Peru Mendivil · Mondragon Unibertsitatea, 2025-26
The hub motor and the IFX007T inverter were given. I designed and sized the measurement boards and did the control between them on an STM32: two PI controllers in cascade, a back-EMF feed-forward, a low-pass filter on the measured current and a speed estimate from the Hall sensors.
| Item | Description | By |
|---|---|---|
| 1 | Magnets, 30 (15 pole pairs) | given, hub motor |
| 2 | Hall sensors, 3 (90 edges per turn) | given, hub motor |
| 3 | Stator winding, 3-phase | given, hub motor |
| 4 | Axle | given, hub motor |
| 5 | IFX007T inverter shield, 10 kHz PWM | given |
| 6 | Measurement boards: voltages, phase currents, Hall signals | Peru Mendivil |
| 7 | Motor characterisation: R, L, τ, Kem, damping | Peru Mendivil |
| 8 | Current and speed control on the STM32 | Peru Mendivil |
| 9 | Battery charger | other team members |
Parts list Items 1 to 4 as marked on Section A-A
The current board carries the Hall-effect current sensors, resistor voltage dividers and op-amp buffers, soldered by hand on perfboard. To reduce electromagnetic noise I used twisted wiring, minimum loop areas between signal and ground, and a separate ground for each Hall output.
For R and L the rotor is locked, so there is no back-EMF. A voltage step then gives a first-order current response, and the time constant is the time to reach 63.2 % of the final value. I measured τ = 3.36 ms on the scope, and L = R·τ. The same test on the Simulink motor gave 3.37 ms.
With Ti = L/R the PI zero cancels the motor pole and the closed loop is first order. I chose τcl = 1 ms, ten times the period of the 10 kHz inverter, which gives Kp = L/τcl = 1.06.
| Rs | 0.3157 Ω | 0.483 V / 1.02 A, rotor locked |
|---|---|---|
| Ls | 1.0607 mH | L = R·τ |
| Kem | 0.0318 V·s/rad | 4.8 V back-EMF at 24 Hz |
| Rev | Change |
|---|---|
| A | The control first ran at 10 kHz. The speed loop measured time by counting the calls to the control function, and that count fluctuated and dropped as the scooter went faster: the STM32 could not finish the calculations in 0.1 ms. We checked it with the professor and reduced the control rate to 2 kHz (0.5 ms). |
With that sampling, the current loop overshoots above Kp ≈ 2 and becomes unstable above Kp ≈ 3.9. Scroll, or move the gain on the scale.
Unstable above Kp ≈ 3.9 with Ts = 0.5 ms
Rev B On the scooter I kept Kp = 1.06 and made Ti twenty times larger (67 ms). The response is slower, but more stable with the noise in the current measurement.
The phase current measured with the Hall sensors still had noise, so I added a first-order low-pass filter on the measurement. The filter adds phase lag inside the loop. Without it the phase margin is 90°. I chose the lowest cut-off that kept an acceptable margin: 250 rad/s, with 28° at 470 rad/s. In the tests, a lower cut-off made the loop unstable.
There is no encoder. The three Hall sensors give 6 edges per electrical cycle, so with 15 pole pairs there are 90 edges per turn. At each edge the STM32 calculates ω = (2π/90)/Δt and keeps the value until the next edge. At 200 rpm there is a new value every 3.3 ms; at 20 rpm, only every 33 ms.
The speed controller is a PI with acceleration feed-forward and anti-windup (Kp = 0.0177 A/rpm, Ti = 2.0 s). Its output is the current reference of the current loop.
In the complete Simulink model, the current follows a 5 A step within a few milliseconds and recovers after each commutation. The speed reaches 200 rpm in about 0.6 s.
On the scooter, the drive was tested on the floor: it could be driven, accelerated and braked correctly.
In Västra Götaland, January to June 2027. I'm doing an Erasmus+ double degree between Mondragon Unibertsitatea and the University of Skövde.
Data: annex E of the project report, Mondragon Unibertsitatea, 2025-26. Charts on this page are computed in the browser from the motor model with the measured parameters. Generated images are marked as such.
