[svenakela]

I'm tuning a bunch of R990H cards I never used with a bunch of Yaskawa/Nidec minertia servo's (cute a powerful small things, by the way). They are rated to 65 V but now when tuning I'm only using 12 V.

When test running the setup with R990H.exe's tune command, I get a really steap and steady curve up to top of the graph and only a overshoot of 14 steps. That is with following config:
Kp: 6016
Ki: 0
Kd: 0

But when I read the suggestions in the nice howto at http://www.eaglemotion.com/home/R9xHbdi/pg5.htm it seems that a servo config should have a Ki-dampening and pretty much Kd. If I test with values closer to the suggestion:
Kp: 6016
Ki: 150
Kd: 4000
The curve becomes a lot more bouncy and I have an overshoot at 80-120 steps.

To me, the first config and graph looks a lot better, but is it normal or okay to run with no Ki/Kd?

Regards,
Sven

[H.O]

Hi Sven,
No Ki or Kd sounds a bit strange. Are you tuning without any load on the motors? Without Ki and Kd the "stiffness" of the motor will be fairly low, I think.

/Henrik.

[svenakela]

Yes, no load. They are on my desk. The servos are really compact and I know some says they are a little odd to run, but no Ki/Kd at all and a perfect tune is strange to me too.

[H.O]

Try loading them just a bit with your fingers or whatever, without any Ki they probably won't reach the target position. Also, when the motor is stationary at try deflecting the shaft, without Ki and Kd it will probably feel "soft and sluggish". But if it works to your satisfaction without Ki and Kd then what the heck...if it ain't broken..... ;-)

/Henrik.

[svenakela]

Check it out.

EDIT: Hmm... Will the characteristic be different when I'm feeding with max voltage? I used 12 V only now, and they're supposed to get 65.

[svenakela]

Well, everything seems fine so... <swedish>...vad tusan. Jag k&#246;r p&#229; det. </swedish>

[H.O]

Hi,
That sure looks like a near perfect step-response! You will definetly need to retune after raising the voltage since that is basically the same as raising the gain (Kp). If it changes the impact on the "no Ki or Kd issue" is hard to say, I guess the only way to know for sure is to try it.

Nice compact motors there, what are you going to use them for?

/Henrik.

[svenakela]

On a steel monster.

[svenakela]

I ended up with Kp at 4000 and still no Ki and Kd. Got a sampling rate less than 0.3 seconds. Pretty awsome acceleration on a servo that small. A lot more overshoot though, but I believe it's a minor issue.

[H.O]

Hi Sven,
Try adding som differential gain (Kd) then, it should decrease that overshoot. When you add Kd you may need to add even more Kp though.

[Karl_T]

...

But when I read the suggestions in the nice howto at http://www.eaglemotion.com/home/R9xHbdi/pg5.htm it seems that a servo config should have a Ki-dampening and pretty much Kd. If I test with ...

I went to this page. Nicest explanation of P, I, and D I ever seen.

5.1.1. Kp – proportional gain. Proportional gain is directly proportional to the torque. It is the main force in the whole PID loop. The kp provides a multiplier for the position error between the commanded position (from the CNC control) and the counter which tracks the feedback from the encoder.

5.1.2. Kd – differential gain. It is like oil in a shock absorber. It makes the servo settle in the desired velocity.

5.1.3. Ki – integral gain. It is like a flywheel in a toy car. You have to push it harder (more Kp) to move it, but it runs smoother then a toy car without a flywheel. The I part of the PID filter integrates the following error and this integral is directly added to the P term and keeps the following error small during the motion or in steady condition. In fact, correctly tuned Ki can keep the following error within +/- 1 encoder count during constant velocity motion (or in steady condition) while the motor can deliver full torque. The other great advantage of Ki is that it can make from a simple trapezoid velocity profile a profile very similar to a typical “S” profile. A trapezoid velocity profile is when the motor acceleration is constant (an increment of velocity) until it reaches the desired velocity. In S profile, the acceleration at the start is rather low, then it exponentially goes up till it reaches the maximum possible acceleration. Then, just before reaching the desired velocity the acceleration gradually decreases. You can see the typical S velocity profile on the faces of the astronauts in the space rocket while taking of.




BTW, I have some servos in a Hardinge HNC that liked almost no I and D

[svenakela]

Hi Sven,
Try adding som differential gain (Kd) then, it should decrease that overshoot. When you add Kd you may need to add even more Kp though.

Nope, it didn't help (I tried that). The overshoot was still the same but with a lot more oscillation before the servo settled. With no Kd the curve is still the best.

Maybe I'll try with more Ki later, but for now I think I'll leave it as it is.

--S

[svenakela]

I went to this page. Nicest explanation of P, I, and D I ever seen.

Very nice, isn't it?

[svenakela]

Ok, last feedbak on the issue. I hope.
I added a high wattage resistor on the DC stage of the power supply, and I had to tweak the settings a little bit again. On top of that my servo's doesnt get warm now. Ok, they were not boiling before, but know they don't get hot at all.
Kp: 4384
Ki: 0
Kd: 1200

Sampling in 0.38 seconds and an overshoot of 6 steps. Sounds pretty darn good to me. At least more than enough.

[H.O]

Cool! One question though, where exactly is this resistor located, what's the value and what's supposed to be the purpose of it?

I've read a Rutex application note where they suggest to put a resistor in series with the motor armature to protect the drive from blowing when used with high power, low resistance motors. Is this what you're doing?

/Henrik.

[svenakela]

I have 5 W 10 k resistors connected between the plus and minus on the DC stage, they're connected together with the capacitor. I'm not sure, but I think they help a bit when the servo is regenerative. But I'm sure though that the servo got a little bit easier to tune and behaves very smooth now. And as I said, six steps in overshoot.

[svenakela]

Oh, I forgot the clip!
Recorded a small movie with the camera when I did some test runs with my X axis (rotating ballnut, byt the way). As usual I like the Rutex boards, but these small servo's are amazing.

Look closely, it is moving:
[ame="http://www.youtube.com/watch?v=NkFSWTjDHjA"]YouTube - 20090416012[/ame]

Not even full speed and the stop makes the whole table shiver. The gantry itself is about 80 kilo and still the accel and decel is awesome (gear ratio 22/60):
[ame="http://www.youtube.com/watch?v=Ejed_tbt4Ho"]YouTube - 20090416010[/ame]

[Mariss Freimanis]

Kp only servos are perfectly stable when Kp is small to moderate. The closer you get to the target the less torque you have because P = 0 at the target. This means you always stop short of the target destination.

Increasing Kp eventually results in oscillation. Imagine you make Kp infinite; you get full torque towards the target up to the target location, inertia and kinetic energy insures you overshoot the target so you get full torque back towards the target until you overshoot it again in the opposite direction and the process repeats. You graph shows you are on the edge of that happening. If motors and loads had no inertia then infinite Kp would work perfectly.

Adding Ki and Kd allows you to crank back on Kp. Ki multiplies even the smallest error with time and restores position. It acts as if you do have an infinite Kp without the instability.

Mariss

[svenakela]

The gantry on this machine weighs about 100 kg, and of course I get a bigger overshoot now when mounted on the machine. But it is still okay and it only "oscillates" one phase before it's holding position. I did try all different kinds of setups when I tuned the servos and never reached a smooth curve when I had Ki involved (that's the reason why this thread was created).
I do have Kd now as stated in a previous post and it helps, but Ki doesn't seems to go on these motors. I know they were supposed to be a little bit tricky to tune. Right know I have a config that seems to work pretty well. It's not every day I will blast at full speed in half turns...