
module consists
einen Bogenof two bilden.
mutually Diesupporting
Stabilität as a fiber-derived product, it follows that the analogy be-
zwischen
timber folded
den Modulen
panels that entsteht
are joined
durchtogether
den Kontaktcon- tween micro-scale fiber structures and timber-derived
secutively
entlang der along
Steckverbindungen.
their slots to build Dieupwesentlichen
an arch. The wooden structures can be explored at the micro and macro
stability
mechanischen
of the Eigenschaften
inter-module connection
der Konstruktion
is provided
wer- scale. The key to our approach is the underlying notion
den
by contact
mittels boundary
der Finite-Elemente-Methode
condition over the sliding unter- that timber’s fibrous nature, historically perceived as a
sucht,
joints. wobei
The fundamental
die nichtlinearenmechanical
Grenzflächenbedin-
properties of liability for a construction material, is in fact a positive
gungen
the structure
an der areKontaktfläche
examined using einbezogen
the Finite Element
werden. feature that should be exploited to increase the material’s
Untersucht
Method considering
werden the auch nonlinear
das statische
contact Verhalten
boundary functional and aesthetic value. Its inherent flexibility al-
condition.
für den Lastfall
The static
Eigenlast
behavior
und isdasstudied
dynamische
under Ver-
the lows it to be folded into robust, lightweight structures
self-weight
halten des loadModells.
case,Entsprechend
as well as the model
den Analyseer-
dynamic that use material very sparingly. The concise observation
gebnissen
response. undAccording
mithilfe todes
analysis
CAD-Parametermodells
results, and with of existing textile techniques and fabrication methods,
the
werden
aid of
strukturelle
a CAD parametric
und geometrische
model, structural
Alternativenand notably here the geometric techniques used to produce
geometrical
vorgeschlagen, alternatives
um die Tragfähigkeit
are proposed zu to
verbessern.
improve modular patterns (as described by Clare Horne 8), com-
Auf
the Basis
structural
dieses performance.
geometrischen A prototype
Prinzips wurde
based einon bined with the investigation of the modular structure
Prototyp
this geometric
hergestellt
principal
und hasaufgebaut,
been fabricated
um die Reali-and presented here, is intended to result in the development
sierbarkeit
assembled im in order
baulichento explore
Maßstab thezufeasibility
untersuchen. of the of a new family of timber constructions based on the logic
concept at a building scale. and principles of textile fabrics.
Räumliches Tragwerk aus Holz, 1.2 Geometry of spatial structures:

Keywords timber Stabflechtwerk,
spatial structure,strukturelle
reciprocal frame a survey of “form-element” relationships
Verbesserung,
structure, Finite-Elemente-
structural system improvement, Over the past decade, four principal directions
Analyse, analysis,
finite elements parametrisches Entwerfen
parametric design dealing with the geometry of spatial structures have
been recognized and researched. Fig. 1 shows an example

Element Assembly Materiality Equilibrium
condition
(connection) Boundary condition
and mesh
Materiality
Engagement
Nexor length
Tension Compression
Fig. 1 b
Forms for spatial structures:

Fig. 1
Form
from a related publication on the subject. The various ap- a) Reciprocal frame structures19
b) Equilibrium forms either in
proaches are described and analyzed below. tension or compression20
Fig. 2 Folding concept for the
– Reciprocal frame structures: This family of modular structure designed in
modular spatial structures consists of interlaced the Atelier Weinand workshop
linear elements, where the final form is a Fig. 1 a
result of a basic module as well as a connection
technology (Fig. 1a). 9, 10, 11, 12
– Form-finding practice: This deals principally

with tensile membranes or vaulted masonry
construction in order to determine a tensile
or compressive geometry. Here, the final form is
the direct result of equilibrium and is greatly
influenced by the materiality and the boundary
condition applied (Fig. 1b).13, 14, 15
– Topology optimization: The last stream can be

Fig. 2
categorized under topology and shape optimiza-
tion problems. These methods, combined with
evolutionary algorithms, often result in free-
form spatial objects, which have been generated In fact, when it comes to free-form design, it generally
on the basis of structural criteria. These fall into starts with a “complex form” that ends up with multiple
the first category, where they are geometrically typologies of elements in a top-down process and “com-
approximated in order to be built. plex connection technologies” where structural elements
follow offsets of the external form. Indeed, we ask the
– Free-form design: Here the free form design question whether the same free-form practice should be
surface is the input and by means of geometric- followed in the context of timber construction. Thus it is
mathematical models, the surface is approximat- necessary to take the materiality and the complexity of
ed by use of a uniform (or typologically uniform) the connection technology as an important feature here,
planar mesh. The mesh generation is clearly in order to make a distinction between the four ap-
constrained by tolerances of the approximation proaches described above. The core idea that will be ex-
problem as well as the economical sensibility plored is an investigation of families of spatial structures,
analysis. The supporting structure closely follows where the focus is on materiality and related connection
this approximated mesh and thus its spatial form technology, rather than on an irregular surface approxi-
is less influenced by the materiality and robust mation. According to this approach, the final form is a
mechanical reflections on the equilibrium and result of the geometry of connected members as well
the mechanical governing laws. The materiality as the employed connection technology. In the context of
and the connection technology clearly come timber engineering, we focus on use of CAM facilities
alongside the free-form design surface.16, 17, 18 to cut complex geometries from thin timber panels and
slender beams, while the connections are kept simple, as
is appropriate when using timber.
71
S2(P1)
P1 = R1(P 0) P1
P1
P0 Q1
γ π – 2γ
0
X
Q1 = R1(Q 0) Q1
Q0 S2
S1(Q1)
S1 π – 2γ
0
1
Q
Q
P0
P1
P1
P1
0
W
R1
R1
B0 B1 Mcx = B1j S1 (B1) Mcv = B1j S2(B1)
S2(P1)
Q1
P1
Q1
P1
S1(Q1)
T1
1
1
Q
P1
P1
P1
P1
R2
ψ
1)
(Q
S1
T1(Mcv) T1(Mcv)jMcx T1(Mcv)j R 2(Mcx)

P1
b
2
P1
P1
P1
b
2 T
2 I1
T
3
T 3 • T1(Mcv)j R 2(Mcx) M‘cv jM‘cx
Fig. 3 Geometric transformation of basic trapezoidal panel (B 0 ) into two slotted V-form modules (M cv jM cx )

The modular structure proposed in this paper is derived
from the exploration of form-element relationships. It is
not only treated geometrically, but also from a mechani-
cal point of view it is demonstrated that its structural
behavior is understood. Moreover, a prototype is realized
to complete the investigation.
2 Presentation of structure
2.1 The folding concept

The folding concept presented in this paper was
initially examined during an architectural workshop Fig. 4 Prototype realization at EPFL
titled the “Atelier Weinand at IBOIS EPFL,” focusing on the
discrete architectural geometry under the supervision of
Yyes Weinand. A V-form basic module is fabricated con-
necting two timber panels and is then spatially multiplied By repeating the same geometric transformation, a dou-
using consecutive spatial rotations to form an arch (Fig. 2). ble cut module is obtained. This “base module,” as shown
The structure can be divided into four principal in fig. 4, is the base unit of the modular structure.
typologies, each consisting of two mirror-image timber As a result of the implementation of the geo-
panels joined together as a V-form module through the metric transformation described above in a CAD setting,
bisector plane by means of two hardwood dowels. They a parametric model of the modular structure is obtained
are placed in the middle as shear keys, and two oval head where the geometry is controlled by means of a set of
screws are inserted close to the borders in order to avoid meaningful scalars.
relative translation and rotation. Fasteners are inserted
in the direction of the normal to the plane of reflection. 2.3 Prototype realization
These modules are then slid consecutively along their A prototype of this structure has been realized
U-shaped incisions to form an arch. The inter-panel sta- at EPFL in order to test the structural feasibility of the con-
bility is provided by roto-rigidity of the slide connection cept as well as to investigate the architectural quality.
and axial contact of the reciprocal panels. Two photos illustrating the project are shown in fig. 4. The
project included development of relative NC codes for
2.2 Geometric and parametric decomposition 5-axis machining, as well as designing constructive de-
of global forms tails and fabrication of the structure for exhibition. All
A trapezoidal plate is introduced as the generat- V-form folded modules have been manufactured from
ing element of the form, denoted as B1 and referred 21-mm-thick, three-layer, cross-laminated panels and cut
to as the “base panel.” In general, this single panel is by means of CNC machines on the EPFL campus.
transformed by means of two classes of operators in or-
der to give shape to the global form. The first operations
are the set of Boolean operators representing the manu- 3 Structural analysis
facturing, connecting, and assembling processes. The
second type of operators, referred to as Geometrical oper- Here the objective is to understand the structural system
ators, introduce rigid structural movements and consist considering the geometric non-linearity by the aid of
of congruent maps employed to place the object in the appropriate numerical models and to improve it. The di-
space. Among Boolean operations, union, intersection, agnostics about the structural system consists of the line-
and remove are used. The geometrical operators we may ar static analysis under the self-weight and the modal
utilize include rotation around a space vector, reflection dynamic analysis, in order to examine the rigidity. Based
against a plane, inversion against a spatial point, or trans- on these observations, improvements are proposed in
lation in the direction of a space vector as examples of section 4.
simple isomorphisms.
The geometric transformation permitted to de- 3.1 Modeling hypothesis and local boundary
termine the slot cuts between two consecutive modules condition analysis
is illustrated in fig. 3. The manuscript letters (e.g. “B,” According to the adapted modeling approach, the
“M”) stand for geometric objects, while bold capital letters thickness of panels becomes relevant in the reality of
(e.g. S, R, T, etc.) symbolize isomorphisms, respectively modeling, enabling the modeling of the slide connection
the symmetry, rotation, and translations. Boolean opera- as it is defined by the geometric configuration. For each
tions are represented by their mathematical symbols of slide connection, five normal contacts are defined between
union, etc. two pairs of surfaces as master and slave surfaces. Contact
73
property is considered to be frictionless. The contact angle of U-joints. This would help to reduce the
boundary condition has to be satisfied along slide joints concentrated stresses (Fig. 6a). Applying these two
underlying finite deformations. A 10-node modified quad- main modifications to the original structure, the
ratic tetrahedron element is chosen to mesh the continu- maximum von Mises stress under the self-weight
um model, referred to as C310DM ABAQUS® solid element. load case is reduced from 14.1 Mpa for the origi-
The timber is considered to be an elastic, homo- nal configuration, to 1.66 Mpa, keeping the same
geneous material throughout the entire thickness. The order of magnitude of maximum total deforma-
Young Modulus, E = 8000 Mpa, Poisson’s ratio, υ = 0.3, and tion. Furthermore, the stress leads to intermedi-
material density, ρ = 500 kg/m3 are calculated from docu- ate elements rather than panels and consequent-
mentation disposed by the industrial provider of cross- ly the concentrated pattern of stress on panels
laminated panels.21 has been resolved. The natural frequency for the
first global mode of the structure is increased to
3.2 Structural analysis of a single arch 0.98 Hz.
Results for the global deformation field and the
von Mises stress driven from a static nonlinear analysis 4.2 Increasing panel interlocking effect
of a single arch under its self-weight load case are shown Consider a CX module, chosen deliberately from
in fig. 6. It can be seen that the geometric configuration the arch. The region of this module is indicated in equa-
of the slotted-together modules leads to a concentrated tion 1 and fig. 7a. According to the geometric principle
distribution of stress at the location of slide cuts. This shown in fig. 4, CX1 is connected to CV1 and CV2 across two
happens primarily due to the bending behavior of the U-joints. The concept is to increase the length of these
structure. current U-joints to make CX1 meet CV−1. (Fig. 7b) The entire
Moreover, a modal dynamic analysis for the iso- two-intersection cubes are then removed from CV−1 re-
lated modular arch has been constructed to have an ini- sulting in two extra U-joints on the external part of
tial estimation of structural rigidity in lateral and trans- the panels. Consequently, CV1, currently connected with
verse loading conditions by comparing natural frequency CX0 and CX1, intersects with CX −1. Removing the two new
values. The first global mode is lateral and has a relative- intersection cubes from CV−1 provides two extra U-joint
ly small natural frequency of 0.59 Hz, compared to practi- connections, in this instance situated in the internal part
cal guidelines, which advise a minimum natural frequen- of the CV module.
cy ranging between 1 and 4 Hz. 22 If we resume, the general idea is to keep the
cut-pattern for the CX module unchanged, although for
the CV module there would be four extra U-form cuts: two
4 Propositions for structural system internal and two external (Fig. 7c).
improvement To achieve this objective, the original geomet-
ric concept has been implemented within a parametric,
Based on our findings in 3.2, in this section we proceed computer-aided design interface. The important para-
with two goals: first, to obtain a more uniform stress dis- meters that determine the geometry of each typology
tribution in panels and to reduce the stress concentration of modules in the original design have been identified.
in U-joints, and second, to increase structural rigidity, Next, the geometric configuration for the montage of the
which is measured by means of the natural frequency of Base modules is set with respect to the height and total
the first global mode. span of the original structure.
4.1 Toward a truss system CV−2 – CX −1 – CV−1 – CX 0 – CV1 – CX 1 – CV2 eq. 1
While already having a geometric superposition

concept, one immediate observation would be to change Increasing the notch length, while keeping the total span
the current beam-like system to a more truss-like system. and height of the structure constant, will increase the
Two main directions are tracked, as follows: number of modules. The original design consisted of
– Addition of intermediate elements: this could be 33 slid modules. In fact, by increasing the length of joint
realized with the aid of additional intermediate by 103 mm to create two more U-joints between panels, 53
elements which are carefully inserted at the modules of nearly the same size are needed to achieve the
mid-plan of the arch to connect consecutive same height and span as the original design. Therefore, it
CX-CX and CV-CV modules to each other. follows that the interlocked version will be 53/33~1.7
– Opening the U-joint: in the initial geometric times heavier than the original one. By increasing the
configuration, each module’s stability is provided number of slide joints and distributing their position
by the locking effect of panels across the U-joints. across the entire length of the panel, we expect a more
In a truss system with additional intermediate uniform load transfer between modules. Indeed, the re-
panels, fixed between them and fixed to the sults for von Mises stress from an elastic, nonlinear anal-
slid modules, it would be possible to increase the ysis, confirm this idea. The maximum von Mises stress for

magnitude u (mm) S Mises (Pa), Av. 75 %
+4.123e–03 +1.412e+07
+3.435e–03 +1.177e+07
+3.092e–03 +1.059e+07
+2.405e–03 +8.237e+06
+1.718e–03 +5.884e+06
+1.031e–03 +3.531e+06
+6.871e–04 +1.178e+06
+0.000e+00 +1.268e+03
Fig. 5 Global deformation field and von Mises stress in a single, isolated arc model
CV2 CV2
CV1
CV1 CX1
CV–1 CX1
CX0 CV–2 CX0

CV–1
CX–1
CX–1 (IV)
opened U-joint
CV–2
a)
(IV) (IV)
CV2 (III)
(III)
CV1 CV2
CX1
CV1
CX0
CX1
CV–1 CV–1 CX0

b)
CX–1
CV–2 CX–1
CV–2
a) b)
(III): Internal supplementary U-joint typology

(IV): External supplementary U-joint typology
Original CV Interlocked CV
(III)
(IV)
(II) (II)
c) (I)
(I)
c)
Fig. 6 Toward a truss behavior a) CX module modification and intermediate Increasing panel interlocking effect: two geometric configura-
Fig. 7
panels b) CV module modification and intermediate panels c) Isolated arch tions a) Original b) Interlocked c) Comparing geometric modification
reinforced with intermediate panels on upper and lower areas brought to CV module as well as its connection typology
75
self-weight load case, reduces to 1.15 Mpa with a maxi- Acknowledgements
mum total deformation of 1.3 mm, which is still accept- The authors would like to express their gratitude to Olivier
able. This is true even though the interlocked configura- Baverel from Navier laboratory of ENPC for his important
tion is 1.7 times heavier than the original one. The main contribution to this study.
gain is in the structural rigidity, where the minimum
natural frequency, calculated from a modal dynamic
analysis, is estimated to be 5.99 Hz. Using the values of References
natural frequency (f ) and total mass (m) for the original 1Thun, M., DETAIL Zeitschrift für Architektur + Baudetail, 2010.
configuration (marked with subscript 0 in equation 2) 50(10): 982–988.
and the improved interlocked version (marked with sub- 2 Weinand, Y. “Innovative timber constructions.” Journal of
script 2), one may compare the relative equivalent struc- the International Association for Shell and Spatial Structures,
tural stiffness (k) between these configurations as repre- 2009. 50(161): 111–120.
sented in (Fig. 3), concluding that the new slide locks stiffen 3 Herzog, T., J. Natterer and M. Volz. Timber Construction
Manual. DETAIL ed. 2000: Birkhäuser Architecture.
the original structure by 165 times.
4 Dunky, M. and P. Niemz. Holzwerkstoffe und Leime:
k 2 m 2 T f2 Y
2
Technologie und Einflussfaktoren. 2002: Springer-Verlag Berlin
k0 = m , 165
f0
eq. 2
0 Heidelberg.
5 Buri, H. and Y. Weinand. “Übersicht Massivholzplatten.”
39. Fortbildungskurs Schweizerische Arbeitsgemeinschaft für
Holzforschung. 2007, SAH Schweizerische Arbeitsgemein-
5 Conclusion schaft für Holzforschung: Weinfelden. 63–84.
6 See note 4 above.
A modular structural concept consisting of folded planar 7Weinand, Y. and M. Hudert. ”Timberfabric: Applying textile
timber elements is presented in this article. It is shown principles on a building scale.” Architectural Design, 2010.
that the form of the spatial structure is determined by 80(4): 102–107.
the geometry of its base module and the connection ge- 8 Horne, C.E. Geometric symmetry in patterns and tilings. 2000:
Woodhead Publishing.
ometry between panels. It represents a novel family of
9 Baverel, O., H. Nooshin, and Y. Kuroiwa. “Configuration
spatial structures that can be interpreted as planar recip-
processing of nexorades using genetic algorithms.” Journal of
rocal frames. The parametric implementation of the con- the International Association for Shell and Spatial Structures,
cept is analyzed and it is established that this parametric 2004. 45(145): 99–108.
model can be used to propose structural improvements, 10Popovic Larsen, O., Reciprocal Frame Architecture. 2008:
notably here by increasing the interlocking effect of con- Architectural Press.
nections. Further research will look at a generalization of 11Douthe, C. and O. Baverel. “Design of nexorades or recipro-
this work based on mathematical exploration of modular cal frame systems with the dynamic relaxation method.”
structural shapes suitable for timber construction. Computers and Structures, 2009. 87(21-22): 1296–1307.
12Baverel, O. and H. Nooshin. “Nexorades based on regular
polyhedra.” Nexus Network Journal, 2007. 9(2): 281–298.
13Bletzinger, K.-U. et al. “Optimal shapes of mechanically
motivated surfaces.” Computer Methods in Applied Mechanics
and Engineering, 2010. 199(5-8): 324–333.
14 Block, P. and J. Ochsendorf. “Thrust network analysis:
A new methodology for three-dimensional equilibrium.”
Journal of the International Association for Shell and Spatial
Structures, 2007. 48(155): 167–173.
15 Tibert, et al. “Review of Form-Finding Methods for
Tensegrity Structures.” International Journal of Space Struc-
tures, 2003. 18(4): 209–223.
16Eigensatz, M., et al. “Paneling architectural freeform
surfaces.” ACM Transactions on Graphics, 2010. 29(4).
17Pottmann, H., et al. “Freeform surfaces from single curved
panels.” ACM Transactions on Graphics, 2008. 27(3).
18Glymph, J., et al. “A parametric strategy for free-form
glass structures using quadrilateral planar facets.”
Automation in Construction, 2004. 13(2): 187–202.
22 Bachmann, H. Vibration problems in structures: practical
guidelines. 1995: Birkhäuser.

77
2. 4
“Discover processes where the parameters

are insufficient”
Interview between Mark Pauly (Computer Graphics and Geometry Laboratory, LLG, EPFL, Lausanne)
and Yves Weinand, February 10, 2015
Yves Weinand: Something I’d like to discuss with parameters. I think these types of problems are inherent-
you is the prototype presented in this chapter, ly changing, because they involve justifications. The
Modular Timber Structures, which has never been problem is well understood when you have continuous
published before. There is also Dr. Ivo Stotz’s space and continuous energy and you want to optimize
thesis about fractals and iterative details, them. But when you have connections and you have
developed by architects and computer scientists many potential ways of connecting them together, I don’t
at IBOIS. We created this pavilion with civil think there is a general solution, where you can simply
engineering students, where we simply changed play with the system to find an optimal solution. The
geometric parameters. That’s what we’re doing all individual piece that you have may be locally optimized,
the time: we try, fail, and retry. It’s an iterative the local shape is interacting with the global geometry
manner of working with civil engineers. This is and that is absolutely fundamental to look at. At this
something we could try to highlight in our point I don’t see a solution for this. We’re still working on
discussion. The first question is: What did we do very specialized solutions: the justifications can’t be
here? We developed this structure as architects, determined in advance, nor can they be fixed with regard
and then we controlled the contact zones by to the shape, the geometry, the size of the elements, the
mechanical means, as we have in the models. manner of changing the connections, or the interaction
Then we improved the geometry by slicing pieces between the elements. I don’t think there is a system that
deeper in relation to one another to add addition- allows you to dynamically modify these things and
al contact points, thereby creating another obtain any kind of meaningful results, because you have
form of global geometry in order to stiffen the a kind of fragility in the sense that if you change things
structure. I can do this when I work with students, very slightly in terms of the connectivity, it’ll have a
but they are not able to reach this point very huge impact on the structural performance.
often. Sina Nabaei, who was working on this
project at the beginning, was aware of this aspect, Y. W.: We probably didn’t really manage to model
because he understood the geometry. He made it the contact zones. It was more a global geometry
very clear at the outset and we haven’t been able change. We added more contact points because
to advance further yet. I would like to know: How we modified the global geometry and this
would you approach this in a different way? This already helped structurally. As a computer
is just one solution; there is no general framework scientist, would you take this as a case study? I
about this geometry and it’s more from the point would actually like to take this project further.
of view of a civil engineer, who was trying to It’s there and nothing has been done, but I think
stiffen the structure a bit further. There’s already this kind of structure—just plates—could be
something here; it’s a case study. Do you see it very interesting in architectural applications,
as defining a general framework to describe this for example in lighting. As directors, how could
type of geometry and to optimize it? we define something like a research plan that
would also be suitable for one of your doctoral
Mark Pauly: First, it is an iterative process—we use our candidates in computer science? Would you
intuition and we try. I think that’s something that will encourage this type of topic, or how would you
remain and that’s probably a good thing, because when define the approach? I would like to take a civil
you design this kind of thing, you’re not only looking engineer and show him this project. He would
at solving a problem, you’re also looking at defining a probably repeat the research until he gained
problem. You often don’t know exactly what you want, rigidity. I think we’re a bit lost, as we don’t have
and this is makes it harder. This is the process of a general framework, as you said, we have not
searching for the right constraints and the right really addressed topological questions yet.

M. P.: The first thing I see—and this is perhaps my to solve one specific problem, you want to keep the
perception as a scientist—is the possibility of abstracting general problems that are more useful for the applica-
this structure into general performance. Basically, tion. I would use a dual approach: you look in very
what I see here is a framework that is constrained. general terms at elements that are connected in a very
You have many small performances that interact with abstract language that could be easily mathematically
each other and these are local interactions, but they represented and, at the same time, come from a bottom-
create a global dependency. Let’s say you have hundreds up approach and use wooden plates or shapes that are
of plates or hundreds of elements, and then you can connected in an abstract way, and bring these two
encode the interrelations towards a topology of the approaches together. You can learn both from the
system. You can define their local interactions, even abstract point of view and the general point of view, but
experimentally, whatever contact forces you have. also from the interaction system, from the concrete
Your aim would be to have a system that allows you to properties of your design, because you won’t necessarily
interactively change the shape of individual elements, be able to think of relevant factors up front. You may
the shape of the overall structure, and all the connectivi- understand once you start to play with these things,
ty, plus get feedback on structural performance and when you break down some of them, or with material
assembly constraints. tests and so on. Then you will realize that there are
I would try to have a general description of this kind of parameters that you haven’t thought about and that you
element; this means a specific way of joining two pieces need to include in your system. To me, this is an
together that have certain properties, geometrically approach I would like to take. It’s often hard to impose
and structurally. However, you could think of many this on students. My experience is that students either
other ways of connecting pieces together. My first have a bottom-up kind of mind-set or a top-down kind of
approach would be to render this in more abstract terms. mind-set, and they like to progress in a linear fashion
On the one hand, this type of construction has a very and find it difficult to look at things in different ways.
general structure beneath; on the other hand, it’s also I think as a tutor you can give them suggestions.
important to make progress to study a specific instance.
You don’t want to investigate all possible ways you could Y. W.: When you speak about parameters, are
connect things and try to solve all these problems they geometric in nature or are they other sorts
together. I think at this stage, as we don’t have a tested of parameters?
solution for these topological assemblies, it would be
good to start with a concrete example and maybe limit M. P.: Of course, many of them are geometric, such as
the variations. By maybe resizing the plates, you can the size, the thickness of these plates, and the angles.
redesign and change them, and so on. You can parame- But you can also put in material properties: the strength
terize the connections in certain ways, so that you have of certain materials, the friction parameters between
a fixed space in which you could operate. The parameters materials, and so forth.
are fixed, but the numbers and types of parameters are
not fixed. This is a design task. At least start with a fixed Y. W.: Do these parameters equal constraints in
set of variables and build a system that could work with this case? When you speak about mechanical
them and then discover processes where the parameters constraints, for example?
you decided to work with are insufficient. You need to
build such a system, a sort of bottom-up approach,
starting with something simple and adding complexity
as you go. But at the same time, keeping the opposite
perspective in mind, the sort of abstract representation
that you could use for this, because you don’t want
79
M. P.: By parameters, I mean aspects you can vary; constrained and you’ll have no more freedom to design,
for instance, the thickness of the plates is a variable. or it will be under-constrained and every solution will
In my case, the geometric constraints would be that the be possible, but then you won’t know which one is good
angle between two plates cannot exceed 45° because of anymore. You need additional thickness criteria in order
the limits of the fabric technology. These are fixed to make a choice. If your constraints don’t restrict the
parameters, for example, in the case of simple con- design space, then you don’t gain anything through an
straints. With more complicated constraints, if you want optimization process, because it doesn’t tell you any-
to assemble these things, then there is only a certain thing. When you design something that looks nice, but
path whereby you can move the pieces and perhaps there doesn’t work mechanically, how would you know what
is a global dependency. In terms of materials, you may you have to change in order to make it work? This is
say the maximum stress that the piece can support would exactly the challenge here, because if you have a multi-
be a mechanical constraint. To me, these are the parame- dimensional space you can have thousands of degrees
ters that define a space. This is a trivial example (he of freedom. Many of the designs will be poor. These are
draws something): You have a two-dimensional space; totally different things. How could you build a system
let’s say you have an x variable and a y variable. Then that lets you understand the different ways to achieve a
you specify constraints as a function, so things are meaningful design? Should you make the plates thicker,
particularly free, and then you say something trivial, for should you lower the height, should you change the
example, the sum of the two would be equal to one. And it degrees, or something else? This remains unsolved.
gives you this two-dimensional space. You start with two Because of this iterative process, what I would like to do
degrees of freedom and you have one constraint. Of in the future is to better understand the consequences
course it’s not always the case that the degrees of of introducing new constraints or removing constraints.
freedom are independent or that the constraints are Sometimes you have constraints that are conflicting
independent. and there is no visible solution. How do you understand
It’s a challenging modeling task. First of all, you have to which of these constraints you need to lose and which
define your degree of freedom and your constraints, and to keep? If you have hundreds of constraints, you cannot
then model them in a mathematical way in order to test simply modify each of them individually. It’s very
an optimization. The challenge for design is that neither difficult from an optimization point of view. We have to
the degrees of freedom nor the constraints are fully come up with another solution. It’s challenging and
understood. You start with something that you think is that’s why I think you have to build it up slowly. The
reasonable, you build your solution, you explore things pavilion is a good example to begin with, because it’s not
and you find while designing that maybe you are overly complex. It’s not combining many different things
actually missing a constraint, maybe you have too many together. There is certain regularity to the problem and
constraints, or perhaps you have to introduce new the scale is appropriate for this type of problem, because
degrees of freedom. For example, in this case, you decide I believe you would have to start small.
at the beginning that the thickness of the plate is fixed,
but you then you realize that with this fixed thickness, Y. W.: Of course, I’m interested in defining
you cannot obtain what you want and you have to mechanical constraints and to see when you
introduce a parameter, that is, the thickness of the plate, compare those two systems: This one is far
so that you have more freedom in design. As you better in vertical loading, or this one is not good
mentioned at the beginning, this whole process needs to at all in horizontal loading. When we think in
be iterative. You won’t know the degrees of freedom at terms of optimization, it’s quite hard to see
the beginning; you won’t know the constraints. If these because every load case is already very different
two elements are not modeled correctly, either your structurally. Different types of structures react
model will be over- differently to different types of load cases. That

would be more like a geometrical study, which M. P.: The first question is: What are your design
gives its structural performance. My feeling is objectives? In this case, maybe you want structural
that the geometrical problem is easier to tackle performance. You have some boundaries and, for
or to define than the mechanical one. example, this is the amount of load we want to tolerate.
You can vary certain things, depending on what you
M. P.: Yes, because the topology is fixed. For example, decide to address at the beginning. You have a list of
let’s say you want to optimize something like this for a parameters that you could change. Is this something that
vertical load, you have this prototype, and you see is difficult to specify at the onset if I want to find the
that it performs badly in terms of sustainable load. What optimal angle to get the optimal performance?
are you supposed to do? This is exactly the kind of thing
you would say, regarding your constraints and how Y. W.: But you can’t say it in advance, it’s really
much vertical load you can tolerate. The connections are strange.
not strong enough, so you have to tie them so that they
can support more load. That’s a very complex constraint, M. P.: So you don’t have the physical model to simulate
because it doesn’t necessarily affect every piece. Let’s this?
say the function to evaluate your design and its effect
gives you the vertical load tolerance, the maximum Y. W.: No. I know this parameter is the most
vertical load you can sustain. This is a very complicated important one, but in the end, I think it’s
function, but you can evaluate it, you can run your probably not solved.
design through a simulation and determine the result.
In order to design, you have to work on that function, M. P.: But you did run simulations, right?
you have to figure out how to modify x to increase the
load. This would be an example of a complicated Y. W.: Yes, from them you have the results and
constraint, as opposed to a simple calculation that can be then, for example, you get deflection. Then we
easily evaluated. Figuring out how you would need to can register the structural basis, the stiffer one
change an axis in order to increase that load optimally is for example. But that’s all. Even the interactions
highly complicated. Currently you would probably do it between the parameters are unclear.
intuitively to make things different, but coming up
with the system does something like this automatically. M. P.: What you have now is a system in which, if you
I think it’s very difficult, unless you clearly fix the change parameters, you can run it and it gives you an
parameters. If you only allow for parameters, let’s say answer?
some geometric parameters like the sizing of the plates,
maybe the thickness, then that’s a tool, as you have a Y. W.: Yes.
sort of continuous space. But as soon as you allow for
topological changes, I have no idea how to project this in M. P.: So, you have an oracle basically. What I mean is,
an optimal manner. you choose a point in your parameter space and it gives
you a quantity indicating how good it is. In theory you
Y. W.: In order to be prudent we should begin by could sample all the possible combinations of parameters
trying to define certain parameters. In terms of and then look for the one that is the most compliant. In
coming up with a research plan, do you think theory, though not in practice.
that would be possible?
Y. W.: Yes.
81
M. P.: That’s already significant, because you can Y. W.: When you speak about form-finding and
evaluate this function, but you can’t invert it, you can’t optimization in a mechanical way, the Block
say which parameter is the best. What you really want in research group, for instance, defines an ideal
optimization in mathematics is to find the argument for function for specific load cases. But if you change
maximizing or minimizing that function. Right now, the load case, you would need another form. So
all you could do is to evaluate f. If x has many dimen- we would rather choose a parabola, or something
sions, it becomes computationally attractive. If you only similar, if the main goal is efficiency for taking
have one parameter, you can simply sample twenty vertical loading. The form optimization—having
simulations. But already, you have this function, but it is more load points—would, however, increase the
often not given or you don’t know the function. In such flexion rigidity for horizontal forces. Thus
a case, I think there are different strategies. One has to geometric modifications of a given form can
make the function evaluation efficient enough to show vary depending on whether the modifications
that your project is valuable, so you can do a search; or tend to be able to carry vertical or horizontal
you can try to simplify this function, for example, by forces. The parameters that need to be consid-
looking at geometric properties that you can evaluate ered can be of a different nature regarding
more easily. I think this last approach is an interesting different load cases. Even if you defend that, you
alternative. The question is: Looking at the simulations, will have a global stability and a local approach
can you gain some geometric intuition? Let’s say your (not rigidity). I still don’t see exactly how I could
energy function looks overly complicated (he draws), but have a global approach here in terms of
you can find a simpler function and then you can decide structural engineering. There is another thesis
to go back to your original function. You can return to a at the ETH in Zurich written by Thomas
complex function, but you can start with a simpler Kohlhammer about reciprocal frames. Here the
geometric formulation you can optimize, one that you optimization process consists of taking out
can search for globally. You’ll never be guaranteed to bending moments.
find this pick. You may find some other picks on this
function, which are not as good, but at least if you do it M. P.: I think a good approach would be to abstract
carefully, you can usually prove that you are going to certain physical properties and map them into a
stay within a certain optimal boundary. There are many geometric space. You generalize the model to optimize it,
optimizations, which are exponentials, so you cannot and it’s beautiful, but you have to challenge this first
hope to find the right solution, but you’ll have a certain abstraction step. You then find the optimal surface
approximation that will guarantee that you to stay for a perfect vertical load, and I guess in practice that
within a certain boundary. That means, though, you’re also means it’s a terrible surface for a slightly vertical
guaranteed to be within five percent of the optimum load, but you lose any solutions that might work.
result. You wouldn’t know if you achieved the best This has to be checked.
possible result, but at least you’d know it’s not the worst. I think the only advantage is that it gives you initial
It’s a challenging and intellectually difficult task to points for further investigation. It doesn’t solve the
simplify such a model; simplifying a model is not whole problem, but the whole problem has to be attrac-
something you can easily automate. tive enough for certain cases. It’s interesting in itself
to look at attractiveness. You can solve problems for
constraints graphically with twenty or fifty increments,
but it’s impossible with a thousand. There are many
solutions that seem to work well on twenty examples,
but they don’t scale. It’s not just that the computer is
faster and you can do all the optimizations that are

possible—no, you have to increase by a factor of a M. P.: We don’t think it’s at all possible. What do you
million times before anything works. This is a challenge. mean by “objectivity”? You could say that we have a
It’s interesting to study the problem of model scale; this shape that has a certain performance, let’s say structur-
is what I mean when I’m also keeping this global ally. We could show that this performance is better than
perspective. You can solve this at a small scale and you any other alternative. Maybe in certain cases, you can
can learn quite a bit, but if you don’t understand the even prove it’s the optimal performance. Even so, it may
global aspect of it, and for example, see that your still need to be designed. I would say the structural
solution has an exponential complexity, then you will performance is just one criterion. At some point, you may
never be able to use it on a real-scale model. You can want to sacrifice structural performance for aesthetics,
argue that it’s still informative to look at it, but the or cost, or other aspects, but I’m viewing it from the
solution is not precise. That’s why perspective of the engineer. At the very least, you want
it’s good to go back to the global function; whatever you to be able to know if this is a well-performing structure
want to apply is a real example. Space parameters start or not. You don’t want to just guess performance by how
from two degrees of freedom to five hundred, raising the it looks, though I don’t know if there is a better solution
question: Is this approach still valuable? Many forms are or not. I would like to fully understand this aspect.
beautiful in two dimensions, but they don’t work when I think this is important, because if you knew these
scaled up by ten, or fifty, or two hundred. This is a things, then you could make meaningful decisions.
challenge, but it may just be important to reflect You could say, “OK, I accept these degrees of structural
complexity for certain tasks. For others you have to performance, because I know I gain something for it,
accept simplification and sometimes you have to hope and I don’t break down the structural performance for
for the best. You have to hope that whatever comes nothing, because I have the wrong design tool or
up—what’s actually tested—is still good enough. I don’t whatever.” This would be something you have already
think there is a way to simulate such a shape in terms of understood through your design: What you sacrifice for
formal, mechanical, and structural properties in a way what gains. I think for that you will need to have a more
that you could invert the simulation for form-finding. It complete picture of the design basis and its performance.
will always be an approximation, anyway. The question
is: Are your approximations relevant? You have to verify Y. W.: We often have a strong correlation between
your experiment. form and structure. At least, we say this because
I think at some point you have to go back to physical we feel it. But we cannot explain it in a deeper
tests. Of course, it would be nice if you could optimize sense.
this link between your physical experiments and virtual
designs to make them more aligned. If your virtual M. P.: Is that really true? For example, we can say that an
production and your physical tests have no relationship, arc would structurally have a more robust performance
then you have no meaning in this model. than any other form.
Y. W.: You see, there is a subjective motivation. Y. W.: I think there is something here, which is
Its design is a little bit different and we would convincing for the correlation between structure
like—with regard to the present publication—to and form, but this is just one take on an enormous
objectivize our research. We would like to number of topological possibilities and it’s not
increase our objectivity. Whether this is actually so clear. I would like to have more clarity here.
possible is something that is still unclear to me.
M. P.: I guess what you’re saying is that there are some
basic principles that can lead to better performance, but
these are only general principles.
83
Y. W.: Yes, and we have the cohabitation of why this image works so well in communication.
different structural systems in one, etcetera, but When it’s more free-style, like the table, for
we select a sort of mixture satisfying different example, it doesn’t have the same impact, even if
criteria, which are for performance motivation it’s made using the same tool. As an introduc-
and design, and we end up having entry in tion, I would like to have something documented
forms without being able to explain this in a about this.
more comprehensive manner. Are you more
attracted by this or that, or is this just the same M. P.: If you have desired constraints, the offset surfaces
research question for you when you consider the come back to the notes “beam torsion free” mathemati-
different scientific aspects? cally. Then it introduces a very clear mesh. Mathematical
theories will help you to design. As soon as you move
M. P.: That looks different to me, in the sense that the away from these types of surfaces, you’re going to run
structure is fixed here. I guess the question for this into this problem. If you don’t know that, then there is
type of thing is, do you just want to optimize the form, no way to fix this with just a cap-modeling tool. You
or would you also like to optimize the topology? really want an optimization that brings you back to this
space to avoid these problems. This is true for many
Y. W.: I would also like to have more information examples. We have local considerations that lead to
about the ideal size of these pieces. global dependencies. You know that you want to change
the position of one of these plates or its orientation,
M. P.: Again, it’s a matter of abstraction. But, as you go because you don’t like it. It’s very difficult to modify
through the abstraction to more concrete sizes, there anything, because somehow, when you modify things
could be some additional topics because the connections, you usually do it from a local viewpoint. You like the
the topology, or the parameters are different. form and you want to just change things, but you can’t,
because when you change something here, it will affect
Y. W.: Years ago, we were fascinated by all that the entire thing. If you do design in such a space without
we could do in terms of introducing new visual/ a global optimization in the background, then your
geometrical factors such as nurbs and meshes. results will almost be left up to chance, because maybe it
But then there was a rebound effect; an overload works out for some strange reason, or you accept that
of images, random images, images without things can change in reality, but you don’t truly
meaning. We didn’t want to continue down that understand it. You don’t know what you could do
route. Now it’s clear that we have a sort of differently to avoid this effect. What you are saying here
general tool and then very subjectively, we is probably also a question of perception.
select a different modification, which we then To be honest, this problem requires a sort of interdisci-
stick to. But this is a very subjective process. plinary effort, which is always a challenge in itself. To
Using this tool, for example, we can add an really make something like this meaningful, we can
architectural or structural solution. We’re far completely ignore all of the physical aspects, the
less enthusiastic about the wide range of structural aspects, the material constraints, and we can
different types of curves we can really produce. purely abstract the question to its geometry, which is
There is this motivation to travel down a very fine if you keep to a toy scale. Thin wood models support
general route for this pavilion, but the selection the structure anyway, so there is no issue. Of course it’s a
of certain images gives a new interpretation to limited approach, which cannot be generalized to
an existing vault structure. You have an atypical large-scale constructions. I think you need somebody to
view of something you know or something that look into the structural aspects; you need somebody to
is a bit familiar to you. I think that’s the reason look at global geometric aspects, and somebody else to

look at material aspects, and hopefully somebody will M. P.: We’re still trying to stay in this world of real-time
solve this optimization problem in some way. It would be feedback. We want to do optimization. If you have an
very challenging to put together the right team, find the insufficient system that take hours to give a single
right people, and make sure that PhD students can make answer, then if you do optimization, you need an answer
a thesis, and that all the constraints are of a university every ten milliseconds and we need to invert that
standard. I think at some level, if you want to have function. We’re coming from a starting point of purely
results that can be applied on a larger scale, you cannot geometric design, to make it interactive and make it
avoid going down this route. The approach based on a work faster. Now we’re adding physics, but without
toy world, where we can abstract physics into geometry, moving too far away from offline computation. We still
simply doesn’t scale. On one hand it’s convenient, want to do something inspired by physics or some
because we abstract a lot of problems and you can do plausible physical behavior added to these interactive
mathematics or some computer science models, but on processes.
the other hand, it’s frustrating, because you’re stuck in You can say here’s the real physics that is too slow, we
this toy world. My ideal approach would be to pull need to make it faster or simplify it in order to make
together the right people and somehow suffer through some interactive process. However, we started from some
this additional complexity of managing a bigger project, interactive process and we added as much as we could
of having highly skilled people and different communi- without losing interactivity. The rationale is that you are
ties and all the things that ensure the likelihood of a still in the early phases of concept design, where you
project’s success. That’s the big challenge here. I don’t would simply like to find a shape or a form that is a
think there is a good model. I can’t point to many starting point towards the final structure.
examples where we could say here is a structural We try to give you as much feedback in terms of physics
engineer, an architect, a computer scientist, and students as we can afford without claiming that it is already
at the university level, working together and making the meaningful for the structural properties.
work successful. I think it’s an interesting question of what you can add,
Perhaps there are more examples in industry, I don’t in what way, to inform your design and your form-
know, but I also doubt there are many. Possibly that’s finding with as much structural physics as possible.
the challenge we have to face. We have to educate people Once again it’s a question of how you design and what
to be able do this research. you want. Right now, we’re probably mostly looking
at this initial form-finding, exploring shapes rather than
Y. W.: While speaking with you and with Helmut finding them with structural properties in mind. But
Pottmann I had the impression that you this to me is the ultimate goal. In the end, you want to
would prefer to start your investigations from get all of the analysis tools you have at your disposal,
a specific situation with its own specific including structural performance, maybe energy
constraints and probably physical constraints. performance, and heat gain—all the things that you
can’t predict in a final given design. You’d like to bring
M. P.: We did move a bit into this direction to add more them into the earlier design stage in some form or
physics, but it’s still toy physics in many ways. another; but this is a lot of work.
It’s still insufficient to bring in simple elasticity models,
gravity forces, from an engineering point of view. Y. W.: From speaking with Matthias Kohler,
I know that there are so many interesting
Y. W.: This is precisely what Sina Nabaei did. processes. The students try to learn something
from the process, but we need a sort of final goal
or a synthesis of one kind or another. We’re
looking more to the synthesis or to the end
85
product. In many cases, process observations
concern additive processes. Here we aim for a
synthesis, where the global form is considered.
M. P.: That’s true, but I think that the approach as a

process is interesting in itself. The process you design or
that you want could be beautiful, but is simply not
realizable. That’s an interesting, but also a very compli-
cated process; you design a logic of connections that
abstracts the problem in some sense. When you say,
“Here’s my connection logic. From this logic I can
explicitly specify what needs to be satisfied for a
structure to be constructible.” If you construct a Lego
model, you know exactly how to connect the blocks. If
you change the design a bit, then suddenly the structure
might become unstable; again, it’s very difficult to
explore this.
Y. W.: Well, we probably will end up having both

considered: the process but also the synthesis.
I’m looking forward to exploring this with all
of you, taking advantage of our NCCR research
collaborations.

