Professional Documents
Culture Documents
Automatica
journal homepage: www.elsevier.com/locate/automatica
Brief paper
LUNAM Universit, Ecole Centrale de Nantes, IRCCyN: Institut de Recherche en Communications et Cyberntique de Nantes, UMR CNRS 6597, BP 92101,1 Rue de la Noe,
44312 Nantes Cedex 3, France
b
University of Nuevo Leon, Department of Electrical Engineering. P.O Box 148-F, 66450, San Nicolas de Los Garza, Nuevo Leon, Mexico
article
info
Article history:
Received 29 July 2010
Received in revised form
1 July 2011
Accepted 8 September 2011
Available online 18 February 2012
Keywords:
Adaptive observers
Practical stability
Backstepping control
Sensorless induction motor benchmark
abstract
In this paper, an observer-based controller scheme is designed to robustly drive a sensorless Induction
Motor (IM) even for the case of low frequencies with unknown load torque. Combining the field oriented
control strategy with the backstepping control method, we introduce additional integral terms to improve
the robustness properties of the controller in spite of uncertainties and perturbations. The estimation
of the fluxes, the speed, the load torque and moreover the stator resistance is given by an adaptive
interconnected observer. The practical stability of the proposed observercontroller scheme is proved.
Finally, some experimental test results are given on the framework of a specific sensorless induction
motor benchmark.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
Generally, for IM complex controls like Field Oriented Control
(FOC), a shaft encoder is necessary. For all power drives, the
sensors maintenance implies difficulty, and moreover for small
and medium power drives, the sensor cost is important. For
these reasons the IM drive without mechanical sensor has had a
considerable interest in the past years. To control an IM, several
difficulties are found, for example the estimation of the state
variables at low frequencies and ensuring the robustness of the
controller against parameter variations, where the most critical
parameter affecting performance at low speed is stator resistance
(see Holtz (2002)). Several efforts have focused on sensorless
schemes in order to solve the IM control problem, taking into
account the robustness with respect to parametric uncertainties
and their performance under operation conditions.
Contribution. By using an adaptive interconnected observer for
estimating the IM flux, speed, load torque and stator resistance,
0005-1098/$ see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.automatica.2012.01.018
ba + bp i + i + m u
rd
rq
sd
s sq
1 sd
barq bp rd isq s isd + m1 usq
ard + (s p )rq + aMsr isd
rd =
rq
1
m(rd isq rq isd ) c Tl
isd
isq
(1)
2R
L2r Rs +Msr
r
Ls L2r
, = 1 (Msr2 /Ls Lr ),
m = pMsr /JLr , m1 = 1/ Ls , 1 =
2R
Msr
r
Ls L2r
. We denote by Rs and
683
( )dt ,
for = , rd ,
mrd isq + c +
z =
rd
Tl
+ K ( ) 1
J
= rd
+ ard aMsr isd + K (rd
rd ) 1rd .
(3)
2
Choosing the following candidate Lyapunov function Vz = 12 z
and by taking the derivative along the trajectories of (3), we get
V z = K z
T
mrd isq + c + l
+ K ( ) 1 .
Following the backstepping methodology, by choosing the virtual
control inputs isq as
T
+ c + l + K z + K ( ) ,
isq =
Combining the advantages of the field oriented controller design with the robustness properties of the backstepping control, a
control law is robustly designed by introducing additional integral
terms. Denote and the smooth bounded
reference signals
it follows that
tively.
Following the strategy of field oriented control (rd =
V z = zrd { rd
+ ard aMsr isd + K (rd
rd ) 1rd }.
rd
+
2
rd
2
rq ,
respec-
2
2 , = 0) and the fact that the stator frequency is given
rd
+ rq
rq
Controller design using a new integral backstepping method. Following the classical backstepping method, the controller design is done
in two steps. First step, the control problem is to choose isd and isq
in such a way to force and rd to track their desired reference
signals and . The second step is devoted to the current loops
design: find the controls usd and usq such that the currents isd and
isq converge fast to desired references isd and isq , respectively.
It is well-known that the control performances of IM is still
affected by the uncertainties of the plant, such as mechanical
parameter uncertainties, external load disturbance, no ideal field
orientation in a transient state and unmodeled dynamics in
practical applications. As introduced before, the backstepping
control robustness of speed/flux and currents controllers can be
increased by the introduction of new integral terms. This design is
detailed below.
