Pick and place machine kit, with vision

[JuKu]

You don't like the machine, so don't buy it. :-) I'd still like to comment on few points:

.. the time it takes to program it and lay out all the parts...

No matter how you do placement, you have to get the parts out of storage and lay them out on your work area, there is no way around that. Attaching component strips with double sided tape is not a big deal, and stripping the cover tape is much easier when the tape is held down firmly. I used to tape part strips on a piece of cardboard when doing hand-placement, because it made handling the tape without the cover tape much easier.

The beginning of the video shows all "programming" you need to do for a new board: Few clicks for overall start, two clicks for each part type.

, you could have just hand placed the board... then it's not a time saver.  And if you get to the point where it becomes worth it because you can set it up and place 10 or 20 boards, then you're at the point where a Quad 4C makes sense.

<place 10 or 20 boards> <feeders> <board auto-loading>

All these imply production. My machine is not good for that, nor do I imply it is. It is for prototypes. Rx8pilot has explained the problem already. I built the machine because I needed it badly and couldn't find one to buy. And because I had the need, I  guessed that someone else might want it too,. So I wrote the documentation, setup the web site and shop and I am in process of setting up kits for shipping. We'll see if my guess was right to justify the extra effort. Ask me again in an year or so, I know by then. :-)

[JuKu]

The machine seems to have a placement to placement time of about 10 seconds.

Thereabouts. First part of a part type is somewhat slower as the machine is measuring the part height.

That's about 3 times slower than a person can achieve.

3 sec per part? That is very, very good. I would think that that requires parts all in one orientation and a board that the person knows by heart (no looking at documentation). I can do a tad under 10 sec with these conditions. I would think 10 secs humanly impossible for an unfamiliar board with parts oriented randomly. But I've seen seemingly impossible feats on Youtube, so I'm just saying that very few people can beat my machine at the task it is built for, although it does not come close to real production machines, nor is it built for production.

In either human or machine placement, you need to prepare the parts, and it will be at least as slow (likely slower) to prep the parts for machine placement.  But then you have the additional task of programming the machine.  Even if you load your BOM and parts locations automatically, you still need to go through and program the machine where the source parts are and where all the destinations are.  If it's a few seconds per part, then in the same amount of time you could have just placed it by hand.

Parts preparing is about the same. Programming is two clicks for each part type, which comes down to a small fraction of a second per part.

If the OP can solve the feeder, on-the-fly vision and board loading/unloading issues, he will have a very valuable product on his hand if it can be <$5,000 (the price of a TM-240), or even better if <$3,600 (the price of a TM-220A).  Just picking and placing from taped down sections of reels is something you can do with Mach8 and a cheap Chinese CNC machine or even an old plotter or motorized XY table.  Solving the vision and feeding problem would be revolutionary.

Feeders and board loading imply production, which is not the point. The machine uses vision for machine calibration(1), part position measurement(2) and fiducials recognition(3).

1) Most done when building the machine. At startup, homing. At run time when the pickup needle is changed, the needle wobble is optically measured and compensated.

2) It is looking for the tape hole, it can't see a black component in a black tape pocket. the tape pocket is bigger than the part, and that is not taken into account currently. That error source is smaller than reflow self-alignment.

3) Measuring the location of fiducials corrects for board position, board placement rotation, scale errors and skew errors. I don't see how a CNC could place parts without these taken into account. Board position and placement rotation are always significant by my experience. Scale errors come to play with self-made boards (laser printers are not dimensionally accurate) and bigger professional boards. I don't really know id skew (squares being slightly diamond-shaped) has any significance, but it is taken into account anyway.

The machine software is not a complete vision solution, but even if I may say so myself, it is a decent step or two in that direction - and available now for hobbyists and company internal use, Github repo is at https://github.com/jkuusama/LitePlacer.

[mikeselectricstuff]

3 sec per part? That is very, very good. I would think that that requires parts all in one orientation and a board that the person knows by heart (no looking at documentation).

Groups of same parts from tape can be well under 2 secs, so an aggregate rate for a design you know well ( because you just designed it) , of 5 secs per part is very doable.

[mikeselectricstuff]

In either human or machine placement, you need to prepare the parts, and it will be at least as slow (likely slower) to prep the parts for machine placement.  But then you have the additional task of programming the machine.  Even if you load your BOM and parts locations automatically, you still need to go through and program the machine where the source parts are and where all the destinations are.  If it's a few seconds per part, then in the same amount of time you could have just placed it by hand.

