Académique Documents
Professionnel Documents
Culture Documents
Abstract: Field-oriented speed control of induction motors (IMs) without mechanical sensors
(speed sensor and load torque sensor) are considered. The methodology is divided into two
parts. First, interconnected high-gain observers are designed to estimate the mechanical and mag-
netic variables from the only measurement of stator current. Secondly, the speed and flux esti-
mation are used by a controller to achieve the speed/flux tracking. The flux regulation problem
is simple and the traditional approach is followed by using proportional integral (PI) controller.
For the speed-regulation problem, it is stated that flux regulation quickly happens by using a high-
gain PI controller to regulate the q-axis current to its reference. Stability analysis based on
Lyapunov theory is proved to guarantee the ‘observer þ controller’ stability. To test and validate
the controller – observer by considering the sensorless control problem of IM at low frequency, a
significant benchmark is implemented. The trajectories of this benchmark are designed to validate
the controller and observer under three operating conditions: low speed, high speed and very low
speed. Furthermore, robustness tests with respect to parameter variations are given in order to show
the performance of the proposed observer – controller scheme.
Remark 2: The choice of the variables of each subsystem Remark 4: B1(Z2)( y2 2 y^2) þ B2(Z2)( y1 2 y^1) ; k[m(f̂rd
has been considered in order to separate the mechanical isq 2 f̂rqisd) 2 m(frdi^sq 2 frqi^sd)] ; k(Te 2 T̃e), where Te
variables (V, Tl) from the magnetic variables (frd , frq). It and T̃e are, respectively, the ‘measured’ and ‘estimated’
is clear that other choice could be considered in order to rep- electromagnetic torque.
resent these subsystems, provided an observer could be
designed. This choice will be highlighted in Section 6.1. Assumption 2:
1. The states X1 and X2 are bounded,
Now, we present an observer design for the sensorless IM 2. A1 (X2 ) is globally Lipschitz with respect to X2 ,
that is based on the interconnection between several obser- 3. A2 (X1 ) is globally Lipschitz with respect to X1 ,
vers satisfying some required properties, particularly the 4. g1 (u, y, X2 , X1 ), g2 (u, y, X2 , X1 ) are globally Lipschitz
property of inputs persistency. with respect to X2 , X1 and uniformly with respect to (u, y).
From the definition, it is possible to calculate the Lipschitz
Remark 3: A regularly persistence input is an input that suf- constant; see the following example. We consider matrix
ficiently excites the system in order to guarantee its obser- A1 (X2 ), the norm of the Jacobian of A1 (X2 ) along X2 is:
vability [14]. k@A1 (X2 )=@X2 k ¼ bp.
IET Control Theory Appl., Vol. 1, No. 6, November 2007 1683
Then, assuming that Assumption 2 is verified, a nominal and u2 can be selected from (16). Furthermore, the
observer for interconnected systems (5) is given by estimation error converges asymptotically to zero if we
8 know exactly the motor parameters. Otherwise, the conver-
>
> Z_1 ¼ A1 (Z2 )Z1 þ g1 (u, y, Z2 , Z1 ) gence of the complete observer admits a small upper bound.
>
>
>
> þ(GS11 C1T þ B2 (Z2 ))(y1 y^ 1 )
>
>
>
> þ(B1 (Z2 ) þ KC2T )(y2 y^ 2 ) Remark 6: When the persistence condition is not satisfied
< _
S ¼ u1 S1 AT1 (Z2 )S1 S1 A1 (Z2 ) þ C1T C1 (in the unobservable area), we will prove the stability of
O: 1
>
> y^ ¼ C1 Z1 the observer in Section 6.1.
