Víctor REPECHO1, Domingo BIEL1,2, Antoni ARIAS1

Universitat Politècnica de Catalunya (1), Institut d’Organització i Control (2)

Sensorless Sliding Mode Control of PMSM Drives Using a High

Frequency Injection Algorithm
Abstract. A high frequency injection (HF) technique for rotor position estimation using Sliding Mode Control (SMC) in Permanent Magnet
Synchronous Machines (PMSMs) is presented. Since SMC is used, instead of the Field Oriented Control (FOC), the injected HF test signals for
tracking the machine’s saliency are naturally eliminated. Such angle information suppression is tackled using ”the equivalent control” principle.
Simulation results confirm not only the proper angle estimation but also the sensorless control at low speed reversal and load impact at zero speed.

Streszczenie. W artykule przedstawiono technikę dodawania sygnału wysokoczęstotliwościowego (HF) w celu estymacji położenia wirnika w
przypadku sterowania ślizgowego (SMC) maszyną synchroniczną z magnesami trwałymi (PMSMs). Od kiedy do sterowania ślizgowego, zamiast
metody FOC, jest dodawany testowy sygnał wysokoczęstotliwościowy, asymetria magnetyczna maszyn zostaje w sposób naturalny wyeliminowana.
Tłumienie informacji o kącie może być rozwiązane przy użyciu zasady „regulacji równoważnej”. Wyniki symulacji potwierdzają nie tylko właściwą
estymację kąta jak również, ale również możliwość sterowania bezczujnikowego przy bardzo niskich prędkościach i nawrocie maszyny..
(Bezczujnikowe sterowanie ślizgowe napędem PMSM z zastosowaniem algorytmu z dodawaniem sygnału wysokoczęstotliwościowego)

Keywords: Sliding Mode Control. PMSM Sensorless Control. High Frequency Injection Methods.
Słowa kluczowe: Sterowanie ślizgowe, bezczujnikowe sterowanie PMSM, metoda dodawania HF.

Introduction signals and therefore the system tends to cancel it as any

Permanent Magnet Synchronous Machines (PMSMs) other perturbation.
have higher power density, higher efficiency and better In this paper, the HF injection technique for tracking the
dynamic performance than Induction Machines. Its control rotor position in a Surface Mount PMSM Controlled by
requires accurate rotor position information in order to Sliding Mode is addressed. The novelty of the paper relies
implement the coordinate transformation and the speed on the use of SMC instead of the well known FOC
(and position) vector control loops. Significant research technique and therefore the issue is to tackle the natural
efforts have been conducted in order to achieve vector suppression of the injected signal. The well-know benefits
control of PMSMs without encoders or resolvers. These of the SMC [14] [15], like fast torque response, robustness,
techniques can be broadly divided into model based reduction system order or high immunity against
techniques, where the back-emf of the machine is used for parameters drift are reached, improving the FOC dynamics.
rotor magnet flux detection, and injection techniques, where Position signals and the injection algorithm are
a test signal, either high frequency (HF) AC voltage or discussed and fully presented. Finally, Sensorless SMC
voltage pulse, is used to detect the rotor saliency speed reversal and load impact results are presented.
(difference between LD and LQ) [1,2].
Model based techniques e.g. [3,4] successfully achieve PMSM Model
sensorless control at medium and high rotor speed but fail The electrical equations that model the PMSM in the
at low excitation frequencies due to the reduction and rotating reference frame (dq) are shown on (1) and (2).
eventual disappearance of the back-emf induced by the
rotor magnets at low rotor speed. diD vD R LQ
Injection methods, on the other hand, detect the angular
(1)    iD  e iQ
dt LD LD LD
dependent saliency of the machine and its rotor position
estimation is therefore fundamentally speed independent diQ vQ R L 
[1,2]. (2)    iQ  e  D iD  e  M
There are mainly two different injection techniques. The dt LQ LQ LQ LQ
first one consists on the superposition to the fundamental
where: vD,vQ - Stator dq Voltages, iD,iQ - Stator dq currents,
voltage vector of a HF injection either in the alpha/beta
LD,LQ -Stator dq Inductancse, R - Stator Resistance, ΦM -
frame [5,6,7,8] or in rotating d/q synchronous frame [9,10].
Permanent Magnet Flux, ωe - Electrical angular speed, p -
The second ones are based on the modification of the
Pole pairs number, B - Friction coefficient, J - Moment of
fundamental PWM pattern to include a voltage pulse test
inertia, TL - Load torque, ωm - Mechanical angular speed.
[11,12,13].
For all injection methods to function, some level of The related electromagnetic torque equation to dq current
machine saliency is necessary. This makes the technique components is expressed at:
straightforward for salient machines such as the interior
3  p  M
permanent magnet ones. Surface Mount PMSM, on the
other hand, only have a saliency due to stator tooth
(3) TE  iQ  ( LD  LQ )  iD  iQ 
2
saturation and it is generally of small magnitude. where:
Such mentioned injection techniques are mostly
implemented using Field Oriented Control (FOC) schemes 3  p  M
where the band width of the inner current control loops is
(4) kT 
2
typically of lower value than the injected frequency. The PMSM motion equation is:
Therefore, the system does not react against such a
perturbation not eliminating the further position information. dm 1 B 1
On the other hand, Sliding Mode Controllers (SMC) have a (5)  Te  m  TL
band width, (which strictly speaking is not defined since it is dt J J J
not a linear controller), within the range of the injected test

PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 1a/2012 16

 From the expressions (3), (4) and (5) the third state Taking a look of the control law expression summarized
space system equation can be obtained. in (16) and (17), it could be noted that the control does not
depend on any system state variable or parameters value
PMSM Sliding Mode Control (R or L ), which proves the high robustness against
The SMC is applied to the system defined by equations parameters drifts.
(1) and (2). The SMC is executed above a control variable, In order to ensure that the control can be successfully
which will take two discrete values at any time, choosing the employed, the sliding domain must be guaranteed. Indeed,
proper one depending on the system state. The control the two variable control discrete values have to be able to
forces the systems states trajectories towards the switching produce the equivalent control values, necessary to satisfy
surface at any time. The switching surfaces are defined by the references values for any given conditions.
the systems errors, where the errors are the difference
between the desirable and the real values of the (18)  vD 0  vDeq  vD 0
controllable variable.
In order to control PMSM, the chosen controllable (19)  vQ 0  vQeq  vQ 0
variables are dq currents, and consequently the control
variables are dq voltages. After the SMC has been deduced and developed, the
The switching surfaces (S) for controllable variables are PMSM can be controlled in terms of dq current with the
* control law obtained.
defined, using directly the error of the reference value (x )
and real value (x) of a given variable x, then: For setting the reference current values, it is necessary to
understand clearly the physical meaning of the dq model.
*
(6) S D  iD  iD  eID The dq motor model, as it was noted before, is a rotating
reference frame system aligned with the motor PM. The d
(7) SQ  iQ*  iQ  eIQ current component is aligned in the flux PM direction, and
the q current component, its 90º phase-shifted from d
Using the expression (1), (2), (6) and (7), and knowing component. From this point of view, the dq PMSM model
that the equilibrium point of the system is forced by the equation is highly similar to a stator equation of a DC motor.
SMC, the errors system dynamics is rewritten as follows: Besides, from this deduction, is clear that the preferred
desired current value on d component has to be set to 0, in
deID vD R LQ order to avoid fluxes that could demagnetize the PM.
(8)    eID  e  eIQ Torque current component, that is completely perpendicular
dt LD LD LD to PM flux, is created by iQ, as it can be deduced from (3).
deIQ vQ R L
(9)    eIQ  e  D  eID Sliding Mode Control and High Frequency Injection
dt LQ LQ LQ a. Analyzing the Control Band Width
The injection of a rotating HF voltages for saliency
The control will be defined by two discrete values, tracking when using FOC has been proposed by several
depending on the S sign. The discrete dq voltages values authors [1,3].
are defined as: At this point, analyzing the SMC proposed on section II, a
big difference with FOC control arises. Meanwhile the
(10) vD  vD 0 ,vD 0  bandwidth of the FOC control can be easily known by the PI
vQ  vQ 0 ,vQ 0 
controllers, the SMC have, theoretically, an infinite
(11) bandwidth (BW).
Typically, HF carrier signal frequency for angle estimation
The SMC has to ensure the system trajectories are is fixed around 1-2 kHz (between fundamental electrical
always directed towards the sliding surface, and this is frequency and PWM frequency). Fig. 1, shows how the HF
achieved when (12) and (13) are fulfilled: injection is inside SMC BW. This means that the SMC will
perform a determined action control against this HF
(12) S D  S D  0 injection, being treated as an external disturbance.
(13) SQ  SQ  0
SMC HF Injection
Using the equivalent control definition ( vDEQ, vQEQ ), G Controller
which is defined as the input values of control variable that
satisfy a given system condition, expressions (16) and (17)
are extracted from (14) and (15). The physical meaning of FOC
the equivalent control is that the system variables are on Control
the switching surfaces and remain there.
SMC BW f
FOC BW
(14) S D  [vD  vDEQ ]  0 (Ideally Infinite)
(15)  
SQ  vQ  vQEQ  0
Figure 1. FOC and SMC Band Widths