87
2. 5
Modular pavilion:
a structure for the Paléo Festival
Marielle Savoyat
Design and Project

IBOIS, Labor
designfür
and
Holzkonstruktionen,
execution: und EPFL, This structure was initially designed by the student
execution
IBOIS,
Eidgenössische
Laboratory Technische
for TimberHochschule
Constructions, Bastien Thorel in the Weinand studio at IBOIS, the Labo-
EPFL,
Lausanne,
SwissSchweiz
Federal Institute of Technology, ratory for Timber Constructions, EPFL (Swiss Federal In-
Lausanne,
Bastien Thorel,
Switzerland
Student, Prof. Yves Weinand stitute of Technology, Lausanne) during the academic
Bastien
und SinaThorel,
Nabaei, student,
Wissenschaftler
Prof. Yves Weinand year 2008/09. The goal of the exercise was to design a
and Sina Nabaei, researchers structure inspired by textile mesh for an architectural
2008–2009
program, which would then be proposed as a structure
Execution 2008–2009
Paléo Festival, Nyon, Schweiz for the Paléo festival in Nyon, a town on the shores of
Location Paléo Festival, Nyon, Switzerland Lake Geneva in Switzerland. Thorel proposed a structure
based on a discrete geometry composed of interlocking
wooden panels. Only two types of panels were used to
create this structure, which also acts as an envelope and
an enclosure. A basic module in the shape of a folded V
was connected to another basic module, also V-shaped
and mirroring the first shape. Each module included four
U-shaped cutouts, commonly called semi-wood, which
slide into each other to create an arc.
The result of this spatial structure—light, lofty,
aesthetic, and easy to install—made it particularly suita-
ble for a temporary function in the context of a summer
festival, like Paléo.
A prototype of this structure was first created at
the EPFL, to test both the structural feasibility of the con-
cept and the resulting architectural language. The project
necessitated the development of a digital cut with five
saws to create the U shapes, as well as the development of
the connection details and the overall structure. The basic
module was composed of 21-mm-thick, 3-ply panels.
A three-dimensional digital model was useful for
studying the structural behavior of the modular arc as un-
dertaken by IBOIS researcher, Sina Nabaei. The structure is
a good example of a complex global final form, built using
simple wood-wood connection techniques and folded and
interlaced modular panels. Its ingenuity lies most notably
in the fact that the modules support themselves reciprocal-
ly. The global form is determined by the local geometry of
the connection.
Fig. 1 Axonometry of the complete arch

Fig. 2 EPFL Campus 2009: Shell prototype
Fig. 1 Fig. 3 Section of a half arch

Fig. 2
vertical plane of symmetry

trimming plane
trimmed part on CV-element
horizontal trimming
plane
Fig. 3
trimmed part on CX-element
89
3 Active bending
3. 1 Actively bent and woven structures 92

Yves Weinand
3. 2 Geodesic lines on free-form surfaces—

optimized grids for timber rib shells 102
Claudio Pirazzi and Yves Weinand
3. 3 Apparent simplicity and modular complexity

in timberfabric structures 108
Markus Hudert
3. 4 “Shaping by bending is very simple” 118

Interview between Jan Knippers and Yves Weinand
3. 5 Geodesic lines for shell forms—

a playground installation at the Vallée de
la Jeunesse (Lausanne, Switzerland) 122
Marielle Savoyat
91
3. 1
Actively bent and woven structures

Yves Weinand
The tests carried out on IBOIS investiga- If the aim is to create a curved or double-
tions can be broadly divided into four curved surface, the above-mentioned
types of structures: structures can be utilized. When progress-
ing with a structural design, it is also
Structures created from flat panels: necessary to take the possible connection
– Origami-like folded structures techniques into consideration. Particular-
– Free-folded panels, e.g. developed ly in actively bent structures, new types
from iterative functions of connections are possible that would be
unfeasible in the structural types previ-
Actively bent structures: ously discussed. It is possible to change,
– Structures resulting from linear or even to do away with, the connection
supporting elements, such as rib node altogether. In the parabolic fence
shells posts depicted in fig. 1, the rotation of the
– Structures formed from planar ele- linear element results in a double-curved
ments, such as braided structures surface. The consequence of this is that
the connections vary in nature.
Fig. 1 a and b Parabolic fence posts
92 Active bending
Fig. 2 Glass cones generated from triangles
3-D chart of tower and axial system of profiles:
Fig. 3
the cross-section of the triangular frame is shown
in green; the cross-section of the inner tube is
shown in red. The axonometric shows linear
connecting elements—which are not incorporated
into the node—the triangular, self-contained
carrying frame, and the glued frame structure of
the glass panes.
Fig. 4 Elevation of the completed frame system
Fig. 2
Fig. 3 Fig. 4
93
Fig. 5 b
Fig. 5 a
Diagram of the nodes connecting a

Fig. 5 a and b
ribbed shell
Fig. 6 The rib is composed of several layers.
With an active connection of the rib, the indivi-
dual layers have a tendency to slide against
each other. The connections should help to
avoid, or at least reduce, this displacement.
Fig. 7 The stiffness of the connection must be
calibrated. The shifting of layers in relation
to one another generates transverse forces in
the connection, which are tested here.
Fig. 6 Fig. 7
In the example shown here, the connec- and replaced by cross connections along
tion is made from rubber, which is able to the sides of the triangle. The flow of
adapt to the specific, local form due to its forces thus no longer passes through a
elasticity. connection node, but is rather directed
around the node. The effect of the forces
Another example of the development of within the surfaces and their trans-
connections is demonstrated on a glass mission along the edges of these surfaces
cone in the Belgian town of Lommel. is preferred to the transmission of
Here, the node connections at the corner the forces through single nodal points.
junctions of the triangles are removed
94 Active bending
The system described here highlights the The series of examples shown here is
relationship between construction prin- based on standardized surface areas. We
ciple and connection detail. Depending on are interested in learning how a certain
the choice of the overall system, the con- size of surface area (or footprint) pro-
nection type varies. To begin with, struc- posed by an architect could be built over,
tural systems made from linear elements using the ribbed dome system. This work
were differentiated from structures com- is described below by Claudio Pirazzi.
posed of planar surfaces. The two types The following fig. 10 a and b illustrate the
of structures resulted in different types of construction of a prototype.
connections; this fact is essential for the
developed structures explained below. The ribbed shell constructions presented
here reference, on the one hand, felt fabric
The most widespread type of structure in and, on the other hand, woven items,
the category of actively bent systems is such as knitting patterns, braiding, and
the ribbed dome. Well-known examples weaving.
of ribbed dome buildings are the halls of
the Federal Garden Show in Mannheim The geometric view of this pattern shows
(Mutschler/Otto), the Polydôme at the Uni- that a stabilizing “system effect” strength-
versity of Lausanne (Natterer), and Savill ens these structures. The close proximity
Garden in London (Happold). The main of the elements to one another results
load-bearing elements or ribs are actively in a high friction effect before the system
bent elements that are not interrupted can be deformed. This property was
at their connection points but continue, of particular interest to us. It raised the
uninterrupted, with components of up to question of whether this material-specific
50 meters in length. Their cross-sections assessment might be used from an
thus fully retain their strength and are engineering perspective to stabilize and
not weakened. Ribbed domes consist reinforce structures.
of continuous supporting elements, as
opposed to disconnected support sys- If you start to “weave” wooden materials,
tems—such as reciprocal systems—where you have to deal with various systems.
the structural parts are interrupted Fig. 13 illustrates a braided structure
and the nodes have to be connected to that develops in the plane. The spatial
one another. A possible variation of interpretation of this woven structure is
a ribbed dome connection is shown in illustrated in fig. 14. A spatial knitted
fig. 5 a and b. pattern is created by the braiding pattern
along the bottom or along the third axis.
However, the extrapolation of this local
principle in a curved structure diverges
95
inserted at the top of the image. As shown
at the bottom of the image, the compo-
nents are forced into both actively bent
and planar forms. Here, the structure
is subdivided into a primary weave and a
cross weave direction, represented by a
linear element.
The targeted structural optimization of

the geometry remains a focus. The weave
presented here shows that deeply woven
Fig. 8
strands not only are interrupted but are
also shifted. This shift happens spontane-
ously, and results in the shifting of the
network strands in relation to one another.
In the fall of 2008, the architecture studio

at the EPFL focused on the theme of
woven structures. Some of the results
Fig. 9
have already been published in Wood:
Fig. 8 The IBOIS team paid special attention to the
Material or Form: Transformation of Struc-
continuous formation of the ribs when they built tural Logic (2014).
their experimental ribbed shell.
Fig. 9 Modeling of geometry as finite element models,
while taking into account the connection stiffness
between the layers.
Markus Hudert started to manipulate
forms of textile modules, then to multiply
them and shift them. One of his first suc-
cessful attempts is depicted in fig. 19 a.
from the original concept. The arched Through the active bending of two planes
truss shown in fig. 15 is unable to and their rotation into a single, unified
replicate the rigidity of the basic braiding system, a compact, seemingly simple form
pattern at this scale. is generated. The wood panel surfaces
were bent and twisted, thereby producing
The example shown here fails because bending and torsion moments, building
exclusively linear components are used. up residual stress.
For this reason, actively bent and twisted
components are also used, as illustrated It is fascinating to examine the standard-
in fig. 16 a and b. Leonardo da Vinci’s ization of this support system. If the
system of reciprocal self-stabilization is panel edges lie in a parallel and horizontal
96 Active bending
Fig. 10 a and b Construction of a free-form ribbed dome at IBOIS, EPFL,
based on the use of geodesic lines on a free-form surface.
97
Fig. 11 Fig. 12
Fig. 13 Fig. 14 Fig. 15
Fig. 11 Summary of the braiding and weaving Spatial representation of basic braiding
Fig. 14 Kerto plates; span 21 m; Atelier Weinand /EPFL
patterns patterns 2009. Student: Sophie Carpienteri
Fig. 12 The three considerations are portrayed Fig. 15 Representation of a stave mill, Fig. 18 Student project from the workshop “woven
diagrammatically. drawn from the basket weave. structures,” EPFL, 2008
Fig. 13 Representations of basic planar braiding Fig. 16 a and b Reciprocal system: actively bent weave
patterns
Fig. 17 Fish Market in Tokyo: woven roof structure
Fig. 16 a Fig. 16 b
Fig. 17 Fig. 18
98 Active bending
Fig. 19 a Fig. 19 b
Fig. 21 a
a) natural interstice
Fig. 20 a
b) bar-shaped
Fig. 21 b connectors
Module
height
Module width
Fig. 20 b c) multi-directional
connector
Module length
Fig. 21 c Fig. 22
Fig. 19 a and b Textile module, EPFL, 2008 Fig. 21 a, b, and c Illustration

of static behavior:
fixed or contact points, applied force increases
Superimposition of nodal points and
Fig. 20 a, b, and c
structural height
offset of network elements in relation to one another
Fig. 20 c Fig. 22 Geometry of fabric module
99
S, Max. (MPa) S, Max. (MPa)
In-plane principal In-plane principal
SNEG, (fraction = –1.0) length/width = 7 = 11.55 m/1.65 m length/width = 8.7 = 11.55 m/1.32 m SNEG, (fraction = –1.0)
(Avg. 75 %) max. principal stress = 46 MPa max. principal stress = 29 MPa (Avg. 75 %)
+1.458e+02 +1.788e+02
+1.289e+02 +1.620e+02
+9.516e+01 +1.284e+01
+6.142e+01 +9.480e+01
+2.769e+01 +6.121e+01
–6.051e+01 +2.762e+01
–2.292e+01 –5.970e+00
–5.666e+01 –2.277e+01
Finite element model of the active connection of a fabric module. Comparison between
Fig. 23
two column widths: 1.65 m and 1.32 m. The narrower the plate, the lower the residual stress state.
Fig. 24 The calculated values on a physical model can be verified with the aid
of a test form.
100 Active bending

position, then the panels are forced into
an inclined position by mutual bending.
The panel reaches a certain height and
forms a kind of hat or triangle in cross-
section. Fig. 21 a shows the contact points
in plan view. The cross-section described
is superimposed in red in fig. 21 b and c.
If an external force is added to this bent
active module, then it increases in length
and contracts in cross-section.
Thus the cross-section gains height, and
therefore structural height. From an
engineering perspective, it is a kind of
active system that increases in rigidity
when the external force is increased.
The geometric position of the nodal points

and the panels in relation to one another
directly determines the final shape of
the fabric module and its stress state.
The contact conditions shown in fig. 1
transform into peripheral conditions if a
layering of textile modules is envisaged.
This observation concerns the primary
weave direction and also leads to the
emergence of the cross weave direction.
Fig. 2 illustrates this cross weave, which is

created with linear slats that serve at
least two functions:
– Rigidity of the woven structure in
depth
– Peripheral conditions of a module
101
3. 2
Geodesic lines on free-form surfaces—

optimized grids for timber rib shells
Claudio Pirazzi and Yves Weinand
In
Mithilfe
order toderoptimize
Berechnungssoftware
grids of timber GEOS,
rib shells
die with
zwi- 1 Introduction
schen
regard2002
to theundbending
2004stress
an derofEidgenössischen
the boards due to Tech
ini--
tial
nischen
curvature,
Hochschule
GEOSLausanne
software (EPFL)
was developed
entwickelt atwur-
the Although concrete and steel are the most common materi-
Swiss
de, lassen
Federal
sichInstitute
Holzrippenschalen
of Technology hinsichtlich
in Lausanneder als for the construction of lightweight spatial structures
(EPFL)
Biegespannungen,
between 2002 denen
and 2004.
die einzelnen
The construction
Bretter in-of in modern architecture, timber has recently had a well-
a
folge
prototype
der Krümmung
in the summer
ausgesetzt
of 2005sind,
proved optimieren.
the relia- deserved renaissance. In addition to glue-laminated tim-
Durch
bility ofdentheBau
assumptions
eines Prototyps
upon which
im Sommer
the program
2005 ber, screw-laminated timber has increasingly been applied
wurde
is based. die Gültigkeit der Annahmen überprüft, auf to rib shells. During the past two decades several spatial
denenThedasfollowing
Programm article
basiert.
describes the steps that structures of this type were constructed.1,2 The transpa-
were Dertaken,
folgende
starting
Beitrag
fromdokumentiert
the generation die of
einzel-
the rency of their bearing behavior and their aesthetic archi-
form
nen Schritte,
and theangefangen
design of the bei optimized
der Formfindung
grid to und
the tecture fascinate experts and laypeople alike.
der
finalGestaltung
construction desofoptimierten
the prototype.
Netzes Loading
bis hintests
zum The ribs are made from laminated timber boards,
endgültigen
were carriedBau out des
to evaluate
Prototyps.theEsstructural
fanden auchcalcula-
Be- which are joined together with the aid of pin-like fasten-
tion
lastungsprüfungen
model. Finally, statt,
the comparison
um das statische
betweenBerech-
calcu- ers, like screws or nails. In contrast to other rib shell
nungsmodell
lated and measuredauf seine
deformations
Genauigkeitwillhinbezudiscussed
überprü- structures, which were made from square timber sec-
fen.
briefly.
Am Schluss des Beitrags werden die berechneten tions, 3, 4 relatively thin boards were used with a thickness
und die gemessenen Verformungen kurz verglichen. of between 16 mm and 35 mm. These laths are inexpensive
and construction can be executed in a relatively simple
way without sophisticated techniques. The extra cost due
Keywords rib shells, active bended structures, to manual labor for the assembly can be compensated
geodesicRippenschalen, biegeaktive
lines, multi-layered Tragwerke,
beam structure for by rationalization methods during the planning and
geodätische Linien, mehrlagige manufacturing process. For geometrically demanding
Balkentragwerke structures this method of construction is a viable alterna-
tive to glue-laminated timber rib shells.
In order to reduce the stress due to initial cur-
vature, the ribs on the surface are arranged according
to geodesic lines. Thus, bending of the boards across the
strong axis, which causes unfavorable stress, can be
avoided. Ideally the boards are only subjected to bending
and torsion across their weak axis. In addition, this ap-
proach allows the use of straight boards. A geodesic line
on a surface is defined as a curve, where the normal vec-
tor of both curve and surface are parallel or non-parallel
at each point. The shortest distance between two points
on a surface is always a geodesic line. In plan, a geodesic
line represents a straight line. The term “geodesic” is de-
rived from the Greek ge (earth) and daiesthai (to divide).
Geodesic lines on simple, regular-shaped surfaces
can be determined by analytical means. Geodesic lines
correspond to helices on cylindrical surfaces and to great
circles on spheres. On free-form surfaces—currently en-
joying great popularity in contemporary architecture—
102 Active bending

the determination of geodesic lines is far more complex. In
order to satisfy this demand and to improve automation
of the production process, GEOS software was developed
in close collaboration with the Laboratory of Timber Con-
struction (EPFL/IBOIS) and the chair of Geometry (EPFL/
GEOM). This software calculates grids of geodesic lines on
free-form surfaces and provides all geometric data neces-
sary for computer-controlled sawing. The project was
financed by the Swiss National Science Foundation (SNF).
In order to examine the reliability of this pro-
gram and its precision concerning the assumptions
made, a free-form shaped timber rib shell prototype has
been designed with GEOS and constructed at the IBOIS.
The realization of this prototype was financed by the
“holz 21” fund, a research program at the Swiss Federal
Environ mental Agency (BAFU).
2 Form design and calculation of

geodesic lines
The form is defined by manipulating control points of

cubic Bézier-polynomials within a certain user-defined
number of parallel cutting planes along the surface.
A Bézier surface is calculated based on these polynomials.
Starting points and end points of each geodesic line
and the connectivity of the lines have to be defined for
the calculation of the grid (Fig 1). The iterative calculation
of the geodesic lines is based on a L-BFGS-B algorithm.
The tangential vector is used as a control variable. This
quasi-Newtonian targeting turned out to be the most Fig. 1
efficient method for calculating geodesic lines on mono-

patched, free-form surfaces.5 Fig. 1 Software GEOS; design of the free-form surface by
means of control polynomials (above); definition of starting
The user has no direct influence on the topology
and end points of the geodesic lines and their connectivity
of the resulting grid, but can manipulate the starting in a u,v-parameter range (below)
points and end points and define the connections of the Design of the timber rib shell prototype; generated
Fig. 2
geodesic lines. Thus, the result is not always compatible mono-patched free-form surface (left); geodesic line model
with structural and architectural demands like the regu- (center); multi-layered model (right)
larity of the grid or the minimum radius of curvature of a
single board. It has to be adapted by an iterative process.
In general it can be stated that the spacing of the meshes Based on the calculated grid of geodesic lines, a multi-
tends to widen in significant convex areas, whereas it layered model is developed according to the number and
tends to narrow in concave hollows. However, the exist- the dimension of the laths (Fig 2). This is achieved by ex-
ence of geodesic grids with homogeneously spaced meshes trapolating the layers normal to the surface inwards and
on complex free-form surfaces is not always evident. outwards. The midlines of the extrapolated layers no
Fig. 2
103
longer correspond to geodesic lines. They are neither part
of the initial surface, nor do they represent geodesic lines
for the enveloping surface to which they belong. However,
the error resulting from this inaccuracy seems to be neg-
ligible with regard to the relatively large ratio of the
effective radius of curvature R0 and the thickness d of
one single board, usually applied in practice. The smaller
this ratio is, the more important the inaccuracy becomes,
with the risk that the boards of a rib will no longer
be parallel and that the predrilled holes at the intersec-
tions will not fit precisely. Therefore, the front section of
the prototype has been designed with a relatively small
radius of curvature, with a minimum ratio R0 /d of 100.
3 Construction of the prototype
Thin timber laths made from Swiss spruce wood (Picea

abies), with a rectangular section of 12 × 60 mm were used
for the construction. Due to the tight curvature of the
front part of the shell, high quality wood for the laths
is imperative. Careful visual grading guaranteed an ex-
cellent strength class with only a few knotholes. The
modulus of elasticity was determined for five random
samples. The average value is 14,800 MPa. The ribs con-
sist of four laths: two of them are continuous and the re-
Fig. 3
maining two are considered as intermediate in-fill layers.
Continuous layers of diagonally crossing ribs intersect
them. Both directions are offset by the thickness of
one lath. No finger joints were required. The longest lath
is 6,740 mm and the shortest is approximately 160 mm. In
total, 792 pieces were mechanically sawn with a five-axis,
computer-controlled Créno saw. The data was exported
from GEOS in dxf. format.
Beginning with the inner layer, the laths are
fixed at their starting points before they are curved
across scaffolds and connected to the corresponding end
points. The five transverse scaffolds have been planed
in order to ensure a maximum number of fixed points
in space. In general, these supports are unnecessary
because the expected form is automatically obtained by
successively connecting the laths at their intersection
points by means of bolts (10 mm diameter). The continu-
ous inner layers in both directions could thus be realized
in a relatively short amount of time, followed by the in-
stallation of the two continuous outer layers (Fig 3). In
general, it can be stated that even after assembly of the
fourth layer, the predrilled holes at the intersections fit-
ted precisely about 85 percent of the time. This confirms
the reliability of the calculation and shows that the re-
quired precision is kept even for significant curvatures.
Fig 3 Construction of the prototype

Fig 4 Intersection points before (above) and after (below)
assembly of the intermediate in-fill layers
Fig. 4
104 Active bending

Fig. 5 Load tests; uniaxial displacement transducer (left); loading with two, single loads (center); load measuring (right)
The intermediate in-fill layers were mounted after plac- sections around the charged points was measured (Fig. 5).
ing all continuous layers in both directions. Screws with The displacement transducers were adjusted so that they
a diameter of 4 mm were used to connect the layers. The measured deformation normal to the shell’s surface and
effective distance of the doubled-rowed fasteners is about parallel to the introduced load in the case of the charged
50 mm. Fig 4 shows a detail before and after the assembly intersections. The test results show that, for relatively
of the intermediate in-fill layers. The in-fill layers are low charges, the structure undergoes hardly any plastic
approximately 2 mm shorter than the distance between deformation at all and reacts absolutely symmetrically.
the continuous laths at both ends. A distinct increase in stiffness could be shown after
mounting the intermediate in-fill layers. This is not only
true for load cases, where the structure shows flexional
4 Data behavior, but also for load cases that are essentially
transmitted by standard forces. Certainly, the flexional
Lath, rectangular section: 12 × 60 mm (602 running stiffness of the ribs increases considerably if the inter-
meters of high quality mediate in-fill layers are mounted and screwed down.
Swiss spruce) Thanks to the double-curved surface, the effect of the
Number of pieces: 792 (including intermediate inter mediate layers on the stiffness of the structure was
in-fill layers) estimated to be far less significant. In most cases exam-
Timber volume: 0.434 m³ ined, the deformations were halved by assembling the
Dimension of the shell: 8000 × 3000 × 2060 mm intermediate in-fill layers.
Developed surface: ≈ 36.0 m²
Base: ≈ 18.6 m²
Number of bolts: 202 M10 6 Structural analysis
Number of screws: ≈ 3000 4.0 mm
The comparison of the measured deformations wmes
with the calculated deformations wcal enabled the eva-
5 Load tests luation of the calculation model based on a framed load-
bearing system. The structural analysis was carried
In order to get a better understanding of the load-bearing out with SAP2000 V9.16 NL software from CSI-Berkeley.
behavior of the structure, load tests were carried out be- Based on the geometric data exported as a dxf. file from
fore and after the assembly of the intermediate in-fill GEOS, the structure was programmed in an external text
layers (construction phases C1 and C2). For both phases, file and then imported. The generation of data by means
three symmetric load cases with single forces on differ- of external text files allows one to carry out modifica-
ent intersection points were examined. The forces, intro- tions in a relatively straightforward way and ensures
duced by a cable winch, were applied normal to the a more general overview of the data.
surface. Uniaxial deformation of a total of seven inter-
105
The load-bearing behavior of the structure is fairly com- connections and given in the technical specifications—
plex. The engineer has to take various effects into consid- cannot be applied without modification. 6 Preliminary
eration. The extent of the effect on the structure is basi- tests were carried out to determine the initial slip modu-
cally unknown. This makes modeling fairly difficult. lus in this specific connection. Furthermore, the initial
Apart from the influence of the rotational stiffness of bending of the laths has a pre-stressing effect on the
the ribs’ intersections (distortion of the diamond-shapes structure’s stiffness. This effect—ideally favorable—is
in their plane), the initial slip modulus of the screwed disregarded.
connection is largely unknown. Due to different displace- As the intersections of two ribs are not entirely
ment behavior of the connection, the characteristic values rigid, both directions are modeled separately in the
of this modulus—empirically determined for standard form of two substructures. The degrees of freedom are
subsequently coupled by means of kinematic linking
Substructures B1 and B2 Substructures A1 and A2 (“local-constraint” option). This allows the release of a
single rotation about the local axis normal to the surface
at each intersection and thus the simulation of the distor-
tion of the diamond-shapes in their plane. According
to the method of shear analogy 7–10 two subsystems—
A and B—are introduced. This method is applied to the
modeling of the construction phase C2 in order to take
the flexible compound cross-section of the ribs into con-
sideration. Subsystem C, introduced for stability problem
by Scholz,11 is neglected due to minor normal forces.
Therefore, the model consists of four substructures, two
Fig. 6 for each subsystem (Fig. 6).
60 The following assumptions were made:

w9 w10 w12 w14
40 − Isotrope and ideal elastic (Hook) behavior of
timber
20
− Geometrical non-linearity (P-delta effect)
Deviation eta (%)
0
− Linear slip behavior of the mechanical fasteners
-20 with an initial slip modulus Kser,1 of 500 N/mm,
-40 according to preliminary load-tests carried out
-60
at the IBOIS 12
− Rotation about the frame 1-axis is considered to
-80
be constant between two intersections
-100
− Initial state of bending stress is not taken into
0.0 0.5 1.0 1.5 2.0
account
Force (kN)
− No fixed restraints of the substructure B
60
w9 w10 w12 w14 (slip between the laths is not hindered at the
40
restraints)
20 − Torsion stiffness due to the build-up compound
section is not activated
Deviation eta (%)
-20
− Weakened sections at the intersections are not
taken into account
-40
-60 For two demonstrated load cases of construction phases

-80 C1 and C2, the relative deviation between measured and
-100 calculated deformation is shown in fig. 7 (before and af-
0.0 0.5 1.0 1.5 2.0 ter fitting the intermediate layers). Observing the former,
Kraft (kN) one can see that with an increasing load level, the devia-
Fig. 7
tion increases noticeably. This is due to nonlinear effects,
which were not taken into account in the structural anal-
Fig. 6 Four substructures used for modeling of the ysis. However, for small charges, a strong accord between
construction phase C2 in the floor plan. Linking was by calculation and reality (with relative derivations mostly
means of kinematic constraints (not in view).
below 30 %) was observed. Even the comparison of the fi-
Fig. 7 Comparison between measured and calculated
nal construction phase C2—far more complex in its struc-
deformations of two load cases of construction phase C1
(above) and C2 (below). Directly charged joints are shown tural behavior and with additional compound stiffness—
in dotted lines. shows satisfactory accordance.
106 Active bending

7 Conclusions Literature
1 Natterer, J., N. Burger, and A. Müller. “The roof structure
The GEOS software, developed to calculate grids of geo- ’Expodach‘ at the World Exhibition Hanover.” Proceedings
desic lines on free-form surfaces, was tested by the con- of the Fifth International Conference on Space Structures.
struction of a timber rib shell prototype. The precision of University of Surrey, United Kingdom, 2002. 185–193.
the calculated geodesic lines of the computer-controlled 2 Natterer, J., N. Burger, and A. Müller. “Holzrippendächer
prefabrication of the laths and its assembly was found to in Brettstapelbauweise: Raumerlebnis durch filigrane
Trag werke” (“Screw-laminated Timber Rib Shell Roofs:
be highly satisfactory and augurs well for an application
The Experience of Space through filigreed Structures”).
of this type of structure at a larger scale. Bautechnik, 2000, Vol. 77. No. 11. 783–792.
Load tests were carried out in order to evaluate 3 Happold, E. “Timber Lattice Roof for the Mannheim
the structural computer model. For the observed load cas- Bundesgartenschau”. The Structural Engineer, 1975, Vol. 53,
es, the tests show that the structure is approximately No. 3. 99–135.
twice as stiff after assembly of the intermediate infill 4 Kelly, O.J., R.J.L. Harris et al. “Construction of the Downland
layers. Good agreement between the measurement and Gridshell.” The Structural Engineer, 2001, Vol. 79, No. 17.
the calculation, at least for relevant loads, has been ob- 25–33.
served. This is even more pertinent with regard to the 5 Rozsnyo, R. “Optimal Control of Geodesics in Riemannian
Manifolds.” Doctoral thesis, EPF-Lausanne. Switzerland,
complexity of the treated structure. Several para meters,
2006.
such as the initial slip modulus of this specific connection
6 Pirazzi, C. “Zur Berechnung von Holzschalen in Brettrippen-
or the level of rotation at the intersections (no in-plane bauweise mit elastischem Verbundquerschnitt” (“On the
shear stiffness) have not yet been completely clarified. Calculation of Screw-laminated Timber Rib Shells”). Doctoral
Further reflection on the influence of these parameters thesis, EPF-Lausanne, Switzerland, 2005.
on the structure’s bearing behavior is necessary. 7 Ibid.
GEOS provides an important tool for the design 8 Kreuzinger, H. “Verbundbauteile aus nachgiebig mitein-
and the realization of timber rib shells. It contributes ander verbundenen Querschnittsteilen” (“Structural Mem-
to clarifying current uncertainties in the design plan- bers made out of semi-rigidly connected built-up Sections”).
Ingenieurholzbau, Karlsruher Tage 2000, Forschung für die
ning process and will greatly improve the confidence of
Praxis Vol. 28, 2000. 42–55.
engineers and architects in conceiving and realizing this
9 Kreuzinger, H. Mechanically Jointed Beams – Possibility of
type of challenging lightweight spatial structure. Analysis and Some Special Problems. Inter national Council
for Research and Innovation in Building and Construction,
Working Commission W18. Venice, Italy, 2001.
10Scholz, A. “Ein Beitrag zur Berechnung von Flächentrag-
werken aus Holz” (“A Contribution to the Calculation of
Spatial Timber Structures”). Doctoral thesis, TU Munich,
Germany, 2004.
11 Ibid.
107
3. 3
Apparent simplicity and modular complexity

in timberfabric structures
Markus Hudert
The
Die Forschungsarbeit
research work “Timberfabric:
„Timberfabric:Applying
Applying Textile
Tex tile 1 Introduction
Assembly Principles for Wood Construction in Archi-
tecture” untersucht,
explores theinwieweit
interplaydas
between
Zusammenspiel
assembly This paper is based on the doctoral research “Timber-
von
procedures
Fügeprozessen
and the properties
und Elementeigenschaften
of the involved com- als fabric: Applying Textile Assembly Principles for Wood
generativer
ponents as wellFaktor
as the
im Architektur-
potential of this
undinterplay
Trag werks-
as Construction in Architecture,”1 which the author under-
entwurf
a generative
genutzt
factor
werden
in architectural
kann. Inspiriert
and structural
durch tex- took at IBOIS between 2007 and 2013. First, this re-
design.
tile Fügeprinzipien
More specifically,
entwickelt
it examines
die Arbeithow princi--
die Grund search explored the interplay between assembly pro-
ples
lagenoffürtextile assemblyBausystem,
ein neuartiges techniques, as applied
in dem to
elastisch cesses and the properties of the components involved,
elastically deformable timber
verformte, mehrschichtige elements,
Holzplatten eine can be
tragen- and second, the potential of this interplay as a generative
employed for theAusgangspunkt
de Rolle spielen. development of ist an innodas
dabei vative
so- factor in architectural and structural design. More spe-
unit based Holzgewebemodul,
genannte timber construction system.
das ausThe starting
zwei sich cifically, principles of textile assembly techniques were
point in this is the
überkreuzenden so-called Timberfabric
Holzplattenstreifen besteht.Module
Durch examined—in combination with elastically deformable
which
die consists ofmehrerer
Kombination two interlaced
solcherstrip-shaped
Module können ele- timber elements as components—to try and find out
ments and
diverse in which various textile
Tragwerkskonfigurationen properties are
unterschiedlicher how they could be employed for the development of an
combined.
Komplexität Through
generiertthewerden.
combination of several such innovative unit-based timber construction system.
modules,Der different support structure
im englischen Originaltext configurations
verwendete With the above-mentioned goal, this research
of varying
Begriff complexity can
„Timberfabric“ ist be
nurgenerated.
schwer ins Deutsche adopts an empirical approach that is subdivided into
zu übertragen. Die Übersetzung „Holzgewebe“ sollte three parts. The first part determines the timberfabric
hier als Kompromisslösung verstanden werden, da module as a basic unit for the modular structural system;
die im Beitrag vorgestellten Strukturen weniger auf this is the developmental goal. The second part examines
Web-,
Keywordssondern eher auf
experimental Flechtprinzipien
timber construction,basieren. the properties of this module and the implications of its
assembly and material driven design, large-scale production. The final section, which forms
material computation the basis of this contribution, systematically explores dif-
Holzgewebe, Makrotextilien, ferent possibilities of combining multiple modules into
formaktive Holztragwerke more complex structures, and examines how the connec-
tivity between their components can be established and
optimized.
The production of physical models and proto-
types plays a crucial role throughout the process. Physical
modeling allows direct measurement of the material’s
elastic deformability and the impact of the proportions
of the basic elements. Furthermore, it contributes to a
general understanding of the geometric foundations of
the developed structures, as well as of their mechanical
properties. The models and demonstrators resulting from
this process are examined with regard to a series of select
evaluation parameters. The overall impact of transfer-
ring textile assembly principles is one of the issues dis-
cussed. Analogies and differences between the developed
and actual textile structures are identified. The insights
derived provide the basis for a modular construction
system for self-supporting building envelopes. It is shown
108 Active bending

that the interplay between material properties and as- latter “[…] are considerably greater in length than in
sembly techniques not only generates a specific architec- width or in diameter.”2 However, by definition,3 braided
tural form of expression, but also contributes to the spa- textiles involve at least three yarn elements, while the
tial and structural qualities of buildings. timberfabric module, which is the basic unit for one-
directional timberfabric, consists only of two elements.
Another ana logy can be found in the twisted fibers in
2 The basic unit of timberfabric structures yarn. The development of the one-directional timber-
fabric is an inter mediate step towards two- and three-
Timberfabric structures are modular, wooden construc- directional structures, where the one-directional struc-
tions that can be configured with varying levels of com- tures act as higher-order units.
plexity. Their basic unit is known as the timberfabric
module. This module was developed based on the unit 4.1 Geometry and structural performance
cells of woven fabrics, as well as on the principle of ar- In an individual timberfabric module, the great-
ranging elements in a helical order, as is applied in yarn est cross-sectional height is located at the module’s center,
twisting and some forms of braiding. The unit cells of whereas the lowest height is located at its extremities.
woven fabrics are comprised of two interlaced sections In assemblies with two or more modules, lower cross-
of yarn and are of interest because of their modular qual- sections are not only located at the support points, but be-
ities and the reciprocal relationship of the components come part of the arch-shaped structure. As a result, the
involved. The timberfabric module is likewise composed structural height of an arch-shaped assembly such as
of two interlaced lamellar elements that support each this varies between zones of higher and lower strength.
other in a reciprocal manner. Taking inspiration from In one-directional configurations, additional elements
the techniques of yarn twisting and braiding, these two are required to connect of modules in the longitudinal
components are assembled to achieve a helical disposi- direction. In addition to their function as connectors,
tion. Along with the components’ properties, this manner these elements also have the potential to reinforce the
of assembly generates the overall shape of the module. zones of lower strength at the modules’ extremities.
3 Multi-module configurations 5 Two-directional timberfabric (2DTF)
Based on the geometrical and mechanical properties of Two-directional timberfabric structures are produced
the timberfabric module, different strategies of how to by aligning several one-directional units perpendicular
combine several modules into more complex structures to the span. Two types of one-directional units can be
were developed. By systematically applying these strat- used that are inherently either able to link or unable
egies, various kinds of one-, two-, and three-directional to link in the lateral direction. Regarding their connec-
configurations were generated. Although the first two tivity and coherence, more requirements arise in two-
configuration types could be implemented as structures directional configurations.
in their own right, the third variety has the greatest
potential as applications for load-bearing building en- 5.1 Additional connectors in the lateral direction
velopes: the use of multiple layers not only provides an In interconnected configurations, coherence in
improved structural performance, it also permits greater the lateral direction is an intrinsic property. In two-
control of the structure’s degree of transparency and directional timberfabric based on self-contained assem-
thus the quantity and quality of incoming daylight. As blies, supplementary connectors are necessary in order to
mentioned above, the timberfabric module is the basic establish a coherent structure. These can be designed as
unit in all cases. Hence, the factors that influence the independent connectors in addition to those that estab-
geometry of an individual module are likewise relevant lish connectivity in the longitudinal axis. Alternatively,
for the overall geometry of multi-module assemblies. according to a more inclusive approach, these connectors
can be employed to fulfill both tasks simultaneously.
4 One-directional timberfabric 5.2 Alignment of laterally adjacent modules

The manner in which laterally adjacent modules
Different methods can be employed to combine several are aligned is relevant for the ways they can be con-
modules into one-directional assemblies. The resulting nected. Two options relating to the properties of the
assemblies are three-dimensional and have a width timberfabric module are considered, where one stronger
that is relatively narrow in relation to the length of the and two weaker zones were identified. Due to their
structure. With regard to textile assemblies, one-direc- geometry4, these two weaker zones are also the most suit-
tional timberfabric can be compared to twisted fibers able positions for connector elements in the lateral direc-
in yarn, or certain types of braided textiles, where the tion to the span. Resulting from these considerations,
109
two possible constellations of laterally adjacent modules approximately described as the linear extrusion of a
can be envisaged: either fully aligned or rotated in rela- cross-section profile. However, one can also imagine
tion to one another by the length of half a module. two-directional structures that are extruded along a
curved line. This could be achieved if the individual arch-
5.3 Minimum distance in a lateral direction shaped units were based on tapering components. As a
to the span secondary effect, the global cross-section of such struc-
In a lateral direction to the span, the minimum tures would have a non-continuous curvature.
distance between the central axes of two modules corre- One example of a two-directional configuration
sponds to the measurement of their width, which is deter- is shown in the following figures. This configuration is
mined by the dimensions of the basic panels. In this case, comprised of mutually rotated, self-contained, higher-
the two structures touch each other at their lateral edges. order units and is implemented with V-shaped connector
To avoid this, the minimum distance is set to twice the elements.
width of the components used in the modules.
In interconnected structures, this distance is de-
fined by the length and angle of the lengthening ele- 6 Observations on one- and
ments, whereas the angle is in turn determined by the two-directional timberfabric
geometry of the timberfabric module. The geometry de-
pends on the connection detail and the proportions of the A number of observations can be made regarding one-
panels. The global geometry of each of the variants and two-directional timberfabric, some of which are
of two-directional timberfabric presented here can be listed below.
Fig. 1 Fig. 2
110 Active bending

6.1 Self-contained and interconnected 6.3 Textile aspects and analogies
configurations One-directional structures can be considered as
If a two-directional structure is based on repeat repeats. Similar to woven textiles, a warp and a weft di-
units that are exclusively composed of modules, addition- rection can be identified in two-directional structures.
al connector elements are required to establish coherence Expandability is perpendicular to the span direction.
in the lateral direction. In structures with laterally con- This can also be considered as analogous to active and
nectable units, this is not the case. However, connector passive “thread” systems. Although one-directional build-
elements in a lateral direction can be employed as re- ups can be used as freestanding structures or as struc-
inforcing elements. tures in themselves, they are more often considered as
intermediate structures that define or describe the char-
6.2 Number of connector types acter of two-directional structures. In that sense, they
Connectors in the lateral direction can be de- can be interpreted as a physical implementation of a re-
signed as a second type of connector in addition to those peat. Repeats are graphic representations or descriptions
that establish connectivity in the direction of the span. of the pattern of a woven fabric.
Alternatively, a single type of connector that is able to
fulfill both functions is conceivable. Each approach has
implications for the assembly procedure and sequence. Prototype of a two-directional configuration with
Fig. 1
multi-directional connector elements
Fig. 2 Connection between modules in longitudinal direction
Fig. 3 Connection between modules in lateral direction
Fig. 4 In this study, the principle of multi-directional connectors
is transferred to a three-directional configuration.
Fig. 3 Fig. 4
111
Fig. 5 Fig. 6
Intermediate-scale prototype of three-directional

Fig. 5
timberfabric, configuration study one, photographed
near completion
Fig. 6 Connection of modules in longitudinal direction
Fig. 7 Connection of modules in lateral direction
7.1 Offset between layers
Fig. 8 Connection of modules in radial direction
In all variants except one, the superposed layers
circumscribe a cylindrical segment and have a common
central axis (that runs in the x-direction) while the exter-
7 Three-directional timberfabric ior layer is perpendicular to the interior layer. The
offset between the two layers is achieved by increasing
The previous section showed how multiple modules the size of the modules (the elements’ width is neglected
and their basic components can be combined into one- here) that are designated to be on the exterior layer.
and two-directional timberfabric assemblies. For two-
directional assemblies, two modes of aligning laterally 7.2 Minimum offset
adjacent modules have been identified with regard to Technically, the value of the offset can be freely
potential additional connector elements in a lateral di- chosen. However, several aspects have to be considered in
rection. In three-directional timberfabrics, these prin- the process. If the offset is too small, the units cannot be
ciples are taken to the next level by superposing and rotated about each other, as this would cause a collision
connecting two layers of two-directional timberfabric of between the low-lying extremities of the upper module
identical or different construction. To make this possible, and the prominent center of the lower module. As a self-
the exterior layer needs to be based on modules with imposed rule of design, the offset has to be large enough
a larger radius or apothem. Several other factors that to allow for two modules to be superposed while being
need to be taken into account are discussed below. mutually rotated against each other by half a module.
112 Active bending

Fig. 7 Fig. 8
7.3 Maximum offset 7.4 Configuration principle of vertically adjacent

In principle, there is no upper limit for the offset modules
distance between two layers. One could even argue that The character of the layers and their alignment
a greater distance is desirable as it results in a higher relative to one another is defined with regard to their
structure overall. On the other hand, the combined height constituent modules. Five basic principles of alignment
of the two layers is only effective if they work as an have been established and form the basis for the timber-
entity, which in turn depends on the connections between fabric configurations subsequently developed. The align-
them. Among other considerations, these connections ment principles are conceived in order to provide for con-
have to be moment-resisting, in order to prevent relative nections between the low- and high-strength zones of the
displacement of the layers in the span direction in the modules.
case of horizontal loads, as such a displacement lowers
the overall rigidity of the structural system. Greater dis- 8 Timberfabric demonstrators
tance leads to greater leverage and hence also increases
the demands on the connections’ resistance. Furthermore, As part of the research, various configuration studies of
increasing the distance also has a visual impact: it de- three-directional timberfabric have been carried out, two
creases the visual coherence of the two layers, which to of which are discussed below.
a certain degree can be understood as analogous to the In this configuration, the layers are mutually
structural logic. rotated around O and shifted in the x-direction. The inte-
rior layer is built up of a series of identical arch-shaped
113
units of higher order that are placed next to each other. the connectors’ cross-sections. The interstice between the
The center-to-center distance of these units in the x-di- two layers and the constellation of their vertically adja-
rection corresponds to twice the width of the modules’ cent modules make a dual connector approach more ap-
constituent elements. Each of these arch-shaped units propriate than an integrated one. Two different types of
consists of three entire modules. The exterior layer is connectors are used for longitudinal and lateral connec-
composed of a series of units similar to those of the interi- tions. The overall configuration can be varied without
or layer. The units of the exterior layer have an overall being constrained by the connectors.
length of three modules. Unlike the units of the interior The approach taken in this study of twisted
layer, they consist of two plain and two halved modules, superposition is to use the same pattern on both layers.
where the plain are located in the center of the unit The overall construction is relatively simple. Each layer
and the halved are located towards the support points. 5 consists of a sequence of independent “interbraided” arch-
The module geometry and the distance between the one- es. The arches of the interior layer equate to a length
directional assemblies together determine the shape of of three timberfabric modules and sit perpendicular to
the edges of the structure. The exterior layer is also
composed of a sequence of “braided” arches, again with
Large-scale implementation of three-directional
Fig. 9
timberfabric, configuration study three, view from above
an arch-length of three textile modules. However, this
layer’s arches are built up quite differently from the
In the longitudinal direction, the modules are
Fig. 10
connected by disc-shaped elements. interior modules. Each arch starts with a half timber-
fabric module, followed by two complete ones, and ends
Fig. 11 In the lateral direction, the modules are connected by
two variants of custom-made connectors: their shape results with a half module again. This approach results in a shift-
from the interstice between the interior and the exterior layer.
Fig. 12 Due to their customized geometry, these elements
can likewise establish the connection between the two layers.
Fig. 9 Fig. 10
114 Active bending

ed pattern on the exterior layer and allows the connec- – Configuration principle
tion of weaker and stronger zones between layers rather Mutual rotation of superposed units around O
than merely within each layer, thus increasing overall and around a central vertical axis.
structural performance. The major difference in compari-
son to the interior layer is that the exterior layer’s arches – Construction of interior layer
are not perpendicular to the edge. The solution proposed The interior layer consists of a series of self-
by this study is similar to the structure that was devel- contained higher-order units of one type, which
oped in the first double-layered prototype. This approach units in turn consist of three entire modules.
succeeds in improving the connectivity of the separate
arches by the simple means of rotating the span direction – Construction of exterior layer
of the exterior layer relative to the interior layer. In the The exterior layer is made up of a series of
original, double-layered prototype, one arch of one layer self-contained higher-order units of one type.
was connected to two arches of the other layer, while in Here, the unit consists of two entire and two
the solution employing rotation, one arch of the exterior halved modules.
layer could be connected to five arches of the interior
layer. In this way, an improvement in connectivity and – Relation of the two layers and their constituent
continuity is achieved. As in the other examples, with modules to one another
the exception of the interlaced-layers solution, the se- The two layers and their constituting units are
quence of the layers is adaptable. This configuration can rotated around each other around a central,
be summarized as follows: vertical axis.
Fig. 11 Fig. 12
115
8.1 Impact of layer configuration on connector
elements
Due to the mutual rotation of the interior and
the exterior layer, the cross-sectional profile of their
interstice limits the use of multi-directional connectors
at certain locations. Here, multiple one-directional con-
nectors are employed.
9 Timberfabric structures with alternative

global geometries
The timberfabric configurations developed thus far share

a fairly simple global geometry. In part, this is due to
the construction of the one-directional configurations,
for which until now—with one exception—only identical
modules have been used. As a consequence, the configu-
rations developed all have an arch-shaped profile with
constant curvature. The two- and three-directional con-
figuration studies are essentially linear arrays of such one-
directional units. Logically, their global cross-sectional
profiles display identical geometrical properties.
In order to obtain differently shaped timber-
fabric structures, several strategies can be imagined.
For example, cross-sectional profiles with non-constant
curvature could be achieved by using modules of dif-
ferent sizes. Another approach could be applied in multi-
directional configurations that are based on independent,
Fig. 13 Different types of connector elements for
longitudinal, lateral, and multi-directional connections one-directional units. By arranging these units along a
curved instead of a straight trajectory, double-curved
structures could be created. Furthermore, one could grad-
ually increase the size of the modules in the respective
units, which in turn would increase their span and height.
This would result in an overall geometry that can be
described as part of a cone. A further approach towards
alternative global shapes could be the use of modules that
are made of specifically formed, custom-made panels.
References
1Hudert, M. “Timberfabric: Applying Textile Assembly
Principles for Wood Construction in Architecture.”
Thesis no. 5553, EPFL, Lausanne, 2013.
2 Seiler-Baldinger, A. Textiles: A Classification of Techniques.
Bathurst: Crawford House Press, 1994.
3 Beech, S.R., C.A. Farnfield, P. Whorton, and J.A. Wilkins, eds.
Textile Terms and Definitions. Manchester: The Textile Institute,
1986.
4 In this context, the term “geometry” refers to the shape
of the cross-section.
5 In order to maintain the geometry of a bisected module,
a triangular spacer piece is placed at the intersecting line
to keep the module’s deformed elements in position.
116 Active bending

117
3. 4
“Shaping by bending is very simple”
Interview between Jan Knippers (Institute of Building Structures and Structural Design,
ITKE, Stuttgart) and Yves Weinand
October 12, 2015
Yves Weinand: Right from the outset at IBOIS, we from ours, as we work with the principle of predicta-
have dealt with fractals, as well as with folded bility based on simulated calculability. I think it's inter-
systems. The latter, however, could not really be esting to analyze natural structures in order to gain
implemented. During the 1970s, there were awareness of one’s own approach to design and construc-
attempts to develop folded structures. I truly tion rather than to make a direct reference. It isn’t our
believe in the potential of folded systems. goal to build a large sea urchin; rather, we strive to
Our goal as engineers is to implement these expand the variety of possibilities and to validate our
types of structures. Architects are rarely able to, approach.
as they have to create 3-D drawings for their
architectural competitions that only super- Y. W.: In publications from the Stuttgart scene,
ficially represent structural optimization or it struck me that bionics is used as an argu-
topological considerations. ment for the generation of form. It wasn’t previ-
In the research represented in this publication, ously clear to me whether nature should be
structures, for example derived from fractal directly copied. You’ve just explained how you
geometries, will be analyzed. Here, we draw on prefer to distance yourself from what you see
subjectively selected iteration forms with sys- in nature and to use what you observe as an
tems that directly imitate existing configura- educational tool, thereby creating more space for
tions found in nature. We adopt a similar reflection in which to define what might be
approach when developing woven systems. constructible.
Although woven systems can be found in nature,
we avoid copying them directly and rather try J. K.: To create my own design approach, to construct, and
to interpret these geometries or systems. Direct- to reflect, I look at structures that are completely
ly copying or interpreting nature are issues different from my own, as a biological tool, so to speak.
that you analyze at your institute. You referred So I challenge my approach to the principle of design
to the structural principle of sea urchins that as an engineer on a meta-level. It’s also interesting to
you have adopted. What do you mean by this? find a connection to woven structures. The theme of
shaping by bending is actually very simple. It is the most
Jan Knippers: Of course, we aren’t trying to directly basic approach to the creation of nonplanar geometries.
translate a biological principle, like the sea urchin, into It was nevertheless never utilized, as one couldn’t calcu-
a technical process, though we do see the benefit in late it. Only now do we have the methods of calculation
testing our approach to design in relation to what we and the simulation tools to calculate large deformations.
see in nature. We can only design what we are able to These tools have only recently become accessible to
simulate, as we always invent something that does the average engineering practice, so this approach has
not yet exist. We have to predict the structural capacity, only now become part of my design process, because
stability, and integrity. And we are only able to do this we are able to calculate it.
if we are also able to calculate and simulate things.
This is our “design space,” so to speak. Basically, it works Y. W.: We proceeded from basics. We bent two
exactly as it already did back in the nineteenth century, panels, but not in any relation to nature. Our
only now it is far more complex and much larger. approach is intuitive; it is an inductive method.
And that is why I believe that it is interesting to work I am particularly interested in discussing appli-
with biological forms, as the criterion of predictability cations of this method—that is, how it might
is totally irrelevant. It all develops in a process of actually be possible, in addition to bending these
self-organization and identification with a few critical panels, to create more efficient structures that
elements. This is a completely different approach are also far more flexible than structures usual-
118 Active bending

ly proposed by engineers. What advantages do becomes more stable, even against wind loads. Here we
you see in these multilayered woven or twisted used the principle of bending to respond to different
systems that you first presented in 2010 at geometric configurations. Additionally, the system be-
the first ICD/ITKE pavilion in Stuttgart? What comes more rigid due to the bending.
advantages do you see in terms of structure?
Y. W.: That’s a good application.
J. K.: We didn’t approach this theme from a structural
standpoint, as you might have expected, but rather from J. K.: Unfortunately, though, none are made from wood.
two different directions. One aspect was related to tex- The most interesting aspect of active elements, however,
tiles, and to the benefits of accessing the potential ener- is that different geometric configurations are accounted
gy in the bias of the textiles. Almost ten years ago, we for. Members are attached to the upper side and act
constructed a simple structure made of fiberglass slats like a compound spring and span the membranes.
and we stretched fabric over it. The other aspect was to The elements can be bent and can thus track the sun.
investigate the use of different geometric configurations The elastic energy is used to pre-stress the textile strips.
in mobile structures. The bending allows one to shift
between different geometric scenarios without altering Y. W.: Can you actually develop support systems
the system. This aspect of relativity, the adaptability of from these support structures, in the traditional
the geometry, and the connection with the membranes— sense of load transfer? Do you have an idea of
these were basically our first forays into the area of the kind of deformation in the upcoming 2016
bending. From this, we developed the idea of applying version of the ICD/ITKE pavilion?
bending, thereby creating geometrically complex config-
urations, such as the ICD/ITKE pavilion from 2010. J. K.: I don’t know that yet. We use bending to create the
geometry of the modules, and we use the adjusted
Y. W.: From a strictly engineering standpoint, rigidity—that is, where the bend is most prominent
what are the structural benefits of these actively and the element at its thinnest—to obtain the least
bent systems? If one does not consider the pro- resistance. The forces are dissipated in the shell, which
cess but rather the result, I can well imagine consists of several layers woven together. We basically
that one could obtain an extremely flexible use many individual boards made up of composite
structure that can bend strongly. Thus you can modules. In principle, our current system is fairly
deal with large deformations. Light structures similar to the previous one, only we no longer assemble
such as these, for example in earthquake zones, each panel from a planar element that is bent. So we
would not even absorb the horizontal energy, as use bending as a way of modeling.
a reinforced concrete structure would. Aren’t
there also formal benefits here too? Y. W.: Woven structures can also be superimposed,
that is, double-layered and offset spatially.
J. K.: In the ICD/ITKE pavilion, we found that by introduc- That is how to obtain the greatest rigidity in the
ing bending stress, the structural capacity increases center of the element, where the structural
and, with it, the tensile strength and system rigidity. height is at its maximum, and you can place this
We utilized this effect in other projects, such as the zone above the weak points of the next layer.
synthetic facade for the Expo pavilion in Yeosu, South You overlay the weak points with strong points
Korea, in 2012. It consists of four strips that are bent in several layers. It is fascinating when a third
upward, thus creating the climatic structure. Due to the weave direction is required. That means that we
upward bending of a strip, a bending stress is produced. first weave twice in a principal direction, similar
Since the strip is only 8 mm thick, it stiffens and to knitting. Then you continue along the third
119
weave direction. This third weave direction would like to work with wood. That is, they
is what defines both the constraints and the would like their projects to be visibly environ-
support conditions of the individual panels. mentally friendly. It is now common knowledge
It would be fascinating to be able to weave on a that steel and reinforced concrete may not be
large scale and to be able to connect the entire as sustainable as timber, even though they are
structure using robotics. great materials. Only recently a contractor
approached me with a request to build a hall
J. K.: I have always wondered how structures deal with using a new system.
local pressure at the contact points. Are there any prob-
lems with pressure across the grain? J. K.: After the timber shells of the ICD/ITKE Research
Pavilion 2011, in 2014 we were able to build the
Y. W.: As a rule, we haven’t had any problems— exhibition hall of the National Horticultural Show in
except when we were working with larger Kerto Schwabisch Gmund by Achim Menges. This development
panels that already broke at 18 N/mm2 bending. allowed us to put our ideas into practice. There are now
They really shouldn’t have failed, but they did, a series of projects on the horizon where this technique
due to the fact that there were coincidentally too will probably be used. But in other cases this step has
many random knots concentrated in one place. proven extremely difficult. For example, our office
We developed a small parametric system in has tried a few times to realize timber bridges, but this
order to be able to select different geometric has proved futile.
parameters, so that changes could be made to
the overall geometry as required. But this has Y. W.: I’d like to return to the material itself.
remained a prototype. You've just shown that you yourselves laminate
here. But for the exhibition hall in Schwabisch
J. K.: We have noticed this even in the simplest office Gmund you decided to use Hess panels. Today
building, where we frequently propose timber structures. there are also Pollmeier panels, which are very
But then at some point in the planning process the use of good. The company, Pollmeier, has built an
timber is rejected, for either structural reasons, thermal impressive new facility. At last there is an alter-
mass, or budget. Although we work with ambitious archi- native to Kerto. Kerto was first introduced to
tects and contractors and start by specifying a straight- Germany in 1995, and we built a small library
forward timber structure, at some point a decision is with Kerto panels at the RWTH in Aachen. As a
made against the use of timber. It is very difficult to find civil engineer, what requirements would you
a project where the conditions enable an innovative step. give the manufacturers of the panels today?
I try to incorporate research from the university into our
practice, for example the composite structures used in J. K.: We’ve built a whole series of projects using cross-
the pavilion. But it is difficult to find the right constella- laminated timber panels. At the IBA Softhouse in
tion of architect, client, and design brief. Hamburg, we tried to work with wooden ceilings in a bid
to become more independent from the manufacturers.
Y. W.: Basically, I agree with you. It has also been The argument we used there is that laminated timber
my experience that many timber projects are construction can be calculated using normal design
not realized, even though the chances of standards, that is to say, independent of specific authori-
implementation initially looked promising. zations, and thus is feasible for smaller companies.
But perhaps something is currently happening. That’s why we try to promote this construction method
I have the impression that many contractors despite its structural limitations.
are strongly influenced by their clients, who
120 Active bending

