Escolar Documentos
Profissional Documentos
Cultura Documentos
P.L.Jansen R.D.Lorenz
Dept. of Electrical and Computer Engineering
University of Wisconsin-Madison
Madison, WI 53706
-
Abstract Rotor flux Observers can provide an complexities and lack of mechanical robustness
attractive means for achieving direct field oriented associated with intrusive sensors located within machine
control of induction machines. This paper air gaps. Furthermore, a correction is required for the
presents a physics-based design methodology and rotor leakage flux if rotor flux field orientation is to be
uses it to evaluate open-loop observers and to achieved. However, a recently demonstrated method of
develop a new closed-loop flux observer. It is measuring the flux angle via the saturation-induced third
shown t h a t the new flux observer is a harmonic in the phase voltages is promising as it
straightforward structure with properties that alleviates the need of intrusive sensors or modifications
combine the best features of known methods. A of the machine.[3]
distinction is made between observers, which use Estimation rather than measurement of the rotor flux is
only integration a n d feedback summation an alternative approach for direct field orientation that
operations, and those estimation methods requiring has received considerable attention.[4-131 As with the
approximate differentiation which are, in essence, third harmonic scheme, only terminal properties
"cancellation" methods. Furthermore, a clear measurement (in some cases including rotor speed) is
distinction is made between accuracy and dynamic required. However, the accuracy associated with such
robustness of the observer. This distinction is estimation schemes is inherently parameter sensitive.
important because the accuracy of flux observers The degree of sensitivity to both machine parameter
for induction machines is inherently parameter estimates and also measured signal noise, e.g.
sensitive. Whereas robustness of observers, in a quantization noise, filter attributes, sample rates, etc., is
controls sense, is not parameter sensitive. dependent upon the particular estimation scheme that is
Moreover, it is shown how flux observers can implemented.
provide robust field oriented control because the Even though observers in general have been around for
flux angle is substantially more correct than the several decades [14-151, a clear and physically-insightful
flux magnitude. A distinctive form of frequency method of developing flux observers has not yet evolved.
response function (FRF) analysis similar to that Furthermore, an insightful method for evaluating the
used in classical control engineering is accuracy implications of parameter errors for applications
demonstrated to be a useful and insightful tool such as rotor flux estimation is still needed. This lack is
even though flux observers a r e multiple-input, perhaps a fundamental problem with the current
multiple-output systems. Finally, the limits of methodology of designing observers based solely upon
such flux observers is experimentally evaluated. Gopinath's minimal order observer theory [14]. This
paper attempts to present in a straightforward manner the
INTRODUCTION accuracy attributes of both open and closed-loop
There are two basic forms of rotor flux field nonlinear rotor flux observers for direct field orientation.
orientation; direct field orientation which relies on direct It attempts to lay a foundation for a physics-based design
measurement or estimation of the rotor flux magnitude and accuracy analysis methodology applicable to both
and angle and indirect field orientation which utilizes an linear and nonlinear observers.
inherent slip relation [ 11. Because indirect field Technically, the word "observer" implies an estimator
orientation is essentially a feedforward scheme it is that employs both inputs to integration process models
naturally parameter sensitive, particularly to the rotor and feedback control for error correction to improve
time constant, and thus has led to the development of estimation accuracy and for defining the error correction
numerous parameter adaption schemes [2]. dynamics. In this paper, such estimators with feedback
The implementation of direct field orientation via air correction controls are denoted "closed-loop Observers",
gap flux measurement has typically been plagued by the while "open-loop observers" refer to estimators that
employ integration process models without feedback Note that 2s)is a stator transient impedance and ab is the
correction. The open-loop observer is then essentially a rotor velocity.
real-time simulation of the physical process, i.e. the Equations (1-4) can be rearranged in terms of the stator
induction machine in this case. As such, one important current and rotor flux in a form similar to a state space
property of the open-loop observer is that it provides model with the current and flux being the state variables.
nearly instantaneous tracking capability. It will be shown Note, however, the system is nonlinear with respect to
how the closed-loop observer can be formed to inherently rotor velocity.
retain this nearly instantaneous tracking capability.
