Speed Sensorless Field Orientation Control

of the Induction Machine

Hirokazu T-4JIMA and k'oichi HORI
Department of Electrical Engineering
The University of Tokyo
7-3-1 Hongo, Bunkyo-ku, 113 Tokyo, Japan
TEL +81-3-3812-2111. FAX +81-3-5684-3986
E-mail tajiniaG kaya. t .U-t okyo.ac. j p

Abstract is instantaneously used for the Flux Observer based

Field Orientation (FOFO) controller which we pro-
We propose a speed estimation method for an induc-
posed some years ago [3]. Through the implementa-
tion machine and its application to the Flux Observer
based Field Orientation (FOFO) control system, which tion using DSP (Digital Signal Processor), and some
we proposed previously. The motor speed is estimated laboratory experiments. we will show that the speed
based on the difference between two flux estimators. sensorless FOFO operates with enough stability and
We analyze its convergence performance and propose
the novel pole assignment method. Next, we apply this
has strong robustness to changes in machine param-
speed estimator to the FOFO controller. We imple- eters such as the rotor resistance.
ment the system centered about the DSP and through
some laboratory experiments we will show that the
speed sensorless FOFO operates with enough stability
and has a strong robustness to rotor resistance varia-
tion.
2 Model of Induction Machine
The state equation of the induction machines are
1 Introduction given by

We know two major techniques for high performance

control of induction machines:
.
x= [ A21
A1z
A22
] [ ]
5+ U,
1. Slip frequency controlled type vector control, and
2 . Field orientation control.
where
Recently the elimination of speed sensor has been one
of the important requirenients in vector control sys- 5 = [is. A,]"
tems, because the speed sensor spoils the ruggedliess
and simplicity of ac motors. We can see some ex-
cellent speed sensorless approaches (for example in
[2] ). but they are applicable only to slip frequency
controlled type systems.
In this paper, we mill propose the speed sensorless
control method applicable to the field orientation COII-
trol. It is well known that the field orientation control
is inherently more insensitive to rotor resistance vari-
ation than the slip frequency coiltrolled system [3].
Howevei, since the field orientation controller has no
concept of frequency. we should estimate the rotor
speed from the instantaneous values of stator volt-
ages and currents. Thr state vaiiables are the ststor current z, =
First, we refer the speed cstiniatioii nicthod pro- ,z( i 6 b ) T and the rotor flux A, = (Ar,. A,+)?'. The
posed by Schander [l]based on the idea of 52RAS input variable is the stator voltage w, = (?isar ziSt,)'.
(Model Reference Adaptive System) and propose its Note that all variables are haidled on the stator CO-
novel pol^ allocation method. The estimated speed ordillate system.

@'/803-0453-S/91$1 .WO1991IEEE 385

 I

y; Voltage
Model
~

1 I
Fig. 2: Speed Estimator Dynamics

h-p= (2(iuc - 1 / ~ , ./1A,I2

)
h-I = W',2/IA,12 (7)
I
From (7) the transfer function from Aw, to A;,. is
Fig. 1: Configuration of the Speed Estimator
A& (2<dC- l/T,.)S + d;
-- - (8)
3 Speed Estimator Ad,. s2 + P(ld,s + w,"
We can see that the speed estimation dynamics are
3.1 Configuration of Speed Estimator characterized by one zero and two poles given by
From (l),we can obtain two well known flux simula-
zero: s, = - 4
tors. One is called the voltage model and the other l/rr
2<W, - (9)
the current model. poles: sp = -(wc f jwJ1-T"
Voltage model:
= L, 4 Flux Observer
A,, = (vs - R,i, - uLSas) (2)
Current model: We designed another flux observer to be used for
the field orientation control by using Gopinath's re-
(3) duced order observer theory, which we proposed pre-
viously [3].
The voltage model of (2) does not involve rotor The observer equation takes the form of
speed W , while the current model of (3) does. We can
then estimate the speed based on the output differ- A,. = A22 A,. + Ail is
ence between these two models. +G(H, - A12Ar - Ailis - B i v , )
2, = (KP+?)€ (4)
= -HA, + (A21 - GAll)i,
- G B l v , + GZ, (10)
where
~~ A A In this equation, the eigenvalues of
AriaArvb - ArvaArib (5)
- H = A22 - GA12 (11)
Fig. 1 illustrates this idea of speed estimation.
Through the linearization with respect to a certain are the observer poles which can be specified freely
operating point, we can obtain the transfer function by the observer gain G.
from the estimation error Aw, -A;, to A€ as follows Fig. 3 depicts configuration of Flux Observer.
111. We specified the observer poles -a j,B as
-N = -CkI+$J
a = Ji/.p2 +r*(i/rp2++:). (12)
3 = -4,

Fig. 2 depicts the whole block diagram for the pro-

posed speed estimation.

