Motor gearbox needed with belt drive?

229 views
Skip to first unread message

T J

unread,
Mar 8, 2020, 11:40:51 PM3/8/20
to OpenPnP
It seems like the resolution of the PnP belt drive is rather poor with typical dimension driving pulleys and a 1.8deg stepper. It seems like at minimum a 0.9deg stepper would be needed. But those are a bit more rare and the torque drops. Apparently BDLC motors (like ODrive system) would have more accurate position control than the steppers, but perhaps not enough for belt drive?

So are you using NEMA 23/34 (planetary) gearboxes to improve motor resolution? If you are using gearbox + feedback, have you installed the encoder after the gearbox? Is that needed to tackle the gearbox backlash?


Jarosław Karwik

unread,
Mar 9, 2020, 1:38:38 AM3/9/20
to OpenPnP
I have seen more often dual belt systems ( with something like 3...4:1 ratio) more often as it is much cheaper and does not need compensation.
Good gearbox is quite expensive (especially low backlash)

T J

unread,
Mar 9, 2020, 2:45:38 AM3/9/20
to OpenPnP
I was thinking something like this for a gearbox for an ODrive BDLC motor for example:
https://www.amazon.com/Planetary-Gearbox-Reducer-Stepper-Precision/dp/B06XDVS95M/

Michael Anton

unread,
Mar 9, 2020, 2:49:46 AM3/9/20
to OpenPnP
Yes, the two commercial PnPs I have use belts for doing the reduction.  Now, these machines use servos, so reducing the motor RPM, increases torque, and it reduces the speed, but a 3000 RPM motor is a bit overkill for driving a belt driven PnP anyhow.  I'm not sure exactly what the reduction is, but it looks to be somewhere from 2:1 to 3:1.

I'd say you would have far less backlash when using GT2 belts, than you would with a gearbox.

Jim

unread,
Mar 9, 2020, 4:36:09 AM3/9/20
to OpenPnP
I'm using on y-axis 3:1 transmission, but it only increases torque for the motor.
It does not help at all to remove belt backlash or linearity problems.

gear.JPG


Marek T.

unread,
Mar 9, 2020, 5:23:37 AM3/9/20
to OpenPnP
My commercial machine has ballscrews for x and y. But ballscrew are driven from dcservos with short (some 25cm) belts. Motors are 3000RPM with encoders. Poor accuracy (non linearity) because of old ballscrew forced me to add one cheap Chinese optical linear scale along the X. Zero backlash. Belts fiberglass enforced.

T J

unread,
Mar 9, 2020, 4:59:27 PM3/9/20
to OpenPnP
How about using microstepping and encoder feedback? Can I assume that with microstepping I could have 10x more true motor resolution, i.e. 0.18deg instead of 1.8deg for example? I'm thinking about the Anthony Webb design, where there is the NEMA 17 hollow shaft stepper used. It narrows down the possible motor options unless the design is radically changed.

Jarosław Karwik

unread,
Mar 9, 2020, 5:01:49 PM3/9/20
to ope...@googlegroups.com
If you get 4..8x then you are already lucky. Mechanical errors and magnetic inaccuracy in stepper do not allow going further

pon., 9 mar 2020, 21:59 użytkownik T J <tuomas...@gmail.com> napisał:
How about using microstepping and encoder feedback? Can I assume that with microstepping I could have 10x more true motor resolution, i.e. 0.18deg instead of 1.8deg for example? I'm thinking about the Anthony Webb design, where there is the NEMA 17 hollow shaft stepper used. It narrows down the possible motor options unless the design is radically changed.

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/d9137e8f-5e50-48a5-95bb-8322e5730a1e%40googlegroups.com.

Michael Anton

unread,
Mar 9, 2020, 6:43:37 PM3/9/20
to OpenPnP
I believe the Smart Stepper drivers would do that: http://misfittech.net/.  I think you can get just over 0.1 degrees with them.

ma...@makr.zone

unread,
Mar 10, 2020, 4:20:12 AM3/10/20
to OpenPnP
I replaced my 0.9° steppers, they were too slow. Using Trinamic drivers you get up to 256 microsteps. In PnP you have no lateral loads to speak of (at least when it matters i.e. when picking or placing parts) so the microsteps will be accurate enough IMHO without encoder feedback.

When I worked on runout compensation and bottom vision I was amazed what the simple Liteplacer (mod) belt machine can resolve. I've observed single digit µm compensation resolution and ~15um repeatability if I remember correctly. IMHO this is plenty for all PnP operation, surely for 0402 and maybe smaller.

