Pick & Place MachineTVM920

[anfang]

Hi Thommo,

In the exchanges above, note where I asked "This format isn't working" and the reply came back from dtf saying "This format I'm pasting here works"...that exchange was simple and direct. I had a similar exchange with the software leader at QiHe over 3 days and approaching a dozen messages. It went in circles over and over, with him asserting that I couldn't modify anything in excel, under any circumstances, it all had to be done in Altium. I asked the question many different ways. I gave examples. I went sentence by sentence. I wrote very short mails.

Some background: I've traveled to Asia more than 130 times (Japan, China, Taiwan) for business and I work on very complex projects with folks over there routinely--all great engineers. I used to think that I could communicate in the language of EE...but I'll tell you it is impossible if the person doesn't have the base language skills. In finally gave up with the QiHi guy and found this place.

I'm afraid the language skills aren't where we need them to be at QiHe for complex technical engagements--but I could be wrong. It's certainly worth a try. But if a person cannot answer a direct question of "what is the file format required for this file?" then I think everything that comes after that is borderline hopeless.

If we can understand their file formats, then we can create an external program (as Harry has done) and make the QiHe software very usable. This should be fairly straightforward. I'll likely go this route regardless of where others head as I need something short term.

If concurrent to this they could make the feeder setup automatic (such that the machine would look for feeders pockets automatically and populate the lists) that'd be awesome.

Ultimately, we need to learn the protocol and ditch their software. It is primitive beyond words.

I hope I'm wrong. But I'm slowly setting aside some time to do the work I need to make this machine useful.

If someone wants to put together a list of asks so that we can vote on them, here are my 3 in order:

1) Explain the machine control protocol. Looks like everything is done over TCPIP and 2 vision inputs.

2) Make the current software smarter so that it can discover the location of feeder pockets on front/back automatically and update the locations in the config. This should be easy. Find a sprocket hole is already happening automatically it looks like.

3) Explain the various file formats

[ar__systems]

I'm afraid the language skills aren't where we need them to be at QiHe

Hehe in my experience in order to make sure people in Asia understand what I'm telling them, I need to repeat the information 3 times using different words.

Unfortunately language barrier is really a barrier. I ordered some CNC parts from China and the girl from there was calling me periodically to follow up. But even though she manages with written English, her spoken Eng was non-existent. I felt so sorry for her, as she was trying.

Back to the topic, is the format of the 920 so difficult to understand? For 802 the format although cryptic, it is fairly simple to reverse.

[mrpackethead]

The language barrier is not that hard. You could learn Mandarin.

[thommo]

Hi 'anfang',

Can I ask who you've been speaking with at QiHe on these topics?

Although we understand the frustration, I must say [albeit with persistence] we've received good results from them to date. We have the benefit of having a Chinese speaking staff member, but this hasn't been required with QiHe so far.

If we all fragment, the amount of effort [individually] will be great. If at all possible, especially now that the 920 numbers are starting to add up, it would be great if we can get QiHe onboard and to take this on so that everyone can benefit.

Cheers - Peter

[anfang]

Can I ask who you've been speaking with at QiHe on these topics?

I just sent you a PM with more info as I don't want to get the guy in trouble. He's very sincere and tries very hard and he's very prompt. It's just that the info that comes back isn't helpful or relevant.

Although we understand the frustration, I must say [albeit with persistence] we've received good results from them to date.

There's no question the hardware these guys have built is phenomenal. I have just loaded a feeder with 0402 caps and the feeding on 0402 is flawless. Placement is a bit wobbly and I still have a bit more diagnosing to do there.

On the software, I've posted twice now my list of wants. If anyone else wants to post their wants I'm happy to have us rank these as a group and pick a person to approach QiHe and get the features scheduled.

