Statistically-based Bed Compensation and Calibration



  • Hello,

    Had an interesting experience with the Smart Effector the other day. I ran bed calibration and got something like this:

    (Shamelessly stolen from TLeTorneau in another thread)
    [https://forum.duet3d.com/assets/uploads/files/1542436419111-bed_level_001_mesh_11172018.png

    What's interesting is that this shows some real mechanical drift of the sensor over time. I had the same problem on the Smart Effector. I doubt that this is an artifact of the print bed, especially since when I changed directions, the lines appeared in the other direction.

    I think an ideal bed probe system would be to implement the following:

    • Randomly determine the probe point order.
      • This would help eliminate sensor drift from the final results.
      • Alternatively, you could probe in reverse directions (although randomly would be better)
    • Perform 3+ measurements at the same point (at random points in time)
      • This gives you mean standard error reduction
    • Implement a localized flatness smoothing
      • Sharp bumps are more likely to be errors than real changes.
      • False sharp bumps can produce problems / deformations in the final print

    Bottom line is that with the statistical methods mentioned above, I think it's possible to achieve a 3X decrease in calibration sensor error. For mechanically based sensors, this could be a real boon and help achieve the accuracy needed for printing.



  • Have you checked out the A and S parameters of set probe type?

    https://duet3d.dozuki.com/Wiki/Gcode#Section_M558_Set_Z_probe_type

    On the upside for the current method testing in a specific order does give you a fighting chance of driving out sources of error. Randomising the order could potentially hide these patterns giving the false impression of it just being a noisy measurement?


 

Looks like your connection to Duet3D was lost, please wait while we try to reconnect.