I have no doubt your post contains much wisdom and experience. However, I also see some statements that are wrong.
For example, you talk about expansion and contraction of the pick-and-place machine due to temperature changes. Of course that is true, because most metals have high co-efficient of thermal expansion. However, the encoder scales are typically Silicon Dioxide (quartz) or Titanium Dioxide (called ULE quartz), and those materials have very close to ZERO thermal expansion coefficients. I know this stuff because I used to design telescope optical systems that had to retain their shape within 100 to 1000 angstroms during entire nights observing outdoors while temperature is constantly changing. And I've designed products with linear and rotary optical encoders too, so this is familiar territory for me.
So the position measurements of the PNP are based upon the quartz scale linear encoders, not the metal parts of the PNP machine.
But that's not all. Anyone who is not stupid puts fiducial marks on their PCBs, especially if they contain tiny components and/or fine-pitch contacts. When the cameras center over those fiducials, the machine knows exactly where those positions on the PCB are... in real time. And the software can re-check the location of those fiducials as many times as it wants during the assembly process. Frankly, a smart pick-and-place machine can chose just about any tiny through-hole on the PCB as a stand-in fiducial mark, so even fiducial-free PCBs should be capable of precise positioning, unless the software in the pick-and-place machine is seriously stupid.
These two facts, and other common techniques, make precision much better and easier than your posts imply.
I also note your comment about the Z-axis precision... about errors in PCB height. Any machine that doesn't have at least 2mm or 3mm of "Z-axis resiliency" is stupid! What I mean is, the placement tip should have at least 2mm to 3mm of spring-loaded or air-pressure-loaded travel. And the machine should detect the first 1mm of tip travel. Then the tip can be removed (with a simultaneous "puff" of air-pressure to assure the part does not cling to the tip).
If current machines really are as stupid and crude as you imply... they should be tossed in the dumpster! And furthermore, if they are that stupid, well, now I understand why you have so many troubles!
BTW, I sent my information to http://www.4pcb.com
and they refuse to assemble PCBs unless they manufacture the PCB too.
Before I continue, understand this. I don't complain about prices. People are free to charge whatever they want. What I complained about was that company giving me a quote, then when I asked them to build the PCBs they quoted, telling me I had to pay 10 times more than their quote. That is FRAUD. And in my case, that completely destroyed my budget calculations (to make my prototypes, not production runs). I don't have unlimited budgets like many companies. That's why I got quotes in the first place!
I believe everything you say about your personal experience. However, I cannot assume that everyone else, with every other set of equipment will have the same experiences. I would hope that new equipment does better than some old piece of junk equipment that people abused for years. This is another huge variable! Some people take care of their equipment, and run it slower to reduce wear and tear, and maybe that would be good equipment to buy used. But other people take no care of their equipment, they abuse it endlessly, and when they sell it used, it is all screwed up and nothing but trouble for the poor fellow who purchased it from them.
I would also hope that new equipment is better. I would hope that all companies learn smarter ways to build their equipment every year... by observing problems in their old equipment, by observing how other companies build their equipment, by collecting statistics on where errors arise (and then focus attention to improve those areas).
You have talked a lot about errors in the assembly process. Yet the manufacturers love to talk about "maximum speed". From my point of view, I would prefer to run the machine at 1/10 speed to increase precision, reduce assembly errors, and make the machine last longer. That won't work for a company that runs all day assembling PCBs, because they'd have to purchase 5 or 10 times as many pick-and-place machines.
Think about how much time is required to inspect PCBs, find errors, rework PCBs to fix errors, debug PCBs that are not working (if that is even possible, which it isn't unless the PCB is your own so you know how it works). Frankly, the kind of people who are smart enough to debug fancy electronic circuits cost so much that it is almost always cheaper to throw away PCBs when they don't work.
So everything you say makes me want to be careful at every step (look carefully at the solder paste on the PCB before you place in the pick-and-place machine), slow down the pick-and-place machine, make sure the reflow oven is optimal, etc. My plan was to have a multiple zone oven, for exactly the reasons you stated... to improve reliability.
You had to throw away your first 5 pick-and-place machines. My guess is, they were used machines. Of course it is possible to get lucky and buy a machine that was very well designed, very well taken care of, and be happy with your choice. I am too afraid to attempt that. I will only buy brand new, up-to-date machines. That approach has treated me well in the past, and so I will continue the practice. Maybe I'm wrong, but my experience tells me to take this approach.
