Add-ons: feeders

[WayOutWest]

My friend has one of these NeoDen TM-240A - i was watching it yesterday - (some videos of it in use on Youtube) - it has quite a novel way of advancing the tape from the feeders - no motors - a needle comes down from the PnP head above where the holes in the tape are, and then it drags some more tape out from the reel, before the needle retracts and it then picks up the components.

I've seen a lot of PnPs that do this. Usually they have what looks like an upside-down pencil attached to the head, in addition to the needle. The eraser part of the upside-down pencil does the paper-dragging in order to not put stress on the pickup needle (which can bend it, influencing accuracy).

But honestly, all this feeder nonsense is strictly for ultimate speed, which is something the liteplacer is not meant for. The liteplacer is never going to win speed competitions, so why fool around with complicated mechanisms whose only purpose is speed?

I've gone totally feederless now... just rip the tape off a dozen meters worth of components, spill them onto the table, and let the vision system sort them into nice rows on gel-pak substrate overnight. Many added bonuses -- you can close up the gel-pak box and swap them in/out MUCH more easily than any tape+feeder system, virtually unlimited number of different components, QR codes to identify gel-pak boxes, place the frequently-used components right next to the work, etc, etc. Of course you need to be using better vision than the Hough Circle Detector for this... OpenCV's MSER plus minimum-bounding-rectangle is really really fast and recognizes all sizes of SMT passives flawlessly using the Andonstar cam; just give it upper/lower bounds on the area and aspect ratio of the component you're interested in. FireSight really opened my eyes to what's possible with good vision.

Feeders are a waste of time for this machine, IMHO. Think out of the box (or reel)!

[mpbrock]

Hi Adam

Have you any more info on this sorting method your using? is it a software mod? I'm intrigued! A photo of the gel-pack setup would be great!!

Thanks

Matt

[WayOutWest]

Hi Adam

Have you any more info on this sorting method your using? is it a software mod? I'm intrigued! A photo of the gel-pack setup would be great!!

I got it working using double-sided Kapton tape shortly before ripping my machine apart for this.

The gelpak is strictly to make sure no residue is left on the component. In theory Kapton could leave resiude, but in practice I think the solderpaste flux will etch it off; that stuff is powerful. So I might not even bother with the gelpak boxes, but if anybody complains about "the sticky goo will leave residue that will mess up the soldering" there is an easy answer to that for just a few dollars extra cost; the chip industry has been shipping around bare silicon on gelpak for decades now, it releases cleanly on much more demanding contamination scales than required here.

I will definitely post a video once I have the machine working again. I really should've taken one before I disassembled it.

The sorter is my own software written from scratch (except for FireSight/OpenCV/libuvc) since I do not consider learning proprietary programming languages (like C#) to be a good time investment. I'm willing to open-source my software but I cannot provide any support at all and I don't want people to think that it competes with Juha+thereza's software, which is much more user-friendly. There is no GUI, just video playback on the monitor and a point-and-click plan view remotely displayed on another machine. So I'll post the video soon and if you really want the software with absolutely-no-guarantee-or-support I will provide it. It's written in Java with JNI linkages to OpenCV, FireSight, DirectFB, and libuvc; the full-framerate video display is Linux-only (since I need DirectFB to get smooth playback at those speeds from four cameras) but everything else should compile+work on pretty much any platform.

[mpbrock]

Thanks Adam, sounds pretty cool

I've not come across that firesight library, i'll have to take a look - id be interested in having a look at your software at some point (running Linux is a great thing) - although with you doing it in Java guess its pretty much platform independent (as long as the other modules are of course!). Been a long time since i touched java, these days seem to spend all my time coding C/C++!

[WayOutWest]

I will definitely post a video once I have the machine working again. I really should've taken one before I disassembled it.

Video is here http://www.liteplacer.com/phpBB/viewtop ... 158&p=1012

[mawa]

Well if your PCB only contains SMD capacitors... but how about resistors and LEDs and FETs?

When you spill these onto your table a random quantity will be lying on their back. How do you intend to turn 'em over?

And how sticky is the gel-pack?

It will probably be sticky on the whole bottom surface of the part. You firmly press the part into the gel-pack.
When you later try to pick up a part your needle will have be very plane /orthogonal with the parts pickup side otherwise there will be a small air gap.
When you pickup a part from a tape or a tray the part is loose and does not adhere to the tape or tray so it can come up by the vacuum suction and tilt to fill the air gap.

So seeing is believing... could you show the second phase where you pick up your parts from the gel-pack and place them on your PCB?

[ameli]

So seeing is believing... could you show the second phase where you pick up your parts from the gel-pack and place them on your PCB?

