Low Cost Pick and Place Machine

[tggzzz]

It is to illuminate the bottom of the IC. Below the lights is a camera that is looking up at the bottom of the IC. This camera corrects for the rotational error of the IC, so when it is placed all the leads from the IC are correctly placed on the pads of the PCB.

How would that work with a board like this https://entertaininghacks.files.wordpress.com/2015/03/designsparkpcb.gif where red is the topside and blue the bottom side? Note that the bottom is mostly a ground plane.

Would it work at all with a 4 layer board, where the inner layers are mostly power planes?

tggzzz, I'm not sure I understand the question. Could you explain a little bit more?

Oh, drat. I misread you point by concatenating two posts. I see you are only illuminating the underside of the ic, not the PCB. Sorry to have wasted your time!

[ProtoVoltaics]

Ah ok I understand now. Yes, we light up the bottom of the IC not the PCB.

No worries thanks for taking the time to read the thread! 

[eneuro]

Anyway what about placing elements at choosen angles lets say 45 degs or any given angle ? The same camera for correction is used?

You do the vision first to establish the offset and rotation of the picked part, then rotate it to the correct orientation, preferably during the journey from camera to placement position

Correct. It is as mikeselectricstuff says. 

Do you mean parts taken directly from feeder, I guess, since in the case of parts lying on working table freely distributed I see no need to use additional correction camera, since I have USB microscope in my machine head and it can see picked element all the time?

I'm considering do tricky thing like, when picked element travels to destination, some kind of background is quickly inserted below picked element, so I have nice well defined backgroud for image recognition purposes, catched part above and visual camera on top adjusting part rotation to fit into final angle.
I see, no reason to create such strange corner with additional lights to perform such simply task like rotating part before placement as seperate task as it was shown on video I've linked earlier

Maybe do you have full PCB assembly video registered starting from brand new PCB and feeded parts with hopefully assembled board ready for "cooking"? 
I'd like to see it at 10 times slower speed to know better challenges we need to overcome to have fully assembled board without manual human work

[ProtoVoltaics]

Eneuro, are you building your own PnP? Yes, there is a lot of work, and lots of thought that have to go into these projects. One fix seems to always lead to one more problem hahaha.

[eneuro]

Yep, but from the begining its purpose it is help me assemble prototype boards and no need for fully automatic mode-just want to be able drive its axes using something like gesture recognition by processing accelerometer and magnetometer as user input interface and mayb even use VR googles to enable stereo mode with USB microscopes support.
Playing with those tiny SMT parts is a huge challenge, especially when they throw from tapes on table and it is boring to use magnifing tools to recognize needed chips...

I believe at some point mechanics and a few drivers will let build this machine PCBs itself in comfortable way, so will be able automate things not everything at once but step by step.
Control visual cameras first, move axis manualy add motors to each axis.... see how they perform in typical assemby tasks and maybe finally I will be able assembly simple rather small  SMT boards.
It could help make inspection of production units too.

It is interesting challenge

BTW: I wonder, how much steps do you have when rotating those parts around Z vertical axis? Trying to calculate this based on typical smallest parts packages I might want to use.
Quick estimation: 0.1mm movement at radius 10mm gives estimate 0.01 [rad] angle, so it leads to ~600 steps per revolution if I wanted such angle precision around Z axis 
Could you reveal how precise your rotational positioning is, eg. when correcting this? Probably a lot of steps, while you are dealing with such small parts, I never attempt to solder at home using my PnP?

[ProtoVoltaics]

Hello all, we were just invited to the World Maker Faire in NYC. This will unfortunately make us unable to attend the Milwaukee Maker Faire as they are on the same days.

Here is a list of the Maker Faires we will be attending in September and October.

Orlando Maker Faire, Atlanta Maker Faire and World Maker Faire NYC.

We are still waiting to see if we get an invite for the Pittsburgh Maker Faire and Maryland Maker Faire. This will complete our Maker Faire schedule for 2015.

