Delta PnP

[Glen English]

I have moved the discussion to this (new) topic, off the robot arm forum.

any of the deltas can be used, just of course, use a vac nozzle  instead of an extruder

The delta doesn't even have to be particularly good (in terms of print performance) to do the job well.

with a spring vac nozzle there is considerable license for Z error

Use a NEMA8 hollow stepper to rotate the part (my choice)
OR
you could rotate the build platform instead..... but that is really just a different complication.
IE on a turntable. but introduces a whole new level of errors and compensation

parts would live around the edge in little containers, stereo vision on the effector for height measurement of a component in a jumble of loose parts in a container  and a better look at things.

the up and down cameras can have IDEAL lens because the Z height  is totally whatever you like....
so something like 12 to  20 mm lens...

for measuring part height (like my Yamaha)  you could have a side mounted camera off the side of the turntable that the part could be bought into the ideal vision location for (against a background on the other side).

[Glen English]

Prototype goals :

I have been watching the google groups delta 3d printer forum for a few weeks.

340mm plate,  100mm Z max height, MGR12 bearings, Magnetic ball 440mm long arms,

2020 with metal corners

Just getting parts so far.....build in March...

Delta with its natural Z movements should be a killer.  I think Firepick has a working one.

Certainly for large build area and from feeders and HEAVY head, ball screw cartesian is the way to go

with CoreXY as a close 2nd for lightweight heads

Use a 'build plate' (the bottom plate)  with printed or laser etched fiducials on it 

[Jason von Nieda]

Hi Glen,

It sounds like you are describing a linear delta. FirePick was a rotational delta and it does not work for pick and place. It is incapable of producing better than about 0.5mm accuracy even with extensive calibration and a pile of interpolation. If you run through the math it's pretty clear why. The FirePick project is effectively dead.

Linear deltas are more accurate, but also require pretty extensive calibration. My understanding is that a lot of that calibration is now built into some of the 3D printer firmwares. There is an inherent "bowl" effect that has to be mapped out and that is done with z-probing. That said, I'll be pretty surprised if you can hit 0.1mm repeatability with one for pick and place. A lot of the error that in inherent in deltas can be pretty easily ignored for 3D printing because the process is so inaccurate to begin with. With pick and place it's a different story.

Jason

On Thu, Feb 9, 2017 at 5:08 PM Glen English <gleneng...@gmail.com> wrote:

Prototype goals :

I have been watching the google groups delta 3d printer forum for a few weeks.

340mm plate,  100mm Z max height, MGR12 bearings, Magnetic ball 440mm long arms, 

2020 with metal corners

Just getting parts so far.....build in March...

Delta with its natural Z movements should be a killer.  I think Firepick has a working one.

Certainly for large build area and from feeders and HEAVY head, ball screw cartesian is the way to go

with CoreXY as a close 2nd for lightweight heads

--
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 post to this group, send email to ope...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/0395ceb4-10c6-47ce-98b1-e3a747c5264c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Glen English]

Hi Jason

I think you are completely wrong about deltas. Have you got one ?

I would say that the firepick guys,  did not try hard enough, or did not understand all that was going wrong...

but rotation delta agreed much more difficult than my planned linear delta. That was probably the 'bite too many'.

The 3d printer guys who know what they are doing can maintain 0.1mm  over a large build area

simple as that..... yes you got to know what you are doing

MY own  3d delta printer (a chinese one with metal corners) makes about 0.15mm no problem with basic cal.....

and better than 0.1mm over say a 100 x 100 x 100 cube.....

(HT-PLA - not ABS)

I personally think old cartesian is a brain dead system, (but suited to heavy heads and LARGE build areas I agree for sure)

[Jason von Nieda]

Hi Glen,

I thought you might say that :) I hope to proven wrong!

I worked on the FirePick Delta project in Shenzhen, China for 3.5 months trying to make it accurate. It was a lost cause. I was there when the project died. I wrote the Beta Calibration guide and designed the calibration kit that was sent to all of the beta customers. FirePick Delta died for a lot of reasons, but one of them was that it assumed that it could get 16 microsteps of resolution out of a basic NEMA 17 stepper and a cheap driver, and that proved impossible. It would have needed to do about 4 times that to get anywhere near what we were shooting for, and it would still require a lot of calibration to fix all the non-linear error that comes with a rotational delta.

