1054 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 28, NO.

5, SEPTEMBER/OCTOBER 1992

Adaptive Speed Identification for

Vector Control of Induction
Motors without Rotational
Transducers
Colin Schauder

Abstract-The paper describes a model-reference adaptive benefit from the closed-loop current-control techniques
system (MRAS) for the estimation of induction motor speed that have evolved for use in vector control systems. Cur-
from measured terminal voltages and currents. The estimated
speed is used as feedback in a vector control system, thus rent control is readily applicable to existing voltage-source
achieving moderate bandwidth speed control without the use of inverters, where it reduces the incidence of over current
shaft-mounted transducers. This technique is less complex and tripping and improves inverter utilization.
more stable than previous MRAS tacholess drives. It has been Once the inverter is current controlled, additional con-
implemented on a 30-hp laboratory drive, where its effectiveness trols must be provided to specify the magnitude and slip
has been verified.
frequency of the injected current vector and hence regu-
late the flux and torque of the motor. Motor speed
feedback is typically required for outer-loop speed control
INTRODUCTION
as well as in the flux and torque control algorithm. This

M ODERN CONTROL techniques for ac motor drives

were developed largely as a result of the search for
low-cost alternatives to high-performance four-quadrant
presents a problem in low-performance systems where
motor speed transducers are not usually available. This
has led to a renewed interest in “tacholess”vector control
dc servo drives. In these applications, the use of shaft- with the objective of providing an intermediate class of ac
mounted tacho generators and resolvers is established drives with enhanced performance and a wider range of
practice, and the digital shaft-position encoder used in the applications than “adjustable-speed”drives but about the
most effective vector-control schemes is considered ac- same cost except for the small additional cost of more
ceptable. Nevertheless, the shaft encoder does present sophisticated control algorithms.
problems. Delicate optical encoders with internal signal- The traditional approach to tacholess vector control
conditioning electronics are widely used. These lower the uses the method of field orientation, either relative to the
system reliability, especially in hostile environments, and rotor flux linkage vector or its time derivative [ll. The
require careful cabling arrangements with special atten- difficulty in either case lies in determining the instanta-
tion to electrical noise. There are also situations where neous orientation of the relevant vector. Tamai et al. [21
the positional feedback is extremely difficult to obtain. have described an approach in which the field-oriented
This is particularly true for the case of linear-motor drives reference frame is identified by means of a model-refer-
such as for transportation vehicles. Finally, the encoder is ence adaptive system (MRAS). This paper presents an
a cost factor since the provision of special motor-shaft alternative tacholess control method that uses an MRAS
extensions and encoder-mountingsurfaces leads to more to determine the motor speed and thereby establishes
expensive machines. vector control of the motor as well as overall speed
Until recently, the rapid developments in vector control control. The new M U S scheme is thought to be less
technology have had little impact on “adjustable-speed” complex and more effective than the previous approach.
ac drives. These are typically simple voltage-source invert- It has been implemented in a 30-hp drive, which has
ers with variable output frequency and are used for appli- proved its viability and robust nature.
cations requiring little dynamic control, such as pumps
and fans. It has now become clear that these drives can CALCULATION OF MOTORSPEED
Many schemes [3] based on simplified motor models
Paper IPCSD 91-144, approved by the Industrial Drives Committee of
the IEEE Industry Applications Society for presentation at the 1989 have been devised to sense the speed of the induction
IEEE Industry Applications Society Annual Meeting, San Diego, CA, motor from measured terminal quantities for control pur-
October 1-5. Manuscript released for publication December 30, 1991. poses. In order to obtain an accurate dynamic representa-
The author is with the Science and Technology Center, Westinghouse
Electric Corporation, Pittsburgh, PA 15235-5098. tion of the motor speed, it is necessary to base the
IEEE Log Number 9203291. calculation on the coupled circuit equations of the motor.

