Neoden 4 pick and place

[mikeselectricstuff]

So where does the feeder plug in ? And how are addresses assigned? Did that feeder come with any instructions ?

[ServoKit]

I'm fairly certain that there must be another controller of some sort in the machine, likely a USB to CAN interface, this one should also show up in device manager. The pics from the motherboard show several USB connections (note text on cables and wire colors typical for USB). We already established the GPIO chip (ITE IT8781F), likely for limit switches, solenoids and the like. I too think that the cameras are plain USB devices.

Regards, Axel

[Ichan]

So did you basically just go through every part and off set the placement data by the amount it was being placed incorrectly?

If you knew what you know now about getting the file setup how quickly do you think you could have had the machine up and running after you got it setup?

1. No, i just scale up the cad data coordinates by 1.002 using spreadsheet (now i am thinking how to install ms excel on  the machine ), including the mark point / fiducial coordinates.

2. Within one day, well add one more day to setup the feeders .

I have to say that N4 software is already workable, yes there are some bug and strange behavior, but all softwares are like that they all will have many revision before become stable, right?

-ichan

[vonnieda]

So where does the feeder plug in ? And how are addresses assigned? Did that feeder come with any instructions ?

All things I'd like to know too It did not come with any instructions or documentation.

[mrpackethead]

I need a prototype / low volume production machine (100~500 boards) that can reliably place 0201 passives, and tiny 0.35mm pitch board to board connectors (e.g. Molex 504622 series). Has anybody had success placing these small parts? Other than Neoden sales propaganda, I couldn't find anything.

I'm very much in support of getting openPNP working with the N4, so if I buy one, Jason you are most welcome to come over and hack and packet-sniff our machine.

Tim.

The SmallSMT ( check the other thread ) folks claim they can place 0201s'  they posted some pics. 

[ar__systems]

The SmallSMT ( check the other thread ) folks claim they can place 0201s'  they posted some pics.

What I noticed from my experience is those pics or videos are very misleading. Sure they populate 20 parts that would look decent. What about 1000 or 10000 parts? How many times will you have a mispick or part lost or installed crooked? I really would like to see those specs from manufacturer.

[mrpackethead]

The SmallSMT ( check the other thread ) folks claim they can place 0201s'  they posted some pics.

What I noticed from my experience is those pics or videos are very misleading. Sure they populate 20 parts that would look decent. What about 1000 or 10000 parts? How many times will you have a mispick or part lost or installed crooked? I really would like to see those specs from manufacturer.

In deed, and its unlikely they are going to post pics that show such events.. Heck even the $250,000 line that i use for my large scale production screws up from time to time.  ( often assisted by operator stupidity )

[thommo]

Congratulations Ichan !!!
I'm so happy for you, and it's great to know you've developed a method that will allow you to proceed to production now.

Looking forward to the next stage where you produce the first batch.
Is the N4 connected to your reflow oven now?

Your mentioned installing Excel onto the machine. Is your intention to try and interface it with the application somehow?

Also, did you get a response from NeoDen about multiple heads loading from the same feeder?

So did you basically just go through every part and off set the placement data by the amount it was being placed incorrectly?

If you knew what you know now about getting the file setup how quickly do you think you could have had the machine up and running after you got it setup?

1. No, i just scale up the cad data coordinates by 1.002 using spreadsheet (now i am thinking how to install ms excel on  the machine ), including the mark point / fiducial coordinates.

2. Within one day, well add one more day to setup the feeders .

I have to say that N4 software is already workable, yes there are some bug and strange behavior, but all softwares are like that they all will have many revision before become stable, right?

-ichan

[timbo73]

The SmallSMT ( check the other thread ) folks claim they can place 0201s'  they posted some pics.

Thanks, I'll check it out. Our boards are tiny, with only a couple of dozen off-chip components (most of the work is done by an ASIC). If all boards need to be visually inspected and 10~20% need an 0201 nudged back into place, that would be sufferable. But if it routinely screws up the placement of the fine pitch connectors, that's a deal-breaker.

[mrpackethead]

https://youtu.be/aeW66c9oN5Y

[Ichan]

We run the machine for about 5 hours non-stop today, finishing 31 panel (124 board). Still some pick failed happen but better than my other machine, really!

Here is the video, taken from several angle - watch something at the end of the video .

https://youtu.be/pBTZm1Lf3WM

Congratulations Ichan !!!
I'm so happy for you, and it's great to know you've developed a method that will allow you to proceed to production now.

