Feedback on calibration and resulting dimensional accuracy

As promised, I did some more measurements of flex, backlash and belt tension. Unless otherwise noted, I used the maslow.yaml with my manual frame measurement and z of the router was so that the distance between the top of the sled’s honeycomb structure and the bottom of the plastic around the linear guide bushings was 43mm.

All numbers are differences between moving left and right or up and down, respectively.

Jogging in x:

  • 0.4mm x-deflection, vertical post, upper end
  • 0.02mm x-deflection, vertical post, lower end
  • 0.4mm x-deflection, vertical guide rod, upper end
  • 0.35mm x-deflection, vertical guide rod, just above the linear bushing
  • 0.3-0.35mm x-deflection, linear guide bushing, upper end
  • 0.15mm z-deflection, linear guide bushing, upper end
  • 0.15mm x-deflection, linear guide bushing, lower end
  • 0.35mm x-deflection, on the sticker of the router (drawn on the opposite side on the image below, because of the BR motor in front of the router’s sticker)
  • 0.5-0.7mm x-deflection, PCB cover on top of the router
  • backlash (sled, not router bit): 1.25mm

Jogging in y:

Belt tension:

Belts were extended approximately 2300mm. I put a straight edge parallel to the belt. Then I pulled the belt horizontally with 2 Newton force, in the middle of the belt. This yielded the following horizontal displacements of the belt:

  • 26mm (one of the “pulling” belts after jogging in x-direction)
  • 42mm (one of the “trailing” belts after jogging in x-direction)

Same setup but using the maslow.yaml with the frame dimensions based on the automatic calibration:

  • 10mm (one of the “pulling” belts after jogging in x-direction)
  • 13mm (one of the “trailing” belts after jogging in x-direction)

If you want to try measuring your anchor points manually, note that the tlX, tlY, trX, trY, etc. numbers are not the distances between your anchor points, so you can’t just take your tape measure and put the readings in the maslow.yaml file. Instead do this:

I have this all in onshape Onshape

thanks Hendrik,
for the heads up, and @bar, here is some feedback as well.

I was just setting up my Maslow 4.0 for the first time, and after calibration I got a fitness of 0.66. Then I measured my anchors with a tapemeasure and doublechecked it in CAD, and the largest deviation was the TR (ofcourse) with 13mm.
first cut a square of 1000x1000mm with the calibration values, on the right side of the bed:
the sides were not equal, 3mm off, top and bottom were not parallel, 3mm off, and the diagonals were not equal, 7mm off.
I then changed the maslow.yaml to my measured values, The result was not better.

then I cut a 900x900 square in the middle of the bed, with the old calibration values. same story.
then put my measured values in the yaml, cut the square, and hooray, it was accurate.

I had been cutting the squares by lowering the z by hand and jogging around the bed, clockwise.
As a last test I cut the 900x900 square by jogging counterclockwise. Now the sides of the square were not equal, 4mm off.

And another observation: I calibrated a few times, frame stiffnes reported different numbers, and on different diagonals as worst: 0.04 and 3.5 on the first calibration and 2.4 and 0.1 on the last calibration.

1 Like

Are you using the arm stacking order I recommended or the one in the M4 assembly guide? (see here for example X and Y off by 1% - #94 by md8n)

And if you’re using the one I recommended have you made the specific changes to the maslow.yaml file for the Maslow_**Z values?

I did not change the order of the arms yet, I will try that.

The key issue is that the highest and lowest arms in the stack should be separated by a long side, and not a short side.

I would add the thought (not tested) that you want opposite arms adjacent to
each other so that as they pull against each other, the router does not tilt.

David Lang

Tested, can confirm that this is real for verticalish frames

Can we please stay with horizontal setups in this thread, in order to avoid confusion?

I have more results that might be of interest. First let me describe two changes I’ve done (yes, simultaneously, bad for systematic analysis, I know …):

PTFE pads on sled

To reduce friction I put some self-adhesive PTFE on the bottom of the sled. We sell this stuff, so it was a no-brainer. I’m not sure how much impact it has, but it surely doesn’t hurt.

Belt tension

Earlier I mentioned that belt tension with my manual frame measurement was noticeably lower than with the automatic calibration, and I wrote:

If Maslow thinks the frame is bigger than it actually is, it releases more belt than necessary. So I scaled the anchor coordinates in my maslow.yaml by a factor of 0.999, but keeping the geometry. This yielded nice, tight belts. I know that my frame measurement isn’t off by that much, but maybe the belts’ pitch is incorrect.

Test results

Using this setup I made a test cut in particle board. One big square, five small squares and five circles. Everything is a bit too big, but there is no overall scaling error. For details check this image:

Pretty close!

2 Likes

That is impressive results! Very encouraging.

1 Like