Choosing a Desktop Pick and Place Machine

[level6]

When the YY1 starts to mount components, it first detects the position of the fiducial marking. If you don't have a formal fiducial mark, then it is possible that the vision is picking some feature that is off from where you centered what you selected as a fiducial. When you start part placements, step through the first stage where it detects the fiducial and see on the left screen the position it finds.

[Friedl.b]

Thanks! 

I will use the test PCB again and see whether I can get that working 100%.  After that I will try the custom PCB again.  I also have more boards I need to assemble that do have fiducials so can test those as well.  Busy loading more components in the machine, will let you know and post some pictures.

Thank you for the help, it is much appreciated!

Regards, Friedl.

[glenenglish]

Dave you can have my TVM920 and all the spares if you come and get it (canberra) . I'd regard it as a good 0603 machine. 0402 I think requires more refined pickup and place head skills.  well 0402 is OK  on it depending on the parts and a bit of patience. I've always thought CL feeders were a bit on their edge on 0402s but they do make 0201 feeders for CL, so it can be done.

[EEVblog]

Dave you can have my TVM920 and all the spares if you come and get it (canberra) . I'd regard it as a good 0603 machine. 0402 I think requires more refined pickup and place head skills.  well 0402 is OK  on it depending on the parts and a bit of patience. I've always thought CL feeders were a bit on their edge on 0402s but they do make 0201 feeders for CL, so it can be done.

Wow, thanks Glen!
I found the dimensions. L1080 * W805 * H750mm is that correct? Will have to measure the car and the dungeon door.

[glenenglish]

needs a small pantec. maybe a wide van.
PM me.

[Friedl.b]

Hi -

I have a couple of YY1 questions, hoping someone can assist.  Pretty sure I am doing something silly

1. I have some optocouplers in one of the flexible feeders.  the Machine proceeds to pick one up, move it over to the camera, verifies and then moves on to where it should be placed.  It then 'hesitates' and then proceeds to throw it away 

2. I have some Otto-triacs in the bulk-feeder.  The Machine tries to pick them up but misses them, no matter what I do.

3. I am using a custom panel (2 x 2).  The first boards seems to work (for the most of it at least).  When it finishes, it then gives the following error "Unable to find fiducial". attached is a photo of the the panel.  Pretty sure I am doing something wrong here.


Would appreciate some advice ?

[newto]

Hi -

I have a couple of YY1 questions, hoping someone can assist.  Pretty sure I am doing something silly

1. I have some optocouplers in one of the flexible feeders.  the Machine proceeds to pick one up, move it over to the camera, verifies and then moves on to where it should be placed.  It then 'hesitates' and then proceeds to throw it away 

2. I have some Otto-triacs in the bulk-feeder.  The Machine tries to pick them up but misses them, no matter what I do.

3. I am using a custom panel (2 x 2).  The first boards seems to work (for the most of it at least).  When it finishes, it then gives the following error "Unable to find fiducial". attached is a photo of the the panel.  Pretty sure I am doing something wrong here.


Would appreciate some advice ?

1) Most likely a contrast issue with the image, my guess would be it can't see all the pins or sees a reflection and thinks it didn't pick up properly. When it rejects a part it always goes over the position first for some reason. If you absolutely can't get it to work with the camera, you can disable the check and use vacuum to ensure you've picked something up, and then manually check placement after.

2) Are you using the bulk feeder in camera mode, or just picking up from a fixed position? I don't think the camera mode works on things more complicated than passive part caps and resistors. If you're not, check the pick height and position of the feeder location. I use a ton of the bulk feeders by manually placing the components into the squares, but I have to make sure I index them against the same corner every time (next revision will be much simpler and not need as many individual refills for the feeders)

3) I haven't tried panels yet, worst case you could create a placement file that does all 4 boards together, instead of trying to use the software panel system. My guess for your panels is that the software wants a fiducial setting put in, and a fiducial on each board to check before starting the second board.

