Anda di halaman 1dari 6

Analysis of the Chosen Speed and Flux Estimators

for Sensorless Induction Motor Drive



Grzegorz Tarchala, Mateusz Dybkowski, Teresa Orlowska-Kowalska, Senior Member IEEE
Wroclaw University of Technology, Institute of Electrical Machines, Drives and Measurements, Wroclaw, Poland
E-mail: grzegorz.tarchala@pwr.wroc.pl, mateusz.dybkowski@pwr.wroc.pl, teresa.orlowska-kowalska@pwr.wroc.pl

Abstract-In the paper chosen rotor flux and speed estimation
methods for the speed sensorless induction motor drive are
presented. Speed estimators based on: the sliding mode theory
(SMO), model reference adaptive systems (MRAS) technique and
two chosen concepts based directly on the rotor slip frequency
calculation are discussed and compared. Stator flux and speed
estimators sensitivity to the motor parameter uncertainties is
investigated. Simulation and experimental results of the sensorless
DTC-SVM without speed control loop are presented. The low
speed region and the dynamic properties of the IM drive with
electromagnetic torque and stator flux feedbacks are tested.
I. INTRODUCTION
Less cabling, lower costs, great reliability and space
minimization, cause that sensorless induction motor (IM) drives
are getting one of the most popular drives in different industrial
branches. In the literature, estimators sensitivity to motor
parameter uncertainties is one of the most discussed problems
[1]-[4]. Thus different, new speed and flux estimation
techniques, less sensitive to the motor parameter changes and
stable in the whole reference speed changes, are proposed [3].
Universal speed and flux estimators should be stable in
whole speed reference range, with and without load torque
changes. Furthermore, these estimation techniques should
reconstruct real speed even in a very low speed region
including zero speed operation.
This paper deals with chosen speed estimation methods for
speed sensorless induction motor drives. Speed estimators,
like: Sliding Mode Observer (SMO) [5], Model Reference
Adaptive System-type observer, based on the Current Models
(MRAS
CC
) [6] and chosen concepts with direct rotor slip
frequency calculation, namely: Stator Flux Speed Observer
(SFSO) [7] and Dual Reference Frame Observer [8],
abbreviated to Dual-Mode (DM) observer, are discussed.
Speed and flux estimators are implemented within the
Direct Torque Control structure (DTC-SVM). This sensorless
control structure can be successfully used in traction drives
and other industrial applications.
Sensitivity of the discussed speed and flux estimators to motor
parameter changes is tested. Low speed operation and dynamic
properties of the IM drive with electromagnetic torque and stator
flux vector feedbacks are investigated. The range of stable work
of the control system with all tested estimators is shown.
II. SPEED AND FLUX ESTIMATION TECHNIQUES
Speed and rotor/stator flux vector can be estimated with
different techniques. Classification of the induction motor
state variables estimation methods is presented in Fig. 1.

Fig. 1. Classification of induction motor state variables estimation methods

All estimation methods discussed in this paper are of
algorithmic type and are marked grey in the Fig. 1. All
analyzed methods depend on motor parameters and thus are
sensitive to motor parameter uncertainty and works stable only
in a specified speed region. They allow estimating rotor or
stator flux vector, motor torque and speed but have different
complication scales and require special digital processors or
FPGA matrices for realization.
Among many different MRAS-based estimators the stator
current-based type (MRAS
CC
) seems to have the best
properties [6] therefore it was chosen for further
comparison. The Dual-Mode observer (DM) [8] is classified
as one of the sliding mode observers it takes the advantage
of sign function in order to estimate the flux vector
components. However, it utilizes the slip speed calculation
method to obtain the rotor speed, similarly as one of the state
simulators, SFSO [7]. Additionally, performances of the
Sliding Mode Observer (SMO) [5] were also compared.
III. MATHEMATICAL MODELS OF THE SPEED
ESTIMATORS
The analyzed speed and flux estimators are based directly
on the mathematical model of the IM [4].
The so-called rotor flux Voltage Model can be expressed
in coordinates, in a per unit system by the equation:


|

\
|
=
dt
d
x r
x
x
dt
d
T
s s
m
r
N
s
s s
u
r
i
i u
(1)

