What do you think of magnetic linear motors like the one in the new Peoploly Magneto X? Does it help achieve substantial improvements in backlash?
BTW I would have expected accuracy would be quoted from a fixed reference point, that would coalesce all the repeatability scenarios you laid out into a single 'worst case', error-does-not-exceed value. (Are you implying with low accuracy but high repeatability you could get good results that are, conceptually, simply offset by whatever margin? Or that you'll get jagged edges / artifacts / incomplete or overdone cuts at scales below the stated accuracy threshold eg. if you have a corner approached by cuts from different ends?)
They still have backlash, but less so, it all depends on the masses of the gantry and the head and the speeds at which it moves. Whether it's a magnetic field or a belt doesn't matter that much, in principle there is some elasticity in the system and the height of the head above the gantry rails is a big factor in how much slop there is.
In practice you'll cut at speeds low enough that these things aren't an immediate issue, though when you start to cut cardboard or paper at near maximum speeds there likely will be some artifacts.
I will try to find the limits of the machine to see at what speeds these issues become apparent. Given the flimsy construction I'm amazed at how well it does, to be honest I had not expected it to be as immediately usable as it is.
Cutting 18 mm plywood with a < 0.5 mm kerf and < 0.1 mm repeat accuracy - especially compared to all of the other tools I have access to - is incredibly precise. In metal it wouldn't be all that impressive, but that's not what this thing is intended to do.
Anyway, the article wasn't intended as a treatment on CNC accuracy issues, there are many more that the GP hasn't touched on (such as: positioning errors due to temperature variation, which with aluminum frames can be considerable, and frames being out-of-true).
It's a dancing pig: it dances, that's the amazing thing, how it holds up compared to industrial machines that cost 400x as much isn't all that relevant.
I do think repeatability is very important. If you try to cut a circle and it ends up not even close to because you can't circle around a point twice without it being close to the same circle, ....
The rest i mostly agree with except maybe the 400x number. Seems high to get to a better level.
For the price point the repeatability is uncanny. Of course this machine is still reasonably fresh so we'll see how it holds up over time but on 10 complete passes over a work piece that spans 80% or so of the total work surface the last cut is dead on on top of the first.
I'm actually quite surprised at this, I did not expect that to be the case. But: this is my first laser (previous CNC tooling: Lathe, Mill, plasmacutter, the latter a homebrew affair at 8x4') and the main advantage that it seems to have over the other tools that I worked with is that the gantry and the head are fairly light in comparison to what you would normally expect. Even a plasmacutter requires a movable Z in order to compensate for warp (or you will definitely have material strikes).
So the head is probably < 1 Kg all together and the gantry < 10. This most likely is the biggest factor in how with such a light and - bluntly - flimsy drive mechanism it works as well as it does. It's got less backpressure than a pen plotter would have, basically just the rolling resistance of the rollers and the drag from the airhose and a thin electrical cable. I did add a segmented chain for the main gantry to ensure the cables and hose can never get tangled.
As for the 400x, an industrial laser from a brand with a good rep runs between 20 and 40K, these open frame lasers sell from anywhere between 500 and a thousand $US, I mistakenly added a zero too much so you are right about that! I spent a whole day on writing that up and was super tired, I slept a bit since and it's much better now.
In principle they do, as do servos, you'll find neither on machines like this. An industrial servo + driver (a single one) costs more than this whole rig.
I've designed CNC gear for a living and I'm aware of the way things are done in industry. I think this low cost approach opens up an entirely now domain and if you're making boxes, wooden toys and decorative pieces the accuracy requirements are much less critical than they would be if you were to make parts for aerospace or such, but nobody is going to attempt to do that using a rig like this.
Steppers with encoders aren't that expensive. Surestep is an example. 150 bucks for a nema34 motor with plenty of torque that won't lose steps. Nema23 is like 50 bucks.
That is the middle ground that won't lose steps but isn't as powerful.
I agree linear motors are well overkill for this. They are mainly useful when you need very high acceleration, which isn't true here
The drive electronics on the budget machines have no easy way to add feedback mechanisms. So you'd somehow have to either roll your own driver electronics or do major surgery on the existing board. Since I don't actually have a feedback issue at all right now I'll just leave it as it is but if this becomes an issue I will definitely look into it. I still have a small form servo set + drivers laying around from another project so it isn't as if I'm wanting for hardware. And it would be nice to finally put that to use, even though it is probably a bit much for this machine.
