O-M-G cheap pick and place machines coming!

[Precipice]

With feeders, alot of them are electrically actuated with a simple voltage, and you can even buy (or make) the simple slot system they mount to.  Quad feeders are a good example... widely available, cheap, easy, reliable, and 100% electrical.

Interesting - I just bought some Quad feeders to feed big tapes to my Versatronics RV1S. Anyone got any idea of the pinout?

[mikeselectricstuff]

I've said before that I think the best solution for any sort of home-brew PnP is to use off the shelf feeders

Yes but availability is pretty random

and nozzles

if there is something readily available and cheap, maybe, but a nozzle isn't a particularly complicated thing, so designing something custom that is cheap to make isn't a big deal.

(and maybe vision as well).

Hell no.
Cheap vision and computing power is the one thing that has changed out of all recognition since the "big iron" machines were designed, and is where there is the maximum scope to get very good performance at negligible cost.
A RasPi + camera or PC webcam +simple illuminator would easily be capable of all the vision performance you need.
P&P vision is very low on the scale of difficulty in vision systems - totally controllable environment, sharp edges, fixed focus distance and the simple task of measuring offset and rotation of a known shape.
The fixed distance means you don't even need fancy lenses - a standard lens screwed out to focus closer will be just fine.
It's also a minimal cost to have as many cameras as you need - e.g. for upward facing fixed cams, have a couple so you can use whichever is closest to the feeder, or seperate cams for large and small parts.

My Versatronics machine uses TV-resolution cameras and a Win95 era PC, and the vision works reasonably OK visioning on the fly between picking & placing down to 0402 and 0.5mm pitch.
A modern webcam or phone camera has 4x the resolution, and you can throw a ton of processing power at it for minimal cost.

[Neil Jansen]

sigh.. You guys make machines for the sake of making machines and tinkering with the machines
..
again : you're not in  run to build and deliver a cheap pick and place machine for hobby usage. you're in the run to have fun tinkering with vision recognition code and you happen to use a P&P as development platform.

I'm going to take the advice that I received yesterday and not engage you directly on this forum, I've been told you've racked up a bit of a history bashing open-source, and a bit of googling proves that.   He also added, "He's an experienced industry engineer but cannot come down from his lofty heights to where people with lesser requirements dwell." 

I'm not one to take trollbait, but please be respectful and do not tell everyone here why Karl and I chose to create this project.  You've never met either of us, you have no idea what our motivations are.  I would love to sit here and have a little debate with you, but I'm gonna get back to work.  I'm not really offended either, I'm just kind of awestruck at how smug you're coming off.   As entertaining as it is, I'm trying to pull off 100-hour weeks right now between this and my day job, so I can't sit here and type out petty rants because someone is wrong on the internet. 

The only thing I'll ask, as a general courtesy to us as human beings, is please leave the ad hominems out.  Feel free to criticize our machine plenty, we'll use that info.  But keep the jabs towards Karl and myself out of it, that just hurts your point.

[EEVblog]

I've said before that I think the best solution for any sort of home-brew PnP is to use off the shelf feeders and nozzles (and maybe vision as well).  The big guys have that stuff figured out 100%, and there's no cheap solutions that will ever come close to what Samsung/Philips/MyData have already figured out years ago. 

Availability would be terrible, and AFAIK prices aren't cheap because these things are sort after. So that's simply not a realistic solution.
Indeed, I think the biggest problem facing the low cost DIY PnP machine is just such a solution.
If someone could tackle that, and that alone, a self contained open feeder mechanism, then a good lot of the issues with realising this dream will go away.
But everyone seems to do it the other way around, they do the PnP head first. Which is understandable, because feeders aren't sexy or as fun.

[EEVblog]

I'm all for people working on these projects because I do think there is some useful stuff that comes out of it. But they seriously need to be more realistic. I think they should be very happy if they can get 0805 parts placed.

Yeah, I wouldn't even bother with 0402, even if you think you can do it. Even the big expensive machines often have a differentiation model or option for going from 0603 to 0402 or below. There are very likely good practical reasons for that.
I'd totally limit the goal to 0603 at best.

