IET Electric Power Applications

Special Issue: Advances in Predictive Control of Variable-Speed Electric Drives

Unipolar PWM predictive current-mode ISSN 1751-8660

Received on 6th July 2016
Revised on 7th December 2016
control of a variable-speed low inductance Accepted on 16th December 2016
doi: 10.1049/iet-epa.2016.0421
BLDC motor drive www.ietdl.org

Rodolfo L. Valle 1,2 ✉, Pedro M. de Almeida 1, Andre A. Ferreira 1, Pedro G. Barbosa 1

1
Electrical Engineering Department, Federal University of Juiz de Fora, Juiz de Fora, Brazil
2
Electronics Department, Federal Centre of Technological Education of Minas Gerais, Leopoldina, Brazil
✉ E-mail: [email protected]

Abstract: This study presents the design and implementation steps of a digital predictive controller to regulate a low-
inductance, three-phase, three-wire permanent magnet brushless DC motor currents. These types of motors are usually
driven by multi-stage converters, switched at high frequencies, or use additional inductances to limit the current ripple.
The motor’s trapezoidal back electromotive force and rectangular currents waveforms make the design and the tuning
process of linear controllers difficult. This task complexity increases when a wide speed range is considered. Digital
predictive controllers are easily implemented in digital signal processors (DSPs), being successfully used to regulate
currents of different types of power electronic converters. A unipolar pulse width modulation predictive controller is
proposed here to regulate the rectangular currents of a brushless DC motor, without the need for any additional filter
or converter. Experimental and simulation results using a 5 kW/48 V three-phase brushless DC (BLDC) motor are
presented to demonstrate the feasibility of this proposal. It will be presented a methodology to compensate the
conditioning and sampling circuits delays as well as the inverter’s semiconductors voltage drop. The control algorithm
was implemented in a TMS320F28335 DSP.

1 Introduction In a similar manner, the low motor’s inductance values also make
the current controllers design for the PMSM and BLDC motors
Permanent magnet synchronous motors (PMSM) and brushless DC challenging. Different types of controllers have been proposed in
(BLDC) motors have been widely used as traction engine for electric the literature where the aim is to design regulators with high
vehicle (EV) and hybrid electric vehicle [1]. These motors are named cut-off frequencies to assure small errors over the wide range of
according to their back electromotive force (EMF) waveforms. the motor speed variation [8].
Regardless their lower volume and weight, which are interesting The successful model predictive control (MPC) applications in
features for embedded applications, they present low values of processing industries associated with high-performance DSPs
inductances, resulting in large current and torque ripples [2, 3]. development have encouraged researchers and engineers to apply
Different converter topologies have been investigated in the MPC to control more complex and fast dynamic systems [9].
literature to drive low inductance motors since they must be Rodriguez and Cortes [10] have demonstrated the MPC algorithms
switched with high frequencies. In [4, 5], the authors use a effectiveness when controlling different types of power electronics
two-stage drive, comprised of a buck-converter and a voltage converters and plants. As the semiconductor switches are turned
source inverter, to control a 36 mH BLDC motor. A large inductor on and off many times per cycle, it is possible to choose a
limits the buck-converter current ripple while the voltage source combination of active switches to optimise a cost function or to
inverter (VSI) is switched with a 200 kHz to regulate the motor force the error, between the reference and controlled variables, to
torque. A similar structure, also based on two-stage converter, was go to zero in a finite number of sampling periods.
used in [6] to control a 3.4 mH inductance BLDC motor. Despite In [11–14], predictive current control algorithms are applied
the limitation of the current through the semiconductor switches, to different motor drives. They can be grouped into: (i)
the series inductor increases the cost, volume and weight of the two-configuration, (ii) direct and (iii) pulse width modulation (PWM)
drive. Besides the reduction of the current ripple and the precise predictive techniques [13]. The two-configuration predictive control
torque control, the high switching frequency makes the use of uses a modulator to apply two different inverter branches
multiple control hardware (field programmable gate array and configurations during the commutation period minimising the current
digital signal processor, DSP) necessary, adds complexity to the error. The direct predictive controller estimates the current future
controller’s implementation and does not allow the use of value for each inverter’s output voltage vector. Then, a cost function
insulated gate bipolar transistors (IGBTs). is used to determine the voltage vector which minimises the current
In [7], a 170 mH inductance BLDC motor is driven by a error [14]. The last type is the PWM predictive control (PPC). It
three-level inverter. Multilevel converters allow the increase of the calculates, at the present time, the necessary change in the control
DC voltage level since four MOSFETs are series connected in signal that makes the current track its reference at the next sample [13].
each VSI branch. At the same time, the use of fast switches, like The PPC has the advantage of not requiring an optimisation
MOSFET, permits high switching frequency, up to 150 kHz. In algorithm. Therefore, it is less computationally complex and
spite of the lower current ripple, the DC-link midpoint voltage exhibits the fastest dynamic response among aforementioned
control is one of the biggest challenges when implementing a predictive controllers. Based on the previous considerations, the
three-level converter. Several non-ideal conditions, such as time PWM predictive is an interesting and promising option to regulate
delays, dead-time, among others, make the midpoint DC-link the currents of low inductance BLDC motors.
voltage control too complicated for low inductance motors On the other hand, the PPC main drawbacks are: (i) steady-state
applications. errors; (ii) effectiveness compromised under parametric