0093-9994/92$03.00 0 1992 IEEE

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
 SCHAUDER. ADAPTIVE SPEED IDENTIFICATION FOR VECTOR CONTROL 1055

Since the motor voltages and currents are measured in a

stationary frame of reference, it is also convenient to
express these equations in the stationary frame:

P.[ ::] [:
= [];

where L l , 2 is the stator, rotor self-inductance, M is the Fig. 1. Open-loop calculationof motor speed.
mutual inductance, R, is the stator resistance, T2 is the
rotor time constant, A, i, U is the.rotor flux, stator current,
and stator voltage, respectively, the d, q subscripts denote
the dq-axis components in the stationary reference frame, I
-t r ~. -1
p . = d/dt., w, is the rotor electrical angular velocity, and
a is the motor leakage coefficient.
Given complete knowledge of the motor parameters,
the instantaneous speed w, can be calculated directly
from measured voltages and currents on an entirely
open-loop basis. First, the angle 4 of the rotor flux vector
and its derivative are defined as follows:

4 = tan-' (2)
'd( pmAq - ( P - Ad
P-4=
(A; + A:) Fig. 2. Speed calculation in the rotor flux vector reference frame.

Then, substituting for p.A, and p.A, in this expression

from (21, we obtain Fig. 2 illustrates the calculation of motor speed using this
method. The angular position of the rotor flux vector 4 is
obtained from an observer based on (1) using a vector
phase-locked loop with good dynamic response. Notice
that the term L 2 / M in (1) has been eliminated because
This result indicates that the instantaneous angular veloc- this observer is only required to calculate the angle of the
ity of the rotor flux vector, as wzll as its instantaneous slip flux vector. The current components in flux vector coordi-
relative to the rotor, can be obtained from a rotor flux nates id, and i,, are obtained through a rotating-axis
observer based on (1). This process is illustrated in the coordinate transformation performed on the measured
block diagram of Fig. 1. currents using the angle 4. The calculation of slip fre-
This open-loop calculation process requires knowledge quency from the transformed currents depends only on
of four constants that depend on motor parameters. The knowledge of the rotor time constant T2. This approach to
parameter sensitivity can be reduced by calculating the the calculation of motor speed thus requires only three
slip in a rotating reference frame that is locked to the constants ( R , , aL,, and T2)that depend on motor param-
rotor flux vector. When (2) is transformed to this frame, eters.
A,, =.p.A,, = 0, and we obtain the following well-known
equations: SPEEDIDENTIFICATION USINGMRAS TECHNIQUES
Fig. 3 illustrates an alternative way of calculating the
p.4 - 0,=
Mi,,
- motor speed by means of MRAS techniques. Two inde-
T2 A d , pendent observers are constructed to estimate the compo-
nents of the rotor flux vector: one based on (1) and the
Mid,
other based on (2). Since (1) does not involve the quantity
'd, =
1 + T2p. or,this observer may be regarded as a reference model of

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
 1056 IEEE TRANSACTIONSON INDUSTRY APPLICATIONS,VOL. 28, NO. 5, SEPTEMBER/ OCTOBER 1992

