© <2021>. This manuscript version is made available under the CC-BY-NC-ND 4.

0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
The definitive publisher version is available online at https://doi.org/
10.1016/j.istruc.2021.08.004
Graphical Abstract
Integrated computational framework for the design and fabrication of bending-active structures
made from flat sheet material
Francesco Laccone,Luigi Malomo,Nico Pietroni,Paolo Cignoni,Tim Schork
Highlights
Integrated computational framework for the design and fabrication of bending-active structures
made from flat sheet material
Francesco Laccone,Luigi Malomo,Nico Pietroni,Paolo Cignoni,Tim Schork

• Novel architectural method for the automatic design and optimization of bending-active structures from an arbitrarily-
curved input shape.
• The FlexMaps algorithm is modified and extended at the architectural scale to provide optimized spiral mesostructures.
• Automated shape-to-factory workflow for advanced manufacturing.
• Application in a 1:1 scale proof of concept demonstrator.
Integrated computational framework for the design and fabrication of
bending-active structures made from flat sheet material
Francesco Lacconea,b,∗ , Luigi Malomoa , Nico Pietronic , Paolo Cignonia and Tim Schorkd
a ISTI,CNR, via G. Moruzzi 1, 56124 Pisa, Italy
b DESTEC, University of Pisa, Largo Lazzarino 2, 56122 Pisa, Italy
c School of Computer Science, Faculty of Engineering and Information Technology (FEIT), University of Technology Sydney (UTS), 15 Broadway Ultimo,

Sydney Australia
d Transformative Technologies Lab, School of Architecture, Faculty of Design, Architecture and Building (DAB), University of Technology Sydney (UTS),

15 Broadway Ultimo, Sydney Australia

ARTICLE INFO ABSTRACT

Keywords: This paper introduces an integrated computational design framework for the design and realization
computational design of arbitrarily-curved bending-active architectural structures.The developed framework consists of a
simulation series of methods that enable the production of a complex 3D structures composed of a set of flat
advanced manufacturing 2D panels whose mechanical properties are locally tuned by varying the shape of embedded spiral-
structural design ing patterns. The resulting panels perform as variable stiffness elements, and they are optimized to
digital fabrication match a desired target shape once assembled together. The presented framework includes all the steps
shell for the physical delivery of architectural objects, including conception, static assessment, and digital
optimization fabrication. The developed framework has been applied to an architectural scale prototype, which
form-finding demonstrates the potential of integrating architectural design, computational simulation, structural
mesostructured engineering, and digital fabrication, opening up several possible novel applications in the building
architecture sector.

1. Introduction lems. A new generation of computational design workflows

takes advantage of geometry processing methods to support
It is estimated that, due to worldwide population growth the exploration of the space of solutions that conform to me-
and urbanization, approximately 2.6 billion people will re-
chanics, fabrication, or budget limitations.
quire new housing by 2050. This demand will be impos-
The quests for compelling free-form shapes in architecture
sible to meet due to the required natural resources and the drove architectural geometry to focus on practical issues, such
current low and stagnating productivity of the construction as material, production, fabrication and assembly [18]. Hence,
industry. Furthermore, the built environment is responsi- one of the relevant problems is that usually the individual
ble for nearly 40% of global greenhouse gas emission [2, 3]. components are considerably smaller than the final structure.
Our future challenge is finding solutions that allow to build
Therefore, the subdivision and discretization of the contin-
more with less material, using renewable resources that cre-
uous target shape represents a key step towards the build-
ate less greenhouse gases while at the same time creating ing feasibility and deserves a large research effort. A com-
high-quality architectural solutions. These solutions include mon strategy is to investigate the piecewise approximation
efficient structural solutions that take advantage of the ma- of complex surfaces or their developability [19, 20, 21].
terial properties. The fields of Architectural Geometry and An alternative approach is the use of mesostructures to ful-
advanced geometry processing offer new approaches to solv-
fill specific shape requirements through mechanical defor-
ing some of these challenges. Architectural Geometry is
mation. Such structures work at an intermediate scale (hence
a new interdisciplinary topic that has gained a significant the term meso-), which is not the material scale (micro-),
boost in recent years [4, 5]. This research field manages and can be appropriately designed to reproduce the mechan-
problems related to the design, optimization, and fabrica-
ical response of a different material by simply using vary-
tion of complex architectural shapes. The complexity orig-
ing geometry. For instance, material can be selectively re-
inates from the different constraints that shape aesthetics,
moved from a solid plate or a volume to create elements with
structural design, and engineering aspects of a desired struc-
a custom deformation capacity. Such a product is also re-
ture pose. Thus, the paneling of free-form shapes [6, 7], ferred as meta-material, since a target behavior is obtainable
the assembly of self-supporting and interlocking structures in spite of the original base-material property. This induced
[8, 9, 10] or the structural design of shells and grid shells non-homogeneity provides the opportunity to achieve a large
[11, 12, 13, 14, 15, 16, 17] have become non-trivial prob-
number of variations by altering the initial geometry, which
∗ Corresponding author in turn means that specialized meta-materials can be adopted
[email protected]; locally to comply with specific required performances. This
[email protected] (F. Laccone) is one of the base strategies of the functionally graded mate-
ORCID (s): 0000-0002-3787-7215 (F. Laccone); 0000-0001-7892-894X
(L. Malomo); 0000-0002-8271-2102 (N. Pietroni); 0000-0002-2686-8567 (P.
rial concept in design [22].
Cignoni); 0000-0002-4583-2439 (T. Schork) While the meta-material approach has been successfully used

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier Page 1 of 18

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

for the production of small scale objects [23, 24], its full by Buckminster Fuller in 1957, which are geodesic spheres
potential in the architectural context is still unexplored. A made from single-curved plywood plates.
key advantage of the use of mesostructures in architecture Applications of these structures have been limited for a long
is that they enable common building materials to achieve a time due to the lack of computational tools and methods ca-
greater deformation capacity, and so they expand the space pable of simulating their complex structural behavior. The
of fabricable designs. On the other hand, mesostructures increase of computing power and further sophistication of
require particular attention on the related structural issues, digital fabrication technology have led to a recent renais-
since they have to achieve a certain deformed shape and at sance and renewed interest in bending-active structures [26,
the same time they should bear external loads. In the archi- 27]. Bending-active structures are a cost-effective construc-
tectural context, similar issues are usually addressed in the tion technique to produce geometrically complex and highly
bending-active structures. efficient lightweight structures [28, 29, 30, 31]. The back-
We adopted the algorithm [1], which is tailored to the bone of the digital design workflow for bending-active struc-
digital fabrication of small-scale objects, and we extended it ture is a new generation of geometry- and material- driven
to work at architectural scale. Shifting the focus to a larger methods [32, 33, 34, 35, 36]. Until recently the majority of
scale poses several exciting new challenges in terms of ma- physically-based methods offer limited or no shape control
terial requirements, structural behavior, fabrication design, to the designer. Newly developed methods make it possible
and assembly. We can recap the main contribution of this to set a target shape. These methods search for the best set-
paper as follows. ting or configuration of the load-bearing elements such that,
once manufactured and deployed, they form a structure that
• we proposed a novel architectural pipeline for the au- closely approximate the target shape.
tomatic design, optimization and fabrication of bending- The work by La Magna [37] exploits the bending stiffness
active structures from smooth, even doubly-curved, of plates and their deformation modes. His first contribu-
input shapes; tion (form-conversion approach) provides a method for cov-
• we extended the FlexMaps method [1] to work at the ering the target shape with bent plates by manipulating the
architectural scale; topology, allowing simple developable elements to map mul-
tiple directions. This is a general method that guarantees
• our pipeline includes an important step based on non- the bendability of the elements and their ability to cover the
linear Finite Element (FE) analysis to assess the pro- whole surface space. The second contribution (embedded
duced internal stress due to form-giving bending and approach) exploits the deformation properties of cellular struc-
due to the design load; tures to design meta-materials that can replicate the base sur-
face’s curvature.
• the proposed pipeline consists in a shape-to-factory The recent X-Shells proposed by Panetta et al. [38] enables
workflow, which eventually generates the path for sub- the computational design of deployable structures made of
tractive manufacturing machines; continuous, elastically bent beams that are connected by ro-
• we tested and applied the developed pipeline in a 1:1 tational joints and assembled in a flat grid configuration.
scale proof of concept demonstrator. A specific actuation system forces the beams to twist and
buckle out-of-plane to reach a desired target position at the
1.1. Bending-active structures static equilibrium. The assembled flat state and the deployed
Bending-active structures are a promising solution to ef- configuration are jointly optimized, while a sparse distribu-
ficiently achieve doubly-curved architectural surfaces out of tion of actuation forces is subsequently optimized to ease the
flat sheets or straight elements. In bending-active structures, erection phase.
the elements are directed to a global self-equilibrating sta- FlexMaps [1] is a fully automated method to optimize the
tus by the elastic deformations caused by joint bending and stiffness of a particular class of mesostructures. The method
stretching. The curvature and the internal stress produced in discretizes the surface into flat, spiral-shaped, deformable
the equilibrium configuration are key drivers of the entire sub-modules. Once these are actively bent, the final struc-
design process. In general, given a base material, the form- ture in equilibrium state results as close as possible to the
defining process evaluates each elements’ shape to match a target shape.
certain equilibrium configuration, as well as a user-provided
set of boundary constraints. Additionally, the employed ma- 1.2. Architectural challenges and objectives
terial needs high-breaking strain capacity [25]. This sce- Originally the FlexMaps method [1] was developed for
nario makes bending-active structures not trivial to design. the design and fabrication of small-scale objects. A further
Modern bending-active structures were pioneered mainly by study attempted to use use this approach in a larger scale pro-
adopting geometry-based approaches, in which the material totype project [39, 40].
status is evaluated analytically based on the estimated ele- The research presented in this paper builds on this work and
ment curvature. Examples of this structures are the Mannheim further extends it by exploring and testing the developed ap-
Multihalle by Frei Otto built in 1974, the largest post-formed proach to a more complex topology. The main aim of the
wooden grid shell; and the Plydome Structures introduced project is to consolidate the use of FlexMaps in advanced