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
688 & The Institution of Engineering and Technology 2017
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
uncertainties, as well as, delays and non-linearities if not taken into
account during the controller design [15]. These constraints may
restrict the PCC application. In [16], the authors added an integral
action to eliminate the current error. However, it changes the
BLDC motor dynamical behaviour reducing the compensated
system stability margins.
This work main contribution is the design of a fast current
predictive control applied to a 5 kW/48 V three-phase three-wire
brushless DC motor with a low winding inductance (14.8 mH).
The control algorithm is implemented without the cost function
minimisation to ensure a fast dynamic for the controller. A simple
methodology, based on the current waveform behaviour, is
proposed to compensate the drive steady-state error without the
need of additional controller. It is used a single-stage two-level
IGBT-inverter to drive the BLDC motor. The 2p/3 rad conduction
mode of the BLDC motor allows to control the VSI as a full
bridge DC–DC converter, increasing the equivalent switching
frequency up to 100 kHz, while each IGBT is commutated with
50 kHz, reducing the current ripple without the need of connection
of any additional series inductance or converter stage.
This paper is organised as follows: Section 2 presents the BLDC
drive analysis showing that the current flows only between two
phases of the motor at every time period. This feature allows to
control the three-phase BLDC drive as a full-bridge DC–DC
converter, and the use of an unipolar PWM strategy to increase its
output equivalent switching frequency. Section 3 addresses the
design steps of the proposed unipolar PWM predictive current
controller. Initially, the current controller is designed to track the
changes in the reference signal in one sampling period. However, Fig. 1 Variable speed BLDC drive
due to the computational limitations, the algorithm is applied in two a Simplified schematic diagram
b Back EMF (continuous lines) and currents (dashed lines) waveforms
consecutive sampling periods to overcome the hardware constraints.
In Section 4, it is analysed how the uncertainties and variations of
the motor parameters affect the system absolute stability. The
theoretical analyses are validated by experimental results. Section 5 current [20]. In the bipolar PWM pattern, the modulating signal is
describes the methodology to compensate the delays and compared with an up-down counter to generate the PWM signals,
non-linearities of the motor drive in the unipolar PPC algorithm. whereas the output voltage is switched between +Vdc and −Vdc .
Section 6 presents the laboratory prototype. Experimental results are On the other hand, the unipolar PWM pattern uses two triangular
used to demonstrate the feasibility of the proposed control strategy. carrier waves, phase shifted by p rad between them, to generate the
Section 7 presents some concluding remarks. PWM signals S1 and S3 as shown in Figs. 2a and b, respectively. The
output voltage vab is switched between 0 and +Vdc with a frequency
two times higher than the frequency of the triangular carrier wave
(Fig. 2c). Fig. 2d depicts the ‘pseudo’ current waveform.
2 The BLDC motor drive It is important to emphasise that the operation strategy described
in this section allows the control of the three-phase BLDC drive as
Fig. 1a shows the schematic diagram of the BLDC motor drive based a full-bridge DC–DC converter, increasing the inverter output
on a three-phase VSI. The voltages ean , ebn and ecn represent the equivalent switching frequency by using an unipolar PWM
phase-to-neutral back EMF while R and L are the resistance and pattern. Consequently, it is also possible to employ different types
inductance of the BLDC phase winding, respectively. This system of predictive controllers, which were originally developed for DC–
is intended to be embedded on a small-scale EV. In Fig. 1a, DC converters. This characteristic is not a disadvantage, but in
coloured lines indicate the phases which are conducting a non-zero fact, an interesting and promising field to be explored.
current during the operation period depicted inside Fig. 1b The technique presented here can also be easily applied to any
highlighted rectangle. multiphase BLDC motors in which the conduction mode is
At a first glance, the back EMF trapezoidal waveform shown in 2p/P rad, where P is the number of the motor phases. In this case,
Fig. 1b can be viewed as a disadvantage. However, the power the inverter (P − 1) branches may be controlled as (P − 1)/2
density and torque increase ∼15% for the same current level. This full-bridge DC–DC converters. In the next section, it will be
feature also simplifies the position sensors placement in the shown how the waveforms of Fig. 2, as well as the duty cycle
machine stator since it will be only necessary to identify six d[k], can be used to derive a predictive control law to regulate the
different positions of the BLDC rotor [2]. the BLDC motor currents.
Another advantage is related to the BLDC motor drive controller
design. Considering that, during each p/3 rad, the current flows only
by two phases of the motor, it is not necessary to design three but
only one current controller to regulate a ‘pseudo’ current
3 Unipolar PWM predictive current controller
calculated by design

