New Pick and Place design ideas

[bootstrap]

Point of attention, how do you visually recognize where you are exactly at the pcb without using fiducials or other markers with each and every component, which is a no go?

#1:  I assume the PCB will have fiducials, and I assume every approach will take advantage of them.

#2:  If we need to, the first step would be to place an empty PCB without solderpaste or components into the machine and run a process that moves the down-looking "nozzle camera" over every component, then snaps a photo of that portion of the PCB (which might be the whole PCB for tiny PCBs).  This gives us images of every pad without solderpaste.  Alternatively and more conveniently the software can extract this information from the gerber file.
 
#3:  I have to assume even crappy mechanics will position the down-looking "nozzle camera" within... say... 0.01" == 0.25mm of any desired x,y position over the PCB (though not even that precision is necessary).

#4:  Unless the camera has an extraordinarily tiny field (which it won't and can't since some components are up to 2" == 50mm square), the image of each component will include quite a number of through-hole pads near and/or within the component area (and the component hasn't been placed, so any through-hole within the component are still visible).  Every one of those through-hole pads has a known x,y distance from the component center, and every component pad (from the images in step #2 and/or gerber file data).

Regarding camera field.  If the camera has a 4096 pixel sensor, and images a 4" square on the PCB, each pixel in the image corresponds to 0.001" == 0.025mm on the PCB surface.  With just a little image processing an effective resolution of twice that is not difficult (0.0005" == 0.012mm).  That should be sufficient, no?

#5:  I know my messages were excessively long, so I understand why you may have missed a phrase or two.  But I said we can't look at pads on the PCB that have solder-paste on them, for the reasons you stated.

#6:  If a PCB is extremely unusual, and some component has less than 3 visible through-hole pads within the 4" by 4" square centered on a component, then we'll have to insist that the PCB contain 1 or more "local fiducials" or [pointless] through-hole pads.  I think that will be extremely rare, if not "never".

Do you still see a problem?

A clean pcb looks totally different from a pcb that was paste stenciled.  The paste could be off a bit also.  And then each paste stenciled PCB will start to look different each and every 5 minutes you look at it, because the paste is oozing after its placed and also when it changes temperature.  Of course when actually placing the components on the PCB, the visual aspect of that PCB will obviously also change.  All those things can be pretty tough on the software to visually determine its exact position.

This is partially answered in #5 above.  To complete the answer, the position of the component pads is determined relative to through-hole pads, which are never covered by solder-paste.  At least I've never seen that happen.  Therefore, the fact that every time a component is placed the PCB looks different doesn't matter.  The software is only finding and looking at the through-hole pads (and treating them like "local fiducials" in a manner of speaking) to determine the position of the component center via the down-looking "nozzle camera".  Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).

Do you agree?

I've always been a very private hermit type, and don't like attention.

You could have fooled me.  I think the advice from mrpackethead was not meant negatively but a relevant suggestion to successfully socially interact on this forum.  The way you are communicating at this point seems to me more fit for a blog (one way direction).

You mean I should give so little information in my posts that nobody has enough information to understand or comment upon the issues and proposed approach (or make at least a vague judgement about whether I have any brains or experience or capabilities)?

Hey, I am quite certain my forum messages are vastly longer and vastly more detailed than most messages.  I don't have a blog, and don't publish anything.  Maybe my messages are "different" because I don't take on trivial (thus simple and short) topics or projects, or maybe because I'm someone who spends [tens/hundreds of] thousands on my projects, as well as months or years.  That kind of seriousness requires more time, care, effort and characters... at least for my lame little brain.

I can assure you I have ONLY been trying to be informative, honorable and helpful... and doing a project that might help a lot of people in the future if I decide to take it on.  From my perspective, it is just a few others who are being rude, snide, non-serious and obstructive.  However, with my tendency to be a non-social hermit, and as someone who is vastly more serious about my work than most people, I accept that people somehow manage to INFER a tone or attitude they don't like.  Nothing I can do about that.  My brain is focused on the technical issues when I write.  I'm not here for social purposes, I'm here for technical purposes, with the intent to take specific actions (buy a neoden4, a LE40V, or build a pick-and-place machine).  My guess is, anyone seriously interested in what I'm discussing for practical purpose and not just to talk or socialize would also be sufficiently focused on the content of the messages than whatever accidental aspect of my communication, language or approach annoys a few people.

[bootstrap]

Bootstrap if you don't change your attitude I doubt you will get any ideas across.

We're frankly not interested in your way of communicating.  Maybe we'll play some jokes on you, pretend to ask questions, annoy you a bit...

And you complain about me?  !!!!!  WOW  !!!!!

And I'm frankly not interested in anyone who has no interest in posting messages with relevant technical content.  Maybe you should find some non-technical forums to attempt to socialize in, or just stick to communicating with folks who write in ways (and about topics) that appeal to you.

