3D-Printed Stay-in-Place

Formwork for Topologically

Optimized Concrete Slabs
Andrei Jipa Mania Aghaei Meibodi
Doctoral Student, Digital Building Technologies, ETH Senior Researcher, Digital Building Technologies,
Zurich ETH Zurich

Mathias Bernhard Benjamin Dillenburger

Senior Researcher, Digital Building Technologies, ETH Assistant Professor, Digital Building Technologies,
Zurich ETH Zurich

INTRODUCTION (e.g. casting and moulding) fabrication processes

Material efficiency is becoming a critical design driver impose significant manufacturing constraints on the
in the construction industry. While many strategies optimized form and its topological features. Three-axis
for improving material efficiency focus on the end of a milling, for example, is limited by tool-head accessibility
building’s lifecycle (recycling materials, reusing compo- and therefore cannot be used to fabricate undercuts.
nents, reducing waste, extending life spans, etc.), there Five-axis milling is more tolerant of undercuts, but
is also great potential for reducing material use in the often parts must be split into smaller subdivisions to
early design phases. This is especially significant for prevent clashes with the larger tool heads. Formative
materials that are difficult to recycle, such as concrete fabrication has directional limitations determined
(Allwood et al. 2011). by the rheology of the casting material and by the
The usual means for achieving material reduction demoulding process (fig. 1).
with concrete are hollow-core construction systems, However, 3D printing, or additive manufacturing, is a
pre-stressing, and the use of lightweight concrete. process that promises almost no fabrication constraints,
Computational methods, such as the optimization of potentially enabling the production of topologically opti-
size, shape, and topology, can also be used to ensure the mized complex geometries. The aim of this research is
efficient distribution of concrete for a given part. While to demonstrate that large-scale parts can be fabricated
significant material reduction can be achieved with these with additive manufacturing. To investigate this hypoth-
methods, the resulting geometries are often so intricate esis, two concrete slab components were designed with
that fabrication becomes problematic (Dombernowsky the aid of topology optimization algorithms, and fabri-
and Søndergaard, 2011). cated using 3D printing.
The designer is confronted with the compromise
between optimal material distribution and fabrication TOPOLOGY OPTIMIZATION
constraints. Subtractive (e.g. milling) and formative Designing for efficient material distribution can be

97 2016 TxA EMERGING DESIGN + TECHNOLOGY 98

 3D-printing facility currently available commer- Given the fact that with PSPF structural components
cially is VoxelJet VX4000, which uses sand to can be fabricated with indefinite geometric freedom,
Figure 1: Topology produce parts in sizes up to 4 x 2 x 1 m3. Binder the questions addressed in this research are these:
optimization with
jetting, the technology used by this machine, is If additive manufacturing allows the prefabrication
fabrication constraints
for different processes: of particular interest to this research. Binder jet- of large, complex architectural parts with optimized
(A) no fabrication ting is a type of 3D-printing technology in which topologies, what is the potential and what are the lim-
constraints (for
reference); (B) 3D
layers of powder material are selectively bonded itations of the PSPF fabrication method with regard to
printing; (C) casting by a resin jet. This type of technology is partic- material optimization?
and demoulding; (D) ularly suitable for the fabrication of optimized
three-axis milling.
Unlike the other case topologies because it does not entail the use of TOPOLOGICALLY OPTIMIZED 3D-PRINTED
studies, 3D-printing auxiliary supports when printing geometries with CONCRETE SLAB ELEMENTS
constraints have a
undercuts and internal voids—the bed of uncon- To explore the suitability of topology optimization algo-
minor impact on the
reference topology. solidated powder provides this functionality by rithms for PSPF additive fabrication, two prototypes Figure 2: Different
computational optimi-
default. Binder jetting produces clean surface fin- were designed and fabricated, investigating: zation processes: size,
ishes and accurate details in the range of 0.2 mm. • design with two different topology optimization shape, and topology.
• While the relatively low bending strength of tools—a free plug-in with limited functionality Size optimization
alters only the size
3D-printed sand (Stutz and de Taisne 2016) makes and a robust commercial application; of members, without
it inadequate for large-scale structural applica- • adaptation of the design to suit specific fabrica- changing their shape.
Shape optimization
tions, this drawback can be overcome if binder tion constraints;
alters the part
jetting is used to 3D print stay-in-place formwork • and the fabrication process using stay-in-place with continuous
(PSPF) for concrete. This fabrication method was 3D-printed formwork for fiber-reinforced concrete. deformations only,
without modifying the
developed recently by the authors to combine topology. Topology
the structural properties of ultra-high-perfor- To demonstrate the applicability of this process to archi- optimization is the
mance fiber-reinforced concrete (UHPFRC) with tecture, the prototypes are large-scale examples of pre- most versatile process,
allowing changes
the fabrication freedom of binder-jet 3D printing fabricated concrete slabs measuring 1.8 x 1 m2—the full in size, shape, and
(Aghaei-Meibodi et al. submitted). size of the Ex-One S-MAX 3D printer bed. topology.

