showstack

The processes

Three ways of making a solid, and how each one fails

Nearly all of it is one of three families, and they are chosen by the failure you cannot accept rather than by the finish you want.

FDM — fused deposition, the familiar filament printer — extrudes molten plastic in layers. It is cheap, the machines are everywhere, the materials range from PLA to nylon and carbon-filled composites, and it produces the anisotropy the next section is about. It is the right answer for most brackets, jigs, cable management and one-off housings.

SLA and DLP cure liquid resin with light, layer by layer. Far finer detail and a genuinely smooth surface, which matters for anything optical, anything that seals, and anything appearing on camera. The catch is that most resins are brittle and many degrade under UV, so a resin part that lives outdoors or in a followspot beam is on a clock. The uncured resin is also a skin sensitiser, which makes it a workshop process rather than a corner-of-the-office one.

SLS and MJF fuse powdered nylon with a laser or a binder plus heat. Genuinely strong, close to isotropic because there are no bonded layers in the same sense, and it needs no support structure at all because the unfused powder holds the part up — so it can make geometries the other two cannot. It is the one you send out to a bureau rather than own, and it is what you use when the part actually has to work.

The one that matters

A printed part is not the same strength in every direction

This is the fact that separates people who print things that work from people who print things that break, and it is not intuitive because the part looks solid.

An FDM part is built from bonded layers, and the bond between layers is weaker than the material within a layer — often dramatically so, because it is a partial re-melt rather than continuous material. Load it in the plane of the layers and you get something close to the plastic’s real strength. Load it across them and you are relying on that bond, and it delaminates like a stack of pancakes.

The consequence is that orientation on the bed is a structural decision, made in the slicer by whoever pressed slice, usually without thinking of it that way. That is worth saying plainly, because the slicer looks like a print-settings dialogue and is in fact where the part’s mechanical properties are chosen. The same file, the same machine, the same spool, turned ninety degrees, is a different part. A hook printed lying down is far stronger than the identical hook printed standing up, and nothing about the model or the settings tells you which one you got.

Two practical habits follow. Design so the load runs along the layers, and if you cannot, say so on the part — a printed component with no indication of its build orientation cannot be safely reproduced by the next person. And treat infill percentage as a stiffness setting rather than a strength one: perimeters carry most of the load, so three walls at 20% infill beats two walls at 60% for most brackets, and uses less material doing it.

Measuring what exists

Scanning, and when a caliper is faster

Reverse engineering on a show usually means making something fit a thing you did not design, cannot take away, and are not allowed to modify. There are three levels of answer and the first is almost always the right one.

Measure it. A caliper, a set of radius gauges, thread pitch gauges and twenty minutes gets a better model than a scan for anything with flat faces, round holes and right angles — which is most brackets, most panels and most mounting plates. A scan gives you a mesh of approximately where the surface was; a measurement gives you the number the original was made to, which is nearly always a round one in some unit. Recognising that a hole is 8 mm rather than 7.94 is the entire skill.

Photogrammetry builds geometry from a set of overlapping photographs. It is free, the camera is in your pocket, and it is excellent for organic shapes, scenic elements and anything textured. It is poor at shiny, transparent and featureless surfaces, because it needs visual detail to match between frames — and it gives you no absolute scale unless you photograph a ruler alongside.

Structured light and laser scanning project a pattern and measure its distortion. Genuinely accurate, genuinely scaled, and it produces a mesh rather than a model: you still have to work out what the designer intended and rebuild it as real geometry, which is the part that takes the time. Scan-to-CAD is not a conversion, it is a reconstruction.

Working like it is production

Because on a show it is

A prototype on a production is not a prototype. It goes in the truck and stays there for eighteen months, so the habits that matter are the boring ones.

Print two of everything and put the spare in the case, because the second one costs the material and nothing else, and the day it is needed there is no printer. Mark parts with the file name and the date, in the model, as embossed text — a part you cannot identify cannot be reprinted. Keep the source file rather than only the mesh, because an STL cannot be sensibly modified and a parametric file can be adjusted in a minute. And write the material and orientation on it, for the same reason you write the gauge on a cable.

Iterate on the joint rather than on the part. Most failures are at the interface — a hole in the wrong place, a fit that is 0.2 mm out — so printing a small coupon of just that feature takes ten minutes instead of four hours and answers the only question that was actually uncertain.

Where this goes next

The neighbouring pages

Drawings, models and BIM covers the modelling side and why an object that knows what it weighs is different from a line. When the venue is a field has the IP ratings that decide whether a printed housing is protecting anything. Hoists and the safety chain is where the boundary above comes from. And before there was a computer in it is the components that usually end up inside whatever you just printed a case for.