[Richard Crowley]

Did I miss it?  Was there any actual video of the gadget actually placing parts in real time? Talk is cheap, and so far, that is all I am seeing.
I just have a deep distrust of that triangular positioning scheme for doing something so critically orthogonal.

[mrpackethead]

...
Yes, a well designed manual system would probably be quite popular, once word gets around that DIY PnP's generally suck, and a well designed manual system can work very well.

I can confirm this supposition. We have a "real" PnP at my business, but setting it up is a PITA so it is only worth firing up for a production run. So to bodge together prototypes or just a few production boards I decided to buy a manual PnP ("eZpick") that is basically just a stabilized vacuum pickup pen: http://www.abacom-tech.com/%2FManual-SMT-Pick-and-Place-Machine-ezPick-P93496.aspx

It's hard to believe so simple a machine could be of any benefit, but it is - I timed the assembly of a couple different board designs alternating doing them by hand (ie - with tweezers) and with the eZpick (so that familiarity with what parts go where wouldn't unduly affect just one method) and using the eZpick was twice as fast and much less prone to smearing the solder paste from inaccurate initial placement.

I did make some modifications to it almost straightaway. It comes with a lazy susan carousel which you are supposed to place the board on - I really don't see how it it benefits anything as you can rotate the pickup nozzle with your thumb and forefinger freely so I ditched that. I also didn't want to have the boards sitting directly on a plastic surface both because of static charge buildup and there being no good way to fix the board in place, so I attached a piece of galvanized steel sheet (coiled flashing for roofing) on top of another piece of acrylic that fits precisely in between the slide bearing supports. That way I could use strips of flexible "fridge magnet" tape to hold the board and ground the working surface (via a 1M resistor and a wire to earth ground), while the extra sheet of acrylic lets me slide the entire work surface left or right, giving me a much larger working envelope. Attached is a picture showing my setup (sort of) in action (ie - this is a staged picture; no solder paste on the board, etc.).

NB - no affiliation with company or product, just a happy customer.

I found the Ezpick to be nothing but a pile of junk.      Regret buying it.   Found a Dima FP600 for sale on Ebay, and bought it for $1200. Awesome.

[EEVblog]

Ditto the software - all the talk of various different libraries, licenses etc. already gives me premonitions of it ending up a big mess held together with sticky tape that takes a week's learning to do anything useful.
At the basic level both the software and hardware it needs to be packaged up into a form that just works out of the box. By all means allow people to fiddle, but don't force them to fiddle just to use it at all.

I agree.
The "winner" in this race will be the one that has a solution that works out of the box, and works repeatably and reliably.
A machine that reliably does 1206 will beat a machine that's a bit dicky but can do 0402.
As with most OSHW, the vast majority of people do not actually want to tinker with it, they just want something that works out of the box. The Open aspect is just a warm fuzzy that they can tinker if they want. How many people build their own RepRap? Very few, bordering on zero compared with the numbers of 3D printers used. Indeed, the success of 3D printers has been that people produced ones out of the box that just worked.

[zapta]

The only thing I'll ask, as a general courtesy to us as human beings, is please leave the ad hominems out.  Feel free to criticize our machine plenty, we'll use that info.

Don't let the nay sayers to distract you, just pick the constructive feedback (and there was some in this thread) and keep going.  What you are doing is perfectly valid. You have a vision and you pursue it on your own time and money. Even if you will fail it would be worth trying and you will learn plenty along the way, both on the technical and business sides.

[Corporate666]

Yes but availability is pretty random

if there is something readily available and cheap, maybe, but a nozzle isn't a particularly complicated thing, so designing something custom that is cheap to make isn't a big deal.
[/quote]

I was thinking more along the lines of a PnP from a company that still makes feeders, so availability would be virtually unlimited.  I know Quad uses the same feeders on multiple machines, and my previous Fuji IP PnP used the same feeders across multiple machines (although electric and pneumatic).  I think many of the MyData ones use all electric feeders and those have been around for ages... there always seems to be a lot of supply on eBay of such feeders.