BTW, just about the most trivial and childish behavior I can imagine is playing jokes on (and trying to waste the time of) people trying to have a serious technical conversation.  I'm perfectly confident that my attitude and communication is vastly more appropriate than my detractors, no matter what oh-so-special social skills I may happen to lack IYNVHO.

[Kjelt]

#1:  I assume the PCB will have fiducials, and I assume every approach will take advantage of them.

Yes ofcourse there are a few fiducials but are two enough to keep the exact location, say we have a 100x160mm pcb having two fiducials will not guarantee your exact location without other than visual guidance. So I guess , see below, we need more unique marking points on the pcb to keep the exact placement.

#2:  If we need to, the first step would be to place an empty PCB without solderpaste or components into the machine and run a process that moves the down-looking "nozzle camera" over every component, then snaps a photo of that portion of the PCB (which might be the whole PCB for tiny PCBs). 

That would take quite some time, that is why it is not done on pro placers I guess, those are all about time, but it could be a viable alternative for your proposed solution.

#4:  Unless the camera has an extraordinarily tiny field (which it won't and can't since some components are up to 2" == 50mm square), the image of each component will include quite a number through-hole pads near and/or within the component area (and the component hasn't been placed, so any through-hole within the component are still visible).  Every one of those through-hole pads has a known x,y distance from the component center, and every component pad (from the images in step #2 and/or gerber file data).

My (unscientific) gut feeling tells me you might need two camera's. One for the tiniest components having a tiny field of view but extremely precise , that could than also be used for the larger components to make the final placements adjustments, and one camera with a wider field of view for the bigger components to allow exact position of the entire component relative to the surrounding.

Regarding camera field.  If the camera has a 4096 pixel sensor, and images a 4" square on the PCB, each pixel in the image corresponds to 0.001" == 0.025mm on the PCB surface.  With just a little image processing an effective resolution of twice that is not difficult (0.0005" == 0.012mm).  That should be sufficient, no?

I think that would do be enough yes.

#6:  If a PCB is extremely unusual, and some component has less than 3 visible through-hole pads within the 4" by 4" square centered on a component, then we'll have to insist that the PCB contain 1 or more "local fiducials" or [pointless] through-hole pads.  I think that will be extremely rare, if not "never".

Instead of the through hole pads I would suggest using the via's, those are more common on smd pcb's.  But my guess is you meant the via's in the first place ?

Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).
Do you agree?

I do not know for sure if the via's are in the pnp file but they should be in order to get the needed accuracy for visual placement. A lot of smd pcb's nowadays do not have any TH components or only in small areas like the power supply.

Do you still see a problem?[/font]

If the via's are in the pnp file or can become part of the pnp file this could work. It will be slower and less for production suited but for small batches for hobbieists it would be an interesting alternative IMO.
Only the amount of time for the software development will be huge, I have no clue how long and how many programmers worked on openPNP but it must be a larger team effort to be succesfull.

I can assure you I have ONLY been trying to be informative, honorable and helpful... and doing a project that might help a lot of people in the future if I decide to take it on.  From my perspective, it is just a few others who are being rude, snide, non-serious and obstructive.  However, with my tendency to be a non-social hermit, and as someone who is vastly more serious about my work than most people, I accept that people somehow manage to INFER a tone or attitude they don't like.  Nothing I can do about that.  My brain is focused on the technical issues when I write.  I'm not here for social purposes, I'm here for technical purposes, with the intent to take specific actions (buy a neoden4, a LE40V, or build a pick-and-place machine).  My guess is, anyone seriously interested in what I'm discussing for practical purpose and not just to talk or socialize would also be sufficiently focused on the content of the messages than whatever accidental aspect of my communication, language or approach that annoys a few people.[/font]

I had a friend who taught me electronics when I was a kid who had a similar attitude, most only tech talk not much into social talk.
I can deal with it but also know the amount of problems, frustration and pain he had interacting with the outside world.
In his case it was not an option, it was how he was, if that is the same for you and you act like you always act I think we can only respect you for being open about it, and "are who you are".

[bootstrap]

Reply to Kjelt.

Yeah, I started designing electronics and laying out PCBs decades ago when I was rather young.  Back then I never read or heard the world "via"... they were always called "through-holes".  So I got in the habit of saying "through-hole" and that stuck even after I started encountering the synonym "via".  I'll try to remember to say and write "via" instead of "through-hole" in the future, which will also shorten my messages a bit (hurray).

To perform a one-time scan of an unpopulated PCB will take much less time than placing components on that PCB.  So that overhead doesn't bother me.  Nonetheless, I believe the gerber files contain all that information.

Yes, the two or three standard fiducials on a medium or large PCB are very insufficient for precise placement of components.  That's why the via pads are necessary to act as "local fiducials".