[ProtoVoltaics]

Official schedule for those who would like to see the machine in person.

Cincinnati Ohio Mini Maker Faire              August 28,29
Orlando Maker Faire (Florida)                  September 12, 13
Maryland maker Faire                              September 20
World Maker Faire (New York, Queens)   September 26,27
Atlanta Maker Faire (Georgia)                  October 3,4
Pittsburgh Maker Faire                             October 10,11

[eneuro]

Official schedule for those who would like to see the machine in person.

I see no youtube schedule to see complete process of example PCB assembly using this thing...

[ProtoVoltaics]

Official schedule for those who would like to see the machine in person.

I see no youtube schedule to see complete process of example PCB assembly using this thing...

We have no planned date for a YouTube video release, but we imagine there will be a full video out before the next Maker Faire.

I'd estimate 2-4 weeks from now.

[ProtoVoltaics]

https://youtu.be/DLqMLY7oD14

Our latest test video. This video is a test to see where errors would occur. We were testing to see if our vision could recognize parts in our feeders, parts in different kinds of trays, and parts from cut tape. We were also testing the automatic tool changer and the upward looking camera (located between the auto feeders). The test placement was successful from all feeder types, and we will now be adding more error checks as well as optimizing for increased speed. This test was NOT meant to show the current speed of the pick and place.

There are some obvious things that need to be fixed/optimized. We are currently working to get these things done. We wanted to wait to post another video till we had those things ready, however, we have had a few people write in wanting to see the progress we have made.

We welcome any and all feedback. 

[ProtoVoltaics]

After lots of testing we successfully did our first small batch run on the machine. We will be making a new video, and starting our KickStarter soon!

[ProtoVoltaics]

Here is our official project video! You can see it all here! Auto tool changing, multi-board placement, IC placement, small part placement, placement of caps, Software/UI, computer vision, and more. Please take some time to watch the video and let us know what you think. This is only the beginning! This is a product we will be using ourselves and we will be continuously updating the software and hardware to keep the machine as awesome as possible!

https://youtu.be/Ny99KzsySHE

[matseng]

So how can you improve the speed?  It looks like it currently does 3-4 parts per minute compared to the old standby TM220 that does over 100.  And at that speed the paste will dry up before a reasonable complex board is populated....

[mikeselectricstuff]

That is the slowed-down demo mode right ..? What's the actual CPH?
Why does it rotate the part while sitting over the camera, instead of measuring the rotation & offsets and rotating on the way to the placement position?
How well do those feeders work with plastic tapes?
What aboiyr really wide (24mm) plastic tapes?

[ProtoVoltaics]

With all the vision corrections, and part detection it slows it down a lot. With everything on, for every part, it runs at about 300 parts per hour. Mostly, it is the angular alignment that takes so long. This is something that we are actively working on, and for the purposes of the video in my previous post we are doing angular alignment on all parts which isn't necessary.

Our machine makes heavy use of computer vision, and there is an opportunity to make the processing faster by having a graphics card perform some of those vision tasks instead of having the computer's processor do them. This development is in the works now, and is expected to greatly speed up the CV processing.

However, with the angular alignment and part detection turned off to mimic more how the TM220 works (no vision) then the speed jumps up to about 850 parts per hour. The "X" and "Y" vision alignments will remain on always. You can see this in the video below. We have since this video made some modifications to the machine, and expect it to be somewhat faster than shown in this video with the said parameters turned off.

https://youtu.be/C7Pi4akL5og

If you need to do angular alignment for all parts the current speed is very slow. For our board in the video we really only need the angular alignment for the IC placements.

The reason it doesn't find the angular correction, and do it on the fly is because there is slack in the A-axis belt. This is a fault in our currently displayed pick head. We have received in the parts for a new pick head that will remedy this issue and allow us to do just that.

As for the feeders, they are actually really great and have lots of torque, but I'll test them out and get back to you on the 24 mm plastic tapes.

