Reliable Placement of 0402-Sized Components

[Michel Goto]

Hello developers and community members,

Can all of you consistently and accurately achieve stable and precise placement of 0402-sized components on a PCB?

I have been using OpenPNP for approximately three years now, primarily for 0402-sized components. However, when it comes to placing these components on the PCB, I often encounter alignment issues. Over these three years, I have been unable to resolve this problem, and I find myself using a microscope to make adjustments with tweezers after component placement, which can be quite exhausting.

I suspect that the root cause may be related to the precision of my DIY machine or calibration and configuration issues with OpenPNP, but I have been unable to identify the exact cause or find a solution. I am truly perplexed about what to do next. I am even contemplating building a new small machine (with a rail length of 400mm×400mm) from scratch to reset everything.

If any of you have been able to reliably place 0402-sized components, it would greatly inspire me. I would sincerely appreciate any information or assistance.

P.S. My ultimate goal is to work with 0201-sized components.

[Patricio Cohen]

Hi,

I'm very interested in this topic, since I have been trying for several years to have a reliable placement of of 0402 components. I have made some progress, but it still needs tweezer work to be done. My wish is to be able to leave the machine unattended someday.

Some of the progress I made was strictly related with identifying  any loose parts involved in the coupling between the motor and the belt. Sub millimeter plays can make a difference when dealing with 0402.

BR

Patricio

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/8f7bcd4e-03dd-4e84-ba0d-d464f9fef23dn%40googlegroups.com.

[mark maker]

Hi,

Since both of you speak of "several years", I assume you used much older OpenPnP versions and nay have legacy machine configurations, so I dare ask the following:

1. Have you tried with a recent test version? There was a lot improved lately, by various contributors.
   
2. Have you done all the proposed Issues & Solutions calibration?
   a) from scratch with a fresh config directory (having copied/backed up the old one), or
   b) on top of the existing configuration?
   
3. Have you made sure to get the new self-tuning pipelines in bottom/fiducial and nozzle tip calibration vision?
   (I should add detection of outdated pipelines to Issues & Solutions too...)
   

Between all the calibrations I believe 0402 should be possible on almost any machine. Yes, it may require some conservative settings like hefty backlash compensation, that may slow down the machine, but this will surely still result in higher CPH than "microscope and tweezers".

And if the calibration and compensation measures do not help, they will likely point to the problem. For instance, you get detailed diagnostics from both Tony's Advanced Camera Calibration and my Backlash Calibration, and from Camera Settling Diagnostics, and from Nozzle Tip Calibration, and from automatic Background Calibration etc.

https://youtu.be/md68n_J7uto

https://youtu.be/Pxg6g3KI5_E

Some features like Simulated 3rd Order Motion Control (Jerk Control) are available (on some controllers at least) and may solve issues with machine shaking etc.

https://youtu.be/cH0SF2D6FhM

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CAA_7zy38KrGhbN1%2B7nSTLRFj%2Bn8FhMtu0igFptpgvbjszo3OOw%40mail.gmail.com.

[Michel Goto]

Hi, Mark
Thank you for your comments.
I always have great respect for your development efforts.
In response to your comments, I respond to the following

1. Regarding the version of openPNP, I am using the latest test version.

2. Concerning Issues & Solutions calibration, I often encounter errors during execution that I cannot resolve.
   So I don't think it's in a perfect state.  It would be helpful to know the order of items to be resolved (order of priority), as my thinking is confused as to where to start.

3. Adjusting the pipeline is very challenging for me. I have only worked with the threshold level.

I believe this is a matter of my own abilities, but understanding and adjusting openPNP is truly difficult.
When I experiment with openPNP at work, I end up spending many time and can't do other tasks. Therefore, I was thinking of building a new machine that is solidly assembled and keeping it at home to calmly learn and experiment during evenings and weekends.

Thank you for sharing the YouTube videos. I would like to review them.

I primarily use Mark's Push Pull Feeder, and I think it works very well.
Thanks to Mark's explanation, I now have the courage to know that I can reliably place the 0402 component.

 By the way, I'm unsure about which nozzle to use for the new machine. My current machine uses a Juki Nozzle, but I'm debating whether to go for the CP40 or  CP45...

My long journey continues.
Thank you very much.

2023年9月5日火曜日 20:54:04 UTC+9 ma...@makr.zone:

[Mark]

Re 2:  >  It would be helpful to know the order of items to be resolved (order of priority), as my thinking is confused as to where to start.

Issues & Solutions has the system of Milestones, to make it a somewhat logical overall progress. Furthermore, it should continuously analyze the current machine setup and automatically propose what comes next, and not list solution, if the prerequisites are not yet done.