Same with nozzles... it's one of those things that sounds simple, but the tolerances are tricky to get just right.  On the Quad's there is the nozzle and a spring loaded piston that provides correct force when picking/placing.  I think the replacements cost $100 or so, and there is supply from both OEM and aftermarket.  Infinitely better than a POS Luer lock syringe tip.

Hell no.
Cheap vision and computing power is the one thing that has changed out of all recognition since the "big iron" machines were designed, and is where there is the maximum scope to get very good performance at negligible cost.
A RasPi + camera or PC webcam +simple illuminator would easily be capable of all the vision performance you need.
P&P vision is very low on the scale of difficulty in vision systems - totally controllable environment, sharp edges, fixed focus distance and the simple task of measuring offset and rotation of a known shape.
The fixed distance means you don't even need fancy lenses - a standard lens screwed out to focus closer will be just fine.
It's also a minimal cost to have as many cameras as you need - e.g. for upward facing fixed cams, have a couple so you can use whichever is closest to the feeder, or seperate cams for large and small parts.

My Versatronics machine uses TV-resolution cameras and a Win95 era PC, and the vision works reasonably OK visioning on the fly between picking & placing down to 0402 and 0.5mm pitch.
A modern webcam or phone camera has 4x the resolution, and you can throw a ton of processing power at it for minimal cost.

Interesting!  Thanks for the info - I didn't realize 'vision' was that basic of a task... Quad/Tyco liked to sell vision upgrades... dark field systems and all that jazz.  I have up and downward vision on my machine, but I only use the down for larger chips... and even then it needs to move it and scan the corners for anything TQFP-100 and up.  I think they wanted something like $3k or $5k to upgrade to the wider FOV dark-field imagine solution.

[EEVblog]

I was thinking more along the lines of a PnP from a company that still makes feeders, so availability would be virtually unlimited.

Yes, but at what cost?
I thought feeders were very expensive, like 4 figures a pop?
You can't rely on the 2nd hand market.

[Corporate666]

I've said before that I think the best solution for any sort of home-brew PnP is to use off the shelf feeders and nozzles (and maybe vision as well).  The big guys have that stuff figured out 100%, and there's no cheap solutions that will ever come close to what Samsung/Philips/MyData have already figured out years ago. 

Availability would be terrible, and AFAIK prices aren't cheap because these things are sort after. So that's simply not a realistic solution.
Indeed, I think the biggest problem facing the low cost DIY PnP machine is just such a solution.
If someone could tackle that, and that alone, a self contained open feeder mechanism, then a good lot of the issues with realising this dream will go away.
But everyone seems to do it the other way around, they do the PnP head first. Which is understandable, because feeders aren't sexy or as fun.

I've had several different brands of PnP machine and the issue you describe above is exactly what makes the Quad feeder solution perfect.  The feeder is self contained and just needs 12V (5V on older feeders).  There is no communication between the feeder and the machine.  The nozzle picking the part breaks a light beam which advances the tape to the next part when the nozzle leaves the beam.  The amount of advance is programmed into the feeder using a couple of buttons.  The Quad machine itself has no knowledge of a feeder area or anything like that.  You just set X/Y/Z pick locations (that you assign numbers to) and X/Y/Z place locations (also assigned numbers) and then your program becomes

PICK 1
PLACE 1
PICK 1
PLACE 2
PICK 1
PLACE 3

and such.  The beauty of it is that the machine doesn't care what you are doing for feeders... you can use Quad feeders, you could use a vibratory feeder, you could have some home grown solution, you could use something pneumatic, or anything else.  The machine also supports offsets per pick which lets you use either waffle trays for chips or strips of cut tape stuck down to a flat surface, if you like. 

These home brew PnP's ought to do the same thing - then they could say "compatible with Quad feeders, MyData feeders, Philips feeders, and our own design of feeder".  Because I totally agree with you - everyone tries to figure out the pick/place head and that's not the hard part - the feeders are the hard part.  So they have these shite solutions for feeding that nobody would ever use in a real environment and the whole machine ends up being a turd.

