You are more or less correct. The idea is to retract bottom two, extend enough to reconnect and tighten and measure. With those two lengths and the known length of the bottom axis, you should be able to derive the other two lengths.
You are right that it all depends on the accuracy of the measurements, but the entire system is dependent on the accuracy of those measurements.
I guess I’m worried about the concentration of error. when you retract/zero all
4 belts, you are then getting 4 errors that hopefully cancel each other out a
bit.
when you only take 3 measurements and then calculate what the other two should
be, you are pushing all your error into those two calculated lengths.
but it’s worth trying. If nothing else as a sanity check for when you want to
tell the system that it hasn’t moved.
It’s too bad that the belt material isn’t preprinted with some sort of registration marks (I’m thinking like a QR code) at regular intervals that uniquely identifies where on the belt you are. That would be really useful for finding itself after a power-down, geo-fencing to keep the sled from going off the work piece, and error detection.
It’s too bad that the belt material isn’t preprinted with some sort of
registration marks (I’m thinking like a QR code) at regular intervals that
uniquely identifies where on the belt you are. That would be really useful for
finding itself after a power-down, geo-fencing to keep the sled from going off
the work piece, and error detection.
Interesting idea, but it would take 4 cameras to then read the codes, and it
would be interesting to find where you could buy belts marked like that.
Agreed. I have never seen belts marked like that. In my imaginary world where it did exist, the factory would churn out miles of belt with these codes and guarantee that the codes would be unique for a stretch of belt up to a certain length (1000 m seems achievable). The Maslow would have to remember the arbitrary offsets and the directionality of each belt.
If you used a linear bar code you could probably use an LED and a photodetector rather than a camera, but you would still need one for each arm.
Thinking about this more… the reason I said “post calibration” is that then we can define and save precisely where the “limit clips” are in terms of belt length (because we can drive the maslow into that corner until it hits the limit, and since we would be in a good state, we’d know the precise belt length that stop exists at). Then we don’t have to rely on human measurements to determine that spot on the belt.
My thoughts keep turning to being able to manually save the current location before shutting it off and manually confirm that it hasn’t moved while turned off when booting up, thus letting the Maslow know it can trust the saved location and save you the tedium of reseating it on the frame just because of a power cycle or overnight shut down.
It already has to trust me to not run it off the frame when I tell it to run G code from the chosen home location, I don’t see this being any worse unless someone makes the mistake of not being correct when they press the ‘I didn’t move it’ button.
Thinking about this more… the reason I said “post calibration” is that then
we can define and save precisely where the “limit clips” are in terms of belt
length (because we can drive the maslow into that corner until it hits the
limit, and since we would be in a good state, we’d know the precise belt
length that stop exists at). Then we don’t have to rely on human measurements
to determine that spot on the belt.
when we are shortening the belt, we don’t need a calibration, we can retract the
belt (to get it to 0 from the anchor), extend the belt, and retract the belt
(but instead of setting zero, use the distance at the time it stops as the
amount to add to that belt)
My thoughts keep turning to being able to manually save the current location
before shutting it off and manually confirm that it hasn’t moved while turned
off when booting up, thus letting the Maslow know it can trust the saved
location and save you the tedium of reseating it on the frame just because of
a power cycle or overnight shut down.
Yes, we do want to do this
It already has to trust me to not run it off the frame when I tell it to run G
code from the chosen home location, I don’t see this being any worse unless
someone makes the mistake of not being correct when they press the ‘I didn’t
move it’ button.
The thing we are paranoid about is that these motors are not a worm gear, so if
there is tension on the belts and enough vibration, they may unspool a bit
without you realizing it.
That’s why I told Bar we need to not only save the length, but also the encoder
angle, when we restart, if the encoder angle hasn’t changed, we can make a
reasonable guess that the belt length hasn’t changed. If the encoder angle has
changed, we can correct for up to ~20mm of error, and that should be enough for
people to notice and tell us that it has changed.
we may want to do a sanity check on movement to a couple locations and pull all
belts tight to confirm, but this is still FAR easier than a full retract/extend
each time.
In all honesty, I’d even be fine with having to screw the sled to the spoil board to take the weight off the belts if it came down to it.
Way less effort than a full re-seat, and I already do basically this while reseating.
I have a bungee around my centered frame cross-beam that hooks the top of the linear rod supports and I stick a screw on either side of the sled so it can nest between them, held against the surface by the bungee.
Not really suggesting that as the solution, just musing about my first thought for removing the potential-unspooling concern.
Or, use slotted half inch PVC pipes that clip over the belts to lock the sled in a reference position and use the PVC as a mechanical stop to reconfirm position at next usage cycle.
Yes, I understand that but what I’m saying is that during calibration (edit: at any time after calibration, but in a good state) we could do another sequence where we have the user drive the sled to each corner, add a clip there, and hit a button to have maslow measure that and we’d have the limit stop values. We could then use those to recover from a power cycle where things have changed or belts slipped or whatever. maybe I’m not understanding some nuance here, but this seems to me like the best way to get the retract values for the limit stops.
Yes, I understand that but what I’m saying is that during calibration (edit:
at any time after calibration, but in a good state) we could do another
sequence where we have the user drive the sled to each corner, add a clip
there, and hit a button to have maslow measure that and we’d have the limit
stop values. We could then use those to recover from a power cycle where
things have changed or belts slipped or whatever. maybe I’m not understanding
some nuance here, but this seems to me like the best way to get the retract
values for the limit stops.
that is what I’m proposing, the problem is figuring out how to drive to those
points.