IPM #

Brief #

ipm estimates the junction temperatures of the power module (IPM) on the F4 board, so the drive can derate and trip on die temperature between the fast overcurrent trips and the slow module NTC. It adds a modelled temperature rise of each of the twelve dies (per leg: high and low IGBT, high and low diode) to the NTC reading. Loaded by conf/template/ipm.txt, which the motor templates (pid, mpid, uf and vf) link at their end, so a config does not link it again (a second link ipm loads an ipm1 nothing reads): its inputs come from hv0 (ipm0.id = hv0.id_fb, …, ipm0.case_temp = hv0.hv_temp, ipm0.pos = hv0.pos) and fault0.ipm_temp = ipm0.temp feeds the derate (conf0.high_ipm_temp) and the IPM_TEMP_ERROR trip (conf0.max_ipm_temp) in fault. The defaults are for the IKCM30F60GD fitted to v5.0 (conf/template/ipm_ikcm30f60gd.txt repeats them); link conf/template/ipm_im06b50gc1.txt after the motor template for the IM06B50GC1.

Component Explanation #

Everything runs in rt. There is no rt_start, so the die temperatures survive a disable (the module does not cool because the drive stopped).

1. Leg currents and duties:

• The angle is read from the pin that finally drives pos: ipm0.pos = hv0.pos links an input to an input, so the component follows the link chain (up to 4 hops) to the driving pin, e.g. vel2.pos_out.
• id, iq, ud, uq are turned into the three phase currents and voltages with an amplitude invariant inverse Park and Clarke transform at that angle.
• Leg duty = CLAMP(0.5 + u_phase / dc_volt, 0, 1); the common mode of the modulation is ignored. dc_volt is taken as at least 1 V.

2. Losses per die (after the CIPOS Mini application note):

• Current out of the leg flows through the high IGBT for the duty and the low diode for the rest; current into the leg through the low IGBT for the rest and the high diode for the duty. Dies not carrying the current get no loss.
• Conduction: duty_share * kvt * (v0 * |i| + r0 * i^2) with vi/ri for the IGBT and vd/rd for the diode, kvt = 1 + v_tc * (Tj - 25).
• Switching: e * f_sw * dc_volt / e_volt * |i| * ket with ei (IGBT Eon + Eoff per amp) or ed (diode Erec per amp), ket = 1 + ei_tc (or ed_tc) * (Tj - 25). Both temperature factors are clamped to 0.5..3 and use each die’s own last junction temperature.

3. Thermal model:

• Each die has a three pole Foster network, one first order lag per pole with the shared time constants tau1..tau3 and the splits fi1, fi2 (IGBT) or fd1, fd2 (diode). The static rise is P * rth_i or P * rth_d. Time constants shorter than the rt period are clamped to the period.

Tj = case_temp + sum over k of lag(P * rth * f_k, tau_k)

• At speed the networks average over the electrical cycle; at standstill the leg carrying DC heats on its own, which a mean value model over a sine would understate about 3.3x.

rth_i and rth_d are the datasheet Rth(J-C) times a factor that makes the model reproduce the module’s SOA chart (1.5 for the IKCM30F60GD, 3.6 for the IM06B50GC1). They do not include the heatsink, which is covered by case_temp from the NTC; trim them against the real hardware. pos is used as linked, without hv0.adv.

Pins #

References #

Source: .c File