New Machine Based on Acro Base

[Greg Wroblewski]

Hello,

I gave it a shot and I built a machine using 2' x 4' OpenBuilds Acro base (https://openbuilds.com/builds/openbuilds-acro-system.5416/), Smoothieboard, and lead screw Z axis. It works nice overall, my first small PCB with Atmel SAML turned out perfect. The Y (shorter) axis is very accurate, it has 2 motors. The X axis though developed a backlash of about 0.2mm. I managed to compensate for it in software down to ~0.1mm, but it is a bummer. It must be coming from the stretching of the belt under the load of added weight of Z axis motor, lead screw, hollow shaft motor, camera, etc. I am 73.1% sure, since it took some time to rule out probably all other possible causes.

Would it be a good idea to get rid of the belt, and put a 900mm lead screw on the X axis? It would add roughly 1lbs of weight, so it is hard to predict how the Y axis would respond.

I will post my CAD files and other details soon.

Cheers,

Greg

[Tasos T]

Hi,

I have the same acro base system but i have put the project on hold because of lack of free time.

I cannot predict the behaviour of X axis with the added weigth but i find it rather strange to have so much backlash with the belt&pinion setup which has half the belt length of a "classic" setup.

Did you try to reduce X-acceleration on smoothieware?

Did you try to tighten the X-gantry wheels on the V-slot rail (check if they get losen-up durring usage)?

Are you sure that this is backlash and not "structural flexing" on X-axis? (you may be able to correct this by adding better support from the machine structure to the bed [to the 4 corners of the acro system])

I would very much like to see your CAD files and a photo of your completed machine.

Tasos

[Marius Liebenberg]

I just looked at the build video and noticed that they use plastic of a sorts for some structural members. That is not a good idea and I think the will find that you dont have sufficient stiffness in the frame of the machine to get good repeatable motion.
You can improve the stiffness by adding some cast aluminum corner brackets or by replacing the plastic parts with aluminum parts.

[Cri S]

The eleks, that is similar works. It require upgrading the beld
holder with 3D printed ones,
Then it can be used for direct laser exposing pcb material and for
pick and place.
Without laser and automatic nozzle changer, the cost is 300$ for a
complete machine +
approx 30$ shipping cost.

> --
> 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/01faf155-1f21-4d41-b2dc-091c78aa9a72%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>

[Greg Wroblewski]

The biggest Eleks I found was roughly 2x2ft, but my X axis is two times longer (~100cm, machine footprint is ~120cm). It works perfect overall, and if you account for the deviation the accuracy is <=0.1mm. It is not a classic backlash though. Let's say you move 100mm to the right, reaching point A. Then you move 0.1mm left - the actual move of the gantry is 0.09mm. If you immediately move 20mm to the left, the new position will be A - 19.8mm. If you move in 0.1mm steps, after 5mm you will see the 0.2mm difference. This happens each time you change direction, with great precision. It's more like a spring effect of some sort. The belt is 1m long, so my hypothesis would be that the belt stretches a little bit, or maybe there is a compression between the motor and the belt happening in the direction of movement (caused by the extra weight?). Changing acceleration, speed, tension of the belt, disconnecting all the cables for pnp, made no difference (like, behavior is 100% the same, even with super slow speed).

Currently I relearn all the vision lessons that you can find in the older threads, mainly because I tried to use common USB microscope cameras (bad idea). I have probably experienced most of the challenges mentioned before, including the weirdness of SOT-23 packages.

Greg

[Daniel Dumitru]

Mach3 has compensation for backlash. Maybe Smoothie has also something similar...

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/73fe1f34-ff84-486f-9b1a-b2c7e05232b2%40googlegroups.com.

[Cri S]

The problem with the eleks before the upgrades are non horizontal belt and stickiness on movements. 

For cheap  camera, you need blur, optional threshold, convert color to gray, blur, normalize , blur . I don't think its in videos.

Sot23 could require different pipeline because of that amount of blur. I need to see source image.

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/73fe1f34-ff84-486f-9b1a-b2c7e05232b2%40googlegroups.com.

[Greg Wroblewski]

That's the problem I hit. I got everything working for different parts, like Sot 23, TQFP 64, and small ones, but then I started placing some darker 0603 capacitors, and they failed roughly 1 in 10 cases. When I adjusted pipeline for them, Sot failed because of random reflections (using black Samsung nozzles does not help). As I understand it is a common practice to use separate pipelines for different parts.