You don't need to tell it where the destinations are, or the orientations,  as this comes from the pick/place data from your PCB software.
You just need to map part types to feed positions.
Similarly the fiducials can be embdded in the Pick/place data (e.g. by having a fid as a part), so you don't need to teach that.

A big part of creating an efficient process for low volumes is to streamline the data preperation as much as possible, e.g. making your parts library has consistent orientation with respect to the tape orientation in feeders.
 

[donotdespisethesnake]

2 - 3 seconds manual placing time? That is remarkable! Also, a rather dubious claim!

Even if that is true, that would put you into the "expert" category. Oh, how to be young again, with steady hands and perfect vision

I am lucky to get 10 times that, even on boards I designed. And I can't sustain the effort very long.

The logic of the objectors is "I don't need this DIY machine, therefore no-one else does" which is obviously flawed, and rather short sighted IMO.

Clearly, the DIY PnP is not designed for expert professionals who have no problem placing 1500cph, or who have a business where $10k layout is not a problem. For the rest of us, amateurs, hobbyists, non-engineers, normal people, any automatic assistance would be helpful.

Tbh, "an expert can do it faster/better" is completely missing the point, and is the kind of luddite reasoning aimed at many automated processes. My friend's wife refused to contemplate owning a dishwasher, said there was no point, it was a waste of money etc. After a few weeks with one, she now refuses to contemplate not having one.

Can we make a deal, if we say this type of DIY machine is "for non-experts", can the experts stop saying it is no use for experts?

[Kjelt]

This is definitely fullfilling a niche i think.

What would be great IMO is if you can also manually control the machine (with joystick?) and that the camera would show the part orientation,nozzle and exact location below it so you can also use it for manually assisted placing as some manual P&P machines can.
Still wondering what that large black disc on the nozzle is or does? It blocks the vision of the nozzle?

[JuKu]

This is definitely fullfilling a niche i think.

What would be great IMO is if you can also manually control the machine (with joystick?) and that the camera would show the part orientation,nozzle and exact location below it so you can also use it for manually assisted placing as some manual P&P machines can.

Can't do without adding another camera. The camera needs to look straight down. Hmm, maybe the software could take a snapshot and overlay it on the video, when operator is happy press a button and machine puts the part down where the image was shown. I need to think this.

Still wondering what that large black disc on the nozzle is or does? It blocks the vision of the nozzle?

The needle/nozzle is calibrated by looking the tip with up looking camera. The black disc blocks the needle attachment from up looking camera, so the camera only sees the tip.

[Kjelt]

Can't do without adding another camera. The camera needs to look straight down.

Not sure about that look at this video at 1 minute 25 seconds. That kind of components would really like to be placed very accurately.
They do it with a 45 degree rotating camera as it looks?

The needle/nozzle is calibrated by looking the tip with up looking camera. The black disc blocks the needle attachment from up looking camera, so the camera only sees the tip.

Ah ofcourse you have an upview camera for parts rotation orientation, didn't realize that. Same as in the machine that Mike has.
You know that could also be used for refining the placement (if the part is a bit off the exact axis of the nozzle), thereby correcting the possible backlash of components in their storage strip.

[mikeselectricstuff]

2 - 3 seconds manual placing time? That is remarkable! Also, a rather dubious claim!

foot-operated vacuum pen.

[rx8pilot]

For an individual component, I can do 2-3 seconds per part when timing from one placement to placement. I would have to know exactly where it was and where it was going and its orientation. This is not even slightly realistic for a BOM with 50 components and a total of a few hundred components. I never measure how long it takes to put a resistor on a PCB, I measure how long it takes to populate the entire PCB and then divide by the number of components.

3 seconds per place is 1200 CPH. There may be one or two humans that could pull that off with a lot of planning and practicing on a specific board but I would say that no human alive or has ever lived could take a PCB (even one they have personally designed) and make the first five pieces at anywhere close to that rate.
I have a new PCB on the way that I designed myself. It will be here on Friday or so. It has about 160 components all of which have been kitted in special holders in the order that they are placed and I have color coded reference documents prepared to help match the components with the pads. This is an ideal setup for manual placing since everything is in place and in order. I have a PCB holder, vac pen, and a microscope together with a few thousand hours of manual assembly experience.

The first PCB from the time it lands under the microscope til it is on the way to the oven will be 90 minutes or so since I have to verify every placement. The second one will take about 45 min if all goes well, and it will eventually stabilize at about 30 minutes per PCB after I have most of it memorized. I am not exactly world record material, but I am faster than most and make very few mistakes. To make the first two PCB's for testing would be about 2hrs 15min of actual placement time - about  146CPH + plus the despicable mental focus necessary.