I think that this is a very nice demonstration of part recognition and placement. The algorithms are there, so you wouldn't necessarily need to go from random spill to gel pack to PCB. You'd just go from random spill to PCB.

You run out of caps, just dump some more down into your container for the machine to find and place.

But I agree that this is fine for parts symmetric caps, but would be interesting how you would handle flipped resistors.

[WayOutWest]

When you spill these onto your table a random quantity will be lying on their back. How do you intend to turn 'em over?

All at once. You spill the random quantity onto a hard rigid plastic plate, have the machine sort the ones that are lying face-up, then the next day a human puts another hard plate on top of the remaining components, flips the "sandwich" upside down to expose the other side, removes the top plate (formerly the bottom plate) and the machine finishes the job. Disclaimer: I haven't coded this yet.

And how sticky is the gel-pack?

It is sold in four different adhesion strengths. I've ordered a sample-pack to figure out which one is right.

It will probably be sticky on the whole bottom surface of the part. You firmly press the part into the gel-pack.

No, that's the magic of gel-pak, it is a membrane; there's no adhesive. People put bare silicon dice on them (in fact, that's what they're for).

So seeing is believing... could you show the second phase where you pick up your parts from the gel-pack and place them on your PCB?

Alright, may take a week or two while the gelpaks arrive and I get my demo-video gear set up again.

[WayOutWest]

So seeing is believing... could you show the second phase where you pick up your parts from the gel-pack and place them on your PCB?

I think that this is a very nice demonstration of part recognition and placement. The algorithms are there, so you wouldn't necessarily need to go from random spill to gel pack to PCB. You'd just go from random spill to PCB.

Yes, that is possible, and I might even do it, but it is much slower and people would complain

In theory the placement of a whole board isn't supposed to take more than 1-2 hours or the solderpaste can dry out. Although the new no-refrigeration solderpaste (which I *LOVE*) can be on the board for a full 24 hours minimum before reflowing; it's only the older stuff that forces you to rush the job.

But I agree that this is fine for parts symmetric caps, but would be interesting how you would handle flipped resistors.

Well, resistors are symmetrical, maybe you meant diodes or LEDs? I don't have any of those on my current board -- all the non-passive parts are in packages with very obvious orientation markings -- but I don't think it's an insurmountable barrier. Worst case you bring the device over to a test-point and electrically test it. I have to do this anyways for one of our parts since it's been reballed and we don't want to assemble broken chips.

[WayOutWest]

But honestly, all this feeder nonsense is strictly for ultimate speed, which is something the liteplacer is not meant for. The liteplacer is never going to win speed competitions, so why fool around with complicated mechanisms whose only purpose is speed?

...

Feeders are a waste of time for this machine, IMHO. Think out of the box (or reel)!

I ought to humble this down a bit.

I've been forced into using feeders (just two). I've got about 3,000 boards to build with 100 placements per board, and 90 of those placements are the same 0603 capacitor for decoupling. My "spill parts onto the table, sort into a gelpak carrier" scheme has been working well for the components that appear only once on each board and are 0603 or larger, but when you want to place 270,000 of something.... well, I confess, feeders are the way to go there. I was (partly) wrong here.

Also the gelpaks I have will reliably release 0603s but with the 0402s I'm getting occasional pickup failures, and the needle I have to use for those is so tiny that the pressure sensor can't detect the failed pickup. There are still some options here: improve the pressure sensor, install the Juki nozzles to see if they allow better pressure sensing (no long needle to pull air through), use the vacuum-release port on the bottom of the gelpak (which I haven't been using), etc... but these are all harder than just setting up a second feeder.

I've imitated vbesemens' "pull back the cover tape with a servo" scheme from here and it's actually working surprisingly well. I'm using the craptastic chinese aluminum reel holder, which actually works okay now that I figured out the right tape routing (see second picture for a hint if you're going to try this yourself). It isn't perfect, certainly not as elegant as Knas' solution, but hey it was $8 and I got it working in one day with parts I already had lying around. Only problem is that while modding the servo for continuous rotation I managed to make it completely flaky. Argh. But at least I can direct-drive it from the TinyG. If anybody wants the firmware patch to do that, let me know.

I do a vision alignment (MSER-based) before every pickup -- I'd already put a huge amount of effort into getting good vision working since I have a lot of BGAs. It is reliable, but slow. That's okay, it just needs to be able to run all night without intervention.

And yes, I still owe you all (the forum, collectively) videos of the gelpak sorting. I keep procrastinating because of how ugly my machine looks right now, in the midst of various upgrades. Wires hanging out all over the place and so on.

downcam.jpeg (25.19 KiB) Viewed 9907 times

needlecam.jpeg (20.82 KiB) Viewed 9907 times

x.jpg (33.07 KiB) Viewed 9907 times