Pick and Place machine brainstorm

[Chris Mr]

Hi All,

I've been thinking for ages that a really low cost pick and place machine would be great but up until recently couldn't come up with something that breaks the stepper motor barrier.  I say barrier because a) they aint cheap, b) they need drivers which aint cheap and c) programs to drive them.

So here's the idea for overcoming the stepper motor barrier.  What I realised, which potentially removes the steppers, was that for a pick and place machine you don't need to know where the business end (picker / placer) is until it's at the target position - once at the target position you need to know it's arrived but that is all.

Ages ago I made a capacitive sensor that measures distance.  You can put the capacitor in an oscillator and measure the frequency really easily.  So if you make a long capacitor (out of PCB say) and have a 'U' shaped secondary plate* that moves with the axis you get to know where the axis is; rather than stepping to a 'feed forward' position using steppers.  Then all you need is a signal generator (read DDS) and output that to a phase detector that's looking at the oscillator whose frequency is dependent on the position of the axis.  Once 'locked' you are ready to pick or place (or whatever else).

The thing is, for this idea you can just use belts and simple DC servo motors, with audio amplifiers driving the servo motors.  In turn that also means that the weight of the whole thing can be small allowing it to move really quickly.  DC servos have more torque than steppers so there's room for some real fun 

Instead of being code driven (to make steps) it would be analogue using PID loops to move the axes.  A processor would send tell a DDS what frequencies to generate and then wait for the 'locked' signals to arrive.

Anyone interested in discussing it further, to come up with something to actually build?

I have built CNC milling machines in the past but not pick and place.  I bet there are much more challenging things to consider than just the movement of the business end.


* three parallel strips, two side strips are grounded with a tapered centre strip.  The U shaped piece couples the ground to the tapered piece so the whole assembly is corrected for slight deviations from centre and shielded.  Centre tapered strip is the oscillator capacitor to ground.

[ataradov]

Stepper motors are probably the cheapest things in PnP machines.  What you are inventing called servo motors already, so using servo motors and capacitive feedback is a bit redundant.

And making this work reliably is not an easy task. Start from estimating accuracy of that capacitive position meter. You will get better luck with inductive meters, like all digital calipers do, but it is still reasonably hard to do.

[macboy]

Reading your brainstorm I can't help but to imagine the FirePick Delta.
https://www.youtube.com/playlist?list=PLc9sTVFTHAE7oFt5UNO2h9-KCpKGqn8bk
It uses 3 arms to position the head in 3-dimensional space.

[ebclr]

Stepper motor's really are extreme cheap, step driver are also extremely cheap, you can have a driver / motor system for less than 10 USD, definitely price  ins't the problem here. Your idea is a kind of DC servo motor who will be fair more complicated, the only advantage will be a better dynamic torque and speed and accuracy due to the close loop nature, besides those  point's will be a easy loser for stepper system that are extremely simple and easy to work with.

[daqq]

I've been thinking for ages that a really low cost pick and place machine would be great but up until recently couldn't come up with something that breaks the stepper motor barrier.  I say barrier because a) they aint cheap, b) they need drivers which aint cheap and c) programs to drive them.

a) Define cheap. You can get reasonable stepper motors for less than 15 EUR/pc.
b) Define cheap. You can get drivers very cheaply.
c) At the end of the day everything needs some software - you need to convert the placement positions to some kind of signal anyway. Unless you want to go full analog, which is a somewhat 1950s approach and you could gain nothing by it.

So here's the idea for overcoming the stepper motor barrier.

There is no stepper motor barrier, unless you are aiming for a sub-50 EUR BOM for the whole machine.

[wraper]

Steppers may cause vibration which can cause parts moving when held by the nozzle.

[Chris Mr]

There is no stepper motor barrier, unless you are aiming for a sub-50 EUR BOM for the whole machine.

Blimey, you have 50 Euros to spend on one 

I come from Yorkshire where if you find a fly in your beer you grab it, hold it over the edge of the glass, and say "now spit it out".

