CH NEO-ZÜRICH EDITION
WEATHER · CLEAR 22°C
BLEND OF THE DAY · 07/ROGUE
EST. 2027
THE AEC CYBER MORNING NEWS

PAZ Kaffi

DESIGN · DEMOLITION · CAFFEINE · DISPATCH
EDITION 0809 · 9 August 2026
BROADCAST 04:42 CET
2,400 BROADSHEETS PRINTED
READ TIME · 47 MIN
The Segmented Shell: How You Fake a Curve and Keep the Physics
BAU
FRAME · 06:50
09-08-2026

The Segmented Shell: How You Fake a Curve and Keep the Physics

How segmented shells keep membrane efficiency while deleting the mould — from Candela to BUGA's 376 timber segments to ETH's demountable concrete floors.

Start with the surface, not the story. A shell is a structure that spans by being curved — load travels through the material as in-plane membrane force, tension and compression running along the surface, with almost no bending. Bending is what a beam does, and bending is expensive: it needs depth, and depth is weight. Curvature is the trick that lets a few centimetres of concrete roof a room. The problem was never the physics. The problem was the mould.

The segmented shell is the answer to that bill. It keeps the membrane behaviour and throws away the continuous formwork by refusing to be smooth — approximating a double-curved ideal with a mosaic of flat or single-curved plates, each prefabricable on a flat table, each joined into a stiff continuous skin on site. This is a foundation move worth understanding from first principles, because the news you will read this decade about low-carbon floor slabs is standing on it.

←TODAY: A double-curved concrete roof still costs more in timber mould than in the shell it shapes; segmentation is how you delete that mould. →3012: Every span in the Zurich-3012 skyline unbolts into flat, re-milled plates — the building is a kit, not a casting. Fulcrum: The curve carries the load, but only the faceting makes the curve affordable — you need both directions in view or you build a beautiful object nobody can pay for or take apart.

What it is: A segmented shell is a bet that you can fake a curve well enough. The target is a surface where load is membrane force, not bending. Instead of one continuous double-curved sheet, you tile it with developable plates — developable meaning a surface you can unroll flat without stretching, like a cylinder or a cone but not a sphere. That single geometric property is the whole industrial argument: a developable panel lies flat on a CNC bed, ships flat on a truck, and springs into its place in the curved assembly without deforming its own material. You approximate a sphere you cannot flatten with pieces you can. The plates are the easy part. The structure lives in the seams.

Why it works: Here is the mechanism, and it is where most photogenic facet patterns quietly cheat. A smooth shell carries pure membrane force. The instant you replace the smooth surface with flat facets, every shared edge becomes a kink — a small dihedral angle between two plates that were supposed to be one continuous tangent. Membrane force arriving at that kink cannot simply continue in-plane; the fold deflects it, and the deflection shows up as a bending moment concentrated on the joint. As the PAZ concept panel on the engineering puts it bluntly: the plate is easy, the edge is the structure. This is why the research obsesses over joint-pattern layout and component differentiation rather than the panels themselves. Coarser segmentation — fewer, bigger plates — makes fabrication trivial and the joints furious; finer segmentation calms the joints and multiplies the parts. There is no free lunch, only a knee in the trade-off curve, and finding that knee is the design work. The deep reason the biological precedent matters is that a sand dollar — the flat sea urchin whose calcite test is a tessellation of plates locked by finger-jointed edges — had already solved plate discretization, graded component sizes, and joint patterning as one integrated problem. It did not optimise the plates and bolt on a connector afterwards. Neither should you.

Bending-active variants push the idea one turn further. Rather than accept flat plates, they actively bend single-curved developable strips into position and vary the bending stiffness along each element, so the target geometry emerges with no mould at all. The elastic energy you store in the bent element becomes part of the form-finding — the structure holds its shape because it is fighting to spring back. It is form-finding with the material as a spring, the same family of thinking as Frei Otto’s hanging models and the catenary, resolved in a laser-cut strip instead of a chain.