The gerber files my PCB layout program (diptrace) generates contain locations of ALL holes (vias and holes for components with pins or leads).  In fact, the PCB manufacturers couldn't drill the PCBs without those locations, unless I'm missing something.  Plus they need the sizes of vias specified, otherwise they'd have to guess, which I don't think anyone wants!

I hope we don't need two down-looking "nozzle cameras".  But per my example, we won't be able to suffice with a low-resolution webcam type camera.  These days image sensors up to about 20 megapixels (5Kx4K) are quite inexpensive.

Yes, part of my design premise is, this machine can be up to 2x to 5x slower than commercial pick-and-place machines if necessary.  And I expect it will be about 2x to 3x slower.  I intend this to be open-hardware and probably openpnp (probably with some additions we'll have to write).  Nonetheless, what I'd call "my target market" in that context are individual engineers or engineers at nano-companies who only make prototypes or demos.  So if the placement process takes 2x or 5x as long but works reliably with 0201s and 0.30mm pitch, I'm happy.

If we can't just add to the openpnp code, then yeah, I suspect software development time will be substantial.  But since openpnp is open-source (I think), adding to openpnp should cut the workload substantially (I hope).  I'm a good, fast, quality programmer (assuming C or assembly-language (or C++ that isn't excessively strange)).  But if I'm lucky someone else will write the software!   

Yes, I am who I am, and I don't really give a damn what people think of me.  And I am indeed an extreme outlier... in many respects, actually.  At the same time, I'm the most benevolent human I know.  So go figure how those aspects of my personality fit together and coexist.  I rarely have problems with the "outside world", maybe because I rarely interact with the outside world, and I'm "live and let live".  Maybe your friend got frustrated because he cared what others think of him.  Unfortunate, that.

[Kjelt]

Yes, the two or three standard fiducials on a medium or large PCB are very insufficient for precise placement of components.  That's why the via pads are necessary to act as "local fiducials".
The gerber files my PCB layout program (diptrace) generates have locations of ALL holes (vias and holes for components with pins or leads).  In fact, the PCB manufacturers couldn't drill the PCBs without those locations, unless I'm missing something.  Plus they need the sizes of vias specified, otherwise they'd have to guess, which I don't think anyone wants!

Ah yes indeed you could use the drill files as input for the visual placement location, clever.
With some easy post processing you can determine also if the created pcb has enough unique of these location identifiers (I perse do not use the term fiducial since that is reserved for the two unique standard location identifiers) and if not create some extra. Although the pcb house might complain if there are small holes in the pcb without traces 

[bootstrap]

Unfortunately, unless you have a via hole through a pad, it won't end up in the gerber files!  Oops!

How do you like the term "viaducial"?      Okay, not that funny!

I like the term "local fiducial" for actual (but possibly/hopefully smaller) fiducial marks that aren't the primary fiducials.

[mikeselectricstuff]

Therefore, the fact that every time a component is placed the PCB looks different doesn't matter.  The software is only finding and looking at the through-hole pads (and treating them like "local fiducials" in a manner of speaking) to determine the position of the component center via the down-looking "nozzle camera".  Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).

Vias are normally covered in resist, so you  won't be able to reliably vision them
Hole positions can have small offsets from the copper , which is why they're not used for positioning   
Any system that relies on additional design features like vias as fids will have limitations - if you're using  0201's, you'll be short on space and won't want to be adding extra marks if there aren't already enough vias.

[bootstrap]

Therefore, the fact that every time a component is placed the PCB looks different doesn't matter.  The software is only finding and looking at the through-hole pads (and treating them like "local fiducials" in a manner of speaking) to determine the position of the component center via the down-looking "nozzle camera".  Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).

Vias are normally covered in resist, so you  won't be able to reliably vision them.

Hole positions can have small offsets from the copper, which is why they're not used for positioning   
Any system that relies on additional design features like vias as fids will have limitations - if you're using  0201's, you'll be short on space and won't want to be adding extra marks if there aren't already enough vias.

I've never seen vias covered by soldermask.  My PCBs aren't that way.  Take a look at the images of two of my PCBs several messages back for examples (near bottom of page 1).

It may be possible to see via pads covered by resist.

If you look at the two PCBs a few messages back, you'll see that the traces are easily visible on the large PCB with medium-to-dark blue soldermask (which I assume you call "resist"... another term I should probably adopt, eh?).

The smaller PCB has a dark black soldermask because the IC in the center is an image sensor, and the idea is to prevent reflected and scattered light as much as possible.  Nonetheless, the traces are barely visible in the photos (easier with human eyes), so via pads might be visible to the camera even then.

If you're right, people will either leave [some] vias uncovered, or not apply a deep dark "resist".