My experience with linear deltas comes from having spent a lot of very long days and nights working with a designer of linear delta robots while trying to figure out the problem with FirePick. He told me quite a bit about the hoops he had to jump through to get his linear deltas to do 0.1mm repeatably. I don't have a lot of hard data on this, so like I said, happy to be proven wrong.

I would challenge you to strap a camera, or some scales to your delta printer and see if it really does 0.15 throughout the built volume. It's easy enough to find that it works at certain points, but a PnP has to be able to hit every point. This is where the FirePick failed. We could make it accurate to a 10x10 grid with barycentric interpolation but the points between the grid points were a mess.

Anyway, I really do hope to be proven wrong. I think a delta PnP would be fun to watch. But I think there's probably good reasons that you never really see it being done.

Jason

--
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 post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/a0ea965f-ad52-4bfb-80a7-cd90566a3703%40googlegroups.com.

[Glen English]

ahh so you have alot of rubber on the road. you are the man to talk to.

I can learn from you.

yeah I agree  a rotational delta is difficult. so this will be a linear delta.

I think my additional techniques with the cameras and fids will reduce the difficulty  I expect.

my worst angle will be around 45deg WRT to Horz IE approx acos ( 320 / 450)  so very conservative which helps alot.

****

Yeah , by assuming 1/16 stepping , that would have been a bridge too far.

I assume 1/4 for reproducable microsteps for open loop stuff.....maybe . 1/8 for a good driver and a well damped system.

(3 phase motors are much better also)

I use the encoder steppers often, $85, but still only 1000 steps in the encoder per rev. which is about like 1/5 microsteps for  the average 200 step fella.

you can get better encoders of course but I think I can use stereo vision to do better along with base fids and some horizontal benchmarks, also

****

short towers makes a bit of a difference for rigidity... panel bracing  essential I think to avoid the corners needing to be super duper.. 

****

I also thought of doing the verticals with  ball screw linear guides but the $ value of the project goes up a bit... (like $350 per vertical!).

I'll only go to that if I cant get the belt/stepper performance

***********

I will read up on the firepick delta- where can I read up ?

cheers

[Glen English]

Post script

(since there is no EDIT POST on GG (that I can find)

I looked at firepick.org, and watched the videos

ahh one of those deltas. rotational , right. Well it has a different build volume shape of course.

I gather you went that way for a reason compared to all the 3d printer paths ?

As for the rotational delta. I gather you were driving to direct drive the wheels ?

hmmm taking a simple geometry condition for 0.1mm  resolution at a guess looking at the sizes you would have needed 3000 steps per rev..

I think that you would have been doomed with a 2 phase stepper with or without encoder...unless it was a 4096 step encoder and even then.. not great.

now, you could have done that with a 5 phase stepper... 0.36deg basic step  and 1/4 micro steps  and not too expensive either...

this is quite a good tutorial

http://www.orientalmotor.com/technology/articles/2phase-v-5phase.html

actually I have lots of papers on 5 phase steppers if u are interested.

my rotating build plate idea (probably bad idea) 

The only thing a rotating build plate would do is eliminate the rotation of the nozzle on the effector.

and  weight less, but a world of extra complexity, and a big slow down for rotating the build plate.

-like just one axis too far...  nema 8 weighs about the same as many of the Bowden extruders.

[Michael Anton]

The FPD did not direct drive the arms.  It was something like a 14:1 pulley ratio, as higher angular resolution was required than what direct drive could do.  The large pulleys attached to the short arms are driven by 16 teeth pulleys.  Neil, the FPD designer, opted for a rotational delta to allow for a square footprint, and to reduce the mechanics required.  It was designed to be very inexpensive, so there were many compromises made.  Too many in my opinion.

The only people that can edit posts in GG are the admins... which sort of sucks.

Jason should be able to fill in anything that I missed.

[Jason von Nieda]

Hi Glen,

It wasn't my design. I was basically just hired to make it work with OpenPnP but when it became clear that it was nowhere near accurate enough to do PnP it turned into my job to try to make it accurate. 

If you want to read more about it, the Google Group is the best source of info: https://groups.google.com/forum/#!forum/firepick

The original project is at https://hackaday.io/project/963-firepick-delta-the-open-source-microfactory

I'm no longer involved, and am not the creator of the project, I was just there for the worst part :)

Moving on... 

Having etched fids on the bed is a good idea, and we tried something similar with FirePick. We produced a number of huge PCBs covered in fids to use for alignment. This was definitely helpful and helped us find a number of sources of error. As you know, I'm sure, it's actually kind of tough to find an accurate grid to work against and this seemed like the best way to do it.