It's just like the CNC machines... if you look at what the home brew guys are doing, they usually totally miss the point.  They add useless features but then they use ACME leadscrews instead of ballscrews, thinking they can program around it.

[sleemanj]

Indeed, the success of 3D printers has been that people produced ones out of the box that just worked.

Mmm, yes, but I'd argue that the "ones out of the box that just worked" are out there at least in part because those who did want to tinker did tinker and their tinkering led to improvements that enabled somebody, not necessarily them, to create the "ones out of the box that just worked".

[GeoffS]

Indeed, the success of 3D printers has been that people produced ones out of the box that just worked.

Mmm, yes, but I'd argue that the "ones out of the box that just worked" are out there at least in part because those who did want to tinker did tinker and their tinkering led to improvements that enabled somebody, not necessarily them, to create the "ones out of the box that just worked".

Just having finished the build of a 3d printer, I'm now at the tinkering stage.
Without being able to refer back to those users who already have their printer working, this would be a long, frustrating task or rather  more of a long, frustrating task.

[miguelvp]

Indeed, the success of 3D printers has been that people produced ones out of the box that just worked.

Mmm, yes, but I'd argue that the "ones out of the box that just worked" are out there at least in part because those who did want to tinker did tinker and their tinkering led to improvements that enabled somebody, not necessarily them, to create the "ones out of the box that just worked".

Just having finished the build of a 3d printer, I'm now at the tinkering stage.
Without being able to refer back to those users who already have their printer working, this would be a long, frustrating task or rather  more of a long, frustrating task.

And all it took was for someone to put them out there to begin with.
I think that was exactly sleemanj's point.

[EEVblog]

I've had several different brands of PnP machine and the issue you describe above is exactly what makes the Quad feeder solution perfect.  The feeder is self contained and just needs 12V (5V on older feeders).  There is no communication between the feeder and the machine.  The nozzle picking the part breaks a light beam which advances the tape to the next part when the nozzle leaves the beam.  The amount of advance is programmed into the feeder using a couple of buttons.  The Quad machine itself has no knowledge of a feeder area or anything like that.  You just set X/Y/Z pick locations (that you assign numbers to) and X/Y/Z place locations (also assigned numbers)
*snip*
These home brew PnP's ought to do the same thing - then they could say "compatible with Quad feeders, MyData feeders, Philips feeders, and our own design of feeder". 

That sounds like a really good solution, and what I had in mind. A completely self contained feeder system. I did picture some sort of interface in addition to the built in control, but if not needed then all the better.
I wonder what the big boys would think of the design being copied and open sourced?

[EEVblog]

Mmm, yes, but I'd argue that the "ones out of the box that just worked" are out there at least in part because those who did want to tinker did tinker and their tinkering led to improvements that enabled somebody, not necessarily them, to create the "ones out of the box that just worked".

Yes, of course, I never meant to imply otherwise.
That initial seed and the development is vital.
My point was that for anything to succeed in the marketplace, and specifically for a new enabling technology like low cost PnP to become mainstream and usable, relies upon someone taking that initial work and building an out-of-the-box solution that works well and reliably enough for the market to take off and set a defacto standard baseline of usability.

[Karl Lew]

As with most OSHW, the vast majority of people do not actually want to tinker with it, they just want something that works out of the box. The Open aspect is just a warm fuzzy that they can tinker if they want. How many people build their own RepRap? Very few, bordering on zero compared with the numbers of 3D printers used. Indeed, the success of 3D printers has been that people produced ones out of the box that just worked.

Agreed. We have found individual tinkering to be inefficient, especially with the broad engineering skillset required for FirePick Delta. Neil is a much faster mechanical/electrical engineer than I am. I have made massively more mistakes in software than he has--hopefully there is some room for innovation in my head. Christian has worked extensively with SMT in his job. Simon has worked 5years with OpenCV, etc. Our frustration with our individual limitations and non-replicable tinkering has brought us together as team. Yet even with that collaboration, it will take multiple iterations to step up to the high standards required of turnkey, especially for the harried small business owner making custom boards. My earlier mention of high school students was to provide an example of an easier, achievable, yet relevant use case. It's a very real use case and I know of at least two high school students out there interested in building FPD today.