The third form of estimator evaluated in this paper is
based on cancellation methodologies, whereby
approximate differentiation of signals is used to cancel
the effects of process integration. In linear systems, this From (11) and (12), the induction machine electrical
approach is also known as pole/zero cancellation. model can be formulated into a nonlinear block diagram
However, differentiation makes such approaches as shown in Fig. 1 with stator voltage and rotor velocity
susceptible to measurement and quantization noise. treated as system inputs.
Fig. 4. Open loop rotor flux observer based upon the voltage model
utilizing measured stator voltages and currents - (in stationaty reference,
For steady state operation, p may be replaced by the using complex vector notation)
excitation frequency, j w e , and the response function can Since the voltage model is essentially an integrator
be expressed as a function of the slip frequency, cos, without feedback, it is sensitive to offset and drift errors.
where as= oe - w ; Thus a major problem with this open-loop observer is
f A \
that it lacks feedback necessary for convergence. In
practice, a low pass integrator is often used to provide
stability.
The corresponding flux estimation FRF is:
"."
0.001 0.01 ' 0.1 1.0
slip @A.) slip @.u.)
Fig. 3. x q d r / A q d . FRF (15): Current model open-loop rotor flux
observer with machine parameter deviations from nominal r ~l-0,, and
Llro
".,
Voltage Model Open-Loop Flux Observers 0.01 0.1 1 2 3 0.01 0.1 123
Rotor Velocity (P.u.) Rotor Velocity (P.u.)
The voltage model utilizes the measured stator voltages
and currents, but not the rotor velocity. It is commonly Fig. 5. & , . / ; b d ,FRF (1 9): Voltage model open-loop rotor flux obsewcr
at rated slip with machine parameter deviations from nominal rso, LO,
used to implement direct field orientation without and Llro.
measured velocity feedback for low cost drive
applications [2,16]. From (1), the stator flux can be Unlike the current model based observer, the accuracy of
estimated by integration of the following equation, this observer is completely insensitve to rotor resistance
but is most sensitive to stator resistance at low rotor
P x q d s = Vqds - Fs iqds (16)
velocities. At higher rotor velocities, the stator resistance
572
IR drop in (16) is less significant relative to the speed method is readily apparent from the various deviations
voltage (back emf), resulting in reduced sensitivity to shown in Fig. 7. This method 2
also especially
stator resistance. However, the low speed sensitivity is a problematic due to the division by Obr when the rotor
well acknowledged limitation of this observer [2,16]. velocity is near zero. In this range the quantization noise
Because the atmbutes of the two open-loop observers due to numerical differentiationof such signals to obtain
are in many ways complementary, there has been at least rotor velocity typical of modem digital position feedback
one attempt to create a better observer by directly drive systems becomes a dominant form of contamination
merging the two models via selected weighting [4]. The for the estimate.
concept is well merited however a better approach will be After evaluating the parameter dependent accuracy of
found in the form of a closed-loop observer. the various open-loop approaches, it now is appropriate
to evaluate what accuracy improvements can be gained
CancellationMethod Open-Loop Flux Estimators
by forming a closed-loop observer.
It is possible to construct a cancellation method open-
loop flux estimator based on the rotor and stator flux
linkage and voltage loop equations as combined in (1 1) CLOSED-LOOPFLUX OBSERVERS
after substituting the estimated machine parameters and A closed-loop rotor flux observer from the current
measured terminal properties as follows. model open-loop observer in Fig. 2 can be developed
directly from the induction machine model of Fig. 1.