_Mark

Jim Freeman

unread,
Mar 10, 2020, 8:25:25 AM3/10/20
to OpenPnP
Hi, Mark. When you say Liteplacer (mod) what mods did you do?  Just the drivers?
Best, Jim

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

ma...@makr.zone

unread,
Mar 10, 2020, 11:31:53 AM3/10/20
to ope...@googlegroups.com

1.8° Steppers and the Smoothieware running Azteeg X5 GT controller with Bigfoot BSD2660 - Based on TMC2660 drivers. Beefed up the Y stepper to NEMA23.

https://makr.zone/pick-place-machine-first-simulated-small-test-run/66/

https://www.panucatt.com/azteeg_X5_GT_reprap_3d_printer_controller_p/ax5gt.htm

_Mark

Jim Freeman

unread,
Mar 10, 2020, 2:19:23 PM3/10/20
to OpenPnP
It looks great! A few of questions: 1) why did you switch out the TinyG ?   2) Which of the mods do you feel is most important to improving the resolution? 3) Did you have any particular procedure for nozzle / down camera / upcamera alignment?
Best,
Jim


--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

ma...@makr.zone

unread,
Mar 10, 2020, 4:13:10 PM3/10/20
to ope...@googlegroups.com

Hi Jim

First, I'd like to clarify that I'm still stuck with tests, I have never placed real full boards! I keep getting sidetracked ... ;-)

> 1) why did you switch out the TinyG ?

TinyG was and still is recommended against on the OpenPNP page:
https://github.com/openpnp/openpnp/wiki/Motion-Controllers#tinyg
See also the "quirks" page linked there.

Plus I wanted one more axis for my PCB conveyor (still work in progress). Plus I liked the Azteeg X5 G and Bigfoot modular drivers with more amps*, Trinamic driver options, SPI control and the USB isolator.

*Note, that I since learned, its not the amps that count for speed, but the voltage!

See also the similar question by Дмитрий Голик on the YouTube page:
https://youtu.be/poLwQDpfeJ8

> 2) Which of the mods do you feel is most important to improving the resolution?

Trinamic drivers!

> 3) Did you have any particular procedure for nozzle / down camera / upcamera alignment?

Yes, I wrote the bottom camera location and rotation calibration plus the tool specific bottom camera offset calibration code in OpenPNP.

https://makr.zone/improved-runout-compensation-and-bottom-camera-calibration/346/

Plus I improved pre-rotate bottom vision to work "smart" i.e. do multiple passes when errors are large (precision) and only one pass when not (speed).

Both together were the "key" for very accurate large 0.5mm pitch part placement for me.

While developing the BlindsFeeder, I also learned that the machine is quite accurate, despite being a very simple belt machine. I believe that many underestimate the accuracy of these machines, just because they are simple and cheap.

But having said that, I also believe that it is difficult to get the belt tension just right.

Because I felt so completely helpless with that, back when I set the machine up, I read a lot on that on the internet. Unfortunately - stupid me - I did not bookmark the best sites and I can't seem to find them again.

I just remember that listening to the frequency of the belt when plucked was the key "aha" moment for me.  On a 750mm machine they must sound very low. The belt feels surprisingly soft when that is the case. So I believe that many people overstretch their belts. And beyond a certain point that's probably damaging and the belt must be replaced.

One indicator: your steps per mm must absolutely match the theoretical value. If that is not the case, go back to square one.

_Mark

Jim Freeman

unread,
Mar 10, 2020, 6:17:04 PM3/10/20
to OpenPnP
Thanks, Mark. Very interesting. Let me digest this. When you say that the steps per mm must match theoretical, I am puzzled. There will be "as built" variation in the pulley size, ... so getting a non-theoretical calibration would be expected.
Best,
Jim


ma...@makr.zone

unread,
Mar 11, 2020, 3:55:10 AM3/11/20
to ope...@googlegroups.com

Hi Jim,

The pulley diameter does not matter (within reasonable tolerances, of course). It is only the number of teeth on the pulley that count, literally.

In the end it only matters, how many teeth the pulley transports per revolution. If the belt stretches a bit or bunches up a bit over the pulley, that may matter for friction and wear an tear, but not for steps/mm. The belt can't slip the teeth. So the only thing that matters for steps/mm is how many mm one tooth spans along the belt, as tensioned in the machine. So the belt tension alone gives the steps/mm. And if the belt is good quality, that does match the theoretical spec values (speaking for the level of precision needed in the field of PnP) as long as you tension the belt correctly.

From my config.txt

alpha_steps_per_mm 80  # Steps per mm for alpha stepper: GT2 2mm pitch 20 tooth pulley @1.8° stepper = 200 steps/40mm x 16 microsteps = 80

On my machine I observe an error of about 0.2mm over 400mm (if the millimeter parer I use is to trust). That is completely acceptable because I'll never have a PCB that is 400mm in one piece. Remember that truth to scale only matters within objects that are in one piece on the machine (and with OpenPNP's Affine Tranform on the PCB, even that does not really matter).

What does matter is repeatability. You capture a coordinate for a feeder's pick location today and you expect the machine to return to the same location precisely tomorrow. And If I say precisely, I mean +/-100µm or so.  That is IMHO sufficient for all PnP at 0402/0.4mm pitch (which is my target). For 0201 I guess you'd have to half that.

There are many discussions about precision here in the group, that may be reasonable for large scale metal working CNC machines but are - sorry - nonsense for PnP. 

Think about how much even a 0402 part can wiggle in the punched paper tape pocket. Or how the parts will self-align on the pads due to surface tension in reflow.

_Mark

Reply all
Reply to author
Forward
0 new messages