However, both of these aspects might not really work 100% for old configurations, where things were calibrated and configured by hand before, and where Issues & Solutions initially "guesses" the Milestone from the old configuration state. The dependencies logic is also not easy, there are sooooo many combinations possible. There are almost certainly cases, where it could misjudge the situation.

Note however, that Issues & Solutions deliberately leaves it to you to chose the next step out of those it thinks are ready. Unlike many other extremely rigid "workflow" systems, Issues & Solutions aims to be a flexible as possible, to accommodate the imperfect and spontaneous DIY world. You can also go back and revisit stuff.

If you want, send the machine.xml. I'll have a look.

Often users also start from scratch, by backing up and removing (or simply renaming) the .openpnp2 config directory. OpenPnP will then create a new one and start fresh.

Although it is much easier and faster than before, it is quite some work (especially if you have many feeders). You should plan for it to take time and energy. But frankly, it is likely worth it. 😁

And having the backup, you can always switch back, in case you're stuck or out of energy.

Re 3. > Adjusting the pipeline is very challenging for me. I have only worked with the threshold level.

Using the new pipelines it should be very easy! That's another reason for a fresh start. Watch this animation:

An perhaps read the Wiki:

https://github.com/openpnp/openpnp/wiki/Bottom-Vision

> By the way, I'm unsure about which nozzle to use for the new machine. My current machine uses a Juki Nozzle, but I'm debating whether to go for the CP40 or  CP45...

Best open a new discussion thread for that separate question.

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/99ea4420-4e8c-4547-bfa5-85a4e303441dn%40googlegroups.com.

[Patricio Cohen]

Hi guys,

After Mark's reply I'm now convinced I should start my config from scratch. I have been working with my config from previous versions and maybe that is the culprit I'm not having a better performance with my machine. I will start fresh.

Thanks Mark for the tips

BR

Patricio 

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/e31e20d9-a300-e034-a18d-cef029869638%40makr.zone.

[Michel Goto]

Hi Mark,

Thank you so much for your detailed explanations. There are very helpful. 
I am now running Issues & Solutions from scratch.
I may have to redo it a few times.
It may take a little more time, but I will report back when I get the results.
I think the behavior of the machine has changed from before.
Hopefully it will work.

I agree with creating a new thread on nozzle selection.

Good luck Patricio !

2023年9月6日水曜日 10:39:01 UTC+9 mpat.co...@gmail.com:

[Michel Goto]

Hi Mark, everyone.

I performed the following operation four times, deleted the .openpnp2 directory and ran 'Issue & Solutions.

As a result, during the execution of 'Advanced camera Bottom calibration,' a 'Too many misdetects' error occurs.
Is the small screen area of the bottom camera in relation to the nozzle's movement position the cause?
Nozzle Tip calibration has passed without any issues.
I attach images and logs.

I would appreciate any hints on the cause of the error.

2023年9月6日水曜日 16:17:42 UTC+9 Michel Goto:

[Mike Menci]

Your Nozzle image seams very dark - try to adjust the vision so that you get more grey tip in image. Radial step settings you have now 32 - try 40. 

[tonyl...@gmail.com]

I wouldn't increase the number of radial lines - leave that at 32. But your nozzle tip does appear to be very out of focus. At the first Z level (which is where your images above were taken), the nozzle tip should be in good focus. You probably need to adjust that.  But, before doing so, you may want to consider lowering the camera to increase its field of view. 

Tony

[Michel Goto]

Hi,

Thank you for your advices.
By increasing the Exposure value of the bottom camera and changing the focal length of the lens from 12mm to 8mm, I was able to eliminate the error in 'Advanced camera Bottom calibration'.
However, when I actually placed the components, the accuracy deteriorated compared to before.
There might be something fundamentally wrong. Hmm.

I intend to start over completely from the beginning.

2023年9月9日土曜日 22:34:56 UTC+9 tonyl...@gmail.com:

[tonyl...@gmail.com]

Can you post your machine.xml file?

[Michel Goto]

Hi Tony,

After running Issue and solution from a clean slate, the accuracy issue may have been resolved. 

But the Up looking Camera image now looks like a fisheye lens.
The same symptom occurred when I ran Issue and solution from a clean state before, and I couldn't fix it, so I had to start Issue and solution all over again.
Is there a solution to this problem?
The camera is an ELP Full HD camera.
Attached is the current machine.xml.