That yields: In FOC, the HF carrier voltage is injected on alpha beta

(αβ) plane, the PI controllers are not sensible at this
(16) vD  vD 0  sign( S D ) frequency and the system does not perform any action
(17) vQ  vQ 0  sign( SQ ) against this disturbance. The HF signals pass through the
PMSM, and the HF currents outputs have the desired angle
information. When the same injection technique is used with
SMC, the HF voltage passing through the PMSM is almost

17 PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 1a/2012

reduced to zero as well as the HF output currents because Note that in a non-salient PMSM ( LD=LQ ), the inductance
of the higher BW of the control and the capability to react versus angle variation is not produced.
against the perturbation. The αβ PMSM model it is defined by (23).
Taking into account the idea explained before, is
reasonable to choose a HF current carrier for the system. In  v  i  d i  d cos( r ) 
the same way, the control will perform against the (23) v   R i   L i    m  sin   
perturbation, but in this case, the current signal generated      dt    dt  r 
by the control, ideally equal to the disturbance with an
opposite signal, are passing through the motor. In this If we assume the disturbance frequency higher enough
injection method, the signal processing to obtain the angle than the phenomena related to the motor rotation, the HF
estimation must be done above the αβ voltages. On Fig. 2, motor model can be simplified as stated in (24).
the proposed system with the HF current injected and HF
voltage output processed is presented.
 v  d  i 
Note that an outer speed control loop has been (24) v   L  
implemented, based on the slower mechanical motion    dt i  
PMSM described on equation (4). A classical PI linear
controller is used, which will give the q current reference to Angle Estimation in Sliding Mode Control Scheme
the SMC, keeping the d current component always equal to The current injection added on the αβ currents is:
zero. If the sliding domain is ensured, we can extract the
dynamics of the system related to the angular speed from
equation (4). iP   sin( i t )
(25) i    AP   
b. HF PMSM System Model  P   cos( i t ) 
In the PMSM with saliency, there are variations of the
inductance values depending on the rotor position. These Under sliding motion the current vector cancelling the
changes are caused by the different flux levels on the disturbances can be modelled as (26):
corresponding PMSM areas, depending on PM position.
The inductance factor change in front of PM position is iP   sin( i t )
modelled: (26) i   AP  
Ls  Ls  cos(2r )  Ls  sin(2r )   P   cos( i t ) 
(20) L   
  Ls  sin(2r ) Ls  Ls  cos(2r ) The equation (24) is solved using as an input currents
Where the ones showed on (26). Solving the proposed expression,
(27) is reached. These voltages can be obtained using a
(21) Ls  LQ  LD  2 proper band pass filter, depending on the injection
frequency chosen.
(22) Ls  LQ  LD  2

Figure 2. Sliding Mode Control with High Frequency Injection.

PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 1a/2012 18

 Fig. 4 are directly the signals obtained at the end of the
vHF  vO  cos(  i t )  v1  cos( 2 r   i t )

(27)  Signal Process, not applying any additional filter. In order to

v HF   vO  sin(  i t )  v1  sin( 2 r   i t )  validate the quality of the position signals the FFT is
performed as shown in Fig. 5. The quality of the signal is
remarkable since any sub harmonic appears which might
where v0    I  AP  Ls , v1   I  AP  Ls cause an error in the estimated angle.
From the αβ position signals the arc tangent function is
executed and the estimated angle is obtained.
1.5

vHF 1

0.5


vHF  i  AP  Ls  e
j i t
 Ls  e
j( 2 r t i t )
 0

-0.5


vHF  i  AP  Ls  Ls  e
j( 2r t 2i t )
 -1

-1.5
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

j ( 2 r t  2i t )
vHF  i  AP  Ls  e Fig. 4. Alpha and beta position signals.
1
0.8
j2 r t
vHF  i  AP  Ls  e
0.6
0.4
0.2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

