Sometimes. REALLY depends on what it is. If it's something like this that demands precision, absolutely. But for example, a little plastic fastener on my mom's shifter linkage broke, decoupling the stick from the push pull cable. The part should have been cheap, but was proprietary and only sold with the entire shifter linkage. I think it was $250. Was able to fix it with a little nylon bushing, a drill, a 1/4" bolt and a stack of washers. Cost? $0 as we had everything on hand. Even if I'd had to buy it all, including the drill, it would have been a fraction of the price lol.
yup. I use this to explain why digital *cannot physically* replace analog, and vice versa. Usually trying to explain to some C.E.O. with a buisness or banking degree all the reasons why thier maliligned attempts at full automation failed miserably. By the time you get through the required input channels for actually simulating an single neruon in the analog brain to any degree of accuracy they start to get the picture, but to be sure I take the width in copper atoms and have them translate that to Km, then Imperial miles. Its 55 km by the way. Per nural connection... Assuming a single width of copper atoms can form a wire (it cannot). Some of that can be made up with faster translation times but. not all. There is a reason ChatGPT takes up a giant wearhouse and loses two bucks for every buck they earn. That analog brain will always be 5% off. But that 5% takes a LOT of resolution to simulate.
@@jakedewey3686 GPT models, and learning are *being sold* has full simulation of brains and neurons etc, and are being funded by people attempting to do so. As such, and perhaps unsuprisingly the money men leading various sectors are expecting results of full simulations of human brains to allow for full automation. Now that said, I absolutly can fully automate most if not all non-biological production with an big IF. (I.E. yes I can build a 3d printer that could print a car, a computer and every nail and outlet needed to build your home. just not mango chunty, wood, or a tomato. Bio stuff.) .... IF production is willing to go with a much MUCH slower method of production and take up orders of magnatude more space. Its honestly rather trivial at this point to do, but nobody wants the responsability to fully fund the Research part of R&D they would rather redevelop over and over in response to miniscule stock market flucutations. Also, ya know what, everyone should know this at this point. Make a 3d printer with slip-stick nanodrivers, they are open source and 3d printable. Thats the gist of it, you need diffrent application methods for diffrent materials but one that works for most is to stick the whole thing in a vacuum and build an emiter of some sort really just a small aprature blocking the path of a heated chamber filled with the materal you want to print with. Another good one is just electrplating through a small little capilliary tube as the printhead. They have enough precision to even print a computer, the layer size is small enough that you can mimic any crystaline structure using eaither of these. But your layers will print very very slowly.
My dude I thought this was a channel a year or two old. If this is truly the first video I am amazed by both quality and the sheer traffic you are getting. Hope you are not too overwhelmed, take your time and enjoy the process your doing something right according to the algorithm don’t be a slave to it quality is likely to carry you through many an algorithmic storm.
Yeh! Imagine my face when I was wo, this channel is awesome. Let me digg all the videos he has... To just realize this is the only one! Hope you can grow and get big with TOT inspiration! Let's go man!
I would've just used a belt reduction. I don't think the torque transfer is great, but you can tighten that belt down so much it reduces almost all backlash
@@Nevir202 this is how almost every 3d printer works, and they last for years with continuous use with high torque and accelerations. a belt reduction would be the way to go. and microstepping does give u finner than 1.8deg precision
I ran into a similar problem when I was trying to make an equatorial goto mount. The RA axis needs ridiculous step precision to ideally keep steps smaller than the area in the sky that a single pixel covers at the given focal length, I can't remember the exact step size, but I had to use 1/32 microstepping(since this is a constant motion, microstepping does help) and add a 224:1 gearbox ontop of that. I used a split ring compound planetary gearbox, neet stuff those are.
Yeah that astronomy stuff really does require a lot of resolution. I did come across those split ring gearboxes and almost tried to make one, but the strain wave just looked so much cooler when running. Did you ever get the mount working?
@@UnnecessaryAutomation yeah it's awesome, and quite ridiculous. It takes some hundred of thousands of steps to rotate fully. And with helical gears it has no noticable backlash.
@@U_Geek Did you consider a brushless DC motor and a high precision rotary encoder instead of a stepper? That sounds like a decent approach if you want controlled constant motion. Full disclosure - I haven't done it, and there might be problems I haven't thought of!
@@paulwomack5866 I have but I really didn't wanna spend too much on it and a stepper with ridiculous gearing was cheaper since I had most of the electronics. If I ever decide to upgrade it that is definetly the motor of choice, but I do have to say I appreciate that I don't need a counter weight to balance.
This video is so good! You've got a great personality for these kinds of videos, you've nailed going into detail while keeping it easy to understand, and the animations are already incredible. The only feedback I can think of improving is a better mic and some color grading/better camera, and after that you'd seriously be on the level of channels with millions of subs.
I agree it's a great video, but please try giving more attention to audio. Various parts have changing levels of high frequency noise and background noise.
I think I heard somewhere that you should record the "silence" of a room to use in between the audio tracks, because absolute silence is always different and distracting
You can try using the microstepping function of most stepper driver ICs to achieve resolution below 1.8 degrees. I've used the DRV8825 with 1/32 microstepping and it theoretically should be able to get down to 0.05 degrees per step.
It clearly said in the video (without explaining though) that microstepping does not increase resolution. The truth of this is presented in the end (without specifically mentioning that microstepping has the same issue, although it is obvious).
It really depends on the driver chip that he is using. When using DRV8825 you have 3 config pins and when you pull them to the highest setting, every pulse on the STEP pin will be 0.05 degrees on a standart 1.8 deg/step motor. A4988 will do that too, but if I recall correctly the microstepping goes only up to 16. As the article says, it will of course have less torque using microstepping - but he even says he will not be rating the torque difference.
