Experimental Harmonic Drive Reducer - 3D Printed

Check out Fantom Wallet at: store.fantomwa... and use coupon code 'BRUTON' for 10% off.
Over the last few weeks I've been investigating 3D printed Cycloid Drives. I thought it was time to take a look at a Strain Wave, or Harmonic Drive reducer. I'm using TPU for the flex-spline in this design, and the rest is PLA. I'm using the same brushless motor and the reducer is a 10:1 ratio to match the cycloidal drive so we can compare them.
CAD for this: github.com/XRo...
You can support me on Patreon or buy my Merchandise:
***************************
Patreon: / xrobots
Merchandise: teespring.com/...
***************************
Affiliate links - I will get some money of you use them to sign up or buy something:
***************************
Matterhackers 3D printing supplies: www.matterhacke...?aff=7500
Music for your KZbin videos: share.epidemics...
***************************
Other socials:
***************************
Instagram: / xrobotsuk
Facebook: / xrobotsuk
Twitter: / xrobotsuk
***************************
CAD and Code for my projects: github.com/XRo...
Huge thanks to my Patrons, without whom my standard of living would drastically decline. Like, inside out-Farm Foods bag decline. Plus a very special shoutout to Lulzbot, Inc who keep me in LulzBot 3D printers and support me via Patreon.
HARDWARE/SOFTWARE
Below you can also find a lot of the typical tools, equipment and supplies used in my projects:
Filament from: www.3dfuel.com/
Lulzbot 3D Printers: bit.ly/2Sj6nil
Lincoln Electric Welder: bit.ly/2Rqhqos
CNC Router: bit.ly/2QdsNjt
Ryobi Tools: bit.ly/2RhArcD
Axminster Micro Lathe: bit.ly/2Sj6eeN
3D Printer Filament: bit.ly/2PdcdUu
Soldering Iron: bit.ly/2DrNWDR
Vectric CNC Software: bit.ly/2zxpZqv
Why not join my community, who are mostly made up of actual geniuses. There’s a Facebook group and everything: / 287089964833488
XROBOTS
Former toy designer, current KZbin maker and general robotics, electrical and mechanical engineer, I’m a fan of doing it yourself and innovation by trial and error. My channel is where I share some of my useful and not-so-useful inventions, designs and maker advice. Iron Man is my go-to cosplay, and 3D printing can solve most issues - broken bolts, missing parts, world hunger, you name it.
XRobots is the community around my content where you can get in touch, share tips and advice, and more build FAQs, schematics and designs are also available.

[@jamesbruton]

Testing both Cycloidal and Harmonic Drive Reducers with an encoder and ODrive is coming up in next week's video, but Patrons and KZbin Channel Members have it already!

[@manujain9995]

Loving this series!

[@lidular]

You may have metioned it, but why not use planeteary gears?

[@jimmyhackers8980]

you could forgo motors altogether with nitinol wire. would make for an interesting muscle design. it would be the equivalent of myomers in mechwarrior lore.

[@Sn0wZer0]

The wave generators in the metal ones use a deformable bearing, while you are using just a single small wheel. Consider using more bearings to give the spline more support, or at least the largest ones that will fit: kzbin.info/www/bejne/qmXLiGmij76Uh80 The other thing is that harmonic drives don't like skipping more than two teeth between the inner and outer. Four is already pushing it, allowing for slippage (in particular when combined with the very flexible spline and not much support). Good luck!

[@tigertalar]

Checkout the back driveable _Abacus Drive_, as an alternative to harmonic and cycloidal.

[@simonmerrett]

Hi James, thanks for the attribution and I'm delighted to see you showcasing your own exploration of this interesting gear type. I'm sure you can optimise even further and I'm very excited about seeing some torque tests when ready. At a high level, I would suppose that your flex spline may be suffering from the same major issue that my timing belt designs did. I don't know about the round profile as I did not fully investigate it. However, tooth deformation was definitely a major contributor to my gears' limits. Therefore I would look to see if you can produce your splined cup from something less flexible and build flexures into the mechanical design (slots, such as in kerf-bent laser cut curved plywood designs). The other (hopefully) quick win you could do is increase the diameter of the bearings on your wave generator. This should produce a more gradual meshing-demeshing of the teeth (friction force spread over a larger angle of rotation) and also reduce the main motor's torque requirement because the pivot of the wave generator bearings are closer to the main motor axle.

[@Marci124]

I've seen slots on some commercially available drives as well.

[@martindinner3621]

I've got a nasty hunch that part of the binding issue is that the TPU part is experiencing, for lack of a better descriptive term, helical deformation. Basically it is distorting from a straight tooth gear to something that looks like the negative of barrel rifling. This would explain the extreme resistance to backfeed.

[@assumptionsoup]

Interestingly, Koal Designs also tried easing the strain wave contact point for their scara robot. Instead of increasing bearing size, they added more bearings to model the falloff. And it seems to be working for them: kzbin.info/www/bejne/pXbHaYyvaayojK8

[@simonmerrett]

@@assumptionsoup thanks. The "many mini" bearings approach is one several people have tried. If your splined cup is stiff enough then it will do a good approximation of a wide curve. However it does nothing to bring the bearing pivot point closer to the motor axis, so I don't know if it realises any improvement in that area (there may be no benefit other than simplicity with a pair of larger bearings).

[@IDoNotLikeHandlesOnYT]

I don't see why moving the axes of the bearings closer to the input axis should reduce the torque requirement. The flex spline still has the same radius, and the tooth ratio is the same. It seems to me that using bigger bearings closer to the input axis is like changing the size of an idler between two spur gears: it doesn't affect the overall gear ratio.

