FLUX_OBS #

Brief #

flux_obs is a hybrid rotor flux observer for an induction motor on an encoder (review item B5). It measures the rotor flux angle and magnitude from voltage and current, without tr, and can trim the current model’s slip so that the field orientation follows the measured flux at speed. Below its crossover it falls back to the current model (acim_flux), so at low speed nothing changes.

Status (14 Oct): shadow (pos_err, psi) and the slip trim are bench-checked on the spindle, cold and warm, without load; untested under a cutting load.

conf/template/flux_obs.txt loads it after acim_foc (link flux_obs) in shadow mode, at rt_prio 9.4, with en = fault0.en_out, r = conf0.r, l = conf0.l, adv = hv0.adv, id/iq/ud/uq = hv0.*_fb, pos_ref = angle0.pos, vel_ref = angle0.vel, psi_cm = acim_flux0.psi, slip_in = acim_flux0.slip and slip_max = acim_flux0.slip_max. flux_obs0.pos_err then shows how far the current model’s orientation is off at speed (tr). To trim the slip with it: flux_obs0.kp_s = 20, flux_obs0.ki_s = 2 and angle0.slip = flux_obs0.slip; flux_obs0.k is then true / modelled slip.

Component Explanation #

All work is done in rt; rt_start clears the observer and sets k to 1.

1. Model (Jansen / Lorenz, stationary frame):

• With l = sigma*Ls and psi = lmr * i_mr (the rotor flux seen from the stator, as acim_flux0.psi):
• Voltage model: d(psi_s)/dt = u - r i.
• Current model: psi_s_cm = psi_cm e^(j pos_ref) + l i.
• Hybrid: d(psi_s)/dt = u - r i + kp (psi_s_cm - psi_s) + ki int(psi_s_cm - psi_s), with kp = 1.4 wc, ki = wc^2.
• Rotor flux: psi_r = psi_s - l i; psi is its magnitude, pos its angle.
• The estimate follows the current model below about wc and the voltage model above it, where only r and l matter and r’s share falls with speed. The current model term holds the voltage integrator’s drift (offsets, r error).
• The resistance used is r + r_w * abs(vel_ref).
• While en <= 0 the stator flux is set to the current model’s and the integral cleared.

2. Frames:

• Like obs: the feedback (hv0.id_fb …) is in the frame pos_ref + vel_ref * adv (angle0.pos, hv0.adv). Run flux_obs before angle0 (lower rt_prio), so pos_ref is the angle the feedback was taken with.
• The voltages are the mean over the rt period before the sample, which the frame turned through, so they sit u_adv off the current’s frame: half an rt period (u_adv 0 = -period / 2, -100 us at 5 kHz).
• Do not calibrate u_adv on idle holds: on the spindle (13 Oct) -30 us zeroed pos_err there, but that hid the current model’s slip error and left 0.17 rad while braking. At -100 us the rotor resistance seen by the observer (slip * psi / iq) is 0.121 to 0.128 in run-up, hold and both brakes, at -30 us 0.09 to 0.15.
• pos_err = pos minus that frame: how far the current model’s orientation is off, 0 below the crossover (and 0 while psi is below 1e-4).

3. Slip trim (off by default, shadow):

• slip = slip_in * k + kp_s * pos_err.
• kp_s pulls the frame onto the measured flux. The pull is bounded to 0.5 * abs(slip_in) + slip_min, so a wrong angle (voltage limit, sensor fault, transient) cannot drive the frame off by more (it was open: +-63 rad/s at kp_s 20).
• ki_s moves k, the ratio of true to modelled slip (tr / true tr), so pos_err goes to 0 in steady state and a load step finds the right slip at once. Flux ahead of the frame (pos_err > 0) means the true slip is larger. k stays within 0.5 to 2.
• Both act only while the gate is open (trim = 1): en > 0, abs(vel_ref) > vel_min, abs(psi_cm) > 1e-3 and psi > psi_min * abs(psi_cm). k moves only while also ki_s > 0 and abs(slip_in) > slip_min.
• With acc_max > 0, k also holds while abs(acc) (the vel_ref derivative, 5 Hz low pass) is above it. With u_adv right k is the same in run-up, hold and brake (spindle 13 Oct: within 0.01 in field weakening), so the gate is only for a doubtful u_adv.
• k starts at 1 when the drive starts (rt_start) and otherwise keeps its value while the gates are shut or en is 0, so slip = slip_in * k still applies there: k tracks the rotor temperature, which changes slowly, so a new enable starts from the last learnt ratio.
• The trimmed slip is clamped to +-slip_max when slip_max > 0, as acim_flux clamps its own.
• kp_s 0 and ki_s 0: slip = slip_in (with k at 1), the observer only reports. Wire angle0.slip = flux_obs0.slip to use it.

4. Results:

• Spindle bench (13 Oct, unloaded): shadow (pos_err, psi) checked.
• Spindle bench (14 Oct, no load, template): shadow within 0.03 rad cold. Trim kp_s 20, ki_s 2 follows the rotor temperature (k 1.0 cold, 1.07 to 1.11 after 20 min at 2352 rpm) and holds pos_err within 0.02 rad.

The slip trim is not tested under a cutting load yet.

Pins #

References #

Source: .c File