The BLDC drive current control must quickly respond to changes on

1

the motor load assuring an acceptable torque ripple. Therefore,
Ip = |i | + |ib | + |ic | , (1)
2 a the predictive control is a suitable technique to meet the
aforementioned requirements.
where Ip is the magnitude of the ‘pseudo’ current synthesised by the Due to the fact that the operation sector shift does not affect the
VSI [17–19]. ‘pseudo’ current behaviour, the control strategy development will
Notwithstanding, having in mind the operation of the motor inside only address the operation within the highlighted rectangle
the highlighted sector of Fig. 1b, it is possible to use two different depicted in Fig. 1b. Therefore, neglecting the resistance of the
PWM switching strategies to control the VSI output voltage and motor winding (R ≪ 2pfs L) and taking into consideration the

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
& The Institution of Engineering and Technology 2017 689
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
It was considered that the mechanical time constant is greater than
Ts in (3) and (4). Therefore, the back EMF value, which depends on
the shaft speed, was assumed constant during a switching period.
Substituting (3) and (4) into (2), isolating d[k] and making
Ip [k + 1] = Ip∗ yields

1
   
Lfs E
d [k ] = + Ip∗ − Ip [k ] + (5)
2 Vdc 2Vdc

where Ip∗ is the reference current, d[k] is the duty cycle for the
unipolar PWM pattern between the intervals kTs and (k + 1)Ts .
The duty cycle range is 0 ≤ d[k] ≤ 1. However, there are some
operation modes, for example the regenerative breaking, where the
control variable may assume negative values. Therefore, making
d [k ] = (m[k] + 1)/2, it is possible to rewrite (5) as follows

2Lfs  ∗
   
E
m[ k ] = Ip − Ip [k ] + (6)
Vdc Vdc

where −1 ≤ m[k] ≤ 1 is the modulation index for the unipolar

PWM signal generator.
The modulation index m[k] is the control signal calculated by the
PWM predictive controller in the instant kTs that is sent to the VSI
modulator to force the ‘pseudo’ current Ip [k ] track the reference
current Ip∗ in the next sample period (k + 1)Ts .
Sometimes, due practical limitations, it is not possible to sample,
execute the algorithm and act as fast as shown in Fig. 2. To overcome
this constraint, (6) is applied to two consecutive periods resulting in
the following control law

2Lfs  ∗ 2E
m[k + 1] = I − Ip [k] − m[k] + (7)
Vdc p Vdc

where Ip∗ = Ip [k + 2] is the reference current and m[k + 1] is the

modulation index for the period between (k + 1)Ts and (k + 2)Ts .
That is, (7) returns the modulation index for the instant (k + 1)Ts
to force the VSI output current to track its reference after two
Fig. 2 Unipolar PWM pattern waveforms samples.
a Modulating signal and triangular carrier waves One constraint of the proposed methodology takes place when the
b PWM signals (S1 and S3 ) motor resistance is not negligible. In this case, the predictive current
c VSI output voltage (vab )
d Synthesised ‘pseudo’ current (Ip )
law given by (7) lose its effectiveness and the design steps presented
in this section should be revalidated considering the effect of this
parameter.
output voltage and current waveforms shown in Figs. 2c and d, the
following relation can be written for the ‘pseudo’ current
4 Absolute stability analysis
Ip [k +1] According to the previous developments, the predictive control

T T T
 (2) algorithm regulates the BLDC currents in such a way that the
=Ip [k ]+2 r2 1−d [k ] s +r1 2d [k ]−1 s +r2 1−d [k ] s reference signal is tracked after two samples. However, the