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✷ ♦❢ ✶✽

 shape approximation.
Using a four-arms spiral pattern as mesostructure has three
main advantages:
• As opposed to simple straight elements, spirals offer
less resistance to bending, making them well-suited to
approximate high curvatures, as they distribute inter-
nal stress in a controlled way.
• The mechanical behavior of a spiral can be locally
tuned by modifying just two geometrical parameters,
the spiral angle and wall thickness (see Section 3.3).
Consequently, keeping the constituting material fixed,
a mesostructured panel’s stiffness can be tuned on a
local basis by acting on the geometry of each spiral.
• The spiral connectivity results very simple and makes
it possible to connect one element with their neighbors
even in case of mesh singularities (Fig. 2).
Eventually we obtain a set of panels containing a lim-
ited number of spirals, namely patches, and each one is op-
timized to have the entirely-assembled structure match the
target shape. Essentially, the algorithm produces a set of
anisotropic patches that, assembled together, are capable of
reaching the desired shape by uniformly redistributing their
internal stress. This stress is the one induced during the as-
sembly phase, by individually bending and mutually con-
architectural design and further integrate and streamline the necting these patches.
workflow that links shape design to digital fabrication. The
integrated computational design framework extends the pre-
vious method and includes shape design, optimization, struc-
tural verification, 2D layout design, and digital manufactur-
ing. The developed framework was tested in a demonstrator
project that was built at the School of Architecture, Faculty
of Design, Architecture, and Building (DAB) of the Univer-
sity of Technology Sydney (Fig. 1). The whole structure is
made out of CNC-milled 15 mm thick plywood components.

2. FlexMaps algorithm
FlexMaps optimizes the shape of individual patterns to 3. Design method
compose 2D mesostructured panels, which deform in a pre-
scribed manner. Each pattern is shaped as a four-arms spiral Besides the design capabilities of the FlexMaps method,
which can be embedded in a quadrilateral mesh face (i.e. the architectural scenario analyzed in this paper poses addi-
four spiraling arms connect the face barycenter to the mid- tional challenges ranging from the shape conception to the
points of the quad edges). Such patterns can stretch and bend final physical validation at the architectural scale. The main
in any direction and also assume double curvature. The ap- steps of the new workflow for the architectural scenario is
proach starts by quadrangulating the input shape. To reduce presented in Fig. 3 and it consists of five steps:
the number of required pieces, a semi-automatic merging shape design In this initial step the architect designs a free-
process connects adjacent quads, which are then clustered form continuous target shape. While modeling, the
and mapped from the input surface onto a planar domain. designer should consider several constraints involved,
A spiral-shaped mesostructure is embedded into each flat such as the maximum size, the overall weight and the
quad. A reduced physical data-driven model is used to es- anchor points of the resulting structure. Moreover, be-
timate the mechanical behavior of each individual spiraling ing a lattice structure the shape should be at least sta-
pattern, with respect to the parameters determining its shape. ble.
This model enables the simulation of the resulting shape in
3D at interactive rates, as well as the automatic optimization quadrilateral remeshing As in FlexMaps, the present pipeline
of the patterns’ parameters for improving the quality of the requires an initial remeshing step that is performed

Preprint submitted to Elsevier

 using a technique similar to [41]. The remeshing is curvature, supports and loading conditions of the target de-
guided by an underlying curvature field and optimizes sign. However, in practice, the achievable accuracy for the
for having quads as regular as possibile. This is a cru- resulting shape approximation is affected by several factors:
cial aspect for the correct characterization of the em-
bedded patterns, as the reduced simulation model is • The resolution of the mesostructure influences the ca-
computed from spirals embedded into regular quads pacity of the structure to approximate high frequency
and results inaccurate for deformed shapes. geometries, such as sharp edges.

spiral form-finding This step optimizes the shape of each • The designer should consider that the material can even-
individual pattern in order to match the expected target tually break if it undergoes an excessive load or an ex-
geometry. As in FlexMaps this step requires two main cessive target deformation.
components: a way to characterize how the shape of • The pipeline will not produce a close approximation of
a spiral affects its physical behavior, and a method to the input design if its shape and boundary conditions
derive the desired elastic properties such that the final (e.g., loading scenario plus anchor points) are far from
assembled structure matches the target shape. Those being physically achievable.
two problems are strictly interconnected. As in FlexMaps
the behavior of a spiral pattern is parametrized with re- Targeting the architectural scale includes also two addi-
spect to its geometric parameters, it is simulated and tional specific challenges:
it is represented using a reduced model for fast for-
ward simulation, while backward derivation is used • From a geometric point of view, architectural-scale
for the global shape optimization. This component is objects, being surfaces with boundaries, are more dif-
similar to the original FlexMaps pipeline [1] but it has ficult to produce because preserving the ‘open-boundaries’
been redesigned to account for a different scale, novel requires additional devices (i.e., edge beams) or fixing
building materials, additional design load (e.g., grav- anchor points. In contrast, for small-scale objects such
ity) that can be considered at this stage, and a more problems are usually neglected.
user-friendly way to specify the scenario’s boundary • Similarly to the point above, the weight of small-scale
conditions. structures is negligible, and stability is not the most
structural analysis A structural analysis is used to validate relevant issue. In contrast, these are fundamental re-
the static performance of the structure, namely once quirements for architectural-scale objects. Indeed, it
submitted to the form-giving bending and self weight. is not possible to reuse a simulation setup targeted to
This FE analysis uses a non-linear orthotropic mate- small-scale structures [34], nor is it possible to re-
rial model to simulate the behaviour of the plywood scale or adjust the cross-sections of the patterns at a
and a well-tested, reliable pipeline to perform it for later stage for a bigger scale, as this has a direct im-
multiple components and large displacements. pact on the bendability and material strength.

fabrication and assembly The final step includes the gen- Consequently, the optimization has to be performed on the
eration of 2D milling instructions for manufacturing object’s real scale and the actual permanent load, which is
and the actual assembling of the entire structure. neglected in [1], has to be considered. However, since the
optimization process does not explicitly manage the input
shape’s stability, an additional design constraint is to con-
3.1. Shape design sider the initial shape definition to be at least stable in the
The architect is free to design complex target shapes to equivalent continuum.
be approximated and fabricated via the proposed pipeline.
In theory there are no restrictions on the topology, Gaussian

Preprint submitted to Elsevier

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

3.2. Quadrilateral remeshing and patch

segmentation
The problem of transforming 2D patches into assembled

w
3D structures is related to developability, namely the prop-
erty of 3D mesh to be unrolled and flattened without distor-
tion on a 2D plane, either entirely or subdivided into patches.
This 3D-to-2D mapping is referred in computer graphics as θ
parametrization [42], and usually the accompanying distor-
tion is a quantity to minimize. Our parametrization proce-
dure starts by tessellating the desired surface geometry (e.g.,
Quad Size (s)
NURBS) into a triangulated mesh. Then, it generates quad
Most Flexible
mesh strips by tracing curvature-aligned geodesic paths di-
rectly onto the triangle mesh, using the method in [41], pro-
ducing a quadrangulation as a byproduct. As explained in
Section 3, FlexMaps assumes that mesostructures are em-

Spiral Angle (θ)

bedded in regularly-shaped quads. Consequently, the more
Selected
regular are the quads, the more accurate is the reduced model range
simulation. For this reason, the method described in [12] is
applied as a post-processing regularization of the quadran-
gulation.
After a shape is converted into a regularized quad mesh, a
semi-automatic process groups adjacent quads, producing a
set of quad patches which are then mapped with low distor-
tion onto a flat domain. In this way it is possible to decom- Most Stiff
pose even non-developable complex shapes into a set of flat Wall Thickness (w)
patches which will become the building blocks of the target
structure. If desired, the user can manually tune this patch ❋✐❣✉r❡ ✹✿ ❙♣✐r❛❧ ♣❛r❛♠❡t❡rs ✭t♦♣✮ ❛♥❞ ♠❛tr✐① ♦❢ s♣✐r❛❧ ♠♦r✲
♣❤♦❧♦❣✐❡s ✭❜♦tt♦♠✮✳
decomposition with a simple visual interface.

