IEEE ISIE 2006, July 9-12, 2006, Montreal, Quebec, Canada

Sensorless Control for Induction Motors via Fuzzy

Observer Design
Kuang-Yow Lian and Cheng-Yao Hung
Department of Electrical Engineering
Chung-Yuan Christian University
Chung-Li 32023, Taiwan
TEL: 886-3-2654815, FAX: 886-3-2654899
Email: lian dgdec.ee.cycu.edu.tw

Abstract-In this paper, sensorless control for induction mo- poses a fuzzy observer scheme to estimate the immeasurable
tors is developed based on fuzzy observer design. First, the T-S variables of rotor speed and rotor flux of an IM. In traditional
fuzzy model is used to exactly represent the induction motor. fuzzy observer design, the premise variables of fuzzy rules
Then, the fuzzy observer of the induction motor is straight-
forwardly constructed to estimate the immeasurable states of are assumed to be measurable. This is a strict constraint and
rotor speed and rotor flux, where the estimation gains are limits its application [13], [14]. For many physical systems,
obtained by solving a set of linear matrix inequalities (LMIs). including IMs, it is inevitable to use immeasurable states as the
This observer scheme gives the estimated states converging to premise variables. The immeasurable premise variables often
the real states exponentially. For the controller design, in light of lead to non void disturbance of which the effect make the
the principles of vector control, a new concept, namely virtual-
desired-variable synthesis is introduced to design the control law. estimation error only remain in a residue set. Fortunately, we
Finally, numerical simulations and experiments are carried out to will point out that the membership functions of fuzzy sets for
verify the theoretical results and show satisfactory performance. IMs satisfy a linear proportion type of property. The benefit
of this type of membership functions is that the estimation
error can converge to zero exponentially. Then, we obtain the
I. INTRODUCTION observer gains by solving a set of LMIs using MATLAB LMI
Induction motors (IMs) have been widely applied as electro- Toolbox.
mechanical actuators because of their ruggedness, low mainte- After finishing the speed and flux estimation design, the
nance, and low cost. Their usage in speed and torque tracking speed tracking control is investigated based on separation
control application is expected to be quite popular in the near principle, i.e., the estimation error is supposed to be zero in
future [1], [2], [3], [4]. For advanced servo applications, the the control design. At first, the speed tracking control is refor-
control technique mainly relies on high performance IM drive, mulated into the torque tracking problem. To achieve vector
where the need of precise rotor speed for feedback is essential. control, a set of virtual desired variables (VDVs) is introduced
Optical encoders are usually used to detect the rotor speed. to synthesize the controller. The VDVs are determined in a
However, these speed sensors weaken the ruggedness, reliabil- straightforward manner based on the goal of achieving all of
ity and simplicity of an IM. It seems quite clear that the speed the desired torque, the constant desired flux, and the system
sensors cannot even be mounted in a hostile environment. stability. In the design, a skew-symmetric property pertaining
In view of these points, many researches have focused on to the dynamics of the IM is utilized to simplify the structure
sensorless control. Due to the high order nonlinearity of IM's of the controller. To demonstrate the effectiveness of the
dynamics, the estimation of the rotor speed and flux becomes proposed scheme, a voltage-fed model of drive system is set
a challenging problem. To overcome these difficulties, various up to perform the task of speed tracking. The simulation and
control algorithms have been proposed in the literature [5], experimental results illustrate nice performance, even though
[6], [7]. In this paper, we propose a new scheme for speed in the case of low speed command is considered.
sensorless control for the full fifth-order model of IMs. The II. DYNAMICAL MODEL OF INDUCTION MOTORS
proposed scheme is mainly based on the fuzzy observer design.
In recent years, fuzzy control has been widely applied to Let (isa, isb), (Aral Arb) and w denote the components of
deal with nonlinear systems. Many researches on this issue the stator current, rotor flux, and rotor speed, respectively. The
are carried out based on Takagi-Sugeno (T-S) fuzzy models induction motor is represented by a fifth-order model [15]:
[8], [9]. The T-S fuzzy approach has been extensively used to Zsa = ( Uf+cL2 )isa + c,2Ara +--
crL A,>b + vUl
model nonlinear systems. The basic idea for the approach is to
decompose the model of a nonlinear system into a set of linear =sb ( +L )isb *L WAra +I U2
Arb-LTA
subsystems with associated nonlinear weighting functions. The karsaA raWArb
-
stability analysis is carried out using Lyapunov direct method
whereas the control problem is then fornulated into linear Arb L Zsb L b + WAra
matrix inequalities (LMIs) [10], [11], [12]. This work pro- =T T-) _w (1)

