[stewi]

with a IR
camera (IR cutoff filter removed from the webcam)

Good idea. Will look into it.
IR creates quite different images. My objective is to differentiate between the component and the tape pocket, in order to pick the component in the center and recognize/correct rotation as well.

[nisma]

When you pick up a component, this component can rotate a small amount,
so it must be checked after pick-up.

[stewi]

When you pick up a component, this component can rotate a small amount,

Interestingly, you confirm the effect, I was mentioning due to mechanical component centering. Apparently, the nozzle vacuum creates an airstream like a low pressure tornado.
However, the Siplace nozzles (post# 565) have multiple chamber openings and did no longer twist the component during pick up.
In order to confirm my theory about the vacuum nozzle twister, components are usually right twisted here in the Northern hemisphere. What about you guys, down under?

[scsi]

Could the down looking camera locate the component in the tape pocket? Than you wouldn't have to inspect the component again.
Else, very nice.

Like nisma already said, very small components can rotate/move upon pickup. Still, the up-looking camera inspection is optional in madell's software. They heavily rely on the fiducial camera to locate the component on tape before the pickup anyway. Without it the cheap tape feeder produces crappy results.

The other important function of the up-looking camera is to verify that the component was in fact picked up successfully.

And another interesting feature I haven't seen mentioned here before is a use of mirrors instead of the uplooking camera for small component centering. Madell has such an option when they mount a pair of long mirror strips just in front of the feeders and skip the up-looking camera. Instead, the fiducial camera can see the reflection of the component on the nozzle when it travels from the feeder to the board.

[stewi]

Instead, the fiducial camera can see the reflection of the component on the nozzle when it travels from the feeder to the board.

Have you seen it working? The object is now in twice the distance than the fiducial and should be out of focus. Also, the camera must be tilted, board thickness and nozzle length should be taken in consideration.

[stewi]

They heavily rely on the fiducial camera to locate the component on tape before the pickup anyway.

That is what I doubt as well. Watching your video I see an offset in X of 0.00 Y 2.1 and theta 0.00. It appears the system is locating the tape pockets, but not the component in the pocket. Also, the offset of 2.1 mm indicates, that the vision looks at two pockets, which of course are not rotated to each other.
Nevertheless, it is working, but does not quite, what I was hoping it would be doing.

[scsi]

Have you seen it working? The object is now in twice the distance than the fiducial and should be out of focus. Also, the camera must be tilted, board thickness and nozzle length should be taken in consideration.

I haven't seen it working myself but remember seeing it on some training video that Madell ships with the software. They use not one but two mirrors each at 45 deg, so that you don't have to tilt the camera. The distance will indeed put the part out of focus, but in many cases it only helps with sub-pixel resolution. They even recommend not to focus cameras exactly on the object. I guess Dalsa's vision works better with subpixel. Camera resolution is only 640x480 but evidently it rresolves offsets much better than that.

[scsi]

That is what I doubt as well. Watching your video I see an offset in X of 0.00 Y 2.1 and theta 0.00. It appears the system is locating the tape pockets, but not the component in the pocket. Also, the offset of 2.1 mm indicates, that the vision looks at two pockets, which of course are not rotated to each other.
Nevertheless, it is working, but does not quite, what I was hoping it would be doing.

Well, that 2.1 offset is in machine steps, not millimeters. Actually, everything in their software is measured in machine steps, which some find unprofessional, but I think it's very DIY friendly. Each step on my machine is 20um or 0.02mm.

When looking for parts on tape, vision looks at the area in yellow rectangle and finds a pattern closest to the center and then sends pick up nozzle to that location. With 0201's it sometimes mistakes an empty socket for a component, but with 0402's and larger it works much more reliably. In fact, when the machine advances the tape and then cannot find the component on tape within the area, it repeats the advancement up to three times without attempting to pickup from an empty pocket.