Measuring the stator voltage allows the stator current to
be estimated and compared to the measured current as
Noting the requirement for stator current differentiation,
-
shown in Fig. 8. Because the rotor flux acts as a form of
this cancellation method estimator would then appear as
physical state feedback on the stator current, the current
in Fig. 6.
error can be driven to zero through the use of corrective
feedback acting on the estimated rotor flux. Ideally the
rotor flux estimate will converge on the correct value
when the current error goes to zero.
-" &r
':
-:m
a -
Fig. 6. Open loop rotor flux estimator based upon the cancellation
method utilizing measured stator voltage and current and rotor velocity.
(in stationary reference, using complex vector notation)
:
The estimation FRF for the open loop cancellation
method is obtained by solving ( l l ) , (12), and (20)
simultaneously to achieve
k
The accuracy of this method depicted in the FRF plot of Fig. 8. A closed-loop,Gopinath style rotor flux observer for an
induction machine based upon the current model and utilizing corrective
Fig. 7 for rated slip operation of the machine. current e ~ feedback
w
1.5 Although topologically similar, controllers used for
1.4 closed-loop observers differ from physical system
0 1.3 . . . ...... . controllers. One important aspect is that no physical
B 1.2
.$
,
Fig. 9. Closed-loop, Gopinath style flux observer after block diagram Nonlinear, Velocity Invariant Observer Controllers
algebraic manipulation to show the "implicit"flux estimation reference, The nonlinear, velocity varying dynamics of the flux
&* ,as produced by an open loop cancellation method estimator observer can be made velocity invariant by different
This analysis shows that the implicit flux reference, observer controller design approaches. The classical
A
hqdr*, for this general Gopinath style flux observer is approach taken by prior researchers has been to form
574
velocity dependent gain terms via algebraic eigenvalue Insight on the velocity invariant observer is enhanced by
solution techniques[8]. In this section an alternative forming the feedforward terms to avoid differentiation,
approach is presented based on techniques which allow resulting in an explicit voltage model shown in Fig. 13.
for substantial insight in how nonlinear physical system
dynamics may be altered by controls design to achieve
invariant closed loop dynamics [17,18].
Fig. 11 introduces such a nonlinear observer controller
alternative which considerably reduces the tuning
complexity for the Gopinath style flux observer.
._-_----
Fig. 13. Closed-loop Gopinath flux observer with speed invariant err01
correction dywnics--showing both cancellationmodel flux reference and
voltage models as the equivalent observer model basis.
This observer formulation may be further simplified by
realizing that the first two terms after the flux summing
junction am not needed to retain the accuracy of the
I model. Removing the low pass term will allow a simpler
controller configuration. Furthermore, the implicit flux
Fig. 11. Closed-loop Gopinah style flux observerwith velocity invariant
dynamics due to the nonlinear controller for the nonlinear flux model. reference requirement can also be met by replacing the
cancellation model flux estimator with the more desirable
This controller achieves velocity invariant (fixed) current model open-loop flux observer. The final
dynamics by using a combination of three nonlinear configuration is shown in Fig. 14.
&s$nAtechniques: nonlinear state feedback d e ~ ling
u
of a&&,nonlinear command feedforward of -%!itjr,
and nonlinear manipulated input decoupling of I /@br,
along with the linear state feedback controller, K
[ 17,181.
In configuring the controller, the feedforward
estimation model is set to retain the estimation properties
of the current model open-loop observer. Thus, the
nonlinear state feedback decoupling term requires that a
corresponding term be added to the command
feedforward controller. The manipulated input Fig. 14. Closed-loop Gopinath style flux observer with speed invariant
error comcZion dynamics--showing implementation with simplified
decoupling is set to remove the velocity varying nature of controller and current model input acting as implicit refexence
the forward loop elements in the closed-loop observer The resulting closed-loop flux observer is seen to provide
model. an automatic transition between the two most desirable
Because this entire observer model is internal to the open-loop flux observer models: the current model at low
control processor, it is possible to algebraically simplify frequencies to the voltage model at high frequencies.
the observer implementation. Fig. 12 shows the result This transition is now determined via the bandwidth of
after simplification of feedback and feedforward terms. the flux loop which is velocity invariant and easy to tune.