The open source approach allows us to share knowledge and resources to put together a minimally useful machine. But that minimally useful machine will be a machine that we can tenaciously iterate and improve collectively or individually towards turnkey. In contrast, today, if you want a TM220 with vision, you have to wait for the Chinese.

[EEVblog]

Agreed. We have found individual tinkering to be inefficient, especially with the broad engineering skillset required for FirePick Delta. Neil is a much faster mechanical/electrical engineer than I am. I have made massively more mistakes in software than he has--hopefully there is some room for innovation in my head. Christian has worked extensively with SMT in his job. Simon has worked 5years with OpenCV, etc. Our frustration with our individual limitations and non-replicable tinkering has brought us together as team. Yet even with that collaboration, it will take multiple iterations to step up to the high standards required of turnkey, especially for the harried small business owner making custom boards.

Yep, unfortunately it's not going to be easy or quick.
Just look at Makerbot, arguably one of the major drivers of the out-of-the-box success of 3D printers. Took them 3+ years, a lot of money, and a lot of people before they started to get it right, and many say they still haven't. My Makerbot Replicator(1) for example is a dog when it comes to extruder reliability.
And 3D printers are relatively simple devices compared to PnP machines, and that's why, realistically many are always going to be quite skeptical about a PnP.
Not that it can't be done, it certainly can, it's just whether is can be done for low enough cost?
Although remember, the 3D printer price points are above $1K or even $2K for decent machine that works well.
No one expect a $300 PnP machine. Heck, you can't even get a half decent 3D printer for that. You can argue it's pointless to even try and meet a price point, better to focus on getting a solution that is robust and works. And then it costs what it costs.

[Karl Lew]

No one expect a $300 PnP machine. Heck, you can't even get a half decent 3D printer for that. You can argue it's pointless to even try and meet a price point, better to focus on getting a solution that is robust and works. And then it costs what it costs.

When I designed FirePick 1, the predecessor to FirePick Delta, my replicable cost goal was $500. This is not end user cost unless the end user is a maker. I adhered to a very strict protocol of multiple, repeatable sources and completely avoided eBay 1-of-a-kind-fire-sales for the simple reason that bragging that I made a cheap machine was irrelevant if nobody else could build the machine. So I tried as hard as I could to make FirePick 1 cost $500.

And I failed.

FirePick 1 cost ended up at about $900. And this is where it gets interesting. Analyzing the cost of FirePick 1 revealed big chunks of change going towards certain things like ESD mitigation. FirePick 1 uses Ponoko anti-static plastic 2D laser cut pieces, collectively $200. Neil called me up and said, "Hey Karl, I have carbon nano-tube 3D extrudable plastic--it's ESD safe." When I heard that, I just started laughing. I would have done cartwheels but I'd have killed myself being a klutz. Cheap ESD safe parts. Wow.

That wasn't the only cost mitigation prompted by Neil. FirePick 1 relies on Shapeways for accurate SLS 3D printed parts under the assumption that extrusion printers were cantankerous and inaccurate. Total cost, for Shapeways FirePick 1 parts was about $200. I might have been right about 3D extrusion printing limitations 2 years ago, but the revolution in home 3D printing now fields machines that rival the accuracy and utility of SLS. Once again, Neil demonstrated a way to shave off a chunk of change and speed up the development cycle for mechanical parts from weeks to hours.

Lastly, FirePick 1 relied on the TinyG motor controller and even then it also requires a custom board as well as a Raspberry Pi, so the electronics costs also hovered in the $200 range.  Neil said, "heck, we can do better."  And several team members have put their heads together to make that custom board happen. Currently we're all using RAMPS/Marlin, but only as scaffolding for future iterations.

In terms of robustness, we hope that the open sharing of the FPD design ensures that we have a scientific basis for comparison since the machines are quite similar (300x300x500mm). We trust that this will enable a natural and rapid process of selection towards robustness and turnkey viability. In practice, I can assert that FirePick Delta is much easier to build and more robust than FirePick 1. We'd like that trend to continue.

