Tips (or a machine?) for placing SMD parts

[JuKu]

Ok, I have satisfactory method for hand printing solder paste and reflow soldering for my prototype lab. See https://www.eevblog.com/forum/manufacturing-assembly/beta-layout-reflow-kit-first-try/. I have done my first real board on that, and my process definitely needs some development.

First, placing the stencil could be easier, but I don't think that is a real problem. My next board will have two small holes on the PCB and on the stencil in opposite corners. Then I'll put same size of pegs trough the holes and tape the stencil down. That should do it. Of course, if you have a better idea, please share!

The real issue is placing the parts. I did pretty much what Mike suggested in his video (), but it was still a chore. The board had 460 components, and besides a couple of connectors, all SMD. For this old guy, that took way too long and became difficult towards the end. My hand started shaking (not much, but with SMDs, you don't want any!) after an hour or so, with two more hours to go. The paste also started going dry (the final board turned out to be fine, though). Definitely something that I'm not looking forward doing again!

Right now, I can't think of anything to improve the manual method. Thank you Mike! You have more or less perfected manual already. Or is he? If you have any tips, hints or ideas, please share!

But I can't help thinking about a machine for this. A hobbyist/prototype pick and place machine could be very simple, basically doing what the manual method does. No automatic feeders are needed. I envision guide slots with the part strips (equivalent of those cardboards (see the video)) and for this purpose, it would be perfectly ok if the control program would do first 40 (or whatever length of the slots are) bypass caps and then ask "please peel 24 more from slot 1". Also, it would be ok if the control program says "First tray finished. Please reload slots: Slot 1: part type xxx (470pF, 0805 capacitor) 24 pieces. Slot 2: ...." The ICs or special components I could place by hand, if the 400+ chip parts would be done by a machine. Or maybe place them pre-oriented to a jig or something.

I would think that with something like this, the 4-5 hours of assembly (with preparation of those cardboard plates) would have been less than one. Not good enough for about any level of production, but a tremendous help for hobbyist or a prototype lab. Not to mention the easy part of it...

I searched, and found some videos and info for DIY projects, but none documented enough to just build one. On the other hand, the videos show that in principle, something like that is feasible. I didn't find any affordable commercial units. Seems to me that this only needs a simplest CNC machine, modified head unit and some software. This project would not pay back to me in time, but it would be so fun.

So, why should't I? Ideas? starting points? (Best starting point that I found with one night searching is http://www.shapeoko.com/). Or if you know something like the above commercially available (say, less than 1000€), let us know, too.

[poorchava]

At a time, i was also thinking about building something like that, so i cna tell you to what conclusions i came.

First you need some kind of tool mount moving in XY plane. Since pick and place doesn't involve significant force, this can be one out of steel or aluminum rods and extrusions. For linear slides you have a few options: furniture drawer sliders (cheap, available), supported or unsupported hardened rods+ linear bearings, linear slides (expensive) or diy kind usually made of some angled extrusions and ball bearing. Either of those can do the job. As for drive system since you want rather fast movement of the placement head, I would go with stepper motors + gears + toothed belt. This provides fast movement, at cost of small force (which is completly fine for this application). Also without significant reduction ratio resolution is relatively small (half inch toothed wheel with 200steps/rev motor gives you 0.4mm resolution, but when you get use a microstepping driver with 1/16 division ratio you can get something like 0.025mm which is acceptable.

That's for XY movement. As for Z movement you need some way to adjust height of nozzle over the PCB. For placing identical parts that can be done manually (eg. separate setting for resistors and capacitors) but automatic adjustment can also be done. i think that ordinary threaded rod with a nut driven by a stepper will do the job. there has to be some kind of shock absorption between the nozzle end and whole machine (maybe some kind of spring). Of course you need to have a vacuum supply and a nozzle.

You need some kind of optical recognition or a few lost steps will cost you dearly. Open loop solutions are not the best way here. You also need to take into account that parts can move a little inside the reel.

On top of that moving parts have to be as light as possible, in order to allow rapid accelerations without vibrating.

[JuKu]

At a time, i was also thinking about building something like that, so i cna tell you to what conclusions i came.

