Sliding Mode Vector Control of Non-Sinusoidal

Permanent Magnet Synchronous Machine

José Roberto B. A. Monteiro, Member, IEEE, Allan G. de Castro and Thales E. P. de Almeida
School of Engineering of São Carlos - EESC
University of São Paulo
São Carlos, SP 13566-590
Email: jrm@sc.usp.br
Abstract—This work presents a sliding mode controller for the possible to produce ideal 120◦ square-wave currents, resulting
speed control of a surface-mount permanent magnet synchronous in electromagnetic torque ripple, and its values can be high,
machine (PMSM) with non-ideal trapezoidal back-EMF, using depending on the application [3]. Lately, in [4] is presented
vectorial non-sinusoidal modeling. The calculus of vector-based a vector control approach of the BLDC to reduce the torque
machine model considers an ideal trapezoidal back-EMF wave- ripple. A discussion showing a comparison between a BLDC
form, i.e., all control calculation is based on ideal back-EMF, but
they are applied to a machine with a real back-EMF waveform,
motor, i. e., the PMSM machine with six-step converter,
which can be considered a distorted trapezoid. For the sliding and a PMSM with vectorial control, but using classical PI
controller, this paper shows the use of a sigmoidal function, the controllers, are made in [4], showing the visible effect of phase
hyperbolic tangent, instead of the signal function and uses a commutation of the six-step converter in a positioning and the
proper sliding surface that presents combined integrative action absence of a equivalent visible effect in the vectorial mode.
and anti-windup effect, which is applied to the vector-based model
of the non-sinusoidal machine. The final system consists in a In this work the vector model based control of surface-
high performance controller, robust to machine parameters and mount permanent magnet synchronous machine with non-
disturbances, and light in its design as shown by the results. The sinusoidal back-EMF is implemented considering second order
chosen sliding surface with integrative control action, combined sliding mode speed controller. The use of the vectorial control
to an anti-windup feature and the torque ripple free vector-based for this kind of machine, allows the compensation of its
model, give zero steady state speed error, minimizes the system non-sinusoidal back-EMF, producing a electromagnetic torque
speed output overshoot and ripple. smoother than using the conventional converter, i. e., three
Keywords—BLDC motor, sliding mode control, vector PMSM phase inverter in six-step mode. Differently from [6] and [7]
control, motor speed control, anti-windup. that employ the conventional six-step mode to fed the machine,
in this work, the non-sinusoidal vectorial model of the machine
is used together with two first order sliding mode controllers,
I. I NTRODUCTION
for direct and quadrature stator current components control
Basically, surface-mount permanent magnet synchronous loops, and a second order sliding mode controller for speed
machines (PMSM) can be classified into two main categories, control loop. The objective of the use of that proposed control
which is related to its back-EMF waveform: sinusoidal ma- topology, applied to trapezoidal machines, is to make their
chines and trapezoidal machines. As expected, there are no use competitive comparing to the use of sinusoidal machines
absolute pure sinusoidal back-EMF waveforms and in the same in servo applications, once surface-mounted permanent magnet
way, there are not purely trapezoidal back-EMF waveforms. synchronous machines with trapezoidal back-EMF are cheaper,
This work deals with trapezoidal machines, or better to say, simpler and have more energy density than their equivalent
non-sinusoidal back-EMF machines. This term is used to sinusoidal back-EMF counterparts [1].
refer to a wide variety of back-EMF waveforms considered
trapezoidal [1]. II. M ACHINE M ODELING
It is useful to make a note about the term brushless The machine model is very studied in the literature and can
DC motor (BLDC motor), which refers to the set composed be easily found, as in [4]. Just for better reference is shown
by a non-sinusoidal (or trapezoidal) surface-mount permanent here:
magnet synchronous machine and its electric power converter,
commonly a three phase machine with a three phase electric
converter (a three phase inverter) [2][1]. Considering an ideal      
va Ls Ms Ms ia
trapezoidal back-EMF waveform and an equally ideal 120◦ vb = Ms Ls Ms d
ib +
dt
square wave stator current, the machine produces an almost vc Ms Ms Ls ic
ripple free electromagnetic torque, as in Fig. 1. In this case (1)
the converter operates in six-step mode where each inverter      
ia ea vn
bridge switch is active during 120◦ electrical, resulting in 2 Rs ib + eb + vn
switches on simultaneously. However, there are no machines ic ec vn
with pure trapezoidal back-EMF waveforms and it is not
978-1-5090-6248-5/17/$31.00 2017
c IEEE where:

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
 ea

ia

eb

ib

ic
ec Fig. 2. Waveforms for Φra , Φrb and Φrc [6].

