You are on page 1of 14

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO.

2, FEBRUARY 1993

Adaptive Input -Output Linearizing


Control of Induction Motors
Riccardo Marino, Sergei Peresada, and Paolo Valigi

Abstract-A nonlinear adaptive state feedback input-output by means of nonlinear state feedback
linearizing control is designed for a fifth-order model of an
induction motor which includes both electrical and mechanical
dynamics under the assumptions of linear magnetic circuits.
The control algorithm contains a nonlinear identification scheme
which asymptotically tracks the true values of the load torque (with B ( x , p ) a nonsingular m X m matrix V p E R9) and
and rotor resistance which are assumed to be constant but nonlinear state space change of coordinates
unknown. Once those parameters are identified, the two control
goals of regulating rotor speed and rotor flux amplitude
are decoupled, so that power efficiency can be improved with-
out affecting speed regulation. Full state measurements are
required. Linear control techniques can then be applied in the
design of the control U in (2). Necessary and sufficient
conditions were determined in [17] and [13] for a system
I. INTRODUCTION (1) to be locally feedback linearizable, i.e., transformable
into (2) via (3) and (4)in a neighborhood of x,. They are
I N the last decade, significant advances have been made rather restrictive from a mathematical point of view. Nev-
in the theory of nonlinear state feedback control (see ertheless, they apply to detailed models of helicopters
1151 and [39] for a comprehensive introduction to nonlin- [371, synchronous generators [3 11, switched reluctance
ear geometric control): in particular feedback lineariza- motors [14] and permanent magnet stepper motors [SI.
tion and input-output decoupling techniques have proved Electromechanical systems are good candidates for non-
useful in applications and applied even before the theory linear state feedback design since nonlinearities are often
was fully developed. significant and exactly known being modeled on the basis
The technique of state feedback linearization [17], [13] of physical principles.
was developed in the effort of designing an autopilot for Whenever outputs to be controlled are defined as
helicopters [37]. It requires measurements of the state
vector x and knowledge of the parameter vector p in y = h(x), y E R" (5)
order to transform a multiinput nonlinear control system
(x E R " , U E R " , p E R q ) the problem of making the input-output map decoupled
and linear by state feedback (3) has been addressed and
solved in [111 and [16], following earlier applications in
robotics. The decoupling state feedback may render some
states unobservable from the outputs.
Applications clearly indicate that even though nonlin-
earities may be exactly modeled, the physical parameters
involved are most often not precisely known. This moti-
vated further studies on adaptive versions of feedback
into a linear and controllable one ( z E R", v E R " ) linearization and input-output linearization, since cancel-
lations of nonlinearities containing parameters are
i = A z + Bv (2) required. For systems (1) which are linear with respect to
the unknown parameters p , sufficient conditions for
adaptive feedback linearization were developed in [38]
Manuscript received June 1, 1990; revised May 17, 1991 and April 6,
1992. under sector type restrictions on certain nonlinearities
Paper recommended by G. C. Verghese. This work was supported and in [441, [181, and [201 under structural conditions
in part by Ministero della Universit; e della Ricerca Scientifica e which do not restrict the type of nonlinearities. Different
Tecnologica.
R. Marino and P. Valigi are with the Dipartimento di Ingegneria approaches for nonlinear adaptive stabilization can be
Elettronica, Seconda Universit; di Roma, Via 0. Raimondo, 00173 found in the survey paper [41]. Adaptive input-output
Rome, Italy. linearization was studied in [43] under global Lipschitz
S. Peresada is with the Department of Electrical Engineering, Kiev
Polytechnical Institute, Prospect Pobedy, 37 Kiev 252056 USSR. conditions on the nonlinearities multiplying unknown
IEEE Log Number 9205180. parameters. More recently, the more difficult problem of

0018-9286/93$03.00 0 1993 IEEE


MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 209

output feedback adaptive control has been addressed in fication scheme which asymptotically tracks the true value
[35] and [19] for single output nonlinear systems. of load torque and, when electric torque is different than
Even before the theory of nonlinear feedback control zero, the true value of rotor resistance as well. The
was fully developed, nonlinear state space change of coor- adaptive state-feedback linearizing control achieves full
dinates (4) and nonlinear state feedback (3) were pro- decoupling in speed and rotor flux magnitude regulation
posed in [31 and [41 for induction motor control in order to as soon as the identification scheme has converged to the
achieve an asymptotic decoupling in the control of speed true parameter values: this allows us to improve power
and flux amplitude (the so called field oriented control). efficiency by adjusting flux levels, without affecting speed
Nested loops of standard PI regulators are used in [27] to regulation. In Section V the effects of parameter uncer-
achieve robustness versus parameter variations. In order tainties on the performance of nonadaptive decoupling
to counteract uncertainties linear optimal control tech- control are analyzed in detail and shown by simulations.
niques and linear model reference adaptive techniques Additional simulations illustrate the performance of the
have also been proposed in [ll and in [281, [61, respec- proposed adaptive control algorithm in speed and flux
tively. Decoupling is obtained only in steady state, i.e., amplitude regulation, showing that voltage supply wave-
when the flux amplitude is kept constant. Coupling is still forms are implementable by actual inverters and currents
present when flux is weakened in order to operate the are within acceptable limits. The main drawback is the
motor at higher speed within the input voltage saturation requirement of flux measurements: however, preliminary
limits [27, p. 2171 or when flux is adjusted in order to simulations reported in [36] indicate that the proposed
maximize power efficiency [26], [22]. A different approach adaptive control maintains good performances even when
which makes use of variable structure techniques and of rotor flux values are provided by the observer given in [46]
new state coordinates was proposed in [42]. Nonadaptive and driven by rotor resistance estimates provided by the
input-output decoupling controls were presented in [29], adaptation law. Preliminary versions of this work were
[30], and [25] using geometric techniques (see also [21]). In reported in [33] and [34]. Additional simulation studies
[25], a fifth-order model which includes the mechanical can be found in [36].
part is used: exact decoupling in the control of speed and
flux amplitude is achieved by a static state feedback 11. INDUCTION
MOTORMODEL
controller. In [291 and [301, a simplified model is used: The reader is referred to [lo] and 1231 for the general
only the electromagnetic part is modeled assuming the theory of electric machines and induction motors, to [27]
speed of a slowly varying parameter. Exact decoupling in for related control problems, and to [91 for digital imple-
the control of electric torque and flux amplitude using the mentations. The symbols used and their meaning are
amplitude and the frequency of the voltage supply as listed in the Appendix.
inputs is obtained in [29] by a dynamic (second order) An induction motor is made by three stator windings
compensator and in [30] by a static state feedback com- and three rotor windings. Krause and Thomas [241 intro-
pensator. An indirect adaptive version of the decoupling duced a two phase equivalent machine representation (see
algorithm proposed in [30] can be found in [45], where two the Appendix for the exact transformation of three phase
electrical parameters are assumed to be unknown. variables into two phase ones used in this paper) with two
The main result of this paper is to develop an adaptive rotor windings and two stator windings. Their dynamics
version of the controller presented in [25], assuming that are described by
load torque and rotor resistance are unknown but con-
stant parameters. In Section I1 a fifth-order state space d*SO
model of an induction motor, which includes both electri- R,i,, +- dt
= U,,