It is absolutely necessary to work off the outside dimension of the pad around the via hole, not the hole itself.

[mikeselectricstuff]

Therefore, the fact that every time a component is placed the PCB looks different doesn't matter.  The software is only finding and looking at the through-hole pads (and treating them like "local fiducials" in a manner of speaking) to determine the position of the component center via the down-looking "nozzle camera".  Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).

Vias are normally covered in resist, so you  won't be able to reliably vision them.

Hole positions can have small offsets from the copper, which is why they're not used for positioning   
Any system that relies on additional design features like vias as fids will have limitations - if you're using  0201's, you'll be short on space and won't want to be adding extra marks if there aren't already enough vias.

I've never seen vias covered by soldermask.  My PCBs aren't that way.  Take a look at the images of two of my PCBs several messages back for examples (near bottom of page 1).

Then you clearly haven't seen many PCBs.

It may be possible to see via pads covered by resist.

Not on a black or white resist.

[bootstrap]

Therefore, the fact that every time a component is placed the PCB looks different doesn't matter.  The software is only finding and looking at the through-hole pads (and treating them like "local fiducials" in a manner of speaking) to determine the position of the component center via the down-looking "nozzle camera".  Since the software knows where those through-hole pads are, it will have no trouble finding them (differences in the appearances elsewhere on the PCB are not even noticed).

Vias are normally covered in resist, so you  won't be able to reliably vision them.

Hole positions can have small offsets from the copper, which is why they're not used for positioning   
Any system that relies on additional design features like vias as fids will have limitations - if you're using  0201's, you'll be short on space and won't want to be adding extra marks if there aren't already enough vias.

I've never seen vias covered by soldermask.  My PCBs aren't that way.  Take a look at the images of two of my PCBs several messages back for examples (near bottom of page 1).

Then you clearly haven't seen many PCBs.

It may be possible to see via pads covered by resist.

Not on a black or white resist.

I've seen thousands of PCBs, and designed dozens.  That doesn't mean I'd necessarily notice whether vias were covered with resist on other PCBs in some cases though.  I just poked around on the web and found that the vias are indeed covered by resist on some PCBs.  Hadn't noticed before.

I do know the default setting for my PCB software has been for resist to NOT cover vias.

I suspect white resist might be even more difficult than black, but I don't have any white PCBs here at the moment to look at.  I can see traces on my PCBs with black resist, but admittedly it isn't very easy, and might not be feasible for most cameras.

I don't consider it unreasonable to ask PCB designers to leave [some] vias uncovered (or not choose black or white resist).  That's not much of a price to pay to save boatloads on your pick-and-place machine, IMO.

[mikeselectricstuff]

I don't consider it unreasonable to ask PCB designers to leave [some] vias uncovered (or not choose black or white resist).  That's not much of a price to pay to save boatloads on your pick-and-place machine IMO.

I'd agree that leaving (some) vias uncovered would be a reasonable compromise, provided the system isn't going to need vias so close or so numerous that they need to be added specifically for the vision stuff to work.

[Spikee]

Placing 0201 / small qfn's or even bga is possible with my liteplacer which cost me 1000 euro. The problem is placing 5/10/100/1000 fully automated.
And yes I have placed bga's and or qfns with it. On my latest board I even placed bga's by hand so it is actually not that hard.

[bootstrap]

I don't consider it unreasonable to ask PCB designers to leave [some] vias uncovered (or not choose black or white resist).  That's not much of a price to pay to save boatloads on your pick-and-place machine IMO.

I'd agree that leaving (some) vias uncovered would be a reasonable compromise, provided the system isn't going to need vias so close or so numerous that they need to be added specifically for the vision stuff to work.

Based on the nominal design, the software would need at least visible 3 via pads in the 4" square around the center of every component.  Plus, at least 2 of the via pads must not be too terribly close together.  Actually it works with only 2 visible via pads, but prudence dictates "always have one extra".  Actually, with a bit more cleverness in the software, it should work with one via pad (though that's going further than my comfort level).

The process is possible with only 2 or 1 via pad because the software knows how much the x,y coordinate system of the PCB is tilted relative to the pick-and-place x,y axes from looking at several other components (not to mention the primary fiducial marks, which are generally separated the most, and thus give the best leverage for that computation).  Nonetheless, I'll continue to remember we need 3 to be safe.

In worst case scenarios there are other options, but they are not "nice".  For example, it is almost certain that solder paste will not flow beyond all four edges of any of the component pads (much less all the component pads).  So with some work, software should be able to find some straight edges on pads, and by that means determine the x or y location of those pad edges.  Of course, if the pads are round, that's a tad more difficult, but not impossible.

But I still would rather not "go there".

[bootstrap]

Placing 0201 / small qfn's or even bga is possible with my liteplacer which cost me 1000 euro.  The problem is placing 5/10/100/1000 fully automated.