[Friedl.b]

Hi -

Thank you for the prompt response

I managed to get the optocouplers working by switching to Mode 4 (large component view)... I actually did test without the camera and it does work, would have been last resort, hehe.

For the bulk feeders I am using the camera yes.  My very limited and brief experience with the machine is that it is most certainly easier to use reels.  I loaded quite a bunch of reels, but some components are just too expensive to buy in reel form or maybe project specific, so I will try your method as well, thanks!

This is not a bad idea as Altium actually exports the panelized PNP file this way.  Saves me a lot of work editing it afterwards.  I am also adding a parameter in my component library so that the exported file will already contain the feeder number for each component (where possible).

Aside from that, enjoying the machine so far.  Just a little annoyed with the fact that I quite often need to fine tune component placement.  Not to bad on 30pcs board, but less fun on 400 component count boards 

Regards, Friedl

[Friedl.b]

Hi -

Just another question about the fiducial.  Is this the wat the camera is supposed to see it.  I have seen some videos where there all black around the fiducial and then only the plated pad is picked up.  In my case it looks quite different.

Regards
Friedl

[newto]

Glad you got most of the issues sorted. I've got two components that I have to place blind, a large inductor and a button cell battery holder, neither work with the camera, but they have huge pads, so accuracy isn't super needed.

For the fiducial image, that should still work, the software is smart enough to ignore the white outside of the circle with the crosshair over it. The reason it doesn't have a nice black field is that you have white solder mask and the camera picks it up like the reflective dot. Whether the fiducial actually *does* anything is another matter. There's definitely some uncertainty on whether it actually does anything before placement...

It's definitely a "you get what you pay for" type situation. Low number of unique boards (or lots of boards that share the same components), relatively small production quantities, minimal quantities of small components (I have an 0201 diode on our main board, I make sure that's placed first so I can check it before it does the rest) and it's pretty good. Go outside of that, and you'll be absolutely miserable with it.

[Friedl.b]

Haha,.. always good to hear something might turn into an absolute disaster 🤣🤣  I will try the new file tomorrow with all the components as apposed to the panel option, should work I guess.

I doubt I will go below 0603… maybe 0402 but not likely. I do use quite a few QFN packages but it seems to handle those just fine. Aside from the tape… of my word I can see those being a ‘challenge going forward.   No issue on the paper tapes but the plastic ones have been challenging. My quite expensive super flat LEDS when flying like popcorn kernels 🤦🏻🤣

Still need to master the Nozzle change as well,  it am sure that will be ok. Looks like the thing is to place components using same nozzle sizes in groups as it only allows 4 changes.

With regards to the fiducial… would be quite nice to know whether it actually does something yes. If it uses an absolute 0,0 origin of the board it should be an able to place just using the X,Y coordinates from the PnP file I guess??

[MR]

Glad you got most of the issues sorted. I've got two components that I have to place blind, a large inductor and a button cell battery holder, neither work with the camera, but they have huge pads, so accuracy isn't super needed.

For the fiducial image, that should still work, the software is smart enough to ignore the white outside of the circle with the crosshair over it. The reason it doesn't have a nice black field is that you have white solder mask and the camera picks it up like the reflective dot. Whether the fiducial actually *does* anything is another matter. There's definitely some uncertainty on whether it actually does anything before placement...

It's definitely a "you get what you pay for" type situation. Low number of unique boards (or lots of boards that share the same components), relatively small production quantities, minimal quantities of small components (I have an 0201 diode on our main board, I make sure that's placed first so I can check it before it does the rest) and it's pretty good. Go outside of that, and you'll be absolutely miserable with it.