achieved through size, shape, or topology optimization running costs, and reducing CO2 emissions. Small-
processes. Size optimizations are contained within a scale 3D-printed prototypes of topologically optimized
fixed shape, while shape optimizations are constrained connectors and hinges for planes, satellites, and Formula
by a fixed topology (fig. 2). Topology optimization pro- 1 racing cars have already been successfully created.
cesses will therefore be considered in this research, as For large-scale construction, topology optimization
they are the most versatile and most broadly applicable, frameworks do exist (Beghini et al. 2016; Tomas and Marti
being capable of improving material distribution in terms 2010; Liang and Steven 2000), but built examples are
of size, shape, and topology (Mijar et al. 1998). scarce. This lack of practical investigation is due in part to:
Topology optimization is an iterative computational • the limited availability of large-scale digital fab-
process that works within a confined, discretized rication facilities;
space. For given loads and supports, the algorithm • the limited compatibility of such facilities with
will refine material distribution to meet a prescribed materials suitable for structural applications;
set of performance targets (fig. 3). There are a num- and
ber of different topology optimization algorithms, • the complexity of setting up physically accurate
including Solid Isotropic Microstructure with Penal- and reliable models for topology optimization,
ization (SIMP), Evolutionary Structural Optimization especially for anisotropic materials such as rein-
(ESO), and Topological Derivatives (Rozvany 2009; forced concrete.
Aremu 2010). Despite the computational differences
between these algorithms, they all produce a family of 3D PRINTING IN ARCHITECTURE
typical geometric features: interconnected networks of The issues noted above have thus far restricted the fabri-
thin ribs and narrow tubular structures with dynamic cation of large and topologically optimized architectural
changes in porosity. The focus of this paper is on how components. Nevertheless, recent developments have
these typical features can be fabricated. motivated the endeavour described in this paper:
The greatest interest in topology optimization cur- • Large-scale 3D-printing facilities that can fabri-
rently comes from the aerospace and automotive indus- cate parts without geometric limitations at no
tries (Rozvany 2009), where weight reduction is a critical additional cost are becoming more accessible
design driver for improving performance, decreasing (Dillenburger and Hansmeyer 2013). The largest

