Servo Tuning Advice

And here is what adding FF1 does




It doesn't realy change the following error like I think it should, it does change the pid.errorI , and I can make that to be nearly 0 by adjusting FF1.

This and the previous plot seem to indicate that something is wrong in the hal file (a large ferror that the integral term is ignoring in the first plot and FF1 acting like FF2 in the second)

Can you plot pid.1.error (it should be the same as ferror)

Does this tell you anything?



Remember the drives are runnin in torque mode if that matters.

This means that the PID loop is not looking at the same parameters as ferror which suggests there is a wiring error since they both should be looking at the difference in commanded and actual POSITION

And of course if the PID comps idea of what an error is is different from what you are trying to minimize, you are going to be running around in circles...

PCW wrote:

This and the previous plot seem to indicate that something is wrong in the hal file (a large ferror that the integral term is ignoring in the first plot and FF1 acting like FF2 in the second)

Then Maybe I should try adding some FF0?

It is not going to be possible to tune this without figuring out whats wrong and fixing it. You can see the the PID loop is doing the best it can to minimize pid.1.error but for some reason (incorrect connections or execution order in HAL maybe) this is not the same as ferror

can you post your current hal file here?

I don't know why but now the pid.z.error is different with out changing anything.


I included both the hal and ini files with the last plot post.

I dont see anything obvious in the hal file but something is really broken

I guess the next step is to find out who calculates ferror and why is it not the same as pid.2.error (is it passed through Axis to motion?)

You can see the as far as the PID loop goes things are under control (pid.z.error nearly 0 )the PID loop does not see the ferror at all.

So either the inputs to the PID loop are not correct somehow, whoever calculates ferror is not getting the right info or some other subtle bug, This is definitely an error of some kind.

Also since the ferror seems proportional to velocity I wonder if theres a pipelining issue where
there a one sample delay between the commanded and actual positions that the ferror calculation uses (is the ferror ~ distance per ms?)