IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO.

5, MAY 2009 1699

Wide-Speed-Range Estimation With Online
Parameter Identication Schemes of Sensorless
Induction Motor Drives
Mohamed S. Zaky, Mahmoud M. Khater, Shokry S. Shokralla, and Hussain A. Yasin
AbstractRecently, the development of speed estimation meth-
ods for sensorless control of induction motor drives has found
great interest in the research community. Parameter adaptation
schemes play an important role for better speed estimation over a
wide range fromzero to high levels beyond the rated speed. There-
fore, parallel identication schemes for both speed and stator
resistance of sensorless induction motor drives are proposed for a
wide range of speed estimation. These estimation algorithms com-
bine a sliding-mode current observer with Popovs hyperstability
theory. Low- and zero-speed operations of the proposed sliding-
mode-observer (SMO)-based speed estimation combined with an
online stator resistance adaptation scheme are investigated. A
modied SMO-based speed estimation scheme for eld-weakening
operation is also introduced. The mismatch problem of magne-
tizing inductance in the eld-weakening region is treated by an
online identication scheme. Magnetizing inductance, estimated in
this way, is further utilized within the SMO, so that the main ux
saturation variation is taken into consideration. The performance
of the proposed SMO and its speed estimation accuracy, with an
indirect eld-oriented controlled induction motor, are veried by
simulation and experimental results over a wide speed range from
zero to high values beyond the base speed.
Index TermsMagnetizing inductance identication, sliding-
mode observer (SMO), speed sensorless, stator resistance
identication.
I. INTRODUCTION
S
EVERAL methods have been recently proposed for speed
estimation of sensorless induction motor drives [1], [2].
They can be classied into two major categories. The rst one
includes the techniques that estimate the rotor speed based on
nonideal phenomena such as rotor slot harmonics and high-
frequency signal injection methods [3]. Such methods require
spectrum analysis which, besides being time-consuming pro-
cedures, allow a narrow band of speed control. The second
category of speed estimation methods relies on the model of the
induction motor. The supremacy of a certain method depends
on its estimation accuracy over a wide speed range. Model-
based methods of speed estimation are characterized by their
simplicity, but sensitivity to parameter variations is considered
Manuscript received January 31, 2008; revised October 23, 2008. First
published November 18, 2008; current version published April 29, 2009.
The authors are with the Department of Electrical Engineering, Faculty
of Engineering, Minouya University, 32511 Shebin El-Kom, Egypt (e-mail:
m_s_zaky@yahoo.com; khatermm@yahoo.com; shokrysaad1@yahoo.com).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TIE.2008.2009519
the major problem associated with them. Stator resistance plays
an important role, and its value has to be known with good
precision in order to obtain accurate speed estimation in the
low-speed region [4].
The interest in stator resistance adaptation came on the scene
much recently, with the advances of speed sensorless systems.
It has also received more attention with the introduction of the
direct torque control technique. An accurate value of the stator
resistance is of crucial importance for the correct operation of a
sensorless drive in the low-speed region, since any mismatch
between the actual value and the set value used within the
model of speed estimation may lead not only to a substantial
speed estimation error but also to instability [5][7]. Therefore,
there is a great interest in the research community to develop
stator resistance identication schemes for accurate low-speed
estimation. An ofine estimation procedure of the motor pa-
rameters at various operating conditions is introduced in [7].
In that work, the stator resistance is observed to be varying
as a function of stator temperature. The rotor resistance is also
estimated ofine for rotor eld orientation as a function of slip
frequency and rotor temperature.
Numerous online estimation techniques are also proposed
for continuously updating parameter values. An online rotor
resistance estimation for rotor eld orientation using a sliding-
mode observer (SMO) and a model reference adaptive system
(MRAS) is proposed in [8][10]. The online stator resistance
identication schemes can be classied into a couple of distinct
categories. These schemes rely on stator current measurement
and mostly require information regarding stator voltages [11]
[15]. The rst category includes different types of estimators
which often use an adaptive mechanism to update the value of
stator resistance [11][14]. The stator resistance is determined