[ttsthermaltech]

What I found on my TVM920, was that the calibration was off by a considerable margin. At this point, there were no abilities to re-calibrate the steps/mm for X and Y, but they should be in your software / firmware. What I did was went and bought a precision metal ruler, and placed it on the machine bed, and used the camera to re-calibrate the X and Y step scaling the configuration options. Once I did that, the positional accuracy was far better (near perfect). The only caveat, is that once you re-calibrate those two settings, it will throw all your other locations / offsets into a tizzy, and you will have to go over EVERYTHING! I also adjusted the nozzle X/Y offsets from camera / etc, which helped a lot also.

The only area of concern that is currently not correctable, is the error introduced by nozzle concentricity. Since the 920 does not have automatic nozzle correction, if you are not using vision for placement, and the nozzle is not perfectly concentric (in build quality, or in the nozzle holder), and rotations of the part will throw placement accuracy out the door.

But in any case, with a good few hours spent calibrating the machine, mine is running really well (so far).

R.

[anfang]

But in any case, with a good few hours spent calibrating the machine, mine is running really well (so far).

After I set my machine up, there seems to be drift over time. That is, in the morning, after I turn the machine back on and do the 'home' operation, the locations for everything is slightly off. Not by a fixed amount. But I have to re-zero the pockets for each feeder, re-zero the nozzles, etc.

Does your machine precisely remember the locations of things over hours/days? Or are you having to re-zero things during operation?

I just ran a test of 100 0402 placements at 70% speed using two nozzles and placing onto sticky tape pcb. This was with 'quick' vision correction. I'd be interested if others could run this.  You can see there are a few places in the placement where the results were very good. And other places not so good. I will continue to tweak and learn.

[anfang]

And one more note as I just looked this up: A 10C degree change in room temp will change the length of a 1m piece of aluminum by 1mm, which suggests that if you want to do 0402, you need to keep room temp swings to a few degrees. But more importantly, it suggests that to keep a calibration on a machine overnight, the machine temps need to be within a few degrees of each other day to day.

[ar__systems]

Sorry, your math is off by a factor of 5. So it is not so bad.

And one more note as I just looked this up: A 10C degree change in room temp will change the length of a 1m piece of aluminum by 1mm, which suggests that if you want to do 0402, you need to keep room temp swings to a few degrees. But more importantly, it suggests that to keep a calibration on a machine overnight, the machine temps need to be within a few degrees of each other day to day.

[thommo]

Hi anfang,


From your observation, what are you putting the cause of these misalignments down to?
It occurs to me that the component is moving on the nozzle after visual alignment.

Have you tried 'stepping' through each phase of the placement?

What result do you get if you slow the machine down even more, or use the more precise vision option?
What nozzle are you using?
Have you checked the nozzle tip for damage due to crashing it into the component/PCB?

1mm displacement seems a lot to me. I'm not certain I've explrienced that kind of shift but we are in an air conditional environment.

I'd be interested in your response.
Thanks for your posts.

Thanks - Peter

[thommo]

Yes, I believe ar_systems is correct

0.2mm shift / 1mtr / 10C

[Smallsmt]

But in any case, with a good few hours spent calibrating the machine, mine is running really well (so far).

After I set my machine up, there seems to be drift over time. That is, in the morning, after I turn the machine back on and do the 'home' operation, the locations for everything is slightly off. Not by a fixed amount. But I have to re-zero the pockets for each feeder, re-zero the nozzles, etc.

Does your machine precisely remember the locations of things over hours/days? Or are you having to re-zero things during operation?

I just ran a test of 100 0402 placements at 70% speed using two nozzles and placing onto sticky tape pcb. This was with 'quick' vision correction. I'd be interested if others could run this.  You can see there are a few places in the placement where the results were very good. And other places not so good. I will continue to tweak and learn.

It looks like a Z axis height problem go in 0.05 mm steps deeper (reduce component height ) and try again and check if it's placing better.
You can reduce max by 0.15mm deeper for 0402.
The PCB maybe not even height so we use the nozzle spring to compensate.

Good luck!