I absolutely disagree ... it is not "you get what you pay for" in that industry, care has to be taken but the chinese are at least low cost. I got scammed by Mechatronika from Poland nearly a decade ago they're still active out there.
I bought their top model (used) but sent it all to them for a complete overhaul including replacing the motors -- , and if you know about those low end mid range models you should know that there's not much behind that, you have a gantry and a pick and place head.
If you have new motors and a new controller on it practically is like a new machine. The only bad thing that could be is that the linear rails are worn out but that was not the case. The oscillating / shaking portal problem came up after they have done their magic upgrade - it wasn't there before.
Other issues were optical recognition based and there were also some vendor calibration issues involved eg. wrong timing triggering component pictures which caused additional component drop offs. I got far enough to had a proof (=pictures which showed that the flying camera triggered too early or too late back then - I even figured out the obfuscated eeprom setting how to adjust that, they did not expose that calibration option to the user)

The machine overall looked like something due to heavy metal used for the frame but the machine is put together close to unserviceable in some areas the entire gantry needed to be lifted in order to access the Y motor (400W Delta Servo Motor - to be fair even Delta tried to support the shit Mechatronika machine I reached out to them told them that Mechatronika is unwilling to resolve severe issues and they tried) In 2020 we stopped completely since I moved to Taiwan. I decided to not move the machine since the value just isn't worth the shipping price (I'm honest about that and I won't let go that machine either since I don't want to scam anyone with it - and I have the knowledge to make something useful out of it; I have a weak plan to put my DIY head on it and convert it to a CNC which can mill Aluminum at least).

The machine cost 40.000$ back then (27.000$ used, ~13.000$ directly to Mechatronika for the service; and they were friendly as long as we were willing to throw money at them.. the return was 2x 500$ Servos and some crappy controller PCBs), and now in 2024 you can still buy those machines for 50-60.000$
I built my own machine in the meanwhile (DIY) since I very much studied their machine and rewrote the PNP Software until 2020 (which worked way better than their one). My software collected the pick issues and let the operator fix the issues at the end of the procedure - this worked reasonably well and fixed one of my biggest issues - the baby sitting.

I had no machining background until 2020, I got into that from 2020 on, CNC turning, CNC milling (CAD / CAM). Aluminum prices where I live is like 5$/kg delivered to the door. Our new machine is multihead 2 heads installed, the head is set for 4 heads. the axis have 3 400W servos installed (accuracy is like 0.01 - 0.02mm).
So whatever the Mechatronika machine has a real value below 4000$, not too many people are buying that machine due to the high price the machine is capable of building PCBs but very uncomfortable to use.
And I can tell only unexperienced people buy such a machine (including me back then).

Someone from Spain bought a feeder box which they sell for 4000 EUR, there are 2 cheap stepper motors inside, I think I will make a teardown about that, in my setup (my feeders) 16 lanes jammed completely due to plastic tapes inside -- again no support nothing available from Mechatronika (and at any time I gave them the option to fix their issue).

Resellers love Mechatronika since the margin to earn is relatively high, there's just no value behind that (again I built my own machine, my knowledge started with Mechatronika). I did not expect resellers to stand with their machines so seriously - anyone listing a Mechatronika clearly shows deep business interest rather than having an honest long term business.
The usual answer others don't have that problem is scam, anyone can say that;
A pick and place machine is a mathematical system a good software can make a cheap machine shine as long as it takes care about all the problems.
If a pick and place machine - after a fixed procedure randomly throws components onto a PCB due to mathematical errors (eg. not taking into account that a PCB is squeezed by a PCB holder and not flat) that just shows that certain cases are not handled by the software.
And doing such a pick and place software is not difficult it just takes time. They have a procedure to measure the PCB distance - well then just be sure that the measurements are taken at multiple points (it's measuring the PCB distance to the nozzle via vacuum sensor sensing - I also did that on my DIY machine, until I put a panasonic distance sensor on it)
Even so as a pick and place manufacturer they should provide equipment to avoid this - it's no problem to manufacture a frame or simply telling the customer what to do - or put it into a manual (their manual is a joke).
The optical recognition cannot even handle HASL based PCBs since the PCB lightning is just a cheap LED ring - they simply just cannot be set up at all. I am also monitoring the OpenPNP community, while I fixed that with better diffuse lightning, OpenPNP uses the Parallax detection method (my software is not based on OpenPNP, and I have never used it but just by reading the documentation I value any high level discussion about certain technical issues)

