I think other kinds of automated fabrication might serve as well as 3D printing or better for general-purpose home use.
There are several kinds of automated fabrication. CNC machining is pretty important. Lithography is pretty important; it’s how both ICs and printed circuit boards are made, and in one or another form it’s how almost all text and pictures get onto manufactured goods. Robot welding has been a major part of fabricating many kinds of metal goods since the 1980s. The first versions of CNC machining were cam-controlled, in the late 19th century; they were called “screw machines”. After setup, you would feed them a piece of bar stock and they would conduct a programmable series of lathing operations on them, repeatedly, under the control of the cams, which is how screws have been made ever since then and why screws are cheap.
The promise of 3D printing was, when 3D Systems was founded in 1986, that product designers could rapidly produce prototypes to experiment with, even if the cost was far too high to compete with mass-produced products. My math teacher told the class about it in 1993; he’d seen a plastic apple produced by stereolithography.
In 2005, Adrian Bowyer founded RepRap as a project to make automated fabrication accessible to the masses. Its goal was to make “rapid-prototyping machines” so cheap that it’s cheaper to make things with them than with injection molding, by making them self-replicate. But it failed at that goal.
RepRap-style 3D printing — 3-axis gantry open-loop single-material FDM without a heated chamber — is really limited in what it can achieve for a variety of reasons.
Most of these are problems that would not be serious if the printer could self-replicate at a low cost, as was the original goal — it could print a new version of itself that solved the problem.
RepRap-style 3D printers typically have 100-micron errors even in the X and Y directions, and due to the nature of FDM, the Z-axis of each piece is quantized to typically between 250 and 500 microns, resulting in a stairstep appearance.
This doesn’t sound like a lot, but it’s common for any ordinary subtractive machining operation to be carried out to 20-micron or 10-micron precision, with 1-micron precision in areas that need it. An ordinary 600-dpi laser printer has 40-micron precision; an expensive 2400-dpi printer has 10-micron precision. 1-micron feedback sensors are easily available, but RepRap-style printers don’t use them. Why? I think it’s a design error, but I think the intention was to lower costs.
The big problem with imprecision is that it makes the properties of the material unpredictable, because you get unpredictable layer adhesion. Poor layer adhesion gives you a tiny fraction of the potential tensile strength of your material in the Z direction.
Two of the main sources of poor layer adhesion are actually bed-leveling problems and filament diameter miscalibration.
Recent RepRap-family printers have ceded somewhat on the open-loop-control front to do automatic bed leveling by touching down on each corner of the bed before starting the print, then twiddling the Z axis dynamically as the extruder moves around in X and Y to compensate for bed tilt. This ensures that the first few layers deposited have a uniform thickness, greatly reducing layer adhesion problems. Without bed leveling (either manual, using screws, or automatic as described above) it’s easy to have a bottom layer that’s squished flat on one side of a piece and barely touching the bed on the other, with lots of roundness in the deposited filament.
Filament diameter miscalibration also affects layer thickness and dimensional precision; if the filament diameter is 10% smaller than expected (for example, 180 microns if your filament is 1.75 mm diameter, the most popular size), then the amount of plastic per linear millimeter is 21% smaller, and so the width of extruded traces will be 21% low, and you may get poor layer adhesion. Many commercial filaments have significant diameter variations even within a single spool, and they are rarely perfectly round, making precise measurement difficult.
Imprecision puts a severe limit on volumetric resolution. I’m particularly interested in volumetric resolution because, for mechanical computation, that’s what counts — if a certain multiplier design requires 4 million voxels, then it will be 10.5 milliliters on a RepRap-style machine or 0.5 cubic millimeters (half a microliter) on a hypothetical 5-micron-resolution 3-D printer.
With FDM without a heated chamber, your materials are basically limited to PLA, ABS, and nylon, and filled or colored versions of these. (Except see below about PVA, elastomers, and PETG.)
These materials are not all bad. They come out of the machine sterilized, which is nice, and PLA and nylon are biocompatible; and the most interesting fillers I’ve seen are brass and fine sawdust, which give you the ability to produce objects that are mostly brass or wood. PLA is very dimensionally stable, ABS is very strong, and nylon is very tough and easily accepts very large deformations (30% elongation at break is typical).
But they do have many disadvantages.
PLA is weak and brittle.
But ABS and nylon offgas toxic fumes during printing, so you likely don’t want to print in them indoors without a good ventilation system. ABS also leaches toxic styrene when in contact with acid, so you can’t use it for food-contact applications. You can’t use PLA for most food-contact applications, where you would expect it to shine due to its nontoxic nature; the problem is that it softens at a very low temperature, as low as 60°, even though you need at least 180° to extrude it.
PLA is hygroscopic and aquadegradable, so by leaving a PLA spool exposed to air, you can run into big problems printing after a few weeks or months. The water in the plastic can boil and make bubbles as it goes through the nozzle, or it can just degrade the plastic, making it even weaker and more brittle than it already is.
ABS and nylon require high temperatures, like 210° and 240° respectively, which PLA can’t withstand; this is a problem if you switch from printing in nylon to printing in PLA, because you have to print at nylon temperatures until all the nylon is out of the nozzle, because it’s frozen solid at the usual PLA temperatures.
