Digital Fabrication: File to Factory

File to factory names a simple but profound shift: the digital file is not a picture of the thing to be made, it is the instruction set that makes it. In a traditional workflow a drawing is interpreted by a person who then makes the object. In digital fabrication the geometry drives the machine directly - a laser, a router, a printer - with far less human reinterpretation in between. The consequence for a computational designer is that how you build the definition now includes how the thing gets made.

The two most accessible machines are subtractive. A laser cutter works on flat sheet: it moves a focused beam in 2D and cuts or engraves acrylic, plywood, card, MDF. A CNC router or mill moves a spinning cutting bit in 2D or 3D to carve sheet or solid stock - the same numerical-control idea, in three dimensions. Both read simple machine files: a laser typically takes 2D vector lines (DXF), a router takes a toolpath (G-code) generated from your geometry by CAM software. The through-line is that your parametric model must ultimately resolve into clean, machine-ready geometry - and building toward that from the start is the whole discipline.

Why do this parametrically? Because when the design is a definition, the fabrication files regenerate when the design changes. Nudge a slider and every panel, every joint, every cut line updates and re-nests - no redrawing two hundred parts by hand. Imagine a perforated screen of two hundred unique panels: drawn by hand, a late client change is a week of re-drafting and a fresh chance for errors; built as a definition, it is one slider move and a re-baked DXF, with the part numbering still consistent. That regeneration is the payoff that makes genuinely complex fabricated work feasible at all - it is not a convenience, it is what makes the ambition affordable.

Most fabrication starts from flat sheet, but designs are rarely flat - so the core operation is unrolling (developing) a 3D surface into the flat pieces you will cut and then reassemble. In Grasshopper the Unroll Brep component (and panelling plug-ins like LunchBox, with its Diamond, Quad and Triangle Panels components) flattens strips and panels and returns a transform that maps each flat part back to its place in the 3D form - so you can keep track of which edge met which and reconstruct the whole. A clean unroll workflow is roughly five moves: divide the surface into panels, unroll each to the world XY plane, tag each with a label, add joints, then nest - and every one of those steps is a component chain you build once and reuse.

Here geometry imposes an honest rule you cannot argue with. Only developable surfaces - ones with zero Gaussian curvature, like cylinders and cones, curved in a single direction - unroll to flat with no distortion. A double-curved surface (curved in two directions at once, like a sphere or a saddle) cannot be flattened without stretching or tearing, which is exactly why you can wrap paper smoothly around a can but never around a ball. So double-curved designs must be rationalised for flat fabrication: approximated by many small flat facets (planar panels), or sliced into developable strips, or cut as a contoured waffle of ribs. Rhino's curvature analysis or a Gaussian-curvature check will show you which regions are which. Choosing the rationalisation strategy is a real design decision with visible consequences - flat facets read faceted, strips read ribbed - and it is far cheaper to make that choice in the definition than to discover it standing at the machine with a surface that will not lie flat.

Alongside unrolling sits its cousin, contouring: slicing a solid into a stack of profiles (for a laser-cut sectional model or a plywood 'waffle' of intersecting ribs) or into ribs plus a skin. Same idea - reduce a 3D form to a set of flat, cuttable parts, with the assembly logic baked in so the pieces can only go back together one correct way.

Flat parts still have to be laid out and joined. Nesting is arranging all your parts economically on the stock sheet - rotating and interlocking them so you waste as little material as possible, because both the sheet and the cutting time cost money. For a handful of parts you nest by eye on a rectangle the size of your real stock; for hundreds, nesting algorithms and plug-ins pack them automatically. Good nesting also respects grain direction (plywood is stronger and cleaner along the face grain), keeps a margin from the sheet edge where clamps and the bed sit, and leaves small bridges where useful so parts do not skitter loose mid-cut.

Then the parts must hold together. Rather than glue everything, computational fabrication loves integral joints cut into the parts themselves: tab-and-slot connections, finger joints along edges, notched cross-lap joints for waffle structures, and mortise-and-tenon logic for frames. The elegance is that these joints are generated by the same definition that made the panels, so they always line up exactly - and because the design is parametric you can label each part and engrave its number and its neighbours' numbers automatically, turning bench assembly into a numbered sequence rather than a jigsaw puzzle. For a waffle, each rib carries slots half its depth so ribs in the two directions simply drop together at their crossings.

Designing joints is really designing for assembly: the order pieces go together, whether a hand or a screwdriver can physically reach a fastener, and whether the half-built structure stands up on its own or needs a jig. This is the step where a beautiful screen model becomes a buildable object - and it is always worth cutting and testing a single joint before committing a whole expensive sheet to the laser.

Two small realities decide whether your parts fit, and beginners meet them the hard way. Kerf is the width of material the tool removes - the laser beam burns away perhaps 0.1-0.3mm, the router bit is several millimetres wide - so a cut is never a zero-width line. Ignore kerf and every slot ends up slightly oversized and every tab slightly undersized; parts that should press-fit end up loose. Precise work compensates for kerf, offsetting cut lines by half the kerf so the finished dimension is what you drew.

Tolerance is the deliberate clearance you design into a fit. Nominal material thickness lies - '3mm' plywood might be 2.7mm and vary sheet to sheet - so a slot cut at exactly 3mm may jam or gape. The professional habit is to test: cut a small coupon with a few slot widths, measure the real thickness and real kerf on your machine and material, and set your fits from evidence, not the catalogue number. A joint that clicks together came from a test cut, not from luck.

Finally the handoff: the laser needs clean DXF. That means closed polylines (open loops confuse cut order), curves at a sensible tolerance, layers separating cut from engrave from score by colour or layer as the shop expects, correct real-world units, and no stray duplicate or zero-length lines. Grasshopper can bake exactly this - flatten your parts to the world XY plane, sort onto the right layers, and export - so the definition produces a file the machine accepts on the first try. Get kerf, tolerance and a clean DXF right, and file-to-factory simply works.