Since I don't know much about linear deltas I'll leave it at that. I'm not particularly interested in them. I think it's just a harder way to do something that should be done with a cartesian, but I understand the allure and wish you luck.

Jason

--

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 post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/d0aa753d-3464-4a69-a909-0a87d95033e5%40googlegroups.com.

[Cri S]

Jason, today i have tested the wasp 40/70 printer because i have retired some 3d stamping and have asked if i could test 5min the accuracy with Webcam. The error was in the order of 1.3pixel 0.037mm always +-, 80 microns error.

[Jason von Nieda]

Hi Cri,

That is really amazing! I'm not familiar with that printer, is this it? http://www.personalfab.it/en/products/deltawasp-40x60-2/

It looks like it runs about $6000 from a quick search. 

Do you know much about the technical details? Rails, drive system, closed loop, etc?

For what it's worth, I think deltas make a lot of sense for 3D printing. You need lots of Z. For PnP, not so much.

Jason

On Fri, Feb 10, 2017 at 9:42 AM Cri S <phon...@gmail.com> wrote:

Jason, today i have tested the wasp 40/70 printer because i have retired some 3d stamping and have asked if i could test 5min the accuracy with Webcam. The error was in the order of 1.3pixel 0.037mm always +-, 80 microns error.

--
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 post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/a908114b-fe9b-4b0c-8e44-0f6edba61670%40googlegroups.com.

[Cri S]

Check out this Video, https://m.youtube.com/watch?v=uG-yoiFV85c
The Motors are ON top hidden, open loop steppers and same pulley AS you See ON Video. Spool is on top on center of the motors. Everything igus. Lot of resolution, only moderate angled on joints.
I have seen delta for PNP for th connectors, because delta can vibrate a lot better as scara resulting 5 time faster. The drawback is that two delta and one rotation table is needed for loading connectors from tray and placing to PCB where scars robots can do both task because of larger workspace.
Beside odd size the connectors and condensators I don't have seen application for delta on pnp,.
There is one emerging , hand soldering, because it can be reached with inverse delta and this make it very flexible, that yes.

[Jason von Nieda]

Very nice! Do you know how they do bed leveling? It looks like they touch the nozzle to the glass bed, so it's not conductive. Stepper feedback? Or something in the head?

Jason

--
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 post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/066c5ecd-1d9b-4039-a75c-b1b6c6be74ad%40googlegroups.com.

[Glen English]

Thanks for the comments and links

Wow you really got subjected to the blowtorch at the pointy end.

I'm going with the linear delta mainly due to  the fact that the wealth of experience in the 3d printing community with linear deltas, and the broad availability with truly off the shelf bits. Nothing to make.... When I look at pictures and videos of industrial rotational deltas doing their job, I see mostly non alignment ciritical jobs like sorting apples on a fast conveyor...

cheers

On Saturday, February 11, 2017 at 2:32:01 AM UTC+11, Jason von Nieda wrote:

Hi Glen,

[Thomas Langås]

We have the 2040 turbo at work, and afaik it doesn't auto-level the bed. We did some calibration work in the beginning and it's been spitting out filament ever since :)

On Feb 10, 2017 20:07, "Jason von Nieda" <ja...@vonnieda.org> wrote:

Very nice! Do you know how they do bed leveling? It looks like they touch the nozzle to the glass bed, so it's not conductive. Stepper feedback? Or something in the head?

Jason

On Fri, Feb 10, 2017 at 12:30 PM Cri S <phon...@gmail.com> wrote:

Check out this Video, https://m.youtube.com/watch?v=uG-yoiFV85c
The Motors are ON top hidden, open loop steppers and same pulley AS you See ON Video. Spool is on top on center of the motors. Everything igus. Lot of resolution, only moderate angled on joints.
I have seen delta for PNP for th connectors, because delta can vibrate a lot better as scara resulting 5 time faster. The drawback is that two delta and one rotation table is needed for loading connectors from tray and placing to PCB where scars robots can do both task because of larger workspace.
Beside odd size the connectors and condensators I don't have seen application for delta on pnp,.
There is one emerging , hand soldering, because it can be reached with inverse delta and this make it very flexible, that yes.

--
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+unsubscribe@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/066c5ecd-1d9b-4039-a75c-b1b6c6be74ad%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

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+unsubscribe@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CA%2BQw0jzJJ-K_-B-aEQ0_jbbauTM-kr8rsXzVNDfU6L32VL%3D0qQ%40mail.gmail.com.