1-4244-0497-5/06/$20.00 © 2006 IEEE 2140

where (u1, u2) denotes the stator voltage; R5, Rr, L5, Lr Plant Rule i:
and Lm are the stator resistance, rotor resistance, stator induc- IF w is Fli and Ara is F2i and Arb iS F3i THEN
tance, rotor inductance, and mutual inductance, respectively; (t) =AiC (t) + Bu (t) + b (6)
a = Ls- L2/Lr; J, D, T1, are the mechanical inertia,
y (t) = Cx(t), i =1, 2, .. 18
natural damping, and loading torque, respectively; and T is
the electromechanical coupling torque expressed as: where W, Ara, and Arb are premise variables which are
immeasurable. The fuzzy sets Fji (j = 1, 2, 3) and the system
n.Lj (Ararsb -Arbsa) '
T = (2) matrices Ai of subsystem i are set as follows:
D1 i
where np = (pole pair) x 3. Let us denote x F= l (z1) = d d+D1+ d F12 (Z1) Di-dil
X3
D1i-d
[X1 X2 X3 x41 = [ isa isb Ara Arb I The F21 (z2) -
d2 +D2 d2 F22 (Z2)
D2
D2 -d2
X4
D2-d2
model (1) can be rewritten as F31 (Z3) =D -3-d3 + Dd D3
D3-d3, F32 (z3)
3 (3 D3 -d3
X5
D3-d3
M3+G(w)x+R(w) = T (3) -A A12 A13 A14'Oi A15
JA + Dw = T-T (4) A21 A22 -A23 O A24 A25
where
Ai = A31 A32 A33 -A34 O A35
A41 A42 A43 O A44 A45
M [ Lr,oI2 0] G(w)- 00 0 L- A51 i A52 'S A53 A54 A55
-J2 J where 01 = D3, V1 D2, O I= D1, 02 D3, 02 D2,
= = =

I2 [I J2
[ 0
1
-1
0
Y02= di, 03 D3, 3 = d2, 503= D1; 04= D3, 4 = d2,
('4 = d, 05= d3, V5= D2,i50= D1, 06= d3, V6= D2,
9P6 = d1, 07= d3, V'7 d2, 07 = D108 = d3, V8 = d2,
508 = di. Using the singleton fuzzifier, product fuzzy inference
ImJ2
_Lm'2
-

I and weighted average defuzzifier, the final output of the fuzzy

L
R(w) =[ LR Rr
'2
r2 Lr system is inferred as follows:
8
T = [ LrUi LrU2 0 0 ]', 7 =
R,
+ c2 ;z (t) E -ti(x(t)) {Aix (t) + Bu (t)} + b
i=l
To express the induction motor (3) and (4) in terms of T-S y(t) = Cx(t), (7)
fuzzy model, we rewrite the equations in the following form: 8
ti (x (t)) = (x (t)) / E (x (t)) with where
;z (t) (x(t)) = H=1 Fji (x (t)). Note that i=1 i ((t)) =1
A (x) x (t) + Bu + b i
y (t)
Cx(t) for all t, where ,ui (x (t)) > 0 for all i = 1, 2, **, 8. Based on
the setting of Fji and Ai, it can be checked that the inferred
where the overall states x (t) = ( X X5 ), i.e., x (t) = output is exactly equivalent to the model of induction motor
[ isa isb Ara Arb wJ ] '; the measurable output y (t) = (5). (-)Moreover, we note that all the membership functions
[isa isb]T; the control input u = [u1 u2]T, and the associated Fij satisfy the following property:
matrix and vector: Fij (x t) Fij (X-(t)) =Tlij (x() ()
All A12 A13 A14w A15 for some constant rij. Since Pi ( (0)
A21 A22 -A23w A24 A25 Fli (x) F2i (x) F3i (x), it follows that
A(x) = A31 A32 A33 -A34w A35
A41 A42 A43w A44 A45 Pi( (t) - /-i('(t))
-AslArb As2Ara A53 A54 A55 i (X - )F2i (X) F3i (X) + Fli () yi22(X -x)F3i (X)
+Fli () F2i(x) Tyi3(X -)
T
B= a
Qoc
I
o a O T , b= [0 0 0 0 _
J AT (X (t) x- (t))
where AT [TIyIF2i (x) F3j (x) + Fli (x) y2F3i (x) +
C =1 0
0 1 0 0 0
0 0 01 Fli () F2i(x) Ty31. Hence the grade functions ,ut have a nice
property:
where AI1 = L2_R L,R, Property: The error of grade functions _t (x(t)))-t (x(t)) is
A22 = (L + ), A31 A42 proportional to the error (x(t) -x(t)).
- = =
Lr 7
A51 = A52 _ n,L, A13 A24 L2 A14 A23 =
JL, 7
= = , =
Before designing the observer and controller, some assump-
,L A33 = A44 = - Lr A34 = A43 1, A55 A
D
J tions are made as follows:
A12 = A15 = A21 = A25 = A32 = A35 = A41 = A45 =:::: A.1 The loading torque T1 is a known constant.
A53 A54 = 0. Then, according to [13], the T-S fuzzy model A.2 The parameters J, D, Lm, L,r R5, Rr are known
representation of (5) can be expressed by the following rules: constants.