in [11] by using a reactive-power-based MRAS. The reactive
power relies on the accuracy of other parameters, such as leak-
age inductance and rotor resistance, which are not necessarily
constant, and the result is prone to error. Adaptive full-order
ux observers (AFFOs) for estimating the speed and stator re-
sistance are developed using Lyapunovs stability criterion [12],
[13]. While these schemes are not computationally intensive,
an AFFO with a nonzero gain matrix may become unstable.
An MRAS for estimating the speed and stator resistance is
developed using Popovs stability criterion [14], [15]. Recently,
two extended Kalman lter (EKF) algorithms for estimating
stator and rotor resistances are utilized in a braided manner,
thus achieving an accurate estimation of a high number of
0278-0046/$25.00 2009 IEEE
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
1700 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY 2009
parameters and states than would have been possible with a
single EKF algorithm [16]. The second category of online
stator resistance identication schemes depends on articial
intelligence techniques in the process of stator resistance adap-
tation. Articial neural networks for estimating stator and rotor
resistances are used for this purpose [17].
High-performance applications of sensorless systems require
high accuracy of speed estimation over a wide speed range
extending from very low and zero-speed operations to high
speeds beyond the rating. Operation of eld-oriented induction
motors below the base speed is usually achieved with constant
ux. Therefore, magnetizing inductance can be regarded as
constant and equal to its rated value. In the eld-weakening
region, the rotor ux reference has to be reduced below its rated
value. Variation of the rotor ux reference implies variable level
of saturation, and consequently, magnetizing inductance of the
machine is varied [18], [19]. Accurate value of magnetizing
inductance is of utmost importance for many reasons. The
rst one is the correct setting of the d-axis stator current
reference in a vector-controlled drive. The second one is its im-
portance for accurate speed estimation in the eld-weakening
region, using machine-model-based approaches, of sensorless
systems. The third reason is the dependency of rotor time
constant identication schemes on magnetizing inductance,
such as the method [22] utilizing reactive power. The accu-
rate estimation of rotor time constant in the eld-weakening
region requires a correct value of the magnetizing inductance to
be known.
Many research works have been devoted to improve speed
estimation of a eld-oriented controlled induction motor in the
constant ux operation region [3][17]. However, little interest
has been given for magnetizing inductance identication to im-
prove speed estimation in the eld-weakening region [18][23].
Magnetizing inductance is represented by a quadratic polyno-
mial curve tting to improve dynamic response in both steady
and transient states [18]. An estimation of magnetizing induc-
tance at various operating conditions is introduced in [7]. A
correlation between the magnetizing inductance and saturation
is observed, and a polynomial to represent this dependency
is proposed. Attempts for reducing the speed estimation error
in the eld-weakening region using a Kalman lter and a
Luenberger observer are given in [19] and [20]. Terminal
signals of stator voltages and currents combined with the
magnetizing curve of the machine can be used for online
magnetizing inductance identication [21][23]. This method
is characterized by its simplicity and accuracy; therefore, it is
used here with the proposed SMO to improve speed estimation
in the eld-weakening region.
This paper presents a speed estimation algorithm based
on a sliding-mode current observer which combines variable-
structure control, Lyapunovs stability, and Popovs hypersta-
bility theories. Since a speed estimation algorithm should be
insensitive to parameter variations, particularly stator resistance
for low- and zero-speed operations, a stator resistance identi-
cation scheme based on the same theories is also developed and
used in parallel with the speed estimator. A modied SMO with
an online identication scheme of magnetizing inductance is
presented to operate in the eld-weakening region. A control
board, based on a TMS320C31 oating-point digital signal
processor (DSP), is used for implementing the proposed SMO
speed estimation algorithm. Experimental results are presented
along with those obtained by simulation to evaluate the accu-
racy of the proposed SMO over a wide speed range from zero
to high values beyond the base speed.
II. SPEED AND STATOR RESISTANCE
ESTIMATION PROCEDURE
A. Construction of SMO
The induction motor can be represented by its dynamic
model expressed in the stationary reference frame in terms of
stator current and rotor ux as follows:
d
dt
_
i
s
s

