Hi,
I did investigate a bit about placement errors.
I changed all the parts of an unused pcb to 0402 resistors,
sprayed some repositionable adhesive on the pcb to make it sticky
and then ran the placement job. The pcb has fiducials at the 4
corners.
Next I moved the camera over some resistors , ideally the
resistors should be centered to the camera view. This is not
always the case!
I selected resistors to form rows and columns, 5 rows for the X
axis and 3 columns for Y
The first plot shows the Y axis errors measured with the camera
and the DRO in relative mode:

Same for X axis:

Accuracy is better at the center of the pcb, Y errors increase
with the same sign. X errors are smaller and the sign change.
It should be noted that errors are more important close to the
fiducials.
More data would be better but this is not a task for a human, the
machine would be much faster and accurate.
That explains the placement errors I noticed with real job but I
have no idea where they come from and how to reduced them. Camera
are both calibrated with the new advanced calibration, X and Y
axis uses linear encoder.
Attached is the measurement data. If someone comes with an
idea...!
joël
--
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/eeac411c-6083-3e36-12d2-02b13b65ae6f%40gmail.com.

The AffineTransformation obtained from the fiducials (inside the
Job) is not applied when you manually position over the part. The
DRO never shows compensated coordinates. So I wonder what you
measured.
The only thing that you should judge is how well
centered the parts are over the actual PCB footprints. I guess
that's hard to measure, though.
_Mark
--
Don't understand what you mean.
--
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/43548c6e-86ae-4ce9-97b5-f1d968d39900n%40googlegroups.com.
with my poor english easier to explain with pictures:
when the placement job is over I visually check the parts
placement with "position the camera at placement location" from
the Job pane.
The camera crosshair is at PCB footprint location where should the
placed part center be too. I measured the offset

To measure I reset the DRO to make relative measurement and jog the camera to the actual part center :

then the DRO shows the offset:

joël
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/CABUTZN_BCbfANoKme3Gmjw95iAghsJMogu3dxsWkjmPESQOfRw%40mail.gmail.com.
Thank's, nice plot!
I rejected the too much misplaced parts from measurements, too
much angle included.
X and Y axis uses linear encoders, a lot of mechanical errors
should be cancelled out.
Maybe I'm wrong, I think mechanical errors should be the same for
the camera and the nozzle or that would mean their offset change
with X Y positions.
Except for some added "noise" but plots doesn't only noise. Some
trends are visible.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/0e88ff08-cfee-43f5-a1ae-3017dce70a9dn%40googlegroups.com.
Is it the same when using "position the camera at placements
location" from the job pane?
joël
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/4f6d903e-f147-a6db-7c03-335e7485afdf%40makr.zone.
> I believe the Affine transform is applied when you use the "position the camera at placements location" from the placements pane. If you then switch the DRO to relative mode and jog over the part, I think you should be able to measure the relative placement error.
Well, I guess you would measure the "relative part from where-fiducial-guided-AffineTransform-and-machine-positioning-thought-the-placement-should-be error". ;-)
The real benchmark would be to first jog to the part center, then
click on the DRO to make it relative, then jog to the copper
footprint center behind the part.
_Mark
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/c6e00cfd-7b95-4669-a36b-5c0fa25f49a0n%40googlegroups.com.
Perfect!
--
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/0050456d-5558-453a-a3b1-8edbcc4c55a8n%40googlegroups.com.
After the Advanced Camera Calibration, that effect should be completely gone. Do you have it enabled?
_Mark
The important thing is to center the chip in the cross-hairs. When the Advanced Camera Calibration is enabled, the tilt elimination should make sure that there is no parallax.
Note, Advanced Camera Calibration has a slider to eliminate all invalid pixels, which crops the image to be symmetric around the cross-hairs. I recommend sliding it all the way, even if you lose part of the view. Otherwise you might be tempted to fit the chip in the view area, i.e. not truly centered, i.e. you will have a parallax.
_Mark
> I believe the Affine transform is applied when you use the "position the camera at placements location" from the placements pane. If you then switch the DRO to relative mode and jog over the part, I think you should be able to measure the relative placement error.
Well, I guess you would measure the "relative part from where-fiducial-guided-AffineTransform-and-machine-positioning-thought-the-placement-should-be error". ;-)
The real benchmark would be to first jog to the part center, then click on the DRO to make it relative, then jog to the copper footprint center behind the part.
_Mark
Yes you are right. I made the assumption of a smaller footprint /
camera error than placed part / camera. I didn't make measurements
but I noticed it every times a placement is off, the camera
crosshair is at the footprint center within a few tens of µm.
Close to perfect for my needs. Next picture shows a placement
close to a fiducial. The camera crosshair is exactly at the
footprint center but the part is not.

Aligning part center and camera crosshair :

And measuring placement offsets :

Fiducials and Affine Transform performs very well, errors comes
with placement and errors are not only random noise. Measurements
shows trends.
It is not very common and maybe it has no connection, the top
camera position offset is not 0:

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/12cd2b0c-fd47-434e-35a2-95dd58695fdf%40makr.zone.
> I didn't make measurements but I noticed it every times a placement is off, the camera crosshair is at the footprint center within a few tens of µm. Close to perfect for my needs. Next picture shows a placement close to a fiducial. The camera crosshair is exactly at the footprint center but the part is not.
OK, that makes it a lot clearer.
The things that come to mind to improve placement accuracy (you might have already reported some of the things below, can't keep track of every user 😅):
_Mark
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/54e2b96e-d885-1098-bb38-0a0ee5b6ff98%40gmail.com.
Forget point 8, that was an editing mishap.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/345808e6-5c00-02f0-d1d9-0aba8833426a%40makr.zone.