[eneuro]

it runs at about 300 parts per hour

Lets make quick calcualtions: 1 hour is 3600 seconds this means that you have ONLY 1 part per 3600/300= 12 seconds ? 
I'd expect something like 10x times faster from machines like those... else at this speed it could be boring to watch such long PCB assembly 

It looks like humans can be faster, with a litte help to handle those parts automatically, so maybe may not low cost, but my cheap PnP will be faster even in manual mode? 

Anyway, thanks for sharing this interesting video-probably I'll try to challenge you in... manual mode with a little help with automated parts handling and... human brain etc, especially for small PCBs, with well known BOM.

BTW: Which could be estimated low cost of your  PnP if it were available to buy?

Update: There is problem with those videos-it looks like you recorded it in very bad room lighting conditions and a lot of reflections from point light  sources 
Wouldn't you created something like ring or sphere with low power LEDs but plenty of them distributed around workpiece machine table to help its CV recognize things?
Probably the first thing I'd like to do is create perfect lighting conditions when CV is implemented into this thing... but ok, this is only my vision how my working table lighting for PCBs assembly will look like...
To much reflections makes those videos very difficult to watch... To avoid any reflection eg. when making amateur photos with flash, I've covered flash with white paper, so no problem with flash from small distances even without great external lighting, so for professional video recording or photography I think, smooth light is a must, else those reflections on PCB destroys image...and I wonder if it doesn't confuse machine CV? 

[mikeselectricstuff]

You seem to be wasting a lot of time hanging around the camera - it should just take a snap and figure out rotation etc. on its way to the placement position.

IMO anytyhing below 1000cph isn't worth spending significant money on. You will be killed by newer Chinese machines with vision.

[rx8pilot]

You seem to be wasting a lot of time hanging around the camera - it should just take a snap and figure out rotation etc. on its way to the placement position.

IMO anytyhing below 1000cph isn't worth spending significant money on. You will be killed by newer Chinese machines with vision.

I would agree, 300cph makes it hard to be excited about. It would be somewhat faster than a human, especially considering the machine does not fatigue mentally and physically, but it it still slow, slow, slow. For pro use - it's useless. For semi-pro/hobby use, it's cost/performance is not there.

Not sure what you are up against, but I would suggest you make it much faster or prepare for very modest sales. How much is this expected to cost?

[Bud]

Please take some time to watch the video and let us know what you think.

With all due respect, i fell asleep after 3 minutes of watching.
Not sure what cv software you using, if it is OpenCV, i just finished a project with it in the other tread and i think it was slow as hell.

[ProtoVoltaics]

Thanks for the feedback, Eneuro quoted a small section of text. The full text is "With everything on, for every part, it runs at about 300 parts per hour."

With 0603's and 0805's and so on you don't really need angular alignment. This can be turned off for these parts. You really only need it for 0402's and smaller and IC's. We left it on for every part during the video to show that it is a capability we have with our PnP.

You seem to be wasting a lot of time hanging around the camera - it should just take a snap and figure out rotation etc. on its way to the placement position.

I tried to answer this in my last post. We are wasting a lot of time over the camera currently. The reason we are is because we have to wait and see if the rotation got applied correctly. Currently there is backlash in the angular alignment belt this is currently being worked on. When this is fixed we will be able to quickly take a snap shot and apply the correct in route to the placement area.

I would suggest you make it much faster

Haha yes, we were not planning on stopping now.

[eneuro]

Not sure what cv software you using, if it is OpenCV, i just finished a project with it in the other tread and i think it was slow as hell.

