Вы находитесь на странице: 1из 8

Electric Power Systems Research 78 (2008) 18811888

Contents lists available at ScienceDirect


Electric Power Systems Research
j our nal homepage: www. el sevi er . com/ l ocat e/ epsr
Nonlinear systems time-varying parameter estimation:
Application to induction motors
Godpromesse Kenn e
a,
, Tarek Ahmed-Ali
b
,
F. Lamnabhi-Lagarrigue
c
, Amir Arzand e
d
a
Laboratoire dAutomatique et dInformatique Appliqu ee (LAIA), D epartement de G enie

Electrique, IUT FOTSO Victor, Universit e de Dschang, B.P. 134 Bandjoun, Cameroon
b
Ecole Nationale Sup erieure des Ing enieurs des Etudes et Techniques dArmement (ENSIETA), 2 Rue Franc ois Verny, 29806 Brest Cedex 9, France
c
Laboratoire des Signaux et Syst` emes (L2S), C.N.R.SSUPELEC, Universit e Paris XI, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France
d
D epartement Energie, Ecole Sup erieure dElectricit eSUPELEC, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France
a r t i c l e i n f o
Article history:
Received 10 April 2007
Received in revised form
12 November 2007
Accepted 16 March 2008
Available online 7 May 2008
Keywords:
Time-varying parameter estimation
Sliding mode observer
Induction motor
a b s t r a c t
In this paper, an algorithmfor time-varying parameter estimation for a large class of nonlinear systems is
presented. The proof of the convergence of the estimates to their true values is achieved using Lyapunov
theories and does not require that the classical persistent excitation condition be satised by the input
signal. Since the inductionmotor (IM) is widely used inseveral industrial sectors, the algorithmdeveloped
is potentially useful for adjusting the controller parameters of variable speeddrives. The methodproposed
is simple and easily implementable in real-time. The application of this approach to on-line estimation
of the rotor resistance of IM shows a rapidly converging estimate in spite of measurement noise, dis-
cretization effects, parameter uncertainties (e.g. inaccuracies on motor inductance values) and modeling
inaccuracies. The robustness analysis for this IM application also revealed that the proposed scheme is
insensitive to the stator resistance variations within a wide range. The merits of the proposed algorithm
in the case of on-line time-varying rotor resistance estimation are demonstrated via experimental results
in various operating conditions of the induction motor. The experimental results obtained demonstrate
that the application of the proposed algorithmto update on-line the parameters of an adaptive controller
(e.g. IMand synchronous machines adaptive control) can improve the efciency of the industrial process.
The other interesting features of the proposed method include fault detection/estimation and adaptive
control of IM and synchronous machines.
2008 Elsevier B.V. All rights reserved.
1. Introduction
The problem of parameter estimation has been of consider-
able interest during the last two decades, as can be seen in [15]
but only partial and quite weak results have been obtained in
terms of robustness with respect to time-varying parameters.
On the other hand, some methods are designed for a particular
application [610,14] and always make the assumption that the
estimated parameters are constant during the identication pro-
cess. But most parameters of practical plants are time-varying
such as the rotor resistance of an induction motor (IM) which
may vary up to 100% of its nominal value due to rotor heat-
ing and this variation can be hardly recovered using a thermal
model. Control algorithms exploiting this type of parameter

Corresponding author. Tel.: +237 77 59 52 19; fax: +237 77 59 52 19.