Greg

[Cri S]

Sot23 require mostly separate pipeline, depending on camera the small
pipeline can work.
There is a small window where the angle is returned wrong and this
need to be avoided.
The small parts pipeline having nozzle mask and mostly doing body
outline recognition.
And for bigger parts where the pins are detected.
It is possible that for higher precision bga have its own pipeline.
This only for 0.8mm pitch
or finer, 1mm pitch don't need it.
Finally one pipeline for components that are bigger as the camera view.
This are the 2-4 pipeline mostly used. Depending on lightning
condition and camera
setup, additional pipeline could be necessary.
Example are aluminium smd caps or coils when the light intensity is too high,
otherwise it could be detected with the pipeline for bigger parts that
detect the pins of
the component.
Basically you probably ends up with 3-7 pipelines for bottom vision
that you copy and past
from a text editor and if that pipeline need to be optimized, you edit
the machine.xml
and replace the same pipeline for all parts at once or use scripting
for select the standard
pipeline and replace it.
The default pipeline is a generic pipeline that works, but rarely it
provides good results
incorporating several tecnics to improve the results.
Pipeline is different if camera is setup with shutter control or with
automatic control.
It is normal that for passive the ratio and size is checked in order
to not place mispicket
passives. It uses a script , without this checks 0402 and smaller
passives can be problematic or need to be extra checked. Hovewer it is
not normal to have different scripts
for passives.
Hope this helps

2018-05-09 19:50 GMT+02:00, Greg Wroblewski <greg...@gmail.com>:
> That's the problem I hit. I got everything working for different parts,
> like Sot 23, TQFP 64, and small ones, but then I started placing some
> darker 0603 capacitors, and they failed roughly 1 in 10 cases. When I
> adjusted pipeline for them, Sot failed because of random reflections (using
>
> black Samsung nozzles does not help). As I understand it is a common
> practice to use separate pipelines for different parts.
>
> Greg
>
> On Saturday, May 5, 2018 at 4:06:48 AM UTC-7, Cri S wrote:
>>
>> The problem with the eleks before the upgrades are non horizontal belt and
>>
>> stickiness on movements.
>>
>> For cheap camera, you need blur, optional threshold, convert color to
>> gray, blur, normalize , blur . I don't think its in videos.
>>
>> Sot23 could require different pipeline because of that amount of blur. I
>> need to see source image.
>>

>> Il sabato 5 maggio 2018, Greg Wroblewski <greg...@gmail.com <javascript:>>

>>> email to openpnp+u...@googlegroups.com <javascript:>.

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

>>> <javascript:>.

>>> To view this discussion on the web visit

>>> https://groups.google.com/d/msgid/openpnp/73fe1f34-ff84-486f-9b1a-b2c7e05232b2%40googlegroups.com
>>>
>>> <https://groups.google.com/d/msgid/openpnp/73fe1f34-ff84-486f-9b1a-b2c7e05232b2%40googlegroups.com?utm_medium=email&utm_source=footer>

>>> .
>>> 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+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/10e03009-1459-429b-a7a6-a7a07f393df6%40googlegroups.com.

[Greg Wroblewski]

Got it, thanks. I use my own software, so I ended up adding five parameters to each tape: light intensity, two threshold values, and two Gaussian kernel values. In practice I use only 4 variants though.

Here is a video of the machine in first continuous action: https://www.youtube.com/watch?v=Xj_rwAG7iLs

Rather slow, but there are plenty of opportunities for optimization. I think that the max speed with vision will be ~600 CPH.

Greg

[ma...@makr.zone]

Hi

Nice! A good film too, very informative and complete.

Nice Z probing before nozzle tip change. This is far up on my list of things to add, out of very bad experience... if your Z axis is off from optimizing speed a bit too much, it gets real ugly real fast. At least with the stainless steel holder of the Liteplacer. Bent my nozzle tip holder badly. In my case I'd need to probe in the unload too (as part of the down move), because that's where the ugly crunching happens, when the tip collar does not go into the tight slot.

Mind If I ask why you rolled your own software?
It seems that your hardware is all fast, but the software takes some time to command the next step. Or are these deliberate pauses for testing?

_Mark

[Greg Wroblewski]

I hit exactly the same problem. My nozzle holder is not straight, but the bottom vision is compensating for it perfectly. In my case moving the nozzle 2mm over the landing part of the nozzles box, pressing against it and then sliding it in, solved the problem (at least I did not see it again so far).

Today I tried pushing the limits of the machine, and I managed to place successfully Max10 FPGA in WLCSP 6x6 0.4mm pitch package (https://www.youtube.com/watch?v=bFpWPysxBMY). I don't think I will be doing it very often though. :)

Going with my own software was not my best idea in the hindsight. Surely I learned a lot. The code is under: https://github.com/gwroblew/CoolPnP

I also uploaded my OpenSCAD and STL files to the GitHub repo.

The build steps should be on OpenBuilds (https://openbuilds.com/projects/acro-pick-and-place-machine.154/), once the project gets out of moderation.

I do have a lot of delay and wait steps in my code. It is not always clear when you can speed up, or when you must slow down. It just requires more testing and some experience.

Greg

[Greg Wroblewski]

Correct link to the build description: https://openbuilds.com/builds/acro-pick-and-place-machine.7127/

I added BOM, steps, pictures, etc.

Greg