New Pick and Place design ideas

[rx8pilot]

You are increasingly convincing me that checking the position of tiny components with the down-looking "nozzle camera" before moving to pick them up... is the safe and reliable way to go.  However, note that each time it finds the component with the camera, it detects (and remembers) the error.  After a few picks of a given tiny component the software should be able to tell whether it can continue without pre-checking with the camera, or needs to check every single time.  Frankly, my money is on "check every single time anyway" for all tiny components.  At the margins (perhaps 0805) it will become clear the mechanical precision is sufficient and camera checking is a total waste of time.

The process of picking the part will move it, so using vision on the feeder will not give you what you need to place the part - at least not with the sensitive ones. Again, not all parts are vision candidates while they are in the tape. That is why each part is scanned after it is picked up and stable. The XYT correction is different each time.

SCREW MODERN BUSINESS.  Be a futuristic business or die.

Oh, and most of that millions bucks profit per year is for your/our time!  Expenses?  Near zero.

I am about as unconventional and forward thinking as they come. I have built an entire factory into a 400ft/2 garage that I call 'Factory 400'. I produce more at my my house than I did when i had a partner that hired a dozen people and had 5000ft/2. I am very automation centric as well as process oriented. That is not why I would not go for a PnP business. It's because I have at least the same or better chance of turning my current products into a $million profit. I am already on the market and selling right now, so I have only a trickle of R&D and the business is already in motion - building.

Expenses are not zero unless you are stealing all the materials, tools, and supplies and live on nothing. Let's say we formed a business and it took a year to develop. The costs would be anything but trivial - prototyping a myriad of mechanics, wiring harnesses, PCB's, software, burning things up, starting over, etc. Let's pretend it takes 4 full iterations to get a reliable enough system to release (generous estimate) and then we get to the end of the R&D year and need to do a pilot run of 15 systems - 5 of which are demos. That is a lot of money that is needed up front. Now, pretend we sell the first 10 units for a gross profit of $10k ea - that may cover 1/4 of our expenses if we are lucky - but there will be issues that we have to cover for free on the early adopter units and hope they are nice enough to promote the product. I doubt that forums that cater to hobbyists will be a great sales tool for a $20k assembly tool that is only needed if you are a full fledged business of just the right size and type. Tiny business can't afford it, larger businesses would want to outsource high volume or buy a much more capable machine. A niche within a niche. So maybe, 40-50 machines in (if all is perfect), we break even. That point may happen 2 years into the effort.

I am not risk averse at all. Not even slightly. All I am saying is that I have just recently found some traction with my current efforts and it would take more upside potential to jump ship. If the realistic potential of the new business is that I may make $1mil over 3 years after likely working double time and weekends - I will easily stick with what I have. I am two years into my current product (contracting is almost completely a thing of the past) and the prospects are looking better than a PnP effort from scratch.

It is a fun and challenging topic that has no end. If I had no need to make money, I would actually make one just for fun. Unfortunately, I need to keep my business going for now so can only talk about the ideas. If I had $500k and wanting to start a new and profitable business - I doubt I would choose this.

[bootstrap]

You are increasingly convincing me that checking the position of tiny components with the down-looking "nozzle camera" before moving to pick them up... is the safe and reliable way to go.  However, note that each time it finds the component with the camera, it detects (and remembers) the error.  After a few picks of a given tiny component the software should be able to tell whether it can continue without pre-checking with the camera, or needs to check every single time.  Frankly, my money is on "check every single time anyway" for all tiny components.  At the margins (perhaps 0805) it will become clear the mechanical precision is sufficient and camera checking is a total waste of time.

The process of picking the part will move it, so using vision on the feeder will not give you what you need to place the part - at least not with the sensitive ones. Again, not all parts are vision candidates while they are in the tape. That is why each part is scanned after it is picked up and stable. The XYT correction is different each time.

You were worried about failures to pick up parts (due to lame feeders or related issues), or components that fall off the nozzle because they were never firmly attached to the nozzle.  The purpose of vision at this point is NOT to place the component precisely (or at all).  The entire purpose is to make sure the nozzle tip comes down exactly in the middle of the component, no matter how far the component is from where it should be.

In other words, I'm very much "on board" with your claim we need to pay a lot of attention to feeder issues.  The whole point of this step is to assure every component is picked up at the exact center of the component, so our machine suffers as few pick-up problems as possible.

As a very important side benefit, this also means that some of the feeder issues that could cause reliability problems or extreme expense... can now be ignored.  The components no longer need to be positioned perfectly by the feeder system.  Result: simpler, cheaper feeders.

I'm not saying feeders don't have other potential problems.  Obviously they do.  But this is a big one, and we can absolutely solve this reliably and cheaply with this vision step.  And we only need to perform this extra step on components that need this extra step (presumably just tiny components and maybe a few rare special cases).

PS:  If bringing the nozzle tip down on the exact center of the component is not sufficient to achieve reliable pick-up, you need to explain.  Yes, obviously the tip needs to press on the component with the appropriate force, the vacuum must engage at the right time (and force), and generally do everything smoothly.  But if there are other obscure reasons for unreliable pickup, you need to explain.  Because I can only guess, not having the experience with these problems that you have.

Or if you just mean we can't depend on the component not moving/rotating on the nozzle as the nozzle touches it and picks it up, then that's "no problem", because the next step is to move over the up-looking "component camera" to learn precisely where the component is.

SCREW MODERN BUSINESS.  Be a futuristic business or die.

Oh, and most of that millions bucks profit per year is for your/our time!  Expenses?  Near zero.