E-mail addresses: gokenne@yahoo.com (G. Kenn e), ahmedali@ensieta.fr
(T. Ahmed-Ali), lamnabhi@lss.supelec.fr (F. Lamnabhi-Lagarrigue),
amir.arzande@supelec.fr (A. Arzand e).
need to be updated on-line in order to increase their perfor-
mance.
In this article, a novel on-line technique for both constant and
time-varying parameter estimationof a large class of nonlinear sys-
tems is proposed. Following the same concepts developed in [11]
for nonlinear systems identication and control with time-varying
parameter perturbation, the proposed method is derived using a
new formulation. The proof of the convergence of the estimates to
their true values is achieved using Lyapunov theories and does not
require that the classical persistent excitationconditionbe satised
by the input signal as in [11]. Application of the proposed method
to the on-line estimation of the rotor resistance of an IM is there-
fore investigated to show its strength with respect to time-varying
parameter estimation. The experimental results obtained demon-
strate that the application of the proposed algorithm to update
on-line the parameters of an adaptive controller (e.g. IM and syn-
chronous machines adaptive control) can improve the efciency of
the industrial process.
Another important applicationareaof theproposeddesignis the
problem of fault detection and estimation. In particular, for IM, if
0378-7796/$ see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2008.03.014
1882 G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888
the value of the rotor resistance is outside some acceptable bounds
for normal operation, then a fault should be identied and the IM
has to be stopped or a different control scheme should be applied.
The remainder of the paper is organized as follows. In Section 2
we describe the nonlinear model dynamics and the problem state-
ment. In Section 3, the proposed algorithm is introduced and its
convergence is proved using Lyapunov-like technique. In Section 4
the proposed design is applied to on-line time-varying rotor resis-
tance estimation of IM. Experimental results with variation of the
rotor resistance and short-circuit simulated are reported in Section
5. Finally, Section 6 is devoted to the conclusion of the paper.
2. Nonlinear model dynamics and problem statement
Consider the following class of nonlinear system
x = F(x, u)0(t) +g(x, u) (1)
where F is a square matrix with elements continuous and bounded
on a compact
F
R; x R
n
is the state vector; uR
m
, m n is the
vector of the measurable inputs and0(t) R
p
, 1 p nis the vector
of the unknown time-varying parameters. The following assump-
tions will be made until further notice.
(i) the state vector x is available, continuous and bounded;
(ii) the vector of the measurable inputs u, is continuous and
bounded;
(iii) the system described by (1) is inputs states stable (ISS);
(iv) 0(t)
0
which is a compact set of R
P
;
(v)

0
i
(t)
0i
, 1 i p where
0i
are known positive numbers;
(vi) the operating points of the system described by (1) and gov-
erned by the vector of the measurable inputs u, are such that
the matrix F is inversible.
Remark 1. Assumption (i) is not a restriction for the proposed
method. It has been used to simplify the study. Indeed, in the case
where onlyone element of the state vector x is available, model-free
observers (e.g. high-gain observer (HGO), sliding mode observer)
can be applied step by step to estimate the derivatives or non-
available states of the class of the nonlinear systems of the general
form:
y
1
= y
2
y
2
= y
3
.
.
.
y
n1
= y
n
y
n
= j (y
1
, y
2
, . . . , y
n1
, y
n
, u(t), 0(t))
(2)
where only y
1
is measured or available. Assuming that the non-
available states are estimated using model-free observers and that
their estimates converge to their true values, the above nonlinear
system (2) can be re-rewritten in the form given by (1).
3. Parameter estimation adaptive observer
The objective is to design an adaptive observer based on the
variable structure theories and to combine this observer with the
estimation of the parameter. The method proposed in this paper
is the variation of an existing method [11]. In the present study, a
formulation for time-varying parameter estimation for nonlinear
systems is proposed and its application to on-line estimation of the
rotor resistance of induction motors.
1
The proof of the convergence
of the estimates to their true values is achieved using Lyapunov
1
The details are given in Section 4.
theories anddoes not require that the classical persistent excitation
condition be satised by the input signal as in [11]. Let us start
the design by considering the following variable structure observer
which has been proposed in [11]:

x = F(x, u)

0(t) +g(x, u) +v
x
(3)
where x,

0 are the adaptive observer of x, estimate of 0(t), respec-
tively, and v
x
= K
x
sign( x x), is the input of (3).
The function sign() : R
n
R
n
is dened as
sign(x
T
) = (sign(x
1
) sign(x
2
) . . . sign(x
n
))
with
x =

x
1
x
2
.
.
.
x
n

and sign(x
i
) =
_
1 if x
i
: 0
[ 1, 1] if x
i
= 0
1 if x
i
- 0
K
x
is a square n n diagonal matrix with positive elements.
Let diag(/), be the column vector whose elements are the diag-
onal elements of a given square diagonal matrix A.
The following notation is also used:
|x
T
|
C
= (|x
1
||x
2
|, . . . , |x
n
|).
By dening c
x
= x x, the vector of the observer error and c
0
=

0 0, the vector of parameter estimation error and taking into con-


sideration (1) and (3), it follows that the dynamic equation of the
observer error is given by
c
x
= Fc
0
+v
x
. (4)
Fromthe fact that F and0(t) are boundedandassuming that