@@ntomata0002 so microsteps don't have a repeatable positioning? I know the sentence is weird but you should get the picture, I have a blockade right now thinking about a better way to say it XD
@@Ramog1000There are a lot of reasons that microsteps are not accurate, some to get the picture: a) The motors are not electromechanically constructed to do accurate microsteps. b) I 'll explain this with an example. Let's assume that you have an 2A, 1Ntm motor and you provide 2A in one phase. The motor aligns in this position. But if you apply some load torque, the axis will deviate from its zero point. Applying more torque will make it deviate more until you get past the 1Ntm limit and the motor will kick.The kick will happen after the motor passed 2 steps away from zero torque position. So, any torque applied to the motor, either load or friction will result with an unknown deviation from the theoretically expected position.
As a historical note (~40 years), there used to be a company making steppers which used the priniciple directly. The rotor was the ellipse, and the stator deformed. The input sequence was different as well, using 8 inputs with 4 consecutive inputs being energized and the 4 shifting by one per input every step. It was called a harminic drive stepper, although the term is not apparently used that way nowadays. It seems to me that, since the shaft/ellipsoid rotates within the inner gear that friction would be an issue and might limit the step rate.
1. Timing belts can be an alternative to complex gearbox with backlash control. 2. Better option is linear scales to provide accurate positioning. Even if the gearbox has no error, it does mean the screw drive is without error, or if the stepper motor skips steps (ie not moving in step with driver step pulses because of too fast, or not enough torque for the motor speed used).
I thought cycloidal and strain wave are basically the same principle, just implemented slightly differently. They both fundamentally come down to a cam and a slight mismatch in the number of gear teeth (or lobes) between the input and output. I think the main difference is a cycloidal is easier to 'back drive' which is a bad thing in many applications
Not sure if this technique applies directly to stepper motors but it may give you an electronic instead of mechanical solution. My experience is with closed loop servo motors. The way the windings are designed you can "Step" them around manually by connecting dc voltage to any 2 of the 3 windings. as an example with the phases labeled U V W, +U-V moves to the first step, +U-W to the second, +V-W to the third, +V-U to the forth, +W-U fifth, +W-V for the sixth position and then +U-V brings it back to the electrical first position in the series. there will be one set of 6 lockups per pole pair in the motor (number of total poles divided by 2) so a 2 pole winding will be steps of 60 degrees mechanically, 30 for a 4 pole, 20 for a 6 pole and 15 for an 8 pole. The thing is you can do half steps by connecting 2 phases to the same leg of the circuit. Basically you connect the next leg in the series before you disconnect the previous one. Going half steps it would be +U-V first, then +U-V-W, then +U-W and so on through the cycle. This lets you get 7.5 degree mechanical steps on an 8 pole servo instead of 15. You may be able to do something similar on a stepper motor to get half steps. In an actual in use servo the drive is not just stepping through positive and negative voltages but actually modulating the voltage to get the precision to hit any arbitrary angle. if its asked to turn the shaft smoothly at 1 RPM the voltage would trace a very slow sine wave on each phase. That level of modulation might be beyond your control boards ability to produce but I'd wadger it could connect the negitive or positive voltage to 2 wires at once to get half steps if the stepper motor behaves the same way. Just something to try.
I need one of these. I'd attach a vertical laser tool to it, then rotate it to hit a target. Then this tool would show the straight line on the floor leading from the origin to that target.
Fun fact: similar to the strain wave gearbox the stepper motor has 2 more teeth on the rotor than on the stator creating a reduction from electrical rotation to mechanical rotation.
9:37 I believe the encoder also has a resolution. In short the next step on the motor is closer to the next step forward on the encoder than to the step back. If you rotate the encoder position you will probably see the lines move up and down in parallel. 10:09 The extra noise might be because of the additional play introduced by the gear box.
You pretty much misread the article that you linked... There is definitely microstep positioning resolution, with SOME distortion just a little with a well behaved driver. The deflection under load is incidental but it's equivalent to backlash basically and doesn't depend on the step size. It's not like a gearbox doesn't introduce a positioning distortion (error) of its own...
Loved the video. A few critiques. Try and get your audio levels normalized between your CAD and talking hands section and keep your hands in frame. I know. Nit picks but thats some easy money to make things a bit nicer.
Darn it! I been looking for this exact kind of mechanism for my digital handheld monochromator for ages. Nobody makes any reduction for a Nema 8 from what I could find.
Kudos for building a strain wave gearbox! I've thought about them myself for another application. I did do a planetary gear set with herringbone gears that when properly meshed had no perceivable backlash. But I did not measure with a high precision device. Have you considered trying a hearing bone planetary gear set?
@UnnecessaryAutomation Well if you decide to try out a planetary set, they're very easy to make. I don't know what CAD software you're using, but in freeCAD, The gears workbench makes it extremely easy to render the gears.
You don't really need a 'high precision" device to measure backlash. Just attach a long pointer perpendicular to the end of the shaft - the longer the pointer, the more precision you'll get.
How cool! I had the notion that these would tear themselves apart of 3d printed! So cool that you figured this out. Could you run out to test longevity.?😊
I'd be curious how much deflection you see applying a small load to the shaft for just-the-motor and the strainwave case- I'd expect the print has more flex / less rigidity to it. Also, a different way to handle the backlash of a geared system would be a preload spring / some-manner-of-preload. Though that may not fit the project you have in mind.
Hello! Nice video! Just a little problem for me is the "background noise", starting from the begining. As it seems constant in frequency and intensity, I guess it might be easy to remove in post prod :-)
Hmm... I used to build very low flow pumps - 5 or so microliters per minute, although it would go down to nanoliters for mixing. I'm on the mechanical end, but our electrical guys said micro step improved the resolution too.
Its important to document your mistakes to avoid overs doing the same, you show all the problems you encountered in making the gearbox to help others who wish to do the same, probably in a differant video.