2023年9月12日火曜日 10:39:50 UTC+9 tonyl...@gmail.com:

[tonyl...@gmail.com]

If you go to the Advanced Calibration tab for your Up Looking camera and scroll down to the Calibration Results/Diagnostics section, you will see you are getting very large errors (over 4mm) in some areas.  These all appear to be occurring at the left and right edges of the camera's field of view. My guess is you are getting a lot of mis-detects in those regions due to bad lighting when the nozzle tip is moved to those regions. You can either try to fix the lighting in those regions or (and probably the easiest thing to do) just crop the image. You will probably never actually use those pixels anyway (and it is a waste of computer resources to process them through your pipelines). I always advise cropping the up looking camera image down to a square - on the Advanced Calibration tab, near the top, set both the Cropped Width and Height to 1024 (your camera resolution is currently 1920x1024) and then rerun the Advanced Calibration.

After calibration is complete and you are satisfied with the results, you probably will want to move the Crop all Invalid Pixels all the way to the left.

Tony

[Michel Goto]

Hi Tony,

Thanks for your comment.
Sorry for the delay in responding.
As you pointed out, the problem was caused by the LED lighting in the four corners of the screen.
I changed the position of the bottom camera and the lights, but could not get the nozzle tip to be detected properly.
Next week, we will reexamine the camera mounting and LED lighting locations and start over from the beginning.

2023年9月13日水曜日 22:10:51 UTC+9 tonyl...@gmail.com:

[Michel Goto]

Hi Mark, Tony, and everyone,

I ran the "Issues & Solutions" process about 10 times from a clean slate. As a result, the accuracy of component placement for 0402 parts did not improve.

To provide a bit more context, it's not that all component placements are inaccurate; some components are placed accurately on the board, while others are misaligned. When we replace the board and start a new job, we get almost the same results consistently. This suggests a high level of reproducibility. Based on these findings, I suspect there may be an issue with the machine's accuracy. If there are any other things we should try, please let me know; it would be greatly appreciated.

Today, I adjusted the component positions with tweezers once again, but I'm not satisfied with the current situation. Therefore, I'm considering building a new, more accurate machine to improve precision.

2023年9月17日日曜日 18:36:58 UTC+9 Michel Goto:

[Mike Menci]

Hi Michel 

To give you a better answer we would need to know more about your PnP - can you send a photo of overall arrangement and the Head - how many nozzles do you have, do you change nozzles during the job and could it be that only one of nozzles is performing inaccurate ..... Picture of how parts look on >PCB after < Job is done and Trace file after Job was completed,.... 

What can you provide ?

Mike

[Mike Menci]

Please add as well images from your Folder - Look for date when Job was performed: 

[mark maker]

send the machine.xml

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/cbb35876-5b4a-4f6d-8083-391b1f7e36c0n%40googlegroups.com.

[Michel Goto]

Hi Mike, Mark

 I am currently in a place without the machine, so I will reply on Monday.  

Thank you.

2023年9月30日土曜日 23:20:48 UTC+9 ma...@makr.zone:

[Michel Goto]

Hi Mark , Tony, Mike

I will upload a video showing the state after placing 0402 size capacitors. I used only one nozzle and placed approximately 30 of them consecutively from a single PushPull Feeder. While some components are accurately placed, many are slightly misaligned and will require adjustment using tweezers.

I will upload the machine.xml file, as well as photos of the entire PNP machine and the machine head. The head was designed and manufactured based on Peter's head.

Thank you.

2023年10月1日日曜日 9:04:12 UTC+9 Michel Goto:

[Michel Goto]

[Mike Menci]

Is this the distance from your down Looking camera to PCB ?

On Monday, 2 October 2023 at 18:13:19 UTC+2 m3452...@dea-ac.com wrote:

[tonyl...@gmail.com]

I watched your video and it shows that the camera doesn't even align very well with component footprints. How many fiducials are you using and where are they on your board? Perhaps you need to tune your fiducial detection pipeline. If you can't get your camera to center better on each footprint, there's little hope in getting components to be placed any better than what your camera can do.

Tony

[Michel Goto]

Hi Mike,

Yes, as you understand.

To capture the pick location of the push-pull feeder, a camera has been installed in this position.

[Mike Menci]

What is the Z distance from Camera lens to your Push-pool feeder ? 

Is your feeder higher than your PCB for placement of parts on it ?

[M. Mencinger]

Can you find this folder and zip up some images and send us to see what your vision is seeing? 

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/b86537f6-fa49-450f-8928-61282723fa96n%40googlegroups.com.

--

Mike

"--

 **************************************************************************************************************

This e-mail may be confidential and it may also be legally privileged. If you are not the addressee you may not copy, forward, disclose or use any part of it. If you have received this message in error, please delete it and all copies from your system and notify the sender immediately by return e-mail. Internet communications cannot be guaranteed to be timely, secure, error or virus-free. The sender does not accept liability for any errors or omissions. 