99 2016 TxA EMERGING DESIGN + TECHNOLOGY 100

 Figure 3: Convergence Two Topology Optimization Strategies
graph for the optimi- Prototype “A” (figs. 4 and 5) was developed through a
zations in Figure 1.
The goal is to achieve hybrid process based on topology optimization and mesh
an 80% reduction in subdivision. A two-dimensional evolutionary algorithm
material use while
minimizing the strain
was used: Millipede, a free add-on for McNeel Rhinoceros,
energy. Convergence and Grasshopper (Michalatos and Payne 2014). The main Figure 5: Prototype “A,”
close-up detail.
is achieved more goal of the optimization process was to reduce material
slowly with additional
fabrication constraints. to a 0.2 set fraction of the initial amount while minimiz-
ing deformations of the slab under uniform surface load.
Boundary conditions were set to three fixed supports.
The design space was discretized into 135,000 nodes,
and the algorithm was run for 500 cycles, producing a
greyscale bitmap representing material distribution.
This bitmap was subsequently vectorized and given a
three-dimensional ribbed topology based on the grey
values corresponding to the underlying nodes. Cat-
mull–Clark and loop subdivision algorithms were finally
applied to achieve a smooth surface and accommodate
fabrication constraints. The subdivision algorithms were
selectively applied to aesthetically differentiate the ribs
and the fields (fig. 6).
Prototype “B” (fig. 7) was developed through the SIMP
topology optimization algorithm of SIMULIA Abaqus, a
commercial structural analysis software package. The
main goal of the optimization process was to reduce The preliminary tests described above were an essential
material to a 0.18 set fraction of the initial volume while step in the design process. Their role was to establish a
minimizing the stress of the slab under uniform surface series of formal design guidelines derived from the fab-
load. Boundary conditions were set to four simple sup- rication constraints. According to these, the geometric
ports located close to the corners. features of the formwork can be dimensioned in relation
The 1.8 x 1 x 0.15 m3 design domain was discretized to both the length and volumetric content of the fibers
into 83,072 nodes with a volume of approximately 3.4 in the concrete mixture. These guidelines informed the
cm3 each. Other tests were done with 270,336 samples final design of the two prototypes with regard to the
at 1 cm3 and with 2,162,688 samples at 0.12 cm3 (fig. rheological constraints of the concrete casting process.
9). While finer discretization marginally improved the In response to these fabrication constraints, geomet-
quality of the result, it also had a major impact on com- ric refinements had to be included in the design process
putation time, and therefore the coarser samples were in order to:
used for designing the prototypes. • smoothen under-sampling and correct artifacts
resulting from the discretized nature of the topol-
From Optimized Form to Printable Geometry ogy optimization process;
To determine the fabrication constraints specific to • filter out geometric features that were too fragile
Figure 4: Prototype “A,” PSPF, a number of tests were performed in collaboration to be 3D printed;
topology optimization with the chair of Physical Chemistry of Building Mate- • filter out geometric features that were too
of a slab with three
supports. rials at ETH Zurich (Stutz and de Taisne 2016). These narrow to permit 3D printing post-processing
tests investigated: (unconsolidated sand removal and surface infil-
• the properties of 3D-printed formwork in relation tration);
to concrete—how the porosity, sorptivity, and • filter out geometric features that were too nar-
capillary absorption of the sandstone influenced row to permit the flow of concrete because of
the setting of concrete; local fiber clogs; and
• new ultra-high-performance fiber-reinforced • accommodate the architect’s design intention
concrete mixes with ductile behavior; regarding surface quality, ornamentation, edge
• the rheological properties of these new mixes as details, etc.
a relation between fiber content and geometric Fabrication constraints were applied to the intricate
features—inner radii, bending radii, and channel geometry resulting from the optimization process, with
length-to-diameter ratios (fig. 10); and tubular structures being hollow only if they were large
• the properties of the sandstone-to-concrete bond. enough to permit the flow of concrete inside.

101 2016 TxA EMERGING DESIGN + TECHNOLOGY 102

 Figure 7: Prototype “B,”
topology optimization
of a slab with four
supports.

