Construction Robotics (2020) 4:217–226

https://doi.org/10.1007/s41693-020-00047-4

ORIGINAL PAPER

Skeletal composites: robotic fabrication processes for lightweight

multi‑nodal structural components
Marshall Prado1

Received: 12 June 2020 / Accepted: 29 October 2020 / Published online: 24 November 2020
© Springer Nature Switzerland AG 2020

Abstract
The research presented in this chapter describes the novel robotic fabrication strategies for multi-nodal structural components
made from lightweight fiber composite materials. The paper contextualizes the research within a larger area of composite
manufacturing in architecture and focuses on the developed methodologies for adaptive, material-efficient production. This
research builds on coreless filament winding processes that eliminate the need for large surface molds and wasted materials for
composite production. This process allows for large geometrically differentiated structural building components to be easily
produced adaptively for architectural applications. The research tests the production of complex components for a vertical
lattice structural system. The multi-nodal structural components enable continuous material and fiber orientations across the
intersections of the lattice while simplifying connections. Key improvements presented in this paper included the robotic
assembly process for the reconfigurable winding frames that reduce assembly times and increase accuracy, computational
techniques for developing winding syntax, and physical simulation of material orientations for robotic path planning. This is
followed by a conclusion and outlook to discuss the tested results on a full-scale demonstrator and the future design potentials.

Keywords Robotic fabrication · Lightweight · Fiber composites · Coreless filament winding

1 Introduction form the material. This effectively streamlines composite

fabrication for serialized production of identical parts while
Fiber composite materials have many advantages for archi- problematizing its use in architectural applications. Serial-
tectural production. Composites are commonly used in many ized or mass-produced architecture has repeatedly failed to
large-scale manufacturing industries such as automotive, aer- become a major mode of building design and construction
ospace, and marine (Barynin et al. 1999; Adam 1997; Davies without severely restricting design freedom or the ability to
and Chauchot 1999). These lightweight, high-performance adapt to the changing demands of the architectural industry
materials are easily formable and can provide a variety of (Knapp 2013). Most buildings are one-off constructions so
material behaviors through control of fiber arrangements more adaptive means of production are necessary to effec-
and orientations. They enable a higher degree of material tively utilize composites for architectural applications.
efficiency, weight reduction, corrosion resistance, and struc- Filament winding offers some potential benefits for archi-
tural capacity than standard building materials. This makes tectural production. It is the most efficient and cost-effective
them an important material solution for many performance- form of composite fabrication (Peters 2011). As an additive
based architectural applications (Kreysler 2017). Despite manufacturing process, it creates very little waste while pro-
the technological advancements that have been made in viding complete control over fiber orientation, alignment,
composite manufacturing, including many robotic fabrica- and density. Filament winding is commonly used for sim-
tion strategies, most of these processes still rely on costly ple symmetrical and straight pressure vessels but has also
formwork, surface molds, or winding mandrels to shape or been used for more complex asymmetric geometries such
as T-shaped fittings or wind turbine blades. The size of the
component geometry you can make with filament winding
* Marshall Prado processes is often limited by the size of the machine but
[email protected] large-scale production of aircraft fuselages shows the pos-
1 sibility of scaling up to the size of architecture (Hiken 2017).
University of Tennessee, Knoxville, TN, USA

13
Vol.:(0123456789)
218 Construction Robotics (2020) 4:217–226

Recent developments in coreless filament winding, a process Nobel 1986). Cacti, such as the Cholla Cylindropuntia sp.
requiring no surface molds or mandrels to fabricate compos- or Saguaro Carnegiea gigantea, have fibrous skeletons that
ite structures, have opened up the possibility for more adap- form tube-like lattice structures (Fig. 2). The tallest of these
tive manufacturing (Minsch et al. 2019; Prado et al. 2014). have been known to reach almost 80 ft in height. These
The goal of the research presented here is to develop alterna- skeletons are perhaps analogous to lightweight perimeter
tive strategies for coreless filament winding that explore new frames that form the structures of many of our modern-day
design and structural opportunities (Fig. 1). skyscrapers. Understanding the abstract biological princi-
ples from the morphology of the cactus structure could,
therefore, be transferred into new ways of designing and
2 Context manufacturing skeletal composite structures for architectural
applications.
In biology, there are many examples of natural fiber systems Lightweight frame structures have been explored exten-
that have a high degree of material and structural efficiency sively in architecture and engineering applications. Popular-
(Chen and Wu 2013; Bobich and North 2009; Gibson and ized by Buckminster Fuller, these structural strategies can

