[Eclipze]

Quite interesting about the feeders stewi. Never even considered something like the tape dust. I previously looked at parting the cover tape up vertically to pick, and then re-join it to exit. I concept I've very much liked. But I do like the idea of having a laser slice the center of the tape cover tape too

[Nails]

>>You may want to look at the Agilis feeder from Mydata.

Wow, that looks ultra-simple. Can't believe I never came across that with all the searching I have done on this topic.

Thanks Stewi!

Now, do you know the specifics about the head designs in pick an place machines? I've been struggling with that. I mean, how are they achieving rotation, vacuum, and z-axis movement on multi-head designs in such a compact space and also control placement force. I'd love to see a detailed picture of the mechanism.

[stewi]

Hi Nails,
the head has a vacuum distribution wheel in the center made with a Teflon surface. 12 holes for the 12 nozzles are tapered. However, during motion the holes are closed for a few ms. It doesn't matter, because there is enough vacuum in the airline, that components don't fall off.
Each of the 12 segments on the head have its own linear bearing. In all other but in the pick and place position all segments are held in a cage. In the pick and place position the segment is pick up by a bearing which moves, belt driven, the segment up and down.
The stroke is calculated by feeder height, board thickness and component height. An additional optical sensor senses, when the nozzle touches the board or component in the feeder. On older machines we used to sense, that the z axis stopped, traveling with low current until there were no more encoder signals.
Each segment contains a hollow nozzle holder shaft, an incremental encoder and a valve (vacuum/air blow). The head cycles through different stations during movement. There is pick and place at 6 and reject at 5 (if a component has a vacuum error the head moves to a dump bin). There are two turning stations rotating the nozzles with the components, one at 4 and one at 8 o'clock. The one at 4 rotates the component before camera inspection at 12, the other rotates the component after inspection at 8.

I show a picture of the older style 12 nozzle head segment. I guess, Siemens would have a fit if I share newer pictures.
Keep on asking or PM me.

[Nails]

It sounds as if you are describing a turret style head. Am I right? I supposed I should have been more specific. I was actually referring to the kind where all heads are in a row vertically and move up and down individually with independent theta adjustments. I looks like they are spaced less than an inch apart, and like 6 or 8 of them in a row. Such as can be seen in the following video.

[ame=http://www.youtube.com/watch?v=cKJ34ATlvK4&feature=related]Building a SMT PCB with Philips Topaz and Emerald machines. - YouTube[/ame]

Can you explain more about the optical sensor that senses when the nozzle touches the board or component in the feeder?

>>On older machines we used to sense, that the z axis stopped, traveling with low current until there were no more encoder signals.

I thought perhaps I could do something similar by using one of these

Miniature Metal Gear Motor - 79 RPM

and control the current with a PWM, then control placement force by increasing the current some amount which would correlate to the desired force.

I think we all are just looking for any ideas that work well for relatively cheap. I still haven't started building mine and am open to looking at all possibilities. I don't plan on having a mult-headed unit, but if I can design something with less weight, that is always a plus.

Also regarding placement force. What can you tell us about that? As I understand it, some machines actualy use force transducers and some just overtravel with a spring-loaded tip a distance proportional to the spring constant and desired force.

It's great to have someone with experience with commercial machines to ask questions of. I think I can speak for all of us that we appreciate your participation and input.

Thanks

P.S. I got inspired enough after seeing the Agilis feeder from Mydata that I bought a reel of capacitors to do some testing to see if I could accomplish the side-peeling.

[stewi]

Yes and no. A torret head machine has a stationary head, Fuji CP 2,3,5, Sanyo, Panasonic to name a few. Siemens has a revolver head, which moves on a X,Y gantry. Advantage, each component can be inspected on the fly, while Philips, Mydata and other have to travel with each nozzle over a camera.
The basic idea of multiple nozzles in line is to have nozzles spaced out such, that each nozzle can pick from feeders at the same time. Indeed for some boards like memory boards, which have only two different components this is perfect.
If you are going for speed, than the sensor is a better idea.
If you look at most of these modern machines, than they all have a micro controller board at the head, controlling functions as vacuum sense, camera light on and so forth. If you look at the current sense, than this signal will have to go all the way to the motor control board.
Siemens can do 100 g to 800 g placement force. The idea was to place PLCC with more force. In practice, nobody is using it and 200 g, spring force is just fine.

[stewi]

Nails,
you shouldn't use a gear motor. The gear makes it less sensitive and too slow.
Here a view of the old style revolver head rear side, with the Z-axis, cam driven.
Motors are Escap (Swi$$ made) with encoder and tacho feedback.