s
r
_
=
_
A
11
A
12
A
21
A
22
_ _
i
s
s

s
r
_
+
_
b
1
0
_
[v
s
s
] = Ax + Bv
s
(1)
where A
11
, A
12
, A
21
, A
22
, and b
1
are given in the Appendix.
With reference to the introduced mathematical model and
considering the stator currents as the system outputs, the
sliding-mode current observer can be constructed as
p

i
s
s
=

A
11

i
s
s
+

A
12

s
r
+ b
1
v
s
s
+ Ksgn
_

i
s
s
i
s
s
_
(2)
p

s
r
=A
21

i
s
s
+

A
22

s
r
(3)
where K is the switching gain.
The error equation which takes into account parameter vari-
ation can be expressed, by subtracting (1) from (2), as follows:
de
i
dt
= A
11
e
i
+A
12
e

+A
11

i
s
s
+A
12

s
r
+Ksgn
_

i
s
s
i
s
s
_
(4)
where
e
i
=

i
s
s
i
s
s
, e

s
r

s
r
A =
_
A
11
A
12
A
21
A
22
_
.
The sliding surface S is constructed as
S(t) = e
i
=

i
s
s
i
s
s
= 0 (5)
whereas the switching function of SMO is dened as
sgn(S) =
_
1, if S 0
1, if S < 0.
(6)
If rotor speed and stator resistance are considered as variable
parameters, assuming no other parameter variations, the matrix
A is expressed as follows:
A
11
=
R
s
I
L
s
A
12
=

r
J

A
21
=0
A
22
=
r
J.
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
ZAKY et al.: WIDE-SPEED-RANGE ESTIMATION WITH ONLINE IDENTIFICATION SCHEMES OF MOTOR DRIVES 1701
The sliding mode occurs when the following sliding condi-
tion is satised:
e
T
i
e
i
< 0. (7)
B. Characteristics of SMO on Sliding Surface
When the estimation error trajectory reaches the sliding sur-
face, i.e., S = 0, then, from (5), it is obvious that the observed
currents will converge to the actual ones, i.e.,

i
s
s
= i
s
s
.
According to the equivalent control concept, assuming that
the observed currents

i
s
s
match the actual currents i
s
s
in the
steady state, then
e
T
i
= e
i
= 0 (8)
from which the error equation becomes
0 = A
12
e

+ A
11

i
s
s
+ A
12

s
r
L. (9)
The estimation algorithm of stator currents is constructed by
a closed-loop observer, as in (2), whereas the estimation of rotor
uxes is carried out by an open loop represented by (3) without
the ux error. Therefore, the real and estimated rotor uxes are
assumed the same

s
r
=
s
r
; thus, the error equation becomes as
follows:
0 = A
11

i
s
s
+ A
12

s
r
L. (10)
C. Stability of the Identication System
Popovs hyperstability theory is well known as stability crite-
rion for nonlinear feedback systems. This theory is applied here
to examine the stability of the proposed identication system.
This requires that the error system and the feedback system are
derived so that the theory could be applied.
In the SMO, using a speed identication error
r
=
r

r
, a stator resistance identication error R
S
=

R
S
R
S
,
and an error signal L = K sgn(

i
s
s
i
s
s
), the error system
from (10) is written as
L = A
11

i
s
s
+ A
12

s
r
. (11)
Substituting the given equivalents of A
11
and A
12
into (11)
yields
L =
R
s
L
s

i
s
s

r
J

s
r
= z
1
R
s
z
2

r
(12)
where
z
1
=(1/L
s
)

i
s
s
(13)
z
2
=(J/)

s
r
. (14)
Popovs integral inequality of (11) is written as follows [14]:
S =
t
0
_
0
L
T
Wdt
2
, = const (15)
Fig. 1. Identication system for speed and stator resistance.
where L
T
is the input vector; W = z
1
R
s
z
2

r
, which
represents the nonlinear block, is the output vector of the
feedback block; and is a nite positive constant which does
not depend on t
0
S =
t
0
_
0
L
T
Wdt =
t
0
_
0
L
T
(z
1
R
s
z
2

r
)dt. (16)
Substituting (13) and (14) into (16) yields
S =
t
0
_
0
L
T
Wdt
=
t
0
_
0
_
L
T
R
s
L
s

i
s
s
_
dt +
t
0
_
0
_
L
T

r
J

s
r
_
dt
(17)
S =S
1
+ S
2

2
(18)
S
1
=
t
0
_
0
_

L
T
R
s
L
s

i
s
s
_
dt
2
1
(19)
S
2
=
t
0
_
0
_
L
T

r
J

s
r
_
dt
2
2
. (20)
The validity of Popovs inequality of (18) can be veried by
means of the inequalities expressed by (19) and (20), provided
that the estimates of rotor speed and stator resistance can be
obtained by (21) and (22), respectively,

r
=K

_
L
T
J

r
dt (21)

R
s
=K
R
_
L
T
i
s
s
dt (22)
where K

and K
R
are adaptive gains.
An identication system for speed and stator resistance
is shown in Fig. 1, which is constructed from a linear
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
1702 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY 2009
Fig. 2. Block diagram of speed, stator resistance, and magnetizing inductance
identication schemes.
time-invariant forward block and a nonlinear time-varying feed-
back block. The system is hyperstable if the forward block is
positive real and the input and output of the nonlinear feedback
block satisfy Popovs integral inequality. Fig. 2 shows the
block diagram of parallel speed and stator resistance estima-
tion algorithms based on a combination of SMO and Popovs
hyperstability theory.
III. ONLINE IDENTIFICATION ALGORITHM OF
MAGNETIZING INDUCTANCE
Magnetizing inductance of induction motors may vary
signicantly when the rotor ux reference varies in the eld-
weakening region. Accurate speed estimation in this region re-
quires the precise value of magnetizing inductance. Therefore,
the structure of the speed observer should be modied in such
a way that the variation of main ux saturation is recognized
within the speed estimation algorithm. This requires the online
identication algorithm of the magnetizing inductance.
The magnetizing inductance is given on the basis of the
known magnetizing curve of the machine with
L
m
=
m
/i
m
(23)
where