Linear block

Fig. 4. MRAS representation as a nonlinear feedback system

Since G, is a function of the state error, these equa-

tions describe a nonlinear-feedback system as illustrated
Fig. 3. Structure of M U S system for speed estimation. in Fig. 4. Following Landau, hyperstability is assured,
provided that the linear time-invariant forward-path
the induction motor, and (2), which does involve w,, may transfer matrix is strictly positive real and that the nonlin-
be regarded as an adjustable model. The error between ear feedback (which includes the adaptation mechanism)
the states of the two models is then used to drive a satisfies Popov’s criterion for hyperstability. It can be
suitable adaptation mechanism that generates the esti- shown that the forward path transfer matrix in the present
mate G, for the adjustable model. MRAS’s of this kind system is indeed strictly positive real (i.e., [ F ( j w ) +
are increasingly being used to identify plant parameters F T ( - j w ) ] is a strictly positive definite Hermitian matrix,
and inaccessible variables. A successful MRAS design can where F ( s ) = [sZ- A ] - ’ ) , but the proof is lengthy and
yield the desired values with less computational error than will not be repeated here. Popov’s criterion requires a
an open-loop calculation and is often simpler to imple- finite negative limit on the input/output inner product of
ment. the nonlinear feedback system. Satisfying this criterion
In designing the adaptation mechanism for a MRAS, it leads to a candidate adaptation mechanism as follows:
is important to take account of the overall stability of the Let Gr = Q2([EI)+ /‘@,([EI)dT.
system and to ensure that the estimated quantity will 0
Popov’s criterion requires that
converge to the desired value with suitable dynamic char-
acteristics. Landau [4] has described practical synthesis
techniques for MRAS structures based on the concept of $ [ . ] ‘ [ W I dt. 2 - -yo2 for all t , 2 o
hyperstability. When designed according to these rules,
the state error equations of the MRAS are guaranteed to where yo2 is a positive constant. Substituting for [ E ] and
be globally asymptotically stable. Landau’s synthesis [ W ] in this inequality and using the definition of G,,
method leads to candidate structures for the adaptation Popov’s criterion for the present system becomes
mechanism but does not establish the dynamics of the
convergence process. To study the dynamic response of a
particular MRAS, it is always necessary to resort to analy-
sis of the system equations, linearized about a selected
operating point. - /hl([ E ] ) d 4 dt. 2 - y;
0
In general, wr is a variable, and the models are linear
time-varying systems. For the purpose of deriving an A solution to this inequality can be found through the
adaptation mechanism, however, it is valid to initially following well-known relation:
treat w, as a constant parameter of the reference model. 1
Subtracting ( 2 ) for the adjustable model from the corre- jblk(p.f(r))f(t) dt. 2 - -k.f(0)2, k > 0.
sponding equations for the reference model, we obtain 2
the following state error equations: Using this expression, it can be shown that Popov’s in-
equality is satisfied by the following functions:

= K 2 ( e q i d- e d i q )= K,(Aqid- A d i q )

Q 2 = K 1 ( e y i d- e d i q )= K l ( A q i d- A d i q ) ,

Fig. 5 shows the block diagram of the speed identifica-

tion MRAS based on this adaptation mechanism. Notice
that the factors L,/M in (1) and M / T , in (2) have
that is, p [ &=
] [AI[&-
] [WI. conveniently been incorporated into the adaptation mech-

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
SCHAUDER: ADAPTIVE SPEED IDENTIFICATION FOR VECTOR CONTROL 1057

Fig. 6. Steady-state error functions.

Dynamic Response of MRAS Speed Identification

I I
In general, the quantities Gr and or are time varying,
Fig. 5. Block diagram of MRAS speed identification system. and each may be regarded as an input to a system de-
scribed by (2). In order to investigate the dynamic re-
sponse of the MRAS speed identifier, it is necessary to
anism gain constants K , and K,. The outputs of the two
linearize these equations for small deviations about a
models thus only represent the rotor flux vector in angle
particular steady-state solution. If this is done in a station-
and have been renamed with primed symbols. As in the
ary reference frame, the resulting linear equations will
system of'yig. 2, the only remaining parameters are R I ,
still be time varying; therefore, it is useful to first trans-
c+L,,and T2.Note that the angular position of the rotor
form the equations to a reference frame rotating syn-
flux vector is not calculated explicitly, and all variables are
chronously with the stator current vector. We thus obtain
expressed in the stationary frame of reference so that the
the following equations:
input voltages and currents are obtained from measure-
ments through constant gain terms only.
The M U S of Fig. 5 can be interpreted as a vector
phase-locked loop in which the output flux vector from
the reference model (A& + jhh) is the reference vector,
and the adjustable model is a vector phase-shifter con-
trolled by Gr. Examination of the adaptation mechanism
shows that the error term E is proportional to the sine of L J

the angle error between the two vector outputs, thus

supporting the notion of a phas:-lock:d loop. The influ-
ence of Gr on the phase of (A& + j A b ) is seen in the
steady-state solution of (21, which leads to the following
phasor relationship between input current and output
flux.
& -

- M(1 - ~ ( W O - mro>T,> I .
A=
(1 + ( W O - mr0l2T,2)