0 is also
bounded, if the gains diag(K
x
i
), i = 1, . . . , n are chosen such that
diag(K
x
i
) : (|Fc
0
|
C
)
imax
(5)
aslidingmoderegimeoccurs onthemanifolds c
x
= 0[15,16]. There-
fore, (4) can be rewritten as
v
xeq
= Fc
0
(6)
where v
xeq
is the equivalent control.
Remark 2. From a practical point of view, it is not possible to
implement v
xeq
because the vector of parameter estimation error
c
0
is not available. To overcome this problem we used the average
control vector as an approximation for the equivalent control v
xeq
[12]:
v
xeq

=
1
1 +zp
v
x
, (7)
where z is a positive constant.
By taking into account Assumption (vi) and Remark 2, the
expression of the parameter estimation error can be derived as
follows
c
0
= F
1
v
xeq
. (8)
Using (8), we propose the following adaptive parameter identier:

0 = K
0
sign(F
1
v
xeq
) = K
0
sign(c
0
), (9)
where K
0
is a diagonal square matrix gains p p with positive ele-
ments. Now the dynamic of the parameter estimation error can be
deduced as
c
0
= K
0
sign(F
1
v
xeq
)

0. (10)
The main result of this part can be summarized by the following
theorem:
G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888 1883
Theorem 1. Given the nonlinear system (1) and the variable
structure identier (3). If the Assumptions (i)(vi) are satised and
condition (5) holds, then the estimated parameters

0 given by (9)
converge to their nominal values 0(t) in nite time.
Proof. Given the following Lyapunov candidate function
v =
1
2
c
T
0
c
0
(11)
and taking its time-derivative along the solution of (10), we obtain

v = c
T
0
c
0
=c
T
0
(

0) = c
T
0
(K
0
sign(c
0
)

0) = |(c
0
)
T
|
C
diag(K
0
)c
T
0

0.
From the fact that |

0
i
|
0
i
, i = 1, . . . , p, we can deduce that

v |c
T
0
|
C
(diag(K
0
)
0
).
If the gains of the adaptive parameter identier (9) are chosen such
that diag(K
0
)
i
:
0
i
,i=1,...,p
, then

v is negative-denite for c
0
= 0.
Therefore, the vector of the parameter estimation errors c
0
con-
verge to 0 in nite time. Consequently, the estimated parameters

0
i
, i = 1, . . . , p converge to their true values 0
i
(t) in nite time.
Remark 3. Note that the sign function may be omitted in (9) if the
unknown parameter is constant or varies slowly. This means that a
constant prediction parameter identier is considered and only the
instantaneous mean value

0 is estimated. Therefore, if the estimate
is used on-line to update a parameter of a control algorithm, the
adaptation scheme will be efcient if it is faster than the variation
of the actual unknown parameter 0.
4. Application to IM rotor resistance estimation
To illustrate the effectiveness of the proposed tool, we consid-
ered the problem of the identication of the time-varying rotor
resistance of an IM.
To achieve the rotor resistance estimation goal, let us consider
the classical dynamic model of an IM, expressed in the stator
reference frame assuming linear magnetic circuits and balanced
three-phase windings [13,8]
d
dt
=
1
e
m


m

1
L
m
(12)
dz
r
dt
=
_

R
r
L
r
l +n
p
j
_
z
r
+
R
r
L
r
Mi
s
(13)
di
s
dt
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_
z
r

1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s
+
1
oL
s
v
s
(14)
where
l =
_
1 0
0 1
_
, j =
_
0 1
1 0
_
, i
s
=
_
i
sa
i
sb
_
,
v
s
=
_
v
sa
v
sb
_
, z
r
=
_
z
ra
z
rb
_
.
In (12)(14), denotes the angular speed of the rotor; z
r
, i
s
and v
s
are the rotor ux, the stator current and voltage, respectively. R
s
, L
s
,
R
r
and L
r
are the stator resistance and inductance, rotor resistance
and inductance, respectively; n
p
is the number of pole pairs; o =
1 (M
2
]L
s
L
r
) is theleakageparameter; Mis themutual inductance;
m is the total motor and load moment of inertia; 1
e
= i
T
s
jz
r
is the
electromagnetic torque with the constant = n
p
(M]L
r
); 1
L
is the
external load torque and is the damping gain.
In order to satisfy Assumptions (i)(vi), it is assumed for the
system described by (12)(14) that:

the stator current and voltage are continuous and bounded with
time-derivatives bounded piecewise-continuous

the rotor resistance R


r

Rr
which is a compact set of R

the rotor resistance is time-varying and can therefore be


expressed as
R
r
= R
rn
+^R
r
with |

R
r
| = |^

R
r
|
Rr
where
Rr
is a known positive number.
Moreover, the following assumption will also be considered
until further notice:
(vii) the stator and rotor resistances vary slowly.
Using all these assumptions and assuming also that the term
containing d]dt is negligible or small with respect to other exist-
ing terms, the differentiation of (14) combined with (13), yields the
following dynamic equation
2
:
d
2
i
s
dt
2
= j
0
+R
r
j
1
(15)
where j
0
and j
1
are given as follows
j
0
=
1
_
dv
s
dt
R
s
di
s
dt
_
+j
_

2
i
s
+
3
v
s
+
1
di
s
dt
_
,
j
1
=
2
v
s

3
i
s

4
di
s
dt
and
1
,
2
,
3
,
1
,
2
,
3
,
4
are IM parameters expressed as

1
= n
p
,
2
=
n
p
R
s
oL
s
,
3
=
n
p
oL
s

1
=
1
oL
s
,
2
=
1
oL
s
L
r
,
3
=
R
s
oL
s
L
r
and
4
=
1
oL
r
.
The advantage of the transformation that leads to Eq. (15) is the
elimination of the non-measurable state z
r
.
The model given by (15) is an adequate representation of the
motor dynamics if
2

d
dt

2
||
1
r
. (16)
This condition can be satised by a reasonably number of variable
speed drive systems.
Remark 4. In [6], using IMmodel expressed in a coordinate frame
rotating with the rotor, an algorithm for parameter estimation and
rotor ux reconstruction is derived using a simplied model of the
IM and it is shown that the validity condition of this simplied
model is |d]dt| (||]1
r
). Hence the rotor speed is allowed to
vary more with the simplied model used in this paper for the
rotor resistance estimation.
Remark 5. Note that j
0
and j
1
are expressed as functions of the
derivatives of the stator current and voltage which cannot be com-
puted directly using numerical differentiation due to the presence
of measurement noise. To estimate these stator current and voltage
derivatives we considered the following high-gain observer

x
i
= 1
i
( x
i
x
i
) = 1
i
c
i
, i = 1u, 1b,

x
l
= 1
j
( x
j
x
j
) = 1
j
c
j
, j = 2u, 2b,
(17)
where c
i
= x
i
x
i
,c
j
= x
j
x
j
are the observer errors; 1
i
and 1
j
are
the gains of the HGO, x
1
= i
s
and x
2
= v
s
. To avoid the inuence
2
For detail, see Appendix A.
1884 G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888
of noise in (17), we approximated the stator current and voltage
derivatives as solutions of the following equations:
z
i
d
dt

x
i
+

x
i
=

x
i
, i = 1u, 1b
z
j
d
dt

x
j
+

x
j
=

x
j
, j = 2u, 2b
(18)
where z
i
and z
j
are positive numbers which tend to 0.
Therefore, (15) can be rewritten as follows
d x
1
dt
=

j
0
+R
r

j
1
(19)
where

j
0
and

j
1
are the estimates of j
0
and j
1
, respectively, com-
puted by replacing the stator current and voltage derivatives by
their estimates x
1
=

x
1
and x
2
=

x
2
derived from (17) and (18).
Assuming that the stator current/voltage andthe rotor speedare
available, the design of the time-varying identier R
r
is straightfor-
ward. It is based on the technique described in Section 3 with the
variable x, matrix F and the function g being replaced by
s
,

j
1
and

j
0
, respectively. The function

j
1
is continuous and bounded since it
is the linear combination of the stator current, stator voltage and
the time-derivative of the stator current which are continuous and
bounded by assumption. Moreover, one has ||

j
1
||
2
: 0 when the
motor operates in the stable region (||

j
1
|| = 0

j
1u
=

j
1b
= 0 which
does not correspond to the normal operation of the motor). In this
case, the matrix F is replaced by the vector

j
1
and the adaptive
observer, the expression of the rotor resistance estimation error
and the dynamics of the rotor resistance estimate are given by
d

x
1
dt
=

j
0
+

R
r

j
1
+v
s
(20)
c
Rr
=

j
T
1
v
seq
||

j
1
||
2
(21)

