this robot that loads pencil leads it really just a software testing device. A research tool. What is new here s not the idea too use two sensors but
1) use a predictive model that says how the system will flex in the future in s given acceleration is apllied. 2) collect data while the machine is running in so that the above model is up to the millisecond calibration data. Notice the robot starts by slowly placing the lead in that speeds up at the end. The author said this is because the machine runs only as fast as it as confidence on the predictive model. Once you start an operation you coect data and can speed up in real time. To do this on a mill you would need glass encoders physically connected to the table, a rotary encoder on the servo motor. But really, you need sensors any place that can inform your physic model. Again what is new here that LCNC doe not do is attempt to predict in advance the effect of changing the motor torque. A PID loop, ow two nested PID loops can ONLY see error after the fact. PID is reactive. But real world machine always have a delay between applied torque and action because the physical material (even cast iron) is elastic. The reason he used these "crappy" hobby servos and long, thin aluminum extrusions was not because he was dumb or broke. But because H wanted a system with loads of flex and backlash so that he could test his ability to measure flex and backlash. Technically what he is doing is using a Kalman Filter to estimate the position of the robot's hand. KF is mathematically complex is really not hard to understand intuitively. The idea is that if you have many independent source of information you can combine those source and the result is much better then any of them. The Wikipedia article explainsbut is not easy reading, look up to tutorials if you like ut basically the KF use a model to predict the current stat from measurements that were taken in the past. It is not a reactive controller lk PID. On Tue, Jan 4, 2022 at 11:27 AM dave engvall <[email protected]> wrote: > Hi, > I ran across an article by one of the servo drive firms that said the > using two sensors/axis improved motion. I think they used a resolver on > the drive and an encoder on the ball screw. While this improves motion > that assumes you have very good ball screws. I think that a better > approach would be to use glass scales and then another sensor on the > ball screw or drive. Right now I have a mill with glass scales on X and > Y with Z being considered. Planned is implementation of stepper-servos > like Gene's setup. First iteration will be stepper-servos plus a DRO > simply to tell me how bad things are. > BTW -- on X using the hand wheel vis a vis an accu-rite 5 um glass scale > was dead on moving X in the positive direction I won't admit how bad > it is on reversal. ;-) Y is damaged but still pretty good going positive > but really bad in Y neg. No scale on Z so no estimate but not perfect. > The X does have a 40 K count/inch in quadrature encoder on it. > Note: Chinese glass scales are now mounted on X and Y since the > Accu-rite scales do not play well with the Chinese DRO. > > Cheers, > Dave > > > > On 1/3/22 6:58 PM, Sam Sokolik wrote: > > This has been done at a bigger scale... (2 feedback loops) > > > > > https://web.archive.org/web/20160222165548/http://jmkasunich.com/cgi-bin/blosxom/shoptask/wichita-trip-02-20-08.html > > > > > http://wiki.linuxcnc.org/cgi-bin/wiki.pl?Combining_Two_Feedback_Devices_On_One_Axis > > > > On Mon, Jan 3, 2022, 8:42 PM Chris Albertson <[email protected]> > > wrote: > > > >> CNC is about precision motion control. Here is a new idea where this > >> builder gets 0.05mm accuracy but uses hobby level R/C servo Not only > that, > >> but he connects three of these in series, one to the next to the next so > >> all the backlashes and poor tolerances add together. Then he uses this > to > >> do precise motion. He loads a mechanical pencil with this chain of > cheap > >> parts. > >> > >> What does this means for LCNC? It means that someone has found a > software > >> solution to backlash. What he does is place a quadrature encoder on > both > >> the motor and the output shaft. The difference in encoder reading is > an > >> exact measure of mechanical backlash and effective gear ratio. He can > >> measure the backlash under different conditions and store the > >> measurements. Then he places a cascaded PID controller and Kalman > filter > >> over this hardware. > >> > >> Technically the problem with backlash control via software is the delay > >> from input to output pays poorly with the PID algorithm. He applies a > >> predictive model. > >> > >> Checkup this video. It is unimpressive if you have a $100,000 CNC > milling > >> machine, but he is using a linked chain of hobby servos. The novel > idea is > >> his software. It is on github, you can read it. > >> https://youtu.be/gq-P39rfRqU > >> > >> He explains it here. Notice in the video he shows the backlash. The > >> gear-slop is at the 1/4 inch level but using his software backlash > >> correction you can see the results in the dial indicator is about 0.05mm > >> (or about 0.002 in American units) Not bad given the truly horrible > >> mechanics. > >> https://youtu.be/SioCwvR_PYY > >> > >> Why do I care? I have a robot-dog leg here on my workbench using hobby > >> servos, let's say performance could be improved. But the anti-lash > >> technique might be used on a real milling machine. Maybe one of the > >> experts here could look and see if it could be applied? I will use > parts > >> of his idea on my dog-bot. > >> > >> -- > >> > >> Chris Albertson > >> Redondo Beach, California > >> > >> _______________________________________________ > >> Emc-users mailing list > >> [email protected] > >> https://lists.sourceforge.net/lists/listinfo/emc-users > >> > > _______________________________________________ > > Emc-users mailing list > > [email protected] > > https://lists.sourceforge.net/lists/listinfo/emc-users > > > > _______________________________________________ > Emc-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/emc-users > -- Chris Albertson Redondo Beach, California _______________________________________________ Emc-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-users