Y. W.: Shouldn’t composite construction methods
be encouraged in timber construction? For ex-
ample, floors built from a combination of con-
crete and timber that provide an acoustically
better result at a lower price? The timber indus-
try often promotes a purely wood-based doc-
trine—in reaction to the other industries. But
don’t we need a stronger relationship to the
existing construction industry? Perhaps some-
thing is happening with general contractors who
increasingly integrate timber into their
schemes.
J. K.: There is a strong lobby for timber construction in

Britain, as well as here in Baden-Württemberg, in the
Black Forest, which has been replanted after large sec-
tions of pine forest were destroyed by a number of
storms. There is a political lobby that strongly supports
timber, but it has not been able to translate this support
into concrete regulations, which would, in turn, result in
the re-evaluation of construction methods. We have had
discussions in Freiburg about where a new football
stadium ought to be built. There is a vague desire to use
timber for the stadium, but when it comes to mainte-
nance, investment, and construction, the use of timber is
immediately called into question by the contractors and
stadium operators. Thus, there is a discrepancy between
the general attitude on a meta-level and the legal re-
quirements that a building project has to meet. What is
missing is an intermediary layer, which would provide
the contractors with the appropriate support.
Y. W.: Timber is a good material and I hope that

we’ll soon be ready to propose larger wooden
structures, such as stadiums. The logistics exist
for the use of timber in housing, but there are
no logistics for contractors that would allow for
innovative or hybrid structures made from
wood. Classical structures with laminated beams
are proposed time and again. Perhaps not
enough has yet happened to allow new struc-
tures to develop.
121
3. 5
Geodesic lines for shell forms—a playground

installation at the Vallée de la Jeunesse
(Lausanne, Switzerland)
Marielle Savoyat
Design Labor für

IBOIS, Laboratory
Holzkonstruktionen
for Timber Constructions/
(IBOIS) an der This architectural structure in curved wood is based on
Eidgenössischen
EPFL, Swiss Federal
Technischen
Institute of
Hochschule
Technology, the geometry of geodesic lines. The idea was developed in
Lausanne
Lausanne,(EPFL),
Switzerland;
Schweiz,Prof.
Prof.
YvesYves
Weinand
Weinand conjunction with Claudio Pirazzi’s doctoral dissertation,1
und
and Dr. Claudio Pirazzi,
Pirazzi (Wissenschaftler)
researcher and undertaken at IBOIS, the Laboratory for Timber
Lehrstuhl
GEOM, Laboratory
für Geometrie
of Geometry/EPFL,
(GEOM) an der Constructions, along with Roland Rozsnyo, 2 at GEOM, the
Swiss
Eidgenössischen
Federal Institute
Technischen
of Technology,
Hochschule Laboratory of Geometry, both at EPFL, the Swiss Federal
Lausanne
Lausanne,(EPFL),
Switzerland;
Schweiz,Prof.
Prof.
Peter
Peter
Buser
Buser Institute of Technology in Lausanne. A geodesic line rep-
und
and Roland Rozsnyo,
Rozsnyo (Wissenschaftler)
researcher resents the shortest line between two points on a curved
surface, or lines made up of simple curves. The curvature
Project Charpentes VIAL SA
execution achieves the weak axis of inertia of each rectangular
2006 section and not the strong axis of inertia. However, the
Completion 2006
geodesic line follows a helicoidal line, where some torsion
Location Vallée de la Jeunesse, Lausanne, Switzerland
Schweiz is also experienced.
In the context of this research project and under
the auspices of the two laboratories, the researchers
developed a program capable of determining the geodesic
networks on free-form shapes These networks, in turn,
enabled the definition of geometric data necessary for a
digital cut. In this way, new options were found in terms
of the use of timber for the creation of curved structures.
Fig. 1 Modelization of the pavilion

View of the pavilion in its
Fig. 2
Kindergarden situation
Fig. 1
122 Active bending

Fig. 2
A large-scale prototype of a ribbed shell made of laminated, The resulting prototype structure from this research,
screwed boards was created to test the potential of this called “Géoline,” was installed in the garden of a day-care
program. The chords of the shell followed geodesic lines, center in Lausanne, where it now serves as a playground
avoiding all double curvatures and reducing the impor- installation and playhouse for children at the Vallée de
tant constraints initially posed by curvatures, such as la Jeunesse.
bending at the fortified axis. The slats are therefore solely
utilized for a unique moment of flexion, based on their
weakest axis. The meshing of optimal wooden ribbed References
structures can be achieved, thereby reducing the risk of 1 Pirazzi, C. “Zur Berechnung von Holzschalen in Brett-
buckling. rippenbauweise mit elastischem Verbundquerschnitt.”
It is interesting to note that different dispositions of Thesis no. 3229, EPFL, Lausanne, 2005.
geodesic lines on the same free-form surfaces are possible, 2 Rozsnyo, R. “Optimal Control of Geodesics in Riemannian
and that a subjective element can appear in the architec- Manifolds.” Thesis no. 3481, EPFL, Lausanne, 2007.
tural design, notably regarding questions of aesthetics,
construction, or spatiality. This paves the way to a wealth of
new potential in the creation of contemporary architecture.
123
4 Form-finding and mechanical investigations
of active bended systems
4. 1 Optimization of double-curved shell structures 126

Yves Weinand
4. 2 Experimental and numerical study of the structural

behavior of a single timber textile module 134
Masoud Sistaninia, Markus Hudert, Laurent Humbert,
and Yves Weinand
4. 3 Generation process and analysis of innovative

timberfabric vaults 148
Etienne Albenque, Markus Hudert, Laurent Humbert,
and Yves Weinand
4. 4 Mechanical form-finding of the timber fabric

structures with dynamic relaxation method 162
Seyed Sina Nabaei, Olivier Baverel, and Yves Weinand
4. 5 „These programs don’t provide easy,

off-the-shelf solutions” 178
Interview between Olivier Baverel and Yves Weinand
4. 6 Braided structures: applying textile principles

at an architectural scale 184
Marielle Savoyat
125
4. 1
Optimization of double-curved
shell structures
Yves Weinand
The scope of research presented here the question of new global forms. Archi-
concerns the interface between architects’ tools are highly representational;
tecture and structural engineering. therefore, the presentation of space is
The simultaneous openness to both in- greatly simplified. Engineers’ tools have
ductive and deductive methods is an a high level of mechanical precision,
important aspect of the manner in which but their spatial representation is less
research is conducted at IBOIS at the EPFL. user-friendly. In other words: even if tools
The functional diagram above shows are increasingly better able to represent
that both empirical and rational tools are space, the process of mechanical verifica-
used for brainstorming; physical experi- tion remains incremental.
ments, digital models, and mathematical
and analytical descriptions are all used In this context, the work of Sina Nabaei
simultaneously to solve problems. presented here aims to develop a simula-
Initial ideas are described in principle tion model for shells and support struc-
by one or another of these methods, and tures, enabling the creation of simulations
jumps in scale are deliberately staged, during the structural design process
where both structural and architectural that are based on physical characteristics.
considerations come into play. The
synthesis is achieved in the context of Nabaei also presents a robust and stable
a structural optimization process. simulation method for calculating
woven timber structures. While the work
Why do hybrid structures no longer ap- presented by Markus Hudert has been
pear in civil engineering textbooks? empirically developed “by hand,” the tool
Why is there no theory supporting the developed by Nabaei uses mechanical
superimposition of different structural constraints to control the form-giving
systems in a single structure? One of process.
the aims of this publication is to stimu-
late the interest of civil engineers in
126 Form-finding and mechanical investigations of active bended systems

Investigation of timber fabrics
Textile structure – documentation
Woven / fabrics –
physical explorations 3D geometry
Digital modeling
Mathematical descriptions
Applicability
Dimensioning real structures
structural engineering Interactivity Ease of use
Architectural and structural
Structural analysis applications
Scaling
Synthesis
Bring it together
Mechanical
Optimization processes precision
Fig. 1 Fig. 2
Fig. 3
Mutual Position in node n α is {Ω(k)

0
> Ω(i)0 > Ω(j)0 }
Functional diagram,
Fig. 1
IBOIS Research Cluster
Diagram illustrating the
Fig. 2
interaction of existing tools
Fig. 3 Textile module formed
by interweaving
Individual strands
Fig. 4 a and b
are formed mechanically
Sint : 6nα(j) dSint ; a mutual position order of solids is set
Sext : 6nα(i) dSext ; x(nα(i)) = (x0) α(i) overall displacement is imposed over Sext
and a mutual position order of solids can also be set.
Fig. 4 a
Fig. 4 b
127
Parallel offset distance = T/2
Final cutting pattern
Developed circular surface S‘
Fig. 5
Fig. 6 c
cross-section profile
corrugation profile
Fig. 6 a Fig. 6 b
Fig. 5 Geometric structure and detail: laying the

edge geometry flat, multi-layered structure,
determination of the central plane
Fig. 6 a) A folded form made from actively curved
panels, b) a box girder made from actively curved
panels, c) Detail of box girder in b)
Fig. 7 Examples of actively curved shell structures
Fig. 7

As the previous examples show, it is clear Even with a constant panel thickness, this
that form and structure can exist in construction demonstrates a tremendous
a close relationship to one another. Two bending stiffness in the corners of
other examples presented here also illus- the frame. In analyzing the causes of this
trate the potential relationship between phenomenon, it was ascertained that
form and structure in order to optimize bending stiffness is ensured by the exist-
structural behavior. The origami design ing structural strength. Because the
tool presented in Chapter 1 makes it surfaces are rotated in relation to one
possible to generate folding processes. another, they cannot rotate around an
Originally, the flat panels created the axis. The bending moments that are
necessary preconditions. During further generally absorbed in the corners of the
experiments, it was found that transverse structural frames are not activated,
and longitudinal profiles are also able to as the tension from the axis of rotation
generate structures with curved surfaces. is deflected by the curved surfaces.
A folded structure formed with curved
panels and an active box girder made A 1 : 5-scale prototype was built with multi-
from composite curved surfaces are layered panels. For this purpose, a form
shown here. was built that corresponded to the chosen
curvature. In order to achieve this,
The ability to construct with panels that several layers of panels are forced into
curve in opposite directions resulted in a the curved shape and glued in place. Thus,
new pavilion project, which was commis- it is not about actively bent panels, but
sioned by the Accademia di Architettura about timber panels that are produced in
in Mendrisio in 2013. curved forms.
Initially, the principle is studied with re- An important aspect of the project
gard to convex and concave panels that development is the formation of timber–
are connected in two opposing directions. timber connections along the edges.
Civil engineering master’s students at The inclination of the panels in relation to
the EPFL were asked to critically observe one another affects the nature of the
the given geometry and to optimize timber connection. Within the framework
the structure using only geometric ad- of this prototype, two different inter-
justments. For this purpose, parameter locking geometries were selected, para-
studies were undertaken where not only meterized, and CNC-milled.
span widths but also, for example, the
depths of the indentations were varied.
The educational aim was to direct
engineers toward a consideration of the
overall form of structures.
129
Fig. 8 Potential applications of the main structural system
L : beam length
S : span
j : reflection angle
w2
i =1.1 i = 1.1 i =1.1
q=1 q = 1/p q =1
p =17/3 p = 17/3 p ≈1.609
j = 90° j = 90° j = 90°
k = 0.05 j h2 k = 0.05 h0 k = 0.05
Fig. 9 Study of the support parameters

Fig. 10
Fig. 10 Test equipment for measuring

deformations
Fig. 11 Test girders made at IBOIS
Fig. 12 Test results of the three volumes
Geometric parameters of the
Fig. 13
pavilion (plywood model)
Fig. 11
F/ 2 F/ 2
4m 8m
Prototype deflection* max load deflection** max load
(mm) (kN) (mm) (kN)
16.3 9.6 14.3

15.9 9.1
16.7 9.0
17.7 8.2 15.0 >37.2

16.5 10.4 15.4 >35.9
17.1 9.5 15.6 >35.6
23.9 7.9 20.3

23.6 8.1
20.8 8.1
* F = 2 kN ** F = 4 kN Fig. 13
Fig. 12
131
Fig. 14 a
Fig. 15
Fig. 14 b Fig. 14 c
Production of curved timber panels:

Fig. 14 a–f
parametric geometry of the timber–timber connection
Fig. 15 Parameter studies on a singular basic geometry
Fig. 14 d Fig. 14 e
Fig. 14 f

On this basis, a pavilion with a span of as the panels are laminated. The 77-mm-
about 13 meters was developed. Convex thick panels are produced under a pneu-
and concave panels exhibit the same matic pressure pad. Subsequently, the
curvature. This measure simplifies the panels are cut along their curved contour.
manufacture of panels, as there is only a The geometry of the timber–timber
single form. The finite element model connection can also be milled in the
shows the panel thickness and the degree same machine path. The resulting time
of deformation of the structure. The timber– savings are relevant in the development
timber connection could be fully modeled of this typology of structure.
Fig. 16 Assembled pavilion
133
4. 2
Experimental and numerical study

of the structural behavior
of a single timber textile module
Masoud Sistaninia, Markus Hudert, Laurent Humbert, and Yves Weinand
This
In diesem
essay Beitrag
investigates
wird an eine
advanced
moderneclass Art of
dertimber
Holz- 1 Introduction
structure
konstruktion termed
untersucht,
timberfabric,
die das which
Potenzial
has potential
für An-
applications
wendungen als in roofing,
Überdachung,facade,Fassadenelement
and bridge construc- und Timber is a versatile construction material that is abun-
Brückenkonstruktion
tion. The developmentbesitzt. of timberfabric
Diese sogenannten
structures dant in many regions of the world. Moreover, it is a re-
originates
Holzgewebefrom basieren
the aufapproach
der Idee,of die
harnessing
strukturellen the newable resource that can be processed and assembled
structural,
und modularen modular,
Eigenschaften
and qualityvon of Geweben
textiles infür timber
den in energy-efficient ways. Recent studies1, 2 indicate that
construction. 9
Holzbau zu nutzen. 9
Timberfabric
Holzgewebetragwerke
structures comprise werden re- the use of timber as construction material results in
aus
petitive
wiederholt
arrangements
angeordneten
of one or konstruktiven
more structural Elemen-
unit buildings with a better environmental performance than
cells,
ten, dencalled
sogenannten
textile modules.
„textilen
When
Modulen“,
properly zusammen-
designed, conventional materials. With regard to present-day con-
gesetzt.
one obtains Bei sorgfältiger
a modular and Planung
lightweight
entstehtstructure
ein mo- cerns over globally increasing energy consumption and
dulares
with interesting
leichtes Tragwerk
and unusual vongeometric
hoher konand strukstruc-
tiver simultaneously decreasing resources, wood holds a dis-
Qualität
tural qualities.
und mit einer ungewöhnlichen Geometrie. tinct advantage over other construction materials such
This
Der paper focuses soll
Schwerpunkt on the single
hier auf timber textile
ein einzelnes as concrete or steel. This, in turn, should increase the
module.aus
Modul Based on the finite
Holzgewebe element
gelegt werden. (FE)Esmethod,
wird ein a interest of the research community in expanding the
reliable procedure
verlässliches Verfahren is auf
proposed
Basis derforFinite-Elemente-
modeling the range of applications of timber structures.
overall
Methode assembly
(FE-Methode)process of the textile
vorgestellt, mit dem module. In
sich der Examples of modern but well-established timber
order to compare,
Montageprozess des textile module
textilen Moduls prototypes
modellieren are architectural forms include folded plate structures, 3, 4
constructed at both large and werden
lässt. Zu Vergleichszwecken intermediate scales
Prototypen lattice structures (e.g. timber lattice roof for the Mann-
with
textilerdifferent
Moduleassembly
im großen conditions.
und mittleren The proposed
Maßstab heim Federal Garden Show), 5, 6 and multi-reciprocal frame
geometrical,
angefertigt. Dank nonlinear FE model allows
des geometrischen, the eva-
nichtlinearen structures.7, 8 Forms such as these present clear advan-
luation of the
FE-Modells stresses
lassen thatSpannungszustände
sich die are induced during aus- the tages over more traditional, flat-surfaced roofing struc-
construction
werten, die währendprocess des and Konstruktionsprozesses
which may affect the tures, in that the efficiency and load-carrying capacity
structural
ein gebrachtintegrity
werdenofund the die
module. In particular,
strukturelle the
Integrität of the structure is increased and its weight reduced.
risk Moduls
des of failure during assembly
beeinträchtigen is identified
könnten. Mit demusing Tsai- A new type of timber structure, called timber-
the anisotropic
Hill-Ver Tsai-Hillfürcriterion.
sagenskriterium anisotropeTheMaterialien
structural fabric, has recently been developed at IBOIS, 9, 10 the
behavior
wird of the timber
insbesondere textile module iswährend
das Versagensrisiko then inves-der particular structural properties of which emanate from
tigated through
Montage ermittelt.bending tests using des
Das Tragverhalten theHolzgewe-
construc- the principle of weaving techniques. The development of
ted prototypes.
bemoduls During thedurch
wird außerdem loading procedure, the
Biegeprüfungen an timberfabric has been driven by the concept of incorpo-
vertical
den deflections
Prototypen are measured
untersucht. at different
Die vertikale loca-
Verschie- rating qualities specific to textiles, such as modularity
tions on the prototype
bung während surface
der Belastung by auf
wird meansder of external
Oberfläche and the mutual support of constituent elements of textile
displacement
des Prototyps transducers.
an mehreren Using Stellenthe FE model,
mittels the
externer fabrics into timber construction. Timberfabric structures
corresponding
Wegaufnehmer deformed gemessen.shapes Mit dem areFE-Modell
simulatedwer- by have a broad potential for architectural applications
applying
den the bending loads on
die dementsprechenden the pre-stressed
Verformungen durch thanks to their versatility, adaptability, and the qualities
textile module.
Aufbringung Experimental displacements
der Biegebelastung auf die vorgespann- and directly linked to their structural make-up. They are
FE
ten predictions
textilen Module are simuliert.
compared Am to Ende
see ifwird theyvergli-
cor- based on the repetition of a structural unit cell, the
respond
chen, ob favorably.
die experimentell gemessenen und die aus textile module (Fig. 1a) which results from applying textile
der FE-Modellierung erhaltenen Verschiebungswerte assembly principles to timber components. Fig. 1b shows
übereinstimmen. the double-layered timberfabric structure, representing
Keywords timber textile module, construction a single example of the many possible configurations of
stresses,Holzgewebemodul,
experimental method, biegeabhängige textile modules.
Eigenspannungen,
finite element modeling. experimentelle The single textile module presented in fig. 1 a
Methode, Finite-Elemente-Modellierung provides a structural shape of particular interest for

Fig. 1 Model of the structure
this study. Briefly, it consists of two mutually supporting 1.
thin panels that become curved during the assembly

process, as illustrated in fig. 2. Consequently, construc-
tion (or residual) stresses are generated during the fabri-
cation of the module. Their amplitude typically depends
on the constitutive material, the size of panels, and on
the assembly conditions. The use of poor quality material
or inappropriately dimensioned panels, or a combination
of both, may even cause premature failure of the textile 2.
module during the assembly process. The construction

stresses can be evaluated by means of a finite element
(FE) model that takes the different fabrication stages into
account (Fig. 2).
This paper focuses on the fabrication process
and structural behavior of a representative single textile
module in a bending-load configuration. The proposed
approach is both experimental and numerical. It involves 3.
the fabrication of two prototypes at two different scales
(intermediate- and large-scale) under different assembly
conditions, as discussed in the first part of section two.
The second part of section two is devoted to the bending
test setup and the measurement equipment required.
Despite the numerous finite element (FE) models
already available for braided textile composites,11–15 the
numerical study of the timber textile module requires
special attention, as its analysis is complicated by the Fig. 2 Design principle for the textile module
135
a)
the textile modules are respectively supplied by the
0° companies Schilliger Holz AG (Switzerland) and Thebault
90°
0°
90°
(France). As schematically depicted in fig. 3, the panels
0° 0° consist of symmetric orthotropic laminates, with a three-
ply [0/90/0] GFP and four-ply [0/90]s Okoumé with a thick-
Middle surface
ness of 33.0 mm and 6.3 mm respectively. As is customary,
the plywood is produced from rotary cut veneers, bonded
b) with an adhesive (synthetic) resin under high-pressure
0° conditions. In each case, the outer layers of the veneers
0°
90° 90° are of equal thickness, with the same grain direction
0° 90°
0°
along the longitudinal axis (L-axis) of the laminate. As
depicted in fig. 3 a, for the three-ply configuration, the
Fig. 3
Middle surface axis of symmetry passes through the center of the 13-mm-
thick core ply, and its grain is directed along the trans-
verse axis (T-axis) of the laminate. For the four-ply model,
the two core plies of equal thickness (i.e. 2 mm) are glued
together with their grain direction running perpendic-
particular geometry and assembly conditions encoun- ular to the longitudinal axes of the face veneers (along
tered. Because of the large deflections and rotations the T-axis). This glued interface corresponds to the plane
ex perienced by the module during the fabrication stages of symmetry (central surface) of the plywood. This even-
(Fig. 2) , a geometrically non linear FE model that aims layered arrangement has been proven to increase effi-
to accurately reproduce the geometrical shape of the tex- ciency in veneer manufacturing and grading (less varia-
tile module is developed (section 3). It is anticipated that bility), although it does increase the cost of production.
this model should permit a representative evaluation Both the GFP and Okoumé panels are composed
of the construction stresses involved in fabrication. In of orthotropic layers (veneers) with their principal mate-
section four, vertical displacements measured at several rial axes coinciding with the longitudinal (L) and trans-
locations of the prototype surface during the bending verse (T) geometric directions of the panels (Fig. 3). On a
tests are compared to the finite element predictions. macroscopic scale, they can be treated as homogeneous
Finally, the structural behavior of the textile module is orthotropic materials, with the L and T axes as the prin-
discussed. cipal axes of the equivalent material.
Homogenized elastic material properties, pro-
vided by the manufacturers under standard conditions
2 Experimental investigations (20 °C temperature and 65 % relative humidity), are re -
ported in Table 1, where the L and T subscripts refer to
2.1 Materials and specimens the longitudinal and transverse directions respectively.
Two textile modules have been constructed for Table 1 also provides the longitudinal and transverse
this study: one is formed from two GFP laminated wood moduli of the Okoumé panels, measured in the envi-
panels with a length of 12.320 m and a width of 0.770 m, ronmental conditions of the laboratory. Tensile tests were
while the other is formed from two TeboPlyTM Okoumé carried out in a standard crosshead testing machine (in
plywood panels with a length of 2.34 m and a width of displacement control with a crosshead speed of 2 mm/
0.24 m. The GFP and Okoumé panels used to fabricate min) approximately eight months after receiving the
Characteristic values of strength at Modulus of elasticity Shear modulus

fifth percentile (MPa) (median values) (median values)
Plywood Bending Compression Tension In plane Tension and In plane shear
panel 2
shear Compression (MPa) (MPa)
f m,L f m,T fc,L fc,T f t,L f t,T f v,L f v,T EL ET GLT

panel 1 GFP 15.8 2.6 7.3 4.7 4.8 3.2 1.5 1.5 6667 4333 720
(3 layers)
Okoumé 35 32.4 10 28.5 6.1 17.4 7 7 2398 6852 552
(4 layers) 4592 a 7288 a
a
Averaged values obtained in the laboratory.
Table 1 Mechanical properties of GFP and Okoumé laminated wood

Fig. 4

Threaded rod
panels from the manufacturer. The average values panel 1 direction
of the elastic moduli reported in Table 1 are calculated
on the basis of measurements obtained using four sam- Threaded rod
ples with the direction of the grain of the face layers direction
panel 2
parallel to the loading direction, and two samples with
the direction of the grain of the face layers perpendicular
to the loading direction. These values will subsequently
be used in the finite element simulations.
The GFP textile module will hereafter be referred Fig. 5
to as the large-scale (TM1) specimen and the Okoumé

textile module as the intermediate-scale (TM2) speci-
men. Besides their differences in size and material prop-
erties, the modules are also assembled differently. For
prototype TM1, the ends of the panels are mechanically
linked by means of a pinned connection, while hard-
wood wedge connectors that restrain all degree of free-
dom are employed to build the second prototype, TM2.
The construction process of the two prototypes is de-
scribed in detail below.
2.2 Construction of the textile modules
2.2.1 Pin connection

A pinned connection is used for the assembly
of textile module TM1. Practically, the two panels are
connected to each other at their ends using a system of
steel-threaded rods with nuts and washers. The starting
Fig. 6 a
point for the assembly procedure is illustrated in fig. 4.
For convenience, the global reference coordinate system
(x,y,z) is introduced, such that the vertical z-axis crosses
panel 1 and panel 2 at points A1 and A2 respectively.
These two points have the same geometric position, taken
as the origin A of the global system. Strictly speaking,
point A1 belongs to the lower face of panel one, while
point A2 is on the upper face of panel 2, with the same in-
plane coordinates L = w/2, T=0 in the local frame (L,T)
of the panels (Fig. 3). The middle surfaces of the panels are
positioned parallel to the plane (x,y) and symmetrically
rotated about the z-axis by an angle α = 10°. Then the pan-
els are linked to each other at points A1 and A2 using the
steel-threaded rod (not represented in fig. 4), thus re-
straining their relative displacement at A while keeping
the rotation free along the threaded-rod axis. Note that
the threaded-rod axis initially coincides with the z-axis,
but will not retain this direction after assembly.
Fig. 6 b
Now the opposite points C1 and C2 (Fig. 4) will be
considered, with local in-plane coordinates (L = l – w/2,
T = 0), such that AC1 = AC2 = l–w = 11.55 m. The line (C1C2)
lies perpendicular to the y-axis, bisecting it at point C'. In
other words, one can trace an isosceles triangle (AC1C2),
the base of which is (C1C2). Practically speaking, the pan-
els are conjoined at points C1 and C2 by means of cables to GFP and b) Okoumé plywood layups and local coordinate
Fig. 3 a)
a third common point C on the y-axis. The distance system (L,T ).

AC is typically less than AC'. The panels that become Fig. 4 Assembly procedure and associated coordinate system.
curved are pin-connected again at C by means of a second Fig. 5 Textile module geometry with curved panels
threaded rod. During this operation, the midpoints B1 Fig. 6 a) Prototype TM1 and steel connector used to fix it
and B2 of two panel edges come into contact at point B, on the floor b) Failure during the assembly process
137
which corresponds to the maximum height of the result- clearly be described by means of six assembly parameters
ing structure (Fig. 5). The maximum height (i.e. distance (θx1, θx2 , θ y1, θ y2 , d, d') as presented in fig. 8. The origin O
BB') and span (i.e. distance AC) for TM1 are 1.50 m and of the reference coordinate system (x,y,z) is placed at the
10.05 m respectively. center of the bottom section of Wedge 1. Fig. 8 illustrates
It is worth noting that the two threaded-rod the final shape of the module, where the vertical z-axis
axes and point B belong to the vertical plane (y, z). By intersects panel 1 and panel 2 at points A1 and A2,
denoting B', the vertical projection of B onto the y-axis, the respective coordinates of which are (0, 0, d + d') and
one has an isosceles triangle ABC of an equal height to (0, 0, d'). Unlike the prototype TM1, these two points do
the line segment BB' and base AC = 2AB'. Moreover, the not occupy an identical position, because of the introduc-
two threaded-rod axes are symmetrically oriented with tion of Wedge 2. It should also be noted that the threaded-
regard to the axis BB'. rod axes keep a fixed orientation corresponding to the
The two ends of the textile module TM1 are firmly vertical z-axis during the assembly. Moreover, A1 and A2
fixed to the ground by means of a special steel connector are centroidal points for the contact areas p1 and p2, whose
(Fig. 6a) , thus restraining all displacements and rotations orientations—with respect to the x-axis and y-axis—are
along the corresponding panel edges. Various assembly given by the two angles θx1, θ y1 and θx2, θ y2 respectively.
conditions can readily be imposed on the textile module Similarly, the opposite points C1 and C2 are
panels, accordingly modifying its final shape and struc- cent roidal points for the contact areas p1' and p2' as de-
tural behavior. The selection of the type of connectors fined in fig. 8. Point C1 is now located under point C2 and
to be used depends on the assembly conditions under their respective coordinates are given by (0, s, d') and
consideration. (0, s, d + d'). Denoting O', the intersection of the vertical
Finally, it should be noted that apparently minor axis passing through these points and the y-axis, one
alterations to the assembly conditions may have a signif- obtains s = OO'. In this case, the angles – θx1 and – θ y1 (i.e.
icant influence on the strength of the structure. An early the angles – θx2 and – θ y2) orientate the plane p2' (i.e. the
prototype (TM0) was composed of panels with character- plane p1') with respect to the x-axis and y-axis.
istics similar to those of TM1 except that the width w was As was the case for prototype TM1, an isosceles
equal to 1.65 m. This prototype was broken in two after triangle OBO' can be constructed where vertex B is de-
the nuts were tightened in the final stage. Both panels fined as the common position of the contacting points
failed in two parts in the middle (Fig. 6b) due to the stresses B1 and B2. Again, B' denotes the vertical projection of the
induced during assembly. vertex B on the opposite side OO’.
Fig. 9 shows a TM2 module that is affixed to a
2.2.2 Wedge connection thick wooden base using the previously described wedge
A second type of connection, referred to simply connections. The maximum height BB' corresponds to
as a “wedge connection,” is employed for the construction 0.36 m, the span s=2.056 m and the assembly parameters
of prototype TM2. As indicated in fig. 7, two identical are chosen as θx1 = 17°, θx2 = 11.5°, θy1 = 14.5°, θy2 = –8°, d' = 76 mm,
sets of three wedge elements are used to connect the d = 47.7 mm.
panels to one another and to simultaneously affix the en-
tire structure to the ground. The different elements are 2.3 Test setups and measurement equipment
manufactured by cutting a 20-cm-long block of hardwood The test setup used for the experiments carried
with a cross-sectional area w/2 × w/4 = 12 × 6 cm2 along out on prototypes TM1 and TM2 is shown in figs. 10 a and
two oblique cutting planes with different orientations 10 b respectively. The ends of prototype TM1 are clamped
(Fig. 7) . The resulting uppermost (Wedge 3) and lowermost on the floor by means of the steel connector shown in
(Wedge 1) elements thus have one cross-section at right fig. 6. A hydraulic actuator from Walter & Bai Company
angles and the other obliquely oriented in relation to the with a maximum force capacity of ±500 kN is used to
lateral faces. Before cutting, three 10-mm-diameter holes load the prototype at mid-span. An additional V-shaped
were drilled along the length of the wooden block, as element (made of hardwood) is attached to the actuator
indicated in fig. 7, for the insertion of a tightening system for transferring the load on each panel, as indicated in
composed of threaded rods and nuts. fig. 10 a and fig. 11 a. During the test, the load is applied
Positioning the structure on the ground is up to the ultimate limit state under displacement control
achieved by means of the supporting element Wedge 1. with a constant displacement rate of 6 mm/min.
The intermediate element (Wedge 2) permits the relative Prototype TM2 and its wooden base are placed
position of the panels to be accurately controlled on on two concrete blocks, as shown in fig. 7. In this case,
the constrained areas (assembly conditions). The third a hydraulic actuator with a maximum force capacity of
wedge element (Wedge 3) is used to connect the panels to- ±300 kN is used to apply the bending load at the proto-
gether by simply tightening the nuts, thus aligning the type mid-span. For an accurate measurement of the force,
three wedges vertically. Moreover, this locking device al- as indicated in the insert of fig. 10, a load cell of 5 KN is
lows the module to be fully clamped on the ground. The added. A custom-made wooden element with two parallel
positions of the two oblique (cross-sectional) planes can steel rods, is fastened to the actuator in order to dis-

1
2·
2
1
Threaded rod axis
Fig. 9
Wedge 3
Wedge 2
Wedge 1 Actuator
V-shaped element
Fig. 7
Panel 2
LVDT
Fig. 10 a
Panel 1
Actuator
5 kN load cell LVDT
Fig. 8
Fig. 10 b
tribute the load on each panel (figs. 10 b and 11 b). Again, the Fig. 7 Textile module assembled with wedge connection
test is run under displacement control (constant rate of Fig. 8 Definition of the assembly parameters
3 mm/min) and the prototype is loaded up to 0.5 kN. Fig. 9 Textile module TM2 with wooden base (2) and wedges (1)
In both cases, the resulting deflections are re-
Fig. 10 Test setup for a) prototype TM1 and b) prototype TM2
corded at several positions along the prototype surface
using external transducers mounted on self-supporting
vertical steel rods, as shown in figs. 10a and 10b. Variable
linear differential transducers (LVDTs) with a measure- transducers are directed perpendicular to the top wedge
ment range of ± 100 mm are employed for measuring connection in order to check its deformation along the
the vertical component of the displacement for proto - horizontal y-axis during loading. As can be observed in
type TM1. Data is typically recorded at three positions the enlarged region of fig. 10 b, small circular pieces of
(1, 2, 3 in fig. 11a) and the displacement of the actuator metal are glued onto the panels to provide a flat mounting
is measured by using a LVDT mounted onto the actuator. surface in order to maintain contact with the transducers
The positions of LVDTs in the local frame (L,T) are: during the test. A grid mesh composed of 78 × 8 elements of
LVDT1 (L = 6.1875 m, T = –0.4125 m), LVDT2 (L = 8.075 m, 3 × 3 cm size (not visible in the image presented), is also
T = 0.4125 m), and LVDT3 (L = 1.25 m, T = 0.4125 m). drawn on the upper surface of each panel. This grid cor-
For prototype TM2, LVDTs with measurement responds to the numerical mesh used in the finite element
ranges of ± 30 mm and ± 50 mm are placed at seven posi- analysis. Finally, the positions of LVDTs in the local frame
tions, as indicated in fig. 11 b. Those with measurement (L,T) are given by: LVDT1 (L = 0.54 m, T = 0.06 m), LVDT2
ranges of ± 50 mm are only used to record the vertical (L = 0.84 m, T = –0.03 m), LVDT3 (L = 1.8 m, T = 0.06 m), LVDT4
displacements at points 4 and 5. For positions 6 and 7, the (L = 1.5 m, T = –0.03 m), and LVDT5 (L = 0.84 m, T = 0.03 m).
139
3 Numerical modeling 3.3 Boundary and assembly conditions
The simulation of the textile module that is
A three-dimensional, geometrically nonlinear, finite ele- pinned and built by wedge assembly requires the con-
ment analysis is proposed at this stage to simulate the sideration of the construction procedure presented in
assembly process of the textile module, including the section 2.2. The same nomenclature is repeated here.
fixation of the pinned and wedge connections. The aim is Thus, rectangular shell surfaces are used to represent the
to calculate the construction stresses (initial stresses) panels, which are positioned in the global coordinate
and to reproduce the structural behavior of the structure system (x,y,z) of ABAQUS, as indicated in section 2.2.
under bending static loads. In particular, they are symmetrically rotated by an
angle α (here α = 10°) about the global z-axis. The assem-
3.1 Material properties bly process is then simulated by applying appropriate
In this study, GFP and Okoumé plywood are mod- boundary and assembly conditions on the panels, as
eled as single-layer orthotropic linear elastic materials, detailed below for the pinned and wedge connections.
assuming that the layers are rigidly bonded together. The pinned connection (prototype TM1) has a modeling
When these simplifying assumptions are taken into ac- procedure as follows:
count, only four independent engineering constants are – The boundary conditions u x = uy = u z = 0 (ui is the
required to fully characterize the material behavior, displacement component in the i-axis direction)
namely: the longitudinal EL and transverse ET Young’s are imposed at points A1 (panel 1) and A2 (panel 2).
moduli, in-plane shear modulus GLT and Poisson’s ratio These points are located at the origin A of the frame.
νLT.16, 17 The equivalent unidirectional lamina is conse- – Only the vertical displacement u z = 0 is imposed
quently assumed to be under a state of plane stress with at points C1 and C2 (Fig. 4). Consequently, these
the principal material directions (L,T) indicated in fig. 3. points are forced to remain in the plane (x,y).
The equivalent material properties given in Table 1 are – The displacement vectors u C1 = –uC ê 1 and
used for the numerical analysis. In the calculation, the u C2 = uC ê 1 (ê 1 is the unit vector in the x-direction)
average value νLT = 0.3 will be considered for Poisson’s ra- are applied to points C1 and C2 respectively to
tio. Volumetric changes (shrinking or swelling of wood correspond with the common point C' of the
panels) due to moisture loss or gain are not taken into y-axis,
account here.
uC = (l – w) sin(α) Equ. 1
3.2 Finite element model

The commercial FE code ABAQUS is used to build where uC is equal to 2005.64 mm and 364.66 mm
a three-dimensional model of the assembly process for for TM1 and TM2 respectively.
the textile module. Because large rotations and deflec- – The midpoints B1 and B2 of the panel edges are
tions, as well as boundary nonlinearities (i.e. sudden simultaneously shifted to the contact point B
change in contact conditions between the panels) are ex- by applying the displacement vectors
pected, a geometrically nonlinear analysis is considered, u B1 = –uB ê 1 + wB ê 3 and u B2 = u B ê 1 + wB ê 3 (ê 3 is the
where the specified displacements and loads are applied unit vector in the z direction) at points B1 and
incrementally. B2 respectively. The expression of uB for this
Each panel is discretized by means of six-node, transformation can be determined as
triangular, thin-shell elements of uniform size (no refine-
ment). Specifically, fully integrated STRI65 elements with u B = (l – w) / 2 · sin(α) – w / 2 · cos(α) Equ. 2
five degrees of freedom per node (three displacements u x,
uy, u z and two rotations θx, θ y) are used. These elements, with respective numerical values of 623.66 mm
adapted for large rotation but small strain, provide accu- and 64.15 mm for TM1 and TM2.
rate solutions in the framework of the classical (Kirch-
hoff) shell theory, where the shell normal remains per- The parameter wB measured by the experimental proto-
pendicular to the shell reference surface (i.e. negligible type is imposed, modifying the y-coordinate of point C'
transverse shear).18–20 that becomes point C (figs. 4 and 5). Note that wB corresponds
An implicit time-integration procedure that re- to the distance BB' in fig. 5.
lies on the Newton-Raphson iterative scheme21 is typical- For the wedge connection (prototype TM2),
ly used to solve the simultaneous incremental, nonlinear the modeling procedure is as follows:
equations. Accordingly, an estimate of the incremental – The boundary conditions u x = uy = 0 are imposed
displacement field satisfying the displacement and trac- at points A1 and A2.
tion boundary conditions is obtained at the end of a – u z = d' is imposed at points A2 and C1, while
generic time increment. In this analysis, the calculations u z = d'+d is imposed at points A1 and C2.
are achieved with a maximum time frame of 0.1 minute For the definition of d and d' see fig. 8.
for each increment.

– θx = ± θx1, θ y = ± θ y1 are applied to constrain the The Tsai-Hill failure criterion is now considered to esti-
nodes on the areas p1 and p2' and mate the macroscopic strength of the textile module.
θx = ± θx2, θ y = ± θ y2 to constrain the areas p2 and p1'. According to the Tsai-Hill failure theory22 the macro-
– The displacements in x-direction are obtained at mechanical failure criterion for anisotropic materials is
points C1 and C2 using the displacement vectors given by
u C1 = –uC ê 1, u C2 = uC ê 1, and equation 1. As previ-
ously, the points are free to move in the σ LL2 σ LLσ TT σ TT2 σ LT2
y-direction to the common point C where AC = s. IF = – + + ≤1 Equ. 3
fm,L 2 fm,L 2 fm,T2 fv2
– For the wedge connection, only one displace-
ment component (in the x-direction) is needed to
position points B1 and B2 at the common point B. where σ LL ,σ TT and σ LT are the local stresses in the ortho-
Thus, the displacement vectors u B1 = –uB ê 1 and tropic material directions. In equation 3, fm,L , fm,T and
u B2 = u B ê 1 should be applied to points B1 and B2 fv (= fv,L = fv,T ) are the bending and in-plane shear strengths
respectively. The expression of u B is given in of laminated wood materials given in Table 1. I F values
equation 2. greater than 1.0 may lead to failure. The Tsai-Hill crite-
rion IF is visualized for each module in fig. 13b. Analysis
3.4 Contact interaction of fig. 13b clearly indicates that maximum values of IF
To prevent interpenetration of the two panels, are located at the contact point B (the most critical
contact regions have to be defined. The general finite- point of the textile module during the assembly process).
sliding, surface-to-surface algorithm of ABAQUS/Standard Table 2 gives the numerical values of the stress com-
is used, allowing arbitrary large sliding, as well as large ponents at point B for prototypes TM0, TM1, and TM2.
rotations and deformations of the surfaces. A frictionless, According to the Tsai-Hill criterion, a module with the
hard-contact, pressure-overclosure relationship is con- same characteristics as prototype TM0 is predicted as
sidered, where the penetration of the slave surface into being likely to fail during the assembly process, as was
the master surface is minimized and no tensile stress experimentally observed in the laboratory.
is transferred through the interface. Master (MS1, MS2)
and slave (SS1, SS2) surfaces that can potentially come
into contact are specified in fig. 12 by subdividing each TM0 TM1 TM2
panel into two regions. Moreover, surface SS1 (resp. SS2) σ LL (MPa) 45.1 18.01 9.61
is located on the lower face of panel 1 (resp. panel 2) while σ TT (MPa) –3.3 –0.302 –0.61
MS1 (resp. MS2) belongs to the upper face of panel 1
σ LT (MPa) 0.65 0.248 0.48
(resp. panel 2). Two contact surface pairings (MS1, SS2) and
(MS2, SS1) can then be identified for the tracking contact
Table 2 Construction stresses values at point B
algorithm. Frictional effects can be included by consider- for TM0, TM1, and TM2
ing the (simplest) classical, isotropic Coulomb model to
describe the frictional behavior between the contacting
panels. Representative values of 0.3 and 0.5 will be taken Although the application of this criterion also strictly
for the (static) coefficient of friction μ in our case. predicts failure for TM1 (Fig. 13b), it should be noted that
the characteristic value of fm,L on the fifth percentile
3.5 Numerical results ( fm,L =15.8 MPa) provided by the manufacturer for GFP
Fig. 13 a shows the initial three-dimensional material and used to calculate the Tsai-Hill criterion, is
shapes obtained numerically for the large-scale and rather conservative. Three-point bending tests, carried
intermediate-scale textile modules. In each case, the con- out on a sample size of 26 GFP plates of size 486 × 50 mm 2
tact opening variable COPEN (in millimeters) corresponds revealed that the median value of the bending resistance
to the clearance between the potentially contacting sur- of the GFP material is approximately 24 MPa. Never-
faces. Negative values of COPEN in fig. 13 a indicate small theless, due to the presence of knots and other irregular-
overclosures of the surfaces. For the large-scale geometry, ities in the wood, a relatively large variability in the
two widths have been considered for the panels (i.e. results was observed, with bending strength values
w = 1.650 m and w = 0.77 m) that correspond to the proto- ranging from 4.3 to 55 MPa. If one takes the median value
types TM0 and TM1. The FE mesh incorporates STRI65 shell of fm,L ( fm,L = 24 MPa), the maximum value of IF obtained
elements with an element mesh size of 110 mm. Prelimi- for TM0 is 5.58, while for TM1 the maximum value of
nary calculations were conducted with several element IF obtained is 0.62. This value corresponds with our
mesh sizes ranging from 80 to 150 mm, revealing a differ- experimental observation that the TM1 module system-
ence of less than 2 % for the in-plane principal stresses. In atically withstands the forces applied to it during its
order to reduce the computational time, a uniform mesh construction.
size of 110 mm (resp. 20 mm) has been chosen for the
simulation of large (resp. intermediate) scale geometries.
141
4 Results and discussion linearly ramped over the step up to 13 kN. Simulated
force-displacement curves are given in fig. 14 at the
To compare the experimental and simulated initial cor responding experimental positions. The actuator dis-
shapes for TM2, the z-coordinates of points 1–5 (Fig. 11b) in placement obtained from the FE model corresponds to
the global frame (x,y,z) are reported in Table 3. It can the displacement of a node in the center of the rectangu-
be seen that the values obtained from FE model and those lar load area. The graphs indicate a nonlinear (elastic) re-
measured with the prototype correspond well. The great- sponse of the structure to bending load. The experimental
est discrepancy is found to be 14 % at point 4. decrease of the load (after 13 kN) is clearly not reproduced
Force-displacement curves are presented in fig. 14 in the simplified, elastic FE model considered.
for prototype TM1, where the vertical displacements are It is observed that the numerical model tends
recorded at locations (points) 1, 2, and 3 (Fig. 11a). Here, com- to underestimate the rigidity of the structure. The dis-
pressive loads are given positive values. The upper left crepancy between the experimental and simulated curves
figure provides a plot of the recorded load versus actua- might be primarily attributable to the material para-
tor displacement. Experimental results indicate a nonlin- meters used. Moreover, the assembly conditions consid-
ear response of the structural element with a maximum ered for the large panels are clearly less controlled than
sustained load of about 13 kN, wich is reached when the those considered for the intermediate-scale specimen.
actuator moves down to 170 mm. At point 1, this maxi- The idealized ‘‘pinned connection’’ could not accurately
mum load leads to a deflection of 120 mm for the timber represent the more complex (real) assembly conditions in
module (fig. 14). Smaller and comparable deflections of this case. Nevertheless, relatively good correspondence
about 80 mm are obtained at points 2 and 3, farther is observed between the experimental and numerical
away from the loaded surface. Note that the value at point predictions.
1 is far more than the maximum deflection (span length/ The force-displacement curves obtained for the
300=33.5 mm) of the serviceability limit state, indicating prototype TM2, when loaded up to 0.5 kN, are plotted in
a degree of inherent flexibility of the structure. fig. 15. Compressive loads are once again given positive
Numerically, the module is loaded at midspan values. To verify good reproducibility, the loading pro-
such that half of the total (vertical) force is uniformly cedure was repeated three times and the correspond-
distributed over a small (600 × 300 mm) rectangular re- ing curves are referred to as Test 1, Test 2, and Test 3
gion of each panel (Fig. 14). In ABAQUS, a second “loading” in fig. 15. A first series of tests was conducted directly af-
step has been added to the first building step (section 3.3), ter receiving the Okoumé material from the manu-
allowing the bending load to be applied on the simul- facturers. A second series of tests (Test 4 and Test 5)
ated (prestressed) textile module. The applied load is were carried out on samples of the same batch of panels
about eight months later, where the same experimental
procedure was applied. The experimental results indicate
an overall nonlinear response of the structure. Figures
labeled “Position i” (i = 1...5) refer to vertical displace-
Point 1 Point 2 Point 3 Point 4 Point 5
ments obtained with transducers positioned at points
Experimental 198 271 135 220 221 i (i = 1...5) as indicated in fig. 11b. Interestingly, it appears
values (mm)
that certain regions of the structure flex upward during
FE values 202 293.6 156.4 257 257.5 the loading test. Specifically, we observed that all the
(mm)
considered points shift downwards (towards the ground),
Table 3 z-coordinate of points 1–5 in global frame (x,y,z) except for point 1, which is shifted in the opposite direc-
tion (positive direction of z-axis). However, for simplicity
of graphic representation, all of the displacements have
a) Load surface
been plotted with positive values in the graphs.
Fig. 15 indicates—particularly in positions 3, 4,
and 5—smaller displacements for Tests 4 and 5 than those
obtained in the first series of tests with the same loading
conditions. This effect might be attributable to a change
in environmental conditions. At 0.5 kN, a displacement
amplitude of 10 mm is measured at point 1 that is compa-
b) Load surface rable to the displacement recorded at the opposite point 3
(for the same panel). The vertical displacement at point 2
remains surprisingly small (less than 1 mm) compared
to the other. Nevertheless, the curves of the second series
of tests indicate the same trend, even though the sudden
increase of the slope observed for the first series after
Fig. 11 Location of measurement points for a) TM1 and b) TM2 (top view) 0.5 mm no longer appears.

TM0
panel 1 panel 2
SS2 MS1
MS2
SS1
TM1
Fig. 12
TM2
Fig. 12. Definition of contact interaction

between the panels
Fig. 13 a) Simulated initial shapes and
visualization of clearance values
between contacting surfaces of TM0, TM1,
and TM2 in millimeters
b) Representation of the Tsai-Hill criterion IF Fig. 13 a Fig. 13 b
The largest displacements (i.e. about 15 mm) are encoun- manu facturer data” in the legend), the numerical model
tered at positions 4 and 5. Potential wedge displacements correctly reproduces the experimental behavior of the
can be recorded with transducers that are positioned textile module at positions 3, 4, and 5. The experimental
horizontally at the two checking positions 6 and 7 (Fig. 11b). trend is also satisfactorily recovered at position 1. Intro-
A maximum (horizontal) displacement of 0.2 mm is reducing the elastic moduli from the laboratory measure-
corded at these points for the wedges. This indicates a ments in the FE model (“FEM-laboratory data” in the leg-
relatively small rotation of the wedge elements due to end), one recovers the experimental load-displacement
the bending load. Numerically, the wedges are fixed and curves of Tests 4 and 5 at positions 1, 3, and 5. It is worth
the displacement is accordingly zero. noting that this material data was obtained during the
Fig. 15 depicts the load-displacement curves ob- period cor responding to the second series of tests, wich
tained with the FE model by considering the manu- were conducted under the same environmental condi-
facturer and laboratory material data. Half of the total tions. The model does not seem to be able to reproduce the
(vertical) force is uniformly distributed over a small cir- small displacements recorded experimentally at position
cular disk with a radius of 15 mm on each panel (Fig. 15). 2 and significantly overestimates them in the load range
The applied load is ramped linearly over the step up considered.
to 0.5 kN. Using the manufacturer‘s data (labeled “FEM-
143
Displacement of actuator Position 1
Load (kN)
Load (kN)
FEM
Test Data
14 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
0 50 100 150 Displacement (mm) 0 50 100 150 Displacement (mm)
Position 2 Position 3
Load (kN)
Load (kN)
14 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
Fig. 14 Experimental and simulated force-displacement curves (TM1 prototype)
The effects of tangential friction between the contacting 5 Conclusion

panels have also been addressed, taking the values μ = 0.3
and μ = 0.5 for the coefficient of friction. The simulations This work has involved the investigation of a novel
indicate that these effects do not influence the global class of timber structures based on the logic and prin-
behavior of the structure in this case. These are not docu- ciples of textile techniques. A geometrically nonlinear,
mented here, while the plots are similar to those pre- finite element model has been developed for the construc-
sented in fig. 15. It is concluded that no sliding occurs tion of a single textile module, including pinned and so-
between the surfaces of the panels. called “wedge connections” for the assembly conditions.
Finally, the introduction of wedge elements sig- For comparison, large-scale and intermediate-scale expe-
nificantly affects the overall rigidity of the structure. rimental prototypes with the previous connections have
Considering the TM2 geometry once again, fig. 16 com- been constructed.
pares the simulated displacement-load curves obtained The proposed analysis first aimed at reproducing
at points 3 and 4 when the structure is assembled with the initial shape of the structure and thus evaluating
pin or wedge connections. In both cases, the same mat- the resulting construction (initial) stresses induced dur-
erial (i.e. Okoumé plywood) is considered and the mod- ing the assembly process. It was shown that the simulated
ules are loaded up to 0.5 kN. As expected, the wedge shape could satisfactorily correspond to the experimental
connecting elements provide more rigidity for the struc- shape at several measurement points. Moreover, the an-
ture than the pin connections. isotropic Tsai-Hill criterion based on the maximum in-

Displacement of actuator Position 3
Load (kN) FEM
Test 1
Test 2
0.8
Test 3
0.6
0.4
0.2
0
Load (kN)
0.8
0.6
0.4
0.2
0
Load (kN)
0.8
0.6
0.4
0.2
0
0 1 2 3 4 5 Displacement (mm) 0 2 4 6 8 10 12 14 Displacement
(mm)
Fig. 15 Experimental and simulated force-displacement curves (TM2 prototype)
145
Vertical Displacement (mm) References
50 1 Langenbach, R. “Resisting earth’s forces: Typologies

Point 4 of timber buildings in history.” Structural Engineering
Pin connection
Wedge connection Inter national: Journal of the International Association for
40
Bridge and Structural Engineering (IABSE). 2008;18:137–40.
Point 3
2 König, H. Bauen mit Holz als aktiver Klimaschutz. Bauen mit
30 Holz: Wege in die Zukunft. München: Prestel Verlag. 2011.
3 Iffland, J.S.B. “Folded plate structures.” ASCE J Struct Div.
20 1979;105:111–23.
Point 4 4 Happold, E. and W. I. Liddell. “Timber lattice roof for the
10 Mannheim Bundesgartenschau.” Structural Engineer.

Point 3 1975;53:99–135.
5 Ibid.
0
6Kelly, O.J., R.J.L. Harris, M.G.T. Dickson, and J.A. Rowe.
“Construction of the downland gridshell.” Structural Engineer.
Load (kN) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 2001;79:25–33.
7 Douthe, C. and O. Baverel. “Design of nexorades or recipro-
Simulated displacements at points 3 and 4 for pin and
Fig. 16 cal frame systems with the dynamic relaxation method.”
wedge connections (TM2 geometry) Computers and Structures. 2009;87:1296-307.
8 Popovic Larsen, O. Reciprocal frame architecture. London:
Architectural Press; 2007.
9 Weinand, Y. “Innovative timber constructions.” Journal of
duced stresses allows one to select safe design para meters. the International Association for Shell and Spatial Structures.
It was observed that a length-to-width ratio l/w = 7.5 for 2009;50:111–20.
the large-scale GFP panels led to failure during construc- 10Weinand, Y. and M. Hudert. “Timberfabric: Applying textile
principles on a building scale.” Architectural Design.
tion, while l/w = 15 was safe. For the intermediate proto-
2010;80:102-7.
type (l/w = 9.75), wedge connections comparatively led to
11D’Amato, E. “Finite element modeling of textile
lower levels of construction stresses and could be adopted composites.” Composite Structures. 2001;54:467–75.
at a larger scale. Second, the structural behavior of the 12D’Amato, E. “Nonlinearities in mechanical behavior of
textile module has been examined under bending tests. textile composites.” Composite Structures. 2005;71:61–7.
For the two geometries being considered, the resulting 13 Kalidindi, S.R. and E. Franco. “Numerical evaluation of
deflections have been measured and calculated at several isostrain and weighted-average models for elastic moduli
locations. They highlight a nonlinear bending response of three-dimensional composites.” Composites Science
of the textile module. Good correspondence is generally and Technology. 1997;57:293–305.
observed between the experimental results and the FE 14Page, J. and J. Wang. “Prediction of shear force using 3D
non-linear FEM analyses for a plain weave carbon fabric in a
predictions at intermediate and large scales. Finally, the bias extension state.” Finite Elements in Analysis and Design.
introduction of wedge elements was found to significant- 2002;38:755–64.
ly improve the overall rigidity of the textile module. 15 Whitcomb, J., K. Woo, and S. Gundapaneni. “Macro finite
element for analysis of textile composites.” Journal of
Composite Materials. 1994;28:607–18.
16 Daniel, I. and O. Ishai. Engineering mechanics of composite
materials. Oxford, UK: Oxford University Press; 1994.
Acknowledgements 17 Bodig, J. and B.A. Jayne. Mechanics of Wood and Wood
The authors would like to thank the Swiss National Science Composites: Van Nostrand Reinhold Company; 1982.
Foundation for supporting this work through contract 18 ABAQUS: ABAQUS Theory Manual Version 6. 9. 2009.
number 200021–126802. They would also like to express 19 Ertas, A., J.T. Krafcik, and S. Ekwaro-Osire. “Performance
their gratitude to Maria Lindqvist for her participation in the of an anisotropic Allman/DKT 3-node thin triangular flat
experimental tests. shell element.” Composites Engineering. 1992;2:269–80.
20 Murthy, S.S. and R.H. Gallagher. “A triangular thin-shell
finite element based on discrete Kirchhoff theory.”
Computer Methods in Applied Mechanics and Engineering.
1986;54:197–222.
21 Reddy, J.N. An introduction to non-linear finite element
analysis. New York: Oxford University Press; 2004.
Reprinted from Engineering Structures, vol. 46, p. 557–568,

Oxford: Elsevier, 2013

147
4. 3
Generation process and analysis of

innovative timberfabric vaults
Etienne Albenque, Markus Hudert, Laurent Humbert, and Yves Weinand
Tragwerke
Modular elements
aus Holzgewebe,
called timberfabric
wie sie ammodules
IBOIS ent-
are connectors are made from planar wooden panels cut by a
generated
wickelt werden,
by curving
sind and
eine connecting
neue Art von twoHolzkon-
slender, numerically controlled machine. The design process of
struktionen,
wooden panels. die This
auf denpaper
logischen
concerns Grundsätzen
the study this prototype is based on an empirical approach, the
of
textiler
vaults
Gewebe
obtained
basiert.
by assembling
Durch das Biegen
severalundtimber-
Ver- principal steps of which are summarized by a simplified
fabric
bindenmodules
von zweitogether.
schlanken A Holzplatten
parametric tool
entstehen
is pre- diagram (Fig. 2).
die
sented
modularen
which can
Elemente,
automatically
auch Holzgewebemodule
generate a three- The complexity of the structural and architectur-
genannt.
dimensional,
In diesem
finite element
Beitragmodel
werden
of a Gewölbekon-
structure for al design of a timberfabric vault is a consequence of
struktionen
a given set ofbehandelt,
initial parameters.
die aus This
mehreren
tool assists
Holz- its spatial configuration, which depends on the geometry
gewebemodulen
the architectural aufgebaut
design of sind.
theseEsstructures,
wird ein Com-
pro- of the timberfabric module, which in turn depends on
viding
puterprogramm
insights into
vorgestellt,
their geometry
das ausasvorgegebenen
well as into the material properties and assembly conditions of the
Anfangsparametern
the construction stresses automatisch
and the ein
kinematic
dreidimen-
con- timber panels. Moreover, construction stresses are gener-
straints
sionalesbetween
Finite-Elemente-Modell
different constitutive
erstellt.
elements.
Dieses ated during the assembly of the timberfabric modules
Programm unterstützt den architektonischen Ent- and those stresses will affect the resistance and the
wurfsprozess für diese Konstruktionen, liefert behavior of the finished structure.
Einblicke in deren Geometrie
Keywords timberfabric und ermittelt
vaults, innovative structures,die The objective of the work presented in this paper
konstruktionsbedingten Spannungen
Python script, generative algorithmsowie die was to develop a parametric tool to assist in the struc-
kinematischen Randbedingungen zwischen den tural and architectural design of a timberfabric vault.
verschiedenen Bauteilen. The research is focused on one particular family of
timberfabric vaults. However, it aims at establishing an
analytical framework for all timberfabric structures
I
1 Introduction
Gewölbe aus Holzgewebe, innovative and describes the current state of a numeric tool that
Konstruktionen, Programmiersprache can be further developed.
In recent years, textiles
Python, have Algorithmus
generativer increasingly served as The developed tool permits the automatic gener-
a reference for architects and civil engineers.1, 2 The re- ation of a three-dimensional, finite element model of a
search project titled “Timberfabric” 3, 4 focuses on the timberfabric vault for a given set of initial parameters.
assembly principles and techniques that are used for the This model can be used to study the spatial quality, as
production of textile structures. By combining these prin- well as the structural behavior and the resistance, of
ciples with the particular properties of laminated timber the projected structure. The elaboration of this tool is
panels, the research sets out to develop a modular con- presented in three steps. Firstly, we present the strategy
struction system for supporting building envelopes. adopted to generate a three-dimensional, finite element
The basic unit for this system was developed during model of one timberfabric module, with given boundary
the first stage of the research. This basic unit, which will conditions, by simulating the deformation of two flat
be referred to as a “timberfabric module,” consists of two wooden panels for which the user can specify both the
interlaced, mutually supporting timber panels. 5, 6 Based dimensions and the mechanical properties. Secondly,
on this unit, a considerable amount of empirical research several geometric properties shared by the resulting
has been carried out in which various vault-shaped, timberfabric modules are highlighted and “intermediate
multi-module configurations were developed.7 In paral- variables” are introduced to describe the geometry ob-
lel, numerical models have been developed to generate tained for the given set of initial parameters. Further-
the deformed shape of several particular configurations more, the results of a parametric study show how the
of one timberfabric module. 8 Fig. 1 shows a prototype initial parameters of the flat wooden panels influence
of a vault built by connecting together entire timber- the “intermediate variables” that describe the geometry
fabric modules with half timberfabric modules. The of the corresponding timberfabric module. Finally, we

Fig. 1
Fig. 1 Timberfabric vault prototype

Fig. 2 Construction process of timberfabric vaults
Fig. 2
149
Parameters of panels
Fig. 3 a)
W t Mid-surface of panels 1 and 2, definition of
b) and c)
geometric entities
L Calculation steps in the process of generating
Fig. 4
a) a timberfabric module
x 21 x 31
x 11 C1
B1
A1
Edge-A1 Edge-C1 Edge-B1
L/2 L/2
b)
Step 1
x 22 x 32
C2
x 12 B2
A2
Edge-A2 Edge-C2 Edge-B2
L/2 L/2
c)
Fig. 3
Step 2
show how those “intermediate variables” can be used to

describe the relative positions of timberfabric modules
in a timberfabric vault system. The numeric tool of the
first part is therefore completed in order to enable the au-
tomatic generation of a full vault from one timberfabric
module. The geometry of the connection pieces and their
behavior are also generated automatically.
The different steps presented in this paper do
not follow the same empirical logic developed before.
In contrast, several initial hypotheses are based on ob-
servations of the built prototypes and their relevance to
the modeling of a physical reality is demonstrated. This
chronology illustrates the synchronicities between em-
pirical and numerical approaches to design.
Step 3
2 Parametric model for generating

a timberfabric module
Here we describe the methodology employed to generate

the geometry and calculate the initial stress state of a
timberfabric module for a given set of geometric para-
meters and material properties of the timber panels. The
methodology is based on an editable Python script writ-
ten under the ABAQUS Scripting Interface environment 9
that builds and runs a non linear, finite element model. Fig. 4

2.1 Geometric parameters and material properties
Thickness t Width W Length L Connector
of the panels (mm) (mm) (mm) height h c (mm)
Timberfabric modules (TM) can be regarded as
8.0 280 2450 80
elementary units that are connected together to form a
more complex spatial structure, such as the vault proto-
type depicted in fig. 1. The module is formed by two
E1 (MPa) E2 (MPa) ν 12 G12 (MPa)
self-supporting, initially flat, rectangular wooden panels
that are deformed and assembled together by connection 4163 5088 0.3 552
pieces at their extremities. The two panels that comprise
a module come into contact at a single point on one of Table 1 Geometric and mechanical properties of the panels
their longitudinal edges. The width, length, and breadth
of the undeformed planar panels are noted respectively
W, L, and T (Fig. 2). Depending on the scale of the timber-
fabric module, the timber panels are either constructed To simplify the operations that simulate the interweav-
with plywood panels (for reduced-scale prototype) or ing of the panels, it is permitted for the model to run
laminated wood panels (for full-scale TMs). In both cases, through intermediate steps that would not be possible
the panels are constituted of orthotropic layers that are in the physical world; for example, the intersection of
considered to be rigidly bonded together. II the two panels (Fig. 4). By comparing the initial simu lated
configuration and the geometry of a real TM, one notices
2.2 Analysis framework that they share several geometric properties. In the mod-
The finite element (FE) code ABAQUS is used to el, these geometric properties are expressed by constrain-
build a three-dimensional model of the deformation process ing the possible relative movements (displacements and
that enables one to generate a timberfabric module from rotations) of several points.V In the virtual space of the
two planar timber panels. Rather than using the graphic model:
user interface, the FE model is generated from the instruc- – A1 the central point of Edge-A1 is fixed through
tions written in a script using the programming language all the subsequent transformations.
Python.9 By automating the construction of the FE model, – All the middle-points of transverse panel edges,
this method allows one to vary parameters with ease. A1, B1, A2 , and B 2 are bound to remain in the same
The initial mathematical settings of the model vertical plane.
have been chosen in accordance with the nature of the
physical problem. III The geometry of the panels is de- 2.4 Interactions and imposed displacements
scribed with reference to their central surface. The posi- Interaction conditions
tion vector of a point/node M of the middle surface is de- In the actual process of fabrication, the panels
noted by rM and the unit vector normal to the middle are deformed in several sequential steps with interven-
surface at this point is denoted by n̂M. Moreover, u M and ing means of temporary fixation. During the final step,
θ M are respectively the displacement and rotation vectors the edges of the panels are forced into position in order
at M, with components and θjM given along a base vector êj. to adapt to certain geometric conditions imposed by con-
In particular, θjM is the component of the rotation of the nection pieces. For the numerical simulation, we chose
normal vector around the base vector êj. to impose some of those conditions beforehand. In the
In the following, the same annotation will be FE model, imposing interaction conditions consists of the
used for node sets and for the corresponding geometric elimination of degrees of freedom (DOF) of a group of
points. Depending on the context, the superscript M will nodes and the coupling of their motion to the motion
either refer to a node or to the geometric point of coordi- of a single node (called the “master node” in ABAQUS).
nates rM = ( r 1M , r 2M , r 3M )T. For example, if all the displacement degrees of freedom of
Let A1, B1, A2 , and B 2 be the mid-points of the the nodes of a straight edge are coupled with the displace-
transverse edges of panels (1) and (2) (Fig. 3). The mid- ment DOFs of one node of the edge, then the nodes will
points of a longitudinal edge of panels (1) and (2) are C1 be aligned for any deformation state of the system. Thus,
and C2 and Edge-A1, Edge-B1, Edge-A 2 , Edge-B 2 , Edge-C1, the effect of a straight, rigid piece of connection pinned
and Edge-C2 refer to the transverse edges corresponding continuously along the edge is simulated.VI As for the
to the points Ai, Bi, and Ci . Again, the same notation is initial boundary conditions, these constraints are de-
used both for the edge as a geometric entity and for the fined by close observation of physical mock-ups and re-
edge as a set of node numbers. produce the kinematic constraints imposed by physical
connection pieces. It is worth noting that these connec-
2.3 Initial position and boundary conditions tion pieces have been developed empirically following
In the initial state of the numerical model, the construction constraints. Therefore, the abstract geo-
two rectangular surfaces simulating the panels are laid metric model is already informed by the constraints of
flat and are placed in the same geometric position.IV the physical construction.
151
Along the extremities of a TM, a rigid connection forces We can thus allow the pair of panels modeled to inter-
the extreme transverse edges of the panels into two par- penetrate each other during the calculation and propose
allel planes. As will be shown in the following section, a virtual assembly process that is far simpler than the
this allows geometric continuity of surfaces belonging to actual fabrication assembly process of a TM prototype.
adjacent TMs forming an arch. In the model, this effect is
translated thus: 2.6 Internal forces and boundary reactions
in deformed state
– The interaction conditions constrain all the The process described above allows one to gener-
edges defined above. ate the deformed shape of two wooden panels with com-
– (Edge-A1, Edge-B1, Edge-A 2 , Edge-B 2 , Edge-C1, and plex boundary and interaction conditions. At this stage,
Edge-C2 ) remain rectilinear. it is legitimate to question whether these particular con-
– The panels are constrained to remain locally ditions are relevant to describing the physical reality of
tangential to planes along their extreme trans- a timberfabric module.
verse edges (Edge-A1, Edge-B1, Edge-A2 , Edge-B2 ), If there is a boundary condition on node M,
while at each extremity of the timberfabric vector R M designates the reaction force at the node. In its
module, the planes for each panel are constrained final deformed shape, the system constituted of the two
to be parallel (they have the same normal vector). wooden panels is subjected to several internal forces
and to six reaction forces at points A1, A2 , B1, B 2 , C1, and C2 .
Imposed displacements By considering the nature of the boundary conditions
With the boundary and interaction conditions and the absence of exterior actions, it can easily be
defined above, the final step to generate the geometry demonstrated that all these forces are collinear to ê2 and
of the TM is to impose displacements on a set of points. R M ¥ ê2 = 0 for M Œ {A1, A2 , B1, B 2 , C1, C2 }.
The first displacements simulate the effect of the connec- Furthermore, if we introduce C—the point such
tion pieces at the extremities of the modules. In the first that rC = ( rC1 + rC2 )/2 — the deformed shape of the panels
step, the displacement of node A2 is constrained along the has an axial symmetry of axis Δ(C, ê3 ). By using the axial
straight line D (A1, n̂ A1) and the displacement of node B1 is symmetry and the absence of external forces, we can
constrained along the straight line D (B 2 , n̂ B2). The norm show that R A1 = –R B2 , R A2 = –R B1 and R C1 =–R C2 .
of the dispacement, hc , is also an initial parameter of the By observing the numerical results, it is pos-
design. It specifies the height of the connection pieces at sible to assimilate the reaction forces R C1 and R C2 to inter-
the extremities of the TM, thus, u A2 = hc n̂ A1 and u B1 = hc n̂ B2 . nal forces. Numerically, we observe that the distance
In the second step, we impose uC1 = W –2 ê2 + W
–2 ê3 and between the positions of C1 and C2 in the deformed state
C2 = –W
–2 ê2 + W
–2 ê3 The imposed displacements on nodes A2
u is inferior to 0.5 % of the length of the panels. Accordingly,
and B1 are maintained. d = ||rC2 – rC1|| < 0.005× L.
The displacements W –2 ê2 and –W
–2 ê2 move respec- By assuming that this distance is negligible, the
tively the points C1 and C2 in the plane, P(A1,ê2), a plane two opposed reaction forces can be assimilated to a con-
that already contains the points A1, A2 , B1, and B 2 . In both tact force between the panels. To confirm this hypothesis,
cases, the last term W –2 ê3 is added to force adoption of one the model was repeated by slightly shifting the initial
of the two equi-probable buckling modes of the panels: position of the points C1 and C2 such that initially
indeed, the displacement along ê2 induces an important
bending moment in the panels along ê3 and if no displace- L–d W
) r C2 = ( 2 , 2 , 0)
T
L+d –W
r C1 = (
T
2 ,0 ,
,
2
ment along ê3 were specified, the panels would continue
to have two equi-probable, lateral buckling modes. For After this iteration, we oberved d’= ||rC2 – rC1|| < 0.001 × L.
reasons of convergence of the algorithm, the norm of The result of the first round of modeling has thus per-
this displacement has the same order of magnitude as the mitted a more accurate prediction of the coordinates of
expected displacement. In the final step, the constraints the contact point.
on Edge-C1 and Edge-C2 are released and the points C1 and Furthermore, in part 3, several variables will
C2 reach their “natural” post-buckling position. be introduced to quantitatively describe the model. The
numerical results of the variables for the two situations
2.5 Example described above is negligible, meaning that the first
For the prototype of the vault described in the approximation is acceptable. Finally, the four remaining
introduction, the panels are made of commercial Okou- reactions are consequences of the descision to impose
mé plywood.VII The deformed shape and displacement the co-planarity of the points A1, A2 , B1, and B 2 . It will be
fields are reproduced in fig. 4 at the beginning and end demonstrated in part 4 that this geometric condition is
of each step of imposed displacements. The objective of necessary for the assembly of several timberfabric mod-
this modeling is to calculate the geometry and the stress ules and that the corresponding reaction forces will be
state of the timber panels that constitute a TM, but is calibrated from one timberfabric module to another.
not intended to simulate the real process of fabrication.

3 Geometric aspects of timberfabric vectors of particular points, so that they can be extracted
modules from the numerical results of the finite element model.
In order to describe the relative positions of several A geometric interpretation of those variables is given in
timberfabric modules that form a timberfabric vault, it fig. 6. More specifically,
is essential to introduce new variables that describe a = |qM2 | = arccos( n̂ M × ê3), M Œ {A1, A2 , B1, B2} is the absolute
the geometry of a timberfabric module. The definition of value of the inclination angle of the tangency planes at the
such “intermediate variables” necessitates some geomet- extremities of a timberfabric module.
ric observations of timberfabric modules. Q = |qC1 – qC1 | = 2|qC1 | = 2|qC1 | is the angle between the panels at
1 2 1 2
mid-span.
3.1 Geometric description Lf = L + u1B and hm = u3C = u3C are respectively the span
2 1 2
The drawings in fig. 5 illustrate a few geometric and the height of a timberfabric module.
consequences of the boundary and interaction conditions
described in Part 2: 3.3 Parametric study
– The points A1, A2 , B1, B 2 , C1, and C2 belong to the Thus far, we have introduced:
plane P (A1,ê2 ). – The initial parameters L, W, t, and hc to describe
– The panels 1 and 2 have tangency planes along the geometry of the timber panels and the
their extreme transverse edges and the planes interaction condition at their ends.
corresponding to each panel are parallel at – The homogenized mechanical properties of
each side of the timberfabric module. In mathe- the panels: E 1, E 2 , n12 , G12 .
matical terms, P (A1, n̂ A1) PP (A2 , n̂ A2) and P (B1, n̂ B1) – The “intermediate variables” a, Lf , hm , and Q to
PP (B 2 , n̂ B2). describe the geometry of the timberfabric
– The four normal vectors n̂ A1, n̂ A2 , n̂ B1, and n̂ B2 , do module.
not have components along ê2 . That is n̂ M × ê2 = 0,
M Œ {A1, A2 , B1, B 2 }. By adapting the script that calculates the deformed con-
figuration of the panels of a timberfabric module, it is
3.2 Identification of intermediate geometric possible to run a parametric study to observe the influ-
parameters ence of the geometric parameters and the mechanical
In order to quantitatively describe the geometry properties on the “intermediate variables” that quantita-
of the timberfabric module, four “intermediate variables” tively describe the geometry of a TM.
are introduced. Those variables are expressed as a The results of this study provide useful informa-
function of components of the displacement and rotation tion for the design of vaults with imposed dimensions.
E2 G12 ν2 Lf hm W2 hc t Lf hm
α (°) θ (°) W ref h cref t ref
α (°) L ref
θ (°) L ref
E ref G12ref ν12ref L ref L ref
0.8 1 1 40.33 0.948 125.89 0.1594 0.8 1 1 32.29 0.964 127.59 0.1333
0.9 1 1 40.31 0.948 125.93 0.1594 0.9 1 1 36.04 0.957 126.56 0.1460
1 1 1 40.29 0.948 125.96 0.1594 1 1 1 40.29 0.948 125.96 0.1594
1.1 1 1 40.28 0.948 125.98 0.1593 1.1 1 1 45.03 0.939 125.57 0.1732
1.2 1 1 40.27 0.948 126.00 0.1593 1.2 1 1 50.25 0.928 125.21 0.1873
1 0.8 1 39.17 0.950 124.90 0.1570 1 0.8 1 40.28 0.951 125.91 0.1564
1 0.9 1 39.74 0.949 125.44 0.1582 1 0.9 1 40.29 0.949 125.93 0.1579
1 1 1 40.29 0.948 125.96 0.1594 1 1 1 40.29 0.948 125.96 0.1594
1 1.1 1 40.83 0.947 126.46 0.1605 1 1.1 1 40.30 0.947 125.99 0.1609
1 1.2 1 41.36 0.946 126.93 0.1616 1 1.2 1 40.31 0.946 126.02 0.1624
1 1 0.8 40.46 0.948 125.99 0.1597 1 1 0.8 42.92 0.943 127.88 0.1651
1 1 0.9 40.38 0.948 125.98 0.1596 1 1 0.9 41.48 0.946 126.86 0.1620
1 1 1 40.29 0.948 125.96 0.1594 1 1 1 40.29 0.948 125.96 0.1594
1 1 1.1 40.20 0.948 125.93 0.1592 1 1 1.1 39.29 0.950 125.16 0.1571
1 1 1.2 40.11 0.949 125.89 0.1589 1 1 1.2 38.42 0.952 124.44 0.1552
Table 2 Results of the parameter study
153
For example, if we want timberfabric modules with a The geometric configuration of the vault is based on
given length, rotation radius, and material, the results of several assumptions and can be described by a new set
the study will help to determine the corresponding length of user-parameters. Here, the study is restricted to the
and width of the panels.VIII family of vaults obtained by assembling identical TMs.
The most relevant output variables are α and Lf as The logic of the assembly and its implementation in a
they will affect the geometry of the vault. The variables script are detailed in the following sections.
hm and Q will affect the geometry of the connection pieces.
For a given length of the panels, we observe that α and 4.1 Arch from timberfabric modules
Lf are mostly influenced by the width W, as shown in Let TMj be a timberfabric module; the notations
Table 2. The thickness of the panels and height of the introduced previously are extended by adding the value
connector have a less critical influence on the geometry, “j” of the module in superscript. Accordingly, the local co-
but greatly influence the maximum stresses. Fig. 7 pro- ordinates system associated to TM1 is written (A11, ê11, ê21, ê31).
vides a graphic illustration of this influence. For brevity, a unit vector at a given point M of TMj will
j
be denoted in the subsequent by n̂M .
In the plane P (A1, e 2 ), a unique circle exists that
1 1
4 Parametric model for timber vaults is tangential to the plane P (A11, n̂A1 ) and passes through 1
the points A1 and B2 . By symmetry, the circle is also tan-

The tool presented in Part 2 is further developed to enable gential to the plane P (B21, ˆnB1 ). From Section 3 and accord-
2
the generation of a Finite Element model of timberfabric ing to the geometric construction of fig. 8, the radius R
vaults. The script that builds the model uses a function of the circle can be expressed as follows:
of ABAQUS that enables us to use the result of a previous
Lf
analysis as the initial state of a new analysis. Therefore, R= ( 2 × sin α
) Equ. 1
the deformed configuration of the generated timberfabric
module can be imported into a new model, copied, rotated, In addition, the height h2 of the arch [ A11, A21 ] is defined by:
and translated in the three-dimensional space. The
timberfabric modules of the new model can then be h2 = R × (1– cos α) Equ. 2
connected to one another and to the ground in order to
form a statically balanced vault.
B1
B2
C1
C2
π(A 2,nÂ2 ) Fig. 5 Geometric characteristics of

timberfabric modules
nÂ2
ê3 Fig. 6 Definition of “intermediate
A2 variables”
ê2 ê1
A2 B1 ê3
A1
nÂ1
B2
C hc A1
C2 1
ê1
π(A 1,nÂ1 )
Section along plane P (A1, ê2)

A2
A1
a
O
, ê 2) C
π( A 1 ê3 ê3
hm
Fig. 5 A2 B1 B1
A1 ê1 a’ B2 _ ê2 B2
Lf
Section aa‘
Fig. 6

L = 240 cm W = 20 cm
hm= 31.9 cm
Initial geometric parameters
L f= 231.4 cm L, W, t, hc
L = 240 cm W = 24 cm
Mechanical properties
hm= 35.0 cm E1, E2 , ν12 , G12
L f= 229.6 cm
Parametric model
L = 240 cm W = 28 cm
hm= 38.3 cm
of the textile module
L f= 227.5 cm
Intermediate variables
L = 240 cm W = 32 cm α, Lf h, Θ
hm= 41.6 cm
L f= 225.4 cm
L = 240 cm W = 36 cm
hm= 45.0 cm
L f= 222.7cm
Fig. 7
B21 = A 21 ê 31 plane P (C1, ê11)

ê 21 ê 33 e hc
ê 2
ê 1 3
ê 22
1
A11 3 C d
a
ê 11 ê 11 _ e2
ê 12 ê 12 hc g
f
h
e c
b
1
C’ ê31 hm
h2
TM1
TM1
2
ê 21 C’1
B
R 2
_ _ _
_ TM2
k
^
: TM 1 _ _
O i^ : TM 2 _ TM2 l1 l2
_ sj
A21 = B12 k
h2
_ _ j
O
A 1 i
1 Lf sj
Fig. 8 Fig. 9 a
Fig. 7 Graphic results of the parametric study

for t = 8 mm and hc = 80 mm (t/tref =1, hc /hcref =1)
Fig. 8 The assembly of timberfabric modules to
form an arch
Fig. 9 Connecting arches together
Fig. 9 b
155
k=1
connectors half-panels
elevation top view
j+1
T T
half-panels
top view
B B
connectors
elevation
T
i+1 i
connectors half-panels
Fig. 10 k=2
elevation top view j+1
B B
j
half-panels
top view
Fig. 10 “Map” of a timberfabric vault H H

connectors
Fig. 11 Kinematic interactions between constitutive elements of

elevation
the model
Fig. 12 Geometries generated for several values of the initial T
and intermediate parameters (hc=80 mm and material
properties are given in Table 1) i+1 i
Fig. 11

In the local coordinates system of TM1, the coordinates of
the circle center O1 are (L–2f , –h1, 0)T with h1 = R – h2 .
If we now introduce (O, î, ĵ ,k̂ ) as the global Carte-
sian coordinates for representing the vault system, TM1
is placed in the global system such that O1 coincides with
O and î, ˆj and k̂ are respectively collinear to ê11, ê12 , and ê13.
TM2 is obtained by rotating TM1 clockwise about the axis
ˆj by an angle 2a, resulting in the following conditions at
the ends of the modules:
nba = 3 nbm = 8 s j = 40 cm
L = 240 cm W = 20 cm t = 8 mm
B 2 = A 1, B 1 = A 2 ,
1 2 2 2
n̂1B2 = n̂ 2A1, n̂1B1= n̂ 2A2 Equ. 3
such that P (B12 , n̂1B2 ) = P (A21, n̂ 2A1 ) and

P (B 21, n̂1B1) = P (A22 , n̂ 2A2 ).
Equation 3 implies that there is a C1-continuity between:

– “panel 1” of TM1 and “panel 2” of TM2
– “panel 2” of TM1 and “panel 1” of TM2
We also observe that the boundary reaction force in B12 nba = 3 nbm = 8 s j = 45 cm
and B 21 are respectively opposed to the boundary reaction L = 240 cm W = 24 cm t = 8 mm
force in A21 and A 22 . These boundary reactions can there-
fore be replaced by interaction forces between TM1 and
TM2. The nature of these interactions and their imple-
mentation in the FE model will be detailed in section 4.5.
This two-module arch is geometrically smooth and has
static equilibrium. Larger arches are obtained by assem-
bling n timberfabric modules in a way that reproduces
the geometry that would be obtained with two continu-
ous panels woven together n times.
4.2 Vaults from arches nba = 3 nbm = 7 s j = 50 cm

In this paper, a cylindrical vault is typically ob- L = 240 cm W = 28 cm t = 8 mm
tained by juxtaposing several arches along the axis (O, ˆj )

of the global coordinate system. The distance between
two arches is fixed by the parameter sj as explained be-
low. The modules are rotated from one arch to the other
by an angle α around the axis (O, ˆj ) (fig. 9). This implies that
the extremity of an arch can be either a full timberfabric
module or half a timberfabric module.
Up to this point, describing the kinematic rela-
tionship between several nodes has simulated interaction
between the panels of a timberfabric module. In order to
nba = 3 nbm = 6 s j = 55 cm
link the arches together, connection pieces are introduced L = 240 cm W = 32 cm t = 8 mm
in the model and referred to as “connectors.”
Let TM1 be a timberfabric module placed in the
global coordinates system, as in section 4.1 As illustrated
in fig. 9, the timberfabric module TM2 is obtained by a
translation of TM1 of vector T = sj ĵ and a rotation of angle
α around the axis (O, ĵ ). The mid-surface of the connectors
belongs to the plane P (C1,ê11): this plane is perpendicular
to the tangency planes P (A21,n 2A1) and P (A21,n 2A2) and in-
cludes the points C1, A21, and A 22 .
The geometry of a connector can be generated
automatically from the previous parameters. IX nba = 3 nbm = 5 s j = 60 cm
Fig. 12 L = 240 cm W = 36 cm t = 8 mm
157
Geometric
parameters of the
panels W, L, t, hc Generative script of the textile module
Generates the deformed shape of a TM
Mechanical properties and extracts intermediate geometric parameters
of the panels
E1p, E2p, ν12
p p
, G12
Intermediate 3D finite elements

variables model of one
α, Lf, hm, Θ textile module
Geometric parameters
of the vault
nbm, nba, sj, tc Generative script of the timberfabric vault
Generates the deformed shape of a TM and extracts
Mechanical properties intermediate geometric parameters
of the connectors
E1c, E2c, ν12
c c
, G12
3D finite elements model of a timberfabric vault
Fig. 13 Synthetic diagram of the complete script
Pieces of wood cut from 25-mm-thick laminated timber in the finite element model by specifying appropriate
panels (class of resistance C30) are assembled to create interaction conditions. As represented in fig. 10, the bot-
the connectors. The detailed construction of these connec- tom section of the timberfabric module is referred to as
tion pieces is not approached in this paper. “half-module k = 1” and the top part as “half-module k = 2”:
each module consists of two panel halves and one connec-
4.3 Map of the vault tor. With this virtual cutting of the timberfabric modules,
The α-rotation from one arch to the other implies the vault can be described with a chess-like diagram.
that some timberfabric modules have to be cut in half When using the script, the user enters a value to
so that all the arches can be connected to the ground at the parameters “nba” and “nbm” to specify the number
the same level. The “half-modules” are shown in fig. 1. of arches and the number of modules (in the study of
In order to give a clear description of the structure, the prototype, we have nba = 3 and nbm = 6). The program
we will generate the vault by only assembling halves automatically computes the corresponding modules. X
of timberfabric modules together. The continuity of the
panels of the full timberfabric modules will be restored 4.4 Geometric assembly process
The geometric assembly is achieved by perform-
ing a double loop on i in [1, 2, .., nba] and j in [1, 2, .., nbm].
Points a b c d For each couple (i, j), the three parts corresponding to
Coordinates* (0,hm ) (–l 1,h 2) (–l 2,h 2+ hc ) (0,hm +h 3) the half module k(i, j) are imported into the model, and
then translated, and rotated. First, the instances of the
Points e r g h parts are placed in the global coordinate system (O, î, ĵ , k̂ )
Coordinates* (l 2,h 2+ hc ) (2sj –l 2,h 2+ hc ) (2sj –l 1,h 2) (l 1,h 2)
such that their reference points A1 (also noted A1j j = 1..nbm
in section 4.1) and point O are coincident and local,
* in ( c '1 , ê12 , ê13 )
and the global axis is parallel. They are then placed in
Table 3 Coordinates of the points defining the geometry the vault system by applying an appropriate translation
of the connectors and rotation. XI

4.5 Kinematical interactions and boundary conditions 5 Conclusion
Finally, to model the vault, it is necessary to
introduce kinematic interactions between the different The numeric tool described in this paper can be specifi-
geometric entities of the model in order to: cally useful for the architectural and structural design of
timberfabric structures. Indeed, the complexity of such
– restore the material continuity of the elements structures resides essentially in the interdependence of
that have been divided (recall that this division the geometrical and mechanical aspects. The primary
was undertaken to facilitate the description of challenge was therefore to find a way to quantitatively
the virtual assembly process) describe the geometry of one constitutive element (a TM)
– simulate the kinematic relationship between the with a small set of variables and to elaborate a procedure
panels and the connectors to calculate them by generating the entire deformed
geometry.
The definition of these interactions is also generated Rather than trying to reproduce the precise
automatically by looping over i in [1, 2, ..., nba] and j steps of deformation necessary to assemble flat panels
in [1, 2, …, nbm–1]. Fig. 11 summarizes the interactions into a timberfabric module and to assemble several
that are defined, depending on the value of k(i, j), and TMs together, the strategy was to directly reproduce the
the letters refer to the nature of the interaction between final geometry. These objectives came from the obser-
the two connected edges. For all the elements that have vation of the results of the empirical approach for the
been virtually “cut” in the model, the corresponding prototype shown in fig. 1. This also means that the geo-
meshes have to be tied back together. In fig. 11, the letter metry of the TMs that had been calculated was not
“T” stands for “tie.” Accordingly, the DOFs of each node of the geometry of self-equilibrated TMs, but rather a geo-
one edge are constrained to the DOFs of the correspond- metry “extracted virtually” from the vault and having
ing node of the other edge. more complex boundary conditions.
For the interactions between the panels and the The diagram shown in fig. 13 synthesizes the
connectors, there are two objectives: ensuring the static actions performed automatically by the scripts and high-
equilibrium of the vault and imposing geometric condi- lights the input parameters that need to be specified by
tions as described in Section 3.1 In practice, the panels the user, as well as the “intermediate variables” that need
are fixed onto the connectors by punctual screws with to be introduced in order to enable the automatic genera-
given characteristics and a given repartition. To model tion of the vault after having calculated one timberfabric
the global effect on the structure’s behavior of these lo- module. It is important to note that this strategy of gener-
cally complex connections, all the nodes of an edge are ating modules from flat panels allows one to calculate the
constrained to the central node of this edge and simple stress state of the panels that constitute a timberfabric
kinematic interactions are defined between pairs of module. This information is essential to study the long-
central nodes. Two different connections are implement- term behavior and the resistance of such structures. Over-
ed, as represented in fig. 11 where “B” stands for “beam” all, the complexity of timberfabric structures calls for the
and “H” stands for “hinge.” The beam constrains all the elaboration of powerful tools. The work presented in this
DOFs of one node to the DOFs of the connected node, paper gives a strategy that generates finite element mod-
whereas the hinge constrains all the DOFs except the els for a rather simple family of timberfabric vaults, but
rotational DOF along the axis of the edge. Finally, the the structural analysis still has to be performed manual-
boundary conditions consist of blocking all the DOFs of ly. Further elaboration of the tool should allow the evalu-
the connectors of the extremities (corresponding to the ation of structures made from timberfabric modules of
couples (i, 1) and (i, nbm + 1) for i in [1, 2, ..., nba]). various sizes and the ability to automatically perform
service and ultimate limit state verifications.
4.6 Application
As was previously argued, the parametric model
is a useful tool for modifying the architectural design
References
of the initial structure. Fig. 12 illustrates how we can
generate a very different geometry by varying the initial 1 Garcia, M. (ed.) “Architextiles.” Architectural Design,
2006;76(6).
and intermediate parameters of the model. This will
Tsukui, N. (ed.) “Cecil Balmond.” A+U Special Issue, 2006.
be of particular interest during the conception phase 2
since architects, planners, and engineers can discuss and 3 Weinand, Y. “Innovative timber constructions.” Journal of
the International Association for Shell and Spatial Structures,
jointly visualize the variation of geometric parameters, 2009; 50(161): 111–120.
such as the curvature radius shown here. The visualiza-
4 Hudert, M. “Timberfabric—Applying Textile Assembly
tion of the global form is crucial for the interdisciplinary Principles for Wood-Construction in Architecture.” PhD
understanding necessary to achieve harmony of structure- Thesis, EPFL, 2013.
function.
159
5 Weinand, Y. and M. Hudert. “Timberfabric: Applying textile V The following boundary conditions are imposed to block
principles on a building scale.” Architectural Design, 2010; certain displacement degrees of freedom of the nodes A1 , A 2 ,
80(4): 102–107. B1 , and B 2 :
6 Hudert, M., and Y. Weinand. “Structural Timber Fabric: – A1 is fixed: u1A1 = u2A1 = u3A1 = 0
Applying Textile Principles on Building Scale.” ICSA2010 1st – B 2 is constrained on line Δ(A1 , ê1): u2B2 = u3B2 = 0
International Conference on Structures & Architecture, – A 2 and B1 are constrained in plane P (A1 , ê 2 ): u2A2 = u2B1 = 0
Guimaraes, Portugal, 2010.
VI The following interacting conditions impose kinematic
7 Hudert, M., and Y. Weinand. “Timberfabric: Innovative
constraints respectively between the degrees of freedom of
Lightweight Structures.” IABSE-IASS Symposium, London,
the node sets Edge-A1 , Edge-B1 , Edge-C1 , Edge-A 2 , Edge-B 2 ,
U.K. , 2011.
Edge-C2 , and the nodes A1 , B1 , C1 , A 2 , B 2 , C2 . To simplify the
8 Sistaninia, M., M. Hudert, L. Humbert, and Y. Weinand. expressions, these constraints are written using the position
“Experimental and numerical study on structural behavior and normal vector of the nodes.
of a single Textile Module.” In: Engineering Structures, 2013; The first set of constraints forces the panels to be tangential
46: 557-568. to planes along their extreme transverse edges (Edge-A1 ,
9 ABAQUS Version 6.11 Documentation: ABAQUS Scripting Edge-B1 , Edge-A 2 , Edge-B 2 ), while at each extremity of the
User’s Manual and Reference Manual. Dassault Systems; 2011. timberfabric module the planes for each panel are con-
10 Daniel, I., and O. Ishai. Engineering mechanics of composite strained to be parallel (they have the same normal vector):
materials. 1994, Oxford, UK: Oxford University Press. – Panel 1 has to remain tangential to the plane P(A1 , n̂ A1)
along Edge-A1 and to the plane P(B1 , n̂ B1) along Edge-B1:
11 Bodig, J. and B.A. Jayne. Mechanics of wood and wood
composites. 1982: Van Nostrand Reinhold Company. (rM – rA1) × n̂ A1 = 0, n̂ M ¥ n̂ A1 = 0, M Œ Edge-A1
12 Reddy, J.N. An introduction to non-linear finite element (rM – rB1) × n̂ B1 = 0, n̂ M ¥ n̂ B1 = 0, M Œ Edge-B1
analysis. 2004, New York: Oxford University Press.
– Panel 2 has to remain tangential to the plane P(A 2 , n̂ A1)
13 Teboply. Okoumé plywood datasheet. www.tebopano.com along Edge-A 2 and to the plane P(B 2 , n̂ B1) along Edge-B 2 :
(rM – rA2) × n̂ A1 = 0, n̂ M ¥ n̂ A1 = 0, M Œ Edge-A 2
(rM – rB2) × n̂ B1 = 0, n̂ M ¥ n̂ B1 = 0, M Œ Edge-B 2
Footnotes
The second set of constraints is imposed on the central
I The following text was initially written for publication in the
transverse edges of the panels (Edge-C1 , Edge-C2 ).
scientific journal Structural Engineering International. In the
The nodes of these edges should remain aligned:
adapted version presented here, several mathematical
notations and some highly specific scientific commentaries (rM – rC1) ¥ (rM‘ – rC1) = 0, M, M‘ Œ Edge-C1
have been replaced or completed by simpler textual descrip- (rM – rC2) ¥ (rM‘ – rC2) = 0, M, M‘ Œ Edge-C2
tions in the core of the text. The original parts that have
been modified are reproduced as end notes to this document. VII The mechanical characteristics of the Okoumé Panel
are E1 = 4163 MPa, E2 = 5088 MPa for the Young’s moduli
II The behavior of a panel at a macroscopic scale is viewed (in traction/compression) and G12 = 552 MPa for the
as being equivalent to the behavior of a single-layer, homo- in-plane shear modulus.13 For convenience, the geometrical
geneous, orthotropic material with principal (material) and material properties of the panels are listed in Table 1.
axes along the longitudinal and transverse geometric direc-
tions of the panel. With those assumptions and considering VIIITo choose the size of the mesh we used for this study, we
a linear elastic constitutive law, only four independent first ran a parametric study on the number of elements per
constants are required to fully characterize the material’s width of the panels. By accepting an error of 0.5 % on
behavior: Young’s moduli in the two principal directions the results of the “intermediate variables,” we chose a mesh
E1 and E2, the in-plane shear modulus G12 and the Poisson’s size corresponding to 12 elements per width of the panels.
ratio ν 12.10, 11 Moisture content and time dependence of
the material properties are not taken into account. IX Considering the sections of TM1 and TM2 by this plane,
the geometry of the connectors is given by the points
III The timber panels are meshed using four-node, quadri- {a,b,c,d,e,f,g,h}. The coordinates of the points in the system
lateral shell elements S4R with reduced integration and six (Cl1 , ê21 , ê13) are given in Table 3 (Fig. 9) . The parameters
active degrees of freedom per node. These general-purpose introduced in Table 3 are given by:
elements are suitable for the analysis of (doubly) curved
l1 = (hm – h2) × tanQ/2
thick or thin shells, allowing transverse shear deformation.
Because large rotations and displacements are expected, l2 = l1 + hc /tanQ2
we run a geometrically non linear analysis for which the 1 p+Q
h3 = hc × (1 + ), Q2 = equ. 4
tanQ/2×tanQ2 4
imposed displacements are applied incrementally.12
In the FE model, those connectors are modeled as two planar
IV Initial position: (O, ê1 , ê 2 , ê 3) is an orthonormal Cartesian shells rigidly bonded along the edge [e,h]: “connector 1”
coordinate system. Initially, the panels are positioned in the and “connector 2” will respectively refer to the planar surfaces
( χ1 , χ2 ) plane with their longitudinal axes along ê1 . For each defined by the sets of points {e,f,g,h} and {a,b,c,d,e,h}.
panel, the initial position of the nodes introduced previously
are such that rA1 = rA2 = O, r B1 = r B2 = (L, 0, 0)T,
rC1 = ( 2L ,-W
2 , 0) , r = ( 2 , 2 , 0) .
T C2 L W T
In the subsequent section, D (M, n̂) will stand for the

straight line through a point M, directed by a (unit) vector n̂
and P (M,n̂) for the plane of normal n̂ at M.