m
=
_

2
dm
+
2
qm
. (24)
The magnetizing ux vector can be obtained as follows:

dm
=
_
(v
ds
R
s
i
ds
)dt L
s
i
ds
(25)

qm
=
_
(v
qs
R
s
i
qs
)dt L
s
i
qs
. (26)
Since the magnetizing ux is now known, it is possible to
estimate the magnetizing inductance using the known nonlinear
inverse magnetizing curve
i
m
=f(
m
) (27)

L
m
=

m
i
m
. (28)
Fig. 3. Magnetizing curve and magnetizing inductance variation of the ma-
chine used in experiments.
Fig. 4. Block diagram of the experimental system.
The block diagram of online identication procedures of the
magnetizing inductance is shown in Fig. 2. The magnetizing
curve of the machine was identied ofine, as described in [21],
and is represented with a suitable analytical function in per unit,
of the form
i
m(pu)
= a
m(pu)
+ (1 a)
b
m(pu)
. (29)
Coefcients a and b of (29) were determined in [21] as a =
0.9 and b = 7. The rated magnetizing current is 0.85 A (rms),
and the rated magnetizing ux is 0.84524 Wb (rms). The rated
magnetizing inductance value is 0.9944 H. Using (29) and the
given data, the required

L
m
is easily obtained. The magnetizing
curve of the machine and the magnetizing inductance variation
are shown in Fig. 3.
IV. SYSTEM IMPLEMENTATION
The basic conguration of the experimental system is shown
in Fig. 4. It consists of an induction motor interfaced with a
digital control board DS1102 based on a Texas Instruments
TMS320C31 DSP for speed estimation. The induction motor
is coupled with a dc generator for mechanical loading. The
rating and parameters of the induction motor are given in the
Appendix. Stator terminal voltages and currents are measured
and ltered using analog circuitry. Hall-effect sensors are used
for this purpose. The measured voltage and current signals are
acquired by the A/D input ports of the DSP control board. This
board is hosted by a personal computer on which mathematical
algorithms are programmed and downloaded to the board for
real-time speed estimation. A direct speed measurement is also
carried out for comparison with the estimated speed. The output
switching commands of the DSP control board are obtained
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
ZAKY et al.: WIDE-SPEED-RANGE ESTIMATION WITH ONLINE IDENTIFICATION SCHEMES OF MOTOR DRIVES 1703
Fig. 5. IFO controller with compensation of ux saturation.
Fig. 6. Speed estimation error for +20% R
s
error in the observer at
(a) 150 rad/s and (b) 3 rad/s.
Fig. 7. (a) Stator resistance adaptation for 10% and 50% R
s
initial
detuning. (b) Actual and estimated speeds, and speed error for 50% R
s
initial
detuning with stator resistance adaptation.
via its digital ports and interfaced with the inverter through
optoisolated gate drive circuits.
V. RESULTS AND DISCUSSION
A. Simulation Results
A sensorless indirect eld-oriented (IFO) controlled induc-
tion motor drive, shown in Fig. 5, is used, where the actual
speed feedback signal is replaced by the estimated one. The
sensitivity to stator resistance mismatch is shown in Fig. 6 for
+20% R
s
error at high and low speeds. These results show that
the speed estimation error at high-speed operation (150 rad/s)
is 1.9 rad/s (1.26%), and that at low-speed operation (3 rad/s)
is 0.32 rad/s (10.7%). Large error at low speeds may cause
instability. In order to avoid this, the online stator resistance
adaptation scheme (22) has been applied. The initial detuning in
the stator resistance takes values of 10% and 50%, as shown
in Fig. 7(a). In both cases, the stator resistance adaptation
was activated at t = 0. It is clear that the stator resistance
Fig. 8. Actual and estimated speeds, and speed estimation error at speed
command of 2 rad/s. Stator resistance adaptation is activated at t = 1 s.
(a)

R
so
= 1.5R
s
. (b)

R
so
= 1.3R
s
.
Fig. 9. Actual and estimated speeds, and speed estimation error at speed
command of 1 rad/s. Stator resistance adaptation is activated at t = 1 s.