R
r
= k
Rr
sign(c
Rr
) = k
Rr
sign
_

j
T
1
v
seq
||

j
1
||
2
_
, (22)
where v
s
= K
x
1
sign(

x
1
x
1
) and v
seq
is the equivalent control
approximatedas reportedinRemark2andassumingthat the motor
operates in the stable region.
5. Experimental results
The experimental setup is illustrated by the block diagram of
Fig. 1 which includes a development system DSP1103 within the
Fig. 1. Block diagram of the experimental setup.
Matlab/Simulink environment software, an input/output electron-
ics board (for analog/digital conversions) and a Personal Computer
(PC). This PC is used to programthe DSP, to store and display exper-
imental data. A 5kW induction motor whose data are reported
in Appendix B has been used. This motor is supplied by a 3-
phase variable power supply 0127V/50Hz and in all estimations
a phase input of 110V/50Hz is applied. The external load torque
is produced by a loaded dc generator. The motor instantaneous
speed is measured by an optical incremental encoder with 1024
lines per revolution. The stator currents and voltages are measured
by Hall-type sensors and isolated voltage sensors, respectively.
All measured electrical parameters are converted by 16-b A/D
converter channels with 1s conversion time. Data acquisition
performed by a DSP1103 within the Matlab/Simulink environment
software, was used to implement in real-time the identication
algorithm with a sampling time of 105.8s.
The parameters of the high-gain observer and the rotor resis-
tance identier used both in experiments and in simulations are:
1
i
= 26, 250, 1
j
= 30, 500, z
i
= z
j
= 0.53ms, K
x
1u
= K
x
1b
= 6 10
6
and k
Rr
= 1.5. The equivalent control v
seq
has been approximated
using rst order low pass-pass lter with a time-constant of 1ms.
The dynamic behavior of the stator current, the observer errors
and the dynamics of the rotor speed and the rotor resistance esti-
mate are depicted in Fig. 2 with under-estimated initial value. Fig. 3
Fig. 2. Experimental results with under-estimated initial value of Rr in the hot case during the start-up. (a) Stator current component (i), stator current HGO error (ii) given
and stator current adaptive error (iii). (b) Rotor speed (i) and rotor resistance estimate (ii).
G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888 1885
Fig. 3. Experimental results with over-estimated initial value of Rr in the hot case during the start-up. (a) Stator current component (i), stator current HGO error (ii) and
stator current adaptive error (iii). (b) Rotor speed (i) and rotor resistance estimate (ii).
shows the same results with over-estimated initial value. In both
cases, the experiments are carried out during motor start-up and
after the motor is operated at rated load torque for over 60min (hot
case).
The results reported in Fig. 4 illustrate the inuence of the stator
resistance variation on the identication scheme when the same
set of the estimation algorithm parameters is used. Note that there
is no signicant effect on the rotor resistance estimate dynamics
for a large range of variation of the stator resistance. Signicant
rotor resistance error appears for R
s
1!. Therefore we didnot use
temperature sensors to update the value of the stator resistance in
the estimation scheme.
Fig. 4. Experimental results illustrating the inuence of the stator resistance variation on the rotor resistance estimation with over-estimated initial value in the hot case
during the start-up. (a) Stator current component (i), stator current HGO error (ii) and stator current adaptive error (iii). (b) Rotor speed (i), stator resistance (ii) and rotor
resistance estimate (iii).
Fig. 5. Experimental results with over-estimated initial value of Rr in the cold case during the start-up. (a) Stator current component (i), stator current HGO error (ii) and
stator current adaptive error (iii). (b) Rotor speed (i) and rotor resistance estimate (ii).
1886 G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888
Fig. 6. Comparative results of the rotor resistance estimate in hot case and in cold
case.