3.2 Pole Assignment of the Speed Es-

t imat or
Assuming that cu', = 0 for simplicity, we can specify
the damping factor ( and natural angular frequency
dc by using h-p and KI in (4) as follows. Fig. 3: Configuration of Flux Observer
 Fig. 4: Pole Allocation of the Flux Observer

1’1

- 1
& (estimated speed)
(stator current)

Fig. 5: Setup of the Experimental System centered

about DSP

This pole allocation keeps a kind of sensitivity

norm to be constant regardless of the motor speed.
This was the key of our success [3]. Fig. 4 shows that
the observer poles move according t o the motor speed
U,. r is the parameter in our pole allocation. In the
following experiments, we used r = 1.0.
vs

iS
L-

- s
Fig. 7: Voltage Model

Volt age
Model

+ 1/T
-+ s 1/T

Current,

Fig. 9: Speed Estimator (Experiment)

Fig. 6 shows the block diagram of the speed-

Xr.,

sensorless field orientation control system. The speed

estimator calculates the speed L, from the stator
voltage and current components, and this estimated
speed is immediately used for the flux observer and
the speed controller.
The DSP performs the calculations required for
the speed estimator and the flux observer (inside the
dashed line box in Fig. 6). The program size is about
900 steps and Its control period is about 135 [ps].
I

5 Experimental Result 5.2 Some Modification in Actual

Speed Estimator
5.1 Experimental setup
In practice the original voltage model of (2) is difficult
Fig. 5 shows the DSP based implementation of the to implement because it requires a pure integrator,
speed sensorless FOFO controller. The DSP used was which has the initial value and drift problems. To
pPD77230 made by NEC, which can perform 32bit avoid these problems, we replaced the pure integrator
floating point mult,iplication. Its clock frequency is by the low pass filter. This modified voltage model
13.333 [MHz] and machine cycle for every calculation has the form of
is 150 [ns]. . I
1-1 L
The tested machine is a 4-pole 2.2 [kW] induction = ---ATu
T
+2
M
(w. - R,i, - aL,;,) . (13)
machine which has the following parameters.
The voltage model and the modified voltage model
R, = 0.8770 R, = 1.47R are illustrated in Fig. 7 and Fig. 8, respectively.
L , = 165.142mH L , = 165.142mH Since the output of the modified voltage model
M = 160.8mH a = 0.0519 of (13) is different from the rotaor flux. we have to

387
 I

I DSP
U,;"
I I
J 21ib I

VR : Vector Rotator IM : Induction Machine

Fig. 6: Speed-Sensorless Field Orientation Control Systeiii

modify the current model simultaneously. The mod-

ified voltage model in Fig. 8 is equivalent to one
that consists of a voltage model and a high-pass filter
(HPF). Hence the modification of the current model 150
is achieved by inserting H P F (see Fig. 9). T is a time Speed
constant of HPF. In experiment we use T = 0.05, [r .p. m.]
then its cut off frequency is about 3.2 Hz. 0

5.3 Experimental Result of Speed Es-

timation I 200ms I

Fig. 10 (a) and (b) show the speed estimation per- ( a ) Experiinental Result
formances. (a) is a experimental result of Fig. 9,
and (b) is calculation result by linear model of (8).
speedlr.p.m.1
In case of = 1.0, wc = 100, PI parameters are
then K p = 399, h-1 = 20408. However when we
use these PI parameters, the estimated speed became
very noisy. Finally we used K p = 100. It is equiva-
lent to the case of = 0.29 and w, = 100.
The agreement of Fig. 10 (a) and (b) is not so good,
because the linear model (8) ignores H P F and the
existence of slip. We can say that the speed can be
estimated effectively with the specified convergence
characteristics by and U,.
0.00 I I
1 time[msl
0.00 200.00 400.00
( b ) Calculation Result

Fig. 10: Dynamic Response of Speed Estimator

(< = 0.29. ~3~ = 100)

388

..
5.4 Experimental Results of Flux Es- 0.5Wrb/d
timation
T h e estimated speed is used in the Hux observer as an
important paranieter of matrices in (10). Figs. l l ( a )
and ( b ) show the flux estimation performances of thp
Hux observers using the measured speed w', and the
estimated speed 2,. respectively.

5.5 Robustness of Torque Control to

the Machine Parameter Variation 'div
Figure 13 shows the stationary torque errors caused
by the variation of rotor resistance R , . The tested in-
0.5\." b/d
duction machine is connected t o a de generator and
the induction machine commands the generation of
constant torque. The rotation is maintained a t a
constant speed by the load dc machine (see Fig. 12).
We can see that the proposed field-orientation coil-
troller using thc estimated speed (it is noted "Speed-
sensorless field orientation type ") is completely in-
sensitive t o R, variation.
This can be f>xplained as follows. The flux ob-
server (10) has & skew-synirnetric structure, so we can
.5Wb/div
interpret its diagonal parts as the real paits and the
(b) Using Estimated Speed 2,
skew symmetric. parts as imaginary parts of the coni-
plex nuniber as follows
Fig. 11: Flux Observation

Torque Control Reference Speed Control Reference

where A , is tlic real flux and Ay
is thc observed flux IT* or*
in steady state. respectively.
The steady state error of the estimated flux can be Extemal
Resistors
calculated using ( l o ) , (14). When the rotor resistance
varies and we use the estimated speed for the Hux Torque Control Speed Control
observer. it takes the form of
Fig. 12: Setup of the hIeasuring Systeni of the Sta-
A , - A,- - tionary Torque Errors t o Changes in the Rotor Re-
-_
A1'
sistance

a - JW,.
CI + jujs (l/T,
-I
- jw,)
( ,;
us- - A+
Torque current(A)
50
where
40

20
R: is the actual rotor resistance. Aw,.the estimation
error of the speed can be calculated using (3). It is 10

given by
0 50 100 150 200 250

Rotor resistance variation @)

Fig. 13: Stationary Torqur Errors Caused by R,

Variation ( u j , = 7 5 r.p.ni.)