2141
A.3. The desired speed wd is a smooth and bounded signal. According to (12), we obtain
A.4. The stator current and voltage are measurable, while the 8
rotor flux and speed are assumed to be unavailable. V0 (e) < E
i=l
pi (x')e (t)' Gie
III. FuzzY OBSERVER OF INDUCTION MOTORS where Gi = (Ai- LiC)T P + P (Ai -LiC) + UTU + PP.
In this section, we will design the fuzzy observer to estimate Hence, the exponential convergence of e (t) is concluded if
the immeasurable states under Assumptionsl-4. According to Gi < 0. Using Schur complement [16] and letting PLi = Zi,
the fuzzy model (5), the fuzzy observer is given as follows: the inequality Gi < 0 can be converted to the following LMIs:
Observer Rule i: [ATPP+PAi -CTZfT -Z'c+U U P <
IF w is Fli and Ara is F2i and Arb is F3i THEN
(8) Vi=1, 2, , 8.
(t) = Ai;(t) +Bu(t) +b +Li (y(t)-(t)) (13)
S (t) = C;z(t) I i = 1, 2,. ,8 Design of the Fuzzy Observer: For the fuzzy observer (8),
where the premise variables W, Ara, and Arb are the estimated suppose that all states and control input are bounded. If there
states of W, Ara, and Arb, respectively; x (t) and y (t) denote exists a common positive definite matrix P and Zi such that
the estimations of x (t) and y (t), respectively; and Li is an the LMIs (13) are feasible, then the estimation error converges
observer gain to be determined. The inferred output of the to zero exponentially by letting observer gains Li = P-'Zi
observer is We can solve LMIs (13) using powerful packages like
MATLAB Toolbox to obtain P and Zi. In turn, the observer
8
:; (t) gains are calculated from Li = P lZi.
E pi (x(t)) {Ai- (t) + Bu (t) + b
i=l
IV. CONTROLLER SYNTHESIS BY VIRTUAL DESIRED
+Li(y (t)-y (t))} VARIABLES
y (t) C:z (t) . (9) The concept of control design for a sensorless induction
Define the state estimation error e (t) = x (t) - (t). Subtract- motor is shown in Fig. 1. For simplicity, we assume the
ing (7) by (9), it leads to fuzzy observer provides a perfect estimation of x, i.e., we
let A = A and w = w. The use of separation principle in the
8
controller design is suggested by the exponential convergence
e (t) = pi (x(t)) {(Ai -LiC) e} + 1 (t) (10) of estimation error, which is endowed with the robustness for
i=l
some amounts of uncertainty.
where rl _1
8
(t) =
I
i=l
(ti (X)l-H(X:))
i {Aix (t)} (1 1) Fuzzy ~0),IA MechanicalI IvvIV
Observer-Based part
The term 1 (t) in (11) arises due to immeasurable premise
variables W, Ara, and Arb. Recalling the property in Section .I Electrical
II, we have I
i part
8 8 L -----__
lTl = CT EAi {Aix (t) } E {Aix (t) } AT e-
i=l i=l vVsu, L/vn Vsw
Supposed that x (t) are bounded (this will be confirmed in
controller design given later), the term 1 (t) has the following _ IM s

bounded fashion
TIl < eTUTUe (12) . . . .E er. . . . . . . . . . . . . . . . . . . . . . .

