Pick and Place Machine TVM802A / TVM802B

[thommo]

Great progress Axel.
Tell me please, is progression of the 'tape' dependant on the 'peeler' effectively pulling it forward, or is the another mechanism to advance the tape which I'm not seeing?
Eg - is there some kind of drag pin positioned in the head unit?

[Bud]

I watched these videos and Youtube recommeded this one:

https://youtu.be/IvUU8joBb1Q

We need this guy to build pick and place machines using that novell concept to feed and drop parts
 

[Koen]

make a concrete slab to bolt it too.
Problem solved.

Or buy a nice block of granite. It's plane, clean, heavy, resistant, stable over time and over temperature changes.

[uncle_bob]

make a concrete slab to bolt it too.
Problem solved.

Or buy a nice block of granite. It's plane, clean, heavy, resistant, stable over time and over temperature changes.

Hi

If you are actually going to spend real cash on a bench top:

http://www.grizzly.com/products/Maple-Butcher-Block-60-x-24-x-2-1-4-/T21247

Is a pretty nice way to go. The stuff looks good. You can drill holes in it to run wires through. It's incredibly sturdy. There are a *lot* of people who sell the stuff, so shop around. The thing to watch is the thickness. You probably don't *need* 2.25" thick, but why not ...

Various people will also sell you "two thousand pound" industrial leg sets to go with a top like that. It then becomes a DIY table kit. With some help on the heavy lifting, you can get it together quite quickly.  I prefer fixed height legs over the adjustable ones. You always need a back brace to go with the legs.

Bob

[ServoKit]

@thommo: There's a prick in the head that comes down on the arc shaped spring to release the tension on the tape. As far as I can see (without having actual reels) the tape advancement comes solely from pulling the cover tape. Interestingly my machine has two motors for this, one on each side of the reel rack (the CHMT48VA has only one). I hope to have parts on reels later this week and will make a close-up video of the mechanism.

Addendum after closer inspection: The prick comes down a little in front of the spring, presumably into a hole of the tape, the head is then moved away from the feeder effectively pulling the tape, at the same time the motors for pulling the cover tape engage. Then the prick comes up again and the nozzle moves over the tape to pick the part. Video will follow.

Regards, Axel

[sedelman]

@ServoKit Sigh, these are again carefully choreographed demonstration videos that I would expect to see from a manufacturer. It would really be interesting to see how the machine handles components that are not perfectly centered in the tray (as is the case in your video) and not pushed against an edge but at odd angles in the tray. Let's see this machine properly make an angle adjustment and place the component correctly. Please?

[ServoKit]

@sedelman: The second video was a test of the "Quick" vision algorithm; this will only work if the parts are already properly aligned, i.e. not skewed in the tray. Also, the very small SOIC-8 parts (which would usually be on a tape) have to be properly located in the respective tray pocket or the nozzle can't pick them up.

Regards, Axel

[ServoKit]

Clarifying on my previous post re. the trays.

The tray pockets are roughly 20x20mm. For each tray you specify the location where the nozzle should come down. One option is to use the center of a pocket at x = 10mm, y = 10mm, and place the parts accordingly, like so:



Ideally this should let the nozzle reliably pick up everything that's put in the proper center of the pocket. However, it's easy to misplace parts this way, especially smaller ones. Only 2 mm to the right or so and the pick can fail.

Better use a point near the top of the pocket, like x = 10mm and y = 16mm. Then you can simply shove the parts towards the edge, like so:



As a bonus this placement will mostly eliminate the need for rotating the part for alignment; only left is adjusting for x,y deviation which is much faster.

Of course, eventually we will use custom trays with proper cut-outs of the correct size (the position, layout and dimensions of up to 10 trays can be specified in the software).

Regards, Axel

[uncle_bob]

Clarifying on my previous post re. the trays.

The tray pockets are roughly 20x20mm. For each tray you specify the location where the nozzle should come down. One option is to use the center of a pocket at x = 10mm, y = 10mm, and place the parts accordingly, like so:



Ideally this should let the nozzle reliably pick up everything that's put in the proper center of the pocket. However, it's easy to misplace parts this way, especially smaller ones. Only 2 mm to the right or so and the pick can fail.

Better use a point near the top of the pocket, like x = 10mm and y = 16mm. Then you can simply shove the parts towards the edge, like so:



As a bonus this placement will mostly eliminate the need for rotating the part for alignment; only left is adjusting for x,y deviation which is much faster.