And yes I have placed bga's and or qfns with it.  On my latest board I even placed bga's by hand so it is actually not that hard.

Can you elaborate?  Are you saying your machine can place 5, 10, 100, 1000 components of 0201 size precisely enough, but "not automatically"?  Or you mean something else?  Have you done something special with your machine or process, or you think everyone who buys one of those machines can place 0201s and 0.30mm ~ 0.50mm pitch components reliably?  I was under the impression they didn't claim to support 0201s or such fine pitch components, but I'll go check again to see if my crappy memory is lying to me.

Also, I am only concerned about placing BGAs/QFNs and other fine pitch components when their pitch is 0.30mm to 0.50mm.  I believe you hand placed 1.00mm pitch BGAs, but find it difficult to believe 0.50mm pitch, and definitely not 0.30mm pitch (unless you are one lucky SOB, or the most talented guy on the planet).  I tried to place 0.50mm BGAs under an excellent stereo microscope but had to give up.

Another difference might be, most of my PCBs are not "tiny" and typically contain 300~500 components.  I know my PCBs are bigger and contain more components than many PCBs, but mine are not that far outside the norm.  So my target is to support large PCBs with hundreds of components.  PS:  When I say "large", I mean roughly 12" x 16" or larger.

The price certainly is right!

Later:  Some people in their forum have problems with 0201s (not pickup, but placement precision).  However, as you point out in their forum, they may be able to add code to support their up-looking camera like neoden4 to improve their precision.

The precision of 4 out of 5 of the 0603 components in their video are fairly bad.  Yet all but one of the 0402 components were placed better than the 0402s!  I infer a consistency issue, perhaps.

I see, no feeders.  For some people I think that's a smart way to go --- keep it cheap.  But for my purposes feeders are probably almost a requirement.  Nonetheless, I very much like the idea of making the machine capable of working without feeders, because some people only make tiny PCBs with not so many components.

[Spikee]

Placing a few of one small component is usually not that hard, doing many automated is. But that really depends on how good the software is and if the people that wrote it actually understand the vision algorithms.

With those super small smd parts the pickup nozzle is also quite important.
On my machine I have only changed the camera's (stock one's really suck), leds and pickup nozzle.
I have done a few bga's and 0402 passives not 0201.

0.5 mm bga is actually not that hard to place. Just put the silkscreen line at the exact size of the part. That way it is easy to put it on the right spot.

0.5qfn and the likes (especially with big center gnd pad) can do decent self alignement. As long as the placement is not that far off.

When I place a qfn / bga with my liteplacer I do this:
1. cal via fiducials
2. goto component tray
3. center on middle of ic via vision , pick up
4. do rotational offset
5. goto location and place

This could be fully auto but the software does not support this yet.

0201 imperial is small but not super small. If you use IPC complaint footprint for the passive than there is quite some margin for error. Passives re-align quite good.

If you have time than do the alignement for each placement. If you do not have time than do it every 5 placements for example.

Doing it fully mechanical (and automated) is quite expensive

A machine like the neoden 4 should be able to place it (I have not used one). But that kinda depends on how good it is made.

Regarding feeders, if the cam just looks (and calculates offset) before picking up the part than there should not be an issue.

btw. If your boards are that large in dimension why bother with 0201 imperial?

The limiting factor I currently have is that somewhat affordable pcb services use quite big minimum via sizes. Unless you pay a lot of money and go for laser drilling.

My latest design contained about 270 components for a 80x100mm size so your component count is mid to high qty.

Whatever machine you end up using I advice you to do a few test runs of you board on one of their machines. To see if it actually works for your application.

It will take some more time until the open source and somewhat affordable commercial software is ready for decent vision implementation. When that is done a relatively cheap machine can do the components you talk about.

At this time it is not there yet. Doing it the commercial route at this time will probably cost you 60+ k for the pnp from one of the bigger mfg's.

[bootstrap]

0.5 mm bga is actually not that hard to place. Just put the silkscreen line at the exact size of the part. That way it is easy to put it on the right spot.

0.5qfn and the likes (especially with big center gnd pad) can do decent self alignement. As log as the placement is not super far off.

When I place a qfn / bga with my liteplacer I do this:
1. cal via fiducials
2. goto component tray
3. center on middle of ic via vision , pick up
4. do rotational offset
5. goto location and place

This could be fully auto but the software does not support this yet.

0201 imperial is small but not super small. If you use IPC complaint footprint for the passive than there is quite some margin for error. Passives re-align quite good.

If you have time than do the alignement for each placement. If you do not have time than do it every 5 placements for example.

Doing it fully mechanical (and automated) is quite expensive

A machine like the neoden 4 should be able to place it (I have not used one). But that kinda depends on how good it is made.