[anfang]

Yes, I believe ar_systems is correct

0.2mm shift / 1mtr / 10C

Yes, thanks to both for correcting.

Does your machine hold the same calibration over 24 hours? That is, if you locate the pocket on feeder F1, and turn the machine off at night, and then wake up in the morning and do a 'home' operation does it go back to the dead center of the feeder pocket again?

[anfang]

From your observation, what are you putting the cause of these misalignments down to?

There is a parameter called "pick delay" and "place delay" and both were at zero. I changed both to 150 mS with some good improvement. See attached.

It occurs to me that the component is moving on the nozzle after visual alignment.

Yes, probably. But how do you know the visual alignment is correct after leaving the camera? I can single step it, but for 'quick' alignment, you still don't get to the 'after' shot. Only the 'before' shot, is that right?

What result do you get if you slow the machine down even more, or use the more precise vision option?

I am thinking the lack of delay for 'place delay' was an issue. What do you set this figure at?

I'm not certain I've explrienced that kind of shift but we are in an air conditional environment.

So, when you start a new day and turn the machine on, is everything aligned just as you left it? Nothing needs adjustment? All feeders are perfect, nozzle alignment is dead on, etc?

Any chance someone could run the csv I posted with some 0402 on their TVM920 and post the results? Just load the 0402 into F1 and you'll be ready to go in 60 seconds.

I'll keep tweaking and tuning. Learning a lot.

[anfang]

The PCB maybe not even height so we use the nozzle spring to compensate.

Hmm. This got me looking at the nozzle in more detail, and there was some xy play in it, and on further inspection I'd not placed the o-rings and so the nozzles had a tiny bit of extra slop. Stupid on my part, I know.

And with the o-rings in place, the results were very good even at 100% speed.

There is still an issue related to Y spacing. All parts should be equally spaced and as you can see they aren't quite. But this is a big step forward. Thanks to all for the advice. Very impressed on 0402 placement at 100% using quick vision. This was about 3 minutes to place 100 parts.

[thommo]

Nice work anfang!

I am not really 'running' my machine yet, so any opinions or advice are coming out of observation or logic.

The machine has a fairly simply job to do, so the results you first reported didn't make sense, especially knowing that its using closed-loop steppers.

Glad the o-rings are helping also.

What nozzle size are you using? Worth taking a look on the official Juki site to check what they recommend I'd say.

But generally speaking, that's a very positive result on those 0402 with more improvement to come.

BTW - what 'vision' option are you using with the 100% speed for that test?

Cheers - Peter

Oppps - I meant to mention that you ought to consider running your machine through a 'warm-up' procedure before use. I have no idea exactly what it'll do for repeat ability [other than to improve it]. The rails will warm up and the lubricant won't be as sticky, the belts likewise will change characteristics, the pistons will warm up, and the motor also.

All CNC machines I know of have a warm-up phase prior to operation

[anfang]

so the results you first reported didn't make sense, especially knowing that its using closed-loop steppers.

Yes, agree. The appeal of this machine (for me) was the feeders, closed loop and feeder count. However, I'm still scratching my head on the location drift over time. I need to get out the dial gauge and understand backlash on the machine and also verify how repeatable homing is.

The smallest nozzle the machine comes with is a Juki 503 (aka #1), which Juki recommends for 0603/0805. The 502 nozzle is for 0402, and i'll order one to try out.

My original thinking was that I'd use 0402 for bypass only to help a bit with board sizes.  The 503 nozzle (made for 0603) will tend to tombstone an 0402 part if not lined up with 0.2mm or so when it picks, and the corner of the part can almost fit into the nozzle when stood on end. But as you see, once everything is lined up, over 200 parts, there wasn't any tombstoning. It was pretty interesting watching the 'quick correction' images splash by on the screen because everything was so consistent the camera appeared almost like it wasn't working. In other words, the pick up operation was so good that the camera was just cleaning up the  +/- 0.2mm shift and +/- 10 degrees or so rotation that was needed. With the o-rings missing, every image on the screen flashing by was drastically different. So, lesson to myself is that if the pick isn't solid, the correction probably won't be able to fix it. Expect the camera tweak the pick, not fix a bad pick.