I am about as unconventional and forward thinking as they come.  I have built an entire factory into a 400ft/2 garage that I call 'Factory 400'.  I produce more at my my house than I did when i had a partner that hired a dozen people and had 5000ft/2.  I am very automation centric as well as process oriented.  That is not why I would not go for a PnP business.  It's because I have at least the same or better chance of turning my current products into a $million profit.  I am already on the market and selling right now, so I have only a trickle of R&D and the business is already in motion - building.

Expenses are not zero unless you are stealing all the materials, tools, and supplies and live on nothing.  Let's say we formed a business and it took a year to develop.  The costs would be anything but trivial - prototyping a myriad of mechanics, wiring harnesses, PCB's, software, burning things up, starting over, etc.  Let's pretend it takes 4 full iterations to get a reliable enough system to release (generous estimate) and then we get to the end of the R&D year and need to do a pilot run of 15 systems - 5 of which are demos.  That is a lot of money that is needed up front.  Now, pretend we sell the first 10 units for a gross profit of $10k ea - that may cover 1/4 of our expenses if we are lucky - but there will be issues that we have to cover for free on the early adopter units and hope they are nice enough to promote the product.  I doubt that forums that cater to hobbyists will be a great sales tool for a $20k assembly tool that is only needed if you are a full fledged business of just the right size and type.  Tiny business can't afford it, larger businesses would want to outsource high volume or buy a much more capable machine. A niche within a niche. So maybe, 40-50 machines in (if all is perfect), we break even.  That point may happen 2 years into the effort.

I am not risk averse at all.  Not even slightly.  All I am saying is that I have just recently found some traction with my current efforts and it would take more upside potential to jump ship.  If the realistic potential of the new business is that I may make $1mil over 3 years after likely working double time and weekends - I will easily stick with what I have.  I am two years into my current product (contracting is almost completely a thing of the past) and the prospects are looking better than a PnP effort from scratch.

It is a fun and challenging topic that has no end. If I had no need to make money, I would actually make one just for fun.  Unfortunately, I need to keep my business going for now so can only talk about the ideas.  If I had $500k and wanting to start a new and profitable business - I doubt I would choose this.

I'm in a similar position.  Like you, there is ZERO question whatsoever that this machine is FAR from the best business opportunity that I have.  I think we both said before, something about pick-and-place machines is interesting, compelling, fascinating or appealing to us in some way.  That's it.

Oh, plus I need one!  Lucky you already has one.  I mean two.  Or more?

And I don't rule out just making a couple or three for us... assuming we come up with a design that turns us on, and we're convinced is worth the trouble.

As a business opportunity though, there are endless ways to play this... especially if we basically just decide to design and build a couple for ourselves.

From past experience (approaches I've taken before):

#1:  Find some capable engineer who is also fascinated with pick-and-place, but doesn't have all the opportunities we have, and let him run the company from his garage (once we set him up in lean, mean, cheapo-machine fashion).  We do ZERO work from then on, and he sends some percentage of gross revenue (10% to 25% perhaps).  Lots of people would LOVE to have their own business, and don't expect anywhere near the monetary rewards you and I do, but don't know how to do it.  For someone to show up in their life and drop something like this onto their lap is like a dream come true.  Yes, we can't just accept anyone who wants the opportunity, because most people are incapable and/or scam artists and will just destroy everything or stab us in the back.  But... we might be able to find someone suitable.

#2:  What almost always happens to me in the past is this.  I develop a new product.  Sometimes someone in an existing company hears about what I created from someone who has seen me testing my prototype.  They contact me and buy the product to add to their product line.  Or if that doesn't happen, once the prototype is done and working reliably, I take a rare trip out of my cave and demonstrate the product to companies that manufacture and market similar products.  For me, this has almost always generated a sale.  Then the product is totally out of my hands, I walk away with a big check in hand, and go work on my next project (which at that point is about 100,000,000,000,000 times more interesting than the "been there, done that" project I just finished).  I bet that's how your brain works to.

Now I must say, I never charged anywhere near what I should have for products, and that's part of the reason it was totally trivial to sell them.  Fact is, I would usually have a list of 3~6 companies to present to, and never did I reach the final company.  So take the above in the proper context... that would can probably get nice checks, but they won't make us rich.  Just let us retire and live a comfortable but frugal life if we wanted.  Which, actually, I can do now if I don't spend too much money to keep developing new products and technologies!  So for me, this would just create a convenient buffer.

I don't count developing expenses in my computations.  That's just not how I think.  I treat the development process (and expenses) separate from the business.  So the cost of running the business is independent in my mind.

As a matter of fact, of course whatever we would receive from the biz is both paying back our development expenses, then paying back something less than the time we spent was worth (given we are the kind of people we are).

Bottom line:

#1:  The main motivation to brainstorm this issue is... because we find it interesting.  PERIOD.

#2:  The main motivation to create physical prototypes and make them work is... because we want to.  PERIOD.

#3:  The main motivation to generate revenue in one way or other is... because "why not" (we got this far).

So I'm not going to try to convince you this is a great business opportunity FOR US, because IT SUCKS.

It flat out IS NOT a business opportunity.  It is a "side interest" for us, for lack of a better term.

