Sensorless PMSM Field-Oriented Control

[General_motor_control]

Dear,

I am currently implementing sensorless control for PMSM using field-oriented control in combination with the AMCLIB_PMSMBemfObsrvDQ and the AMCLIB_TrackObsrv for the ARM Cortex M7F. I am using as a reference the DRM148 Rev.1, 02/2016. Looking at Fig. 23 a constant q-axis current is applied during start-up to accelerate the PMSM from standstill, where after the catch-up speed closed-loop operation is enabled. 

I implemented just this. However, referring back to Fig. 23 from the DRM148, the estimated observer speed is controlled near the open-loop predicted speed during start-up. In my situation I have a sensor mounted to the shaft which does the closed-loop control over the full speed range. In the background I have an observer running which does e.g. 20000 RPM, whereas the actual shaft does only e.g. 150 RPM. When accelarting the shaft to double the speed (300 RPM) the observer estimated mechanical speed raises to 40000 RPM, so there seems to be a linear correlation. I was expecting, based on Fig. 23 from the DRM148, that even when using a sensor the observer in the background would converge somewhere close to the true value. This is in no-load condition. When applying load to the motor shaft the estimated mechanical speed from the observer decreases rapidly. 

What are possible causes for this difference in RPM between the true sensor mechanical speed and the estimated observer mechanical speed when using sensored closed-loop control?

[RafaR]

Hello General_motor_control

Good day!

First of all, I apologize. I had to delete the reference manual you uploaded because to access it, you need to create an account, and uploading it to the community makes it completely public.

In DRM148 the observer is actively forced and corrected during startup.

In your implementation the observer is free‑running and decoupled, so it never converges.

In DRM148 (Fig. 23), during open‑loop startup:

• The angle and speed used by the FOC are generated by a predictor (open‑loop ramp)
• The BEMF observer is explicitly driven using this same angle/speed
• The tracking observer (AMCLIB_TrackObsrv) is used to lock the estimated speed to the open‑loop reference
• Only after convergence does control switch to observer feedback

Some reasosn of your issue conld be:

• Q, The speed is ramped with the constant acceleration while the Q current is kept at the defined level. When the catch-up speed is reached, the merging ratio is incremented with the merging coefficient.

Verify that your Q value remains fixed and there is no variance.

• Check the scaling; if any point is significantly affected, it can give you an indication that more voltage/current is being perceived at a certain point, which is causing this variation.

If you need more detailed assistance and would like to share more documentation with us, I recommend opening a case directly with us. This way, we won't have to delete any documents you provide.

I hope this information has helped you, please let me know if you need help with anything else.

Have a great day and best of luck.

[General_motor_control]

Dear @RafaR,

Thankyou for your reply. 

Based on your input I implemented the following. In the AMCLIB_PMSMBemfObsrvDQ I use the sensor speed output (converted to angular electrical rotor speed) as an input to the BEMF observer. Furthermore, the requested voltages (Udq*) and currents (Idq*) are used for the BEMF observer input(s). I convert the AMCLIB_TrackObsrv output electrical rotor speed to mechanical speed in RPM and visually compare this to the actual rotor/sensor mechanical shaft speed. This all happens in closed-loop manner (sensored). However, the AMCLIB_TrackObsrv output speed still does not converge on a long time frame (e.g. >10 seconds) while keeping the rotor speed constant (e.g. 150 RPM). 

Am i still overseeing something?