2 2 2 uncertainties or variations on the motor parameters may
compromise the algorithm effectiveness given in (7). Neglecting
the harmonics generated by the switched operation of the VSI, the


where 2d [k ]−1  period between kTs and (k +1)Ts in which
is the
following dynamic equation can be written for the active branches


vab is +Vdc ; 1−d [k ] is the period between kTs and (k +1)Ts in
which vab is 0; Ts is the sampled period; k = 0,1,2,3...; r1 and r2 shown in Fig. 1a
are, respectively, the ‘pseudo’ current rise and fall rates, which are
given by d 1

   
R 
I (t ) = − I (t ) + vab (t) − eab (t ) (8)
dt p L p 2L
vab [k]−eab [k] +Vdc −E
r1 = = (3)
2L 2L where Ip (t ) = ia (t ) and ib (t ) = −ia (t ).
Applying the zero-order hold (ZOH) [21] to discretize (8), the
and following state-space representation is obtained


Ip [k + 1] = FIp [k ] + G vab [k ] − eab [k ] . (9)
v [k]−eab [k] E
r2 = ab =− (4)
2L 2L
where
where
L is the  motor per phase inductance and
eab [k]= ean [k]−ebn [k] =E. F = eATs = e−(R/L)Ts (10)

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
690 & The Institution of Engineering and Technology 2017
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
and case of state feedback controller in which the compensated poles are
placed in the origin of the complex z-plane.
tk+1
1 − e−(R/L)Ts The Jury’s stability criterion is very useful to determine whether
G= eA(tk+1 −t) B dt = . (11) the uncertainties on Lc affect the compensated system pole
tk 2R
location by moving it to the instability region [21]. Thus, to
guarantee that the poles of (14) are inside the unit circle, the
Substituting vab [k ] = m[k ]Vdc and eab [k ] = E into (9) and applying follow inequality must remain true
the Z-transform in the resulting difference equation yields

1 1 − e−(R/L)Ts
1 + e−(RTs /L)

Ip (z) =

M (z)Vdc − E . (12) 0 , Lc , RTs . (15)
2R z − e−(R/L)Ts 1 − e−(RTs /L)

In the same way, applying the Z-transform in the difference equation The inequality in (15) is reduced to 0 , Lc , 2L when R  0. This
of the PWM predictive controller given by (7) returns condition implies that the compensated system will be unstable if
the parameter Lc exceeds twice the value of the motor inductance
1 2Lc fs  ∗ 2Ec L. The discrete closed-looped function in (14) can also be used to
   
M (z) = Ip (z) − Ip (z) + . (13) investigate how the motor parameters R and L affect the
(z + 1 ) Vdc Vdc
compensated system poles location.
The subscript ‘c’ associated to Lc and Ec in (13) was introduced here On the other hand, parametric errors due to measurement or
to differentiate the estimated parameters used in the control law from variations on the motor operation point may lead the designer to
the real ones. pick wrong Lc values when tuning the predictive controller. Figs. 3b
From (12) and (13), it is possible to draw the control block and c show the locus of the poles considering the variations on the
diagram depicted in Fig. 3a. Assuming the BLDC voltage E is motor resistance and inductance, respectively. As can be seen the
effectively compensated by the feed-forward signal Ec , the poles do not remain exactly at the origin of the complex z-plane. By
following closed-loop transfer function can be written the analysis of the aforementioned figures, it is possible to conclude
that, variations of the resistance does not affect significantly the
Ip (z) (Lc /RTs ) 1 − e−(RTs /L)

 poles location, while changes in L can make the system unstable.
= . From Fig. 3c, considering L , Lc , the compensated system will
Ip∗ (z) z2 + z 1 − e−(RTs /L) + (Lc /RTs ) 1 − e−(RTs /L) − e−(RTs /L)



present an underdamped characteristic until L = Lc /2, when it will
(14) become unstable. In contrast, for L . Lc , the compensated system
will show an overdamped characteristic.
An interesting condition to be investigated takes place at the end
Considering
Lc = L and R  0, the discrete transfer function (14) is of the highlighted rectangle of Fig. 1b. During a short overlapping
reduced to 1/z2 . In this way, the PPC can be viewed as a discrete period, the three phases of the VSI are active and the equivalent

Fig. 3 Compensated system poles locus under parametric variation

a Variable speed BLDC drive (PPC and BLDC) block diagram
b Assuming Lc = L and varying R
c Assuming R = 0 and varying L

