Rationalization for Fabrication

The central problem, as we met in lesson 6.1, is that a doubly curved surface cannot be tiled with flat quadrilaterals - a quad's four corners generally won't share a plane. Since flat glass and flat sheet metal are dramatically cheaper than curved, the first rationalization move is usually to make the panels planar. There are three honest routes.

Triangulate. Any three points define a plane, so triangular panels are always flat. This is the bulletproof option and wraps any surface, at the cost of more panels, more framing, three-way nodes and a busier, more crystalline look. Planar-quad (PQ) meshing. With more effort you can find a quad mesh whose every face is (nearly) planar - visually calmer than triangles and structurally efficient - but only for surfaces that permit it, and it needs a solver. Physics relaxation. Tools like Kangaroo can take a quad mesh and iteratively nudge its vertices until each face is planar within a tolerance, balancing planarity against staying close to the design surface. You are negotiating: perfect smoothness versus perfectly flat panels, and the tolerance you accept is the deal.

The key mental shift is that the smooth NURBS surface is now a target, not the deliverable. The deliverable is a mesh of flat (or single-curved) panels that approximates it within an agreed deviation. Rationalization is the art of keeping that approximation faithful enough to read as the intended form while flat enough to build.

Flat is cheapest, but there is a valuable middle ground: single curvature. A surface that curves in only one direction - like a rolled sheet or a cylinder - is developable, meaning it can be unrolled flat without stretching or tearing. That matters enormously for fabrication, because a developable panel can be cut flat from a sheet and then rolled to shape by a simple bending machine, no expensive double-curved moulds required.

The rationalization strategy here is to approximate a freeform surface with a series of developable strips - long ribbons, each single-curved, that together follow the form. The Gehry-office-style shingled metal facades are the famous example: what looks like a continuous doubly-curved skin is actually a family of developable strips and flat shingles, each unrollable to a flat cutting pattern. In Grasshopper you can unroll developable pieces with Unroll Surface (Unroll) to get the exact flat blank the cutter needs.

So a mature rationalization has a hierarchy of ambition and cost: flat panels (cheapest), then single-curved / developable (roll a flat blank - moderate), then doubly-curved (moulds or curved glass - expensive), used sparingly only where the design truly demands it. A good computational designer consciously pushes as much of the surface as possible down toward the cheap end of that ladder, spending double curvature only where it earns its place.

Here is the fact that reshapes how you design: a facade's cost is driven less by the number of panels than by the number of _unique_ panels. A thousand identical panels share one cutting file, one mould, one setup, one learning curve - the factory makes them fast and cheap. A thousand different panels means a thousand setups, a thousand drawings, a thousand chances to fabricate or install the wrong one. Uniqueness, not quantity, is the cost driver.

So a huge part of rationalization is clustering: grouping near-identical panels and replacing each cluster with a single representative type, accepting a small geometric deviation in exchange for a large drop in unique-part count. Where 900 subtly different panels can be built from, say, 30 types within tolerance, the saving is transformative. Computationally you sort panels by their key dimensions (width, height, corner angles, curvature) and snap similar ones to shared types - the same logic behind standardising a structure to a few member sizes.

This is a genuine design negotiation, not just post-processing. Sometimes you deliberately adjust the form - regularising the grid, relaxing curvature, aligning panel sizes - so that far more panels fall into shared types, trading a little geometric freedom for a lot of buildability. The most elegant rationalizations are designed in from the start, with the surface shaped so it wants to become a small kit of repeating parts. Retrofitting repetition onto a wilfully unique surface is far harder.

Two final, unglamorous disciplines turn a rationalized model into a delivered building. The first is tolerance. Nothing is built to the exact millimetre of your model: steel is fabricated within a range, concrete moves, panels are hoisted by crane onto a structure that itself has a survey deviation. So the joint between panels is not a zero-width line - it is a designed gap sized to absorb the accumulated tolerance while staying weathertight. A rationalization that assumes perfect fit will fail on site. You design the reveal, the gasket range and the fixing slots to swallow real-world error, and you keep tolerance as an explicit dimension in the definition, not a hope.

The second is scheduling and labelling. Every panel needs a unique identity - a code like P-042 - carrying its type, its exact dimensions, its angles, its position on the building and its orientation ('this edge up'). From the parametric model you generate a panel schedule (a data table, one row per panel) and, for each type, a labelled cut file the factory machine reads directly. This is file-to-factory: the Grasshopper definition outputs not a pretty image but dimensioned geometry, DXF cut patterns, and a numbered schedule that flows into CNC routers, laser cutters and the installers' setting-out drawings. The unique IDs are what stop panel P-042 being installed where P-140 belongs - the difference between a facade that goes up cleanly and one that becomes a site dispute.

Done well, rationalization closes the loop the whole course has been building toward: the definition that generated the form also generates the parts list, the cut files and the assembly labels. Design and manufacture become one continuous, parametric chain - which is exactly the promise of computational design made real.

The most important lesson about rationalization is when it happens. The weak workflow treats it as clean-up: design a wilful surface, freeze it, then desperately try to planarize and de-duplicate an already-committed geometry. That almost always ends in pain - either the design gets butchered to become buildable, or the buildability gets butchered to preserve the design. The strong workflow rationalizes from the first sketch, shaping the surface so it inherently wants to become a small kit of flat, repeating parts.

Concretely, that means keeping buildability metrics live in your definition while you are still exploring the form. Show planarity deviation, unique-panel count and the flat/developable/doubly-curved breakdown as real-time readouts, coloured on the model, so that as you push and pull the surface you watch the cost consequences move. A slight regularizing of the grid, a curvature relaxed just here, a fold aligned just there - each is cheap while the design is fluid and ruinous once it is frozen. This is design-space thinking (Module 8's territory) pointed at manufacture: you are exploring not only how the surface looks but how buildable each version is, together.

The other half of rationalizing early is talking to the people who build it early. A fabricator's real constraints - their maximum sheet size, their minimum bending radius, their standard glass thicknesses, their genuine cost drivers - are worth more than any generic rule, and they can only shape the design if you learn them before the geometry sets. Rationalization, at its best, is not a solitary geometric exercise performed at the end; it is a conversation with fabrication, held from the beginning, encoded in a definition that keeps the ideal and the buildable visible side by side. That is the habit that lets computational designers deliver ambition intact.