Professional Documents
Culture Documents
I. INTRODUCTION Motor
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
1346 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS,VOL. 28, NO. 6, NOVEMBER / DECEMBER 1992
de
&(k + 1) =
( 6J
1--q k )
6KN
+ -?(k)
H I':[
_ -
- w (3)
dt
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
SEPE AND LANG: REAL-TIME OBSERVER-BASED CONTROL OF A PM SYNCHRONOUS MOTOR 1347
de?otes an estimated quantity, and a(&k), k ) , and cannot be so located. The IP velocity controller is given in
i( O(k),k ) are, respectively, the measured motor voltages discrete time by
and currents transformed into the rotor frame using >he
observer-estimated position. Note that the sym_bol is x,(l + 1) = x , ( l ) + 6,( o * ( l )- w(1)) (11)
used to denote an estimate, whereas the symbol is used
to denote the transfoTmation of a measurement into the
estimated frame of 8. Without direct measurement of i:(l) = F , x , ( l ) - Fp w ( 1 ) +-
KN
sgn ( ~ ( 1 ) ) (12)
rotor position, the stator-frame-to-rotor-frame and rotor-
frame-to-stator-f;ame transformations must use the esti-
mated position 8 in place of the actual position 8. Thus,
the closed-loop controller and observer operate in an
estimated rotor frame. where x , is the integrator state, w* is the velocity com-
mand, 6, is the sampling period of the velocity controller,
I v . CURRENT AND VELOCITY CONTROL 1 is the index of time step a, and F, and Fp are gains.
This section contains a brief presentation of the rotor- Here, 6, = 86,.
frame current controller and of the velocity controller.
Further development can be found in [7] and [SI. V. OBSERVER-BASED
CONTROL
This section integrates the observer with the current
A. Current Control controller and the velocity controller. Adaptation is not
To achieve a fast dynamic response and zero steady-state yet implemented. The fast inner-loop rotor-frame current
error to constant command inputs, the inner-loop vector controller uses the estimated currents i from the observer
current controller is implemented as a proportional- from (4) in place of i in the current control law (8)-(9). In
integral (PI) controller. The discrete-time current con- addition, the predicted velocity from ( 5 ) and predicted
troller is current from (4) are used in (10) to globally linearize the
nonlinear motor system, and the predicted position from
x k ( k + 1) = x i ( k ) + ? $ ( i * ( k ) - i ( k ) ) (8) (6) is used to transform the rotor-frame voltage com-
u(k + 1) = C , x , ( k ) + G , ( i * ( k ) - i ( k ) ) + G,u(k) mands u(k + 1) to stator-frame voltage commands that
can be applied to the inverter. The slower outer-loop
(9) velocity controller requires estimates of velocity i, to
replace w in its control law (11)-(13) [SI.
In the observer-based control system, estimated current
is used as the feedback signal. Note that this is necessary
because the feedback of rotor frame current! is necessary.
+ K N w ( k + 1)
[:'I + u ( k + 1) (10) The two available rotor-frame currents are i and i. Here,
i is chosen because this observer estimate is filtered and
because it is computed as a prediction, which is required
where i" is the current command vector, x i ( k ) is the
integrator state vector, u ( k ) is a delayed plant input, and in (10).
G I , G,, and C , are gains. Note that (8)-(10) globally
linearize (1) by cancelling back emf and that they accom- A. Analysis
modate the delay of their computer-based implementa- Successful operation of the control system requires that
tion; see [SI and [SI. In particular, C, is the gain on the the gains for the velocity controller, the current con-
delay state needed for full-state control. troller, and the observer result in a closed-loop stable
system with the desired performance. The choice of gains
B. Velocity Control is complicated by the fact that the inner-loop current
Because the closed-loop bandwidth of the current controller operates at a sampling rate of 2.44 kHz, whereas
controller is much larger than the desired closed-loop the outer-loop velocity controller operates at a sampling
bandwidth of the velocity controller, the velocity con- rate of 305 Hz, by the nonlinear nature of the motor
troller issues commands to the current controller through modeled by ( 1)-(3), and the severely nonlinear observer
i* and assumes that the desired currents are achieved modeled by (4)-(6).