with a constant matrix U. This undesired term 1 (t) will Fig. 1 The concept of control design for sensorless induction
affect the estimation performance. Its effect, however, can be motor.
exponentially attenuated to zero by suitably choosing observer
gains Li. Now, we apply Lyapunov method to get the observer A. Speed Tracking Control
gainsLi,i =1,2, ,8.
Choose the Lyapunov function candidate V0 (e (t)) = First, denote the speed tracking error as w w- Wd. The
4 (t) Pe (t). Taking the time derivative, we have tracking error dynamics can be rewritten as
8 Jw + (D + kw) = T -Td + (Td- YO + kw), (14)
Vo (e) < +Zi=l Hi P()e'K(Ai -LiC)P+P(Ai LiC)] e
where Td denotes the desired torque which produces
+C Tppe + lTl. the desired speed; k, is an adjustable damping ratio;

2142
Y= [1 CDd wdl is the regression vector; and the parameter For the last term, we rewrite it as (T- T) w = _X';,
vector 0= [Ti J D]T. For speed tracking control, the desired where L, =
[ X4d X3d X2 -x1 .Since R (w)
torque is selected as depends on w, we further re-express (20) as
Td = YO -kw, -LrJ2w ] ~+ f
V7c = -s X(+
kw[ 0
This yields the following dynamics
error
- (D +kwA) 6J2 + f;T
Jw + (D + kw) = T -Td (15) where
Since the damping ratio D of induction motors is usually Lro7yI2 + K-'2 &LtR I2
L,R, R
small, the damping term k,Cw plays a dominant role on Lr r2 1-
improving the transient response for speed tracking. If T -Td
is driven to zero, the rotor speed will converge to the desired After setting p= [ - 12x -, we obtain
value. In other words, the speed tracking control problem has
been reformulated into the torque tracking problem, where the V (t),w) xRx -(D + k,)2
objective is T Td. The concept of virtual desired variables
--

(VDVs) is introduced in the following to achieve the objective. It can be checked that the matrix R > 0 by choosing
B. Concept of VDVs K >-LrRs.

The VDVs consist with the virtual desired current Hence the exponential stability is shown if the VDVs are well
(X1ld X2d) and virtual desired flux (X3d, X*4d). They will be defined.
specified by satisfying
(i) the desired torque D. Specijying VDVs
In the remaining design procedure, control law u and VDVs
nL. (Xx2dX3d-XldX4d);
-Xd are chosen such that (P = 0. The perturbed term (19) is
Td = (16)

(ii) the constant desired flux explicitly rewritten as:

2 2
c= X3cd + X4d;
2
(17) p
= [ (pl (p2 (p3 (p4 I

(iii) the stability when the true state variables (t) track the x LrU1 LrUi7;d + LmWX4d -XL(7yld +L 3d
VDVs. Lr U2 LrU7;2d- LmWX3d- LrO77X2d + X4d
Notice that the condition (17) for desired flux is to achieve -X4d - JX4d + L Xld -RL X3d 1-(P
the optimal torque [4]. Define the error signal for the electrical --'C4d + bJXf3d +LLR, X2d -R
L 4d
part as 5 = -d, where Xd = [ Xld X2d X3d X4d ] -
x (21)
The control objective of steering T to track Td can be achieved To satisfy (p3 = (p4 = 0, we can obtain
if -÷ 0. To this end, the equation (3) is rewritten in terms
of as X3d X3d LmRr 1-ld

';4d (_ RrI2 + JJ2)

X4d
+
Lr X, 2d
Mx+G(w)x+R(w) = p + p (18) I J

X1
+J L
Lr PJ2 (22)
where (P is regarded as a perturbed term with the following [
Xf2
form
From (17), ( X3d X4d ) = ( coCs (p (t)) csin (p (t)) )
(p = T [Mid + G (w)Xd+ R (w) Xdl ~p. (19) for variable p (t) to be determined later. It follows that
We will intend to set (P = 0 to specify the VDVs, which will X3d PX4d = X3d
(23)
be addressed in Subsection D. X4d J PX3d J X4d J

C. Stability Analysis Substituting (23) into (22), we obtain

To analyze the stability of the system and design (P as well, Xld 0 = 1 (r wp)J2+I2)[X3d]
we choose the following Lyapunov function candidate:

VC (t)i ) 2 (t)T M (t) + ijsJ2

-wnpJ2
C 1 (24)
Rr L 2
The time derivative of Vc along (I15) and (I18) is Since the desired states also satisfy (16), substituting (24) into
(16) yields
VC (x (t), w) (t)T R (w)j2 (t)(T+ (t)T -(
(D + kwg) + -Td) (20) Lc 'A
-

