Proceeding of the 2004 American Control Conference WeP19.

3
Boston, Massachusetts June 30 ~ J u l 2,2004
y

An Indirect Variable Structure Model Reference

Adaptive Control Applied to the Speed Control
of a Three-phase Induction Motor
Josenalde B. de Oliveira, Aldayr D.de Araujo
Department of Electrical Engineering, Federal Universiq of Rio Grande do Norte. RN-Brazil
j n a l d e w c a .ufrn.br, [email protected]

of uncertainty is the value of the rotor time constant that

Absfrocf-In this paper an indirect approach to the variable varies with the operation conditions, changing the system
structure model reference adaptive control (VS-MRAC) is behavior. Then there is the necessity of methods of adaptive
proposed. This approach is applied to the speed control of a and/or robust control, which can he applied to systems that
three-phase induction motor. In the direct approach, the present parametric uncertainties.
switching laws are designed for the parameters of the In adaptive control, a common approach is the model
controller, and in this case the main difficulty is to find the
bounds which must be used in the control laws. On the other reference adaptive control (MRAC) [IZ], where is proposed
hand, in the indirect approach, the switching laws are a reference model (transfer function) that describes the
designed for the plant parameters, which are uncertain, but desired I/O properties of the closed-loop plant. This plant
their bounds can be easily found, since they are related to adaptation to model, generally, is slow, generating an
physical parameters of the system. The idea is to design an undesirable transient.
indirect variable structure approach and to apply it to the On the other hand, the variable structure control
speed control of a three-phase induction motor. approach has its roots in relay control, and consists of use a
switching control law as a function of system state
I. INTRODUCTION variables, and, in its common configuration, in order to
In recent days the induction motors have been restrict the system dynamics to a surface referred as a
increasingly taking place of the DC motors in high sliding surface [16].The variable structure systems have as
performance electrical motor drives [ I l l . In the case of main characteristics the fast transient and robustness to
motors with squirrel’s cage rotor, its main advantage is the parameter changes and disturbances (in a range stipulated
elimination of all sliding electrical contacts, resulting in an on project).
exceedingly simple and rugged construction. Induction Thereby, it was proposed a controller that makes use
machines are made in a variety of designs with ratings of a of the characteristics of both techniques [7], i.e., a
few watts to several megawatts. The induction motors can controller with very fast transient, robustness to parameter
be used in adverse atmospheres since that they don’t have changes and disturbances, utilizing only plant inputbutput
commutator and, consequently, there isn’t a possibility of measurements. This controller was named VS-MRAC, that
sparking. With the progress of the power electronics and the was posteriorly modified by simplifications in the original
appearance of low cost and very fast microprocessors, the algorithm proposed by Araujo [l], facilitating the practical
induction motor drives have reached a competitive position application of this controller [8]. The VS-MRAC controller,
compared to DC machines. For the DC motors, the speed in its direct approach, has been successfully applied on
control can be carried out in a simple way, since the torque control of DC machines [15] as well as on control of
and the flux can be decoupled. The technique of vectorial induction machines [41, [51.
control based on the rotor field orientation applied to the In a recent work was presented the indirect approach to
induction motors [Z], [lo], [ I l l , when the motor is fed by the VS-MRAC [13]. In the present paper, in order to
ideal current sources, provides the decoupling between the confirm its feasibility, an application on a three-phase
torque and flux in a similar way to the DC machine. This induction motor is shown.
technique is known as Field Orientation Control (FOC).
The choice of the rotor flux as reference for the d axis
facilitates the decoupling between motor torque and flux U. MODEL OF THE INDUCTION MOTOR
[3], [IO], [ I I]. In this control strategy, an important elemenl
In this section we use a vectorial technique for modeling the
induction motor, which is very important to study field

0-7803-8335-4/04/$17.00 02004 AACC 1946

orientation control [3], [ll]. We define a system of flux in DC machine and the component of stator electrical
complex orthogonal axis, d and q, where the rotor flux is current vector on the quadrature axis is equivalent to the
the reference for de d axis. The motor vectorial diagram is armature current in a DC machine. Additionally, if the
presented in Figure 1, where component of the rotor flux is kept constant, the torque
can be controlled only by the component of the stator
6 stator electrical current vector angle related to the rotor
~ electrical current vector on the quadrature axis.
flux;
p - rotor flux angle related to stator phase 1 axis; ID. PARAMETERlZATlON AND ASSUMPTIONS
ws - stator electrical current vector angular speed; The plant and model reference are defined as follows
vRd( 1 ) - rotor flux related to the d axis; Plant
E - angle between axis of stator phase 1 and rotor phase 1;
is(') - stator electrical current vector; Y(S) = W(s)u(s)

isd,isq - stator electrical current vector components on

direct and quadrature axis, respectively;
dN)
w ( t ) = -- rotor angular mechanical speed.
dt