A toy motor can be had for a Euro or two and an audio amplifier can just be an opamp and two transistors.  If everything is light, and small, then the motors can be small too.

Another idea is to use a rotary table as one axis, and the other moves across the rotary table - picking up and putting down, so the components would be arranged in a circle around the outer edge.  That would mean that the PCB would be on the table so it would move - not sure if that's ok but it's another idea.

I wasn't suggesting that it would be the fastest in the world, just light and accurate.  Say the PCB area were for a 200mm square as a starting point.  The sort of thing a school could build for fun and kids could make things - you know, to get them interested and excited 

[ataradov]

A toy motor can be had for a Euro or two

Toy motors barely have torque to turn themselves.

Another idea is to use a rotary table as one axis, and the other moves across the rotary table - picking up and putting down, so the components would be arranged in a circle around the outer edge.  That would mean that the PCB would be on the table so it would move - not sure if that's ok but it's another idea.

What are you winning by doing this?

The sort of thing a school could build for fun and kids could make things - you know, to get them interested and excited 

Yes, operating flaky PnP machine is super exiting. I doubt there is a school project that can't be build by hand way faster than you can teach the PnP machine. Especially if you are building only one board.

[Chris Mr]

Toy motors barely have torque to turn themselves.

Rubbish - that's like saying toys don't work

    Another idea is to use a rotary table as one axis, and the other moves across the rotary table - picking up and putting down, so the components would be arranged in a circle around the outer edge.  That would mean that the PCB would be on the table so it would move - not sure if that's ok but it's another idea.

What are you winning by doing this?

It reduces the plan area and number of linear bearings.  A single central bearing with a platter with the PCB on it would be light.  The PCB would be mounted in the centre.  Same capacitor idea, this time the two plates are half circles.

<edit>
I forgot to mention that the rotary table idea means that the business end can then be mounted directly to the base plate meaning that there aren't so many of those expensive moveable cable management bits
<edit\>

The capacitor idea has another neat feature in that it can have an abrupt change to its normal direction - so you can find two reference points without needing limit switches.

[C]

Blind leading the blind.

With steppers or a servo with attached optical encoder you are doing motor position not pick head position. To try to make motor position work the costs of the axis increases and you still have error or limits.
If you know pick head position then axis costs can decrease.

Here I read the main part of idea as a capacitive sensor with direct connection between the axis part that moves and the fixed part. A cheaper axis that has backlash and other errors would have less effect the capacitive sensor.
In addition It would not have to be linear, You just need a function that returns the frequency for position X.
The direct sensing is what is important to lower costs and also allow even more accuracy when positioning of the pick head.

[ataradov]

Rubbish - that's like saying toys don't work

No, it is not. Toys do work, but that's basically all those motors can do.

It reduces the plan area and number of linear bearings.

Yes, but you will need really high angular resolution. There is no way you are achieving that cheaply.

Your are thinking in ideal terms, once you include all imperfections and backlash issues, your design cost will go though the roof. There is a reason why everyone is using classical Cartesian system.

I mean, you can keep on dreaming, it is a good exercise, but I doubt you will find any serious supporters on this.

[Chris Mr]

Yes, but you will need really high angular resolution. There is no way you are achieving that cheaply.

On what basis are you making that comment?  By the way, vernier calipers are capacitive, not inductive as you stated earlier.

https://blog.adafruit.com/2010/11/08/digital-calipers-tear-down/

and that accuracy (+/-0.001") is with tiny plates

You are underestimating what I am suggesting

[Koen]

You could team up with bootstrap : New Pick and Place design ideas and eneuro : Cheap recognition & pick & place & reflow oven solder machine for tiny PCBs.

[Chris Mr]

Koen, thank you for the suggestion - I'll read those threads through.

There is another way I have up my sleeve too.  In 1997 I was involved in making an ultrasonic positioning system which has a patent that runs out soon.

https://worldwide.espacenet.com/publicationDetails/biblio?DB=EPODOC&II=1&ND=3&adjacent=true&locale=en_EP&FT=D&date=20020611&CC=US&NR=6404703B1&KC=B1

That's for a single point to point system but you can use them in parallel, occupying different ultrasonic frequencies, to get more dimensions.