3.3. Spiral patterns optimization

The scope of this step is to tune the parameters of the grees of freedom. At the coarse level, such simulation data
mesostructures such that the globally assembled shape ap- is used to produce a simple beam model that captures ax-
proximates as closely as possible the designed target shape. ial, bending and twisting deformation modes and can be ef-
The FlexMaps algorithm (refer to [1] for the details) re- ficiently integrated into a sensitivity analysis for global shape
lies on three local geometric parameters for each spiral pat- optimization. The data resulting from this phase consists of
tern (Fig. 4 top): the pattern size (s), which is implicitly set the full set of patches containing optimized spiral mesostruc-
in the quadrangulation phase; the spiral twist angle (𝜃); and tures, laid out as flat surfaces.
the spiral wall thickness (𝑤). By varying the spiral parame-
ters within a fixed quad size, a matrix of spiral morphologies 3.4. Structural analysis
with differentiated material behavior can be created (Fig. 4 The optimization routine gives no control over punctual
bottom). For example, a combination of low wall thickness stress and provides only limited control over the resulting de-
and a high spiral twist angle results in a highly flexible spiral, formation. With an acceptable accuracy loss, the procedure
required, for example, in areas with high curvature. In con- employs an efficiently reduced model that allows the overall
trast, large spiral wall thickness and low spiral angle result in computation to be tractable. However, at the architectural
a stiff spiral, which makes sense for flat regions. Simulations scale these considerations raise the need for a detailed struc-
are performed for a large set of patterns and different load tural analysis. This active shell is modeled employing solid
scenarios and the resulting precomputed data is used to pop- finite elements to obtain its exact deformation, to verify that
ulate a parametric database with a sampling of the achievable the stress is within the capacity of the employed material,
spiral shapes and their relative mechanical behavior. and to estimate the overall safety level. This verification
The optimization algorithm determines the heterogeneous workflow has been built for the ANSYS software environ-
distribution of spirals by interpolating patterns from this pre- ment [43] and consists of two simulation phases, which are
computed database. A data-driven two-scale model is used conceptually depicted in Fig. 5.
to predict and then optimize, in a continuous space, the ap- In the first phase (Fig. 5a), every flat patch is rigidly moved
proximate physical behavior of our structures. At the fine from the horizontal plane, in which it is generated, to a po-
level, the algorithm parametrically generates an accurate ge- sition that is as close as possible to its corresponding de-
ometric model of a single spiral structure and simulate its formed target shape. This patch positioning improves the
behavior using a non-linear FE model with thousands of de- convergence of the analysis since a limited number of time

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✺ ♦❢ ✶✽

 target bonding
shape

boundary
conditions

steps are needed to complete the nonlinear simulation. This damental. Defects should be avoided as they can affect the
phase approximates the form-giving bending behavior that shape and or propagate in failures. On the other hand, sub-
each individual spiral patch will be subjected to during as- tractive manufacturing is better suited as it can rely on a large
sembly. A displacement is imposed at the extremities of the number of materials to be actively bent and produced in form
spirals’ free arms to move them into their target deformed of flat sheets, i.e., aluminum, plywood, glass; but a drawback
shape. Consequently, all spirals will independently assume is that the gaps between the spirals became material waste.
their pre-stressed actively-bent configuration (as in Fig. 5b). Good quality can be obtained with a reduced manufacturing
The purpose of this initial procedure is not to simulate the ex- time using CNC machines, such as laser and water-jet cutters
act assembly sequence but to provide a solution that approx- or milling machines, and integrating the digital fabrication
imates the strain of each spiral during the formation process into the current pipeline.
(all patches are simultaneously bent independently). Mod- The manufactured structure reaches its final shape under ap-
eling the assembly sequence is a much harder problem. It propriate boundary conditions once all patches, carved from
requires considering the actual assembly sequence, the col- flat sheets, are individually bent and tied together. In par-
lisions, and all the deformations imposed to bring the panels ticular, the connection between the patches should be able
to their final position. to guarantee full-strength and preserve this capacity during
After this first step, the resulting states of the panels (de- the time, i.e. not going into creep. An illustration of the
formed shapes and accompanying strains) are set as starting construction system is shown in Fig. 6.
point of the second phase, in which it is simulated the behav-
ior of the structure as a whole. The individually-bent patches
are linked, and boundary conditions are applied (Fig. 5b).
The patches are then released so they can uniformly redis-
tribute their accumulated stress, the gravity, and other loads
applied under, such as its self-weight or in general all loads
that are assumed as permanent (Fig. 5c).
For performing such simulations it is recommended to de-
duce the material properties to plug into the model from
experimental tests on real-scale patches prototypes (Fig. 5
shows an example of such prototypes for calibrating the ma-
terial properties). This is particularly critical as subtractive
manufacturing processes may lead to a reduced and more
scattered stiffness in the case of composite or natural ma-
terials. Moreover, the stiffness and strength of the connec-
tors must be evaluated experimentally via bending and ten-
sion/compression tests with the objective of conceive a detail
that restores the cross-section characteristics.

3.5. Fabrication and construction system

FlexMaps enables both additive and subtractive manu-
facturing. Additive manufacturing is undoubtedly a material
saving strategy, but it is not recommended at the architectural
scale since the local material strength and continuity are fun-

Preprint submitted to Elsevier

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

4. Architecture of the computational model processes, in which the provided mesh can be easily 3D printed.
On the other hand, subtractive fabrication processes need a
To connect the operations of the presented methodology,
dedicated subroutine since the essential part is not the solid
a centralized computational model has been developed as
piece but its outline (indeed, the outlines identify tool paths).
shown in Fig. 7. All aspects of the design and development
In this subroutine, at first the meshes’ outlines are converted
process are integrated into it. This model is made of a core
into NURBS for a better interoperability with most CAM
part and several subroutines. Moreover, to initiate the design
softwares, then the connectors are generated at the endpoints
process, some preliminary actions are required, i.e. the sta-
of patch-boundary spirals (the core process ignores the pres-
bility check of the equivalent-continuum target shape and its
ence of connectors). Additionally, cutting tolerances can be
quad remeshing. General considerations are also important
embedded, as well as the specification of using different ma-
in this phase to inform the computational model.
chine tools for different cuts (i.e. different diameters tools
can be selected based on the desired precision and stored as
Shape design Practical
different layers).
The computational model has allowed an easy exploration
PRELIMINARY

constraints
Stability
check
site specification, tools and
precision, material, detailed of the demonstrator feasibility and has been of fundamental
design and connections etc.
help in supporting decisions and coordinating all the design
Experimental
Tests
Quad Remeshing selection of and fabrication choices required for such a complex project.
target quad size s
at material level, at meso-
structure level (i.e. arch)
(as regular as possible) The model shows also wide flexibility for what the geomet-
ric design space is concerned. Conversely, the material input
material restrictions is restricted to a specific database, which needs to be recom-
parameters OBJ on θ, w
puted each time a new material is considered.
CORE

Segmentation/ Morphologies
Simplified model Patching Computation
5. The Demonstrator
The method feasibility was tested and evaluated at the
material flat patches OBJ patches SVG architectural scale by building a full scale prototype (Fig. 1).
SUBROUTINES

parameters in 3D and labels

FE Analysis Visualization Fabrication As project constraints, this demonstrator could not rely on
deformed
shape hanging supports from the ceiling or any rigid fixing. The
STL obj with patches and labels 2D tool paths (nesting for
system has been conceived to be in equilibrium by just stand-
best fitting the raw sheet)
APDL solid detailed model
ing on the floor. Gravity load is considered as acting in both
the form-giving process and as the only load during the lifes-
ANSYS Rendering, Assembly instructions CAM
pan of the demonstrator. It consists of:
❋✐❣✉r❡ ✼✿ ❙❝❤❡♠❛t✐❝ r❡♣r❡s❡♥t❛t✐♦♥ ♦❢ t❤❡ ✢♦✇ ❝❤❛rt ✐♥ t❤❡
• a set of mesostructured flat panels that are optimized
❝♦♠♣✉t❛t✐♦♥❛❧ ♠♦❞❡❧✳
through the algorithm;