I think the offset may come from the backlash and may be a measurement of the backlash. When the teeth push fireards, and then go in reverse, moving the distance of the blacklash and then hitting the reverse direction tooth. Maybe that's the cause? Idk tho
I was really surprised to see so much noise in the gearbox output, what do you think was the cause of that? I was thinking that the periodicity of the noise would match the number of teeth in one of the gears, but I counted ~40 peaks in the noose plot and only I think 32 teeth in the outer ring gear. Hmm - but you said it’s a 20:1 ratio, so maybe the noise frequency is 2x the ratio? I suspect it has to do with how the gear teeth engage and disengage with each other. Maybe they should have a different profile than normal spur gear teeth, tailored to how they engage as the peak of the strain wave approaches, to maintain more of a constant torsional force as they engage? 3D printed strain wave gearing seems like a potential gold mine for a lot of applications needing affordable precision and good torque, and you’ve done a great job of designing this assembly - I’d love to see you dig into it more deeply. (But then you were working on an application, not a research project 😁) Great work, thanks for sharing - this is the first time I’ve seen a plot of angular precision like this!
Stepper motors don't have magnets on the surface of the rotor. They just have iron teeth that line up with the iron teeth on the stator. Most are "hybrid" steppers that have one large axially magnetized magnet in the center, and then one set of rotor teeth at each end, slightly offset from each other. This is why you feel the cogging effect even when the motor has no power. Also why you can generate moderate voltages from spinning a stepper.
Amazing production quality! Is it possible to get a closeup / slomo shot of the gears? It was a bit hard to follow while turning so quickly. Looking forward to seeing what you use the gearbox for!
In both cases the error you're seeing is relatively predictable. More-so without the gear box, where you could pretty easily overshoot and reverse back to choose between the 2 error directions. But even with the gearbox those spikes are pretty repetitive, you could probably probably predict them in software with some reasonable accuracy, and spend some of that extra accuracy you now have to compensate for the predicted error a bit. It might behave differently under load and in other conditions, so that might mess up any predictions though, it might take some work to get right.
Wouldn't adding feedback loop improve precision? You already have the sensor, so you can use it to correct for positional errors. As a bonus it will also account for lost steps. I guess the viability of this approach depends on what you are building this for.
that article was made in 2016, micro-stepping today with a tmc2209 or similar would yield usable results, especially in higher resolutions than 1/16 The 2209 supports 1/256 microsteps
As he pointed out, at higher microstep ratios the incremental torque becomes less, and is more likely to cause problems under load, especially direction changes.
If you want to make a harmonic drive just say you want to make a harmonic drive. Don’t say that microstepping is not actually increasing the resolution of the motor as that is false. You can create magnetic states between the poles through mircostepping that allow for higher resolution.
Great video! :D You should note that magnetic encoders are very sensitive to alignment, and do have a natural hysteresis, and with the 3d printed bearings, I wouldn't put too much trust in the magnetic encoder being held as well as it could be. you might find significant improvement with better bearings + somehow mounting the magnet perfectly concentric, but it won't fix the natural hysteresis in the encoder. A lot of the issues with magnetic encoders are buried or outright omitted in the encoder chip's datasheet to many engineer's frustrations 😅
Thanks! Yeah I came across a lot of those issues with the encoder and ended up doing some software stuff to compensate for the alignment error. I had some of that in an early draft of the video but I couldn't get it to flow well so it got cut. I didn't know about the hysteresis though, that would explain some other things. Like you said, not much about this in the datasheet 😂
Next time you want to measure small angles with very high accuracy without doubt, do it like the old masters. Use a laser and a mirror. Attach the mirror on the axis of rotation, set the laser to aim the mirror and reflect to a wall. The reflection of the dot to a wall 2m away will give about 70mm movement for 0.5° of rotation. Easy to visualize and get accurate resolution of even 0.01°.
Yeah I didn't catch that until too late. The gearbox I used for that demo was an older prototype with a different ratio. But the final version on printables is the 20:1
Not sure if you already know that, but to get the audio quality up (and noise down) you need to either buy a decent mic or at least remove background noise with software like Audacity (free). For the video it is mostly about your white balance. Your hands certainly are not blue. Perhaps get a grey card to get you started. Once you have a feel for it, you can do it by hand. Or just always use a fixed setting on the camera and make sure that all your light sources are the same to avoid any post-processing with that.
@@UnnecessaryAutomation I recommend essentially any video here on youtube on that topic. Search specifically for white balance, as color correction is something different and unlikely to be needed here.
The wobble in presicion most likely comes from an uneven gear surface. If you have access to it, you might want to try this design with resin printed parts, since they have a much better surface finish, than fdm parts.
I actually did try a resin version, well except for the flex gear since I didn't have a good flexible resin. It didn't perform that much better so I stuck to FDM
@UnnecessaryAutomation have you tried undercuring tough resin? Formfutura tough resin can be hard as pla or gummy like tpu depending on the exposure time and temperature during print (hard >7 sec on an elegoo mars 2 pro, soft ~6 sec). Aftercuring and storage for at least a year won't change the propperties noticable.
The Strain Wave Gearbox is a cool approach and I may try this with one of my automation projects. One question I do have is about the gearbox temperature and ware over time?
Interesting! I was intrigued by the wave pattern of the resulting precision, went back to the number of teeth and found the circular spline to have 32 teethe and the flex spline 30 but the number of waves is 40. I assume the waves are produced by one of the gears. Did you use a different one for the testing than shown in the video to demonstrate the mechanism? To get a reduction of 20 one would need a 40 tooth flex and 42 tooth circular gear, so is it the flex spline causing the waves? Just a guess: could the 3d print surface imperfections be the cause of the waves? Print lines interfering with each other in other words. Maybe printing the 3 components with different layer heights might reduce the waviness.
In industry, there is a leap to servo motors with encoders. Yes, they are not as cheap as steppers, which is why steppers are so common. A typical motor used has a 500 line encoder. With an incremental encoder, the sine and cosine, called the A and B encoder signals, one full line of the encoder has 4 steps of the movement of a single encoder line, for a native resolution of 2000 steps per revolution. In photolithography equipment used in exposing semiconductor layers, decades old equipment resolved to 0.5 microns. Newer equipment to register modern semiconductor layers, sub nanometer resolutions are used. When in fast motion, the encoder signals are well into the RF range of several megahertz. Disclaimer, I do work in this industry. Laser interferometery is used for positioning far more accurately than optical line based encoders could ever do. 32 bit resolution on a linear motor is typical of older than 10 year equipment. New stuff is proprietary trade secrets.