cal and mechanical dynamics, is given. In Section 111


previous control schemes are reviewed and it is shown d*Sb
that field oriented control can be viewed as a feedback R,i,, +- dt
= uSb

transformation which achieves asymptotic input-output


decoupling and linearization. It is also established that the
model is not state feedback linearizable and that the
dynamics made unobservable by a state feedback input-
output linearizing control (zero dynamics) are due to the
rotation of the flux vector. In Section IV an adaptive (7)
version of the exact decoupling and linearizing control
given in [251 is developed, which covers the more realistic where R , i, a), U, denote resistance, current, flux linkage,
situation in which the load torque and the rotor resistance and stator voltage input to the machine; the subscripts s
are not known. Rotor resistance may have a range of and r stand for stator and rotor, ( a , b ) denote the compo-
variation of +50% around its nominal value due to rotor nents of a.vector with respect to a fixed stator reference
heating. Even though available nonlinear adaptive results frame, ( d ' , q ' ) denote the components of a vector with
do not apply as such to the model given in Section 11, the respect to a frame rotating at speed n p w ; and np denotes
key idea in 1181 leads to a second-order nonlinear identi- the number of pole pairs of the induction machine and o
210 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO. 2, FEBRUARY 1993

the rotor speed. Let 6 denote an angle such that terms of rotor fluxes and stator currents as
d6
_
dt -
-nPw, 6(0) = 0.
L r

We now transform the vectors ( i r d , i, , ,), ( @ r d t , @ r q . ) in the so that the rotor dynamics are
rotating frame ( d ' , q'> into vectors ?irU,irb),($r,,+r,,) in
the stationary frame ( a , b ) by

where J is the moment of inertia of the rotor and of any


tool attached to it and TL is the load torque.
By adding the rotor dynamics (15) to the electromag-
netic dynamics (13) and rearranging the equations in state
space form, the overall dynamics of an induction motor
Applying transformations (91, (10) and using equation (8), under the assumptions of equal mutual inductances and
(6) and (7) become linear magnetic circuit are given by the fifth-order model:
d*s, dw n,,M
R,i,, +-dt
= U,, -= -(*raisb - $rbiso) --
TL
dt JL, J
d*Sb
R,i,, +- dt
= U,h

d*ru
R,i,, +-dt
+ n pw&,, = 0

Under the assumptions of linearity of the magnetic cir-


cuits and of equal mutual inductances and neglecting iron
losses, the magnetic equations are (see [23, p. 1721)
Ccsa = ' s i , , + Mi,,
@sh = L s i s b +
$ru = M'su + ' r i r u
$rb = + Lrirb ( 12)
where L,, L , are autoinductances and M is the mutual
inductance; as we shall see, the assumption of linearity where i, +,uJ denote current, flux linkage and stator
will be enforced by a control action which will keep the voltage input to the machine; the subscripts s and r stand
absolute value of the rotor flux below the nominal value. for stator and rotor; ( a , b ) denote the components of a
The purpose of introducing the transformations (9) and vector with respect to a fixed stator reference frame and
(10) is precisely to obtain (12) which is independent of 6: = 1- (M~/L,L,).

in fact fluxes and currents in (6) and (7) are related by From now on we will drop the subscripts r and s since
6-dependent auto and mutual inductances. we will only use rotor fluxes (+,,, q5rb) and stator currents
Eliminating i,,, i,, and qbSu71+9,,, in (11) by using (12), (i,,, is,) as state variables. Let
we obtain
= ( $0, @b7 ib)T (17)
be the state vector and let
>U

R,i,, + - -+
L,
dlC',b

dt
(L,- g)% = U,,,
P = ( ~ 1 3 ~ =
2 )( T ~L - T L N ~ R ~

be the unknown parameter deviations from the nomi-


nal values TLN and R,, of load torque TL and rotor
-', IT (18)

resistance R,. TL is typically unknown whereas R , may


have a range of variations of *50% around its nomi-
nal value (see [27, p. 2241) due to rotor heating. Let U =
(uU,uh)' be the control vector. Let (Y = (Rr,,,/Lr), p =
(M/wL,L,), y = (M2R,,/~L,L:) + (R,/(TL,),
p = ( n ,M/JLr), be a reparameterization of the induction
The torque produced by the machine is expressed in motor model, where a , p, y , p are known parameters
MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 21 1

depending on the nominal value R I N .System (16) can be Since cos p = ($,/I I
) I), sin p = (&/I $I>, with I$/ =
rewritten in compact form as d m , from (24) and (25) we have
= f ( x )+ uaga + ubgb 'Plfl +P2f2(') (19) $ai, + $bib
ld =
where the vector fields f, g,, gb, f l , f 2 are I*I

We now reinterpret field oriented control as a state


feedback transformation (involving state space change of
coordinates and nonlinear state feedback) into a control
system of simpler structure. Defining the state space
(21) change of coordinates
o = w