[@PPYTAO]

“This is black this time instead of red because I’ve ran out of red” I feel like that was anticipating comments asking haha.

[@DivineWerezwolf]

Would like to see a planetary gear setup next just curious to see all three performing against eachother.

[@JohnPM]

If you don't want to wait I'd check out this guy's build, it's pretty neat kzbin.info/www/bejne/aJWqkHlvp7dgoK8

[@kensmith5694]

Yes, a herringbone geared planetary gear set can all be printed in one printer run as an assembled item that can't come apart. This seems to be a natural for a 3d printed reduction gear. You do have to "run it in" a bit to wear down the roughnesses on the surfaces. You can also do "differential gears" to get a greater reduction in a small place. Imagine a gear with 10 teeth rolling around inside a 100 tooth internal tooth gear. Now imagine the 10 tooth gear has an 11 tooth gear on the same shaft. The 11 tooth is slightly bigger and runs in a slightly bigger 101 tooth gear. For every trip the small gear pair makes around the circle, the 100 tooth and 101 tooth gears rotate slightly.

[@IDoNotLikeHandlesOnYT]

@@kensmith5694 Gear Down for What made a lot of 3D-printed gearboxes like you describe a few years ago. I haven't checked to see if that channel's still active, though I haven't seen any new videos in a few years. The old videos are presumably still good, though (but I don't recommend using an Arduino as a coprocessor to a PC to calculate gear ratios by trial and error).

[@DALBYNATOR]

Me: say ‘strain’ James: ‘strain’ Me: say ‘wave’ James: ‘wave’ Me: say ‘strain wave’ James: ‘strain wEave’

[@azyfloof]

"Isn't it strain wave?" "Nah mate, it's strain weave" "I'm pretty sure it's strain wave" "I've never heard of that word in my entire life mate"

[@HandbrakeBiscuit]

James had too much Cool Hwhip on his dessert the day he made this video...

[@Alansshows]

Strine wive?

[@dominicrichardson5546]

Ahh it hurts every time I hear it

[@iliya-malecki]

try printing the flex spline out of nylon (obviously also making it thinner). That would reduce friction and would waste less energy deforming

[@nophead]

Yes my thought exactly. It doesn't need to be rubbery, just thin enough to flex like the metal ones.

[@mufassarwaheed6160]

Printing in PP is even better i guess. It would stay flexible "forever"

[@mitri5389]

​@@mufassarwaheed6160 but harmonic drives are not made to be flexible like rubber, it mainly deforms well under the max yield strength curve just under the elastic limits of the material. it should be a fairly ridged material that is allowed to deform yet elastic enough that no permeant deformation takes place. I also think that in this video the teeth are too pronounced and it looks like the contact is shared across multiple ring teeth at once, on the real drive the teeth are very fine and shallow in height and the teeth mesh only with a single tooth per side. likely another coinciding possibility is that the bearings are too wide stretching too much area of the flex shaft

[@WaylanEE]

@@mitri5389 This is another reason the cyclo is better. More tooth engagement means higher torque capacity. With the cyclo all the teeth are engaged all the time. These have been around for a long time and are commonly used in low speed very high torque applications where the extra tooth engagement is essential. Dump trucks use two and often times three counter-shafts to spread the load over more than just the one or two teeth engaged. That is the weakest point in any transmission. Planetary gears handle higher torque loads with more planets. The more planets the more torque tolerant the gear set.

[@mitri5389]

@@WaylanEE planetary gears are not that great for robotics due to backlash its hard to make planetary gears that have little backlash compared to cycloidal and these ones.

[@TheAnoniemo]

I think the strain wave drive is being funky because the teeth profile is very shallow so combined with the flexible material of the spline can squish and slide over each other when put under load. I'd try with a steeper tooth profile. Also, I think the applications for both drives are different, the strain wave will come out on top on higher reduction ratios, because you can easily scale the teeth down to make a larger reduction. I've been looking at making a 3D printed star tracker and using a strain wave drive for it, but that is a totally different use case from yours, requiring high precision and reduction but very low speeds.

[@fatgamer919fan4]

These cyclodial drive and harmonic drives looks so ttrippy

[@duncanmcgregor4973]

trippy? What do you mean?

[@ronnetgrazer362]

@@duncanmcgregor4973 Like, psychedelic, man! It's pretty far out, compared to your average mechanism. It was invented in 1957 but applications started popping up in the 60s. Imagine seeing this for the first time while turned on and tuned in.

[@stocchinet]

@@ronnetgrazer362 Man, it's like a gear, but it flexes, it's weird man

[@crackedemerald4930]

I still get scared when you grip those bolts with your hands.

[@ddegn]

At least he's not grabbing the finger amputator this time. The cycloidal drive looked like it would have taken his hand off.

[@OwenBuckingham11]

It terrifies me too

[@BarioIDL]

i can only use saw cuz of my angle grinder trauma

[@dogfarmer2662]

@@user-mh1ds9os5p no mercy, killing intelligent life is only acceptable in self defence, this witch is evil

[@crackedemerald4930]

@EZ Z cus the threads are sharp and might bruise his fingies

[@Game0verxD]

I think flex spline doesnt need to made on elastic material it can be printed with ABS which can elastically deform. and it will result much better tooth engagement which can result much more resilient to forces. OFc its needs to be thin wall and needs to be calculated or experimented

[@ihydf]

In commercial designs and successful 3d printed ones the strain wave generator itself also maintains the ovular profile of the belt driving the entire surface of the belt, not just at the major axis kzbin.info/www/bejne/rp3Rf51ve5hqg6M

[@mervynlarrier9424]