389
 current to increase. Thus in turn causes a propor-
I Im
tional increase in slip frequency and also source fre-
quency causing flux simulator output to improve.

6 Conclusion
We proposed the speed sensorless FOFO (Flux Ob-
server based Field Orientation) controller for the in-
duction machine. The motor speed is estimated based
on the difference between the outputs of two flux sim-
ulators. The torque generated by this system is ex-
Fig. 14: Vector diagram of Flux and Current tremely robust to rotor resistance variations. We im-
plemented the system using a DSP and showed its
efficacy through laboratory experiments.
Fig. 14 depicts the vector diagram of flux and cur-
rent. Under the field orientation control, the station-
ary torque error is given by Acknowledgment
I would like to express my best acknowledgment to
Professor Yoichi Kaya for his guidance, all kinds of
where T* is torque reference, T is actual torque, re- wise advice.
spectively.
Eqs.(l5) and (17) tell us that the flux estimation
error is zero. This means that the angle 4 in Fig. 14
References
is zero. Then from (18) we can notice that the sta- C. Schauder : Adaptive Speed Identification For
tionary torque is insensitive t o R, variation. Vector Control Of Induction Motors Without
Rotational Transducers, IAS '89, pp 493-499
5.6 Speed Control Response
T. Ohtani, N. Takada ,K.Tanaka : Vector Con-
Fig. 15 and Fig. 16 show the speed control responses. trol of Induction Motor Without Shaft Encoder,
(a) is of the slip frequency controller using the mea- IAS '89, pp 500-507
sured speed w,, and (b) is of the speed sensorless
FOFO controller using the estimated speed G,. The Y. Hori, T. Umeno : Implementation of Robust
waveforms are speed reference (U;), actual speed Flux Observer Based Field Orientation Con-
(w,),estimated speed (G,.) and flux simulator outputs troller for Induction Machines, IAS '89, pp 523-
(A,,,, A r i a ) , respectively. Speed reference was 0.5Hz 528

-150 -
rectangular waveform of 0 150 [r.p.m.] (Fig. 15)and
N

150 [r.p.m.](Fig. 16).

The estimated speed in Fig. 15 (a) and Fig. 16 (a)
I. Miyashita, Y. Ohmori : Speed Sensorless High-
speed Torque and Speed Control Based on In-
stantaneous Spatial Vector Theory, IPEC- Tokyo
are calculated but not used for speed control. The '90, pp 1144-1151
estimated speed is in good agreement with the actual
speed in the higher speeds, but in the lower speed re- X. Xu, D. Mi. Xovotny : Implementation of Di-
gion the estimated speed has a considerable transient rect Stator Flux Orientation Control on a Ver-
error. It is caused by H P F used in the speed esti- satile DSP Based System, IAS'90, pp 404-409
mator (see Fig. 9). Due to the presence of HPF, the
flux simulator output signal a t low frequency is highly M. Vklez-Reyes, E;. Minami, G. C. Verghese :
attenuated and so speed estimation performance is Recursive Speed and Parameter Estimation For
poor. Induction Machines, IEEE IAS '89, pp 607-611
In contrast, in Fig. 15 (b) and Fig. 16 (b), the esti- U. Baader, M. Depenbrock, G. Gierse: Direct
mated speed is immediately used for speed controller Self Control of Inverter-Fed Induction Machine,
feedback signal and the flux observer. The estimated A Basis For Speed Control Without Speed-
speed corresponds with the actual speed even a t lower Measurement, IEEE IAS '89, pp 486-492
speed range and the speed itself can be controlled
fairly well. Estimated speed feedback causes torque
Speed
Reference
1r.p.m.l O
1 11 w: T-L

Estimated 150
Speed
1r.p.m.1 O

15
Torque
Current 0
[AI
-15

(a) Using Measured Speed ( b ) Using Estimated Speed

Fig. 15: Speed Control Response (0 t-1 150 [r.p.m.])

Reference
Ir.p.m.1
0
i ---I w;

1 ,/-kr ,/I
-150 [r.p.m.l

Actual
Speed
15:
1 Actual

Ir.p.m.1
-150 [r.p.m.l
-150
I
Estimated 150
Estimated
Speed
[r.p.m.l

Torque
Current

[AI
0 ] x z Gr

IT
Speed
[r.p.m.l

Torque
Current
0

-150
l5
0
-15 __ ~ --
-15

(a) Using Measured Speed (11) Using Estimated Speed

Fig. 16: Speed Control Response (-150 - 150 [r.p.m.])

391