We don't even need to take step #3.  However, if we develop the sucker, and it works great, and costs what we hope, then why not sell it to some company, or set someone worthy up to have the kind of business we can create almost in our sleep, but they can't.  I know you know something that I shouldn't say here, because it will just piss a bunch of people off.  But given our histories and something [perhaps undefinable] about our nature, we can do things that the vast majority can't.  So our motivations are different.  That's just the way it is.  As a business opportunity, this project is something like silly or lame.  But like you, I've had that itch (or little voice in the back of my mind) saying to me "you should build one of those... you can do better and it will be interesting and satisfying".  I think you do to.  And that's the only legit reason we should given the facts of the matter as I see them.

I feel it is way too early to get serious about business.  For the moment, this project is just to scratch an itch, work out a jigsaw puzzle, satisfy our curiosity, enjoy the process of problem solving.

PS:  I'm no multi-millionaire, or even a singular millionaire.  But I'm on the way, have no debt (never have), low expenses, and know how to live both a very comfortable and inexpensive life simultaneously.  Frankly, I'm very frugal about everything... except the quality of my development equipment and tools that let me "do my thing".  My goal is not to become rich or "captain of industry".  My goals are much simpler (follow my interests) and much more grandiose (yet practical with technology I already developed).  So much more grandiose that I shall refrain from starting any chaos here by elaborating on that here.

[Smallsmt]

The expensive machines that cannot place tiny components do not fail because they are moving too fast.  How do I know?  Because they could slow down for problematic/tiny components if that was all they needed to place them precisely.

Interesting but totally wrong you need much better mechanic and high precision nozzle and vision system. A belt machine can't do because need linear measurement system on x an y axis for position feedback. Next point is the vibration problem for a stepper motor system and head swinging because of using belts.

[bootstrap]

The expensive machines that cannot place tiny components do not fail because they are moving too fast.  How do I know?  Because they could slow down for problematic/tiny components if that was all they needed to place them precisely.

Interesting but totally wrong you need much better mechanic and high precision nozzle and vision system.  A belt machine can't do because need linear measurement system on x an y axis for position feedback.  Next point is the vibration problem for a stepper motor system and head swinging because of using belts.

I don't understand how your answer relates to my question.  So I'll ignore that aspect and just comment on the content of your message.

When you say "need much better mechanic", I assume you mean "much better position precision on x,y axes".  If not, please elaborate.

When you say "a high precision nozzle", I have to ask those who pick 0201 and smaller components, what does this mean?  I guess my assumption is, a nozzle with a tiny tip and hole is designed for 0201 components and thereabouts, and would therefore automatically be "high precision".  Otherwise, what would the point be of making a tiny nozzle that doesn't work reliably?  If the bottom line is that some tiny nozzles aren't high enough precision and some are, then we would obviously have to support some existing high precision nozzle (for those tiny sizes at least), or make our own.

Tell me what aspects of the nozzle must be high precision?  It seems obvious to me the tip must be precisely centered on the axis of the nozzle, and the hole be precisely in the center of the tip.  And I assume the bottom must be precisely flat, the ID and OD edges must have no burrs or irregularities, and not be sharp edges (45s).  For the tiny sizes, I'd definitely chose some kind of non-corroding hardened steel (or ceramic/glass) and make them ground instead of turned.

But better for you to tell me what you mean by "high precision" than for me to guess.

Please explain further why a belt machine won't work.  How about a belt machine with linear glass scales?

As for stepper motors, I personally abhor them for a few reasons.  I've been a major fan of DC servos + encoders for ages, though in a few special cases I have designed in steppers.  We won't be able to afford my favorite motors (DC brushed pancake motors (servos)), but plenty of other cheaper DC [servo] motors will work.

I know we have too long a topic here for everyone to read everything, but somewhere far back there is a discussion of how we can make the equivalent of glass scale linear encoders for super-cheap.  That option is still on the table.

If you could describe in MUCH greater detail the various specific failure cases and causes, I would GREATLY appreciate that.  Thanks.

[Smallsmt]

When you say "need much better mechanic", I assume you mean "much better position precision on x,y axes".  Please explain further why a belt machine won't work.  How about a belt machine with linear glass scales?

You need linear guides and a closed loop position system which is working fast!
Servo motors combined with ball screw spindle can do the job without adding a flexible component like belt.

When you say "a high precision nozzle", I have to ask those who pick 0201 and smaller components, what does this mean?  I guess my assumption is, a nozzle with a tiny tip and hole is designed for 0201 components and thereabouts, and would therefore automatically be "high precision".  Otherwise, what would the point be of making a tiny nozzle that doesn't work reliably?  If the bottom line is that some tiny nozzles aren't high enough precision and some are, then we would obviously have to support some existing high precision nozzle (for those tiny sizes at least), or make our own.

The rotation and the height tolerances have to be perfect to place small components. You receive the biggest problem on pickup for the placement you can use the visual system.
Expensive machines use index on A – axis to do an individual calibration for each nozzle and take care you need to restore the position if you use a nozzle changer. The Samsung nozzle we use where build using a ceramic steel combination. Maybe you receive higher precision if the nozzle has no spring element and the system use a force feedback sensor.

I know we have too long a topic here for everyone to read everything, but somewhere far back there is a discussion of how we can make the equivalent of glass scale linear encoders for super-cheap.  That option is still on the table.

But your camera solution need to be very fast!

[bootstrap]

When you say "need much better mechanic", I assume you mean "much better position precision on x,y axes".  Please explain further why a belt machine won't work.  How about a belt machine with linear glass scales

You need linear guides and a closed loop position system which is working fast!  Servo motors combined with ball screw spindle can do the job without adding a flexible component like belt.

Remember the target market and our goals:  precision == YES : speed == NO.  Sure, if we find we don't have to slow the machine down too much, great.  But we are willing to accept 2x to 5x slower speed to meet the other goals.