Maybe someone is aware of milling PCBs, there's free software out there which measures the height of the PCB raster based and interpolates a linear milling path. So not doable is just an excuse (again the machine is quoted 5 digit value and doesn't offer anything advanced).
Luckily I do not depend on my Mechatronika anymore. Mechatronika is even sending their dealers a mail about us (I know they will read this too since I always send them the links whenever I write something about them), just a big warning if they want to end up in the GDPR violation database just keep going scamming more people.
That machine has cause financial loss to us until we got it up and running with our software.

The machine is not in use anymore if they think they can do it better they can still reach out and fix that machine up to what I call usable quality level (I would document it accordingly online).

Attached the picture of their staging camera (of the MX80 and M10V of another customer), in the meanwhile other Mechatronika customers also confirmed that they have this style of camera installed. (the upgrade quoted "new" camera back then... so that's what you get .., I replaced it with a 10$ chinese CVBS camera already and that just works as a drop in replacement, the current modules usually support AHD and CVBS and offer an on screen menu to adjust various settings)

And writing the customer should fly to them for training, honestly .. a single head pick and place machine doesn't need advanced training because there's nothing advanced behind that (again I built my own multihead machine, I know it all; my machine is not perfect but exceeds the capabilities of the Mechatronika machines at least).
For multihead calibration some training might be useful if someone has never used a machine before - multihead is a different story.
Set component/feeder location, PCB location and PCB distance and that's it. If an entire portal starts to shake due to loosely installed pulley and errors out servos that's something a manufacturer should deal with and not bug the customer about getting some training

Or telling their resellers and customers .. that guy bought it for the wrong purpose.. No I did not want to mix yoghurt with this machine if that's what that the "shaking portal" including watch it until it's mixed properly (=babysitting) feature is supposed to do with oscillating I'm very sorry I did not see that in the spec sheet.

That's my comment about "you get what you pay for.". No you don't even in 2024. A higher price yet doesn't mean that you will get something better.

All European resellers should be ashamed of listing those machines and actively ripping off people. For that money they were always able to get way better machinery. And those who have those machines work for their purpose their requirement is very low and they're lucky - and yet they could still have way better equipment for a lower price infront of their nose.

[mikeselectricstuff]

 

My software collected the pick issues and let the operator fix the issues at the end of the procedure - this worked reasonably well and fixed one of my biggest issues - the baby sitting.

This is an important  feature that seems to be consistently missing from low-end machines - it's obvious that it's the correct way to do it (at least as an option) but for some reason teh Chinese makers don't understand.
 I have a manual from a DOS-based machine that's maybe 35 years old that does this right.

It's so frustrating that there are now so many machines with OK hardware, but which could be improved very substantially with a couple of days' software work

[Friedl.b]

As I am VERy new to this, I guess I will still get around to finding out what actually annoys me about the low end machines....  Having said that, I can already see why there is such a hype about electronic feeders.  While mine works well with the paper reels, the plastic ones have been a pain.  It is as if those little friction wheels grabbing the tape is not grabbing it properly resulting in the tape cover not being pulled off when the tape is being pulled forward.

[Friedl.b]

Hi -

I have seen on some videos that you need to adjust (fine tune) each component after setting the first component and fiducial.  At first this sounded plausible as the crosshairs seems a bit off when viewing the components in EDIT COMPONENT MODE.

All of that said, I have not had much success getting them placed properly and then enquired at Neoden.  I received the following response: " dont need to adjust each component after that.  The image displaying here just a reference, not so accurate" 

What is you experience with this, do you need to finetune component centers after setting the fiducial?

Regards,
Friedl.

[br]

I have received my YY1 late Nov 2023.

