Anisotropic Scaling and Offsetting
-
I'm working through fine calibration and offsetting for my main printer using the following calibration cross:
https://www.thingiverse.com/thing:2484766/files
The spreadsheet needs a little work to work in the typical Duet setup as we (or at least I!) don't mess with steps/mm or extrusion multipliers for accuracy but with a little fiddling you can get the required scaling and offset values to use in your slicer.
I consistently find different values for X and Y axis. Is this usual, or indicative of an issue?
This is a P3Steel with GT2 belts & 16t pulleys/idlers.
X Scale 1.002289
Y Scale 1.001725X Offset 0.0235
Y Offset 0.0187Currently I'm testing a just do it approach as to try and fully understand the interplaying factors is a task not to be underestimated, but on the face of it you can't help but think they should be the same?
Not tried Z axis yet as I thing the tower part on it's own would on average be too hot. I also appreciate that a flat thing firmly mounted on the bed won't give the same results as an intricate thin wall part, but you have to start somewhere!
-
As far as I understand it belt tension shouldn't come into play.
Perhaps alignment of the hotend relative to the bed? In other words the axis of the nozzle slightly off from aligned with the bed's normal vector?
-
@DocTrucker said in Anisotropic Scaling and Offsetting:
P3Steel
That's a 0.06% difference. I would not expect a "bedslinger" to have identical scale on X and Y. Far, far too many differences in the mechanics between those two axis.
For example, if the pulley on the motor under the bed, and the idler under the front of the bed, are off even a tiny fraction in VERTICAL alignment, you'd get a difference. Or if the attach point for the belt lifts the belt out of plane. Or...
-
@Danal said in Anisotropic Scaling and Offsetting:
That's a 0.06% difference.
Or for the offset more like 0.4-0.5% of wall thickness when looking at the offset. The real easy point to invalidate the offset investigation is that the positional resolution of the motion axis is only 0.01:
((16 * 2)/(200 * 16))
So based on that alone the merit of applying a correction that is around a fifth of the minimum movement the machine can make is minimal.
@Danal said in Anisotropic Scaling and Offsetting:
For example, if the pulley on the motor under the bed, and the idler under the front of the bed, are off even a tiny fraction in VERTICAL alignment, you'd get a difference. Or if the attach point for the belt lifts the belt out of plane.
I did have that issue originally and I believe I corrected for it, although I will double check. The error would present in a non-linear manner, and so should be easy enough to spot on the plots.
I will also model the errors to see what degree of error is required to produce the results shown in order to establish what is a controllable error.
Target is +/-0.2mm accuracy across the 200mm square bed. Ideally I would be using a larger calibration part but my calipers limit me to a maximum measure of 150mm.
-
Running the offset through a simple model gives the following results.
Edit: offset being the distance in the z-axis between the ideal belt clamp location on the y-axis belt and it's actual location.
Offset Target Length Actual Length 0.0 150 150 0.2 150 150.0002 0.5 150 150.0011 1.0 150 150.0044 1.5 150 150.0100 5.0 150 150.1112 The two scaling values predict that a difference in length between two samples printed as 150mm long in x and y axis would be 0.08mm. This would need an bed belt clamp offset error of around 4.35mm. I'm confident my belt clamps aren't that bad. The actual X-Y difference I'm seeing is 0.04mm shorter in Y than X over 150mm.
To be honest I'm loosing a little faith in the method I'm using. Given the fine tolerances and marginal difference between what I can reasonably measure with my vernier I think my sample set is far too small. I will need a further think on this!
-
Yeah, hold on this for a while. I've no confidence in the method. Suggest corrections have made the tolerances worse rather than better. More samples needed, varying geometry, and a close look at the spreadsheet!
