Pick and Place Machine TVM802A / TVM802B

[protoneer]

- fiducials with no solder mask around are undetectable by the down-camera. Thought it was a good practice but now ended up permanent-marker-filling in the space around the marker metal.

Thanks for the nice update. With the  Fid's ,what kind of plating did you have on your board?

[Smallsmt]

Report from the trenches:

Populated 6 "half-PCBs" today (just one side, hence half). 130 components populated per board, majority of them 0402. Learnings:

When the process is down, it works quite well. Last 2 PCB's had only 1-2 rejected pickups and the last PCB took 8 mins to populate, including a few hand-placed parts. That could probably be even faster running at full speed (I'm at 70) and actually using both nozzles (machine comes with only one "0402-compatible" nozzle, so it's mostly a single nozzle show for now).

Notes:
 - up-facing vision system is useless for 0402 components, at least for caps. In fact, it actually messes things up, badly, due to incorrect angle detections. Once I turned it off for 0402 components they were placed like a charm (once alignment was good, see below)
 - nozzle alignment is absolutely key for each and every reel. Even for bigger components, misalignment means occasional needle jams. For 0402, it's the difference between perfect repeatable placement and components placed on their side or 50m across the lake from the board
 - digikey digireel is shit and has a very good chance of breaking the machine - the connection point between the cut tape and the "header tape" often jams, and with thick enough tape can probably bend/break the needle.
 - I'm mystified by how the tray counting algorithm works but it works
 - CSV import from Altium: lots of magic. I ended up creating a spreadsheet to translate CSV directly into machine-compatable CSV, specifying nozzles, speeds, and so on. I imagine everyone out there will end up making some version of this for themselves
 - occasional 90d rotations - maybe because things get juggled at pickup? Not quite sure but a quick inspection before baking helps detect and correct these (happened to 3 components out of the lot)
 - the fucking tape collection system - it's the problem everytime, after alignment. Basically if you did the alignment but things are still wrong, it's 90% a problem of tension in the tape collector.. or the used tape not running smoothly out. Eventually some magic tiny turns to certain friction reels seem to have resulted in a well-working system.. for now.
 - fiducials with no solder mask around are undetectable by the down-camera. Thought it was a good practice but now ended up permanent-marker-filling in the space around the marker metal.

My stenciling needs improvement. Using T3 paste right now and it's really either all over the place or barely on the pad. Feels like it's too thick.  Ordered T5 paste, hope that improves things.

Overall once everything is setup it's a wonderful thing to watch. Small-leaded components are no problem so far. Watch out for those digi-reels. Rigging takes forever.

Hope this is useful. Pictures of the setup once I regain consciousness.

Your fiducial detection problem is maybe caused by your ambient light. Don't place a light above the machine and test under different light conditions.

[Koen]

Hello, I'm time-constrained and ordering a TVM802A would mean starting production tests earlier than assembling one, learning openpnp and vision stuff. So, I'd like to ask the owners : if you had to redo it, would you buy the TVM802 again ? Thank you !

[mikeselectricstuff]

Hello, I'm time-constrained and ordering a TVM802A would mean starting production tests earlier than assembling one, learning openpnp and vision stuff. So, I'd like to ask the owners : if you had to redo it, would you buy the TVM802 again ? Thank you !

Any new machine or process is going to involve significant time to master the process. The only difference between cheap Chinese stuff and the traditional players is the latter will usually do on-site installation and training.
Of course time spent working around badly designed and/or poorly documented software doesn't help but that's going to be the same for all the Chinese machines.
   

[ServoKit]

@Koen: Yes.

Regards, Axel

[sergeil]

Yup, it's not perfect, but it does a great job for the price. The only thing that would change that is if it went up in flames tomorrow because of shoddy parts/assembly, but from the parts I've seen it's pretty solid.

[protoneer]

My machine finally arrived this afternoon and here are a few photos of the winders that drive the reels. The Clutch system now makes more sense...

[ar__systems]

A brief update on our protocol reverse engineering effort.

The controller turns out to be not so primitive after all, at least not as primitive as the software

It allows independent control of acceleration, deceleration, and max speed on 6 axis. (X, Y, Z, R1, R2, and strip motors.) So yes, 18 independent parameters. Which is kind of nice. I think acel/decel values are even more important than speed. And we can change them at every step. I.e. move fast accel fast when moving empty and slower when carry part. Original s/w links all three together into a single control.

Also for Z we can accel fast moving down but need to decel slower to avoid losing the part. And moving up, vice versa, we need to accel slowly to keep the part on, but can decel fast.

I don't have time to write up the entire protocol, so I think we'll just publish the source code for our test program. We'll just clean up and organize it a bit. Should be able to post it next week. There are still bits and pieces in the protocol that we don't know what they do (seemingly nothing) but what we do know allows full control of everything that is required to start working on the actual placement software.

corrected.

[ar__systems]

My machine finally arrived this afternoon and here are a few photos of the winders that drive the reels. The Clutch system now makes more sense...

Lucky you have those plastic inserts. Mine came with rubber inserts that provide way, way too much friction. I had to take them apart and leave maybe quarter of the circle to make it more or less reasonable.

[Royce]

There are still bits and pieces in the protocol that we don't know what they do (seemingly nothing) but what we do know allows full control of everything that is required to start working on the actual placement software.

Awesome work!

[ServoKit]

I have the same inserts as protoneer. Seems to be another difference between A and B.

Regards, Axel

[protoneer]

I don't have time to write up the entire protocol, so I think we'll just publish the source code for our test program. We'll just clean up and organize it a bit. Should be able to post it next week. There are still bits and pieces in the protocol that we don't know what they do (seemingly nothing) but what we do know allows full control of everything that is required to start working on the actual placement software.

