On Wed, 2010-07-07 at 20:43 -0500, Chris Radek wrote: > On Thu, Jul 08, 2010 at 10:32:50AM +1000, Jake Anderson wrote: > > We have converted our mill to CNC and EMC2 and all is fairly well. > > http://wiki.linuxcnc.org/cgi-bin/emcinfo.pl?JakeAndRussells > > ... > > > Unfortunatly when I change that cruise speed (g1 F200 vs G1 F100 say) > > the ferror goes up, and I start either leading or lagging the set point. > > I can tune to hover around 0 error at any given speed, but as the speed > > changes I start getting error again. > > > Any tips? > > Yes but beware they're all half-baked. > > I see you're using bare H bridges which means you don't have a > velocity loop. I think this problem with FF1 is because of that, and > it is a fairly fundamental problem with a torque-mode setup. > > In hand-wavey terms, a full velocity mode servo amp has a commanded > velocity (output of emc's pid in our case) and actual velocity (from a > tachometer) and it uses the difference between the two to determine > the torque/current to apply to the motor. > > It seems like you can fake this up by using the mesa encoder's > velocity output as your pretend velocity feedback. Use sum2 (one gain > negative) to calculate the difference between your pid output and your > velocity feedback. Pick your scales carefully so everything matches > up. Use this resulting difference to drive your pwmgen. > > I haven't tried this and the whole idea came out of a late-night chat > session with Kim K. Unfortunately I can't find the archive of it. > > I'd be thrilled to hear whether this gives you a stable loop and > whether it tunes up better. Please do keep notes and report back if > you try it. > > One step more complicated is to have dual pid loops, a torque loop > inside a velocity. I'm not sure what you'd use for torque/current > feedback, though. Normally that's current sensing in hardware, and > you have no equivalent that I can see. > > Simpler is to leave your pid as-is and try using I instead of FF1. If > you can increase your I by a lot and still keep the loop stable, > you'll get correct following at any (steady) speed. The tradeoff is > sloppy settling at speed changes, like at the beginning and ending of > moves. > > Chris
Hi all, I is not that I know anything about tuning torque mode. Just did one a few weeks ago. I started out like usual as for a tach/velocity mode tune. Not the best idea. Take a look at the parameters used on the Mazak conversion. ~/emc2/emc2-trunk/configs/demo_mazak/demo_mazak.ini or something close to that. Those values look pretty radical but it gives you an idea what is needed for tuning torque mode. Maybe try 1/20th or less of those values and see what happens. Pretty much you get it going with a reasonable P and then use I and D get stability and then increase everything in steps to get decent stiffness and response. At least on my machine FF(n) didn't do much good and believe me I did try them. Linear scales on machines that are not really low in backlash can be a challenge. I got better results with encoders on the ballscrew. The mill was a Cinci contourmaster with servos that had, and still does, about 0.003" backlash. In terms of tuning, linear glass scales, encoders on the servo motors then encoders on the ballscrews; encoders on the ballscrews being by far the easiest to tune. SEM MT30H4s with tachs. for X and Y and a torque mode amp for the Z. Hope this helps. Dave > > ------------------------------------------------------------------------------ > This SF.net email is sponsored by Sprint > What will you do first with EVO, the first 4G phone? > Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first > _______________________________________________ > Emc-users mailing list > Emc-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-users ------------------------------------------------------------------------------ This SF.net email is sponsored by Sprint What will you do first with EVO, the first 4G phone? Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users