At 10 s/part (360CPH) the first PCB would take about 28min. The second one would take 28min. The setup time is probably no different the programming effort would be very similar to the color coded paperwork I do for manual. We have the same task done over an hours sooner and it was a lot less frustrating in the process. As you see, it will take about the same amount of time after I have the PCB memorized, but by then the prototype effort is done and it moves on to more proper production. This is not liekly a machine that can go unattended for long, but I can be prepping tapes and doing other ancillary tasks as to puts down resistors automatically.

This is all assuming the machine itself is working and reliable which I cannot speak to, I am just saying that 360CPH is a VAST improvement over 100% hand placing. Humans are not made for this task, machines are. At the price, plus whatever mods and effort I may put in - I only have to do a few PCBs for it to make sense. I still have my bigger plan to get a 4C, but I would rather have both of them for different tasks. This little machine can be right next to my bench.

[JuKu]

This is definitely fullfilling a niche i think.

What would be great IMO is if you can also manually control the machine (with joystick?) and that the camera would show the part orientation,nozzle and exact location below it so you can also use it for manually assisted placing as some manual P&P machines can.

Can't do without adding another camera. The camera needs to look straight down. Hmm, maybe the software could take a snapshot and overlay it on the video, when operator is happy press a button and machine puts the part down where the image was shown. I need to think this.

I made some experiments towards this. Obviously the snapshot image quality needs enhancement, but this image tells the idea:

The machine knows very well the relationship of the needle and camera, so you don't really need to see where the needle is going, the camera image is enough. The machine can take a look at a difficult part, build a snapshot and take the image over the placement at roughly the correct orientation and position. The user fine-tunes the placement (there are already several methods of moving it manually) and when happy, the machine places the part where the image shows it is going. This would work for BGAs too by taking the snapshot with the up looking camera.

(Edit: clearer explanation)

[Kjelt]

That looks very promising 

[rx8pilot]

Yes, if it has the potential to place QFN's and QFP's it would be fantastic.

[JuKu]

Yes, if it has the potential to place QFN's and QFP's it would be fantastic.

Clearly, the _potential is there. I am confident that I can also convert the potential to actual cabability. :-) (BGAs, too)

[rx8pilot]

Clearly, the _potential is there. I am confident that I can also convert the potential to actual cabability. :-) (BGAs, too)

I would love to have a reliable way to prototype with BGA's. I have been pushing my PCB layouts to the limits and the next step is incorporating blind/buried vias, via in pad, and using smaller passives with minimum pads. At the moment, most of my designs have been in the realm of hand soldering (although I almost never hand solder anymore). I now have a proper stencil printer and a professional convection reflow oven. Very happy to see this project progressing.

[JuKu]

KiCad* and DipTrace are now supported.

*Note to KiCad users: The current version of KiCad allows you to put spaces to component value. The KiCad generated .pos file is space separated, and the field width can change. These don’t fit together, and this will change in a future version of KiCad. So, spaces in values are not accepted without quotes around them (this is how I’m told the KiCad will behave in the future).

[rx8pilot]

I just purchased a Quad IVC for a very good price so I will not be doing this machine yet. The goal was to buy/build this first and get the Quad later, but the opportunity to get the Quad with 55 feeders came up and had to jump on it.

I still think this is a great idea and will still benefit me as an addition to the Quad pick and place machine.

[Spikee]

Do you have a video of it placing parts on a bigger board?
I kinda want to buy one just to make my work a bit easier.

[JuKu]

I don't, but it would look much to same. What is your concerns about big boards?

[Spikee]

I don't, but it would look much to same. What is your concerns about big boards?

No real concerns, It would just make me feel better about it 

[Spikee]

Bought one. Will do a review off my experience with it when it arrives.

[zeke]

JuKu, I am impressed. Well Done!

I watched the placement demo video and now I have two questions:

• Do you have a demo video of it placing things like QFNs or power inductors?
• Do you think you could crank up the speed of the servos more like the TM220A?
• Would it tear itself apart if the speed was increased?

[JuKu]

JuKu, I am impressed. Well Done!

I watched the placement demo video and now I have two questions:

• Do you have a demo video of it placing things like QFNs or power inductors?
• Do you think you could crank up the speed of the servos more like the TM220A?
• Would it tear itself apart if the speed was increased?

1. Will do at some point. Right now I need to prepare some kits.
2. Not that fast. Beefier motors and power supply would bring up the speed but I haven't tried it.
3. The motors would run out of torque first. Ok, I guess if you put in really big motors and really big power supply you could make it go bang. 

[JuKu]

Now in stock.

[rx8pilot]

Any movement? It seems like a fantastic project.