I had my Maslow for a while now, it is great but I get a consistent -5% error in the Y axis (and a smaller +1% error in the X axis)
I was sure it is a matter of calibration but seem that either manual or auto calibration gives the same error, the only solution seems to be X/Y scale
I did not think much about it, but when browsing the forums I found other posts of other users having the same error, seems to be that a lot of users get the same deviation.
Bigger cal-grids reduce the error in both axis somewhat linearly, but still leave around 1% err
Same as my case, Manual calibration does not help as well
I have a guess, maybe , horizontal mode has some bug in it, maybe some gravity compensation that should not be there that make this returning arrow. Maslow is symmetrical so I do not see any other reason for X and Y to have consistent opposite error signs
Can it be? Would love to hear other ideas, I think calibration issues as these are very important as they can make Maslow unfit to projects until solved
Thanks,
Eldar
EDIT:
I copy pasted this post as a GitHub issue, maybe the clanker will come up with something
P.S
I’m glad we have a forum and not a discord so returning issues like these can be tracked instead of buried under 1000s of other posts
One thing that has been on my todo list to investigate is the ratio of the belt teeth to length. Our precision comes from counting belt teeth, but the tooth spacing on the belts could be off by 1% and I wouldn’t be shocked.
I did do some testing and experimentation on this a while back, but because it is such an important topic for precision I think it is very worth revisiting. It’s just a very involved topic because each change requires re-running calibration and then re-cutting a test pattern and we are only now getting to the point where the whole system is stable enough to do that.
Edit: Now that I’m thinking back I remember that Anna actually had some time to test this and didn’t find any improvement there. It could still be an area for research, but we did explore it and didn’t find any obvious issues
I had my Maslow for a while now, it is great but I get a consistent -5% error in the Y axis (and a smaller +1% error in the X axis)
please check to see if the error is consistant/linear or not. We had one person
do some extensive testing and found that almost all the error showed up within a
few inches of the edges
Bigger cal-grids reduce the error in both axis somewhat linearly, but still leave around 1% err
This is to be expected, the bigger grid results in more accurate anchor
locations
Same as my case, Manual calibration does not help as well
what do the manual calibration anchor locations look like compared to the
automatic calibration?
can you try grabbing the belt lengths from near the end of the automatic
calibration and put them into this calculator that uses a different algorithm to
calclulate the anchor locations and see how the results differ? (both how they
differ in the anchor locations, and how the resulting anchor locations result in
different x/y error)
The error is completly *constant*, a square will become a perfect rectangle for example, I tested it in different locations and the error is consistent across the cutting area, center and edges. Seems to be the same case in the 4 topics I refered too.
The error is also the same after manual calibration
I realy think this error is not fault of the calibration proccess itself, but a result of some sort of differense between horizontal and vertical mode, so only some users get it, but can be something else.
The error is huge so I don’t think that a 1% belt error is it
The bot came up with a solution regarding the z of the anchors, can it be a valid cause?
The bot came up with a solution regarding the z of the anchors, can it be a
valid cause?
If the Z offsets for the anchors are wrong for your frame, I believe that can
cause grief (that is what I was suggesting when I talked about playing with the
wasteboard/workpiece values, they get added to the z offset to determine the Z
coordinate of the anchors)
I’m also struggling with the constant tuning of the axis scale. I’ve already tuned it so that on each axis there is an error of max up to 1mm. I don’t have much time to try all the options for comparing the locations of the calculated or found anchors. I settled for the last anchor finding in v1.15 and fine-tuned the scales. I need to cut projects!
it’s tedious…
if a person has SW problems, then HW problems start to get involved and nothing will be cut exactly
Could you tell me more about that?
Still weird that users constantly have the same +x -y error though
How one should tune the achor heights?
I have my Maslow the floor, the anchors are at hight 0 (the belt end tough the floor, so maybe the belt z center at the achor is 5mmm) and above it have 18mm spoilboard and above it the workpiece, ususally another 18mm
BTW, is there a way to ask maslowbot questions about the codebase? I could learn how it works better so I could try and help diagnose the issue and maybe future ones
I have my Maslow the floor, the anchors are at hight 0 (the belt end tough the
floor, so maybe the belt z center at the achor is 5mmm) and above it have 18mm
spoilboard and above it the workpiece, ususally another 18mm
The default Z offsets assumed that the anchors were at the same height as the
spoilboard. The old way was to change the height of each of the anchors, but a
few versions back we added the ability to specify a wasteboard and workpiece
thickness.
Assuming your floor is flat, in your case you would want to set your wasteboard
thickness to 18mm and your workpiece thickness as appropriate (18mm when using a
workpiece, 0 if running calibration on the bare wasteboard)
go to the fluidnc tab, then the maslow settings (maslow.yaml) and you should be
able to find the wasteboard and workpiece thickness variables. Set them, save
them, save the maslow.yaml and then test
BTW, is there a way to ask maslowbot questions about the codebase? I could
learn how it works better so I could try and help diagnose the issue and maybe
future ones
you can create an issue and tell it to create a report about whatever you want,
then when it does so in a PR, keep asking it questions and reading it’s report
(you can close the PR when you are done if you want, or tell bar to merge it if
you think it will be useful for others, just have it create the report in the
docs directory)
I expect that is because almost everyone is using a rectangular frame that’s wider than it is tall - which means you need more force on the belts to move it vertically than horizontally.
That does seem to have been valuable. I don’t think it was the issue that you were seeing, but the AI did spot that the “material thickness” setting wasn’t being applied in the math to locate the anchor points so I have now fixed that.
Not the main bug we are hunting but a valuable find none the less
I’m working on updating my firmware and making a pen plotter and then I could do some more testing and maybe learn more, for now if that ok I’m keeping the PR open so I could ask copilot more questions