> _1
>
>
> Z ¼ A2 (Z1 )Z2 þ g2 (u, y, Z1 , Z2 ) þ S21 C2T (y2 y^ 2 )
> _2
> T T Proof: Let us define Vo ¼ V1 þ V2 a candidate Lyapunov
>
: S2 ¼ u2 S2 A2 (Z1 )S2 S2 A2 (Z1 ) þ C2 C2
> function where V1 and V2 are, respectively, the candidate
y^ 2 ¼ C2 Z2 Lyapunov function for each dynamics (11), (12) given by
(10) V1 ¼ eT1 S1 e1 and V2 ¼ eT2 S2 e2 . Next, taking the time deriva-
tive of Vo and replacing the suitable expressions, (13) and
Remark 5: It is worth noticing that kS1 k and kS2 k are (10) we have
bounded for u1 and u2 large enough due to the persistency
property of inputs; more details can be found in [9]. V_o ¼ eT1 u1 S1 (2S1 GS11 1)C1T C1 2S1 B02 e1
þ 2eT1 S1 A1 (X2 ) A1 (Z2 ) X1 þ 2eT1 S1 DA1 (X2 ) X1
4 Stability analysis of observer under
parameters uncertainties þ 2eT1 S1 g1 (u, y, X2 , X1 ) g1 (u, y, Z2 , Z1 )
Defining the estimation errors as þ 2eT1 S1 Dg1 (u, y, X2 , X1 ) þ eT2 u2 S2 C2T C2 e2
e1 ¼ X1 Z1 and e2 ¼ X2 Z2 þ 2eT2 S2 A2 (X1 ) A2 (Z1 ) X2 þ 2eT2 S2 DA2 (X1 ) X2
the estimation error dynamics is given by þ 2eT2 S2 g2 (u, y, X1 , X2 ) g2 (u, y, Z1 , Z2 )
e_1 ¼ [A1 (Z2 ) GS11 C1T C1 B2 (Z2 )C1 ]e1 þ 2eT2 S2 Dg2 (u, y, X1 , X2 ) 2eT1 S1 (B01 þ K 0 )e2
þ g1 (u, y, X2 , X1 ) g1 (u, y, Z2 , Z1 )
where B01 ¼ B1 (Z2 )C1 , B02 ¼ B2 (Z2 )C2 and K 0 ¼ KC2T C2 .
þ [A1 (X2 ) A1 (Z2 )]X1 From Assumption 2, the following inequalities hold
(B1 (Z2 )C2 þ KC2T C2 )e2 (11)
kS1 k k1 , kS2 k k5 , kX1 k k3 , kX2 k k7
k g1 (u, y, X2 , X1 ) g1 (u, y, Z2 , Z1 ) k k4 ke2k þ k9 ke1 k
e_2 ¼ [A2 (Z1 ) S21 C2T C2 ]e2 kA1 (X2 ) A1 (Z2 )k k2 ke2 k
þ [A2 (X1 ) A2 (Z1 )]X2 k A2 (X1) A2 (Z1 ) k k6 ke1 k
þ [g2 (u, y, X1 , X2 ) g2 (u, y, Z1 , Z2 )] (12) k g2 (u, y, X1 , X2 ) g2 (u, y, Z1 , Z2 ) k k8 ke1 k þ k10 ke2 k
kB01 k kB1 , kB02 k kB2 , kK 0 k kk 0
Now we consider that the motor parameters are known
with uncertainties. So (11) and (12) become Using the norm, from Assumptions 1 and 3, and by
regrouping with respect to ke1 k and ke2 k, the time deriva-
e_ 1 ¼ [A1 (Z2 ) GS11 C1T C1 B2 C1 ]e1 tive of Vo can be rewritten as follows
þ g1 (u, y, X2 , X1 ) þ Dg1 (u, y, X2 , X1 )
V_o (u1 þ 2kB1 2k1 k9 )eT1 S1 e1 þ 2m1 ke1 kke2 k
g1 (u, y, Z2 , Z1 ) þ [A1 (X2 ) þ DA1 (X2 ) A1 (Z2 )]X1
þ 2m2 ke1 kke2 k (u2 2k10 k5 )eT2 S2 e2
(B1 C2 þ KC2T C2 )e2 (13)
þ 2m3 ke2 kke1 k þ 2m4 ke2 kke1 k þ m5 ke1 k
e_ 2 ¼ [A2 (Z1 ) S21 C2T C2 ]e2 þ m6 ke2 k (14)
þ [A2 (X1 ) þ DA2 (X1 ) A2 (Z1 )]X2
þ g2 (u, y, X1 , X2 ) þ Dg2 (u, y, X1 , X2 ) where m1 ¼ k1 k2 k3 kB2 k1 kk 0 k1 , m2 ¼ k1 k4 , m3 ¼ k5 k6 k7 ,
m4 ¼ k5 k8 , m5 ¼ 2(k1 k3 r1 þ k1 r3 ) and m6 ¼ 2(k5 k7 r2 þ k5 r4 ):
g2 (u, y, Z1 , Z2 )
Now, consider that the following inequalities are satisfied
where the terms DA1 (X2 ), DA2 (X1 ), Dg1 (u, y, X2 , X1 ) and
Dg2 (u, y, X1 , X2 ) represent the uncertain terms of A1 (X2 ), lmin (Si )kei k2 kei k2Si lmax (Si )kei k2 , i ¼ 1, 2
A2 (X2 ), g1 (u, y, X2 , X1 ) and g2 (u, y, X1 , X2 ), respectively.
