1226 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO.

6, JUNE 2000

Robustness Margins for Indirect Field-Oriented Control where x1 and x2 represent, respectively, the quadracture axis compo-
of Induction Motors nent and the direct axis component of the rotor flux, w is the rotor
speed, u1 , u2 , and u3 stand for the inputs—the slipping frequency, the
A. S. Bazanella and R. Reginatto direct axis stator current component, and the quadrature axis stator cur-
rent component, respectively; Tm is the load torque, which is assumed
constant, and the “c” parameters are all positive. In particular, c1 rep-
Abstract—The influence of the rotor time constant mismatch on the sta-
resents the inverse of the rotor time constant, which is a critical param-
bility of induction motors under indirect field-oriented control is analyzed.
The results of [14] are generalized. A Lyapunov function which provides a eter for indirect field-oriented control. The model considered in [14] is
global stability test and allows us to compute robustness margins is given. expressed in a reference frame fixed in the rotor and is, upon a non-
Different mechanisms for the loss of stability are detected by means of bi- singular (u1 -dependent) change of coordinates, equivalent to (1)–(3)
furcation analysis. Robustness margins and design guidelines are derived with c3 = 0. See [37], [41], and [40] for further details regarding the
from these results.
derivation of the induction motor model.
In speed regulation applications the indirect field-oriented control
I. INTRODUCTION strategy is usually applied along with a PI speed loop. This control
Indirect field-oriented control (IFOC) is a well established and strategy is described by the following equations [37], [14]:
widely applied control technique when dealing with high performance u3
u1 = c^1 (4)
induction motor drives [37], [41], [40]. Yet, it was not until very u2
recently that this control scheme has been provided a firm theoretical u2 = u2
0 (5)
foundation [15], [23], [14]. t
The comissioning of an IFOC requires the knowledge of a single u3 = kp (wref 0 w) + ki ref ( ) 0 w( )) d
(w (6)
motor parameter, namely the rotor time constant, which can vary 0
widely in practice [36], [22]. The robustness of this control strategy where c^1 is an estimate for the inverse rotor time constant c1 , kp , and
against mismatches in this parameter has been analyzed in [14], where ki are the gains of the PI speed controller, wref is the constant reference
the robust global stability of the operating point has been established velocity and u20 is some constant which defines the flux level.
from a qualitative standpoint. A range of the parameter mismatch that If c^1 = c1 , that is, if we have a perfect estimate of the rotor time
guarantees the uniqueness of the equilibrium point for any loading constant, we say that the control is tuned, otherwise it is said to be
condition has also been given. detuned. Accordingly, we define
In this paper we analyze the effects of parameter mismatches on the
behavior of induction motors under IFOC in speed regulation tasks. In =
1 c^1 (7)
Section II the system modeling and the control equations are given. c1
Then we generalize and deepen the results of [14] in two directions. as the degre of tuning. It is clear that  > 0 and the control is tuned if
First, in Section III, the robustness of the global stability property is and only if  = 1.
given a quantitative measure by means of a Lyapunov analysis sim- The closed-loop system (1)–(3) with the control (4)–(6) is a fourth-
ilar to that in [14]. Second, in Section IV, a bifurcation analysis is pre- order system that can be described as follows:
sented in terms of two parameters, namely the motor load and the rotor
x_ 1 = 0 c1 x1 + c2 x4 0 c 1 x2 x4
u20
time constant mismatch. Conditions for the ocurrence of saddle-node (8)
bifurcations are given, providing bounds on the two parameters that
avoid such bifurcations and guarantee uniqueness of the equilibrium.
0 c1
x_ 2 = 0 c1 x2 + c2 u2 + 0 x1 x4 (9)
Finally, a discussion of the results presented is given in Section V, de-
u 2
riving some design guidelines. All symbolic manipulations and plots 0
x_ 3 = 0 c3 x3 0 c4 c5 (x2 x4 0 u2 x1 )
presented have been performed with MATLAB V.
0Tm 0 cc34 wref (10)
II. PROBLEM STATEMENT
x_ 4 = (ki 0 kp c3 )x3
The dynamic model of a current-driven induction motor expressing
the rotor flux and the stator currents in a reference frame rotating at 0 kp c4 c5 (x2 x4 0 u20x1 ) 0 Tm 0 cc34 wref (11)
synchronous speed is given by [37], [41], [40]
where we have defined the new state variables x3 = wref 0 w and
1
x_ 1 = 0 c1 x1 0 u1 x2 + c2 u3 (1) 1
x4 = u3 .
x_ 2 = 0 c1 x2 + u1 x1 + c2 u2 (2) We shall analyze the stability properties of the closed-loop system
w_ = 0 c3 w + c4 [c5 (x2 u3 0 x1 u2 ) 0 Tm ] (3) (8)–(11) and its dependence on the loading conditions and the degree
of tuning .

