Preview of the HeapFeeder
244 views
Skip to first unread message
Team14
Oct 19, 2017, 7:08:55 PM10/19/17
to OpenPnP
Unfortunately our machine is picking and placeing now reliably.
You can see a demo video here: https://www.youtube.com/watch?v=rYFK_RLNU-g
What it is all about:
Its aim is to provide as much as possible different type of parts and as many parts for each type
to the machine at the click of a button. It should reduce the time needed for human workers for
seting up a pick and place job.
On our table we could put over 600 part types with up to 4000 pices of each type. In total we could
have nearly 2.5 million parts on "stock" directly reachable by the machine.
The HeapFeeder consists of following components:
1) The DropBox: it is used for separation and flipping the parts
2) BoxTrays: They consist in our case of 16 SubBoxes with a size of 13x13x35mm inner volume.
They are surrounded by corridors for parts transportation to eliminate the risk of mixing
up differnt types of parts.
3) The ChipFlipper: it is used for sideview via a mirror and can flip parts automatically.
The procedure ist for now the most simplest approach:
1) If the feeder has changed sice last HeapFeeder's feed, check if we have to clean up the DropBox
2) Fetch new parts from the defined SubBox. This is a critical stage to get reliably parts from a
heap of parts.
3) Transport the partd through the predefined corridors
4) Drop the parts in the DropBox and categorize them for upside up, upside down and anything else.
5) a) Pick the upside up parts and place them.
b) If no upside up parts could be found, pick a flipped part and try to flip it by picking
it up and dropping it repeatedly.
c) anything else is left for clean up
If you want to have a look into the code, you can find it on GitHub, but be aware - it is a hack for now. There are still many todos
https://github.com/KarlZeilhofer/openpnp/tree/feature/heapfeeder
Greets, Karl
from Team14
You can see a demo video here: https://www.youtube.com/watch?v=rYFK_RLNU-g
What it is all about:
Its aim is to provide as much as possible different type of parts and as many parts for each type
to the machine at the click of a button. It should reduce the time needed for human workers for
seting up a pick and place job.
On our table we could put over 600 part types with up to 4000 pices of each type. In total we could
have nearly 2.5 million parts on "stock" directly reachable by the machine.
The HeapFeeder consists of following components:
1) The DropBox: it is used for separation and flipping the parts
2) BoxTrays: They consist in our case of 16 SubBoxes with a size of 13x13x35mm inner volume.
They are surrounded by corridors for parts transportation to eliminate the risk of mixing
up differnt types of parts.
3) The ChipFlipper: it is used for sideview via a mirror and can flip parts automatically.
The procedure ist for now the most simplest approach:
1) If the feeder has changed sice last HeapFeeder's feed, check if we have to clean up the DropBox
2) Fetch new parts from the defined SubBox. This is a critical stage to get reliably parts from a
heap of parts.
3) Transport the partd through the predefined corridors
4) Drop the parts in the DropBox and categorize them for upside up, upside down and anything else.
5) a) Pick the upside up parts and place them.
b) If no upside up parts could be found, pick a flipped part and try to flip it by picking
it up and dropping it repeatedly.
c) anything else is left for clean up
If you want to have a look into the code, you can find it on GitHub, but be aware - it is a hack for now. There are still many todos
https://github.com/KarlZeilhofer/openpnp/tree/feature/heapfeeder
Greets, Karl
from Team14
Bernd Walter
Oct 19, 2017, 8:00:48 PM10/19/17
to OpenPnP
On Friday, October 20, 2017 at 1:08:55 AM UTC+2, Team14 wrote:
Unfortunately our machine is picking and placeing now reliably.
You can see a demo video here: https://www.youtube.com/watch?v=rYFK_RLNU-g
I really don't like the motor sound.
Such resonance frequencies are super bad for motor torque and could lead to lost steps.
Not saying that it happens, but it is a bad thing and something I wouldn't risk.
I've noticed this on a lot of liteplacer youtube videos, so you are not alone.
Well - since you've ordered a smothieboard, you shouldn't be worried too much about it, since this will very likely go away with better drivers and microstepping, which the smoothieboard supports.
I think the problem is because they use the TI DRV stepper drivers in a configuration, which is bad for microstepping.
Pretty awsome system.
Not very fast, but still awsome.
I've build my machine big to have as many parts on the machine as possible, but that fact caused me a lot of troubles.
Personally I'd be a bit worried if a part drops in a nearby container when refilling them.
Nevertheless I might give it a try one day.
Jason von Nieda
Oct 19, 2017, 8:53:37 PM10/19/17
to ope...@googlegroups.com
Amazing work Karl! This type of system is really going to be a game changer for prototyping, I think. This is a great progression of the concept of a loose part feeder and I can't wait to see how far you take it. Well 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/c19542ae-cb63-486f-8d2b-0c0bea5d2ba5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Team14
Oct 20, 2017, 3:31:09 AM10/20/17
to OpenPnP
@Bernd: yes, it is not the fastest, but that was not the goal at all.
Regardin mixing up parts on refilling: we have thought about that too, and came up with this:
@Jason: Thanks. We also think this is a game changer in prototyping. Very important for that is though, that the complete database in the background works properly.
So when you put a part into your schematic, the PnP machine allready has to know, where it gets that part from. We have solved this with a simple CSV database and some python scripts.
Regardin mixing up parts on refilling: we have thought about that too, and came up with this:
@Jason: Thanks. We also think this is a game changer in prototyping. Very important for that is though, that the complete database in the background works properly.
So when you put a part into your schematic, the PnP machine allready has to know, where it gets that part from. We have solved this with a simple CSV database and some python scripts.