Relative to a given input current vector with constant

angular frequency w 0 , the phase angle of the output flux
vector swings monotonically from -90" to +90" as the
value of wr0 is taken from --CO to +m. Fig. 6 shows the
error function E calculated for a 30-hp motor under three
different load conditions in the steady state, assuming
constant rotor flux vector magnitude and correct motor The error function E has the form of a vector inner
parameter values. The hyperstability synthesis guarantees product that is independent of the reference frame in
that the estimated speed will converge to the actual speed which the vectors are expressed. It may thus be repre-
from any initial value, and this can be clearly seen from sented by the following linearized expression:
the sense of the error functions in Fig. 6. These functions
always maintain the polarity required to drive the esti-
mate Gr0 towards the correct value wr0.
 1058 IEEE TRANSACTIONSON INDUSTRY APPLICATIONS, VOL. 28, NO. 5, SEPTEMBER / OCTOBER 1992

& ( f r o m MRAS)

Fig. 7. Block diagram representing MRAS dynamic response.

Reference
-1 Im

Stator Phase
Current References

Fig. 9. Vector control system using M U S speed estimate.

Fig. 8. Sketch of closed-loop root locus (for case kZ/kl > 1/T2).
'd
From these equations, we can obtain the transfer function Iq

relating A orto A E as follows:

A E I
G ( A w , = O
=-
I
"rlAw,=O
-
(
ii2+I?
s - .lAOl2

'
REFERENCE
(a)

MODEL
s+ - +(wo-wro)2
l9
'd
ADJUSTABLE xd -
= G,(s).IA,I~ P +'/T MODEL

?here lA0I2 = ($,, + Ate0), and we assume that A,, = '9 ,.

"r
AqO, and A d o = A d o . This allows us to draw a block dia-
(b)
gram of the speed identification as shown in Fig. 7. The
closed-loop root locus sketch of Fig. 8 illustrates that the Fig. 10. M U S ' S using auxiliary variables.
achievable bandwidth with which the actual speed can be
tracked is limited only by noise considerations. in the value of T, will reflect as an error in the speed
feedback, which will affect the accuracy of the speed
VECTORCONTROL USINGMRAS SPEEDESTIMATE control slightly.
The speed estimate Gr can be made to track the actual
PRAC~ICAL
CONSIDERATIONS
speed very closely and can be used both for speed loop
feedback and for orienting the injected stator current Use of Auxiliary Variables
vector for torque and flux control. A block diagram illus- In practice, the rotor flux observer based on (1) is
trating this use of the speed estimate is shown in Fig. 9. difficult to implement because of the pure integration of
An important feature of this system is that, provided the sensed variables that is required. This leads to problems
same value of T, is used in the MRAS adjustable model with initial conditions and drift. The MRAS structure has
and in the function blocks of Fig. 9, perfect orientation of advantages in this regard because the model outputs need
the injected current vector is achieved, in theory, even if not be the actual motor flux components but can be
the value of T2 used is quite wrong. If the MRAS success- auxiliary variables related to them. Although it is desir-
fully maintains nearly zero error, then the adjustable able to eliminate the pure integrals, it is also desirable to
model accurately replicates the dynamic relationship be- retain some measure of low-pass filtering in their place.
tween the stator current vector and the rotor flux vector This helps to normalize the model outputs and attenuates
that exists in the actual motor. Examination of (2) shows the high-frequency components normally found in the
that this is only possible if motor terminal voltages. Fig. 10 shows two possible ways
of modifying the system of Fig. 5 to achieve this objective.
'2( 0 - 'r 1 = ',(actual)( 0 -or)
In each case, an identical linear transfer matrix is inserted
where w is the instantaneous angular velocity of the into both the reference and the adjustable model. This
stator current vector, that is block must appear in the output of the reference model
since the input cannot be altered, but in the case of the
T2 u s 1 = ',(actual) wsl(actua1).
adjustable model, it may appear in either the input or the
Thus, if T2 # T2(actua,), then wsr # o ~ ~ ( ~ but
~ ~ since
~ ~ , only
) , output as shown.
the product T2wJI is used in setting up the injected In Fig. 10(a), the reference and adjustable models have
current vector, this is always done correctly for the actual identical structures, but this is no longer a hyperstable
slip, even under dynamic conditions. Naturally, the error system since it does not meet the conditions for hypersta-