Manuscript received July 27, 1998; revised January 20, 1999 and July 7, 1999. III. LYAPUNOV ANALYSIS
Recommended by Associate Editor, L.-S. Wang. 1
Let us define a change of coordinates z = x xe , where xe = 0
A. S. Bazanella is on leave from the Department of Electrical Engineering, e e e e T
[x1 x2 x3 x4 ] represents a generic equilibrium point. The system can
Universidade Federal do Rio Grande do Sul, Av. Osvaldo Aranha 103,
90035-190, Porto Alegre, RS, Brazil. be described in these coordinates as
R. Reginatto is with the Department of Automation and Systems, Universi- e
dade Federal de Santa Catarina, 88040-900 Florianópolis, SC, Brazil.
z_1 = 0 c1 z1 + c2 0 x2uc0 1 z4 0 z2 (z4 + xe4 ) c
u0
1 (12)
Publisher Item Identifier S 0018-9286(00)06022-0. 2 2
0018–9286/00$10.00 © 2000 IEEE
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000 1227

z_2 = 0 c1 z2 + c 1 [xe z + z (z + xe )]
u02 1 4 1 4 4
(13) The Lyapunov derivative (19) can be put in a quadratic form as

z_3 = 0 c3 z3 0 c4 c5 [z2 (z4 + xe4 ) 0 z1 u20 + xe2 z4 ]

shown in (20), at the bottom of the page.
(14) Define also the quadratic polynomial
z_4 =(ki 0 kp c3 )z3 0 kp c4 c5 [z2 (z4 + xe4 )
0 z1 u20 + xe2 z4 ]: (15) p(m) = p2 m2 + p1 m + p0 (21)
2 0 e 2
p2 = 0 c1 3 x1 c10 + c2 u2 0 0x2 c1 
e
Define also, for convenience of notation 2u2 2u2 (22)

=1 c1 > 0 2 0 e
p1 = 0 1 3 x1 c10 0 2c1 3 14 c2 u2 0 0x2 c1 
e
u20 c4 c5
1 2 ki > 0
2u2 2u2
k2 = e 2
c2 + 132 x21uc10 + c12 3 4 (23)
k3 =  c3 kp > 0:
1 2 2
ki 2 0 13
p0 = c1 (1 3 4 0 3 14 2 4 ): (24)
We will study the stability of the origin z = 0 by means of the Then we have the following result, which provides a robust stability
quadratic Lyapunov function candidate
test and establishes the robust global stability in the tuned condition.
V (z ) = 12 zT (P1 + mP2 )z (16) ( )
Theorem 1: Let p m have distinct real roots m1 , m2 ordered such
that m2 > m1 . If
where 2
kp2 + k 0 0k2 0kp  m2 > 13 0 1 3 =
1m
0 (25)
c1 3
P1 =
0 0 0 0
0k2 0 kp2 k3 + k2 0kp k3
(17) then the origin of the system (12)–(15) is globally asymptotically
0kp  0 0kp k3 k3 + 2 stable. }
1 0 0 0 Proof: It is clear from (20) that if there exists an m such that
all the leading minors of the symmetric matrix Q are positive then the
P2 =
0 1 0 0
0 0 0 0 (18) origin will be globally asymptotically stable. The leading minors of
0 0 0 0 first and second order are always positive, whereas the third and fourth
order minors are given respectively by
and m is a positive scalar to be assigned. The matrix P1 +
mP2 is
0
symmetric positive definite for any m > and any possible operating 13 (m) = mc1 [(1 + mc1 )3 0 132 ] (26)
condition. 14 (m) = mp(m): (27)
The time derivative of this Lyapunov candidate can be calculated as
If the roots of p(m) are real and distinct, then p(m) > 08m 2
V_ (z ) = 0 1 z12 0 3 z32 0 4 z42 0 213 z1 z3 + 214z1 z4 (m1; m2 ), since p2 is negative; then 14 (m) is positive in this interval.
e
0 mc1 z12 0 mc1 z22 + m x1uc01  z2 z4 On the other hand, 13 (m) > 0; 8m > m0 , so that all the leading
2 minors will be positive for m 2 (maxfm0 ; m1 g; m2 ). Condition (25)
e
+ m c2 0 x2uc01  z1 z4 (19)
guarantees that this interval is not empty. }
2 The robust global stability of the IFOC close to the tuned condition
for general systems can be established as a direct consequence of The-
where orem 1.
1
1 = c1 ( + 1)(k2c + k ) > 0
p 2
Corollary 1: The equilibrium of the system (12)–(15) is globally
i
c2
c3 2 (k2 c + k ) > 0
asymptotically stable provided that  =1 or sufficiently close to this
1
3 = }
p 2 i value.
1
c2 Proof: As the tuned condition  ( = 1) is approached, p2 tends
4 = c2 kp  > 0 to zero by the left and p1 tends to a finite positive value, since xe1 and
xe2 approach 0 and c2 =c1 u20 , respectively. Hence one of the roots of
( )
1
13 = 0 2 cc12 ( + 1)ki  + cc32 (kp2c2 + ki ) 0 kp ki ( ) +
p m tends to 1 and, as m0 remains finite, there exists an interval
1 1 [k  + k2 c + k c ( + 1)] > 0 of values of  containing the point  =1 such that condition (25) is
14 = }
2 i p2 p 1 satisfied.