π 2π
0 3 3 π
θe
Fig. 1. Brushless DC motor ideal electromagnetic torque generation (θe :
electrical position of the rotor) [6].
ea , eb and ec : induced voltage of stator phases
a, b and c, respectively, due to rotor magnets
movement, as in (2);
ia , ib and ic : stator phase currents a, b and c,
respectively;
Ls : stator phase self-inductance;
Ms : stator phases mutual inductances;
Rs : stator phase resistance;
va , vb and vc : a, b and c stator phases applied
voltages, respectively;
vn : stator neutral terminal voltage (this terminal is
usually not connected).
Considering Φra , Φrb and Φrc the linked magnetic fluxes
between rotor magnets and stator winding phases a, b and c,
respectively:
Te
A. PMSM Vectorial Modeling
4π 5π
3 3 2π
The vector-based model of this machine has the main
objective to obtain the electromagnetic torque equation pro-
portional to a stator current component, as in the vector-based
model for sinusoidal machines. In order to achieve that, a spe-
cial transformation is used additionally to the transformations
used for the sinusoidal one, i.e., αβ and dq transformations.
This transformation is referred here by dqx transformation [4]
and is applied following to the dq one, as shown in Fig. 3 and
is written as:
xαβ = ax ejθx ejθe xdqx (4)
where:
ax : dqx size of axes, relatively to dq axes, which are
considered equal to the unity;
θx : angle of dqx axes , relatively to dq axes.

        Fig. 3. dqx transformation represented by blocks.

Φra d Φra 1 ea 1 d Φra
Φrb = Φrb = eb = Φrb The αβ transformation is:
Φrc dθe Φrc ωe ec ωe dt Φrc
(2)   
2  xa
2π 2π
where: xαβ = 1 ej 3 e−j 3 xb (5)
3 xc
ωe : electrical rotor speed.
   
2 √ √ √  xa
TEL = npp (Φra ia + Φrb ib + Φrc ic ) (3) x0 = 2 2 2 xb (6)
3 2 2 2
xc
where:
where:
TEL : machine-generated electromagnetic torque;
npp : number of machine’s pole pairs; xαβ : quantities written over αβ axes (complex num-
Since the magnets are mounted in rotor surface, inductance ber);
variation due to rotor position can be neglected [2]. x0 : zero component;
xa , xb and xc : quantities of a, b and c phases,
The phases back-EMFs are ideally trapezoidal, therefore respectively.
Φra , Φrb and Φrc are also trapezoidal as shown in Fig. 2,
considering (2), where ΦM is their amplitudes and θe is the Applying those transformations to machine electric and
rotor angle in electrical degrees. torque equations, (1) and (3), and choosing ax and θx as:

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
 
3 Φm ΦM real back-EMF
ax = (7) trapezoidal back-EMF
2 Φ2 + Φ2
rα rβ

−Φrα
θx = tan−1 − θe (8)
Φrβ 0

The electromagnetic torque equation in dqx axes is:


3 −ΦM
TEL = npp Φm i q x (9)
2 0 π 2π
π 4π 5π
2π
3 3 3 3

And the electric equations: θe

Fig. 4. Ideal trapezoidal back-EMF and measured back-EMF, of the real

didx machine, waveforms.
vdx = Rs idx + (Ls − Ms ) +
dt

 1.2
a real ax
(Ls − Ms ) ωe idx x − iqx (1 + θx ) 1
trap. ax
ax 0.8
0.6
diqx 0.4
vqx = Rs iqx + (Ls − Ms ) +
dt 0.2