Looking forward to the next stage where you produce the first batch.
Is the N4 connected to your reflow oven now?

Your mentioned installing Excel onto the machine. Is your intention to try and interface it with the application somehow?

Also, did you get a response from NeoDen about multiple heads loading from the same feeder?

Thank you. Yes the pcb goes to the conveyor oven on the first floor, before that some manual placement done plus visual inspection and correction.

I am not very serious about the Excel, just quite annoyed that i have to go back and forth to another pc.

I do not have many conversation with Neoden so far, besides i know they follow this thread closely - let they decide themselves want to win the competition or not.

In general i am fairly happy with the machine, not in big amount but for me i win the bet - it will be a jackpot if later i can use the feeders on the other machine.

Jackpot? I bought also N4 extra feeders 8x8mm + 4x12mm + 4x16mm, the cost of similar feeders for the LS60V will be enough to buy N4 machine with full feeders! (using their 2010 price when i bought the machine).

-ichan

[rwb]

It's really nice to see the machine up and running consistently for you now

I see that you placed some weights or quarters on one of the part lines. Tell us more about this.

Are any particular parts being miss picked consistently?

Are those small USB no lead .5mm pin chips being placed accurately every time?

Is anything needing fixed up on each board by hand consistently?

Now I did find a place that does small scale fabrication. They have a online pricing tool and I was wondering if you could load your design into to and see what kind of cost they would charge you to run the same 2000 boards vs what you think your cost is by doing it in house. The cost savings of having a Neoden 4 machine is something that is really hard to calculate and its something I would love to see more comparison on.


The company that does this small scale fab is here: https://macrofab.com/

[vonnieda]

Here is the video, taken from several angle - watch something at the end of the video .

This is a really great video, thank you for posting it! I noticed several interesting things:

* They are not using linear interpolation on the motion planner. This tells me that they are probably not using one of the common open firmwares available for CNC machines since most of these don't implement G0 properly. They could be using TinyG, I suppose, since I think that it does implement it properly, but based on how jerky the motion is I doubt it.

* It looks like they can only image all four nozzles for very small parts. A little tough to tell, but it looks like the upper limit is SOT-23. This seems to be the case for the nozzle calibration as well. During startup it images all four nozzles once and then appears to image the larger nozzle separately.

* It appears the camera only captures one image at a time, rather than a stream. I noticed that as it's loading a board with the conveyer it's strobing the top LEDs instead of just leaving them on. This makes me think their camera API consists of something like "strobe and capture" versus "turn on strobe", "capture", "turn off strobe". I would not be surprised to find that the LEDs are wired to the cameras for triggering.

Very cool to see all this in action. Maybe not that interesting to a machine owner/user but for someone who develops PnP software it's very enlightening!

Jason

[mrpackethead]

On the cost equation..  For me at least, i know that getting a machine, and loading  more than afew 10's of boards will actually cost me MORE than having them fabricated in China, in a PCBA factory.   Our labour costs kill it every time a monkey.

The reason for me to do this, is the sub < a few 10's of boards, where speed and time is of the essense.  Yes, they will cost more, but the cost of time is way more valuable.  I figure i'll be able to save 30% in elapsed time on projects from design --manufacturing sampel.

It's really nice to see the machine up and running consistently for you now

I see that you placed some weights or quarters on one of the part lines. Tell us more about this.

Are any particular parts being miss picked consistently?

Are those small USB no lead .5mm pin chips being placed accurately every time?

Is anything needing fixed up on each board by hand consistently?

Now I did find a place that does small scale fabrication. They have a online pricing tool and I was wondering if you could load your design into to and see what kind of cost they would charge you to run the same 2000 boards vs what you think your cost is by doing it in house. The cost savings of having a Neoden 4 machine is something that is really hard to calculate and its something I would love to see more comparison on.


The company that does this small scale fab is here: https://macrofab.com/

[mikeselectricstuff]

* It appears the camera only captures one image at a time, rather than a stream. I noticed that as it's loading a board with the conveyer it's strobing the top LEDs instead of just leaving them on. This makes me think their camera API consists of something like "strobe and capture" versus "turn on strobe", "capture", "turn off strobe". I would not be surprised to find that the LEDs are wired to the cameras for triggering.

but it doesn't do this when looking at fids. Wonder if it's due to alternating move-rail and take-image commands.

* They are not using linear interpolation on the motion planner. This tells me that they are probably not using one of the common open firmwares available for CNC machines since most of these don't implement G0 properly. They could be using TinyG, I suppose, since I think that it does implement it properly, but based on how jerky the motion is I doubt it.