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
 SCHAUDER ADAPTIVE SPEED IDENTIFICATION FOR VECTOR CONTROL 1059

REGENERATING MOTORING

1
T I
~~~~~~ ~ ~~

Fig. 12. Speed identification dynamics for Fig. lO(b).

drive is allowed to dwell in this region. The MRAS of Fig.

10(b) is believed to be generally superior to that of Fig.
10(a), and it was therefore used in the experimental
system.
Fig. 13 shows a block diagram of this system as imple-
mented in a microcomputer for control of a 30-hp drive.
Notice that some functions are performed in analog cir-
-2 -1
-3 -2 -1 0 cuitry external to the computer so that the signals sam-
pled through the A/D interface have limited high-
Fig. 11. Pole/zero loci for G,(s). frequency components. The analog circuitry requires am-
plifiers with extremely low offset voltages to eliminate the

where wsro= ( w o - wr0).

Fig. 11 shows the parametric pole/zero locus of this
transfer function for a 30-hp motor under two load condi- Motor Parameter Sensitivity
tions with T = 1 s and T2 = 0.76 s. In each case, the loci As has been mentioned, the parameter T2 has negligi-
start at w o = 25.12 rad/s (4 Hz) and extend to wo = 0 in ble influence on the operation of the overall MRAS/vec-
steps of 0.157 rad/s (0.025 Hz). Notice the complex tor control system. The terms a L , and R, influence the
pole/zero pairs in the region of the excitation frequency. accuracy with which the observer based on (1) represents
Although the MRAS operates .well under most conditions, the phase angle of the rotor flux vector. An error in the
it is possible to excite these resonant poles through tran- angle of the reference model output is tracked by the
sient changes. A more serious problem is evident in the adjustable model with a corresponding error in h,. This
steady state under extremely low-frequency motoring, results in the machine operating at a different flux level
where right-plane zeroes indicate probable closed-loop from that intended, with a consequent change in the
instability. In practice, this MRAS has been found to work maximum torque capability. The effect of an error in R,
well around and through zero frequency, although the is usually quite negligible at high excitation frequency but
speed estimate is not usable if the excitation dwells in the becomes more serious as the frequency approaches zero.
danger zone for more than a few seconds. This occurs because the term R,i/p. becomes relatively
In Fig. 10(b), the reference and adjustable models have larger as the frequency decreases. In contrast, an error in
different mathematical structures, although it can be crL, has roughly the same effect at all excitation frequen-
shown that the system is indeed hyperstable for the spe- cies, where it is zero when the slip frequency is zero and
cial case where wr is a constant. The closed-loop dynam- becomes worse under load. It is therefore very important
ics of this MRAS are identical with those described for to minimize the error in the value of aL1 used in the
the system of Fig. 5. However, when w, is a variable, new reference model.
dynamics are introduced into the reference path of the In the experimental system, the motor parameters are
speed identification block diagram, as shown in Fig. 12. In automatically measured when the drive is at rest by means
effect, the MRAS of Fig. 1Nb) removes the excitation- of injected currents that produce stationary or pulsating
frequency pole/zero pairs from within the high bandwidth fields and, hence, do not produce net torque. This subsys-
identification loop and places them instead in the refer- tem uses a number of MRAS's and is illustrated in Fig.
ence path, where they are less likely to be excited in 14. R, is identified with SW1 closed and with an arbitrary
practice. At very low excitation frequency, the right-plane d-axis injected dc current. SW1 is then opened, and the
zeroes now appear in the transfer function between the value of a L , is read from an analog MRAS that capital-
actual and estimated speeds, and this may be expected to izes on the high-frequency content of the motor voltage.
lead to instability of the outer speed-control loop if the The d-axis current is then pulsed to produce the test

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
 1060 IEEE TRANSACTIONSON INDUSTRY APPLICATIONS, VOL. 28,NO. 5, SEPTEMBER / OCTOBER 1992

0-

Time (ls/div)
I--- - 7 - - 7 --
--7- -,

Fig. 15. Measured response of T2 identification.

Motor (from shaft-mounted