I know you definitely have more experience than me so I'm not going to try to "teach you", but if I may offer some comments on the strain wave gearing vs. the cycloidal: 1. The strain wave gears are typically, as you know, high precision gearing. But they're also not really suited for excessive or counter loading. One of their most common failure modes is teeth stripping because, especially for the high-precision ones, the teeth are very fine and easy to strip. 2. Cycloidal drives, on the other hand, ARE suitable for excessive loading because of the geometry of their "teeth." Because all the energy is transferred via rolling action, they tend to be more durable and capable of higher loads than comparable strain wave gearing. But, they're not quite as precise and, unlike strain wave gearing, are back-driveable. 3. As a tip for your strain wave gears in the future, the flex spline is always the most difficult part to nail because you don't want it to be too flexible but you also still need it to be flexible enough. This can be more easily achieved, I think, with thin metal versus polymers of variable shore hardness and different fillers to give different mechanical properties. So it will really have to be tuned. That said, I think it would be well worth your time to try out, as troublesome as it can be to print, a nylon instead of a TPU for the flex spline. It'll still be flexible (assuming you use straight nylon and not one filled with carbon fiber, although I've heard glass fiber-filled nylons are typically just as flexible as non-filled but stronger) but also provide more rigidity to the flex spline. So instead of worrying about your wave generator, try switching the flex spline material to nylon. You may see an improvement in functionality.

[@Skythedragon]

Maybe look at Paul Gould's cycloidal drive designs, they are quite compact, and have had a lot of development behind them

[@HumanlyRobotic]

Pretty sure his first cycloidal drive videos credited him.

[@TheRainHarvester]

Cheaper bearing cost too: $30 .

[@recurvestickerdragon]

this design: - can't be back-driven - can be stalled - is heavier - will probably wear out sooner, with the flexible material In all, seems like the cycloids will work much better for you

[@recurvestickerdragon]

also, I'm pretty sure your flex material is squishing, leading to the skipping and stalling

[@nonchip]

and we can even already see wear in the video, at e.g. 11:05. and by the pattern i'd guess not much of the inner TPU gear actually engages with the outer PLA one, or at least only the bit with the bearing right behind is actually involved in transferring much force. looks like that might be where the "stalling" originates from, simply the gear deforming in an unintended way

[@syedsulaiman8380]

Yeah in this case cycloidal drive is better but harmonic gear boxes are really next level technology and really the best in certain use casses

[@euantubely1517]

Strain "weave"?!

[@MrPesht]

You may be correct, but it's because of the materials he's forced to use due to 3d printing it. The design itself is extremely popular in robotics because it's very effective when using the proper materials and design, but 3d printing it has always cause issues unfortunately.

[@tylerwright6006]

this one man equals the entire boston dynamics company :p

[@tylerwright6006]

@@kruleworld You know, it would probably still be the same basement home brew content, but with better video editing and lighting :p maybe some drone footage xD

[@RandallStephens397]

I like that even though the cycloidal drive seems like the obvious winner here, you're still going ahead with developing the harmonic drive anyway. Not only because it could still be useful in other applications, but doing the tests and publishing the results is useful to people in the future. The only difference between messing about and doing science is writing shit down.

[@Zhus118]

Cool stuff!. Actual harmonic drives are back-drivable with little losses to friction. They key is to have the wave generator fully support the path of the elliptical deforming gear. Otherwise it will just try to flex into itself and cause lots of friction. I am testing a 3d printed version with large offset bearings similar to the center shaft of your cycloidal, but have both of those support the elliptical gear. Also, since you have very large teeth thus large deflections, consider the pancake version of the harmonic drive. You won't lose as much energy where the flexspline meets the output as the whole flexspline stays deformed.

[@marvinzapf8352]

Nice Harmonic Drive man! Unfortunately not backdrivable :/ You should look into the new abacus drive and inception drive! Those would perfectly fit into your gear reduction video series!

[@StevenIngram]

I'd never heard of either of those. I think the inception drive looks very interesting. :)

[@marvinzapf8352]

@@StevenIngram the latest and greatest ;)

[@Blox117]

what about the warp speed drive

[@stephenhumble7627]

​@@StevenIngram Neither had I as they were only recently invented and the patented by SRI - this is a real surprise for a new kind of mechanical gear mechanism to be invented The inventor certainly deserves whatever SRI are paying him these are really great inventions that use a principle of variable radius motion and angular contact between moving parts in ways not realised before. The nuVinci transmission used on some bicycles is another fairly new innovation that also uses variable angular contact for changing gears but cant stop or go backwards like the inception drive. The downside to the abacus drive is the rollers geometry contact area to the variable width inner and outer tracks is relatively small so it places quite high demands on the contact surfaces of the tracks and rollers - so these have to be well made to handle the contact pressure. The inception drive like a variable speed transmission is relying on surface to surface adhesion - so you need a powerful actuator system to ensure there is high applied contact pressure or the drive is liable to slip and cause rapid wear. It will be interesting to see the application they find for these in future. Had they invented the inception drive 30+ years ago it would have been well suited to adapting the power of combustion engines which have limited speed range compared to modern electric motors but now everything is going electric and the applications may be quite niche. Modern electric motors and controllers are so good the need for a variable speed gearbox is greatly reduced.

[@nathan471]

This is so cool! Every time I see flexible prints my mind is blown!