I suspect the 502 nozzle for 0402 will be a bit more forgiving on the alignment. Will report when it arrives.

[thommo]

Hi anfang,

Can you confirm that you are using the updated Software App which has provision for calibrating the axis?

Do you 'still' see the Y drift now that the O-Ring is in place? Could it have been attributed to nozzle movement without the O-Ring perhaps?

What are the variations you are seeing over these tests, and how are you measuring it?

Please attach the photos of the steel scale you are testing with illustrating the variation. I assume you are making this conclusion based on the 'down-camera' images - am I correct? Or are you looking at actual parts that have been placed?

Cheers - Peter

[anfang]

I found another issue: The X axis servo (the axis driven by a single servo) was loose enough to rock ever so slightly. One of 4 screws was missing completely, another was backed out halfway and a 3rd was barely seated. Fixing this has helped too. I found this after the picture shared above.

The 'drift' issue is best replicated as follows: Align Nozzle 1 perfectly with the up-looking camera and 'capture' the location. Then re-home the machine. Place a few boards. Do whatever. And then go back and look again at nozzle 1 via the up-looking camera. On mine, it needs to be re-centered after, say 60 minutes of use. How much it needs is perhaps +/- 5 clicks (+/-0.05mm) in any direction. At least I'm pretty sure it was any direction. Maybe this was a manifested of the loose servo.

I'm getting the dial gauge out today and will track this down further. My current theory is that the home operation is a bit sloppy (+/- 0.05mm).

[thommo]

Can I suggest you do this that using the down looking camera and the Stainless Steel scale fixed to the riser blocks?

That way you will be able to isolate the area of movement to either the axis in general, or in the heads nozzles..
It could be an issue weiththe nozzle not being concentric, and it is rotating to that location.

You could also do your same test but ensure there are no nozzle rotations in the programme. Of course looking for other loose fixtures in the head is a good idea also.

I must say that at 5/100 mm, I think you are close to hitting the practical limitations of the design. What resolution tolerance are you expecting or needing it to achieve anfang?

Cheers - Peter

[glenenglish]

Anfang,  do the stepper/servo attachment screws have a flat washer + spring/shakeproof washer ?

At least we know they did not over-tighten the screws- which is a good idea for steel screws in an (soft) Aluminium thread....

Maybe a case for flat+spring washer and precise torque and Loctite on the threads. and maybe where there are nuts, change to Ny-Loc nuts

glen

[Smallsmt]

Anfang,  do the stepper/servo attachment screws have a flat washer + spring/shakeproof washer ?

At least we know they did not over-tighten the screws- which is a good idea for steel screws in an (soft) Aluminium thread....

Maybe a case for flat+spring washer and precise torque and Loctite on the threads. and maybe where there are nuts, change to Ny-Loc nuts

glen

The easy solution is use Loctite 648 glue to fix the screws!

[anfang]

Can I suggest you do this that using the down looking camera and the Stainless Steel scale fixed to the riser blocks?

Using nozzle as reference is the same, isn't it? And it makes it easy to ask the machine where is the marked location

You could also do your same test but ensure there are no nozzle rotations in the programme. Of course looking for other loose fixtures in the head is a good idea also.

Yes, agree this is very important. When you zero nozzle locations, make sure all rotations are zero degrees. The pick will happen always with 0 degree rotation, and on the way to the camera or pcb, the rotation will occur. And then the place will happen, and on the way back to the next pick, the rotation will be reset to zero.

So, for consistency and max pick accuracy, always zero nozzles and verify zero degrees. The smallest (Juki 503) nozzle has an ID of 0.6mm, and I'd estimate over the full rotation is also the concentricity error. The Juki nozzle error is very likely a small fraction of that, with the way the Juki is mounted to the stepper being the larger sourch

I must say that at 5/100 mm, I think you are close to hitting the practical limitations of the design. What resolution tolerance are you expecting or needing it to achieve anfang?