 SAVE THE NATURE - SAVE PAPER - THINK BEFORE YOU PRINT!  "

[Michel Goto]

Hi Tony

I'm using three fiducials on the PCB, as shown in the uploaded photos. 

I ran the job after confirming that all three reference points are correctly recognized.
It seems like it might be an issue with mechanical precision, after all. 

I'm curious if your PNP machine can trace footprints more accurately than my Down-looking camera. I would love to know."

2023年10月3日火曜日 2:01:55 UTC+9 tonyl...@gmail.com:

[tonyl...@gmail.com]

> I ran the job after confirming that all three reference points are correctly recognized.

How did you confirm that? Can you post a video (or still pictures) showing how well the camera aligns over the three fiducials after the fiducial check has been performed? Remember with three fiducials, any small positional error in detecting their locations gets magnified as you get further away from the fiducials. It's too bad you don't have a fourth fiducial on the other corner of your board.

[Michel Goto]

Hi  Mike,

I believe the distance from the Dwon looking camera lens to the pick-up location of the push-pull feeder is approximately around 100mm.
Please note that this is just an estimate, as I haven't measured it myself; I'm currently at home.

Additionally, the height of the pick-up location of the feeder and the surface of the PCB are the same.

I attempted to copy the VisionSolutions image data to take home, but unfortunately, I failed.
I will try to upload it tomorrow. Please bear with me for a little while.

2023年10月3日火曜日 2:17:25 UTC+9 mike....@gmail.com:

[Michel Goto]

Hi Tony,

Before starting the job, I use "Position the Camera at the placement location" to ensure that the crosshair aligns with the center of each fiducial.
Afterward, I begin the job with "Check Fids" turned off. Is this approach acceptable?

In fact, there is a fourth fiducial, but due to the footprint of the surrounding components, it isn't recognized properly, and I currently have it disabled.
I've tried adjusting the pipeline in the "Fiducial Vision Settings" under the "Parts" tab, but I haven't been successful.

2023年10月3日火曜日 3:11:04 UTC+9 tonyl...@gmail.com:

[tonyl...@gmail.com]

> Before starting the job, I use "Position the Camera at the placement location" to ensure that the crosshair aligns with the center of each fiducial.

I assume you mean you are doing this after you've run the fiducial check, correct? If so, we would like to see a screen shot of each fiducial as you do this (including the fourth unused fiducial). Use your mouse scroll wheel to zoom in so we can better see how well it is aligned.

> Afterward, I begin the job with "Check Fids" turned off. Is this approach acceptable?

It should be assuming the earlier fiducial check went well. But there really shouldn't be any need to turn off "Check Fids" prior to starting the job either.

Regarding your fourth fiducial, also send a screen shot of it with the camera view fully zoomed out so we can see what is near it.

Tony

[Michel Goto]

Hi Mike,

I will upload four images in the VisionSolutions directory. I feel that image recognition with the Up looking Camera is working correctly.

Thank you.

[Michel Goto]

Hi Tony,

> I assume you mean you are doing this after you've run the fiducial check, correct? If so, we would like to see a screen shot of each fiducial as you do this (including the fourth unused fiducial). Use your mouse scroll wheel to zoom in so we can better see how well it is aligned.

Yes, you are correct. I will upload a video that includes screenshots of each fiducial (FD1, FD2, FD3, FD4). Is these images zoom level acceptable?

FD3 is often excluded from the fiducial check because it frequently fails to be recognized."

> It should be assuming the earlier fiducial check went well. But there really shouldn't be any need to turn off "Check Fids" prior to starting the job either.

That's correct. However, the probability of successfully recognizing all three fiducials (FD1, FD2, FD4) during Fiducial Check is low.

Therefore, after a successful Fiducial Check, I turn off 'Check Fids' before executing the job.
I will upload a video of the Fiducial Check execution. I have been trying to adjust the Exposure value for the Down Looking Camera, but it hasn't been successful.

One possible reason might be that the surfaces of some fiducials are not flat. I believe that some adjustments to the pipeline are necessary, but I don't understand the exact method.

>Regarding your fourth fiducial, also send a screen shot of it with the camera view fully zoomed out so we can see what is near it.

I will upload an image of FID3, which is the one that fails recognition the most. It is usually excluded from Fiducial Check because it is affected by the adjacent USB Type C connector's elongated hole.

Thank you.

[Michel Goto]

[Michel Goto]

[mark maker]

Go to Machine Setup / Vision / Fiducial Locator and set Max. Distance to 1mm. The default is currently at 4mm and that's way too much for packed PCBs. This default setting will have to be revised in a future version.