You can also get closed loop steppers these days, which are servo motors too. Anything with a 'closed loop' motion controller is actually a servo. I think closed loop steppers are cheaper than servos generally as the motor controllers are a little simpler
Seems like the gearbox adds some phase lag to the reference angle. Should be pretty easy to account for it I wonder what the error is when doing mean variance...
Do you have a high torque requirement? If not, why not just get a 0.9 degree stepper and a simple belt drive for the reduction? Or a cycloidal drive? Strain wave is really neat and I admire your grit working through all those prototypes!
Most stepper drivers allow for micro stepping. Thus dividing a 1.8 degree step by 10. Increasing your steps per revolution 10 fold. Also timing belts and pulleys is another way for gear reduction without introducing backlash. Last option is spring loaded worm drives.
Microstepping gives a reduction in torque, especially holding torque. Microstepping is driving two sets of coils of the motor partially to move the net magnetic field between the usual steps - because the coils are driven with less than full current, you end up with less than full holding force too
Hello, I just wanted to say that I really enjoyed your video! I was wondering if you could kindly let me know what animation software you use for your videos. I’d really appreciate it. Thank you!
Looking at the trace from the gearbox it's clear that the error is cyclical, probably from every rotation of the cam. The 0.6 degree error is total but at each position the actual error is very very small. You should be able to compensate for most of it in software and get 0.1 degree precision or better.
Ok but according to the hachaday article, microsteps only weakness is error in open loop and torque. You could mitigate the first problem with hall sensors, there are some closed loop solutions for 3d printers for example. And for torque - what do you need the stepper for? I feel like for tiny and precise movements you wouldn't need too much torque, but again, idk what you need it for.
@@UnnecessaryAutomation There are closed loop nema 17 motors that have an encoder built into the back which take up less length than a gearbox while also providing more torque than a standard stepper as it can briefly overload the amps when needed, also it is worth noting that the hackaday article is fairly dated and does not test more modern drivers like the tmc 5160 or 2209.
@@UnnecessaryAutomation i got a 3D printer with 0.9 degree steppers and TMC 2209 drivers and with 16 microsteps the torque is enough to not let me move the gantry. And as someone mentioned, modern closed loop systems would take much less space than a gearbox.
This New Tony
Less nail biting for sure
Just needs some jump cut edits ✂️. Edit: 7:08 seems I spoke too soon! New Old Tony confirmed 👍
So I guess this is now the This is the New New Old New Tony show?
This new Tony...
Electronic This Old Tony is that you ?
I wish I was that good lol
@@UnnecessaryAutomation You just need to work on your Kung Fu.
No shit, huh? He's been practicing and it shows. He's nailed it good
I thought This Old Tony made another channel 😅.
Same immediate thought but the hands are different. Good video!
Nice to see that This Old Tony found the fountain of youth. Welcome This Young Tony :D
All that time travel is showing some side effects!
The algorithm knows… everyone here thought the same thing when they first opened the vid. This young Tony. Keep up the good work.
When you said a lot of trial and error, I didn't expect a full box of non-working prototypes. That's crazy.
And that's not even all of them, as I've been cleaning my workshop I've been finding more discarded prototypes that didn't make it to the box
I expected way more tbh
more like trial and terror
@@UnnecessaryAutomation You definitely need a "Box of shame ™" like Inheritance machining does but larger by the looks of it.
That box was incredibly tame.
Remember if a premade solution is expensive doing it yourself the first time will be waaaaay more expensive! *
Sometimes. REALLY depends on what it is. If it's something like this that demands precision, absolutely.
But for example, a little plastic fastener on my mom's shifter linkage broke, decoupling the stick from the push pull cable. The part should have been cheap, but was proprietary and only sold with the entire shifter linkage. I think it was $250.
Was able to fix it with a little nylon bushing, a drill, a 1/4" bolt and a stack of washers. Cost? $0 as we had everything on hand. Even if I'd had to buy it all, including the drill, it would have been a fraction of the price lol.
pedantically,
Precision: Repeatibility, defines minimum meaningful resolution
Accuracy: Deviation from expected value
They are not the same
Yes - cheap 2m tape measures usually have *resolution* down to a single mm, but are very unlikely to be *accurate* to a single mm over the whole 2m
I try to use relative and absolute error instead, most people thinking about it should figure the difference out.
yup. I use this to explain why digital *cannot physically* replace analog, and vice versa.
Usually trying to explain to some C.E.O. with a buisness or banking degree all the reasons why thier maliligned attempts at full automation failed miserably.
By the time you get through the required input channels for actually simulating an single neruon in the analog brain to any degree of accuracy they start to get the picture, but to be sure I take the width in copper atoms and have them translate that to Km, then Imperial miles.
Its 55 km by the way. Per nural connection... Assuming a single width of copper atoms can form a wire (it cannot). Some of that can be made up with faster translation times but. not all. There is a reason ChatGPT takes up a giant wearhouse and loses two bucks for every buck they earn.
That analog brain will always be 5% off. But that 5% takes a LOT of resolution to simulate.
@@AnonymousAnarchist2 What kind of automation are you working with where people are trying to implement automation by fully simulating a brain??
@@jakedewey3686 GPT models, and learning are *being sold* has full simulation of brains and neurons etc, and are being funded by people attempting to do so.
As such, and perhaps unsuprisingly the money men leading various sectors are expecting results of full simulations of human brains to allow for full automation.
Now that said, I absolutly can fully automate most if not all non-biological production with an big IF. (I.E. yes I can build a 3d printer that could print a car, a computer and every nail and outlet needed to build your home. just not mango chunty, wood, or a tomato. Bio stuff.)
....