X The nature of the half-module corresponding to a module “j”
of the arch “i” is described by the variable k for which
possible values are 1 or 2. The choice of the value of k for i = 1
and j = 1, i.e. k (1,1) ∫ k11 = 1 or 2, determines the value of k
for any other integer couple (i, j) as
k(i,j) = mod (i + j + (k11 – 1), 2) + 1, 1 £ i £ nba, 1 £ j £ nbm equ. 5
mod (a ,b) being the positive remainder in the division of a

by b, where a and b are positive integers.
XI The translation vector T and rotation angle b about the

global axis ĵ are defined below:
Lf
T=– î + (i –1) × sj × ĵ + (R–h2) × k̂
2
nbm
b = (j + k(i,j) – 1 – )×a equ. 6
2
Note that by taking nba = 1 (i.e. one arch) and nbm = 2 (i.e.
two modules) in equ. 6, the positioning of the two modules
TM1 and TM2 in Section 4.1 in the global coordinate system
is recovered.
Reprinted from Structural Engineering International, vol. 24,

num. 2, p. 254-264.
161
4.4
Mechanical form-finding of the timber fabric

structures with dynamic relaxation method
Seyed Sina Nabaei, Olivier Baverel, and Yves Weinand
Timber
In Holzgewebetragwerken
fabric structures (TFS) vereinen
combine
sich textile 1 Introduction
Fügungsprinzipien
principles with recent mit den
industrial
neuesten developments
Entwicklungen in
in
producing
der Produktion
cross-laminated
von Brettsperrholzplatten.
timber panels. Several Dabei
in- 1.1 Background
dividual
werden mehrere
timber strips
Holzbretter
are wovendergestalt
according
miteinander
to a pat- A timber fabric module (TFM) (fig 1 b) was devel-
verwoben,
tern, therebydasscreating
innovative
innovative
Raumtragwerke
spatial structures.
entste- oped through a research project at the EPFL-IBOIS by
The
hen. three-dimensional
Die so erhaltene dreidimensionale
geometry obtained Geometrie
can be applying textile principles at an architectural scale in
kann
regarded
als asspannungsfreie
the stress-free Anordnung
configurationverformter
of de- the particular context of timber structures.1, 2 This new
Platten
formed panels
unter under
bestimmten
the imposed
Randbedingungen
boundary condi- be- concept of structures is derived from the combination
trachtet
tions. Wewerden.
thus propose
Wir schlagen
a form-finding
deshalb ein pro
Formfin-
cedure of fabric production techniques—the micro-scale struc-
dungsverfahren
that reproduces vor,this das
deformed
diese verformte
configurationAnordnung
as the ture of textiles, weaving, braiding, knitting, etc.—and
steady
als Beharrungszustand
state of a pseudo-transient,
eines pseudotransienten,
constrained, dy- lightweight architecture using timber panels.
dynamischen
namic problem. Problems
The corresponding
mit Randbedingungen
nonlinear prob- dar- The common denominator of different inter-
stellt.
lem involves
Das entsprechende
a finite rotationnichtlineare
regime and Problem
contact be- lacing techniques is the knot (fig. 1 a). The key to the
inhaltet
handling die
through
Berechnung
the cross-section
von endlichen on Rotationen
both panel approach employed in the timber fabric project is to rein-
und
faces.die
To effectively
Bestimmung dealder
withKontaktzonen
nonlinear constraints,
entlang terpret the “knot” by replacing yarns with panels to re-
der
a new Seiten
modified
beiderdynamic
Platten. relaxation
Um nichtlineare
method Randbe-
is uti- produce a more complex overlap (fig. 1 b). Generalization of
dingungen
lized, which effektiv
combinesbehandeln
elastic material
zu können,
behavior with wird the weaving concept is straightforward: continued weav-
ein
a fictitious
neues, modifiziertes
stiffness proportional
dynamischesdampingRelaxations-
into an ing with two strips leads to a braided arch. Fig. 1 c shows
equivalent
verfahren verwendet,
fictitious, viscous
bei dem material
elastisches
model.Werk-The three parallel braided arches.
stoff
procedure
verhalten
is implemented
mit einer fiktiven
as ansteifigkeitspropor-
ABAQUS/Explicit Markus Hudert1 has also investigated the spatial
tionalen
user Dämpfung
subroutine VUMAT in einem
and theäquivalenten fiktiven,
overall accuracy of extension of braided arch structures, both in-depth, as
viskosen Werkstoffmodell
the numerical results has been vereint wird.for
studied Das Verfah-
a number double-layer, braided arches, in transverse direction,
rengeometrically
of wird in der VUMAT-Subroutine von ABAQUS/
nonlinear, shell benchmark prob- and with multiple parallel arches connected together.
Explicit
lems. Thisausgeführt, und für is
numerical approach einige geometrisch
then employed to Various different double-layering and transverse multi-
nichtlineare
simulate theBenchmark-Probleme
assembly process forder a Schale
timber wurde
fabric plication techniques were proposed based on the analysis
die Gesamtgenauigkeit
module (TFM), which isder annumerischen Ergebnisse
interlaced assembly of of the local connection technology and the structural
untersucht. Dieser The
two timber strips. numerische
simulatedAnsatz
geometry wirdforange-
the performance. Double-layer, braided arches have different
wendet, um
deformed den Montageprozess
surfaces is then extracted einesandHolzgewe-
is com- weaving patterns for the upper and lower layers to im-
bemoduls,
pared with das aus zwei miteinander
a three-dimensional verflochtenen
processed surface prove structural performance and reinforce weak areas.
Holzstreifen
mesh obtained besteht, zu simulieren.
from scanning Um das Simu-
a built prototype with Markus Hudert1 has also examined connection pieces that
alationsverfahren
non-contact laser zu scanner
validieren,
armwird die simulierte
to validate the sim- integrate transverse with lateral load patching. He con-
Geometrie der verformten Flächen dann extrahiert
ulation procedure. ducted prototype fabrication in medium and large scales.
und mit einem dreidimensionalen Flächennetz ver- This morphological investigation has mainly been based
glichen, das durch das Abscannen eines gebauten on an investigative approach with prototypes built in in-
Prototyps mit einem
Keywords dynamic berührungslosen
relaxation, Laserscanner
dynamic explicit, timber creasing scales. Although it is helpful to create prototypes
erhalten fabric
wurde.structures, three-dimensional finite to understand and resolve conceptual issues (mainly
elements method, three-dimensional mesh related to local assembly aspects), the newly discovered
processing morphology can hinder the designer’s creativity and his
dynamische Relaxation, dynamisch or her handicraft skills.
explizites Verfahren, Holzgewebetragwerke, Broadly speaking, the fabric structure concept is
dreidimensionale Finite-Elemente-Methode, situated in an interdisciplinary context between topo-
Verarbeitung dreidimensionaler Netze logy/knot theory, structural engineering, and wood tech-

a) b) c) d) e) f)
Fig. 12-D representation of TFS:

a) A simple two-strand knot, TFM
b) Two-strand braid: Three braids placed side by side
c) Three-strand braid
d) Multi-strand knot
e) Three-strand fan
f) Five-strand fan
nology. In this context, further conceptual structures can try condition imposed to keep panels “connected” (with-
be introduced with multi-thread braids (not only limited out collisions) at this particular point. As a consequence,
to two as in Hudert1 ) and with a greater number of over- we are limited to reproducing interlaced shapes for
laps. However, these complex structures quickly become which we ascertain the exact position of edge-edge con-
too complex for an empirical approach; the remaining tact points from a previous experiment with prototypes.
conceptual potential indicates a need for a form-finding/ Furthermore, the position of this edge-edge contact point
tool procedure. is supposed to be given and has not been treated as an
unknown, thus limiting the validity of the approach.
1.2 Mechanical form-finding Thus, we propose two primary contributions to the TFS
The first analysis regarding the structural be- form-finding problem:
havior of TFS was undertaken by Masoud Sistaninia and
his colleagues3 in order to propose a simulation proce- – Reformulating the TFS form-finding problem:
dure for the TFM form-finding problem through a nonlin- The deformed state of the timber fabric
ear, static analysis. In order to validate the simulation, structures can be regarded as the stress-free
the experiment proceeds with two vertical uniformly- configuration of a flat, initial state under a
distributed loads applied to the mid-span. The deforma- set of imposed boundary conditions (BC).
tion is then measured over a number of marked points on These constraints can be either i) the order of
panels, in order to compare them with the simulation re- panels or their offset at a particular overlap, or
sults. The main drawbacks of the approach taken in Sis- ii) an imposed displacement/rotation on a de-
taninia 3 is that the edge-edge contact is not treated in the gree of freedom for a structural node. Interlac-
static, nonlinear, implicit solver employed, and the bounding and then connecting panels together makes
ary condition at the mid-contact point is a simple symme- them deform and initiates a pre-stressed mod-
163
1.0 M
= 0.2
Complex conjugate roots M Mmax
= 0.4
Mmax
Distinct real roots
M
°h L° = 0.6
Mmax
°hL°min Two equal real roots M

= 1.0 undeformed mesh
Mmax
0
1− a 2 g
a Δt
Fig. 2
a)
1
Material prop. Geometry
E = 1.2 × 10 6 l = 12
ν=0 b = 1.0 0.8
h = 0.1
0.6
Mmax
M
const. distributed moment 0.4
along the edge
2πE I
mid plane Mmax = 0.2
l
Clamped edge
analytic solution
u x = uz = 0 0
uyy = 0 modified DR
0 5 10 15
Fig. 3
b) Tip deflection u(A)
l
Fig. 2 Relation between and m E . Fig. 4
Dt
Fig. 3 Clamped strip problem, section 3.2.1
Fig. 4Clamped strip simulation results:

a) Deformed mesh for loadcases
b) Tip deflection u x(A), u z(A) versus end moment
ratio M/Mmax .
ule with active bending behavior. We postulate strands are disconnected locally at each overlap.
that knowing the flat configuration and the The continuous curve passes on top of the others
overlaps interpreted as mechanically meaning- in this particular overlap node. Each strand is
ful BCs suffices to numerically calculate the numbered with positive integers starting from 1,
stress-free form of TSF. In order to distinguish once where they start and once where they end.
different interlacing patterns, a new graph For inter mediate nodes, the order of overlapping
representation is introduced. As seen in strands is noted using the > notation: 1 > 2 > 3
two-dimensional knot graphs in knot theory, stands for a three-strand overlap where strand 1
it helps to distinguish strands and also to take is woven above strand 2 and strand 3 passes
into account a notation to represent the order underneath. Examples of this structural typo-
of overlapping strands in a particular interlace logy with more complex nodes are illustrated
node. Each strand/panel is represented by along with their corresponding graph
a two-dimensional smooth curve and the lower representation in fig. 2.

The two-dimensional graph enables an effective 2 The dynamic relaxation method
description and illustrates a braided configura-
tion in an abstract topological representation, 2.1 A review of dynamic relaxation
regardless of its dual three-dimensional, stress- Dynamic relaxation (DR) is a numerical iterative
free shape. The first step toward a numerical method to find the solution to a system of nonlinear equa-
approach to studying TFS morphologies is to be tions. It is also referred to as “pseudo-transient dynamic
able to differentiate one from the other. analysis” in the relevant literature. This method is used
in structural mechanics to solve the static equilibrium
– Use of explicit procedure to effectively deal with com- of the system by integrating the damped wave equation
plex edge-edge contact handling: to find the steady state of the equivalent dynamic prob-
Once considered as a mechanical simulation, the lem. Interestingly, the method was introduced during
interlacing/weaving of panels involves shell, the 1960s by Otter and Day 4–7 and has its origins in the
nonlinear mechanics in finite deformation. The “second order Richardson method,” developed by Frankel 8
appropriate finite element framework and the in 1950.
nonlinear analysis method have to be deter- Since 1970, DR has been used to solve numerous
mined in order to proceed with the simulation. engineering problems: nonlinear analysis of plates and
The complex assembly of the panels calls for the shells, 9–13 buckling and post buckling analysis,14–16 mem-
use of an existing finite element package. In brane and cable-net form-finding and analysis,17–19 form-
terms of the numerical method, pseudo-tran- finding and analysis of tensegrity structures, 20 inflatable
sient, explicit dynamic analysis has been effi- structures, 21 reciprocal frame structures,22 and medical
ciently used to find a static solution for complex, applications, 23 among others. In addition, several publica-
nonlinear, static shell problems. In such an tions have focused on the convergence and the stability
analysis, the initial static problem is considered of the method and proposed accelerated rates through an
as being pseudo-transient dynamic by adding improved viscous damping matrix, improved conditions
fictitious, dissipative features, such as damping for set-up of the fictitious mass matrix, or by introducing
and mass scaling. The static solution will be the kinetic damping. 24–35
steady state of the equivalent pseudo-dynamic The governing discretized equation for structur-
problem. However, evaluating fictitious energy al static equilibrium can be written as Kx = F, where the
dissipative parameters is a demanding task, solution is x* = K-1F. In material/geometrically nonlinear
as different parameters are usually interrelated, problems, the estimation of the tangent stiffness matrix
thus making the manual calibration difficult. K can be costly computationally. In order to obtain the
The main goal of this study is to establish a static solution using the DR method, fictitious mass and
valid simulation procedure which can handle damping is introduced and the equivalent dynamic
the complex modeling requirements of braided equation is integrated using the central difference explic-
structures such as these, while also enhancing it technique. Let the incremental displacement vector be
an automat ically adapted evolution of fictitious x n = x(tn), the dynamic equilibrium equation at nth time
dissipative parameters for pseudo-transient step will be
analysis.
Mẍ n + Cẋ n + Kx n = Fn equ. 1
The paper is structured in the following manner: the dy-
namic relaxation method (DR) is briefly reviewed and According to the central difference integration and sup-
modified in section 2, and the dynamic relaxation method posing a constant time step, mid-step velocity vector,
is presented in section 3, where the DR method is reformu- step velocity vector, and step acceleration vector can be
lated as a fictitious, viscous, elastic material model. This written as equation 2.
fictitious material model is implemented as an ABAQUS
1 1
xo n + 2 = Dt Q x n – x n–1 V, xo n = 2 Q xo n + 2 + xo n– 2 V
1 1 1
user subroutine material called VUMAT. Results are cross-
referenced with a number of nonlinear shell benchmark 1 equ. 2
xp n = Dt Q xo n + 2 – xo n– 2 V
1 1
problems for validation. Regarding the TFS form-finding

problem, in section 4 the proposed modified DR is used to
simulate an arbitrarily chosen configuration of timber Substituting equation 2 values into Equation 1, gives an
fabric module (TFM). Finally, deformed surfaces from iterative linearized equation for the mid step velocity
simulation and prototype are processed and analyzed in vector (equ. 3). The DR algorithm can then be resumed in
section 5. algorithm 1.
–1
M C M C
xo n + = S Dt + 2 X SS Dt + 2 X xo n– + F n – Kx n X
1 1
2 2
1 equ. 3
x n + 1 = x n + xo n + Dt
2
165
Algorithm 1 DR method 3 Modified dynamic relaxation method
Require: C, M and δ as admissible error 3.1 Mathematical formulation
1: Initialize with n = 0 , x 0 and ẋ 0 = 0 The idea here is to add the fictitious damping
2: while (rn = Fn – Kx n) ≥ δ do contribution to the material elastic response and to treat
the dynamic relaxation concept as a fictitious viscous
3: if n ≥ 0 then
material model. Revisiting the Rayleigh damping equa-
4:
1 Dt
xo ! 2 M –1 r 0
2 tion, C = cM + κK, a stiffness proportional damping matrix
is supposed as C = κK and by following the eigenvalue
5: else analysis of the error vector, a near critical damping co-
6: M C –1 efficient is obtained.
xo n + ! S Dt + 2 X
1
2
Let M be a diagonal lumped mass matrix with di-
SS M + C XX xo n– + F n –Kx n
1
2
agonal elements mii, D = M-1K and C = κK. Substituting these
Dt 2 into equation 3 and rewriting the whole with regard to
7: end if the step displacement vector x, results in equation 4.
8: 1
x n + 1 ! x n + xo n + Dt
2
a 1 x n + 1 + a 2 x n + a 3 x n–1 – F n = 0
9: n < n+1
M S lDt
10: end while a1 = 2 I + 2 DX
Dr
11: return x a2 =
M
2 Q Dt D – 2I V
2
Dr
M S lDt
a3 = 2 I – 2 DX equ. 4
Dr
2.2 Fictitious mass and damping
Different choices for the damping C and mass M
matrices lead to distinct DR-derived variants where the Following Papadrakakis39 and Lynch 26, if x* is the solu-
rates of convergence and stability conditions will be dif- tion for the static equilibrium equation K x = F, then the
ferent. The mass matrix M is usually scaled in order to incremental displacement error vector can be introduced
increase the time-step size, although it can also be left as as εn = x n – x*. Furthermore, successive error vectors are
a lumped mass diagonal matrix. Some examples of the supposed to be linearly dependent via a matrix E as
fictitious mass proposed in scientific literature are as εn+1 = Eεn. Let λE be the largest eigenvalue of E (including
follows: a mass matrix computed equation from different the non-real ones), then εn+1 = λE εn = λE2 εn–1. In order to have
directional densities, 36 a diagonal mass matrix consisting the iterative method converge, λE should have its complex
of the diagonal elements of the stiffness matrix,26 a mass norm less than one, |λE | < 1. Substituting an incremental
matrix based on the stiffness matrix proposed with an error vector into equation 4, λE can be calculated as the
insight on the Gerschgorin theorem,27 a mass/scale factor solution of a quadratic equation, where m, λD are respec-
determined by the incremental rate of change of the ki- tively any eigenvalue of M and D.
netic energy upon internal energy for the structure 37
and, finally, a diagonal lumped mass matrix that is scaled m 1 m 2E + m 2 m E + m 3 = 0 &
at each node to align with the maximum values of its
elements. 23 A typical equation approximation for C is –m 2 ! m 22 – 4m 1 m 3
mE =
the Rayleigh damping, which presumes the damping to 2m 1
be a linear combination of mass and stiffness matrices, m lDt
m1 = 2 S 1 + 2 mD X
C = cM + κK. The other main alternative is to introduce ki- Dt
netic damping, as introduced by Cassell et al. 38 The idea m
m2 = 2 Q Dt m D – 2 V
2
behind kinetic damping is to observe the evolution of the Dt

kinetic energy of the system and to neutralize the veloci- m lDt X
m3 = 2 S1– mD
ty components as soon as a local kinetic energy peak is Dt 2 equ. 5
observed. Supposing the Rayleigh damping, the main
approach has been to suppose a particular construction
for damping matrix C and then to adapt the damping It can be shown that the minimum |λE | occurs when the
coefficient to a critical damping value at each iteration quadratic equation 5 has two equal real roots that corre-
via the eigenvalue analysis of the iterative error vector. spond to a near critical damped system. (fig. 2)
(Similar to Lynch 39 and Papadrakakis26 in several succes-
sive publications.) This approach is also called “adaptive
damping.”

Material prop. Geometry Material prop. Geometry
E = 21.0 × 10 6 ri = 6.0 E = 6.825 × 107 ri = 10.0
ν = 0.0 re = 10.0 ν = 0.3 h = 0.04 x–y plan symmetry BC
h = 0.03 θ = 18°
uz = 0
uxx = uyy = 0
Clamped edge
u x = u y = uz = 0 const. distributed moment
uxx = uyy = uzz = 0 along the edge AB z–y plan symmetry BC
Pmax = 0.8 ux = 0
uyy = uzz = 0
mid plane
mid plane undeformed mesh
Fig. 5 Annular ring problem, section 3.2.2 Fig. 7 Buckling of a pinched hemisphere problem, section 3.2.3
undeformed mesh P P P = 100 P = 200

= 0.2 = 0.4
Pmax Pmax
a) P = 300 P = 400
P P P
a) = 0.6 = 0.8 = 1.0
Pmax Pmax Pmax
400 Sze et al.
Buetcher et al.
Simo et al.
0.8 Sze et al. (among others) Modified DR
Modified DR
300
0.6
Load P
200
Load P
0.4
100
0.2
0 0
0 5 10 15 0 2 4 6 8
b) vertical deflection uZ b) radial deflection
Fig. 6 Annular ring plate: a) Superimposed deformed mesh for Buckling of a pinched hemisphere:
Fig. 8
load cases, b) Vertical deflection uZ versus free edge shear force P a) Deformed mesh configuration for P = 100; 200; 300; 400
b) Radial deflection versus radial load P
167
In the case of explicit time integration, maximum stable Here, L(e) is the element characteristic length (or if simpli-
time step must satisfy the condition fied, the “smallest” dimension of the element),
2
Dt cr # ~ max c= S m + 2n X
t
for an undamped system with ωmax as the maximum and λ, μ are the Lamé constants, ρ being the density.
frequency of the system and
In the numerical implementation of explicit FEM solvers,
2 2 the estimation of system frequencies at each increment to
Dt cr # ~ max # ~ max Q p2 + 1 – pV
determine the maximum stable time step size is a costly
task, so the Courant-Friedrichs-Lewy condition is practi-
in case of damping, where 0.0 ≤ ξ ≤ 1.0 is the fraction of cally verified for the smallest element in order to deter-
critical damping in the mode with the highest frequen- mine the Δtcr.
cy. 40 The relation for the maximum time step size can also This, along with the fact that we have related the
be written as a function of λmax = ω 2max and the maximum stiffness proportional damping coefficient κ to the time
eigenvalue of the system, as step size with equation 6, enables us to adapt the fictitious
stiffness proportional damping contribution to a near
2 critical value at each time step, based entirely on the time
Dt cr =
m max step size estimation. In fact, the ratio
Following the notation used in equation 5, it is con- min Dt (cre)

venient to recall that λmax is related to the maximum Dt cr
eigenvalue of D = M-1K, ( λD)max 40, 41 and since equation 5 is
true for any eigenvalue, one can suitably replace λD as is equivalent to the coefficient γ in equation 6, which
a function of maximum time step size Δtcr. If we then should be selected deliberately by the user in our algo-
assume that the actual increment time step Δt is also rithm. This ratio can be calculated knowing the finite
proportional to this stable time step size by a real coef- element’s mass and stiffness matrices and the problem
ficient 0.0 ≤ γ ≤ 1.0d as Δt = γΔtcr, then all terms of the configuration. The finite element’s mass and stiffness
discriminant in equation 5 can be rewritten with respect matrices give us the maximum eigenvalue of the element,
to γ, Δt and κ. If then, it is set to zero to obtain the fastest and the maximum stable time step size chosen for the
convergence condition, the following equation can be element is
established between the stiffness proportional viscous
damping coefficient κ and the time step: Dt (cre) =
2
m (max
e)
l 1 – c2
= (see Tablei 43 for a demonstration of thin-shell problems).
Dt c equ. 6
Here we take the approximate value of γ = 0.9 for our
numerical simulation. The main interest of a stiffness
Regarding the discretized system, in explicit central proportional matrix over the other proposed options is
difference FEM, the maximum eigenvalue of the assem- that it can be implemented as a viscous elastic material
bled system λmax is bound with sufficient accuracy, by model and used as an existing time step estimation algo-
the maximum eigenvalue of any element rithm inside an explicit procedure, without explicitly
having access to assembled mass and stiffness matrices.
max or Dt cr =
m (e)
2
# Dt cr #
2 The viscous stress is generated proportionally to
m(e)
max mmax the strain rate in the form of σvis = κ Cε̇ . The total damped
stress σ = Cε + κ Cε̇ is then used to evaluate the amount of
where Δt(e)cr is the element’s stable time step size.41 On internal force necessary to compute the accelerations for
the other hand, the maximum stable time step for any velocity update. The above-mentioned idea is resumed
element must satisfy the Courant-Friedrichs-Lewy condi- under the following algorithm:
tion, 42 which is based on the smallest transit time of a dil-
atational wave across any of the elements in the mesh.
The Courant-Friedrichs-Lewy condition in our context of
structural mechanics can be taken as
L(e)
Dt (cre) # c # Dt cr

Material prop. Geometry Load
Algorithm 2 Incremental total stress and internal EL = 3300 ri = 10.0 P max = 3000
force update algorithm ET = 1100 l = 2× 254
νLT = νTT = 0.25 h = 0.04
Require: γ, C, t, Δt(n), σ(n)
p , δε p , f int
(n) (n) GLT = 660 θ = 0.1 Rad
1: if t = 0 then
2: for all integration points p do hinged BC
u x = u y = uz = 0
3: σ(0)
p ← C δε p
(0)
4: end for
5: else
6: for all integration points p do
7: ε(n+1)
p ← ε(n)
p ← δε p
(n) mid plane
1 – c2 Y
8: vQpn + 1 V ! CfQpn + 1 V T c CdfQpn V undeformed mesh
v (pn + 1) + v (pn)
9: f Qintn + 1 V ! f Qintn V S 2
X dfQpnV
10: end for

11: end if
12: return δε(n+1)
p , f int
(n + 1)
Viscous damping parameter, Dlt = 1c– c

2
γ
C Material stiffness matrix Fig. 9 Hinged cylindrical laminate problem, section 3.2.4
Δt (n) Time step size in nth increment
σ(n) Cauchy stress vector in nth time increment
δε(n) Incremental Strain vector in nth increment
ε(n) Green Strain vector in nth increment
f(n)
int Internal force in nth increment
1.2
Reddy et al.
Modified DR
3.2 Numerical implementation 1
In this section numerical results obtained from
the simulation with the VUMAT user subroutine are ex-
0.8
amined for accuracy. In particular, verification is real-
ized using results presented in Sze, 44 Arciniega, 45 Buech-
ter, 46 and Simo, 47 among others. 0.6
Four problems are tackled here: The isotropic
Pmax
P
elastic behavior is assumed throughout section 3.2.1 and

0.4
section 3.2.3, and an orthotropic laminate case is studied
in section 3.2.4 The case study in section 3.2.1 is a classical
benchmark problem with an analytical solution to test 0.2
the large rotation of shells. The bending/torsional cou-
pling is tested in section 3.2.2, and inextensible bending
0
behavior is examined with the buckling hemisphere
problem in section 3.2.3 This problem is considered as be-
ing both a linear and a nonlinear benchmark problem. 0 10 20 30
Here loads are increased until they reach the large defor- vertical deflection at mid point C
mation state. Finally, the post-buckling behavior of an
orthotropic laminate cylindrical roof is studied in section Fig. 10 Hinged cylindrical laminate problem: Vertical
deflection of the mid point versus relaxed reaction force
3.2.4 for double lamination, in order to demonstrate the
generalizability of the proposed approach.
The ABAQUS S4, 4-node, quadrilateral shell finite
element with explicit time integration scheme and second-
order accuracy is used for all simulations. The VUMAT
user subroutine describes the fictitious viscous material
behavior and replaces the software’s default elastic
169
behavior in order to determine the stress and internal 3.2.2 Annular ring plate under transverse shear edge force
energy update for each increment. According to the An isotropic open ring is clamped on one edge
communicated syntax, VUMAT must define the Cauchy and a uniformly distributed edge shear force P is applied
stress and the rotations are applied by the framework to the other edge along the z-axis. (fig. 5) The strip under-
with the co-rotational formulation based on the Green- goes a spring-like oscillation under bending and tor-
Naghdi rate. The rotation, however, should be implement- sion caused by the sheared edge force until it reaches the
ed if a rate other than the default rate is required. equilibrium steady state. The plate dimensions are speci-
The analysis is run over time until it reaches a fied in fig. 5 and the thickness is intentionally h = 0.03.
steady state for the desired degree of freedom. The only Material elastic properties are E = 21 × 106, ν = 0.0. The strip
dissipative feature is the fictitious stiffness proportional is simulated under P/P max = 0.1, 0.2 the two-dimensional
damping with algorithm γ = 0.9, as detailed in algorithm 2. 0.4, 0.5, 0.6, 0.8, 0.9, 1.0 load cases and vertical deflection
The default linear, bulk viscosity parameter is thus of points A and B (two ends of the loaded edge) are ex-
turned off. The proposed approach is based on a pseudo- tracted to plot the load-deflection curve.
transient dynamic analysis, so wherever the density is The final load case (P = P max = 0.8) involves a de-
not specified in the case studies, we assumed the deliber- formation of ≈ 3 × r i . Deformed configurations are plotted
ately chosen value of 500 kg/m3. without magnification in fig. 4 and the simulated load-
deflection curve for both end points are compared with
3.2.1 Roll up a clamped plate strip results presented in Sze, 44 and Arciniega, 45 among others.
A distributed moment is applied on one end of a
plate strip, while the other end is clamped. This problem 3.2.3 Buckling of a pinched hemisphere
has an analytical solution for the tip deflection in 2D case Two equal point loads P are applied to a pinched
(in xz plan): hemispherical shell with an opening on top. The problem
is simulated in one quadrant due to the symmetry bound-
EI MI
u 1 Q A V = M sin S EI X –1 ary conditions at opposite ends (fig. 7). This problem is a
EI MI test to examine the method’s ability to represent inexten-
u 3 Q A V = M S 1 – cos S EI XX equ. 7 sible bending behavior and rigid body modes and is used
as both a linear and nonlinear benchmark, depending on
2rEI the applied load level. Here, the load is increased to cause
If , M = M max = I the timber strip is rolled into a full a deformation level up to ≈ 60% of the radius to be placed
circle. The problem configuration is shown in fig. 3 in a large deformation regime. The material is assumed
and the material properties are set as E = 1.2 × 106, ν = 0.0. to be linear elastic with E = 6.825 × 107, ν = 0.3 and shell
The results obtained for simulations with M/Mmax = 1.0, thickness is h = 0.04.
0.9, 0.8, 0.6, 0.4, 0.3, 0.2, 0.15, 0.1 are compared with the Radial deflections under the applied load curves
computed analytical solutions. The deformed configura- are extracted for load cases P = 20, 40, 60, 80, 100, 120, 140,
tions are also extracted for a number of the load cases 160, 180, 200, and are compared to transcribed results in
in fig. 4. the literature. Particularly, modified DR results agree
non-accessible zone
a)
A1
C1
b)
Fig. 12
Fig. 11 TFM built prototype

Reconstruction of 3-D surface mesh corresponding to
Fig. 12
the top-surface of Panel 1
a) The scanned point cloud of the prototype after registration
b) The 3-D mesh surface reconstruction of the point cloud
Fig. 11 (see Fig. 13 a)

w
C‘2 w
2
2
C‘1
C2 C1
Zone coupled with the
control point A2
3
1
2 3
1
A‘1 2
A‘2
A2 A1
Material orientation
w
2
w w
4 2
panel 2
panel 1
l
d
a) d‘
p1
A1
θx1 θx
p2 θy1 2
threade
d rod
A2
θy2
wedge 3
A1
wedge 2
wedge 1
d
A2
d‘
bolt
O
b)
Fig. 13
Fig. 14 a) 0.0 ≤ t ≤ 2.0 b) 2.0 ≤ t ≤ 4.0
S, Mises
TFM prototype geometry and assembly,
Fig. 13 a) SNEG, (fraction = –1.0)
b) Connection relative angles and offset, (Avg. 75 %)
wedge connection detail +2.7e+07

+2.4e+07
Deformed snapshots for simulation Step 1
Fig. 14 +2.0e+07
+1.8e+07
0.0 ≤ t ≤ 2.0, see column a), view from top to bottom +1.3e+07
and Step 2 2.0 ≤ t ≤ 4.0, see column b) view from top to bottom +6.7e+06
+2.2e+06
Simulation results at the end of Step 3
Fig. 15 +6.2e–01
a) The von Mises stress on mid-surface panels [Pa]
b) Deformed snapshot for t = 14.0 Fig. 15 a) b)
171
140 with values ratified by Simo et al. 47 which are also report-
ed by Buechter and Ramm, 46 and Jiang and Chernuka. 48
120 These curves are different; their values are shown in
Sze44 (fig. 8b).
100
Kinetic Energy (Nm)
3.2.4 Hinged cylinder roof:

80 post-buckling behavior of an orthotropic laminate
A vertical point load “P” is applied to the mid
60 point of a cylindrical shell with hinged boundary condi-
tion assumed for lateral edges. The other two transverse
40 curved edges remain free. The problem involves buckling
and post-buckling of a lightly curved laminate shell (see
20 Step 1 fig. 9 for problem configuration).
Step 2 Step 3 Two laminate configurations are considered:
0
[0°/90°/0°] and [90°/0°/90°]. All lamination plies have an
equal thickness of h/3 and ply material assumed proper-
0 2 4 6 8 10 12 14
ties are assumed as EL = 3300, ET = 1100, GLT = 660, and
Time (s)
νLT = ν TT = 0.25. Ply direction 0° is attained once the fibers
run along the longitudinal edge. Viscous material defini-
140
tion can be used with both imposed load and/or imposed
displacement regimes. Here, the problem is simulated
120
with the imposition of the various displacement levels at
the midpoint. Instead of the point load, the vertical dis-
100
placement is imposed at the midpoint and vertical reac-
Kinetic Energy (Nm)
tion force is extracted once steady state is reached for the

80
reaction force. The load-deflection curve is extracted for
the midpoint displacement -u Z(C)2.5, 5, 10, 12.5, 15, 17.5,
60
20, 23, 25, 27.5, 30, 31.66, and is compared with the results
40
reported by Sze et al. 44 and Arciniega and Reddy45 (fig. 10).
20
4 TFM prototype simulation
0
4.1 Prototype geometry and assembly process
0 2 4 6 8 10 12 14
A medium-scale prototype has been built from
Fig. 16 Time (s) 2340 × 240 × 6.7 mm laminated wood panels to verify
simu lation results (fig. 11). In order to ensure that relative
panel offset and spatial angles are respected at their
correspondent connections, three wooden wedges per
connection point are exploited. The module is fixed
with three bolts per wedge into a rigid panel. The set
of three wedges at one end is mirrored on the other set,
corresponding to the mid-plan normal to the y-axis. The
geometric parameters for the prototype are as following
(notations illustrated in fig. 13).
Fig. 16 Energy evaluation during the TFM simulation steps

Fig. 17Deformed surface curvature comparison:
a) Simulation extracted surface
Fig. 17 a) Mean curvature (m –1) b) Gaussian curvature (m –1) b) the prototype reconstructed-scanned surface

l 2060 mm θx1 17° 4.4 Finite element simulation
w 240 mm θx2 11.5° The mid-surfaces for each panel are simulated
d 47.7 mm θ y1 14.5° using quadrilateral shell finite element, S4R ABAQUS/
d’ 76 mm θ y2 -8° Explicit, with reduced integration scheme and enhanced
hourglass control. Each FE node has six degrees of free-
dom: three translational and three rotational. ABAQUS/
In order to assemble the woven timber module from Explicit user subroutine VUMAT described in Section 3 is
flat panels, first A1 and A2 are positioned respectively at exploited for this simulation with constant γ = 0.9.
A'1 and A'2 and are fixed. Subsequently, C1 is positioned Four fixed points, A'1, A'2, C'1, C'2, are defined in
at C'1 and fixed, and finally C2 is brought to position, three-dimensional space, which represent where cor-
C'2 and the entire structure is connected to the base steel responding points for each panel should be positioned.
profile (fig. 13a). Two nodes are marked on each panel as positioning points
to deform the panel, with A1, C1 for panel 1 and A2, C2
4.2 Material properties for panel 2. These positioning points help to impose trans-
Cross-laminated TeboPlyTM Okumé plywood pan- lational/rotational boundary conditions forced by con-
els, supplied by Thebault (France) are used to build the nector wedge blocks.
prototype. The panels are 6.7 mm thick. It is a 7-mm, four- A rectangular zone (corresponding to the wedge
ply [0/90]s laminate. Orthotropic material orientations cross-section) is selected per positioning point and is
are illustrated in fig. 13 a and homogenized material con- rigidly coupled with it over all six degrees of freedom.
stants are E1 = 4163Mpa, E2 = 5088Mpa, ν 12 = 0.3, G12 = 552MPa, ABAQUS/Explicit general contact feature is used to handle
and G13 = G 23 = 91MPa, which correspond to the values ob- face and edge contact during all simulation steps. Only
tained from the examples tested in ambient temperature non-penetrative, hard normal contact is considered, and
and humidity. the tangential component is ignored. Form-finding simu-
lation is realized in three steps. Since the analysis is
4.3 Surface measurement using laser scanner FARO pseudo-dynamic, a notion of time length is assigned
arm and mesh processing to each analysis step, but the duration of each step is
In order to compare simulation results and vali- chosen to be long enough for imposed boundary condi-
dation, the top surface of panel 1 (fig. 13a for panel 1 at RHS) tions to be applied smoothly (see figs. 13 a and 13 b for notations) .
from the prototype is scanned using a three-dimensional,
non-contact scanning technique with a FARO laser arm – Step 1: (0.0 ≤ t ≤ 2.0)
scanner. Depending on the size of the prototype, scanning The panels are positioned with exact relative
is performed through six snapshots with sufficient super- offset d, d' and their left-hand side handles A1,
position to ease the registration. However, due to the tight A2 are angled into fixed points A'1, A'2 respec-
over-closure between panels along the connection zone tively. θx1, θ y1 and θx2 , θ y2 are imposed respective-
between C'1 and C'2 handle points, the reconstructed ly for A1, A2. All degrees of freedom for these
panel is slightly smaller than the flat initial rectangle. points remain fixed during this first step. Keep-
After every six snapshots, the point cloud is ing A1, A2 fixed during a dynamic/explicit step,
“cleaned” to contain only the vertices corresponding to the handle point C1 is brought to position C'1
the top surface of panel 1, and filtered to merge all the and spatial angles – θx2 , – θ y2 are imposed.
vertices together that are smaller than a 10-mm thresh- (Wedges are mirrored.) Rotational and dis-
old. Duplicate vertices are also removed. These point placement BCs are imposed simultaneously.
clouds are then aligned and merged using the three- The left midpoint of panel 1, noted as B1, is
dimensional shape registration technique based on the also fixed in the x-axis in order to ease the
ICP algorithm. 49 Subsequently, a three-dimensional, tri- overlap that will occur in the following steps.
angular surface mesh is reconstructed to match the
aligned and merged point cloud using the Poisson sur- – Step 2: (2.0 ≤ t ≤ 4.0)
face reconstruction technique. 50 The reconstructed mesh, Positioning points A1, A2, and C1 are kept at
output of Poisson reconstruction, extends beyond the their deformed positions A'1, A'2, and C'1 respec-
initial point cloud boundary and thus has to be split tively. All degrees of freedom for these three
with poly-lines that correspond to approximate surface positioning points are fixed during this step
edges. This trimmed and reconstructed surface mesh except θzz to let panels reposition respective to
is also re-meshed with uniform 30mm mesh size target each other. The fixed boundary condition for
length, using an isotropic, surface-based incremental positioning point B1 is also released. Starting
remeshing technique. 51 This remeshed, polygonal, three- with Step 2, the point C2 is brought to position
dimensional mesh (fig. 13) is used for comparison with C'2 and spatial angles – θx1, – θ y1 are imposed.
simulation results in section 5. During this step, panels will intersect over the
173
cross-section and will reposition themselves values, particularly in surface zones extending between
into the relaxed configuration. the mid-contact point and the connection wedge, which
means that the TFM deformed surface cannot be devel-
– Step 3: (4.0 ≤ t ≤ 14.0) oped in its precise definition. Furthermore, the deforma-
Having four positioning points at deformed tion map from the flat panel onto the deformed surface of
position and exact angles, all degrees of freedom the TFM is not isometric. Meanwhile, the absolute value
for these four positioning points are fixed for the Gaussian curvature remains relatively small.
during this step and the module is left to dissi- The distribution for Gaussian curvature follows the intu-
pate extra kinetic energy. ition that flat, cut wedges impose planarity (zero Gaussi-
an curvature) on a locally restrained zone at both ends.
Deformed snapshots extracted at equal time intervals il- Analysis of the mean curvature leads us to the important
lustrate the deformation of panels during Step 1 and impact of the mid-contact point between panel edges,
Step 2. During Step 3 deformed geometry does not vary which imposes the local flexural deformation in this
much from the deformed state at the end of Step 2, but area. Transverse stability of each panel in the TFM assem-
it is essential to ensure that the system has reached the bly relies on in-plane stiffness at this mid-contact point.
stationary state to have an accurate stress state. At the Distribution of both Gaussian and mean curvature are
end of Step 3, deformed coordinates of nodes for the top not symmetrical, which limits the simplification of the
surface of Panel 1 are extracted and reported for surface simulation.
reconstruction and comparison with the scanned proto-
type geometry (Section 4.3).
6 Conclusion and future work
5 Results and comparison Pseudo-dynamic, explicit finite element methods, such as

dynamic relaxation, can effectively deal with geometri-
With our primary objective being form-finding of TFS, we cally nonlinear problems of form-finding and initial
progress with a comparison of deformed three-dimen- stress prediction for space structures, where the objective
sional shapes, issue of modeling, and prototype scanning. is the free manipulation of thin shells undergoing finite
Here, the main focus will be on the geometric comparison rotation.
of these surfaces, rather than on the more classic ap- Timber fabric structures (TFS) are novel con-
proach of loading the prototype and deformation meas- ceptual structures that originate from a reinterpretation
urements as tracked in Sistaninia 3. of knot and braided structures with thin timber panels
that can be generalized into a wide family of spatial
5.1 Simulation surface processing structures where the form and structural behavior are
Deformed coordinates of nodes of the upper sur- strongly interrelated.
face of Panel 1 are ascertained from the simulation result Mesh processing pipeline (from shape laser scan-
database for the last frame of the simulation, Step 3. ning up to three-dimensional mesh surface reconstruc-
A three-dimensional, triangular mesh is patched onto tion and discrete curvature estimation techniques) seems
these vertices and remeshed based, on the same tech- to represent a promising approach for complex geometry,
nique and criteria described in section 4.3. structural form investigation and may efficiently replace
traditional loading experiments and reshape the special
5.2 Surface comparison focus on the deformed shape comparison.
Surfaces are aligned with the help of the ICP reg- The simulation procedure with geometrically
istration algorithm, in order to obtain an optimum fit nonlinear, shell FE, although precise, is expensive and
between two, three-dimensional meshes. A maximum less responsive to design changes. Consequently, we are
normal distance of 9 mm is found between two surfaces. currently working on the application of physically based
Further inquiry on correspondence between surfaces modeling techniques for discrete thin shell and elastic
would be to inspect their local curvature properties, the rod simulation, to approach a more interactive, finite
mean, and Gaussian curvature. These local properties strip simulator with an aim to integrate it into a physi-
on three-dimensional mesh surfaces can be estimated cally based NURBS/CAD tool for active bending form
using the Algebraic Point Set Surface (APSS) approxi- exploration.
mated curvatures of the best fitting spheres to the mesh
point set. 52
Results for theses approximated curvatures are
respectively shown in figs. 17 a and 17 b and indicate a
good correlation between numerical and experimental
results for both absolute value and distribution. The esti-
mated value for the Gaussian curvature takes non-zero