The Current Model of the rotor flux vector is described by:


978-1-4244-9311-1/11/$26.00 2011 IEEE 525


(

+ =
i
r
i
r s
i
r
i
m m
r
r
N
j ) (x
x
r
dt
d
T
(2)

Stator flux vector and motor torque can be calculated by:


s s
i u
s s N
r
dt
d
T =
(3)


s s s s e
i i m = (4)

where:
r
=
r
+j
r
,
s
=
s
+j
s
, u
s
=u
s
+ju
s
, i
s
=i
s
+ji
s

rotor and stator flux vectors, stator voltage and current
vectors, respectively,
r
s
, r
r
, x
s
, x
r
, x
m
stator and rotor resistances, stator and rotor
reactances, magnetizing reactance,
T
N
=1/(2f
sN
), f
sN
=50Hz, =1 x
m
2
/(x
s
x
r
)
m
e
,
m
motor torque and speed.
The general scheme of all analyzed speed and flux
estimators is presented in Fig. 2.

m

s
i
s
u

s
i

s r
,
s
i

Fig. 2. Schematic diagram of the analyzed speed estimators.

The biggest difference between presented solutions is a
methodology of the rotor speed calculation:
SFSO and DM use directly the rotor slip frequency
r
to
estimate the rotor speed:


r s m
=
(5a)


|

\
|
=

r r r r
r
N
s
dt
d
dt
d T

2
(5b)


( )

s r s r
r r
r m
r
i i
x
r x

2
=
(5c)


This solution is very sensitive to the IM parameter
changes, especially to the magnetizing reactance. Moreover
the estimated speed must be filtered before it can be used in a
control system (due to differentiation utilized in (5b)).
Variable structure system theory in the SMO is used
estimated speed must be filtered due to the chattering effect
(sign function introduces rapid changes in the output signal).
In the MRAS
CC
estimator the PI controller in the speed
adaptation mechanism is used. In this case the output filter is
not necessary.
A. SFSO Estimator
Stator Flux-based Speed Observer SFSO was proposed in
[7]. This estimator is based on modified stator flux estimator (3):


( )
s s AB s s s N
k r
dt
d
T i i i u
s

=
(6)

The stator current vector can be calculated from algebraic
relationship between stator and rotor flux vectors:


|
|

\
|
=
r s
i
1

r
m
s
s
x
x
x
(7)

After substitution (7) into (6), the stator flux vector can be
calculated from:


( )
s s AB s
r
m
s
s
N
k
x
x
x
r
dt
d
T i i u
r s s

+
|
|

\
|
+ =

(8)

Virtual rotor flux vector is then introduced:


( )
s s s N
r a
dt
d
a T i u
(I)
r
=
2 1

(9)

The relationship between rotor flux vector and the virtual
rotor flux vector can be directly obtained from (7):


1

a
s
i

(I)
r r
= (10)
where:

r s
r m
x x w
w
x
a
w
x
a


= = = ; ;
2 1
(11)

The main drawbacks of this estimator are correlations
between estimated stator and rotor flux vector components
leading to algebraic loops. Estimator can therefore be
unstable. Moreover equations (9) and (3) are identical.
The estimator scheme is presented in Fig. 3. The motor
speed is calculated from (5).

( )
s s
r
i

f
=
r s
m

=

Fig. 3. Schematic diagram of the SFSO speed estimator [7].

