Wanted to buy: Moderate pick and place line.

[mikeselectricstuff]

The problem is if you do the feedback on the rails, you need  rigidity on the gantry. If there was a way to measure the position of the tip in free space, then you could use a really cheap/simple drive with closed-loop feedback.

[flynnjs]

However I have no idea how feasible it would be to measure this - laser TOF/interferometry is probably the only viable option, which maybe rather tricky to do in practice.

Inkjet printers can do insane positioning with low cost mechanics. They're fairly fast too. I suspect there's a lot to learn there.

Just been mulling the comment about rocket science... I must be in the fairly small intersection of having home built a space satellite and also owning a P&P  machine   

[rx8pilot]

great, now you guys are encouraging me to build a P&P just to see if I could do it. 

It would be a fun project to have a group of smart people work out a low-cost approach based on existing commodity tech. I have loads of ideas for feeders and placement but always short on time. Most of my ideas are based on commercially successful designs I have done in the past, not wishful thinking. I do have 5 axis CNC mills,

Back on topic...I need a P&P ASAP to get my regular work out the door.

[mrpackethead]

great, now you guys are encouraging me to build a P&P just to see if I could do it. 

It would be a fun project to have a group of smart people work out a low-cost approach based on existing commodity tech. I have loads of ideas for feeders and placement but always short on time. Most of my ideas are based on commercially successful designs I have done in the past, not wishful thinking. I do have 5 axis CNC mills,

Back on topic...I need a P&P ASAP to get my regular work out the door.

Join the fire pick delta project.  Its coming along nicely..   it will be a sub $500 project.  Sure it will have lots of limitations, but its going to do a job.

[Precipice]

Inkjet printers can do insane positioning with low cost mechanics. They're fairly fast too. I suspect there's a lot to learn there.

Horizontal is just watching optical marks as the head moves and firing ink at precisely the right time - it's not accurate positioning, it's firing ink when the head's in the precise position.
Vertical is accurate paper feeding - and that _is_ magic.
I fear that neither of these translate well to pnp, though, where managed acceleration while maintaining precision is the name of the game...

(That said, if we're going to degenerate this thread into 'things that would be cool in pnp-land...
How about a stacked component dispenser. Load it, possibly offline, with a tall stack of a component you use a lot of (decouplers, 1K pullups, whatever), then when you want to place them, just hold the dispenser in the right place, press the stack from the top, and have the bottom one stick to the paste. Lift off, and repeat on the next position. No need to go back to the feeder, no need for vision, since they're all held in the stack, and the bottom is 4 ptfe springy walls or something, so the parts are in the right place.)
Ah well, engineering hell, I suspect. I just get bored waiting for my head to go back to get yet another decoupler, and I can't begin to afford one of those multi-nozzle chipshooter machines that place faster than the eye can see...

And yes, join the fire pick project! I would, if I had any time at all.

[mrpackethead]

they are going to make a kit some time soon,  so it will be the IKEA of the PNP world. :-)      Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

Not something that you would seriously use in a commercial set up, and i wonder how long parts printed on a 3d printer would work. but if your making 2 project boards per month.. it probably will last a long time.  And theres just the "hey thats cool factor"

The quads are an interesting idea..

Space is a premium in our shop,  the Autotronika 70 is looking good.   Mmm.  Decisions Decisions.   

[rx8pilot]

Join the fire pick delta project.  Its coming along nicely..   it will be a sub $500 project.  Sure it will have lots of limitations, but its going to do a job.

That looks fun and I need a hobby. I may look into what I could contribute. The options for small businesses are dismal since the market goes from toys to mega-machines with very little in the entry level area to kick start a business. We have painfully struggled to work up to being able to acquire a $50k P&P which is still low-end. Now I have all the resources and experience to build whatever I need/want, but it nearly killed me to get here.

For today's challenge, I have pretty much decided that the Quad 4000C from PPM [Updated Quad 4C] is the perfect solution for our modest needs with room to grow tomorrow. They now offer a way to do cut-tape which is very common for me doing small batches with parts that cost $7-$10 each means that I am always in a cut-tape world. Splicing onto reels or paying $7 for each Digi-Reel is not needed.