I did some testing during couple of days and noticed it could not place lqfp128 package (Big IC mode). My guess was that the YY1's vision is confused about presence of Exposed Pad (EPAD) under the LQFP128 package (the pad reflects light).

I posted a video to youtube and reported the problem to NeoDen:

They did a firmware fix for me, but instead of sending me a link to the new firmware, they have shipped to me a new motor controller board with updated firmware. Weird, but I don't mind to have a backup board.

Instead if swapping my current board to a newly received one, I extracted firmware from it, and programmed my current board. Saved some time (the boards are identical).
The fix did not help however, even worse some new bugs appear on feeder setup page.

I did some disassembly of their firmware but could not find anything interesting.

The soc on the board is STM32F407. I have some experience with it and I also have a small real-time os developed from scratch during my life and work in Cambridge.
Spent a couple of days with a multi-meter to track down all connections to stm and bring up my OS on it.

The vision is done on the "Edge", meaning that it is done directly on camera module PCB the image sensor is located on. Both Up and Down looking cameras feature stm32h7 microcontroller (firmware is identical I checked). Camera sensors handled by DCMI (Digital Camera Interface) STM32's peripheral device.

There is no USB in entire YY1 system. Cameras process the image and send result over low speed serial to the motor control board. I could not remap those serial interface pins to USB because stm32 does not allow that for those particular pins used. Each camera module has a connector for the front display, they are also low speed serial (probably SPI).

Long story short, I have a new firmware written from scratch for YY1 and I managed to assemble a few boards with OpenPnP!

Also I replaced both cameras. I have some cad models and links to aliexpress if someone interested.

[Kean]

Long story short, I have a new firmware written from scratch for YY1 and I managed to assemble a few boards with OpenPnP!

Wow, congrats.  I am sure others would be keen to try that.

[glenenglish]

The hardware in the chinese machines is generally OK.

Yes, the video can be a bit slow. I converted all my Kayo  analog video machines to USB2 and USB3 video when I wrote my own SW for it.  THis is because the  multi video analog capture card that came witth the machine wouldnt talk to anything.

I've got a colleague with the latest and greatest 6 head Kayo machine,--- again hardware is quite good. but software is a complete P.O.S. ...   Just a joke... Better to buy a 30,000 hour  used  Yamaha me thinks, if you can fit it and give it the air it needs, would be total cheaper than the chinese machine including a big scroll compressor.

Jon in Melbourne bought my three Kayo machines and got OpenPNP going on them, so that is a plus. (these days I have a big Samsung - Hanwha SM482) .

[48X24X48X]

The hardware in the chinese machines is generally OK.

Yes, the video can be a bit slow. I converted all my Kayo  analog video machines to USB2 and USB3 video when I wrote my own SW for it.  THis is because the  multi video analog capture card that came witth the machine wouldnt talk to anything.

I've got a colleague with the latest and greatest 6 head Kayo machine,--- again hardware is quite good. but software is a complete P.O.S. ...   Just a joke... Better to buy a 30,000 hour  used  Yamaha me thinks, if you can fit it and give it the air it needs, would be total cheaper than the chinese machine including a big scroll compressor.

John in Melbourne bought my three Kayo machines and got OpenPNP going on them, so that is a plus. (these days I have a big Samsung - Hanwha SM482) .

Kayo was basically using OEM machine from other manufacturer, last I know was from Bovi. In fact quite a few is doing this sort of business like YX and ZB (until 2019) from HWGC. Hence, the software are usually 1 or 2 major revision behind the real manufacturer (probably intended). You will also notice Kayo's machine line up didn't really grow like others.

[glenenglish]

ahhh. yeah they seem to have their own software people since it seems they can do the odd fast bug fix if they want, albeit with zero testing LOL.
AFter I reverse engineered all the RS485 bus commands (from the PC to the controller board)  and wrote my own software , I still couldnt figure out a couple of commands out of maybe 100, and they just sent me the whole documentated command list.