R
so
= 1.3R
s
.
estimator quickly removes the initial stator resistance error and
consequently eliminates the large speed estimation error. A
considerable reduction of the speed error is observed with stator
resistance adaptation due to 50% initial R
s
mismatch in the
observer, as shown in Fig. 7(b).
Fig. 8 shows the actual and estimated speeds as well as the
speed estimation error for two initial levels of stator resistance
mismatch at a speed reference of 2 rad/s under no-load condi-
tion. The initial detuning in the stator resistance takes values of
+50% and +30%. The same simulation results are obtained at
a speed reference of 1 rad/s under no-load condition, as shown
in Fig. 9, with +30% initial detuning in the stator resistance. As
shown, there exists a substantial speed estimation error between
the actual and estimated speeds. The stator resistance adapta-
tion scheme is turned on at t = 1 s. It is clear that activation
of the stator resistance adaptation scheme quickly compensates
for the initial error in the estimated stator resistance value and
therefore eliminates the initial speed estimation error.
The performance of identication schemes is also tested
during speed reversal in the low-speed region. Fig. 10(a) shows
the actual and estimated speeds as well as the speed estimation
error during speed reversal from 0.5 to 0.5 rad/s. The correct
value of stator resistance leads to elimination of the speed
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
1704 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY 2009
Fig. 10. Actual and estimated speeds, and speed estimation error with stator
resistance tuning (a) during speed reversal from 0.5 to 0.5 rad/s and (b) at
zero speed.
Fig. 11. Actual and estimated speeds, and speed estimation error for step
change of speed command to 200 rad/s (a) with constant parameter SMO and
(b) with modied SMO.
estimation error, and the actual and estimated speeds are in
very good agreement in the steady state with a considerable
reduction of the speed error (12.5%) during transients to zero
in 0.25 s. Fig. 10(b) shows the actual and estimated speeds
as well as the speed estimation error at zero speed with stator
resistance tuning. As shown, the proposed speed observer with
stator resistance adaptation achieves good speed estimation.
Furthermore, results conrm that due to the accurate stator
resistance estimation, the drive does not lose stability during
operation at low and zero speeds. Finally, simulation results
prove the supremacy of the parallel speed and stator resistance
identication schemes to provide an accurate speed estimate in
the very low speed region and also at zero speed.
The IFO controller of Fig. 5 is also tested for the operation in
the eld-weakening region. Simulations are at rst performed
using the SMO of Fig. 2 with the magnetizing inductance set
to the constant rated value which is suitable for the base-speed
region. Speed reference is rst applied to its rated speed value
which equals 150 rad/s. At t = 1.2 s, the speed command is
increased in step change to 200 rad/s, as shown in Fig. 11(a).
This gure shows that the estimated speed tracks the actual
speed in the base-speed region with minimum speed estimation
error. However, once the eld-weakening region is entered at
t = 1.2 s, a signicant speed estimation error occurs due to
variation of the magnetizing inductance that is neglected in the
speed estimator. As shown, the speed estimation error changes
Fig. 12. Actual and estimated speeds, and speed estimation error for step
change of speed command to 250 rad/s (a) with constant parameter SMO and
(b) with modied SMO.
from 0.6 to 2.25 rad/s at the nal speed of 200 rad/s. The
proposed modied SMO with online magnetizing inductance
identication is now used instead of the constant parameter
one. The observable differences of the actual and estimated
speeds when compared with those of Fig. 11(a) are illustrated
clearly from the speed estimation error shown in Fig. 11(b).
It is observed from comparing the speed estimation error of
Fig. 11(a) and (b) that the application of the modied SMO
with online magnetizing inductance essentially eliminates the
speed estimation error and gives accurate speed estimation in
the eld-weakening region. The same simulation results are
repeated during step change from 150 to 250 rad/s in both
cases, with constant magnetizing inductance [Fig. 12(a)] and
with online magnetizing inductance identication instead of
the constant parameter one [Fig. 12(b)]. Simulation results of
Figs. 11 and 12 prove the ability of the modied SMO to
adapt to the actual saturation level in the machine and therefore
provide an accurate speed estimate for any operating point in
the eld-weakening region.
B. Experimental Results
Since motor heating usually causes a considerable varia-
tion in the winding resistance, so there is often a mismatch
between the actual stator resistance and its corresponding set
value within the model used for speed estimation. For this
purpose, the proposed SMO with a stator resistance identi-
cation scheme is rst tested under different values of stator
resistance to represent this parameter mismatch. Fig. 13 shows
the estimated speed at 10 rad/s at +50% stator resistance mis-
match with activated stator resistance adaptation. It is observed
that the estimated speed preserves its value unchanged under
this parameter mismatch. This test proves that the proposed