IF production is willing to go with a much MUCH slower method of production and take up orders of magnatude more space. Its honestly rather trivial at this point to do, but nobody wants the responsability to fully fund the Research part of R&D they would rather redevelop over and over in response to miniscule stock market flucutations.
Also, ya know what, everyone should know this at this point.
Make a 3d printer with slip-stick nanodrivers, they are open source and 3d printable. Thats the gist of it, you need diffrent application methods for diffrent materials but one that works for most is to stick the whole thing in a vacuum and build an emiter of some sort really just a small aprature blocking the path of a heated chamber filled with the materal you want to print with. Another good one is just electrplating through a small little capilliary tube as the printhead. They have enough precision to even print a computer, the layer size is small enough that you can mimic any crystaline structure using eaither of these. But your layers will print very very slowly.
My dude I thought this was a channel a year or two old.
If this is truly the first video I am amazed by both quality and the sheer traffic you are getting.
Hope you are not too overwhelmed, take your time and enjoy the process your doing something right according to the algorithm don’t be a slave to it quality is likely to carry you through many an algorithmic storm.
Your like and subscriber count are about the same😂 1.7k
finally a adequate enough step angle for 0.2mm 3d printer nozzles
That is the most concise explanation of a stepper motor I have ever seen. I'll have to copy your homework when my kids ask me about them.
strain gears are so cool took me a time to realize it wasn't spinning just stretching
Oh yes, I didn't really understand them until after I made the first prototype
very good channel, here before it gets famous, subscribed!
Really appreciate that!
Yeh! Imagine my face when I was wo, this channel is awesome. Let me digg all the videos he has... To just realize this is the only one! Hope you can grow and get big with TOT inspiration! Let's go man!
Getting in on the ground floor.
I have a feeling that I will love this channel ,love editing, explanations and the project itself ,keep up with the good stuff 😁😁
Love the pencil and graph paper optic of the plots at 10:00 minutes. Very neat presentation.
What software do you use for creating the presentation?
its like This Old Tony but electronics. Hell yeah
Thanks! He's definitely a big inspiration for me
Holy crap, i just saw the subscriber count. This channel needs much more. if you keep this quality you are going far my man
Congratrulations on graduating on This Old Tony TH-cam CInematography Academy!
I would've just used a belt reduction. I don't think the torque transfer is great, but you can tighten that belt down so much it reduces almost all backlash
space efficiency, also drop in solution
Cheap and easy, torque transfer is no issue, I use that with NEMA 34 for up to 4:1 reductions (don't need more.)
With high lateral load like that, aren't you begging to destroy the bearings on a little stepper like this?
@@Nevir202 I would only worry about the bearings if they showed slop beforehand and at that point your misalignment would be the main issue.
@@Nevir202 this is how almost every 3d printer works, and they last for years with continuous use with high torque and accelerations. a belt reduction would be the way to go. and microstepping does give u finner than 1.8deg precision
The stepping explanation is really well done! Congrats!
Thanks! Glad you like it!
Wow this channel only has 61 subscribers ??!! It's just great !
Thanks!
@@UnnecessaryAutomation not any ,more haha bros video boom up in his first upload that means he's making good videos
Great video, I thought I knew stepper motors pretty well but learn a few things. Interested to see the project .
Thanks! Hopefully it won't be too long on the next video
ThisOldTony i can sense HERE
I ran into a similar problem when I was trying to make an equatorial goto mount. The RA axis needs ridiculous step precision to ideally keep steps smaller than the area in the sky that a single pixel covers at the given focal length, I can't remember the exact step size, but I had to use 1/32 microstepping(since this is a constant motion, microstepping does help) and add a 224:1 gearbox ontop of that. I used a split ring compound planetary gearbox, neet stuff those are.
Yeah that astronomy stuff really does require a lot of resolution. I did come across those split ring gearboxes and almost tried to make one, but the strain wave just looked so much cooler when running. Did you ever get the mount working?
@@UnnecessaryAutomation yeah it's awesome, and quite ridiculous. It takes some hundred of thousands of steps to rotate fully. And with helical gears it has no noticable backlash.
@@U_Geek Did you consider a brushless DC motor and a high precision rotary encoder instead of a stepper? That sounds like a decent approach if you want controlled constant motion.
Full disclosure - I haven't done it, and there might be problems I haven't thought of!
@@paulwomack5866 I have but I really didn't wanna spend too much on it and a stepper with ridiculous gearing was cheaper since I had most of the electronics. If I ever decide to upgrade it that is definetly the motor of choice, but I do have to say I appreciate that I don't need a counter weight to balance.
Great video with good simple explanations. Can't wait to see the next part!
Glad you like it, I think the next one will be pretty cool
This video is so good! You've got a great personality for these kinds of videos, you've nailed going into detail while keeping it easy to understand, and the animations are already incredible. The only feedback I can think of improving is a better mic and some color grading/better camera, and after that you'd seriously be on the level of channels with millions of subs.
Great video. Looking forward to the next to know more about what the project is about.
Thanks!
Great production quality and explanations, great first video, keep it up
Thanks! Already got the next one in the works!
I agree it's a great video, but please try giving more attention to audio. Various parts have changing levels of high frequency noise and background noise.
I think I heard somewhere that you should record the "silence" of a room to use in between the audio tracks, because absolute silence is always different and distracting
Yeah, I'm not entirely happy with the audio. I've got a couple of changes planned with my setup for the next one that should solve those issues
You can try using the microstepping function of most stepper driver ICs to achieve resolution below 1.8 degrees. I've used the DRV8825 with 1/32 microstepping and it theoretically should be able to get down to 0.05 degrees per step.
didn't he say in the video that microsteppin doesn't work? I am confused.
It clearly said in the video (without explaining though) that microstepping does not increase resolution. The truth of this is presented in the end (without specifically mentioning that microstepping has the same issue, although it is obvious).