̂ r 1
0.8
0.6

Figure 3. Signal Processing Algorithm steps 0.4

0.2

The resultant output voltages have the same structure 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

than the output current voltages resultant when a HF

voltage signal is injected in FOC application. Therefore, the Fig. 5. Position signals’ FFT.
signal processing, shown in Fig. 3, for angle estimation (θe) b. Speed Reversal
is almost equal. Fig. 6 illustrates the performance under a step speed
Results change (+/-10 rad/s) at full load torque (12.2Nm), under
sensorless conditions.
The simulation results showed have been obtained using
the PMSM characterized by the table II parameters. The 20

speed PI control and SMC were adjusted for a good 10

dynamic response and acceptable error on the angle

0

-10

estimation obtained. -20

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

TABLE I. PERMANENT MAGNET SYNCHRONOUS MACHINE 15

Surface Mount PMSM 14

Rated power 3.8 kW 13

Poles number 6 12

Nominal speed / Rated torque 314.15 rad/s / 12.2 Nm 11

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

R / Ld / Lq 0.94 Ω / 7 mH / 8.3 mH
Magnetic flux linkage [Wb] 0.2515 400

Friction Coefficient [N·m·s] 0.03833 300

2
Moment of Inertia [kg·cm ] 20.5 200

100
TABLE II. ANGLE ESTIMATION BLOCK PARAMETERS 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Injection Frequency 1 kHz 1

Injection Signal Amplitude 250 mA 0.5

4th Order Butterworth Type 0

Band-Pass Filter LP = 800Hz HP= 1200Hz -0.5

1st Order Butterworth Type -1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

High-Pass Filter HP= 60 Hz

Fig. 6 Sensorless Speed Reversal (10 rad/s) at full load (12.2
a. Position Signals Nm). From top: Estimated (green) and reference speed (blue)
Fig. 4 and Fig. 5 correspond to steady state conditions, at (rad/s); current torque demand (A); estimated angle; angle error. All
1% of the nominal speed (3.14 rad/s) and full load torque angles in electrical degrees.
(12.2 Nm), which implies that the frequency of the position The angle error is kept within 0.5 electrical degrees
signals is at 3 Hz, considering the number of pole pairs and
(0.16 mechanical degrees) even during the speed reversal
the fact that the position signal frequency is twice the
and when the speed is set to zero. The angle estimation is
fundamental electrical frequency. The results showed on