Acknowledgments 18Adriaenssens, S. and M. Barnes. “Tensegrity Spline Beam
and Grid Shell Structures.” In Engineering Structures, 23(1),
This research work was sponsored by the Swiss National
29–36, 2001.
Science Foundation (SNSF) under grant No. 20002_137884/1.
This support is gratefully acknowledged. 19 Ye, J., R. Feng, and S. Zhou. “The Modified Dynamic Relaxa-
tion Method for the Form-Finding of Membrane Structures.”
In Advanced Science Letters, 4(810), 2845–2853, 2011.
References 20 Bel Hadj Ali, N., L. Rhode Barbarigos, and I. Smith. “Analy-
1Hudert, M. “Structural Timber Fabric: Applying Textile sis of Clustered Tensegrity Structures using a Modified
Principles in Building Scale.” PhD Thesis, IBOIS, EPFL, Dynamic Relaxation Algorithm.” In International Journal of
Lausanne, 2012. Solids and Structures, 48(5), 637–647, 2011.
21 Rodriguez, J., G. Rio, J. Cadou, and J. Troufflard. “Numerical
2Weinand, Y., and M. Hudert “Timberfabric: Applying Textile
Principles on a Building Scale.” In Architectural Design, 80(4), Study of Dynamic Relaxation with Kinetic Damping Applied
102–107, 2010. to Inflatable Fabric Structures with Extensions for 3-D Solid
Element and Nonlinear Behavior.” In Thin Walled Structures,
3Sistaninia, M., M. Hudert, L. Humbert, and Y. Weinand. 49(11), 1468–1474, 2011.
“Experimental and Numerical Study on Structural Behavior
22 Douthe, C. and O. Baverel. “Design of Nexorades or
of a Single Timber Textile Module.” In Engineering Structures,
46(0), 557–568, 2013. Reciprocal Frame Systems with the Dynamic Relaxation
Method.” In Computers & Structures, 87(21), 1296–1307,
4 Otter, J. and A. Day. “Tidal Computations.” In The Engineer,
2009.
1960.
23 Joldes, G., A. Wittek, and K. Miller. “An Adaptive Dynamic
5 Day, A. “An Introduction to Dynamic Relaxation.” Relaxation Method for Solving Nonlinear Finite Element
In The Engineer, 219, 218–221, 1965. Problems. Application to Brain Shift Estimation.”
6 Otter, J., A. Cassell, R. Hobbs, et al. “Dynamic Relaxation.” In International Journal for Numerical Methods in Biomedical
In Inst. Civ. Engrs. Proceedings, vol. 35, 633–656, 1966. Engineering, 27(2), 173–185, 2011.
7 Otter, J. “Computations for Pre-Stressed Concrete 24 Brew, J. and D. Brotton. “Nonlinear Structural Analysis by
Reactor Pressure Vessels using Dynamic Relaxation.” Dynamic Relaxation.” In International Journal for Numerical
In Nuclear Structural Engineering, (1), 61–75, 1965. Methods in Engineering, 3(4), 463–483, 1971.
8 Frankel, SP. “Convergence Rates of Iterative Treatments 25 Wood, WL. “Note on Dynamic Relaxation.” In International
of Partial Differential Equations.” In Mathematical Tables and Journal for Numerical Methods in Engineering, 3(1), 145–147,
Other Aids to Computation, 4(30), 65–75, 1950. 1971.
9 Frieze, P., R. Hobbs, and P. Dowling. “Application of Dynamic 26 Papadrakakis, M. “A Method for the Automatic Evaluation
Relaxation to the Large Deflection Elasto-Plastic Analysis of of the Dynamic Relaxation Parameters.” In Computer Methods
Plates.” In Computers & Structures, 8(2), 301–310, 1978. in Applied Mechanics and Engineering, 25(1), 35–48, 1981.
10 Pica, A. and E. Hinton. “Transient and Pseudo-Transient 27 Underwood, P. “Dynamic Relaxation (in Structural
Analysis of Mindlin Plates.” In International Journal for Numeri- Transient Analysis).” In Computational Methods for Transient
cal Methods in Engineering, 15(2),189–208, 1980. Analysis, Amsterdam, 245–265, 1983.
11 Pica, A. and E. Hinton. “Further Developments in Transient 28 Cassell, A. and R. Hobbs. “Numerical Stability of Dynamic
and Pseudo Transient Analysis of Mindlin Plates.” In Interna- Relaxation Analysis of Nonlinear Structures.”
tional Journal for Numerical Methods in Engineering, 17(12), In International Journal for Numerical Methods in Engineering,
1749–1761, 1981. 10(6), 1407–1410, 1976.
12Ramesh, G. and C. Krishnamoorthy. “Geometrically 29 Alwar, R., N. Rao, and M. Rao. “An Alternative Procedure in
Nonlinear Analysis of Plates and Shallow Shells by Dynamic Dynamic Relaxation.” In Computers & Structures, 5, 271–274,
Relaxation.” In Computer Methods in Applied Mechanics and 1975.
Engineering, 123(1), 15–32, 1995. 30 Tong, P. and Ty Tsui. “Stability of Transient Solution of
13Kommineni, J. and T. Kant. “Pseudo-Transient Large Moderately Thick Plate by Finite Difference Method.”
Deflection Analysis of Composite and Sandwich Shells with a In AIAA Journal, 9(10), 2062–2063, 1971.
Refined Theory.” In Computer Methods in Applied Mechanics 31 Rezaiee-Pajand, M., M. Kadkhodayan, and J. Alamatian.
and Engineering, 123(1), 1–13, 1995. “Timestep Selection for Dynamic Relaxation Method.”
14 Ramesh, G. and C. Krishnamoorthy. “Postbuckling Analysis In Mechanics Based Design of Structures and Machines, 40(1),
of Structures by Dynamic Relaxation.” In International Journal 42–72, 2012.
for Numerical Methods in Engineering, 36(8), 1339–1364, 1993. 32 Alamatian, J. “A New Formulation for Fictitious Mass of
15Papadrakakis, M. “Postbuckling Analysis of Spatial Struc- the Dynamic Relaxation Method with Kinetic Damping.”
tures by Vector Iteration Methods.” In Computers & Structures, In Computers and Structures, 9091(1), 42–54, 2012.
14(5), 393–402, 1981. 33 Rezaiee-Pajand, M., S. Sarafrazi, and H. Rezaiee. “Efficien-
16 Alamatian, J. “Displacement-based Methods for Calculat- cy of Dynamic Relaxation Methods in Nonlinear Analysis
ing the Buckling Load and Tracing the Postbuckling Regions of Truss and Frame Structures.” In Computers and Structures,
with Dynamic Relaxation Method.” In Computers and Struc- 112113, 295–310, 2012.
tures, 114115, 84–97, 2013. 34 Rezaiee-Pajand, M., M. Kadkhodayan, J. Alamatian, and
17Barnes, M. “Form Finding and Analysis of Tension Struc- L. Zhang. “A New Method of Fictitious Viscous Damping
tures by Dynamic Relaxation.” In International Journal of Determination for the Dynamic Relaxation Method.”
Space Structures, 14(2), 89–104, 1999. In Computers and Structures, 89(910), 783–794, 2011.
175
35 Rezaiee-Pajand, M. and S. Sarafrazi. “Nonlinear Dynamic Analysis and Design, 40(11), 1551–1569, 2004.
Structural Analysis using Dynamic Relaxation with Zero 45 Arciniega, R., and J. Reddy. “Tensor-Based Finite Element
Damping.” In Computers and Structures, 89(1314), 1274–1285, Formulation for Geometrically Nonlinear Analysis of Shell
2011. Structures.” In Computer Methods in Applied Mechanics and
36 Rushton, K. “Large Deflection of Variable Thickness Engineering, 196(4), 1048–1073, 2007.
Plates.” In International Journal of Mechanical Sciences, 10(9), 46 Buechter, N. and E. Ramm. “Shell Theory Versus
723 –735, 1968. Degeneration a Comparison in Large Rotation Finite Element
37 Chung, W., J. Cho, and T. Belytschko. “On the Dynamic Analysis.” In International Journal for Numerical Methods in
Effects of Explicit Fem in Sheet Metal Forming Analysis. Engineering, 34(1), 39–59, 1992.
Engineering Computations.” In Journal for Computer Aided 47 Simo, J., D. Fox, and M. Rifai. “On a Stress Resultant
Engineering, 15(6), 750–776, 1998. Geometrically Exact Shell Model.” In Computer Methods in
38 Cassell, A., P. Kinsey, and D. Sefton. “Cylindrical Shell Applied Mechanics and Engineering, 79(1), 21–70, 1990.
Analysis by Dynamic Relaxation.” In: Inst. Civ. Engrs. 48 Jiang, L. and M. Chernuka. “A Simple Four-Noded
Proceedings, vol. 39. 75–84, 1968. Co-Rotational Shell Element for Arbitrarily Large Rotations.”
39 Lynch, R., S. Kelsey, and H. Saxe. “The Application of In Computers and Structures, 53(5), 1123 –32, 1994.
Dynamic Relaxation to the Finite Element Method of 49 Besl, P. and N. McKay “A Method for Registration of 3-D
Structural Analysis.” Tech. Rep.; Technical report No. Shapes.” In IEEE Transactions on Pattern Analysis and Machine
THEMISUND681, University of Notre Dame, 1968. Intelligence, 14(2), 239–256, 1992.
40 Hughes, T. “Analysis of Transient Algorithms with 50 Kazhdan, M., M. Bolitho, and H. Hoppe. “Poisson Surface
Particular Reference to Stability Behavior.” In Computational Reconstruction.” In Proceedings of the Fourth Euro-graphics
Methods for Transient Analysis, Amsterdam, 67–155, 1983. Symposium on Geometry processing. Euro-graphics Associa-
41 Belytschko, T. “An Overview of Semi-Discretization and tion, 61–70, 2006.
Time Integration Procedures.” In Computational Methods for 51 Botsch, M. and L. Kobbelt. “A Remeshing Approach to
Transient Analysis, Amsterdam, 1–65, 1983. Multi-Resolution Modeling.” In Proceedings of the 2004
42 Courant, R., K. Friedrichs, H. Lewy. “über die Partiellen Eurographics/ACM SIGGRAPH symposium on Geometry Process-
Differenzengleichungen der Mathematischen Physik.” ing. ACM, 185–192, 2004.
In Mathematische Annalen, 100(1), 32–74, 1928. 52 Guennebaud, G. and M. Gross. Algebraic Point Set Surfaces.
43 Tabiei, A. and R. Tanov. “A Nonlinear Higher Order Shear ACM Trans Graph, 26(3), 2007.
Deformation Shell Element for Dynamic Explicit Analysis.” In
Finite Elements in Analysis and Design, 36(1), 17–37, 2000.
Reprinted from International Journal of Sapce Structures,
44Sze, K., X. Liu, and S. Lo. “Popular Benchmark Problems for vol. 28, No. 3 & 4, 2013.
Geometric Nonlinear Analysis of Shells.” In Finite Elements in

177
4. 5
„These programs don’t provide easy,

off-the-shelf solutions”
Interview between Olivier Baverel (Laboratoire Navier, Ecole nationale des Ponts et Chaussées, Paris)
and Yves Weinand
May 26, 2015
Yves Weinand: Olivier, we both come from the Y. W.: Rib shells are actively bending structures
academic world and we’re particularly interest- that fascinate us. We share this common passion,
ed in the practical and applied potential of our but I’m under the impression that there’s a large
research. However, I am under the impression step to be made in terms of convincing firms to
that our activities have not garnered the interest share our passion for bending structures.
of a large audience. We remain marginal in Shigeru Ban, for example, proposed this type of
combining research with practical applications. structure, but was unable to construct works
Together, we have already had the opportunity that are efficient or maximized from a structural
to address the question of active bending. In the perspective. These structures seem to evolve
classes I give at EPFL,1 I try to clarify the notion from a more formal conception, rather than
of active bending by presenting built examples from a structural perspective. In that sense, his
that have left their mark on the history of work remains modernist. It does not correspond
construction. Such as the structures that Colonel exactly to what we are attempting to discuss in
Emy2 designed for the French military, or the the fourth chapter of this book, namely: stronger
creation of glue-laminated wood. 3 I’ve noticed harmony between form and structure, and
that there are few applications of these types about structures that are actively flexed. Do you
of structures. What is your opinion about this? think there’s a way to motivate people around
these types of structures, to convince them
Olivier Baverel: Most likely it’s because we’ve not yet through objective arguments—for example, in
perfected our tools, and so long as they’re not user-friendly, relation to flexibility and in relation to seismic
we won’t have many users. It’s important to remember activity? Do you think it’s possible to convince
that the truss beam already appeared 500 or 600 years people in a pragmatic manner, and not from a
ago, but no one used it until the moment when mathe- kind of researcher’s idealistic motivation?
matical tools allowed for it to be better understood.
It was only around 1820 that we began to understand O. B.: Yes, that would be possible the day a designer has a
and began to be capable of predicting the behavior of tool in his or her hands that is easy to use. The relevance
a truss beam: fifty years later, the Eiffel Tower was of these extremely slim structures is to take exposed
erected. There’s a similar story in relation to Frei Otto beams made from small wooden sections, assemble
and metallic-textile structures, which until the 1970s, them on the ground and shape them to create a three-
existed solely to cover small spaces. From the moment dimensional form. These structures have a low environ-
we understood the theory and created tools to predict mental impact, as they don’t need a complex trans-
forms and their applications, large-spanning projects formation of wood, no bending or bonding, and little or
appeared, such as the Munich stadium in 1972. no material waste. There is currently nothing simpler
As far as we’re concerned, we’re close to having a final today than to make a lattice and set out the lattices
and predictive tool, but there’s still a lot of work to one next to the other. Why look for more structural
be done. Once these tools can easily be used by a sub- optimization when latticework already has tremendous
stantial group of engineers and architects, that’s when capacities? Lattice will consume a bit more material
I believe these structural typologies will fully emerge. than a ribbed shell, but today, in relation to the overall
Until now, if not for the last ten years or so, it was not cost of things, it’s barely significant. If we begin to
at all conceivable. Neither the digital tools nor the consider environmental costs more seriously, and not
calculation capacity were sufficiently developed, solely financial costs, these slim, optimized structures
from either a theoretical or mechanical point of view. that don’t involve excessive transformations and impor-
We’re not there yet, but we’re getting closer. tant losses would become truly worthy of interest.

In parallel, these last fifteen years have seen the rise of O. B.: There are two answers to your question: the first is
digital tools like Archicad or Vectorworks in architecture. that we need to employ young people who will use Sina
De Casteljau and Bézier, engineers at Citroën and Nabaei’s tool, take it further, and potentially create a
Renault, developed algorithms called NURBS for the large-scale model at the EPFL or at the Ècole des Ponts to
shaping of car bodies. With embossing techniques, using demonstrate the capacity of these algorithms.
a thirty-ton press, any kind of metal plate can be shaped. You also need to have some luck and find the opportuni-
However, these tools have not yet been adapted to ty. If we look at the gridshells created by Frei Otto, it was
our industry, which is primarily based on the assembly pure chance that a flower exhibition at Mannheim ne-
of different metals and not by embossing a shape out cessitated a construction over a sixty-meter range. There
of a forty-by-forty-meter piece of metal. We have spent was sufficient intellectual impetus for it to be produced,
the last fifteen years using a tool that is a poor example and enough luck to make it happen. Insofar as the Sina
of what we should be fabricating. This use of this in- Nabaei tool is concerned, we are not mature enough yet.
appropriate tool for the task means a fairly significant First and foremost, Sina Nabaei designed a complex
over-consumption of material in many contemporary engineering tool of a computer scientist. We would need
examples. This over-consumption is testimony to a lack to build a prototype at the university level with a range
of capacity in representing the elements that need to of about ten meters. Then, we would need the opportuni-
be fabricated. Thanks to tools like Rhino 4 and Grass- ty to test it in a real project. We are still missing these
hopper, 5 industry has progressed from that perspective. two stages. We had a few opportunities with our grid-
But there’s still a lot of work to do in order to have tools shell structures, which we constructed out of composite
that are completely adapted to our industry. The rep- materials at the Ècole des Ponts. This really showcased
resentational capacity means that structures such as our representational capacities and know-how in terms
ribbed shells will no doubt be far more interesting than of calculations. We had two concrete opportunities, one
classical structures in the coming years. Remember that for a festival and the other for an ephemeral cathedral
to build an airplane, Dassault Systèmes proposes a suite (a temporary structure), using composites. We had to
of more than 150 programs; but only five years ago only deal with the French planning office for this large-scale
two or three programs were available to help in the construction, who are very competent, but to whom we
construction of a building. We were lagging very far had to explain our structure. At that time there was no
behind. Today, diversification is taking place on many official Eurocode for composite materials. We therefore
levels thanks to programs like Rhino and Grasshopper, took the double risk of designing an atypical structure
and associated plug-ins. with a material that had never before been used in
our industry. Thankfully, the contractor and auditor
Y. W.: I feel the time is ripe for considering appli- were supportive and it was a success.
cations; the tool developed by Sina Nabaei 6 for
example, but also for the application of actively Y. W.: Many people seem to be fascinated by
flexed structures. The NCCR7—where we are braided structures, but this fascination seems
active—could allow for it, though we lack engi- limited to the aesthetic perspective. The struc-
neers to be able to take Sina Nabaei’s work fur- tural advantages of this type of construction
ther. His work only includes one practical appli- are not perceived, nor are they commented on.
cation to date (a spider form). This sole example We have not yet been able to create a wider
is not sufficient for illustrating the tremendous acceptance for this type of structure. It seems to
potential of this tool. I’m under the impression me that the practical application proposed by
that, at that level, very few people would be Sina Nabaei is still somewhat awkward or unfin-
capable of helping us at the moment. ished in its global form, and that it would be
179
necessary to add an architectural aspect to it. powerful lobby, when compared to the producers of
I believe that we should deploy this type of other materials where the lobby is well organized. There
structure quickly and efficiently in earthquake- is true production potential, but there are few large
prone areas, or in response to natural disasters, sawmills, compared to Austria, for example. Small enter-
for example, rather than aiming for a specific prises like Simonin do their best and are keen to inno-
commission. Rather than building a pavilion, vate, but they don’t have the same capacity as the large
a pragmatic application would no doubt be very concrete or steel factories.
useful towards bringing these structures into
the mainstream, in terms of how they are under- Y. W.: From an academic perspective, I have tried
stood in relation to their structural efficiency. to motivate engineering students to be interest-
ed in the global form. Should we consider trying
O. B.: One has to pay attention to the goal—with the to attract these types of students, specifically to
concept of a pavilion: some are more focused on rep- further develop Sina Nabaei’s tool and to make it
resentation, while others are more interested in the easier to understand? Do you have any ideas or
structure. Certainly if we move towards pure aesthetics, proposals in relation to this?
interest will quickly wane. You have to clearly demon-
strate the structural advantage—or at least the combina- O. B.: Engineers lack a culture of structures and often
tion of technological and structural advantages—as remain far from the line of research we seek to advance.
these are always the two pillars on which we base our They can’t understand because they’re not taught these
constructions. This work remains to be undertaken. aspects during their training. In general, the engineer
It’s about creating a prototype that has an architectural calculates the structure and justifies it through the
function to be determined, an interesting rendering, Eurocode, but doesn’t question the form. I believe this is
but above all a true mechanical capacity, whether at the a serious error in relation to how work is organized.
level of the assembly typology chosen, or for the zones At the EPFL, at the École des Ponts, and at other universi-
with high risk of seismic activity, or simply thanks to ties, there are courses on conceptual design and struc-
the mesh effect that is placed between the slat layers. tures that should be further developed and taken up by
With Sina Nabaei’s thesis, we have a tool and a working other institutions. It’s also about sensibilities: for an
method, but we have not developed further in demon- engineer, there’s nothing more comforting than making
strating how this type of typology is relevant in terms of calculations and applying rules such as Eurocode. Of
mechanical behavior. With a good design team, it would course taking things further, questioning and innovat-
be possible to have an extraordinary result, with true ing, is bound to be more complicated. Our engineering
added value, where there is a reasonable consumption of students are primarily recruited for their mathematical
materials and an implementation of technology that skills. An emphasis on curiosity and innovation should
would no doubt be fairly straightforward. also be integrated into their curriculum.
Y. W.: Are there any industrial partners in France Y. W.: But if we consider that the IASS 8 is organiz-
for developing this technology? ing an event in Amsterdam this summer, where
pavilions will be built, I’m under the impression
O. B.: As you know, France has an annual production rate that these pavilions will be consistent with the
of around 65 million cubic meters of timber. That’s more representation of form, rather than remaining a
than a cubic meter per person, but unfortunately the purely mechanical problem. In terms of final
system is not well organized. There are 3.5 million pri- form-finding, I believe Sina Nabaei’s tool is very
vately-owned forests that are on average only 2.6 hec- strong. The depth achieved by his tool is not
tares in size. Currently, French timber does not have a rendered solely through the construction of a

pavilion. Freestanding structures are regularly Y. W.: When we met up at the IASS conference in
published, but rarely present their mechanical Venice in 2009, we noted that engineers are start-
characteristics and performances. Aspects like: ing to embrace forms. Indeed, the engineer
Is there friction or not? What was the original should not be indifferent or neutral to the form
situation of equilibrium? What does friction proposed by the architect. It is necessary for the
bring as added value? Is there friction? We easily engineer to take a stance in relation to this form.
talk about a synthesis between form and struc- He should be interested in this global form from
ture—morphogenesis—but we don’t detail the his own point of view. Certain engineers claim
mechanical performance. It therefore becomes that they can calculate any forms! This may be
auto representation. true, but today I consider this stance with some
skepticism. I prefer to look for work where there
O. B.: Yes, I have exactly the same impression. But this is a strong coherence between form and structure,
also comes from the fact that understanding structures as with Félix Candela’s work, for example, origi-
is complex, even with a simple, static diagram. People nally from the IASS. Certain new trends at IASS,
prefer to create an aesthetic form, but to move from which I see as weaknesses, don’t seek this synthe-
there and to take the effort to understand the structural sis that I aspire to, and to, which I would like to
behavior, to rationalize the construction, to limit the attract more adherents—engineers in particular.
losses…there are not very many people who think along
these lines. I saw this among my architectural students. O. B.: Yes, I agree. But then, let’s be prudent with our
There’s a plug-in in Grasshopper called Karamba, 9 which comparisons. If, for instance, we consider the IASS in the
enables you to do structural analysis calculations. 1970s, history has forgotten mediocre achievements:
All of the students were enthusiastic, but as soon as I told we only retain the significant ones. When it comes to
them it was about making mechanical structures, eva- the capacity to calculate all forms without limit, there’s
luating the supporting conditions, what is the load, what certainly a lack of culture on the part of the engineer
are the sections, what types of calculations should be on the critical question of form. As far as I’m concerned,
undertaken, etc., all of these questions are immediately when I discuss this issue with an architect, I try to
less “glamorous.” You have to ask yourself many understand the substance of the discourse around his
questions, you have to anticipate, examine, and these or her intentions, rather than solely the form. It is work
programs don’t provide easy, off-the-shelf solutions. that architects and engineers have to collaborate on.
What many people tend to forget is that the mechanics of What’s important is to convey the architect’s concept,
structures only began to be mastered during the past rather than the formal object he has rendered, recogniz-
180 years or so. Galileo posed the question: how to find ing that the tools he is using are not adapted for present-
the stresses in a cantilever beam? But he didn’t ing the form that he wants.
manage to find an answer. Galileo was a pioneering Incidentally, regarding representational tools, with
thinker who initiated an intellectual revolution Nurbs, it is practically impossible to make folds or to
throughout Europe, and yet he was unable to solve the create discontinuities in curves, or local singularities
simple question of the cantilever beam! This clearly on surfaces. In fact, architecture today is a prisoner of a
proves that it’s complicated and not necessarily accessi- formal universe, which does not represent all of the
ble to everybody. I’m not surprised by what you describe; possible forms, but only those of Nurbs. Today, there are
we see many things, but we don’t explain them suffi- many other solutions, which we don’t allow ourselves
ciently. In any case, it’s a problem; we sell the image to create, because we don’t have the proper representa-
without really understanding the content. tional tools. Those who believe they can represent
everything with Nurbs are fooling themselves. There’s
a lot of work to be done by the entire community to
181
develop new representational tools. Furthermore, we Y. W.: This is precisely what we have defined in
also began working on a thesis along this theme two our upcoming research topics. I’m fascinated by
years ago. We are advancing well and are capable of the fact that reinterpretation of Japanese joinery
describing formal realms with particular, “non-Nurb” may have an effect on the way future logistical
structural typologies. (…) concepts in the wood industry evolve.
We can continue our discussion with the massive impact
of robots in architectural production. I would like your
opinion on this. You have several ongoing theses on this
topic, is that correct? References
1“Conception de structures” (Structure conceptions),
Y. W.: Yes, but I’m not convinced that all systems 3rd-year bachelor, architecture section, EPFL.
have to be additive systems. We also know 2 Traité de l’art de la charpenterie (Treatise on the art of
that robots are imprecise. There are sequential carpentry), Colonel Emy, 1878.
aspects that must be respected. I’m interested 3 Otto Hetzer, 1906.
in advanced building processes in architecture. 4 Rhino is a 3-D drawing program for graphic design and
These processes do not only relate to robotics, modeling.
but to all technologies that are available to 5 Grasshopper® is a graphic editor of 3-D algorithms, entire-
serve the construction and architecture markets. ly integrated with Rhino modeling tools. Unlike RhinoScript,
We would like to actively fold panels by using Grasshopper does not require programming
or scripting knowledge, but allows designers to build form
several robots, but I’m not convinced that should
generators from the simple to the grandiose.
be done by robots. We can test it, but we’ll even-
6 Sina Nabaei. “Mechanical Form-Finding of Timber Fabric
tually find other means. The history of ribbed Structures.” Thesis n° 6436, EPFL, Lausanne, 2014.
shells demonstrates that there are other ways to 7 The National Centre of Competence in Research (NCCR)
approach this. What interests me is the follow- Digital Fabrication—Advanced Building Processes in Archi-
ing question: How to interest industrial partners tecture is hosted at the ETH Zurich, directed by
in this type of structure, generally speaking, in Matthias Kohler, in collaboration with 13 laboratories—
order to advance further? 8 laboratories from ETHZ, 2 laboratories from EPFL, 2 from
EMPA, and 1 from BUAS. Included is the Laboratory for
Timber Construction, IBOIS, directed by Prof. Yves Weinand.
O. B.: In my opinion, for that you would need to look for a
8 IASS 2015 Annual International Symposium on Future
long-term partner. Apropos robots, I’m lucky enough Visions, 17–20 August, 2015: The continuous development of
to be mentoring a young researcher who is developing a design, analysis, and construction techniques for the
robot manipulation tool from Grasshopper. The robots built environment and shell and spatial structures.
will help us to think about how to produce complex http://www.iass2015.org
structures today. A new technological tool will appear 9 Developed by Bollinger + Grohmann engineers.
that will be taken up by some like a religion, but will be
considered an enemy by others. The thesis is about creat-
ing concrete formwork systems with a very particular
robot technology, which would not be feasible without a
robot. I think there are many techniques that have been
abandoned, as they were too time-consuming and ex-
pensive, but with robotics, we can once again ask ques-
tions about their feasibility. For example, we can revisit
Japanese wood joinery and build with wood without
using any metallic connectors at a more reasonable cost.

183
4. 6
Braided structures:
applying textile principles at an architectural scale
Marielle Savoyat
Design IBOIS—Laboratory
IBOIS, Labor für Holzkonstruktionen,
und EPFL, The research undertaken between 2007 and 2013 by
EPFL,
Eidgenössische
Swiss Federal
Technische
Institute
Hochschule
of Technology, Markus Hudert within IBOIS, the Laboratory for Timber
Lausanne, Switzerland
Schweiz Constructions at the Swiss Federal Institute of Techno-
Prof. Yves Weinand and
und Dr. Markus Hudert logy, analyzes the use of textile techniques at an archi-
(researcher)
(Wissenschaftler) tectural scale. It soon became apparent that principles
of knitting, braiding, and weaving offer great potential
Research 2007–2013
and for varying structural possibilities when applied to the
completion scale of architecture.
The common denominator between all of these
textile techniques is one basic element: that of a thread
interlaced with another thread. This starting principle
can be transferred onto two interlaced planks of wood.
To put this concept into practice, a first prototype called
a textile module was created, which demonstrated how
the application of a textile technique, when combined
with the properties of wooden material, could lead to a
particularly efficient freestanding structure.
Fig. 1 Axonometry of a braided arch

Fig. 2 Underview of the model
185
The geometry of the whole was automatically conditioned Textile principles are therefore of interest both in struc-
by the assembly techniques. What’s more, when pressure tural and architectural terms. By twisting the material,
is exerted on the structure,1 it lengthens and becomes internal tensions are created and stiffening becomes ap-
narrower, while the middle section of the basic module parent. The properties and the aspect created by interlac-
gains in height, thus making the structure more rigid. ing elements are not only highly aesthetic, but also em-
Indeed, superimposing two thin wooden planks and body great structural strength and high potential for
vertically aligning the screw holes cutout at each extrem- spatial quality.
ity automatically generates the curvature without rely- The coherence between structure and space re-
ing on a mold. A structural resistance becomes apparent. sults in high quality architecture. It is important to note
To obtain structural capacity, the finer layers have to be that these structures remain limited to single-story build-
fixed together. ings. The question remains: How can these structures be
One of the structural advantages of textile is that roofed over to further their architectural form? The link-
it is composed of a great number of simple, interconnect- ing elements could also hold a covering membrane to-
ed elements, which work together to form a whole. Thus, a gether. A great deal of further exploration remains open,
weakness in one of the elements does not affect the struc- though these first research endeavors hold great promise
ture in its entirety in any way. In order to create the same and are an excellent spring board.
effect at an architectural scale, the creation of a stable
structure composed of a multitude of elements also be-
comes essential, so that the deformation of certain ele- Reference
ments would not compromise the entire building. 1 Hudert, M. “Timberfabric: Applying Textile Assembly Princi-
Markus Hudert’s research was applied to explo- ples to Wood Construction in Architecture.”
ring the ways that the textile module could be used as a Thesis no. 5553, EPFL, Lausanne, 2007.
basic unit in a large-scale structure. By aligning several
modules in the same axis, a structure in the form of an arc
is created. A sequence of several of these arcs then pro-
duces a vault-structure. The disadvantage of this approach
resides in the fact that the modules remain independent
from one another. To obtain large structures, connections
would need to be added to the chain mail at the upper and
lower points.
Fig. 3 Fig. 4

Fig. 3 Test procedure
Fig. 4 Spatial definition of stress points
Fig. 5 Large module under compression test
The module on show during the Timber Project
Fig. 6
exhibition at EPFL
Fig. 5
Fig. 6
187
5 Customized construction
5. 1 Innovative wood–wood connections 190

Yves Weinand
5. 2 Interlocking folded plates—integral mechanical

attachment for structural wood panels 200
Christopher Robeller and Yves Weinand
5. 3 Rotational stiffness on the ridges of timber

folded-plate structures 210
Stéphane Roche, Geoffroy Mattoni, and Yves Weinand
5. 4 “Digital fabrication leads to a new building culture” 226

Interview between Matthias Kohler and Yves Weinand
5. 5 Curved-panel wood pavilion 234

Marielle Savoyat
189
5. 1
Innovative wood–wood connections

Yves Weinand
The technology of the connections is es- dovetail in the European tradition, in the
sential in all design processes. In timber Japanese tradition they are inserted
construction, half the costs are deter- diagonally across the dovetail. In addition
mined by the volume of material used to the insertion angle, of interest is the
and the other half by the installation and principle of different types of integral
connection costs. Unlike structural sys- connections, which are determined by
tems of reinforced concrete or steel, timber different geometric boundary conditions.
construction requires a greater focus on
integrated planning. For this integrated If the connection has geometric properties,
planning process, the preselected geo- then the following consideration can be
metry of the connection technology has to added: assembly sequences can be defined
be incorporated into the planning process that cause specific elements in specific
in advance. In timber construction, places to be locked at any given moment
connection techniques are transmitted of the installation phase. Depending on
by adhesive bonds (chemical forces), the sequence definition, so-called “lock
by welded connections (physical forces), graphics” arise that describe the locking
and by mechanical or integral mechanical of an element in the overall system.
connections (mechanical forces).
Building on this accumulated knowledge,
Regarding the latter category, with integ- a number of avenues of research were
ral mechanical connections such as undertaken at IBOIS that examined the
dovetail joints, Christopher Robeller has relationship between the overall geo-
demonstrated that European dovetail metry, the assembly, and the connection
connections can be distinguished from typology of the individual joints. The
Japanese equivalents by the selected slid- investigations focused on the insertion
ing layer. While galvanized wood–wood angle, the number of degrees of freedom,
connections are inserted alongside the tool-related conditions for milling, and

Fig. 2
a)
Fig. 1
b)
faces along edge faces across edge only faces

across edge
a) b)
c)
Fig. 3
v
v
Fig. 4
P5
P2
P1 v P5 Fig. 1 Folding structure developed with the
v origami tool
Fig. 2 Failure mechanism of the shell:
z opening of bolted joints and connections
y P4 P1 P2 Fig. 3 a) Steel corner connections, two orthogonal
x plates b) Screwed corner connection of two
P3 orthogonal plates c) Screwed corner connection
of two cut plates with angle variations
v
Fig. 4 Diagram of a sliding vector
P4 P3 a) European tradition b) Japanese tradition
Fig. 5 Fig. 5 Lock diagram
191
S1
E1
S2
E2 S1 S2 S3
S3
E3
Fig. 6 b
Fig. 6 a, b, c, and d Summary based on research at multiple scales undertaken

at IBOIS
Fig. 6 a Fig. 6 c
Fig. 6 d

the mechanical properties of the inno- Locally defined geometric constraints lead
vative construction. By combining struc- to connection types, which in turn affect
turally strong veneer plywood panels and the shape as a whole. Civil engineers are
efficient wood–wood connections within interested in the overall form and how the
predefined spaces and where the insertion detailed design of elements of the overall
angle for one or more panels is provided, form can lead to “integrated” structures,
new sustainable timber structures can be where the relationship between form and
realized. Ideally, these should consist structure is tight and inextricably inter-
exclusively of wood, which also simplifies connected. This “woven quality” gives the
the future sorting process for disposal. formed structures a special beauty.
In retrospect, a new structural design Tribute to Eladio Dieste

concept emerged. Usually, a form is first In numerous constructions, Uruguayan
proposed and then discretized or divided, architect and civil engineer Eladio Dieste
and finally an overall shape is proposed, (1917–2000) demonstrated how form and
generally by the architect. The size structure can function efficiently together.
of each element and the nature of the For his department store in Montevideo,
connection of the elements are then, Dieste envisioned the assembly of a double-
however, usually determined by the en- curved shell structure using discrete
gineer. The subdivision and connection elements—in this instance, bricks. Steel
methods are therefore determined reinforcement was inserted into the joints
by whoever designs the overall form. to absorb the tensile stress. The transfor-
mation of the cross-section of an arc from
The assembly process is ultimately addi- a line to an S-shaped cross-section in the
tive in nature. The overall shape is gene- apex of the arc made the double curva-
rated by the connections of the individual ture necessary. The S-shaped cross-section
elements and the mechanical connections. provides an increased moment of inertia
The realization of nonstandard architec- and thus a higher buckling resistance at
ture of this kind generally fails due to the apex of the arch.
high costs. However, when the conception
of structures is undertaken as described The manufacture of the double-curved
here, the process is reversed. Considera- surfaces also poses a challenge in timber.
tions are not regarded in separate, In the project shown here, Dieste’s bricks
successive phases—first global and then were replaced by laminated veneer
local. Rather, the interaction of both lumber panels. The reinforcing steel was
the detailed scale of the connections and replaced with dovetail connections, which
the global scale of the overall form are bind the boxes together. A fundamental
considered and evaluated simultaneously. geometric difference is that the curvature
193
Fig. 7 b
z D/2
x
Ti
y
ni–1
T i+1
ni+1 ni–1
ntool, i+1
Li
ni+2
L i+1
Ln Tn TCP ntool pnotch

Fig. 7 a Fig. 7 c D/2
Milling, geometric description of the

Fig. 7 a–c
machine path that determines the angle of the
cutter head to the plate
Fig. 8 The time required for the production of
geometrically complex shuttering decreases:
a) single-shell element with its pre-milled wood–
Fig. 8 a wood connections; b) shell structure; c) complete
molded fiber-reinforced concrete shell
Development of a concrete form to a timber–
Fig. 9
concrete shell. The idea of a completely maximized
surface structure is formed by two superimposed
layers of wood and concrete, which achieve the
necessary adhesion solely due to geometric
constraints. Since the wood–wood connections
along the edges of the plates penetrate into the
concrete, these assume the integrating function
typically provided by pins.
Fig. 10 Axonometric and section of the double-
curved shell proposed by Dieste.
Fig. 8 b Fig. 8 c

VL
ber L
RC
UHP
er lum
rete/
vene
co n c
ate d
lamin
Fig. 9
s-shaped
crown line
Z
y catenaries symmetry
x
profile
linear
bearing
fixed bearing
Fig. 10
of the Dieste hangar is achieved by vary- The subdivision of the arcs into individ-
ing the strength of the mortar joints. ual segments was generated by a plug-in
All the bricks have identical dimensions, and the tool Grasshopper. Subsequently,
but the joints compensate for the variati- all the planes were arranged in an ortho-
on of the curvature. Here, then, the di- gonal grid, with all surface ribs positio-
scretized frame construction must hold ned vertically. Thus, vertical planes could
the angular position locally between the be created that clearly demarcate the
frames both transversely and longitudi- start and end of each arc. Then the solver
nally. Thus, the frame geometry is twisted had to smooth all areas, as the design
in the local inclination. In total, twenty- was composed of flat panels. This plana-
three 6.5-m-wide sheets should be created rization amounted to calculating the
with a variable span of between 32 m optimum position of the flat surfaces in
and 52 m and a height attained of 13 m. the space, so that the various facets
The scope of the building is approxima- could follow the double curvature as
tely 170 m. closely as possible.
195
workshop courtyard workshop offices
Fig. 11 Plans and model:
architecture collective
Valentiny hvp architects
and Weinand student
bureau,
Floor plan, scale 1:1000,
Longitudinal section,
scale 1:1000
Fig. 12 Planarization of
the surfaces
Axonometric
Fig. 13 a)
and parameterized
surface of the plug-in
system b) Milling code
c) Installation of a
prototype at IBOIS
Fig. 11

First, a system was developed that, as a
plug-in system, required a sequential
assembly. A frame does not exist by itself,
but rather abuts the next frame, so that
at least one lateral surface belongs to two
frames (or extends beyond the frame and
is integrated into the next). In this man-
ner, a woven system is created. However,
a disadvantage was that the modules Fig. 12
were not prefabricated, but rather the

structure was created by a continuous plug-
in system, which could also be classified
as a reciprocal assembly method.
From this, the idea arose of combining

both requirements. Thus, on the one hand
a modular system was developed that
Fig. 13 a
simplified the assembly process, and on
the other hand an interwoven structural
system was developed. The overall geo-
metry is subdivided again, not in recti-
linear modules but rather on the basis of
a subdivision principle reminiscent of a
herringbone pattern. The vertical planes
remain. The newly selected subdivision
principle affects the subdivision in plan.
Fig. 13 b
The insertion angle is no longer perpendi-
cular to a side surface, but continues dia-
gonally, so that two side surfaces of a
frame are simultaneously affected by it.
This diagonal insertion angle is reminis-
cent of the Japanese tradition of dovetail
connections presented above, where
the insertion angle is selected beyond the
orthogonal space. This measure makes it
possible to combine the rational principle
of serial assembly of modules practiced in
Fig. 13 c
197
the West with the Eastern principle of
interlocking the modules. The interlo-
cking is significant, as both tensile and
compressive membrane forces are
absorbed into the double-curved shell.
In the interlocking here, local displace-
ment or disengagement in relation to the
adjacent frame is prevented.
arch no. 1
Parametric design tool to generate the
Fig. 14
herringbone pattern
Fig. 15 Parametric design tool generates a pat-
tern similar to the herringbone for the
subdivision system of the frames. The diagonal
insertion angle relates simultaneously to at
least two sides of the same frame.
Fig. 16 Visualization of the herringbone pattern
using the example of three primary arches
rectangles
Fig. 17 Visualization of the interlocked frames.
The degree of freedom of the interlocking must
be subordinate to the global plug-in sequence.
Fig. 18 Frame prototype images shown at scale
1: 1. Two frames are connected or interlocked
here.
Mechanical test body. Here, the dovetails
Fig. 19
rhombi are tested in shear.
View of the shell construction from
Fig. 20
above and below
Fig. 14

1. definition of a module by profile curves
2. NURBS-surfaces
3. segmentation using
the IBOIS-tool
4. average size of the

elements
5. largest element
Fig. 15 Fig. 16
Fig. 17 Fig. 18
Fig. 19
Fig. 20
199
5. 2
Interlocking folded plates—integral mechanical

attachment for structural wood panels
Christopher Robeller and Yves Weinand
Automatic
Die automatisierte
joinery hasHerstellung
become a von common Holzverbindun-
technique 1 Introduction
gen
for thehat jointing
sich im ofHolzrahmenbau
beams in timber bereits
framing
etabliert.
and
Damit
roofing.konnten
It has revived
althergebrachte
traditional,Zimmermannstech-
integral joinery, Architectural designs are often inspired by folded shapes
such
nikenas wie
mortise-and-tenon
die Zapfenverbindungenconnections.
wiederJustneu rbelebt
ecent- such as Japanese origami; however, folding principles like
werden.
ly, the automatic
Doch erstfabrication
seit Kurzemoflassen
traditional
sich diese
cabinet-
tra- this can rarely be directly applied to building structures.
ditionellen
making joints Verbindungen
has been introduced
aus dem Tischlerhandwerk
for the assembly Rather, many folded plates have been cast as concrete thin-
of
auchtimber
für Schalenkonstruktionen
panel shell structures. ausFirst
Holzplatten
prototypesma- shells in the 1960s. These constructions were labor inten-
schinell
have been herstellen.
assembled Es entstanden
with such integrated
erste Prototypen
joints sive and required elaborate formwork for in-situ casting.
for
mit solchen
the alignment
integrierten
and Verbindungen,
assembly of components,
die auch als Prefabricated constructions with discrete ele-
Positionierhilfe
while additional beiadhesive
der Montage bonding
der Bauteile
was used
dienten;
for ments made from fiber-reinforced plastics were researched
eine
the load-bearing
zusätzliche Verklebung
connection.verstärkte
However, gluedderenjoints
Trag- in the 1960s.1
cannot
wirkung.beJedochassembled
könnenon Klebeverbindungen
site due to lack ofnicht con- Folded plates built from laminated timber panels
vor
trolled
Ort conditions,
hergestellt thus
werden,
resulting
woraus in several
sich wiederum
design were presented by Regina Schineis19 (Glulam) and Hans
diverse
constraints.
gestalterische Einschränkungen ergeben. Ulrich Buri 2 as cross-laminated timber. These designs
In this
diesempaper,Beitrag
we propose
stellenthe wir
use Schwalben-
of dovetail combine the elegant and efficient shape of folded plate
schwanz
joints without
verbindungen
adhesiveohne
bonding
Verklebung
in the case
vor, die
study in shells with the advantages of structural timber panels,
einer
of a timber
Fallstudiefolded-plate
an einem structure.
Holzfaltwerk Through
untersucht
their such as CO2 storage, and a favorable weight-to-strength
single-degree-of-freedom
wurden. Da sie aufgrund ihrer (1DOF)
Geometrie
geometry,nur these
einen ratio. However, a major challenge in the design of a tim-
Freiheitsgrad
joints block the besitzen,
relativelassen
movement dieseofVerbindungen
two parts in ber folded plate is presented by the joints: Since timber
die
all but
relative
one direction.
Bewegung This
derpresents
beiden Teile
the opportunity
in lediglich panels cannot be folded, a large number of edgewise joints
eine
for anRichtung
interlocking
zu. Somit
connection
ermöglichen
of plates,sie as
zwar
well eine
as has to provide two main functions. One of these functions
a
formschlüssige
challenge for the Verbindung
assemblyder of folded-plate
Platten, stellenshells,
je- is the load-bearing behavior, where connector features of
doch
wherefür multiple,
die Montage
non-parallel
der gefalteten
edges per Schalen
plate have
eine the joints have to provide sufficient stiffness and rigidity.
Herausforderung
to be jointed simultaneously.
dar, weil an jeder Platte mehrere, The second main function of the joints is the assembly of
nicht parallele Kanten gleichzeitig ineinandergefügt the parts, where locator features of the joints are essen-
werden müssen. tial for the precise, fast positioning, and exact alignment
of the parts.
Keywords integral attachment, timber folded-plate Benjamin Hahn 5 examined the structural behav-
structures, digital fabrication, design for ior of the first timber folded-plate shell, which was built
assembly
Integrale Verbindung, Holzfaltwerke, from plywood and assembled with screwed miter joints.
digitale Fertigungsprinzipien, He concluded that the load-bearing performance could
Schalenkonstruktionen, Montage be improved significantly with a greater number of
resistant connections.
Inspiration for such improvements may be found
in integral mechanical attachment techniques, the oldest
known technique for the jointing of parts, where the
geometry of the parts themselves blocks their relative
movements.13 Such integrated joints have recently been
rediscovered by the timber construction industry. Since
1985, mortise-and-tenon joints have not been used in
timberframe and roof constructions.7 Only very recently,
integrated joints have also been proposed for the edge-

Fig. 1
qE q q
i
j+3 j+3
j+3
j+2 j+2
j+2
n0
j+1
n1 j+1 j+1
u 2 u1 w1
u3 w2 w3 v
L2 Xj
F+0 L3 L1
F0 p L0 F+1 p p
F-0 F1
a) F-1 b) c) d) e)
Fig. 2
Fig. 1 Folded thin-shell prototype built from 21-mm LVL panels,

assembled with single-degree-of-freedom dovetail joints
wise joining of timber panels. In the ICD/ITKE Research without adhesive bonding. Components interlock with one
Pavilions 201112 and 201311, finger-joints have been ap- another.
plied to plywood panes and an application of dovetail Fig. 2 Joint geometry a) Basic parameters b) Intersection planes
joints for cross-laminated timber panels (CLT) was pre- (gray) normal to pq c) 3DOF joint d) Rotated intersection
planes (gray) normal to wj e) 1DOF joint
sented in the IBOIS Curved Folded Wood Pavilion 2013.18
In these prototype structures, the integrated joints have
played a vital role for the assembly of the components.
They have also contributed to the load-bearing connec- In this paper, we propose the use of dovetail joints with-
tion of the parts, though additional adhesive bonding out additional adhesive bonding, in the case study of a
was needed. With few exceptions,6 such glued joints timber folded-plate shell (Fig. 1).
cannot be assembled on site, because they require a curing Through their single-degree-of-freedom (1DOF)
period under controlled conditions, with a specific con- geometry, these joints block the relative movement of
stant temperature and humidity.15 Therefore, their ap- two parts in all but one direction. This presents the
plication is limited to off-site assembly of larger compo- opportunity for an interlocking connection of plates,
nents, which complicates both transport and handling as well as a challenge for the assembly of folded-plate
while still requiring additional connectors for the final shells, where multiple, non-parallel edges per plate have
assembly. to be joined simultaneously.
201
1.1 Dovetail joint geometry and mechanical
performance
Using polygon mesh processing, we describe an
edgewise joint based on its edge E. From the mesh connec-
tivity, we obtain the edge vertices p and q and the adja-
cent faces F0 and F1 with their face normals n0, n1. The
polygon mesh is used to represent the mid-layer of tim-
ber panels with a thickness t and offset F1 and F2 at ± –2t
to obtain the lines L (Fig. 2a). From a division of E, we obtain
the points Xj for a set of reference frames {u1, u 2 , u3 }, where
" and u | | n (Fig. 2b) . A finger-joint geometry is ob-
u1 | | pq 2 0
tained from an intersection of planes located at Xj, normal
to u1, with four lines L.
Without additional connectors, finger-joints are
a kinematic pair with three degrees of freedom (3DOF),
also called planar joints. They can resist shear forces
parallel to the edge and in-plane compressive forces. How-
ever, depending on the plate geometry, thickness, and
most of all rotational stiffness of the connection detail,
bending moments are also transferred between the plates.
Furthermore, due to the rotation of the plate edge caused
by bending, in-plane traction forces appear perpendicu-
lar to the edge line and their magnitude increases under
asymmetrical loads. Such forces, which occur as a result of
out-of-plane loading, cannot only be supported by shear
and in-plane compression-resistant joints.
On a dovetail joint (Fig. 2d,e), the intersection planes
on the points Xj are normal to a rotated vector w1. It is
obtained from a rotation of the reference frame {u1, u 2 , u3 }
about u3 at an alternating angle ±θ 3. The resulting rotated
side faces reduce the dovetail joints, degrees of freedom
to one translation w"3 (1DOF). Simek and Sebera20 have rec-
ommended θ 3 = 15° for spruce plywood panels. Such pris-
matic joints can only be assembled or disassembled along
one assembly direction " v = w"3. In addition to the finger-
joints resistant to shear and compressive forces, dovetail
joints can, without adhesive bonding, also resist bending
Fig. 3
moments and traction forces that are not parallel to " v.
Due to the inclination of the side faces of the joint, resist-
ance to these forces can be improved significantly. In this
manner the inclined faces assume the role that the glue
would have in a finger-joint. (Fig. 4)
1.2 Fabrication constraints

One of the main reasons for the resurgence of
finger and dovetail joints is the option of having auto-
Fig. 4 matic fabrication. However, the mechanical performance
of the joints depends on fabrication precision. At the same
time, fast machine feed rates are important for time-
Fig. 3 FEM analysis (plan view) of a 3x3 m, 21-mm Kerto-Q
folded-plate, thin shell assuming fully stiff joints. efficient production. We have fabricated joints such as
Distribution of traction (red) and compression (blue) these with a robot router and a gantry router, achieving
stresses in the y direction5. Top: gravity load case. higher precision with the gantry machine, which is
Bottom: asymmetrical snow load stiffer and provides a higher repeat accuracy.
Fig. 4 FEM simulation of bending on a dovetail joint The variability of the machine-fabricated joints
connecting two 21-mm, Kerto-Q, LVL panels. The bending
is enabled by the 5-axis capability of modern routers.
moment applied is transformed into compression, normal,
and shear forces parallel to the inclined contact faces. Although traditional edgewise joints in cabinetmaking
were used for orthogonal assemblies, both the finger

βmax βmax
φmin
TCP
TC
CP
TCP
βmax
50° 90° 130°
φmax
Fig. 5 Fig. 8
E1
E1 E2
e2
e1 ×3 E2
Fig. 9
e0
e1 e2
e0 e1
×2
2, max
Fig. 6
a) faces along edge faces across edge b) only faces

across edge
Fig. 10
v
v
Fig. 7
S1
Fabrication constraints. Side-cutting techniques used for
Fig. 5
the automated fabrication of 1DOF edgewise joints with common
5-axis CNC routers. The maximum tool inclination max is a result E1
of the tool and the tool holder geometry. From this we obtain the
range of possible dihedral angles between panels.
S2
The assembly of a folded plate from discrete elements (left)
Fig. 6
requires the simultaneous assembly of non-parallel edges (right).
The insertion direction of our 1DOF joints is rotated, to make the E2 S1 S2 S3
insertion vectors of simultaneously jointed edges parallel.
We chose a hexagon reverse fold pattern, which only requires
moderate rotations.
S3
Fig. 7 a) Dovetail joint, b) Nejiri Arigata Joint
Fig. 8 2D vector subset
E3
Fig. 9 2D simultaneous assembly
Fig. 10 3D vector subset
Fig. 11 3D simultaneous assembly Fig. 11
203
Fig. 12
Fig. 12 Folded-plate arch prototype built from 12-mm birch plywood