First you need some kind of tool mount moving in XY plane. Since pick and place doesn't involve significant force, this can be one out of steel or aluminum rods and extrusions. For linear slides you have a few options: furniture drawer sliders (cheap, available), supported or unsupported hardened rods+ linear bearings, linear slides (expensive) or diy kind usually made of some angled extrusions and ball bearing. Either of those can do the job. As for drive system since you want rather fast movement of the placement head, I would go with stepper motors + gears + toothed belt. This provides fast movement, at cost of small force (which is completly fine for this application). Also without significant reduction ratio resolution is relatively small (half inch toothed wheel with 200steps/rev motor gives you 0.4mm resolution, but when you get use a microstepping driver with 1/16 division ratio you can get something like 0.025mm which is acceptable.

Makerslide for slides. Slides, rollers, bearings belts etc at https://www.inventables.com/categories/innovative-materials/components, a couple of hundred at most. Based of the results that the CNC and 3-D printer folks get out of that system, that is more than sufficient (better than hand, and that is all we aim for).

That's for XY movement. As for Z movement you need some way to adjust height of nozzle over the PCB. For placing identical parts that can be done manually (eg. separate setting for resistors and capacitors) but automatic adjustment can also be done. i think that ordinary threaded rod with a nut driven by a stepper will do the job. there has to be some kind of shock absorption between the nozzle end and whole machine (maybe some kind of spring).

The nozzle mount needs to be designed and built; we need rotation as well. The height adjustment need is only a few mm's. A spring load so that the downforce is few ten grams, and I don't think there is a need for a more complicated adjustment.

Of course you need to have a vacuum supply and a nozzle.

That would not be difficult. And aquarium air pump and a solenoid valve to turn vacuum on and off works. That is more or less what I have now, although the pump is in Aoyue box.

You need some kind of optical recognition or a few lost steps will cost you dearly. Open loop solutions are not the best way here.

I don't think lost steps are a big problem, otherwise stepper motors would not work for so many applications than they are used. Besides, keeping it simple: Every few components, have the nozzle press a switch. If the trigger signal does not come at the step we expect it to, alert the operator and ask manual verification of the few last components. Or have a calibration PCB on work area and a probe needle: If the needle does not hit a 6mil track, we have a problem.

You also need to take into account that parts can move a little inside the reel.

The resistors and capacitors are the real problem in manual method. But they do self align during the reflow; I would think that the error inside the reel plus the machine tolerance is still more than good enough.

Looking at the board I did this week: 450 or so SMD components. Four SMD relays and one 0.5mm QFP; if I would need to place those manually, no problem. Two power pack regulators that the ultra-low-end placement machine may or may not be able to handle. Ok to do those by hand, too. Three SOT-223 regulators and five electrolytic caps; I would be disappointed if those would not work, but if not, not a bid deal yet. All the rest are capacitors (most 0805, some 1206), 0805 resistors, SOT-23 diodes and transistors and SOIC ICs; if the system works at all, it can do those.

On top of that moving parts have to be as light as possible, in order to allow rapid accelerations without vibrating.

That's a software issue. It would be far faster than manual in any case, and if the system knows when it is placing a resistor or a capacitor (fast) and when it is placing something else (careful), speed will not be an issue.

You didn't discourage me to drop the idea yet.

[mikeselectricstuff]

The problem is finding a sweet spot on the usefulness versus complexity/cost curve. I've yet to be convinced there is one.

As soon as you limit the useable part types, you reduce the range of jobs for which it is useful.
However clever the hardware solution, you still need to do some setup/programming, the time taken for which eats into the minimum component count for which it is a time saver over manual placement.
Even with a very streamlined process to take PCB data and create a placement position/rotation list, you still have to tell it which parts are where in whatever feeder arrangement you have. Without a feeder bank big enough to dedicate lanes to common parts, you need to do some setup for every job.
 
As soon as you start spending money on any sort of motorized system you have to ask yourself how much usage does it need to get to cover costs and/or save time, compared to paying someone else to do it.

It may be practical to make a fairly cheap semi-automatic/manually assisted system that maybe auto-picks from a few lanes of tape, but its usefulness would be limited to small jobs and prototypes which are complex enough that setup time is small compared to placement time.

I think the only possible way someone might achieve anything generally useful is to throw cheap computer vision at the problem rather than expensive mechanics, and maybe leverage some cheap mass-market mechanical device and use vision to improve its performance

Despite a few crude attempts seen on youtube nothing has changed my opinion that there will never be a truly useful DIY/OS pick/place machine.
  The market for it is too narrow, and the complexity of doing it too high.