[flynnjs]

   Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

WTF? My 15yr old P&P visioning is almost instantaneous on a Pentium MMX 233MHz. With the GPU goodies that a Pi has ... that is criminal!

[Precipice]

I'd suspect some combination of: unnecessary resolution (Pi-cam is 3 or 5 Mpixel?), naive coding (let's face it, there are a million ways to make image processing go slowly), and the word 'java' just won't leave my mind

If this ever becomes a real issue, I'm sure it'll be sorted pretty quickly. I'd rather people concentrated on getting a good design, than tuning a crappy design for speed...

[mrpackethead]

Yeah, i'm sure it will get better and faster.  It may be quite a few things that are contributing to it..

[nctnico]

they are going to make a kit some time soon,  so it will be the IKEA of the PNP world. :-)      Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

I agree. If you have vision you can do the fine positioning based on the vision system instead of relying on accurate mechanics.

[Corporate666]

That looks fun and I need a hobby. I may look into what I could contribute. The options for small businesses are dismal since the market goes from toys to mega-machines with very little in the entry level area to kick start a business. We have painfully struggled to work up to being able to acquire a $50k P&P which is still low-end. Now I have all the resources and experience to build whatever I need/want, but it nearly killed me to get here.

For today's challenge, I have pretty much decided that the Quad 4000C from PPM [Updated Quad 4C] is the perfect solution for our modest needs with room to grow tomorrow. They now offer a way to do cut-tape which is very common for me doing small batches with parts that cost $7-$10 each means that I am always in a cut-tape world. Splicing onto reels or paying $7 for each Digi-Reel is not needed.

I did the exact same thing as you - looked at the market (in my case, I bought a few used machines) and after lots and lots of pain, trial and error and money wasted, I ended up with Quad 4C's.  I know I keep shilling the machines like I have a stake, but I don't... the nice thing is that IF "the shit hits the fan", the people from PPM are always a phone call away.  There are 2 Quad 4C's on eBay right now with a few dozen feeders each for the $4k range... I'm tempted to buy one of them just for the feeders, although I already have quite a few 5V feeders already.

Have you talked to PPM?  May I ask what they quoted you for the 4000C?  I am pretty close to their shop and I've been there a few times and talked to them lots.  They want $7500 for the Windows upgrade to the 4C as a self-install kit, but they also want you to send back the old parts to them.  And they want $26k for a 4C refurbed and converted to a 4000C, or $19k for a 'refurb' 4C.   I got a machine that was just refurb'ed by them and used by the client for a few months for $6k.  I also bought two other 4C's for around the same price... and frankly, other than the refurb machine, I probably paid way too much considering other 4C's are available on eBay for $4k.

I keep saying it but the nice thing about 4C's is they are dumb.  You tell it to go to A, pick, go to B, place.  So you can write all sorts of creative programs for dispensing or picking from different locations (waffle trays, strips, or even parts just taped to the table top).  The whole table top is just an array of drilled/tapped holes so you can mount anything anywhere. 

I am still very tempted to do the windows upgrade, just because the software/interface is a pain in the ass sometimes.  But for $7500 and it doesn't even include new motor drivers or anything mechanical... IMO is ridiculously expensive.

[rx8pilot]

The all-in 4000C is $28k. I also saw the machines on eBay and a few brokers that are, of course, as-is. Troubleshooting and replacing parts on a machine that has been around the block is not going to help me get PCB's out the door. I have no experience with these and I am sure the learning curve is not trivial. I did almost all of the maintenance on my big CNC's including repair/rebuilding a 400 volt vector drive for the 30hp spindle - but that was only because I had to.

I like the support that PPM can offer and if money is tight, I may ask for the older DOS version. I am well versed in a DOS world so that part does not scare me too much. If I can deal with the money end, I will just get the newer, updated 4000C that I would not likely need to fiddle with for a long time. The windows interface will also allow me to have operators up and running much faster than throwing someone half my age at a DOS machine.

This will be my first and only P&P machine so reliability and ease of use are very important. I hope its the right call. $50k is cheap for a P&P, but a lot to lose if it is a rotten decision. It seems like I should be safe though.

