Pick and Place machine brainstorm

[ebclr]

If you really want's to go to on the closed loop dc motor approach here a IC that will make your live easier

http://www.ti.com/product/fdc2212-q1

But will never surpass stepper motor open loop in simplicity and cost

[mikeselectricstuff]

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. 

What use is that - you need an accurate posiiton. Ballpart isn;t useful

It can also see where the outline of the board is. 

How is that useful ? You need the fiducial positions.

[Brumby]

This discussion seems to have two rather distinct views on the subject.

One comes from the conceptual perspective, floating a 'new' idea, backed up by a few techniques that have shown qualitative value.

The other comes from the practical experience of actually making a real-world mechanism perform.

The chasm, nay ABYSS, between these two is nothing short of galactic!


Just because you have an idea for motion and another for measurement, does not mean you will be able to produce a mechanism that can perform in the real world.  It doesn't mean you won't, either.

The big difference comes from all that stuff that comes under the heading of "product development" - something which is completely absent from the "concept view" camp.  It's in this part of the process that issues like backlash, torque, overshoot, acceleration and a hundred other parameters are thrashed out ... and where limits of components are discovered and alternatives are examined.  If an existing technology cannot achieve the necessary objectives, then the success will rest on the development of new technology that can - if that is at all possible.

This is where the "practical experience" camp has weighed in.  There are some practical issues that the "concept view" camp have not even raised, but will have to address and resolve.  Ones that, in many cases, members of the "practical experience" camp have already been burned and learned.


Until the "concept view" camp have actually ventured down the path of a practical implementation, opinion between the two camps will not converge.


The "concept view" camp has the optimism - and the "practical experience" camp has reservations, if not doubt.  Who will win out will never be settled until the new idea has been put through the development wringer.  The big question is ... will that happen?

[Chris Mr]

ebclr, thank you for the suggestion, unfortunately that device is limited to under 14k samples per second.  I was thinking of using a cascode (great isolation) oscillator where the capacitor is in a tank (with an inductor) - oscillating say a few MHz to give very fast update rate.  That would go into a psd (4046) whose other input is the DDS (which are normally 32bits, 28 bit resolution go to 0.05Hz) to a PID and then audio amplifier.

Suppose you go for 1/2 octave change from 7.5-10MHz, even with a 28 bit DDS, you get 2,500,000 / 0.05 = 50,000,000 gradations of 200mm would be 4 nano meters - well into detecting the imperfections of the edge of the tapered capacitor.  Clearly the range has to be more than 200mm (that's just the size of the board) but the figures tell you that it's well within reason.

For a turntable type axis you only need half the distance as the business end only needs to go (precisely) through the middle.  You can turn the turntable to get to the other side.

What use is that - you need an accurate posiiton. Ballpart isn;t useful

It is if you want 'belt and braces' - "My position system is no-where near where my visual feed back says it should be - yipes! now would be a good time to stop" - before it impales itself into the opposite wall 

How is that useful ? You need the fiducial positions.

If someone removes the board, or something lands on it, or someone moves their arm nearby - belt, braces & safety pin(s)



I agree Brumby,

Maybe one issue is lack of information about my character.  I am not Meredith Perry, I don't go blindly into things and think I can do it just because I think I can.  I am more cautious than that.  I am confident about the position measurement because I have used it many times.  One version was used at Leeds university to measure the amount piezoelectric devices move when stimulated - that could resolve well under a 100nM.

I have built CNC machines that are rigid - none of this gantry stuff, real "built in a box" devices.  I am fully aware of backlash which is why a servo system offers so many advantages.

Somehow text doesn't convey the necessary information - but I am trying  (very trying  )

[TheSteve]

I would suggest using/operating a few different pick and place machines before trying to design one. It's all the little things that end up being a pain.
Like it or not everything Mike has said is bang on.

[C]

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.

How is that useful ? You need the fiducial positions.

Using a image of board with solder paste and a image of bare board it should not be too hard to get a mask of solder paste locations.
This results in two outputs that could be used separately or together.
1. Does image of board less masked area have enough information remaining to place parts?
2. Does the solder paste have enough information to place parts?

So you know for a fact that #2 above can not be used as extra fiducial positions in an optical positioning system?

This is a huge change from old school to using machine learning to do the job..

I see cheap light weight optics as a foundation to other changes.
These optics could make a huge difference in the feeder area.
 

[Chris Mr]

This is a huge change from old school to using machine learning to do the job..