It's clear though the chinese mfrs tend to not have any ways of updating firmware etc in their sub boards. Like Kayo, Neoden,  QiHi have all sent new controller boards to update software in their controllers. (instead of providing a tool).

For the suppliers that just use an OEM machine (like as you say Kayo) , they hold onyto their software/firmware mods very tightly, they're scared of copying.  There's no concept of locks, keys, etc to limit distribution, its all really low tech.

That's why I think if we ever go to war with china, none of their sh1t will work. LOL.

[br]

ahhh. yeah they seem to have their own software people since it seems they can do the odd fast bug fix if they want, albeit with zero testing LOL.
AFter I reverse engineered all the RS485 bus commands (from the PC to the controller board)  and wrote my own software , I still couldnt figure out a couple of commands out of maybe 100, and they just sent me the whole documentated command list.

It's clear though the chinese mfrs tend to not have any ways of updating firmware etc in their sub boards. Like Kayo, Neoden,  QiHi have all sent new controller boards to update software in their controllers. (instead of providing a tool).

For the suppliers that just use an OEM machine (like as you say Kayo) , they hold onyto their software/firmware mods very tightly, they're scared of copying.  There's no concept of locks, keys, etc to limit distribution, its all really low tech.

That's why I think if we ever go to war with china, none of their sh1t will work. LOL.

Chinese make a good cheap (and sometimes working?) hardware, but on the software side is complete fail.
Fitting machine vision into restricted (in terms of resources) environment of stm32h7 is a boldly attempt (and failed on YY1).

Unexpected Maker in one of their videos mentioned that the firmware of NeoDen IN6 does not do its job properly and they wanted to find ST-Link connector on it (to reprogram stm32 I guess)
https://youtu.be/PQ7eh93AQRk?t=412

[br]

Long story short, I have a new firmware written from scratch for YY1 and I managed to assemble a few boards with OpenPnP!

Wow, congrats.  I am sure others would be keen to try that.

Thanks. I have published my firmware:
https://github.com/mdepx/neodenyy1

It lacks documentation but I'm working on it slowly.

[glenenglish]

yeah, there is no way a STM32 is going to have the CPU grunt required to perform the optical work in a robust manner.
You need a big PC for that or other highly customized FPGA hardware.

also, use of color cameras show that they dont really know what they are doing, compared to a hi-res, NON bayer filter B&W imager. 

The color cam throws away almost 75% of the otherwise useful resolution.
and then there is the chromatic abberation of lenses. Illumination with mono chromatic light helps a bit there (red). but white illumination  for fiducial marks with a color camera is bollocks.....

[MR]

Colour cameras have their advantage that you can pick up details which you might not see with BW cameras.
Add a software filter and you'll have your BW picture for analysing it, it's not a big deal with opencv, even without (eg. YUYV (strip the chroma UV), since most eg. usb cameras actually do not submit RGB but YUYV or UYVY it's a simple operation, some of them do some pre-filtering on the controller). Basically all variations of cameras are out there nowadays.

Most of those cheaply built pnp machines use a simple LED ring for illumination only, there are various mechanisms available.
https://www.vision-doctor.com/en/illumination-techniques/diffuse-incident-light.html

pick and place machines do not do heavy optical recognition, it's more about the algorithm which is used for a certain components.
For optical recognition I mostly use 640x480, and all is fine with it. Most cameras are able to deliver far more than that.
Some cameras have luma filters inside the controllers / sensors. You can get acceptable results with lowest end cameras.
In case of Mechatronika they used a BW CVBS camera (interlacing is even worse than having no pure BW picture, since you need to wait for a stable frame (2 half frames in still position, 40 milliseconds at least) to get the "full" resolution. Those cameras cost 10$ plus PCI controller (I also used it via USB at some point).

[glenenglish]

yeah but the Bayer filter and chomatic abberation ruins your potential resolution and edge detection. IE more demands on the lens, light source.