Speed and flux loops. Let us consider the reduced model of IM given
by
2
V z = K z
z 1 ,
K > 0.
isd =
aMsr
[ rd
+ ard + K zrd + K (rd
rd )],
(5)
finally, we get
V z
rd
Krd > 0.
Current loops. Now to design the control laws for the complete
model (1), one introduces the following tracking errors
zi =
is
is +
zi ,
zi
Ki
(is is )dt ,
(6)
for = d, q. Kiq and Kid are positive constants. Next, consider the
following candidate Lyapunov functions
1
1 2
1
2
ziq
+ ziq2 ,
Vzid = Vz + zid
+ zid2 .
(7)
rd
2
2
2
2
Taking the time derivative of Vziq along the trajectories of (3) and
by replacing (4), it follows that
Vziq = Vz +
(2)
(4)
2
V ziq = K z
+ ziq
Tl
Tl
mrd isq c J + 1 m, c , J
=
rd
ard + aMsr isd + 1rd (a, Msr )
mrd
disq
dt
disq
dt
(8)
di
1
m1
disq
dt
(9)
684
It follows that
2
Kiq ziq2 + {ziq + ziq }Kiq (isq isq ) z 1
V ziq = K z
(10)
where Kiq , Kiq are positive constants with Kiq > Kiq and K1 =
min{K , Kidq Kiq , Kiq }.
Now, taking the time derivative of Vzid along the trajectories
of (3) and replacing (5), it follows that
disd
dt
disd
dt
zrd 1rd .
(11)
From (1),
= bard + bp rq isd + s isq + m1 usd and by
working out the control usd
disd
dt
usd =
1
m1
disd
dt
, (12)
it follows that
2
V zid = K z2rd {Kid Kid }zid
Kid zid2 zrd 1rd
(13)
where Kid , Kid are positive constants with Kid > Kid , and K2 =
min{K , Kid Kid , Kid }.
X 1 = A1 (
X2 , y)
X1 + g1 (u, y,
X2 ,
X1 ) + Tl
1 T
1 T
1 ) + KC2T (y2 y 2 )
+(
S
)
C
(
y
1 y
1
3
1
T
T
S3 = 3 S3 + T C1T C1
= (A1 (
X2 , y) S11 C1T C1 ) +
X 2 = A2 (
X1 )
X2 + g2 (u, y,
X1 ,
X2 ) + S21 C2T (y2 y 2 )
O2 :
(19)
T
S2 = 2 S2 A2 (
X1 )S2 S2 A2 (
X1 ) + C2T C2
Vc = Vziq + Vzid
1
2
2
z
+
1
2
2
ziq
+
1 2
1 2
1
1
ziq + z2rd + zid
+ zid2 .
2
2
2
2
(14)
Taking its time derivative and replacing the suitable terms, after
straightforward computations, one has
V c = Vc + ,
(15)
and =
12
22 2
rd
,
i (0, 1) i = 1, 2.
Rs ]T and
with y 1 = C1
X1 , y 2 = C1
X2
X2 = [isd
X2 = [isq rd rq ]T
are the estimated state variables respectively of X1 and X2 . 1 , 2 , 3
are positive constants, S1 and S2 are symmetric positive definite
matrices, with S3 (0) > 0, B1 (
X2 ) = km rd , B2 (
X2 ) = km rq ,
T
T
= diag(1, 1, ), KC2 = (kc1 , kc1 , 0) , k, kc1 , kc2 , and are
positive constants. Note that ( S31 T C1T + S11 C1T ) and KC2T
are the gains of observer (18) and S21 C2T is the gain of observer (19).
One now assumes that all parameters of IM are uncertain
and bounded with well-known values. In order to analyze the
robustness of the observer under parametric uncertainties, system
(16)(17) is now written in the following form
1, :
(16)
(17)
2 , :
(20)
(21)
3 = Tl Tl . From (20)(21) and (18)(19) and applying the transformation 1 = 1 3 , the dynamics of the estimation errors
are given by
1 = [A1 (
X2 , y) S11 C1T C1 + 2 ]1 + [A1 (X2 , y)
+ 1A1 (X2 , y) A1 (
X2 , y)]X1
+ g1 (u, y, X2 , X1 ) + 2 3
+ 1g1 (u, y, X2 , X1 ) g1 (u, y,
X2 ,
X1 ) + (2 K )2
2 = [A2 (
X1 ) S21 C2T C2 ]2 + [A2 (X1 )
(22)
where 1 = B1 (
X2 )C2 , 2 = B2 (
X2 )C1 and K = KC2T C2 .