I see cheap light weight optics as a foundation to other changes.
These optics could make a huge difference in the feeder area.

Couldn't agree more.

Seeing the deepmind agent play space invaders & finding the solution to breakout got me going - and it only took a few hundred attempts to get super human.

Imagine giving the machine a few different pick tools to try on different components - it would very soon work out which was the best for any given component.  That would lead to making better pick tools and so on.

You would imagine pick and place machine learning being for one machine but what if you network many machines that are attempting the same task !

[Koen]

So, what's your plan ? What are you going to work on first ?

[mikeselectricstuff]

Imagine giving the machine a few different pick tools to try on different components - it would very soon work out which was the best for any given component.  That would lead to making better pick tools and so on.

[Chris Mr]

Hi Koen,

I'm still wondering what the best layout is for it getting access to components and board.  I really like the idea of a ring around the outside of the board area that can rotate 360 degrees with a gantry fixed to it (or even two orthoganal grantries) that goes through the centre of rotation (so it can reach any position on the central area).

But there are many possibilities..

[Chris Mr]

Good to see mike is having a good old think about it.  Let's all join in

 

[Kjelt]

If you even had half the P&P experience Mike has you would be very 

[Chris Mr]

Hi Kjelt,

In machine learning terms what you are talking about is "grandmothered" - lots of experience that only allows one way thinking.

This is what new generations are for.  They don't look at the problem the same way as they don't have the experience - so they come up with different solutions; or they make a mess of it; either.

But they do something

[Kjelt]

They are designing these kind of machines for over 40 years, it is as with cars, the biggest innovation of the last decade is change the petrol engine for an electric engine, but there are still 4 wheels and the elementary principle has not changed since the t-ford.
If you want to innovate a P&P you will probably have to change other "standards" that are fixed and blocking in the industry such as the packaging of the components, or the reflow process it self.
So yes there are a lot of SciFi innovations possible such as floating transport carriages that place the individual components and then weld it to the pcb and hundreds working simultaneously, but given the XY placing problem and camera usage therefore I doubt there is a more optimal solution, still surprise us but patent it first 

[Chris Mr]

Hi Kjelt,

The subject of patents is probably another 10 threads (underestimate).  I have had lots of experience of them so am probably grandmothered!

It would seem,  from my experience, that patents are a tool used by large companies against either small companies or other large companies.  It isn't the cost of getting them that's the barrier (even though that is considerable) it's the cost of testing the patent in court - hence why it is used by large companies.  Owning a patent doesn't really have much value; once you've been to court, and won, it has a lot of value.

Large companies tend to be slower to change - so you can counter getting a patent by changing quickly.  Put another way, if by the time someone copies what you have done you are in front of them, you win 

[Kjelt]

Yes lets not go into patents indeed. 

[Koen]

Guys, I have a brilliant idea. Hear me out. Floor space is expensive. Vertical space is cheap. Industry needs more space-efficiency. Let's reinvent the reel. Instead of a small center hole, let's make the hole a meter wide. In this space, let's have a pick and place : a gantry surrounded by a 2 meters stack of horizontal reels. The gantry rotates, its base moves up and down. The PCB is introduced on top. All the components of reel 1 are placed. The base plate lowers itself one level. All components of reel 2 are placed, boom it goes down to reel 12. After all useful reels are done, the PCB continues a few steps lower into a warm bed of vapour and reflows. Then it shoots back to the top to be packed. It naturally warms going down to activate the flux and naturally cools when going up, brilliant. I've assembled a quick proof of concept :

[Chris Mr]

proper brainstorming material there

By removing the vapour phase stage and make it separate, you could then process a board at every level - in parallel, think of the throughput of that lot.  Obviously you would have to have components like 100n on more than one level, else it would be down to how fast you process them, but so what!

A ring of reflow ovens pointing out from the centre to take the boards away.

It is interesting to think that whilst components are delivered on reels, with nicely spaced holes etc, ready to go on machines, there is no reason why they couldn't go onto "bespoke processing jigs" (AKA reel with Meter wide hole in it or whatever) first.  Of course this is what a feeder is - a bespoke processing jig - but it could be completely different.



I remember doing brainstorming training.  We were tasked with designing something that's impossible to make.  After about 20 minutes of heated discussion, and one member of the team saying nothing, we came up with a box with 3 dials and one button.  One dial was "where you want to be", another was "when you want to be there" and the last was "how big you would be when you got there" - the button was to make it happen.

We were all pretty pleased that it would indeed be impossible to make.