SMO with online stator resistance tuning is dependable and
accurately gives the same behavior as the actual speed under
stator resistance mismatch.
The effectiveness of the stator resistance adaptation algo-
rithm is tested during reversing transients; this corresponds to
a very short term operation at zero speed. Fig. 14 shows the
actual and estimated speeds for a reversing transient from 100
to 100 rad/s with and without stator resistance adaptation. The
results with stator resistance adaptation mechanism illustrate
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
ZAKY et al.: WIDE-SPEED-RANGE ESTIMATION WITH ONLINE IDENTIFICATION SCHEMES OF MOTOR DRIVES 1705
Fig. 13. Estimated speed at 10 rad/s under +50% mismatch of stator resis-
tance with stator resistance adaptation.
Fig. 14. Actual and estimated speeds during speed reversal at 100 rad/s
(a) without stator resistance adaptation and (b) with stator resistance adaptation.
Fig. 15. Actual and estimated speeds at zero speed (a) without stator resis-
tance adaptation and (b) with stator resistance adaptation.
that a good estimated speed during reversing transient through
zero speed is achieved compared with the same situation in
Fig. 14(b). Persistent operation at zero speed is possible experi-
mentally, as shown in Fig. 15, with and without stator resistance
adaptation. The stator resistance identication scheme is also
examined at very low speeds. Fig. 16 shows the experimen-
tal results during steady-state operation at 2.5 rad/s (0.8 Hz,
0.016 p.u.). This gure shows good speed estimation due to the
introduction of stator resistance adaptation.
Fig. 16. Actual and estimated speeds during steady-state operation at 2.5 rad/s
(0.8 Hz, 0.016 p.u.) with stator resistance adaptation.
Fig. 17. Actual and estimated speeds during startup operation at (a) 150-rad/s
and (b) 200-rad/s speed commands with modied SMO.
The introduced results demonstrate that the system exhibits a
good robustness and high speed estimation accuracy under dif-
ferent operating conditions and with stator resistance mismatch.
This is possible due to the strong robustness of the SMO with a
stator resistance adaptation scheme.
Aseries of tests are also performed experimentally in order to
verify the accuracy of the modied speed observer for operation
in the eld-weakening region. The base speed, at which the
eld-weakening region is initiated, is set in all the tests to
150 rad/s. The starting operation of the drive is rst examined.
Fig. 17 shows the experimentally actual and estimated speeds
during startup operation for speed command equal to 150 and
200 rad/s. It is observed that a good agreement between the
actual and estimated speeds is achieved. Fig. 18(a) shows the
actual and estimated speeds for transient from 150 to 200 rad/s.
The speed command is rst applied, so that rated speed oper-
ation is achieved. At t = 4.7 s, the speed command is further
increased to 200 rad/s so that the eld-weakening region is
entered. As shown, the estimated speed tracks the actual one
with good agreement. Fig. 18(b) shows the actual and estimated
speeds during speed reversal from 185 to 185 rad/s. The
experimental results conrm good speed estimation in the eld-
weakening region with magnetizing inductance identication.
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
1706 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY 2009
Fig. 18. Actual and estimated speeds, with modied SMO (a) for step change
of speed commands from 150 to 200 rad/s and (b) during speed reversal from
185 to 185 rad/s.
VI. CONCLUSION
In this paper, parallel speed and stator resistance identica-
tion schemes of sensorless induction motor drives have been
introduced to overcome the problem of resistance variation.
Estimation algorithms have been obtained based on a sliding-
mode current observer combined with Popovs hyperstability
theory. It has been found that activation of the stator resistance
adaptation mechanism quickly compensates the initial error in
the estimated stator resistance value and therefore eliminates
the initial speed estimation error. As a consequence, the actual
and estimated speeds become in very good agreement. Low-
and zero-speed sensorless operations have also been investi-
gated by the proposed SMO combined with the online stator
resistance adaptation scheme.
The supremacy of a certain speed estimation method is
weighted by its successful operation and accuracy over a wide
speed range. For this purpose, a modied SMO for speed
estimation in the eld-weakening region has been introduced.
The mismatch problem of magnetizing inductance due to ro-
tor ux reference variation in the eld-weakening region has
been treated by an online identication algorithm. Using this
algorithm, the magnetizing inductance has been continuously
determined from the measured stator voltages and currents,
enabling the correct calculation of rotor ux component at any
operating point taking saturation level into consideration.
The superiority of the modied SMO over the constant
parameter one for wide-speed-range estimation, from zero to
high values beyond the base speed, has been proved by simula-
tion results. This is also illustrated by the experimental results.
APPENDIX
A. List of Symbols
L
m
Magnetizing inductance.
L
s
; L
r
Stator and rotor leakage
inductances.
R
s
Stator resistance.
T
r
Rotor time constant.