The core part is a script and include the steps developed in • a base made of three supporting edge beams and an
the work FlexMaps [1]. It consists of the implementation of interlocking floor system to tie these beams together;
the reduced model, the shape optimization routines, and the • a set of slender edge beams to preserve the relative
system for producing the flat quad patches. position of connectors located on the surface’s open
The FE Analysis subroutine is aimed at generating input for boundary.
the ANSYS software, namely a geometry file and a list of
instructions in the form of APDL commands. For instance, For these purposes we selected plywood as building mate-
this latter contains deformation vectors to bring the points rial, as it represented a balancing point between ease of man-
of the connectors from the flat shape to their candidate de- ufacturing, cost and mechanical requirements. Therefore, all
formed shape (first phase of the analysis, Section 3.4). For components of the demonstrator have been CNC-milled out
generating the geometry of the panels, all parametrized 2D of 15 𝑚𝑚 thick plywood sheets, and this process has been
mesh patches are moved to a position close to their ideal de- carried out entirely within UTS faculty facilities. The manu-
formed shape. Then, a Grasshopper [44] script has been de- factured panels resulted lightweight and easy to handle even
veloped to convert discrete meshes into NURBS and to gen- for unskilled workers and did not require any special han-
erate solid geometries, which can be subsequently processed dling equipment. Among common building materials, ply-
in the FE analysis. wood is listed as suitable to be used in active-bent contexts
The output of the visualization subroutine is a geometrical due to a good proportion of flexural strength and stiffness
model to be used for navigation and for rendering. Also a [25, 45].
mesh model with distinct annotated patches is automatically
5.1. Experimental tests
built and it is aimed to guide the assembly of the structure.
The fist step required to initiate the design process is to
The raw result of the core process is an array of solid meshes.
perform experimental tests to characterize the building ma-
This data is inherently compatible with additive fabrication
terial and calibrate the base material properties. A good test

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✼ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

case is a three-panels arch, made of 3 × 9 all-equal spirals design is a non-developable vaulted surface with three outer
and similar to the example of Fig. 5. This simple struc- arch profiles and a central hole. The arches have different
ture has been manufactured and tested in two setups with heights and are linked to a central hole using saddle-shaped
same boundary and loading conditions, but with two differ- surfaces that merge reciprocally. The Gaussian curvature of
ent spans of 1.5 𝑚 and 2.0 𝑚 (Fig. 8 (a) and (b) respectively). the surface flips as it approaches the ground on three different
The supports are cylindrical hinges, so they allow rotations supporting lines (Fig. 10a). The whole shape fits a volume
and avoid vertical and horizontal displacement. For each of 3.41 × 3.02 × 2.71 meters, which was compatible with the
setup, the three mid-span nodes are loaded in three steps of available exposition space.
0.5, 1.0 and 1.5 𝑘𝑔 per node. For each load increment the The initial quadrangulation and patch decomposition has
deformations are measured through a Artec handheld scan- been performed according to the method described in Sec-
ner. The load is materialized as steel blocks hanged from tion 3.2. At first, a smooth cross-field has been computed
the center of each of the three mid-span spirals (as visible following the main curvature directions and aligned with the
in Fig. 8). The objective is to have two structures with dif- borders (Fig. 10b). Then, a mainly regular quadrangulation
ferent stiffness to be tested at different loading rates, having has been derived (Fig. 10c), except for few singular vertices,
fabricated only three patches. The stiffness of the spirals de- where a valence of 3 or 5 is necessary to comply with pos-
pends on various factors related to their geometry and to the itive or negative Gaussian curvature. The patch layout has
material locally. For this specimen the spirals are designed been generated using the semi-automatic process that itera-
to be representative of the mean properties of the demon- tively merges quads together into flattened patches by keep-
strator, so they are inscribed in 0.25 × 0.25 𝑚 quads, and ing the introduced distortion below a given threshold. In the
have fixed parameters 𝜃 = 125◦ and 𝑤 = 15 𝑚𝑚. The speci- demonstrator, for aesthetic and practical reasons, some patch
men in the assembled and developed configuration measures boundaries have been manually enforced using the dedicated
0.75 × 2.25 𝑚. The material properties have been obtained graphical tool (Fig. 11a).
by reverse fitting the measured deformation on a FE model Due to fabrication constraints, we have selected a target quad
as per Fig. 5 (developed in two phases, following the same edge of 𝑠 = 0.26 𝑚. This led to a spiral size that is manage-
strategy presented for the full model in Section 3.4), given able during fabrication, and also easy to handle and bend
as target points the mid-span nodes deformations in the three during the assembly. Moreover, for aesthetic reasons and
load steps. The material properties are obtained by multiply- to avoid local weakening, the optimization space of spiral
ing reference values by a factor until simulated and measured parameters has been restricted to a constant width of 𝑤 =
displacements result as close as possible. 15 𝑚𝑚 and a twist range of 𝜃 = 40 − 180◦ (red frame in
A first FE model uses an isotropic material, which is a sim- Fig. 4). Eventually, each spiral arm can have a different
plification needed to calibrate the reduced model adopted in length but a constant cross-section of 15 × 15 𝑚𝑚 to guar-
the core process. This is a requirement for guaranteeing a antee a minimum stiffness everywhere. The upper limit on
consistent computation of the optimized spirals morpholo- the twist angle serves instead to avoid excessively-dense spi-
gies. Besides, a second FE model uses a more realistic or- rals, whose arms may result too close to the milling toler-
thotropic material, whose parameters are intended to be used ance. The angle lower bound, instead, avoids aesthetically
in the full simulation of the demonstrator. Hence, a higher unpleasant spirals, resulting similar to a cross.
accuracy is needed to assess the internal stress and predict
the actual deformed shape. 5.3. Detailed design
In any case, these experiments can be regarded as a test bench The connector between the patches is a fundamental de-
of the whole pipeline and are fundamental to familiarize with tail of the system. This connection is required to be full-
the constructional system, to verify the detailed design and strength and therefore substain axial, shear, bending and tor-
in general to gather essential knowledge to face all practi- sion. At the same time, the connection mechanism has to be
cal issues that may arise while working on larger structures. easy to assemble and aesthetically pleasant. The adopted key
Since the selected target size of quads and all the details, connection (Fig. 12a bottom) is made of two antisymmet-
such as the connectors, are the same of the demonstrator, this ric interlocking parts. The contact surface friction and the
tests set the benchmark for the tools, the speed and precision shape interlocking provide a sufficient resistance in prepara-
of machine milling as well. tion for the ultimate lock of the pieces by means of a couple
of through-hole M5 bolts (Fig. 12a top). The bolts work as
5.2. Shape design and optimization a key that avoids any relative movement. The drilling pre-
For the design of the structure a set of NURBS-based cision and fabrication tolerance are extensively calibrated as
shapes have been considered as candidates, in particular, to they have a major impact on the connection stiffness and,
challenge the physical production of a complex topology within therefore, on the success of the assembly.
a limited indoor space (Fig. 9). The shapes are either manu- The open boundaries of the structure have been stiffened
ally modeled or obtained through form-finding techniques. with edge beams. Each edge reinforcement is made of a
Eventually we selected the red-framed geometry in Fig. 9 couple of plywood lamellae, which are point-fixed on the
because it has a significant variation of curvature, despite its end-point of the spiral free arms located at the boundaries
small size, and has no symmetries. The selected anticlastic (Fig. 12b). Here, the spiral arms are equipped with two

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✽ ♦❢ ✶✽

 1.5
1.5 𝑚 2.0 𝑚

of the whole boundary path, to preserve a smooth profile

and avoid kinks. The primary function of these edge beams
is to preserve a correct inter-distance between the boundary
connectors. Additionally, they help the patches to preserve
their target deformed shape during the assembly. The pro-
files are geometrically modeled based on the target deformed
shape, i.e. the result of the phase one of the structural anal-
ysis. For manufacturing, a 6 𝑚𝑚 plywood panel has been
used, so each edge lamella cross-section is 6 × 30 𝑚𝑚. It
is worth noting that edge beams are also actively bent but,
since their bending-stiffening contribution to spiral patches
has been observed to be low, their contribution has not been
considered in the FE structure simulation.
Concerning the floor support structure, the base spirals ter-
minate with T-shaped extremities that are embedded into
ground beams made of three 15𝑚𝑚 plywood plies (Fig. 12c).
The first two plies provide the T-extremities with the right
shape tangency, while the top ply avoids their uplift. Each
central ply of the ground beam hides two ballast iron discs
that have been designed to provide a sufficient bending support-
reaction. Once all these components are dry assembled, the
base plies are screwed together.
holes for the bolts that hold the two lamellae. The seams All bottom layers of the ground beams have an interlocking
of the two lamellae are interleaved between the two sides design that allows them to connect to the floor. The latter