$200,000 for a "feeder box for 0201 components"? What the hell is a "feeder box"?
As I understand this, the MYDATA machines are all-day every-day production machines, not relatively slow machines designed to build prototypes quantities. Which is probably why that machine is 5 times as expensive as the ddmnovastar machine I'm looking at. If we include feeders, it appears like the mydata is $250K and the ddmnovastar is about $40K~$45K. On the other hand, apparently you have about 10 times as many feeders as I plan to have!!! So maybe the entire difference in cost is feeders, not the basic machine! Hahaha.
Though I think you exaggerate some issues, I take your comments seriously. I am thinking about how to approach this situation. Tell me what you think of this approach. What if I approach the ddmnovastar pick-and-place company where I was planning to get ALL my equipment (automatic stencil printer, pick-and-place machine, 3-zone reflow oven) and offer them the following deal. I will buy your machines IF... we run 25 of my ice_quad PCBs and 50 of my ice_eye PCBs and all components are placed correctly and all solder joints are good. In other words, no assembly errors.
What do you think would happen? Do you think they would refuse? What if they accept? Am I asking them for too much? Should I allow 1 or 2 faults out of those 75 PCBs to be fair? I'm serious. If these companies are BIG FAT LIARS, then why not find out? Why would they refuse?
Another variation of this would be to ask for a recent customer with the same equipment, then I could call them up and ask them to build my PCBs while I watch them. Of course, I'd pay them for doing the work, but I would say I'm thinking about buying this equipment and want to see it work before I decide.
Because... if your stories are even close to real experience (with brand new machines)... then I'm not even slightly interested in buying this equipment or assembling PCBs.
Frankly, your messages got me thinking. I've been designing and prototyping high-tech electronic devices for over 30 years now. Back before SMT was necessary, I'd build all my own PCBs by hand, and they'd all work. ALL of them. Okay, maybe not the first year I started, but after that when I had enough experience. Unfortunately, now most of the components I need are only available in SMT packages, so I have no choice but design with SMT.
But you have me thinking. My old PCBs were also about 300 to 400 components (though on larger PCBs). The reason they worked was... because I could personally see what I was doing. Which means, a very alert, careful, observant person could reliably build PCBs with virtually 100% success every single time for year after year after year.
So I wonder... is there a way to make a system today that works the same way, with the same reliability, for PCBs full of fine-pitch SMT components? I already tried a fully manual approach, and that doesn't work. Even though I had plenty good vision with my stereo microscopes, other problems made the process impractical. For example, there is a slight adhesion (or static cling) of components to tweezers (no matter what material they are made of). So once you release the grip on the component and move the tweezers away... the cling will drag the component a bit. I'm sure you know all about the troubles of trying to place components manually. Sure, it is possible to place an 0201 or fine-pitch QFN component with great effort and plenty of time (and multiple attempts), but that is too time consuming for building a whole PCB.
But... if a technique can be found to manually verify placement component by component, with as much time as necessary for the human viewer to inspect each placement... that is probably sufficient for assembling small prototype runs (like one or two or a few PCBs of each type).
I guess my first thought along these lines is the following. You set up the pick-and-place machine to pause an array of 5 cameras over each component after it places the component, but before it picks up the next component. The human viewer can then look at the component from directly above, and from four 45-degree angles (from N, S, E, W). I would guess that 90% to 99% of bad placement or goofy-looking solder-paste situations would be visible and caught by the human operator (assuming they are me, not some bored jerk of an employee at a regular company).
Now, the first response might be... "that's too labor intensive".
My answer to that would be... everything you wrote! Because if what you wrote is true (of new machines too), we ALWAYS spend an ENORMOUS amount of time finding and fixing screw-ups of the automatic processes.
My final comment is this. If SMT assembly really is as bad as you portray, then SMT was one of the most enormous screw-ups in the history of modern technology! The effort should have gone into finer pins and through-holes... or some other approach to increasing package densities. Because what you describe is ABSURD compared to previous generations of electronics.
It is a bit difficult for me to believe that the entire SMT equipment industry is one gigantic FRAUD. Of course, that is possible, for there are a great many huge frauds in the world today, so I don't rule that out entirely. I certainly need to find out whether that is the case, or perhaps you just suffer from problems associated with old models and badly misused used equipment.
Thanks for all your observations.