However, I would add another "goal" here after all the conversations with rx8pilot and others.  While we willingly allow the machine to be slow as necessary (to achieve low price), we pay great attention to operator efficiency.  What does this mean in practice?

For example, in the nominal version of the machine, once the operator starts the job, he need not hang around.  He can go do something else.  Perhaps he will load other feeders and feeder banks for the next job, or something completely unrelated.  But we don't want him to hang around to help the machine.  This may not be an absolute 100% necessary goal (if it adds too much cost), but it is an extremely desirable goal that we shouldn't give up unless we absolutely cannot find a way around it without greatly increasing the price.  However, depending on the final configuration, we might have a "stripped down" model for engineers that simply cannot pay more than the price of the stripped-down model, and are willing to waste hours to help the machine along at certain steps (like maybe there is only one active feeder, which must be manually changed after those components are placed).

The consequence of all these considerations is... the machine can run slow, as long as it can do so without being attended.  The goal of the machine is VERY LOW VOLUME... for assembling 1~5 "prototypes", for assembling a run of 5~20 "demo-PCBs" for alpha and beta testers of your product... for assembling 1~20 specialty products that are part of products you sell once in a blue moon.  This is not intended to be a production machine!  For that, you spend $40K to $500K (depending on volume requirements), not the $10K to $20K we ask.

When you say "a high precision nozzle", I have to ask those who pick 0201 and smaller components, what does this mean?  I guess my assumption is, a nozzle with a tiny tip and hole is designed for 0201 components and thereabouts, and would therefore automatically be "high precision".  Otherwise, what would the point be of making a tiny nozzle that doesn't work reliably?  If the bottom line is that some tiny nozzles aren't high enough precision and some are, then we would obviously have to support some existing high precision nozzle (for those tiny sizes at least), or make our own.

The rotation and the height tolerances have to be perfect to place small components.  You receive the biggest problem on pickup for the placement you can use the visual system.

Expensive machines use index on A – axis to do an individual calibration for each nozzle and take care you need to restore the position if you use a nozzle changer. The Samsung nozzle we use where build using a ceramic steel combination.  Maybe you receive higher precision if the nozzle has no spring element and the system use a force feedback sensor.

I forgot to mention one of my ideas about precision of nozzle-tip location after changing nozzles.  Since by definition the nozzles have a hole in the end of the tip, we can put an LED at the top of the nozzle spindle.  So after the machine (or operator) changes nozzles, it can move over the up-looking "component camera", and that camera can very precisely see the aperture on the nozzle-tip, determine the position precisely, and until the next nozzle change, apply the correction.  Since I don't know about the dozens of pick-and-place machines on the market, I have no idea whether this is a novel idea or not, but should be quite cheap, since we already have the up-looking "component camera".

About vertical position being so critical.  Are you saying the nozzle is not spring loaded along the z-axis?  Are you saying the nozzle tip must just barely touch the top of the 0201 component because it is not spring loaded?  I can see that the spring loading must be extremely low force for tiny components like 0201, otherwise risk causing the component to dislodge or shift.

One other idea I had a long time ago but forgot until now.  If the bottom of the component tape slides along on a piece of metal, the first time the machine goes to pick up a component from a new feeder, it could slowly lower the nozzle tip onto the metal.  The moment the nozzle tip touches the metal, that would complete a circuit, and the software could record the exact position of the bottom of the component tape.  If the height of the top of these components above the bottom of the component tape is known, we now know precisely how far to lower the nozzle to pick up those parts.  Frankly, if that dimension is not part of the spec for the component, we can require the operator measure this for each component and add it to his database.  That only needs to be done once per specific part.  Also, probably ALL 0201, 0402, 0603 or 0805 resistors [from any given manufacturer] are identical height, which will save a lot of time.

Actually, if we want to be extra-elaborate and convenient, we can have a lightly spring loaded metal clip of known thickness on the opposite side of the tape that can be rotated over the component on the tape and rested on top during this "component feeder sync" process.  Then the machine would perform the same action again except this time on top of the metal clip on top of the component and have an exact measure of the component thickness (after subtracting the known clip thickness).  I can think of a couple automatic ways for the machine to accomplish this without the operator moving the clip back and forth, but they don't seem quite as robust and foolproof as I'd like.  Maybe someone else can invent something more clever (or maybe I can with more time and thought).

I know we have too long a topic here for everyone to read everything, but somewhere far back there is a discussion of how we can make the equivalent of glass scale linear encoders for super-cheap.  That option is still on the table.

But your camera solution need to be very fast!

Yes and no.  If it was necessary to continuously keep track of the position to full resolution, to implement a linear encoder with the camera would indeed require insanely if not impossibly fast camera and software.  However, if we assume the machine has cheapo rotary optical encoders on the servo motors (or gag, steppers), that should be sufficient to know where the nozzle-head is within around 0.010" == 0.250mm == 250u with even lame mechanics.

Only when the head stops somewhere important do we need know position to high precision.  And the way the current design works, the machine and algorithms are not at all concerned with the absolute position of the nozzle-head.  Why?  Because it looks at the PCB surface and makes all position decisions based upon fiducials or component pads on the PCB surface.

In other words, what matters is the ability to move the nozzle head slowly to exactly center the nozzle-camera over the component center (or alternatively know exactly what point on the PCB near the component center the nozzle-camera is currently over).  That becomes the "reference point", and the next motion is to move the nozzle-head the exact known distance between "component camera" and "nozzle tip" to center the component over the PCB.  That motion is not very far (probably 50mm to 80mm).  Even this move can be done crudely and quickly at first (based on the less precise rotary encoders), and then getting the exact position (from the camera-driven fake linear encoder) must be done very slowly anyway (compared to normal motion, though still only a fraction of a second presumably).