In control systems which require rotor flux vector knowledge
SFSO works simply like the rotor flux Voltage Model (1).
B. SMO Estimator
The Sliding Mode Observer [5] is based on the mathematical
model of the IM (1)(2). The speed
m
and an auxiliary
variable (introduced to minimize the rotor time constant
526

mistaken identification) are high-frequency signals obtained
using the sign function. The rotor flux vector is calculated as:


r r s r
r
i

C j
x
r x
x
r
dt
d
T
m
r
r m
r
r
N
+ + =
(12)

Stator current can be estimated by:


|
|
|
|
|

\
|

+
=
r r
r s s s
s

i i u
i

2 2
2

r
m
m
r
m
r
r m
r
m r
s
s
N
x
x
x
x
j
x
r x
x
x r
r
x dt
d
T
(13)

The rotor speed and additional variable are calculated from:

s
m
sign
0
= (14)

s sign
0
= (15)

while the suitable switching functions are:

( ) ( )


r s s r s s
i i i i s

= (16)

( ) ( )


r s s r s s
i i i i s

+ = (17)

where: C,
0
,
0
in (12), (14) and (15) positive constants to
be chosen.
The estimator scheme is presented in Fig. 4. The Current
Model and Current Estimator are adjusted by the high-
frequency speed signal.

f m,

s
i


K K ,
s
i
s
u
s
i r


s s ,
,
m
1
1
) (
) (
,
+
=
s T s
s
f m
f m


Fig. 4. Schematic diagram of the SMO speed estimator [5].

C. Dual-Mode estimator
Estimator proposed by I. Boldea in [8] is based on the
mathematical model of the IM in stator coordinates (-) and
in rotor coordinates (x-y). Current model expressed in a
synchronous frame (x-y) is independent on the rotor speed.
Stator and rotor flux models (18) and (21) are tuned using the
sliding-mode switching functions. Rotor and stator flux vectors
and stator current are expressed in the stationary frame -:


i u
s s s 1
K r
dt
d
T
s N
+ =
(18)

( )
s s
s
r
i
s
m
r
x
x
x
=
(19)

|
|

\
|
=
r s s
i
1

r
m
s
x
x
x
(20)

Rotor flux expressed in the x-y frame is:


r
x
r
rx
r
r r
sx
r s
r m r
rx N
K
x
r

x x
r x

dt
d
T
2
+ =

(21)

where:

( )
|
|

\
|
+ = =
p
k k s s sign
I p
1

), (
s s
i i
, (22)

where: K
1
, K
2
, k
p
, k
I
positive constants, 0 =
r
ry
.
The PI regulator in (22) is introduced in order to set to zero
the current estimation error. Speed is calculated from (5).
Rotor flux angle is calculated based on (19):

( )
s
r
s
r s
arctg

= (23)

The DM estimator, slightly more complicated than
previously presented structures, is shown in Fig. 5. Four
parameters must be determined: K
1
, K
2
, k
p
, k
I
.

s
i
s
u

s
i

, (
s s r
i f =

y x
y x
) , (

r s s
i f =
r s
m

=
( )
s
r
s
r

1
tg

s
r

s
i
2 1
, K K I p
k k ,
1
1
) (
) (
,
+
=
s T s
s
f m
f m

f m,

Fig. 5. Schematic diagram of the Dual-mode (DM) speed estimator [8].

D. MRAS
CC
estimator
MRAS-based estimator is one of the most popular ways of
the speed estimation [6]. It uses the PI regulator as an
adaptive mechanism. The schematic diagram of the MRAS
CC

estimator is presented in Fig. 6.
s
u
s
u s
i

s
i
s
i

s
i
s
i

s
i

s
i

s
i
m


Fig. 6. Schematic diagram of the MRAS
CC
speed estimator [6].
527

IV. SIMULATIONS AND EXPERIMENTAL RESULTS
All presented estimators were implemented in the Direct
Torque Control structure shown in Fig. 7. Torque and stator
flux vector were estimated by the Voltage Model (3)(4)
(during simulations parameters in this model were exactly the
same like in the IM).
S
b
Sc
SVM
x-y
Estimator
3
m
IM
Induction Motor
(Load Machine )
Inverter
Load
machine
En
ref
m
PI PI
PI

R
ref
s
s

s
d
u
S
a
sA
i
sB
i
d
u
, s
i
, s
u
s

SMO,MRAS,
Dual-Mode,
SFSO
abc
m

e
m
ref
e m
Torque generator

Fig. 7. The diagram of the DTC-SVM control system.

