Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Harmonic drive tests for PE

  • Please log in to reply
16 replies to this topic

#1 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 14 April 2024 - 12:01 AM

Preliminary results...

 

A couple of weeks ago I was able to spend a day with Dan Gray and Danial Wynn in the Sidereal Technologies' shop testing a couple of Harmonic drives.  The drivers are #20 frame size (?) and 160:1 (well 161:1 mechanically).  The test setup was on the axis of one of Dan's direct drive motors with 2^26 tick encoders on both input and output (~67m ticks).  So the input shaft to the motor has the input encoder on it: no gears belts are flex adapters.

 

We have more to do.  These are brand new drives (so break in or different lube may help) and operating in no load.  The drives appear very well made and are well documented (drawings, models etc.).  Yes, these are out of China

 

The adapter is also out of square by 0.0005" on a 25mm interface which wobbles the 200mm encoder by 0.005" (oops, my bust).  Not enough to be very meaningful but we want to correct that and try again. Data is completely raw, no PEC or other correction has been applied and there's no smoothing of the graphed data.

 

Graphs and setup below:

  Nominal Periodic error RMS ~ 45 arc-sec

  Nominal Peak to Peak           120 - 145 arc sec

 

Vertical scale on the graphs are in arc-seconds.  I think the Horizontal period is 2.25 degrees per period it may be half that.  Will resolve that in the next test though it's in the numbers if I look hard enough.

 

Test setup.jpg

 

Drive-1.jpg

 

Drive-2.jpg

 

 

If anyone has ideas on how to smooth these out, please let me know.

 

Thanks,

 

Greg


  • TopherTheME likes this

#2 C. Thieullet

C. Thieullet

    Vendor - astromist

  • -----
  • Vendors
  • Posts: 68
  • Joined: 18 Dec 2004

Posted 14 April 2024 - 03:51 AM

Hello Greg,

 

The result are very unusual for an reducer of this size.

With a 14 size PE can be around +/-15 at the best for an hole reducer and up to +/- 40

With a 17 the regular value is +/- 20

With a 20 it is around 10

With a 32 below 10.

 

I tested the fourth on real mounts. They are accurate in a repetitive way.

 

Even using second hand reducer, My "old" Crux 320 with its 32 reducer is around +/- 5 without PEC correction and guide at 0.3/0.2 total rms on stable sky with PEC enable with more than 40kg OTA load.

 

So the rule : the bigger the better.

 

The PE depends greatly of the accuracy of to position of the wave generator. 0.1mm variation in longitudinal position can create a larger PE. There is a sweet spot.

Lateral error introduce a dissymmetric shape in the PE graph a part of the sinusoid is lower than the other part. At the extreme, it is flat. That's call dedoidal moves. That can destroy the reducer.

 

That means if we put a geared motor reducer before, we take the risk to have a play in longitudinal that will enable the wave generator to move its longitudinal position upon the model of the harmonic reducer.

With right rotation, the wave generator will pull on the motor reducer axis and will go deeper, With right rotation, it will push and it will go back.

If the axis of the motor reducer has a small tilt play or if the motor attachment is a bit tilted, this will degrade the wave of the PE upon the load or sky position.

After one year of such process a reducer can have its native PE to go from +/- 20 to +/- 80.

I did this observation on several 14 reducer models.

 

Last, small geared motor reducer have usually a huge backlash (>1°), and upon the load and position on the sky this backlash will produce unwanted spike and peak into the guiding that can be huge.

 

Several mount maker solved this issues with belt. Other by using an other harmonic reducer instead of the geared reducer and a new one by linking the motor shaft directly to the reducer.

 

Of course adding an encoder at the end enable to hide all the trouble but this do not solve all the degradation in the long terms due to bad alignment of the wave generator.

 

That's why hollow shaft reducer are the best for us. Everything is under bearing and at the best position for years. Two drawbacks : they are heavier and there is no way to adjust anything.

 

That's the kind of model you choose ;-)

Only Zwo seems to understood that's. And they decided to test all there reducer. That's the only way to avoid putting a bad adjusted one in a mount.

Several (All ?) other did not and client have various trouble from time to time (peak in guiding, slip, irregular PE).

 

So in your case with this hollow 20 reducer, I am afraid you have no way to make any adjustment.

According to me such PE might indicate a product with a wrong wave generator longitudinal position, or a really bad / cheap product ? Last option would be a second hand reducer but yours is new.

 

On the model I tested, new Laifual hollow reducer 14 had a PE around +/-20.