immediately. Following the procedure detailed in [SI, the system is
Performance goals for the velocity controller are speci- analyzed by first linearizing the inner-loop system consist-
fied in terms of risetime and overshoot. An IP controller ing of the current controller, the motor, and the observer.
is used to meet those objectives. The IP controller does Next, the linearized inner loop is projected onto the
not have a zero in its command-to-output transfer func- slower time scale of the velocity controller and connected
tion as does the PI controller. Thus, the performance of with the velocity controller. Finally, an eigenstructure
the IP controller is completely specified by its pole loca- analysis is performed on the entire system. The gains of
tions, which can be located to give the desired risetime the system are used to place the closed-loop poles in the
without overshoot. Note that the zero in the PI controller desired locations.
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
134s IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 28, NO. 6, NOVEMBER / DECEMBER 1992
I
'0 50 100 1% 200 250 300 350 400
k-samples
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
SEPE AND LANG: REAL-TIME OBSERVER-BASED CONTROL OF A PM SYNCHRONOUS MOTOR 1349
1 direct-am clmalt
I
::
: I t
0
3,
50 100 150 200
k-samples
250
quadrature-am current
300 350 400
I-samples
400 MO
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
1350 IEEE TRANSACTIONS O N INDUSTRY APPLICATIONS, VOL. 2X, NO. 6, NOVEMBER / DECEMBER 1992
VII. CONTROLLER
A N D OBSERVER
REDESIGN A. Analysis
The closed-loop mechanical dynamics are dominated by The success of the observer-based adaptive controller
two complex pole pairs. Participation factor analysis used depends on the ability of the observer to supply the
in [71 and [SI demonstrates that the first pole pair mainly parameter estimator with an accurate velocity estimate
describes the dynamic interaction between the motor ve- even before adaptation has occurred. Otherwise, the pa-
locity and the velocity controller, whereas the second pole rameter estimator will likely give incorrect parameter
pair mainly describes the dynamics of the error between estimates and, at best, poor adaptive controller perfor-
the true and estimated motor velocities. In the adaptive mance. To demonstrate the problem, a simulation is per-
setting, the location of the first pole pair is kept invariant formed. A square wave in velocity from 0 to 100 rad/s is
by adjusting the velocity controller gains according to the commanded to the observer-based controller with param-
estimated parameters. The location of the second pole eter estimation. However, controller and observer re-
pair gives acceptable performance when the estimated design using (22)-(25) are inhibited. The inertia of the
parameters are used to adjust the observer so that there is motor is increased to three times its nominal value,
no parameter mismatch between the observer and the whereas the observer uses the nominal motor inertia. Fig.
motor. Although it may also be advantageous to adjust the 6 shows the result. The estimated velocity behaves as
gains of the observer in the presence of parameter desired, but the actual motor velocity waveform is much
changes, the present lack of an analytical procedure to different. Because the estimated velocity transient is used
determine the gains precludes this possibility. Both con- by the parameter estimator, the estimated parameters are
troller and observer redesign are performed whenever a characteristic of the nominal motor. Thus, the closed-loop
valid set of parameter estimates are received. system does not recognize the fact the motor inertia has
For the I-P velocity controller, the proportional and the changed. Adaptation therefore cannot improve the system
integral gains are computed from performance; it is blind to parameter changes because of
FP = (1 + a PI -
poor velocity information from the observer.
- (22)
To improve the transient tracking ability of the ob-
Ff = (1 - F1 - F2 + Pl F Z ) / ( b % ) (23) server with parameter mismatch, the observer gains G,
are increased one hundredfold. This allows accurate ve-
where p,, = 0.6662 k 0.26521' are the desired closed-loop locity transient tracking for inertial mismatch up to about
pole locations. Finally, (12) must be updated with C via two times the nominal inertia while not compromising the
KN stability of the closed-loop system. As long as the inertia
c = -c.b varies less than a factor of two before adaptation takes
place, the observer redesign can keep the inertial mis-
A detailed explanation of (22)-(23) is given in [7] and [81. match within acceptable bounds, Of course, the price for
As seen in Section V-B an inertial mismatch can lead to increased observer bandwidth is an increase in the noise
closed-loop instability. The observer redesign algorithm content in the state estimates and in the closed-loop
desensitizes the system to inertial change. This requires velocity control system. One possible accommodation is to
the on-line estimation of the continuous-time inertia H have two observers: one for control and one for parame-
for use in (5). For reasonably fast sampling, (20) becomes ter estimation.