Looking forward to seeing it. If you put it on Github we can start helping with the documentation part...

[uncle_bob]

My machine finally arrived this afternoon and here are a few photos of the winders that drive the reels. The Clutch system now makes more sense...

Lucky you have those plastic inserts. Mine came with rubber inserts that provide way, way too much friction. I had to take them apart and leave maybe quarter of the circle to make it more or less reasonable.

Hi

Consider some sort of lubrication if you have to much friction. The tough part is finding something that does not mess up the rubber or wind up all over everything in sight.

Bob

[ar__systems]

Consider some sort of lubrication if you have to much friction. The tough part is finding something that does not mess up the rubber or wind up all over everything in sight.

Tried, does not help much. I'm in the process of completely replacing cover tape collector.

[ar__systems]

 - fiducials with no solder mask around are undetectable by the down-camera. Thought it was a good practice but now ended up permanent-marker-filling in the space around the marker metal.

That is strange, for me unmasked fiducials actually works the best. Have tried adjusting threshold in the PCB setup? Are you using FR-4 boards or something exotic? Masked fiducials is not a good idea since the camera can't distinguish between circle of the fiducial and circle of the mask. It will lose some accuracy.

[protoneer]

Did my first placement last night and I am pretty happy with the results.



Results of placement attached.

I have placed all of my production boards by hand up to now. My intention with this machine is increase volume and to free up some time. It has raised an interesting situation for me.

When manually P&P'ing a proportion of time is spent on making sure parts are placed correctly. With these machines focus have been shifted and I can see a need for extra quality control.

This has pointed me to Automated Optical Inspection for:

• Checking solder paste coverage
• Missing/misplaced components before reflow
• Misplaced components after reflow
• Checking obvious shorts after wave soldering

Found a really nice blog on the topic by Bob Baddeley
http://bobbaddeley.com/2015/12/creating-an-automated-optical-inspector-for-50/

The python script works with OpenCV and brings up a life video stream that magically focuses on issue parts.
(Needs a small amount of training but works amazingly well)

[Bud]

On machines of this class is changing reels painful and requires any re-alignment or is it easy (when building two sided boards) ?

[ar__systems]

Regarding the pressure sensor in the machine. I initially wrote that it has a mechanical pressure switch, but I was wrong. It actually has an electronic pressure switch. Then I'm wondering why is working so unreliably... Maybe the threshold is set incorrectly. I wish they transmitted digital pressure value to the PC, but alas, they only send the result of comparison with the setpoint.

[protoneer]

On machines of this class is changing reels painful and requires any re-alignment or is it easy (when building two sided boards) ?

I have only had a few hours play and doing double sided boards should not be an issue. On the board side alignment is done with fiducials and it works really well. (Jobs can be stopped half way if a nozzle change needs to be made and the machine will just do a fiducial check on restarting.)

As for changing reels. Once you understand its working it should not be to hard to change. With my first few runs I noticed that the nozzles kept picking up resistors off center. The placement as seen above is okay for 0805 but could be better.

Yesterday I played with a reel-less off cut of Sot-23's. Had the same issue with the component not being picked up in the centre. With SOT23's misplacement was a bigger issue and not acceptable. Played around and the problem ended up being the offset between the nozzle and the needle that advances the reels.

The files that the machine saves includes all of the config and offsets settings. As I play with , it is starting to get more and more accurate.

I use Kicad and getting the component positions out is pretty easy. Things to consider is the component orientation in the tray.

Micro's are also not that bad to work with. I played with an ATMEGA328P-AU (0.8mm pitch) at an 45degree angle. The camera system aligns well but as mentioned before the system is very sensitive to light sources. I am thinking of making shades for the bottom camera as soon as the machine has been moved to a permanent spot.

[ar__systems]

The files that the machine saves includes all of the config and offsets settings. As I play with , it is starting to get more and more accurate.

A bit of correction: the placement file only contains locations of the fiducials. (I mean, obviously, aside from actual components). All of the offsets and other data, such as camera-to-nozzle, nozzle-to-needle, location of the feeders and everything else, the s/w stores in the n/v memory of the machine, which has space to store at least 400*4 bytes of data. Maybe more, but the s/w uses about that many.

[protoneer]

I have attached an example file that the machine software exported/saved. (For example it includes the nozzle-to-needle offset at 1.60. So it looks like these settings get sent to the machine when the files gets loaded)

Actually makes sense cause things like exposure settings could change for board colors and should be saved as part of the config.

I am starting to like this little machine...   

[ar__systems]

I have attached an example file that the machine software exported/saved. (For example it includes the nozzle-to-needle offset at 1.60. So it looks like these settings get sent to the machine when the files gets loaded)

Actually makes sense cause things like exposure settings could change for board colors and should be saved as part of the config.

I am starting to like this little machine...   

Nozzle to needle offset is two coodinates, not one 1.60 is a magic number that nobody really knows what it is. It is also 1.60 in the files that I generate and I copied it from mathias's blog

You can try and change numbers in the System Config, where actual nozzle-to-needle is located. See if it will change in the file you save

You can change parameters in the sys config and not save any file at all and yet they will be restorered next time you open the program.

[protoneer]

Haha... Lots to learn and discover.

Tonight I will try some old boards without fids. Hopefully I can use a pad om some kind to align it.

[ar__systems]

Guys, how is the focus on your down cameras? I just noticed that the focus get significantly better if PCB is lowered a bit.

[alexanderbrevig]

random thought but 1.6 is a pretty common pcb thickness...?