%I The FRF of flux estimation for this observer has the
1
1
following form:
'
' A
Lr
:- - J
- A I I which is shown graphically in Fig. 15.
G, X@ I II
575
10 . . . . . .
hsr
mlm
5 @xu.)
h
0
.......\ L ... .I $ 0
E
-1
4 -; . ............
@xu.) 1 .............
.:. ...........
0.6
-1s
-20
~.
0..
1 ...
.
_:.
.
.
.............
. 1:. :;:: __i
.
. . _ _ ..
I
& coodinale
Flux Angle
Calculator T d o m A t . . . . . .
Rotor Flux
Observer Velocity
Obsclvn
Fig. 16. Direct field oriented controller based upon either open or ... .:. ........ . . .
closed-loop rotor flux observers.
0.2 . . . .. . . . ... . . . . . . . .
The direct field oriented controller results for the three ., .
.
forms of rotor flux estimation are shown in Fig's 17-19. 0 100 200 300 4000 100 200 300 400
Tme (msec) Time (msec)
Fig. 19. Rotor flux, torque, and velocity waveforms of tuned and
detuned (kfi is l x rro ,right is 2x rro) direct field oriented drives based
upon the closed-loop, velocity invariant rotor flux observer with current
model input and 20 Hz eigenvalues.
The results generally substantiate the expected behavior. REFERENCES
The current model is overall the best estimator in that the [l] Novotny, D.W. and Loren& R.D., (eds.), Introduction to Field
least dynamic distortion occurs. The voltage model is the Orientation and High Pelfomnce AC Drives, tutorial book from
1985& 1986 EEE-IAS Annual Meetings.
next best estimator in that distortion occurs mostly at low
[2] Hung, K.T. and Lorenz.R.D.. "A Rotor Flux Error-Based Adaptive
speeds. It is also very sensitive to the quality and Tuning Approach for Feedforward Field Oriented Machine Drives",
bandwidth of the low pass integrator. Proc. IEEE-IAS Annual Meeting, October 1990.
The closed-loop, velocity invariant observer with p ] Moreire, J.C. and Lipo. T.A., "Direct Field Orientation Control
Using Third Harmonic Component of the Stator Voltage",
current model input provides as good a dynamic result as International Conference on Electric Machines (ICEM), Boston,
the current model observer at velocities well below the MA, August 1990.
bandwidth of the observer. At higher velocities the [4] Vagati, A. and Villata. F., "A.C. Servo System for Position
Control". Proc. ICEM, Lausanne,Switzerland, 1984, pp. 871-874.
observer moves into the transition region set by the
[5] Hori, Y., Cotter, V., and Kaya, Y.,"A Novel Induction Machine
eigenvalues and the parameter sensitivity varies as Flux Observer and Its Application to a High Performance AC Drive
predicted by the FRF (22). System". IFAC 10th Triennial World Congress, Munich,FRG, 1987,
pp. 363-368.
[6] Verghcse. G.C. and Sanders, S.R., "Observers for Flux Estimation
UIInduction Machine", IEEE Trans. Industrial Electronics, Vol. 35,
CONCLUSIONS No. 1, F c b ~ p r y1988, pp. 85-94.
This paper has attempted to contribute the following: D] Kubota, H. and Matsuse, K., "Flux Observer of Induction Motor
A clear, physically meaningful accuracy analysis of with Parameter Adaption for Wide Speed Range Motor Drives",
Proc. IPEC-Tokyo, 1990. pp. 1213-1218.
flux observers [SI Hori, Y. and Urneno, T., "Flux Observer Based Field Orientation
A methodology for understanding the source of (FOFO) Controller for High Pexfomance Torque Control", Proc.
estimation errors in Gopinath style observers IPEC-Tokyo, 1990, pp. 1219-1226.