As I said somewhere before, the ironic fact is this.  The very first PCBs I need to make are "robotics vision system cameras" that are superb for this application (but frankly way, way, way higher performance than we need for this machine).  So if we can't find any perfect cameras, we can adopt something cheap and lower performance for the time being, then substitute our own custom cameras later, once they're available.  A lot of tiny, high-resolution and not too expensive cameras have come out recently, so we have lots to choose from (just look at quad-copters these days and pay attention to the cameras they have on them).  What is the best final choice in cameras... I have no idea yet.

[bootstrap]

Here is a terrible idea that will probably make everyone puke.  But I toss it out there anyway in case it stimulates ideas in others who know factoids or have expertise to replace or circumvent the terrible parts, or else make them work.

At the moment the assumption is this:

#1:  We have an x,y stage like most pick-and-place machines, but it must be inexpensive, so even with all our expertise, we cannot assume it will be precise enough to pick-up and place 0201 discretes and 0.30mm pitch BGAs/QFNs/etc "blind"... where "blind" means "just based upon the inexpensive rotary encoders coupled to the x,y motor shafts and without vision".

#2:  So we plan to leverage inexpensive but high resolution leverage cameras AKA "machine vision" in a few clever ways to achieve the precision we need.  Though not entirely accurate, we basically replace "precision mechanics with linear glass scale encoders on the x,y axes" with "machine vision".

#3:  To place components precisely we stop the "nozzle camera" over the center of the component as best we can based upon our inexpensive rotary encoders and modest mechanics.  Even with lame mechanics and drive system, we should be able to get within 0.010" == 0.250mm of desired positions.

#4:  Once we are [roughly] over the next component on the PCB, the down-looking "nozzle-camera" captures an image of the area of the PCB where the current component will be placed.  The image processing software finds and identifies known features within the field of view... "global fiducials", "local fiducials" component pads (if edges are free of solder paste), etc.  From that information the software precisely determines the position on the PCB that is at the exact center of the down-looking "nozzle-camera" image.

#5:  In front of the nozzle-camera lens are two flat glass windows tilted at angles to faintly reflect ghost images of the "x-rail" and "y-rail" of the pick-and-place machine.  These rails are the fixed x,y rails along which the "nozzle-head" is moved.  On these rails are tiny 0.0005" == 0.012mm aperture slits backlit by LEDs, roughly 50mm apart.  As the "nozzle-head" is moved in x and y, the images of these slits move.  Because the scale of this camera is 0.001" == 0.025mm per pixel at the focus distance (distance to PCB surface), the image processing software can distinguish slit positions to about 0.0005" == 0.012mm precision.

#6:  Once we know the "nozzle-camera" is precisely over the center of the component, we can move the "nozzle-head" any x,y distance we wish within 0.0005" == 0.012mm precision by means of the slit image positions.  Since we know the precise fixed distance between down-looking "nozzle camera" and the "nozzle head", we can now move the "nozzle-head" the exact x,y distance to bring the component center directly over the point on the PCB that the component center must be placed.

#7:  The component can now be placed.

-----

The potential problem that many previous messages discussed is... how can the software be certain it can always find "local fiducials" or "unsmeared component pads" in the field of view of every component (small or fine pitch enough to require precision placement)?  The idea of treating via pads as fiducials was kicked around, but eventually rejected.  The idea of requiring "local fiducials" was proposed and received mostly positive responses.  But the drawback of this requirement is that some existing PCBs (without "local fiducials") cannot reliably be assembled.  This is not a "killer", but is an annoying drawback.

How can this be resolved?  How can all PCBs be supported, with or without "local fiducials"?

One approach is to require solderpaste application to the PCB to be done well enough that solderpaste is not smeared beyond the edges of component pads.  In this case, the software can determine the center and rotational orientation of each component by its component pads before the component is lowered into place.

And so, this is a request for everyone who assembles PCBs to respond below with your estimate of how often this requirement would be met (only worried about tiny discretes (<= 0603) and fine-pitch components (<= 0.80mm pitch) you prepare for assembly.

Note that with more effort the software can probably work reliably if the solder paste extended beyond the edge of some component pads.  As long as the software can find at least 2 pads with clean edges on both sides of the pad in the x direction, and at least 2 pads with clean edges on both sides of the pad in the y direction, the software can infer where the component center is... with some effort.  So a few instances of solder paste hanging over the edge of pads should be tolerable (but require more software effort).

The consequence of this consideration is... the chances both pads of 2 or 3 pad discrete components are sufficiently free of smeared solder paste is much lower than components with many pads.  On the other hand, if a few other "clean" component pads (for discretes or otherwise) can be found within the field, the software can determine position from them.  This is somewhat problematic, however, because as more components are placed, more component pads are covered by components, leaving fewer for this purpose.

To be sure, a LOT of extra software work can greatly improve the reliability of this process.  For example, before component placement begins, the software could scan the PCB, find all component pads with clean sharp edges, and compute whether the default order of component placement will always result in success (sufficient position references available and uncovered for every component).  This will always be the case if application of solder paste is neat and clean (not smeared over the edges of pads), but not necessarily otherwise.  If it finds the nominal order doesn't work, it can attempt to find an order of component placement that works.  If that fails, it can BEEP and tell the operator to clean the PCB and apply solder paste again.