where lmin (Si ) and lmax (Si ) are minimal and maximal eigen-
Assumption 3: We assume that the uncertain terms satisfy values of Si independently to u
the following inequalities
kei k2Si ¼ eTi Si ei , i ¼ 1, 2
kDA1 (X2 )k r1 , kDA2 (X1 )k r2 , kDg1 (u, y, X2 , X1 )k r3
kDg2 (u, y, X1 , X2 )k r4 Inequality (14) can be rewritten in terms of functions V1
and V2
with ri . 0, for i ¼ 1, . . . , 4.
Theorem 1: Let us consider system (5) and assume that V_ o (u1 þ 2kB1 2k1 k9 )V1 (u2 2k10 k5 )V2
Assumptions 1 – 3 are satisfied. Then, system (10) is an pffiffiffiffiffipffiffiffiffiffi
interconnected observer for system (5). The parameters u1 þ 2m̃ V1 V2 þ m5 ke1 k þ m6 ke2 k (15)
X
4 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0 1
mfrd isq cV
m̃ ¼ m̃ i , min (S) ¼ min (S1 ) min (S2 ) 0 1 B C
i¼0 _
V B afrd þ aMsr isd C
B C B M C
mi B ḟ rd C B pV þ a sr isq C
m̃ i ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi, i ¼ 1, 2, 3, 4 B C B C
min (S) B ṙ C ¼ B frd C
B C B C
pffiffiffiffiffipffiffiffiffiffi B C B Msr 2 C
Using the following inequality, V1 V2 (e=2)V1 þ @ isd A B B gisd þ abfrd þ pVisq þ a
i
frd sq
C
C
B C
(1=2e)V2 , 8e [ ]0, 1[ isq @ M A
gisq bpVfrd pVisd a sr isd isq
frd
V_ o (u1 þ 2kB1 2k1 k9 )V1 þ m̃eV1 0 1
1
þ
m̃
V (u2 2k10 k5 )V2 þ m5 ke1 k þ m6 ke2 k B 0 0 C
J 0 1
e 2 B C
B 0 0 0 C Vsd
B CB C
Consequently
þB 0 0 0 C @ Vsq A (20)
B C
B C
@ m1 0 0 A Tl
V_o (u1 þ 2kB1 2k1 k9 m̃e)V1
0 m1 0
m̃
u2 2k10 k5 V þ m5 ke1 k þ m6 ke2 k
e 2 where isd , isq and Vsd , Vsq are the stator currents and stator
voltages in axes of dq frame, respectively, and r the flux
with d ¼ min (d1 , d2 ) and m ¼ max (m5 , m6 ), where angle.
d1 ¼ u1 þ 2kB1 2k1 k9 m̃e . 0 and d2 ¼ u2 2k10 k5 Note that the electromagnetic torque is
m̃ =e . 0.