[stewi]

You buy cheaper dummy components for your test Dummy Components. Genuine TopLine Daisy Chain Dummy Lead Free Dummy Components for SMD training, Machine evaluation. Pb Free. RoHS Compliant. Best dummy components in town. BGA, CSP, WL-CSP, WLP-CSP, eWLP, QFN, QFP, TQFP, LQFP, PLCC, SO, SOIC, SSOP,

The problem is, you may get it to work for one reel and it doesn't work on the next.
Tape width can change from 8.1 mm to 7.9 and the position of the cover tape is not even specified. With that, it is difficult to position the peeler, that it works just right for all "8 mm" tape.
Another tip, the tape should go in a small valley in order for the cover tape to build a small hill.

[Nails]

I wish I knew what the heck I was looking at in the pictures!

The Dummy components link is an ultra-cool find. That's awesome to know.

[stewi]

For simplicity, you may want to look for a hollow shaft stepper for theta.
Hollow Shaft Stepper Motor offers design flexibility
Imagine you stick a vacuum nozzle at the end of the shaft and move the entire motor up and down for Z.

Here is an example of a turret head chip-shooter (Sanyo), using hollow shaft steppers for nozzle/component rotation:
[ame="http://www.youtube.com/watch?v=cZAhMuMDir8&feature=related"]Some of my stuff getting assembled at SMD factory - YouTube[/ame]

[stewi]

Also, to my embarrassment, I should have known better, RomanLini responded on the audio-tape linear scale in another thread.
At least, I hope, the new lower cost linear motors are new to this forum:
Design News - Features - New Class of Linear Motors

[RomanLini]

Yeah the linear motors look nice! It would be good to get linear motors for a couple hundred dollars or less, with cheap steel rail. Very cool.

You could make a linear motor by gluing those tiny cheap ebay magnets into holes in a plastic or aluminium rail, and making a coil assembly. I've wanted to make one for a while but it's a lot of work and probably wouldn't give a very high resolution. And would still need some type of linear bearing as well.

Stewi have you seen the great forum thread on the "rack meshed belt" principle? I think you did something similar in your low cost CNC build thread;
http://www.cnczone.com/forums/linear...e_ever_if.html

http://www.cnczone.com/forums/attach...&d=1243495440&

The main thing I have against belts is the tendency to behave like guitar strings, but that belt setup means the belt length is never held like a taut string.

[stewi]

Thanks, RomanLini,
the guitar string effect doesn't have to sacrifice in accuracy.
I used to be impressed by the accuracy of the granite base, air bearing, Zevatech Micron
Buy ESEC / ZEVATECH: Micron 2 DIE ATTACHERS at CAE
until I saw a machine at the show in Hong Kong placing dies with a belt drive machine and aluminum extrusion frame. The head was swinging over the placement position visibly, but than after some calculations, apparently when X and Y swayed through zero, the head placed the die dead on the money.
$500,000 - $80,000
I also built a custom linear motor for a tape feeder experiment, but it is hard to compete with cheap motors used for printers or pager/cell phone vibrators.

However, this thread makes me think again, if I should get back into SMT or continue with CNC router.

What vision system are you guys plan to use?

[neilw20]

Neat heads.!!

Have a look at the placement heads:
2:07 and 2:27 is a good place to pause and browse.
[ame=http://www.youtube.com/watch?v=Va3Bfjn4inA&feature=related]How to make a Motherboard - A GIGABYTE Factory Tour Video - YouTube[/ame]

[Nails]

>> What vision system are you guys plan to use?

Semi-custom with opencv.

A 5mp camera module from Terasic

Terasic - Daughter Cards - Video & Image - 5 Mega Pixel Digital Camera Package

A camera controller interface board, the DE0 nano.

Terasic - FPGA Main Boards - Cyclone IV - DE0-Nano Development and Education Board

And a somewhat simple custom board with a socket for the DE0 nano. This board will have a hdmi connector on it to transfer the camera signals via high speed lvds. All three of these boards will be stacked. I originally thought it might be nice to make it cameralink compatible, but the the cameralink cables are over $150 each. I will settle for a $5 hdmi cable. I'll probably end up with another board for a controllable LED light source.

As for the main processing, I'm not using a general purpose computer. I'm planning on using a pandaboard and Linux. This should be here this week.

Pandaboard

I will have another custom board underneath the pandaboard connected to its general purpose memory interface expansion connector. This board will have sockets for two more DE0 nano boards. One to act as a frame grabber, and the other to act as a dedicated motion controller.

The pandaboard is nice because the user application can run on one of the two 1 Ghz ARM cores, and any vision processing routines can run without any OS overhead on the DSP core -- where they belong.

Needless to say, I have a lot of programming work ahead of me before I can even begin the physical build. A daunting task for sure, but I'm too far into it to back out and have just enough experience to believe I can pull it off.

[RomanLini]

...
What vision system are you guys plan to use?