Preprint submitted to Elsevier

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

is a wooden tiling milled from the same 15 𝑚𝑚 panels em- material properties adopted in the analyses are derived from
ployed for the spirals and the ground beams. The design of the experimental tests performed on the arch specimen (Sec-
the floor is based on the aperiodic Penrose pattern shown in tion 5.1), from which orthotropic material properties are de-
Fig. 13. The rhombus tiles are modified with an interlock- rived (Table 1).
ing joint to hold the edge beams of the bases and counteract The demonstrator patches are generated as 2D meshes,
the thrust forces introduced by the structure. The connec- which are then moved to their starting positions and orien-
tion mechanism is dry and demountable. The interlocking tation (as shown in Section 3.4), from which their solid ge-
tiles have redundant connectivity because they are designed ometry is generated. The latter is imported in the ANSYS
to provide a diffuse restraint to the thrust forces induced by software package and is re-meshed by means of finite brick
the base spirals. elements. Each solid patch is equipped with a local reference
system to correctly orient the material anisotropy. This infor-
mation has been extracted from the fabrication subroutine, in
which the patches are nested and oriented on the raw panels
to be processed by the milling machine. It is worth men-
tioning that the patches are not located on the milling plane
to pursue a mechanical criterion but rather a cost- and time-
saving one. As for the manufacturing, the spiral optimization
does not consider the material anisotropy and therefore the
✭❛✮ ✭❜✮ ✭❝✮ material strength and the shape accuracy have to be carefully
checked via FE analysis.
❋✐❣✉r❡ ✶✷✿ ❉❡t❛✐❧❡❞ ❞❡s✐❣♥✿ ✭❛✮ ❝♦♥♥❡❝t♦r ❜❡t✇❡❡♥ ♣❛t❝❤❡s❀
✭❜✮ ❡❞❣❡ ❜❡❛♠ ❝♦♥♥❡❝t♦r ❛♥❞ ❧❛♠❡❧❧❛❡❀ ✭❝✮ ♣❛♥❡❧ ❢♦♦t✐♥❣s✳
The analysis results are shown in Fig. 14. In phase one, the
patches are bent through the imposed displacement of up to
115 𝑚𝑚. As predicted, this phase produces stress peaks in
the connection points, which are expected to mutually bal-
Basic Penrose Tiles Deformation with Interlocking Edges
ance once spiral arms are connected inter-patch. Phase two
simulates the actual behavior of the assembled demonstrator,
inheriting deformation and strains from the previous phase.
Tile A Tile B
Rhombus A Rhombus B Some modeling simplifications are introduced, especially for
the connections details. The patches connector (Fig. 12a) is
reduced to a face-to-face coupling, in which a bonding con-
straint applies to simulate a full-strength connection. The
connection of the spiral arm to the edge beam (Fig. 12b)
is replaced by springs that link pairs of neighboring arms.
This modeling is conservative since only the axial stiffness
is simulated. The lateral stiffness, provided by the active-
bending of the lamellae, would be hard to estimate without
the explicit modeling of the edge beams, therefore it is ne-
glected. The T-shaped spiral footing (Fig. 12c) is not mod-
eled and has been replaced by a fixed connection on all the
faces pointing down. This is also a simplification since sev-
eral nonlinearities, such as unilateral contacts, are there lo-
cated. All these connections are activated in phase two, so
the patches are internally joined and the bottom faces are at-
tached to the ground. The structure then behave as a whole
and redistributes the stress. Being an indoor installation,
only the gravity load with characteristic value is applied in
phase two. This is also the scenario for which the demon-
strator has been optimized.
Stress peaks occur at sharp vertices and kinks, but these are
❋✐❣✉r❡ ✶✸✿ ❘❤♦♠❜♦✐❞ P❡♥r♦s❡ ❛♣❡r✐♦❞✐❝ ♣❛tt❡r♥ ❣❡♥❡r❛t✐♦♥ not deemed to reproduce the reality and are simply a conse-
✭t♦♣✮ ❛♥❞ ♣❛✈✐♥❣ ❞❡s✐❣♥ ✭❜♦tt♦♠✮✳ quence of the automatic generation of the model. To provide
a qualitative idea of the material utilization, Fig. 15 reports
a histogram the equivalent von Mises stress, which shows
5.4. Modeling and structural analysis results that the material is employed in a small spectrum of val-
With the exception of edge beams, all the demonstra- ues. However, von Mises stresses are not meaningful for
tors’ wooden components have been machined from 15 𝑚𝑚 an orthotropic material, therefore the stress is recalculated
plywood, composed of hardwood and softwood plies. The for each patch accounting for the fibers direction in which

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✵ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s
❚❛❜❧❡ ✶
❖rt❤♦tr♦♣✐❝ ♠❛t❡r✐❛❧ ♣r♦♣❡rt✐❡s ❛❞♦♣t❡❞ ✐♥ t❤❡ ❋❊ ❛♥❛❧②s✐s ♦❢ t❤❡ ❞❡♠♦♥str❛t♦r
❙❡❝t✐♦♥ Pr♦♣❡rt✐❡s ❉❡♥s✐t② ❘✐❣✐❞✐t② ❊❧❛st✐❝✐t②
◆♦♠✐♥❛❧ t❤✐❝❦♥❡ss ★ ♣❧✐❡s 𝐺 𝐺 𝐺 𝐸 𝐸 𝐸𝑜𝑢𝑡
✭𝑚𝑚✮
0 90 𝑜𝑢𝑡 0 90
3 2 2 2 2 2
𝑘𝑔∕𝑚 𝑁∕𝑚𝑚 𝑁∕𝑚𝑚 𝑁∕𝑚𝑚 𝑁∕𝑚𝑚 𝑁∕𝑚𝑚 𝑁∕𝑚𝑚2
✶✺ ✼ ✺✷✵ ✷✹✽✳✹ ✷✶✸✳✻ ✻✵✵ ✶✶✵✻✼✳✻ ✾✾✸✷✳✹ ✻✵✵

✭❛✮ ✭❜✮
❋✐❣✉r❡ ✶✹✿ ❋❊ ❛♥❛❧②s✐s r❡s✉❧ts✿ ✭❛✮ ❞✐s♣❧❛❝❡♠❡♥ts ✐♥ ♣❤❛s❡ ✶ ❛♥❞ ✭❜✮ ♣❤❛s❡ ✷✳ ❚❤❡ s♣r✐♥❣s
❛r❡ ❛❝t✐✈❛t❡❞ ❛❢t❡r t❤❡ ♣❤❛s❡ ✶✳

the strength is tested. Fig. 16 reports these results (along 5.5. Fabrication and assembly
the grain, across the grain and in-plane shear) for an ex- The interface between optimized design and fabrication
emplary case of one of the patches attached to the ground is managed through a Grasshopper workflow. The process
bent towards the outer direction (Z in the patch local frame), starts with a set of planar mesh, which is the outcome of
showing one of the most stressed areas of the structure. Due the optimization and parametrization (flattening) step. Each
to bending, inner (Z positive) and outer (Z negative) faces mesh boundary is converted into a NURBS, and the free ex-
present high, negative or positive respectively, stress val- tremities of the spiral arms are replaced with the three differ-
ues. The material strength has not been experimental inves- ent connector types: a) inner joint to mutually connect the
tigated, however the obtained stress is within the usual nom- patches; b) an open boundary connector to enable the instal-
inal capacity of this kind of material, if peaks due to mod- lation of the edge beams; c) a footing to connect the spirals
eling are neglected. Moreover, as a further confirmation of to the ground beams (Fig. 17). Additionally, the workflow
this result, no visible failures appeared during the assembly marks each spiral with a numeric label to be milled at its
and in the as-built structure. center.
At the fabrication level, the setting of milling precision, tool
depth, and tolerances have been considered to avoid surface
damage. These variables have a significant role since they
affect both the node’s structural capacity and its interlocking,
especially for the inner connectors. These cutting lines are
set to be processed with a 3 𝑚𝑚 diameter tool. The calibra-
tion of tolerances between the extremities of the connectors
is set to 0.12 𝑚𝑚, which has been deduced experimentally
among a set of mock-ups. This choice is a compromise be-
tween the necessity to restore the stiffness of a continuous
element, which would have required the tightest tolerance
possible, and the necessity to be easily installed, which con-
versely is facilitated by a large tolerance. All the other cut-
❋✐❣✉r❡ ✶✺✿❋❊ ❛♥❛❧②s✐s r❡s✉❧ts✿ ❤✐st♦❣r❛♠ ♦❢ ♣❤❛s❡ ✷ ❡q✉✐✈❛❧❡♥t ting lines are performed with a 6 𝑚𝑚 diameter tool. The
✈♦♥ ▼✐s❡s str❡ss✳ patches and the floor tiles are then nested automatically to
minimize the waste of material and sent out for CNC pro-

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✶ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