[@basement_robotics]

Great job! I work at a robotics company and have designed and built robots that use both types of gearboxes on mobile platforms and fixed robot arms. Your videos perfectly show the operational and engineering pros and cons of both reduction types. Harmonic drives are the preferred gearbox for applications that require no backlash and high efficiency while cycloidal drives are actually notorious for being inefficient given all of the surfaces they have rubbing against one another, but they are typically much cheaper than harmonic drives so there is a design tradeoff. I will say though that I have been able to back drive harmonic drives with ratios up to 40:1 without issue, so I think your back driving problem with your own harmonic drive lies somewhere in your design (to be expected though given how temperamental they are). There are companies with large teams of engineers designing these things, and the fact that you were able to make one that works this well on your first try is nothing short of amazing! Looking forward to more videos on your projects!

[@guyincognito-1]

I think the improper meshing of the teeth is most likely what is happening. Such a cool mechanism none the less!

[@flyerminer]

I learned about harmonic drives for the first time during a robotics internship I was on last year. Fascinating stuff!

[@JulianMakes]

Great video, i was wondering when you’d do strain wave! PS robot dog needs a face and tail :)

[@tylerwright6006]

watch Adam Savages Tested video where he gives a face and shoulders to his boston dynamics dog :p

[@Mrbulletfire]

It would be cool to have some components in a clear material so we can see what’s going on inside!

[@bjrn-oskarrnning2740]

3D printed materials aren't very see-through, unfortunately: even transparent filaments are so hazy they might as well be white. You can post-process the outside, but the infill will still be mostly opaque. I guess plexiglass can be used in some cases, but that's somehow flexible AND brittle at the same time, so it's not exactly ideal, either.

[@Mrbulletfire]

@@bjrn-oskarrnning2740 maybe a laser cut acrylic

[@rahul38474]

I’ve seen a few other harmonic drives on KZbin where the back plate (not the output plate) is made from acrylic so you can see the precession of the flex spline.

[@pvic6959]

@@Mrbulletfire does he have a laser cutter?

[@Mrbulletfire]

@@pvic6959 not sure, could be something to look into and fun new branch of making to discover

[@apbosh1]

It was worth a go and a great printing excersise if nothing else but looks like a total nightmare in terms of reliability or consistent performance. Thanks for the interesting video.

[@aftenstjernemix]

Commercial Strain wave gears absolutely are backdrivable - even at 1:100 ratio. They are used extensively in collaborative robots for this very reason - otherwise you could not move the robot in teach mode. Internal friction in this prototype combined with the low stiffness of the flexspline is causing geometric binding. I think the ideal ratio for backdrivability is closer to 1:30 or 1:40. A lot of knowledge on this topic can be found in Harmonic Drive’s older (10+ years) catalogs.

[@batlin]

"wave generator... strain weave... wave generator" Man that grinds my gears.

[@georgemathieson6097]

I don't know if off the top of my head, but there is a really lovely equation for working out the number of teeth required for each sides of the harmonic drive - maybe a mismatch is causing it to skip. Also, I'm almost certain you can make a harmonic drive back-drivable (again, I can't remember how). Don't give up on it just yet James, a robot made to use harmonic drives would be really impressive. Everyone else on KZbin just does a cycloidal gearbox you see, it would really set you apart.

[@nicoladc]

Seeing results from Levi Jansen you may want to give a look at his harmonic compliant drive.

[@andrewpienaar4522]

Harmonic drives are particularly used in 5 or 6 axis robots due to the fact that they don't back-drive. When there is a E-stop (Emergency stop) or power failure you don't want them to go limp and drop their payload. Most of the robots (Kuka, Fanuc and Kawasaki) we use for spot weld and manupulation applications have an 165 to 200kg payload and you don't want that flopping around. Thanks for your tinkering with this however, as it has given me food for thought and I can definitely use your cycliodal gearbox for something I have in mind.

[@mschnuth]

I'm a mechanical engineer who has worked with Harmonic drives professionally and they are absolutely back driveable, (although I admit I struggle to wrap my head around how) but they require very tightly toleranced components and are very stiff. The reason your printed one likely is not backdrivable is likely due to the compliance of the TPU flex spline teeth lowering your ratcheting torque and also absorbing rather than transfering the necessary forces back to the wave generator. Cycloidal drives are much more impact resistant due to the larger surface area in contact between components transferring forces in the gear train which is why they're used in big industrial robots like Kukas. That's probably your best bet for 3D printing as a fabrication technique and for the application of the Open Dog legs.

[@alifleming2256]

There are some interesting designs that have a planetary gear set withint the flex spline to further improve the gear ratio. Jeff Kerr has a good video demonstrating this

[@Lordtheobald]

the flex drive has baring that runs along the inside of the flexible belt. Because of this, I believe, that when you stop it that lets the barrings spin freely. There is nothing stopping the inside barings from spinning feeling along the inside of the flexible belt.

[@leozendo3500]

Try planetary gear box next? You can 3d print multi stage ones and buy commercial mental ones for cheap since they are commonly used on drills

[@kantpredict]

The point of these two reduction types is that you get a very large reduction in a single stage.

[@leozendo3500]

@@kantpredict thank you. As far as I know you have covered the only two in-line single stage reduction gearboxes - plus one with many moving parts patented by a company basically like a wave drive gearbox but use flutes undead of gear.

[@JohnDlugosz]

@@leozendo3500 undead ?

[@drblitzzz]

The skipping is possibly due to the concentrated forces on the single stack of wave generator bearings. the flexible gear is deforming around the bearing- so spreading the load and reverting to a double stack should help.