Speed tachometer)
Fig. 13. Microprocessor implementation block diagram.

Ap 0

1
- (ob)
4 Current

Time ( O . l s / d-r
ivd
r I I I I I I

Fig. 16. Measured response of experimental system.

I
- I Xd
tachogenerator for reference purposes. Motor current
feedback signals are obtained using Hall-effect current
sensors, and the motor terminal voltages are sensed with
resistive voltage dividers followed by differential ampli-
fiers.
ld Under operating conditions away from zero speed, the
U performance of the drive is very satisfactory, and speed
Fig. 14. Adaptive systems for parameter identification.
loop bandwidths of 10 to 15 rad/s are easily obtained.
This performance is maintained to quite low speeds
(around 2-Hz excitation) and, in fact, is available right
excitation for the MRAS that identifies T2. The adapta- through zero speed on a transient basis. However, if the
tion mechanism for this M U S is obtained by hyperstabil- drive is allowed to dwell at zero frequency excitation for
ity synthesis using (2) with wr = 0, and its convergence more than a few seconds, the amplitude of the MRAS
from an initial zero estimate to the correct final value is model outputs goes to zero, and speed control is lost.
illustrated in the oscillogram of Fig. 15. Work is presently underway to determine suitable se-
quencing techniques that will make this condition accept-
PERFORMANCE OF THE EXPERIMENTAL SYSTEM able in a practical application. Figs. 16-18 illustrate the
An experimental drive system was built and tested using dynamic response of the experimental system.
the MRAS/vector control technique, which has been
described. Both the MRAS of Fig. 13 and the vector CONCLUSION
control algorithm of Fig. 9 are executed in a 16-b micro- This work was aimed at developing a low-performance
computer that delivers current references to the PWM tacholess vector control system that would allow existing
inverter current-control system. The test machine is a "adjustable-speed" drives to benefit from closed-loop cur-
standard 30-hp, 230-V, 60-Hz cage induction motor that is rent-control techniques. In fact, the performance of the
coupled to a dc motor of similar rating. The inertia of the system has exceeded initial expectations and approaches
set is, thus, more than twice that of the induction motor that of vector control systems using shaft position en-
alone. A speed measurement signal is available from a dc coders. Despite some limitations in the region of zero-

7
- -

Authorized licensed use limited to: UNIVERSITY OF PLYMOUTH. Downloaded on October 10, 2009 at 09:01 from IEEE Xplore. Restrictions apply.
SCHAUDER: ADAPTIVE SPEED IDENTIFICATION FOR VECTOR CONTROL 1061

speed operation, it is expected that the MRAS/vector

control technique will find practical use in a large number
of ac motor drive applications.

REFERENCES
R. Jotten and G. Maeder, “Control methods for good dynamic
performance induction motor drives based on current and voltage
as measured quantities,” IEEE Trans. Industry Applications, vol.
IA-19, no. 3, pp. 356-363, 1983.
S. Tamai, H. Sugimoto, and M. Yano, “Speed sensor-less vector
control of induction motor applied model reference adaptive sys-
tem,” in Conf. Rec. IEEE/MS Ann. Mtg., 1985, pp. 613-620.

Lo I I I I , ,
Time (40ms/div)
I I , , I I I ,I A. Abbondanti, and M. B. Brennen, “Variable speed induction
motor drives use electronic slip calculator based on motor voltages
and currents,” IEEE Trans. Industry Applications, vol. IA-11, no. 5,
Fig. 17. Model outputs during rapid reversal. pp. 483-488, 1975.
Y. D. Landau, Adapiive Control--The Model Reference Approach.
New York Marcel Dekker, 1979.

1
Motor (from shaft-mounted
Speed-tachometer)
Colin Schauder received the B.Sc (Hons.) and
Ph.D. degrees in electrical engineering from the
University of Cape Town, South Africa, in 1972
and 1978, respectively.
From 1978 to 1983, he was employed by GEC
0 Electrical Projects Ltd. and GEC Industrial
Controls, Rugby, Warwickshire, England, where
he worked on the development of high-perfor-
mance ac motor drive systems. Since 1983, he
has been with the Westinghouse Electric Corpo-
ration at the Science and Technology Center,
Pittsburgh, PA, where he is presently an Advisory Engineer in the Power
Time (O.ls/div) Electronics Department. His work involves the design and development
of advanced power conversion systems, including motor drives and static
Fig. 18. Measured response of experimental system. compensators for utility power line applications.