Large-scale 3D printing promises architecture the seductive ideal of the seamless, continuous curve — a building that flows from digital model to physical form without interruption. But the robots that print these structures live inside hard physical limits. Six-axis robotic arms can only reach so far, print beds only measure so wide, and composite materials must be deposited and cured in sequential stages. The result is that every seemingly continuous geometry must be divided into segments, and every division creates a seam. A new research article published in Architectural Intelligence on 1 June 2026 confronts that seam head-on, treating the joint not as a flaw to hide but as a tested, engineered architectural detail in its own right.
The study, led by Alan ‘Ho Kyeong’ Kim with Stuart Bateman, Phillip Crothers and Roland Snooks of the RMIT Architecture | Tectonic Formation Lab, presents an iterative, research-through-design methodology for advancing joint design in multi-material additively manufactured fibre composite systems. The team developed seven joint iterations across two experimental stages, combining robotic additive manufacturing of polymer shells with carbon fibre reinforcement through repeated cycles of prototyping, mechanical testing and full-scale application. Their central finding is stark: joint performance is governed by the interaction of reinforcement continuity, fastener integration and interface control — and above all, by whether the internal reinforcement path remains unbroken.
The first experimental stage produced three prototypes built on PETG polymer substrates. The large-scale surfaces were fabricated using robotic Fused Granulate Fabrication with a KUKA KR-150 robot and a custom extruder, printing SK Chemicals Skygreen S2008 PETG pellets through a 3.00 mm nozzle at 2.20 mm layer height and 220 °C. Internal reinforcement cores were printed on a desktop Fused Filament Fabrication printer at finer tolerances, and two fibre types — E-glass and carbon fibre, both 200 g/m² plain woven fabric — were integrated through manual wet lay-up and vacuum bagging with a West System epoxy, cured at 20 °C under −100 kPa gauge vacuum for 24 hours. Where mechanical fastening was needed, M4 steel clinch floating nuts were embedded in the spine component and bonded with Araldite 420 A/B aerospace-grade epoxy, post-cured at 50 °C for five hours.
Iteration 01 served as a baseline: a PETG spine with carbon fibre reinforcement integrated within the joint interface, assembled with M4 bolts and nuts but no external plate, leaving the seam visually continuous and translucent through the printed polymer. Iteration 02 substituted three continuous E-glass fibre sheets within the spine and masked the seam behind an external plate. Iteration 03 replaced the fibreglass with carbon fibre to maximise strength and stiffness, again concealing the seam beneath a carbon fibre plate. It was Iteration 03 — the carbon fibre spine and plate configuration — that was selected for full-scale implementation in the project’s architectural demonstrator, named Ghost Tectonics.
Ghost Tectonics is a large-scale architectural demonstrator developed at RMIT Architecture | Tectonic Formation Lab in partnership with Boeing Aerostructures Australia, Hightex and the Leibniz-Institut für Polymerforschung Dresden. The project merges robotic large-format 3D printing with carbon fibre Tailored Fibre Placement (TFP), deploying 150 k tow carbon fibre strands directly onto the substrate surface along the main structural stress paths, tying discrete joint sites into a continuous reinforcement field. Spine geometry was parametrically varied across the assembly — thickened at structurally critical intersections to optimise load transfer, reduced at apertured or tapering regions to ease assembly and fit. Externally, custom plate-based detailing covered the seam regions, adjusted according to local stress analysis, producing a diamond-like articulation that both reinforced the joint and rendered the logic of connection legible on the building’s envelope.
Crucially, the full-scale demonstrator changed the direction of the research. During fabrication and assembly, three problems became clear. First, placing carbon fibre spines within curved and irregular geometries demanded far more precision than the specimen tests had suggested; small misalignments degraded bonding quality and slowed assembly. Second, vacuum curing and post-processing introduced small deformations that affected the fit between printed components. Third, aligning three joint elements simultaneously — internal spine, printed substrate and external plate — proved difficult, because each carried separate tolerance demands. These observations redirected the research question from which material achieves the highest flexural strength toward how a joint must also manage tolerance, sequence and assembly to remain buildable at architectural scale.