ABS, because of its larger thermal shrinkage and high stiffness, has a terrible tendency to curl as it cools without an enclosed build chamber; worse, this depends on the ambient temperature, which means your ABS printing may be going fine all day and then fail during the night. If it curls up off the bed, it can collide with the nozzle and be knocked entirely out of place, but even if that doesn’t happen, you can get delaminations which totally vitiate the strength of your print.
PLA is vulnerable to glass-style fatigue; a piece of PLA held in a stressed position may crack through after weeks or months with no motion. Glass does this too, but mostly only if it’s underwater. In both cases there seem to be critical stresses below which the fatigue cracks don’t propagate. Probably whether this happens to a given specimen of PLA depends in part on the amorphousness (glassiness) of the PLA, which in turn depends on the composition and cooling rate.
I don’t have any experience printing in nylon, but it’s very soft, which limits its applications somewhat. And the higher temperature makes the hotend fail more often.
FDM prints unavoidably (?) come out with a stairstep surface because of Z-axis quantization. If they are ABS, they can be smoothed by brief exposure to acetone vapor; this is harder to achieve with PLA, because PLA is invulnerable to almost all organic solvents. People have had success using dichloromethane, which is outrageously toxic, and reportedly ethyl acetate, which is several times less toxic than table salt. But the ethyl-acetate-based nail polish remover I tried didn’t do the job, it just swelled the plastic. Other people just recommend painting the piece.
Thermoplastic elastomers like “Filaflex”, “Ninjaflex”, and “Flexifil” are a promising new addition that I don’t have any real knowledge of; they don’t have the level of superelasticity that rubbers do. Ninjaflex is a thermoplastic polyurethane; I don’t know what the others are made of. Supposedly Filaflex has 70% elongation at break, which is quite a bit more than nylon’s 30%, and Ninjaflex 660%, but neither equals an ordinary rubber band, which can easily hit 1000%.
PETG filament, like Form Futura’s HDglass, is another newish product that claims to let you use FDM to fabricate transparent objects. I don’t have any experience with it. Being PETG, it ought to be food-safe, too.
Standard RepRap-style machines have a single extruder nozzle, fed by a single filament, in order to lower the cost and improve reliability. That means it can only print in a single material at a time, and changing between materials is far from instant. This creates a number of problems.
Most obviously, unless you paint the piece, it’s a single solid color, or it’s the uncontrolled sequence of colors that the filament had — like when you knit with variegated yarn.
I think the single-material limitation is the main thing that prevents RepRap replication from really taking off. The single most failure-prone part of a RepRap-style printer is the hotend, and that’s the part it can’t print, because it’s made of brass, Teflon, Kapton, and a cartridge resistance heating element, which don’t melt at the temperatures it’s exposed to.
(Somewhat less obviously, the hotend has a dimensional tolerance that’s tighter than RepRaps are usually capable of achieving: the diameter of the nozzle aperture.)
Dual-extruder FDM printers are capable of doing several things that single-extruder models can’t manage, aside from being theoretically better suited to self-replication. One is printing objects that have controllable visible patterns on their surfaces. Another advantage is the ability to print support material in a soluble form: polyvinyl alcohol (PVA), a nontoxic water-soluble plastic that melts at a conveniently low temperature, is a favorite choice. But ABS, which easily dissolves in acetone, is a reasonable alternative for supporting PLA, which is invulnerable to acetone.
Dissolvable support material is an enormous help in making preassembled mechanisms. The usual RepRap-style support material has to be cut away from the finished piece using a knife, which requires manual intervention, leaves crap on the surface, and is difficult to impossible inside enclosed spaces. Also, it’s easy to cut your hands that way.
Here’s an example of using a dual-extruder FDM machine for dissolvable PVA support material.
Finally, there are a lot of kinds of machinery that inherently require two or more different materials to be deposited in intimate contact. Circuitry requires both conductors and insulators; PLA filled with graphite, lead, or more expensively, brass, silver, or gold, is totally adequate as a conductor for many circuits, but you need insulators too. Electric motors could conceivably be 3-D printed, but as far as I can tell, practical motors need many thin layers of conductors, insulators, and high-magnetic-permeability materials. (Those are typically a poorly-conducting kind of steel, deposited in layers with insulating epoxy in between to reduce eddy-current losses.)
Many of the more interesting possible applications of 3D printers are in metamaterials; by careful design of a repeating 3-D structure, for example, you can combime two materials with different thermal coefficients of expansion to get a bulk “metamaterial” with a TCE that’s orders of magnitude lower than either of the two “phases” or materials that compose it. That’s super important for precision measurement, feedback, and control.
As I said above, the single-extruder RepRap design is one of the major obstacles to self-replication. You could imagine an alternative RepRap design that used two different materials for two different nozzles, plasticized using two separate mechanisms, such that either nozzle could print the other nozzle — for example, one material plasticized using heat, and the other plasticized with a solvent.
But multiple extruders is not the only possible solution.