V_ (z ) = 0 zT Qz 0
014 0 c2 u2 20u0x2 c1  m
e
1 + c1 m 0 13
xe c1  2
Q= 0 c1 m 0 0 21u0 m
02
(20)
13
0 e
0 3
014 0 2 20u0x2 c1  m
c2 u xe c1 
1
0 2u0 m 0 4
2 2
1228 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000

Given the complexity of the expressions involved, it is impossible Collecting (33)–(35) and the definition of r we can write the equi-
at this point to make general statements about the conditions of The- librium point as

c2 u20 1 0 
orem 1, unless in the important case of the corollary above. However,
this theorem provides a simple robust stability test for any particular r
xe1 c1 1 + 2 r2
system, allowing us to establish robustness margins by the verification
xe2 0
c2 u2 1 + r 2
of condition (25) for different values of . :
c1 1 + 2 r2
= (39)
xe3
0
xe4
IV. BIFURCATION ANALYSIS u20 r
A. Parameterization of the Equilibria
The equilibrium is now parameterized in terms of a single di-
We start our analysis by studying the equilibria of the system mensionless quantity r , which satisfies (38). This is a third order
(8)–(11) and their dependence on the degree of tuning . The equi- polynomial equation whose coefficients—which are dimensionless as
libria of this system are the solutions xe1 , xe2 , xe3 , xe4 of the system of well—depend only on the degree of tuning  and the motor load as
equations denoted by r3 .
0 c1 xe1 + c2 xe4 0 c 1 xe xe = 0
u02 2 4
(28)
B. Uniqueness of the Equilibrium
0 c1 xe2 + c2 u20 + c 1 xe xe = 0 The real solutions of (38) give the equilibrium values of r for any
u0 1 4
(29)
2 given degree of detuning——and any given load—r3 . It is clear that
0 c3 x3 0 c4 c5 (x2 x4 0 u20xe1 ) 0 Tm 0 cc34 wref
e e e
=0 (30) (38) has at least one and at most three real solutions, depending on the
particular values of  and r3 .
(ki 0 kp c3 )x3e 0 kp c4 c5 (x2e xe4 0 u20 xe1 ) 0 Tm 0 cc34 wref = 0:
Theorem 2: Assume that

(31) > 3 (40)