$d= &&-??
1 '
__ *b
J p = arctan -
*a
0
fdx) = 0 '
0
0
$sib - $bia
1, =
I*I
and the state feedback
-1

f2(x) =

the system (16) becomes


dw -
_ TL
dt
- p$diq - 7
111. INDUCTIONMOTORCONTROL -*d- - -
dt
+ aMid
A. Field Oriented Control
A classical control technique for induction motors is by did i2
-= -yid + f f p & + n p w i , + a M 4 + -ud
now the field oriented control. First introduced by dt *d uLs
Blaschke 131, [4] in 1971, it involves the transformation of
i,i,
the vectors ( i a ,ib),( + a , in the fixed stator frame ( a , b ) di, -
_ -yi, - pnpwt+!td- n p w i d - aM- + -U,1
into vectors in a frame ( d , q ) which rotate along with the dt *d uLr
flux vector (+,, defining
*b
dP = n p w
- +~ M L . (29)
p = arctan - (23) dt *d
*a
Defining the nonlinear state feedback control
the transformations are
i2
(j: j [ =
cos p
-sin p
sin p
cos p ] (ja)' (24) -npwi, - ( Y M -~apqd
*d
+ vd
(30)
idi,
(25) np + npwid + aM- + U,
212 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO. 2, FEBRUARY 1993

so that (28) becomes If w and t,hd are defined as outputs, field oriented control
achieves asymptotic input-output linearization and
decoupling via the nonlinear state feedback (281, (341,
(36):PI controllers are then used to counteract parameter
variations.
During flux transient the nonlinear term i,bdiq in (32)
makes the first four equations in (32) still nonlinear and
coupled: as a result speed transients are difficult to evalu-
ate and may be unsatisfactory, Flux transients occur when
the motor has to be operated above the nominal speed: in
this case flux weakening (for instance = ( k / w r e f ) )is
the following closed-loop system
required in order to keep applied voltage within inverter
do- TLN ceiling limits [27, p. 2171. Even when the motor is oper-
_
dt - p@di9 - ated below the nominal speed, flux may be varied in order
to maximize power efficiency (see [26],[22]).
di,
_ -
dt
- - yi, + U,
B. Input-Output Decoupling
As shown in [25](see also [21]),field oriented control
-*d- - - f f $ b d
dt
+ aMid can be improved by achieving exact input-output decou-
did pling and linearization via a nonlinear state feedback
- = -yid
dt
+ U, control which is not more complex than (31). We now
summarize this technique which will be made adaptive in
dP i next section. The following notation is used for the direc-
-=npw +~ M X (32) tional (or Lie) derivative of state function +(XI: R" + R
dt *d
along a vector field f ( x ) = ( f , ( x ) ; - . ,f n ( x ) )
is obtained. System (16) is transformed into (32) by the
J+
feedback transformation (271, (31).
System (32) has a simpler structure: flux amplitude
~ f =
$ C Gfi(x).
i= 1
(37)
dynamics are linear
Iteratively, we define L$$ = L f ( L ' f -')$I.
-
d*d - - -a+d + aMid The outputs to be controlled are w and :b,t + i,!~,'.

dt Define the change of coordinates (see also [42])


did - Y, = M X ) = w
_
dt
- -yid + U, (33)
TLN
and can be independently controlled by for instance via
Y2 = L f + l ( x ) = p.(cGhib - *bia) - -
J
a PI controller, as proposed in [27] Y3 = $Ax) =*: *; +
= -kdl(*d - *drref) - k d Z / f ( * d ( T )
0

When the flux amplitude $d is regulated to the constant


- q d r e f ) dT.

(34) y, = arctan (t) = $3

reference value I , ! J ~ ~rotor


~ ~ , speed dynamics are also lin- which is one to one in Q = (x E R5:$2 + @', # O} but it
ear is onto only for y 3 > 0, -90 s y , I 90. The inverse
transformation is
_
dw -
- p*dref iq - -
TL
w =Yl
dt J
di, +a = & COS Y ,
_ - - -yi, + U, (35)
dt *h = 6sin Y ,
and can be independently controlled by U,, for instance by
two nested loops of PI controllers, as proposed in [271 i,
1
= -(COSY5(
6
Y +2aY,
42aM ) -
1
-pY5(Y2+
TLN
7))
MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 213

The dynamics of the induction motor with nominal para- The input-output linearizing feedback for system (40)
meters are given in new coordinates by is given by
Y l =Y2

9 2 = L2f41 + LgaLf4 l U a + LgbLf4 l U b


Y3 = Y 4
where U = (ua,ub)T is the new input vector. Substituting
Y 4 = L2f42 + LgaLf4 2 u a + LgbLf 42ub (47) in (40) the closed-loop dynamics become in y-coordi-
nates
Y 5 = Lf43. (40)
Y1 = Y 2
The first four equations in (40) can be rewritten as
3 2 = ua

Y3 = Y 4

Y 4 = ub

- Pnpo(rCdia + *bib) Equations (45) or (46) represent the dynamics which have
L2fc#J2= ( 4 2 + 2 a 2 p M ) ( *2 + *;) been made unobservable from the outputs o and $2 + +:
by the state feedback control (47). In order to track
+ 2 a M n p o ( Gaib - & , i n ) desired smooth reference signals or,,(?) and I + f e f ( t ) for
the speed y 1 = w and the square of the flux modulus
- ( 6 a 2 M + 2a')'M)( +aia + + b i b ) y , = 42 + &, the input signals U, and q, in (47) are
+ 2a2M2(i,2 + i,") designed as
and D ( x ) is the decoupling matrix defined as

D(x) =
[ Lgkf
Lgkf
'1

42
LgbLf

Lg&f 42
"'1 va = - k a l ( Y l

= -kal(
- o r e f ( t ) ) - ' a 2 ( ~ 2 - 4 e f ( t ) ) + &ref

- orer - ka2 ( P( *sib - I(lbia)


('1

[
P P
-- *b Z * a
U Ls

2aM = -k,i(y3 - I+IL) - k b 2 ( y 4 - I4IL) + Ili;I:ef


= ?@a -$b
c+LS = -kbl(lGb2 + - l+l:ef) - kb2(2a(M($aia + *bib)
Since -(*; + +)I; - I4I:ef) + Ili;Ifef (49)
where ( k a , ,k a 2 )and ( k b l ,k b 2 )are constant design param-
eters to be determined in order to make the decoupled,
linear second-order systems
D ( x ) is nonsingular everywhere in Q.
The dynamics of the flux angle y , = 4,(x> are d2 d
d43
- - dY5
--
aM z (- 0 oref) = -ka1( - wref) - k a 2 ~ w( - o r e r )