Then the guy who had remained quiet said "and it all has to run from a PP3 battery for a year" 

[macboy]

Consider that a hard drive head is positioned using a linear (non stepping) motor, to a precision orders of magnitude higher than necessary for a P&P, and it hits that mark within single-digit milliseconds.

[mikeselectricstuff]

Consider that a hard drive head is positioned using a linear (non stepping) motor, to a precision orders of magnitude higher than necessary for a P&P, and it hits that mark within single-digit milliseconds.

But it uses closed-loop servo feeedback from data on the disk platter, and operates over a much smaller distance.

[Chris Mr]

Floor space is expensive. Vertical space is cheap.

Has Koen hit the nail on the head - just point the hammer at a different nail?

Time taken from getting the business end from picking up the component to placing it will always be affected by the distance traveled, the further the distance the longer it takes.  The plan area of the board can't be changed (apart from by redesign) but the way the components are presented can.

Ridiculous as it sounds, can the components be removed from the reels / waffles etc and then stacked vertically - the expensive floor space (that takes time for the business end to traverse) could be reduced by many times if the parts are right up to the edge of the board?

[mikeselectricstuff]

Floor space is expensive. Vertical space is cheap.

Has Koen hit the nail on the head - just point the hammer at a different nail?

Time taken from getting the business end from picking up the component to placing it will always be affected by the distance traveled, the further the distance the longer it takes.  The plan area of the board can't be changed (apart from by redesign) but the way the components are presented can.

Ridiculous as it sounds, can the components be removed from the reels / waffles etc and then stacked vertically - the expensive floor space (that takes time for the business end to traverse) could be reduced by many times if the parts are right up to the edge of the board?

Most boards have a fairly uneven mix of parts, so you get close to this by simply putting the most-used parts in the closest feeders.

For boards with a very large number of a particular part, e.g. LED strips, mounting a feeder on the head would make some sense - I've seen a video of what AFAIR is a homebrew machine that does this - can't immediately find it.
 
De-taping parts ahead of placement isn't very sensible for a number of reasons, including that you don't necessarily know exactly how many will be used, due to mis-picks/vision fails etc.
Also handling of loose parts would be difficult - getting hard ceramic parts of varying sizes with sharp edges to move reliably is going to be even harder than making tape feeders.

I have occasionally wondered about something that resembles a PEZ dispenser for chip parts.... 

Here's one impressive system using non-taped parts

[Koen]

This might be the video you were looking for Mike :

[mikeselectricstuff]

This might be the video you were looking for Mike :

Yes that's the one.

[rx8pilot]

This is where the "practical experience" camp has weighed in.  There are some practical issues that the "concept view" camp have not even raised, but will have to address and resolve.  Ones that, in many cases, members of the "practical experience" camp have already been burned and learned.

Yes.

My first adventure in precision motion control was for an advanced stereo 3D camera system. I designed it conceptually without any real experience. It did not work. Not even close. By the time I dealt with all the real world issues, it was vastly more complicated and expensive and the capability was about half of what was originally dreamed.

My view of P&P before I got one was very Utopian and simplified. One I purchased one and started making PCB's, the practical reality smacked me like a 10-ton truck. Designing a P&P machine with no previous experience will lead you toward a very expensive and time consuming learning curve. The most important thing to remember is that P&P exists to solve a business problem - not a hobby problem. if you need to assemble 3 PCB's, you get a microscope and a vacuum pickup. If you are making 100's of PCB's - they are probably not just for fun. In the context of business, a P&P needs to be very solid and reliable to have any value. Speed and many other whiz bang marketing features are useless if the machine mis-picks just one part on every cycle.

The established methods have been established after massive amounts of experimenting and practical every-day lessons. At the end of the day, it is a fools errand to make a cheap P&P. The 'cheapness' gets VERY expensive when it does not get the job done. The goal of P&P should be to deliver properly assembled PCB's every time at a tolerably fast speed. Making the first requirement of a new design -'CHEAP' ensures failure. When that requirement is combined with a design group that has never used or designed anything like it - makes a successful outcome not likely.

I would look at existing successful designs - big-small-expensive-cheap. You really cannot simulate experience, so you need to listen very closely to those that do this all the time. Some make PCB's 24hrs/day while some (like me) only make a handful of PCB's per month. Some make the same board over and over while some are constantly changing the setup.

The time and cost of P&P is NOT the motion control system. It is the feeders, the operator, software, change-over speed, reliability, and other things. A novel way of moving and detecting position will not be much of an added value in the context of a full assembly system.