The equilibrium values of the fluxes can be obtained rewriting (28) and consider the inequalities shown in (41) and (42), at the bottom of
and (29) as the page.
1 If (41) and (42) are both satisfied strictly, then the system (8)–(11)
c1 c1 r xe1 r
c2 u20 has three equilibrium points. If either (41) or (42) holds with equality,
0c1 r c1 xe2
=
c2 u20
(32)
then the system has two equilibrium points. Otherwise, the equilibrium
1
point of this system is unique. }
where we have defined the dimensionless variable r = (x4e =u20 ). Proof: Rearranging (38) we have r3 as a function of r
Solving (32) for xe1 , xe2 yields
r2 + 1
x1 = 2
ec
u20
1 0 r3 = f (r) = r 2 2
 r +1
: (43)
c1 1+ 2 r2 r (33)
c2 1 + r
2 Equation (38) has a unique real solution whenever f (r) is bijective. It
e
x2 = u20 :
1 +  r2
2 (34) is clear that f (r) is continuous and injective, so that it will be bijective
c1
if it is strictly monotonic
Now, from (30) and (31) it is clear that
df (r) 2 r2 + 1)(3r2 + 1) 0 r(r2 + 1)2r2
(
xe3 = 0 =
(2 r 2 + 1)2
(35) (44)
1 T + c3 w : dr
c5 (x2e r 0 xe1 )u20 = Te =
c4 ref  r + (3 0 2 )r2 + 1
2 4
m (36)
= :
(2 r 2 + 1)2
(45)
Substituting (33) and (34) into (36) we get
c2 0 2 r3 + r Te The sign of the derivative depends only on the sign of the polynomial
(u ) :
c1 2 1 + 2 r2
= (37)
c5 in the numerator, since the denominator is always positive, as well as
. Thus the function f (r) is bijective for all values of r3 such that
1
Let us define also r3 = (Te c1 =c5 c2 (u20 )2 ), which is a dimensionless
quantity that represents the system loading, since it is proportional to 2 r4 + (3 0 2 )r2 + 1 > 0: (46)
the electrical torque developed in steady-state Te . Also note that r3
equals the value of r in the tuned condition. Then, from (37), it follows The points at which the derivative changes sign are the roots of the
that r must satisfy the following polynomial equation: polynomial equation

r3 0 r3 2 r2 + r 0 r3 = 0: (38) 2 r4 + (3 0 2 )r2 + 1 = 0: (47)

jr3 j  [ ( 0 1)( + 3) + 0 3)][3(2 0 1) + ( 0 1)( + 3)( + 1)( 0 3)]

( + 1)(
22 (2 0 1 + ( 0 1)( + 3)( + 1)( 0 3))
(41)

 0 3)][3(1 0 2 ) + ( 0 1)( + 3)( + 1)( 0 3)]

jr3 j  [ ( 0 1)( + 3) + 2 ( + 1)(
2 (1 0 2 + ( 0 1)( + 3)( + 1)( 0 3))
(42)
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000 1229

ditions lie in between the two curves and a unique equilibrium outside.
At each one of the two curves two equilibria coincide.

C. Local Stability Analysis

The Jacobian matrix of the system (8)–(11) is given by (49), as
shown at the bottom of the page, whose determinant is
2 r4 + (3 0 2 )r2 + 1
det(J (; r)) = c1 c2 c4 c5 u02 ki  :
1 + 2 r 2
(50)

We know from the stability analysis in the previous section that for
the tuned condition the Jacobian has all its eigenvalues in the closed left
half-plane. Accordingly, we can verify from (50) that det(J (1; r)) >
08r . As the parameters  and r vary, loss of stability can be detected
3
by the eigenvalues of the Jacobian, as either a pair of complex eigen-
values or a single real eigenvalue cross the imaginary axis toward the
right half-plane. If a pair of complex eigenvalues crosses the imaginary
axis, then a Hopf bifurcation takes place. Given the complexity of the
expressions involved in an eigenvalue analysis, we cannot at this point
provide generic conditions for the ocurrenc of Hopf bifurcations. We
Fig. 1. Locus of the points in the parameter space where the number of
point out however, that it can happen for certain choices of PI gains, as
equilibria changes.
shown by an example given in [14].
On the other hand, the crossing of a real eigenvalue through zero can
The four solutions of this equation can be written as a function of  be detected just by looking at the sign of det(J (; r)) in (50). This de-
terminant is the product of positive quantities and a fourth-order poly-
r=
1
2
[ 6 ( 0 1)( + 3) 6 ( + 1)( 0 3)] (48) nomial in r . Thus it is zero if and only if the polynomial is zero