dr dr
= npo + =( *sib - * b i a )
d2
-dr2
(I+I~ - I+I:ef) = -kb1(l+l2 - IqItef)
(45) d
- k b 2 z ( l $ 1 2 - I+I:ef) (50)
The difference between flux angular speed C$3 and rotor
speed n p w is usually called slip speed, w s , which can be asymptotically stable and to shape their responses.
expressed, recalling the expression of a, as Remarks:
1) The closed-loop system (48) is input-output
+sib - $bia
*;
RrNM
4, - n p w = os= - decoupled and linear: the input-output map consists of a
L, *2 + pair of second-order systems. This allows for an indepen-
dent regulation (or tracking) of the outputs according to
(50). Transient responses are now decoupled also when
is varied, even independently of oref. This is an
214 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. VOL. 38, NO. 2, FEBRUARY 1993

improvement over the field oriented control (see also controlled by a third input U, so that
[251).
2) State space change of coordinates both in the field dP i
oriented control and in the decoupling control [i.e., (27)
- = n,]w +~ M U,.
L + (53)
dt *d
and (39)] are valid in the open set R = { x E R': $2 +
$: # 0); notice that +: +;
+ = 0 is a physical singularity The model consists of five states (id, i,, qd,I)~, p ) and is
of the motor in starting conditions. driven by three controls. A nonlinear state feedback con-
3) As in field oriented control, while measurements trol is designed which decouples and linearizes the three
of ( w , i,, i,) are available, measurements of ($,, $ b ) control actions of regulating torque, rotor flux amplitude
require installing flux sensing coils or Hall effect trans- and of forcing the reference frame ( d , q ) to rotate along
ducers in the stator which is not realistic in general- with the rotor flux vector while making p unobservable
purpose squirrel cage machines. from the outputs. Assuming two electrical parameters
4) Easy computations show that the induction motor unknown, an indirect adaptive version of this control was
model (16) is not feedback linearizable. The necessary and developed and simulated in 1451.
sufficient conditions given in 1171 fail; in fact the distribu- A n input-output decoupling control law in the d - q
tion gi = span {ga,g,, adfg,, ad,gb} is not involutive reference frame has also been proposed in [211. The
since the vector field [adfg,, adfg,] does not belong to 27, control algorithm is based on flux estimates provided by
(ad,Y or [ X , Y ] denotes the Lie bracket of two vector an open-loop flux simulator. Following [12], rotor resis-
fields; one defines recursively adkY = ad,(ad;l ' Y ) ) .Fol- tance errors are computed in [21] on the basis of steady-
lowing the results in [32], since F0= span {g,, gh) is invo- state regulation errors and used in the output feedback
lutive and rank FI = 4, it turns out that the largest feed- control algorithm in order to reduce the steady-state
back linearizable subsystem has dimension 4. This shows regulation errors. The performances of the overall system
that are verified by simulations and experimental tests.
the control (47), (49) provides the largest linearizable
subsystem in the closed loop. INPUT-OUTPUT
IV. ADAPTIVE LINEARIZATION
5 ) The state feedback control (47), (49) is essentially When a decoupling control algorithm (47), (49) is used,
the one proposed in [25]. The only additional contribution variations in load torque TL and rotor resistance R , cause
in this section is to make clear that the decoupling control loss of input-output decoupling, steady-state tracking
makes the angle +3 unobservable from the outputs and errors and deteriorated transient responses. This calls for
that (16) is not feedback linearizable: this will be impor- an adaptive version of (47), (49) which will be developed
tant in the design of an adaptive control in next section. in this section under the assumptions that TL and R , are
Exact input-output decoupling controls for induction unknown constant parameters. In this section an adaptive
motors are also proposed in [29], [30] with reference to a tracking problem is addressed for a class of reference
simplified model: the mechanical dynamics in (16) are not signals satisfying the following assumptions.
considered and w is viewed as a parameter in the last Assumption I : The reference signals w J t ) and I$l;ef
four equations of (16). In [29] the inputs are constrained ( t ) are required to be C 2 bounded functions with bounded
to be of the form derivatives such that
U, = vcos 0
lim wrCf( t ) = c,,
t'X

ub = Vsin 0 (51)
and two integrators are added
with c , , c 2 E R .
dV
- = UI Let us rewrite system (19) in the y-coordinates defined
dt by (38); since the Lie derivatives L f Z 4 1L, f , L f 4 1 ,L f , & ,
d0 - Lf1L,+2,L f 1 4 3 > L f 1 L f 2 4 2Lg"439
? Lgh43 are all equal to
- U? zero, we have

so that (ul, U , ) are the new inputs. The outputs are chosen Y l = Y2 + PlLfl4I
to be the electric torque ( n , M / L , ) (+,i, - and the
rotor flux amplitude square $2 + 0;. A nonlinear state
feedback control for uI and u2 is obtained in [291 which
achieves input-output decoupling and linearization while
making a two dimensional state space submanifold unob-
servable from the outputs. In [30] flux and currents are
expressed in a rotating frame according to the transforma-
tions (24) and (251, where the dynamics of p ( t ) are
MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 215

where on the parameters estimate @ ( t )


21 = Y l

22 =Y2 +81Lf141
2 3 =Y3

2 4 = Y 4 +82Lf242

=y5. 25 (57)
In z-coordinates system (19) becomes
T 2, =22 + ep,Lf,41
i 2 = L2f4l + P 2 L f 2 L f 4 1+ Z Lf141
d8,

+ L g k f4 l U a + LghLf 4 l U b
i 3 = 24 + ep2Lf242

2, = L2f42 + P2Lf2Lf42 + 7
4 L
2 f242

+82LfLf242+82P2L2f242
+ U0(Lg0Lf42
+B2Lg.Lr242)
+ Ub(LgbLf42 +82LgbLf242)

is = Lf43 +P2Lf243’
6aL,ML, + 2 M 3
-
($ai, + $bib>? Let

(a]
UL,L)

(55)
(2) +

An adaptive version of input-output linearizing controls


was proposed in [43],which requires an overparameteriz-
ation and global Lipschitz property for the nonlinearities
multiplying the parameters. Adaptive versions of feedback
linearizing controls were developed in [381 under sector
LghLf41