Right now I'm just trying to learn limits.

Here's a test I did:

1) Place 400 parts to warm up the machine. These were "air" placed (no pcb, no  feeder, quick vision).
2) After placement, zero nozzle 1
3) Go home, run 400 more parts, nozzle location is dead on the entire time. Homing is very consistent.
4) Turn machine off, wait 15 minutes, turn machine back on, check nozzle location. Picture attached ("after powercycle")
5) Air place 200 parts
6) Check location. It is mostly unmoved ("after placement" on picture)

Some observations/theories for others bringing up their machines:

1) Warmup matters a lot. I'd estimate +/- 0.5mm. Run the machine for 5 mins or so before zeroing.
2) Homing repeatability is very good. I'd estimate +/- 0.02mm
3) Board to board repeatability is very good.
3) Something I don't yet understand is what happens between step 4 and 5 above. Something gets a bit out of whack and in the pictures, it's mostly an xaxis issue. Will monitor in the future.

Finally, there is a parameter called "delay before vision" or similar that is important at 100% speed. If you are air placing, then you can always see the system's idea of the "absolute zero" nozzle location during a 'quick' vision check: Every image that flashes on the screen should show the nozzle centered exactly in the crosshairs. If the delay value is set to 0, lots of times an image won't even register. I'm assuming the system treated the recognition as failed since it didn't display anything. If you set it to 50 mS, then you will see the nozzle is't in the center of the cross hairs during each recognition. This most likely means it hasn't settled yet--the various mechanical dampings are still occurring even though the system thinks its stopped. But at 150 mS, the nozzle is dead centered every single time.

So, if you are aiming for small parts and accurate placement, you should be looking at 100 to 150 mS for pick delay, place delay, and pre vision delay. This will ensure the system has settled. And when you are watching images flash by during quick vision, make sure that each image shows the nozzle mostly in the cross hairs. If not, then you have a concentricity problem or the system isn't settled enough. If the nozzle is mostly in the cross hairs and the part isn't, then you have a pick problem.

PS. I also attached my best effort on 0402 placement. It's where I need it.

[anfang]

Anfang,  do the stepper/servo attachment screws have a flat washer + spring/shakeproof washer ?

At least we know they did not over-tighten the screws- which is a good idea for steel screws in an (soft) Aluminium thread....

Maybe a case for flat+spring washer and precise torque and Loctite on the threads. and maybe where there are nuts, change to Ny-Loc nuts

No washer, no lock washer. But the threads go into an octagonal anodized aluminum spacer, so probably there is ~30 threads engaged, so little risk of stripping.  I will monitor.

[anfang]

Of course looking for other loose fixtures in the head is a good idea also.

I've studied this some more. If you remove the nozzle and lower the head, the nozzle 'receiver' will come into focus and you can spin it. Mine looks to be very accurately machined. And I inspected the Juki nozzle some more--there is almost no lateral play in the tip. But when you put the two together, the play materializes.

You would think the shoulder of the receiver should mate with the shoulder of the Juki. That would give a metal on metal contact. But in fact, there is a gap between all the nozzles shoulders and the receivers on my machine. It ranges on my machine from about 0.2mm to 1 mm. This gap is the source of the play I suspect. As you wiggle the nozzle, you can see the gap opening and closing.

Ideally, there'd be a second oring on the Juki to keep it centered and prevent the rock. Looking up into the receiver with a camera, I can see another oring up there it looks like. That seems to be the thing that prevents the shoulder to shoulder mating. Note the mating isn't firm--it's a bit squishy.

I wonder if a precise 1mm spacer/washer between the Juki and the receiver might help this problem.

Unrelated...the nozzles are all different lengths. Where does that get accounted for in the software? Does the spring just soak up the ~2mm difference in length?