Of course, none of this is required when PCBs have "local fiducials" per the instructions that come with this machine.

-----

Finally we arrive at the terrible idea that may or may not be something we can find ways to make practical.

That is... solder paste is not applied to the PCB before component placement.

Instead, we make the machine apply solder paste to the pads of the component we are about to place, just before we place the component.

The process would work like this:

#1:  We follow the process as described earlier in this message.  When the time comes to find fiducials or component pads to determine where the down-looking "nozzle camera" is over, we always perform this process on the bare (solder-paste free) component pads rather than looking for "local fiducials" or component pads without smeared solder paste.

#2:  After we sync up position of the "nozzle-camera" (and thus "nozzle-head") with the PCB surface (which means, move the "nozzle-camera" precisely over the component center and consider this our <0.00000, 0.00000> reference position for this component, we move a "paste-nozzle" to each component pad lower the tip almost to the pad, then squirt out the appropriate quantity of solder paste onto the pad.  Obviously this "paste-nozzle" is also at known, fixed x,y distance from "nozzle-camera" and nozzle-tip.

#3:  Once the solder paste is applied the down-looking "nozzle camera" can move over the component center again, capture an image, and run a quality control check of the applied solder paste... and reapply to any pad that didn't receive enough paste if the software can be made that smart.

#4:  The component can then be moved to the appropriate position and placed.

This approach has some benefits and some drawbacks.

APPROACH BENEFITS

#1:  No stencil printer need be purchased (one time $$$ savings).
#2:  No stencil need be purchased (once per PCB $$$ savings).
#3:  No solder paste wasted, presumably (once per PCB $$$ savings).
#4:  No need for operator to apply solder paste to PCB (once per PCB time savings).
#5:  Reliable and precise placement of solder paste (if we do our job well).
#6:  Volume of solder paste can be variable for special pads/cases/components.
#7:  No need for any "local fiducials", so machine can reliably assemble all PCBs.
#8:  add more here ... ??

APPROACH DRAWBACKS

#1:  Extra complexity of machine.
#2:  Extra complexity of development (could be substantial).
#3:  Higher price of machine --- must add less to price than quality stencil printer.
#4:  add more here ...  ??

-----

Well, that's my terrible idea.

If someone here just finished 5 years of experiments to get to the point he can reliably apply solder paste with some not-to-expensive nozzle or gizmo he developed... well... maybe this idea looks a bit less terrible.

Otherwise, this whole "apply solder paste" activity could be a huge bag of worms!  I'm pretty good with mechanics, optics, electronics, software and lots of other things, but I have very little experience "squirting" or applying pastes to anything.  Or even liquids for that matter.  So... I'd be starting from ground zero.  I read some forum messages in the past 3 or 4 years from people playing with solder paste application with nozzles and such, but they were mostly tales of woe.  Haven't read much in a year or more though, so maybe someone has made progress.  Or maybe there is even an inexpensive product on the market to eject solder paste that we could integrate into the machine.  A lot can happen in a year or two in this industry, but... Murphy's Law probably applies to our wish there is such an inexpensive device available that we can just adopt.  If anyone knows better, please yell loudly and soon!

-----

As always, comments welcome, and ideas, suggestions and brainstorming especially welcome.  No need for excessive negativity, since that's already presumed in advance on this idea!

Oh, and everyone with first-hand experience assembling PCBs, do answer that question above... about what percentage of component pads with solder paste applied still have clean edges.

[mikeselectricstuff]

Solder paste dispensing  and 0201's - never gonna work.

[bootstrap]

Solder paste dispensing  and 0201's - never gonna work.

Just because, huh?

You can explain the reasons you believe that?  Proprietary information?

[mikeselectricstuff]

Dispensing systems are notoriously hard to get to work reliably, even for larger geometries. Issues like surface  tension, adhesion, viscosity dependence on temperature, paste uniformity all affect the result.
The size of nozzle you'd need for fine pitch QFPs would be too small to get any reasonable flow rate due to the metal particles in the paste.
If it was viable for fine pitch, Mydata wouldn't have any market for their very expensive paste-jet printers.

[bootstrap]

Dispensing systems are notoriously hard to get to work reliably, even for larger geometries.  Issues like surface  tension, adhesion, viscosity dependence on temperature, paste uniformity all affect the result.  The size of nozzle you'd need for fine pitch QFPs would be too small to get any reasonable flow rate due to the metal particles in the paste.  If it was viable for fine pitch, Mydata wouldn't have any market for their very expensive paste-jet printers.

I guess 0.20mm is too big but a little?  But this seems to be just "some guy" doing this six years ago.

-----

I know it's easier to talk than do, but maybe it is possible to flash-heat just enough solder paste just inside the end of an appropriate size needle to vaporize the flux and eject it out and onto the PCB (carrying the metal in it).  I guess that's heading in the direction of the MyData, but that machine is extremely fast, probably has oodles of heads, and is actually shooting out dots of solder paste before it gets to the pad, and computing how far it will travel in order to hit the pad.  That's so far beyond what I'm talking about in sophistication, it is dangerous to pretend they are the same.

Found the link here:  https://forums.adafruit.com/viewtopic.php?f=42&t=16273

Not sure, but maybe that would work for 0.50mm pitch but not 0.30mm - however that's just a guess.

Oh, I see you posted on that page!  How did that research go?  Very bad, I have to assume!

[rx8pilot]

Once you see the price and consider the process complications - you might realize that stencils are the best option. If you were making huge numbers of very time sensitive one-off, it may make sense. Stencils are very cheap and very reliable. Very few practical use cases for fine pitch solder jet dispensing IMHO.