r
;
r
Actual and estimated ro-
tor speeds.
Leakage coefcient.
i
s
s
= [i
s
ds
i
s
qs
]
T
;

i
s
s
= [

i
s
ds

i
s
qs
]
T
Actual and estimated
stator current vectors.

s
r
= [
s
dr

s
qr
]
T
;

s
r
= [

s
dr

s
qr
]
T
Actual and estimated ro-
tor ux vectors.
v
s
s
= [v
s
ds
v
s
qs
]
T
Stator voltage vector.
A
11
=aI A
12
=cI + dJ A
21
=eI
A
22
= A
12
b
1
=bI
I =
_
1 0
0 1
_
J =
_
0 1
1 0
_
a =
_
R
s
L
s
+
L
2
m
L
s
T
r
L
r
_
c=
1
T
r
d=

r

e=
L
m
T
r
=
L
s
L
r
L
m
b =
1
L
s
= 1
L
2
m
L
s
L
r
T
r
=
L
r
R
r
.
B. Induction Motor Parameters
TABLE I
INDUCTION MOTOR PARAMETERS
REFERENCES
[1] J. Holtz, Sensorless control of induction motor drives, Proc. IEEE,
vol. 90, no. 8, pp. 13591394, Aug. 2002.
[2] M. S. Zaky, M. M. Khater, H. Yasin, S. S. Shokralla, and A. El-Sabbe,
Speed-sensorless control of induction motor drives, Eng. Res. J., vol. 30,
no. 4, pp. 433444, Oct. 2007.
[3] Q. Gao, G. Asher, and M. Sumner, Sensorless position and speed control
of induction motors using high-frequency injection and without ofine
precommissioning, IEEE Trans. Ind. Electron., vol. 54, no. 5, pp. 2474
2481, Oct. 2007.
[4] H. Tajima, G. Guidi, and H. Umida, Consideration about problems and
solutions of speed estimation method and parameter tuning for speed-
sensorless vector control of induction motor drives, IEEE Trans. Ind.
Appl., vol. 38, no. 5, pp. 12821289, Sep./Oct. 2002.
[5] J. Holtz and J. Quan, Sensorless vector control of induction motors
at very low speed using a nonlinear inverter model and parameter
identication, IEEE Trans. Ind. Appl., vol. 38, no. 4, pp. 10871095,
Jul./Aug. 2002.
[6] G. Guidi and H. Umida, A novel stator resistance estimation method for
speed-sensorless induction motor drives, IEEE Trans. Ind. Appl., vol. 36,
no. 6, pp. 16191627, Nov./Dec. 2000.
[7] A. B. Proca, A. Keyhani, and J. M. Miller, Sensorless sliding-mode
control of induction motors using operating condition dependent models,
IEEE Trans. Energy Convers., vol. 18, no. 2, pp. 205212, Jun. 2003.
[8] A. B. Proca and A. Keyhani, Sliding-mode ux observer with online
rotor parameter estimation for induction motors, IEEE Trans. Ind. Elec-
tron., vol. 54, no. 2, pp. 716723, Apr. 2007.
[9] D. P. Marcetic and S. N. Vukosavic, Speed-sensorless AC drives with the
rotor time constant parameter update, IEEE Trans. Ind. Electron., vol. 54,
no. 5, pp. 26182625, May 2007.
[10] S. Maiti, C. Chakraborty, Y. Hori, and M. C. Ta, Model reference adap-
tive controller-based rotor resistance and speed estimation techniques for
vector controlled induction motor drive utilizing reactive power, IEEE
Trans. Ind. Electron., vol. 55, no. 2, pp. 594601, Feb. 2008.
[11] G. Edelbaher, K. Jezernik, and E. Urlep, Low-speed sensorless control of
induction machine, IEEE Trans. Ind. Electron., vol. 53, no. 1, pp. 120
129, Feb. 2006.
[12] H. M. Kojabadia and L. Changb, Comparative study of pole placement
methods in adaptive ux observers, Control Eng. Pract., vol. 13, no. 6,
pp. 749757, Jun. 2005.
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.
ZAKY et al.: WIDE-SPEED-RANGE ESTIMATION WITH ONLINE IDENTIFICATION SCHEMES OF MOTOR DRIVES 1707
[13] J. Maes and J. A. Melkebeek, Speed-sensorless direct torque control of
induction motors using an adaptive ux observer, IEEE Trans. Ind. Appl.,
vol. 36, no. 3, pp. 778785, May/Jun. 2000.
[14] H. M. Kojabadi, Simulation and experimental studies of model reference
adaptive system for sensorless induction motor drive, Simul. Model.
Pract. Theory, vol. 13, no. 6, pp. 451464, Sep. 2005.
[15] V. Vasic, S. N. Vukosavic, and E. Levi, A stator resistance estimation
scheme for speed sensorless rotor ux oriented induction motor drives,
IEEE Trans. Energy Convers., vol. 18, no. 4, pp. 476483, Dec. 2003.
[16] M. Barut, S. Bogosyan, and M. Gokasan, Experimental evaluation of
