BLDC motor control

[SPIGUY1]

We have a 24V BLDC motor that requires low cogging and torque ripple. We designed a motor with a 15 slot/4 pole configuration with the rotor magnets skewed by one slot (approx. 6 degrees). After testing two samples with a marked reduction in ripple, we made 25 more and shipped them to our customer. The customer is reporting that the cogging and torque ripple are now worst than before when we used a 12 slot/4 pole design. We are also now seeing a difference in ripple based on motor rotation. I think the difference in torque ripple based on direction should be a result of some sort of asymmetry in how we are driving the rotor. The asymmetry could be a result of our encoder alignment or of some sort of offset in the output of the control.

We do not do a mechanical alignment of the halls as one might traditionally do. We actually do a so called magnetic alignment. We energize one coil which places the rotor in a known location. While the rotor is locked in place, we zero the encoder which sets the location of the “hall effect” sensors for trapezoidal commutation. Theoretically, this type of alignment method should be much more accurate than a mechanical or designed in alignment. This is true because we are aligning to the magnetic fields rather than some mechanical location on a component. This procedure should produce a very accurate alignment of hall effects to coil location but we still must verify that this is correct.

My question is this. When the control switches from trapezoidal to sinusoidal modes, is the alignment of sinusoidal outputs based on a trigger from one of the hall effects? If so, which one?

 

[TurbineGen]

You can eliminate much of the cogging torque and ripple by choosing an N/P (stator pole/magnet) configuration that is very close. I tend to prefer the 12N14P configuration or the 12N10P, but I find it odd that the 15N/4P has a torque ripple issue with the slots skewed.

That said, I'm not sure that the sinusoidal controllers reduce the torque ripple in the odd N/P configurations. They certainly help in the 3/2 and 3/4 cases such as in your 12/4 design.

Is your customer using the same controller you are testing with? How clean is your customer's power supply compared to yours?



------------------------------------------------------------------------
If it is broken, fix it. If it isn't broken, I'll soon fix that.

 

[cswilson]

How -- and whether -- the controller switches from "trapezoidal" commutation to sinusoidal commutation depends totally on the controller and the strategy employed, and often on the motor as well.

The most common case for switching commutation strategies occurs when you are controlling a sinusoidally wound motor with an incremental quadrature encoder and Hall-style commutation sensor.

At power-up, the control's knowledge of the absolute rotor angle comes from the coarse Hall sensors, which only split the commutation cycle into 6 states. This means that the measurement error could be +/-30 degrees (electrical). Many controls will not try to employ sinusoidal commutation at this point off the high-resolution quadrature feedback, and just use the switching "six-step" commutation typically employed for trapezoidally wound motors.

However, when these controls get a better sense of the rotor angle, they switch over to a sinusoidal commutation strategy. Many controls do this as soon as they see the first Hall edge (whichever of the 6 edges it is). Some will wait until they see the once-around index pulse.

In your case, you are not doing this approximate then accurate phasing, so there is not the need for this kind of switchover. However, I do have some issues with your strategy and your assertions about it.

You are establishing your phase reference by forcing current into a phase and letting the motor settle. (We call this the "stepper-motor phasing search" because at this point you are driving the motor like a stepper motor.) You believe this is more accurate than using a Hall-sensor edge, but I'm not so sure this is true with a loaded motor.

We have many users who use this method on an unloaded motor to mechanically align the Hall sensors (or to calculate the offset of the Hall sensor as is). However, in actual use with a loaded motor, they will use the Hall sensors -- first just state, and then edge, for more accurate phasing.

In many (most?) applications, your method on a loaded motor is subject to phasing error due to external loads. With iron core motors, the first thing you notice with small phasing errors is increased torque ripple (which can be different in opposite directions). Our field guys are very adamant that when someone uses your method to establish the initial phase angle, they should always do a sensor-based correction.

Another, fairly independent, issue has to do with your back EMF waveforms. I think you are saying you employ "trapezoidal commutation", which usually means a simple six-state switching commutation algorithm. Is your motor trapezoidally wound? That is, when you spin the motor up mechanically, are your phase back EMF waveforms substationally trapezoidal?

The lowest ripple comes from sinusoidal commutation on a sinusoidally wound motor (and minimal or no cogging torque). If you want to employ "six-step" commutation, you should use a trapezoidally wound motor (although you will never get perfectly trapezoidal windings, and you will get a perturbation every time you switch states).

Curt Wilson
Delta Tau Data Systems