[@JackRyanRobtics]

Things that helped my design: Printing the flex spline in PETG Higher diametral pitch allowed for a lower pressure angle and lower deformation to engage. The commercial drives have flex splines made of steel with a really high pitch and only deform by about a millimeter. I imagine this critical as it would massively reduce the amount of energy required to deform as energy required is proportional to the square of deformation, (E=1/2 K*d^2) while also reducing the friction (Ff=mu * N; Fs=-kd)

[@sreynoldshaertle]

A harmonic drive's spline is more efficient the better it is at retaining its circumference without losing engagement with the exterior teeth. This suggests several optimizations: * Print out of a rigid material and keep the deformation juuust this side of plastic deformation. This allows you to make the spline much thinner without compromising on resistance to tension. * Print out of a rigid material. You're likely losing a ton of energy into compressing the *teeth*, which costs energy directly and allows the teeth to engage badly. * Introduce little metal belts. These will resist tension without introducing resistance to bending. * Don't extend the teeth all the way down the outside of the spline. Corrugations are *fantastic* for introducing resistance to bending, which is what you don't want.

[@evergrumpygeorge]

I think the reason that the output can be stalled is due to the fact that the flexible filament can be deformed meaning that the engagement points on the inner gear can bend and 'squish' and allow the bearing to rotate around without the inner gear advancing to the next tooth.

[@ninehundreddollarluxuryyac5958]

Your cycloidal drive has bearings instead of plastic teeth where it meshes. I think thats why the cycloidal one runs so much more efficiently. It converts less work into heat when running and has less friction to overcome when starting.

[@cibisto]

VESC does support many encoders, even FOC and position control

[@panakon366]

From what I understand, the reason that you can stop it with your hands is because when you hold it the flex gear expands (because its flexible) and locks in place while the inner piece just rolls around freely.

[@elderdavidyoung]

A bit new to these types of drives my self. I've noticed that the harmonic drives has less surface contact/engagement area amongst the sets of teeth. That would result in putting a lot more force between a small amount of teeth. Also, it's likely your flex teeth are also being heavily compressing between the wave bar bearings and the wall of outer teeth. Which could be allowing the flex teeth to compress enough for slippage to happen. If available, you could try another more solid, but flexible material that can handle the force between the low area contact with the teeth sets. Overall, it looks like a more finicky drive to figure out the tolerances in design. It's pretty cool that your trying it out and having fun with it! I've been enjoying the proxy ride along with your adventures into different types of reduction drives. As well as your robotic projects in general.

[@OrvisEvans]

I've worked with harmonics a bunch before in a previous job - those gear boxes were back-drivable in my application. I'm guessing your motor can be stalled and can't be back driven due to the spline being able to stretch. That would expand the spline and allow more teeth to mesh at once than would mechanically allow for the teeth to skip properly. The thin metal splines in harmonics can deform, but not stretch. If you possibly print or use a spline material that can elastically deform without stretching, that may give you better results. One option is putting a thin-walled spring-steel tube inside your spline.

[@manojdixit6924]

Gear transmission is positive type, which means power transfer is guaranteed as there are no flexible parts. In case of this harmonic drive, as transmission body is flexible, power transmission as not guaranteed (motor shaft can turn when output shaft is stalled) In which case the encoder needs to be attached to output shaft rather than at motor shaft. Nice design btw!

[@karlspencer1493]

I agree, the problem should be the flex spline overflexing. In general I think the splineintroduces to much friction. A solution to this problem could be the usage of pins, like in the Galaxy-Drive, a reconfigured Harmonic-Drive. It would be really nice if you could try to build that kind of drive. Really love you videos, keep up the good work!

[@Aleks_Z]

Harmonic drive has "mushy" gear. When rotational force is applied to teeth, their output force is towards the middle, increasing the gap between teeth and making them skip. To reduce this, you could try making a cross (it should have shorter sides than length) with 4 bearings, instead of a single bar with 2. Or even a 6 point star, where the 2 main bearing would remain at the same position as before, but the 4 side bearings would allow flex gear to move certain ammount. It should act as a tensioner for flex gear and make it have less free space to buckle in. Ideally, the best solution would be a parabolic curve slider, from one bearing to another, on both sides, to really stop flex gear from having extra space. But that would increase friction drastically. Also, another thing, Cycloidal drive has so much more contact patches, while harmonic drive essentially has only 2 (at bearing points).

[@rx8tom]

I think the harmonic drive keeps jamming up because the gear teeth flex, and don’t go where they are supposed to. Also, it probably doesn’t back drive because the teeth are locked together, and you would have to force the teeth out of position, which doesn’t really happen because the bearings/rollers are in the way. Changing tooth profiles and experimenting with smaller rollers might produce different results. Great vid!

[@44R0Ndin]

One other thing about those harmonic drives with the flexible metal strain wave "gear". They almost certainly run in lubricant, which does wonders for the intrinsic friction of the mechanism. Also, as a whole, harmonic reduction drives aren't really that useful at these gear ratios like 10:1 where it's easy to back-drive things. They're much more suited to being used at ratios of 50:1 or greater, and at those ratios they have a few advantages compared to other types of reduction drive: - Greatly reduced number of moving parts, especially compared to a high reduction ratio planetary gear-train. ("the best part is no part") - Relatively low backlash compared to most other offerings, which is always a plus. - Can be made much more compact (shorter) than a planetary reduction gear-train of the same reduction ratio and torque capacity. - Difficult to back-drive (this can be an advantage in some if not most applications) However, it does have some drawbacks: - The lower the reduction ratio, the bulkier it becomes (theoretical minimum diameter must include a "gear tooth height * 4" component, which at low reduction ratios will become the majority factor of the diameter of the whole mechanism, but at high reduction ratios it trends toward zero). - Extremely high friction in the back-driving mode, often to the point that the mechanism will find other ways to permit the output shaft to rotate rather than drive the rotation of the input shaft (as you saw in your prototype "slipping" instead of back-driving). - The use of a flexible component creates a nearly-inevitable point of fatigue cracking failure. - High overall frictional losses in the mechanism (this can be avoided by clever selection of materials and use of bearings, but even then the frictional losses will still be higher than most other kinds of gear reduction).