Is this actually a problem? I can see how it could actually be a benefit, as it means each axis always goes through a consistent speed profile. With interpolation, speeds will be very variable, which may cause issues with resonances at particular speeds, causing possible part movement/drop issues.
 
As noticed before, at slow speeds it still goes slow when there is no part - I can't see any excuse for this and suggests to me that the controller is pretty dumb and has only a global speed control. ISTR some mention that Z speed couldn't be independently set from XY speed, which is even more dumb.

Can you set it up so you can place the next PCB(s) in the rail and have it auto-load the next when the last one is done? Can PCBs be loaded end-to-end to fill as many as will fit the rail?
 

[vonnieda]

Is this actually a problem? I can see how it could actually be a benefit, as it means each axis always goes through a consistent speed profile. With interpolation, speeds will be very variable, which may cause issues with resonances at particular speeds, causing possible part movement/drop issues.

Nope, not a problem at all. I've been preaching for years that we (OpenPnP) need a dead simple motion controller that skips all interpolation and just does per axis acceleration and and velocity.

[thommo]

Thanks for publishing this MrP.

Well done Jason. The Fids interface looks very clear - something we can't say for the N4 at present.
And the machine vision is coming along great too.
How close do you feel you are to estimate work on the N4? Are you still waiting on those serial bus dumps?

Are you able to turn off the top camera LEDs while you're using the bottom camera yet?refucing extraneous light sources will definitely improve resolution and definition.

I made a comment in SmallSMT just before about the bottom camera LED array being square/rectangular, instead of circular. Is this what you are doing? You will get MUCH better results illuminating pads and legs etc if the light source is parallel to the edges rather than in an arc. All the large commercial machines appear to use box configurations of LEDs rather that circular or tube arrangements.

Image sensors and components are square/rectangular after all. Only lenses are circular ;-)

https://youtu.be/aeW66c9oN5Y

[vonnieda]

Thanks for publishing this MrP.

Well done Jason. The Fids interface looks very clear - something we can't say for the N4 at present.
And the machine vision is coming along great too.
How close do you feel you are to estimate work on the N4? Are you still waiting on those serial bus dumps?

Are you able to turn off the top camera LEDs while you're using the bottom camera yet?refucing extraneous light sources will definitely improve resolution and definition.

I made a comment in SmallSMT just before about the bottom camera LED array being square/rectangular, instead of circular. Is this what you are doing? You will get MUCH better results illuminating pads and legs etc if the light source is parallel to the edges rather than in an arc. All the large commercial machines appear to use box configurations of LEDs rather that circular or tube arrangements.

Image sensors and components are square/rectangular after all. Only lenses are circular ;-)

Hi thommo,

Yes, the serial dumps are the main thing I need to be able to estimate the work. Still waiting and hoping

Re: turning off the LEDs: Not yet, although I've got a plan for it. Just need to do some wiring. Right now I'm finding I don't need it so I am holding off until I finish the first pass at bottom vision.
Part of the pipeline that processes the image for bottom vision masks out everything within the green circle that you see in the video, so the LEDs from the top camera are not considered in the image.

In case you are curious, here is my current bottom vision processing pipeline: http://pastebin.com/fCp7JbkB
The filtering portion of the pipeline primarily masks out the circle and then subtracts green hues from the image, which pulls out the large disk above the nozzle as well as the green part of the nozzle itself. This is working very well so far, although I have to add scaling of the circle mask for tall parts since the disk "shrinks" with respect to the image when the nozzle is higher.

I do want to switch to rectangular array of LEDs but I have not been able to find one pre-made. The rings were cheap on eBay. I'm probably just going to design one and make it a low angle light at the same time. Just another thing on the TODO list

Jason

[TheSteve]

We had a very successful build today - the best yet. Powered the machine off for a bit, when we turned it back on it made a thunk sound(pretty normal) and then died. It is currently dead, no signs of life.
We're opening it up now.

[mrpackethead]

We had a very successful build today - the best yet. Powered the machine off for a bit, when we turned it back on it made a thunk sound(pretty normal) and then died. It is currently dead, no signs of life.
We're opening it up now.

Ewww....

[TheSteve]

Well nothing too scary in the end, just the main fuse. It is only rated at 3 amps which seems awful low to us. I wonder if Neoden has spec'd different fuses for 120 and 240 VAC. The fuse is integrated into the IEC mains input and actually holds a spare.

