Hi David,
Do not unduly increase the Pick Tolerance, i.e. reduce it. There seems to be a different problem.
In order to diagnose from afar, please send log at TRACE level and machine.xml
https://github.com/openpnp/openpnp/wiki/FAQ#where-are-configuration-and-log-files-located
Also send some native camera images of the alignment:
https://github.com/openpnp/openpnp/wiki/FAQ#how-can-i-get-a-native-camera-image
_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/060a3965-33ea-4840-89fc-f52baf395ebcn%40googlegroups.com.
Hi David
as Jan already pointed out do not use the vision offsets, they are only used in very exotic cases.
From your log it becomes evident, that the threshold
(between dark background and bright pins) is not right:
2023-10-21 14:35:54.201 MaskHsv TRACE:
Fraction actually masked = 0.990929459600556
This means that almost the whole image is being masked, i.e. it does not actually recognize just the pins.
First make sure to use the newest test version of OpenPnP. It has the best support for bottom vision.
Then make sure you have the new stock settings and pipelines for bottom vision, as described here:
https://github.com/openpnp/openpnp/wiki/Bottom-Vision
If you come from an older machine configuration or OpenPnP version, chances are you don't have them yet.
Then you should be able to adjust the threshold, as shown in this animation:
_Mark
> The placement accuracy has improved although it is still
about 1/2 a pin width of to the right (so approx 0.15mm)
May points to either
_Mark
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/eb82d15b-edeb-4995-b08c-d41f5dbd1cb2n%40googlegroups.com.
Tony is right and the Strip Feeder should also continuously recalibrate itself using sprocket hole vision. Even if not, bottom vision should compensate for any pick inaccuracy.
> Interestingly the Issues and Solutions does not help with this part of the calibration.
However it is not so easy. Unless very carefully and strictly made safe, it would very likely create more problems in "real life" than it would solve. Especially when used from an easy to use UI such as I&S, where people don't actually need to know what they are doing. 😇
Changing the machine scale like that, essentially renders all coordinates already captured in OpenPnP invalid. So it would only be allowable at the very beginning of a machine build, where you have no stored coordinates yet (or just a few that are known and could be adjusted by OpenPnP).
But the experience doing support here on the list tells me that
such a strict limitation to the very beginning is utterly
unrealistic/useless. People do shortcuts in I&S, reuse old or
bad machine.xml from other people,
where I&S is already done or dismissed, only realize they need
to improve the accuracy when the machine is already finished and
all feeders already setup, etc. Or they have a bad machine crash,
or a modification of the machine, and need to re-calibrate.
So to be useful, this scaling and squareness calibration would need to be available at all times. Therefore, if we ever wanted to introduce it, we would need to have functionality that is able to enumerate all stored coordinates, and apply a before-and-after-transformation. Until then we better leave it as it is.
This is acceptable, because all the most critical "absolute
scale" areas in OpenPnP have their own local Affine
Transformation calibration or similar: the PCB/Panel with
fiducials, BlindsFeeder array with fiducials, Strip Feeders with
sprocket holes vision, PushPullFeeder with sprocket holes vision,
nozzle tip changer template image vision, etc. pp. This "local
area" calibration also covers any errors when things are
mountable/pluggable on the machine, including tinkering. The
overall absolute machine scaling and squaring therefore becomes
mostly irrelevant, i.e., any absolute scaling or squaring error is
already present when you capture a coordinate so it does
simply not matter, when you later position back to it.
PnP is much more lenient that way than other NC applications.
Yeah nah. I'm using strip feeders so the more accurate the scaling the better the picking will be.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/8b3f4c4a-6005-4437-8da7-ea1e4a5dc3a6n%40googlegroups.com.