[mikeselectricstuff]

How about a stacked component dispenser. Load it, possibly offline, with a tall stack of a component you use a lot of (decouplers, 1K pullups, whatever), then when you want to place them, just hold the dispenser in the right place, press the stack from the top, and have the bottom one stick to the paste. Lift off, and repeat on the next position. No need to go back to the feeder, no need for vision, since they're all held in the stack, and the bottom is 4 ptfe springy walls or something, so the parts are in the right place.)

Pick & Pez...!
Should be viable to have it load itself from tape out of a feeder.
Sort of like a multiple head, but stacking parts vertically.
Speaking of double heads, it would seem sensible to look at having a dual head, as that would give the biggest delta speed vs. delta cost - more than 2 will have diminishing returns.

[mikeselectricstuff]

they are going to make a kit some time soon,  so it will be the IKEA of the PNP world. :-)      Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

I agree. If you have vision you can do the fine positioning based on the vision system instead of relying on accurate mechanics.

You can't do fine positioning with downward vision as here will be paste on the pads. You could use vision to calibrate out fixed errors, linearity issues etc. by imaging a reference grid

[mikeselectricstuff]

   Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

WTF? My 15yr old P&P visioning is almost instantaneous on a Pentium MMX 233MHz. With the GPU goodies that a Pi has ... that is criminal!

You beat me to it! Vision on a RasPi should be pretty instant, however once you get it down to a few hundred mS it is no longer in the critical path as you have the time between taking the image and travelling to the placement position before you need the results - delayed detection of a mis-pick is the only case where slower vision would waste a little time.

[Corporate666]

The all-in 4000C is $28k. I also saw the machines on eBay and a few brokers that are, of course, as-is. Troubleshooting and replacing parts on a machine that has been around the block is not going to help me get PCB's out the door. I have no experience with these and I am sure the learning curve is not trivial. I did almost all of the maintenance on my big CNC's including repair/rebuilding a 400 volt vector drive for the 30hp spindle - but that was only because I had to.

I like the support that PPM can offer and if money is tight, I may ask for the older DOS version. I am well versed in a DOS world so that part does not scare me too much. If I can deal with the money end, I will just get the newer, updated 4000C that I would not likely need to fiddle with for a long time. The windows interface will also allow me to have operators up and running much faster than throwing someone half my age at a DOS machine.

This will be my first and only P&P machine so reliability and ease of use are very important. I hope its the right call. $50k is cheap for a P&P, but a lot to lose if it is a rotten decision. It seems like I should be safe though.

I'm happy to help you out with programming and setting up your 4C if/when you get it. 

As far as the Windows software version - the thing to keep in mind (other than pricing) is that it only replaces the firmware/software and the internal CPU of the 4C... so all of the vision, motor drivers, sensors and everything else stays the same.  In other words, it's 100% about ease of use, not performance or reliability.  They do have some features that aren't in the stock 4C like inventory management, but not much.  That said, it is a nice upgrade... although it is not compatible with anything higher than Windows XP, which recently (or soon?) is having support ended for it, so in a few years might be like having a Windows 95 machine.

You definitely will not be unhappy with the 4C.  Like I said, I've had several PnP's and the 4C just keeps chugging along.  I haven't had any problems with mine.  Only thing about PPM is boy, will they charge you for everything and to the max as well... so I'd make sure everything they promise you is 100% spelled out in writing.  Not that they are dishonest, just their prices are sky high for parts and service.

Something else to keep in mind, if you decide to save a few bucks... they sell refurb machines, but really, it's just taking it all apart, repaint and new plastic guards, etc.  They will also sell you a "gone through" DOS machine for substantially less than a full refurb DOS machine (like $9-10k).  Just something to consider.

And it probably goes without saying, but get as many feeders as you can up front - they are cheap up front, expensive later. 

[Precipice]

You can't do fine positioning with downward vision as here will be paste on the pads. You could use vision to calibrate out fixed errors, linearity issues etc. by imaging a reference grid

With lots of processing, and a downwards-looking  global shutter camera running at high frame rates, you might be able to do something useful. Sort of terrain matching as you go in - match on lots of pads, not just the one you're aiming at, and the effect of solder paste should be less. Some serious processing needed in real time, though, and I'm sure there would be cases where it failed - parts in the middle of nowhere, etc.