Im d (s)=s
n
+as
n-I n-l n-1-i
P 1 + c q+]s
i=l

I Reference Model

'sq 'Sd ' Phase 1

E Statorkvls
Fllw 1
n-l n-I-i
Fig. 1. Motor V c e t o d Diagram n,(s)=s + Zpmis
i=l
From Figure 1 we have d,(s)=s
n
c(Y . s n-I-i
+ amlsn-l + n-1
i=l m,r+l (7)

is ( t ) = (isd ( t ) + jisq (t))," (l) In order to simplify the notation, let us define the plant
parameter vector as
(2)
* T T T
Using the vectorial analysis with the rotor flux orientation Bp = [ k p , P ,al,a 1 (8)
[3], [ I I], we obtain the following expression for the torque where

Te(t)= -P-VRd
2 Lm (')isq( t ) (3) - P ER"-l is a vector that contains the elements
3 Lr pi ( i = n - 1,...,I) of (4)
where
L, - rotor inductance by phase; - al E R is the element al O f ( 5 )

L, - magnetization inductance by phase; - (YER"-lis a vector that contains the elements

P - number of poles pairs. a;+l(i = n - 1,J) of ( 5 )
The equation (3) describes the induction motor torque in a
similar way to the DC machine. The component of the
and, likewise, it,s defined pm and am , with respect
rotor flux vector on direct axis is equivalent to the field to reference f"k.

1947
 To obtain the control algorithm, the following T
u =8 (t).w(t)
assumptions are made:

- -the plant is monovariable, controllable, observable and where 8 ( t ) is the adaptive Parameters Vector.
* If the previous assumptions are satisfied, one has that
has relative degree n = 1 ;
- * *T t Y(S)
the degree n of d p (s) is known; 3!0 such that U = 0 w = U implies at -= M ( s ) ,
r(s)
- k , and k p have the same signal (positive, without
i.e. , y + y , (matching condition) where
lost of generality);
- the reference model is strictly real positive (SRP)and * *T *T *
has the same plant relative degree; U =ev1 v1 + 8 , y c e
"2
v2 ie2,r (15)
- n o @ ) , n m ( s ) and d m ( s ) are Hunvitz monk -1
-
polynomials;
d p (s) is a monic polynomial;
Withe = 8
* [ i l T
8*
n
e*
v2
*
82n
]
*
- the upper bounds for k p , pi ( i =l,...,n-l), "land Obviously, 8 only can be calculated if the plant
parameters are known, but when this is not possible,
a i (i = 2,...,n) are known. e * turns into B .

The adaptation algorithm needs 2n variables. Once only

the plant inputloutput are known, it's built a set of state B. Conlrol Law to Matching Condition
variables from U and y , that are filtered signals of plant The expressions to the controller parameters at the
inputloutput [ 9 ] . matching condition [14], when plant parameters are
unknown or partially known, are
A

A. Plant Input/Output Filters

The following filters are defined
n

v2 = A.v2 + g.y k p (f)

A n

where

g=[O . . 0 y]' E Rn-'

YE R
v1 ,v2 E Rn-l

The vector w with the 2n variables is named regressor

where (:) is the estimate of (.) at each instant t
vector and is denoted by

T N.VARIABLE STRUCTURE MODEL REFERENCE DIRECT

w=[v T y v 2 ~r 1 ADAPTIVE CONTROL (VS-MRAC)
-1 - (13)
In this section will be reviewed the strategy of adaptive
and, thus, the control law U is given by control that needs only plant inputloutput measurements,
with switching laws to the controller parameters [I], [7].

