On Tue, Mar 2, 2010 at 6:40 AM, John Kasunich <jmkasun...@fastmail.fm>wrote:
> > > On Mon, 01 Mar 2010 19:10 -0800, "Flying Electron Inc" > <sa...@flyingelectron.com> wrote: > > Hi all, > > > > Hoping some people that have tuned their servos with EMC2 might > > be able to share some of their insights that they learned during > > their tuning with me. > > I built a pick and place machine with a servo on the X axis and I > > started > > running tests to tune the PID loop for the servo. Right now, I have > > everything set to zero, a deadband set to 0.00001 and I am adjusting > > the Pgain to find the optimal Pgain before I start varying the other > > parameters. > > 0.00001 inches seems like a very small value for deadband. What is > your encoder scaling (counts per inch)? Normally, the deadband should > be set to about 1 or 2 counts. Do you really have 100,000 counts per > inch? > Thanks for the help! I had no idea how to set the deadband so I just made up a number. Deadband being 1 or 2 counts of the encoder makes perfect sense to me now that you point it out. The machine has 20807.68 counts per inch (10cm/turn ballscrews with 8192 counts per rev encoders). I'll try setting the deadband to 2 counts, which should be 0.000048 inch. > > > The test I am using is to start from a standstill at X=0, move to > > X = 8 inches with an acceleration of 25in/sec^2 and a maximum > > velocity of 9in/sec, come to a full stop and wait for 0.5 seconds, > > then reverse back to X=0 with the same acceleration profile. > > That seems like a reasonable profile to tune with. > > > I tried different Pgain values of 4, 8, 10, 12, 13, and 14 to see > > what the error looked like. at Pgain of 16 the machine began > > oscillating badly. I've attached a link to an image of the > > screenshots i took from hal scope for the different Pgains that I > > tried. To my inexperienced eye, it looks like Pgain 12 is the best, > > but I'm not sure if the error graphs are even supposed to look like > > this. > > > > http://i.imgur.com/hefCf.png > > > > If anyone can tell me if these error graphs look normal or not, I'd > > appreciate it! > > > > No, they don't look normal. There is way too much wiggling going > on there, even at the lowest Pgain setting. > > Those Pgain numbers seem very small to me, but that might be because > of the way you have things scaled. I recommend that you always scale > things such that you can understand the traces on halscope in physical > units. > > Looking at your traces, motion.current-vel is in inches per second. > The trace clearly ramps up to 9 inches/sec, just like you commanded. > It is easy to understand what that trace means. > > All position signals are normally in machine units - inches in your > case. So the pid.2.error trace shows a maximum error of about 0.035" > in the Pgain = 12 trace. Probably not what you want, although maybe > OK for a pick-n-place machine, since you really only need to be > accurate at the endpoint of the move. > I agree with your thinking, error during the move will probably not be a problem for the machine, just trying to nail the final position at the end of the move. > > What is not clear from your traces is the scaling of the pid output. > The pid.2.output trace is scaled at 0.2 "units" per div, and at the > fastest part of the move, when the axis is moving 9 inches/sec, the > pid output is at 2 divisions, or 0.4 units. That means a "unit" on > of pid output would make the axis move at roughly 22 inches per > second. > The servo drives are driven by a PWM signal and a direction signal. The pid.2.output is scaled from 0.0 to 1.0 corresponding to 0% duty cycle and 100% duty cycle. The servo amps are brushless dc servo amps (more about them below). I'm hoping there is some linear relationship between the pwm output and either torque, velocity, or position, but I'm not sure since there are no current or velocity loops inside the amps themselves. > > I don't know what kind of drives you have, and it makes a difference > in how you tune. If you sending pid.2.output to a PWM generator and > directly driving an H-bridge, then the average motor voltage will be > proportional to the duty cycle, and motor speed will roughly track > the voltage, with some droop under load. If you are sending the > output through an analog output to a torque-mode drive, then motor > current, torque, and acceleration will be proportional to the > output value. (That doesn't seem to be the case here.) Finally, if > you are sending the output through an analog output to a velocity > mode drive, the motor speed will fairly accurately track the output, > but only if the velocity loop inside the drive (NOT in EMC) is tuned. > > Judging from the traces, I suspect you have velocity mode servo amps, > and I suspect their velocity loops aren't tuned. It doesn't make > sense to tune EMC's position loop when the drive velocity loop isn't > right - the position loop builds on the foundation provided by the > drive. > > If you can tell us about your drives (and ideally post a link to a > copy of the drive manual and/or specs), we can probably help more. > The servo drives are from Advanced Motion Controls. They are Direct PWM Brushless DC Servo Drives Model# BD30A8. The datasheet can be read at http://www.a-m-c.com/download/datasheet/bd30a8.pdf . I talked to their tech support and there is no current loop or velocity loop in the drive. They told me the PWM signal directly controls the amount of time the MOSFETs are turned on. If it was a brushed DC motor, I understand that the PWM duty cycle would have a linear relationship to the average voltage and hence average speed of the motor, but since it's a brushless DC motor, I don't understand if that same relationship holds or not. > > Regards, > > John Kasunich > > -- > John Kasunich > jmkasun...@fastmail.fm > > > ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
