TVM920 + OpenPNP Status

[BendRocks]

I had a kick-ass weekend that was about 20 hours in the shop without a single interruption. A short update on where things are with the TVM920 and OpenPNP.

https://youtu.be/Es060QcynB0

The machine is very usable to me now. There are some driver glitches that require a re-start, and a long list of things I will eventually do to make it more friendly. But the summary is that it's at a point where things are very usable for me. I build a lot of 1-off boards and will use the TVM to place the QFP and bypass and then most everything else I will hand place because the time it takes to hand place a single part if far less than the time it takes to machine place a single part. 

What sucks right now is centering each and every feeder before use. I have a hardware fiducial (located on the uplooking camera). That gets pretty close--probably 0.2 mm. Should be good enough. But each feeder has a bit of play when installed and each feeder is slightly different. So, the upshot is that you need to align each feeder before use. What I'd really like is the ability to to push a button and have the machine find all the feeders that are installed, grab the location and update the feeder entries automatically. If that existed, then it'd make sense to have the machine place a single part on a 1-off board as long as you have lots of feeders.

I'm in the process of replacing the nozzle holders on the TVM. In the vid you'll see the runout on the TVM nozzle holder is about 0.001". Definitely less than 0.002. The problem is that to use the Juki with this holder, QiHe requires you to install an oring. And it's just one o-ring. The nozzle can then rock in the holder. And the runout of the nozzle tip is whatever you want it to be: Nudge it a bit and you could easily get 0.4mm. And you'll really see this is you have a string of  bypass caps with 0 and 180 rotation: Instead of a straight line, it looks like a zipper.

It's also very time consuming to set up the nozzles. I'd like to do the silly putty thing on the first nozzle, and then have the up-look camera learn the offsets for nozzle 1-4 and do the math automatically. 

I'd also like to add code into the tvm920 driver config panel to let the hardware fiducial be manually tweaked.

There's a noticeable lag between operations you can see in the video. The TVM PC is slow, and what you see is running under the debugger and with logging off. But still, the machine has to think a bit before figuring out what to do next (eg. move to pick location, think, lower head, think, pick on, think, raise head, think, etc). It's very low on the list of things to fix (and it's definitely a driver issue). But it is a very big difference compared to the factory TVM. 

While I've posted everything on github, I want to caution against anyone reading this and thinking if they get a tvm920 it'll be easy to run right away under openpnp. You can do it with the github source, but I warn that things aren't super polished. 

Finally, the TVM920 serves as a very good template for a basic machine design that can be readily replicated. Having built several machines with 8020/steppers/DIY feeders, the more I study the TVM, the more I appreciate the design. We've all built things with bolted together 8020. And it's fine for learning. But I think there's a few ground-breaking things the QiHe folks figured out in a cheap machine. First, how to get a lot of CL feeders supported for cheap. Second, the inherent accuracy of this machine is very good. And it is derived from a thick base plate.rather than a bolt together frame. Seems trivial. but the more I study the more I understand how key this is (and I say this having made many bolt-together machines). 

If you start with machine tooling plate (a special type of cast aluminum with a very tight thickness tolerance--search on ATP5) you can readily mount rails to it and get the flatness required to meet the specs required by the rail makers. You also get a clean mounting for a bank or two of CL feeders. In the attached drawing is a concept I'm eager to try. It will use tooling plate base and belts. The backlash from the GT2 belts on the TVM is almost zero. To me, there's not much reason to do screws given the light loads. 

[Trampas Stern]

I would love to see how the TVM920 did the CL mounting plate for the air connections. 

[Jason von Nieda]

Great update Matt, thanks for posting!

You may find https://www.youtube.com/watch?v=brLDBEzPWy0 interesting. This shows how to use a script to setup and load feeders using QR codes. The script source is in the description of the video. May help you with what you are looking for regarding quickly setting up feeders. 

FWIW, I completely agree re: MIC6 base plate. I think this is the way to go for a "next level" machine when you need more accuracy than extrusion.

Jason

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/7a458bf7-485e-4bb8-aa09-fb7f35fd145f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[BendRocks]

Thanks for the link on QR codes. That's very cool.