Cri S
Oct 20, 2017, 4:47:27 AM10/20/17
to ope...@googlegroups.com
I don't remember the dimension, but from memory you have 15x15mm and 7mm space,
for 10 containers with 16 bins is 30*10+9*7=363mmx127mm counting 160 bins..
Instead if using 112 bins, on the same overall dimension , every bin
could have it's own
drop area and fuirther it allows to exclude resistors that don't sit
on nozzle correctly.
The designs suppose two rows and drop size of 2 times 7x7mm with
different heights and
flip nose. Supposing you could have 3 such areas of 363x127, this are
480 vs. 336 bins.
that could be loaded at same time inside machine.
Further having this 7mm corridor, if it's larger, example 10mm, it could be used
as drop box too, with conical/round constriction to 7/6mm on all sides.
This allows 144 bins on the same size.
2017-10-20 9:31 GMT+02:00, Team14 <off...@team14.at>:
> @Bernd: yes, it is not the fastest, but that was not the goal at all.
> Regardin mixing up parts on refilling: we have thought about that too, and
> came up with this:
>
>
for 10 containers with 16 bins is 30*10+9*7=363mmx127mm counting 160 bins..
Instead if using 112 bins, on the same overall dimension , every bin
could have it's own
drop area and fuirther it allows to exclude resistors that don't sit
on nozzle correctly.
The designs suppose two rows and drop size of 2 times 7x7mm with
different heights and
flip nose. Supposing you could have 3 such areas of 363x127, this are
480 vs. 336 bins.
that could be loaded at same time inside machine.
Further having this 7mm corridor, if it's larger, example 10mm, it could be used
as drop box too, with conical/round constriction to 7/6mm on all sides.
This allows 144 bins on the same size.
2017-10-20 9:31 GMT+02:00, Team14 <off...@team14.at>:
> @Bernd: yes, it is not the fastest, but that was not the goal at all.
> Regardin mixing up parts on refilling: we have thought about that too, and
> came up with this:
>
>
> @Jason: Thanks. We also think this is a game changer in prototyping. Very
> important for that is though, that the complete database in the background
> works properly.
> So when you put a part into your schematic, the PnP machine allready has to
>
> know, where it gets that part from. We have solved this with a simple CSV
> database and some python scripts.
>
> important for that is though, that the complete database in the background
> works properly.
> So when you put a part into your schematic, the PnP machine allready has to
>
> know, where it gets that part from. We have solved this with a simple CSV
> database and some python scripts.
>
> --
> 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/141aa5f2-83e2-4336-8233-60cccdc145d8%40googlegroups.com.
> 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
FredG
Oct 20, 2017, 3:19:15 PM10/20/17
to OpenPnP
Hello Karl, nice work!
What do you think about to go to the bottom camera first? Maybe you are lucky, have picked up on the right side and can save time?
Btw. we are also working with liteplacer + OpenPnP in Vienna on our opensource cinema camera project apertus.org
Manfred
Karl Zeilhofer
Oct 20, 2017, 4:19:23 PM10/20/17
to OpenPnP
@Fred: as said before, this version is the most simple implementation of a complete feed cycle.
I think, the feeder could decide this by itself based on statistics.
Also could the feeder save all the positions of the upside up parts and the flipped parts in the dropbox, and picking the next ones directly without a prior vision cycle.
@Cri: your ideas sound very interesting. Thanks for your input. In first step we will improve this first version until a properly usable state, ready for release.
The basic idea is also, that the boxes will never be exchanged - since that's the key feature. There is no need to exchange anything. Everything is predefined in software.
What I wanted also to mention is, that this feeder is more or less a complete software solution. What's needed in hardware is:
1) a pressure sensor for picking up from the heaps
2) a DropBox with a proper background for reliable vision with any parts
3) some boxes to store the parts in.
Greets, Karl
I think, the feeder could decide this by itself based on statistics.
Also could the feeder save all the positions of the upside up parts and the flipped parts in the dropbox, and picking the next ones directly without a prior vision cycle.
@Cri: your ideas sound very interesting. Thanks for your input. In first step we will improve this first version until a properly usable state, ready for release.
The basic idea is also, that the boxes will never be exchanged - since that's the key feature. There is no need to exchange anything. Everything is predefined in software.
What I wanted also to mention is, that this feeder is more or less a complete software solution. What's needed in hardware is:
1) a pressure sensor for picking up from the heaps
2) a DropBox with a proper background for reliable vision with any parts
3) some boxes to store the parts in.
Greets, Karl
Bernd Walter
Oct 20, 2017, 8:43:39 PM10/20/17
to OpenPnP
On Friday, October 20, 2017 at 9:31:09 AM UTC+2, Team14 wrote:
@Bernd: yes, it is not the fastest, but that was not the goal at all.
Clear - for small volume, like prototypes this is the perfect solution.
No need to configure feeders for every single board, since all parts can stay on the machine.
I've just noticed how time consuming feeder configuration is.
I really love your idea.
Unfortunately I intend to run small volume production cycles.
Just did a small test job today - 100 LEDs took 12min.
Regardin mixing up parts on refilling: we have thought about that too, and came up with this:
You've thought about everything.
Also good to avoid dust to get into the bins when the machine is not in use.
Karl Zeilhofer
Oct 21, 2017, 5:33:32 AM10/21/17
to OpenPnP
The feed cycle is by now far away from optimized and we achieve already 150 cph.
So your 120 LEDs would have taken about 48 minutes.
So your 120 LEDs would have taken about 48 minutes.
Reply all
Reply to author
Forward
0 new messages