Fibre Reinforced Facade Encyclopedia: Manufacturing
Robotic manufacturing in architectural facade work is the use of industrial robot arms and multi-axis machine tools to carry out operations that were done by hand: milling a mould directly from the digital model, spraying a cementitious or resin laminate to a controlled thickness, and trimming and drilling the cured panel. It is not an automated factory. Architectural panels are made in runs of tens, not tens of thousands, so the economics that justify robots in automotive production do not transfer. Automation pays in this industry where it removes geometric risk or removes a repetitive, unpleasant, variable manual task, and it does not pay anywhere else.
Milling one mould for a doubly curved panel takes a five axis router several hours of unattended machine time. Building the same mould by hand from ribs and battens takes a skilled mouldmaker days, and the result is only as accurate as the setting out.
At a glance
| Property | Typical value | Note |
|---|---|---|
| Mould milling | 5 axis CNC router or robot arm on a linear track | The most established and most economic robotic step |
| Milled substrates | High density polyurethane board, EPS, MDF, tooling paste | Machined direct from the NURBS model |
| Typical milling tolerance | ±0.5 to ±1.0 mm on the machined face | Typical machine capability, before finishing and sealing |
| Robotic spray | 6 axis arm carrying a spray head over the mould | Consistent thickness and fibre distribution; best on repeats |
| Skin thickness consistency | Target thickness ±1 to 2 mm | Manual spray varies more across a large mould; typical |
| Trimming and drilling | Robot or CNC bed for edge trim and fixing holes | Removes the main dust and silica exposure task |
| Economic repeat count | Mould milling from one unit; robotic spray from tens | Indicative; geometry dependent |
| Programming input | The same model used for shop drawings and FEA | Digital continuity is the real saving |
Robotic mould milling
This is where automation in architectural composites has genuinely settled. A five axis router or a robot arm on a linear track cuts the mould face directly from the model, in high density polyurethane board, expanded polystyrene, MDF or a tooling paste applied over a rough substrate. The machined surface is then filled, sealed and polished to the finish required on the panel face.
The advantage is not machine speed. It is that the mould is a direct expression of the model, so the geometric error between what the architect drew and what the factory built is a machine tolerance rather than a setting-out error accumulated across ribs, templates and battens. On double curvature and freeform geometry this is decisive, because there is no practical manual method of building a doubly curved mould to a verifiable tolerance. It also means a revised model produces a revised mould without re-deriving the geometry by hand.
Robotic spray and lay-up
A six axis arm carrying a spray head can traverse a mould at a fixed standoff and a fixed rate, which is difficult to do by hand across a large or deeply profiled mould held at arm’s length. The results are more consistent skin thickness, more consistent fibre content and less overspray, which flows through to less variation in modulus of rupture between panels and less material waste.
The qualifications are significant. Programming the path for a new mould takes time, so robotic spray pays where the same mould is used many times or where a family of moulds shares a path strategy. Roller compaction after spraying, which is what removes air and develops the bond between layers on a cementitious laminate, is still largely a hand operation. Deep returns, tight internal corners and undercut geometry are hard to reach with an arm and are finished manually. In practice a robotic spray cell in architectural work is a hybrid: the robot lays the bulk of the skin, an operator finishes the difficult areas and compacts.
Trimming, drilling and the tasks worth automating
The strongest case for automation in this industry is not speed. It is removing tasks that are repetitive, dusty and variable.
- Edge trimming a cured panel to the model line, which is slow, noisy and easy to get wrong by a few millimetres
- Drilling and routing fixing holes and anchor pockets, where position accuracy determines whether the panel hangs on the substructure without site rework
- Cutting silica-bearing cementitious material, which is a respirable crystalline silica exposure that is better done in an enclosed cell
- Surface preparation and finishing passes on a mould, which are long and monotonous
- Scanning a finished panel against the model for dimensional quality control
Where automation does not pay
An honest account has to include the cases where robots are the wrong answer, because the industry’s marketing rarely does.
Low volume is the core problem. A typical facade package might have forty distinct panel types in runs of five to thirty. Programming, fixturing and proving a robotic path for a run of five costs more than the labour it replaces. Hand lay-up and hand spray remain the default for short runs and will continue to.
Complex geometry cuts both ways. The same deep returns, undercuts and re-entrant folds that make a panel worth making in a moulded composite are what a robot arm cannot reach. Reach, singularity and collision avoidance are real constraints on an arm working inside a 4 m mould.
Finishing is still craft. Acid etching, sandblasting, integral colour matching and defect repair are judgement tasks performed against a control sample. Colour consistency across a facade is achieved by a person comparing panels under controlled light, not by a machine.
The realistic position is that digital fabrication has transformed mould making and quality control in architectural composites, and has improved but not replaced the wet processes. That is a meaningful change; it is not a robot factory.
Digital continuity is the actual benefit
The value that robotic manufacturing delivers is less about the robot than about what feeding it requires. If a mould is milled from the model, the model has to be complete, correct and coordinated. That discipline pulls the whole package forward: shop drawings are generated from the same model, panel setting out is derived from it, finite element analysis of the panel uses the same geometry, and the as-built scan is compared back against it.
Standards and references
- ISO 10218 parts 1 and 2, Robots and robotic devices, safety requirements for industrial robots
- ISO/TS 15066, collaborative robot safety, where an operator shares the cell
- AS 4024 Safety of machinery series (Australia), for guarding and cell design
- ISO 2768 and ISO 286, general tolerances, as the basis for machined mould tolerance specifications
- GRCA Specification for the Manufacture, Curing and Testing of GRC Products (fibre content and thickness control apply equally to robotic spray)
- Safe Work Australia Code of Practice, managing respirable crystalline silica in the workplace
Where this matters on a ShapeShift project
Spanda: A large freeform public artwork of the kind that cannot be set out accurately by hand. Geometry of this sort is resolved in the model and cut into the tooling rather than developed on the shop floor.
Related terms
Questions specifiers ask
Does robotic manufacturing make facade panels cheaper?
On mould making for complex geometry, yes, and often from the first unit. On the wet processes, only where a mould is used many times. Architectural run lengths are short, so most of the saving sits in tooling and quality control rather than in production labour.
Can a robot spray a whole GRC panel?
It can lay the bulk of the skin at a consistent thickness, but roller compaction, deep returns and internal corners are still finished by hand. Robotic spray cells in this industry are hybrid, not autonomous.
What tolerance can a milled mould hold?
Machine capability on the milled face is typically ±0.5 to ±1.0 mm, before the surface is filled, sealed and polished. The finished panel tolerance is wider and is set by the material and the demoulding process as well.
Is robotic fabrication necessary for double curvature?
In practice yes. There is no reliable manual method of building a doubly curved mould to a verifiable tolerance, so freeform and double curvature work is milled from the model.
What is the safety benefit?
Enclosing the cutting and trimming of cementitious panels in a machine cell removes the largest respirable crystalline silica exposure in the factory, and takes operators out of repetitive overhead spraying.
Freeform or doubly curved geometry is resolved in the model and cut into the tooling. Send the model early. Talk to the ShapeShift team
Reviewed by Richard Halvorsen, General Manager, ShapeShift Technologies. Last updated September 2026.