[rx8pilot]

I'm happy to help you out with programming and setting up your 4C if/when you get it. 

Thank you so much for the practical wisdom. I am definitely thinking about cost/benefit of the various options. I am planning on a bunch of feeders - enough for all the boards I regularly would do with some left over for prototypes.

Do you have any way to estimate programming time in the DOS version? Is is possible to look at the BOM lines and get a sense of the setup time? Obviously a newbie P&P question, but any wisdom in that area would help me understand what I am in for. Is editing an existing program reasonably easy - to accomodate a PCB revision where a few of the parts move.

Will the machine alarm on a mis-pick or does it try again and keep going?

Is it practical to machine place big parts like 16 position headers (44mm) tape, push-buttons (24mm tape), etc. We have been getting these parts on tape, but placing them manually.

[IconicPCB]

We have been placing push button switches ( on tape ) from a cut tape feeder on our M10V.

[Corporate666]

I'm happy to help you out with programming and setting up your 4C if/when you get it. 

Thank you so much for the practical wisdom. I am definitely thinking about cost/benefit of the various options. I am planning on a bunch of feeders - enough for all the boards I regularly would do with some left over for prototypes.

Do you have any way to estimate programming time in the DOS version? Is is possible to look at the BOM lines and get a sense of the setup time? Obviously a newbie P&P question, but any wisdom in that area would help me understand what I am in for. Is editing an existing program reasonably easy - to accomodate a PCB revision where a few of the parts move.

Will the machine alarm on a mis-pick or does it try again and keep going?

Is it practical to machine place big parts like 16 position headers (44mm) tape, push-buttons (24mm tape), etc. We have been getting these parts on tape, but placing them manually.

Sorry for the length, I type fast

A tip on feeders... there are 5V feeders and 12V feeders.  Go with 12V if possible.  The only real difference is the 12V feeders index faster, but if you place a frequently used component in a 5V feeder near the PCB, it's possible it may not have indexed to the next part by the time the head comes back.  12V is standard on anything recent, but sometimes the OEM's will try to offload 5V feeders on those who don't know better

As far as programming, the machine works as such... you program picks and places as two separate actions.  So pick 1 might be a 10k 0805 resistor.  You might have 10 of those on your board.  So that would be 1 pick and 10 placements.  Your program would be (literally)

PICKUP 1
PLACE 1
PICKUP 1
PLACE 2
PICKUP 1
PLACE 3
..
..
etc.

So as you can see, you can program picks independently of placements.  With this setup, it is ideal to keep often used components in fixed locations on the machine, however if you wind up using more 20k 0805 resistors and want to give them the prime spot, you can just move your 10k 0805 feeder somewhere else, program it as a new PICKUP and edit your program in about 30 seconds.  And as you can see, editing a program then becomes super simple.  You can add lines, remove lines, edit pickups/placements, or you can easily globally edit all coordinates in a program.  There are also board and panel pattern repeats for picks or places.  So if you have (say) a line of 20 0805's in a row all spaces 0.250" apart, you don't need to program each individual placement.  Likewise, if your boards are panelized, you program the first board and then tell the machine the X and Y distance to the next boards and how many in X and Y.  I have a board with 12 different parts per board, 20 placements per board, 10 boards per panel.  The program for all this is maybe 30-40 lines long and that includes a nozzle change.

As for programming, I do it with the handheld device.  You program am offset (in X and Y) between the camera zero and the nozzle zero.  The manual tells you how to calibrate this offset... you never really need to change it.  Then you program your pickups and placements by moving the camera crosshairs over your pick and place locations - the machine takes care of factoring in the offset between the camera and nozzle.  There is also a Z height value for picks and places, but Z generally won't change for a given component/feeder even if you remove the feeder and re-insert or change out the tape.  Likewise, Z doesn't change for placements either.  And since your PCB will be level, once you program the Z height for, say, an 0805 resistor on your board, you can just copy-paste the Z value for all other 0805 resistor placements.  The Z nozzle is on a spring loaded assembly so it's not all that crucial.