First, the estimators operation with ideal parameters
knowledge is investigated. Rapid and very long speed reverses
operations with load torque changes was tested and presented in
Fig. 8 and Fig. 10 (for the reference speed and load torque
changes shown in Fig. 9a).
0 0.5 1 1.5 2
-0.2
0
0.2
SMO
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
SFSO
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
MRAS
CC
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
Dual-mode
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m

Fig. 8. Measured, reference and estimated speed, simulation results.
2 4 6
-1
0
1
t [s]

m r
e
f

[
p
.
u
.
]
a)
0.5 1 1.5 2 2.5
0
0.5
1
t [s]

m r
e
f
,

m
o

[
p
.
u
.
]
b)
m
o

m
ref

Fig. 9. Test signals for estimators tests: a) long reversions, b) rapid reverse
with load torque change.
2 4 6
-5
0
5
x 10
-4
MRAS
t [ s ]

m

[

p
.
u
.

]
2 4 6
-5
0
5
x 10
-3
SMO
t [ s ]

m

[

p
.
u
.

]
2 4 6
-5
0
5
x 10
-4
SFSO
t [ s ]

m

[

p
.
u
.

]
2 4 6
-5
0
5
x 10
-4
Dual-Mode
t [ s ]

m

[

p
.
u
.

]

Fig. 10. Speed estimation errors for very long speed reverse operation

In Fig. 10 speed estimation errors for long speed reverse for
MRAS
CC
, SMO, SFSO, and DM are presented, respectively.
All estimators work correctly errors between the real and
estimated speeds are almost zero and can be neglected.
Afterwards, the analyzed estimators sensitivity to motor
parameter variations is tested. Like previously, the estimators
are tested for rapid and very long speed reverses. In Fig. 11 the
rapid reverse operation with simultaneous +10% stator and
rotor resistances mismatch is presented. The estimators do not
lose stability. SMO and MRAS
CC
reconstruct the rotor speed
with smaller error than SFSO and DM estimators. Oscillations
of estimated speed are visible in SFSO and DM operation
even for so small resistance parameter mismatch. For bigger
changes, especially stator resistance, SFSO looses the
stability, so in the next tests only small parameter changes
were introduced to obtain the stable operation of the system.

0 0.5 1 1.5 2
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
SMO
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
SFSO
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
MRAS
CC
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m
0 0.5 1 1.5 2
-0.2
-0.1
0
0.1
0.2
0.3
Dual-mode
t [s]

m
,

m e
s
t
,

m r
e
f

[
p
.
u
.
]



m
est

ref
m

m

Fig. 11. Reference, measured and estimated speeds under windings
resistance mismatch: rs
obs
= 1.1rs
R
, rr
obs
= 1.1rr
R
, simulation results
528

1 2 3 4 5 6 7 8
-0.02
0
0.02
MRAS
t [ s ]

m

[

p
.
u
.

]
1 2 3 4 5 6 7 8
-0.02
0
0.02
SMO
t [ s ]

m

[

p
.
u
.

]
1 2 3 4 5 6 7 8
-1
0
1
SFSO
t [ s ]

m

[

p
.
u
.

]
1 2 3 4 5 6 7 8
-1
0
1
Dual-Mode
t [ s ]

m

[

p
.
u
.

]

Fig. 12. Speed estimation errors for very long speed reverse operation under
windings resistance mismatch: rs
obs
= 1.1 rs
R
, rr
obs
= 1.1 rr
R

2 4 6 8
-0.2
0
0.2
MRAS
t [s]

s

[
p
.
u
.
]
2 4 6 8
-0.2
0
0.2
SMO
t [s]

s

[
p
.
u
.
]
2 4 6 8
-0.5
0
0.5
SFSO
t [s]

s

[
p
.
u
.
]
2 4 6 8
-0.5
0
0.5
Dual-Mode
t [s]

s

[
p
.
u
.
]