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
& The Institution of Engineering and Technology 2017 691
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 4 Behaviour of the ‘pseudo’ current for a step change on the reference signal Ip∗ from 15 to 20 A
a Lc = L
b Lc = 1.5L
c Lc = 0.5L

phase inductance is reduced to (3/4) L. Thus, replacing L by (3/4) L Although this condition happens six times per cycle, the poles
in (12) and substituting Lc = L into √ (13), the poles of the remain inside the unity circle and (14) stays stable.
compensated system move to +j(1/ 3), as shown in Fig. 3c. Figs. 4a–c show the experimental step response of the ‘pseudo’
current when the parameter Lc is chosen equal to L, 1.5L and
0.5L, respectively. The reference current signal Ip∗ is changed from
15 to 20 A. As predicted by the previous analysis, the choice of
Lc . L produces an underdamped current response (Fig. 4b),
while for Lc , L an overdamped current response is obtained
(Fig. 4c). Ideally, when Lc = L (Fig. 4a), the current tracks its
reference signal in two sampling periods.
All the results presented in this section were obtained using a
motor which parameters are R = 6.2 mV, L = 14.8 mH and
Krpm = 0.0125 V/rpm. The VSI switching frequency is 50 kHz
and the DC link voltage 48 V. More details about the variable
speed BLDC drive prototype are presented in Experimental results.

5 Driver parameters compensation

Despite the good dynamic performance and robustness of the
predictive controller, the currents synthesised by the VSI exhibit
steady-state errors due to non-linear effects such as blanking-time,
voltage drops on the semiconductor switches, delay time of pulse
drive and sampling delay time [13]. In the following sections is
presented how the aforementioned undesired effects can be easily
compensated in the PPC algorithm [22].

5.1 The blanking time

Fig. 5 Pulse drive delay effect on the ‘pseudo’ current synthesise by the VSI In practice, due to the finite turn-on and turn-off times, the active switch
a Inverter output voltage
of each VSI branch is turned off before the other is turned on. This time
b Real and delayed currents delay is chosen in a way to avoid a short-circuit or cross-conduction
c Triangular carrier current through the VSI branches [20]. As a consequence, the output

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
692 & The Institution of Engineering and Technology 2017
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
where TBT is the blanking time; DvxN is the inverter output voltage
increasing or decreasing factor, x [ {a, b, c} and N is the negative
terminal of the VSI.
Therefore, having in mind the unipolar PWM switching pattern,
the rates r1 and r2 can be recalculated substituting vab [k ] by
(Vdc − Dvab [k]) and −Dvab [k ] into (3) and (4), respectively, to
correct the
predictive control  law given by (7), where
Dvab [k ] = DvaN [k ] − DvbN [k ] .

5.2 The delay time of pulse drive

Figs. 5a–c show the pulse drive delay time effect on the VSI output
voltage and current. The current Ip [k ] is sampled every time the main
triangular carrier waveform reaches its peak value as shown in
Fig. 5c. This choice provides a good noise immunity on the
current sensor signal since the average value of the ‘pseudo’
current is sampled without the need of any passive filter.
Comparing the drawings (dashed and solid lines) are shown in
Fig. 5, it is possible to conclude that the delay time will cause an
offset error on the ‘pseudo’ current. A possible solution is to
calculate a correction factor for the sampled current as follows

DI ′p = r2 TDD (17)

where TDD is the delay time of pulse drive; r2 is the current fall rate
of Ip [k ].
Since r2 is negative, the addition of the factor DI ′p in the sample
Ip [k ] current will force the predictive control law (7) to correct the
index m[k + 1] to compensate the undesired offset on the
synthesised current.

5.3 The sampling delay time

The delay time of the sampling and the conditioning circuits produce
an offset error on the sampled current similar to that shown in
Fig. 5b. As in the previous case, this delay time can be
compensated by adding a factor in the sample current Ip [k ] given by

DI ′′p = r2 TSD (18)

where TSD is the delay time of the sampling and conditioning

circuits.
Fig. 6 Experimental setup
a BLDC current controller block diagram
b Picture of the BLDC motor and load (DC generator)
c Picture of the three-phase inverter and auxiliary circuitries 5.4 Voltage drop of the semiconductor switches

Similarly to blanking-time, the voltage drop of the semiconductor

switches reduces the VSI output voltage. The polarity and the
voltage of each VSI branch is reduced or increased, depending on the value of the voltage drop will depend on the direction of the
direction of the current flow, by following factor current flow and on which switches are active.