All X/Y/Z coordinates are just numbers in thousands.  So a pick might be X = 8564 Y = -12345 Z =5483.  Those numbers are in 1/1000th's of an inch from the origin, so if you ever notice a component is slightly off, it's super easy to move it 0.010" to the left, for example. 

There is also an auto-programming utility... it's a simple DOS based utility that takes either GERBER data or perhaps standard PCB output files (I forget, I never use it) and outputs your PnP program in coordinates.  Because the programming is just in 1/1000th of an inch, you should be able to export the data from your CAD software also, and easily add in the offset from the machine home position to the X/Y zero position of your PCB.

As for mis-picks.. you set the X/Y/Z size of your component and 2 tolerance values, in absolute 1/1000" and in percentage.  So an 0805 might have a tolerance of 20 and 10 and 10%/10%.  If the vision system does not measure the part in this range, it moves the head to an X/Y location on the table (that you set) and drops the part.  So I just have it go home, and put a small plastic tub in that area.  It will try to re-pick the component 3 times on it's own.  Then if it still fails, it will re-set the Z height of the nozzle (basically it re-homes the Z height using the same laser used to center components) and try to re-pick.  If it fails to pick successfully again it will move the camera over the component it can't pick and start beeping.  You hit stop - machine goes home... fix the problem - then hit start, it will pick up where it left off.  It will even pick up where it left off if you shut the machine off overnight.

Sorry for the length - any other ??'s, just ask.

[Corporate666]

Oh, as for placing large or odd shaped parts... no problem.  That is one of the advantages of a gantry type PnP as opposed to a chip-shooter style machine.  Gantry is slower, but less acceleration/deceleration so they can work with larger parts.  You can also adjust traverse speed if you find a large part is coming loose when the head moves... but if that happens, you almost certainly have the wrong nozzle.

There are different nozzles spec'ed to work with different size parts.  For bigger stuff like TQFP's, the nozzle is of course larger and has a rubber seal to maintain maximum vacuum against the part.  It uses a venturi to create vacuum from air pressure, so you can crank up the air pressure for more suction as well.  You can also buy (or have made at a local machine shop) custom nozzles if necessary.  I needed to place PLCC-4 LED's with a dome lens, so it was easy to get a nozzle made with a conical opening in the tip which came down and auto-centered over the LED and also held it flat against the base instead of trying to pick it from the domed portion.

[coppice]

   Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

WTF? My 15yr old P&P visioning is almost instantaneous on a Pentium MMX 233MHz. With the GPU goodies that a Pi has ... that is criminal!

You beat me to it! Vision on a RasPi should be pretty instant, however once you get it down to a few hundred mS it is no longer in the critical path as you have the time between taking the image and travelling to the placement position before you need the results - delayed detection of a mis-pick is the only case where slower vision would waste a little time.

A few hundred ms might be a bit long, if you aim to place a part per second, which I think is a reasonable minimum rate for a serious project. 100ms might be a reasonable goal.

The firepick project page makes it look like a project going nowhere. Instead of using the best available free stuff they seem to be working hard on the development of a wheel. Presumably this wheel has to be a complete new design so it can be in pure Java.  Are they really starting from scratch with vision, and is it really being developed by people who consider this a really really hard problem, like the web site page says?

[rx8pilot]

Sorry for the length, I type fast

Wow! That is very useful information. I am going to look at the less expensive DOS versions knowing that I can always upgrade myself. The concept sounds very easy. I am totally comfortable in a DOS world and 3.5" floppies, I have spent 6 years programming my 5 axis milling machines that will tear my arm off if I push the wrong button, so I should be able to deal with [PICK(X,Y)] [PLACE(X,Y)]

These machines sound very flexible and certainly capable of doing what I need them to do. Cant wait to put my vacuum pen away for a while.

[nctnico]

   Its amazing that a raspberry PI can do the Vision.  It does it, its just slow..  ( couple of seconds per part ).. but hey it will be $300!!

WTF? My 15yr old P&P visioning is almost instantaneous on a Pentium MMX 233MHz. With the GPU goodies that a Pi has ... that is criminal!

The SoC on the Pi is optimised to be a setup box. So it is just good at decoding video. Not processing it. The CPU in the Pi isn't very beefy so I'm not surprised it is slow. The algorithm on your old P&P is probably better optimised as well.