With modern GPUs OpenCV can be very fast even on tablets with a lot of GPU cores available in some NVidia crated for gamers, but of course it require design this CV system with this acceleration support in mind to be able  recompile it  with GPU enabled or disabled eg. in complile time, just by passing to makefile -DGPU or as environment variable in Unix/Linux something like this which could help optimize written OpenCV code  and speed up critical parts of CV system;)

Anyway, when watched 2nd video we have on top right list in YT this robot and this is really great concept which creates brains storms howto DIY at home something like this, but probably we need a lot of steps in those four motors on top, to be able translate its angular rotation of those arms to decent precision needed to place things, but yeah it looks great and... fast 
High-Speed Pick & Place Robots by Bastian Solutions



How do you want compete with those 100 pick & places per.... minute? 6000 per hour? 

High Speed Adept Quattro Robot handling steel balls on conveyor


BTW: Looking at this I've improoved my PnP machine mechanics a lot and... I've now clear vision howto build its axis, so after some brake to let brain find solution to this problem which puzzled me for a few weeks now it is time to  my concept shown in other thread and keep it going to speed up manual assembly first, than try automate this and make faster than humans can do it- like those amazing robots linked above 

[matseng]

However, with the angular alignment and part detection turned off to mimic more how the TM220 works (no vision) then the speed jumps up to about 850 parts per hour. The "X" and "Y" vision alignments will remain on always. You can see this in the video below. We have since this video made some modifications to the machine, and expect it to be somewhat faster than shown in this video with the said parameters turned off.

Still, that nearly an order of magnitude slower than the TM220 that does 7000 cph...   

The X and Y rapids seems quite fast, so it must be the dead slow Z axis that does it.  Most hobbyist-style machines puts the components *really* slowly and carefully - this in contrast to the "real" machines that slams down the component with such speed that I'm amazed that the solderpaste doesn't end up at the walls in the room. ^_^

[rx8pilot]

Just for reference, I just hand placed 7 PCBs with 120 parts per board at about 200cph. This is the average speed - NOT the burst speed. The parts are very well organized and sequenced and I have the PCB memorized. I also have a carefully worked out placement bench. There are about 10 fine pitch .4mm parts that are slower and some SOD-523's that are slower than the 0805's. This info is only to provide a reference for how a slow PnP can compare - it is still faster and less tedious while placing parts.

I will still say that even though 300cph from an automatic machine that can do it all day long is always going to be better than hand placing - it would be nice to see it go up to 800-1000cph. The TM-220 (which I have not used or own) seems like a machine that can only do the big parts without crossing your fingers and relying on 'Hope'. Therefore I will discount that model as a viable option to be a complete solution.

I have a Quad 4000C that is just getting up and running. It is rated at 3600cph, but in reality probably around 3000 or maybe a little less depending on the optimization of the feeders. At that rate, it is faster than I can print the solder paste and get them through the batch oven. The good part is that I can have 80+ feeders on it at once so setup is super easy. All the typical parts are on the machine all the time. Tube feeders, strip feeders, matrix trays, custom trays, in addition to all the reel feeders. This is where the time savings are for me. If I have to load the machine for every PCB, I would be super slow the and 3000cph would barely be noticed after loading and verifying the machine which takes a while if done from scratch. Another reference point is that this machine was around $6k or so.

Still trying to figure out the target audience for this machine though. I got a used commercial model because my living will depend on this machine. The only reason I got it was because I was hand placing tens of thousands of parts per month. For a hobbyist, why would you need (want) this? For a professional, is it helpful/reliable enough?

Again, anyone know what the target pricing is for it? Is LitePlacer getting anywhere with theirs?

[eneuro]

I have a Quad 4000C that is just getting up and running. It is rated at 3600cph, but in reality probably around 3000 or maybe a little less depending on the optimization of the feeders. At that rate, it is faster than I can print the solder paste and get them through the batch oven.

This one  in action?
PPM Quad 4000C IVC 4C pick and place demo & feeder setup


This kind of machines can fill pads with solder paste or it could be too slow, so it is printed instead to speed up things?

[quantalume]

I have a Quad 4000C that is just getting up and running... this machine was around $6k or so.

At that price point, what sort of shape was the machine in when you obtained it? How much work was involved in getting it running? Thanks.