Anders Wallin wrote: > Hi all, > > I got my x-axis servo motor mounted today and ran some tests. > There is a video and some hal-scope screen shots here: > http://www.anderswallin.net/2008/04/x-axis-test/ > > This is a system with no motor current, motor voltage, or velocity > feedback. An m5i20 generates PWM output for pico-systems DC-brush amps > and 4000 pulse/rev encoders provide feedback. > This might make the system a bit different to tune from other systems > which run another loop on current/voltage/velocity. > > I found FF1 worked quite predictably and reduced the cruise-phase error > to very little. Only P and FF1 was needed for that. (2nd pic on page) > > Then I added a lot of I-gain which helped a bit for the acc/dec phases. > > With FF2 I became confused on a higher level (3rd pic on page). It seems > to either fix the acc. phase or the dec. phase but not both. > I couldn't find your serial number in my data base, so I don't know what version of it you have. The earliest models had a little bug in them during deceleration. The drive uses a sign-magnitude scheme where it shorts the motor whenever it is not applying power from the DC supply. If the current exceeds the current limit, it goes back to shorting. This is fine during acceleration, but makes the problem worse during deceleration. These are the rev A through C versions.
So, I had to change the circuit to go to open-circuit on current limit, since the current sense logic can't tell acceleration from deceleration. Rev D and E have this fix. > The last pic shows moves at different feed-rates. All of them show > around 16 to 25 count spikes during the acc/dec. phases. Presumably at > the moment when acceleration changes and jerk is infinite. How much have > other people been able to reduce these spikes? Well, that tells me either the FF1 is correcting too little or too much, or you are demanding more accel/decel current than your current limit is set for. It is VERY easy to over-correct with FF2, so I always check for having too much. Since you are direct-driving with a small motor, it is also easy to ask for too much torque from the motor. See http://wiki.linuxcnc.org/cgi-bin/emcinfo.pl?PWM_Servo_Amplifiers for some of my tuning results. The last picture actually represents a fair number of encoder counts, but at the resolution of 128,000 counts/inch, it is still a far smaller movement than the machine's accuracy. Jon ------------------------------------------------------------------------- This SF.net email is sponsored by the 2008 JavaOne(SM) Conference Don't miss this year's exciting event. There's still time to save $100. Use priority code J8TL2D2. http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users