[@romandegtiarev]

-So, how many bearings do you need for your gear boxes? -Yes

[@sabrinazwolf]

You can stop this one because I think you got a lot stronger since the last video. :)

[@tekeguy68]

There's just the inescapable fact that you will always overcome the flexibility of even triangular teeth with even a small load. It needs to be able to flex over its diameter but still have rigid teeth---which is why a metal flexspline works.

[@jameshamaker9321]

The circular disc drive system, uses a set up, that looks like a spirograph. This new set up, has a similar feel. Even then, it looks, awesome.

[@MetroidChild]

Love these unconventional reduction designs, would love to see how a typical planetary gearbox would fair against them in the future.

[@kensmith5694]

... or even an non-typical one would be interesting.

[@Will_Huff]

I really like watching you working through these designs! Keep it going!

[@genijable]

I just had a genius explanation of this one in my youtube recommendation. Just a few days ago. KZbin is amazing, can even predict the future already. They for sure know that I will watch James, so to make me better understand it they pushed an explanation of that one.

[@PraxZimmerman]

OskarPuzzle made a similar gear reduction called The Grinding Gears that omitted the flexible spline in favor of a solid one. Could be worth a look.

[@anomamos9095]

The reason you can stall a strain wave is that there is no direct connection with the motor and the gears. The bearings are distorting the flexible drive causing it to move in a similar manner to a snake not actually driving the gear. So when you stall it the bearings simply squeeze the flexible material without imparting any drive.

[@marshallwilkes2272]

When you try to stall the cycloidal drive, you are trying to stall something that is truly at 10:1, but when you try to stall the harmonic drive, you end up binding the drive bearing assembly, which means that is is stalled at a 1:1 ratio, not the 10:1 output ratio. To get a better understanding of how different components can be stopped and their impact on torque and gear reduction, check out the working concepts of a planetary gear reduction. You may connect a few dots :)

[@NeutronicalGaming]

Interesting but here's an idea, adding an expanding collar to the top of the flex drive cup so it can deform the shape of the teeth without overly effecting the top part of the cup, a v shaped collar at the top should be enough to let the varying geometries to work together nicely without adding as much resistance to deformation, or rotation, along with shaving a few microns off the overall width of the flex drive, and supporting the teeth better with more bearings, or perhaps a barrel roller instead of bearings, so the whole tooth is pushed into engagement for less slippage?

[@codyhouse6436]

Your creative use of different 3D printed materials + your through explanations of how something works (or "should" in theroy, lol) is why I keep looking forward to your amazing content!

[@jgeating]

VESCs can in fact use encoder feedback from a number of different sources such as ABI encoders and AS5047 magnetic encoders - even the older 4.8 hardware shown in this video

[@ddegn]

Have you seen Levi Janssen's *Compliant Harmonic Drive (3D Printed)* video? I thought it looked promising and I liked how much smaller the drive was compared with other options.

[@punkrock4401]

start this video: wow thats a cool gear Next screen: fully 3d printed boston dynamic clone: "OH SH-"

[@lilhazerd]

I know this is a bit of an old video but if you see this comment and ever want to come back to SWG drives I think I know what is causing your low stall torque and erratic operation. For an SWG drive the difference in the tooth count between the flex gear and the stator gear has to be a difference of 2. By having a difference of 4 your drive is trying to move 2 teeth every 180 degrees of rotation which shifts your contact patches away from where your wave generator is trying to hold the teeth together. With a difference of 2 teeth your flex gear shifts 1 tooth every 180 degrees which keeps it from binding with it's 2 contact points. for a 10:1 ratio you would need 20 inner teeth and 22 outer teeth. You can achieve this by either shrinking the part or enlarging the teeth. You can also increase efficiency by redesigning your wave generator to help your flex gear hold shape at high speeds.

[@rj7855]

i think the problem is the stretching of the flex part, it might help to have a second rotor (slightly shorter to hold the flex part in the correct "off" position) or somehow integrate a metal wire, or a belt, inside the flex gear.

[@surfcello]

I imagine the wave generator should not just be a narrow stick with two bearings at the end but rather an oval that conforms to the inside of the flex spline. This could reduce buckling of the spline and thus slippage. Also, consider using a chain rather than a flexible spline for more stiffness. That would, of course, make coupling with the output cup more difficult. Lastly, increasing the diameter of the bearings on the wave generator that push the spline outwards into the ring gear would increase meshing and thereby reduce slippage.

[@BRUXXUS]

My intuition tells me printing the flex spine from a soft material is what’s causing the slipping. I think it would be worth trying it with other, more rigid plastics. I’m also curious if you could add a second flex spine, with the rollers 180 out of phase to get more engagement? Would add friction, but could be even stronger holding power. 🤔 I love it all, very fascinating!

[@timseguine2]

Yeah, I have a feeling the teeth are getting squeezed out of the ring gear because it is too flexible.

[@fionnmerz]

Very cool. Having worked with collaborative robots during my day job, I can confirm that the full spec Harmonic Drives are backdrivable, and have the encoder mounted motor side. The wave generator is elliptical though, with a ball bearing race between it and the spline IIRC. Might be printable if you use a flexible material for the ball cage. And then bathe the whole thing in lubricant. At that stage the cycloidal reduction starts to look a lot more feasible. Looking forward to seeing where this goes next!