If someone really wants to do it they need to forget about messing around with X/Y systems and solve the problem of cheap and reliable tape feeders.  Most DIY P&P attempts fail to realise that feeders are the problem that needs solving.

I think however there may be scope for some useful tools to be developed to speed up manual placement, but not a generally useful automatic pick/place machine.

For example how about this as an idea as a placement aid to speed up and improve accuracy of manual placement with no programming or precision  mechanics needed : 
A simple manually operated gantry (Pantograph style, maybe made from PCB?), where the only automation is a cheap downward-pointing camera and an x/y/z actuator (RC servos) capable of small movements (10mm?) , to home in on part positions and placement locations when manually pointed to them.

I.e. you point it manually towards the part you want, within 50%  of the pick position and the actuator homes in on the  true position and picks it accurately, and then you point it roughly at the placement position, and again it homes in via vision and places it. (The latter requires a clean enough paste print to use for position guidance)
  You'd  also need manual rotation to at least 90 deg increments.
 As soon as you reduce the required freedom of mechanised motion, then it becomes viable to use very cheap  servos, and with a limited field of view, cameras and lenses also approach negligible cost.
With a pantograph type mechnical arrangement, some clever mechanical design could allow the manual XY control and the motorized parts to act at different pivot points, so you simply hold the manual control stationary (e.g. drop it onto the table as a brake) to allow the mechanics to make fine adjustments at the pivot point.
A reason for doing it this way would be to allow the X/Y mech to not be on the base and not the moving arm, reducing weight and hence required strength.

[JuKu]

If someone really wants to do it they need to forget about messing around with X/Y systems and solve the problem of cheap and reliable tape feeders.  Most DIY P&P attempts fail to realise that feeders are the problem that needs solving.

The feeder might only need some slots. The system in this video has horribly slow Z movement, but the feed idea is clever:
. The problem then is to get rid of the cover tape. If I accept the controller to say every now and then "Please peel 32 units more from slot 4" (with unit markers on the slot, so I know how long is 32), the problem is more or less solved. Putting strips to slots is in any case far faster and less tedious than putting the parts to PCB. For reels, something like http://www.goodluckbuy.com/smt-smd-feeder-for-diy-prototype-pick-up-place-5-way.html, and for strips just the slots would be even cheaper. At those prices, one can afford lots of trays.

The market for it is too narrow, and the complexity of doing it too high.

This is the crucial point. I don't think the complexity is necessarily too high, and the cost of the product would be acceptable. However, the effort does not pay back for any company to do something like this for internal use, you certainly are right here. The hobbyists seldom commercialize the results, so I'm not expecting anything from that direction either. I would pay 1000€ for a machine that places most parts from SOIC and down, so I would think there is market for such a system. You think there is not. I guess we won't find out until some company tries it and either fails or succeeds. (They would have at least one customer!) I think the price point is achievable, too (300-500 in parts, retail around 1000 or so, and there is some  profit and development payback).

I think however there may be scope for some useful tools to be developed to speed up manual placement, but not a generally useful automatic pick/place machine.

For example how about this as an idea as a placement aid to speed up and improve accuracy of manual placement with no programming or precision  mechanics needed : 
A simple manually operated gantry (Pantograph style, maybe made from PCB?), where the only automation is a cheap downward-pointing camera and an x/y/z actuator (RC servos) capable of small movements (10mm?) , to home in on part positions and placement locations when manually pointed to them.

I.e. you point it manually towards the part you want, within 50%  of the pick position and the actuator homes in on the  true position and picks it accurately, and then you point it roughly at the placement position, and again it homes in via vision and places it. (The latter requires a clean enough paste print to use for position guidance)
  You'd  also need manual rotation to at least 90 deg increments.
 As soon as you reduce the required freedom of mechanised motion, then it becomes viable to use very cheap  servos, and with a limited field of view, cameras and lenses also approach negligible cost.
With a pantograph type mechnical arrangement, some clever mechanical design could allow the manual XY control and the motorized parts to act at different pivot points, so you simply hold the manual control stationary (e.g. drop it onto the table as a brake) to allow the mechanics to make fine adjustments at the pivot point.
A reason for doing it this way would be to allow the X/Y mech to not be on the base and not the moving arm, reducing weight and hence required strength.

