General Help

Can you link to the instructions you followed? I haven't seen any guides that say to tune P to oscillation then increase _I_ to remove oscillation.

Is the jump due to FF0 or FF1 perhaps? What numbers do you have there?

The bump could be FF1 if overall scaling is wrong. Try an FF1 of zero and see if the bump disappears.

tjd wrote:

can you explain what you mean by scaling?

Assuming a velocity-mode servo drive, if a velocity command of 10 units/sec into the 7i77 analogue stage gives 10 units/sec at the motor, then the scaling is correct.

So, then, if an input of 10 units/sec into the PID from the cmd-deriv input (or internally calculated) when mutiplied by the FF1 Gain (of 1) you should get exactly what is requested as the output velocity.

The PID output is just the sum of 6 terms:
FF0 * position command
FF1 * velocity command
FF2 * acceleration command
PGain * position error
IGain * integral of position error
DGain * derivative of position error.

P, I and D do have units, but you never really know what they are, so they always need tweaking. But FF0, FF1 and FF2 have the same units as the machine, so a gain of 1 is only correct if the output scaling gives the same units.

tjd wrote:

So with that in mind, should I continue with the 'Increase P till oscillation' then increase I to stop oscillation.....' method to tune the axis?

No, because nothing ever says to do that
Some guides say to increase _D_ to kill the oscillation, but none ever say to use I, as far as I know.

On another HAL PID Tutorial it uses Halscope to get the axis tuned correctly - is this something you're familiar with?
gnipsel.com/linuxcnc/tuning/servo.html

I think this is the best way. It lests you see exactly what effect each parameter is having.

andypugh wrote:

tjd wrote:

So with that in mind, should I continue with the 'Increase P till oscillation' then increase I to stop oscillation.....' method to tune the axis?

No, because nothing ever says to do that {edit} Actually, it does say that, doesn't it? I have never heard of that approach, and I would be surprised if it worked well.
Some guides say to increase _D_ to kill the oscillation, but none ever say to use I, as far as I know.

On another HAL PID Tutorial it uses Halscope to get the axis tuned correctly - is this something you're familiar with?
gnipsel.com/linuxcnc/tuning/servo.html

I think this is the best way. It lests you see exactly what effect each parameter is having.