I haven't encountered other happy customers of it, but for a bolt-on closed loop upgrade I'd recommend MKS SERVO42 series. It's a hobbyist type product that comes from AliExpress no enclosures or safety protections, and slots right into RAMPS driver ports. It just needs an initial calibration, no software changes needed at the host side.
i feel like if brother can make an inkjet printer for a hundred bucks that gets positioning repeatability well below 100 microns with an optical servo, laser engraver makers can do the same thing. my capacitive digital calipers cost like ten dollars and are also in that precision range. optical, magnetic, or capacitive servos don't need to cost thousands
You have no idea how much this question has been vexing me! I gave up on the development of a public good product because I couldn’t answer that very question (low cost braille reader). I couldn’t get it to work seamlessly without high precision and couldn’t achieve high precision at low cost. Bought a couple of cheap inkjets and stripped them for parts, found proprietary optical strips and encoders, but still couldn’t figure out how they managed to machine/manufacture the plastic/nylon/POM parts to such high precision and still make a profit. In the end, I surmised they don't make a profit off the parts (though selling at a loss is illegal in the EU?) and rely on the cartridges to make money, but the bigger part of the equation is that they probably have those parts manufactured in massive numbers and with highly tuned and optimized designs carefully matched to the manufacturing process and the application.
I even put out ads trying to hire someone that’s worked as an (electro)mechanical engineer at an inkjet company to hire on a contract basis but got no responses. It’s possible those are mainly outsourced - or that the know how turned into domain knowledge that can’t be reproduced these days!
i don't think the nylon and delrin parts have to be high precision; the way i see it, all that matters is that the displacement between the optical sensor and the print head is constant and that the plastic tape isn't stretched so far that the printed page looks wrong, and that the print head stays more or less the same height off the paper and, more importantly, exactly the same angle
backlash, variable friction, motor power variation due to voltage, belt stretch, most flexions of the frame — all of that should just be 'external disturbances' that the negative feedback system automatically corrects. only the position feedback itself (and the time of actuation of the inkjets) has to be precise, that's the magic of negative feedback
as for the optical strips and encoders, i figured that a 600dpi laser printer printing on laser printer transparency film should be able to print a light/dark transition every 42μm, though it might take some fiddling to get that to actually work. supposedly 1200×1200 dpi laser printers also exist on the market for US$300. the standard way inkjet printers do this seems to be with a slit that's only slightly wider than the size of a single stripe, but a second transparency with the same 50% black pattern would also work, producing a moiré pattern (though with a viewing angle of only 25° or so due to the thicknesses of the transparent films). let me know if this is unclear, i'll make an animation or something
with a quadrature cycle (as the inkjet printer sensors seem to use, according to the datasheets i've managed to find) every 84μm you get a full cycle, so you get a transition every 21μm and you know your position ±10.5μm. that's half a thou, good enough for machining a piston
if you don't truncate the brightness to one bit, though, you can measure the phase within the cycle to probably within a tenth of a cycle, so you get ±4μm
as for who did the mechanical engineering, i suspect that it's something like ten people in the world, half of them retired. dissecting printers from different decades i see an astounding degree of similarity from one decade to the next
The one exception is probably if you want to screw bolts straight in without any kind of prep work (tapping) on the hole. Then accuracy matters, too much slop and your bolt won't hold or it will strip the material, too little and you may well end up snapping the bolt, especially a thin one.
Apropos machining pistons: the bigger issue with anything that needs a reliable 'Z' dimension on any kind of cutter like this (essentially a two-dimensional device) is that that third dimension is really only well specified at the point of focus. Outside of that it is more or less conical depending on the kind of cutter and the optics in case of a laser. Waterjet, plasma and laser all have different characteristics depending on what you cut with and in case of a laser the construction of the head and the kind of optics installed. Plasma also has work hardening effects that can not be ignored.
The only economical way to accurately cut large pieces of thick material is by using a heavy gantry mill or an EDM machine. Both will still be very costly and this sort of use is probably outside of the hobby arena anyway. If you need that kind of work piece I would suggest outsourcing it.
I read your response until near the very end and was itching to replying "but EDM!" before I got to your last paragraph! I actually was lucky enough to be able to use EDM for my initial prototype and I remain absolutely confounded as to what degree of accuracy, precision, and repeatability we're able to get out of this fairly old machining technique (and one that also avoids the z-depth issues you pointed out), but it has its drawbacks. It's insanely slow (though I don't know if machines made this side of 1990 are appreciably any faster) and it's too expensive for anything other than prototyping or one-off bespoke designs, and of course there are limitations to what materials you can cut.
> The only economical way to accurately cut large pieces of thick material [..]
Fortunately for most real-world applications the old maxim about size and required precision being inversely correlated tends to hold.
I was a die-hard subtractive machining zealot but I've slowly come around to appreciating 3D printers and they've made incredible strides in terms of capabilities and accuracy over the past decade. The hobbyist stuff still has some ways to go, but the exponential improvements are hard to ignore and I think it's become a viable suggestion for a lot of things were 2D machining used to reign king, at least where the end goal is to make something and not specifically to machine something.
The pairing of 3D printers and small lasercutters is like a fabbing super power. It's absolutely amazing what I can cook up in a matter of hours here on effectively 3 square meters.
But I still miss my machine shop :)
EDM is as slow as it was in the past, there have been incremental improvements but nothing that would make you go 'oh' and of course the waste in the wire is still as much a factor (and one that makes me dislike EDM but the capabilities are off the scale in terms of precision, cut depth and consistency, in machining everything has its price).