[bootstrap]

Once you see the price and consider the process complications - you might realize that stencils are the best option. If you were making huge numbers of very time sensitive one-off, it may make sense. Stencils are very cheap and very reliable. Very few practical use cases for fine pitch solder jet dispensing IMHO.

Can you answer the question in my long post?  What's the statistics on how many pads have "clean edges" (not overflowed by solder paste)?

BTW, you say "stencils are very reliable", but so many people I talked to who actually have pick-and-place machines and actually assemble PCBs say... most of their bad boards are from bad application of solder paste with stencils.  And they say that is the biggest hassle of the whole process too.

I'm not contradicting anyone here.  I just don't give up to fast and easy!

PS:  Anyone know whether solder paste can be heated up some (for a few seconds) to make it less viscous without having the flux and solder balls separate?  Then maybe application isn't such a hassle.

Found two "close but no cigar" units:  fisnar: PDV-1000 and RV5000DP
http://www.fisnar.com/products/positive-displacement-valves/pdv1000-2

Unfortunately, smallest dot size is 0.020" == 0.50mm (2~3x too big for 0201 or 0.50mm pitch).

[rx8pilot]

BTW, you say "stencils are very reliable", but so many people I talked to who actually have pick-and-place machines and actually assemble PCBs say... most of their bad boards are from bad application of solder paste with stencils.  And they say that is the biggest hassle of the whole process too.

My first efforts with stencils did not go well. So I did some reading and experiments and discovered that printing solder paste is not a freebie. The design of the stencil is critical. The quality of the apertures. Then you have the technique mixed with the actual paste chosen. It takes a few bad prints to get it right, but when you do - it is very easy and very fast and very cheap. Just do not underestimate the learning curve if you are doing fine pitch printing. There is very little room for error but like riding a bike - it seems nearly impossible at first and then you get it for life.

I have no doubt that jet printing will work just fine, but the economics don't seem to add up. There is no practical benefit to doing a jet print/place - jet print/place scheme that I can think of. Print it, place it, bake it. What is faster or more economical? If economy is not the goal, jet printing and a really nice in-house PCB fab could save a bit of time, but at a wildly huge cost. The reality today is that PCB's and stencils are super cheap and fast. So much so, that the viable market for stencilless printing is pretty small. Maybe an assembly house that specializes in prototype assembly? Not sure.

[bootstrap]

BTW, you say "stencils are very reliable", but so many people I talked to who actually have pick-and-place machines and actually assemble PCBs say... most of their bad boards are from bad application of solder paste with stencils.  And they say that is the biggest hassle of the whole process too.

My first efforts with stencils did not go well.  So I did some reading and experiments and discovered that printing solder paste is not a freebie. The design of the stencil is critical. The quality of the apertures. Then you have the technique mixed with the actual paste chosen. It takes a few bad prints to get it right, but when you do - it is very easy and very fast and very cheap. Just do not underestimate the learning curve if you are doing fine pitch printing. There is very little room for error but like riding a bike - it seems nearly impossible at first and then you get it for life.

I have no doubt that jet printing will work just fine, but the economics don't seem to add up. There is no practical benefit to doing a jet print/place - jet print/place scheme that I can think of. Print it, place it, bake it. What is faster or more economical? If economy is not the goal, jet printing and a really nice in-house PCB fab could save a bit of time, but at a wildly huge cost. The reality today is that PCB's and stencils are super cheap and fast. So much so, that the viable market for stencilless printing is pretty small. Maybe an assembly house that specializes in prototype assembly? Not sure.

Yeah, this idea is a long shot.  But, I consider long shots until I realize they won't work or are impractical.   

Also, the bottom line is that all this hassle only solves one "shortcoming" of the proposed design... the need for "local fiducials" (or take your chances the PCB design doesn't have enough reference marks of other kinds for the software to determine position precisely).  Seems like nobody thought requiring "local fiducials" was a killer requirement, and I tend to agree with that.  So yeah, this probably is a wild goose chase.

Looks like Norton has a DV-01 syringe valve that can lay down dots of fine solder paste, but unfortunately no smaller than 0.20mm.  That is just a tad too big for 0201 and maybe adequate for 0.50mm pitch, but nothing smaller (and 0.30mm is my "if possible" goal).

My guess is, someone will release a cheap version of the spider or scorpion machine in 5 years.  Or if it comes from China, maybe 2 or 3 years?

[forrestc]

In relation to dispensing solder paste:

Figuring out how to do this reliably and consistently and inexpensively would be FAR more marketable than any P&P you could come up with.

I have a high end paste dispenser on my P&P (time and pressure).   I use it for the first run prototypes and for production of lower volume products.  Generally, every product I have has had the first hundred or so manufactured using it.  (If you're curious, the brand I have is Martin.  I have one version prior to the current products).

It's fine for larger pitch components.  Once you get down to 0.5mm TQFP's and similar, you're entering into the realm of it doesn't really work all that well (although it's good enough if you're ok with occasional/frequent rework).  It definitely wouldn't come anywhere close to six sigma at 0.5mm.

[bootstrap]

In relation to dispensing solder paste:

Figuring out how to do this reliably and consistently and inexpensively would be FAR more marketable than any P&P you could come up with.

I have a high end paste dispenser on my P&P (time and pressure).   I use it for the first run prototypes and for production of lower volume products.  Generally, every product I have has had the first hundred or so manufactured using it.  (If you're curious, the brand I have is Martin.  I have one version prior to the current products).