(23)
= s = p + a
Msr
rd
(28)
[ S31 T C1T C1 + 2 ]1 1 2
Tl
i
(
,
T
,
rd sq
l
rd
.
=
J
rd
ard + aMsr isd (rd ) 1rd
3 = [ S31 T C1T C1 + 2 ]3
V o Vo +
685
t
1
m
1
0
rd +
(
Rs )isq +
s isd + isq
ab rq + bp
t
1
u
=
(isd isd )dt
K
(
i
i
)
+
K
K
sd
id
sd
id
sd
id
m1
0
ab rd bp
rq +
(Rs )isd s isq + isd .
isq .
(24)
(29)
In terms of tracking error dynamics for speed and flux: z and zrd ,
(29) can be rewritten in the following form
z = K z + 1 ( , Tl , , Rs ) 1
zrd = K zrd 2 ( , Rs ) 1rd
(30)
, T l = Tl
where =
Tl , = rd rd , Rs = Rs Rs ,
,
1 ( , Tl , , Rs ) = mrd {isq (
Tl , rd ,
Rs ) isq ( , Tl , rd , Rs )},
rd
, and assume that they are differentiable and bounded. Then, the
closed-loop system given by system (1) with the speed, flux and
current tracking laws (27)(28) using the estimates provided by the
adaptive interconnected observer (18)(19), is strongly uniformly
practically stable.
The proof follows a similar procedure as Proposition 1.
(25)
6. Experimental results
= p
aMsr isq
rd rd
rd +
ks
1 rd
isq
(26)
1
T
+ c
+ l + (K + K )(
)
i
=
sq
m rd
+ K K
)dt
(
0
sr
+ K K
(rd
rd )dt
(27)
The experimental results are obtained using a experimental setup described in Glumineau, Boisliveau, and Loron (2005). The motor parameters values are: nominal rate power 1.5 kW, nominal
angular speed 1430 rpm, number of pole pairs 2, nominal voltage
220 V, nominal current 7.5 A. The identified parameters values are
Rs = 1.633 , Msr = 0.099 H, Rr = 0.93 , J = 0.0111 Nm s2 /rad,
Ls = 0.142 H, fv = 0.0018 Nm s/rad, Lr = 0.076 H. The parameters of the observercontroller scheme are chosen as follows. For
the observer design: = 50, = 10, k = 0.16, kc1 = 450,
Kc2 = 0.5, 1 = 5000, 2 = 7000 and 3 = 109 . For the controller design, the gains are chosen as follows: K = 200 s1 ,
K = 1100 s1 , Kid = 550 s1 , Kid = 20 s1 , Kid = 34 s1 ,
Kiq = 1200 s1 , Kiq = 10 s1 , Kiq = 35 s1 , ks = 90 rad s 2 .
The sensorless trajectories of the benchmark are such that: after
that the reference speed is carried to 20 rad/s and from 1.5 to 2.5 s
the load torque is applied. This first step allows to test the performances and the robustness of the controller without mechanical
sensors at low speed but under observable conditions. From 3 to
4 s, the speed is carried to its nominal value (100 rad/s) and remains constant until t = 6 s. Again the load torque is applied from
5 s. This second step is defined to test the controller behavior during a great transient speed. Then, the motor is driven to reach a
negative constant low speed value from 7 s until 9 s. This speed is
chosen to obtain a stator pulsation equal to zero. This last step allows to illustrate the induction motor unobservability phenomena
(from t = 7 s to t = 9 s). Finally, the induction motor is driven in
order to leave the unobservability conditions (see Fig. 1).
686
Fig. 3. Int. backstepping experimental results for 50% Rr case versus time (s). a,
c: measured speed and torque; b, d, f estimated speed, torque and flux; e: reference
flux.
Fig. 2. FOC\PI experimental results in nominal case versus time (s). a, c: measured
speed and torque; b, d, f estimated speed, torque and flux; e: reference flux.
Fig. 4. P.I. field oriented control experimental results for 50% Rr case versus
time (s). a, c: measured speed and torque; b, d, f estimated speed, torque and flux;
e: reference flux.
687
Jesus de Leon received the Ph.D. degree in Automatic Control from Claude Bernard Lyon 1 University in 1992. Since
1993, he is a Professor of Electrical Engineering at Universidad Autonoma de Nuevo Leon, Mexico. He is currently
working on applications of control theory, electrical machines, nonlinear observers and power systems.