As a completely separate thought ( I did title this …approach(es)…)
I think there could be two levels of the existing calibration process-
A loose one, used where the dimensions are only approximate and the system has to provide a large amount of slack to avoid interference etc.,
and
A tight one that is only used when one has a usable set of anchor point coordinates (sufficiently precise to allow continuous motion with all four belts in play). Regardless of the method used to reach this point, this would allow the existing calibration methodology (which as noted seems to be advantageous specifically due to the large number of points captured over the entire area) to be used iteratively to achieve higher precision, rather than starting from scratch every time. This could be-
Faster, since it could travel directly from point to point.
Safer, with less manual oversight, since it would require only slight slacking and only (I think) while parked at each point.
These two items would make it practical to re-run the calibration more often in use for verification and confidence.
Potentially higher precision- due to the previous two factors, it would be more practical to-
Measure vastly more points.
Re-check points determined to be outliers.
Adjust the density of points measured in different areas.
Potentially perform series of measurements to calculate factors such as Z axis angling of the belts.
With the machine parked on an anchor point, that would not be necessary
I think difference in diameter of spindle and anchor is a fixable problem, and so is the difference in z-height, because right now the software is taking that into account
The real issue is that we are asking a lot more of the user to calibrate this way, with more things that can go wrong (although the belts being eaten is not one of them)
It would be cool if the option to calibrate this way would come available though.
My apologies- I misunderstood. Your approach would be to place the sled at/on each of the anchor points and measure from there to the other three anchor points.
I agree that the anchor/collet issue can be addressed, and the Z height issue can, if nothing else, be accounted for. Balancing the sled on the 3D-printed anchors can be overcome with some blocking, but now we’re adding a lot of error-inducing elements into the process.
For a 12 ft x 8 ft (roughly 3600mm x 2400 mm anchor pattern) frame, the diagonal distance would be at, near, or past the full extent of the belt length, but if nothing else the diagonals are the case where the sled can easily be placed on the existing support. and the distance measured with two belts.
I completely agree that this lacks the elegance of having the machine calibrate itself, but I would temper that disappointment with three observations;
The existing process is itself both problematic and requires manual input (although I believe with refinement it might be as easy as envisioned).
The process of building a frame probably involves at least as much measurement as is being asked of the user in this process, which could have some automated assistance.
This is envisioned as a bootstrap process, with further iteration being done via a process that would hopefully be even more automated.
I think the simple expedient of having to manage only one or two belts initially is potentially a significant advantage.
The six-measurement method you started the thread with seems like a good candidate for the loose calibration. My guess is that it would be pretty good.
Ideally, Maslow could interface with a secondary position tracking reference, camera based, time of flight based, other, etc and base calibration on multimodal.
How about an injection moulded strip with a steel insert pin and a hole in the other end that’s around 350mm long…?
Fasten the pin into the router, then hook the strap onto the anchor and you’ve a known measurement, it’s not balancing, and then repeat for all the other corners …?
Given the challenges with using belt tension including the limits it places on frame size vs workpiece size, it would be interesting to consider alternative tracking systems, optical, attaching an X and Y axis laser that shoots low to a sub sidewall on X and Y frame to get ~3mm absolute position accuracy of the sled:
Given the challenges with using belt tension including the limits it places on frame size vs workpiece size, it would be interesting to consider alternative tracking systems, optical, attaching an X and Y axis laser that shoots low to a sub sidewall on X and Y frame to get ~3mm absolute position accuracy of the sled:
Sub-mm using belt tension alone seems challenging to me given changes in belt after calibration and over time due to elasticity change from wear, temperature, vibration and resonance on the sled during cuts, dust accumulation in belt and gears etc.
use secondary tracking along with belt tension may get to sub-mm in a reduced frame footprint