1948
 This technique had its origin on the necessity of find a
controller that could turn the closed-loop system robust to
plant uncertainties and disturbances, with transient
performance better than MRAC strategy.
At the scope of this work, the emphasis will be given to
*
plants with n = 1 and the MRAC adaptation integral laws
will be replaced by switching laws.
Hence the adaptive parameters Bi (16) tum in

(18) where

- 1.1
0;> 0; ;i = 1,2,...,2n
<, =--P"V1 --P
A T

VI
where e,, = y - y , is the system output error. U
Y Y
At implementation level, the values of 0;must be chosen
in terms of a determined percentage of nominal values of
*
parametersBi , to avoid generating a very high control <pi = y
- VI ,i
,I = I , ..., n-1
signal. Here is where we may have difficulties to find these

6;,because the evaluation of (17) must be done manually, ri =Y-- P".,Y

what it's rather tedious for high order systems. Another
option is to use an auxiliary computer program to make this
evaluation, but this isn't recommended whereas a good
algorithm should be as self-contained as possible. rei= - , i = 2
v2.i-l ,..., n
By Lyapunov's Stability Theory, is shown that the closed-
loop system presents asymptotic global stability. Also is Y
proved that the null error, chosen as sliding surface, is
reached in a finite time [7]. The stability proof is shown in [14]. The relays'
amplitudes must satisfy
V. VARIABLE STRUCTUREMODEL REFERENCE INDIRECT
ADAF'TNE CONTROL

In a recent work [13], it was developed a controller similar

to VS-MRAC, but with the switching laws made on the
plant parameters. This was made in order to simplify the
controller design, avoiding the problems mentioned above,
and whereas the uncertainties at the plant parameters can be
known easier than in the traditional VS-MRAC, since they
are related with uncertainties in physical parameters, such
as resistances, capacitances, inertia moments, friction
coefficients, etc. The only remark is that now the algorithm
is a bit expensive from the computational point of view, The introduction of k (nominal) in the
since the evaluation of (17) is made directly in the P , nom
algorithm h

Therefore the innovation of this technique is on the design k P expression (19) is necessary to guarantee that
facility, i.e., once deduced the matching condition A

expressions (17), the work consists on defining the range of k p doesn't assume negative values and thus guarantee
pertinence of each plant parameter and on the design of the
switching control laws. the system stability [14]. Therefore, the switching is
Thus, the following switching laws were proposed [13] made only on parameter uncertainty. This procedure is
close to what occur on practice, where one has an
uncertainty on the nominal value of the parameter, like,
e.g. the uncertainty range of a resistor value.

1949
 A

At implementation level, p should be calculated before

A

k p , and, in the expression of c, (ZO), instead of use U ,

it should be used a filtered signal of U (Ua,,).
The control signal obtained in variable structure systems
has a high frequency switching and shouldn't be used on
practical problems, due to limitations in the driver system.
As solutions we could have a change at the relay function
Tachometer
by the introduction of a linear region [ 5 ] (Figure 2) and/or
techniques of smooth control 161. The first one causes error h&tien
Mache
in steady-state, but should be used together with a PI
controller. Fig. 3. Driver System
B. Results

The dynamic of the induction motor is represented by the

following equation

where

J moment of inertia of the rotational mass;

Fig. 2. Lincar region B damping constant;
T, induction motor torque;
The fimction sgm(x) is defined as
T, load torque.

The induction motor model introduced here ( ( 3 ) and

(23)), for a certain operating point, considering the rotor
flux constant and the motor parameters given in [3], yields
to a first order model given by

3798
VI.EXPERIMENTAL RESULTS
W ( s )= P x -
s+a s+11.3
In this section will be presented the results obtained for
the application of the technique described above to the The reference model time constant was chosen near to
speed control of a three-phase induction motor. plant time constant, because high order models with faster
dynamics could generate high control signal, without
A . Driver System necessity. Thus, the chosen reference model was

The driver system used to implement the variable k 12

structure model reference indirect adaptive control is M(s) m- __
showed in Figure 3. It is composed by an 0.25 HP induction s+am s+12
motor fed by a three-phase VSVPWM inverter with current
control by hysteresis window. In the current control, Hall
effect sensors are used to measure the currents of two In analyzing (24), the values used in the algorithm were
- -
phases of the motor. One microcomputer receives the motor k,,,, = 3798, k , = 250, a = 40. The sample period
speed using a tachometer and, by a control solbare in C
language, sends the necessary signal to the inverter. was h = 0.00042s. A linear region with A = 1500 was
introduced in order to smooth the control signal. With these
values the following graphic was obtained (Figure 4), where