which yields real solutions only for  > 3, thus establishing the ne- 2 r4 + (3 0 )r2 + 1 = 0: (51)
cessity of (40). In this case, two positive and two negative solutions
exist. Since f (1) is odd, the analysis can be restricted to r 2 [0; 1)
Since this equation is the same which describes the points where the
number of equilibria changes (47), the following fact has been estab-
and extended to negative r by symmetry arguments. Let the positive lished.
solutions be r1 and r2 , r2 > r1 ; then the right-hand side of (41) and Fact 1: System (8)–(11) presents turning points (saddle-node bifur-
(42) are obtained by evaluating f (r2 ) and f (r1 ), respectively. By the cations) at the solutions of (51). }
sign of the derivative, it follows that r1 is a local maximum and r2 Then the curves in Fig. 1 represent bifurcation surfaces in the pa-
is a local minimum, and hence f (r) will not be bijective for r 3 =
f (r) 2 [f (r2 ); f (r1 )]. This is precisely the condition stated in (41)
rameter space, that is, they give the locus of all the turning points in
1
and (42). Indeed, the set [r 1 ; r 2 ] = fr: r3 = f (r) 2 [f (r2 ); f (r1 )]g
this space [1], [2]. It is instructive to look at the branching diagram ob-
tained varying r3 for a given . Fig. 2 presents the branching diagram
is well defined and such that 0 < r 1 < r1 < r2 < r2 , since for  = 4. Two turning points can be seen, associated to the crossing
f (1) is continuous, unbounded, f (r) > 0; 8r > 0, and f (0) = 0.
1
Observing that f (1) is strictly increasing in A = (r 1 ; r1 ) [ (r2 ; r 2 )
of each one of the two curves in Fig. 1. This branch topology is known
to cause the occurrence of jumps and hysteresis [2].
1
and strictly decreasing in B = (r1 ; r2 ), it follows that (38) has three An additional local stability property follows from the above devel-
distinct solutions whenever r3 2 (f (r2 ); f (r1 )), thus establishing the opment.
first part of the theorem. Now, suppose (42) is satisfied with equality. Fact 2: Assume no Hopf bifurcation takes place for a given positive
Then, r3 = f (r1 ) and, since f (1) is strictly increasing in (r2 ; 1), range of values of r3 and . Then system (8)–(11) has at least one
f (r2 ) = f (r1 ). No other solution exists since r1 is a local maximum locally exponentially stable equilibrium point for all positive values
and f (1) also is strictly increasing in [0; r1 ). With a similar reasoning of  and r3 within this range. Moreover, equilibria such that (46) is
applied to the case in which (41) is satisfied with equality, the second satisfied will be locally exponentially stable, whereas other equilibria
part of the theorem follows. The theorem is proved by noting that f is will be unstable. }
bijective outside the set [r 1 ; r 2 ]. } By Theorem 1 it is clear that there exists a nonempty range of param-
This result is ilustrated by Fig. 1, where the limits described by (41) eters around the tuned condition which satisfies the assumption of Fact
and (42)pare plotted. The point where the two curves intersect is  = 3, 2. The occurrence of Hopf bifurcations depends on the setting of the
r3 = ( 3=3). For large  the upper bound (42) tends to r3 = 0:5 and PI speed controller. Conditions on the parameters kp and ki to avoid
the lower bound (41) tends to zero, as can be seen by analysis of (41) the Hopf bifurcation in the zero load torque case have been given in
and (42). The system has three equilibria if the tuning and loading con- [14]. For small load torque the behavior is expected to be similar, so

0c1 0c1 r 0 c2 (1 0 )=(1 + 2 r2 )

c1 r 0c1 0 c2 (1 0 )=(1 + 2 r2 )r
J (; r) = c4 c5 u20 0c4 c5 u20r 0c3 0c4 c5 cc21 u20(1 + r2 )=(1 + 2 r2 ) (49)

ki 0 kp c3 0kp c4 c5 2 u20 (1 + r2 )=(1 + 2 r2 )

c
kp c4 c5 u20 0k c4 c5 u20r
p
c1
1230 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000

Fig. 3. Simulation for varying load torque in the tuned condition.

Fig. 2. Branching diagram for  = 4.

these conditions should be useful also for this case. However, it is still
unclear how to predict and avoid the occurrence of Hopf bifurcations
for higher loading conditions.

V. DISCUSSION
The conditions for saddle-node bifurcations in Section V are given
in terms of the normalized load/flux rate r 3 and the normalized pa-
rameter mismatch . The range of parameters for which these bifurca-
tions occur represents a low load/flux rate, which implies that operation
under high flux yields smaller stability margins regarding saddle-node
bifurcations. In any case, no saddle-node bifurcation occurs for  <
3. Hence, in terms of robustness alone, it might be a good idea to
“aim low” at the parameter c1 when commissioning an IFOC, thus pre-
venting a high value of . From this point of view it is also advisable
to use low flux levels in order to avoid the saddle-node bifurcations.
When three equilibria are present, and under the assumptions of Fact
2, two equilibria are locally asymptotically stable and the third one is Fig. 4. Simulation for varying load torque with  = 4.
unstable—Fact 3. The stable equilibrium which appears with low r in
the branch diagram (Fig. 2) is the operating point, and the other stable
equilibrium presents a much larger current level. In actual operation, derived. The robustness property given in [14] has been given a quan-
as the load varies passing through a saddle-node bifurcation the system titative measure, applicable for any given system.
will jump from the operating point to the other equilibrium. These results reveal important structural stability/robustness proper-
Some simulations are provided to illustrate the system behavior for ties of the IFOC. From a practical point of view, in most cases tem-
different mismatches in the rotor time constant. The model parameters perature variations inside the rotor can cause the rotor time constant
for the case simulated are c1 = 13:67 s01 , c2 = 1:56 H 1 s01 , c3 = to vary more than 50% but not more than 100% [36]. Hence,  < 2
0:59 s01 , c4 = 1; 176 kg011 m02 , c5 = 2:86, taken from a 1-cv in most practical cases, so that the robust global stability in the tuned
squirrel-cage induction motor. The simulations presented in Figs. 3 and condition along with the uniqueness of the equilibrium for  < 3
4 show the system’s behavior with  = 1 and  = 4 as the load have strong practical implications. Yet, aiming low at c1 and at the
torque is slowly increased from Tm = 0 up to Tm = 4 N 1 m. In the flux level may still be advisable in practical IFOC comissioning in
tuned condition the system behaves properly, but for  = 4 the loss of order to keep the saddle-node bifurcation as far as possible. Closeness
the equilibrium at which the system was operating causes a jump to a to a saddle-node bifurcation tends to deteriorate the system’s perfor-
much higher current level. Although speed regulation is not lost in the mance and efficiency. Furthermore, much larger degrees of tuning 
simulation, in actual operation the motor would most likely stall due are likely to temporarily occur when adaptive techniques (like [31]) are
to overcurrent protection actuation. It is also worth noticing that even employed, reaching the region where multiple equilibria are present.
before the jump occurs the system may experience stability problems, Thus, the results also have practical implications regarding the stability
since the existence of a nearby unstable equilibrium reduces the size of analysis and the design of adaptive IFOC drives.
the region of attraction of the operating point. It is important to stress that neither the existence of the robust global
Global stability of the IFOC has also been studied in this paper. A stability property in the tuned condition nor the bifurcation topology
new Lyapunov function, which generalizes the one given in [14], has given in Section IV depend on the design of the PI controller. On the
been given in Section III. By means of this Lyapunov function, a test other hand, both the size of the robustness margin as given in Theorem 1
for the robust global stability in different operating conditions has been and the occurrence of Hopf bifurcations depend on the PI parameters kp
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000 1231