Fig. 1 Photo of the demonstra-

tor

13
Construction Robotics (2020) 4:217–226 219

reduce the material required for a structural system (Cheung these earlier prototypes explore adaptive structural com-
and Gershenfeld 2013). Often these frame structures are posite components (Magna et al. 2014), modular compos-
constructed from a series of self-similar struts connected ite fabrication (Knippers et al. 2016), and hybrid compos-
at custom hubs. The struts are commonly simple metal or ite structures (FabUnion Homepage 2020), showing the
wood members made from industrialized stock material. potential for coreless filament winding and discretized
These hubs are usually more complex, imbedded with the structures. At the fiber scale these projects, which are
logics of assembly or the geometric configuration of the freed from the constraints of surface molds, or the use
system. These hubs are generally the weakest parts of the of standardized stock material, form open lattice meshes
system since forces accumulate and are transferred across with complex geometries. This reduces the amount of
the hub, the struts are typically discontinuous at the hub material required to make the structure by only placing
and the joints, which are frequently non-planar connec- it where it is needed; however, the systems employed
tions, may introduce unwanted bending stresses for some in these projects still form structured surfaces. Further
configurations. reduction of materials may be possible within a skeletal
Unlike their architectural counterpart, which might structural system. Recent projects such as the BUGA
have hundreds or thousands of discontinuous joints, the Fiber Pavilion 2019 by the University of Stuttgart (Buga
skeletal structure of the Cholla cactus is formed by a lat- Fiber Pavilion project 2020) and the Urban Micro Climate
tice system of vascular bundles that has continuous mate- Canopy (UMCC) by TU Munich (Urban Micro Climate
rial across its intersections (Bobich and North 2009; Gib- Canopy project 2020) utilize a skeletal structure instead
son and Nobel 1986). Though the use of fiber composites of structured surfaces. The UMCC takes advantage of
does enable the ability to make more material continuity in multi-nodal components though the system developed uti-
lightweight construction, at the scale of architecture, this lizes identical components and standardized fabrication.
goal is unfortunately still limited. However, alternative The BUGA Fiber Pavilion 2019 begins to explore differ-
strategies for discretizing a lattice system into a series of entiated component fabrication but functions similarly to
multi-nodal components and novel fabrication strategies connected struts with discontinuous intersections rather
for manufacturing them may solve many inherent issues than a multi-nodal system. These projects are novel in
with common lightweight frames structures including their application and development but do not fully address
standardized planar connections, fewer fabricated compo- the potential of skeletal composites structures introduced
nents, and increased material continuity across the lattice through this research. The following sections highlight
intersections (Fig. 3). the novel developments for the design and robotic fab-
Component-based composite construction has been rication of a lightweight fiber composite demonstrator,
used in recent years to develop several research-based the UTK Filament Tower that take advantage of adaptive
demonstrators for architectural production. These projects fabrication systems, multi-nodal component geometries,
have been produced mainly at the University of Stuttgart and lightweight lattice structures.
in Germany and Tongji University in China. Several of

Fig. 2 Fiber orientation and

continuity across lattice
intersections in Cholla cactus
skeleton

13
220 Construction Robotics (2020) 4:217–226

Fig. 3 Diagram of lattice discretization strategies showing strut and hub vs. multi-nodal