Figure 6: Design pro- Fabrication: Binder Jet 3D Printing and Fiber- necessary geometric adjustments for fabrication have
cess for prototype “A.”
Reinforced Concrete only a minor impact on the calculated optimal topology
From left to right: grey-
scale bitmap resulting For both prototypes, the fabrication process began and are significantly less intrusive than with other pro-
from Millipede, where with binder jetting the sand formwork followed by cesses (e.g., completely eliminating undercuts for three-
each pixel represents
an optimization node;
post-processing the prints (fig. 11). This involved axis milling, subdividing the part into multiple elements
vectorization with removal of unconsolidated sand and infiltration with for five-axis milling tool-head access, or eliminating
hierarchical differenti- epoxy resin in order to increase the strength of the cantilevers for extrusion 3D printing).
ation of ribs and fields;
three-dimensional material. A strategy to avoid damaging the friable For PSPF fabrication, geometric features become Figure 8: Prototype “B,”
close-up detail.
mesh with depth formwork during post-processing was tested. It problematic at a scale of around 20 mm. Such fabrica-
based on greyscale
involves the integration of a protective bed of uncon- tion constraints are close to the material limitations of
values of underlying
pixels; and selective solidated sand contained within a closed 3D-printed concrete anyway (below 20 mm, the structural integ-
subdivision algorithms box (Aghaei-Meibodi et al., submitted). rity of concrete begins to suffer), and such detailed
to achieve desired
aesthetics.
This auxiliary protective box also provides support features have little relevance for large-scale building
for the formwork during the casting of the UHPFRC components.
(fig. 12). The special concrete mix contains 2.75 vol.
% steel fibers 10 mm long and 0.16 mm in diameter. OUTLOOK AND CONCLUSION
The average concrete thickness achieved, 30 mm, With the two successful large-scale prototypes
indicates that weight reductions of up to 70% are pos- described above, this paper can conclude that PSPF
sible. The initial structural tests performed so far by fabrication is sufficiently tolerant of geometric com-
applying a 2,500 kN/m2 distributed load on Prototype plexity to enable the design of architectural components
“B” empirically confirmed the validity of the topology directly through topology optimization. This is possible
optimization algorithm. because of some particular interrelations between the
different aspects of the project:
FABRICATION CONSTRAINTS • The use of 3D printing enables the accurate fab-
While generic digital fabrication—including some rication of precise topology optimization details,
3D-printing technologies—will require significant design while concrete provides the structural strength
alterations to permit the fabrication of large-scale topo- necessary for large-scale components.
logically optimized geometries, this paper shows that • Empirical observation suggests that UHP-
PSPF, a hybrid fabrication process, requires only mini- FRC has an isotropic behavior which is easily
mal changes to the topological optimum. modelled digitally for topology optimization
In general, PSPF for concrete is a fabrication method algorithms; when using fiber reinforcement,
which is not as permissive as pure binder-jet 3D-printing anisotropic behavior and fabrication constraints
fabrication. Nevertheless, compared to the other fabri- resulting from reinforcement bars do not have to
cation methods discussed in the introduction, PSPF is be considered.
very generous in terms of its fabrication constraints. The • Topology optimization and 3D printing both have

103 2016 TxA EMERGING DESIGN + TECHNOLOGY 104

Figure 9: Topology
optimization of Proto-
type “B” with different
numbers of nodes.

Figure 10: Rheology

studies investigating
the relation between
fiber reinforcement
and geometric
features.

potential applications in the realm of one-of-a- performance for specific material reduction
kind, non-standard building components rather targets. Nevertheless, optimization criteria are
than in mass-production. not limited to structural performance. Acoustic
performance or heat transfer, as well as any
In applications of PSPF for larger building components, combination of two or more criteria, can con-
such as entire concrete slabs, structures would need stitute optimization targets. This opens up the
to be assembled from multiple prefabricated parts. In possibility of integrative optimization strategies
order to achieve this, further research must address the for the design of smart building components.
following challenges:
• Reinforcement considerations. Steel-fiber rein- Concrete is one of the most consumed products in
forcement was sufficient for the prototypes, the world, with 10 billion tonnes being produced every Figure 11: Post-
year. Optimizing the use of concrete in prefabricated processing of the
but in order to increase the structural spanning
3D-printed formwork.
capabilities, traditional reinforcement bars or components can have a global impact in reducing
pre-stressing strategies are considered. Again material costs and the carbon footprint of buildings
demonstrating its suitability, 3D printing can and infrastructure. This research draws attention
be used to fabricate guiding features for the to this major potential and proposes a fabrication
precise integration of reinforcement. Topology method based on additive processes which is viable
optimization strategies will have to account for at a large scale. A harmonious compatibility exists
the anisotropy introduced by the direction of the between this additive fabrication process and topol-
reinforcement. ogy optimization used for form-finding purposes. The
• Additional functionality. This paper highlights the authors regard this research as the first steps toward
significant potential of using 3D printing to fab- a new, fully integrated approach to construction driven
ricate large-scale parts with optimal structural by material economy.

105 2016 TxA EMERGING DESIGN + TECHNOLOGY 106

 Søndergaard, Asbjørn, and Per Dombernowsky. 2011.
“Unikabeton Prototype.” In Fabricate: Making Digital
Architecture, edited by Ruairi Glynn and Bob Sheil.
Cambridge, Ontario: Riverside Architectural Press.

Stutz, Felix and Neil Montague de Taisne. 2016. “3D

Sand-Printed High Performance Fibre-Reinforced
Concrete Hybrid Structures.” ETH Civil Engineering
thesis supervised by Nicolas Ruffray and Mathias
Bernhard. Zurich.

Tomás, Antonio and Pascual Martí. 2010. “Shape and

size optimisation of concrete shells.” Engineering
Structures 32 (6): 1650–58.

ACKNOWLEDGMENTS

The authors would like to thank a number of partners

and collaborators whose dedication helped to realize the
projects described in this paper:
• Prof. Dr. Robert J. Flatt, Nicolas Ruffray and Dr.
Timothy Wangler (Physical Chemistry of Build-
ing Materials, ETH Zurich)
• Heinz Richner and Andi Reusser (Concrete Lab,
ETH Zurich)
• Felix Stutz and Neil Montague de Taisne (Bach-
elor Thesis, Engineering, ETH Zurich)
• Hyunchul Kwon, Victoria Fard, Nicholas Hoban,
Michael Thoma, and Philippe Steiner (Proto-
Figure 12: Casting REFERENCES Feringa, Jelle, and Asbjørn Søndergaard. 2012. “Design types and Documentation
UHPFRC inside the
and Fabrication of Topologically Optimised Structures; • Christen Guss AG (Production Partner)
3D-printed formwork.
Allwood, Julian M., Michael F. Ashby, Timothy G. An Integral Approach – A Close Coupling Form Gener-
Gutowski, and Ernst Worrell. 2011. “Material effi- ation and Fabrication.” In Digital Physicality - Proceed- This research was supported by the NCCR Digital
ciency: A white paper.” Resources, Conservation and ings of the 30th eCAADe Conference - Volume 2, edited Fabrication, funded by the Swiss National Science
Recycling 55 (3): 362–81. by Henri Achten, Jiri Pavlicek, Jaroslav Hulin, and Dana Foundation (NCCR Digital Fabrication Agreement
Matejovska, 495–500. #51NF40-141853).
Aremu, Adedeji, Ian Ashcroft, Richard J.M. Hague,
Ricky Wildman, and Christopher John Tuck. 2010. Liang, Qing Quan, Yi Min Xie, and Grant P. Steven. 2000.
“Suitability of SIMP and BESO topology optimisation “Optimal Topology Design of Bracing Systems for
algorithms for additive manufacture.” In Twenty-First Multistory Steel Frames.” Journal of Structural Engi-
Annual International Solid Freeform Fabrication Sym- neering 126 (7): 823–29.
posium, Austin, 679–92.
Michalatos, Panagiotis and Andrew Payne, Millipede
Beghini, Lauren L., Alessandro Beghini, Neil Katz, William 23.03.2014. Windows. [Link], 2014.
F. Baker, and Glaucio H. Paulino. 2014. “Connecting
architecture and engineering through structural Mijar, Anand R., Colby C. Swan, Jasbir S. Arora, and Iku
topology optimisation.” Engineering Structures 59: Kosaka. 1998. “Continuum topology optimisation for
716–26. concept design of frame bracing systems.” Journal of
Structural Engineering 124 (5): 541–50.
Dillenburger, Benjamin, and Michael Hansmeyer. 2013.
“The resolution of architecture in the digital age.” Rozvany, George I.N. 2009. “A critical review of estab-
In International Conference on Computer-Aided lished methods of structural topology optimisation.”
Architectural Design Futures: Global Design and Structural and Multidisciplinary Optimisation 37 (3):
Local Materialization, edited by Jianlong Zhang and 217–37.
Chengyu Sun, 347–57. Berlin: Springer.

107 2016 TxA EMERGING DESIGN + TECHNOLOGY 108