[scsi]

Let me correct that. Just reviewed my video again and the 2.1 offset you are referring to is actually 21 steps. That was the distance of the object from center before correction that the vision found on tape. It's 0.42mm in this case and yes, those kind of offsets are quite common with this sloppy feeder system. That tape advancing rod tries its best to move the tape into desired position, but it does not work always exactly the same way becasue of varying friction, debris and stuff. In other words, the repeatability of this tape advancement method sucks, but is well compensated for by vision.

[flippersplace]

SCSI:

Your feeder video gives me many many ideas.. Thanks much for sharing. Your machine movement is very impressive and smooth and should be able to place at an acceptable rate

I have a used dual 8mm SIPlace feeder Fedex'ing its way to me now to play with but I think I will experiment with your solution as an extremely cost effective option as well...

I've made a few hundred boards on my machine now and I am well tired of manually peeling and pulling forward component tapes. It all works 98% of the time but I am not enjoying it.. So feeders are becoming more of a priority.

I'd be curious to see how well the advance mechanism works with plastic tape parts, ie SOT23's etc as I have always these found these more susceptible to jumping out of exposed pockets when advancing & picking

[scsi]

flippersplace,

I will try to find some SOT-23 on tape to see how the feeder works on them. Right now the closest part I can locate is a 0805 capacitor on embossed plastic tape. Should be very similar in behavior.

By the way, that pneumatic actuator I'm using for tape advancement deserves some attention. Madell uses some pretty high-quality guided actuators from SMC for this purpose but I was wondering if a regular cylinder with spring retraction will do the trick. Went to Grainger store and picked up one of their hous brand cylinders with 2" travel for $12, mounted it on the machine and sure thing, it's been doing the job very well for about two yers now. Needless to say this tape advancement method heavily relies on vision assistance.

The rest of the machine is super high end and it can move ten times faster than that on the video. Up to 2m/s if I'm not mistaken. 1 tonn mass helps a lot with stability, but it's not a requirement for a primitive PPM such as Madell's SX series which I'm copying. Their smallest machine weights only around 40 pounds and you better bolt that to a table of some kind.

[scsi]

Here's a couple more videos with smallish SOT-323 (smaller than SOT-23)packages on embossed tape as well as 0805 capacitor, also embossed. Actually these are screen captures that show the output of the vision system. The challenge is the same as with 0201. Parts fly off the tape if the region of tape that holds exposed parts is subjected to shock. To solve the problem we need to isolate this region of tape with some kind of damper that would prevent the shock from getting into the region. Simplest solution is the same as I did with 0201 on previous video. Two pieces of paper tape 0.45mm thick stuffed under the crossbar. Looks ugly, but no more flying parts.

2stewi and others who pay attention to details. On these two videos, the numbers on the top of the screen appear two times for the same part. First when the top camera looks for the part on tape, the coordinates indicate current offset from programmed pickup location, which will then be used to correct the pickup action. Second time, when the video switches to camera 2 (up-looking), the numbers will again indicate the offset, but this time the values are significantly smaller. Typically around 1-3 machine steps, which is approx. 20-60 micrometers on my machine. This means that the pickup sequence is actually grabbing the part very close to its center regardles of the offset within the pocket, as expected. It is also interesting to watch the last two SC-323 diodes being picked up from tape. The parts are severely rotated in tape pockets but vision recognizes the pattern pretty well. Note that the video is tuned out of focus for better results with subpixe resolution.

As usual, fullscreen gets you higher resolution. In this case 480p...
[ame=http://www.youtube.com/watch?v=gMQHy7xEtto]0805 capacitor embossed - YouTube[/ame]

[ame=http://www.youtube.com/watch?v=1FPzvkMrz4k]SOT 323 embossed - YouTube[/ame]

[Eclipze]

That's pretty good!!!

[stewi]

I have a used dual 8mm SIPlace feeder Fedex'ing its way to me now to play with

PM me and I can likely give more info.
Siemens feeders run on 24 VAC (rectifier inside the feeder).

[stewi]

Parts fly off the tape if the region of tape that holds exposed parts is subjected to shock.

