What options do you have for controlling BLDC motors?

Introduction

Brushless motors or electronically commuted motors can be seen as DC motors, where the mechanical commutator with brushes is replaced with an electronic switching circuit (servo system).

Sensor-based commutation

For sensor-based commutation, a dedicated sensor to measure the rotor position is used. The advantage of this approach is that it is simple to obtain the motor position. The disadvantage is that a sensor is needed, which adds additional costs, additional volume, and weight to the motor and provides an additional point of failure. The two sensor types that are typically used, are hall effect sensors and encoders.

Hall sensors have two states only and measure the presence of the magnetic field of the permanent magnet rotor. Therefore, they are relatively cheap. Typically, three hall sensors are used – giving a 60° resolution of the rotor position. This is sufficient for the so-called “trapezoidal” or block commutation. In combination with a “trapezoidal-wound” motor, high efficiencies under low cost can be achieved.

Encoders on the other side typically give a high resolution of the rotor position – e.g. 1° or even far lower. Hence, encoders are more expensive and are often only used for positioning applications where the position information in such a high resolution is needed anyway. With encoders more sophisticated control methods such as field-oriented control (FOC) are also enabled, which provides advantages for sinusoidal wound motors.

Glance on implementation details

Typically, the hall edges align with the commutation instants. Therefore, the commutation happens in the same instant as a hall sensor edge occurs. For this purpose, the main software approach is interrupts, whereby each rising hall edge triggers an interrupt. In the interrupt service routine, a lookup table is evaluated that directly leads to commutation. For high-speed applications, this can become very challenging for the processor. For such applications hardware support, e.g. an FPGA – is recommended. For low-speed applications polling the hall sensor state with a fixed sampling rate can also be considered.

Advantages/Disadvantages

+ Simple and easy to implement

+ Motor startup is easy as rotor position is always known

+ Can be used for all speeds

+ Good for high dynamic operations under different loads

– Sensors need volume/mass/costs and provide a point of failure

Sensorless

If no dedicated position sensor is used, the rotor position is deduced from the phase terminal voltages and/or currents. In the following sections, the three methods back EMF measurement, observer-based methods, and variable inductance method are presented.

Back EMF sensing

When trapezoidal control is used, two phases are always connected to a defined potential, and one phase is floating. On the floating phase, the induced voltage by the rotor magnet (often called “back EMF”) can be measured. Simply stated, the zero crossing of the measured back EMF corresponds to the magnetized north pole of the rotor passing the winding. By sensing this zero-crossing therefore the motor position can be obtained. The resolution for this method is also 60°, which is sufficient for trapezoidal control.

There are a few issues that must be corrected when this method is used. First, the back EMF signal is speed-dependent. Therefore for low speeds, it is difficult or almost impossible to sense. Moreover, without any motor movement, the back EMF is zero. Thus, for startup and low-speed applications, a different strategy must be used. Second, the back EMF signal sensing is coupled to the PWM frequency and PWM method and thus creates a system engineering constraint. If a bipolar PWM is used, the zero crossing is easier to detect as the floating phase is nearly free of switching artifacts. If a unipolar PWM is used, the zero-crossing signal is affected by PWM artifacts. Filtering can be used, or the back EMF sensing must be accordingly timed in synch with the PWM. This can become challenging if high PWM frequencies are used. Moreover, bipolar PWM is not desirable for modern low-inductance motors as it creates a higher current ripple.

Glance on implementation details

Typically, the measured voltage needs to be filtered because of noise and PWM artifacts. The filter-time constants should be kept as low as possible to avoid delay and accurately detect the zero crossings. Moreover, directly after commutation, the back EMF should not be measured as the body diodes of the inverter are still conducting. Thus, for a short period called “blanking time” the zero-crossing detection should be disabled. Also, for back EMF sensing the motor neutral point voltage is needed and normally not measured. Different approaches like – creating a virtual neutral with physical resistors or by filtering and averaging in software – can be used. Finally, as the zero crossing points do not correspond to the actual commutation point, the commutation point needs to be predicted. This can be done by setting hardware timers or by integrating the back EMF, which is independent of the motor speed. A third option would be the usage of phase-locked loops (PLL).

Advantages/Disadvantages

+ No sensor needed

+ Simple and reliable method for appropriate speed range

+ In contrast to the following sensorless methods: no current sensing is necessary

– Does not work at low speeds/standstill. A startup procedure and a minimum speed are required

– Phase voltage measurements are needed

– Limited to trapezoidal commutation

– Commutation instant not at same time instant as zero-crossing and detection jitter. Difficult for high dynamic applications with changing loads

– Not good for very high speeds / high loads as the zero-detection time window decreases and measurement delays start to dominate.

Observer-based methods

These methods require a mathematical model of the motor, describing the relation of the phase terminal voltages \( u_{a,b,c} \) to the system state \( x \) (e.g., position and speed) and the phase currents \( i_{a,b,c} \). Typically, models of the following form are used

\( u_x = Ri_x + L \frac{di_x}{dt} + e_x \)

Whereby the back EMF is modelled as a constant disturbance \( \frac{de_x}{dt} = 0 \), or a waveform model is assumed: \( e_x (0) a f (0) \). Using observer theory, the system state can be reconstructed from the measured phase currents and voltages. To observe the system state, it is necessary to measure the phase currents and the phase voltages. The rotor position can then either be inferred from the zero-crossings of the observed back EMV voltages or directly from the estimated rotor angle.

Typically, observers that are used for BLDC control are the Kalman Filter and Extended Kalman Filter. Also, Sliding-Mode observers are used as they are more robust to model inaccuracies.

Advantages/Disadvantages

+ No sensor needed

+ Allows field-oriented control

+ Can be used for high-dynamic applications

– High computational cost

– Phase currents must be measured

– Does not work at a standstill / minimum speed required

– Modelling and engineering effort

– Dependent on motor parameters

Magnetic anisotropy method / variable inductance sensing

The inductance of the individual phases is dependent on the rotor position as the “magnetic path” through the motor changes with the rotor position. By injecting pulses into the windings, the phase inductance can be sensed, and the rotor position can be calculated. This method is especially valuable at very low speeds, where the back EMF sensing and the observer-based methods fail.