It really depends on the driver chip that he is using. When using DRV8825 you have 3 config pins and when you pull them to the highest setting, every pulse on the STEP pin will be 0.05 degrees on a standart 1.8 deg/step motor. A4988 will do that too, but if I recall correctly the microstepping goes only up to 16. As the article says, it will of course have less torque using microstepping - but he even says he will not be rating the torque difference.
@@ntomata0002 so microsteps don't have a repeatable positioning?
I know the sentence is weird but you should get the picture, I have a blockade right now thinking about a better way to say it XD
@@Ramog1000There are a lot of reasons that microsteps are not accurate, some to get the picture:
a) The motors are not electromechanically constructed to do accurate microsteps.
b) I 'll explain this with an example. Let's assume that you have an 2A, 1Ntm motor and you provide 2A in one phase. The motor aligns in this position. But if you apply some load torque, the axis will deviate from its zero point. Applying more torque will make it deviate more until you get past the 1Ntm limit and the motor will kick.The kick will happen after the motor passed 2 steps away from zero torque position. So, any torque applied to the motor, either load or friction will result with an unknown deviation from the theoretically expected position.
Awesome video!!
Thanks!
As a historical note (~40 years), there used to be a company making steppers which used the priniciple directly. The rotor was the ellipse, and the stator deformed. The input sequence was different as well, using 8 inputs with 4 consecutive inputs being energized and the 4 shifting by one per input every step. It was called a harminic drive stepper, although the term is not apparently used that way nowadays.
It seems to me that, since the shaft/ellipsoid rotates within the inner gear that friction would be an issue and might limit the step rate.
loved how well and simply you explained this, thanks- instant sub!
Really a compact gearbox. Awesome work. Thanks for sharing ❤
Great explanations and clearly a lot of work to produce a quality video. Your T.O.T. Impression is … well … impressive!
1. Timing belts can be an alternative to complex gearbox with backlash control.
2. Better option is linear scales to provide accurate positioning. Even if the gearbox has no error, it does mean the screw drive is without error, or if the stepper motor skips steps (ie not moving in step with driver step pulses because of too fast, or not enough torque for the motor speed used).
Subscribed and notifications enabled. Great work! Can't wait to see more.
Banger first video
Also a cycloidal gearbox might fit your needs better with way less hassle
I thought cycloidal and strain wave are basically the same principle, just implemented slightly differently. They both fundamentally come down to a cam and a slight mismatch in the number of gear teeth (or lobes) between the input and output. I think the main difference is a cycloidal is easier to 'back drive' which is a bad thing in many applications
Great job on the animations and explanations.
This was great. Looking forward to more!
Tony's doppelganger
Not sure if this technique applies directly to stepper motors but it may give you an electronic instead of mechanical solution.
My experience is with closed loop servo motors. The way the windings are designed you can "Step" them around manually by connecting dc voltage to any 2 of the 3 windings. as an example with the phases labeled U V W, +U-V moves to the first step, +U-W to the second, +V-W to the third, +V-U to the forth, +W-U fifth, +W-V for the sixth position and then +U-V brings it back to the electrical first position in the series. there will be one set of 6 lockups per pole pair in the motor (number of total poles divided by 2) so a 2 pole winding will be steps of 60 degrees mechanically, 30 for a 4 pole, 20 for a 6 pole and 15 for an 8 pole.
The thing is you can do half steps by connecting 2 phases to the same leg of the circuit. Basically you connect the next leg in the series before you disconnect the previous one. Going half steps it would be +U-V first, then +U-V-W, then +U-W and so on through the cycle. This lets you get 7.5 degree mechanical steps on an 8 pole servo instead of 15. You may be able to do something similar on a stepper motor to get half steps.
In an actual in use servo the drive is not just stepping through positive and negative voltages but actually modulating the voltage to get the precision to hit any arbitrary angle. if its asked to turn the shaft smoothly at 1 RPM the voltage would trace a very slow sine wave on each phase. That level of modulation might be beyond your control boards ability to produce but I'd wadger it could connect the negitive or positive voltage to 2 wires at once to get half steps if the stepper motor behaves the same way.
Just something to try.
The manerisms, the framing, it's him!
Excellent work
awesome video man, good stuff.
great first video!
Merci !!! belle réalisation le partage de savoir Pas a pas !!
why not use a tmc 2209 with micro stepping?
right
or any other brand and model that supports microstepping
I need one of these.
I'd attach a vertical laser tool to it, then rotate it to hit a target. Then this tool would show the straight line on the floor leading from the origin to that target.
Fun fact: similar to the strain wave gearbox the stepper motor has 2 more teeth on the rotor than on the stator creating a reduction from electrical rotation to mechanical rotation.
9:37 I believe the encoder also has a resolution. In short the next step on the motor is closer to the next step forward on the encoder than to the step back. If you rotate the encoder position you will probably see the lines move up and down in parallel.
10:09 The extra noise might be because of the additional play introduced by the gear box.
You pretty much misread the article that you linked... There is definitely microstep positioning resolution, with SOME distortion just a little with a well behaved driver. The deflection under load is incidental but it's equivalent to backlash basically and doesn't depend on the step size.
It's not like a gearbox doesn't introduce a positioning distortion (error) of its own...
would be cool to see this in an equatorial mount!
Good video. No background music = Very good start 👍
Very informative.
Great video, you are doing good job m8
Thanks!
amazing video)
nice video
Loved the video. A few critiques.
Try and get your audio levels normalized between your CAD and talking hands section and keep your hands in frame. I know. Nit picks but thats some easy money to make things a bit nicer.
Darn it! I been looking for this exact kind of mechanism for my digital handheld monochromator for ages. Nobody makes any reduction for a Nema 8 from what I could find.
Kudos for building a strain wave gearbox! I've thought about them myself for another application.
I did do a planetary gear set with herringbone gears that when properly meshed had no perceivable backlash. But I did not measure with a high precision device. Have you considered trying a hearing bone planetary gear set?
I did come across those, but the strain wave seemed more fun!