-
p (t)
= np c2 Td + +
L (X1X3d +
X2X4d),

2143
where p (t) is thus defined. On the other hand, to satisfy (P1 the power inverter. Meanwhile, a set of I/0 modules are
(p2= 0 in (21), the control law is formulated as follows: constructed in the card for the voltage/current measurement,
encoder interface and some protections. For the purpose of
U2 Xld lFXldlI XIi
comparison, the motor's instantaneous speed is measured by
+ Lr
U2 [ X2d [X2dJ [X2J an optical incremental encoder with 1000 pulse/revolution.
+L'wJ
The software we adopt is Simulink 3.0 and Matlab 5.3. In
[ (('L + L, ) J2 addition, the system combines the motor control card with the
L_ R, _2 [X3d (2
Simulink/Real Time Workshop Toolbox such that the setup
of the control law in the simulation can be directly applied to
L

T
X4d

the experiment. The overall execution time interval is set as

Indeed, the implementation of the control law (25) is com- 6 sec for both simulations and experiments.
plicated due to the first term on the right-hand side, which Consider tacking of speed wd 30 + 20 sin(X) rad/sec. The
includes the time derivative of Xld and X2d. But, we note control parameters are chosen as: k, = 0.47, c = 0.53. The
that the exponential stability shown in Subsection C give a simulation and experiment results of speed tracking for desired
very nice robustness to the uncertainty. This property permits and actual speed are shown in Fig. 2(a). The maximal tracking
us to approximate the 'id by using aid Xid-Xid, where error are 0.3 and 1 rad/sec for simulation and experiment,
Xid + Xid = Xid. The resulting simplified control law will be respectively. The desired and estimation rotor speed are shown
adopted in our simulation and experiment. in Fig. 2(b). The maximal tracking error are 0.4 and 0.5
V. EXPERIMENTAL RESULTS rad/sec for simulation and experiment, respectively. The actual
and estimation rotor speed are shown in Fig. 2(c). The speed
In this section, the performance of the control scheme will estimation error (w -) is shown in Fig. 2(d). The experiment
be verified by numerical simulations and experiments. For results of stator voltage and stator current for one phase are
comparison, numerical simulation results are put next to asso- shown in Figs. 3.
ciated experiments results. The specifications and parameters From these figures, we can find that the tracking error
of the induction motor are listed in Table I. will tend to zero asymptotically when time goes to infinity.
Furthermore, the stator current response, stator voltage, are
TABLE I satisfactorily. From the results, we can conclude the perfor-
THE SPECIFICATION AND PARAMETERS mance of the proposed control scheme.
OF THE INDUCTION MOTOR
VI. CONCLUSION
Rated Specification A fuzzy observer-based controller has been proposed to
Pole Pair 3 achieve speed tracking. To this end, some new concepts, such
as virtual desired variables and Lipschitz-like condition are
Power 0.4 kW
introduced to benefit the control design. Here, the general
Voltage 120 V fuzzy model of induction motors is used to accomplish the
Current 3.4 A design. The T-S fuzzy observer algorithm has been developed
Speed 1500 rpm for the estimation of the rotor speed and the flux of an
induction motor. The observer gains are obtained by solving
Parameters a set of LMIs. The two-stage design technique is applied to
R, 2.85Q construct a controller for speed tracking control. The numer-
Rr 4.0Q ical simulations and experimental results have illustrated the
expected performance and indicate that the integration of the
LI 0.19667 H fuzzy observer and VDV-synthesis controller are very suitable
Lr 0.19667 H in induction motor applications.
Lm 0.1886 H ACKNOWLEDGMENT
J 0.001 kg n2 This work was supported by the National Science Council,
R.O.C., under Grant NSC 93-2213-E-033-008.
According to LMI (13), where we let U REFERENCES
diag {0.9, 0.9, 0.9, 0.9, 0.9}, the observer gains obtained [1] J. Li, L. Xu and Z. Zhang, "An adaptive sliding-mode observer for
via LMI toolbox of Matlab. The induction motor is driven induction motor sensorless speed control," IEEE Trans. Ind. Appl., vol.
41, no. 4, pp. 1039-1046, Jul./Aug. 2005.
by a PC-based DSP controller. The overall system consists [2] A. M. Lee, L. C. Fu, C. Y Tsai, and Y C. Lin, "Nonlinear adaptive speed
of the PC, DSP controller card, power inverter, encoder, and torque control of induction motors with unknown rotor resistance,"
and an induction motor. For implementing the control law, IEEE Trans. Ind. Electron., vol. 48, no. 2, pp. 391-401, Apr. 2001.
[3] K. K. Shyu, L. J. Shang, H. Z. Chen, and K. W. Jwo, "Flux compensated
the DSP card provides not only a high speed floating-point direct torque control of induction motor drives for low speed operation,"
computation but also a 100kHz PWM signal generation to IEEE Trans. Power Electron., vol. 19, no. 6, pp. 1608-1613, Nov. 2004.