[Glen English]

I have found that here is much to be learned by  just listening at the group :

https://groups.google.com/forum/#!forum/deltabot

[Mark Harris]

I have to admit.. i'm not entirely sure what the point is here.. if it's just "because I can" - awesome.

Starting a design off with something that needs heaps of initial calibration, loads of processing to be run to make it remotely accurate (even if that code is already built into smoothie), and a system that really is worst case scenario for Z accuracy seems a little crazy to me? Basically, why start with something mechanically unsound for moving parts on X, Y, Z and making up for it in software, when you can move a part on an X axis, Y axis and Z axis directly?

When I worked at SolarBotics we had a bunch of Delta printers around from OpenBeam (Kossels) - they were very entrancing to watch run (amazing to have on the corner of a booth at comic-con!) but I was never overly impressed with the print quality - cartesian printers do it much better.

But like I say, if its just to try out/because you can.. awesome - looking forward to seeing it. Deltas are really fun to watch.

On 10 February 2017 at 17:13, Glen English <gleneng...@gmail.com> wrote:

I have found that here is much to be learned by  just listening at the group :

https://groups.google.com/forum/#!forum/deltabot

--
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+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/85e87503-946f-4b4a-b961-7de627578b1f%40googlegroups.com.

[Glen English]

The WASP  deltas are quite impressive. notice how conservative they stay on the angles...

What I need to get a practical understanding , that is real world material limits, of is how much you can hang off the effector before things go wry....

The mag bearings wont let go easily, that is for sure.

I'll use a juki nozzle, and am keeping the angles quite conservative so it can push down the spring in the nozzle at the extremes.

On Saturday, February 11, 2017 at 2:58:46 PM UTC+11,

[Genie Kobayashi]

Hi Glen,

FYI, in 2013 Midwest RepRap Festival, Jeff Mcalvay demonstrated P&P using modified Rostock MAX. It looked partially successful as far as seen in the short video shooting.

https://youtu.be/bLxJjdroT0U?t=12m30s

I'm not sure if it runs stable for completing job but worth a try. It'll be suitable for small PCB though. IMO, a set of JUKI nozzle + brass holder + NEMA 8 or 11 stepper + swivel horse joint won't exceed the weight limit of delta effector.

Genie

[Trampas Stern]

The NZS has a 16384 steps per rotation encoder. It is easy to get +/- 0.05 degrees, with a 0.9 degree stepper and running slower you could get better that 0.01degrees.

Trampas
Misfittech.net

[Glen English]

Well, with the one corner rods straight down, belt movement ~= Z movement 

so....16 teeth @ 2mm pitch = 32mm travel .per revolution.

or 32/360 = 88uM per degree.

OK so maybe go 20 tooth a bit easier on the belts., 40mm per revolution, 111uM per degree

1/4 microsteps on 1.8deg motor.. no problem.

the 3d printer guys seem to like about 4x better reliable resolution than the print layer resolution...

[Douglas Pearless]

Hi,

A quick update, several of us are revisiting the FirePick delta (Nov 2018), specifically to implement the calibration routine that was not done; this allows for real-time correction of the "bowl effect" that all rotary deltas have (by their very nature) and is (or I guess should be) present in all commercial rotary deltas.

My goal is to see if we can get 0.1mm repeatability using TMC2130 256 micro step drivers with 0.9 degree steppers and the aforementioned software routine.

Whether we succeed or fail is yet to be seen :-)

Cheers

Douglas

[Marek T.]

It's gotten by Marlin in 3D printers without problems using 1.8deg motors, isn't it?
So why not with 0.9?
But more important to get 0.01, 0.1 is far from good in pnp machine.

[Douglas Pearless]

I too have had 1.8 with Marlin; I have moved to Smoothie a while ago as trying 256 micro steps with 0.9 degree steppers at speed was not working for me with a Rotary Delta; plus I found Marlin too difficult to modify.

You are probably right I should be looking for 0.01 not 0.1mm repeatability 😄

[Marek T.]

0.01 you need for sure.
I have Marlin in printer and it doesn't look some specially complicated however it seems to me the Merlin is not supported by Openpnp any more as Smoothie is...
Unfortunately I know nothing about Deltas problems :(.

[Marek T.]

Does Smoothie has implemented TMC2130 or 2208 SPI bus support at all?
http://forum.smoothieware.org/forum/t-2140731/tmc-2130-spi
Maybe it's history already...