19 PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 1a/2012

good enough and the SMC used eliminates properly the degrees. Further research is focused on modelling the VSI
perturbation included. The iQ current has an acceptable and a final implementation in order to obtain experimental
ripple value caused by the SMC, and any undesired effect results.
caused by the perturbation signal appears. The external PI
controller speed loop works properly in speed reversal as REFERENCES
shown in Fig. 6. [1] J . H o l t z . , Sensorless Control of Induction Machines- With
or Without Signal Injection?, IEEE Trans. on Ind. Electron..
c. Load Impact Vol.53, pp.7-30, Feb. 06.
The second experiment involves a Sensorless load [2] P . P . A c a r n l e y a n d J . F . W a t s o n . , A Review of
impact test with the speed reference set at zero. Fig. 7 Position Sensorless Operation of Brushless Permanent-
shows the speed response for a load impact of 100% and Magnet Machines, IEEE Trans. on Industrial Electron.. Vol.
back to 0 again. The angle error is kept at smaller values 53, pp. 352-362, April 06.
[3] R . W u , G . R . S l e m o n , A Permanent Magnet Motor Drive
than 4 electrical degrees (<1.3 mechanical degrees) Without a Shaft Sensor, IEEE Trans. on Industry Appl., vol.
taking into account the worst situation under the transient 27, pp. 1005-1011, Sep. / Oct. 1991.
response. [4] R . D h a o u a d i , N . M o h a n , L . N o r u m , Design and
Implementation of an Extended Kalman Filter for the State
400
Estimation of a Permanent Magnet Synchronous Motor, IEEE
300
Trans. on Power Electron., vol.6, pp.491-497, Jul.1991.
200
[5] C. Silva, G. M. Asher, M. Sumner, K.J. Bradley,
100
Sensorless control in a surface mounted PM machine using
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 rotating injection, EPE Journal. Vol. 13, No. 3, Aug. 03.
[6] A . A r i a s e t a l . , Use of a Matrix Converter to Enhance the
4
Sensorless Control of a Surface-Mount Permanent-Magnet
2
AC Motor at Zero and Low Frequency, IEEE Trans. on Ind.
0
Electron.. Vol. 53, pp 440-449, April 06.
-2 [7] J . C i l i a , e t a l , Sensorless Position Detection for Vector-
-4
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Controlled Induction Motor Drives using an Asymmetric
15 Outer-Section Cage, IEEE Trans. on Ind. A p p l . , V o l . 3 3 ,
10
pp.1162-1169, 1997.
5
[8] N. Teske, G.M. Asher, M. Sumner, K.J. Bradley,
0
Encoderless Position Estimation for Symmetric Cage
Induction Machines under Loaded Conditions, IEEE Trans.
-5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
on Ind. Appl., Vol. 37, no. 6, pp. 1793-1800, Nov/Dec 2001.
100
[9] P . L . J a n s e n , R . D . L o r e n z , Transducerless Position and
50
Velocity Estimation in Induction and Salient AC Machines,
0
IEEE Trans. on Ind. Appl., Vol.31, pp.240-247, March 1995.
[10] J . - S . K i m , S . - K . S u l , High performance PMSM drives
-50
without rotational position sensors using reduced order
-100
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 observer, IEEE-IAS Annual Meeting, pp 75-82, 1995.
[11] C.S. Staines, C. Caruana, G.M. Asher, M.
Fig. 7. Sensorless at zero Speed response to 100% load impact. S u m n e r , Sensorless Control of Induction Machines at Zero
From top: estimated rotor position (electrical degrees); angle error and Low Frequency Using Zero Sequence Currents, IEEE
(electrical degrees) ; iQ current(A); mechanical speed (rad/s) Trans. on Ind. Electron., Vol. 53, pp. 195-206, Feb. 06.
[12] J . H o l t z , J . J u l i e t , Sensorless Acquisition of the Rotor
Conclusions Position Angle of Induction Motors with Arbitrary Stator
This work has introduced an angle estimation algorithm Windings, IEEE Trans. on Ind. Appl.. Vol.41, pp. 1675-1682,
for Permanent Magnet Synchronous Machines (PMSMs) at Nov./Dec. 05.
low and zero speed based on the high frequency injection [13] P . L . J a n s e n , R . D . L o r e n z , Transducerless Position and
Velocity Estimation in Induction and Salient AC Machines,"
technique to track the machine saliency. The novelty of the
IEEE Trans. on Ind. Appl., Vol.31, pp.240-247, 1995.
work relies on the fact that the PMSM drive is under SMC [14] V . U t k i n , J . G u l d n e r , J . S h i , “Sliding Mode Control in
for the inner current loops instead of the traditional Field Electromechanical Systems” CRC PRESS 1999.
Oriented Control (FOC), based on PI controllers. [15] Z . Y a n , C . J i n , V . I . U t k i n , “Sensorless Sliding-Mode
The traditional voltage injection process used under FOC Control of Induction Motors” IEEE Trans. on Ind. Electron.,
schemes does not work in SMC due to the ideally infinite Vol. 47, pp. 1286-1297, Dec. 00.
band width of SMC controllers. Alternatively, the high
frequency test signal has been injected in the feedback
alpha- beta currents. Despite the elimination of the injected Authors: Mr Victor Repecho, Dr Domingo Biel and Dr Antoni Arias
are with the Institute of Organization and Control and the
HF current, the angle information can be obtained from the Department of Electronics Engineering, all from the Universitat
equivalent control voltage values generated by the SMC. Politècnica de Catalunya, Spain. e-mail: domingo.biel@ upc.edu,
Simulation results (speed reversal and load impact) [email protected], [email protected].
shows the validity of this angle estimation algorithm since
the angle error value is at all times less than 4 electrical

PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 1a/2012 20