 Considering the active branches depicted in Fig. 1a, it is possible
+
TBT /Ts Vdc , ix . 0 to calculate new values for output voltage vab [k ] as
DvxN = (16)
− TBT /Ts Vdc , ix , 0

Ts (1 − d[k])/2 ≤ t , Ts d[k]/2,
⎧

⎪ +Vdc − 2VG ,
T (2 − d[k])/2 ≤ t , Ts (d[k] + 1)/2.
⎪
Table 1 BLDC motor characteristics
⎪
⎧ s
⎪
⎨
vab [k] = ⎨ 0 ≤ t , Ts (1 − d[k])/2,
Parameter Value
⎩ −2VG , T d[k]/2 ≤ t , Ts (2 − d[k])/2,
⎪
⎪
⎩ s
⎪
⎪
rated power 5 kW Ts (d[k] + 1)/2 ≤ t , Ts .
rated voltage 48 V
rated current 100 A
(19)
maximum efficiency 89.1%
resistance per phase 6.2 mΩ
inductance per phase 14.8 μH
where VG is the voltage drop in each active switch. To simplify the
Krpm 0.0125 V/rpm compensation factor, the voltage drop of all semiconductors switches
number of poles 8 is considered equals.
weight 11 kg Substituting (19) into (3) and (4), respectively, it is possible to
maximum torque 13.92 Nm
rated speed 3532 rpm
recompute the predictive control law taking into account the
voltage drops in the semiconductor switches.

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
& The Institution of Engineering and Technology 2017 693
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 7 Unipolar PWM three-phase currents of BLDC motor (Ip = 15 A, fs = 50 kHz)
a Steady-state PWM waveforms
b Harmonic spectrum of phase ‘a’ current

Fig. 8 Bipolar PWM three-phase currents of BLDC motor (Ip = 15 A, fs = 50 kHz)

a Steady-state PWM waveforms
b Harmonic spectrum of phase ‘a’ current

6 Experimental results synthesised current is 6 A. Figs. 8a and b show the current

waveforms and the phase ‘a’ harmonic spectrum, respectively,
Fig. 6a shows the block diagram of a variable speed BLDC motor when a bipolar PWM predictive controller is used [23]. Although
drive prototype built in the laboratory to demonstrate the both controllers synthesise currents with the same steady-state
feasibility of the proposed predictive control. In this figure, the value, the current ripple of the bipolar controller (Fig. 8a) is 280%
phase selector block diagram is also responsible for determining higher than the ripple produced by the unipolar.
the branches of the VSI which will be activated during each p/3 The comparison of Figs. 7b and 8b shows that the lower-order
rad sector of the operation cycle. Fig. 6b shows the picture of the switching harmonics of the unipolar strategy come in side bands
BLDC motor manufactured by Golden Motor Inc. whose main around 100 kHz, whereas the switching harmonics of the bipolar
parameters are given in Table 1. Fig. 6c shows the picture of the strategy arise around 50 kHz. Thus, considering the commutation
three-phase VSI built with SEMiX101GD066HDS Semikron
module. This system is intended to be embedded into a small EV
which the electrical energy will be supplied by a 48 V/100 Ah Table 2 Delays and voltage drop of the BLDC drive
lithium-ion battery.
The control algorithm of the BLDC motor as well as the detection Parameter Value
rotor position and speed algorithms were implemented in a
TMS320F28335 DSP from Texas Instruments Inc. Fig. 7a shows blanking time (TBT ) 600 ns
the steady-state three-phase current waveforms of the BLDC while sample delay time (TSD ) 1.2 μs
delay time of pulse drive (TDD ) 600 ns
Fig. 7b shows the harmonic spectrum of phase ‘a’ current. The voltage drop of switches (VG ) 1.45 V
reference ‘pseudo’ current signal is 15 A and the ripple of the

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
694 & The Institution of Engineering and Technology 2017
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 9 Phase ‘a’ BLDC current for a step change in the reference signal from 15 to 30 A
a Without parametric compensation
b With parametric compensation

frequency is the same in these two strategies, the unipolar PWM The mechanical load was adjusted to force the BLDC motor to
predictive current controller will result in a better torque response drain the same ‘pseudo’ current in both cases. The comparison of
[24] since the current ripple is reduced and the frequency of the the waveforms shows that the current is not fully rectangular in the
switching harmonics is two times higher than the commutation second case. This behaviour is due to the fact that the back EMF
frequency of the VSI (50 kHz). is not thoroughly trapezoidal for the operation of the motor at high
Incorporating the compensating factors (16), (17), (18) and (19) speeds since the airgap magnetic flux does not vary abruptly. As a
into (7) yields the following the compensated predictive control law consequence, there will be an error between the real back EMF
and the estimated one.
2Lfs  ∗ 2E 4TBT One way to improve the current waveform is shown in Fig. 10b
m[ k + 1 ] = I − Ip [k] − m[k] + + can be achieved by measuring and storing the points of the back
Vdc p Vdc Ts
EMF waveform, for different motor speeds, by addressing these