LgbLf42+ 82Lg,Lf242 1
type restrictions on certain nonlinearities, in [44] under
structural matching conditions, in [18] under extended
matching conditions and, more generally, in [20],under
pure feedback conditions which do not require Lipschitz
or sector type restrictions. No one of the above tech-
niques apply in our case since the nonlinearities involved
are not globally Lipschitz and the system is not feedback
linearizable. However, we will use the adaptation tech-
nique proposed in [18] under the so called “extended
matching” structural condition and show directly the con-
vergence both of tracking errors and of parameter
estimation errors.
Let 80) = (b,(t>, fi2(t)>Tbe a time-varying estimate of
the parameters and let
Ka = [ 0
-kal
1
-ka2]. Kb = [ 0
-kbl
1
- k b 2 ] (62)
are asymptotically stable; w,&), I $ f e r ( t ) are reference
signals which satisfy Assumption 1. Since

be the parameter error. Following [18] we introduce a


time-varying state space change of coordinates depending
~

216 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO. 2,FEBRUARY 1993

the decoupling matrix is singular not only when nite symmetric matrices. Consider the quadratic function
($2 + $);
= 0 as in the nonadaptive case but also when
$,(t) = -R,,,,; this additional singularity has to be taken v = eTPe + eiTe, (71)
into account in the design of the adaptive algorithm.
Define the reference models where r is a positive definite symmetric matrix. The time
derivative of V is

If we now define

or, equivalently,

The model reference tracking error is defined as


which defines the dynamics of the parameter estimate
e (2] - Z I M 7 '2 - '2M 7 '3 - '3M 9 '4 -'4M) (65) @(t),and use (701, then (72) becomes
and its dynamics are given by dV
_ -- -eTQe.
(75)
dl = e2 + e,,Lf,+l dt
4= -kulel - kuze2 + e,2LfzLf+l This guarantees that e ( t ) and e,,, and therefore jXt>, are
bounded and that e ( t ) is an L2 signal. Under Assumption
4 = e4 + f3,,2Lf2+2 1 for ( w r e f ( t )l$I?,,f(t)),
, it follows from asymptotic stabil-
e4 = -kble3 - kb2e4 + 'p?( Lf2Lf+2 '$2 L;2+2)
ity of (64) and from (65) that the first four state variables
(z1;.-, z 4 ) are bounded. We are guaranteed to avoid the
4 = Lf+3 + P 2 L f 9 3 (66) singularities z3 = 0 and b2 = - R r N for the decoupl-
ing matrix, and therefore for the control (59) as well,
with ei(0) = 0, 1 Ii I 4.
when the initial conditions (e(0) = 0, e,(O)) are in S =
While the dynamics of zs are
{ ( e ,e,) E Rh: eTPe + e i r e , I v , v > 01, the largest set
entirely contained in ((e, e P )E R 6 : epz < R,, e3 > - c 2 } ,
with c 2 given in Assumption 1. Since W ( z ,k 2 ) is continu-
ous, contains only bounded functions of z5 (sine and
the dynamics of the vector e can be rearranged as cosine), and ( z l , z 2 , z 3 , z 4 ,F 2 ) are bounded, it follows that
W(z, is bounded and therefore e and $ are bounded
as well. Now, since e is a bounded L2 signal with bounded
derivative t , by Barbalat lemma ([40], p. 211) it follows
that
lim
t'X
(1 e( t ) 11 = 0, (76)

i.e., zero tracking error is achieved, both with respect to


the reference model and to the reference signals. Further-
more, since, according to (671, is is bounded for ( e , e,) E
S and the parameters are assumed to be constant, it
follows that e = ( d / d t ) [ Ke + W(z , $ )e ] is bounded as
where 2 . p
well. Hence, e being bounded, t? is uniformly continuous
K = block diag ( K, , Kh) (69) and (76) implies, by Barbalat lemma again, that
W ( z ,i2)is called the regressor matrix and is a function of lim IIi( t ) 11 = 0, (77)
the x-variables (and therefore of the z-variables). t+=

Let P = block diag(P,, Pb) be the positive definite therefore it must be


symmetric solution to the Lyapunov equation
K T P + PK = -Q (70)
with Q = block diag<Q,,Q,>, Q,, and Qb positive defi-
MARINO el al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 217

Equation (78) implies, from (55) and (681, that where


1
lim Lfl+le,,(t) = lim - -e,,(t) = 0,
"t , t-m J
V. SIMULATIONS
M2
Iim Lf,Lf+lep,(t)= lim - The proposed control algorithm has been simulated for
t-m t-m a 15 KW motor, with rated torque 70 Nm and rated speed
220 rad/s, whose data are listed in the Appendix.
=o The simulation test involves the following operating
i.e., sequences: the unloaded motor is required to reach the
lim e p , ( t ) = 0, rated speed and the rated value of 1.3 Wb for rotor flux
t-m amplitude ]$I, with the initial estimate of rotor resistance
and, since by virtue of Assumption 1 limt-m T ( t ) = TL,
R , in error of +50%. At t = 2 s. a 40 Nm load torque,
when TL f 0, which is unknown to the controller, is applied. This implies
a reinitialization at t = 2 s. and the theorem given in
lim e p , ( t ) = 0. previous section applies from 0 to 2 s. and from 2 s. on. At
"I
t = 5 s. the speed is required to reach 300 rad/s., well
Remark 6: The assumption that the initial conditions above the nominal value, and rotor flux amplitude refer-
belong to the set S is due to the existence of singular ence is weakened according to the rule I$lref(t) =
points for the determinant (63). It follows that the largest ( k / q e f ( t ) ) .The reference signals for flux amplitude and
z3(0)= l$I2(O) and R r m i ,are the largest are the allowable speed, reported in Fig. 1, consist of step functions,
initial condition errors (e(O), e,(O)). It is therefore more smoothed by means of second-order polynomials. A small
convenient to start the adaptive control algorithm when time delay at the beginning of the speed reference trajec-
the squared flux amplitude l$I2 is far from zero. On the tory is introduced in order to avoid overlapping between
other end it can be seen from the model (16) itself and flux and speed transients.
from (43) that large values for R , make the control task Both nonadaptive (471, (49) and adaptive (59), (641, (74)
easier. control laws have been simulated, with TLN= 0 and
In conclusion the results obtained can be summarized R,N = 0.15 R.
as follows. In the nonadaptive case, we observe from simulations
Theorem: Consider the closed-loop system given by the (compare Figs. 2, 3, and 4, where dashed lines stand for
induction motor model (19) and the adaptive dynamic reference trajectories, solid lines correspond to simulated
state feedback control (591, (641, (74). If behavior and dotted line represents the electric torque)
a> the unknown parameters p , and p2 are constants, that the parameter errors cause a steady-state error both
b) the reference signals ( w r e f ( t )I ,$lfef(t)) satisfy in speed and flux tracking; they also cause a coupling
Assumption 1, between speed and rotor flux which is noticeable both
c) the initial conditions ( 4 0 ) = 0, e,(O)> E S = during speed transient and at load insertion ( t = 2 s.). In
( ( e ,e,,) E R6: eTPe + e;Te < v, v > O), the largest order to clarify the effects of unknown parameters, con-
set entirely contained in (Le,)
E R 6 : ep2 < R,,e3
sider the closed-loop system, obtained applying feedback
control (471, (491, for the simple case of a regulation
> -cJ,
problem, to the motor (19) [recall (5411:
then:
lim ( o ( t )- q e f ( t ) ) = 0, (82)
"t