and ki . Conditions to be satisfied by these parameters in order to guar- [24] R. Ortega, C. Canudas, and S. I. Seleme, “Nonlinear control of induction
antee stability have been given in [14] for the zero load case only. The motors: Torque tracking with unknown load disturbance,” in American
study of a number of different examples suggests that Hopf bifurcations Control Conf., 1992, pp. 206–210.
[25] L. Zai, C. L. DeMarco, and T. A. Lipo, “An extended Kalman filter ap-
are avoided by a correct classical PI tuning, but this is still an open issue. proach to rotor time constant measurement in PWM induction motor
drives,” IEEE Trans. Indus. Applicat., vol. 28, pp. 96–104, Jan./Feb.
1992.
[26] T. Ohtani, N. Takada, and K. Tanaka, “Vector control of induction
REFERENCES motor without shaft encoder,” IEEE Trans. Indus. Applicat., vol. 28,
pp. 157–164, Jan./Feb. 1992.
[1] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical [27] R. J. Kerkman, T. M. Rowman, and D. Leggate, “Indirect field-oriented
Systems and Bifurcations of Vector Fields, 2nd ed. New York: control of an induction motor in the field-weakening region,” IEEE
Springer-Verlag, 1986. Trans. Indus. Applicat., vol. 28, pp. 850–857, July/Aug. 1992.
[2] R. Seydel, Practical Bifurcation and Stability Analysis: From Equilib- [28] E. Bassi et al., “A field orienttion scheme for current-fed induction
rium to Chaos, 2nd ed. New York: Springer-Verlag, 1994. motor drives based on the torque angle closed-loop control,” IEEE
[3] R. Chen and P. P. Varaiya, “Degenerate Hopf bifurcations in power sys- Trans. Indus. Applicat., vol. 28, pp. 1038–1044, Sept./Oct. 1992.
tems,” IEEE Trans. Circuits Syst., vol. 35, pp. 818–824, July 1988. [29] Y. Kuroe, “Design of a new controller for induction motors based on
[4] H. O. Wang, E. H. Abed, and A. M. A. Hamdan, Voltage collapse dy- exact linearization,” in Proc. IECON’91—Industrial Electronics Conf.,
namics and control in a sample power system, . Oct./Nov. 1991, pp. 621–626.
[5] V. Ajjarapu and B. Lee, Bifurcations theory and its application to non- [30] R. Krishnan and A. S. Bharadwaj, “A review of parameter sensitivity and
linear dynamical phenomena in an electrical power system, . adaptation in indirect vector controlled inductior motor drive systems,”
[6] E. H. Abed, Control of bifurcations associated with voltage instability, . IEEE Trans. Indus. Applicat., vol. 6, pp. 695–703, Oct. 1991.
[7] A. I. Mees and L. O. Chua, “The Hopf bifurcation theorem and its ap- [31] T. M. Rowan, R. J. Kerkman, and D. Leggate, “A simple on-line adap-
plications to nonlinear oscillations in circuits and systems,” IEEE Trans. tion for indirect field orientation of an induction machine,” IEEE Trans.
Circuits Syst., vol. CAS-26, pp. 235–254, Apr. 1979. Indus. Applicat., vol. 27, pp. 720–727, July/Aug. 1991.
[32] G. O. Garcia, R. M. Stephan, and E. H. Watanabe, “Controle de Ve-
[8] R. Aloliwi, E. G. Strangas, and H. K. Khalil, “Robust speed control of
locidade de um Motor de Induccão através das Técnicas Campo Orien-
induction motors,” in American Control Conf., Albuquerque, NM, June
tado Indireto e Escorregamento Controlado,” in Anais do 8 Congresso
1997, pp. 955–960.
Brasileiro de Automática, UFPA-Belém, PA, 1990, pp. 955–960.
[9] R. Noguchi, S. Kondo, and I. Takahashi, “Field-oriented control of an in- [33] A. de Luca and G. Ulivi, “Design of an exact nonlinear controller for in-
duction motor with robust on-line tuning of its parameters,” IEEE Trans. duction motors,” IEEE Trans. Automat. Contr., vol. 34, pp. 1304–1307,
Indust. Applicat., vol. 33, pp. 35–42, Jan./Feb. 1997. Dec. 1989.
[10] C. C. De Wit and S. I. Seleme, “Robust torque control design for induc- [34] S. Ogasawara, H. Akagi, and A. Nabae, “The generalized theory of in-