Regarding feeders, if the cam just looks (and calculates offset) before picking up the part than there should not be an issue.

btw. If your boards are that large in dimension why bother with 0201 imperial?

The limiting factor I currently have is that somewhat affordable pcb services use quite big minimum via sizes. Unless you pay a lot of money and go for laser drilling.

My latest design contained about 270 components for a 80x100mm size so your component count is mid to high qty.

Whatever machine you end up using I advice you to do a few test runs of you board on one of their machines. To see if it actually works for your application.

This is interesting, and the price is sooooo very "right" that I'm moderately impressed.

Just setting a 0.50mm QFN down at the correct position without smearing solder paste seems like a super-human feat to me!  The BGA is easier because maybe you don't necessarily need solder paste, so you can push it around while it is on the PCB.

From the neoden4 topic in this forum, it seems the neoden4 places about 98% of 0201 components precisely enough to work.  Unfortunately, a 2% error rate is too poor for my purposes.  If you look at the 150mm square PCB posted in my message near the bottom of page 1 of this topic, you'll see it contains about 50 of the 0201 components, and absolute boatloads of 0402 components.  A 2% error rate would mean most of my PCBs would fail due to bad 0201 placement... far too high!

I agree that placing 0201s and 0.50mm pitch fully mechanically and automatically is difficult... which is one reason machines that can do that are so expensive.  This is also why my approach is also vision based, albeit somewhat different.

BTW, it sure appears like the pads on their PCBs for 0201, 0402, 0603, 0805 components are significantly larger than the pads others (including me) put on the PCB for those size components.  While that definitely makes the placement process less difficult (less precise position requirement), that also somewhat defeats the purpose of tiny packages.  In fact, in most cases where I have 0201 components, they span from pad to pad under BGA components.  You can see that clearly in the 150mm square PCB I mentioned above.  The problem is, those pads cannot be made as big as they would be on their test board (assuming the follow the pattern established for the other sizes).

My current PCBs aren't that large, typically 3" to 8" square (or rectangular).  But at least two of my planned PCBs will be considerably larger (probably about 10" x 12" or 12" by 16").  Not as large as most of the PCBs I used to make in the "old days", which were typically 12" x 16 to 20" x 24").  But I also made some small PCBs back then too.

Also, most of my planned PCBs are "size sensitive".  Which means I need to keep them as small as possible (especially the larger ones).  Hence the even more urgent need for 0201s on large PCBs.

[vonnieda]

Hi folks,

I'm the author of OpenPnP and I feel like I need to clear up a couple misconceptions about my software.

1. OpenPnP is not magic software. It's current vision algorithms are pretty basic and bottom vision is brand new. It's working well so far but I've only tested it with a few dozen different components.

2. I do not know of anyone using OpenPnP to place 0201s and < 0.5mm pitch components. Maybe someone is, but I haven't heard of it. That is not my target market. I'm primarily interested in 0603s and 0.5mm+ pitch components. The software is geared towards hobbyists, prototyping and small businesses. That isn't to say it can't do it, but for components that small it's more a matter of how precise your machine is versus how fancy the software is.

3. OpenPnP does not "require" steppers. As noted, the hardware interface is abstract. All OpenPnP cares about is whether your machine can move, pick, and place. It doesn't really care about how it goes about any of those steps and people have used it with machines that range from sticks, bubble gum, and hope to dual head 14 nozzle 2.5m/s linear servo monsters.

Hope this helps. If you have questions about OpenPnP or need anything clarified, we're a friendly bunch. Join the mailing list and introduce yourself: http://groups.google.com/group/openpnp

Thanks,
Jason

[rx8pilot]

While not claiming to have the ability to jump ship and become a pick and place manufacturer - I do enjoy considering the issues. One of the biggest things to consider for anyone wanting to design and build one is to have some experience with the practical issues. I had designed and build a manual machine (very basic) and considered making a machine on a similar level to Lite Placer to help with my high density prototypes and short runs.

One day, I stumbled across my opportunity to buy and refurb a Quad machine. I threw myself into the pick-n-place fire instantly and had to deal with a mountain of issues from motion control to feeders and everything in-between. After getting the machine running in tip top shape, I was faced with the challenge of managing the process as a whole. This very steep learning curve totally changed what I would consider for building my own machine. Like @bootstrap said, I don't want to build the same old machine - I would only be motivated if I could see a path to delivering an innovation of some sort. Something that is uniquely helpful.