  0
ax  3 1 -0.2
(Ls − Ms ) ωe iqx + idx (1 + θx ) + Φm 2 ω e 0 π 2π
π 4π 5π
2π
ax 2 ax 3 3 3 3
(10) θe
where:
Fig. 5. Coefficient ax for the measured back-EMF of the real machine and
 for an ideal trapezoidal back-EMF.
3 Φm
ax = (11)
2 Φ2 + Φ2 4
rα rβ
3 real θx (◦ )
2 trap. θx (◦ )
−Φrα 1
θx = tan−1 − θr (12)
Φrβ 0
-1
-2
The measured back-EMF waveform for one phase together -3
with an ideal trapezoidal back-EMF waveform are shown in -4
π 2π 4π 5π
Fig. 4, where the real back-EMF is from a Siemens machine, 0 3 3 π 3 3 2π
model 1FT5 062. The coefficients for those back-EMF wave- θe
forms are shown in Figs. 5 and 6, for ax and θx respectively.
For the vector control, the coefficients for the trapezoidal back- Fig. 6. Coefficient θx for the measured back-EMF of the real machine and
EMF are used, but the machine model employs the real back- for an ideal trapezoidal back-EMF.
EMF waveform. This causes a difference in dqx currents, but
with minimal influence on electromagnetic torque, as seen in
Sec. IV. The current controllers are first order sliding mode con-
trollers, as in [6], so their sliding mode variables are:
III. S LIDING M ODE C ONTROL
σdx = i∗dx − idx
A block diagram of the complete drive system is shown (13)
σqx = i∗qx − iqx
in Fig. 7, where there are three controllers: the stator current
controllers, Cdx and Cqx , for direct and quadrature controllers;
and the rotor speed controller, Cω . As the time constant of where
mechanical system is far greater than the time constant of i∗dx : stator line direct axis current reference,
electrical system, rotor speed control loop are weakly coupled i∗qx : stator line quadrature axis current reference,
to current control loops, which are strongly coupled to one idx : stator line current for direct axis, and
another. Thus, it is possible to make separate analysis for the iqx : for quadrature axis.
controllers of speed loop (Cω ) and for current loops (Cdx and
Cqx ) [5]. And the stator current controller is:

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.


 
  




 








Fig. 7. System complete control block diagram (KT = 2npp ΦM , Cω : rotor speed controller, Cdx : stator direct current controller, Cqx : stator quadrature
current controller).

σω : sliding variable for rotor speed control loop

C dx : vdx = tanh kI · σdx (Cω );
(14)
C qx : vqx = tanh kI · σqx ω = ω ∗ − ω: shaft speed error;
ω ∗ : rotor speed reference;
The reachability and the sliding mode conditions of the ω: measured rotor speed.
current controller for the first order sliding mode controller
(14) are both shown in [6]. The dynamic mechanical load equation is:

dω B 1 1
A. Rotor Speed Sliding Mode Controller = ω̇ = − ω − TL + TEL (17)
dt J J J
The used sliding variable for the speed control loop is [7]:
where:

r−1
d B: equivalent frictional coefficient, composed by
σ= λ( ) + dt (15)
dt rotor shaft bearings and load frictional losses;
J: combined inertia momentum of machine rotor and
where: load;
TL : load torque;
= x∗ − x;
r: is the degree of sliding surface;
B. System convergence
x: vector of state variables;
x∗ : references for state variables; A general non-linear system can be written as:
The sliding surface is defined as σ = 0 and one must
ẋ = f (x) + g(x)U
note that it differs from the commonly generic sliding variable (18)
shown in [8]. y = h(x)

Due to the use of the integral operator, λ must be a function where:

of x∗ − x. For this application, it is considered a sliding
surface of order 2 (r = 2). The function λ must be chosen U : system input;
according to the application in order to satisfy the reachability f , g, h: linear or non-linear functions, characterizing
and attractability criteria. Therefore, the speed loop controller the system;
is a sliding mode of order 2, and its sliding surface is: y: system output.
 The system input function U can be written as a sum of a
σω = λ( ω ) ω dt + ω = 0 (16) continuous function and a switching function:

where: U = Ueq + Uc (19)

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
where: 500

Ueq : is a continuous function and is referred as 400

“equivalent control”, representing the operation

300
point where the sliding regime occurs [9];

λ( ω )
Uc : is the switching control, representing the variable 200
ω ∗ = 210rd/s
structure of the system, responsible to the attrac- 100 ω∗ = 0

tiveness of the system to the sliding regime. ω∗ = 0

0
ω ∗ = 210rd/s
A well accepted method to prove the convergence of that -40 -20 0 20 40

system to the operation point is by the definition of an energy ω

Lyapunov function and giving the conditions to be satisfied,

the final inequality is [7]: Fig. 8. Hyperbole delimiting reachability condition.

