| Episodes | 137 |
| Mentions | 291 |
| Cited here | 36 |
| First — last | #4 — #727 |
| Top guests | Jeff Keyzer, Chris Anderson, John Saunders |
| Related | 3d printing · kickstarter · open source hardware · pick and place machine · maker faire |
A 3D printer is a machine that fabricates solid objects additively, building them one layer at a time from a digital model under toolpaths expressed in G-code.[251][503] The most widespread class, fused-filament fabrication, drives motors in X and Y, heats plastic filament and feeds it through a head and nozzle onto a platform that must stay level, with stepper motors that must not slip.[326] Print speed on such machines is bounded by the thermal properties of the plastic rather than the motion system: each layer must harden just enough before the next is laid down.[105] The technology’s weight in practice sits in professional product development, where consumer products designed today make heavy use of printing, and in the fabrication of parts — housings, jigs, fixtures — that no other process would serve.[244][421]
History
Before the machine designs were opened, a 3D printer could not be had for less than about twenty thousand dollars, and stereolithography machines ran to a couple of hundred thousand dollars.[4][424] Publishing the designs created the low-cost segment, after which professional machines fell into the one-to-five-thousand-dollar range.[4] The best-known low-cost machine was derived from the RepRap project, and its kit form was discontinued in favour of a fully built unit once the market moved.[79] By 2012 a fused-filament machine for the home had appeared at about $2,400, and stereolithography machines fell to around ten thousand dollars in the 2010s.[424]
The scale of the best-known low-cost vendor was small in unit terms — roughly 22,000 printers over about four years, at modest margin — so a four-hundred-million-dollar acquisition of that business was not for the financials, nor for technology that was largely open, but for the brand and the community.[151] The most widely recommended open machine descends from the same lineage: its founder encountered 3D printing while building Arduino projects as a teenager, combined the two into his own printer designs, and the company grew to around a thousand employees building printers that remain largely open source.[726]
Not every vendor embraced openness. One printer denied the user any control of temperature and other process parameters so that only the vendor’s material would work; the restriction was reverse engineered into an aftermarket temperature controller.[424]
Operating principles
Fused-filament fabrication
A fused-filament machine comprises motors moving in X and Y, plastic that has to be heated, filament fed through a head and nozzle, a platform that has to stay level and aligned, and stepper motors that must not slip and have a maximum usable speed.[326] This complexity is irreducible in a way that resists cost reduction: applying money and time does not make such a machine a hundred times faster, simpler or more reliable.[326] The speed bound is thermal rather than mechanical: each layer of extruded plastic must harden just enough before the next is laid down, or the new layer smears the one beneath it, and this physical limit is why tuned machines remain slow.[105] The Z axis is typically the slowest on a printer because it is driven through a lead screw.[369]
Resin processes
Curing a liquid resin with a laser does not carry the same thermal constraint, which is why the resin processes were expected to be the next step in speed.[105] Stereolithography also holds the surface-quality high ground: fused-filament output does not match the finish of industrial stereolithography, which limits its use wherever the finish is part of the deliverable.[224]
Motion platform and control
The motion platform is common across the category of desktop fabrication: large printers, laser cutters, plasma cutters and CNC machines are the same gantry with a different tool at the end.[251] The commonality goes deeper than the mechanics. On his open benchtop pick-and-place machine, Hawes runs Marlin — printer firmware — making the machine in effect a 3D printer with no nozzle and no bed and considerably more axes.[686] A machine designed for speed rather than cost becomes an exercise in motion control: one such build paired an injection-moulded enclosure with the designer’s own motion engine, PID control and path algorithms to handle the dynamics at high acceleration.[282] Printer motion is also repurposed outright: for a robot that taps phone screens, built on printer motion, Huggins added a solenoid alongside the lead-screw Z axis, reaching around 800 taps per minute.[369]
Every machine in the class is driven by G-code and is useless without software that hides it; the hardware makes the process accessible and the software is what makes it usable.[251] A printed model is described to the machine as STL, a layer-by-layer representation handed to the machine that will build the object up one layer at a time — analogous to GDS in chip fabrication, where the same role is played by the layout file.[503]
Toolchain
The CAD toolchain was the bottleneck for a long time: free packages variously failed to import STEP files correctly or to export usable STL, and the one that worked had an explicit export-for-printing path.[122] One such tool then removed direct STL export in a later release, redirecting the user to order the part from approved vendors on the premise that nobody has a printer at home.[122] For the common case — enclosures for electronics printed on a mid-cost fused-filament machine — an open-source parametric CAD package is rough to learn but capable of the work.[574]
Between the model and the machine sits the slicer, and the slicer is where the accumulated intelligence of the field sits: selecting a common material and dropping in a model produces a working job without the user configuring anything.[686] The material a newcomer most needs is generic across machines: what a slicer is, how the additive process works, and how to diagnose a first layer that will not stick.[675]
Dimensional compensation belongs in the toolchain rather than in the designer’s model: the machine knows the plastic will ooze and that circles will come out smaller than programmed, so the drawing should carry the intended dimensions and the software should apply the offset.[94]
Limitations and failure modes
Printed features come out undersized because the extruded plastic spreads: a 42-millimetre hole modelled half a millimetre oversize still did not fit the part it was cut for.[94] A machine that appears to be working can still be misconfigured in several independent ways at once — streaming a job over USB can lose buffer synchronisation where printing from the memory card does not, temperatures can be out of calibration, and belts can slip or sit misaligned.[94] Adhesion depends on the state of the build surface, not only on the settings: a print that needed a raft to stick, and whose raft could not then be separated from the part, turned out to need the tape on the bed cleaned before starting.[94] Low-end machines are nonetheless adequate for flat work such as a front panel, provided the hole diameters are checked against the parts that must pass through them.[94]