Of course, eventually we will use custom trays with proper cut-outs of the correct size (the position, layout and dimensions of up to 10 trays can be specified in the software).

Regards, Axel

Hi

One alternative to fully custom trays are drop in inserts for the trays you have or something very much like them. More or less: A rectangle that fits the opening with one corner chomped out for the part. The stock trays may or may not be suitable for this sort of insert. Unless you have a very set mix of parts, it is a lot easier to organize inserts than to order up fully machined multi cavity trays for each job. The main downside of the insert approach is pulling the setup off the machine and storing it ... not so much.

Bob

[mikeselectricstuff]

Those trays are pretty big -can you redefine the tray locations to use a more densely packed tray     or strip  of taoe? Can you define arbitary tray/strip locations in the placement  area?

[ServoKit]

@Mike: Yes, you can define up to ten trays with arbitrary layout (10x1, 1x12, 5x3 etc.) and size within the work area. When the software is running it keeps track of a "tray counter", takes chip from designated tray for PCB #1 from first pocket, chip for PCB #2 from second pocket etc. Strips should work too if you can keep them straight and flat. - As a test, I will 3D print a tray with pockets for the TSSOP-28 we are going to use later this week.  And here's something where the mfg. could easily improve the machine: add a grid of threaded holes to the work area to fix up stuff.

@Bob: Inserts are another option, I could problably cut them out of heavy stock with my Cameo.

Regards, Axel

[sedelman]

@ServoKit It would be nice to see how the machine behaves when parts are aligned non-optimally. I realize the trays provide a form of mechanical centering/alignment when you place the ICs against a corner or edge, etc. but a NeoDen TM220A without vision can do the same. Can you show us how fast the vision works (and if it works) when ICs in the tray are rotated 45DEG in the tray? Many thanks for your consideration.

[ServoKit]

@sedelman: here you go, unedited, except for reencoding:

http://www.servokit.com/images/Produce.mp4

Vision at accurate, effective speed 30%.

BTW, if you think about it: 45 degrees or more won't work in this configuration (looking from bottom). The system tries to rotate the chip to fit a rectangle which sides are parallel to the X/Y axes of the machine. If the chip is at 45° the machine can't tell what side goes where (Likewise the software doesn't know where Pin 1 of the chip is. In effect the chip must be in the tray already very close to the desired orientation.)

Addendum: In the part list you can specify a rotation angle but that is applied after the chip has been aligned.

Addenum II: It seems to me that ideally you wouldn't have to use vision at all, save for locating the fiducials (something the TM220/240 can't). Every trip to the camera costs time; a bespoke tray would eliminate this need. Vision would be a last resort, like when you only make a few PCBs where a custom tray can't be justified.

Regards, Axel

[H.O]

Question:
If the vision system always orients the part to 0 degrees is it then able to place it accurately on the PCB at an aritrary angle?

The obvious answer is yes of course but here's what I'm thinking about:

The part is picked up "off center", ie the nozzle isn't centered on the part. The part is moved to the camera where it gets centered up and oriented to 0 degrees. But now, when it's rotated to the arbitrary angle it's supposed to be placed at on the PCB the parts center "wobbles" since it isn't concentric with the nozzle.

Does the vision system and control software compensate for this wobble or am I overthinking this?

[uncle_bob]

Question:
If the vision system always orients the part to 0 degrees is it then able to place it accurately on the PCB at an aritrary angle?

The obvious answer is yes of course but here's what I'm thinking about:

The part is picked up "off center", ie the nozzle isn't centered on the part. The part is moved to the camera where it gets centered up and oriented to 0 degrees. But now, when it's rotated to the arbitrary angle it's supposed to be placed at on the PCB the parts center "wobbles" since it isn't concentric with the nozzle.

Does the vision system and control software compensate for this wobble or am I overthinking this?

Hi

Unless you rotate the part over the vision system and then do a "second look" at it, you will get a skew from an off center pick followed by a random rotation. Simple answer: don't lay out the board so giant / precise parts are rotated 45 degrees.

Bob

[ServoKit]

@H.O: Let's assume the part is oriented properly, i.e. unskewed, but shifted by some millimeters in the tray. The job of the vision system then is to find the offset of the part's center to the center of the nozzle. When this offset is known, the software can a.) correct the path to the part's position on the board by this offset and b.) adjust the center of rotation accordingly if the part has to be rotated before placement.