A second experimental stage followed, producing Iterations 04 through 07, each varying a single parameter to clarify its contribution. Iteration 04 tested a plate-only configuration with self-tapping screws; the screws pulled out of the printed material under load, damaging the substrate and preventing reassembly — a decisive drawback for demountable or relocatable structures. Iteration 05 split the spine into two segments separated by a 10 mm gap, bridged externally by a carbon fibre plate; it produced the lowest flexural strength of the entire series at just 1.90 MPa, confirming that an uninterrupted internal reinforcement path is essential for load transfer. Iteration 06 returned to the plate-only strategy but with bonded M4 clinch floating nuts, reaching 28.93 MPa — nearly identical to Iteration 04’s 29.21 MPa, showing that better fasteners cannot compensate for a missing spine. Iteration 07 removed the external plate and overlapped the printed substrate instead, retaining the internal carbon fibre spine.
The mechanical evaluation used the ASTM D790 standard three-point bending method on specimens 48 mm wide and 500 mm long, with a 400 mm support span and three to four replicate specimens per iteration. The specimens were deliberately oriented to load the interlayer interfaces — the build-normal, z-direction — probing the weakest axis of the printed material. The results were emphatic. Iteration 03 achieved the highest average flexural strength at 38.17 MPa, with Iteration 01 close behind at 36.67 MPa, and both sustained mean peak forces above 2,000 N — Iteration 03 reaching a mean of 2,686 N. Configurations that omitted the spine entirely or interrupted its continuity recorded mean peak forces below 166 N, regardless of whether an external plate was present.
Yet the numbers demand careful reading. The difference between the spine-only and spine-plus-plate configurations was roughly 4 per cent — but with standard deviations of ±6.38 and ±3.46 MPa respectively, the specimen ranges overlap, so the small increase cannot be confidently attributed to the plate alone. The internal spine appears to account for most of the flexural capacity. The plate-only configuration, at 28.93 MPa, sat approximately 21 per cent below the spine-only joint, indicating that the external plate functions as secondary reinforcement rather than the primary structural element. Iteration 07, though not the strongest, accumulated load progressively across an extended displacement range with small stepped force reductions followed by recovery, suggesting sequential damage at individual reinforcement interfaces and relatively high energy absorption. The authors also note that the span-to-thickness ratio of roughly 14.5:1 fell slightly below the ASTM-recommended 16:1, a logistical compromise that may have increased shear effects but does not affect comparative validity among specimens.
The authors are candid about the limitations. The iterations formed a directed design progression rather than a fully controlled experiment, so the contribution of any single variable cannot be attributed with full statistical rigour. The coupon-scale specimens, tested in three-point bending with three to four replicates per iteration, do not replicate the multi-axial stress states, curvature or distributed loads of real architectural application — a gap confirmed by the Ghost Tectonics evaluation. Future work will develop intermediate-scale specimens approximating demonstrator curvature and loading, and will systematically compare external plate systems in place of internal spines. What the study establishes, within its scope, is a quantified hierarchy: continuity of internal reinforcement dominates joint performance, fastening and interface geometry matter, and the joint itself can be designed as a site where structural behaviour and tectonic expression are developed through the same assembly logic — neither purely revealed nor concealed, but distributed across the material divisions that fabrication inevitably imposes.
Subject of Research: Joint design and mechanical testing for multi-material additively manufactured fibre composite structures in architecture
Article Title: Composite amalgam: development of joining approaches for additively manufactured fibre composites in complex geometry in architecture
Article References: Kim, A. H. K., Bateman, S., Crothers, P., & Snooks, R. (2026). Composite amalgam: development of joining approaches for additively manufactured fibre composites in complex geometry in architecture. Architectural Intelligence, 5(1), Article 12. https://doi.org/10.1007/s44223-026-00119-5
Image Credits: AI Generated
DOI: 10.1007/s44223-026-00119-5
Keywords: additive manufacturing, carbon fibre, composite joints, 3D printing, architecture, robotic fabrication, flexural testing, PETG, Tailored Fibre Placement, tectonics, Ghost Tectonics, research-through-design
News Source: Denise Maddox. (October 29, 2026). Carbon fibre spine joints tested for 3D-printed architectural composites. Scienmag.