[Jason von Nieda]

I can't help but ask... why?

What is it about this design that makes people like it so much? It offers no benefits over a much simpler Cartesian platform for PnP and it has been proven repeatedly that it doesn't work.

Do yourself a favor and spend a few minutes working through the math. The numbers simply don't add up unless you use very high resolution servos, and even then you still have all the other sources of error - which you have to multiply by 3.

Microsteps do not increase resolution past about 4x. It doesn't matter what microstep driver you throw at it. This is something everyone wants to be false because it's so appealing to think you can actually get 16x (or more!) resolution but it just doesn't work. The FirePick was designed around this faulty assumption and it killed the project.

There's a reason that there are no commercial delta PnPs, and the reason is that it doesn't work.

Jason

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/a715d84a-2396-4202-b685-39e8306a94c0%40googlegroups.com.

[Douglas Pearless]

I have a Cohesion 3D Smoothie compatible board running TMC2130’s at the moment, though I do need generate a pull request to add the code back into the main GitHub repo.

You received this message because you are subscribed to a topic in the Google Groups "OpenPnP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/openpnp/5koTWtJalXQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to openpnp+u...@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CA%2BQw0jwDRfiROXSEOkzYqBBKsGNc9teg4gGnBkrJRUF6Z0un4g%40mail.gmail.com.

[Douglas Pearless]

Hi Jason,

Yes I watched some of the issues from the sidelines and the fallout 😞

One of the reasons i believe that it failed was only half of the Rotary Delta solution was implemented ( I was the person who ported the solution to Smoothie), it was missing the calibration routines that correct the bowl effect and that is what we are looking at.

I have managed to get hold of the paper from one of the original “inventors” of the mechanism which deals with the required methods and maths to correct for this and have had some maths gurus ( my maths are several  decades behind me 😄 ) look at it and we are working out how to turn that into C++ code.

My goal is to implement this into Smoothie and see if the issue can be solved, or not.

As to the why, we’ll I would like to see an open source micro factory, my FPD acts as a 3D printer but only for small things, but I suspect the real reason is I don’t want to give up on it until the software has been finished, so to speak.

Understand about the stepper resolution and may need to go to 0.36 degree steppers but they are 5 phase and are USD54 plus a driver that is about twice that!

Cheers

Douglas

On 11/11/2018, at 04:49, Jason von Nieda <ja...@vonnieda.org> wrote:

You received this message because you are subscribed to a topic in the Google Groups "OpenPnP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/openpnp/5koTWtJalXQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to openpnp+u...@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CA%2BQw0jwDRfiROXSEOkzYqBBKsGNc9teg4gGnBkrJRUF6Z0un4g%40mail.gmail.com.

[Jason von Nieda]

Hi Douglas,

The bowl effect is really the least of the problems. Resolution and error stackup are the real issues. I really do implore you to take a few minutes and work through the kinematics by hand with the actual resolution you can get from your motors. You will see that even 0.36 steppers will not be enough to get it anywhere near useful for PnP.

Jason

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/9D6481B5-01CD-4D01-A174-5B697BBBB94C%40gmail.com.

[Douglas Pearless]

Hi Jason,

OK, I have someone working through the maths model at the moment and is generating predictions of what it should be for the forward and inverse kinematics (and nice graphs, etc), what were the resolution issues and error stackuo from your knowledge and perspective?

Cheers

Douglas

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CA%2BQw0jy-34aoyRfB5VXwb6AGq1JwTwQ13%2BF2eqwtQn%2BA-ur7Og%40mail.gmail.com.

[Michael Anton]

I'd agree with Jason on this one, but I didn't want to be the first to chime in on it for a change...

When I initially saw the project, I was doubtful that all of the error sources in the mechanism could be corrected for using math alone, since that is very difficult to do when there are all sorts of mechanical problems.  It's not just the kinematics that are a problem.  There are all sorts of backlash and hysteresis issues with that design as well, in part because it really is not stiff enough.  It is going to be much harder to correct for these in a delta configuration that it would be in a Cartesian design, which is truth is very difficult as well.  I was able to make some changes on my Firepick Delta to make it stiffer, but it probably is still not enough, when you compare it to the stiffness of linear deltas.

Why don't you get a set of Smart Stepper drivers, or Mechaduinos that implement closed loop control on the motor?  At least then you can get some reasonable resolution without having to swap to some exotic motor.