 E 4VG points as a function of the rotor position during the machine
+ TDD + TSD fs + . (20)
Vdc Vdc operation. However, from Fig. 3a, assuming that voltage E (real
back EMF) is not entirely compensated by the feed-forward
where TBT , TDD , TSD and VG can be obtained experimentally and signal Ec (estimated EMF), due to the aforementioned error, it is
their values are given in Table 2. possible to write a disturbance transfer function which poles will
Fig. 9 shows the step response of the current of the phase ‘a’ for the be at the same location as those given in (14). Therefore, if the
predictive control laws given by (7) and (20), without and with the estimation error of the back EMF is bounded, the current of
parametric compensation, respectively. In these figures, the red the compensated system will present a stable behaviour since the
dashed line indicates the instant in which the ‘pseudo’ reference poles of the disturbance transfer function are placed inside the
current Ip∗ is varied from 15 to 30 A. As expected, the current of the unity circle [21].
BLDC tracks its reference signal in a fast way, two sampling periods, Finally, an outer speed loop is added and the ECE-15 (European
regardless of the control algorithm and the instant in which the Driving Cycle 15) is used to investigate the performance of the
reference signal is changed. The comparison between the current predictive current controller. Fig. 11a shows the BLDC reference
waveforms depicted in Figs. 9a and b demonstrates that the signal and speed (upper curves) and the phase ‘a’ current (bottom
methodology proposed in this work compensates the steady-state curve). In Fig. 11b, the ECE-15 was modified replacing the smooth
error without affecting the transient response of the system. variations of the speed reference signal by step changes. Both results
Figs. 10a and b show the current of the phase ‘a’ considering the demonstrate that the predictive current control of the BLDC drive
operation of the BLDC motor at 500 and 2000 rpm, respectively. presents a quick response, assuring a good speed tracking.

Fig. 10 Steady-state current waveform for the operation of the BLDC at

a 500 rpm
b 2000 rpm

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
& The Institution of Engineering and Technology 2017 695
 17518679, 2017, 5, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-epa.2016.0421, Wiley Online Library on [13/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 11 Speed BLDC control: reference signal and motor speed (upper curves) (vert. scale: 500 rpm/div), phase ‘a’ BLDC current (bottom curve)
a European driving cycle (ECE-15) benchmark
b Modified European driving cycle benchmark

