How perfect do the dimensions of the frame need to be?
I’m asking because my calibrations keep failing. I’m getting good enough calibration runs that the thing is able to calculate a goodness of fit, but the fit seems to degrade as the calibration progresses. Going off memory here, but I think last time I got a goodness of fit of like 0.62 on the first calculation. But then the calibration process continued and the next calculation was down to 0.49, then 0.33, then I got the “do not use these values” message.
I originally used the 12x8 frame made with four anchors drilled into my garage floor. I just measured using a tape-measure, and loosely converted those measurements to [mm] for the setup file. I don’t think those four anchor points made a perfect rectangle. I never even got a calibration value for that frame. Just the “Do not use these values” on every attempt.
Next, I placed the anchors at the corners of a 4x8 sheet of plywood, measured with an accuracy of probably +/- 1/8 inch, and converted to [mm] for the setup file. That one produced the degrading calibration values mentioned earlier. The frame is pretty square, and I think the precision is decent. I have another idea using precision-cut metal frame, but I’d like to know it will be worth the cost. Do I need to measure with something better than a tape measure?
I don’t think that your initial measurements are the issue here. It’s normal for the value to get lower with each set of measurements (it’s easier to find an answer which fits less data) but it shouldn’t go that low.
There are two things which come to mind which could cause this.
First is flex in the frame. If the frame has any flex to it you will see behavior like this.
Second is a magnet slipping. If one of the magnets isn’t fully glued in place and can rotate slightly it will cause the quality of the measurements to degrade over time.
you can use this onshape page to put in the measured dimensions (including
diagonal measurements) and have it give you the coordinates to put in the
maslow.yaml file
I finally fixed my connection issues and got to try this suggestion. I tightened up the frame, and I also reduced the belt force back to the default 1300N (Was 1500N because the belts weren’t fully retracting when I first got it assembled). I think the anchors are pretty secure now, and the first calibration fitness value is usually ~1.5 to 1.7. Second is usually around 1.0, third is around 0.7, but then it goes a fourth round, and it’s back around 0.33
How many times does it go around and calculate fitness? What do the fitness values even mean? I was surprised when I saw values over 1.0 I would have thought that was 100% fit. Also, if checking for a slipping magnet is my next step, how do I go about that?
How many times does it go around and calculate fitness?
it depends on the grid size you specify, the default 9x9 takes 4 rounds
is it possible that the default calibration size is too large for your frame?
try specifying something smaller
What do the fitness values even mean?
fitness = 1/average error calculation so a value of 1 says that as it averaged
all the belt lengths from each measurement point to where it thinks the anchors
are, it is off by ± 1mm
Also, if checking for a slipping magnet is my next step, how do I go about
that?
if you retract, then extend and retract again and get any belt lengths off by
more than a fraction of a mm you probably have a magnet slipping
there is no way to be sure other than to take the arms apart and physically
check them.
Any luck fixing your problem? I too am unable to complete the calibration process using concrete anchors in my garage floor. Dimensions are accurate, but I keep getting the WARNING FITNESS TOO LOW, DO NOT USE THESE CALIBRATION VALUES. My “Fitness” is 0.14 for 100-1000.
Yes! It finally worked. I’m not sure which steps were necessary, but here’s what I did, more or less in chronological order:
switched to a 4x8 sheet of plywood to guarantee my grid was square (I think my concrete anchors were more of a parallelogram). - calibration failed because plywood was not rigid
Reinforced plywood frame with some 2x4s and went through a couple rounds of failed calibrations messing with the anchor mounts to get them really rigid - Calibration still failing. maybe grid was too big?
Reduced from 9x9 grid to 7x7 grid, which also required me to reduce the calibration grid size. It worked! Final calibration value was something like 0.567, which I guess was good enough.
That seems to have worked! Final calibration value was 0.567, which apparently was good enough. I cut out a puzzle piece example g-code to test it. My next questions will probably be about generating g-code, whenever I get back to this after the holidays. Thanks for the help!