3 Methods axes and frame components were produced to alternate the

winding and post-curing processes.
3.1 Fabrication setup
3.2 Reconfigurable frame
The test demonstrator was fabricated in the University of
Tennessee fabrication lab. A nine-axis robotic setup was The winding frames are setup using the assembly effector
used that included a large industrial robotic arm (Kuka and standard off-the-shelf reusable parts (Fig. 5). Each node
KR150 R3100), an 8-m linear rail (Kuka KL1500), and of the component is assembled separately. There are two
a two-axis positioner (Kuka DKP 400). Attached to the node sizes for the main structural axis and the secondary
two-axis positioner was a 4-in.-diameter × 10-ft-long steel connections. Large nodes have 15 winding points while the
pipe, which served as a horizontal rotating support axis for smaller nodes have 12. Each node is held in place by the
fabrication (Fig. 4a, b). This setup roughly approximates assembly effector through a preprogrammed movement. For
industrial filament winding machines that have a rotating the larger nodes, three steel plates are attached to the assem-
horizontal axis and a custom articulate sled to place fibers on bly effector. A custom milled wooden guide is slid onto the
the mandrel. Instead of utilizing a surface mandrel, a custom central axis, aligned with the assembly effector, and attached
reconfigurable frame was developed that could be adjusted to the steel plates. A large shaft collar is attached to the
to various component sizes and configurations. Two custom wooden guide and locked on to the central axis. This secures
robotic effectors were developed for assembling reconfigur- the node into the correct position and ensures that the wind-
able winding frames and then subsequently coreless filament ing geometry is accurate. For the smaller side nodes, the pro-
winding. An adjustable pneumatic tension device and cus- cess is almost identical except for an added steel extension
tom-fabricated resin bath was used to impregnate between arms. Various standard length extension arms are used to
1 and 4 glass rovings (tex 2400 each) or 1–6 carbon fiber enable a wide variety of node configurations. These arms are
rovings (50 k filaments each) for a wet winding process. A attached perpendicular to the central axis using a pipe clamp
temporary temper chamber with a 128 ft3 capacity was also assembly. A smaller shaft collar is used to attach the angled
created to post-cure composite components. Two complete wood guides to the extension arms. Three and four-node
components were tested though more node configurations

13
Construction Robotics (2020) 4:217–226 221

Fig. 4 a Robotic fabrication setup with a horizontal winding axis, reconfigurable frame, and winding effector. b Detail of fabrication

are possible. Once the node components are installed and the point and later protect the composite material from the abra-
assembly effector is removed the winding points are attached sion of mechanical fasteners while assembled. A 1″-diameter
to each node. The winding points consist of ½″-diameter x washer caps the end of each winding point, which provides a
¾″-long aluminum spacers, which are embedded in the com- flat smooth surface through which the finished components
posite. These set how much material can be wound on each can accurately connect.

