Panelization Strategies

Every NURBS surface in Rhino carries its own two-directional coordinate system, called u and v. Think of them as latitude and longitude painted onto the surface: any point is addressed by a (u, v) pair, each running conventionally from 0 to 1 across the surface regardless of its real size. This is the foundation of almost all panelization, because it lets you divide the surface evenly even when the surface is curved and its edges are unequal.

In Grasshopper the workhorse is Divide Domain (Divide Domain2) followed by Isotrim (nicknamed SubSrf). Divide Domain2 chops the surface's u,v domain into a grid of U by V smaller domains; Isotrim then cuts the surface into one sub-surface per cell. Feed it 10 and 20 and you get a 10x20 field of panels that follow the curvature exactly. Change either number and every panel re-solves - the parametric promise, applied to cladding.

A few honest cautions. A UV grid is even in parameter space, not always in real space: if the surface is stretched, the panels bunch where the surface is dense. And the grid inherits the surface's flow - if the isocurves run awkwardly, so will your seams. Sometimes the first real move in panelization is rebuilding or re-lofting the surface so its u,v directions run the way you want the joints to run.

A sub-surface is just a patch of the surface. Real panels have content: a frame, a folded metal cassette, a perforation, an opening. You don't want to model that thousands of times. Instead you model one panel unit in a neat box and let Grasshopper morph a copy of it into every grid cell, bending it to fit the local curvature.

The tool is Box Morph. You build your panel once inside a unit Bounding Box (a 'reference' box, 0..1 in each direction). Divide Domain2 can output Surface Box cells - little twisted boxes sitting on the surface, one per grid cell. Box Morph then maps the contents of your reference box into each surface box. A mullion in the reference box becomes a mullion that follows the surface; a bump becomes a bump that leans with the curvature. Model once, populate thousands.

This is the engine behind most 'designed' panels - operable louvres, kinetic shades, ornamental screens. The discipline it demands is thinking in normalized coordinates: your reference panel lives in a tidy 0-to-1 world, and the surface's curvature is applied afterwards. Get comfortable moving between that clean unit box and the messy real surface and you can panelize almost anything - though remember the morph will happily warp a flat panel to fit, which is exactly the planarity problem we deal with next.

You rarely have to build the grid maths from scratch. LunchBox, a free and hugely popular Grasshopper plug-in, has a 'Panels' family that takes a surface and a U,V count and hands back ready patterns: Quad, Diamond, Triangle, Hexagon, and staggered brick-like variants. Feed a surface, get panel geometry - it is the fastest way to try five panelization languages in a minute.

The choice is not cosmetic. Quads are efficient and read calmly, but on a doubly curved surface a quad's four corners generally do not lie in one plane - so quads warp. Triangles are the safe workhorse: any three points define a plane, so triangular panels are always planar, which is why triangulated glazing wraps the most aggressive blobs. The cost is more nodes, more edges, sharper visual busyness, and awkward acute corners. Diamonds (rotated quads) and hexagons give a distinctive rhythm and can reduce node complexity, but hexagons on curvature warp even more than quads and often need to be flat-faceted or subtly split.

So the pattern decision is really a negotiation between look, structure (how many members meet at a node) and fabrication (can the panel be made flat and cheap?). A good habit: generate the same surface as quad, diamond and triangle, then measure planarity on each before you fall in love with one.

Panelization is not finished when the surface is covered - the panels have to meet each other. Between every two panels is a joint: a gap for a gasket, a shadow line, a structural mullion, a weather seal. Those joints are where a facade is waterproofed, where thermal movement is absorbed, and, visually, where the geometry either sings or looks cheap.

Computationally this means you almost never want panels that touch edge-to-edge at full size. You offset each panel edge inward by a few millimetres (Scale each panel about its centre, or offset its boundary) to open a consistent reveal for the gasket and to give tolerance for the fact that nothing is built perfectly. That small, deliberate gap is one of the marks of a facade that was detailed rather than merely modelled.

Seams also carry meaning. A continuous vertical seam reads as calm and tall; a staggered, brick-bonded seam reads as textured and can hide slight panel-to-panel variation. Where seams cross - four-way nodes - you get the most tolerance-sensitive, most expensive condition, which is one more reason triangular or diamond patterns (three-way nodes) are structurally kinder. Panelization, done well, is as much about designing the lines between panels as the panels themselves. Keep that reveal in your definition from the start; retrofitting it later is painful.

There is a practical dividend, too: because the reveal is parametric, you can dial it to match the fabricator's real gasket sizes and the building's expected thermal movement, and the whole facade re-solves to suit. A joint is not a leftover; it is a component with its own dimensions, its own tolerance, and its own contribution to how the elevation reads at a distance and up close.

With several tools in hand - UV grids, Box Morph, LunchBox patterns, meshes - the real skill is choosing between them for a given surface and brief. Work through four questions in order and the choice usually settles itself.

First, how curved is the surface? If it is flat or single-curved, quads stay planar and you have full freedom of pattern; if it is aggressively doubly curved, you are already being pushed toward triangles or a planarized mesh, and that constraint should lead. Second, what is the panel material? Flat glass, folded metal cassettes and GFRC each have different size limits, weights and cost curves - a pattern that yields 1.5m panels is a different project from one that yields 4m ones. Third, how many members meet at a node? Quads and hexes create four-way and three-way nodes with different structural and waterproofing consequences; triangles concentrate more members at each vertex. Fourth, what does the seam say? The joint pattern is a major part of the elevation's character, and it should be a positive choice, not a by-product of whichever component you reached for first.

The honest workflow is to run the surface through two or three strategies, measure each on planarity, panel count, unique-panel count and node type, and then decide with data on the table. This is exactly the parametric promise applied to a design decision: because each strategy is a re-solvable definition, comparing them costs minutes, not days. A designer who panelizes by reflex picks a pattern and hopes; a computational designer generates the alternatives and chooses. Keep the comparison itself in your definition, so that when the surface changes - and it will - the whole comparison re-runs.