+/-10V or closed loop step/dir?

[m_c]

Here's a servo flow diagram courtesy of Dynomotion -


That's essentially how most servo's will work, it's just depending on the type, certain parts of that diagram may be handled by different components, and certain parts may be duplicated, or just not exist.
It is essentially black magic involving lots of mathematics!

P(roportional) controls how much output to generate for the given difference between position and destination (destination being where you want the servo to be, and position being where the servo actuall is). This is the key output that does the most of the heavy lifting.
I(ntegral/integrator) is how quickly to increase output for positional difference. This is the output that tries to compensate for following error, but can cause loss of stability if set too high (aka it'll cause wild oscillations)
D(erivative) is like a damper on the PI output, to increase stability, but go too high and the servo loop will become overdamped and increase following error.

Then you have the various Filters. The aim of these are to smooth the generated outputs, so you can push the PID values a bit further. I.e. say you have a stability problem at a certain frequency, you can add a filter that will smooth the output at that certain frequency (pay particular attention to my previous comment about mathematics and black magic!).

Feed Forward is then essentially a bypass function, and often seen a bit like a sticking plaster similar to backlash compensation, however it can be useful.
It essentially bypasses the PID loop, which can help improve following error. If you know the PID loop is slow to respond, you can use some FF to 'kick start' motion change, so by the time the PID loop has responded, the following error is already pretty minimal. However the drawback is, if the motion has some kind of change/interruption, the PID loop will still be slow to respond, which will give you a spike in following error.



Using the KFlop for example In +/-10V operation, all that functionality is internal to the KFlop, however the servo drive will also contain some of that functionality.
In Position mode, the drive will be using PD to create a stable speed output for any given input, regardless of servo load.
In Torque mode, PD in the servo drive is mostly bypassed, and the drive essentially operates like an old fashioned amplifier drive. Torque mode gives the controller the most control over the servo, however it's naturally more unstable than speed mode, as the generated motor speed is directly related to the motor load

To give a practical example, if the motor load changes during travel say due to cutting load changes.
In speed control mode, the servo drive will mostly compensate for that change in load to keep the motor speed constant, with the KFlop only needing to give minor correction to maintain position.
In torque control mode, the servo drive will not compensate, so it's up to the KFlop to change the required torque demand, to keep motor speed constant, which will involve quite a major output correction to maintain position.

To give a good analogy, torque control is how the accelerator pedal in a modern vehicle works. You want to speed up, you put your foot down, you want to slow down, you lift your foot up. To maintain speed, you hold the pedal partially pressed. This is why you'll get maximum revs before the pedal hits the floor with no load.
If the accelerator pedal worked in speed mode, moving the pedal slightly would cause full acceleration until the new target speed was reached.


Run open loop step/dir, everything after the trajectory planner is handled by the drive.

Run closed loop step/dir, you end up with two of those loops, which allows the controller to fine tune position and try and compensate for performance changes. However, with just closed loop just with the motor, once tuned, closed loop step/dir doesn't really have much benefit, other than the controller knows exactly where the motor is. When you have an outer loop closed using linear scales, it gives the ability to potentially compensate better for minor positional variances. I say potentially, because although it can improve things, it still won't compensate for things like excess backlash or stiction.


It is all essentially one big mathematical balancing challenge.