Fig. 5 Diagram of reconfigurable frame and assembly process

13
222 Construction Robotics (2020) 4:217–226

3.3 Winding syntax the motion of the robot nor the orientation of the fiber on
the composite surface. A post-processing step was needed
Multi-nodal winding frames presented new challenges in to create the robotic path planning and to visualize accurate
developing a winding syntax. In precedent research, winding fiber orientations. In previous coreless winding examples, the
frames had at most two nodes or simplified topologies. Wind- component geometries were limited to hyperboloidal surfaces.
ing syntaxes consisted of a sequence of indices that denoted Exact control over fiber placement was unnecessary as fibers
the order of winding or a continuous polyline that represented under tension would find the shortest path or geodesic curve
the winding sequence (Prado et al. 2014). Fiber directions to follow. This path only slightly deviated from a simplified
were determined by the shortest path between the two polyline robotic path. In some instances, however, if the deviation of
points. Multi-nodal winding provides a wider variety of pos- the fiber path and the robotic path was too great, the filament
sible winding scenarios given the same starting and ending would drag across the surface potentially damaging the com-
points. Nodes that are not being connected by fibers can still posite layers below. This became increasingly more evident
be used to control the direction of the fibers on the compos- as the multi-nodal surfaces became more complex, harder to
ite surface. This requires a much more complex system of predict and a larger variety of winding syntaxes were possible.
instructions for determining unambiguous fiber directions. In-
depth physical testing was utilized at full scale to characterize 3.4 Robotic path planning
component typology and variation of winding syntax.
The overall form of the demonstrator as well as individual A new method for robotic path planning became necessary
component geometries were structurally analyzed in three that could more closely articulate the fiber orientations and
stages to determine the material layout. A centreline model winding syntaxes. To generate an accurate motion sequence
was first tested with a single beam representing each tube for the robot required more accurate simulations of com-
section of the structure. This was used to predict approximate ponent geometries and fiber directionalities on the surface.
material amounts, general component diameter, and overall Previous simulation models for component geometries were
force distribution. With these key geometry decisions made, either computationally expensive and, therefore, could only
a surface model of the structure was generated with the pro- be used as a final verification of structure or quick and inac-
posed surfaces represented as a mesh of the approximately curate representations that could be used for visual design
correct thickness (the fibrous structure has varying thickness verification (Waimer et al. 2013). Neither technique could be
across its form). This model was used to determine buckling used effectively for robotic path planning on the multi-nodal
risks and overall stress trajectories within the surfaces to composite geometries. A new physical simulation model was
inform fiber density and orientation. Finally, a high-resolu- developed that considered the winding sequence and hetero-
tion beam model was made with each key fiber bundle mod- geneous tension direction of the composite geometries. This
eled to check for the performance of the proposed structure in model was adjusted to more accurately and quickly simulate
detail. Non-destructive compressive and buckling tests were the composite forms being created.
conducted to verify the computational analysis. By more accurately simulating where the fibers will be
The initial syntax development and the structural analy- laid on the composite surface, robotic fiber laying paths
sis were used to formulate a computational design tool to could be generated that eliminated excessive dragging of
analyze and generate custom winding codes for each com- material across the surface. Several simulation techniques
ponent. The winding code, which could consist of hundreds were tested including shortest path algorithms, live phys-
of connected points, was broken down into individual wind- ics simulations, and toy car simulations. These simulations
ing paths segments for robotic programming. Each wind- proved to be computationally expensive and time-consuming
ing path includes the initial winding point procedure and for components with several hundred winding paths. A light-
travel instructions to the next node location. A series of four weight easily controllable simulation was required that could
positional references were determined at quadrants around simultaneously generate fiber directions for multiply wind-
each node. The travel instructions use these reference points ing paths. To do this, the initial topological polyline repre-
to navigate around nodes (Fig. 6b). The order of reference senting the travel instructions was subdivided into 50-mm-
points in the travel instructions determined the rough fiber long segments. A nearest-neighbor smoothing algorithm was
topology for each path. Any two winding points could have then iteratively applied to the polyline vertices. While the
a variety of potential travel instructions enabling a dense vertices reorganized, if a resulting point moved inside the
fiber composite mesh and an array of fiber orientation to be simulated composite surface it would be replaced with the
formed with fewer required winding points. closest point on the composite surface. The smoothing itera-
The travel instructions can be represented by a topologi- tions would stop once the average deviation from the com-
cal polyline curve to visualize the winding syntax for each posite surface was less than 2 mm. This resulted in a very
fiber path but this polyline is neither accurate in representing quick simulation capable of articulating hundreds of winding

13
Construction Robotics (2020) 4:217–226 223

Fig. 6 a Component, b motion sequence, c simulated fiber path, d safe travel boundary, e tangent lines to the simulated path, f robotic path

paths within a few seconds. The resulting smoothed polyline 4 Results and conclusion
also accurately represented the fiber path along the surface
under tension between two winding points given specific The UTK Filament Tower is a full-scale demonstrator of
travel instructions (Fig. 6c). a novel multi-nodal coreless filament winding process. It
Once the fiber path was simulated a more accurate robotic stretches to a height of 30 ft making it one of the tallest
path could be determined that minimized dragging and
fiber damage. A custom revolved form was generated for
each component using the node centers and the central axis
of rotation. This volume was used as a robotic safe travel
boundary for each component (Fig. 6d). A robotic path along
this surface would be free from any potential collisions. The
simulated fiber path was divided into points. A tangent line
from each point along the curve was intersected with the
safe travel boundary surface. The resulting collection of
points from the tangent line intersections were connected
into a new polyline, smoothed, and simplified to generate
the robotic fiber-laying path (Fig. 6e). This path proved to
be more accurate as fibers were placed along their simu-
lated fiber path without dragging or damaging the composite
material. By uncoiling the robotic path along the component
axis, the rotational data could be determined for the two-axis
positioner. The programmed motion of the robot could then
be simplified to follow a planar, collision-free curve along
Fig. 7 Detail of interior view through center of the tower
the safe travel boundary surface (Fig. 4a).

