Correct, unfortunately it turned out to be iterative for me. In ideal situation, it should spit out the correct values in one iteration. That was my goal. But it did not happen to me. My guess would be that mostly because of measurement errors. Since getting the data to the notebook without a z-probe will be even less precise, then I'm not sure how much it makes sense to try to feed the data with just manually finding the touch point and copying the data from the get current position command to the notebook. Maybe you can test the precision of your manual method by measuring a set of spots more times and check whether you are getting the same results. This would also require you can move each tower carriages by one micro-step (pronterface has minimum of 0.1 mm which is about 8 micro-steps for me). But even with imprecise data it may give you better hints how to move tower positions. I did not try it so I do not know.Quote
payala
Am I right that the process is iterative? You have to measure, feed the notebook with measured data, get new params, measure and start again?
But it did converge quite well for me (3 iterations were needed) and it did converge for Matt Peterson (I do not know how many iterations did he use). This are two cases I know about which showed good convergence. Richie did not finish his attempt to use the notebook. He did not use either maxima or the notebook correctly. Richard wanted to use it but he has given up before actually starting. Nobody else reported any feedback. The only difference (I was able to notice) between these people is that Matt was really good at math. Probably better than me (his responses were quick and insightful). He even made a Sage port of the notebook since he likes Sage Mathematics more than Maxima. This indicates that some math intuition to guess what to make it to converge may be needed.
If it does not converge for you then it can be also because you have a printer geometry error which the notebook cannot fix (e.g. skewed tower or build plate far from perpendicular to towers).
Richie tried to find some heuristic to do this after he had given up his attempt with the maxima notebook. I know his first attempt had some errors I pointed out to him. He reported fixing it but I do not know how and I did not review his code. If you add a z-probe you can just try his firmware.Quote
payala
Anyway, this has got me thinking about the problem of auto calibration in general, you just need to optimize your parameter vector to minimize z error across the build plate. And I was thinking about implementing it in firmware, so the printer gets in autocalibration mode and does the iteration to optimize the parameter vector. So, in fact, I was thinking about implementing the optimization with a genetic algorithm. As bad as the printer calibration can be, you start from a not so bad situation usually, and you just want to find a better point, the problem is that it is not intuitive what is the effect of each parameter change, and especially, the amount of change. Maybe an iterative optimization algorithm could find a good solution for itself. This would be very useful for people that cant get through the math to calibrate their printers, or people that have children like me and have no time ;-) What do you guys think?
If you (or anybody else) would want to implement this then there are at least two options:
- Simulate manual callibration in firmware. This is what Richie is doing. This is not computationally intensive but it will take a lot of iterations. If you like this approach it may be a good idea to just branch his firmware.
- Use the approach in the maxima notebook. This is computationally expensive. You need to do it in a PC application communicating with the firmware but it should take only a few iterations to calibrate the printer.