KN B. Peqormance
H=-S
b "
The observer-based adaptive controller is implemented
Because the permanent magnets are samarian-cobalt in real time on a computer system based on the 25-MHz
magnets, K is essentially constant. Motorola 68020 microprocessor. The nonadaptive part of
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
SEPE AND LANG: REAL-TIME OBSERVER-BASED CONTROL OF A PM SYNCHRONOUS MOTOR 1351
120 -
100 -
80 -
k-samples
‘U-
20 -
parameter estimares
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.
1352 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 28, NO. 6, NOVEMBER / DECEMBER 1992
velocity range. Finally, the turn-on transient of the ob- Raymond B. Sepe (M’YO) received the B.S. and
M.S. degrees in 1983 and the Ph.D. degree in
server-based controller is sensitive to initial observer er- 1990, all in electrical engineering, from the
ror, and the steady-state currents of the observer-based Massachusetts Institute of Technology (MIT),
controller are sensitive to unknown load torque. Cambridge.
From 1983-1985, he was an electrical engi-
REFERENCES neer for GenRad Corporation, West Concord,
MA. He has conducted postdoctoral research at
M. Ehsani, I. Husain, and A. B. Kulkarni, “Elimination of discrete MIT in the areas of adaptive control, estimation,
position sensor and current sensor in switched reluctance motor and computer architecture, with applications to
drives,” in Proc. U S Ann. Mtg., Oct. 1990, pp. 518-524. ac drives, dc drives, and flexible structures. He is
L. A. Jones and J . H. Lang, “A state observer for the permanent- currently with Electro Standards Lab., Cranston, RI.
magnet synchronous motor,” IEEE Trans. Ind. Electron., vol. 3, pp. Dr. Sepe is a member of Eta Kappa Nu.
374-382, Aug. 1989.
P. C. Krause, Analysis of Electric Machinery. New York:
McGraw-Hill, 1986.
S. Low, “Design of an observer for the permanent magnet syn-
chronous motor,” S. B. thesis, Dept. Elec. Eng. Comput. Sci., Mass.
Inst. Technol., Cambridge, MA, May 1991.
S. R. MacMinn and T. M. Jahns, “Control techniques for improved
high-speed performance of interior PM synchronous motor drives,”
in Proc. IEEE IAS Ann. Mtg., Oct. 1988, pp. 272-280. Jeffrey H. Lang (S’78-M’80) received the B.S.,
S. K. Panda and G. A. J. Amaratunga, “Switched reluctance motor M.S , and Ph.D. degrees in electrical engineer-
drive without direct rotor position sensing,” in Proc. IAS Ann. ing from the Massachusetts Institute of Tech-
Mtg., Oct. 1990, pp. 525-530. nology (MlT), Cambridge, MA, in 1975, 1977,
R. B. Sepe, “Adaptive control of the permanent-magnet syn- and 1979, respectively. From 1977 through 1979,
chronous motor,” Ph.D. thesis, Dept. Elec. Eng. Comput. Sei., he held a Fannie and John Hertz Foundation
Mass. Inst. Technol., Cambridge, MA, Sept. 1990. Fellowship.
R. B. Sepe and J. H. Lang, “Real-time adaptive control of the In 1980, he joined the Department of Electri-
permanent-magnet synchronous motor,” IEEE Trans. Industry A p - cal Engineering and Computer Science at MIT,
plications, vol. 27, pp. 706-714, July/Aug. 1991. where he is now Professor of Electrical Engi-
R. Wu and G. R. Slemon, “A permanent-magnet motor drive neering. His research involves the analysis, de-
without a shaft sensor,” in Proc. IAS Ann. Mtg., Oct. 1990, pp. sign, and control of physical systems, and currently concentrates on
553-558. electromechanical systems with application to traditional electric ma-
P. Young, RecursiLse Estimation and Time-Series Analysis. New chine systems, microsensors and microactuators, and flexible structures.
York: Springer-Verlag, 1984. Dr. Lang is a member of Eta Kappa Nu and Tau Beta Pi.
Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on January 30,2024 at 04:08:33 UTC from IEEE Xplore. Restrictions apply.