@UnnecessaryAutomation Well if you decide to try out a planetary set, they're very easy to make. I don't know what CAD software you're using, but in freeCAD, The gears workbench makes it extremely easy to render the gears.
You don't really need a 'high precision" device to measure backlash. Just attach a long pointer perpendicular to the end of the shaft - the longer the pointer, the more precision you'll get.
Wave drives man, wild.
How cool! I had the notion that these would tear themselves apart of 3d printed!
So cool that you figured this out. Could you run out to test longevity.?😊
Thanks! I did leave it running for a couple days continuously and it was fine, but I'm probably not going to do more longevity testing
This was the best explanation of a stepper motor I have seen. Also great you measured the precision, I was wondering what it would be.
Where is the next video😫?! Oh I see this video was aired 3 days ago 👀I will wait for the vi then 🙃keep it up pal! U R Awesome 👍😎
I'd be curious how much deflection you see applying a small load to the shaft for just-the-motor and the strainwave case- I'd expect the print has more flex / less rigidity to it.
Also, a different way to handle the backlash of a geared system would be a preload spring / some-manner-of-preload. Though that may not fit the project you have in mind.
after 30s of video my first thought "This Old Tony wannabe"
Hello! Nice video!
Just a little problem for me is the "background noise", starting from the begining. As it seems constant in frequency and intensity, I guess it might be easy to remove in post prod :-)
Thanks for the feedback! It should be better in the next one
Hmm... I used to build very low flow pumps - 5 or so microliters per minute, although it would go down to nanoliters for mixing. I'm on the mechanical end, but our electrical guys said micro step improved the resolution too.
right here before your first 1k subscribers . we will watch your career with great interest.
Thanks! Hopefully not too long till 1k lol
Its important to document your mistakes to avoid overs doing the same, you show all the problems you encountered in making the gearbox to help others who wish to do the same, probably in a differant video.
Wow this is great thank you
I think the offset may come from the backlash and may be a measurement of the backlash. When the teeth push fireards, and then go in reverse, moving the distance of the blacklash and then hitting the reverse direction tooth. Maybe that's the cause? Idk tho
for anyone who is interested in more information about stepper motors
Lost In Tech has a great 3 part video series on electric motors
I was really surprised to see so much noise in the gearbox output, what do you think was the cause of that? I was thinking that the periodicity of the noise would match the number of teeth in one of the gears, but I counted ~40 peaks in the noose plot and only I think 32 teeth in the outer ring gear.
Hmm - but you said it’s a 20:1 ratio, so maybe the noise frequency is 2x the ratio?
I suspect it has to do with how the gear teeth engage and disengage with each other. Maybe they should have a different profile than normal spur gear teeth, tailored to how they engage as the peak of the strain wave approaches, to maintain more of a constant torsional force as they engage?
3D printed strain wave gearing seems like a potential gold mine for a lot of applications needing affordable precision and good torque, and you’ve done a great job of designing this assembly - I’d love to see you dig into it more deeply. (But then you were working on an application, not a research project 😁)
Great work, thanks for sharing - this is the first time I’ve seen a plot of angular precision like this!
Strainwave may not have backlash but it does have deflection
I've made a stackable belt-gearbox recently it's state is 1:4 with very little to no backlash if you're interested
That's pretty cool! I'd love to see a link to that!
Tried out a Flex Gear printed in PETG, cracked while attempting to install. Got lucky and found a spool of nylon stashed away and now trying that
Nylon worked great
Stepper motors don't have magnets on the surface of the rotor. They just have iron teeth that line up with the iron teeth on the stator. Most are "hybrid" steppers that have one large axially magnetized magnet in the center, and then one set of rotor teeth at each end, slightly offset from each other. This is why you feel the cogging effect even when the motor has no power. Also why you can generate moderate voltages from spinning a stepper.
Amazing production quality! Is it possible to get a closeup / slomo shot of the gears? It was a bit hard to follow while turning so quickly. Looking forward to seeing what you use the gearbox for!
Thanks! I might make a short with some slow shots of the gears
In both cases the error you're seeing is relatively predictable.
More-so without the gear box, where you could pretty easily overshoot and reverse back to choose between the 2 error directions.
But even with the gearbox those spikes are pretty repetitive, you could probably probably predict them in software with some reasonable accuracy, and spend some of that extra accuracy you now have to compensate for the predicted error a bit.
It might behave differently under load and in other conditions, so that might mess up any predictions though, it might take some work to get right.
True, I've solved a similar problem in software on a different project so I might try that here too.
Wouldn't adding feedback loop improve precision? You already have the sensor, so you can use it to correct for positional errors. As a bonus it will also account for lost steps.
I guess the viability of this approach depends on what you are building this for.
that article was made in 2016, micro-stepping today with a tmc2209 or similar would yield usable results, especially in higher resolutions than 1/16
The 2209 supports 1/256 microsteps
As he pointed out, at higher microstep ratios the incremental torque becomes less, and is more likely to cause problems under load, especially direction changes.
If you want to make a harmonic drive just say you want to make a harmonic drive. Don’t say that microstepping is not actually increasing the resolution of the motor as that is false. You can create magnetic states between the poles through mircostepping that allow for higher resolution.
Great video! :D
You should note that magnetic encoders are very sensitive to alignment, and do have a natural hysteresis, and with the 3d printed bearings, I wouldn't put too much trust in the magnetic encoder being held as well as it could be. you might find significant improvement with better bearings + somehow mounting the magnet perfectly concentric, but it won't fix the natural hysteresis in the encoder. A lot of the issues with magnetic encoders are buried or outright omitted in the encoder chip's datasheet to many engineer's frustrations 😅
Thanks!
Yeah I came across a lot of those issues with the encoder and ended up doing some software stuff to compensate for the alignment error. I had some of that in an early draft of the video but I couldn't get it to flow well so it got cut. I didn't know about the hysteresis though, that would explain some other things. Like you said, not much about this in the datasheet 😂
Next time you want to measure small angles with very high accuracy without doubt, do it like the old masters. Use a laser and a mirror. Attach the mirror on the axis of rotation, set the laser to aim the mirror and reflect to a wall. The reflection of the dot to a wall 2m away will give about 70mm movement for 0.5° of rotation. Easy to visualize and get accurate resolution of even 0.01°.