TABLE I. BLDC MOTOR AND MECHANICAL LOAD PARAMETERS USED

∂σ IN SIMULATIONS .
g(x) < 0 for σ = 0 (20)
∂x
Motor Load
The use of a sigmoidal function does not alter the inequal- Rs = 2.3Ω J = 4.2 · 10−3 kg m2
ity (20) as well. (Ls − Ms ) = 12.5mH B = 3.032 · 10−3 kg m2 /s
npp = 3
In order to prove the sliding mode of the system, the Φm = 0.12Wb
equivalent control must conform to systems limits and in order
to prove the reachability condition, so (20) must be satisfied.
Using (17), the equivalent control of the system input (TEL )
value (2.2Nm) and to its maximum in the opposite direction
is:
(-2.2Nm). Fig. 9 shows the machine accelerating and reaching
1000rpm of rotor speed. The details of machine speed is shown
TELeq = Bω + TL (21) in Fig. 10, where it decreases when torque load increases and
vice-versa.
Considering the bounding limits of operation, in the maxi-
mum desired speed, maximum machine electromagnetic torque Fig. 11 shows details of electromagnetic torque and its
must equals the product of maximum speed and shaft friction reference for the quadrature current component. It can be seen
coefficient plus maximum load torque. that produced torque is almost free of ripple, even considering
a different back-EMF waveform, as the coefficients ax and θx
In order to prove that the system stays in the adopted where evaluated using a pure trapezoidal waveform, which is
sliding surface, considering g(x) = 1/J in (16), (20) leads different from that of the machine under test.
to:

4Nm Te
λ( ω ) ω λ( ω ) ω ω̇
−1<0⇒ < 1 ⇒ λ( ω ) < (22) 2Nm
ω̇ ω̇ ω
TL
0
B 1 1 B 1
Using: ω̇ = − ω − TL + TEL = ω − TL − 1000rpm ω∗
J J J J J
B ∗ 1 ω
ω + TEL and considering limit operational values for TL 0
J J
and i, which gives maximum allowed TEL : 40V

ea 0
1 B ∗
B (TEL − TL ) ω -40V
λ( ω ) < + J − J (23) 5A
J ω ω

That results in a hyperbole depending on ω ∗ . When ω <

0, TEL < 0 and considering TEL the preponderant factor of ia 0
the equation, then the hyperbole is always positive. So λ( ω )
must be inscribed in that hyperbole in order to satisfy the
reachability condition (Fig. 8).
-5A

IV. R ESULTS 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
t(s)
The used machine and its load has the parameters shown
in Table I. The load torque (TL ) is 2.2Nm for this machine. Fig. 9. Machine operation at 1000rpm, with control using ax and θx
Simulation results are shown in Figs. 9 to 10. In the figures, coefficients evaluated for an ideal trapezoidal back-EMF waveform but applied
in a real back-EMF machine used.
the load torque is varied abruptly from 0 to its maximum

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
 105.5
105 4Nm Te
(rad/s)

104.5
2Nm
104 TL
ω∗
103.5 ω 0
103
0.06 0.08 0.1 0.12 0.14 0.16 0.18 -2Nm
t(s) ω∗
2000rpm

Fig. 10. Machine operation at 1000rpm, detail of speed during load torque 1000rpm
transients. ω
80V

40V
Te∗
ea 0
2Nm TL
-40V

0 -80V
40V 5A

ea 0

-40V ia 0
4A

ia 0
-5A
0.2 0.25 0.3 0.35 0.4 0.45
-4A t(s)
0.16 0.165 0.17 0.175 0.18 0.185 0.19 0.195 0.2
t(s)
Fig. 12. Machine operation at 2000rpm, with control using ax and θx
coefficients evaluated for an ideal trapezoidal back-EMF waveform but applied
Fig. 11. Machine operation at 1000rpm, detail of reference torque chattering. in a real back-EMF machine used.