Fig. 13. Stator flux magnitude estimation errors for very long speed reverse
operation under resistance mismatch: rs
obs
= 1.1 rs
R
, rr
obs
= 1.1 rr
R
2 4 6 8
-0.2
0
0.2
MRAS
t [s]

s

[
r
a
d
]
2 4 6 8
-0.2
0
0.2
SMO
t [s]

s

[
r
a
d
]
2 4 6 8
-0.5
0
0.5
SFSO
t [s]

s

[
r
a
d
]
2 4 6 8
-0.5
0
0.5
Dual-Mode
t [s]

s

[
r
a
d
]

Fig. 14. Stator flux angle estimation errors for long speed reverse operation
under windings resistance mismatch: rs
obs
= 1.1 rs
R
, rr
obs
= 1.1 rr
R

In the Fig. 12 speed estimation errors for a long speed reverse
operation under the same resistances mismatch: r
s
obs
= 1.1r
s
R
,
r
r
obs
= 1.1r
r
R
are presented (where upper superscript R means
nominal or real parameter value). Flux angle and magnitude
errors are presented in Fig. 13 and Fig. 14, respectively.
It can be seen that SFSO and Dual-Mode estimators are
more sensitive to both resistances uncertainties than SMO
and MRAS
CC
. Estimated speed and flux oscillations in error
signals are visible. Those observers can be unstable for
bigger parameter changes.
For more accurate sensitivity study the insightful tests were
performed and the maximal and steady-state (bold fonts)
estimation errors are shown in the Table I. The test signal for this
study is shown in Fig. 9b. During this test the load torque
nominal value was turned on and off. The three main motor
parameters were changed increasingly: r
s
, r
r
, and x
m
this
situation corresponds to the heating of the motor windings and
magnetic saturation occurring.
TABLE I
ESTIMATION ERRORS FOR PARAMETERS UNCERTAINTIES
Estimator MRAS
CC
SMO SFSO DM
error type max
steady-
state
max
steady-
state
max
steady-
state
max
steady-
state
m [p.u.] 0.027 ~0 0.047 ~0 0.299 0.23 0.305 0.23
s [p.u.] 0.092 ~0 0.092 ~0 0.390 0.20 0.390 0.15
rs=
+10%
s [rad] 0.052 0.004 0.051 0.004 0.334 0.20 0.39 0.20
m, [p.u.] 0.024 ~0 0.052 ~0 0.015 ~0 0.019 ~0
s [p.u.] 0.010 ~0 0.002 ~0 ~0 ~0 ~0 ~0
rr=
+10%
s [rad] 0.006 0.005 0.002 0.004 0.001 ~0 0.001 ~0
m, [p.u.] 0.029 ~0 0.055 ~0 0.008 ~0 0.008 ~0
s [p.u.] 0.10 0.018 0.099 0.018 ~0 ~0 ~0 ~0
xm=
+20%
s [rad] 0.19 0.19 0.19 0.19 ~0 ~0 ~0 ~0

Due to oscillations arising, the steady-state error was
calculated as a mean value of the absolute error. The obtained
investigation results (Table I) divides presented estimators into
two groups. The first group is formed from MRAS
CC
and SMO.
This group is characterized by a relatively small steady-state
errors (only the angle estimation error for magnetizing reactance
mismatch is noticeable) and slightly bigger dynamical errors.
The second one is formed from SFSO and DM estimators. Main
characteristic feature of this group is high sensitivity to the stator
resistance erroneous identification and low sensitivity to the
magnetizing reactance changes. Maximum errors for these two
estimators for windings resistance changes are also bigger than
for the first group. It should be mentioned that for bigger
resistance mismatch these estimators can even loose stability.
In the Fig. 15 the speed estimation errors for magnetizing
reactance mismatch is shown for all discussed estimators. It can
be seen that after transient caused by the reference speed change
(in t=1.5 s), this error disappears completely for MRAS
CC
and
SMO, while a very small steady-state errors occur for SFSO and
DM estimators.
529

1 1.5 2 2.5
-0.1
0
0.1
MRAS
t [ s ]

m

[

p
.
u
.