Might work, but I see the complexity level about the same. Maybe I'm just not as comfortable with machine vision as I am with microcontrollers and stepper motors! Btw, even my first paste print using a manual scraper turned out to be perfect. Any system that can home in on pads can home in on that.

[poorchava]

Cost of mechanics doesn't have to be high.

I've built my cnc-mill at cost of around 2000-2500 PLN (that's like 600 euro). And I had to use massive supported rods, linear bearings, stepper motors with like 2A/phase 0.9 Nm of torque, hardened trapezoidal threaded rods etc (and ofc like 40+ kg of plywood).

I believe the mechanics for such p&p machine could be built for around 150 euro.

As for steppers loosing steps it is an issue, especially when using direct belt drive (instead of screw drive). At high speeds if your tool makes a sudden turn or stops and then accelerates rapidly (eg. when picking up component from tape) the stepper can loose steps because of the sudden force. Especially that a torque of stepper motor is decreasing as the speed increases. That's why all steppers have torque specified as 'holding torque' which means that's the torque when motor doesn't rotate (eg. speed is zero).

Also, microstepping indeed increases RESOLUTION, but there's also a limiting factor of POSITIONING PRECISION, which is a property of motor itself rather than a driver (eg. uniformity of stator windings etc). This means, that when you make motor gor 2 steps clockwise and then 2 steps counterclockwise in pratice it returns to its base position only with some precision. Same as goin 6 steps cw and then 3 steps ccw doesn't really put the rotor ideally in half of the distance, there's always some error.

This is less relevant in screw drives, where one rotation moves the carriage whatever the thread pitch is. For direct belt drive distance per revolution equals to the outer diameter of the tothed drive cog on the stepper it may make a significant difference.

[gregariz]

If someone really wants to do it they need to forget about messing around with X/Y systems and solve the problem of cheap and reliable tape feeders.  Most DIY P&P attempts fail to realise that feeders are the problem that needs solving.

Yes ... its all about tape feeders. Unfortunately, and it seems particularly the case for the small hobby p&p machines, vendors have figured out that they can often earn alot more by selling feeders at 500 a pop than the machines. This adds up when you need 30 feeders for a job.
My personal opinion is that you are probably going to do better by picking up an old mainstream machine like a Philips (Assembleon) or Juki or similiar. Problems are: you'll need the room for it and you'll need to put up with wear and maintenance. But your chances of being able to pick up a box of feeders for cheap are improved because of the number out there.

The feeder might only need some slots. The system in this video has horribly slow Z movement, but the feed idea is clever:
. The problem then is to get rid of the cover tape. If I accept the controller to say every now and then "Please peel 32 units more from slot 4" (with unit markers on the slot, so I know how long is 32), the problem is more or less solved. Putting strips to slots is in any case far faster and less tedious than putting the parts to PCB. For reels, something like http://www.goodluckbuy.com/smt-smd-feeder-for-diy-prototype-pick-up-place-5-way.html, and for strips just the slots would be even cheaper. At those prices, one can afford lots of trays.

The market for it is too narrow, and the complexity of doing it too high.

This is the crucial point. I don't think the complexity is necessarily too high, and the cost of the product would be acceptable. However, the effort does not pay back for any company to do something like this for internal use, you certainly are right here. The hobbyists seldom commercialize the results, so I'm not expecting anything from that direction either. I would pay 1000€ for a machine that places most parts from SOIC and down, so I would think there is market for such a system. You think there is not. I guess we won't find out until some company tries it and either fails or succeeds. (They would have at least one customer!) I think the price point is achievable, too (300-500 in parts, retail around 1000 or so, and there is some  profit and development payback).

The redfrog kit does the same feed trick of pulling the tape with the needle. He uses fishing sinkers dangling from the tape cover to keep them pulled back. A bit of fine tuning necessary there I suspect.

http://buildyourcnc.com/PickandPlaceMachineTheredFrog.aspx

My problem is that I look at the price of it and then look at the price of what a used machine is. Then you can look at the price of the manual and semi-auto systems. If you are careful the prices all start to converge around 3 or 4k and then one has to ask themselves which way to jump.

[mikeselectricstuff]

But your chances of being able to pick up a box of feeders for cheap are improved because of the number out there.

For any old machine that is still out there and being used, feeders tend to hold value better than machines. An old machine without feeders has little value, but a bunch of feeders for an old machine will always have value to anyone still using those machines.