The 3D version (which I have in a box) could resolve 0.001" over >40 Metres!  It has two receivers on each side of a pair of glasses that can work out orientation (polar) as well as X,Y and Z. That thing updated at 1,000 times per second which at the time was quite something - probably still is!

In 1996, when I started that one, doing square roots in a little processor was not possible (for real time work as it took too long) so I had to work out how to do the position calculations without square roots from the lengths and known fix station positions.  How things have changed.  All that can be done in the blink of an eye using nothing more than an arm processor these days with very little analogue.

[C]

 Chris Mr
in addition to the linear cap, you might think of a second cap that has a repeating triangle pattern to assist main cap.
See it serving two functions.
The Max frequency change would let you compute the plate hight.
The frequency change between the two limits should let you add more resolution.

Calibration could be just moving at a almost constant speed and recording the results.

Edit:
A sheet medal shop should be able the cut some linear tapers to test with.

[mikeselectricstuff]

X/Y motion is way down the list of problems. Steppers are cheap and work just fine, assuming you're not aiming to place 01005's on a 1 metre board.
Pick & Place is not one big problem.
It's lots and lots and lots of little ones, any one of which, if not done right,  can reduce yield enough to make the machine basically useless.
The biggest problem that nobody has yet found a cheap yet reliable answer to is tape feeders.

[Chris Mr]

C: (Is that how to pronounce your username or do you prefer just C)

If the capacitor plate is constructed like a sawtooth, but only one long tooth and then two bits at either end, the abrupt change, which is in the wrong direction, is easy to spot and gives you two fixed points of reference.  Moving towards the thin end increases frequency so a sudden drop is odd etc.  The plates don't have to be wedge shaped either, they can be long thin rectangles and just positioned so that one part is out of view of the two ground plates.  If you need more accuracy then just increase the angle of the taper.

Glad mike mentioned the feeder issue; and it'd be good to hear more things to be aware of.

[mikeselectricstuff]

Glad mike mentioned the feeder issue; and it'd be good to hear more things to be aware of.

Unless you've actually used, or tried to build a pick & place you can barely begin to know all the issues involved. Most of them aren't to do with x/y positioning - that's a well-solved problem.

[Chris Mr]

Unless you've actually used, or tried to build a pick & place you can barely begin to know all the issues involved. Most of them aren't to do with x/y positioning - that's a well-solved problem.

Ah, I see.  So the answer is to build something, re-iterate, re-design, re-iterate and so on.  Everything done in a closed room where no-one else can see or offer their input?

The thread does have "brainstorming" in the title - rather than "bashing"

[daqq]

Actually Mikes comments are reasonable - an XY table doing the positioning is the least of your worries here. XYZ movers of various shapes and sizes are becoming dirt cheap thanks to the onslaught of dirt cheap 3D printers. I just checked ebay, the cheapest 3D printer I saw at first glance was ~200 USD/pcs. This included various bits that are not required for a pick and place machine. See: http://www.ebay.com/itm/2016-Upgraded-Full-Quality-High-Precision-Reprap-Prusa-i3-DIY-3d-Printer-/281960334092?hash=item41a625430c:g:XugAAOSwtJZXXrqr&autorefresh=true

Maybe knowing your actual target budget would help.

Yes, you can use DC motors for precise positioning, but you will pay with pricey feedback.

If you want to go really dirt cheap, the cheapest, simplest (in terms of hardware) feedback system is to have no direct feedback from the motors, but rather from a camera that would look at the nozzle and determine the position of the nozzle based on what it sees - first the board could be scanned, compared with the image provided from the gerbers and a positioning "map" be made. Then it should be possible to determine the position of the nozzle based on which part of the board the camera is looking at. This would be obscenely impractical, computationally expensive, but probably reasonably precise I suppose.