From a programmer's standpoint this would probably come down to simply concatenating two matrices, one translation matrix for the offset and one rotation matrix that contains the part's desired angle.

Regards, Axel

[ServoKit]

Some notes about adjusting vision contrast and max. usable part height:

http://servokit.com/blog/2016/03/07/notes-on-vision-contrast-fiducials-and-max-part-height-with-the-tvm802a/

Regards, Axel

[H.O]

Thanks, I understand that it's possible to compensate (it's "just math" right), question is does this machine do it?

As for not laying out boards with parts at 45 degree angle....I think it's fairly common to do exactly that with larger parts (TQFPs etc) since it can make the fan-out easier. Besides, it doesn't have to be 45 degrees or some arbitrary angle. If the vision system always aligns the part to 0 degrees then you'll get the possible COG shift if it's rotatated to 90 (or 180 or 270) degrees as well, right?

But yes, surely the software must compensate for this.

[ServoKit]

I just tried this out and here's what the machine does:

It picks the part, rotates nozzle to apply the desired rotation, then moves to the camera and does the aligning, then places the part on the PCB. Even if the part is shifted several mm in the tray this results in correct placement.

A 45° rotated part would have to be placed either close to square into the tray and have the rotation applied or could be put in a custom diamond shaped tray pocket with no further rotation.

Regards, Axel

[mikeselectricstuff]

Some notes about adjusting vision contrast and max. usable part height:

http://servokit.com/blog/2016/03/07/notes-on-vision-contrast-fiducials-and-max-part-height-with-the-tvm802a/

Regards, Axel

re. the red resist - fiducials should have a resist-free area around them ( to avoid a misaligned resist overlap ofsetting the position) so shouldn't be an issue if your fids are done properly and the machine's vision is sensible
 

[ServoKit]

Yes, besides the one hole being to close to the fid, the res-free circle around both fids is very small and the resist layer is misaligned by at least 2/10mm so that the fids don't stand free properly.

Regards, Axel

[sedelman]

@ServoKit thanks for providing the video showing the vision system in action.

I am sure that you looked at the CHMT48VA as well as the SmallSMT offerings. What motivated you to select the TVM802A over the other choices?

[ServoKit]

@sedelman:

CHMT48VA

1. It has more of the large feeders (16 & 24). That you can define your own trays on the TVM802A makes partly up for that.

2. It has encoders on the steppers. However, I reckoned that for the small boards we make plus having fiducials on each PCB losing steps wouldn't be much of an issue (Let alone that encoders on the steppers are NOT a true closed loop, you can still have slippage with the belts etc. What you really want are optical encoders on the rails, only then does the software know where the head is at any time; AFAIK none of the sub-10K machines has this.)

3. It runs stand-alone while controlled with the touchscreen. Not sure if this is really practical, would love to hear from someone who has used this.

SmallSMT

Judging from the photos the machines look sound. Didnt't look too much into that as I was confused about the offering, which is also apparent in the Neoden 4 thread: Sometimes the machines are called SmallSMT, then Yushengtech, then AutoSMT, then VisionPlacer. Who exactly is the manufacturer? Who do I talk to if I have a problem? Why the flip is the software dongle protected? Confusion is not good when buying big item tickets.

Regards, Axel

[rwb]

@servokit I have been thinking about the Charmhigh machine vs your machine which also has vision.

From the looks of it you don't really need the built in LCD screen if you can do everything on the PC.

The Charmhigh does have encoders on the steppers but I'm not sure how beneficial that is with the belt driven machine. I do have a belt driven 3x4 foot laser cutter machine and its very precise using belt's which have been on the machine for years now with zero maintenance. I'm sure the encoders can't hurt though.

Other than that it looks like they machines are almost identical other than a few larger feeders.

For the price Axel paid I have to say it looks to be a bargin now that the machine has vision built in.

Only time will tell how good and reliable it is once it Axel puts it to work building tons of boards.

I'm just curious how capibable these machines are at placing .5mm pitch 32 pin TSSOP chips cause I have a few of those that would need to be placed.

[ServoKit]

Yes, we have two 3D printers in the office, both belt driven, no problems. Losing steps is more of an issue with machines where the head exerts physical force, like with a router or mill. But then again, you would use leadscrews rather than belts on that kind of machine.

Regards, Axel