Rr cold case (dotted line) and

Rr hot case (solid line).
Toshowthemerits of theproposedalgorithminthecaseof time-
varying parameter estimation, experiments have been carried out
at the start-up of a cold motor (cold case) at rated load torque in
order to compare the results with those obtained previously in the
hot case. The experimental results in the cold case are reported in
Fig. 5. The comparative results of the rotor resistance estimation in
both cases are shown in Fig. 6. The average rotor resistance value is
0.47! for the cold case and 0.54! for the hot case.
Moreover, real-time variation of the rotor resistance has been
carried out by adding a known resistance (by using a 3-phase vari-
able rheostat for rotor wound inductionmotor) during 8s inthe hot
case when the motor is in steady-state operation with rated load
torque. The value of the known additional resistance is 0.4!in the
cold case operation of the motor and 0.47! in the hot case. The
experimental results are reported in Fig. 7. These results conrm
the good behavior of the proposed algorithm to track the varia-
tion of the parameter and therefore estimate the fault for a given
normal operating range of the IM. Note that this on-line varia-
tion corresponds approximately to 87% of the mean value of the
estimated rotor resistance in the hot case. The steady-state error
of about 4% remains on the value of the rotor resistance and is
Fig. 8. Simulation results with rotor resistance variation and short-circuit occurred
at time t = 8s. Rotor speed (i) and rotor resistance estimate (ii).
almost the same before, during and after the variation. The resid-
ual parameter estimation error is due to the measurement noise,
mismatching between the motor and the model parameters, ohmic
heating during experiments and unmodeled dynamics. The steady-
state estimation error of about 4% is indicative of a very good match
of the data with the simplied model of IM (15) compared to the
parameter estimation error obtained in [6] where the same type
of simplied model of IM has been used but without variation of
the rotor resistance. Therefore, the residual parameter estimation
error of about 4% is acceptable in practice and will not affect signif-
icantly the efciency of the IM adaptive controller if the algorithm
is used in on-line updating of the value of the rotor resistance of an
IM adaptive controller.
Fig. 8 presents the simulation results corresponding to the
experimental results shown in Fig. 7 (case of variation of R
r
) but
withshort-circuit fault simulatedat timet = 8s. This result demon-
strates the capability of the proposed algorithm to detect and
estimate parameter fault.
Note that the adaptive error in (3) in the experiment is par-
ticularly higher due to the presence of high order harmonics and
time-delays in the time-derivative as shown in Fig. 9, however,
this does not affect the rotor resistance estimate. The small tran-
sient phases of the estimates obtained both in simulations and
Fig. 7. Experimental results with rotor resistance variation in the hot case when the motor is in the steady-state period under rated torque. (a) Stator current component (i),
stator current HGO error (ii) given by and stator current adaptive error (iii). (b) Rotor speed (i) and rotor resistance estimate (ii).
G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888 1887
Fig. 9. Reference and adaptive time-derivatives in the hot case with over-estimated
initial value of Rr.
in experiments conrm the quick convergence of the method
proposed.
The main drawback of the proposed method is the possibil-
ity to obtain negative rotor resistance estimate during start-up
and during the transient process of identication when an under-
estimated initial value is used. It is therefore recommended to
use either over-estimated initial value or nominal value of R
r
during the start-up and during the transient process of iden-
tication if the algorithm is used in an adaptive controller
of IM.
6. Conclusion
A novel scheme for time-varying parameter estimation of a
large class of nonlinear systems has been proposed. The algo-
rithm developed is potentially useful for adjusting the controller
parameters of variable speed drives and can be used for both con-
stant and time-varying parameter estimation. The proof of the
convergence of the proposed design has been achieved using the
Lyapunov theories without using the classical persistent excita-
tion condition for the input signal. The proposed method is simple
and easily implementable in real-time. Application to on-line esti-
mation of the rotor resistance of an induction motor, shows a
rapidly converging estimate in spite of measurement noise, dis-
cretization effects, parameter uncertainties (e.g. inaccuracies on
motor inductance values) and modeling inaccuracies. The robust-
ness analysis for IM application also revealed that the proposed
scheme is insensitive to the stator resistance variations within a
wide range. The merits of the proposed algorithm in the case of
time-varyingon-line rotor resistance estimationhave beendemon-
strated via experimental results in various operating conditions of
the IM. The results obtained demonstrate that the application of
the proposed algorithm to update on-line the parameters of an
adaptive controller (e.g. IM and synchronous machines adaptive
control) can improve the efciency of the industrial process. The
other interesting features of the proposed method include fault
detection/estimation and adaptive control of IM and synchronous
machines.
Acknowledgment
The main part of the experimental setup used in this work has
been supported by the D epartement Energie, Ecole Sup erieure
dElectricit e, Gif-sur-Yvette, Paris, France.
Appendix A. Details for IM dynamics (15) and validity
condition (16)
Let us recall (13) and (14)
dz
r
dt
=
_

