Speed and position closed-loop servo drive modes

warning

Due to memory constraints, extended modes are not always available. If you do not see folder /driver/pid_speed or /driver/pid_position, the extended modes are not present in your firmware.

If one of these bits is specified in the requested drive mode, the PID loop is activated. We call these "extended" modes.

Bit	Dec	Name	Reference value
4	16	Speed loop	rref
5	32	Position loop	pref

examples

Drive mode = 17: The speed PID loop is active. The motor is driven in voltage mode (17 - 16 = 1)

Drive mode = 34: The position PID loop is active. The motor is driven in current mode (34 - 32 = 2)

Parameters

For /driver/pid_speed and /driver/pid_position : both folders contain the settings for the corresponding PID regulator.

Parameter	Description
P	Proportional term for the PID regulator
I	Integral term for the PID regulator
D	Derivative term for the PID regulator
Td	The time constant for the input filter for the D term. The D term is sensitive to noise, using the filter can help to reduce the noise.

tip

The Td parameter is LPF (low-pass filter) time constant. More about it.

Tuning the loops

Tune one loop at a time, on a test stand where the motor can spin freely, with conservative iref and rref limits. Small steps, one parameter at a time.

Start from a well-tuned current loop. The speed and position loops act on top of the current control, which must be fast and stable first. Its pid_iq / pid_id gains are identified automatically by identlin; re-run it whenever the motor or the supply voltage changes substantially.

Apply a step and watch the response. Command a moderate step in rref (or pref) and observe /driver/motor/rpmf (or the position readback). The textbook criteria apply: rise time, overshoot, settling, and steady-state error. Any of the standard methods gives a usable starting point:

• Manual (heuristic) tuning - raise P until the response oscillates, back off to roughly half of that value, add I until the steady-state error disappears, then add a little D (with Td set so the derivative does not amplify sensor noise) to damp the overshoot.
• Ziegler-Nichols - find the ultimate gain Ku (the P at which the loop sustains a constant oscillation with I = D = 0) and the oscillation period Tu, then take the tabulated fractions of Ku and Tu for a P, PI or PID controller. The Tyreus-Luyben variant uses the same Ku/Tu with more conservative fractions and suits loads that must not overshoot.
• Cohen-Coon - derive the gains from an open-loop step response (process gain, dead time and time constant); useful when a closed-loop oscillation test is not acceptable on the machine.
• Relay (autotuning) test - the modern form of Ziegler-Nichols: a small bang-bang excitation around the setpoint yields Ku and Tu without driving the loop to the stability limit.

Textbook formulas express the integral and derivative action as time constants (Ti, Td) in seconds; the driver's I and D are gain terms. Use the methods for the procedure and for the ratios between the terms, then scale the absolute values by experiment on the step response. Validate the final set at several speeds or positions, with and without load, and in both directions.

Design choice: the integral term does the recovery

Both loops leave saturation as soon as the error allows it, with no windup delay. That recovery is carried by the integral action, so I is a working part of the controller rather than a refinement on top of a P-only loop. Running a loop with I = 0 is possible, but it changes what the loop regulates: after a saturation event it holds the change in the measured value since the loop was engaged instead of converging on the reference.

For the position loop this means the motion stops after

delta = 1 / P electrical radians, which is 1 / (P * 2 * PI * rpole) motor turns

regardless of pref; halving P doubles the distance travelled. This can be used deliberately as a "move by 1/P" behaviour. For reference tracking, keep I non-zero, including during the first step of a tuning session.

A loop saturates on its very first evaluation whenever P * err exceeds 1; for the position loop engaged with a step command that is P * 2 * PI * pref * |cmd| > 1, which is easy to meet with a large pref. Ramp cmd, or use a motion profile, so the loop is not presented with a full-scale step.