I suspect that if this had been just a math problem, it would have been solved long ago.  Here is a commercial delta that has a working envelope similar to an FPD: https://www.fanuc.eu/at/en/robots/robot-filter-page/m1-series/m-1ia-1h, with a repeatability of +-0.02mm (which is actually far better than the last time I looked at these a few years ago).  This is done with servos, and an all metal assembly, designed by a professional company with decades of experience.  How do you think it would be possible to be even better than this, with a 3D printed assembly?

[Douglas Pearless]

Hi Anton

I am planning on replacing most of the plastic parts with aluminum ( good excuse to play with my Wazer ( https://www.wazer.com/ ) ) too; and may well dispose of the large pulleys too though undecided in that at the moment.

I have already developed some magnetic position encoders that attach to the stepper ( or pulley ) so I can have feedback on the actual position of the shafts ( I am planning to pull some code from another of my projects, a USB controlled servo and to add this to Smoothie as well).

I agree the level of plastic in the FPD is probably far too high.

If you want to get more details as this progresses, the hw_motionctrl slack channel for the Firepick is alive again and that is where most of the discussions will be posted.

It will be interesting to see if the design can be improved or not.

Cheers

Douglas

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/9dda2d14-ca90-460c-b3ba-ca71d6ea0418%40googlegroups.com.

[Brynn Rogers]

Funny you mention the Fanuc delta robot.  I took this video at a show last week:   https://youtu.be/RTbzi4so7M4

I've got a Rostock max V2 delta 3d printer which I need to upgrade to do the auto bed-leveling.   I think delta's are fun to watch, but probably are not the best choice for a PnP.

The biggest drawback I see is that they have a relatively smaller build volume than some XY  Gantry type, or whatever you call the machine I am building  (X on fixed gantry, Y bed(s) move)

Except for Z, a delta PnP would have way more Z travel than needed.  Obviously if they have huge problems getting to .01mm or even .1mm that just makes them worse.

Brynn

[Jason von Nieda]

Hi Douglas,

For resolution, if you read through Neil's original analysis he assumes that using 16x microstepping gets 16x resolution and then uses that to justify that the machine will be accurate enough. See: https://hackaday.io/project/963-firepick-delta-the-open-source-microfactory/log/3588-6152014-delta-mechanism-simulation-and-accuracy-determination

If, instead, you assume that 16x microstepping gets you 4x resolution, which is what it does get you in reality, then by extension the error in the calculations is 4x greater.

At the end of the project I created a spreadsheet that worked through the kinematics at various resolutions but I can't find it. It may have been in my FirePick Google account.

As to the error stackup, it's similar to any other precision machine. The precision you get out of it is a function of the precision you put into it. On a Cartesian it's easy to measure the errors and cancel them out with simple math. Unless the machine is just junk there are only a few possible error sources, and they are all linear, so simple to cancel.

In a rotary delta you have WAY more error sources (arm length, motor mounting angle, position of cups on arms, end affector angles, etc.), and then they are all 3x because of 3x arms AND none of it is linear and no one (that we found) could even come up with the math to begin to cancel them out.

So, since we couldn't math our way out of it we thought we'd computer vision our way out. We produced a giant PCB the size of the bed and used computer vision to map the whole thing out, then used Barycentric interpolation in the positioning code. Except the machine was so inaccurate and the error so non-linear that the interpolation would just end up taking to us to some other closer, but still wrong point.

Jason

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/5068F715-8DD4-4ECA-AE63-40B5F98E244C%40pearless.co.nz.

[Michael Anton]

Tin Whiskers build an all metal version in China to see if that would solve their problem.  It didn't.  Perhaps Felix would loan you one to play with for testing, as I think he still has at least one of those machines.  He sent me pictures at one point, and it looked like a really nice machine.

As I recall, the pulleys were pretty important to increase the resolution.  If you don't use them, you need even higher resolution out of the steppers, and you lose a bunch of torque.

I wonder if real shaft encoders would work any better than the magnetic ones.  The magnetic ones need calibration, as otherwise they are nowhere good enough.  Perhaps the calibration changes with temperature, which would be a pain.

[SMdude]

I have to ask the question, why?
It is hard enough to get a Cartesian system to behave, so why make it even harder on yourself by revisiting the delta system?
When I first started looking into diy pnp, the firepick did look very attractive, especially given all the other functions it was meant to be able to perform. However, I am very glad that I went with a Cartesian system as it was not that hard to build and now works quite well. If I had gone delta I would have a pile of junk just sitting there and my Mrs would never stop reminding me of the money I wasted on the piece of junk!
Please don't burden Jason with delta crap. He has more important issues to solve like nozzle calibration and auto board scaling ;)

[Douglas Pearless]

Hi Michael, 

Thanks for the information; the more the better.