Origins: The membrane pedigree runs through the mid-century masters — Félix Candela, Heinz Isler, Eduardo Torroja — who proved a surface a few centimetres thick could span if you let it carry load in-plane. They also proved the mould could bankrupt you. The modern segmented answer matured at the University of Stuttgart, where the ICD and ITKE teams (Schwinn, Sonntag, Grün, Nebelsick, and the pairing of Jan Knippers and Achim Menges) read the concept straight out of the sand dollar. The ICD/ITKE Research Pavilion of 2015–16 was the proof-of-concept: a robotically fabricated segmented plywood shell derived directly from sea-urchin morphology — a marine skeleton read as a fabrication manual. The 2019 BUGA Wood Pavilion in Heilbronn scaled it from pavilion-toy to public roof: 376 hollow-cassette timber segments approximating a double-curved shell over the Federal Garden Show, with finger-jointed cassette edges doing the structural talking. Then the idea started moving downward and southward — to the 2026 Segmented Timber Shell at Alp in the Catalan Pyrenees, where MPDA master students at UPC Barcelona ran an antifunicular rib design of planar n-gon modules from form-finding to standing structure inside a single construction week. The important word is week.

In practice — Atelier: Atelier: In the PAZ Atelier we treat the segmented shell as a lesson in honest compromise, not a styling move. The temptation is to import the sand-dollar imagery and stop — a facet pattern glued onto a slab for the render. We push the other way and start from the seam: before a single panel is drawn, we ask what the joint can carry, how it comes apart in twenty years, and whether the plate lies flat on a truck. A shell you cannot disassemble is, for us, an unfinished thought. The Monday move: take any curved-roof or coffered-slab geometry currently sitting in your model and run one segmentation pass in Grasshopper with Karamba3D — tessellate it, then colour-map the joint moment along every shared edge before you commit to a panel count. Let the connector, not the silhouette, decide the resolution.

Hack: Make the trade-off visible instead of theoretical — for every shared edge, read the dihedral kink you are asking the joint to resolve, so you can see the fabrication-versus-bending curve with your own eyes rather than trust a render. Grasshopper/Rhino with Python (or Karamba3D) is the single environment the Alp team rode from form-finding to CNC. Sweep the panel count and watch planarity deviation fall as the edge moment climbs; the good segmentation is the knee, never either extreme.

import Rhino.Geometry as rg
for a, b in shared_edges(panels):                     # adjacent panel index pairs
    kink = rg.Vector3d.VectorAngle(normal(a), normal(b))   # dihedral angle, radians
    edge_moment.append(N_load * edge_len(a, b) * kink)     # colour-map this per edge
# Flat panels drive the kink to ~0; coarse facets open it and the joint fights back.

Move: The real trajectory is downward — from roof to floor slab, the everyday concrete plate that pours roughly a third of a building’s embodied carbon. The 2023 prototype low-carbon segmented concrete floor system from an ETH-led research consortium is the tell: a shell prefabricated off-site in transportable, demountable segments with columns and tie rods. Not a roof, a floor. Once segmentation logic can undercut a flat slab on embodied carbon and hand you a structure that unbolts, the technique stops being architecture’s showpiece and becomes an argument you make to a quantity surveyor. From the vantage of hindsight, the parametric work worth regretting is never the faceted form — it is the form whose derivation went dark when the plugin did. Write down the objective, name the method, keep the joint logic legible on paper so a 25-year-old can rebuild the shell from the principle and not the file. Whoever standardises a segment-to-segment connector that is stiff, cheap, and reversible will do for the segmented shell what the standard steel connection did for the frame — and the sand dollar will have out-engineered us by a few hundred million years, quietly, from the sea floor.

FILED FROM
CO-SIGNERS
PAZ Academy
CONFIDENCE
HIGH
REPRINTS
© PAZ - PARAMETRIC ACADEMY ZURICH · ALL RIGHTS RESERVED

PAZ Kaffi · multidisciplinary editorial, led by PAZ Academy

⚑ REPORT AN ERROR · SUBMIT A CORRECTION
◂ BACK TO FRONT PAGE · PAZ KAFFI

© 2026 PAZ Academy.