I can give you the contact I have by pm

 

By the way say hello to Dan for me next time you see him ;-)

 

Cyrille



#3 luxo II

luxo II

    Cosmos

  • *****
  • Posts: 9,531
  • Joined: 13 Jan 2017
  • Loc: Sydney, Australia

Posted 14 April 2024 - 04:11 AM

Close-Loop Control Systems 101.  (Yes I am an EE).

 

That’s not PE, that’s oscillation because you have too much gain in the feedback path.

 

- reduce the maximum guide pulse duration,

- reduce the guide rate to 0.5x or even 0.25x,

- reduce the aggressiveness in the guide software.

- use BIN2 in the guide camera instead of BIN1.

 

Reducing any of these by factor of 2 should tame the oscillations.


Edited by luxo II, 14 April 2024 - 04:15 AM.

  • Lumix.guy likes this

#4 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 14 April 2024 - 10:23 AM

Cyelle, Great information.  Yes, PM the contact at at Laifual, they are their drives.    I was hoping for < 20, but so far not the case.  When we tested we ran the input at about 6 rpm, as tracking rates would take all day.  Could that be part of the issue, doesn't seem like it should?   Will pass on the hello to Dan.

 

Luxo, Gain?  Motor encoder, solid shaft, harmonic drive, readout encoder... No guiding, strictly mechanical.   Unless there is backlash and full reversal, even oscillations would not matter with this setup.

 

 

Greg



#5 C. Thieullet

C. Thieullet

    Vendor - astromist

  • -----
  • Vendors
  • Posts: 68
  • Joined: 18 Dec 2004

Posted 14 April 2024 - 01:37 PM

Greg,

 

Thanks :-)

 

I did usually PE measures at x15 sideral rate and PE remains consistent.

So I would be surprised as well if 6 rpm as input can have an impact. But why not.

You can do the test during 30mn at sideral rate and see the results.

If the PE is lower it might be a factor to adjust somewhere in the computations.

 

I sent you the contact.

 

Cyrille



#6 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 14 April 2024 - 02:34 PM

Cyrille,

 

Will try that on the next run.  We did not try the encoders without the Harmonic drive either, that would help verify a few thing including our math. 

 

From your results it sounds like we are out by almost a factor of 10... always disconcerting :-)

 

For those reading this, I would be much happier to find out I made a mistake... that would be much easier to deal with.

 

Greg

 

Edit- I just went over the spreadsheets again and if there is a math error, I sure don't see it.  I would think any external mechanical error would show up as something larger than the arc seconds.  Will add a few more indicators to the next attempt.


Edited by gregj888, 14 April 2024 - 03:00 PM.


#7 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 16 April 2024 - 11:43 AM

Quick update-  I think we have an issue with the test setup.  Can't be sure until we try another run, probably next week.

 

Do not rely on out results above at this point!!!

 

Greg


  • PrestonE likes this

#8 Sean Cunneen

Sean Cunneen

    Let Me Think

  • *****
  • Moderators
  • Posts: 4,709
  • Joined: 01 Aug 2007
  • Loc: Flossmoor Il.

Posted 21 April 2024 - 11:20 AM

We are locking this thread for the moment as OP wants to make sure their numbers are accurate

#9 Sean Cunneen

Sean Cunneen

    Let Me Think

  • *****
  • Moderators
  • Posts: 4,709
  • Joined: 01 Aug 2007
  • Loc: Flossmoor Il.

Posted 25 April 2024 - 01:33 PM

Game on... Proceed....



#10 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 25 April 2024 - 03:46 PM

Thank you Sean.  OK, the saga continues with a detour though my email's spam blockers (long story).

 

First a shout out to Kathy at Laifual, She/they have been great to work with.   Long story short, Laifual tests the PE on their drives, at least they did with mine and provided their graphs.  Bottom line is on drive has 29 arc sec of PE, the other has 33 arc seconds.

 

Testing.  Laifual sent me a short video of their test setup.  It's an industrial quality setup that is effectively the same as ours. Encoders on the input and output so much more. IMHO, reliable than on sky measurements as other mount variables like sag and seeing do not make it into the measurements.

 

The graphs are pretty interesting.  The actual short term PE is 10-20 arc sec and if you orient the drive correctly it's about 10 arc seconds.  If I understand it correctly, that's from a published paper(s), a Byers worm has about 3-10 arc-seconds of raw PE.  Another paper put an LX200-16 at 40 arc seconds as received.

 

   "Astronomical Telescope Mounts Without Periodic Error" -- Byers

    Sorry, can't find the other at the moment.