So that pMsr
Te ¼ f i (21)
Lr rd sq
m̃
u1 . 2k1 k9 þ m̃e 2kB1 , u2 . 2k10 k5 þ (16)
e
For the new model (20), by holding constant the magni-
Then it follows that tude of the rotor flux, there is a linear relationship
between isq and the speed dynamic.
pffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffi Moreover, to eliminate the nonlinear terms effect, a poss-
V_o d(V1 þ V2 ) þ m( V1 þ V2 ) dVo þ mc Vo
ible strategy is to force a current-control mode using high-
(17) gain feedback [13].
p pffiffiffiffiffi pffiffiffiffiffi That is, one uses PI current loops of the form
where c . 0, such that c ðV1 þ V2 Þ . V1 þ V2 .
If the motor parameters are exactly known, then m ¼ 0 ðt
and V_o dVo . This system (10) is an exponential observer Vsd ¼ Kivd (isd isd ) dt þ Kpvd (isd isd ) (22)
for system (5). 0
Otherwise, if m = 0, inequality (17) can be rewritten as ðt
Vsq ¼ Kivq (isq isq ) dt þ Kpvq (isq isq ) (23)
0
V_o (1 6)dVo 6Vo þ mckek, 1.6.0 (18)
Finally, we have to force isd and isq to, respectively, track their corresponding
references isd and isq . The PI current loops result in fast
mc responses by using large feedback gain. As a result, isd
V_o (1 6)dVo , 8kek (19) and isq are considered as the new inputs and the system
6d
reads
By choosing u1 and u2 , inequality (19) shows that in spite 0 1
of perturbations due to motor parameters variations, the Tl
_
observation errors will converge to a small upper bound. V ¼ @ mfrd isq cV JA (24)
A ḟ rd afrd þ aMsr isd
Remark 7: Inequality (16) depends on the Lipschitz con-
stants introduced in Assumption 2. From Lipschitz con- and ṙ ¼ pV þ a(Msr =frd )isq .
stants, we can calculate the minimum value of u1 and u2 . Before carrying on the design of the controllers, let us first
Then we tune u1 and u2 in order to accelerate the conver- examine how to estimate the stator frequency (vs ). For the
gence of the observer. flux-oriented field frq ; 0, so that vs ¼ pV þa(Msr =frd )isq .
To avoid the uncertainties of IM parameters in the obser-
5 Flux-oriented control ver and achieve our goal (frq ; 0), we define
0
l1 l2
V_ oc do kek2Su (hf l1 j1 )kxf k2Pf For flux-oriented control, we have frq ¼ 0. Let us define
j1 j2
e frq ¼ frq 2 f̂rq ) f̂rq ¼ 2e 2frq, e frd ¼ frd 2 f̂rd ) frd ¼
(hV l2 j2 )kxV k2PV frd ¼ frd 2 f̂rd ) frd ¼ e 2frd þ f̂rd ṽs ¼ 2e vs þ vs , e Tl ¼
Tl 2 T^ l , B2(Z2) ¼ 2kmL2e frq
Let us define q1 ¼ (hf l1 j1 ) . 0, q2 ¼ (hV Consequently, (11) becomes
l2 j2 ) . 0, q3 ¼ (d l1 =j1 l2 =j2 ) . 0 and take a1
q ¼ min(q1 , q2 , q3 ), it follows that ė isd ¼ geisd e þ kc1 eisq þ abefrd þ vs eisq
det (S1 ) isd
V_ oc qVoc þ evs isq þ bpVefrq
d1 1
Hence, the estimations and the tracking errors of the aug- ė V ¼ ceV e e þ kc1 eisq
mented overall system converge asymptotically to zero as t det (S1 ) isd J Tl
tends to 1. þ mefrd isq þ mf^ rd eisq þ mefrq isd
Proof that controller (37) are well defined for all t 0.