[@troelsnielsen2848]

Strain wave, yes, but in the video you pronounced is as weave, it's sounded a bit odd

[@albertox0404]

In theory you can connect 3 hall effect sensors to the VESC by default to get position feedback of the motor, or you can modify the source code to be compatible with your favourite encoder. Or my memory is wrong? 🤔😳 Thanks for your amazing videos from Spain 🇪🇸😁

[@rossveenstra3415]

I think the lower you try to flex that belt the harder it is to flex the belt intirely. Especially because the teeth are locked at the bottom. The higher up the easier it can flex and the easier it becomes for the motor to handle it. Try putting the bearings higher up. Because of the two teeth on either side sitting in the inner ring it is almost impossible the drive it back no matter how high the bearings are placed

[@markhorstmeier8734]

In the flex spine you are looking for a flex not a deformation. The long and short chords of the wave generator need to be calculated from the tooth profile. That is, at the long chord, the teeth of the circular spline and flex spline are fully engaged, but along the short chord, the teeth are just barely separated. The fact that you could rotate the flex spline when the wave generator was not inserted is probably not good. Diagrams often exaggerate the flex to show the concept, but the implementation is rarely so eccentric. With the minimum clearance, you can get more teeth at each major axis to share the load. More teeth with smaller depth maximizes the efficiency so you can get up to %60 total teeth sharing the load. Because you are flexing the surface, I think you can use a much more aggressive pressure angle than normal gearing. The reason that metal flexsplines are deeper cups are to allow the needed flex, but that is balanced by the twist of the cup which can distort the tooth geometry. I think that is why you were able to stop the rotation with your hands. If you are going to continue to use a TPE, you should use the hardest (PCTPE/TPA), switch to involute teeth, with an aggressive pressure angle, and calculate your wave generator ellipse dimensions based on your tooth working depth. Creating a wave generator with bearings all around the ellipse (at least at the major and minor axis) should also help keep your flexspline from deforming rather than flexing. If you go for a more rigid material for your flexspline, balance the depth of the cup to allow for flex, but minimize twist.

[@AaedMusa]

Very nice, maybe consider a planetary gear drive next!

[@DEtchells]

Several others have mentioned Levi Janssen’s approach using a compliant structure for the flex spline. I think this could solve some of the problems for you. Making the flex spline out of a harder material should greatly decrease frictional forces between the two splines, and harder teeth should also greatly reduce skipping as well. A rigid material normally wouldn’t work for the flex spline, but the compliant structure Levi designed fixes that. As an added bonus, the compliant structure also greatly (!) reduces the height needed for the flex spline, because the compliance is handled radially, rather than relying on the height of the spline to accommodate the needed amount of flex.

[@rewolff2]

Hi James. I was surprised when you were able tot turn the flex inside the teeth when the deformer wasn't installed. I'd think that it would work a bit better if you make the teeth catch the other teeth by say 30% in the "undriven" state. Also the teeth being triangualr, any force on the teeth will translate to a radial force on the flex part. If you make the teeth semicircular each tooth has a flat part that engages another flat part, so when you backdrive the thing you don't have the flex trying to move out of the teeth. That said: I think your conclusion is solid: the cycloidal drive is much better.

[@copy4862]

I have waited for this, James Bruton.

[@KurinP]

I'm curious if the TPU 'teeth' are just mushing to get out of the way in a stall condition, rather than being properly engaged. Perhaps making a tooth 'cavity' that you can press rigid teeth into, made out of PLA, so that each individual tooth is hard and positively engages with the outer gearing, but the space between each tooth is flexible allowing the whole mechanism to flex? It seems like having the whole mechanism flexible would cause issues under load.

[@anomamos9095]

If you want to use the motor as a spring and protect the drive train from shearing forces make a sprung gear. Using the flexible material make a tyre of sorts. Attach a drive hub to the motor and some bearings if needed then a spoked wheel made of flexible material then the hard gear. Having a flexible wheel will allow the gear to be driven backwards by a small amount without damage. There should be a choice of flexible material some that have a high rate of spring back and others that are more absorbent like Sorbothane.

[@KryptLynx]

I seen video of harmonic drive using complaint mechanism printed out of PLA as inner part. It destroyed itself on torque test, but part itself had design issues - complaint joints broke. Looks like it worth investigating.

[@danredebeer9733]

Personally, I would go for the cycloidal drive. As you said, there is less friction involved, and also, the deformable spline requires pressure to deform, which is lost forever, whereas the cycloidal drive does not. Apart from this, the cycloidal drive is backdriveable, and doesn't skip when this is tried, due to non-deformable parts. Of course, the harmonic drive might need extremely specific tolerances, however as far as i can understand engineering, use the thing that works well for what you need and has as few moving/unpredictable parts as necessary

[@MusikCassette]

two Ideas: 1. Make the flexdrive with dualextruding, with ridged teath. 2. The harmonic drives I have seen before don't mount the output to directly to the flexdrive, but have a second ridged tothring, that gets alined with the stator ring by the flexdrive.

[@Duraltia]

@12:56 _"Yooo tell me what you want what you really really want"_

[@neur303]

I'm pretty sure that although it is called flexdrive it should flex as little as possible, hence why it is usually made out of metal and has only little teeth. So maybe use some hard material and make the teeth smaller. Also experiment with the toothshape. I'd expect trapezoidal/square teeth to be beneficial since they hold the torque in direction of the material and would be less squished into the wave generator.

[@rklauco]

My suggestion - make the wall of the flexible insert thinner to remove the tolerance capacity. I believe, what is happening is that you simply squeeze the wall of the flexible insert so much that the bearings are able to pass around and will simply turn. So thinner wall should address this - give much less capability for the wall to flex thinner and allow the bearing to run around. But it's just a guess...