lim ( h ( t ) - href(t))= 0, (83)


t-m

lim
t+m
(I $(t)l - I$lref(t)) = 07 (84)

where
i s = Lf43
where 2 = (Cl, e,, e,, Z4IT = , y3 ;
( y l .- q e fy,,
Moreover if TL # 0, then
I$l;ef,y4IT, is the regulation error, with qefand I$Iref
constant, K has the structure given in (62), (691, while
W * p takes into account the effects of parameter uncer-
b
218

'c)

v)
:[TI
200

100
L~rpl
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO. 2, FEBRUARY 1993

Flux Am litude Reference

0 0.8 I -0
0 2 4 6 8 0 2 4 6 8 0 2 4 6 8 0 2 4 6 8
Time (sec) Time (sec) Time (sec) Time (sec)

(a) (b) (a) (b)

:
Fig. 1. 8o True & Estimated Load o,2 True & Estimated Resistance
I I

-20 '
0 2 4 6 8
I - 0
0 2 4 6 8
Time (sec) Time (sec)
(C) (d)
Fig. 5.

p 1 = 0, the electric torque T is zero (excepting for a short


transient after the first smoothed step in desired speed,
Flux Amplitude
1.5
I when a coupling is noticed) and rotor flux dynamics reach
,...~~......
a steady state [Fig. 2(b)]: this implies zero steady-state
.............
error according to the above analysis. This is confirmed by
h
simulation; we see (Fig. 2, 0 5 t 5 2) that speed and flux
-1
-2
100
steady-state error is zero even if rotor resistance is in
error of + 50%. Starting at load insertion (at t = 2 s.), the
v)

0 2 4 6 8
- 0
0 2 4 6 8
electric torque and p , are different that zero which cause,
Time (sec) Time (sec) according to (901, coupling and steady-state errors, as
(a) (b) confirmed in Figs. 2(a) and 2(b). Notice that even if
Fig. 3. the load torque were known (and therefore p1 = O),
rotor resistance error ( p 2 # 0) would still cause a speed
steady-state error due to the entry L f 2 L f 4 1which is
proportional to the electric torque [see Figs. 3(a) and
3(b)]. The dynamic responses when both parameters are
e known are reported in Fig. 4 for comparison.
The adaptive case simulations are reported in Figs. 5.
Speed and flux amplitude behavior is shown in Figs. 5(a)
and 5(b), respectively, where solid lines represent actual
variables, dashed lines the corresponding reference values
0 2 4 6 8 0 2 4 6 8
Time (sec) Time (sec)
and dotted line the electric torque. In Figs. 5(c) and 5(d)
load torque and rotor resistance are respectively given,
(a) (b)
where solid lines represent true parameter values and
Fig. 4.
dashed lines the corresponding estimates. The Q matrix
in (70) has been chosen equal to the identity matrix,
tainties [see (18)l. Matrix W * entries are given in (55), the gain matrices K , and K , have been chosen as
from which it is easy to see that Lj,+1 is constant and ( k u 1k, a 2 )= (900,60), ( k , , , k b 2 ) = (900,60) and the para-
L J 2 L f 4 1 is proportional to the electric torque T . The meter update gain matrix r-' has been chosen as r-' =
entry L f 2 + 2is proportional, via a nonzero constant, to the diag ( l / y l , 1 / y 2 ) = diag (0.4,8 X The dynamic
derivative of the squared flux amplitude and therefore, performances of the adaptive control law are satisfactory:
once flux steady state is achieved, LjZ+2 = 0. The entry no steady-state errors occur and transient responses are
L J 2 L J & .can be rewritten as L J 2 L J &= c,(d141~/dt) + decoupled, excepting for an initial short time interval.
c 2 T 2 ,with c1 and c2 nonzero constants. When electric During the first speed transient, due to a wrong initial
torque is zero and flux amplitude steady state is achieved, resistance estimate, a small flux error occurs. At the same
LJ2LJ$b2= 0. Up to 2 s., there is no load torque so that time, due to the torque required to increase speed, rotor
MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 219

400 Applied voltage ua 400 Applied voltage ua provides, on the basis of rotor speed and stator current
I I
measurements, rotor flux and stator current estimates to

-
the adaptive control (591, (741, while rotor resistance
estimates are provided to the observer by the identifica-
tion algorithm (74): those preliminary simulations show
...
.,,.... . .............,.......................,...,., .... that a good performance is still maintained.
-400
0 0.5 1
Time (sec)
1.5 2 -4wi 2:s 3
Time (sec)
3:5 !
VI. CONCLUSION
(a) (b)
400 Applied voltage ua 400 Applied voltaae ua In this paper we propose, for a detailed nonlinear
1 I
model of an induction motor, an adaptive input-output
decoupling control which has some advantages over the
classical scheme of field oriented control. With a compa-
rable complexity exact decoupling between speed and flux
"""."".".",-..
"",
regulation is achieved and two critical parameters (rotor
- 4 4 4'5 ;
Time (sec)
5'5 6 .""6 6.5 7
Time (sec)
1.5 8
resistance and torque load) are identified by a converging
second-order identification algorithm. The main drawback
(0 (4 of the proposed control is the requirement of flux mea-
surements. However, nonlinear flux observers from stator