Belt and motor slippage can be partly compensated in software — driving the axes more slowly, or averaging — but not enough to make a polished product; closing the loop properly requires an encoder, which raises complexity and lowers speed, producing a different machine rather than a firmware change.[337] A printer that produces a part also gives a false sense of completion: the prototype prints, the excitement of it working masks defects such as layers not adhering, and those get deferred to a software fix that does not exist.[337]
For a cheap machine the failure mode is not that information is unavailable but that feedback is slow: hours pass before it is clear whether a change worked, and the time is spent watching prints fail.[329] An intermittently used machine decays into unavailability as new slicer software stops working with old firmware; getting such a printer operational again for a single part can consume a day.[625]
Strength is a further limit. Printing a part that is merely pretty is easy; printing one that carries load is not, and the tools do not default to it: strength comes from settings the user has to learn — more outside perimeters, and different infill patterns and densities — and the documentation for that is poor.[472] Soluble support material stayed an industrial feature and did not reach hobby machines, and the evidence that it is not a prerequisite is the quality of parts printed without it.[472] Keyzer’s own trajectory followed this path: by building parts that failed he learned how to build better parts, the first brackets being mechanically unsound, and he judges that learning alone worth the machine.[472]
A printed part’s success also says nothing about its manufacturability by other means: a printer will happily produce a feature that no volume process can reproduce, so printing a hole successfully says nothing about whether the part could be injection moulded — a lesson that has to be learned once.[127]
Economics of the machine
The mechanical part count sets a price floor that electronics do not have: boards and components can be driven ever cheaper through commodity parts and low-cost labour, but a machine with many mechanical parts costs what it costs, which is the argument behind the conclusion that a sub-three-hundred-dollar printer is not buildable.[243] Warranty and return costs are routinely omitted from these calculations, and on a machine this complex a replacement rate of one in twenty or one in thirty units is plausible.[243] One crowdfunded machine at an unbuildable price raised about a hundred and fifty thousand dollars against the roughly one and a quarter million the arithmetic said it needed.[219]
Buying and owning
The specifications that matter when buying are reliability and the size of the user community, not peak performance: the objective is that every problem encountered has already been solved by somebody else and written down.[610] A machine chosen on that basis becomes a tool rather than a project — one much-supported model needed calibrating once and has produced hundreds of prints since without further attention — and the skill worth building is the modelling rather than the tuning of the machine, a reversal of the earlier position where the printer itself absorbed the effort.[625] Build volume is routinely over-bought: a first machine of 200 by 200 by 300 millimetres proved far larger than needed and traded resolution for a size that never got used.[655]
Upgrade paths have their own structure. Input shaping produces a large speed improvement, but it needs a controller capable of running it: a machine on an older eight-bit board cannot simply be given the feature, and the upgrade path also changes the extruder and therefore the stock of nozzles.[655]
The buying decision is properly framed as a comparison against a service bureau: for the price of a good desktop machine, how many parts could be printed by someone else?[312] The recommended test is to send five parts to a printing service; if the need does not survive that, it will not sustain a machine.[329] Two distinct service models exist: a bureau operating its own factories and industrial machines, and a brokerage that matches a file to whoever nearby has a machine and a suitable nozzle, acting purely as a transaction layer.[313] Board houses now offer printing alongside fabrication at prices that undercut owning a machine, with the part coming off industrial equipment rather than a cheap desktop unit; what ownership still buys is immediate feedback on fit and finish and a part in the hand today.[625]
Use
Product development and the workshop
The technology’s centre of gravity settled in professional product development rather than the household: consumer products designed today make heavy use of printing during the development process.[421] The technology’s practical value at a deadline is turnaround: when a contracted enclosure did not arrive in time for a trade show, Cyr’s team designed and printed a housing in five days so there were units people could pick up.[475] The uses that justify the machine are the ones no other process would serve: a holder for pogo pins is not something anybody would injection mould, and it exists because someone needed one.[244] Bench organisation is another such use — mounting hardware for DIN rail, cases for single-board computers and generic clip-on carriers — with the limitation that very small items still need a custom design each time.[705] Dentistry and oral surgery are among the heaviest professional users: a surgeon scans the patient’s mouth and prints jigs for each part being cut, so the procedure is set up in advance.[589]
Production
One answer to the slowness of individual machines is the farm: a queue that distributes jobs across many individually slow commodity printers, on the model of building large computing capacity from commodity servers, an approach that treats digital fabrication as a manufacturing method rather than only a prototyping one.[121] A heavy user likewise runs several machines rather than reconfiguring one: Bruton keeps eleven printers dedicated respectively to flexible material and to wide-nozzle bulk parts, because swapping material and head on a single machine costs more than owning another.[416]
Production use exposes the limits of the process. On a crowdfunded sensor product, Schappi’s operation printed hundreds — approaching a thousand — enclosure cases plus accessory housings for each sensor variant, which meant buying more machines and running them continuously.[189] The process also carried environmental requirements the workshop plan had not accounted for: the fumes drove the printing outdoors onto a balcony, which then made the operation dependent on the weather.[189] Raw material cost, at least, favours the process: costed per unit mass, filament runs roughly seven to ten times cheaper than construction-toy bricks of the same ABS, and pellets fed to an extruder are cheaper again by a similar factor.[369]