2144
 [4] K. Y Lian, C. Y Hung, C. S. Chiu, and P. Liu, "Induction motor control
with friction compensation: an approach of virtual-desired-varuable cn
synthesis," IEEE Trans. Power Electron., vol. 20, no. 5, pp. 1066-1074, II

Sept. 2005.
[5] H. T. Lee, L. C. Fu, and H. S. Huang, "Sensorless speed tracking
of induction motor with unknown torque based on maximum power II II
transfer," IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 911-924, Aug. 3 4 5 6
2002. 1 1~~~~~~~~~~~~~~~~~~~~~~~~~
[6] M. Feemster, P. Aquino, D. M. Dawson, and A. Behal, "Sensorless rotor
velocity tracking control for induction motors," IEEE Trans. Contr Syst. cn
Technol., vol. 9, no. 4, pp. 645-653, Jul. 2001.
[7] J. Holtz and J. Quan, "Sensorless vector control of induction motors
at very low speed using a nonlinear inverter model and parameter II
identification," IEEE Trans. Ind. Appl., vol. 38, no. 4, pp. 1087-1095,
Jul./Aug. 2002. 3 4 5 6
[8] T. Takagi and M. Sugeno, "Fuzzy identification of system and its (c) Time(s)
applications to modeling and control," IEEE Trans. Syst., Man, Cybern.,
-1-
vol. SMC-15, no. 1, pp. 116-132, 1985.
[9] Q. Gan and C. J. Harris, "Fuzzy local linearization and local basis cn
function expansion in nonlinear system modeling," IEEE Trans. Syst., --1-
Man, Cybern. B, vol. 29, no. 4, pp. 559-565, Aug. 1999. I1
3 4
[10] K. Tanaka and H. 0. Wang, Fuzzy Control Systems Analysis and Design:
A Linear Matrix Inequality Approach. New York: Wiley, 2000.
[11] X. J. Ma, Z. 0. Sun, and Y Y He, "Analysis and design of fuzzy
controller and fuzzy observer," IEEE Trans. Fuzzy syst., vol. 6, no. 1, Il
pp. 41-51, Feb. 1998.
[12] A. Jadbabaie, M. Jamshidi, and A. Titli, "Guaranteed-cost design of cn a .w_
Times
continuous-time Takagi-Sugeno fuzzy controller via linear matrix in-
equalities," in Proc. FUZZ-IEEE, May 1998, pp. 268-273.
[13] K. Y Lian, C. S. Chiu, T. S. Chiang, and P. Liu, "Secure communications II l
of chaotic systems with robust performance via fuzzy observer-based
design," IEEE Trans. Fuzzy syst., vol. 9, no. 1, pp. 212-220, Feb. 2001. 0 1 2 3 4
[14] B. S. Chen, C. S. Tseng, and H. J. Uang, "Mixed H21Ho, fuzzy (d)
output feedback control design for nonlinear dynamic systems: an LMI
approach," IEEE Trans. Fuzzy syst., vol. 8, no. 3, pp. 249-265, Jun. Fig. 2 The simulation and experimental results: (a) desired
2000.
[15] B. K. Bose, Power Electronics and AC Drives. Englewood Cliffs, NJ: (-) an d actual (--) speed, (b) desired (-) and estimated (--)
Prentice Hall, 1986. speed, (c) actual (- -) and estimated (--) speed, (d) speed
[16] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
Inequalities in System and Control Theory. Philadelphia, PA: SIAM, estimation error.
1994.

50 100 -1-M--- --

cn

I~~ . . I
0
-100 -I1- - - - - I - -

U) o
0 1 2 3 4 5 6

-cn50 0v
CZ

cn
U)
0 1 2 3 4 5 6 0 1 2 3 5 46
(a) Time(s) Time(s)
Fig. 3 The experimental results of stator voltage and stator
50 , current for one phase.
,

II
-cn 0
0 1 2 3 4 5 6

50

,. . .~~~~~~~~I
0 1 2 3 4 5 6
(b) Time(s)

2145