1950
the plant and reference model speeds are given in rpm and University of Rio de laneim, Ph.D. Thesis, Rio de Janeiro-RJ (in
pWtUguesc).
the control signal U = i .is given in mA. [2] Bose, B. K. (ZOOO), ”High Pedormance Control of Induction Motor
S¶ Driver”, New Technologies, Department of Electrical Engineering,
PLANT OUTPUT The University ofTennessee, Knoxville, U.S.A.
MODE OUTPUT [3] Cunha, C. D.(2001) “Dual-Mode Robust Adaptive Controller”, MSc.
CONTROL SIGNAL
Thesis, Federal Universily of Rio Grande do Narte, Natal-RN (in
pomguese).
[4] Cunha, C. D., Aralijo, A.D., Barbalho, D. S. and Mota, F. C. (2002),
“A Dual Mode Adaptive Robust Controller Applied to the Speed
I Control of a Threc-Phase Induction Motor”, 7“ International
Workshop on Variable S t ~ c t uSystems,
~ Sarajevo-Bosnia and
Herzegovina, Vol. I , pp. 253-264, JulyI2002.
[5] Fumnato, A. F. A., Salarar, A. 0. and Araujo, A. D. (2001) “Speed
Control Using Sliding Modes to a Three-phase Induction Motor”,
Control & Automation Magazine, Vol.12, N. 2, pp. 148-155.
[6] HE”, L. (1997) “Smooth Sliding Control of Uncertain Systems Based
an a Prediction Error”, 1. I. Robust and Nonlinear Control, Vol. 7,
pp, 353-372.
[7] Hsu, L. and Casta, R. R. (1989) ”Variable S t ~ c t u r Model
e Reference
Adaptive Control Using Only Input and Output Measurements - Part
I”, International Journal OfCantrol, Vol. 49, N. 2, pp. 399-416.
[8] Hsu, L. and Costa, R. R. (1996) “Adaptive Control with Sliding
Modes: Theory and Applications”, Proceedings of the XI CBA,
Minicourses, pp. 39-60.
Fig. 4. Expenmental result [9] loannou, P. A. e Sun, J. (1996) Robust Adaptive Control, Prentice
Hall.
In the practical test the motor and the reference model [IO] Larrea, I. A. S. (1993) ‘%amparative Study of Vectorial Control
outputs start with zero initial speed. The reference is Methods to Induction Motors”, MSc. Thesis, Federal University of
Rio de Janeiro, Ria de Janeiro-RI (in pormguese).
initially assumed as 1000 rpm, after a certain time it’s [Ill Leanhard, W. (1996), “Control of Electrical Drives”, Springer-
increased to 1200 rpm, and finally it’s returned to 1000 Verlag, second edition.
rpm. After that, it’s introduced an external disturbance. The [I21 Narendra, K. S. and Valavani, L. S. (1978) “Stable Adaptive
plant follows the reference in each case (the model gain is Controller Design-Part I-Direct Control“, E E E Tons. Automatic
Control, Vol.AC-23, N. 4, pp. 570-583.
one). The disturbance effect is felt only by increasing the
[I31 Olivcira, I. B. and Araujo, A. D.(2002) “Variable Structure Model
control signal during the time interval in which the Reference Indirect Adaptive Control: Indirect VS-MRAC”,
disturbance is applied. The transient behavior is fast and Proceedings of the XN CBA, Natal-RN, pp. 2557-2562 (in
with no oscillations as it was expected. pomguese).
[I41 Oliveira, J. B. (2003) “Variable Stmcture Model Rcferenco Indireet
Adaptive Control”, MSc. Thesis, Federal University of Rio Grande
VU. CONCLUSION do Narte, Natal-RN (in pormguese).
[IS] Silva, W. A. C. M., Aralija, A. D. and Salazar, A. 0. (1996) “A
Variablc Structure Adaptive Control Applied to the Spced Control of
In this work, the new technique of variable structure model DC Machines”, Proceedings of thc XI CBA, Vol. I , pp. 45-50.
reference adaptive control was applied to the speed control 1161 Utkin, V. I. (1978) “Sliding M o d s and Their Applications in
of a three-phase induction motor, a motor with wide range Variable St~ctureSystcms”, Mir Publishers, Mosco03-vidmar
applications in industry. This technique has as objective and
main advantage turn the design easier, once that the
switching laws are designed to the plant parameters that
have uncertainties known easier than in the direct approach,
once they are physical parameters. The results have
confirmed that the proposed technique presents a
performance similar to direct case.
As suggestions to the continuity of this work we could
mention the generalization to plants with relative degree
bigger than one, a deepest study of linear regions and
smooth control, as well as applications in several areas like
robotics, process control, etc.

REFERENCES

[ I ] Aralija, A. D. (1993) “Contribuition to Theory of Variable Structure

Model Refercnce Adaptive Control: An InpuVOutput”, Federal

1951