Improve the cycle time of AdvancedLoosePartFeeder

[Fern528]

Hello everyone,

This email is meant to be a brainstorming thread for possible improvements. I understand if this is the limit of openpnp. Still very impressive openpnp can do all of this now with the test branch I am using. Deferring errors saved my machine from being stored labeled "not autonomous" or  "not for production". Runs autonomously now but is still slower than loading by hand. 

Huge thanks to whoever contributed to the defer error feature. great work!  

Situation: 

Currently loading a 612 placement job with the advancedloosepartfeeder and takes about 40 minutes. By hand takes about 20 minutes on average.   

I am using the error defer and reduced all vision and pick retries to one. the actual cycle time for one part from tray to board is 4 seconds. Now I am chasing optimization but 4 seconds just might be the limit. Given the machine has to jog over, run feed pipeline, pick, jog to placement, and place. 

I was asked "Can the machine just take a picture of the loose part tray and just know where the parts are?" I can not imagine how openpnp could accomplish this. unless it could scan prior to running the job and store all the imagines it takes. run the feed pipeline, store 1, 2, 3, .. matches and keep the location data for the best matches. then make a pick list of all the stored location data. 

Another improvement would be if openpnp could find the next part while it is placing a part. However, this would require another gantry system for the top camera itself probably. 

Thanks,

Fernando 

[Toby Dickenson]

Hi Fernando,

> This email is meant to be a brainstorming thread for possible improvements

Please share:
* a description of your machine hardware.
* your machine.xml
* a video of your machine running several placement cycles. We might
be able to spot some areas for improvement

> This email is meant to be a brainstorming thread for possible improvements. I understand if this is the limit of openpnp. Still very impressive openpnp can do all of this now with the test branch I am using. Deferring errors saved my machine from being stored labeled "not autonomous" or "not for production". Runs autonomously now but is still slower than loading by hand.
> Huge thanks to whoever contributed to the defer error feature. great work!

I am pleased this is working for you. Thanks for the feedback.

> Situation:
> Currently loading a 612 placement job with the advancedloosepartfeeder and takes about 40 minutes.

That doesnt sound too bad. If I understand right, that is 900 parts
per hour. I dont think I have seen any reports of
advancedloosepartfeeder being used in production and running at that
rate, so I am definitely interested in seeing a video.

> By hand takes about 20 minutes on average.

What does "by hard" mean here?

> I was asked "Can the machine just take a picture of the loose part tray and just know where the parts are?" I can not imagine how openpnp could accomplish this. unless it could scan prior to running the job and store all the imagines it takes. run the feed pipeline, store 1, 2, 3, .. matches and keep the location data for the best matches. then make a pick list of all the stored location data.

... and hope they dont get moved. But they are "loose".

Toby

[Fernando Corona]

I will take a video next time we run it. 

>What does "by hard" mean here?

"by hand" means tweezers or vacuum pen picking loose into our "pcb" ; it's just a large tray with machined pockets for the parts. think giant waffle pack tray. 

> ... and hope they dont get moved. But they are "loose".

that makes sense the parts could move during picking from the tray. 

--
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 visit https://groups.google.com/d/msgid/openpnp/CAH35urdj_LOhcONRFaZqVBGQa--Kq1TZ8hMpV%3D2Lxbft%3DumATw%40mail.gmail.com.

--

- Fernando