R
r
L
r
l +n
p
j
_
z
r
+
R
r
L
r
Mi
s
(23)
di
s
dt
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_
z
r

1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s
+
1
oL
s
v
s
(24)
with
l =
_
1 0
0 1
_
, j =
_
0 1
1 0
_
, i
s
=
_
i
sa
i
sb
_
, v
s
=
_
v
sa
v
sb
_
,
z
r
=
_
z
ra
z
rb
_
.
By differentiating both sides of (24), yields
d
2
i
s
dt
2
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_
dz
r
dt

M
oL
s
L
r
n
p
d
dt
jz
r

1
oL
s
_
R
s
+M
2
R
r
L
2
r
_
di
s
dt
+
1
oL
s
dv
s
dt
(25)
Using (23) combined with (25), we obtain
d
2
i
s
dt
2
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
___

R
r
L
r
l +n
p
j
_
z
r
+
R
r
L
r
Mi
s
_

M
oL
s
L
r
n
p
d
dt
jz
r

1
oL
s
_
R
s
+M
2
R
r
L
2
r
_
di
s
dt
+
1
oL
s
dv
s
dt
. (26)
If the term containing d]dt is negligible with respect to other
existing terms, then (26) can be rewritten as follows.
d
2
i
s
dt
2
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
___

R
r
L
r
l +n
p
j
_
z
r
+
R
r
L
r
Mi
s
_

1
oL
s
_
R
s
+M
2
R
r
L
2
r
_
di
s
dt
+
1
oL
s
dv
s
dt
. (27)
The expression of the rotor ux z
r
is deduced from (24) as
z
r
=

oL
s
L
r
M
_

R
r
L
r
l+n
p
j
_
1

_
di
s
dt
+
1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s

1
oL
s
v
s
_
=
oL
s
L
r
M((R
2
r
]L
2
r
) +(n
p
)
2
)
_
R
r
L
r
l +n
p
j
_

_
di
s
dt
+
1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s

1
oL
s
v
s
_
(28)
Multiplying both sides of (28) with ((R
r
]L
r
)l +n
p
j) gives
_

R
r
L
r
l +n
p
j
_
z
r
=
oL
s
L
r
M((R
2
r
]L
2
r
) +(n
p
)
2
)
_

R
r
L
r
l +n
p
j
__
R
r
L
r
l +n
p
j
_

_
di
s
dt
+
1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s

1
oL
s
v
s
_
=
oL
s
L
r
M
_
di
s
dt
+
1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s

1
oL
s
v
s
_
(29)
1888 G. Kenn e et al. / Electric Power Systems Research 78 (2008) 18811888
Computing (27) with the result of (29), we obtain
d
2
i
s
dt
2
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_

oL
s
L
r
M
_
di
s
dt
+
1
oL
s
_
R
s
+
M
2
R
r
L
2
r
_
i
s

1
oL
s
v
s
_
+
R
r
L
r
Mi
s
_

1
oL
s
_
R
s
+M
2
R
r
L
2
r
_
di
s
dt
+
1
oL
s
dv
s
dt
. (30)
Eq. (30) can be re-arranged as follows.
d
2
i
s
dt
2
= j
0
+R
r
j
1
(31)
where j
0
and j
1
are given as
j
0
=
1
_
dv
s
dt
R
s
di
s
dt
_
+j
_

2
i
s
+
3
v
s
+
1
di
s
dt
_
j
1
=
2
v
s

3
i
s

4
di
s
dt
and
1
,
2
,
3
,
1
,
2
,
3
,
4
are IM parameters expressed as

1
= n
p
,
2
=
n
p
R
s
oL
s
,
3
=
n
p
oL
s

1
=
1
oL
s
,
2
=
1
oL
s
L
r
,
3
=
R
s
oL
s
L
r
and
4
=
1
oL
r
.
In order to determine when the simplied model (15) or (31) repre-
sents adequately the motor dynamics, we returnto(25) andrewrite
it as follows

M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_
dz
r
dt

M
oL
s
L
r
n
p
d
dt
jz
r
=
1
oL
s
dv
s
dt
+
d
2
i
s
dt
2
+
1
oL
s
_
R
s
+M
2
R
r
L
2
r
_
di
s
dt
. (32)
Using (23), the left-hand side of (32) can be rewritten as