✭❛✮ ✭❜✮ ✭❝✮

❋✐❣✉r❡ ✶✻✿ ❋❊ ❛♥❛❧②s✐s r❡s✉❧ts ♦♥ ♣❛♥❡❧ ✻✺ ✐♥ ♣❤❛s❡ t✇♦ ✭❜♦tt♦♠ ❢❛❝❡s ❛r❡ ❛tt❛❝❤❡❞ t♦
❣r♦✉♥❞✮✿ ✭❛✮ ♥♦r♠❛❧ str❡ss ❛❧♦♥❣ t❤❡ ❣r❛✐♥❀ ✭❜✮ ♥♦r♠❛❧ str❡ss ❛❝r♦ss t❤❡ ❣r❛✐♥❀ ✭❝✮ ✐♥✲♣❧❛♥❡
s❤❡❛r str❡ss✳ ❚❤❡ ❣r❛✐♥ ❞✐r❡❝t✐♦♥ ✐s ✐❞❡♥t✐✜❡❞ ❛s ❨ ✐♥ t❤❡ ♣❛t❝❤ ❧♦❝❛❧ ❢r❛♠❡✳

duction. A CNC machine with 3.00 × 1.60 meters bed size FlexMaps panel. Once this area has been completed, it was
has been used. Eventually, the only manual post-processing secured and stiffened by edge beams. The assembly work
work was to sand all the cut surfaces to remove the wood moved later on the two other saddles, which have been pro-
shavings (Fig. 18). gressively formed and later stiffened by the edge beams (Fig.
19). At this point, the shape has become stiff enough to be
locked onto the ground beams, while the paving and the first
layer of the ground beams have been easily assembled as a
large puzzle of interlocking tiles.
The entire prototype (Fig. 20) has been designed, opti-
mized, and engineered in about four weeks. However, it
has to be considered that some design aspects are not in-
tegrated into the main automated pipeline. The fabrication
only has taken one week; the assembly employed on average
1220 mm

four people for about 18 hours. All the assembly operations

have been manually performed and not less than 3-4 people
are required in order to bend and connect a couple of patches
simultaneously while securing the connectors by bolts. Dur-
ing these operations, ‘low-tech’ scaffolding, like tables and
2440 mm
chairs, have been employed. This was allowed as the shape
❋✐❣✉r❡ ✶✼✿ ❋❧❛t ✷❉ s❤❡❡t t♦ ❜❡ ❝✉t ✇✐t❤ ❤✐❣❤❧✐❣❤t❡❞ t❤❡ ❞✐✛❡r✲ accuracy relied entirely on the properties of the spirals, the
❡♥t ❝♦♥♥❡❝t♦r t②♣❡s ✭♦♣❡♥ ❜♦✉♥❞❛r② ❝♦♥♥❡❝t♦r✱ ✐♥♥❡r ❝♦♥♥❡❝✲ precision of the connections, and the guiding provided by
t♦r ❛♥❞ ❢♦♦t✐♥❣✱ r❡s♣❡❝t✐✈❡❧②✮ ❛♥❞ t❤❡ ❧❛❜❡❧✐♥❣ s②st❡♠✳ the ground and edge beams.

Table 2 includes a summary of the pieces, material, and 5.6. Verification and discussion
quantities needed for the assembly of the demonstrator. The The development of a full scale prototype in a constrained
assembly phase has been assisted by a 3D model showing time and space represented a real test for the proposed method-
the patches and the label of each quad, which corresponds to ology from both a design and a construction point of view.
the milled label on each spiral. Thus, its position could have The computational model has been informed by the experi-
been uniquely recognized from its adjacencies. mental tests on the arch specimen, and during this moment
Due to time constraints, a staged analysis of the prototype as- the team has familiarized with the basic assembly operations
sembly process has not been performed. Conversely, since and has gained confidence on how these procedures may af-
the phase one of the structural analysis evidenced no failures, fect the final shape of the built structure, as well as any initial
the panels would remain within the material strength range imperfection or supports misalignment. However, the arch
as long as every single patch stays close enough to its target test has been opportunely designed to avoid such issues and
position (deformation state) during the entire assembly pro- to be easily controlled due to its simple shape and its man-
cedure. ageable size.
Therefore, the demonstrator has been assembled starting from An interesting verification can be carried out by compar-
the larger saddle, sequentially bending and fastening each ing the shapes of the actual built demonstrator and the de-

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✷ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

❋✐❣✉r❡ ✶✽✿ ❋❛❜r✐❝❛t✐♦♥ ♦❢ t❤❡ ❞❡♠♦♥str❛t♦r✿ ❈◆❈ ❝✉tt✐♥❣ ❛♥❞ ❝♦♠♣♦♥❡♥ts ❛❢t❡r s❛♥❞✐♥❣✳

❚❛❜❧❡ ✷
❉❡t❛✐❧s ♦❢ t❤❡ ❝♦♠♣♦♥❡♥ts ❡♠♣❧♦②❡❞ ✐♥ t❤❡ ❛ss❡♠❜❧② ♦❢ t❤❡ ❞❡♠♦♥str❛t♦r✳

▼❛t❡r✐❛❧ ❚②♣❡ ◗✉❛♥t✐t②✴❉❡s❝r✐♣t✐♦♥ ❲❡✐❣❤t Pr♦❞✉❝t✐♦♥ t✐♠❡

❙♣✐r❛❧ ♣❛♥❡❧s ♣❧②✇♦♦❞ ✶✺ ♠♠ ✭✇✐❞t❤✮ ✼✽ ❜❡s♣♦❦❡ ♣❛t❝❤❡s ✷✹ ❦❣ ✭t♦t❛❧✮ ✺ ❞❛②s

P❛✈✐♥❣ t✐❧❡s ♣❧②✇♦♦❞ ✶✺ ♠♠ ✭✇✐❞t❤✮ ✼✾ t✐❧❡ ❆✱ ✹✸ t✐❧❡ ❇✱ ✸✹ ❜❡✲ ✶✷✳✶ ❦❣ ✭t♦t❛❧✮ ✭✇✐t❤ ♣r❡✈✐♦✉s✮
s♣♦❦❡

●r♦✉♥❞ ❜❡❛♠s ♣❧②✇♦♦❞ ✶✺ ♠♠ ✭✇✐❞t❤✮ ✸ ❜❡❛♠s ♠❛❞❡ ♦❢ t❤r❡❡ ✶✽✳✼ ❦❣ ✭t♦t❛❧✮ ✶ ❞❛②
❧❛②❡rs ✭t❤❡ t♦♣ ♦♥❡ ✐s s❡❣✲
♠❡♥t❡❞✮

❊❞❣❡ ❜❡❛♠s ♣❧②✇♦♦❞ ✻ ♠♠ ✭✇✐❞t❤✮ ✶ ❞❛②

❇♦❧ts ❙t❛✐♥❧❡ss st❡❡❧ ▼✺ ✲

❙❝r❡✇s ❙t❛✐♥❧❡ss st❡❡❧ ▼✽ ✲

❇❛❧❧❛st ❞✐s❝s ■r♦♥ ✲ ✶ ❦❣ ✭❡❛❝❤✮ ✲

formed model obtained as outcome of the FE analysis. Dur- wards the ground beams, producing a shifting of the fixed
ing the assembly, the demonstrator could not rely on step- supports of the model that is reasonably responsible of most
by-step shape checks and on a verified minimal-distortion of the difference between the two shapes. Larger discrepan-
procedure. The latter would have indeed required a dramatic cies are located at the top of the structure. Indeed, this imper-
increase of computational time to run combinatorial staged fect position of the boundary constraints makes the structure
simulations. Thus, the accuracy of the final shape is totally softer, so its inner part next to the hole can displace down-
demanded to the capacity of the panels of matching exactly ward, while the external border is held in place by the edge
the target shape they are designed for, to the redistribution beams.
of internal stress, and not least to the team working ability. To support this evidence another FE model has been recom-
The ground beams and edge beams supported this phase by puted displacing the ground supports to the new position
restraining the inner distance between the edge connectors. detected by the scanner, given the same orthotropic mate-
To acquire the actually built shape a scan session has been rial properties. The resulting shape is compared with the
performed one week after the installation through a Artec scan data in Fig. 22. Once the macroscopic imperfection
handheld scanner. The comparison between the FE deformed introduced by floor tiles has been included in the model,
shape and the as-built object is shown in Fig. 21. the overall accuracy has been largely improved, and the de-
The as-built structure suffers from a large misalignment of formed shape is matched satisfactorily. However, deviations
the ground beams in the range of 10-30 𝑚𝑚. This is not re- can be observed locally due to the necessary model simpli-
lated to the acquisition procedure but to the tiles’ fabrication. fications. In fact, the FE model is not represented in all the
In that phase the tolerance between the tiles has not been set details that the built structure is endowed with and so it did
properly, and the tiles needed to be filed and forced to fit in not embed all the features in both geometry and mechanics:
place while assembled. Therefore, this error propagates to- i.e. the connectors, the edge beams and the ground beams

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✸ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