[Kjelt]

P&P target markets are discussed numerous times, the result was that there are at least three markets: high end manufacturing, small time pro manufacturing and very small hobbieist market.
You are targeting the ultra cheap hobbieist nothing to spent market, since the other two markets don't care what it costs as long as it is fast and reliable with the emphasis on the latter.

The problem with this cheap target group is indeed as Mike mentions how to reliably and relatively fast sort the components that need to be placed without using feeders.
Now if you find a solution to that problem that would be of great value to the community.
A different cheaper XY placement HW is quite uninteresting since there are already a couple on the internet priced between $500 and $1000 and at least 100 persons have done and thought about this before you. Just my two cents.

[Chris Mr]

Hi daqq,

Maybe knowing your actual target budget would help.

Not sure it would - any sound solution could be scaled.

I have worked in R&D virtually all my life and the essence has always been to limit the scope at the end - to keep as much as possible in flux until you have had time to test, try, find failure points and so on.  Only when you are happy with your "birds nest" do you commit it to finished hardware.

If you want to go really dirt cheap, the cheapest, simplest (in terms of hardware) feedback system is to have no direct feedback from the motors, but rather from a camera that would look at the nozzle and determine the position of the nozzle based on what it sees - first the board could be scanned, compared with the image provided from the gerbers and a positioning "map" be made. Then it should be possible to determine the position of the nozzle based on which part of the board the camera is looking at. This would be obscenely impractical, computationally expensive, but probably reasonably precise I suppose.

Or several cameras, mounted at different heights, one high up to feedback where the PCB is etc.  Computers are getting more and more powerful so vision systems will get more and more useful.  Might even be able to do it on a smartphone (recycle old ones!).  The great thing about cameras being that you can log the error conditions visually - look at them after the event rather than sitting around waiting for the event to happen and hope to see it.

Storage space is also getting cheaper.  I think we are already at the point where the cost of deciding to delete something is more than the cost of the storage, so having a complete log of operations would be a great idea - enter those little pill cameras.

[mikeselectricstuff]

Unless you've actually used, or tried to build a pick & place you can barely begin to know all the issues involved. Most of them aren't to do with x/y positioning - that's a well-solved problem.

Ah, I see.  So the answer is to build something, re-iterate, re-design, re-iterate and so on.  Everything done in a closed room where no-one else can see or offer their input?

The thread does have "brainstorming" in the title - rather than "bashing"

I've spent way too much time explaining to people who think they have "solved" pick/place why they are wrong. Plenty of threads here to look though.
If you can find a cheap, accurate, consistent way of deternmining the position of the head in X/Y to around 0.05mm across the whole bed than maybe that would allow cheaper mechanics.
Picking is at least as hard as placing, more so as parts get smaller.
And don't even start with the idea that you can vision the board to get the placement position - it's covered in solder paste.

[Chris Mr]

I've spent way too much time explaining to people who think they have "solved" pick/place why they are wrong.

and the reason you haven't done something about it?

To put it another way, why take (way too much) time to explain when you could take time to solve given that you have the information in the first place?

And don't even start with the idea that you can vision the board to get the placement position - it's covered in solder paste.

I wasn't, what I was thinking was that a camera above the placement head could see the position of the head; that's all, like a ballpark view.  It can also see where the outline of the board is.  That was all due to not being fussed about where the head is in-between fixed points.

If your experience tells you that things go wrong then they need analysing and improving to get some new experience until they don't go wrong anymore.  That's what brainstorming is all about...

"What about the situation when the jack box spring deflects more than expected due to friction in the scrunion pully" etc.  Ever tried teflon on the scrunion pully bearing?


What is experience worth if you never use it

[mikeselectricstuff]

I've spent way too much time explaining to people who think they have "solved" pick/place why they are wrong.

and the reason you haven't done something about it?

To put it another way, why take (way too much) time to explain when you could take time to solve given that you have the information in the first place?

I don't have the answers, just trying  to save wasted time on hopeless approaches