We were looking to flux estimator based on voltage model so far and improved performance by adding PLL structures. Still we are wondering if a sliding mode observer could further improve the behaviour in corner cases. What is your experience and is there other position estimation algorithms, we should have a look at? Any help is very appreciated.

Hi Arno. I'm looking for a company that might be interested in a great new sensorless motor controller that can be applied to PMSM and stepper motors. The new patent-pending controller has been developed at the University of Technology Sydney and offers many advantages over other sensorless techniques. If you are interested in hearing more let me know. Thanks.

Hi Martin, of course we are interested in looking to sensorless control techiques. Could you please share some more details? Thanks.

There is another method that needs to inject high frequency signal to sensor rotor position. It is similar with resolver theory. But it requires machine saliency, i.e. IPM.

Statistically compensated voltage model (SCVM) in IFO (dq syatem) is one of the methods for sensorless control.

I've implemented a motor drive using a PLL to track the phase and speed of the calculated BEMF signal to use for FOC control. After completion of the project, I think the PLL may not be the best solution. It works very well at higher velocities, but at low motor speeds the PLL bandwidth needs to be pushed very high resulting in large reference feedthrough. I can't say too much as it is a proprietary design. We managed to speed up the PLL at low velocities to achieve sensorless speeds down to 200RPM on a 3-pole motor but it was tricky.

I can say a PLL can be made to work on a low processing power platform (~$2 microcontroller / digital signal controller), but if I were to start the project again I would look more closely at the so-called slide mode controller before choosing the PLL.

Best regards,

- John

I think the various choices are also influenced by the application (high-speed vs. low-speed, general-purpose vs. known motor, fan/pump application vs. high-performance, ...).
For example, for a known motor and heavy low-speed operation, you're forced to try with HF techniques mentioned by Yichao, since they could be the only solution.

Quadrature PLL is an efficient way (one of the few, AFAIK) for tracking sinusoidal signals frequency and phase with low-pass filtering and no steady-state error.
I wouldn't expect too much from sliding-mode, you can see it as a non-linear feedback. It surely introduces chattering, which needs to be filtered, and it is also difficult to analytically deal with, being non-linear.

As Sandro said, quadrature PLL techniques work very well, and can be made to work at quite low speeds. I've implemented both sliding-mode and PLL-based techniques, and can't say that one is much better than the other. I think the PLL method offers more flexibility, but the sliding-mode observer might be easier to tune.

The trick with any PMSM speed-sensorless technique is the startup, of course.

Hi Arno, If your flux observer works well, and you know your motor's parameters (especially inductance) you can use the current model estimate the angle. I prefer the well-known Kalman or Luenberger observer (depends if you have stochastic info of the process or not), using prediction and innovation/update. You can get good results both in low and high velocity range. Working in the rotor d-q reference frame, the model is linear, thus a linear observer can be used (you could say that the transformation from the stator to the rotor is non-linear, but if you have small angle errors, then you'll see it's "quasi-linear"). using non-linear functions (compensators, observers etc.) is always tricky, since convergence/stability for all working points is hard to prove.

Thanks a lot for all your feedback. We are especially looking into low end drives like fans or pumps as well as aircon compressors. The MCU is a CORTEX-M0, but is equiped with a cordic co-processor.
We were looking into PLL structures, but are still in a phase of feasibility investigations. Sandro, could you please give some more details or links into these "quadrature PLL techniques"?
John, thanks for your input as well, can you give a bit more hints where to look into detail?

Best Regards,
Arno

Thanks for the insight Sandro. I wasn't aware of that technique. Did you consider what happens if the armature reaction is large (then both stator and air-gap flux linkage vectors have different angles) in sense of the PLL ?

Sorry, I'm not sure I understand your question... PLL is fed with estimated back-EMF alpha-beta signals, so it is care of the observer making those signals a good estimate of the PM induced voltage, that is always 90 deg leading the PM flux.

Sandro,you're definitely right about the bemf of the PM, but what about the following term - j*w*L*i (subscript are missing...) - which is the armature reaction bemf in a non-salient motor. this term, depends on the current vector and especially on the inductance. As you can see, there could be situations that this term has a component along the bemf of the PM. This is what my question was about - since I didn't see that the PLL is taking care of this (however, I read it quite quickly so maybe I missed something).

By the way, it's a pity that I cannot paste a picture here (maybe there is some way ?) , as you could see the situation very well in the space phasor diagram...

The back-EMF observer (or estimator) does nothing but separating the voltage induced by the PM rotation from the whole phase voltage:
E = V - R*I - s*L*I (Laplace domain, stationary reference frame)

BTW, yes, it would be nice if there was a way to post images!