I assume, you have a horizontal parallax set? Is it really needed.

https://github.com/openpnp/openpnp/wiki/Fiducial-Locator#parallax-operating-principle

_Mark

On 04.10.2023 09:17, Michel Goto wrote:

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/d87a214c-6cde-484e-8897-72d7ee9efacbn%40googlegroups.com.

[Michel Goto]

Hi Mark,

Thank you for the advice.
I tried it right away.

Setting "Max. Distance" to 1mm resulted in the successful recognition of all four fiducials.
However, when I proceeded to run the job, there wasn't much improvement in the placement of the 0402 components.

Next, as you suggested, I adjusted the Parallax settings and then ran the job again.
This led to a significant improvement in component placement.
Before the adjustment, I had to manually correct approximately 50% of the components with tweezers, but after the adjustment, it seems to have decreased to about 20%.
I was surprised by this.

Upon closely observing the 20% that still required correction, I noticed that most of them were shifted in the Y direction.
My PNP machine has only one Hiwin-made MGN12 rail on the X-axis, which could potentially cause some head movement in the Y direction.
I plan to investigate this further tomorrow, such as by reducing the speed, to see if it makes a difference.

[mark]

have you done the y axis backlash calibration? what are the results?

On 4 Oct 2023, at 16:10, Michel Goto <m3452...@dea-ac.com> wrote:

Hi Mark,

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/e779136e-73e7-456c-80e4-1e30a0d599e1n%40googlegroups.com.

[Michel Goto]

Hi Mark,
I'm attaching an image of the results of the Y-axis backlash calibration that I performed immediately after the report I sent yesterday.

Today, I ran the Y-axis backlash calibration several times, and the results are significantly worse than before.
I wonder why that is.

[tonyl...@gmail.com]

What's going on every 36mm? Does that distance correspond to one rotation of something (stepper motor or pulley)? Perhaps you have something loose in your mechanical linkage.

Tony

[Michel Goto]

Hi Tony,

I'm having trouble imagining the cause of this 36mm cycle issue. I measured the dimensions of the moving parts of the PNP machine, but... One thing I noticed is that the X-axis is also fluctuating by the same 36mm. Why could this be?

[Mark]

Hi Michael,

Tony likely meant to say "36 steps", which is dependent on your axis Resolution (e.g. "steps per mm"). See under Resolution, here:

https://github.com/openpnp/openpnp/wiki/Machine-Axes#controller-settings

If you haven't set the true axis resolution yet (i.e. left it at the default 0.0001mm), the Backlash Calibration takes 0.01mm as the step unit, which would correspond to having the snag at every 0.36mm.

Needless to say this cyclic error is a grave problem that must be solved on the controller/driver/motor/mechanical side. Check motor currents, belt tension (not too much) etc.

_Mark

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/ecfb9908-c35a-42c5-a44c-6deba116f690n%40googlegroups.com.

[tonyl...@gmail.com]

Ahh, yes - I didn't realize the units on that axis of the graph is steps instead of millimeters.

Tony

[Michel Goto]

Hi Mark,

Thank you for your advice. 
I will try setting the axis resolution to the appropriate value next week.

[Michel Goto]

Hi  Mark, Tony

I've achieved good results today.

By correctly configuring the "steps per mm" for the X and Y axes and revisiting the offsets and calibration in the "Issues & Solutions" section, I was able to achieve results those shown in the video.
This outcome appears to be consistently reproducible, which means I might no longer need to fine-tune the positioning with tweezers.
I've also attached images of the results for "Backlash Compensation" for the X and Y axes.

I intend to continue using this state for the foreseeable future.
Thank you for all the support and assistance you've provided.

[mark maker]

Hi Michael,

I glad to hear you got good results. 

However, the Tolerance ±, you set in the Backlash Calibration is likely off by a factor of 10. I guess you wanted ±0.036mm not ±0.36mm. A tolerance of ±0.036mm is okay for 0402, but rather on the high side (there are other errors to contemplate in the "error budget"). 

Please also make sure the 0.36mm do not come from the axis resolution itself, where it would really hurt the accuracy of the machine.

So I guess you need to revisit the Backlash Calibration for best results. The Backlash Calibration might even choose similar settings in the end, there are both "conservatism" and "complacency" brought in by a high value in Tolerance ±, so they might actually cancel themselves out. But you should still redo the calibration to be sure.

Plus, once you've done this, we will be able to see how the machine really performs.

_Mark

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/e779136e-73e7-456c-80e4-1e30a0d599e1n%40googlegroups.com.