It's fine for larger pitch components.  Once you get down to 0.5mm TQFP's and similar, you're entering into the realm of it doesn't really work all that well (although it's good enough if you're ok with occasional/frequent rework).  It definitely wouldn't come anywhere close to six sigma at 0.5mm.

Is this the unit you mean?

http://www.martin-smt.de/en/dispensing/products/dotliner.html

Amazingly, they claim the dot can be as small as 0.10mm, which should be adequate for 0.50mm pitch, 0.40 pitch, and maybe even 0.30mm pitch if true.  The only negative is, they don't seem to sell just the valve.

Unless...   Maybe this is a nozzle: <HT00.8008>  ??

[C]

I am reading optic camera, but not seeing some uses it could help with.

PCB set on a conveyor belt would let the system be more useful.

One problem is where the PCB is and it's rotation.  Optics handles this but a Pin that could be lowered to catch the PCB board edge would let the board be rotated on the belt for better placement of parts.

By using the camera more you could then feed a blank PCB with out paste and shift to a different PCB by scanning the blank and comparing stored images of boards. This could also check that the place commands matches the PCB.
A new blank PCB could then start the process of creating the parts place files if one did not exist.

A scan of a PCB with paste before placing parts could check for proper paste.
A scan of a PCB after placing parts could check for proper placement.
An image of the PCB could get enough copper detail through the solder mask to allow placing of parts.
 

[mikeselectricstuff]

Amazingly, they claim the dot can be as small as 0.10mm, which should be adequate for 0.50mm pitch, 0.40 pitch, and maybe even 0.30mm pitch if true.

"can be as small as 0.1mm" doesn't mean " will reliably produce 0.1mmm dots every time".
Manufacturers will always quote the best possible performance in the ideal circumstances, acheiving it in practice is another matter.

[rx8pilot]

Manufacturers will always quote the best possible performance in the ideal circumstances, acheiving it in practice is another matter.

Like the placement rate of a P&P system.

[bootstrap]

I am reading optic camera, but not seeing some uses it could help with.

PCB set on a conveyor belt would let the system be more useful.

One problem is where the PCB is and it's rotation.  Optics handles this but a Pin that could be lowered to catch the PCB board edge would let the board be rotated on the belt for better placement of parts.

By using the camera more you could then feed a blank PCB with out paste and shift to a different PCB by scanning the blank and comparing stored images of boards. This could also check that the place commands matches the PCB.
A new blank PCB could then start the process of creating the parts place files if one did not exist.

A scan of a PCB with paste before placing parts could check for proper paste.
A scan of a PCB after placing parts could check for proper placement.
An image of the PCB could get enough copper detail through the solder mask to allow placing of parts.

We can't have a conveyor belt, because we need to support PCBs with components in both sides.  But we can support rails like in the neoden4.  They are quite "cool" as far as I'm concerned, and I can see conveniences (like being able to queue up several PCBs, then leave and let the machine assemble them all).  Depending on cost, this could be standard or option.

I don't see how rotating the PCB on the belt is very helpful.  It is convenient for PCB surface axes to align with machine axes (via <fiducial inspection, tweak, loop> iteration), but that's not necessary (position and coordinate adjustment is typically done in software and works fine).

The only version of "rotate the PCB" that I consider useful is... if the PCB can be rotated 360-degrees (like on a rotary table).  In that case, rotating the PCB can replace rotating the nozzle (or they can complement each other).  On a machine where x,y motion was implement by moving the PCB, putting rotation under the PCB can work, but is more expensive and more complex software than nozzle rotation.

Yes, we have a lot of messages in this topic, and those camera ideas you mention are part of the current design plan mentioned previously.  As much as possible has been pushed off onto the camera and vision software in order to be able to create an inexpensive machine that can pick-up and place tiny and fine-pitch components very precisely.

For various processes, it is good to have one image of one bare "reference" PCB, an image of each PCB with solder paste applied, and image of the assembled PCB.  As you note, much information can be gleaned by intelligent software, and many errors in solder paste application and assembly can be found by automatic software.  Plus, those images can be inspected by human eyes for double checks when desired.

[mikeselectricstuff]

We can't have a conveyor belt, because we need to support PCBs with components in both sides.  But we can support rails like in the neoden4.  They are quite "cool" as far as I'm concerned, and I can see conveniences (like being able to queue up several PCBs, then leave and let the machine assemble them all).

And assemble longer boards than the placement area

[rx8pilot]

I have a PCB conveyor and nearly all of my boards are double sided. I love it and would be bummed if the machine did not have it.

[forrestc]

[Is this the unit you mean?

http://www.martin-smt.de/en/dispensing/products/dotliner.html

No, I have a clever dispense 05.

The main problems with this include:  The variabilty of the paste at different temperatures (partially solved with the heated nozzle).  The variability of the paste as it is worked via air (each squirt of air bangs on the tube of paste, causing rheology effects).  Paste nozzle clogging and/or accumulation of paste crud on the nozzle.  And board-to-board variations (pad thickness, board flatness, etc).

On a typical 0.5mm TQFP we get 1-2 bridges when dispensing, more if things aren't quite adjusted well.   Typically zero with stencil printing.   A worse defect are dots being skipped altogether (or mostly skipped), as those are hard to find and/or fail in the field.

[ServoKit]

Working on a dispenser myself, I've been reading up on the topic and also stumbled upon the Martin devices. Dave made a video about them some years ago:



In the comment thread someone mentions that they are in the $30K range... I thought about using an auger for controlling the flow (see attached for a first attempt)

Regards, Axel