[@Anonymouzor]

Harmonic drives are mainly used in industry robots and from what i heard they are just so very very very reliable that it almost sounds too good to be true

[@karlpfalzner9648]

Video title prediction for next month: "Experimental Compound Planetary Reducer - 3D Printed"

[@xaytana]

From what I can tell, harmonic drive designs seem to favor a tooth difference of 2, and they seem to favor high reduction (high tooth counts), which makes a low reduction harmonic fairly rare. Most of the issues more than likely come down to tooth profile, just as involute teeth on gears are the optimal profile, there's more than likely an optimal profile for harmonic drive teeth that's dependent on tooth offset, as a different number in offset requires a different movement of the strain wave, where a design meant for an offset of 2 may not necessarily scale to an offset of 4. It's just not as simple as having a triangular or trapezoidal profile, as the flex spline itself is moving in and out while also rotating; again, involute gear teeth are optimal for their movement, a flex spline will also have an optimal tooth profile. There's also loads of different designs. One uses a compliant mechanism for the flexing rather than transitioning from a flex spline to a hub via a solid piece. Another design uses multiple sets of bearings to better mimic the desired shape of the flex spline, rather than relying on just two bearings to generalize the flex spline shape. Material choice is also important. Tolerances are also important. Etc. Deformation is an issue that plagues printed variations of harmonic drives, there's a reason why actual ones are all metal, either to the teeth and/or the wave generator bearings not being square. The downfalls seen here could easily just come down to geometric issues that need solving. It could just as easily be a tolerance issue. Like all developmental projects, this just needs refining and finessing to get it to work. A cycloidal drive is probably the better choice for the reduction you're wanting, there's a reason why low reduction harmonics are rare, if not entirely unheard of, and I don't think any hobbyist wants to reinvent the wheel just for a singular project when other wheels are more readily available. Maybe save looking into harmonics when you actually need a high reduction drive, where they're most commonly used.

[@knethen]

strain *wave* not strain *weave* , right?

[@swannschilling474]

I designed a 40:1 Cycloidal Drive which is still backdrivable and super compact, the only drawback is the cost and also the weight of the bearings if you would use it in a robotic application like a quad...keep up showing us more reducer options! This is super interesting!! 😊

[@simonmerrett]

Please could you show us @Swann Schilling?

[@swannschilling474]

@@simonmerrett Hey, you can find the content on my channel, please feel free to check it out! 😊 I also made all the printing files available...but it seems like I cannot post any links here, so just check the description of the video on my channel...hope you'll like it and please share your thoughts in case you'll make one for yourself!

[@zenithparsec]

You could put a slightly compliant coupler with a strain gauge on it between the output and the axle, and then you could use that emulate backdriving the motor. Just a thought.

[@jamesbruton]

But then I can't dynamically control the springyness with software

[@eldricliew6223]

If you apply too much force, the tpu buckles. You probably want a stiffer material that can deform a few mm at most to prevent slipping

[@jawawookie]

Harmonics are meant to higher motor speeds with higher ratio numbers like 1:40+ if the output speed is the same as the cycloidal it should be a lot closer in results

[@noiwonttellyoumyname.4385]

Harmonic drive has a *very* high starting torque, due to the deformation that you've got to cause. I've got an actual branded Harmonic I picked up a while back, and it requires something like 40 oz-in *just* *to* *start* *turning,* so you have to subtract that amount from the input torque when you're trying to figure out your output. For the motor I was using (430 oz-in NEMA 23) it was no problem, as that left me 390 oz-in at 100:1, so plenty of output power. You're running at much higher speed than me, as well much lower reduction. (And probably much lower torque as well.) I'm not surprised that you could stall the strain wave gearbox.

[@seandaugherty8931]

The teeth are climbing each other as rotation is inhibited, squishing them at point to point because the leverage is significantly against them. You probably have minute warping as the inner PLA shell is warping also to allow. Maybe if the material was a denser, non flexible type and lubricated so the teeth are forced to slide into being seated into the outer teeth.

[@TheDudeITA]

Wave generator shaped like and cross with a long section doing the driving and a short section keeping the tension throughout the flexi gear. I think this could help with the high speed issues

[@kinzieconrad105]

One day he is going to get a crazy amount of torque and he is going to grab those bolts and get a cut to remember!

[@Brandon_The_Engineer]

Been loving these videos on reducers. I was trying to make a 3d printed harmonic drive myself a while back. I think one thing you could try is to add additional bearings onto the wave generator so the bearings form a sort of ellipse shape to keep the flex spline from flexing in an undesirable way. I think this may solve the issue you were having when the motor would rotate but the output was able to be held in place.

[@omniyambot9876]

Sir I recently discovered your channel and this is perfect for learning robotic concepts. But may I ask if you even rest sir? You're so intelligent and productive.

[@matsv201]

A active brain need something to think about to rest.

[@streaky81]

Might be handy for CNC projects, motors that should be stopped being driven by external force is a constant pain and commercial drives are super expensive. This has the speed and some torque to it, maybe the torque could be improved, plus a decent driver and the encoder and you're onto a winner.

[@Keksstar]

When James printed the cyclodial drive I already thought: I could bet, the next thing he does is an harmonic drive. And he did! :-)

[@mitchelldalton7405]

Id say theres less surface area in the harmonic drive where the teeth meet , which means less holding pressure, coupled with the flexibility of the inner gear, the teeth are probably deforming under load. Id say the cycloidal drive is much more efficient and more durable, though I suppose the cost is a bit more. You could always swap some of the bearings out for nylon with metal studs and some silicone grease, to keep the cost down. Either way good stuff once again.