What are your thoughts on finding feeder locations automatically? As a starting point, I'd like to find an empty pocket in an 8mm tape. And then once found, back up 4mm from that to get the next pick location. I assume this entails first defining an image pipeline that can find the empty pocket. The starting point will be known +/-2mm. Any other hints? I'm thinking there would be a button in the TVM920Feeder panel that would say "Look for feeders...." and then I'd enter a range of slots to tell it where to start. 

Any insight would be appreciated. It looks like the cut tape feeder might be a good starting point, though that seems to look for holes instead of pockets. 

PS. ATP5 is even flatter than MIC6, and pretty cheap. A 3x2' x 5/8" sheet is under $200. I'll report back if I do the experiment with rails on ATP5.

[BendRocks]

Here's some more detail. Take this more as a concept rather than the gospel as some measurements are likely fudged a bit. At the north end of the plate is a 6.1mm hole. This is the indexing hole. The air hole is at the bottom of the plate. That is a 6.5mm hole for the air pin, and then there's a gasket on top of that. The gaskets tend to get torn up a bit through use, but it's not hard to replace them.

Note shown is a small steel plate that covers the air holes. They have an M3 screw between each airhole to keep the gaskets in place. The plate is probably 0.5mm or so. 

The big plate thickness overall is 16mm as shown on the drawing. This is critical. The first batch of machines they sent out where 17mm thick. And everyone installed their feeders, and tried to push the handle down to the lock position and bent the feeder. Then they sent out new base plates with 16mm thickness. Those work great on new feeders, but my old feeders that had been on the 17mm thick plate are too lose to use on the 16mm plate. 

The air hose is connected by a simple M4 hose adapter. Usually these are 5mm threaded to screw into the plate, but you might have other sizes available to you. 

That's it. Pretty simple. Note how the feeder plate sits under the main plate. I think that ideally the feeder plate should be dropped another 2-3 mm with spacers to give better clearance on the TVM. They have the board height set to high IMO. 

Good luck on your build. The CL feeders are the S**T, wickedly consistent, and never fail. And the best part is the feeders can stay with you as you iterate through machines. 

[Trampas Stern]

I was working on my machine with CL feeders, I had to remove the feeders and adjust mounting point.  What I found is a lot of dropped parts on the base plate, and I realized that they could have easily fallen into the air hole. I was wondering if this was why the stock Yamaha design had the hole go clear through with two o-rings?  Of course if part fell into air hole, you could turn air on and blow it out... 

I love the CL feeders on my machine!  However as I tweak my mechanics I keep having to realign my pick locations, however the issue here is that my machine mechanics are not what they should be. Hence, I agree with BendRocks and something like an ATP5 base would have saved me lots of time and headaches. 

[SMdude]

Trampas, could be that your endstop trigger at slightly different locations.
Mine seem to do that a little. Then every time I restart my machine I have to check my feeder board and bottom camera location. I must set up fiducial homing...

[Trampas Stern]

Well I have been doing mechanical modifications.  My base plate mount for the CL feeders had play in it so I had to remove and reinforce, which means resetting pick locations. 

My machine is based on Anthony Webb's design and I have made motor and pulley brackets out of 6mm HDPE (and tried 3D printing).  The results is the plastic has to much give/flex: https://goo.gl/photos/5NyVhhvMGQmuuQ2P8

I need to mill the brackets out of aluminum, this also includes the head plate and such.  

I have also found that one of my NEMA 11 with the Juki quick release adapter has large run out (~1mm) I have clocked the adapter and gotten it down to around 0.2mm, I need to try clocking some more and see if I can get down lower, ie clock in a vise and test before installing. 

My limit switches are mechanical, and should be replaced with optical switches, but overall it is not the biggest problem.  Eventually I would like to have a homing that uses vision to align to feeders and PCB, that is hope to limit switches and then fine tune using vision kind of like LitePlacer does but then check the CL feeders and everything else.  

Generally what I have found is that I get compound tolerance errors stacking up, from cameras not being parallel to XY build surface, to heads not being perfectly square to XY build, to brackets flexing during moves, etc. The brackets and other machine flexing dynamics different when head is in motion verses head being stationary. That is I can move machine around and come back to same point when measured statically, ie stopping for several seconds machine and measuring location as compared to placing parts in fraction of a second. So I can run machine slower to get more accurate placement. 