That is indeed a very nice video and machine.
Still not happy: when the camera inspect components in the tape it should also detect accurate location of the perforation holes. With it, the advance pin could enter the perforation hole dead center not creating any vibration on the hole edges. After tape advancement the moving axis should go in opposite direction for 0.1 mm or so before the pin retracts, preventing tape upwards movement and components flying.
Forcing the tape downwards works in your case with paper tape. With plastic tape, there is the risk that the pockets deflect upwards and eject the component. You should have a groove in the feeder plate and leave the pockets unsupported.
Also, the tape should be supported from underneath and not from the top in order to keep the z-height consistent. If you support the tape from underneath with a leaf spring, than the spring should have two tongues supporting the side of the perforation holes and 0.7mm wide on the other side not touching the tape pockets.

Let's see what flippersplace will tell us about his feeder experiment and I can find some more.

[scsi]

...it should also detect accurate location of the perforation holes.

That's an interesting idea. I have been thinking about it as well. Will ask madell if they can implement something like this. They've been pretty open for suggestions in the past.

Also, if you look closely on 0201 video, you will see that I have the groves for pockets machined in the feeder guides and the improvised paper spring forces down only on the edges of the tape and the pockets remain unsupported. Will see if I can find sutable and more permanent damper solution to replace the paper tape padding. Thinking about making some bronze leaf springs...

[jcdillin]

Hello All!

I'm glad I found this thread, I'm in the process of finishing up my own pick and place machine. Currently my only items left on the to-do list is nozzles and some sort of feeders. I was curious before I reinvent the wheel if anyone had some drawings of the simple drag style feeders they could share?

I'm currently just thinking of buying the maddel nozzles as they seem pretty cheap and easy to use.

I've attached a photo of my machine as it stands at the moment.

[scsi]

I've attached a photo of my machine as it stands at the moment.

That machine looks very promising! The construction looks similar to madell's PX series. Great to see you are using industrial cameras in there. Is that Tamron lens I see? I love those little tamrons and also use them on my machine - 16mm, 25mm and 35mm. What kind of frame grabber and vision software are you going to go with? Writing software yourself?

Madell's nozzles do work fine. They mate with a standard 5mm stepper shaft nicely, but yours doesn't seem to have enough shaft exposure at the front. You will need about 20-25mm of bare shaft sticking out at the business end for the nozzles to mate properly.

I will see if I can dig out my drawings for the feeder strip later today. What CAM software are you using there? I can either get you DXF's or Vectric Cut2D CAM files.

[jcdillin]

Hello scsi

I'm not sure who makes the lens, it came with the camera in a lot of machine vision stuff I bought. I currently am using a Leutron vision PCI card. I have a friend that is working on a software package for it that will work with the grbl motion control code.

I currently have 22mm of shaft sticking out, it's just a standard nema17 that I drilled out.

I currently use cambam so dxf is fine.

Thanks!

[scsi]

Okay, here's my feeder drawing attached. The CRV file is from Vectric Cut2D, you can download their free trial and open the file to see the pocket depth for various feeders. The smallest feature can be machined with 3/16" endmill. I'm using 3/8" thick aluminum stock material.

This design includes 12x 8mm, 4x 12mm, 2x 16mm and 1x24mm feeders on one 12" long bar and can be seen in this video below.

[ame=http://www.youtube.com/watch?v=snNU85SP7dw]0201 resistors on homemade pick and place machine - YouTube[/ame]

I just went and measured the exact length of the stepper shaft that is sticking out past the magnet ring and it is 16.5mm. I think you'll be fine with Madell's nozzles as long as you use fairly slim magnet ring on the shaft. Their nozzle holder (brass part) has a depth stop at what appears to be 16mm, so that each time the nozzle is changed, it would sit at the same exact height on the shaft. Also, it's good to leave an air gap between the ring magnet on the shaft and the butt of the nozzle holder, so that they don't touch each other. If they do, the magnetic coupling force will be quite substantial.