CV2
(9-layer, I-I-I-I-I). Assembled without adhesive bonding or metal CV3
fasteners. (Span 1.65 m, self-weight 9.8 kg)
P i+3
Series of 3-point flexural tests on a small-scale interlocking
Fig. 13 P i+4
arch prototype built from Metsawood 12-mm birch plywood panels. P i+2 ni+3
ni+4
Fig. 14Double-curved folded plate: The radius (R=17m) of the trans- ni+2
verse curvature is determined by the folded plates‘ maximum CV1 P i+5
ni+5
amplitude h 2 , which is inversely proportional to the number of seg-
ments m of the cross-section polyline (grey). We obtain this polyline P i+1 CC2
from a circular arc divided into segments of equal length. ni+1 Pi+6
The interior angle γ = ((m −2) · 180) −m of this polyline is proportional
to all fold angles φ. The geometry of our prototype was fabrication- Pi CC1
constrained to a maximum component length B ≤ 2.5 m CC3
CC: Concave
Force (N)
1000
CV: Convex
Load cycle 1
Load cycle 2
A
800
600 B B
400
A
A B
200
0
0 1 2 3 4 5 6 7 A: min length
Fig. 13 Displacement (mm) Fig. 14 B: max length

and dovetail joint can also be applied to non-orthogonal With this method, we are able to search for a joining
fold angles, which was essential for the reference projects solution for the prototype in fig. 6. We compute rotation
mentioned before. However, there are certain fabrica- windows S1, S2, S3 for the edges E1, E2, E3 and overlay them
tion-related constraints for machine-fabricated dovetail at their center point. Fig. 11 shows that there is a common
joints. In order to integrate the joint fabrication directly vector sub-set S1kS2kS3 between these three edges, in
with the panel formatting, we use a side-cutting technique,8 which one can choose an assembly direction.
which is limited to a tool inclination βmax. We obtain this As a result of these limited rotations, the angle be-
limit from the specific geometry of the tool, tool-holder, tween neighboring, simultaneously joined edges cannot
and spindle used for the joint fabrication (Fig. 5). be very acute. Folded-plate patterns like the Herringbone,
The parts can be assembled in two ways, as the Diamond, or the Hexagon pattern, which we chose for
shown in fig. 5, which allows a larger range of di- our prototypes, (Fig. 6) work well for our joining technique.
hedral angles φ. From this we obtain the fabrication- Another essential feature provided by these reverse-
constrained most acute fold φ min = 90° − βmax and most folds are the acute fold angles, which easily satisfy the
obtuse fold φmax = 90° + βmax . With standard cutting tools, fabrication-constrained range of φmin = 50° to φmax = 140°.
this technique allows the jointing of acute folds up to
φ = 50°, which is ideal for folded plate structures. Highly
obtuse fold angles φ ≥ 140°, which might be required 2 Interlocking arch prototype
for smooth segmented plate shells, cannot be fabricated
with this method. In an assembly of multiple components (Fig. 12), a step-
by-step sequence must be planned for the assembly of
1.3 Simultaneous assembly of multiple edges parts. The completed structure can only be disassembled
The assembly of double-corrugated folded plates piece by piece in the reverse order of assembly. In this
requires the simultaneous joining of multiple edges per way, the elements interlock with one another, like in a
component (Fig. 1), which has implications for both the burr puzzle. 22
shell and the joint geometry. Each joint consists of two parts, which must
For multiple 1DOF-jointed edges, simultaneous be parallel during assembly. We therefore chose a folded-
assembly is only possible if the individual assembly direc- plate geometry with relatively short edges. The manual
tions " v are parallel. With a normal dovetail joint geome- assembly of long edges may be more difficult but can be
try (Fig. 7a), this is not the case. A simultaneous assembly is simplified with a modified joint geometry. It is important
only possible for parallel edges, which only allows recti- to know the approximate direction of insertion for each
linear assemblies, such as for drawers or cabinets. part, as this is not easily visible through the joint geo-
In order to simultaneously join non-parallel metry. Deformations of the arch during the assembly
edges, the assembly direction v of the joints has to be ro- should be minimized. We have assembled this first proto-
tated to make them parallel. This technique is known type lying on its side. However, larger assemblies may
from Japanese cabinetmaking, 9 where certain joints, like require temporary punctual supports. Although the in-
the Nejiri Arigata joint (Fig. 7b), are assembled diagonally plane dimensional stability of the Kerto-Q panels is very
along a vector that does not lie on either one of the two high, panels may be slightly warped and some force
planes. European dovetail joints, on the other hand, form may be necessary during assembly. While we have sim-
a prism with a single tab, using faces both across and ply used a rubber hammer, more advanced techniques
along the edge; the Nejiri Arigata joints form a prism could be applied.
using multiple, differently shaped tabs. To understand the mechanical behavior of the
This Japanese technique is extended to a vector built prototype, we have applied a vertical load at the
subset of possible assembly directions. Fig. 8 shows that mid-span of the arch and measured the vertical deflec-
the rotation about the edge line is constrained to 180°– φi . tion at the same point. The total load of 821 N was applied
The vector subset is large for acute and small for obtuse in two identical load cycles consisting of four loading/
fold angles. This is particularly important when joining unloading sub-cycles. First, a vertical load of 117 N was
multiple edges simultaneously, as an intersection has applied in seven steps, after which the load of the previ-
to be found between multiple vector subsets (Fig. 9). If ous four steps was removed. The loading and unloading
there is an intersection, the parts can be joined simulta- of the previous four steps was repeated three times, after
neously along any direction within the intersection of which the complete load was removed and the residual
the subsets. deflection was measured (Fig. 13).
Finally, this concept is extended to a three- Under a vertical load equivalent to the arch’s
dimensional rotation (Fig. 10). This is possible through a sec- dead weight of 9.8 kg (98 N), the deflection measured at
ond rotation θ 2, which is constrained to a maximum value mid-span was 2 mm. From this we obtain a span-to-
of ±θ 2;max . The limitation results from various other corre- deflection ratio of L/750, and the arch’s structural effi-
lated parameters, such as θ1 and βmax . We call the result- ciency reaches 8.6 when loaded with 821 N (ratio of the
ing three-dimensional vector subset “rotation window.” maximum load over the dead weight of the arch).
205
100 107 91
71 70 69 68 Blocked Fig. 15 Partial connectivity, assembly, and
99 98 90 89 blocking graph of the folded-plate shell
prototype (left-to-right assembly). Large
65 66 68
89 88 87 86 numbers represent mesh faces; small numbers
35 36 41
represent mesh edges.
Free
18 Fig. 16 Left-to-right assembly of the
44 7 59
107 106 105 104 interlocking folded-plate shell prototype.
Built from Kerto-Q structural-grade
LVL panels (7-layer, I-III-I).
Fig. 15
v107
v100
v99 v94
v98
v90
v89
v84
e2 v65
v68
e1 v66
F86 v43
v70
v18
v35
v36
F86 v41 = v68 = v89
v44 v42
v41 v35 = v65 = v98
v59
v7 = v36
v13 = v42
e3
e2 v7
v13
e1
Fig. 16

50
Force (N)
3 Interlocking shell prototype
3.1 Automatic geometry processing 40

Using the RhinoPython application program-
ming interface, a computational tool has been developed
that enabled the instant generation of the geometry 30
of the individual components, as well as the machine

G-Code required for fabrication. The tool processes arbi- 20
trary polygon meshes, and generates 1DOF joints for all
non-naked edges where the fold angle φ is larger than
φmin and smaller than φmax shown in fig. 5 (non-smooth 10
meshes). It also requires the input of edge identifier pairs,
identifying those edges which have to be joined simulta-
neously, and a value for the thickness of the LVL plates. 0
0 20 40 60 80 100
By exploiting this geometric freedom, we have tested Displacement (mm)
our computational tool on the design of a folded-plate
Fig. 17 Load-displacement curve of the shell prototype.
shell prototype with an alternating convex-concave trans- A longitudinal line load was introduced along the top
verse curvature. The shell spans over 3 m at a thickness of the shell. Vertical displacement was measured at the
of 21 mm, using Kerto-Q structural grade LVL panels center point.
(7-layer, I-III-I). (Fig. 14)
Comparing this double-curved folded plate with
a straight extrusion (as tested by Hans Ulrich Buri2), it 3.3 Completed shell prototype and load test
can be concluded that the slight double curvature proves Fig. 17 shows the completed folded-plate proto-
to be highly beneficial when it comes to global deflections, type, with a span of 3 m and a shell thickness of 21 mm.
for example, those caused by wind loads. Deflections for Boundary conditions that restrain displacements of the
the double-curved shell geometry in the vertical direc- supports in every direction, but allow rotations, were ap-
tion are up to 39 % smaller, and up to 13 % smaller in the plied on both sides. A longitudinal line load was intro-
lateral direction than those with a straight extrusion. duced along the top of the shell and vertical displacement
was measured at the center point (Fig. 18).
3.2 Assembly The prototype structure was also modeled in FE
Due to the different assembly directions of its analysis software (ABAQUS) and loaded in the same way.
239 joints, the 107 components in our prototype inter- The plates were modeled using shell elements, where the
lock with one another in a manner similar to a burr mid-surface is used to represent the three-dimensional
puzzle. 22 Fig. 15 shows a section of a so-called “non- plate, and transverse shearing strains are ignored. Con-
directional blocking graph” (NDBG), developed by Wilson nections between the plates were considered as being
and Latombe. 21 completely rigid in order to obtain minimal displace-
In a NDBG graph, single arrows indicate that in- ments of the structure. By comparing displacements of
dividual parts can be removed from the assembly. Two the structure with completely stiff joints with those
opposite arrows between parts indicate that the connec- measured on the prototype, we obtained information
tion is fixed. In order to remove fixed parts, the fixed about the actual semi-rigidity of the joints. The results
parts have to be removed first. Our graph illustrates a obtained from the testing of the large scale prototype
left-to-right assembly. Part number 86 is being inserted showed that the load of 25 kN, which corresponds to the
on the right side. It connects to three other plates and proportional limit of the load-displacement curve, causes
blocks all other parts in the graph. In such a configura- a vertical displacement of 23 mm. In the FE model,
tion, the final part, called the key, remains removable. the load applied in the same manner caused a vertical
Fig. 16 shows the parts from fig. 15 in three displacement of only 2.6 mm.
dimensions, demonstrating how the component based
on mesh face F86 is inserted. Its three edgewise joints
E41, E68, and E89 have to be assembled simultaneously. 4 Conclusion
The three assembly vectors of the edges v" "
41, v68 , and
"
v89 have been rotated to be parallel. The same applies A timber folded-plate shell combines the structural ad-
for the adjacent edges on the left side of the faces F67, vantages of timber panels with the efficiency of folded
F69, F88, F103, and F105 (Fig. 15). Within the rotation window plates. However, in such discrete element assemblies, a
of the edge, we can freely rotate " v for these edges (the large number of semi-rigid joints must provide sufficient
greater the angle between " v and the main direction of support for the adjacent plates in order to ensure an effi-
traction e1, the better). cient load-bearing system. This remains a challenge with
great potential for improvements. 5
207
Integrated edgewise joints present an interesting addi- ticular advantages, possibilities, and challenges of 1DOF
tion and an alternative to state-of-the-art connectors: joints for timber folded-plate shells. We have demon-
Unlike adhesive bonding, such joints can be assembled strated how this joint geometry helps in resisting the
rapidly on site. Moreover, compared with costly metal forces that occur in such structures. In addition to the
plates and fasteners, typically required in large quanti- load-bearing connector features, the joints provide loca-
ties,14 the fabrication of integrated joints does not in- tor features, which allow precise positioning and align-
crease costs. The replacement or reduction of metal fas- ment of the parts through the joint geometry. This im-
teners with an integrated mono-material connection proves both accuracy and ease of assembly.
includes advantages such as improved aesthetics, ease of Furthermore, we have presented a solution for
recycling, or a homogeneous thermal conductivity of the simultaneous assembly of multiple edges per panel,
the parts, which can reduce condensation and decay. 4 which is essential for the application of 1DOF joints in
Another particular advantage is the ability to connect a folded-plate shell structure. The per-edge “rotation
thin panels. The current technical approval for the window” introduced in section 1.3 integrates the joint
Kerto-Q panels does not permit screwed joints on panels constraints related to assembly and fabrication. It can
with a thickness of less than 60 mm. 3 be processed algorithmically and provides instant feed-
Recent experimental projects, introduced in back on whether or not a set of non-parallel edges can
Chapter 1, have already demonstrated initial applications be joined simultaneously. This provides a tool for the ex-
of integrated edgewise joints for timber panels. This ploration of a variety of alternative folded-plate shell
paper followed up on these projects, examining the par- geometries.
Folded-plate shell prototype, built from 21- mm LVL panels. With a self-weight of 192 kg,
Fig. 18
the prototype with a span of 3 m was tested with a line-load up to 45 kN.

The prototypes presented in this paper already suggest 10Krieg, O., T. Schwinn, A. Menges, J.-M. Li, J. Knippers, A.
possible patterns and demonstrate the reciprocal rela- Schmitt, and V. Schwieger. “Computational Integration of
tionship between the geometry of the plates and the Robotic Fabrication, Architectural Geometry and Structural
Design for Biomimetic Lightweight Timber Plate Shells.” In:
joints. Two built structures allowed us to test and verify Advances in Architectural Geometry. 2014. Springer Verlag,
the proposed methods for fabrication and assembly while 2014.
providing valuable information about the load-bearing 11 Krieg O. D., Z. Christian, D. Correa, A. Menges, S. Reichert,
capacity of the integrated joints. K. Rinderspacher, and T. Schwinn. “Hygroskin: Meteorosensi-
For application in a large-scale building structive pavilion.” In: Fabricate 2014 Conference Zurich, 272–279,
ture, further research is required to determine if the 2014.
integrated joints can replace additional connectors en- 12La Magna, R. et al. ”From Nature to Fabrication: Bio-
mimetic Design Principles for the Production of Complex
tirely or reduce their number. A possible combination of
Spatial Structures.” International Journal of Spatial Structures,
integrated joints with additional metal fasteners has re- vol. 28, no. 1:27-39, 2013.
cently been demonstrated in the LaGa Exhibition Hall.10 13 Messler, R. W. Integral Mechanical Attachment: A Resurgence
Another possibility would be a combination of the 1DOF of the Oldest Method of Joining. Butterworth Heinemann,
joints with integrated elastic interlocks.17 2006.
14 Neuhaus, H. DIN EN 1995 (Eurocode 5) – Design of timber
structures. DIN Deutsches Institut fur Normung e. V., 2004.
15Purbond. National Technical Approval Z-9.1-711 / Single-
Component Polyurethane Adhesive for the Manufacture of
Engineered Wood Products. Deutsches Institut für Bautech-
nik, 2011.
Acknowledgments 16Robeller, C. “Integral Mechanical Attachment for Timber
Folded Plate Structures.” PhD Thesis, ENAC, Lausanne, 2015.
We would like to express our gratitude to Andrea Stitic and
17 Robeller, C., Paul Mayencourt, and Yves Weinand. “Snap- t
Paul Mayencourt for their support with generating the finite
joints: Integrated mechanical attachment of structural
element models and load-testing the prototypes, and to
timber panels.” In: Proceedings of the 34th International
Gabriel Tschanz and Francois Perrin for assisting us with the
Conference of the Association of Computer-aided Design in
fabrication and assembly of prototypes. We would also
Architecture ACADIA, Los Angeles. Riverside Architectural
like to thank Jouni Hakkarainen and the Metsa Group for
Press, 2014.
supplying valuable information and material.
18 Robeller, Christopher, Seyed Sina Nabaei, and Yves Weinand.
“Design and Fabrication of Robot-Manufactured Joints for a
Curved-Folded Thin-Shell Structure made from Clt.” In:
Robotic Fabrication in Architecture, Art and Design 2014, edited
References by Wes McGee and Monica Ponce de Leon, 67–81. Springer
1Benjamin, B. S. and Z. S. Makowski. “The Analysis of Folded- International Publishing, 2014.
Plate Structures in plastics.” In: Proceedings of a conference 19 Schineis, R. “Gefalteter Klangkoerper Musikprobensaal
held in London, 149–163. London: Pergamon Press, 1965. Thannhausen/Thannhausen Rehearsal Room.“ In:
2 Buri, H. and Y. Weinand. “BSP Visionen – Faltwerkkonstruk- 10. Internationales Holzbau Forum (IHF), Garmisch-
tionen aus BSP-Elementen.“ In: Grazer Holzbau-Fachtage, Partenkirchen, 2004.
2006. 20 Sebera, V.,and Milan Simek. “Finite Element Analysis of
3 DIBt. Allgemeine bauaufsichtliche Zulassung Kerto-Q Dovetail Joint made with the use of CNC Technology.
Z-9.1-100, Paragraph 4.2 and Attachment no. 7, Table 5. In: Acta Universitatis Agriculturae et Silviculturae Mendelianae
Deutsches Institut für Bautechnik, 2011. Brunensis, vol. LVIII. 321–328, 2010.
4 Graubner, W. Holzverbindungen. Gegenüberstellung 21 Wilson, R. H. and Jean-Claude Latombe. “Geometric
von Holzverbindungen Holz in Holz und mit Metallteilen. reasoning about mechanical assembly.” In: Artificial
Stuttgart, Deutsche Verlags-Anstalt, 1986. Intelligence, 71(2):371–396, 1994.
5 Hahn, B. “Analyse und Beschreibung eines räumlichen 22 Wyatt, E. Puzzles in Wood. Bruce Publishing Co., 1928.
Tragwerkes aus Massivholzplatten.“ EPFL, Master's thesis,
Lausanne, 2009.
6 HESS. Hess limitless, 2014. www.hess-timber.com/de/
produkte/
7Hundegger. H. Hans Hundegger Maschinenbau GmbH,
Hawangen, Germany, 2014. www.hundegger.de/en/
machine-building/company/our-history.html.
8 Koch, P. Wood Machining Processes. Wood Processing.
The Ronald Press Company, 1964.
9 Kondo, T. Dento kanehozo kumitsugi: Sekkei to seisaku no
jissai. LLP Gijutsushi Shuppankai, Tokyo: Seiunsha, 2007.
209
5. 3
Rotational stiffness on the ridges of

timber folded-plate structures
Stéphane Roche, Geoffroy Mattoni, and Yves Weinand
Folded-plate structures
Plattenfaltwerke aus dünnen
provide Furnierschichtholz-
an efficient design 1 Introduction
using
plattenlumber
sind effiziente
panels of Konstruktionen.
thin laminated veneer.Inspiriert
In-
spired
vom japanischen
by JapaneseTischlerhandwerk,
furniture joinery, the wurdemultiple
eine The first generation of folded-plate roofs was constructed
Zinkenverbindung
tab-and-slot joint entwickelt,
was developedmit deren
for the
Hilfemulti-
sich in the mid-1950s using plywood.1 However, technical
assembly
mehrere Platten
of timber mitpanels
nicht with
parallelen
non-parallel
Kantenedges
ohne problems arose from variations in the plywood’s struc-
Klebstoff
without any oder
adhesive
metallische
or metalVerbindungsmittel
joints. As the global zu- tural properties. The aesthetic of the outer layers was
sammenfügen
analysis of ourlassen. origamiDa structures
die globalereveals
Analysethatun- paramount, while its characteristic strength could not
serer
the rotational
Origami-Konstruktionen
stiffness at the ridgeszu der affects
Erkenntnis
global be fully controlled. Single-fold plate roofs behave like a
führte,
behavior,
dasswedie
propose
Rotationssteifigkeit
an experimental an and
den numeri-
Rippen series of V-section beams. The plywood skin transmits
cal
dasstudy
globale
of this
Verhalten
linear, interlocking
beeinflusst, connection.
entschlossen
The shear like a web and transfers in-plane forces to the roof
geometry
wir uns zu iseiner
governed
experimentellen
by three angles
und numerischen
that orien- edges, where the walls and transverse stiffeners bear the
Studie
tate thedieser
contact linearen,
faces. Nine
gezinkten
combinations
Verbindung.of these
Die vertical and horizontal load components respectively.1, 2
Geometrie
angles were wirdtested,
durchand drei the
Winkel
rotational
an den slip
Kontakt-
was Fifty years after the first attempts at folded-plate roof
flächen
measured bestimmt.
with two Wirdifferent
haben neun bending
Kombinationen
scenarios: structures,1 Jaksch et al. 3 proposed a similar pitched roof
closing
dieser Winkel
or opening
untersucht
the foldundformed
die Rotationssteifig-
by two panels. made from lightweight, cross-laminated timber (CLT)
The
keit nonlinear
für zwei unterschiedliche
behavior was conjointlyBiegeszenarien
reproduced ge- rather than plywood. More recent attempts have adopted
numerically
messen: das using
Zusammendrücken
the finite element
und das method
Auseinand
an- glue and nails, or cold-formed thin steel plate, to create
continuum
derdrückendamage
einer aus mechanics.
zwei Platten bestehenden a rigid joint at the ridge. Today, laminated veneer lumber
Falte. Überdies haben wir das nichtlineare Verhal- (LVL), CLT, and solid wood panels (SWP) are the most
ten auch in numerischen Untersuchungen mittels commonly used materials for highly stressed structures.
Finite-Elemente-Methode
Keywords semi-rigid, connection, undmoment-rotation,
Kontinuumsschädi- These high-performance panels inspired other architects
gungsmechanik nachgebildet.
folded-plate and researchers to develop novel solutions for reviving
folded-plate structures. In 2008, Buri et al. described
the application of origami paper folding to timber fold-
halbstarr, Verbindung, Moment- ed-plate structures at the 10th World Conference on Tim-
Rotationsbeziehung, Plattenfaltwerk ber in Japan. 4 The structure’s 21-mm plywood panels
were assembled with 5-mm, self-drilling screws arranged
in staggered rows at the miter joints of the panels. Both
sides were beveled at a 60° angle, to form a 120° corner.
Because the relative position of the parts had to be set
before screwing, pre-mounting with jigs was required.
Failure during a loading test occurred through tearing
at the ridges and valleys, a direct result of the weak-
ness of the connection. 5 Although origami folded-plate
structures are promising complex timber structures,
the connections of their thin panels have to be improved.
Robeller et al. described the first approach to the problem
in 2014. 6 A double-curved, folded-plate prototype was
constructed using a five-axis CNC machine. The cut LVL
plates had integrated jointing at their edges. Based on
linear connections used in furniture (e.g., dovetail or
Japanese Nejiri Arigata joints), these multiple tab-and-

Tj
–Mj +M j
Nj ,Q j
a) q b) q c) q
d) e) f)
M M M
Mj j
-
+
+
+
w
M
-
-
Fig. 1 a) Origami double-curved, folded-plate

prototype analyzed by Robeller et al. 6
b) Bending moments on the ridge
c) Forces along and perpendicular to the ridge
d) Hinged joint at the ridge
e) Semi-rigid joint at the ridge
f) Rigid joint at the ridge
slot joints (MTSJ) with no adhesive bonding allow the However, the fastened connections usual in timber or
simultaneous assembly of up to three ridges or valleys. steel structures actually behave like semi-rigid connec-
The panels can then be directly interlocked in situ at tions.7, 8 These works led EC5 9 to take fastener stiffness
their final location. All these previous studies noted a into consideration in the design of timber structures. The
lack in the connection stiffness model, a deficit that could new rules are nonetheless unsuitable for traditional car-
affect the global behavior. Literature on timber connec- pentry joints. Research programs on the moment-rota-
tions rarely addresses the structural behavior of panel tion behavior of these types of connections include a pro-
connections. Most research on connections during the last ject by the Czech Ministry of Culture, which is dedicated
thirty years has examined joint stiffness within the con- to investigating the timber joints in historical structures
text of frame structures. Among many others, H.J. Larsen, (“Design and Assessment of Timber Joints of Historical
A.J.M. Lejiten, A. Kevarinmaki, and D.B. Engstrom, Scan- Structures,” which was launched in February 2012). A
dinavian members of the Timber Joints Working Group study of the bending stiffness of a dovetail joint for
of the European Cooperation in Science and Technology plywood panels has also been analyzed. Again, friction
(COST action C1, semi-rigid behavior of civil engineering is implemented in the three-dimensional, solid, finite-
structural connections), have published specific research element model, but the studied orthotropic material (used
on the rotational stiffness of timber joints.7 Their work for furniture) was only considered in its elastic stage,
showed that the inclusion of semi-rigidity in the struc- as the final displacement was limited to 2 mm.10 Robeller
tural analysis gave more realistic frame stress and de- et al. 6 presented a linear interlocking connection of plates
flection results, thus demonstrating its significant effect for the in situ assembly of folded plate shells. They intro-
on stability. Before the application of limit-state codes, duced a geometric solution for simultaneously joining
such as Eurocode (EC), to the design of structures, joints plates with multiple, non-parallel edges and calculated
were commonly considered as either hinged or rigid. the mechanical performance of the overall structure by
211
a finite element method (FEM) analysis of perfectly (for the purpose of assembly) that has tabs inserted into
rigid joints. In conclusion, a local simulation of bending slots (Fig. 2e). “Multiple” refers to the interlocked tabs
on a dovetail joint connecting two panels was undertak- and slots, which are repeated along the common edge of
en. The bending moment was transformed into com- two connected panels. The geometry of the connection
pression, normal, and shear forces parallel to the locking defines the relative positioning of the panels and allows
faces. Conjointly, Roche et al.11 conducted numerical and a certain degree of load-transfer between them. The
experimental work to test the assumptions of semi-rigid assembly of the two panels is directed along the vector of
behavior of such a connection. Three-layered box beams insertion. The locking faces of each part belong to the
assembled at the inter-layers using dovetail joints with same contact plane after insertion. The insertion vector
differing tab lengths and tab angles were loaded to fail- and the vector normal to the locking face are obtained by
ure point in three-point bending tests. The connection the sequence of rotations (equ. 1) following the convention
showed promising stiffness results. For instance, a of the Bryant angles (Fig. 2a–d).
110-mm tab-length dovetail joint was stiffer than 110-mm
spaced screws. The previous results confirmed the good {P i ,F = (u 1, u 2, u 3)}
strength/stiffness ratio of the MTSJ in shear. The shear R(u 1, θ1).
due to the inter-layer slip is transferred by compression
{Pil , F l= (u l1 = u l1, u 2l , u 3l )}
on the locking faces. The moment-resisting performance
discussed in Robeller 6 should be also confirmed. R(u 2l , θ 2).
This paper presents a study of the rotational {P im, F m= (u 1m, u 2m = u 2l , u 3m)}
stiffness of a MTSJ as a structural connection in architec- R(u 3m, ±θ 3; i).
ture. A dedicated folding machine was developed to test
{Pni , F n= (u n1; i, u n2; i, u n3; i = u 3m)} Equ. 1
the Japanese pattern of the MTSJ under bending moments.
A parametric experimental study explored the rotational
stiffness through different combinations of the three If n 0 and n 1 are the normals of the two panels, the frame
angles governing the joint geometry. Nonlinear behavior of the joint F = (u 1, u 2, u 3) can be calculated as u 1 = n 0 × n 1,
is reproduced by finite element analysis using an adapta- u 2 = n 0, and u 3 = u 1 × u 2 . Subsequently, the line segment
tion of the continuum damage model of Sandhaas.12 representing the intersection of the two panel, mid-
planes can be uniformly divided into N points (Xi)(i = 1…N);
the distance between points will be the tab length Lj of
2 The multiple tab-and-slot joint as a the joint. The plane P i will be the plane of normal u 1, con-
structural panel connection taining the point X i (Fig. 2a). Thus, by three successive rota-
tions of the plane, P i, and its attached frame, F (Fig. 2a–d),
2.1 Forces at ridges and valleys the final joint will be obtained, defining the planes, Pni for
When a folded-plate structure is uniformly and both the tab and slot as the locking faces and the vectors,
vertically loaded, slab and plate action induce transverse u n1;I , and u 3m = u 3n;i as their normals and insertion vector
and longitudinal action in the folds. 3 The ridges typically respectively (Fig. 2d) . This elemental sequence of rotations
experience the bending moment, Mj (Fig. 1b), forces perpen- is represented by the Bryant angles, θ1, θ 2, and θ 3. In this
dicular to the ridge in the panel plane, Nj, and beyond the study, the principle of the nejiri arigata joint described
panel plane, Q j (Fig. 1c), and forces along the ridge, Tj (Fig. 1c). by Robeller 6, 13 is adopted, where +θ 3 is applicable from
The study is limited to behavior under the bending P1 to PN/2, and – θ 3, from PN/2 to PN.
moment, and at this stage ignores the effect of transverse This geometry allows a multi-edge assembly of
forces. As the model of this joint is assumed to be semi- one panel with three adjoining plates. Four plates k (k = 0;
rigid (Fig. 1e) during bending of the plates, an additional 1; 2; 3) are taken from the origami folded-plate structure
rotation is induced at the ridges. This is contrary to (Fig. 3a and b) . Plate 0 (i.e., k = 0) shares edges 01, 02, and 03
the rigid model, where the initial angle, φ, between the with plates 1, 2, and 3 respectively. For each edge, the eli-
panels remains unchanged after deformation (Fig. 1f). The gible subset of insertion vectors is defined as E 01, E 02, and
MTSJ offers a moment-resisting connection that cannot be E 03 (Fig. 3c). The fabrication constraint, βmax, referring to the
provided by a hinge (Fig. 1d). Here we seek to understand maximal tool inclination, sets the limits of the dihedral
the way that geometry, particularly the angles directing angle φ 0k and the angles θ10k, θ 20k, and θ 30k that govern
the normal to the locking face, affects this semi-rigidity. the insertion vector subsets and the tab angle respective-
ly. 6, 13, 14 For example, for a 12-mm-diameter milling tool
2.2 Description of the MTSJ with a cutting length of 28 mm, βmax is 30°, which prevents
A detailed description of this “integrated mechan- spindle collision when machining 21 mm panels. If the
ical attachment” is given in Robeller; 6 thus, in the next dihedral angle, φ, is set to 120° (Fig. 2f) or if θ 3 is set to
section a brief introduction to the geometric parameters 30° (Fig. 2g), then the maximum tool inclination is already
of the joint is included. The ridge (or valley) connection reached and θ1 and θ 2 cannot be different from 0°. Finally,
employed here is a one-degree-of-freedom connection θ1, θ 2, and θ 3 can only all have non-zero values if βmax < 30°

u2
u2 P
Lj
//
u1 Slot // TP
u3 Xi u1 Tab //
u3
//
// u'''1
n1 u'''2
Notch
Locking face
{Pi ,F=(u1,u2 ,u3 )} u'''3
a) R =(u
(u1,θ1) e)
P’ n0
u'2 =30°
u'1=u1
Xi
u'3 1
n1
1
= 120°
{P"i ,F"=(u1 =u1,u'2 ,u'3 )}
b)
(u'2,θ2)
R =(u' θ1 =θθ2 =θθ3 = 0°
f)
n0
P’’
2 u''1 =30°
u''2=u'2
Xi
2
u''3
n1
=90°
{P""i ,F""=(u''1 ,u''2=u'2 ,u''3 )}
θ1 =θθ2 = 0°
c) R =(u''
(u''3,±θ3,i ) θ3 = 30°
g)
<30°
-θ3 +θ3
Insertion vector
P'''i n0
u'''2,i
u'''1,i
P'''i +1
Xi u'''1,i+1
u'''2,i+1
u'''3,i =u''3 Xi+1 θ1 ,θθ2 ,θθ3 =/ 0°
n1 h)
u'''3,i +1=u''3
3 Rotation sequence for designing the locking face

Fig. 2 a–d)
e) Terminology of the MTSJ
3 f) Fabrication constraint due to the dihedral angle
g) Fabrication constraint due to the tab angle
{P"""i ,F"""=(u'''1,i ,u'''2,i ,u'''3,i =u''3 )} h) Constrained combination of 1, 2, and 3
d)
213
and 60° < φ < 120°. Geoffroy Mattoni showed that the interconnected panels, the tests of pushing F0 and F1 were
dependency between the Bryant angles, dihedral angle, each repeated twice. A 20 kN cylinder pulled two cables to
and tool inclination limits the possible combinations.14 drive the rotation of two moment-transmitting pulleys.
By intersecting the three constrained local vector subsets, Subsequently, the plate supporting two 5 kN cell loads
a unique vector subset is finally obtained, from which the acted as a lever arm on the vertical panel F0 (or F1) of
insertion vector has to be selected. the sample. The horizontal F1 (or F0) panel was rigidly
Given our intention to analyze the rotational clamped onto the base plate. The location of the rotation
stiffness of the joint, angle θ 3 will be taken to be at least center of the rig could be set to coincide with the rotation
10° where the locking faces are clearly not parallel and center of the joint. Rotation and loads were recorded by
offer a certain resistance to rotation about the ridge axis. averaging the values of two inclinometers and adding
Detailed analysis determining the parameters for our data from the cell loads respectively. This method was ap-
sample is given in Mattoni.14 plied in both closing (S01) and opening (S02) tests. A total
of 88 specimens were tested.
3 Experimental study
4 Numerical model
3.1 Description of the samples
Eleven series of two, 21-mm LVL Kerto-Q panels 4.1 Material model
with spruce ply (0-90-0-0-0-90-0) were assembled with The LVL Kerto-Q material is multi-layered due to
eleven particular sequences of joint elements. The geo- the circular notch (Fig. 6), which is required for the digital
metric properties of the specimens are summarized in manufacture of the joint, reducing the contact surface to
Table 1 and fig. 5. No glue was utilized during the assem- a few layers. As through-thickness stress cannot be ig-
bly of samples (a) to (i), which correspond to nine charac- nored, the model uses linear, hexahedral elements with
teristic combinations of Bryant angles.14 All had the same reduced integration (C3D8R). Each of the seven layers
tab length of 50 mm. The two panels F0 and F1 were has its own orientation (0-90-0-0-0-90-0). The Coulomb
200 mm long (along the ridge) and 150 mm wide and had friction coefficient is set to 0.3.
two and one full tab respectively. One series (j) was
screwed using three Würth ASSY screws (4 mm × 70 mm) 4.2 Elastic behavior
with 50 mm spacing. The last series (k) was bonded by Each northern spruce veneer is typically con-
PUR gluing. The last two series had butt joints. sidered orthotropic and behaves elastically in its local
orientation frame (Fig. 7). Interfaces are considered to be
3.2 Method rigid, thus inter-laminar failure is not addressed. Elastic
Four specimens of each sample were tested using properties are summarized in Table 2.15, 16 Reliable values
a dedicated folding-test machine (Fig. 5). To reflect the for spruce veneer are relatively rare in the published
asymmetry of the joint along the bisector plane of the literature.
n3
03 03 03
3 E 03 1 2 3
03
02 02 02 Insertion direction
1 2 3
n2
n0 E 02
02
E 01 E 02 E 03
1300
2 Intersection
0
30
00
01
01 01 01
1 2 3
0
300 n1 E 01
a) b) c)
1
Fig. 3 Assembly constraints a) Plates in situ in the global geometry b) Three-edge simultaneous assembly
c) Selection of the insertion vector into the intersection space

Parameters Symbol Unit Value Property Symbol Value Unit
Insertion angle (1) θ1 °deg → Elastic modulus 11 E11 6200 Mpa

Insertion angle (2) θ2 °deg Elastic modulus 22 E 22 210 Mpa
Tab angle θ2 °deg Elastic modulus 33 E 33 210 Mpa
Dihedral angle φ °deg 90 Shear modulus 12 G 12 350 Mpa
Tab length Lt mm 50
Shear modulus 13 G 13 200 Mpa
Panel thickness Tp mm 21
Shear modulus 23 G 23 40 Mpa
Panel density ρk kg/m3 480
Angle edge to grain α °deg 0 Poisson’s ratio 12 V12 0.61
12-mm notch r t or b b Poisson’s ratio 13 V13 0.60
Tool angle T241 β °deg 30 Poisson’s ratio 23 V 23 0.50
Table 1 Sample parameters Table 2 Elastic properties of spruce
Fig. 4 Combination of Bryant

Fig. 5 Bending tests: a) closing mode b) opening mode
c) sample geometries
Fig. 6 Reduced contact on the locking faces
F1 F1 F1 F1 (j) 50 F1
=
(a ) (b ) (c) (d)
=
=
=
F0 F0 F0 F0 F0
3×
3×Schraube
screw 4 4×
×7070mm
mm
F1 F1 F1 F1 (k) F1
0° 0° 0° 45°
0° F0 30° F0 0° F0 0° F0 F0
±10° ±10° ±30° ±10° PUR-Klebstoff
PUR glue
(e ) F1 ( f) F1 (g) F1 (h) F1 (i) F1
F0 F0 F0 F0 F0
F1 F1 F1 F1 F1
45° 45° 90° 90° 90°
31° F0 30° F0 0° F0 30° F0 0° F0
±10° ±38° ±10° ±10° ±30°
Fig. 4 Fig. 6
slip slip
F0 F1
110
110
F1 F0
a) S01.00X odd S01.00X even
slip slip
F0 F1
120.5
120.5
F1 F0
b) S02.00X odd S02.00X even 2× Moment-transmitting pulleys 2× 5 kN cell loads
c) 2× Pulling cables 2× Inclinometer

Fig. 5
215
Radial
Z,R,3 4.3 Damage approach
To approach nonlinear behavior of the joint, we
use a wooden model based on continuum damage me-
chanics (CDM) (generously provided by C. Sandhaas and
described in detail in references 12 and 15). A main fea-
ture of the model is its penalization of the elastic moduli.
The compliance matrix Cdam, derived for the elastic com-
pliance matrix, defines the modified behavior of dam-
aged elements (equ. 2–5).
Longitudinal
X,L,1
ε = Cdamσ Equ. 2
Tangential
Y,T, 2 RS f 11 VW RS v 11 VW
SS W SS WW
SS f 22 WWW SS v 22 WW
SS W SS WW
SS f 33 WWW C 1 0 SS v 33 WW
dam
SS WW = # & SS WW
SS 2f 12 WW 0 C 2 SS v 12 WW
dam
SS 2f 13 WW SS v 13 WW Equ. 3
SS W SS WW
3 S 2f 23 WW S v 23 W
2 T X T X
1
3 RS 1 –o 1 2 –o 31 VW
SS WW
2 SS (1– d 11) E 11 E 22 E 33 WW
1 3 SS –o 12 1 – o WW
2
= SS WW
32
3 2 C dam
1
1 3 SS E 11 (1– d 22) E 22 E 33 WW
1 2 SS –o 13 – o 1 WW
1 SS 23
W
(1– d 33) E 33 W
Equ. 4
E 11 E 22
3 T X
2
1 RS 1 VW
SS 0 0 WW
3 2 SS (1– d 12) G 12 WW
S 1 WW
1 C 2dam = SSS 0 0 WW
SS ( 1– d 13 ) G 13 WW
SS 1 WW
SS 0 0 W Equ. 5
Fig. 7 Material frame and orientation (1– d 23) E 23 W
T X
dij are the damage variables, which depend on the strain

state. Here, we recall the failure modes Fm15 that control
the onset of damage.
Property Symbol Value Unit
4.3.1 Brittle behavior
t
Tensile strength // to grain f 11 f t;0 78 Mpa
c
Compressive strength // to grain f 11 fc;0 50 Mpa – In tension:
t
Tensile strength to grain f 22 f t;90 2.2 Mpa
v 11
c
Compressive strength to grain f 22 fc;90 10 Mpa Ft,1 = #1 Equ. 6
f t, 0
t
Tensile strength to grain f 33 f t;90 2.2 Mpa
c
Compressive strength to grain f 33 fc;90 10 Mpa
Longitudinal shear strength f12 fv 6.9 Mpa – In tension and shear:
Longitudinal shear strength f13 fv 6.9 Mpa
Q v 22 V2 Q v 12 V2 Q v 23 V2
Rolling shear strength f 23 froll 2.5 Mpa Ft,2 = + + #1
Fracture energy tension // to grain Gf;0 1.6 N/mm
Q ft,90 V2 Q fv V2 Q froll V2 Equ. 7
Fracture energy tension to grain Gf;90 0.5 N/mm

Q v 33 V2 Q v 13 V2 Q v 23 V2
Ft,3 = + + #1
Fracture energy longitudinal shear Gf;v 1.2 N/mm
Q ft,90 V2 Q fv V2 Q froll V2 Equ. 8
Fracture energy rolling shear Gf;roll 0.6 N/mm
Table 3 Mechanical properties of spruce

– In shear under compression: However, for brittle failure, stress diminishes with the
strain increment (Fig. 8). This decrease is defined by the
Q v 12 V2 Q v 23 V2 failure energy (the area below the curve). In this case,
Fv,2 = + #1
Q fv V2 Q froll V2 Equ. 9 the damage parameter, dm, can be calculated as:
Q v 13 V2 Q v 23 V2 2
2gf E
–
Fv,3 = + #1 f max
Q fv V2 Q froll V2 lm
Equ. 10
d m = 1– 2
f max – 2gf E Equ. 15
4.3.2 Ductile behavior where the fracture energy is
– In compression: Gf
gf = l Gf Equ. 16
e
-v
Fc,1 = 11 # 1 Equ. 11
fc,0
and le is the characteristic length of the element in the
-v numerical model (CELENT parameter in ABAQUS). Re-
Fc,2 = 22 # 1
fc,90 Equ. 12 placing Gf with g f in equation 15 minimizes the mesh
dependency.15 κm is the track parameter of the loading
-v 33 history, defined as κm t
= max (1; Fm; κ mt –1). The mechanical
Fc,3 = #1
fc,90 Equ. 13 properties are given in Table 3.15, 16
Damage in tension or due to shear will be brittle (Ft, 1, 4.4 User-defined field (USDFLD) in ABAQUS
Ft, 2 , Ft, 3, F v, 2 , F v, 3), whereas modes in compression will be To modify the compliance matrix, we implement
ductile (Fc, 1, Fc, 2, Fc, 3). While for Fm ≤ 1 or (Fm – κm) ≤ 0, the our CDM by an USDFLD subroutine, whereas Sandhaas de-
material is in the elastic range, and the damage para- veloped a user material subroutine ( UMAT ). The elastic
meter dm (κm ≤ 1) = 0. Damage for mode m initiates when properties of the material were described in ABAQUES
Fm ≥ 1; thus the history parameter, κm, and consequently as “field dependent.” Six fields are defined to represent
the damage parameter, dm, increase. When dm (κm → ∞) the damage: Eii = (1 – f i) E 0ii for 1 ≤ i ≤ 3, Gij = (1 − f i + j + 1) G 0ij
approaches 1, the material is fully damaged; it is calcu- for 1 ≤ i ≤ j ≤ 3, Vij = (1 − f i) V 0ij for 1 ≤ i < j ≤ 3, where f1 = d11,
lated according to the failure modes. Fig. 9 indicates that f 2 = d22, f 3 = d33, f4 = d12, f5 = d13 and f6 = d23.
for ductile failure the stress will remain constant after The subroutine fixes the value of the field according to
failure initiation, and dm is expressed as: strain and stress verifications. Fig. 10 presents the simpli-
fied algorithm of the subroutine.
1
d m = 1– l m Equ. 14
el el
fmax fmax
1 1
E E
(1-d)E
1
(1–d)E
1
max
d=0 0<d<1 d=1 d=0 0<d<1
Fig. 8 Brittle behavior Fig. 9 Ductile behavior
217
90
Get strain at time t
εt 80
70
Ultimate moment (Nm)

60
Stiffness (Nm/°)
Get elastic stress
σ t, el = (C ) –1εt
26
50
24
22
40 20
18
16
30
14
Get value of yield function 12
Fm (σ t, el ) 20 10
8
6
10
4
2
0 0
Actualise history parameter Fig. 11 a b c d e f g h i j k
t t, el t–1
κm = max (1, Fm (σ (, κm ) Sample
Get damage variable

dm (κmt ) 90
80
Actualise field 70
fi
Ultimate moment (Nm)
60
Stiffness (Nm/°)
26
50
24
22
Compute 40 20
C dam,σ t= (C dam )–1εt 18
16
30
14
12
20 10
Fig. 10 8
6
10
4
Fig. 10 Algorithm of the USDFLD subroutine 2
Fig. 11 S01—Ultimate moment and stiffness 0 0
Fig. 12 a b c d e f g h i j k
Fig. 12 S02—Ultimate moment and stiffness Sample
5 Results and discussion 5.1.2 Nonlinear moment-rotation curves

Figs. 13 (S01, closing) and 14 (S02, opening) show
5.1 Experimental results the average moment-rotation full curves. The sample
5.1.1 Replicate comparison names in the legend correspond to those in Tables 4 and 5.
The sequence of rotations that defined the final After failure, a residual moment remains for
geometry of each sample gave an asymmetry with re- all the samples. It never drops below 50% of the ultimate
spect to the bisector plane of the two connected panels. moment except in the case of sample (c), which it does in
For each sample, four replicates were tested by pushing both directions (S01 and S02). This sample later showed
the F0 panel for the two first samples and F1 for the last the best compromise of stiffness in closing and opening
two. Figs. 11 (S01, closing), and 12 (S02, opening) present (0°, 0°, 30°). Samples with the most rigid connections
the ultimate moment and stiffness of each replicate. generally showed quicker softening after damage. The
Despite the asymmetry of the joint, we did not observe hardening differed between closing (S01) and opening
a large difference between pushing on F0 or F1, both for (S02), with only sample (h) hardening in S01 and all but
S01 and S02. Therefore, the rest of the analysis considers (f) and (d) hardening in S02. The MTSJ is globally fairly
the average of the four replicate values. ductile, although its stiffness in bending is far weaker

moment (Nm)
moment (Nm)
(a)
(b)
70
(c) 30
(d)
(e) 60
25
(f)
(g) 50
(h) 20
(i)
40
(j)
(k) 15
30
10
S01—Moment-
Fig. 13 20
rotation average
nonlinear curves for
bending while closing 5
10
S02—Moment-
Fig. 14
rotation average
nonlinear curves for
bending while opening
0 10 20 30 40 0 2 4 6 8 10 12
Rotation (°) Rotation (°)
Fig. 13 Fig. 14
moment (Nm)
moment (Nm)
70
30
60
25
50
20
40
15
30
10
S01—Moment-
Fig. 15
20
rotation average linear
curves (between 10 %
and 40 % of m max) for 5
10
bending while closing
Fig. 16 S02—Moment-
rotation average linear
curves (between 10 % 0 10 20 30 40 0 2 4 6 8 10 12
and 40 % of m max ) for
bending while opening Fig. 15 Fig. 16
219
S 02 S 01
F F F F0 F F1
F0 F1
Uplift Uplift
S 01 S 01
S 02 rigid S 02 rigid
Fig. 17 High-stiffness effect on glued and screwed joints
Moment max Stiffness Moment max Stiffness

Sample θ1–θ 2–θ 3 mean std mean std Sample θ1–θ 2–θ 3 mean std mean std
Nm/° Nm/° Nm/° Nm/°
a 00-00-10 29 2.8 2.0 0.2 a 00-00-10 16 1.2 1.7 0.6
b 00-30-10 25 3.3 2.2 0.3 b 00-30-10 28 2.7 2.8 0.4
c 00-00-30 29 2.7 3.1 0.2 c 00-00-30 30 2.9 2.8 0.8
d 45-00-10 15 1.4 0.5 0.1 d 45-00-10 17 1.3 3.2 0.7
e 45-31-10 28 5.7 3.1 0.4 e 45-31-10 22 2.0 2.4 0.4
f 45-00-38 26 2.4 1.4 0.6 f 45-00-38 33 5.7 3.0 0.6
g 90-00-10 23 4.1 1.3 0.3 g 90-00-10 15 2.1 1.6 0.7
h 90-30-10 30 4.3 2.0 0.5 h 90-30-10 25 0.9 2.4 0.5
i 90-00-30 26 2.1 2.7 0.4 i 90-00-30 28 1.7 2.6 0.7
MTSJ 25.5 4.4 2.04 0.9 MTSJ 23.8 6.41 2.5 0.6
j screwed 49 2.5 4.3 0.1 j screwed 68 15.2 8.9 1.1
k glued 43 9.7 11.3 1.0 k glued 77 5.8 25.3 2.9
Table 4 S01—Mean and standard deviation Table 5 S02—Mean and standard deviation
Mean S01 Dev. Mean S02 Dev.