tion motors: The minimum energy approach,” Automatica, vol. 33, no. direct vector control for AC machines,” IEEE Trans. Indus. Applicat.,
1, pp. 63–79, 1997. vol. 24, pp. 470–478, May/June 1988.
[11] C. Caminhas, R. Takahashi, P. Peres, and H. Tavares, “Observadores [35] E. Y. Y. Ho and P. C. Sen, “Decoupling control of induction motor
de Estados do Motor de Indução com Rejeição de Perturbações drives,” IEEE Trans. Indus. Electron., vol. IE-35, pp. 253–262, May
Paramétricas,” in XII CBA-Congresso Brasileiro de Automática, Rio de 1988.
Janeiro, 1996, pp. 1125–1130. [36] R. Krishnan and F. C. Doran, “Study of parameter sensitivity in high-
[12] P. Vedagarbha, D. M. Dawson, and R. Burg, “Velocity tracking/setpoint performance inverter-fed induction motor drive systems,” IEEE Trans.
control of induction motors with improved efficiency,” in IFAC 13th Tri- Indus. Applicat., vol. IA-23, pp. 623–635, July/Aug. 1987.
ennial World Congr., San Francisco, CA, USA, 1996, pp. 37–42. [37] D. W. Novotny and P. C. Sen, “Introduction to field orientation and high
[13] R. J. Kerkman et al., “A new flux and stator resistence identifier for performance AC drives,” in IEEE Indus. Applicat. Soc. Annual Meeting,
AC drive systems,” IEEE Trans. Indus. Applicat., vol. 32, pp. 585–593, Sept. 1986.
May/June 1996. [38] L. J. Garcés, “Parameter adaption for the speed-controlled static AC
drive with a squirrel-cage induction motor,” IEEE Trans. Indus. Ap-
[14] P. A. S. De Wit, R. Ortega, and I. Mareels, “Indirect field-oriented con-
plicat., vol. IA-16, pp. 173–178, Mar./Apr. 1980.
trol of induction motors is robustly globally stable,” Automatica, vol. 32,
[39] F. Blaschke, “The principle of field orientation as applied to the new
no. 10, pp. 1393–1402, 1996. transvektor closed-loop control system for rotating-field machines,”
[15] R. Ortega, P. J. Nicklasson, and G. E. Péres, “On speed control of induc- Siemens Rev., no. 5, pp. 217–220, 1972.
tion motors,” Automatica, vol. 32, no. 3, pp. 455–460, 1996. [40] B. K. Bose, Power Electronics and AC Drives. Englewood Cliffs, NJ:
[16] S. H. Kim and S. K. Sul, “Maximum torque control of an induction Prentice-Hall, 1987.
machine in the field weakening region,” IEEE Trans. Indus. Applicat., [41] W. Leonhard, Control of Electrical Drives. Berlin: Springer-Verlag,
vol. 31, pp. 787–794, July/Aug. 1995. 1985.
[17] D. Borgard, G. Olsson, and R. D. Lorenz, “Accuracy issues for param-
eter estimation of field oriented induction machine drives,” IEEE Trans
Indus. Applicat., vol. 31, pp. 795–801, July/Aug. 1995.
[18] K. Ide et al., “Vector approximation method with parameter adaptation
and torque control of CSI-Fed induction motor,” IEEE Trans Indus. Ap-
plicat., vol. 31, pp. 830–840, July/Aug. 1995.
[19] R. W. De Doncker and D. W. Novotny, “The universal field oriented
controller,” IEEE Trans. Indus. Applicat., vol. 30, pp. 92–100, Jan./Feb.
1994.
[20] L. Huang, Y. Tadokoro, and K. Matsuse, “Deadbeat flux level control
of direct-field-oriented high-horsepower induction servo motor using
adaptive rotor flux observer,” IEEE Trans. Indus. Applicat., vol. 30, pp.
954–962, July/Aug. 1994.
[21] H. Kubota and K. Matsuse, “Speed sensorless field-oriented control of
induction motor with rotor resistence adaptation,” IEEE Trans. Indus.
Applicat., vol. 30, pp. 1219–1224, Sept./Oct. 1994.
[22] R. Marino, S. Peresada, and P. Valigi, “Adaptive input-output linearizing
control of induction motor,” IEEE Trans. Automat. Contr., vol. 38, pp.
208–221, Feb. 1993.
[23] R. Ortega, C. Canudas, and S. I. Seleme, “Nonlinear control of induction
motors: Torque tracking with unknown load disturbance,” IEEE Trans.
Automat. Contr., vol. 38, pp. 1675–1680, Nov. 1993.