Our goal here is to prove that f^ rd is different to zero at ė Tl ¼
g1 a
e kmf^ rd eisq kmefrq eisd
t ! 1. Let us define det (S1 ) isd
(47)
ecomf ¼ f frd )frd ¼ f ecomf
The coefficients of Riccati equation (S1) defined in (8) are
eobsf ¼ frd f^ rd ) f^ rd ¼ frd eobsf solutions of
S
S_ 23 ¼ u1 S23 þ 22 þ bpefrq S13
Equation (29) is the dynamic of flux tracking error, by J
choosing Kpfrd and Kifrd so that Āf is Hurwitz then 2S
e comf ! 0 at t ! 1. S_ 33 ¼ u1 S33 þ 23
J
Next,
p consider (17) and the following change of variable
v ¼ 2 Vo . The time derivative of v is given by So that a1 ¼ S22S33 2 S223, d1 ¼ S13S23 2 S12S33 , g1 ¼
S12S23 2 S13S22 and det(S1) ¼ S11 * a1 þ S12 * d1 þ S13 * g1 .
v_ dv þ cm (45) From (48), we note that S11 not tends to zero at t ! 1,
even if u1 . 0, ) det(S1) = 0. At t ) 1, the other coeffi-
where the solutions of (45) are cients (Sij) are very small because they depend to e frq . In
unobservability area, vs ¼ 0 and V is constant (V :¼ kV ,
cm in our case kV ¼ 20.5). Consequently, observer O1 (8)
v(t) v(t0 )ed(tt0 ) þ (1 ed(tt0 ) ) (46) may be seen as an estimator. Equation (47) becomes
d
ė isd ¼ geisd þ kc1 eisq þ abefrd þ evs isq þ bpkV efrq
By taking into account that the reference flux is
f ¼ 0.595, it follows that if e obsf = 0.595, then this 1
implies f̂ rd = 0 at all t . 0. A ė V ¼ ceV eTl þ kc1 eisq þ mefrd isq
J (49)
þ mf^ rd ei þ mef isd
sq rq
Our goal here is to prove the stability of the observer þ From (1), with V constant, we have
controller when the inputs are not persistent (i.e. unobserva-
bility area). T e ¼ f v kV þ T l (50)
S_ 11 ¼ (2g u2 )S11 þ 1
fv kV þ T^ l eT
mf^ rd ¼ mefrd þ l (51)
Jisq Jisq S_ 12 ¼ (g þ a u2 )S12 bpeV S11 peV S13
S_ 13 ¼ (g þ a u2 )S13 þ abS11 þ peV S12
Using (51), (49) can be rewritten as
S_ 22 ¼ (2a u2 )S22 þ 2bpeV S12 2peV S23
ė isd ¼ geisd þ kc1 eisq þ abefrd þ evs isq þ bpkV efrq S_ 23 ¼ (2a u2 )S23 þ bpeV S13 peV S33 þ peV S22 abS12
estimated motor speed (Fig. 1b) converges to the measured It appears a static error when the motor is under unobserva-
speed (Fig. 1a) near and under unobservable conditions. It is ble condition (between 7 and 9 s) (Figs. 2a and b). However,
the same conclusion for estimated flux (Fig. 1f) with respect the static error increases a little when the load torque is
to reference flux (Fig. 1e). The estimated load torque applied at time 1.5 and 5 s (Figs. 2a and b). In conclusion,
(Fig. 1d) converges to the measured load torque (Fig. 1c), we can say that the increase in the rotor resistance value
except under unobservable conditions (between 7 and 9 s). slightly affects the performance of the speed trajectories
Nevertheless, it appears a small static error when the tracking.
motor speed increases (between 3 and 6 s). In terms of per- A second test is made with a 250% variation on rotor
turbation rejection, we have noted that the load torque is resistance value. The experimental results are shown in
well rejected excepted at the time when it is applied Fig. 3. For the speed, flux and load torque estimation, the
((Fig. 1a, b, e and f ) at time 1.5 and 5 s) and when it is conclusion is the same as þ50% variation case (Fig. 2).
removed (Figs. 1a, b, e and f at time 2.5 s). But for this robustness test, the control induces noise.
The robustness of the observer þ controller is confirmed This can be seen on the measured torque (Fig. 3c).
by the result obtained with rotor resistance variation A new robustness test is made by a variation of þ10% on
(þ50%) applied to the observer and controller parameters rotor self-inductance value. The results of the test are shown
(Fig. 2). in Fig. 4. By analysing Fig. 4, we can see that the rotor self-
These figures display similar experimental results for the inductance variation does not affect the performances of the
rotor resistance nominal case under observable conditions. Control þ observer scheme. Nevertheless, it appears as a