[9] Franceschini, G.. Tassoni, C., and Vagati. A., "Flux Estimation for
Development of a superior design basis for combining Induction Servo-Motors", Proc. IPEC-Tokyo, 1990, pp. 1227-1234.
the current and voltage models in a closed-loop [lo] Sangwongwankh, S.,Yonemoto. T., Furuhashi, T., and Okuma, S.,
observer with velocity invariant dynamics "Design of Sliding Observer for Robust Estimation of Rotor Flux of
Implementation and experimental evaluation of the Induction Motors", Proc. IPEC-Tokyo, 1990, pp. 1235-1242.
[ll] Franceschini. G., Pastorelli, M.. Profurno, F., Tassoni, C., and
flux observer alternatives Vagati, A., "About the Gain Choice of Flux Observer in Induction
The primary conclusions on the observer topologies are: Servomotors",Proc. IEEE-IAS Annual Meeting, 1990, pp. 601-606.
The current and voltage model open loop flux observers [12] Kubota, H.. Matsuse, K.. and Nakano, T.. "DSP-Based Speed
Adaptive Flux Observer of Induction Motor", Proc. IEEE-IAS
and the closed-loop, velocity invariant flux observer all Atvmal Meeting, 1991. pp. 380-384.
provide flux angular position estimate errors which are [13] Bb'cker, I. and Jatming, J., "Discrete.-Time Flux Observer for P W M
small. The net differences in torque characteristics are Fed Induction Motors". Proc. EPE, Firenze. 1991, pp. 171-176.
thus small, except at low frequencies (low speeds) [14] Gopinath. B.. "On the Control of Linear Multiple Input-Output
where the current model and the velocity invariant flux Systems", The Bell System Technical Journal, Vol. 50, No. 3,
March, 1971, pp. 1063-1081.
observers are superior to the voltage model observer. [15] Luenberger. D.G.. "An Introduction to Observers", IEEE Tram. OR
The cancellation method open-loop estimation method Automc3tic Control, Vol. AC-16. No. 6. Dec. 1971. pp. 596-602.
is the least desirable due to its inherent susceptibility to [16] Xu, X.. De Doncker, R.. and Novotny, D.W., "A Stator Flux
Oriented Induction Motor Drive". Proc. PESC, Kyoto, April 1988.
low speed noise.
[17] brenz. R.D. and Lawson, D.B.. "Performance of Feedforward,
The closed-loop, velocity invariant, flux observer has Current Regulators for Field Oriented Induction Machine
the desirable low speed attributes of the current model, Control1ers"JEEE Trans. Ind. Appl.. Vol 1A-23, No. 2,
and the desirable high speed attributes of the voltage July/August, 1987, pp 597-602.
[18] Lorem, R.D. and Novotny, D.W., "A Contmls Systems Perspective
model. The bandwidth of the observer establishes the of Field Oriented Control for AC Servo Drives",Proc. Controls
frequency (speed range) of the transition region. Such Engr. Conf., June 6-11, 1988. pp XVIII-1 - XVIII-11.
deterministic properties make this observer ideally [19] Lorenz, R.D. and Van Patten, K., "High Resolution Velocity
Estimation for All Digital, AC Servo Drives"JEEE Trans. Ind.
suited for wide speed range applications requiring both Appl.. V0127, NO. 4, July/August, 1991, pp 701-708.
zero speed and field weakening operation.
APPENDIX
ACKNOWLEDGEMENT Induction Machine Parameters
Westinghouse TEE II, Frame 215T, 10 Hp, 3 phase, 460/230v,
The authors wish to acknowledge the financial support 12.2/24.4A, 1750 rpm, rated 0s = 0.028 p.u.
and motivation provided by the Wisconsin Electric rs = 0.20 Q LIS = 1.5 mH & = 32.3 mH
rr = 0.20 Q Lb = 1.5 mH J = 0.045 kg-mz
Machines and Power Electronics Consortium (WEMPEC)
of the University of Wisconsin-Madison.
577