braided EKF for sensorless control of induction motors, IEEE Trans. Ind.
Electron., vol. 55, no. 2, pp. 620632, Feb. 2008.
[17] B. Karanayil, M. F. Rahman, and C. Grantham, Online stator and rotor
resistance estimation scheme using articial neural networks for vector
controlled speed sensorless induction motor drive, IEEE Trans. Ind.
Electron., vol. 54, no. 1, pp. 167176, Feb. 2007.
[18] M. S. Wang and C. M. Liaw, Improved eld-weakening control for IFO
induction motor, IEEE Trans. Aerosp. Electron. Syst., vol. 39, no. 2,
pp. 647658, Apr. 2003.
[19] M.-H. Shin and D.-S. Hyun, Speed sensorless stator ux-oriented control
of induction machine in the eld weakening region, IEEE Trans. Power
Electron., vol. 18, no. 2, pp. 580586, Mar. 2003.
[20] T.-S. Kwon, M.-H. Shin, and D.-S. Hyun, Speed sensorless stator ux-
oriented control of induction motor in the eld weakening region us-
ing Luenberger observer, IEEE Trans. Power Electron., vol. 20, no. 4,
pp. 864869, Jul. 2005.
[21] E. Levi, M. Sokola, and S. N. Vukosavic, A method for magnetizing
curve identication in rotor ux oriented induction machines, IEEE
Trans. Energy Convers., vol. 15, no. 2, pp. 157162, Jun. 2000.
[22] E. Levi and M. Wang, Online identication of the mutual inductance for
vector controlled induction motor drives, IEEE Trans. Energy Convers.,
vol. 18, no. 2, pp. 299305, Jun. 2003.
[23] E. Levi and M. Wang, A speed estimator for high performance sensorless
control of induction motors in the eld weakening region, IEEE Trans.
Power Electron., vol. 17, no. 3, pp. 365378, May 2002.
Mohamed S. Zaky was born in Minouya, Egypt,
on January 11, 1978. He received the B.Sc. (with
rst-class honors), M.Sc., and Ph.D. degrees in
electrical engineering from Minouya University,
Shebin El-Kom, Egypt, in 2001, 2005, and 2008,
respectively.
He is currently a Lecturer with the Department
of Electrical Engineering, Faculty of Engineering,
Minouya University. His research interests include
control of electrical machines and power electronics
applications.
Mahmoud M. Khater was born in Minouya,
Egypt, on July 16, 1959. He received the B.Sc.
and M.Sc. degrees in electrical engineering from
MinouyaUniversity, Shebin El-Kom, Egypt, in1982
and 1988, respectively, and the Ph.D. degree in
electrical engineering from Minouya University, in
cooperation with HeriotWatt University, Edinburgh,
U.K. (Scientic Channel System), in 1994.
He is currently a Professor with the Department
of Electrical Engineering, Faculty of Engineering,
Minouya University. His research interests include
power electronics, electrical drive systems, and control of electric machines.
Shokry S. Shokralla was born in Minouya, Egypt,
on December 13, 1955. He received the B.Sc., M.Sc.,
and Ph.D. degrees in electrical engineering from
Minouya University, Shebin El-Kom, Egypt, in
1978, 1982, and 1987, respectively.
He is currently a Professor and the Head of the De-
partment of Electrical Engineering, Faculty of Engi-
neering, Minouya University. His research interests
include power electronics, electrical drive systems,
and control of electric machines.
Hussain A. Yasin was born in Minouya, Egypt,
on July 13, 1945. He received the B.Sc. degree from
Minouya University, Shebin El-Kom, Egypt, in
1967, the M.Sc. degree from El-Mansoura Univer-
sity, El-Mansoura, Egypt, in 1979, and the Ph.D.
degree from the Energy Institute, Moscow, Soviet
Union, in 1983, all in electrical engineering.
He is currently a Professor with the Department
of Electrical Engineering, Faculty of Engineering,
Minouya University. His research interests include
control, planning, and stability of power systems.
Authorized licensed use limited to: University of Allahabad. Downloaded on January 26, 2010 at 11:00 from IEEE Xplore. Restrictions apply.