Harmonic Drive

Just for fun.....

Harmonic Drive looks like a fun way to get zero backlash, high stiffness reduction for cheap higher speed motors.

[en.wikipedia.org]

The docs etc suggest metal components but I feel this should be doable with plastic components very easily.

The flexible spline could be made from a wide timing belt.

The Circular spline could be moulded using polymorph or some such with a longer timing belt (turned inside out with a disk as internal support) as a plug.

The rotor or wave generator could possibly be two wheels either end of arms the arms driven centraly from a cheap motor.

What do you guys think ??

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Hi Andy,

i have a precise plastic-model as demonstrator for a circular harmonic drive.

It works, but has noticeable inherent friction, so you'll receive heavy wear over time in continuous mode.

I think you must select the materials of the gears and spline very carefully and need much higher surface-accuracies than the actual reprap-resolutuion ...

Viktor

Here is a variant with planetary gears that looks like it would be lower friction and more easily replicated as it does not need the elliptical ball race.

[www.orlin.co.uk]

---

[www.hydraraptor.blogspot.com]

Hi Chris,

it's not so easy replicable too - some years ago i made tech-transfer with HarmonicDrive ( [www.harmonicdrive.de] )

... something with developing a microasembly-station exactly for this gearbox (so i have one prototype left from this time)

The flex-spline and the gear-rings were made by LIGA with X-ray and galvanics with extremely high surface-accuracies for lowering the outwear.

And for the friction: - the two outer rings are displaced by form-shaping with pressing the flexspline on them - here is the maximum of friction and force-transfer.

Viktor

Hurro nop & viktor

I like the ideas on the orlin website, hadn't thought of the planetary gear idea for input nor on the output method. Clever.

I was just free wheeling some thoughts whilst waiting for my lathe spares to arrive (I broke it, or at least the wussy plastic drive gears, have ordered some metal replacements, and doing the bearing upgrade at the same time)

Agreed, the reprap surface accuracy may cause problems. I guess this was why I was thinking along the lines of timing belts. Much as I want to move away from them because of their expense.

If we could find a way to make it fly the high ratio's, high stiffness and lack of backlash are desirable qualities.

Cheers

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Greetings all,

In my experience (built a robot wrist with them in grad-school), harmonic-drive mechanisms don't exactly have backlash, but they have something similar -- nonlinear stiffness, with significantly lower (softer) stiffness around zero torque. This can cause control-loop problems if not handled correctly. However, having said that, they're still useful for getting big reduction ratios into a small volume. Just remember, they're clever, but not magic.

---

Larry Pfeffer,

My blog about building repstrap Cerberus:
[repstrap-cerberus.blogspot.com]

Hurro Larry

Just ever so slightly off topic. I got hold of some stainless steel M6 Long Nuts in the end (18mm long). Turned down the middle 6mm to 8mm ish dia and made a thermal break plug M6 by 6mm long out of ptfe (with a 3mm hole through the middle) and put this in the middle of the turned down long nut. Broke my lathe turning the aluminium heat sink for the cool zone but will do more when I get it fixed. How are your extruder experiments going ???

Back on topic the observations re your experiences of harmonic drives I think I understand. Can you explain what you experienced as softer stifness around zero torque ??

Cheers

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Andy,
I am having trouble visualising what you are describing, any chance of a pic?

BTW,
I just got the all stainless steel thermal break idea working Larry. [hydraraptor.blogspot.com]

---

[www.hydraraptor.blogspot.com]

Sorry guys

This may help.

It is a bit of each of your designs mangled together. There will me a 10mm by 1mm peek thermal washer between the heat sink and long nut when the lathe is back in commission.

The hot end will have a peek thermal shield/insulation around it (no drawings etc yet as I need to sort out the heater end, but got the resistors ready.

Cheers

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

aka47 Wrote:

>
> Back on topic the observations re your experiences
> of harmonic drives I think I understand. Can you
> explain what you experienced as softer stifness
> around zero torque ??

Andy,

Pretty much any mechanical transmission has a finite stiffness and a finite mass. Together these limit the frequency of vibration that the transmission can transmit. Some transmissions have backlash, which "consumes" some of the amplitude of motion a transmission can transmit (to get one radian of motion out, one needs to put in more than one radian's worth of input motion.)

The flexspline in a harmonic drive gives something of a hybrid of these two effects, especially around zero torque. (One can think of backlash as analogous to the stiffness approaching zero -- although it isn't quite that bad.)

Where this gets to be a pain is when one is closing a servo loop from the motor to a sensor on the other side of the harmonic drive. In this case, the change in stiffness results in a change in the resonant frequency in the transfer function from motor (current, say) to whatever the outboard sensor is. Designing a high-performance servo loop, in the presense of this variation is a tremendous PITA!

I'm a bit under the weather today (I think I have the flu, !@#$%)
But some of the basics are touched on in an old paper of mine: [homewood.stanford.edu]
In this case, I *deliberately* designed to avoid the problems of harmonic drives, but one can do through a similar derivation, letting the gear ratio become larger (and similarly let the damping "b" increase.) The resulting transfer function becomes a bastard to control, especially if K varies wildly, as it does for harmonic drives. If memory serves, this was one of the problems that NASA's flight telerobotic servicer had to contend with.

With some hindsight (and more computational power), I'd try to use either non-linear control (e.g. gain scheduling) or adaptive control.

Hope this sheds some light (as opposed to more confusion),,

-- Larry

Hi, guys,
Nice topic! Especially since I've been dabbling in harmonic drives for almost a year. I'll take all your replies and respond individually, where I see the need for a reply:

@aka47: I happen to know exactly what metals are used in harmonic drives. high-end, martensitic-hardened stainless steels. hard-core shit and for good reason. plastic is OK as long as you only need a working toy. if you really need arc-minute-backlash under torque, forget about plastics.

@Victor (vdx) is absolutely right

@nophead: I think the compound planetary design from ORLIN is actually not their design (my guess is they're manufacturing under patent) - it's the design used by the micro/nano division of HarmonicDriveSystems (see here (http://www.micromotion-gmbh.de))

@Larry_Pfeffer: actually, harmonic drives DO have backlash. Even the guys at HDS admit it in certain documents - should you look deep enough. "Zero backlash" is nothing more than a marketing gimmick. Furthermore, they tend to have more backlash in one direction than in the other (for instance: 0.67 arc-minutes CCW and 0.76 arc-minutes CW) - and this, coming from "the guys" themselves. However, other than this observation, your notes are very insightful, intriguing and lecture inspiring. Too bad your site is down

Now, to complete with my own 2 cents:
I too built a spin-off strain-wave generator reduction (similar principle, but eliminating the flex spline altogether) and achieved as far as 0.55 arc-minutes of backlash (tested with a laser beam mounted on the output shaft and pointing at a target 8 meters away). But that "miracle" only lasted for 2-3 hours of continuous running. After that, the gears (currently made of brass) ground and widened enough to induce as far as 20 arc-minutes of backlash. To give you a better picture: I've done the calculations and every 3 microns of radial error induce - in my case - one arc-minute of backlash. That's good enough reason to have a new pair of gears currently made, out of 55 HRC hard steel, this time Keeping fingers crossed here!
(our current CNC only works with hundredths of millimeters, not microns - so for now I need to accept lower resolutions and hence larger backlash in the prototypes - but I'm hoping to find better machining solutions some time soon)

Anyway, here's my current prototype in action, grinding away on a lathe.