With OnStep, good drivers, 256 microstep and gear reduction, we drive telescopes to under 1 arc/sec rms with +50lbs load.
6:37 I counted the inner ring rotating 16 times for the outer to rotate once
Yeah I didn't catch that until too late. The gearbox I used for that demo was an older prototype with a different ratio. But the final version on printables is the 20:1
Not sure if you already know that, but to get the audio quality up (and noise down) you need to either buy a decent mic or at least remove background noise with software like Audacity (free).
For the video it is mostly about your white balance. Your hands certainly are not blue. Perhaps get a grey card to get you started. Once you have a feel for it, you can do it by hand. Or just always use a fixed setting on the camera and make sure that all your light sources are the same to avoid any post-processing with that.
Thanks for the tips, I'm definitely getting a better mic for the next one, do you have any good guides for color correction?
@@UnnecessaryAutomation I recommend essentially any video here on youtube on that topic. Search specifically for white balance, as color correction is something different and unlikely to be needed here.
The wobble in presicion most likely comes from an uneven gear surface. If you have access to it, you might want to try this design with resin printed parts, since they have a much better surface finish, than fdm parts.
I actually did try a resin version, well except for the flex gear since I didn't have a good flexible resin. It didn't perform that much better so I stuck to FDM
@UnnecessaryAutomation have you tried undercuring tough resin? Formfutura tough resin can be hard as pla or gummy like tpu depending on the exposure time and temperature during print (hard >7 sec on an elegoo mars 2 pro, soft ~6 sec). Aftercuring and storage for at least a year won't change the propperties noticable.
The Strain Wave Gearbox is a cool approach and I may try this with one of my automation projects. One question I do have is about the gearbox temperature and ware over time?
Interesting!
I was intrigued by the wave pattern of the resulting precision, went back to the number of teeth and found the circular spline to have 32 teethe and the flex spline 30 but the number of waves is 40. I assume the waves are produced by one of the gears. Did you use a different one for the testing than shown in the video to demonstrate the mechanism? To get a reduction of 20 one would need a 40 tooth flex and 42 tooth circular gear, so is it the flex spline causing the waves?
Just a guess: could the 3d print surface imperfections be the cause of the waves? Print lines interfering with each other in other words. Maybe printing the 3 components with different layer heights might reduce the waviness.
In industry, there is a leap to servo motors with encoders. Yes, they are not as cheap as steppers, which is why steppers are so common. A typical motor used has a 500 line encoder. With an incremental encoder, the sine and cosine, called the A and B encoder signals, one full line of the encoder has 4 steps of the movement of a single encoder line, for a native resolution of 2000 steps per revolution. In photolithography equipment used in exposing semiconductor layers, decades old equipment resolved to 0.5 microns. Newer equipment to register modern semiconductor layers, sub nanometer resolutions are used. When in fast motion, the encoder signals are well into the RF range of several megahertz.
Disclaimer, I do work in this industry. Laser interferometery is used for positioning far more accurately than optical line based encoders could ever do. 32 bit resolution on a linear motor is typical of older than 10 year equipment. New stuff is proprietary trade secrets.
You can also get closed loop steppers these days, which are servo motors too. Anything with a 'closed loop' motion controller is actually a servo. I think closed loop steppers are cheaper than servos generally as the motor controllers are a little simpler
Seems like the gearbox adds some phase lag to the reference angle. Should be pretty easy to account for it I wonder what the error is when doing mean variance...
Do you have a high torque requirement? If not, why not just get a 0.9 degree stepper and a simple belt drive for the reduction? Or a cycloidal drive? Strain wave is really neat and I admire your grit working through all those prototypes!
And so it begins . . . . . [Kosh]
Old tony took the Tardis back in time.
Amazing 😀👍
Most stepper drivers allow for micro stepping. Thus dividing a 1.8 degree step by 10. Increasing your steps per revolution 10 fold.
Also timing belts and pulleys is another way for gear reduction without introducing backlash.
Last option is spring loaded worm drives.
Microstepping gives a reduction in torque, especially holding torque. Microstepping is driving two sets of coils of the motor partially to move the net magnetic field between the usual steps - because the coils are driven with less than full current, you end up with less than full holding force too
Hello,
I just wanted to say that I really enjoyed your video! I was wondering if you could kindly let me know what animation software you use for your videos. I’d really appreciate it. Thank you!
Thanks! It's a custom animation package I made. I might release it some day, but it's in no state to share right now
Looking at the trace from the gearbox it's clear that the error is cyclical, probably from every rotation of the cam. The 0.6 degree error is total but at each position the actual error is very very small.
You should be able to compensate for most of it in software and get 0.1 degree precision or better.
Ok but according to the hachaday article, microsteps only weakness is error in open loop and torque.
You could mitigate the first problem with hall sensors, there are some closed loop solutions for 3d printers for example. And for torque - what do you need the stepper for? I feel like for tiny and precise movements you wouldn't need too much torque, but again, idk what you need it for.
True, but I don't really have space for an encoder unfortunately, and it will be resisting, admittedly small, cutting forces so I do need the torque
@@UnnecessaryAutomation There are closed loop nema 17 motors that have an encoder built into the back which take up less length than a gearbox while also providing more torque than a standard stepper as it can briefly overload the amps when needed, also it is worth noting that the hackaday article is fairly dated and does not test more modern drivers like the tmc 5160 or 2209.
nonetheless this is still a really cool gearbox and good data I'm just putting in my 2 cents.
@@UnnecessaryAutomation i got a 3D printer with 0.9 degree steppers and TMC 2209 drivers and with 16 microsteps the torque is enough to not let me move the gantry. And as someone mentioned, modern closed loop systems would take much less space than a gearbox.