Row Samples θ1 θ2 θ2 N m/° % N m/° %
1 a,b,c 0 2.44 20 2.44 –2

2 d,e,f 45 1.67 –18 2.86 14
3 g,h,i 90 2.01 –2 2.21 –12
4 a,c,d,f,g,i 0 1.85 –9 2.50 0
5 b,e,h 30 2.42 19 2.52 1
6 a,b,d,e,g,h 10 1.86 –9 2.35 –6
7 c,f,i 30 2.41 18 2.82 13
8 a,c 0 0 2.59 27 2.26 –10
9 b 0 30 2.16 6 2.82 13
10 a,b 0 10 2.09 2 2.25 –10
11 c 0 30 3.15 54 2.83 13
12 d,f 45 0 0.97 –53 3.11 24
13 e 45 30 3.09 51 2.36 –6
14 d,e 45 10 1.82 –11 2.78 11
15 f 45 30 1.39 –32 3.02 21
16 g,i 90 0 2.00 –2 2.12 –15
17 h 90 30 2.02 –1 2.39 –5
18 g,h 90 10 2.00 –2 2.12 –15
19 i 90 30 2.69 32 2.62 5
20 a,d,g 0 10 1.29 –37 2.17 –13
21 c,f,i 0 30 2.41 18 2.82 13
22 b,e,h 30 10 2.42 19 2.52 1
Table 6 Comparative analysis of the combination of Bryant angles

than that of the screwed and glued connections. The vorable for adhesive bonding in that direction. As the
fastened and bonded samples showed different experi- glued and screwed samples were stiffer in S02, their mo-
mental results for closing and opening.These types of ments in S02 were higher than in S01 for a given rotation.
connections are far more rigid than the MTSJs; they were The highly brittle failure of the glued connection is clear-
stronger than the panel strength and stiffness during ly visible in figs. 13 and 14. The less rigid screwed joint
deformation. Bending occurred in panels F0 and F1, and behaved similarly to the glued joint, but with smaller
fig. 17 explains this effect on the stiff joint. differences between S01 and S02. Both connections were
In S02, the reference panel was pressed onto the tested as elements of comparison. After testing, the sam-
support with increasing load, reducing local bending ples connected with MTSJs showed no bending within
near the joint. In S01, the reference panel was lifted by the panels. The relative weakness of the connection re-
the loading, and was free to bend. Under a given load, the sulted in no high-stiffness effect. The rest of the analysis
sample deflected more in S01 than in S02. In both cases, will only focus on the MTSJs.
the glue was subjected to tensile stress, which was unfa-
35 35
moment (Nm)
moment (Nm)
30 30
25 25
20 20
Model Model
15 15
S01.a.F0.01 S01.c.F0.01
10 S01.a.F0.02 10 S01.c.F0.02
5 S01.a.F1.01 5 S01.c.F1.01
S01.a.F1.02 S01.c .F1.02
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Fig. 18 S01—Numerical curves versus experimental results, a) Sample a (0°; 0°;10°), b) Sample c (0°; 0°; 30°)
35 35
moment (Nm)
moment (Nm)
Model
30 30
S02.a.F0.01
25 S02.a.F0.02 25
S02.a.F1.01
20 20
S02.a.F1.02
Model
15 15
S02.c.F0.01
10 10 S02.c.F0.02
5 5 S02.c.F1.01
S02.c .F1.02
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Fig. 19 S02—Numerical curves versus experimental results, a) Sample a (0°, 0°, ±10°), b) Sample c (0°, 0°, ± 30°)
221
Fig. 20 MTSJ: open slot (left), closed slot (right).
5.1.3 Linear moment-rotation curves 5.3 Comparative analysis of stiffness

Figs. 15 and 16 were obtained by taking an aver- To evaluate the effects of the Bryant angles and
age of the linear fits of four replicates of each sample. The their partial combinations on the semi-rigidity of the
limits of the linear range were chosen between 10% and MTSJ, mean values of stiffness are listed in Table 6.
40% of the ultimate moment. The curves are displayed in For each group of samples that has the same values for θ1,
this range. The sample names in the legend correspond to θ 2 or θ 3, or any combination thereof, the mean value of
those in Tables 4 and 5. stiffness is compared with the mean of all samples (see
Tables 4 and 5) for both S01 and S02. The deviations from
5.2 Global analysis the global means are listed in the columns headed “dev.%,”
The total average stiffness in closing (2.04 Nm/°) and they are considered to be meaningful when greater
was 18.5 % lower than the stiffness in opening (2.5 Nm/°), than 10 %. The full combinations (θ1 θ 2 θ 3) are listed in
whereas the ultimate moments were not significantly dif- Tables 4 and 5, but trends cannot be clearly extracted
ferent (25.5 Nm ←→ 23.8 Nm, less than 8 %). The range of from them. Here, we analyze trends revealed in the ex-
stiffness on S02 was also narrower than that of S01. This perimental results. Each sample was tested in four repli-
is confirmed by the standard deviations listed in Tables 4 cates. For clarity, the combinations are written as (θ1 θ 2 —),
and 5. In the tables, “MTSJ” refers to the group of nine (θ1 — θ 3) and ( — θ 2 θ 3) in the text. Any value greater than
MTSJ samples (36 samples per loading case). or equal to 30° is noted as 30° in the brackets. θ1 = 0° gives
the best stiffness in closing (S01), whereas θ1 = 45° reduces
the stiffness in opening (S02). Connections with θ1 = 90°
are possibly weaker, particularly in S02. The effect of
θ 3 is obvious, as increasing the value to 30° or more in-
creases the rigidity in both loading directions. For θ 2,
Model stiffness Exp. Stiffness
Sample θ1–θ 2–θ 3 mean mean std
which seems to improve the stiffness in S01, the analysis
Nm/° Nm/°
must be completed in combination with θ1. (00 00 — ) and
(45 00 — ) provide opposite results, with stiffness increas-
a 00–00–10 1.4 2.0 0.2
ing in S01 and decreasing in S02 for the former, and vice
b 00–30–10 1.7 2.2 0.3
versa for latter. In contrast, (00 30 — ) was stiffer in S02
c 00–00–30 2.6 3.1 0.2
and (45 30 — ) was stiffer in S01. Combining θ 2 with θ1 = 90°
d 45–00–10 0.6 0.5 0.1
did not greatly alter the results relative to the total mean.
e 45–31–10 1.4 3.1 0.4
The combinations ( — θ 2 θ 3 ) confirmed the positive effect
f 45–00–38 1.1 1.4 0.6
of θ 3 on the stiffness, except in the case of sample (f)
g 90–00–10 0.7 1.3 0.3
h 90–30–10 1.0 2.0 0.5
(θ 3 = 38°), which showed a rather low stiffness in closing.
i 90–00–30 1.8 2.7 0.4
The tabs were stretched in S01 for this configuration.
Above a certain value (>30°), the normal of the locking
Table 7 S01—Model stiffness versus experimental stiffness faces approaches the orientation where the tension per-

pendicular to the grain starts reducing the strength and 6 Conclusion
stiffness. The sensitivity to large θ 3 angles could be ex-
plored in future works. ( — 00 30) and ( — 30 10) both have Producing strong edge-to-edge connections in thin pan-
a similar effect on S01, and again highlight the beneficial els is challenging. Usual metal fasteners such as screws
effect of θ 3 = 30°. However, θ 2 ≥ 30° did not have any major may not comply with the required edge distance. The
effect during opening (S02). Finally, the best compromise MTSJ presented in this paper is an innovative solution. It
of the nine samples is θ1 = 0°, θ 2 = 0°, θ 3 = 30°, which takes allows the in-space, multi-edge interlocking connection
advantage of the positive effects described above. Accord- of panels. The joint is milled during cutting. Its shear per-
ing to the trends shown in this study, a joint definition formance has already been assessed mechanically11, and
with θ1 = 0°, 10° ≤ θ 2 ≤ 30°, 15° ≤ θ 3 ≤ 30° could behave effi- the semi-rigidity of the MTSJ has demonstrated proper-
ciently. As described in section 2.2, the available combi- ties competitive with wood screws. The bending stiffness
nations are limited by the tool inclination and the multi- on the ridges of the folded-plate structures is considered
edge assembly constraints. The range of possible tool in- here. The first tests on the rotational stiffness of MTSJ
clinations could be increased by altering the shank length with nejiri arigata design revealed ductile behavior
and the tool-holder geometry. with relatively low stiffness compared with screwed and
glued joints. The joints were tested in their minimum
5.4 Numerical model versus experiments configuration with one and two tabs per panel. The effect
A numerical model was built to reproduce the ex- of the tab angle θ 3 is significant, and its combination with
perimental set-up as faithfully as possible. For analyzing the other angles is promising for further analysis. The
the rotational stiffness, the elastic range would have sample with θ1 = 0°, θ 2 = 0°, θ 3 = 30° behaved well, but joints
been sufficient. As the ultimate moment of the connec- defined as θ1 = 0°, 10° ≤ θ 2 ≤ 30°, 15° ≤ θ 3 ≤ 30°could be even
tion is monitored experimentally, comparison with a nu- more efficient. A numerical model based on continuum
merical value would be useful, and a damage model was damage mechanisms showed promising results, although
implemented to achieve this. Some preliminary results it also showed limitations that must be addressed. These
are presented as follows: preliminary results will facilitate the development of
In both S01 and S02, numerical strain and dam- stiffer joints, and further work is already under way.
aged element were well observed experimentally. Consid- Fig. 20 presents the evolution of the MTSJ with closed
er, for example, sample (a) in fig. 18: the numerical model slots; the sample (a) (θ1 = 0°, θ 2 = 0°, θ 3 = 30°) is shown in a
matched the experimental curves well under closing. joined position for comparison. Bending experiments
However, at this stage, our model aborted before reaching have already demonstrated its stiffness to be equivalent
the ultimate strength when the samples were tested to that of a screwed joint. The connections have already
while opening (Fig. 19). The experiment halted before been used in a double-layer, curved shell.
reaching the yield point, thus preventing the observation
of the ultimate moment. The continuum damage model
worked well, but convergence problems were encoun-
References
tered during the contact step, so that only the stiffness
values of the model in S01 were listed in Table 7 and com- 1Wardle, T. “Old Growth: Our Timber Engineering Heritage.
Design Notes for Plywood Folded-Plate Roofs.” In: New
pared to experimental values. They were generally sig-
Zealand Timber Design Journal, Issue 1, Vol. 19, 548–551, 2001.
nificantly lower than the averages of the experimental
2 McDowall, M., M. Leigh, and M. Punton. “Exotic Structural
results. This was probably due to the deviation from actu- Forms—Folded plates.” In: Structural Plywood & LVL Design
al values of both material properties and the friction co- Manual, Chapter 12, Engineered Wood Products Association
efficient (set to 0.3 in our case). The development of the of Australasia, ,154–158, 2007.
numerical model remains ongoing, and pending issues 3 Jaksch, S., A. Fadai, and W. Winter. “Folded CLT structures—
have to be solved by: Developments in Design and Assembly Strategies.” In: World
Conference on Timber Engineering, Vol.1, 18–22, 2012.
– Re-defining the contact step and the Coulomb 4 Buri, H., and Y. Weinand. “Origami—Folded Plate Struc-
tures, Architecture.” 10th World Conference on Timber
friction coefficient
Engineering, Miyazaki, Japan, 2008.
– Refining the elastic and mechanical properties
5 Buri, H. “ Origami—Folded Plate Structures.” Thesis, Swiss
by material testing on Kerto-Q panel and Federal Institute of Technology Lausanne, 2010.
spruce veneer
6 Robeller, C., A. Stitic, P. Mayencourt, and Y. Weinand.
– Identifying dislocated elements with large “Interlocking Folded Plate—Integrated Mechanical
rotation and mesh modification Attachment for Structural Wood Panel.” In: Advances in
– Controlling the time increment Architectural Geometry, University College London, London:
– Conducting mesh optimization Springer, September 18, 2014.
– Migrating from ABAQUS/Standard to 7 Haller, P., Cost C., eds. Semi-rigid Timber Joints‚ Structural
ABAQUS/Explicit. Behaviour, Modelling and New Technologies.

European Commission, 1999.
223
8 Jaspart, J.P. “General Report: Session on Connections.”
In: Journal of Constructional Steel Research, Vol. 55, 68–89,
2000.
9 EN 1995-1-1, Eurocode 5: Design of Timber Structures‚ Part
1-1: General-Common Rules and Rules for Buildings. European
Committee for Standardization, Brussels, 2004.
10Sebera, V., and M. Simek. “Finite Element Analysis of
Dovetail Joint Made with the Use of CNC Technology.” In: Acta
Universitatis Agriculturae Et Silviculturae Mendelianae Brunensis,
Vol. LVIII, Nr. 3, 321–328, 2010.
11 Roche, S., C. Robeller, L. Humbert, and Yves Weinand.
“On the Semi-Rigidity of Dovetail Joints for the Joinery of LVL
Panel.” In: European Journal of Wood and Wood Products, 2015.
12 Sandhaas, C. “3D Material Model for Wood, Based on
Continuum Damage Mechanics.” In: Stevinrapport 6-11-4,
Stevin II Laboratory, University of Technology Delft, 2011.
13Robeller C. “Integral Mechanical Attachment for Timber
Folded Plate Structures.” Thesis, Swiss Federal Institute for
Technology Lausanne (EPFL), 2015.
14 Mattoni, G. “Folded Plate Structure, Design and Analysis
of Woodworking Joints for Structural Timber Panels.”
M.A. Thesis, ENPC and EPFL, 2014.
15Sandhaas, C. “ Mechanical Behaviour of Timber Joints With
Slotted-In Steel Plates.” Thesis, University of Technology
Delft, 2012.
16Ivanov, I., T. Sadowski, M. Filipiak, and M. Knec. “Experi-
mental and Numerical Investigation of Plywood Progressive
Failure in CT Tests.” In: Budownictwo i Architectura, 79–94,
2008.
Reprinted from International Journal of Space Structures,

vol. 30, No. 2, 2015.

225
5. 4
“Digital fabrication leads to a new building culture”

Interview between Matthias Kohler (chair of Architecture and Digital Fabrication, ETH, Zürich) and Yves Weinand
February 17, 2015
Yves Weinand: When I look at your work, it’s clear technologies is at the core of our research. Since our
that you have a strong interest in the process objective is not only an outcome, but also the precise
and in the exemplification of the process. Of control of each progressive step, it is often the case that
course you also consider constraints, such as a the local constraint, i.e. the detail, is significant in the
flat panel or a sheet of paper, for instance. search for entirely new processes of creating physical
As a general goal, we are looking for a structural form. So through digital fabrication we establish a direct
form. We are looking for a connection between connection between material and its processing. Precise-
structure and form. I don’t know exactly how ly this design of processes—of how you make things—
to describe it, so I tend to speak about a “close creates new opportunities for architectural design. How-
relationship” linking structure and form. ever, let’s not be naïve; neither establishing processes,
We realize that when we integrate local obser- nor controlling digital data is a guarantee of good de-
vations, the connections also become stronger. sign. Nevertheless, it opens up interesting possibilities
In general, we think in terms of global form- for design, or the creation of new processes of creating
finding tools. But I believe that there is a architecture in a generative computational manner,
link between the overall geometric situation informed equally by specific small-scale details as by
(geometric parameters) and the local situation. large-scale design intentions.
We have different angles and different ap-
proaches. Could you first develop the process Y. W.: When we look at the brick wall projects
orientation of your own work and later con- you have been working on in the past, and then
centrate on how an architectural synthesis when we look at Eladio Dieste’s work, who was a
might be achieved? structural engineer using bricks (he also created
curved walls that are structurally informed and
Matthias Kohler: That’s interesting, as I think this rejuve- are highly resistant), I ask myself how your
nated relationship between global form and local perfor- work might be informed by Dieste. You’re not
mances—which in our case particularly relates to the interested in the global form to begin with;
process of making—is a debate that we need to have in you’re more interested in the purpose. It’s simi-
future architectural discussions. Our research is cen- lar for engineers: they don’t tend to be interest-
tered on the question of how digital fabrication princi- ed in the global form at first.
ples might lead architecture to a new contemporary
building culture. By using digital design methods that M. K.: Are you interested in the global form from the outset?
seamlessly inform new construction techniques, the
design of details becomes an essential part of an archi- Y. W.: Yes, of course. Dieste looks at how the geo-
tectural design. You addressed our interest in process—it metric positions of the bricks give structural
is true, we start from taking a material and exploring its efficiency to the wall, while you are looking at
properties, associating it with specific fabrication meth- the process and how the process influences
ods and construction systems in order to see how these material and form. I would like to clarify this
relate back to the material’s inherent capacity, and ulti- and to see how to synthesize these ideas. Dieste
mately trying to gain an understanding of the limits does not speak about the global form either.
of what we can achieve with it. By the careful analysis of I think his work is very impressive, because it
digitally controlled processes, we seek expanded design includes content between global form, local
opportunities, which are different from what is achiev- geometry, and structure. This aspect, which has
able with traditional building techniques. This highly to do with synthesis, will also form the focus of
constructive mode of thinking about design with digital our research within the NCCR.

M. K.: Our interest in the global form remains relatively M. K.: As an architect, you generally start to shift priori-
strong. You see that a lot of our work has quite an expres- ties during the design process. You might start with
sive, formal capacity. However, our true interest lies in initial assumptions and then you choose some specific
how we can explore and develop these expressions; it is constraints as being important to your reading of a
not in the form per se, nor in a preconceived narrative or situation. This is what you’re going to focus on, and then,
even a pictorial approach. In fact, I think that your after a while, perhaps after a few initial sketches, other
origami structures are not far from that, as you’re also aspects might capture your attention and become more
not designing the global form first. But with the origami important. Issues that you might have been interested
structures, the expression is inherently set. You are in at the start become subsumed by new, constantly
working with a far more constrained system and there- evolving perspectives. Therefore I ask myself: How do
fore the converse question would be: Why do you look for you deal with this when you use computational design
such tightly constrained systems, while we are always processes? Because, generally speaking, in a computa-
trying to extend the degree of design freedom? tional design process you only define your system once.
You have parameters that you can tweak, but basically
Y. W.: Well, it’s true that I already had two aspects your system is defined. You either re-program the entire
in mind when we started the origami project. system, or otherwise you develop the initial system
First, it grew from an ongoing fascination with further. Nowadays, I wonder if this is a limiting factor.
folded structures. When I began my studies as a Are we artificially constraining ourselves by not shift-
civil engineer, I was already drawn to folded ing perspectives within the design process? Or will we
structures. I was always interested in this and have computational design tools one day that are more
in the “origami aspect” of it. I like to inform open to shifting design perspectives?
the subjective design process with objective
elements, which are more related to nature or Y. W.: I don’t believe it, as the richness lies in the
scientific observation. The constraints of fact that you have something similar to case
the developed tool are that you can flatten it. studies.
M. K.: When you generate specific structural designs and M. K.: What don’t you believe in?
when you develop the tools needed to generate them, do
you have a preconceived idea of what they should look Y. W.: That you can do this, that you can have a
like? Or are you primarily interested in exploring the computational framework, which you can
possibilities of an abstract, constrained space? change according to the progress of the design.
It’s more like you still have to choose, as a
Y. W.: It’s like a series of trial-and-error attempts carpenter does, the right tool to do a specific job.
in different directions. Geometrically speaking,
there is no overall framework to it. There is M. K.: But then, you’re talking about having a sequence of
still no answer to this. I am unsatisfied by the tools?
fact that the synthesis cannot be understood
as a geometric synthesis in any way. It remains Y. W.: Yes.
open—it’s more like a body of work that is
presented in a specific way. M. K.: If this is a given, it might also be contrary to the
notion that you could have a more integral and
consistent computational design process throughout
an agency.
227
Y. W.: Mark Pauly spoke about this contradiction lishes a different relationship between the technician
with regard to a case study. Although he found it and the real, material world. For example, if you have
very interesting, he said that if you want to machines that can calculate how material behaves and
analyze this in a global manner, it would be far at the same time work with these materials—is this idea
too complex. of a geometric ontology still a powerful and central
architectural model?
M. K.: As a computer scientist, he says it’s an impossibility
and, as an architect, you also believe that it’s not clever Y. W.: I wish that it had been possible. I spoke
to overload computational generative tools by having about our work with a few French mathemati-
expectations that are too high. cians and they indicated that you could order
these elements in some kind of geometric frame,
Y. W.: My aim would have been to find some kind but it has not been done yet.
of geometric frame.
M. K.: Is this all just an ideal?
M. K.: Why geometric?
Y. W.: Yes, it’s an ideal. It’s been done very
Y. W.: I believe it could put some things in order. intuitively.
But that order would not be based on a computa-
tional tool definition. In construction, it always M. K.: You care a lot about details in your work, while at
comes back to geometry. the same time, you say that in the end it’s just a geomet-
ric model. I would like to challenge this a little bit, hav-
M. K.: Though, in order to provoke you a little, I would say ing visited your lab, and seen all the models around and
this is a déformation professionnelle, a bias introduced all the physical explorations. What role does physical
by your profession. Because, rather than geometry, in and empirical exploration play in your view? Because,
the end it always comes back to physical construction alternatively, you could also say that you concentrate on
and materials. theoretical work and the exploration of geometric space,
but I believe this wouldn’t be of interest to you, right?
Y. W.: Well, you can say déformation profession-
nelle, but it’s true that I have the impression Y. W.: Yes, I think there is clearly an architectural
that I can achieve the synthesis between struc- motivation. But, this form needs to be structur-
ture and form within geometric constraints. If ally informed. Sometimes I have to indicate that
you look at the history of construction in archi- there is structural information in our architec-
tecture, you always have very important points. ture that we have been able to demonstrate.
Sometimes, it’s the geometry that gives us clues,
sometimes it’s on another level. It has remained M. K.: What’s the role of aesthetics for you? When you
on the geometric level several times. make decisions, for example, to what degree are
you driven by performance and to what degree are
M. K.: Clearly, geometric processing allows a high level of you driven by aesthetics?
control—not only in architectural design, but also in
structural analysis. Yet I wonder if the relationship Y. W.: Clearly we are aware of the aesthetic qua-
between geometric calculation—taking geometry as a lity. I think it’s more a result of a synthesis
foundation for design decisions—and architecture is not of structure and form that we are able to achieve
outdated and superseded by the computer, which works something stronger, for instance in its ex-
according to numerical processing, and therefore estab- pression, but I’m unsatisfied with the word

“tectonics,” for example. It’s related to mechanics, what you are requesting. You are requesting
but it’s unrelated to what you are saying about something beforehand, an earlier stage, some
tectonics in architecture, which is more related sort of information set.
to a kind of objective aesthetic criteria, where
you can play with local and global geometry. We M. K.: For me this is simply a matter of intellectual curios-
have that, but again, I can’t explain it. If I had ity. Can computational design processes be radically
to explain it, I’d look at the geometry. The project distinct from classical iterative design procedures? If
“fractal geometry,” for instance, was clearly a you define the scope of what you want to include in such
speculation about similarities at various differ- a “synthetic process” and make it comprehensive and
ent scales. I find this idea fascinating. logical, is it actually feasible? Or do we run the risk of
shifting architecture towards technocratic, determinist,
M. K.: My assumption would be that you deliberately and optimization-driven design procedures, which leave
focus on finding the limits of these architectural spaces, out key aspects of creativity? These are tricky questions.
while at the same time you’d like to explore the respec-
tive aesthetic capacities. In contrast to our research, Y. W.: Well, you have to clarify the vocabulary:
where we are less interested in constraining and more What do you mean by “constraints,” “degree,”
interested in maximizing degrees of freedom when etc.? I could answer: “Well, you have a maximum
designing. Honestly, we don’t care so much about struc- of 6 degrees of freedom.” It’s like an engineer
tural optimization. We look for project-specific drivers that would see this degree of freedom locally,
that are conceptually interesting for the design. but I’m sure that you have a larger interpreta-
When we work with our partners, we realize that a tion of what you call “degrees of freedom,” as
certain aspect that we are looking for is, in fact, you were saying earlier. Clarification of the
unavailable. For example, when we were designing the vocabulary would be needed. “Constraints”
robot ically fabricated brick wall for the Venice Archi- are also different for a civil engineer than for a
tecture Biennale (2007/08), we wanted to explore if it mathematician, for example.
would be possible to extract specific rules regarding
the structural behavior of this double-curved entity. M. K.: I agree. When I talk about “degrees of freedom,” I’d
However, we could not develop more than a rule like to refer to this in a very broad and open sense, or in
of thumb with the engineer. It is precisely this detailed other words, in a conceptual way as an architect. So, for
knowledge about how complex structures ultimately example, for the Venice Biennale we left the form open
perform that would be interesting to analyze from a right up to the day of production, because the form could
design perspective. be generated at any moment. This “degree of freedom”
was built into our design concept. Nevertheless, architec-
Y. W.: (Looking at the modular timber structure tural questions need to be answered: What would the
pavilion) We designed and built this wall installation look like? How would it appear to the public?
structure as a prototype, then I used it as a basis What would its formal capacities in conjunction with its
for a civil engineering class, to explore design structural performance be? All these aspects had to be
optimization and geometry structurally meticulously designed, even though the final geometric
speaking. In the end, some students were able to definition of the global form itself was deliberately left
change the global geometry in order to optimize open. This is an example of what I mean by “conceptual
the structure. I asked Mark Pauly, how we degree of freedom.” That’s what I’m particularly inter-
could do this in a different way; other than in ested in. But there are also physical degrees of freedom,
an iterative way? Pauly said that we have to use which we need to figure out. That’s the issue of new
a kind of iterative process, which is contrary to fabrication technologies. Where is it possible to liberate
229
ourselves from certain traditional constraints or models? that you cover both arguments is fascinating.
Are these liberations just interesting as a limited gain, Low-tech parametric design could be used
or do they have a disruptive potential for a factual all over Africa. Then, people could adapt the file
impact on the overall building practice, the building to specific local conditions. I think this is an
culture, and its design? architectural strategy.
But coming back to our previous discussion: What about
the pavilion’s structure you were talking about? M. K.: So, you are also conceptually working with degrees
of freedom and adaptation. I think this is interesting,
Y. W.: Yes, it was flexible. because there are indeed multiple levels where we
could use these new technologies. Let’s focus a bit more
M. K.: What do you mean by “flexible” here? on the material research you are pursuing. You’re using
large-scale timber, mainly LVL and CLT panels.
Y. W.: Well, according to wind load, we had to Wood is basically a renewable resource, a natural
stiffen it, as we only adjust one, single contact material, but at the same time, you’re using it in a highly
point and then we analyze it. What I would processed, highly engineered form. As you mentioned,
like to do is to return to the global geometry you’re interested in mass-produced, large-scale-products,
observation and change the position of each and how to work with them very efficiently. Where
piece and allow them to delve far deeper into do you see important development paths, and what do
each other, in order to increase a static height. you think the main advantage of these large-scale
Thus, it was actually the manipulation of products would be?
the global form, in order to achieve structural
optimization. It’s an iterative project. Y. W.: I think they could provide very efficient
constructions; for instance, speeding up con-
M. K.: I would like to go into more detail regarding struction time.
timber. But first I would like to ask where you see
your work being applied? M. K.: Do they use these panels in Rwanda?
Y. W.: The most appropriate application would be Y. W.: No, not yet.
for public and sports infrastructure. I have
recently been collaborating on a project to build M. K.: Could you tell me a bit more about the advantages
drone ports (airports for drones) throughout of these engineered timber products in comparison
Africa. I thought we should develop a timber to the use of local resources, which certainly have lower
modular product in Rwanda using parametric structural capacities? Are you interested in the topic
design tools inspired by vernacular aspects of of resources, and where do you see the advantages and
local patterns. potential developments?
M. K.: Are you talking about aesthetic and decorative Y. W.: I believe that timber has not been devel-
aspects, or of inherent construction aspects? oped to its full potential because of its formal
instability when you use massive wood.
Y. W.: There isn’t only one solution; even structur- In Switzerland some people have managed to
ally, there are several. You can adapt the read- use it for centuries, because they have a keen
ing of the mechanically working connection— understanding of where to place pieces and how
in aesthetic terms—to local patterns. So, they do they could transition over time in order to pre-
both: they function and they integrate. The fact vent damage to the building due to movement.

But this is not possible in terms of thermal M. K.: Indeed, this simplicity is also very appealing to me.
insulation and our construction methods today, What’s your strategy in pursuing this kind of research,
in terms of ventilation. I believe these products, and what new findings are you collecting there?
for example LVL panels, are the right solution,
in order to have a formal stability. In addition, Y. W.: When you look at civil engineering struc-
I appreciate the mechanical performance of tures, like in textbooks, you have massive steel
these elements, because you remove the problem beams, or concrete beams, or you have steel
of natural anomalies as you have it in massive plates, etc. The idea is to take all of these out,
timber. This gives you triple the resistance because you pay for the beam and you pay for
values compared with massive timber. the connection, which is where so many prob-
lems occur. We deduct the price of the beam
M. K.: Do you think there will be an increase in the use of in timber constructions. When you construct a
panel products in developed countries, and do you think sports hall from glue-laminated beams, you
they will be used as structurally integrated components? halve the price in cubic meters of laminated
timber and halve the price of assembly, as well
Y. W.: Yes, and they would be understood by as costs and hassle due to connections. This
architects who would relate the panels to needs integration; there needs to be an integral
architectural concepts that are also suited to attachment that’s part of the system, rather
this material. than something added to it. The idea is to re-
move this when you have the structural draw-
M. K.: Did you ever consider intervening in the material ings and the production drawings, so we are
process chain? Are you interested in discussing able to fuse these two drawings into one. This is
how these products are made and maybe customizing something we have to convince industry about.
the way in which these products are manufactured?
M. K.: In robotic fabrication, we attempt to collapse most
Y. W.: Of course. We tried to produce welded, processing steps into one, continuous robotic process. So
cross-laminated timber panels. The welding by different means we are also looking for new efficien-
process by friction proved to provide sufficient cies, new simplicities, integrating these processes in a
resistance. The shear resistance is less important complex product. I’m not so sure about large panels in this
than in a glued-laminated beam. We approach regard. How will you explore the efficiency of assembly?
it from the construction side, rather than from
the material science side. Y. W.: I’m still analyzing the current situation, on
how industry produces small attachments.
M. K.: What about the role of customized timber joining I’m pragmatic about this; I don’t really feel the
techniques in your overall design research at IBOIS? necessity to conceptualize some kind of assem-
How do you want to explore this further in the wake of bly procedure beforehand, and then to look
Christopher Robeller’s thesis? at what we actually want to build.
Y. W.: When I look at Markus Hudert’s work with M. K.: If I take the example of the curved brick wall that
wooden structures, we spent a lot of time trying we built at the Architecture Biennale, we had to reinvent
to find and understand the geometry of the a new joining and support technique due to the fabrica-
connection of those two panels. We would like to tion process. This was not part of the original design
develop all this out of timber in a very simple way. proposition. Should the assembly not be a vital part of
your development of integral attachments?
231
Y. W.: Yes, of course! We want to use thin panels Y. W.: We have often discussed discrete elements,
that can’t be screwed, as we don’t have sufficient like bricks for instance, and how to attach them.
distance to the edge. That’s the reason why Also, regarding timber construction, we have a
everything breaks or pulls apart. The connec- number of discrete elements that we need to
tions we are currently working on no longer assemble. We need to figure out how every piece
have these problems. can be mounted. This process can be helped by
the use of robots. For example, if you take our
M. K.: I was talking about flat joints, where you need to woven structures, you could imagine two or
introduce extra pieces within the manufacturing three hands with robotic arms, because they are
process. That being the case, you are realistic and prag- self-supported structures, they could be connect-
matic about what’s available. ed at one point with two arms, and then there
would be another arm, which could be working a
Y. W.: Matthias, don’t you think that the program short distance away, etc.. It means that the
of the pavilion has become obsolete? form-finding procedure would be a single stage
during the construction process. There, I see
M. K.: Yes, absolutely! something very different that could be done by
robots.
Y. W.: What do you propose?
M. K.: In our research, the critical question is always
M. K.: I propose to focus our energies on buildable, multi- whether robots add something to the construction pro-
story constructions. How else can we challenge ourselves cess, or if they only replace human labor. Regarding bent
to be more relevant to the actual building practice? elements, for example, robots could clearly bring an
And furthermore your work—like ours—is primarily added value to this process. Robots could sense and
mono-material. Of course, we are all aware that buildings validate how the deformation is physically performing,
are not made of only one material. Thus, I would also and then adapt the next panel to the actual deformation.
put forward the question of how different materials and There are substantial skills that robots can add in this
technical subsystems could become integrated. I think particular case.
these questions are important because many people I have one more question about research culture: What
still only recognize the formal aspects in these kind of aspects do you think are specific to the research culture
pavilion-explorations. The actual impact of these new of architects and engineers, where do they differ? What
technologies at a larger scale is what I think remains aspects do you think we should focus on in an architec-
open and represents a larger, more fundamental tural research culture?
question.
Let’s return to robots and what is pragmatically possible. Y. W.: I think it has to do with what you men-
We were discussing very thin panels and—because we tioned before; for example, that you allow your-
are working with very small-scale elements that could self to have different interpretations on the
be handled by robots, such as bricks or small timber degree of freedom. This opens up the discussion
members—our focus is on the automated assembly of and also allows interdisciplinary collaboration.
such small elements, while your focus is on the geo- I’ve been writing a short article about this in
metric conditioning of larger panels. Therefore, do you the “Best of” publication, where I explained the
have ideas about how there could be synergies between advantages of working in an interdisciplinary
these two research approaches? manner on a scientific level to EPFL’s scientific
community. It is like removing certain para-
meters which are understood in a certain way in

References
different disciplines and trying to connect
them in a different, more organic, sometimes 1 The National Centre of Competence in Research (NCCR)
Digital Fabrication—Advanced Building Processes in Archi-
empirical way. There is a lot for the scientific
tecture is being hosted at ETH Zurich, directed by Prof.
community to learn about the way architects Matthias Kohler, in collaboration with 13 laboratories—8
work if they work on a certain level. laboratories from ETHZ, 2 laboratories from EPFL, 2 from
EMPA, and 1 from BUAS. Included is the Laboratory for
M. K.: You also mentioned what can be learned from Timber Construction, IBOIS, directed by Prof. Yves Weinand.
engineers and what kind of specific qualities we need as 2 Stotz, I. “Iterative geometric design for architecture.”
architects in particular, in research? When would you Thesis no. 4572, EPFL, Lausanne, 2009.
consider something relevant for research in architec- 3 Hudert, M. “Timberfabric: Applying Textile Assembly
Principles for Wood Construction in Architecture.” Thesis no.
ture? In engineering, it seems that there are clear crite-
5553, EPFL, Lausanne, 2013.
ria, which we lack in architecture. What’s your personal
4 Weinand, Y., “The Architecture Studio as a model?”
opinion on this? In: Veillon C., and N. Maillard (eds.), Best of 2012.
Editions Archizoom, Lausanne, 2011., pp. 15-16.
Y. W.: It is true that architects are inspired by a
broader spectrum of elements in design.
There is a shift in the manner of how you judge,
or find, or identify yourself with the production
of architecture. I relate to architects and accept
their openness to a broad range of influences.
I think there is an attitude on the part of
the architects, which is related to the fact that
the individual position is more frequently
questioned by existing techniques.
M. K.: It’s a kind of contextualizing of the individual and

also educating the individual in such a way that you are
able to position yourself within a more complex world.
Y. W.: Yes, and I see the opportunity for that.
233
5. 5
Curved-panel wood pavilion

Marielle Savoyat
Project IBOIS,
IBOIS—Laboratory
Labor für Holzkonstruktionen,
an der A recent research project undertaken by IBOIS, the Lab-
design
EPFL,
EPFL—Swiss
Eidgenössische
Federal Technische
Institute ofHochschule
Technology, oratory for Timber Constructions at the Swiss Federal
Lausanne,
Lausanne,Schweiz
Switzerland/Prof.
/Prof. YvesYves
Weinand,
Weinand, Institute of Technology in Lausanne, was focused on com-
Christopher
ChristopherRobeller,
Robeller,Sina
SinaNabaei
Nabaei,und
andHans
HansBuri plex geometries and their application to wood and, more
Buri specifically, on the curve potential of this material.
Project MerkTimber
Merk TimberGmbH,
GmbH,Deutschland
Germany In order to explore these questions, a delicate prototype
execution pavilion in cross-laminated timber (spruce) panels was
Completion2013
2013 created in autumn 2013 in the Academy of Architecture’s
gardens in Mendrisio, Switzerland. Such a slender render-
Location Accademiadidiarchitettura
Accademia architettura(Akademie
(Academy für
of Archi- ing of panels, as adopted for this pavilion, had not
tecture), Universita
Architektur), Universitadella Svizzera
della Italiana
Svizzera Italiana yet been attained at this scale. The resulting level of
(University of
(Universität derItalian Switzerland),
italienischen Mendrisio,
Schweiz), resistance and lightness achieved by the curve factor
Switzerland
Mendrisio, Schweiz was ground-breaking, displaying technical prowess and
paving the way for new architectural forms. The light-
weight construction spans over 13.5 meters, with a panel
thickness of a mere 77 millimeters. It was presented in
parallel with the Timber Project exhibition, held at the
Accademia gallery from September 20 to October 27, 2013.
The initial concept behind the pavilion emerged as a re-

sult of Hans Buri’s thesis on origami (completed in 2010),
where he recognized that the Japanese art of folding
could be transferred to curved folds, creating new possi-
bilities for wooden structures. While wood does not lend
itself to double curvature, new surfaces can be created
based on curved lines. The researchers who succeeded
him, Christopher Robeller and Sina Nabaei, demonstrated
that by replacing two zigzag polylines with two wavy
lines, new types of forms applicable to timber construc-
tions could be generated. This realization led to further
questioning and explorations, which were then rendered
concrete in relation to wooden material through the
study and construction of a prototype pavilion. The same
software application developed for origami allowed the
development of deformed or pre-deformed structures.
As a result of this program, a greater comprehension of
geometry was achieved through observing different
parameters, such as the curve radius, overall dimensions,
or number of connections. The various measurements
that emerged through computer simulations during the
Fig. 1 Detail of the connection modeling of the deformation processes, including those

Fig. 2 The pavillon by night
caused by natural elements such as wind or gravity, also digital calculations and simulations, while the construc-
offered a broad spectrum of variations for determining tion firm confirmed the feasibility in the execution of the
the dimensions of the form. project. The elegance and slenderness achieved results in
The other novelty of this pavilion, other than its a controlled aesthetic, with a sense of coherence between
curved, folded-form resistance, resides in the wooden form and structure mastered through their geometric
dovetail joints, which allow a more precise connection be- parameters.
tween the curved elements without the use of metallic
connectors. The geometry of the curves is rendered digi-
tally, thus determining the form of the connections. A
close collaboration between IBOIS and the German con-
struction firm Merk pushed the limits of technology: the
research laboratory proved theoretical results through
235
The Research Laboratory IBOIS
at the EPFL Lausanne
The domination of steel, and later reinforced concrete My approach regards the mechanical requirements of
during the past two centuries in the research, the practice form and structure as attributes that can gain full mean-
of civil engineering, and materials science has led to a se- ing and sense only in the context of the geometrically
vere lack of research regarding wood as a construction scaled phenomenon on which they depend. I consider the
material. The intuitive knowledge of carpenters and our use of the digital representation of architecture as an in-
skilled predecessors has been lost since the profession valuable tool that ought to strengthen the integration of
of “ingénieur des ponts et chausses” was evolved in the structure, form, and material within our design concept.
eighteenth century. Today, many engineers do not use However, digital modeling cannot replace the study of
timber as a building material as they assume a priori that physical reality, which is crucial when designing form
timber is less strong than steel and concrete. and space and, integral to them, structure.
My duo profile as both architect and civil engineer allows Since 2004 I am professor and head of the IBOIS Labora-
me to focus on interdisciplinary aspects of architectural tory for Timber Construction at the Ecole polytechnique
design and construction and thus to develop synergies fédérale de Lausanne (EPFL), where I direct an interdis-
between them. Since I have conducted pioneering re- ciplinary group of architects, engineers, mathemati-
search in both structural engineering as well as in con- cians, and computer scientists, who perform research
struction, my perspective of various aspects differ from work in the fields of timber rib shells, folded timber plate
most theorists and practitioners who only specialize in structures, woven timber structures, integral mechani-
one of these areas. Since I practice, teach, and research, cal wood–wood connections, and robotically assembled
I am able to provide students with skills claimed by timber structures.
architects, such as subjectivity and aesthetics, whilst at
the same time I am also able to communicate in-depth I would like to thank my doctoral students Claudio
structural and technical knowledge. My research focuses Pirazzi, Hani Buri, Johannes Natter, Ivo Stotz, Markus
on technical, design, material-based, and structural is- Hudert, Sina Nabaei, Christopher Robeller, Andrea Stitic,
sues that—with some exceptions since the Renaissance— and Stéphane Roche for their valuable work.
have been neglected or delegated by architects in favor of
achieving their aesthetic aims. I take into account the Yves Weinand, October 2016
fundamental links between art and science, as well as the
specific constraints of the observed phenomena and their
concrete implementation. The effect of scale is often sim- Yves Weinand has served as professor and head of the
ply ignored in the field of structural analysis for building department at the Laboratory for Timber Constructions IBOIS
at the EPFL Lausanne. He also established the Bureau
construction. d’Etudes Weinand in Liege/Belgium. He is currently working on
the Vidy theater project, where timber is used as a double-
layered folded plate structure (pp. 29).
236
If not otherwise credited, the illustrations are taken from the IBOIS, the
laboratory for timber construction of the EPFL, Lausanne.
Alain Herzog: Fig. 11, p. 55; Fig. 4, p. 73; Fig. 2, p. 123.

Architecture collective Valentiny-Weinand: Fig. 11, p. 196.
Barnsley M. F., Fractals Everywhere, Academic Press, 1988: Fig. 4, p. 60.
Barthel, Eladio Dieste: Form und Konstruktion, 2001: Fig. 10, p. 195.
Bureau d‘Etudes Weinand : Fig. 8, p. 10; Fig. 4, 5, p. 15; Fig. 6, p. 16;
Fig. 10 (a, b, c), p. 20; Fig. 2, 3, 4, p. 93.
Christopher Robeller / IBOIS: Fig. 16, p. 133.
C. Robeller and Y. Weinand, RobArch 2016: Fig. 33, p. 29; Fig. 34, 35, p. 30;
Fig. 12, p. 197.
C. Robeller and Y. Weinand: Fig. 6b, p. 128; Fig. 14 (a, b, c, d, e), p. 132; Fig. 14,
p. 198; Fig. 15, 16, 17, 18, 20, p. 199.
M. Dedjier and Y. Weinand: Fig. 19, p. 199.
Fred Hatt / IBOIS: Fig. 16, 17, 18, 19 (a, b), p. 22; Fig. 2, p. 45; Fig. 3, 4, 5,
p. 46; Fig. 6, p. 47; Fig. 6, p. 187; Fig. 1, p. 230; Fig. 2, p. 234.
Gloss 1999: Fig. 2, p. 8.
Group of architects: Local Architecture / Danilo Mondada architecture firm
and Shel – Architecture, Engineering and Production Design: Fig 1, p. 44;
Fig. 7, 8, p. 47.
H. Buri, C. Robeller, and Y. Weinand: Fig. 16, p. 133.
H. Buri and Y. Weinand: Fig. 28, p. 26; Fig. 30 (a, b, c, d), p. 28; Fig. 31, p. 29;
Fig. 6a, 7, p. 128; Fig. 13, p. 131.
Ivo Stotz / IBOIS: Fig. 8, p. 61; Fig. 21 (a, b), p. 67.
I. Stotz and Y. Weinand: Fig. 3, 4, p. 52; Fig. 5, 6, 7, p. 53; Fig. 5, p. 128.
I. Stotz, H. Buri, and Y. Weinand: Fig. 8, p. 130.
L. Humbert and Y. Weinand: Fig. 9, p. 130; Fig. 10, 12, p. 131.
Markus Hudert / IBOIS: Fig. 2, p. 89; Fig. 11, 12, 13, 16 (a, b), p. 98;
Fig. 19–22, p. 99; Fig. 1–13, p. 110–116; Fig. 1, p. 122; Fig. 1, p. 135;
Fig. 1, p. 184; Fig. 2, p. 185.
P. Niemz, 1993, and P. Gloss, 1981: Table 1, p. 8.
PhD Thesis C. Robeller, 2015: Fig. 32, p. 29; Fig. 7, 8, 9, 10, p. 38; Fig. 3, 4, 5,
p. 191; Fig. 6 (a, b, c, d), p. 192; Fig. 7 (a, b, c), p. 194.
PhD Thesis H. Buri, 2010: Fig. 22 (a, b), p. 23; Fig. 26, 27, p. 25.
PhD Thesis I. Stotz: Fig. 2, p. 51.
PhD Thesis J. Natterer, 2010: Fig. 6, 7, p. 94; Fig. 8, 9, p. 96.
PhD Thesis M. Hudert, 2013: Fig. 20 (a, b, c), 21 (a, b, c), 22, p. 99.
Philibert de L‘Orme, Traités d‘architecture: Nouvelles inventions pour
bien bastir et à petits fraiz (1561); Premier tome de l‘architecture (1567):
Fig. 6, p. 10.
P. Mayencourt, C. Robeller, Y. Weinand 2015: Fig. 8 (a, b, c), p. 194.
S. Nabaei and Y. Weinand: Fig. 10, p. 55; Fig. 23, p. 100; Fig. 3, 4 (a, b), p. 127;
Fig. 14 f, p. 132.
S. Nabaei, F. Del Zotto, C. Robeller, and Y. Weinand: Fig. 13 (a, b, c), p. 197.
Shel Architecture, Engineering, and Production Design: Fig. 7, 8, 9, p. 19;
Fig. 10 (b, c), p. 20; Fig. 14, 15, p. 21.
Schickhofer 2004: Fig. 3, 4, p. 9.
W. Feyferlick, S. Fritzer, and Y. Weinand: Fig. 21, p. 23.
Yves Weinand / IBOIS: Fig. 8, p. 10; Fig. 2, 3 (a, b), p. 15; Fig. 36, 37 a, p. 30;
Fig. 1 (a, b), p. 92.
Cover: PhD Thesis C. Robeller, 2015; Anna-Lena Walther
237
Concept Yves Weinand
Translation and Copyediting Anna Roos
Project management Alexander Felix and Lisa Schulze
Production Heike Strempel
Layout, cover design and typesetting Anna-Lena Walther
Paper 120 g/m² MultiOffset
Printing Grafisches Centrum Cuno GmbH & Co. KG, Calbe,
printed in UltraHDprint®
Library of Congress Cataloging-in-Publication data

A CIP catalog record for this book has been applied for at the
Library of Congress.
Bibliographic information published by the German National Library

The German National Library lists this publication in the
Deutsche Nationalbibliografie; detailed bibliographic data are
available on the Internet at http://dnb.dnb.de.
This work is subject to copyright. All rights are reserved,

whether the whole or part of the material is concerned, specifically
the rights of translation, reprinting, re-use of illustrations,
recitation, broadcasting, reproduction on microfilms or in other ways, and
storage in databases.
For any kind of use, permission of the copyright owner must be obtained.
This publication is also available as an e-book

(ISBN PDF 978-3-0356-0490-0;
ISBN EPUB 978-3-0356-0513-6)
and in a German language edition (ISBN 978-3-0356-0560-0).
© 2017 Birkhäuser Verlag GmbH, Basel

P.O. Box 44, 4009 Basel, Switzerland
Part of Walter de Gruyter GmbH, Berlin/Boston
Printed on acid-free paper produced from chlorine-free pulp. TCF ∞
Printed in Germany
ISBN 978-3-0356-0561-7
987654321 www.birkhauser.com

2016 Advanced Timber Structures

Uploaded by

2016 Advanced Timber Structures

Uploaded by

Advanced Timber Structures

Yves Weinand (Ed.)

1 Folded plate structures 13

2 Advanced architectural geometry 49

4 Form-finding and mechanical investigations

5 Customized construction 189

The Research Laboratory IBOIS at the EPFL Lausanne 236

Picture credits 237

How can a schedule and a technologically

Within the framework of the Department of Timber

a broken weakest link. Engineered wood—a dual system

The anatomy of wood – It makes sense to produce timber materials

Fig. 3 Tangential section

longitudinal (EL) tangential (E T )

Shear modulus G – shear deflection: Layer

Elasticity modulus E (in tangential, Shear modulus G (in tangential,

Fig. 9 Intercations between

1. 1 Folded plates made from timber panels 14

1. 2 Structural optimization of timber folded-plate

1. 3 “It’s important to progress

1. 4 The Chapel of the Deaconesses of St-Loup

Folded plates made from timber panels

In this essay, we will look at possible ways

14 Folded plate structures

Fig. 1 Folding principle: rigidity is achieved by folding.

Fig. 6 Solution of the support system: the panel is transfor med

geometric position during the folding lumber panels—then this results in a

16 Folded plate structures

18 Folded plate structures

Fig. 9 Finite-element model of a

deformation d x (mm) deformation d y (mm)

Membrane effort N x (kN) Membrane effort Nz (kN)

150 nail perforations per rm,

Fig. 11 Fig. 12 Fig. 13

20 Folded plate structures

Fig. 16 Installation of the panels.

22 Folded plate structures

In summary, the geometric adjustments

edge regions by simply adapting

thereby avoiding the need for

24 Folded plate structures

Internal forces Deformations

26 Folded plate structures

28 Folded plate structures

The strengthening of the panels along the

30 Folded plate structures

Structural optimization of timber

32 Folded plate structures

fold slab plate

Fig. 1 Load-bearing behavior of a single fold, combination of

a) Simple b) Spot or facet c) Rhombus-based d) Anti-prism-based folded form

Fig. 2 Classification of folded-surface structures;

34 Folded plate structures

3 Structural considerations 3.2 Geometry of the structures

36 Folded plate structures

38 Folded plate structures

“It’s important to progress from research

40 Folded plate structures

42 Folded plate structures

The Chapel of the Deaconesses of St-Loup

Design Local Architecture/Danilo

44 Folded plate structures

Fig. 3 Sun and shade expression of lateral facade

46 Folded plate structures

els and 60-mm-thick horizontal panels in spruce, are all

2. 1 Regular and fractal-shaped,

2. 2 Iterative geometric design for architecture 58

2. 3 Geometrical description and structural analysis

2. 4 “Discover processes where the parameters

Regular and fractal-shaped, discretized

In the previous chapter, it was shown ple, by wind or earthquakes—produce

Comparison of standard and folded areas.

d Zmax = 220 mm d Zmax = 45 mm

50 Advanced architectural geometry

Fig. 3 The tool describes the overall

52 Advanced architectural geometry