]
1 1.5 2 2.5
-0.1
0
0.1
SMO
t [ s ]

m

[

p
.
u
.

]
1 1.5 2 2.5
-0.01
0
0.01
SFSO
t [ s ]

m

[

p
.
u
.

]
1 1.5 2 2.5
-0.01
0
0.01
Dual-Mode
t [ s ]

m

[

p
.
u
.

]

m,max

m,max

m,max

m,max

Fig. 15. Sensitivity study results for magnetizing reactance mismatch
xm
obs
= 1.2 xm
R

The MRAS-type and SMO estimators were tested for
different drive conditions in the laboratory set-up with
control and estimation structure realized using DS1103 card.
In the following figures chosen experimental results for
DTC-SVM control structure are presented. The estimators are
tested for cyclic reference torque changes (Fig. 16).
Estimated speed can be used in the diagnostic process and in
the field weakening algorithms. Both estimators depend on
the estimated speed (they are adjusted by this value) so the
control structure depends indirectly on this value.

a) b)

Fig. 16. Experimental transients of the sensorless DTC SVM control with
SMO (a) and MRAS
CC
(b) for mref = 0.5

a) b)

Fig. 17. Measured, reference and estimated speed for: (a) MRAS
CC
for
m=0.001mN and (b) SMO for m=0.01mN


a) b)

Fig. 18. Measured, reference and estimated speed for m=0.05mN for
a) MRAS
CC
and b) SMO

Both estimators work stable for different torques and in the
whole speed range including zero speed region. In Fig. 17
and Fig. 18 a possibility of speed estimation in a very low
speed region is illustrated.

V. CONCLUSIONS
In the paper the chosen speed and flux estimators were
presented SMO, MRAS
CC
, SFSO, and DM. Their
sensitivity to the IM parameter uncertainties and stable work
range were tested. All solutions work stable (with minimal
errors) for nominal parameters but for mismatched values
only two first are stable in the whole speed region. SMO and
MRAS
CC
speed and flux estimators were tested in the
labolatory set-up. MRAS
CC
estimator can work stable for
lower speed region than SMO, but both solutions are good
proposition for industrial applications especially for torque
control methods DTC-SVM.

ACKNOWLEDGMENT
This research work was partially supported by the Ministry of Science
and Higher Education (Poland) under Grant N510 334637 (2009-2011)
REFERENCES
[1] P. Vas, Sensorless vector and direct torque control, Oxford University
Press, New York, 1998.
[2] M.P. Kazmierkowski, F. Blaabjerg, and R. Krishnan, Control in Power
Electronic Selected Problems, Academic Press, USA, 2002
[3] J. Holtz, Sensorless control of induction machines - with or without
signal injection?, IEEE Trans. on Industrial Electronics, vol. 53, no. 1,
pp.7-30, 2006
[4] T. Orlowska-Kowalska, Sensorless Induction Motor Drives, Wroclaw
University of Technology Press, 2003
[5] V. Utkin, Z. Yan, Sliding Mode Observers for Electric Machines An
Overview, Proc. of the 28
th
Annual Conf. of the Industrial Electronics Soci-
ety IECON2002, Seville, Spain, vol. 3, no. 2, pp. 1842-1847
[6] T. Orlowska-Kowalska, M. Dybkowski, Stator Current-based MRAS
Estimator for Wide Range Speed-Sensorless Induction Motor Drive,
IEEE Trans. on Industrial Electronics, vol. 57, no. 4, April 2010, pp.
1296-1308
[7] H. Abu-Rub, N. Oikonomou, Sensorless Observer System for Induction
Motor Control, Proc. of the IEEE Power Electronics Specialists Con-
ference, PESC2008, Rodos, Greece, pp. 30-36
[8] I. Boldea, C. Lascu, F. Blaabjerg, A Class of Speed-Sensorless Sliding-
Mode Observers for High-Performance Induction Motor Drives, IEEE
Trans. on Industrial Electronics, vol. 56, no. 9, September 2009, pp.
3394-3403
530

Anda mungkin juga menyukai