❋✐❣✉r❡ ✶✾✿ ❆ss❡♠❜❧② ♦❢ t❤❡ ❞❡♠♦♥str❛t♦r✳

are simplified, bolts, screws and washers, contact nonlin- ing multiple plies and grain in different directions, plywood
earities are not included etc. The wood relaxation during broadly mitigates these effects (i.e. the Young’s modulus
time has not been considered. Most importantly, disregard- differs by about the 10% along the two directions). This
ing these factors it appears that the proposed method is ef- is the reason why plywood has been considered compati-
fective in terms of matching the target shape. There are no ble with this specific application and with an optimization
local mismatches that are attributable to the surface quad- routine that is designed for isotropic material behavior. An
rangulation or mesostructures form-finding. Therefore, the additional weak point is represented by the edges status. In
present pipeline can be considered suitable to produce com- particular a careful attention has been paid in avoiding local
plex shapes for architectural purposes through bending-active cracks since they can be starting points for tensile failures.
components. This has been achieved by selecting appropriate milling tools
The performance of the demonstrator is also promising and by tuning their velocity. Due to the layered nature of the
from the structural system perspective, even though the choice material, other critical parts are the interfaces of the connec-
of layered and anisotropic material such as plywood has brought tors where the risk of delamination during the assembly is
several issues, as opposed to its strengths the affordable cost very high.
and ease of fabrication and handling. From the material stiff- As main result, the present method has demonstrated fea-
ness point of view, the selection of wood is in contrast with sibility from a technical and a fabrication viewpoint, and
carved unconventional geometric patterns for two reasons. good capacity of matching a target shape under certain loads
First, the direction along the grain is stronger than the other (i.e. gravity) requiring limited assistance during the assem-
across; the result is that the stiffness locally depends on the bly and few devices to keep the structure in place, such as
grain orientation. Second, weak points are also produced if edge lamellae and ground beams. However, the scaling of
the grain runs across a narrow width, namely the so-called this system towards larger spans has to currently deal with
‘short grain’ effect; the cuts break the continuity of fibers a limitation that concerns the behavior of the structure un-
so the material could not attain its nominal strength. Hav- der new loading scenarios. Once the self-shaping process is

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✹ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

❋✐❣✉r❡ ✷✵✿ P❤♦t♦s ♦❢ t❤❡ ❝♦♠♣❧❡t❡❞ str✉❝t✉r❡ ❛♥❞ ❝❧♦s❡✉♣ ♦❢ t❤❡ ♠❛✐♥ ❞❡t❛✐❧s ♦❢ t❤❡
s②st❡♠✳

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✺ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

❋✐❣✉r❡ ✷✶✿ ❙❤❛♣❡ ❝♦♠♣❛r✐s♦♥ ❜❡t✇❡❡♥ t❤❡ t❤❡♦r❡t✐❝❛❧ ❞❡❢♦r♠❡❞ s❤❛♣❡ ✭❝♦❧♦r❡❞✮ ❛s ♦✉t❝♦♠❡
♦❢ t❤❡ ❋❊ ❛♥❛❧②s✐s ❛♥❞ t❤❡ r❡❛❧ s❝❛♥ ❞❛t❛ ✭❣r❛②✮✳

❋✐❣✉r❡ ✷✷✿ ❙❤❛♣❡ ❝♦♠♣❛r✐s♦♥ ❜❡t✇❡❡♥ t❤❡ t❤❡♦r❡t✐❝❛❧ ❞❡❢♦r♠❡❞ s❤❛♣❡ ✭❝♦❧♦r❡❞✮ ❛s ♦✉t✲
❝♦♠❡ ♦❢ t❤❡ ❋❊ ❛♥❛❧②s✐s ❛♥❞ t❤❡ r❡❛❧ s❝❛♥ ❞❛t❛ ✭❣r❛②✮✳ ❚❤❡ ❋❊ ♠♦❞❡❧ ❤❛s ❜❡❡♥ ✉♣❞❛t❡❞
❝♦♥s✐❞❡r✐♥❣ t❤❡ ♥❡✇ ♣♦s✐t✐♦♥ ♦❢ t❤❡ ❣r♦✉♥❞ ❢❛❝❡s✱ ❛s s✉r✈❡②❡❞ t❤r♦✉❣❤ t❤❡ s❝❛♥ s❡ss✐♦♥✳

concluded, the structure has to develop membrane actions as morphology and the physical behavior of the individual spi-
a lattice shell and to avoid any further bending. The strate- rals. The framework enables a new way to fabricate bending-
gies of triggering the membrane behavior due to new loading active structures with differentiated material behaviors.
and its evaluation require future studies. This method can be considered an integral approach to de-
sign bending-active structures [25] as it introduces shape
and material control in the design phase, as well as files-
6. Conclusions to-factory pipeline for subtractive fabrication. To demon-
This paper introduces a proper architectural pipeline to strate the method effectiveness and validate the proposed
design and fabricate actively-bent complex structures exhibit- technique a free form structure has been fabricated out of 15-
ing double curvature through cost-effective and time-saving mm thick milled plywood panels. This structure shows the
technology. This pipeline modifies and extends the FlexMaps enormous potential of this approach in architecture as it can
technique, initially tailored for small-scale objects. These pave the way for larger structures, providing excellent shape
structures are provided with custom spiraling mesostructured expressiveness while employing common building material.
panels, so that a differentiated local stiffness is achieved by
controlling the spiral parameters.
The developed tool allows the designer to reproduce a given Acknowledgments
target geometry by controlling the structure deformation be- The authors would gratefully thank Dane Voorderhake,
havior. This is performed by varying the mesostructure’s Lucy Wang, Nathan Gonsalves, Ella Williams and the staff

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✻ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