From a business perspective - I have never gone down the path of designing products to be a cheap as possible. I tend to go after the 'cost-effective' business solutions that are not the lowest cost, but have an excellent return. With PnP machines, a LOT of value is correlated with reliability AND speed. That is where the MyData class machines fit. They are reliable and extremely fast at the same time. Of course this also means that the entry level MyData machine is very expensive and only targets full time production. Sacrificing speed does indeed save a lot of money and I have always wondered what is the market size of a $20k PnP solution. At $10k, I think there are too many sacrifices to yield a machine suitable for a business. At $20k, I believe there is enough to work with to make a machine that has the right ingredients for small business in-house production work with modest volumes. That means the (in rough terms) the BOM plus labor would need to max out at under $10k if sales are direct and any profit is expected. The volume of sales has a lot to do with how that ratio is derived, but it is at least a rough starting point.

I would focus on a low cost feeder design. The machines core motion is relatively easy. A feeder scheme that can handle a very wide variety of parts with high-reliability and still be low cost will be key. I do, have a number of practical ideas (based on real concepts that I have already designed and built for other projects) to build a machine with very high feeder count, accuracy, and low cost to deal with low volume / high mix / fine pitch / low-modest speed. I have read a lot of 'overthinking' in this thread so far, the only way to get a low cost machine is by examining the real everyday issues and creating the simplest solutions possible to solve them in order of importance. Every nut, bolt, camera, connector, servo, etc adds to the challenge. Part presentation from the feeders is super important since the feeders are duplicated so many times on any given setup.

My current line of business is totally unrelated and it would require a career change to do anything with that. I have done a LOT of contract work in the past where my employers pay me by the hour to give up all of my concepts and ideas and they have literally made many 10's of $millions. That is my motivation to design my own products and get the upside benefit. My contract employers from the past essentially had the development capital and sales chain to take advantage of my ideas/designs/engineering/and implementations. These were great learning opportunities and I am grateful to have been able to see how a product goes from zero to hero. I am now saving all of my IP until a I can engage in a business opportunity that has an upside benefit. My eyes are always open to a new opportunity, but it has to be significantly better than the path I am currently on.

With that said - I am scared of going down the path of building machines with huge BOM counts and complex supply chains. Right now, I see more success in simpler projects (for me personally).

[Spikee]

I think that this design would be a good place to start from:

https://hackaday.io/project/9319-diy-pick-and-place

Belt driven (5mm  width or so), hiwin 2 contact 12mm wide rails, alu profiles.

This is what I would make of it:


For the Y axis 15mm hiwin 4 contact (HGR) bearing. Why? because it is pretty cheap, 4 contact and would allow the option to convert to small router or something like that.

For the X axis 35mm hiwin 4 contact (HGR) bearing. Why 35 mm? because the price difference is so low ... why not.  20-25mm would probably also be fine.
Instead of a small width belt the top design has and also my litplacer I would just put a oversized HTD 8M 30mm belt on it since it is not that much more expensive and will help to minimize backslash significantly.

For motors a 4 axis set of leadshine closed loop steppers. I have these on my cnc and they are great. I bought mine from ebay and they were not that much more expensive than a non closed loop stepper and leadshine driver. The encoder could be upgraded to 12 bit (is 10 bit std. , software supports 12bit).

Two motors on each of the y axis. Because 1. why not , 2. would allow possible cnc /... conversion.
 
For the head I would use (2 or 4 heads):
http://www.robotdigg.com/product/674/PNP+Machine+Nema11+Headset
with the samsung cp45 nozzle / holder instead of the Juki one's. As these are better nozzles in my opinion and allow auto change without buying any special adapters / actuators. Just pull the nozzle off with z-movement.

The x-axis profile is 80x60 x 660mm 8E LP , y-axis profile 80x160 8E SP 500mm (can be sized-up if needed).
The 30mm HTD belt would be enclosed in the 80x60 / 80x160 tube. Chain-guides to handle all the wires/pneumatics going to the head.

Design a enclosure around the frame to block outside light as this greatly influences the vision performance.
Have diffused white leds in the c-profile's , bottom of x-axis portal and even on the enclosure to ensure a "even" lighting condition at the base.

One could even use a t-slot plate in the middle of the base to allow easy mounting of pcb fixtures and the likes
I would use off the self (whatever works / is affordable / easy to use) feeders for this system. Using something like the openfeeder or whatever greatly increases time and manufacturing cost.

Design a machine like that would cost the following:
2x 500 mm hiwin HG15R rail =  80 euro
2x HGW15CC blocks = 68 euro
1x 660mm hiwin HG35R rail = 85 euro
1x HGW35CC block = 73 euro
HTD 8M 30mm belt is 20 euro per meter (4kN max allowed force)
Using a t-slot base would probably cost around 100-200 euro when using ISEL brand.
4x closed stoop stepper + driver is probably around 400 euro
Stepper shaft couplers = 100 euro
Then you need various pulleys, screws, washers ... = 100 euro
led stips probably another 100 euro
laser cut pvc/lexan/ .. enclosure probably 100-200 euro
Tiny G or whatever driver board 100 euro
various alu / plastic milled adapter plates 200 euro (when doing a group buy / china)
Since there is good even light a reasonable up/down cam is good enough = 100-150 euro
China inductive endstops = 50 euro
----
so for about 2000 euro ex tax via (all genuine parts) local  webshop (relatively expensive)
A really reasonable base can be constructed. With some shopping around / leaving t-slot out / downsizing 35mm rail one could even subtract around 300-500 euro from this.