211
Fig. 12 shows the machine accelerating from 1000rpm 210.5
and reaching 2000rpm. During this interval, load torque is
(rad/s)

210
209.5
applied against machine motion (positive value) and favoring 209 ω∗
its motion (negative value). Those load torque changes causes 208.5 ω
some transients in machine speed. After machine achieves its 208
0.3 0.32 0.34 0.36 0.38 0.4 0.42 0.44
set point speed, it occurs a little overshoot due to load torque
t(s)
favoring its motion (Fig. 13). In t = 0.35s, load torque is
released, so machine speed falls but integrative action of the
Fig. 13. Machine operation at 2000rpm, detail of speed during load torque
controller put it back to its set point. In t = 0.4s, a load torque transients.
against its motion is applied, so machine speed falls again, but
due to integrative action, it reaches its set point after 50ms.
When this load is released, machine accelerates but it is back sliding mode control and causes non-zero steady state error, as
to its set point again. pointed out by the literature. The anti-windup effect achieved
by the proposed surface is similar to the conventional method
V. C ONCLUSIONS called conditional integration, but with the difference that the
Considering the results, the ripple in speed is practically integrative portion is gradually activated as the error decreases,
nonexistent due to almost absence of torque ripple, even so the integrator is active even for large error values. This is
considering an ideal trapezoidal back-EMF for the controller a good feature once the conventional method can fail if the
and a real back-EMF (distorted trapezoidal waveform) for the error does not fall into a range.
machine. As the back-EMF waveform of the machine has
The combination of the vector-based model plus the sliding
minor variation in its normal region of operation, the control
mode control shows excellent results for the control of machine
presents a good performance.
rotor speed, i. e., fast speed response, no steady state error,
This work combines a sliding surface and non-sinusoidal minimum rotor speed overshoot and no chattering.
vector-based model in order to minimize electromagnetic
torque ripple and eliminate the steady state error of the
system which is introduced by the use of analog functions ACKNOWLEDGMENTS
as switching functions, instead of the signal function. The use
of an analog function, as the hyperbolic tangent function, can The authors would like to acknowledge the grant
reduce significantly system chattering, but it is not properly a #2006/04226-0, São Paulo Research Foundation (FAPESP).

Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
 R EFERENCES
[1] T. J. E. Miller, Bushless Permanent-Magnet and Reluctance Motor
Drives. Oxford: Claredon Press, 1993.
[2] S. A. Nasar, I. Boldea, and L. E. Unnewehr, Permanent magnet, relu-
tance, and self synchronous motors. Boca Raton: CRC Press, 1993.
[3] J. Holtz and L. Springob, “Identification and compensation of torque
ripple in high precision permanent magnet motor drives,” IEEE Trans-
actions on Industrial Electronics, vol. 43, no. 2, 1996.
[4] J. R. B. A. Monteiro, A. A. Oliveira, M. L. Aguiar, and E. R. Sanagiotti,
“Electromagnetic torque ripple and copper losses reduction in permanent
magnet synchronous machines,” European Transactions on Electrical
Power, vol. 22, no. 5, pp. 627–644, June 2012, doi: 10.1002/etep.594.
[5] B. M. Patre, V. M. Panchade, and R. M. Nagarale, Sliding
Mode Control. InTech, 2011, ch. Sliding Mode Control of DC
Drives, available from: http://www.intechopen.com/books/sliding-mode-
control/sliding-mode-control-of-dc-drives.
Authorized licensed use limited to: Instituto Tecnologico Metropolitano ITM. Downloaded on March 30,2021 at 20:46:36 UTC from IEEE Xplore. Restrictions apply.
[6] C. M. R. Oliveira, J. R. B. A. Monteiro, and M. L. Aguiar, “Sliding
mode control of brushless dc motor speed with chattering reduction,”
in Proceedings of the 2015 IEEE 24th International Symposium on
Industrial Electronics (ISIE). Búzios, Brazil: IEEE, June 2015, pp.
590–595.
[7] J. R. B. A. Monteiro, C. M. R. Oliveira, T. E. P. Almeida, and M. J.
Cezare, “Pseudo sliding mode control with integrative action applied to
brushless dc motor speed control,” in 2015 IEEE 13th Brazilian Power
Electronics Conference and 1st Southern Power Electronics Conference
(COBEP/SPEC), Nov 2015, pp. 1–6.
[8] J. Slotine and W. Li, Applied nonlinear control. Englewood Cliffs:
Prentice Hall, 1991.
[9] M. Mahmoudi, N. Madani, M. Benkhoris, and F. Boudjema, “Cascade
sliding mode control of a field oriented induction machine drive,” The
European Physical Journal Applied Physics, vol. 7, pp. 217–225, 1999.