[rwb]

We had a very successful build today - the best yet. Powered the machine off for a bit, when we turned it back on it made a thunk sound(pretty normal) and then died. It is currently dead, no signs of life.
We're opening it up now.

How many boards did you guys run? Were they the same red boards that were pictured in the post you made not that long ago?

Are you have any other feeder issues now that you have put some weight on end of the reels that were giving you some trouble in the past?

What exactly is that weight accomplishing?

It's stating to sound like everybody has finally figured out how to program the machines to the point they can run thousands of boards without to much trouble which is very promising.

Having looked at the more expensive and stable machines on the market it becomes very clear how revolutionary the NeoDen 4 is when it comes to price, weight, and size.

[TankSparks]

We had a very successful build today - the best yet. Powered the machine off for a bit, when we turned it back on it made a thunk sound(pretty normal) and then died. It is currently dead, no signs of life.
We're opening it up now.

How many boards did you guys run? Were they the same red boards that were pictured in the post you made not that long ago?

Yes, same red boards,  We ran 5 panels with 560 parts on each.  So over 2500 parts,  no stoppage.  It was nice to be able to get some work done while the machine ran.   I can hear it in my office upstairs so I know if stops.  One panel takes about  24 minutes.   I was running most parts at 60% speed.   2 different nozzle sizes,  3 small and 1 larger one for the IC.
0603 parts and .65mm pitch IC
I'm using 2 fiducials bottom left and top right.

Last time we had a 12mm feeder problem with parts jumping out.  at the time didn't know if was a bad feeder or a crappy feeder design problem...turns out it was just a bad feeder/peeler.  I used a different feeder this time and it went very smooth no parts jumping and tape peeler working like it should and no retries.   
Also had feeder issues with the 24mm feeder with another board,  still need to test again with the 2nd 24mm feeder that I have. 

[thommo]

Hi TankSparks,

Did you say 24mm N4 feeder?

I asked recently and they said they no longer support the 24mm feeder. It has some 'undefined issues' and they are trying to redesign it [at least that's what I took from the conversation].

Great news about the general stability all the same! Thanks for the feedback too.

We had a very successful build today - the best yet. Powered the machine off for a bit, when we turned it back on it made a thunk sound(pretty normal) and then died. It is currently dead, no signs of life.
We're opening it up now.

How many boards did you guys run? Were they the same red boards that were pictured in the post you made not that long ago?

Yes, same red boards,  We ran 5 panels with 560 parts on each.  So over 2500 parts,  no stoppage.  It was nice to be able to get some work done while the machine ran.   I can hear it in my office upstairs so I know if stops.  One panel takes about  24 minutes.   I was running most parts at 60% speed.   2 different nozzle sizes,  3 small and 1 larger one for the IC.
0603 parts and .65mm pitch IC
I'm using 2 fiducials bottom left and top right.

Last time we had a 12mm feeder problem with parts jumping out.  at the time didn't know if was a bad feeder or a crappy feeder design problem...turns out it was just a bad feeder/peeler.  I used a different feeder this time and it went very smooth no parts jumping and tape peeler working like it should and no retries.   
Also had feeder issues with the 24mm feeder with another board,  still need to test again with the 2nd 24mm feeder that I have.

[Ichan]

It's really nice to see the machine up and running consistently for you now

I see that you placed some weights or quarters on one of the part lines. Tell us more about this.

Are any particular parts being miss picked consistently?

Are those small USB no lead .5mm pin chips being placed accurately every time?

Is anything needing fixed up on each board by hand consistently?

Now I did find a place that does small scale fabrication. They have a online pricing tool and I was wondering if you could load your design into to and see what kind of cost they would charge you to run the same 2000 boards vs what you think your cost is by doing it in house. The cost savings of having a Neoden 4 machine is something that is really hard to calculate and its something I would love to see more comparison on.

The company that does this small scale fab is here: https://macrofab.com/

1. The weight mean to add a tension on the tape. Small parts in plastic tape like to do circus flea actions, that because the plastic tape had some springy action when some (unwanted) force applied making the parts jumping out of the pocket. The weight i use is a small spanner held in position by some magnets, adding some tension will damp that springy action.

2. That is the usb chip who like to do the acrobat, the others are all calm.

3. On these boards we found about 1 out of 5 panel need correction on those two small ic, seems it caused by not good reference point (fiducial) used / available. Can be seen on the video sometimes the fiducial recognition is not good finding the center point of the drilled hole. I do not want use manual recognition, i just want to enjoy pressing that tempting green start button and go.

4. About outsourcing, if you have good control over them then it would be a good choice, IMO.

-ichan