None. I'm a minimalist. To me it's better place a lot of parts fast and easy (ie blind) even if the parts are a tiny bit larger and the PCB needs to be a tiny bit larger. I can't see any problem placing 0805 and SOIC fast and blind.

Who really NEEDS smaller than that? If it's a special case of a couple of fine pitch chips then I will just use a mechanical alignment pit as Flippersplace and others have done.

[Eclipze]

Due to product size constraints, most new designs of mine are based on 0603. Seems to work far better than 0805 when it comes to layouts with fine pitch components. Only some of the larger caps are 0805. Otherwise, 0805 and 1206 are used when power disipation is a factor. 0605 are a bit cheaper too. So I'm definitely keen to acheive 0603 placement without vision assisting... hopefully without the need to bump align. I haven't had much luck with solder paste and 0.5mm pitch TSSOP and QFN, but I'm not using an accurate enough or thin enough stencil for pasting with enough accuracy. I'd still expect to get good placement with bump alignment, which seems practical given the relative low quantity of these placed vs chip components.

The greatest difficultly I see is the PC side pick and place software.

[Robin Hewitt]

What vision system are you guys plan to use?

None I have left room at the top of the pick head stroke for a pair of component tapping jaws.

OTOH I don't require any vast speed.

[stewi]

You have an impressive list of hardware, Nails, which will require many lines of software code.
I guess, you'll have both a fiducial camera and a component camera and switch, multiplex between these two.
Are there any additional optics you can attach? If you search for the fiducials, than your field of view should be pretty much limited to ~6X8 mm, else a lot of other things on the board look like a fiducial. Than of course, you have to write the algorithm of pattern offsets from 2 and 3 fiducials and everything else related to frame grabbing. Where do you keep the fiducial and component library information? Do you have a host computer, or do you program your machine directly into the Panda board? Are all fiducials and components taught or can you program package dimensions and lead count, pitch?
Since you are going for ultimate, are you considering using bar-code reader for component reels?
However, you don't have to answer all these questions. I was hoping a webcam would do, free or inexpensive windows software. Anything else is beyond my capabilities and I would have to go with blind placements as well.

[Nails]

I'm quite keen on FPGA's. When using a ball grid array package, usually the decoupling caps are 0402 on the underside of the board. See attached pic which is the underside of the DE0 nano board. The pictures are from my usb camera, hand-held, so it could be clearer. The closer pic is about 1 inch away from the board. Each of the vias are 1 mm apart. That's why I am striving for vision. Without it, I'd certainly have to use a bump alignment pit, and even then, I still think you'd at least want a fudicial camera if not just to use it to set the board positions manually without any fancy software automation.

>> I guess, you'll have both a fiducial camera and a component camera and switch, multiplex between these two.

Yes. Each each one will be trasmitting constantly on a dedicated cable, but only the selected one will be buffered to RAM. Switching the cameras will be instant. I've heard there may be some time delay problems when switching between USB Cameras somewhere earlier in this thread. You can see the USB Microscope does give some really good images. I'd go with that instead of a webcam.

>> Are there any additional optics you can attach?
No. I've got the cameras already and tested them out. If I remember correctly, they were focusable to a field of view about 1.2 inches at about 2 inches away. There is some warping of the picture, but that can be fixed in software. As you can see, the USB microscope doesn't have that problem. A webcam will.

There will be no host computer. Everything will be done on the Panda board. Files can be trasferred to the Panda wirelessly.

>> Are all fiducials and components taught or can you program package dimensions and lead count, pitch?

I haven't done any application software development. I still have a ways to go with other critical building blocks before starting that. But it will be done whatever way makes the most sense to me. There is no limit as I am writing the application in a general purpose programming language.

>>Since you are going for ultimate, are you considering using bar-code reader for component reels?

That's an easy add on. Bar-code readers usually use a keyboard wedge that basically emulates the keyboard as if you typed the information on the barcode. There is no programming involved.

>> I was hoping a webcam would do, free or inexpensive windows software.

OpenCV is free for both Windows and Linux. USB microscope is cheap. But you will have to know how to write software in C to use OpenCV (OPEN source Computer Vision).

[Nails]

Check this out:

----------------------------------
From: "[email protected]" <[email protected]>
To: [email protected]
Sent: Monday, September 26, 2011 6:02 PM
Subject: MyUniversal Account Information


Your request for access to MyUniversal has been denied. You may contact
our Customer Support Center at [email protected] for more information.

Regards,
Universal Instruments Online
607-779-5000 (direct)
800-842-9732 (toll free in the U.S.)
--------------------------------

I wanted to get some dimensional prints for a lightning nozzle as they are really cheap (like approx. $30) to see if I could incorporate them for use in my design. I'd like to just be able to buy one that I know is made to tight manufacturing tolerances rather than make one myself.

Pretty stingy with their information if you ask me. I've got one ordered anyway, so I'll just have to break out the calipers.