[mikeselectricstuff]

A quick look on ebay shows used Quad feeders are at least $100 per lane, and pretty much only available in the US.
OK you might occasionally find a big pile but this is not sustainable.

One question when designing a feeder is whether to go single or multi-lane.  My thoughts are probably the latter as it has scope for sharing some electrical & mechanical parts. However it is a bad idea to go too overboard in terms of feeders per bank, as this starts limiting flexibility, and is a pain of you have to unload all lanes to clear a jam in one - My machine used 8 lane feeders, which is too many.
I think the optimum number is around the 4-5 mark (8mm) , with the capability to adjust the widths - e.g. where you have 2 lanes of 8mm, it should be possible to move the guide between lanes and use it for one lane of 12 & 16, and remove another guide for 24mm

On a similar subject, does anyone know of an existing feeder that's good for loading short tapes? I believe Mydata have short-tape feeders but how re-purposable are they?

[mikeselectricstuff]

I was thinking more along the lines of a PnP from a company that still makes feeders, so availability would be virtually unlimited.

That would always be way too expensive. These feeders are designed for high reliability under heavy usage for years on end, and would be way over-engineered for something like this.
 I doubt you could but a single new feeder for any current machine for less than the target price of this machine.

there always seems to be a lot of supply on eBay of such feeders.

Prety much only in the US, or maybe the Far-East, and the weight would make shipping expensive

Same with nozzles... it's one of those things that sounds simple, but the tolerances are tricky to get just right.  On the Quad's there is the nozzle and a spring loaded piston that provides correct force when picking/placing.  I think the replacements cost $100 or so, and there is supply from both OEM and aftermarket.  Infinitely better than a POS Luer lock syringe tip.

Again. $100 is way too expensive for a machine like this

[EEVblog]

It always comes down to the feeders again and again every time someone starts or shows off one of these PnP machines.
Someone needs to come up with a clever cheap solution for this, that works, and works solidly.
A good niche business to sell them ready made and tested.
Then let every fight it out to bring a low cost PnP to market.

[mikeselectricstuff]

It always comes down to the feeders again and again every time someone starts or shows off one of these PnP machines.
Someone needs to come up with a clever cheap solution for this, that works, and works solidly.
A good niche business to sell them ready made and tested.
Then let every fight it out to bring a low cost PnP to market.

Absolutely.
A standalone feeder that was cheap and easy to use would even be useful for hand-placing.
Obviously the shape would need to be amenable to sitting on a flat bench with a low feed position, but that would also be a requirement for a cheap P&P, as people would want to sit it on a table, not have a dedicated stand that allowed feeder parts to hang underneath.
 

[mikeselectricstuff]

Just thinking about the minimum requirement for a standalone feeder, and taking a bottom-up approach.
What's the cheapest suitable actuator?
RC micro servo has to be it by very long way. Has the right amount of movement and torque, and a doddle to control very cheaply.
OK it's not up to industrial duty cycles, but so cheap that heavy users can regard it as a consumable - just design it to be easily replaceable.

I'd imagine one issue is wear on the feedback pot.

I can visualise that it would be possible to make a very clever linkage that moves a pin up, pulls along, then drops down again from a single servo.
But maybe keep it really simple  and just use 2 servos, one for pin up/down, one for drag-along.

If you did use 2 servos, you could potentially share the pull-along one across multiple lanes.

Even with micro servos, you'd struggle to get single lane width down to get good density for 8mm lanes, so maybe start with a width to accommodate 24mm tape, which is the largest common size, and have it able to  be split into 1x16, 2x12 or 3x8mm lanes. Maybe by an interchangeable guide plate.

That leaves cover tape pull-back. The approach on the cheap Chinese unit of a common shaft with friction drive is one possible, but imposes mechanical constraints on positioning.
My P&P uses a reel driven by a belt which is friction driven from the top of the main tape - I think this could be a viable approach, or something else which uses the same mechanism for the pull-along and cover take-up. Maybe a ratchet drive from a wire linkage driven by the pull-along servo.