You could also imagine a multi-step manufacturing process, where for example you first use FDM to fabricate a shape in PLA, then make a plaster cast around the PLA and use it to cast, for example, brass. You can find videos of people doing this on YouTube (“lost PLA process”).
Or first you fabricate a shape in stainless-steel-filled PLA, and then you sinter it in an oven, or something. There are 3-D printing companies that do this with selective laser melting of stainless-filled thermoplastic.
Or you could use a material whose cradle-to-cradle life cycle includes more than two steps like PLA’s melting and hardening. Plaster of Paris, for example, which you can deposit as a thixotropic liquid freshly mixed from powder and water, then allow it to harden. The full cradle-to-cradle process for plaster of Paris involves calcining the resulting hydrated gypsum at 150° to 180° and then grinding the resulting calcium sulfate hemihydrate to a fine powder. As long as the “nozzle” isn’t doing the calcining, it won’t have any trouble making an alternate nozzle. Which is a good thing, because once you let plaster harden in the nozzle, you’ll need to replace it.
A fully autonomous self-replicating system will probably need to be able to make ceramics from clay and fire them, because essentially all industrial processes depend on ceramics. This might be a good place to start.
RepRap-style printers use open-loop stepper-motor control. This results in three problems: imprecision, cost, and unreliability.
Open-loop stepper motors have poor precision (or, let’s say, poor bandwidth) for reasons I don’t fully understand. This is one of the sources of the problems in the “Imprecision” section above.
The beefy NEMA 23 or NEMA 17 stepper motors used in a RepRap are rare to find at salvage and quite expensive to buy new, because they’re mostly used in industrial machinery, not consumer goods. Older 2-D printers used steppers, but much wimpier ones; newer ones use DC motors with closed-loop control.
The printer uses such large motors because the motors have to be strong enough to move the dynamic loads of the printer with basically no effort, because they cannot adjust the effort to the load without feedback. (A certain limited amount of feedback is provided by the variation in slip in the stepper.)
Finally, steppers add unreliability, because if the printer ever encounters resistance and misses a step, all subsequent movements will be offset by the missed step, resulting in a failed print, typically spaghetti.
The gantry construction of RepRap-style printers amplifies the problem of imprecision.
Traditional CNC machine tools use a "gantry" geometry in which high-rigidity linear actuators at precisely aligned right angles determine the relative positioning of the workpiece and the cutting tool. With a gantry, you can make precisely parallel and right-angled cuts by hand, without doing a lot of numerical calculation. But gantries require immensely stiff beams and joints to achieve reasonable precision, linear slides (which are difficult to make and even more difficult to make precise), and linear actuation, which typically adds backlash and therefore imprecision.
Fortunately, less stupid geometries are coming into fashion, including deltabots, parallel SCARA topologies, and polar robots.
Gantry construction synergistically causes problems with open-loop control, because the precision of open-loop control depends critically on rigidity, since the open-loop control can’t correct for mechanical deformation any more than it can correct for missed steps.
My favorite robot geometry, the Stewart platform, has not become popular.
CNC laser cutters are actually far more accessible than 3D printers today, in the sense that there is likely a CNC laser cutter physically closer to you that you can use, and if you make something on it, it will cost you a lot less than doing the same thing on a 3D printer. It’s just that it’s limited to planar fabrication (as are CNC plasma torch tables, CNC waterjet cutters, and lithography per se.
Things like the Othermill at US$2200 are only a little more expensive than RepRap-descended 3D printers; the PocketNC FR4 is a fully-funded Kickstarter for a US$449 three-axis CNC milling machine, which is cheaper than many 3-D printers. The FR4 is milled from epoxy-glass-fiber composite circuit boards, which are very rigid, light, and cheap, compensating for its stupid gantry geometry (see below for the problems with gantries).
You can see that this FR4 is capable of reproducing its own circuit boards and the parts of its own structure, and has 5-micron precision in all three axes. That is about 20×20×50 = 20000 times better volumetric resolution than any open-loop-control FDM 3D printer.
Norbert Heinz built his 100-micron-resolution CNC machine entirely out of recycled junk, including closed-loop control of the underpowered DC motors he used. In that video you can see him using it to engrave glass with a diamond cutter, cut Styrofoam with a soldering iron, and mill gears from acrylic or aluminum, all of which are materials that are out of reach with FDM.
With FDM, you can’t make anything transparent, anything as strong as glass or acrylic or aluminum, or anything as light as Styrofoam. And, at the 100μm precision achieved by both Norbert’s junk heap and a thousand-dollar FDM 3-D printer using open-loop control, your gear surfaces suck, and you need a lot of extra clearance to print preassembled gear assemblies. (Although, of course, with CNC cutting, you can’t print anything preassembled.)
Norbert’s machine is 500×500×?? mm rather than the 200×200×150 of a typical RepRap-derived 3D printer, so it can make things that are about ten times as big.
Michal Zalewski has written is a superb introductory guide to CNC machining and resin casting. Remember what I said about a multi-step manufacturing process? His multi-step process is that he cuts a mold or a positive, casts a mold if necessary, and then casts the part. He gets 2-micron accuracy.