I do not think that adding (cheap) ball-screws would add much speed/accuracy to this system.
The closed loop stepper driver can go up to 65535 micro-stepping.
On my machine it is at the default value of 1024 or so and I get 0.1mm repeat-ability with ease on a much heavier machine. Without doing any tuning.

This machine would be higher spec than quite a lot of hobby cnc routers I have seen. Thus it could also be used for various other configurations.

A machine like this should be able to place 0402 / 0201 mechanically without much problems. Doing visual calibration once in a while should make 0201 imperial placement have a low error rate of <1%.

If one want more feeders the width of the machine can be increased or the 'C' profiles can be changed to something else and allow feeders to come from all four sides.

Software would be openpnp or the likes with possible modifications.

[rx8pilot]

X/Y positioning is simple relative to the details of feeders, vision, nozzle changers, part release, pickup fault detection, fast part change over, tall parts, tube parts, etc are where things get challenging. All of those challenges start getting in the way of other things and the machine gets crowded and complicated very quickly. It seems that most discussions focus on the placement of the parts, where I would focus on how to deal with delivery and pickup of parts which where the real challenges are.

The PCB has fiducials and sits perfectly still the whole time. In a practical sense, there is no concern for vision focus, flex, or temperature on the placement end of the process. The feeders, on the other hand, have a much more challenging task of moving the tapes to the exact position and picking up the parts without them flipping on end or diagonally. The pickup part of the effort is where I have all of my problems - not putting the parts down. Most (if not all) high-end machines put an enormous effort into the feeders to reduce the mis-picks to a very low number. This effort makes the feeders very expensive and sophisticated. In my case (and many other businesses like mine) we need a lot of feeders on the machine so saving money here is a way to drastically reduce the overall solution cost.

Also, feeder banks would be on my ideal machine so that I can be loading a job of 40 feeders while the machine is using 40 different feeders. When it is time to switch - 2 banks of 20 feeders are swapped and ready in a minute, instead of loading each feeder one by one. Each feeder should have a QR code or bar coded that allows the machine to identify it and associate the part that was loaded. The QR code should be visible to the placement vision camera so it adds next to no cost at all. The software (including offline software for setup) is where the codes get matched with the parts.

The placement speed is generally not a limitation in a small entry level environment - it is machine setup and parts management. If I could have 40-60 parts in the machine on banks and swap out the banks in a minute or two to run a totally different job - the machine would be gold. The feeders would have to be very clever to achieve the precision and the low-cost which would allow a working set and a standby set of feeders to swap super fast.

I have made arc-second accurate positioners and very repeatable mounting fixtures for imaging (optical) - This is where I would put most of my efforts. After a feeder/pickup scheme is developed - I would then move on to the XY gantry and other systems.

[forrestc]

I've never seen vias covered by soldermask.  My PCBs aren't that way.  Take a look at the images of two of my PCBs several messages back for examples (near bottom of page 1).

Every board I make has, by design, all of the vias covered with mask.  If you look at many professional boards they have them covered as well.  There are many good reasons to cover the vias, and few reasons not to.

However, if you know your placement machine relies on the vias then you can choose that at manufacturing time for the boards.

I also don't see why you can't/shouldn't use the footprint that you're placing the component on in your vision system.  Even with paste on it it should provide enough contrast that you should be able to place using the pad.

[rx8pilot]

Using pads with paste is an unnecessary risk. Vias are not all that great either since they are crowded by solder mask and are mostly small. Traditional fiducial marks are super reliable and provide more than enough information for nearly all placements. In very critical areas, you can always put a pair of local fiducials to better ensure success.

They are so easy to put in the PCB design and so easy for the machine to see - why would you avoid them? I have used vias and other holes as fiducials on PCB's I did with only hand assembly in mind, but it was more accurate and reliable when I add fids.

[alexanderbrevig]

This discussion (of using vision for accurate mapping to a surface) reminds me of this https://glowforge.com/

[rx8pilot]

PCB's are a 2D world (right now anyway)

The requirements for vision are very straightforward in the big scheme of things. My smartphone has more than enough power to run an entire PnP machine.

As for that glowforge product, it looks like an expensive toy without much value.
 

[alexanderbrevig]

PCB's are a 2D world (right now anyway)

Are you placing mass-less components?

I guess the glowforge has value to those who find it valuable. Much like hobbyist PnP machines