Certainly food for thought; one idea I had was to create an absolute optical encoder around the outer part of the large pulley; I have also considered taking an optical encoder out of an inkjet printer as these seem rather good as well, but I do not know it’s resolution.

I don’t have Felix’s contact details; could you send these privately to me?

Cheers

Douglas

Sent from my iPhone

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/97f5265e-df2d-4135-99b7-ac1fa218932c%40googlegroups.com.

[Brynn Rogers]

Hi Doug,

The optical resolution I've found in the printer tapes, 180 lpi Encoder-Tape-for-Wide-format-printer , always seems to be 180 LPI.

Which might end up on my Z axis someday.   That I believe gets you .03527777 mm resolution.

If used on X and Y it might be okay, might not be enough,  I don't know.   I do know that the 'good' encoder tapes are running between $500 and $1000 for tape and read head, which seems just insane.   There are chinese linear encoder sets that are down to a couple hundred per axis.  I think the chinese are etched glass scales,  the german stuff is reflective etched on steel tape.

Brynn

[Trampas Stern]

The Smart Steppers are really close to providing 0.11 degrees accuracy, which is 16x with a 200 steps/rotation stepper.  The accuracy is dependent on the calibration and placement of magnet and can be checked by calibrating unit and then testing calibration. 

[Brynn Rogers]

Hi Trampas,

    Don't get me wrong, I don't want to be critical, but I think you mean resolution.   

Accuracy is more like repeatability.       Move to a random coordinate, measure how accurately it positioned.   do this many times, then you know what accuracy you can achieve.

Have you measured the actual repeatability or accuracy, or are you assuming the accuracy is the same as resolution?

Brynn

[TheCunningFellow]

The magnetic encoders are 14 bit so the input resolution is 0.02 degrees.

I'd say you should be able to get 11 bits stable.

that would be 0.18 degrees.  That would be repeatability.

Accuracy would require alignment of the magnet with the hall sensor and I imagine you could be whole degrees off the real world if that is bad.  But the fact that the stepper motor can give you a local reference means you should be able to accommodate for bad alignment some what.

[Trampas Stern]

The magnetic encoders are 14bits, I have done some testing and the noise is Gaussian on the encoders and therefore I have found that averaging does lead to increased resolution, and found the limit averaging gives is around 16bits.  Hence the encoder native resolution is ~0.02 and with averaging gets down to ~0.01. 

The accuracy from my testing is limited not by the encoder's resolution but rather by the calibration.  That is the magnet is never centered on the motor shaft or on the IC as such there error. To compensate for this the boards will move the motor one full step and measure the encoder, then repeat.  This gives the encoder reading per 1.8 degrees (assuming the stepper is perfect 1.8 degrees).  The firmware then does a 3 order interpolation between steps.  So you end up with several errors, the main one is assuming the stepper is 1.8 degrees.  So after calibration I will test the calibration and find that I can get within +/-0.05 degrees of the calibration.  As such I do not assume and accuracy any better that 0.11 degrees.  

I have been debating on how to measure the absolute accuracy, it is really hard to find any encoder that has a accuracy better than 0.1 degrees.  On the smart stepper forum we have debated on how to do it. The best we came up with is using a laser and large circle around the motor such that accuracy can be measured using distance.  However I have not set up this experiment yet to be sure. 

[bert shivaan]

a 10,000 CPR encoder gives 40,000 counts per rev.

thats .009 degrees per count

They are in stock at us digital for about $95 usd.

This sounds way easier then the laser test

--

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 post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/63033eeb-9572-42d5-aa34-58be42a7ebcf%40googlegroups.com.

[Douglas Pearless]

Do you have a link for these encoders? :-)

You received this message because you are subscribed to a topic in the Google Groups "OpenPnP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/openpnp/5koTWtJalXQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to openpnp+u...@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CA%2BKNHNzaA4q5t2oRkQa%3D5dnh5Menee_xTYQyOPNrsnNAOnk1Gg%40mail.gmail.com.