,
Fig. 6.
currents and rotor speed measurements have been
obtained in [46], and preliminary simulations show a satis-
factory performance of the proposed algorithm even when
,StatorC,went ia, Stator Current ia
I I flux signals are provided by the observers given in [46].
50 Additional research should analyze the influence of
3 sampling rate, truncation errors, measurement noise, sim-
- 0
plifying modeling assumptions, unmodeled dynamics and
saturations. Moreover, the induction motor control prob-
-50
lem should motivate additional research on nonlinear
0 0.5 1 1.5 2 2 2.5 3 3.5 4 multivariable output feedback adaptive control, since only
Time (sec) Time (sec) single output systems have been so far considered in the
(a) (b) nonlinear adaptive literature (see [19] and [351).
Stator Current ia Stator Current ia

APPENDIX

Induction Motor Data

6 6.5 7 7.5 8
stator resistance (0.18 Q)
4 4.5 5 5.5 6
Time (sec) Time (sec) rotor resistance (0.15 0)
(C) (4 stator current
stator flux linkage
Fig. 7.
rotor current
rotor flux linkage (1.3 Wb) rated
voltage input
resistance estimate quickly converges to the true value angular speed (220 rad/s) rated
and complete decoupling is achieved. Fig. 6 shows the number of pole pairs 1
control input signal U,. Control action consists in varying angle of rotation
amplitude and frequency of the applied voltage. Voltage stator inductance (0.0699 H )
supply signals are well within the capabilities of actual rotor inductance (0.0699 H )
inverters and therefore can be easily implemented by mutual inductance (0.068 H )
current power electronic technology (see [91). Fig. 7 shows rotor inertia (0.0586 Kgm2)
the i, current waveform. load torque (70 Nm) rated
As already remarked the availability of flux measure- electric motor torque (rated power 15 Kw)
ments is an unrealistic assumption. However in the litera-
ture several asymptotic flux observers have been proposed
[7], [8], [2], [461, which are rather sensitive to rotor resis- The changes of variables which transform the motor
tance variations. In [361 a simulation study is reported for equation in the original three phase system to the equiva-
a control scheme in which the observer proposed in [46] lent two phase reference frame (see [23, p. 135 and 1701)
220 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 38, NO. 2, FEBRUARY 1993

are given by:

1 cos p cos p +-
3
2rr
cos p - -
2T
3
[21] D. Kim, I. Ha., and M. KO, “Control of induction motors via
feedback linearization with input-output decoupling,” Intemat. J.

I
Contr., vol. 51, no. 4, pp. 863-883, 1990.
2rr 2?r 1221 D. S. Kirschen, D. W. Novotny, and T. A. Lipo, “Optimal efficiency
K, = -sin p -sin p +-
3
-sin p - - . (91)
3
control of an induction motor drive,” IEEE Trans. EnergV Conu.,
vol. EC-2, no. 1, pp. 70-75, Mar. 1987.
1 [23] P. C. Krause, Analysis of Electric Machinely. New York
- McGraw-Hill, 1986.
2 - [24] P. C. Krause and C. H. Thomas, “Simulation of symmetrical

where D
, is given
” bv (23).
at the 10th IFAC World Congress, pp. 349-354, Munich, 1987.
REFERENCES A. Kusko and D. Galler, “Control means for minimization of
losses in AC and DC motor drives,” IEEE Trans. Indust. Appl., vol.
A. Bellini, G. Figalli, and G. Ulivi, “A microcomputer based IA-19, no. 4, pp. 561-570, July/Aug. 1983.
optimal control system to reduce the effects of the parameter W. Leonhard, Control of Electrical Drives. Berlin: Springer-Verlag,
variations and speed measurements errors in induction motor 1985.
drives, IEEE Trans. Indust. Appl., vol. IA-2, no. 1, pp. 42-50, 1986.
C. M. Liaw, C. T. Pan, and Y. C. Chen, An adaptive controller for
-, “Analysis and design of a microcomputer-based observer for current-fed induction motor, IEEE Trans. Aerosp. Elect. Syst., vol.
an induction machine,” Automatica, vol. 24, no. 4, pp. 549-555, 24, no. 3, pp. 250-262, 1988.
1988.
F. Blaschke, “Das prinzip der feldorientierung, die grundlage A. De Luca and G. Ulivi, “Dynamic decoupling of voltage fre-
f i r die transvector regelung von asynchronmaschienen,” Siemens- quency controlled induction motors,” presented at the 8th Int.
Zeitschriji, vol. 45, pp. 757-760, 1971. Conf Analysis Optimiz. Syst., pp. 127-137, INRIA, Antibes, 1988.
-, “The principle of field orientation applied to the new -, “Design of exact nonlinear controller for induction motors,”
transvector closed-loop control system for rotating field machines,” IEEE Trans. Automat. Contr., vol. 34, no. 12, pp. 1304-1307, Dec.
Siemens-Rev, vol. 39, pp. 217-220, 1972. 1989.
M. Bodson and J. Chiasson, “Application of nonlinear control R. Marino, “An example of nonlinear regulator,” IEEE Trans.
methods to the positioning of a permanent magnet stepper motor,” Automat. Contr., vol. AC-29, pp. 276-279, Mar. 1984.
in Proc. 28th Int. Conf Decision Contr., Tampa, FL, 1989, pp. -, “On the largest feedback linearizable subsystem,” Syst. &
531-532. Con@.Lett., vol. 6, pp. 345-351, Jan. 1986.
C. C. Chan, W. S. Leung, and C. W. Ng, “Adaptive decoupling R. Marino, S. Peresada, and P. Valigi, “Adaptive partial feedback
control of induction motor drives,” IEEE Trans. Indust. Electronics, linearization of induction motors,” in Roc. 29th Conf Decision
vol. 37, no. 1, pp. 41-47, Feb. 1990. Contr., pp. 3313-3318, Honolulu, HI, 1990.
Y. Dote, “Existence of limit cycle and stabilization of induction -, “Adaptive nonlinear control of induction motors via extended
motor via new nonlinear state observer,” IEEE Trans. Automat. matching,” P. V. Kokotovic, Ed., Foundations of Adaptitre Control,
Contr., vol. AC-24, no. 3, pp. 421-428, June 1979. (Lecture Notes in Control and Inf. Sciences). Berlin: Springer-
-, “Stabilization of controlled current induction motor drive Verlag, pp. 435-454, 1991.
system via new nonlinear state observers,” IEEE Trans. Ind. Elect. R. Marino and P. Tomei, “Global adaptive observers and output-
Contr., Instrum., vol. IECI-27, pp. 77-81, May 1980. feedback stabilization for a class of nonlinear systems,’’ P. V.
-, Servo Motor and Motion Control Using Digital Signal Proces- Kokotovic, Ed., Foundations of Adaptiue Control, (Lecture Notes in
sors. Englewood Cliffs, NJ: Prentice-Hall, 1990. Control and Inf. Sciences). Berlin: Springer-Verlag, pp. 455-493,
A. E. Fitzgerald, C. Kingsley, Jr., and S. D. Umans, Electric 1991.
Machinery. New York: McGraw-Hill, 1983.
R. Marino and P. Valigi, “Nonlinear control of induction motors: A
E. Freund, “The structure of decoupled nonlinear systems,” Int. J.
Contr., vol. 21, no. 3, pp. 443-450, 1975. simulation study,’’ in Proc. 1991 European Contr. Conf , Grenoble,
France, 1991, pp. 1057-1062.
L. J. Garces, “Parameter adaptation for speed-controlled static AC
drive with a squirrel-cage induction motors,” IEEE Trans. Indust. G. Meyer, R. Su, and L. R. Hunt, “Application of nonlinear
Appl., vol. IA-16, no. 12, pp. 173-178, Mar. 1980. transformation to automatic flight control,” Automatica, vol. 20,
L. R. Hunt, R. Su, and G. Meyer, “Design for multiinput nonlinear no. 1, pp. 103-107, 1984.
systems, in R. W. Brockett, R. S. Millman, and H. J. Sussmann, K. Nam and A. Aropostathis, “A model reference adaptive control
Eds., Differential Geometric Control Theory. pp. 268-298, scheme for pure feedback nonlinear systems,’’ IEEE Trans.
Birkhauser, Boston, 1983. Automat. Contr., vol. 33, pp. 803-811, 1988.
M. Ilic-Spong, R. Marino, S. Peresada, and D. G. Taylor, “Feed- H. Nijmeijer and A. J. van der Schaft, Nonlinear Dynamical Control
back linearizing control of switched reluctance motors, IEEE Systems. Berlin: Springer-Verlag, 1990.
Trans. Automat. Contr., vol. AC-32, no. 5, pp. 371-379, May 1987. V. H. Popov, Hyperstability of Control Systems. Berlin: Springer-
A. Isidori, Nonlinear Control Systems. Communications and Con- Verlag, 1973.
trol Engineering Series. Berlin: Springer-Verlag, second ed., 1989. L. Praly, G. Bastin, J. B. Pomet, and Z. P. Jiang, “Adaptive
A. Isidori, A. J. Krener, C. Gori Giorgi, and S. Monaco, “Nonlin- stabilization of nonlinear systems,” P. V. Kokotovic, Ed., Founda-
ear decoupling via feedback A differential geometric approach,” tions of Adaptbe Control (Lecture Noes in Control and Inf. Sci-
IEEE Trans. Automat. Contr., vol. AC-26, pp. 331-345, 1981. ences). Berlin: Springer-Verlag, pp. 347-433, 1991.
MARINO et al.: ADAPTIVE INPUT-OUTPUT LINEARIZING CONTROL OF INDUCTION MOTORS 221

