Is it possible to use the pnp head down vision to improve placement accuracy (especially for 0201)

65 views
Skip to first unread message

Visti Andresen

unread,
Oct 26, 2025, 3:56:04 PM (10 days ago) Oct 26
to OpenPnP
Hi is it possible to use the down vision system to update / improve the position of a footprint using the pads on the pcb?

I'm asking as for the most parts the camera is placed quite accurate over a 0201 footprint, but sometimes the position seem to be off by maybe 0.1mm which is a bit much for such small parts.

The PnP used is a Pandaplacer A1 (a belt driven machine)

It might be worth noting that after locating the (three) fiducials both manually and automatic (it seems to be doing that quite well), when positioning the camera over them again, the crosshair doesn't hit dead center on the fiducials.

Toby Dickenson

unread,
Oct 26, 2025, 4:35:51 PM (10 days ago) Oct 26
to ope...@googlegroups.com
It might be worth noting that after locating the (three) fiducials both manually and automatic (it seems to be doing that quite well), when positioning the camera over them again, the crosshair doesn't hit dead center on the fiducials.

This is a key problem.

How accurately can your fiducial vision pipeline locate your fiducials? Try pressing the "Test Fiducial Locator" button on the Fiducial Vision Settings page. Are the crosshairs lined up on the fiducial? After the first attempt click on the status bar position indicator, and it will switch to showing a position relative to the first find. So now you can press "Test Fiducial Locator" many times again and observe the variation. And do the same from offset starting positions. If your board Z position is off then fiducial position is vulnerable to parallax errors, and you might benefit from reducing the "max linear offset" - this will do a second pass from exactly above the detected location to eliminate parallax.

If your fiducial locator is not behaving right then please share a screenshot showing your fiducials.

I find that the fiducial locator is pretty good but not perfect, and I therefore benefit from having a larger number of fiducials. More measurements cause errors to get reduced by averaging. Do you have any other board features that could be recognised by the fiducial locator? I often end up using the circular edge of a copper pour surrounding a via.

 

--
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 visit https://groups.google.com/d/msgid/openpnp/4c65a3ce-a6b1-4f12-a96b-db17e1b5382an%40googlegroups.com.

Visti Andresen

unread,
Oct 27, 2025, 4:05:03 PM (9 days ago) Oct 27
to OpenPnP
I think the fiducial locator works quite well

20251027_20h42m49s_grim.png

Deltas
1) 0.006, 0.004
2) 0.003, 0.000
3) -0.001, 0.000
4) 0.000, 0.000
5) -0.005, -0.003

It may actually be caused by backlash in the PnP

An excellent C25 (0201) after moving to it from a close by C24 (0402)

20251027_20h51m07s_grim.png

A problematic C31 after stepping though the other capacitors

20251027_20h53m52s_grim.png

A quite useful C31 after a direct move from C36
20251027_20h54m27s_grim.png

An overview of the 0201 locations
20251027_21h02m22s_grim.png

Mike Menci

unread,
Oct 27, 2025, 4:56:31 PM (9 days ago) Oct 27
to OpenPnP
There ia a way - but you will need to manual input all the respective new data in the "board (PCB) " file  for X and Y. But this is not recommended !

Mike Menci

unread,
Oct 27, 2025, 5:00:12 PM (9 days ago) Oct 27
to OpenPnP
I suspect your travel is not accurate -  Steps/rotation !?? - If you Jog 100 mm (200mm)  on a ruler  (and reverse) do you get cross hair exactly on the spot on ruler in Y and Y directions ?

Mike Menci

unread,
Oct 27, 2025, 5:06:37 PM (9 days ago) Oct 27
to OpenPnP
Watch how finally smt parts align https://www.youtube.com/shorts/cXoA-KmBvpI

tonyl...@gmail.com

unread,
Oct 27, 2025, 5:38:34 PM (9 days ago) Oct 27
to OpenPnP
If the alignment of the crosshairs with the pads of a particular placement is changing depending on where you start the movement, that would indicate some kind of mechanical issue with your machine. Have you run the backlash calibrations? How do the results look? Could you post your machine.xml file so we can look at it?

Visti Andresen

unread,
Oct 27, 2025, 5:43:05 PM (9 days ago) Oct 27
to OpenPnP
160 steps per mm 
on my caliper 100mm is maybe 0.2mm off in the x direction over the 100mm (the calipers are rotated 0.15mm over the 100mm length)
the return to the start sometimes is off by maybe 0.05mm

Mike Menci

unread,
Oct 27, 2025, 5:51:07 PM (9 days ago) Oct 27
to OpenPnP
Calculation = 
  • Steps per revolution: 400 (for a 0.9° stepper motor).
  • Microsteps: 16 (often the default on Smoothieboards).
  • Belt pitch: 2 mm (standard for GT2 belts).
  • Pulley teeth: 20 (standard for many setups). 
Your Belt tension ? - Than  run the backlash calibration... Check all pulleys are tight to shaft properly..... ! 
Ste_mm.png
Reply all
Reply to author
Forward
0 new messages