13
224 Construction Robotics (2020) 4:217–226

Fig. 8 Photo of demonstrator

installed onsite

structures of its kind. It is open at the top forming a tube- 2.5 and 5 h per component and the frames were typically
like lattice structure that covers an area of 85 sqft. The tower assembled in under an hour.
structure consists of six interconnected columns (Fig. 7). Using a reconfigurable frame can often lead to trade-offs
These columns are made from 27 highly differentiated fiber of accuracy for adaptability and variation but the developed
composite components. Each component weighs between frame proved to be very accurate and adaptable. The use of
6 and 25 lbs and range from 3 to 7 ft long. The composite readily available off-the-shelf parts provided a surprisingly
structure weighs roughly 340 lbs in total and consists of over low-tech solution to the challenge of making unique, compos-
50 miles of glass and carbon fiber. During fabrication, one ite components without the need for individualized surface
component per day was manufactured, post-cured for 1.5 h molds. The potential for human error was minimized through
at 90 °C, and removed from the winding frame. The post- the use of the robotic assemble process. This increased accu-
curing process additionally strengthens the components by racy also significantly reduced assembly problems.
about 30%. Winding times were also reduced compared to The tower was installed onsite in Columbus, IN, with
the previous modes of fabrication by about 40% to between very few unforeseen problems (Fig. 8). Because of the

13
Construction Robotics (2020) 4:217–226 225

Fig. 9 Photo of composite

structure in cold conditions

desire to have all connections hidden in the final dem- tower was installed in less than a week. The composite
onstrator, the mechanical fasteners were only accessible pieces were lit internally with LED lighting that was pre-
through open nodes or openings in the mesh composite installed and connected during the installation. The fiber
of each piece. This was not always convenient but it was composite components were supported by three large-scale
possible given the openness of the mesh. This may not 3D-printed bases made of carbon fiber-reinforced ABS
be a robust solution for future iterations that have higher plastic that are 4 ft tall and around 8 ft in length. These
structural requirements and denser composite meshes. bases serve as ballast and foundations for the rest of the
Future refinement of design geometry could make it easier tower as well as seating for the users. Each one weighs
if secondary nodes fall within arm length of each other. approximately 250 lbs and collectively weighs more than
This would enable more access to internal connections all the fiber-wound composite components combined.
and attaching mechanical fasteners. Testing of embed- Additionally, each base had four earth screws anchoring it
ded threaded connections, which have higher precision to the ground and connected to the lower composite com-
requirements but can be blindly connected from one side, ponents through tension rods. Though each component has
would also simplify the assembly of components. The hid- a high volume to mass ratio, they were able to nest effec-
den connections did not ultimately affect assembly as the tively for transportation. All the composite components

13
226 Construction Robotics (2020) 4:217–226

and 3D-printed basses fit into a standard 15-ft box truck. References
The demonstrator was on exhibition between August and
December. During this time, it was exposed to high heat, Adam H (1997) Carbon fibre in automotive applications. Mater Des
cold, sun, rain, wind, and snow (Fig. 9). The coreless 18(4–6):349–355
Barynin VA, Bunakov VA, Rasin AF, Vasiliev VV (1999) Aerospace
wound components showed no visible structural damage composite lattice structures. In: Proc. of the 120th int. conf. on
from the elements although the carbon fiber-reinforced composite materials, Paris, France
ABS bases had begun to fade from UV exposure. Bobich EG, North GB (2009) Structural implications of succulence:
The testing of these novel fabrication processes through architecture, anatomy, and mechanics of photosynthetic stem
succulents, pachycauls, and leaf succulents. In: Perspectives in
the production of the demonstrator has shown that highly biophysical plant ecophysiology, A Tribute to Park S. Nobel,
differentiated and structurally efficient multi-nodal fiber pp 3–38
composite components can be produced effectively. The Buga Fiber Pavilion project website (2020) https://www.icd.uni-stutt
refinement in the reconfigurable frames and assembly pro- gart.de/projects/buga-fiber-pavilion/. Accessed 16 Feb 2020
Chen J, Wu G (2013) Beetle forewings: epitome of the optimal
cess produced a higher level of precision than was achieved design for lightweight composite materials. Carbohydr Polym
on precedent research with reconfigurable frames. The 91(2):659–665
assembly process showed that the fiber composite compo- Cheung KC, Gershenfeld N (2013) Reversibly assembled cellular com-
nents could interface with several other building and mate- posite materials. Science 341(6151):1219–1221
Davies P, Chauchot P (1999) Composites for marine applications.
rial assemblies effectively as well. The multi-nodal system Mechanics of composite materials and structures. Springer,
has shown tremendous potential for not only tower scenarios Dordrecht, pp 249–260
but also full composite load-bearing systems. Future investi- FabUnion Homepage (2020) https://www.fab-union.com. Accessed
gations will explore additional structural scenarios, embed- 16 Feb 2020
Gibson AC, Nobel PS (1986) The cactus primer. Harvard University
ded connectors, further integration of composite winding Press, Cambridge
with additive manufacturing processes, architectural poten- Hiken A (2017) The evolution of the composite fuselage-a manufac-
tials of multi-nodal fabrication, additional multi-node topol- turing perspective. SAE Int J Aerosp 10:77–91
ogies, and scaling up the production process further. Knapp C (2013) “The End of Prefabrication,” Australian design
review. https://www.australiandesignreview.com/architecture/
the-end-of-prefabrication/. Accessed 16 Sept 2020
Acknowledgements The author would like to acknowledge the help Knippers J et al (2016) Modular coreless filament winding for light-
and support from the many research assistants and students that con- weight systems in architecture. In: Proceedings of the 8th inter-
tributed to the design and production of the research demonstrator. national conference on FRP composites in civil engineering,
Development, Fabrication and Construction: Shane Principe, Sarah CICE
Wheeler, Courtney St. John, Alex Stiles, Nadin Jabri, Geng Liu, Pete Kreysler W (2017) Qualifying FRP composites for high-rise building
Paueksakon, Tyler Sanford, Michaela Stanfill, Michael Mckever, facades. In: Menges A et al (eds) Fabricate 2017. UCL Press,
Michael Swartz, Hollywood Conrad, Teig Dryden, Howard Fugitt, London, pp 130–137
Kristia Bravo, Bridget Ash, Kevin Saslawsky, Michael Vineyard, Zane La Magna R, Waimer F, Knippers J (2014) Coreless winding - a novel
Smith, Josh Mangers, Patrick Dobronski, Joe Gauspohl and with the fabrication approach for FRP based components. In: Building con-
support of Craig Gillam and the UTK Fablab. The author would also struction. Proceedings of the 7th international conference on FRP
like to acknowledge the research collaborators that enabled the concep- composites in civil engineering, CICE 2014, Vancouver, Canada
tual development, engineering and production of the research demon- Minsch N, Müller M, Gereke T, Nocke A, Cherif C (2019) 3D truss
strator. In collaboration with Oak Ridge National Laboratory—Manu- structures with coreless 3D filament winding technology. J Com-
facturing Demonstration Facility Format Engineers Ltd., Fiber and pos Mater 53(15):2077–2089
Composite Manufacturing Facility, Entomology and Plant Pathology. Peters ST (ed) (2011) Composite filament winding. ASM International,
Cleveland
Author contributions Marshall Prado contributed to the study con- Prado M et al (2014) Core-less filament winding. Robotic fabrication
ception and design, material preparation, data collection and analy- in architecture, art and design 2014. Springer, Cham, pp 275–289
sis. Marshall Prado also wrote the first draft of the manuscript and Urban Micro Climate Canopy project website (2020) https ://www.
final manuscript. Design development and production of the design ar.tum.de/en/gtla/research/exper iment al-buildings/urban-micro
demonstrator was lead by Marshall Prado and contributed to by the climate-canopy/. Accessed 16 Feb 2020
acknowledge research assistants listed above. Waimer F, La Magna R, Knippers J (2013) Integrative numerical tech-
niques for fibre reinforced polymers-forming process and analysis
Funding This study was funded in part by Exhibit Columbus in of differentiated anisotropy. In: Proceedings of IASS annual sym-
Columbus, IN and in part by the University of Tennessee, Knoxville. posia. International Association for Shell and Spatial Structures
The research included material sponsorship from Teijin Carbon, Owens (IASS), vol 2013, no 10
Corning, Techmer PM, McLube, SGL Carbon and Hutch and Kevina
Schumaker. Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Compliance with ethical standards

Conflict of interest The corresponding author states that there is no

conflict of interest.

13