[42] A. Sabanovic and D. Izosimov, “Application of sliding modes to Sergei M. Peresada was bom in Donetsk, USSR,
induction motor control,” IEEE Trans. Indust. Appl., vol. IA-17, no. on January 14, 1952. He received the Diploma
1, pp. 41-49, Jan./Feb. 1981. of Electrical Engineer from Donetsk Polytechni-
[43] S. S. Sastry and A. Isidori, “Adaptive control of linearizable cal Institute, Donetsk, in 1974 and the Candi-
systems,” IEEE Trans. Automat. Contr., vol. 34, pp. 1123-1131, date of Sciences degree in electrical engineering
1989. from the Kiev Polytechnical Institute, Kiev, in
[44] D. G. Taylor, P. V. Kokotovic, R. Marino, and I. Kanellakopoulos, 1983.
“Adaptive regulation of nonlinear systems with unmodeled dynam- From 1974 to 1977 he was a Research Engi-
ics,” IEEE Trans. Automat. Contr., vol. 34, pp. 405-412, 1989. neer in the Department of Electrical Engineer-
[45] A. Teel, R. Kadiyala, P. Kokotovic, and S. S. Sastry, “Indirect ing, Donetsk Polytechnical Institute. Since 1977
techniques for adaptive input-output linearization of nonlinear he has been with the Department of Electrical
systems,” Int. J. Contr., vol. 53, pp. 193-222, 1991. Engineering, Kiev Polytechnical Institute, Ukraine, where he is currently
[46] G. C. Verghese and S. R. Sanders, “Observers for f lw estimation Docent (the equivalent of Associate Professor in the US). From 1985 to
in induction machines,” IEEE Trans. Indust. Elect., vol. 35, no. 1, 1986 he was a Visiting Professor in the Department of Electrical and
pp. 85-94, Feb. 1988. Computer Engineering, University of Illinois at Urbana, Champaign. His
research interests are applications of modem control theory (nonlinear
control, adaptation, VSS control) in electromechanical systems, model
development, and control of electrical drives and internal combustion
engines.
Riccardo Marino was born in Ferrara, Italy, in
1956. He received the degree in nuclear engi-
neering, in 1979, and the Master’s degree in
systems engineering, in 1981, both from the
University of Rome “La Sapienza,” Rome, Italy. Paolo Valigi received the B.S. degree in elec-
He received the Doctor of Science degree tronic engineering from the University of Rome
in system science and mathematics from “La Sapienza,” in 1986 and the Ph.D. degree
Washington University, St. Louis, MO, in 1982. from the University of Rome “Tor Vergata,” in
Since 1984, he has been with the Department 1991.
of Electronic Engineering at the University of From 1986 to 1991 he has been at the Depart-
Rome “Tor Vergata,” Rome, Italy where he is ment of Electronic Engineering, University of
currently Professor of systems theory. He visited the University of Rome “Tor Vergata.” In 1989 he was a visiting
Illinois at Urbana-Champaign, IL, during the academic years of 1985-86 scholar, at the Coordinated Science Laboratory,
and 1988-89 and the University of Twente, The Netherlands, in 1986. University of Illinois at Urbana Champaign. His
His research interests include theory and applications of nonlinear research interests are in nonlinear control,
control and, more recently, nonlinear adaptive control. robust, and adaptive control, queueing systems, and robotics.

You might also like