[Douglas Pearless]

I was looking at trying the optical strip that inkjet printers use to track the head movement and attach it to the circumference of a large wheel (which the particular rotary delta - FirePick Delta - I am playing with uses this via a belt to the 0.9 degree stepper to increase the resolution, off the top of my head I cannot recall if it is 1:10 or 1:14 or something else in terms of the reduction), these seem to be 180 lines-per-inch but I need to sit down and do the maths to see if it will work.  These are available for as little as $3 plus the electronics to read it. ent 

Cheers

Douglas

-- 
You received this message because you are subscribed to a topic in the Google Groups "OpenPnP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/openpnp/5koTWtJalXQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to openpnp+u...@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/63033eeb-9572-42d5-aa34-58be42a7ebcf%40googlegroups.com.

[bert shivaan]

https://www.usdigital.com/products/encoders/incremental/rotary/shaft/S6

You will have to set the configuration for it, drop downs on the right hand side.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/B85C4F77-16B6-4EE3-9D88-56EF7A116AF4%40gmail.com.

[Luiz doleron]

Hi Jason,

I'm currently working on a statistical model to describe the accurancy issues for discrete actuators used in rotary deltas. Based on my computer simulations, I do confirm what you have said here so far. Just to illustrate, the following plots describe the error distribution along the bed for 3 scenarios:

Anyway I believe that it is possible to determine analiticaly/statiscally when a specific RDR geometry/control fits or not fits a predefined set of tasks. It would be great to know the problematic scenarios you have faced when you worked in the FPD calibration guide/kit.

Regards,

Luiz

[Jason von Nieda]

Hi Luiz,

Very interesting work! Probably a bit beyond me, to be honest. :)

Before I comment further, I want to clarify that the error I am most concerned with is X/Y, not Z. Z is, of course, important too, but my goal was to get X/Y to be "close" when I was working on the project.

With that said, when I was working on this project my only goal was to be able to place parts reliably. We eventually settled on trying to create a grid points using computer vision and a large calibration PCB and then using barycentric interpolation to try to be more accurate. I believe Karl Lew is the person who suggested barycentric interpolation.

What I found was that if I ran through calibration and then applied my interpolation, and then used OpenPnP to try to go to positions defined on another PCB, I was often off by as much as 0.5mm. This was better than without interpolation. I forget how much further it was off, but I do remember the interpolation helped.

I did my tests by mounting a camera on the end effector, commanding OpenPnP to go to a particular position on the PCB and then using relative coordinates to measure how far from that position I ended up. I verified that things were "working" by telling it to go to the grid points on the calibration PCB and those would line up quite accurately.

My conclusion, then, was that we'd need a might tighter grid to get anywhere near the accuracy we wanted, and that was basically infeasible. Just running the 10x10mm grid took several hours if I remember correctly.

Please understand that it's been 2+ years since I worked on this, and the project ended in such a spectacularly terrible way that I've done my best to forget it.

Jason

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/9796c1cc-3732-4d53-91ce-1e82a63b83d3%40googlegroups.com.

[Luiz doleron]

Hi Jason,

firstly, thank you for so kindly reply.

Yes, like in Z, XY accurancy issues are mathematically relevant as shown in the following plot:

I've build a toy delta and I have found the same XY issues experimentaly. Anyway, the experimental routine that you just described looks promissing. My current setup is pretty old style (I have to check the tracks in the sheet with a packmeter). I'm building an FPD and I'll try to replicate your procedure with it soon.

I really sorry for make you remember something bad for the project.

Best regards,

Luiz

Em terça-feira, 13 de novembro de 2018 14:44:13 UTC-2, Jason von Nieda escreveu:

Hi Luiz,

Very interesting work! Probably a bit beyond me, to be honest. :)

Before I comment further, I want to clarify that the error I am most concerned with is X/Y, not Z. Z is, of course, important too, but my goal was to get X/Y to be "close" when I was working on the project.

With that said, when I was working on this project my only goal was to be able to place parts reliably. We eventually settled on trying to create a grid points using computer vision and a large calibration PCB and then using barycentric interpolation to try to be more accurate. I believe Karl Lew is the person who suggested barycentric interpolation.

What I found was that if I ran through calibration and then applied my interpolation, and then used OpenPnP to try to go to positions defined on another PCB, I was often off by as much as 0.5mm. This was better than without interpolation. I forget how much further it was off, but I do remember the interpolation helped.

I did my tests by mounting a camera on the end effector, commanding OpenPnP to go to a particular position on the PCB and then using relative coordinates to measure how far from that position I ended up. I verified that things were "working" by telling it to go to the grid points on the calibration PCB and those would line up quite accurately.

Good approach. I'll try to replicate it in my experiments. I made a toy delta and I've found experimentally