7 Conclusion 5 Fang, J., Zhou, X., Liu, G.: ‘Precise accelerated torque control for small inductance
brushless dc motor’, IEEE Trans. Power Electron., 2013, 28, (3), pp. 1400–1412
6 Li, H., Zheng, S., Ren, H.: ‘Self-correction of commutation point for high-speed
This paper proposed a novel PWM predictive controller to regulate sensorless bldc motor with low inductance and nonideal back emf’, IEEE Trans.
the currents of a low inductance 5 kW/48 V three-phase three-wire Power Electron., 2017, 32, (1), pp. 642–651
brushless DC motor. The key idea is to control the three-phase 7 De, S., Rajne, M., Poosapati, S., et al.: ‘Low-inductance axial flux bldc motor drive
for more electric aircraft’, IET Power Electron., 2012, 5, (1), pp. 124–133
VSI as a full-bridge DC–DC converter which allows increasing the 8 Chau, K., Chan, C., Liu, C.: ‘Overview of permanent-magnet brushless drives for
inverter output equivalent frequency switching up to 100 kHz electric and hybrid electric vehicles’, IEEE Trans. Ind. Electron., 2008, 55, (6),
while each individual switch is commutated with 50 kHz. This pp. 2246–2257
strategy made it possible to use an IGBT-based VSI to drive the 9 Mohamed, T., Bevrani, H., Hassan, A., et al.: ‘Decentralized model predictive
based load frequency control in an interconnected power system’, Energy
BLDC. It also made possible the predictive algorithm optimisation Convers. Manage., 2011, 52, (2), pp. 1208–1214
due to the fact that it is only necessary to run one control 10 Rodriguez, J., Cortes, P.: ‘Predictive control of power converters and electrical
algorithm to regulate a ‘pseudo’ current. Furthermore, it avoids the drives’ (John Wiley & Sons, 2012), vol. 40
cumbersome cost function minimisation. Despite of the simple 11 Wang, X., Wang, X., Fu, T., et al.: ‘Predictive instantaneous torque control for disc
coreless permanent magnet synchronous motor with the current source chopper’,
processing requirement, the robustness and reliability of the IEEE Trans. Power Electron., 2015, 30, (12), pp. 7100–7112
predictive controller is strongly influenced by model mismatches 12 Sozer, Y., Torrey, D.A., Mese, E.: ‘Adaptive predictive current control technique
and parameter variation. An analysis of the absolute stability was for permanent magnet synchronous motors’, IET Power Electron., 2013, 6, (1),
performed under parametric uncertainties. A methodology to pp. 9–19
13 Morel, F., Lin-Shi, X., Rétif, J.-M., et al.: ‘A comparative study of predictive
eliminate the steady-state error due to blanking time, pulse driver current control schemes for a permanent-magnet synchronous machine drive’,
delay, sampling delay and switches voltage drop was presented IEEE Trans. Ind. Electron., 2009, 56, (7), pp. 2715–2728
and incorporated in the modified predictive control law. The 14 Arashloo, R.S., Salehifar, M., Romeral, L., et al.: ‘A robust predictive current
experimental results obtained with a laboratory prototype were controller for healthy and open-circuit faulty conditions of five-phase bldc drives
applicable for wind generators and electric vehicles’, Energy Convers. Manage.,
presented to validate the theoretical analyses. The performance of 2015, 92, pp. 437–447
the predictive current controller was tested for different motor 15 Alexandrou, A.D., Adamopoulos, N.K., Kladas, A.G.: ‘Development of a constant
speeds, demonstrating a fast response and a good current tracking. switching frequency deadbeat predictive control technique for field-oriented
The unipolar PWM strategy resulted in a better frequency response synchronous permanent-magnet motor drive’, IEEE Trans. Ind. Electron., 2016,
63, (8), pp. 5167–5175
since the current ripple was reduced and the switching harmonic 16 Na, W., Park, T., Kim, T., et al.: ‘Light fuel-cell hybrid electric vehicles based on
frequency was doubled. predictive controllers’, IEEE Trans. Veh. Technol., 2011, 60, (1), pp. 89–97
17 Fakham, H., Djemai, M., Busawon, K.: ‘Design and practical implementation of a
back-emf sliding-mode observer for a brushless dc motor’, IET Electric Power
Appl., 2008, 2, (6), pp. 353–361
8 Acknowledgment 18 Dixon, J.W., Leal, L.: ‘Current control strategy for brushless dc motors based on a
common dc signal’, IEEE Trans. Power Electron., 2002, 17, (2), pp. 232–240
The authors thank FAPEMIG, CNPq, INERGE and SEMIKRON for 19 Berendsen, C.-S., Champenois, G., Bolopion, M.: ‘Commutation strategies for
brushless dc motors: influence on instant torque’, IEEE Trans. Power Electron.,
the financial support and semiconductor donation for this work. 1993, 8, (2), pp. 231–236
20 Mohan, N., Undeland, T., Robbins, W.: ‘Power electronics: converters,
applications, and design, number v. 1 power electronics: converters, applications,
and design’ (John Wiley & Sons, 2003)
9 References 21 Åström, K.J., Wittenmark, B.: ‘Computer-controlled systems: theory and design’
(Courier Corporation, 2013)
1 Asano, K., Inaguma, Y., Ohtani, H., et al.: ‘High performance motor drive 22 Cortés, P., Kazmierkowski, M.P., Kennel, R.M., et al.: ‘Predictive control in power
technologies for hybrid vehicles’. 2007 Power Conversion Conf.-Nagoya, 2007 electronics and drives’, IEEE Trans. Ind. Electron., 2008, 55, (12), pp. 4312–4324
2 Krishnan, R.: ‘Permanent magnet synchronous and brushless DC motor drives’ 23 Damasceno, A., Braga, H., Barbosa, P.: ‘Battery charge system based on
(CRC Taylor & Francis, 2010) bidirectional dc-dc converter employing a digital current-mode controller for
3 Krah, J.-O., Holtz, J.: ‘High-performance current regulation and efficient PWM photovoltaic applications’. Proc. of Brazilian Power Electronics Conf., 2007,
implementation for low-inductance servo motors’, IEEE Trans. Ind. Appl., 1999, pp. 388–394
35, (5), pp. 1039–1049 24 Song, J.-H., Choy, I.: ‘Commutation torque ripple reduction in brushless dc motor
4 Fang, J., Zhou, X., Liu, G.: ‘Instantaneous torque control of small inductance drives using a single dc current sensor’, IEEE Trans. Power Electron., 2004, 19,
brushless dc motor’, IEEE Trans. Power Electron., 2012, 27, (12), pp. 4952–4964 (2), pp. 312–319

IET Electr. Power Appl., 2017, Vol. 11, Iss. 5, pp. 688–696
696 & The Institution of Engineering and Technology 2017