[Toneski]

Hi Tom/everybody,


I have recently made the jump and decided to change the stepper motors on the mill to servo's.
I have a single servo setup with a new kflop/kanalogue combo in a test configuration - my thinking to be to get it going on the bench, then transplant it to the mill with a known good starting point already setup.





However, the tuning of the servo has proven to be a tricky beast to tame and I was wondering if you could cast your knowledgeable eye upon what I had achieved so far and whether it could be further improved (which I'm sure it could be).


After much tuning (and reading of the WIKI) I have arrived at a set of values which seem to have the least following error - however, I am still getting a following error of > 1100 counts.





There also seems to be quite a bit of noise of the rising/falling edge of the output which is caused by D being set to 64, together with some under/over shoot and I'm wondering if this is normal. The D value also causes a little bit of grumble - is this normal?





I am running the servo with NO filtering at this point as TBH the bode plot output is somewhat confusing - how do I interpret its output and do I need any filtering?




So - in summary:-
1. Can I improve on the current set of parameters at this stage?
2. Can I reduce the following error or is 1100-1200 counts an acceptable figure?
3. How do I interpret the bode plot and do I need any filtering?


I'd just like to re-iterate what I've said before - the kmotion system really is top notch - please keep up the great work.


Regards,
Toneski

[TomKerekes]

Hi Toneski,

1. Can I improve on the current set of parameters at this stage?

You should be able to do much better. Note when adding load on the motor the tuning will change.

The error is large because only your P Gain is doing most of the work. With only P gain an error is required to create any output:

Output = error x P

So to have a DAC output of ~1800 to command the amplifier to move at your top speed

1800 / 0.6 = 3000 counts of error would be required.

In order to have small errors and not to rely on large error and P Gain something else must provide the required output.

Possibilities are Velocity Feedforward (VFF) or Integrator (I Gain).

Your I Gain is too small to do much within the time of the move.

Your VFF is helping and reducing the error to about half.

2. Can I reduce the following error or is 1100-1200 counts an acceptable figure?

I'd suggest removing the Acceleration Feed Forward and increasing the VFF to reduce the error. If the error is negative indicating the motor is lagging behind, increase the VFF. If the error goes positive indicating the motor is surging ahead, decrease the VFF.

The VFF needs to counteract two things:

#1 - drag introduced by the way KFLOP implements the D Gain of 64. The amount of VFF for this is:

64 x 90us = 0.00576

#2 - provide the DAC output of ~1850 counts at a speed of 450000 counts/sec. The amount of VFF for this is:

1850 / 450000 = 0.00411

Total expected ~ 0.00987

3. How do I interpret the bode plot?

Please read this. I believe your stimulus is too small and too high frequency so there is little movement and the result is all noise.

do I need any filtering?

With D Gain you should add a low pass filter to remove spikes in the output. See the section here that begins with "When using D Gain...".

HTH

[Toneski]

Hi Tom,

Apologies for the delay in replying - life has got in the way. I'll be right back onto this problem - hopefully later today.

Kind regards,
Toneski.

[TomKerekes]

Hi Toneski,

I understand. No rush from our end.