I am poking around in the kinematics math, and I am trying to get an idea of where we might be able to tweak the math and knowing what our errors in the best case scenario currently look like would be helpful.
The details I am looking for:
What is our horizontal error across the entire work surface? If you make a cut on the lower left side and one on the lower right, what is the error in the horizontal distance between them?
What is our vertical error across the entire work surface? If you make a cut on the far right at the bottom and one on the far right at the top, what is the error in distance between them?
I understand that there are other problems that may cause errors, but I am trying to think about our math errors. My understanding is that in the sweet spot we are getting errors of 1/16 or less.
really all the errors are either math errors or the machine was given the wrong
info
there is a topic āsources of errorā where we go into exhaustive detail about
what can cause error (and most of them then translate into math errors)
the calibration routines are designed to show the different error sources in the
biggest way to try and tweak the match to correct for them.
the model implemented in the kinematics is very limited (a tad less so in the
holey calibration version than in the stock version, but still limited)
by the way, the errors tend to be curves, not linear, and symmetrical around the
center, so you will probably have more vertical error between and edge and the
center than the two edges.
somewhere in GC there is a simulator (it was a stand-alone program at one point)
that lets you enter error factors and see the resutling position vs the desired
position. find and explore that a bit to get a feel for how errors in the
math/dimensions translate into position errors
sorry this does not answer your question, but, My opinion as a kit seller is that most errors are due to not being set up correctly,not calibrating correctly, or hardware limitations. Web control is a big step forward, but how many people are actually using it out of the thousands of kits sold? Instructions for setting it up only got translated to ānon geekā this year. A quick youtube search shows a few 2 year old videos on calibration and one recent webcontrol one. None of them really address how to correct for ovals instead of circles , etc.
If a longer 12ā bar gives better results why arenāt all kits just sold with longer chain. 2 extra meters is $4 more in whole sale costs , maybe $12 retail mark up.
I ran an experiment and sold 100 top bars cut with motor mounts all the same distance in hopes that calibration data would be shared, no one did.
my next thought is including a tape measure in each kit, they are cheap and would ensure everyone is using the same measurements. motor brackets could even be made that have slotted holes so motors could be micro adjusted to be exactly at the same spot for all machines.
Iām with you on this. One of the main reasons I purchased a Metal Maslow was because I was hoping a community would arise and be able to share similar information. This alone would go a long way to making this specific unit more consistent among users.
so instead of a show off a picture of your sled, you would rather have folks post their calibration numbers? That isnāt a bad idea so we can see what the variance is based on frame size / setup. Maybe that could be a user forum profile field to fill in the calibration dataā¦ or maybe have one of those āsend data to help update the productā participation deals where your calibration info along with your other cal numbers could be added to a github spreadsheet every calibration.
there are only a few things that vary from machine to machine the motor to motor distance being the most critical and this is rather easy to get consistent from one build to another. Sled weight data could also be shared. The overall goal is to eliminate as many variables as possible so setup is not frustrating.
I agree with @dlang I can see some aspects of the machine that are not captured by our model. I was thinking of trying to figure out if there was a missing element that we could account for that might get us a large enough pay off to be worth adding to the model. As it is currently implemented, the model doesnāt account at all for the ~15 degree lean of the work surface.
Once I get my machine finally up and running. My plan is to run the holey calibration, but to save the inputs and the measurements. If I know the chain lengths at each of the hole locations, I can play around with the math to see if anything would get me better results.
A crazy idea that is basically duct tape for kinematics, is to take @madgrizzleās idea of finding a best fit function, but only apply this function to the last bit of error that we have. Basically, approximate things as close as possible with the geometric and catenary equations that we have, and then fudge the last little bit of the unknown with a best fit function. The chainElasticity function kinda looks like this already to me.
I would post data, except that my Maslow is sitting in a box right now. I must do all my woodworking outdoors. The rainy seasons in Florida tend to put a damper on summertime fun. Having a roof collapse and storing all of the contents where I try to do my hobbies didnāt help either.