of the Advanced Fabrication Lab (UTS Faculty of the Archi- [16] F. Gil-Ureta, N. Pietroni, D. Zorin, Reinforcement of general shell
tecture), Stefano Nuvoli for the support during the assem- structures, ACM Transactions on Graphics 39 (2020). URL: https:
//doi.org/10.1145/3375677. doi:10.1145/3375677.
bly, and Hamish McIntosh and Francesca Giuliani for the
[17] F. Laccone, L. Malomo, M. Froli, P. Cignoni, N. Pietroni, Automatic
demonstrator photos. design of cable-tensioned glass shells, Computer Graphics Forum 39
(2020) 260–273. URL: https://onlinelibrary.wiley.com/doi/abs/10.
1111/cgf.13801. doi:10.1111/cgf.13801.
Funding [18] H. Pottmann, Architectural geometry and fabrication-aware de-
This research was partially developed during the Short sign, Nexus Network Journal 15 (2013) 195–208. doi:10.1007/
s00004-013-0149-5.
Term Mobility (STM) of Francesco Laccone, hosted by Nico [19] Y. Liu, H. Pottmann, J. Wallner, Y.-L. Yang, W. Wang, Geometric
Pietroni at the UTS. The STM scholarship was funded by the Modeling with Conical Meshes and Developable Surfaces, ACM
Italian National Research Council (CNR). Transactions on Graphics 25 (2006) 681–689. doi:10.1145/1141911.
1141941.
[20] K. Gavriil, A. Schiftner, H. Pottmann, Optimizing B-spline sur-
References faces for developability and paneling architectural freeform surfaces,
[1] L. Malomo, J. Pérez, E. Iarussi, N. Pietroni, E. Miguel, P. Cignoni, Computer-Aided Design 111 (2019) 29 – 43. doi:10.1016/j.cad.2019.
01.006.
B. Bickel, FlexMaps: computational design of flat flexible shells
for shaping 3D objects, ACM Transactions on Graphics (2018) 241. [21] C. Jiang, C. Wang, F. Rist, J. Wallner, H. Pottmann, Quad-Mesh
doi:10.1145/3272127.3275076. Based Isometric Mappings and Developable Surfaces, ACM Transac-
[2] F. Barbosa, J. Woetzel, J. Mischke, Reinventing Construction: A tions on Graphics (2020). URL: http://hdl.handle.net/10754/663008.
Route of Higher Productivity, Technical Report, McKinsey Global doi:10.1145/3386569.3392430.
Institute, 2017. [22] J. Lienhard, P. Eversmann, New hybrids – from textile logics towards
[3] U. Nations, World urbanization prospects: The 2014 revision, high- tailored material behaviour, Architectural Engineering and Design
lights. department of economic and social affairs, Population Divi- Management 0 (2020) 1–6. doi:10.1080/17452007.2020.1744421.
sion, United Nations 32 (2015). [23] J. Panetta, Q. Zhou, L. Malomo, N. Pietroni, P. Cignoni, D. Zorin,
[4] H. Pottmann, Architectural geometry, volume 10, Bentley Institute Elastic Textures for Additive Fabrication, ACM Transactions on
Press, 2007. Graphics 34 (2015). doi:10.1145/2766937.
[5] S. Adriaenssens, P. Block, D. Veenendaal, C. Williams, Shell struc- [24] C. Schumacher, B. Bickel, J. Rys, S. Marschner, C. Daraio, M. H.
tures for architecture: form finding and optimization, Routledge, Gross, Microstructures to control elasticity in 3d printing, ACM
2014. Transactions on Graphics 34 (2015) 136:1–136:13. doi:10.1145/
2766926.
[6] M. Eigensatz, M. Kilian, A. Schiftner, N. J. Mitra, H. Pottmann,
M. Pauly, Paneling Architectural Freeform Surfaces, ACM Trans- [25] J. Lienhard, H. Alpermann, C. Gengnagel, J. Knippers, Active bend-
actions on Graphics 29 (2010) 45:1–45:10. doi:10.1145/1778765. ing, a review on structures where bending is used as a self-formation
1778782.
process, International Journal of Space Structures 28 (2013) 187–196.
[7] Y. Liu, W. Xu, J. Wang, L. Zhu, B. Guo, F. Chen, G. Wang, General doi:10.1260/0266-3511.28.3-4.187.
Planar Quadrilateral Mesh Design Using Conjugate Direction Field, [26] M. Fleischmann, A. Menges, ICD/ITKE Research Pavilion: A Case
ACM Transactions on Graphics 30 (2011) 1–10. URL: https://doi. Study of Multi-disciplinary Collaborative Computational Design, in:
org/10.1145/2070781.2024174. doi:10.1145/2070781.2024174.
C. Gengnagel, A. Kilian, N. Palz, F. Scheurer (Eds.), Computational
[8] M. Deuss, D. Panozzo, E. Whiting, Y. Liu, P. Block, O. Sorkine- Design Modelling, Springer Berlin Heidelberg, Berlin, Heidelberg,
Hornung, M. Pauly, Assembling self-supporting structures, ACM 2012, pp. 239–248.
Transactions on Graphics 33 (2014) 214:1–214:10. doi:10.1145/ [27] K. Crolla, Bending Bamboo Rules: Beyond Century-Old Typologies,
2661229.2661266.
Journal of Architectural Education 72 (2018) 135–145. doi:10.1080/
10464883.2018.1410669.
[9] M. Skouras, S. Coros, E. Grinspun, B. Thomaszewski, Interactive sur-
face design with interlocking elements, ACM Transactions on Graph- [28] C. Gengnagel, E. L. Hernández, R. Bäumer, Natural-fibre-reinforced
ics 34 (2015) 224:1–224:7. doi:10.1145/2816795.2818128. plastics in actively bent structures, Proceedings of the Institution of
[10] Z. Wang, P. Song, F. Isvoranu, M. Pauly, Design and Structural Opti- Civil Engineers-Construction Materials 166 (2013) 365–377. doi:10.
1680/coma.12.00026.
mization of Topological Interlocking Assemblies, ACM Transactions
on Graphics 38 (2019). doi:10.1145/3355089.3356489. [29] J. Lienhard, R. La Magna, J. Knippers, Form-finding bending-active
[11] J. N. Richardson, S. Adriaenssens, R. Filomeno Coelho, P. Bouillard, structures with temporary ultra-elastic contraction elements, WIT
Coupled form-finding and grid optimization approach for single layer Transactions on The Built Environment 136 (2014) 107. doi:10.2495/
MAR140091.
grid shells, Engineering Structures 52 (2013) 230–239. URL: https:
//www.sciencedirect.com/science/article/pii/S0141029613000813.
[30] P. Cuvilliers, J. R. Yang, L. Coar, C. Mueller, A comparison of
doi:10.1016/j.engstruct.2013.02.017. two algorithms for the simulation of bending-active structures, Inter-
[12] N. Pietroni, D. Tonelli, E. Puppo, M. Froli, R. Scopigno, national Journal of Space Structures 33 (2018) 73–85. doi:10.1177/
0266351118779979.
P. Cignoni, Statics Aware Grid Shells, Computer Graph-
ics Forum 34 (2015) 627–641. URL: https://onlinelibrary. [31] M. C. Phocas, K. Alexandrou, S. Athini, Design and analysis of an
wiley.com/doi/abs/10.1111/cgf.12590. doi:10.1111/cgf.12590. adaptive hybrid structure of linearly interconnected scissor-like and
arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1111/cgf.12590.
cable bending-active components, Engineering Structures 192 (2019)
[13] C. Jiang, C. Tang, H.-P. Seidel, P. Wonka, Design and Volume Op- 156 – 165. URL: http://www.sciencedirect.com/science/article/
pii/S0141029618332796. doi:10.1016/j.engstruct.2019.04.102.
timization of Space Structures, ACM Transactions on Graphics 36
(2017) 159:1–159:14. doi:10.1145/3072959.3073619. [32] P. Nicholas, M. Tamke, Computational strategies for the architectural
[14] N. Pietroni, M. Tarini, A. Vaxman, D. Panozzo, P. Cignoni, Position- design of bending active structures, International Journal of Space
based Tensegrity Design, ACM Transactions on Graphics 36 (2017) Structures 28 (2013) 215–228. doi:10.1260/0266-3511.28.3-4.215.
172:1–172:14. doi:10.1145/3130800.3130809. [33] J. Lienhard, Bending-active structures: form-finding strategies using
[15] M. Kilian, D. Pellis, J. Wallner, H. Pottmann, Material-Minimizing elastic deformation in static and kinetic systems and the structural
Forms and Structures, ACM Transactions on Graphics 36 (2017). potentials therein, 36, Institut für Tragkonstruktionen und Konstruk-
doi:10.1145/3130800.3130827. tives Entwerfen (ITKE), Universität Stuttgart, 2014. URL: https:

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✼ ♦❢ ✶✽

 ■♥t❡❣r❛t❡❞ ❝♦♠♣✉t❛t✐♦♥❛❧ ❢r❛♠❡✇♦r❦ ❢♦r ❜❡♥❞✐♥❣✲❛❝t✐✈❡ str✉❝t✉r❡s

//elib.uni-stuttgart.de/handle/11682/124. doi:10.18419/opus-107.
[34] J. Lienhard, J. Knippers, Bending-active textile hybrids, Journal of the
International Association for Shell and Spatial Structures 56 (2015)
37–48.
[35] D. Sonntag, S. Bechert, J. Knippers, Biomimetic timber shells made
of bending-active segments, International Journal of Space Struc-
tures 32 (2017) 149–159. URL: https://journals.sagepub.com/doi/
10.1177/0266351117746266. doi:10.1177/0266351117746266.
[36] M. Tamke, Y. S. Baranovskaya, F. Monteiro, J. Lienhard,
R. L. Magna, M. R. Thomsen, Computational knit – de-
sign and fabrication systems for textile structures with cus-
tomised and graded CNC knitted fabrics, Architectural Engi-
neering and Design Management 0 (2020) 1–21. URL: https://
doi.org/10.1080/17452007.2020.1747386. doi:10.1080/17452007.2020.
1747386. arXiv:https://doi.org/10.1080/17452007.2020.1747386.
[37] R. La Magna, Bending-active plates: strategies for the induction
of curvature through the means of elastic bending of plate-based
structures, 43, Institut für Tragkonstruktionen und Konstruktives
Entwerfen (ITKE), Universität Stuttgart, 2017. URL: https://elib.
uni-stuttgart.de/handle/11682/9406. doi:10.18419/opus-9389.
[38] J. Panetta, M. Konaković-Luković, F. Isvoranu, E. Bouleau, M. Pauly,
X-Shells: A New Class of Deployable Beam Structures, ACM Trans-
actions on Graphics 38 (2019). doi:10.1145/3306346.3323040.
[39] F. Laccone, L. Malomo, J. Pérez, N. Pietroni, F. Ponchio, B. Bickel,
P. Cignoni, FlexMaps Pavilion: a twisted arc made of mesostructured
flat flexible panels, in: Proceedings of IASS Annual Symposia, vol-
ume 2019, International Association for Shell and Spatial Structures
(IASS), 2019, pp. 1–7.
[40] F. Laccone, L. Malomo, J. Pérez, N. Pietroni, F. Ponchio, B. Bickel,
P. Cignoni, A bending-active twisted-arch plywood structure: compu-
tational design and fabrication of the FlexMaps Pavilion, SN Applied
Sciences 2 (2020). doi:10.1007/s42452-020-03305-w.
[41] N. Pietroni, E. Puppo, G. Marcias, R. Scopigno, P. Cignoni,
Tracing field-coherent quad layouts, Computer Graphics
Forum 35 (2016) 485–496. URL: https://onlinelibrary.
wiley.com/doi/abs/10.1111/cgf.13045. doi:10.1111/cgf.13045.
arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1111/cgf.13045.
[42] M. S. Floater, K. Hormann, Surface Parameterization: a Tutorial and
Survey, in: N. A. Dodgson, M. S. Floater, M. A. Sabin (Eds.), Ad-
vances in Multiresolution for Geometric Modelling, Springer, 2005,
pp. 157–186.
[43] ANSYS, ANSYS®. Academic Research Mechanical Release 18.0.,
2018.
[44] R. McNeel, Associates, Grasshopper generative modeling for Rhino,
Computer software, http://www. grasshopper3d. com (2018).
[45] N. Kotelnikova-Weiler, C. Douthe, E. L. Hernandez, O. Baverel,
C. Gengnagel, J.-F. Caron, Materials for actively-bent structures, In-
ternational Journal of Space Structures 28 (2013) 229–240. doi:10.
1260/0266-3511.28.3-4.229.

▲❛❝❝♦♥❡ ❡t ❛❧✳✿ Preprint submitted to Elsevier P❛❣❡ ✶✽ ♦❢ ✶✽