With this said I can still place parts pretty close and up vision helps with part picking tolerances,  however I think to get the quality I want I will need to start with a more stable foundation.

So what I have learned is that you want to remove as much plastic as possible and use aluminum or other rigid materials. I even find that the 3/4 MDF that the CL feeders base plate and machine are attached to has play.  That is my design uses a table machine is mounted to:

Since the feeder base is not attached to the machine except through the 3/4 MDF it means with the CL feeders loaded (especially being long creating a moment) the MDF flexes and causes slight error between the machine and feeders. 

So again having a good solid foundation is important and helps keep things consistent. 

[Michael Anton]

You could probably eliminate most of the flex on that stepper mount, but putting another piece of plastic on the other side of the extrusion to support the stepper pulley, and idler.  That would really stiffen it up.  A single 3D printed part would probably work best, as it would also tie both sides together.

[Rich Obermeyer]

You are using 3Dprinted components as though they were steel  (or aluminum) plate.

3D plastic is not a good replacement for this use.  It will bend over time when under constant pressure.

Using 3D plastic to hold those idler gears shows how it bends over time and changes your calibration and alignment.

If you have to use this configuration, you will need steel (or aluminum) or even wood to retain.

Designing with 3D plastics requires design changes to compensate for its weakness.

Some simple additional cross members to support the plastic from opposing direction would help a lot.

I have been using PLA with carbon fiber built-in and it is significantly stiffer and just as easy to print.

--

You received this message because you are subscribed to the Google Groups "OpenPnP" group.

To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+unsubscribe@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/ea31a296-208d-4bb6-94e8-d62df8a95467%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Rich

[BendRocks]

Hi Trampas,

> I need to mill the brackets out of aluminum, this also includes the head plate and such.

You will find a night and day difference in CNC machining costs if you can describe your work to the CNC folks in 2D. In other words, if you deliver a DXF which is a plate with some holes in it, you can get a lot of stuff CNCd for very little money. I've had 20 pounds of aluminum parts CNCd (and entire machine) for about $350. And then as I refine the design, I draw on each part with a magic marker notes to myself. And then when I have another batch of changes ready, I just read all the notes off the actual parts, revise the design and do it again.

In your drawing, specify the size of drill you want used on every hole. If you specify a hole size that isn't a standard drill size, then it is a little more involved. So, specify all holes and standard drill bits. Tap them yourselves if needed. It's very easy to do with lube, tap and drill. Tapping is a "value added" charge on CNC usually and it's much cheaper to do yourself. 

2D CNC is so cheap because the guy programming the machine is just clicking on lines in the DXF and telling the software "Inside cut" or "outside cut". And they can make this part on the first shot.

3D, on the other hand, requires a person making $60/hour to sit in front of the computer and really understand the design and go into a lot of detail. They might have to make a few parts first, which means their expensive machine is tied up and the $60/hour guy is dialing everything in.

I've never really understood the allure of 3D printers. For my money, a laser cutter is a much better investment because I can cut an entire design out of 3/8" acrylic, make sure everything fits as expected and actually run the machine...and then when it's working as expected, take the same DXF and send it to the CNC place and get it replicated in aluminum.

> What I found is a lot of dropped parts on the base plate, and I realized that they could have easily fallen into the air hole.

Think of a drain on a sink:  Lots of stuff falls into the drain and it gets stuck in the trap (the loop of pipe). I don't think you have to worry about parts falling into the air hole as long as its not a straight shot to the control valve. The part will just get stuck in the trap (loop) and remain there until you decide to clean it out. 

Congrats on getting the CL feeders going!

[Trampas Stern]

Yeap I have a couple of CNC machines in my garage, I just need to power up and do the CAM on the designs and run them.  

On the CL feeder and parts falling in, when I looked at the stock design I wondered why the designed it with more seals than needed, my initial thought was a design like yours and so I could not figure out why they would cross drill the air holes and add more o-rings.  I usually assume when I see such things there must be a reason as it is rare a reasonable engineer adds something to a production design that is not needed and the part falling in the hole was the only answer I could come up with.