M
oL
s
L
r
_

R
r
L
r
l +n
p
j
_
dz
r
dt

M
oL
s
L
r
n
p
d
dt
jz
r
=
M
oL
s
L
r
_

R
r
L
r
l +n
p
j
___

R
r
L
r
l +n
p
j
_
z
r
+
R
r
L
r
Mi
s
_

M
oL
s
L
r
n
p
d
dt
jz
r
=
M
oL
s
L
r

_
n
p
_
d
dt

2
1
r

_
jz
r
+
1
1
2
r
(z
r
Mi
s
) (n
p
)
2
z
r
+
1
1
r
n
p
Mji
s
_
,
where 1
r
= (L
r
]R
r
) is the rotor time-constant. We can then deduce
that (15) adequately represents the IM dynamics if

d
dt

2
||
1
r
. (33)
Appendix B. Induction motor data
Rated power 5kW
Rated speed 1500rpm
Rated torque 32Nm
Rated frequency 50Hz
Rated current 22.9A
Stator resistance RsN = 0.22!
Rotor resistance RrN = 0.52!
Stator inductance LsN = 0.052H
Rotor inductance LrN = 0.0516H
Mutual inductance MN = 0.0495H
Number of pole pairs np = 2
Motor-load inertia m = 0.12kgm
2
.
References
[1] J. Xu, H. Hashimoto, Parameter identication methodologies based on variable
structure control, Int. J. Control 57 (1993) 12071220.
[2] E. Walter, L. Pronzato, Identication de mod` eles param etriques, Masson Pub-
lishing, 1994.
[3] I.D. Landau, Identication des syst ` emes, in: Collection p edagogique
dautomatique, Herm` es, 1998.
[4] I.D. Landau, B.D.O. Anderson, F. De Bruyne, Algorithms for identication of
continuous time nonlinear systems: a passivity approach, in: A. Isidori, F.
Lamnabhi-Lagarrigue, W. Respondek (Eds.), in: Nonlinear Control in the Year
2000, vol. 2, Springer Verlag, Paris, 2000, pp. 1344.
[5] M. Niethammer, P.H. Menold, F. Allg ower, Parameter and derivative esti-
mation for nonlinear continuous-time system identication, in: Proceedings
of the Fifth IFAC Symposium Nonlinear Control Systems-NOLCOS01, Saint-
Petersburg, Russia, 2001.
[6] J. Stephan, M. Bodson, J. Chiasson, Real-time estimation of the parameters and
uxes of induction motors, IEEE Trans. Ind. Appl. 30 (1994) 746759.
[7] K. Akatsu, A. Kawamura, On-line rotor resistance estimation using the transient
state under the speed sensorless control of induction motor, IEEE Trans. Power
Electron. 15 (2000) 553560.
[8] R. Marino, S. Peresada, P. Tomei, On-line stator and rotor resistance estimation
for induction motors, IEEE Trans. Control Syst. Technol. 8 (2000) 570579.
[9] A.V. Pavlov, A.T. Zaremba, Real-time rotor and stator resistances estimation of
an induction motor, in: Proceedings of NOLCOS-01, St.-Petersburg, 2001.
[10] G. Alessandro, D.S. Nunes Ivan, S.A. Jose Paulo, Load torque identication in
induction motor using neural network technique, Electric Power Syst. Res. 77
(2007) 3545.
[11] T. Ahmed-Ali, F. Floret, F. Lamnabhi-Lagarrigue, Robust identication and con-
trol with time-varying parameter perturbation, J. Math. Comput. Modell. Dyn.
Syst. 10 (2004) 201216.
[12] V.I. Utkin, Sliding Modes in Optimization and Control, Springer Verlag, 1992.
[13] W. Leonhard, Control of Electric Drives, Springer Verlag, 1984.
[14] P. Castaldi, W. Geri, M. Montanari, A. Tilli, A new adaptive approach for on-line
parameter and state estimation of induction motors, Control Eng. Practice 13
(2005) 8194.
[15] J.J. Slotine, W. Li, Applied Nonlinear Control, Prentice Hall, New York, 1991.
[16] V.I. Utkin, Sliding mode control design principles and applications to electric
drives, IEEE Trans. Ind. Electron. 40 (1993) 2636.

Вам также может понравиться