One thing that I've noticed the last couple of months is that you need to change your way of thinking about this stuff. If you 'can't make it' you need to think of what you can make and then adapt your design to that. This is far more productive than to stick to the 'proper' way of doing things. Suddenly two - admittedly - fairly crappy machines outperform my old shop in many ways. I really miss the ability that my 12 KW plasmacutter gave me in terms of cutting metal with accuracy and speed. But material hardening was a drawback as was the metal vapor and the conical kerf. By re-working some of the designs to use wood instead of metal and 3D printed parts where the 2D process isn't enough I find I can make almost anything that I could make before as long as it is for indoors use and strength isn't the main factor. Nothing beats metal and welding in that department.
Before getting a 3D printer and the laser cutter I would still cut metal, grind and weld pretty much regularly. But now it's a rarity, and I suspect that once I get the hang of high tech plastics that it will become even more so.
yeah, those inkjets always seem to use metal machine screws to hold everything together. i don't know how to tell how the screws (and, in many cases, nuts) are made but they do seem pretty precise
but i didn't mean to say that cheap inkjets contain no tight tolerances; they contain lots of tight tolerances. (the ones on the nozzles and on the traces on the integrated circuits are a lot smaller than the ones on the screws.) i meant to say that the in-operation movements of most of the parts of the printer don't have to be precise because negative feedback compensates for any errors they introduce
the printer doesn't make any screws or any holes in anything or screw in any screws, it just squirts ink onto paper, so there isn't a question of how precise the holes it makes are
Regarding your point about the two transparent strips: they'd be 180⁰ out-of-phase and directly atop of one another? Or would they have an angular offset wrt one another instead? I'm just not sure how the light sensor and light source would be arranged with respect to those colaminated strips. I do get the point about the viewing angle limitations, though. (Super-cool sidebar: I just learned there are optical encoders that use either of the Moiré effect or the Lau effect to make optical encoders that can track position in two dimensions simultaneously.)
The operating principles of my original prototype [0] needed at least some degree of precision in the mechanical components because I had actual mechanically interfacing/interlocking parts, unlike a CNC/laser/inkjet where the head is effectively traveling "unobstructed" in free air (in the case of a CNC, creating its own void to "float" in as it goes along). There were two separate positions that needed to be tracked, the linear position (this discussion) and rotary position (for which a basic rotary optical encoder or a servo could be used).
The design of the prototype itself (machining issues aside) was sufficient for its time (late '00s) where it would have taken the place of a (then) $10-25k braille reader PC attachment, offering more characters while being available for orders of magnitude less but the world has changed so drastically in such a short time that I've had to rethink the design to be less of a PC attachment and more of a standalone "braille eReader" sort of thing, significantly complicating the mechanics and increasing the precision machining requirements. It would be a "page" composed of multiple such braille reader rows, belt-driven and either (somehow) individually drivable so one motor could drive all the rows or (preferably, if the optical encoder BOM costs could be driven down cheap enough) with a separate motor per row allowing for faster "page refreshes" (esp. important because it takes ~no time at all for a user to finish a line of text).
Here the complication becomes switching from internally actuated to externally actuated "braille discs" in a way that allows manipulating each "cell" sequentially with a drive head that moves from the start of line to the end — but also leaves the cells in an immobile position so they're not free floating and don't change when a user glides his or her finger over them to any degree in the y-axis (instead of purely in the x-axis). Additionally the size of the optical encoder element becomes an issue because there is simply not much room to cram things between each row of braille text.
My first thought to allow me to solve all these in one go was to mount each braille disc on an "electromagnetic clutch" of sorts, but I was left aghast at the price of those -- and none were miniature enough for my needs. I then tried to go old-school and use an arrangement of actual miniature magnets embedded into each braille disc so they would maintain their position until externally actuated with enough torque to overcome the magnetic inertia, but failed to prototype that with sufficient precision and couldn't find magnets that would hold strongly enough while being small enough to embed in a braille disc (and forget obtaining them within budget, at least at retail values).
Had (and still have) other ideas but the time/cost difficulties in prototyping and the limitations on mechanical tolerances of the available prototyping methods really put a damper on things.
Thanks for the offer - I certainly would be happy to do that.
It's funny, I used to post about this on HN deliberately off and on for years and that never went anywhere at all but this chance response has led to the most fruitful conversation I've had on it here!
I've been in-and-out of machining and materials science for decades, it has come in very handy for the paid work I've done over the years but other than the windmill that I've built I feel that most of the tricks of the various trades have been wasted so if there is a worthwhile project to expend it on then I will be the one to be grateful.
The description of your machine has already made me wonder if it isn't feasible after all.
BTW I would have expected accuracy would be quoted from a fixed reference point, that would coalesce all the repeatability scenarios you laid out into a single 'worst case', error-does-not-exceed value. (Are you implying with low accuracy but high repeatability you could get good results that are, conceptually, simply offset by whatever margin? Or that you'll get jagged edges / artifacts / incomplete or overdone cuts at scales below the stated accuracy threshold eg. if you have a corner approached by cuts from different ends?)