 

So at the moment, I'm not sure I'll burn more of Dan's time to retest unless he is really interested.  An RMS number and a look at the actual wave form would be interesting.  In any case, PEC is probably a good thing though it may need a long training time.  It might be better to model the entire system or simply but an encoder on the axis and correct to that.  It does look like there's not a lot of functional difference between the harmonic/strain wave reducers and any other gear based reducer in that all need some form of correction/ guiding/modeling for highest precision.

 

精度2.jpg



#11 duck

duck

    Apollo

  • -----
  • Posts: 1,487
  • Joined: 11 Jun 2020
  • Loc: madera ca

Posted 25 April 2024 - 05:39 PM

the period of the primary 10 arc-sec error is?  Can't tell from the graph...too time compressed.  Is the period even measured? It would depend on the axles speed during the test, correcto?


Edited by duck, 25 April 2024 - 05:43 PM.


#12 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 26 April 2024 - 12:56 AM

Duck, the time period will vary with speed but I think the peak to peak deviation will be pretty consistent.  160:1 reducers so 2.25 degrees or 6.6 minutes for one input shaft revolution, if I did the calculations correctly.   The graph shows the angle deviation from where a perfect reduction would be based on the input drive for one full rotation of the output.  This would be the same periodic error seen in a worm or any other gear drive.  In this case the important thing is the envelope though some analysis of the wave form would be interesting.  We kind of got that from the first tests we ran.

 

These aren't great when it comes to PE but they are compact, strong, less expensive in most cases and seem easier to integrate than say a worm.  If we can build and control a direct drive system, then these too should be controllable.  In effect it moves the stress from the mechanical tolerances to the electronics/software side of thing.

 

The question now is does the control system cost as much as a DD system, if so then the Harmonics might be a poor choice.  My current guess is they will not but that's TBD.

 

There are other questions too.  Looking at the graphs I wonder how they will age and will aging take down the  single rotation input and the single deviations some.  Would other lubricants and super clean assembly help? 



#13 duck

duck

    Apollo

  • -----
  • Posts: 1,487
  • Joined: 11 Jun 2020
  • Loc: madera ca

Posted 26 April 2024 - 12:45 PM

the period of the unresolved hash in the plot looks a lot shorter than 6.6 minutes if the total duration represents 24 hr.  There would be only about 220 periods in the plot, which would have been resolved.



#14 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 26 April 2024 - 01:56 PM

That plot came off their tester.  The drive is 160:1 so 160 revolution of the input.  No idea on the plot resolution.  This graph was probably generated in a minute or two minutes, much higher speed than we would have run it.  In our plot we did 30 revolutions of the input, theirs shows 5x the number of revolutions.

 

If I put the drive on an equatorial mount and I picked the correct placement I could get an overhead 120 degree wedge with pretty good, ~ 14 arc sec, PE.  In that case at tracking speed one short period (look at our graph) would be about 6.6 minutes.  The long period of their graph would be 360 degrees, so one day or work in the circumpolar regions.

 

I'm not suggestion this as a solution nor stamping my feet that it's a problem.   Just information for the decision tree.  I'm pretty sure at this point that I will go ahead with them for my 20" mount, but that's not certain either.

 

Greg



#15 Laplase

Laplase

    Sputnik

  • -----
  • Posts: 41
  • Joined: 23 Nov 2023
  • Loc: Russia, Ural

Posted 26 April 2024 - 02:01 PM

1 period is half the rotation of the ellipse
24*60*60/160/2 = 270 sec, or 4.5 min

#16 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 26 April 2024 - 04:28 PM

Laplase,  correct, sorry and thank you.  Not sure what I fed to the calculator for the 6.6...  but agree with the 4.5 min.

 

The harmonic drive moves 2 teeth per input resolution (the ellipse). So you get 2 "cycles" per input revolution.

 

Greg



#17 gregj888

gregj888

    Gemini

  • -----
  • topic starter
  • Posts: 3,108
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 13 May 2024 - 11:30 AM

The continuing saga.  After reviewing the graphs from the vendor and talking with Dan, I wondered if a break-in period might be appropriate.  So a simple fixture with a couple of 3D printed timing belt sprockets and capstan with a belt I had on hand, a stepper and Arduino CNC board... in business. 

 

Coming up on 24 hours of run time. The weight is 2.5 lbs, light but the thought is more polish than deform.   With the gear ratio it can do a lot more but this should tell me (us) if it matters.  3 turns to unwind 3 to wind the other way then reverse.  It takes about 7 minutes for cycle or 1 min 10 sec per revolution of the output.

 

breakin-front.jpg

 

Breakin-back.jpg




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics