PCB fiducials finding issue

[Josef Plasil]

Dear OpenPnP enthusiasts,
I am trying to solve an issue with finding fiducials on the PCB(s). Everything other works like a charm (fiducial finding during homing, feeders holes detection, etc.) , but when I try to find fiducials on the board, it failes = places fiducial in different place (example attached) or it didn't find fiducial in area set for finding, I don't know why. It is true, the fiducials are not mandatory, but I as I can see on the web, others are successful. I am attaching 2 pictures ... one without detection and after detection. The system detects fiducial on the place, where I would not assume any point, as there is no differences in picture. Why?
From the settings: at Z-40 I have got 0.028 units per pixel and at Z-65 (as 2nd measure) it is 0.054. I am using camera with OV9732 (1280x720), working plane is in Z-44 (see picture).
Thank you very much for your advises in advance.
Kind regards, Josef.

[Josef Plasil]

Pictures yet ... ;-)

Dne neděle 28. května 2023 v 14:44:42 UTC+2 uživatel Josef Plasil napsal:

[mark maker]

Thanks, but I need native images, as explained in the other thread:

And while you're at it, please proceed as described heren and send native image and units per pixel, and report the fiducial diameter:

https://github.com/openpnp/openpnp/wiki/FAQ#how-can-i-get-a-native-camera-image

And please provide a full log and machine.xml.

And please state the board Z.

As you have the Default Working Plane Z not equal Units per Pixel Z I guess you've calibrated these by hand? You do know Issues & Solutions can calibrate this for you, right?

https://youtu.be/md68n_J7uto

https://github.com/openpnp/openpnp/wiki/Issues-and-Solutions

_Mark

On 5/28/23 14:47, Josef Plasil wrote:

Dne neděle 28. května 2023 v 14:44:42 UTC+2 uživatel Josef Plasil napsal:

Dear OpenPnP enthusiasts,

--
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/cee37e26-7cb5-49c4-b1a7-3d5267bbb727n%40googlegroups.com.

[Josef Plasil]

Dear Mark,

thank you really so much for reply.

I repeated the primary and secondary fiducials calibration in "Issues & Solutions", but without success :-(, I had to change camera (on head) orientation after that, because it has automatically changed to approx. 100 degrees (and picture was rotated) and again, detection of fiducials is unsuccessful.

I took some video from  the process of detection (unfortunately 2 videos are too long for this message, I am sharing with you link ...) as well as 2 main pictures pictures from screenshot folder.

Perhaps you will have some idea what I am doing in wrong way.

Thank you for reply in advance.

Regards, Josef.

[mark maker]

Hey Josef

if you want help, please also follow the instructions I sent earlier.😁

1. And while you're at it, please proceed as described heren and send native images, and report the fiducial diameter:
   https://github.com/openpnp/openpnp/wiki/FAQ#how-can-i-get-a-native-camera-image

1. And please provide a full log and machine.xml.

2. And please state the board Z.

_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/c0817921-173a-4c2d-9610-f8d24965d6dbn%40googlegroups.com.

[Josef Plasil]

Hi Mark,

I am very sorry, I tried to send you the native images in my previous message ... It is very complicated to take snapshots by double clicking during the process. There is also information about the debug folder "openpnp/org.openpnp.vision.pipeline.stages.ImageWriteDebug ", but unfortunately I am not able to send pictures from this folder, as there is no such directory in my computer.

Fiducial pad diameter is 1mm, mask has diameter of 2mm.

The whole configuration is in zip file attachment.

Board Z is -44 mm.

Hopefully I provided all the needed in

 

ne 28. 5. 2023 v 17:35 odesílatel mark maker <ma...@makr.zone> napsal:

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/85465b94-12cc-0b2e-70e0-bc4f145cbc2d%40makr.zone.

[mark maker]

I don't see any such pictures in previous messages or online.

If you want to get them during the process, then go into the pipeline and enable the "deb0" stage (see you earlier screenshot).

Then restart the process.

After that, they should be in the openpnp/org.openpnp.vision.pipeline.stages.ImageWriteDebug

Also you still haven't sent the log.

https://github.com/openpnp/openpnp/wiki/FAQ#where-are-configuration-and-log-files-located

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CAFPGrag%3DLobtvoPs4XFEM-M5F-Omk4wS4GYMBh5BQr71Lhaqmg%40mail.gmail.com.

[tonyl...@gmail.com]

I think you need a "2" after openpnp so the directory is  .openpnp2/org.openpnp.vision.pipeline.stages.ImageWriteDebug

[tonyl...@gmail.com]

> I had to change camera (on head) orientation after that, because it has automatically changed to approx. 100 degrees (and picture was rotated)

All the images you posted of your board shows it at some strange angle relative to the image.  Do you really have your board fixed to your machine at that angle or is it your camera that is really rotated 100 degrees?

Tony

[Josef Plasil]

Dear all,

thank you. Of course the folder is with "2" in the end of name, it is clear. But the reason for not having folder/pictures was disabled the "deb0" stage, thx Mark for this!

Unfortunately, just pictures from beginning = homing are there. (in attachment). But it worked and it works always perfectly.

After that, when I press "Perform a fiducial check ..." the camera moves to first fiducial ... recognizes the point approx 4 mm besides and then moves to second fiducial and finishes with dialog ... :-(.

I am sending log as well ...

Again, thank you for your effort.

Regards, Josef.

Dne neděle 28. května 2023 v 18:19:08 UTC+2 uživatel tonyl...@gmail.com napsal:

[Josef Plasil]

Dear Mark,
don't you have any idea, what's going wrong in my case? Perhaps it could be some context with automatic calibration failure (with camera orientation angle)? BTW I have got board really fixed to my machine at that angle  I also tried yesterday advanced calibration for camera (top one) yesterday, again with failure :-( and the angle of camera was again modified by this process ...

Thank you very much for your help in advance.

Regards, Josef.

Dne neděle 28. května 2023 v 18:54:25 UTC+2 uživatel Josef Plasil napsal:

[mark maker]

Hi Josef,

> BTW I have got board really fixed to my machine at that angle  I also tried yesterday advanced calibration for camera (top one) yesterday, again with failure :-( and the angle of camera was again modified by this process ...

If Issues & Solutions preliminary camera calibration or Advanced Camera Calibration tell you there is a camera rotation angle, then there is very, very likely in-deed such a camera rotation angle. It is almost inconceivable how it could be wrong about that.

The rotation and flipping of the camera must be as follows.

Through your "Top" camera view you are looking down onto the XY plane, towards Z-negative. So what is X-positive must be "right" on the camera view, what is Y-positive must be "up". Your rotation C axis goes counter-clockwise.

Through your "Bottom" camera view it is like looking through your nozzle onto an imaginary mirror under your machine table, the reflection looking up at the underside of the X/Y plane, towards Z-positive, i.e. at the underside of your part held on your nozzle tip. Because you are looking at the underside, i.e. back at you, either X or Y, and rotation C would be going the wrong way, hence we need the mirror. In reality, the camera view is  mirrored/flipped either in X or Y (depends which way the bottom camera is mounted). It's a bit like a selfie-camera, it too points back at you, and it too is mirrored, otherwise you would hardly be able to align the camera.

This is all done automatically for you, if you use the Issues & Solutions steps, and do not change things afterwards.

Why is all this important?

If the camera has the wrong angle (you reported 100°, let's assume 90° for simplicity), and it sees the fiducial for example 1mm to the right, then it will move the head with the camera 1mm to the right, to better pinpoint it. But because of the camera angle, the fiducial will now move approx. 1mm towards the top of the camera view instead. The fiducial is then further away than before! It does that a few times to try and center in on it, then finally gives up, because the wrong camera rotation leads it stray:

We see something like that in your log:

2023-05-28 18:52:03.483 ReferenceFiducialLocator DEBUG: Looking for Fiducial_1mm_Mask2mm-Fiducial at (319.767988, 275.744950, -44.000000, 38.405595 mm)
2023-05-28 18:52:05.057 ReferenceFiducialLocator DEBUG: Fiducial_1mm_Mask2mm-Fiducial located at (321.978824, 274.436128, -44.000000, 38.405595 mm)
2023-05-28 18:52:05.842 ReferenceFiducialLocator DEBUG: Fiducial_1mm_Mask2mm-Fiducial located at (324.515286, 272.800100, -44.000000, 38.405595 mm)
2023-05-28 18:52:06.516 ReferenceFiducialLocator DEBUG: Fiducial_1mm_Mask2mm-Fiducial located at (326.948919, 274.126144, -44.000000, 38.405595 mm)
2023-05-28 18:52:06.772 MessageBoxes DEBUG: Error: java.lang.Exception: Fiducial Fiducial_1mm_Mask2mm-Fiducial detected too far away.

Conclusion: Let Camera Calibration adjust the camera rotation angle!

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/13e03d7c-303b-4fb2-b61d-21ca6fab8e51n%40googlegroups.com.

[mark maker]

Added that to the Wiki too, for future users.

https://github.com/openpnp/openpnp/wiki/Setup-and-Calibration_General-Camera-Setup#set-rotation-and-transforms

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/e3aab9f3-0700-91f0-8854-653c3d4946be%40makr.zone.

[Josef Plasil]

Dear Mark,
finally I completely setup newly machine, everything went (during guides) well, top camera is now in advanced mode, but the problem with fiducials on the board remained :-(. I am totally unaware of reason. The interesting thing is machine can find homing fiducial without any problems, despite it is on surface of desktop (below focus = blured), but when machine should find fiducials on the board (very sharp and clear picture), there is problem in both fiducials and they are located incorrectly, i.e. board is virtually moved to incorrect position :-(. I am sending the settings again and selected pictures (couple of them, others are similar with circles on various positions ... :-o) from debug folder. If something go to your mind, please, just let me know. I am thinking about the color of mask (black) in my case, isn't it problem for fiducials recognition? Have a nice evening. Kind regards, Josef

[mark maker]

Just to be sure:

You said earlier

> I had to change camera (on head) orientation after that, because it has automatically changed to approx. 100 degrees (and picture was rotated)

This did no longer happen, right?

You did not change the rotation back, right?

_Mark

On 5/30/23 20:58, Josef Plasil wrote:

Dear Mark,
finally I completely setup newly machine, everything went (during guides) well, top camera is now in advanced mode, but the problem with fiducials on the board remained :-(. I am totally unaware of reason. The interesting thing is machine can find homing fiducial without any problems, despite it is on surface of desktop (below focus = blured), but when machine should find fiducials on the board (very sharp and clear picture), there is problem in both fiducials and they are located incorrectly, i.e. board is virtually moved to incorrect position :-(. I am sending the settings again and selected pictures (couple of them, others are similar with circles on various positions ... :-o) from debug folder. If something go to your mind, please, just let me know. I am thinking about the color of mask (black) in my case, isn't it problem for fiducials recognition? Have a nice evening. Kind regards, Josef

--
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/9a4ee193-c31d-4a8c-a07c-390a21e64c51n%40googlegroups.com.

[Josef Plasil]

Dear Mark,

you are right ... the unexpected rotation didn't happen again. Perfect. Thank you for advise to setup everything newly. But unfortunately the issue with defect PCB fiducials detection remained.

Also (independent issue) I would like to let you know, that it would be perfect to have nozzles in safe position (safe Z level) between homing and calibrating of nozzles above bottom camera, you know, I have got the feeders on the way and in case of not having safe Z there is problem with all the homing process and nozzles can crash into feeders (see picture, please).

And I have got the suspicion the compensation of components position in axis Y by bottom camera are in opposite way, I don't know why ... But it is still unconfirmed. The bottom camera was setup automatically (newly from yesterday evening) and except exposure (it was need, because the message about differences between measurement occurred and there was impossibility to continue with job PnP process) it seems to work well ...

Unfortunately I will leave tomorrow and will be back on June 14th, therefore I will be unresponsive from tomorrow :-(.

Regards, Josef.

Dne středa 31. května 2023 v 9:24:44 UTC+2 uživatel ma...@makr.zone napsal:

[mark maker]

> Also (independent issue) I would like to let you know, that it would be perfect to have nozzles in safe position (safe Z level) between homing and calibrating of nozzles above bottom camera, you know, I have got the feeders on the way and in case of not having safe Z there is problem with all the homing process and nozzles can crash into feeders (see picture, please).

Please open a new thread for this. But I can assure you it does move at Safe Z. So something else must be acting up.

> And I have got the suspicion the compensation of components position in axis Y by bottom camera are in opposite way, I don't know why ... But it is still unconfirmed. The bottom camera was setup automatically (newly from yesterday evening) and except exposure (it was need, because the message about differences between measurement occurred and there was impossibility to continue with job PnP process) it seems to work well ...

Hmmm... perhaps you got the Y axis inverted? Camera calibration would then also flip the top camera Y, so it would actually look normal through the camera, but the axis would move the other way physically.

Please double-check the X/Y sense.

https://github.com/openpnp/openpnp/wiki/Machine-Axes

If you press the "up arrow" button on the Jog Controls, the head should move away from you. When you press the "right arrow", it should move to the right.

https://github.com/openpnp/openpnp/wiki/Setup-and-Calibration_General-Camera-Setup

_Mark

On 5/31/23 10:50, Josef Plasil wrote:

Dear Mark,

you are right ... the unexpected rotation didn't happen again. Perfect. Thank you for advise to setup everything newly. But unfortunately the issue with defect PCB fiducials detection remained.

Also (independent issue) I would like to let you know, that it would be perfect to have nozzles in safe position (safe Z level) between homing and calibrating of nozzles above bottom camera, you know, I have got the feeders on the way and in case of not having safe Z there is problem with all the homing process and nozzles can crash into feeders (see picture, please).

And I have got the suspicion the compensation of components position in axis Y by bottom camera are in opposite way, I don't know why ... But it is still unconfirmed. The bottom camera was setup automatically (newly from yesterday evening) and except exposure (it was need, because the message about differences between measurement occurred and there was impossibility to continue with job PnP process) it seems to work well ...

Unfortunately I will leave tomorrow and will be back on June 14th, therefore I will be unresponsive from tomorrow :-(.

Regards, Josef.

Dne středa 31. května 2023 v 9:24:44 UTC+2 uživatel ma...@makr.zone napsal:

Just to be sure:

You said earlier

> I had to change camera (on head) orientation after that, because it has automatically changed to approx. 100 degrees (and picture was rotated)

This did no longer happen, right?

You did not change the rotation back, right?

_Mark

On 5/30/23 20:58, Josef Plasil wrote:

Dear Mark,
finally I completely setup newly machine, everything went (during guides) well, top camera is now in advanced mode, but the problem with fiducials on the board remained :-(. I am totally unaware of reason. The interesting thing is machine can find homing fiducial without any problems, despite it is on surface of desktop (below focus = blured), but when machine should find fiducials on the board (very sharp and clear picture), there is problem in both fiducials and they are located incorrectly, i.e. board is virtually moved to incorrect position :-(. I am sending the settings again and selected pictures (couple of them, others are similar with circles on various positions ... :-o) from debug folder. If something go to your mind, please, just let me know. I am thinking about the color of mask (black) in my case, isn't it problem for fiducials recognition? Have a nice evening. Kind regards, Josef

--
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/9a4ee193-c31d-4a8c-a07c-390a21e64c51n%40googlegroups.com.

--
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/f334991d-0a66-47c0-9995-5d32744c079bn%40googlegroups.com.

[Josef Plasil]

Dear Mark,
thanks for your advises.
The movements on the bottom camera are following: when I press the UP arrow (of Y axis) on jog panel, the picture goes up, means indicating box is going down on the picture (to Y-) ..., when I press left arrow of (X axis) on jog panel, picture is going left, means indicating box is going to the right.
These actions are completely reversed on top camera.
All of these settings are result of Issues & Solutions steps, except exposition I didn't change anything
Is this right, please?

Safe Z level is really not achieved after  homing to fiducial. Immediately after homing to homing fiducial the nozzles go to the bottom camera to calibrate them, means Z is increasing from reference position (approx. -62 in my case) to -47 mm (the focus of bottom camera in my machine) together with axes X and Y to achieve target position above the camera. Could I solve this by some setting?

But I would like to solve the main problem ... and this is the incorrect recognition of fiducials on PCB. Is there any possible explanation, please?
Thank you so much for your support in advance.
Regards, Josef.

Dne středa 31. května 2023 v 13:06:00 UTC+2 uživatel ma...@makr.zone napsal:

[Josef Plasil]

Dear Mark,

finally I tested it again and from Safe Z point of view: head / nozzle goes from homing fiducial point (Z approx. -62 in my case) to the bottom camera position and Safe Z(Z -30 in my case), it means fluently X&Y&Z together = all the axes finish on the X and Y of bottom camera and Z in Safe Z, and immediately after that Z just goes to Z set for focus of bottom camera (Z-47). This is the reason why trajectory cross the feeders position. Have a nice evening. Regards, Josef

Dne středa 31. května 2023 v 21:57:02 UTC+2 uživatel Josef Plasil napsal:

[mark maker]

Dear Josef,

please read this carefully. Try to answer/check/configure exactly what I'm describing.

> The movements on the bottom camera are following: when I press the UP arrow (of Y axis) on jog panel, the picture goes up, means indicating box is going down on the picture (to Y-) ..., when I press left arrow of (X axis) on jog panel, picture is going left, means indicating box is going to the right.
> These actions are completely reversed on top camera.
> All of these settings are result of Issues & Solutions steps, except exposition I didn't change anything
Is this right, please?

I don't understand what exactly you mean by "picture" and "indicating box". Maybe this is a translation engine artifact?

In the bottom camera view the visible nozzle must go up, when you press the "up arrow". It most go right when you press the "right arrow". It must rotate as the rotate arrow indicates.

https://github.com/openpnp/openpnp/wiki/Setup-and-Calibration_General-Camera-Setup#bottom-camera

In the top camera, it is not the subject (the nozzle) that is moved by the head, but the camera is moved by the head. So yes, it is inverted so to speak.

If you look at a fiducial, it must go down when you press the "up arrow", it must go left when you press the "right arrow".

But if this is appearing correctly in the camera view, it is not surprising, because as I said, the calibration will make it so, even if you have the physical axis reversed.

Therefore I asked about how the head goes, physically.

When you press the "up arrow" the head must physically (not in the camera) move away from you (Y axis positive).

When you press the "right arrow" the head must physically (not in the camera) move to the right (X axis positive).

> Safe Z level is really not achieved after  homing to fiducial. Immediately after homing to homing fiducial the nozzles go to the bottom camera to calibrate them, means Z is increasing from reference position (approx. -62 in my case) to -47 mm (the focus of bottom camera in my machine) together with axes X and Y to achieve target position above the camera. Could I solve this by some setting?

According to your log, that is not so, the move to safe Z is done:

2023-05-30 20:19:22.331 ReferenceFiducialLocator DEBUG: FIDUCIAL-HOME located at (3.014084, 2.185289, -45.200000, 0.000000 mm)
2023-05-30 20:19:22.332 AbstractHeadMountable DEBUG: Top.moveTo((3.014084, 2.185289, -45.200000, 0.000000 mm), 1.0)
2023-05-30 20:19:22.351 GcodeDriver DEBUG: [serial://COM9] >> M204 S54 G1 X5.4908 Y2.132    F139 ; move to target, 20000
2023-05-30 20:19:22.353 GcodeDriver DEBUG: [serial://COM9] >> M400 ; Wait for moves to complete before returning, 20000
2023-05-30 20:19:22.447 GcodeDriver DEBUG: [serial://COM9] >> G92 X4.2606 Y2.0000    ; reset coordinates, -1
2023-05-30 20:19:22.449 ReferenceNozzle DEBUG: N1.home()
2023-05-30 20:19:22.450 ReferenceNozzle DEBUG: N1.home() nozzle tip JUKI503 calibration neeeded
2023-05-30 20:19:22.452 ReferenceHead DEBUG: H1.moveToSafeZ(1.0)
2023-05-30 20:19:22.453 AbstractHeadMountable DEBUG: N1.moveToSafeZ(1.0)
2023-05-30 20:19:22.454 AbstractHeadMountable DEBUG: N2.moveToSafeZ(1.0)
2023-05-30 20:19:22.455 AbstractHeadMountable DEBUG: Top.moveToSafeZ(1.0)
2023-05-30 20:19:22.455 AbstractHeadMountable DEBUG: Top.moveTo((1.783863, 2.053480, 0.000000, 0.000000 mm), 1.0)
2023-05-30 20:19:22.458 AbstractHeadMountable DEBUG: N1.moveTo((323.554000, 50.667631, -30.000000, -180.000000 mm), 1.0)
2023-05-30 20:19:22.463 GcodeDriver DEBUG: [serial://COM9] >> M204 S511 G1 X302.2804 Y132.767 Z-30.000 A-180.00  F10016 ; move to target, 20000
2023-05-30 20:19:22.468 GcodeDriver DEBUG: [serial://COM9] >> M400 ; Wait for moves to complete before returning, 20000
2023-05-30 20:19:25.041 AbstractHeadMountable DEBUG: N1.moveTo((323.554000, 50.667631, -47.000000, -180.000000 mm), 1.0)
2023-05-30 20:19:25.052 GcodeDriver DEBUG: [serial://COM9] >> M204 S298 G1   Z-47.000   F4273 ; move to target, 20000
2023-05-30 20:19:25.056 GcodeDriver DEBUG: [serial://COM9] >> M400 ; Wait for moves to complete before returning, 20000
2023-05-30 20:19:25.536 ReferenceNozzleTipCalibration DEBUG: [nozzleTipCalibration]starting measurement; angleStart: -180.0, angleStop: 180.0, angleIncrement: 60.0, angleSubdivisions: 5

However, it seems your physical controller Z homing seems to leave the nozzle lowered. That is unexpected, normally it be left balanced on a dual head machine. So OpenPnP thinks it is already at Safe Z and does not move it.

This means the Home Coordinate is not configured right in OpenPnP on the Z1 axis. Enter the raw controller coordinate the Z axis has after homing:

If you don't know it, enter

G28

M114

in the Console of the driver, and read the reported Z, and enter it in the Home Coordinate of the axis:

Note if you continue following the Issues & Solutions suggestions into the Advanced Milestone, you will be guided to enable advanced motion control and Location Confirmation, which will (as a side effect) automatically get the reported location after homing, i.e. OpenPnP will always be in sync with the controller and this will not happen, even if the Home Coordinate is not correct.

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/e9fd2ddc-6192-4c5c-aef3-90f807b4d07fn%40googlegroups.com.

[mark maker]

See the other answer about this.

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/129ad374-60c3-4504-afd6-f9cb428a34ban%40googlegroups.com.

[Josef Plasil]

Dear Mark,
again, thank you. I can confirm all movements are rightly set and are the same as you declared in your response.
From Safe Z I have to check it, but as I mentioned previously, I will return on June 13th, therefore I will test it after that.
Just one main problem remains = fiducials detection on PCB. Don't you have any idea, what I should change to have machine working properly, please?
Thanks so much.
Regards, Josef.

čt 1. 6. 2023 v 10:03 odesílatel mark maker <ma...@makr.zone> napsal:

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/700ba212-f9c8-595a-e359-10de61c95466%40makr.zone.

[mark maker]

Well, I'm out of ideas, sorry.

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CAFPGragR2nVjeG9VNhW0rfPFx2rMW97%2BqWG-d-YhLm%2BhJ_pWFA%40mail.gmail.com.

[tonyl...@gmail.com]

Josef,

 I have one more idea you can try: Edit your Fiducial Vision Pipeline, select the DetectCircularSymmetry stage and set maxTargetCount to 2 as shown below. I've also had occasions where the DetectCircularSymmetry stage would rank some random non-descript point as having a better score than the obvious good circle that is near the very center of the image. By setting maxTargetCount to 2, the pipeline outputs both points and the fiducial locator software then decides which is the best one (the one closest to the image center). I never had a chance to investigate why DetectCircularSymmetry does that, but this was a successful work-around for me.

Tony

[mark maker]

Yes, good idea, Tony.

Using Josef's latest machine.xml and source image, I was able to reproduce it:

Turns out, there a barely visible shadow artifact in the background that the plain CircularSymmetry stage finds more attractive:

I think this happens due to an unlucky combination of effects:

As you see in the original and contrast enhanced pics above, the inner darker fiducial reflection is uneven. Could be the HASL surface that is uneven, or the hole in your diffuser is too large or asymmetric, or your diffuser is not diffusing enough (HASL is generally difficult).

You can reduce the relevance of the inner reflection, by reducing the innerMargin property of the pipeline from 0.4 to 0.2:

This did nail the Fiducial, but as you see in the heat map, there is still quite some "attractiveness" (yellow) in the background shadow:

There is also another problem. It seem the Units per Pixel at given Z are not right. At your stated PCB height (-44mm) the maxDiameter is 14 px. But the actual fiducial diameter in the image (once detected) is exactly 14 px, so there is no space left (outerMargin) to actually detect the edge contrast of the fiducial.

If I set the Working Plane (acts as PCB Z) to -38mm then the minDiameter and maxDiameter appear about correct and the match becomes more robust (four times better score than the shadow).

Finally, there are new symmetryScore methods, that I added lately, albeit for feeder vision:

Turns out these help here too, if you set it to RingMedian it will better reject the shadows:

See also:

https://github.com/openpnp/openpnp/wiki/DetectCircularSymmetry#general-properties

So this should be the setting for HASL:

I will think about making this the new stock default, or a separate HASL setting. The smaller innerMargin needs more accurate Units per Pixel at PCB Z, but I believe we can now assume this a given. Hmmm...

@Josef, once you come back, adjust the pipeline as shown above and double-check your PCB Z. Tony's advice of setting the maxTargetCount to 2 or 3 is another valid method that you can combine with the above.

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/d7f8c859-448c-4074-9d7a-7bf9ecfd880cn%40googlegroups.com.

[Josef Plasil]

Dear Mark & Tony,
really appreciate your advices, help and support. Thx for this!
Once I come back, I will apply all your settings. Also I can try to improve my light diffuser, as I can really recognize the pattern of LED diodes reflection on pictures, which BTW I didn't notice before at all :-o.
I will let you know the result.
Greetings! Regards, Josef.

Dne sobota 3. června 2023 v 14:40:13 UTC+2 uživatel ma...@makr.zone napsal:

[Josef Plasil]

Dear Mark,

just to be sure from the "Z safe" point of view ... is the "Z safe" level the number on the Z axis, when the nozzle(s) is / are touching the PCB, please?

I've got both of the nozzles set really symmetrically, therefore both of them touch PCB at -44mm. Means when the both nozzles are in the same level, the Z is 0 on my machine. But camera itself is little bit higher and I focused it to PCB (level) before everything started ...

Thanks for clarification in advance.

Regards, Josef.

so 3. 6. 2023 v 7:40 odesílatel mark maker <ma...@makr.zone> napsal:

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/f731f78d-9bd9-1796-131b-eaf0d8d6569a%40makr.zone.

[Mark]

Your sentence is a bit confusing 😉

It has nothing to do with "Z safe" or "Safe Z" as we usually call it. Safe Z is when the nozzle is retracted to a safe height, so when you move around in X/Y it will not collide with anything on the machine table, including with parts already placed.

But yes, what I was talking about, is the Z coordinate indicated in the DRO when your nozzle tip barely touches the PCB.

But of course this will only work correctly, if you did not in any way change the camera mounting, the camera video mode, the PCB mount, the way the Z axis is homed, etc. since you last calibrated the camera.

Also, please ask Tony about your advanced camera calibration (the machine.xml is in this thread somewhere). I see errors in Y that seem large to me. So maybe the scaling of the fiducial is wrong because of that.

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CAFPGrag2FAgEG535teWhKcPfsqboYv2iiyb3WnbHBG-rfvkACA%40mail.gmail.com.

[Josef Plasil]

Dear Mark,

you're right, it was confusing from my side, but you simply replied exactly, what I meant ;-). I wanted just to say, that every coordinates are derived from DRO.

Could be errors (backlash compensation) in Y axis caused by rubber belts? I can try steel core belts, as they should be more rigid. I'll let you know, once I'll be back (after 13th). Thank you. Regards, Josef.

út 6. 6. 2023 v 12:51 odesílatel Mark <ma...@makr.zone> napsal:

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/d866c99e-96ef-e09e-5378-6e53ef4d6958%40makr.zone.

[Josef Plasil]

Dear Mark,
I can confirm you I was able be successful with fiducials detection with their vision settings as per attached picture.
Thanks for the guidance.

Just I realized the bottom camera light switching on after the whole process is finished. Could you advise me, where I can turn off this feature, please?

BTW, it is consistent with log (I am sending last 5 lines):

2023-06-17 21:54:35.862 ReferencePnpJobProcessor$Finish INFO: Job finished 5 parts in 58.4 sec. This is 308.4 CPH
2023-06-17 21:54:35.909 GcodeDriver DEBUG: [serial://COM9] >> M204 S313 G1   Z23.900   F4714 ; move to target, 20000
2023-06-17 21:54:35.913 GcodeDriver DEBUG: [serial://COM9] >> M204 S578 G1 X7.477 Y4.946    F12663 ; move to target, 20000
2023-06-17 21:54:35.923 ReferenceActuator DEBUG: LIGHT_BOTTOM.actuate(true)
2023-06-17 21:54:35.923 GcodeDriver DEBUG: [serial://COM9] >> M803 ; actuate camera light ON, 20000

Thank you so much in advance.
Regards, Josef.

Dne úterý 6. června 2023 v 21:59:07 UTC+2 uživatel Josef Plasil napsal: