A SIMPLE ADAPTIVE SCHEME FOR INDIRECT FIELD ORIENTATION OF

AN INDUCTION MOTOR
P. J. Costa Branco

Abstract

A self-adaptive scheme is described for automatically tracking the temperature and saturation-

dependent rotor time constant (Tr ) in indirect vector control. This method is based on a monitoring

signal injected into the process. A cross-correlation function between the monitoring signal and the

speed error indicates the Tr changes in direction (increase or decrease) and magnitude. The

proposed identification algorithm presents unique characteristics: it doesn't use current or voltage

sensors; it is not motor parameter dependent; the amplitude of the monitoring signal can be

reduced but with an increase on the speed error sensibility to control interference in drive's

performance and it works either on steady- and dynamic transient-state. It will need some load to

operate but this condition is satisfied in all practical cases. Tests were conducted with a 1,2kW

four-quadrant AC-drive. The adaptive indirect vector-control was implemented in a microcomputer

with the motor supplied by a voltage-source inverter with current control using IGBTs.

Experimental and simulated results using this adaptive digital scheme for different loads and values

of model Tr (representing a detuned operation) are presented showing good performances.

1 Introduction

Field-oriented control, or vector control, gives good performance when applied to induction

motors in variable-frequency adjustable speed drives. This technique maps the nonlinear-

multivariable control AC-drive system into the DC-drive decoupling model [1]. The objective is to

decouple the stator current into flux and torque producing components, regulated separately, to

obtain a very simple AC-machine control. For this purpose, all space-phasors are referred to a

system whose real axis coincides with the rotor flux space-phasor. Without regard to the type of

inverter, field-oriented principle can be applied to VSI (voltage-source inverter), CSI (current-

source inverter), pulsewidth modulated (PWM) voltage-source inverters, as well as cycloconverters.

The recent development of high-power bipolar transistors, with inherent turn-off as well as turn-on

1
capabilities, has made the so called current-regulated PWM voltage-source inverter an attractive

choice in fast response AC-servo applications [3]. This approach employs a fast-switching to

control the current supplied to the motor on an almost instantaneous basis with a local current

feedback loop.

Vector control schemes are classified as direct and indirect ones. In the direct method, the

magnitude and angle of the flux linkage are measured or estimated with a flux observer. In the

indirect vector control, a feedforward path determines the rotor flux position. In practice, indirect

vector control is the most popular because of its relative simplicity and cost of implementation

compared to the direct method.

The complex-structure of AC-drives has made it difficult in the past to control them. This called

for a complex control scheme with many nonlinear electronic components, function generators, etc.

which are difficult and expensive to tune and service. With the evolution of microprocessor

technology, software instead of hardware is applied, so that tasks of almost unlimited complexity

can be handled rapidly and at a reasonable cost. In this paper, an induction motor is controlled by a

microcomputer employing the indirect vector control. The motor is supplied by a PWM voltage-fed

source inverter (VSI) with current control using as power switches the IGBTs.

Despite the indirect vector control simplicity, its accuracy depends highly on the value of the

rotor time constant used to obtain the instantaneous position of the rotor flux. If this position

deviates from the correct value (due mainly to changes in the rotor resistance), decoupling of flux

and torque is lost and both transient and steady-state responses are degraded. Low-frequency

response, loss of input-output torque linearity and speed oscillations are the principal consequences

of this detuned operation [5]. Therefore, parameter adaptation schemes are important in motor

drives in minimizing these effects. For this reason, an adaptive scheme composed by an on-line

identification and a compensation for Tr changes is implemented for an indirect field oriented

control. The Tr identification is determined by adding a low-level characteristic signal in a form of a

Pseudo-Random-Binary-Signal (PRBS) without use of additional current or voltage sensors. It is not

motor parameter dependent, the amplitude of the additional signal can be reduced with consequent

increase in speed error sensibility and so the interference in drive's performance can be controlled, it

works either on steady- and transient-state and it needs some load to operate although this condition

is satisfied in all practical cases. The correlation-function between monitoring signal and the speed

2
error indicates the direction of changes in the rotor time constant, presenting a value proportional to

its magnitude. This signal is then used for Tr compensation. The presented technique has already

been reported in literature for a direct field-oriented control [3]. In this work, implementation of that

technique in indirect vector control is proposed and discussed with new aspects commented. The

main advantage of proposed implemented adaptive scheme is that it merges two simple techniques

(indirect vector control and Tr identification) without use of additional sensors, while adding only a

little more software overhead. Simulated and experimental results show then the good performance

of proposed combination.

2 Control of Induction Motors in Field Coordinates

Neglecting saturation and iron losses and with the usual simplifications such as smooth stator

and rotor surfaces (no slots), symmetrical three-phase windings with sinusoidal distribution, a

cylindrical two-pole motor can be described by a set of four nonlinear differential equations [3]:


R s i s + Ls

dis
+ Lo

d ir e jε(  )
= u s (t ), (1)
dt dt


R r i r + Lr

d ir
+ Lo

(
d i s e jε  )
= u r (t ) , (2)
dt dt

J
dω
dt
2   *
(
= md − ml =   Lo Im i s ir e jε  − ml ,
3  
) (3)

dε
= ω. (4)
dt

The instantaneous phase currents and voltages form complex time-varying vectors in the plane

perpendicular to the motor axis:


is (t ) = is1 (t ) + is2 (t )e jγ + is3 (t )e j 2 γ , γ =
2
(5)
3


u s (t ) = u s1 (t ) + u s2 (t )e jγ + u s3 (t )e j 2 γ . (6)

3
 For the rotor currents and voltages corresponding definitions are applied. The model equations

are valid for any waveform of currents and voltages if

is1 (t ) + is2 (t ) + is3 (t ) = 0 . (7)

Rotor flux may be characterized by an equivalent stator-based magnetizing current containing a

component for magnetic leakage,

  
imr ( t ) = i s + (1 + σr )ir e jε = i mr e jρ . (8)

Rewriting the differential equations with this reference and assuming imposed stator currents

results in following equations:

dimr r
Tr + imr = Re ise jρ = isd , (9)
dt

 1    i sq 
dρ
dt

= ωmr = ω + 
(
 r mr 
T i )
 i
Im s e jρ
= [
ω 
+
(
]
 r mr 
T i )
 , (10)

 2  Lo 
md ( t ) = ki mr i sq , k =   ,
 3  (1 + σr ) 
(11)

representing a flux model for motor control with the stator currents and speed serving as input

signals. The new field-coordinates for motor control are presented in Fig. 1. Equations (9) and (11)

are identical to the ones that describe a DC-motor. Therefore, the well-known methods used to

control this kind of machine can be applied to control induction motors. The difficulty is just to

obtain the rotor flux position. In indirect vector control, eq. (10) is used to solve this problem. The

basic scheme for indirect vector control are shown in Fig. 2 showing the observer for slip frequency.

4
 Im ωs is
i sq Re
i mr
Field-oriented reference
ξ
i sd ω
ρ
δ
Rotor reference
ε

Stator reference

Fig. 1. Field-coordinates for motor control.

PI isq*
ω ref ∆ω
+ + F1 = ( isq* )2 + ( isd* )2 |I s |
- ARW

Tan -1 ( isq* / isd* ) δ +

F2 = ξ
+
ρ
ω
isq* ωs +
F5 = dt
Tr isd*
+

P
ARW: Anti-Reset Wind-up
ω

Fig. 2. Scheme of the indirect vector control.

2 Implementation of the Digital Control Scheme

The digital control scheme implemented can be seen in Fig. 3 in form of a block diagram. The

system is composed by:

- a microcomputer;

- an interface card with A/D and D/A converters programmed by software;

5
- a power block, presenting an AC three-phase voltage-source, a diode bridge-rectifier, a DC-filter,

and a bridge-inverter with current control using IGBTs;

- an optical absolute position sensor with 10-bit resolution in Gray code. These signals are

transfered to the microcomputer by a printer port to be computed the motor speed and position;

- a four-pole, 1.2kW induction motor;

The control algorithm is executed with 1ms sampling time. During this period, the sampled

speed signal is compared with the speed reference, the adaptive control algorithm is executed and,

finally, the microcomputer provides the current reference signals for the inverter. The adaptive

control algorithm is implemented in PASCAL and the computations are done in fixed-point

arithmetic due time execution constraints.

DIGITAL SPEED
CONTROL
AC
Voltage-Source

PC/XT

PORT D/A

DC - Filter

CURRENT
CONTROL

−∆

IM
PS

Fig. 3. Block diagram of the digital control scheme

Results obtained with implemented digital control are presented. Test results were done with

200V in the DC-link voltage and 10kHz of maximal inverter frequency. The waveforms of speed

and current for an inversion from -900rpm to +900rpm are presented in Fig. 4. An acceleration is

presented in Fig. 5 and a deceleration in Fig. 6. The speed signals present a constant torque and a

6
time symmetry, presenting 65ms for acceleration and deceleration and for speed reverse the time of

132ms.

Speed (rpm)

+900
0
-900
Current (A)
+3.25
0
-3.25

0 Time (ms) 512

Fig. 4. Speed inversion from -900rpm to +900rpm

Speed (rpm)

+900
0
-900
Current (A)
+3.25
0
-3.25

0 Time (ms) 512

Fig. 5. Speed acceleration from 0rpm to +900rpm

7
Speed (rpm)

+900
0
-900
Current (A)
+3.25
0
-3.25

0 Time (ms) 512

Fig. 6. Speed deceleration from -900rpm to 0rpm

3 Tr Identification Algorithm

The absence of flux sensors and the easy of operation at low speed compared to the direct vector

control scheme has increased the popularity of the indirect vector control. However, in indirect

vector control is important to know the rotor time constant to determine the instantaneous position

of the rotor flux. If this position is not correct, flux and torque decoupling are lost and both transient

and steady-state responses are degraded. Low-frequency response, speed oscillations and loss of
input-output torque linearity are the principal consequences of this detuned operation, decreasing

the efficiency of the motor drive.

To overcome the detuned operation, different parameter adaptation schemes have been proposed

in literature. Some are based on an error function controlling the Tr compensation. This error

function can be generated using vector product of measured and calculated values of the induced

voltages [7], a function Fo (estimated and calculated one) which is a modified expression of the

reactive power [8] or by comparing estimated rotor flux in the control circuit with that one

calculated) [9]. Other schemes are based in the injection of an external signal [10] using additional

transducers and some without them. Most of the schemes are themselves dependent on machine

parameters (stator resistance, rotor parameters, leakage inductance) or use additional sensors or

implemented algorithms are very complex presenting critical problems in execution time. The

8
proposed adaptive scheme merges for the first time two simple techniques (indirect vector control

and Tr identification by PRBS injection) without use of additional sensors, while having little

increase in the software time.

The cross-correlation function as a Tr estimator

The cross-correlation function which measures the influence of one signal over another one, is

given by

1 T
φ xy ( τ ) = lim ∫− T x( t ) y( t − τ ) dt = x( t ) y( t − τ ) . (12)
T →∞ 2T

The correlation signal is a weighted sum of the product of both x and y signals [2]. The diagram

representing the cross-correlation function is presented in Fig. 7. The implemented algorithm

injects, based on this idea, a monitoring signal in the form of a Pseudo-Random-Binary-Signal

(PRBS) in flux channel (d model channel) and correlating it with the observed signal in torque

channel (q motor channel) by the speed error (∆ω) (Fig. 8). A possible connection between these

signals is detected by the cross-correlation function. If the d model and q motor channel axes are

orthogonal, ∆isd injected noise (PRBS) will not appear in the q motor direction and the cross-

correlation function will disappear. If not so, this signal will indicate Tr changes (increase or

decrease) and will be proportional to its magnitude.

The injected PRBS signal and the observed speed error will be correlated. Depending on the

PRBS level, the speed error is multiplied by +1 (high level) or -1 (low level). The discrete cross-

correlation function will be

φ xy =
∑ (± 1)∆ω , (13)
n

where n is the number of collected data. The delay time τ is chosen to maximize the sensibility of

φxy to the relation between signals. The time of 7ms was found to maximize φxy in the experimental

tests. Fig. 9 is a simplified scheme of the decoupling between flux and torque channels with the

rotor flux error influency and the injected signal PRBS. The ∆isd influence in the torque through

the magnetizing current imr will be filtered because the PRBS frequency (143 Hz) is much greater

than Tr−1 (8.4Hz). The PRBS amplitude was chosen 15% of the nominal current and the additional

9
time for the algorithm represented an increase of 34%. In the following, simulated and experimental

results are presented for Tr identification algorithm with 50% of nominal load.
y(t) x(t)
G(s) Weighted Sum φ xy(τ)

y(t-τ)

1
τ

Fig. 7. Cross-correlation function diagram.

i sq (motor)
q(motor)

q(model)
i sq (model)

is
δ ∆isd
d(model)
i sd (model)
ε i sq (motor)
Correct direction of rotor flux
d(motor)
i sd (motor)

Fig. 8. Vector diagram for rotor time constant identification.

PRBS ml
isd imr md -
isdref + + Tr J
cos(e ) - +
+

isq
ω
+ +
sin(e ) ωref -
+
∆ω

Fig. 9. Simplified scheme for the decoupling error (e) between channel flux and torque.

10
 r identification results

Fig. 10 shows a simulated result presenting the cross-correlation function for different values of

model Tr (Tr* ). It is noted that the function is positive if Tr* > Tr and negative if Tr* < Tr . Moreover,

its amplitude is proportional to the rotor time constant variation.

Cross-correlation signal
+2.0
+1.5
2.0 T r* 1.75 T r*
+1.0
+0.5
0.0
1.5 T r*
-0.5
-1.0 0.5 T r*

-1.5
0.25 T r*
-2.0
0.0 Time (ms) 2.56

*
Fig. 10. Simulated results for different Tr

The identification process for same error in Tr* but for different loads is presented in Fig. 11. The

response time and magnitude of the correlation signal depend on the machine loading ; for a smaller
load, a smaller and slower cross-correlation function occurs and vice-versa. This seems acceptable,

since at light load the adjustment of Tr is uncritical, whereas fast tracking is desired when the motor

is heavily loaded.

11
 Cross-correlation signal
+1.0
50% nominal load
+0.75
+0.5
40%
+0.25
30%
0.0
-0.25
-0.5
-0.75
-1.0
0.0 Time (ms) 15.0

Fig. 11. Simulated result for different loads

In Fig. 12, the identification algorithm is applied to the drive system for different values in the

model Tr (Tr* ) representing a detuned operation with half nominal load. Fig. 12a and Fig. 12b

present the cross-correlation signal for a model with a rotor time constant greater than the real one.

For the two signals, the cross-correlation has a time to separate the PRBS signal from the noise and

other possible influences until it stabilizes. The final signal of Fig. 12b is bigger then Fig. 12a

because the Tr deviation in it is greater. The same occurs in Fig. 12c and Fig. 12d but now to model
Tr minor then the motor one. It is noted that the correlation follows changes in Tr in direction

(increase or decrease) and magnitude.

Cross-correlation (T r*=1.5 Tr)

+0.3

0.0

-0.3
(a) 0 Time (s) 52

12
 Cross-correlation (Tr*=2.0 Tr)
+0.3

0.0

-0.3
(b) 0 Time (s) 52

Cross-correlation (T r*=0.85Tr)
+0.3

0.0

-0.3
(c) 0 Time (s) 52

Cross-correlation (Tr*=0.75Tr)
+0.3

0.0

-0.3
(d) 0 Time (s) 52

Fig. 12. Experimental results for Tr identification

It should be noted in Fig. 13. that the response time and magnitude of the identification algorithm

depend on machine loading . For a smaller load and with same Tr deviation of 25% seen in Fig. 12d

the correlation stays slower and smaller confirming simulated results.

13
 Cross-correlation with (T r*=1.5 Tr)
and 25% of nominal load
+0.3

0.0

-0.3
0 Time (s) 52

Fig. 13. Correlation signal for model Tr 25% minor with 1/4 of nominal load

4 The compensation scheme

The correlation signal is used to compensate the misalignment between the flux- and torque

channels by a PI regulator. In Fig. 14., the compensation scheme is presented. The cross-correlation

signal is processed by the regulator that acts as a slip compensation (∆ω s ) added to the detuned slip
frequency (ω s ). Then, the compensation scheme will bring the correlation signal to zero setting

back flux and torque channels decoupling. The complete adaptive scheme can be now presented

showing the identification and compensation in Fig. 15.

φxy PI ∆ωs

∆ωs
i sq* ωs +
i sq* ωs + ∆ωs
Tr* . i sd* +

Fig. 14. Compensation scheme

14
 PRBS

∆i sd*
∆ω PI
i sq*
ωref + F1 = (isq* ) 2 + (isd *) 2 |Is |
+- ARW
∆i sd*

i sq* δ
F2 = Tan -1 ( i sd* ) +
ξ
+
ρ
ω
i sq* ωs
F5 =
Tr i sd *
+ + z dt

φxy P

∆ω s
PI

Fig. 15. Complete adaptive scheme proposed.

Fig. 16. presents the simulation of the complete adaptive scheme. There are two examples: one

for a model Tr greater and other for one minor than motor one. In the two cases the regulator acts

compensating the model's misalignment taking the correlation signal to zero.

Cross-correlation

0.0 PI regulator signal

0.0
Correlation signal
0 Time (s) 15

(a) Compensation for a +75% model Tr error

15
 Cross-correlation

0.0 PI regulator signal

0.0
Correlation signal
0 Time (s) 15

(b) Compensation for a -50% model Tr error

Fig. 16. Complete simulated adaptive scheme.

In Fig. 17. experimental results of the proposed identification-compensation scheme are shown

with good performance.

PI regulator

0 Time (s) 52

Cross-correlation
+0.3

0.0

-0.3
0 Time (s) 52

(a) Compensation for a +50% model Tr error

16
 PI regulator

0 Time (s) 52

Cross-correlation
+0.3

0.0

-0.3
0 Time (s) 52

(b) Compensation for a -35% model Tr error

Fig. 17. Experimental results for complete adaptive scheme.

5 Conclusion

The use of AC-motors as controlled electrical drives, in spite of their nonlinear multivariable

characteristic, may be attractive using field orientation and microprocessors. A 1,2KW four-

quadrant AC-drive has been developed, in which the control is implemented with a microcomputer

and the motor is supplied by a voltage-source inverter with current control using IGBTs.

Indispensable for the high performance indirect vector control, the rotor time constant identification

is determined by adding a monitoring signal (PRBS) without use of additional sensors, while adding

only a little more software overhead. The use of a correlation function between the monitoring

signal and the speed error indicates the direction of Tr changes, presenting a value proportional to

its magnitude. A compensation scheme was implemented using a PI regulator. Experiments using

this adaptive digital scheme for different values of model Tr (representing a detuned operation) are

presented showing good dynamic response.

17
6 List of Symbols

ARW Anti-Reset Wind-up

d flux channel

e decoupling error between channel flux and torque

D/A digital to analogic signal conversion (Fig. 3.)

F1 , F2 , F5 functions calculated in indirect vector control (Fig. 2.)

G(s) laplace representation of a system example (Fig. 7.)

r
imr magnetizing current vector
r r
is , ir stator and rotor current conjugate complex vector

isd , isq direct and quadrature components of stator current

isdref reference for direct component of stator current (Fig. 9.)

is1 , is2 , is3 instantaneous stator current in phase one, two and three, respectively

IM induction motor
Is magnitude of the stator current

J lumped inertia

Ls , Lr self stator and rotor inductance

Lo mutual inductance
md ,ml driving and load torque

n number of collected data

P number of pole-pairs

PI proportional plus integral speed regulator

PRBS pseudo random binary signal

PS position sensor

q torque channel

Re, Im real and imaginary component

Rs , Rr stator and rotor resistance per phase

T time period
Tr rotor time constant

Tr* rotor time constant value used in the computer model

r r
us (t ) , ur (t ) stator and rotor voltage conjugate complex vector

18
us1 , us2 , us3 instantaneous stator voltage in phase one, two and three, respectively

ε rotor angular position

ω rotor angular speed

δ load angle

ξ stator current angle

γ angular coordinate with a value of 2/3
r
ρ argument of vector imr

τ delay time

∆ hysteresis magnitude for current control (Fig. 3.)

r
ω mr instantaneous angular speed of imr

ω ref speed reference

ωs slip frequency

∆ω speed error

∆ω s added slip compensation

σr rotor leakage factor

∆isd monitoring signal added in direction of flux channel d

φxy cross-correlation function between x and y signals

(− )* value of variable (-) in the computer model

References
[1] Blaschke, F.: The principle of field orientation as applied to the new TRANSVECTOR-closed
loop control systems for rotating field machines. Siemens Review (1972), pp.217
[2] Leonhard, W.: Statistische Analyse linearer Regelsysteme. Stuttgart: B. G. Teubner, 1973
[3] Leonhard, W.: Control of Electrical Drives. New-York: Springer-Verlag, 1985
[4] Gabriel, R.; Leonhard, W.: Microprocessor control of induction motor. Int. Semiconductor
Power Converter Conference (1982), pp. 385-396
[5] Heinemann, G.; and Leonhard, W.: Self-tuning field oriented control of an induction motor
drive. IPEC, Tokyo/Japan (1990), pp. 465-472
[6] Branco, P.J.C.: A didactic explanation of field oriented systems sensitivity 5th Eur. Conf. on
Power Electron. Appl. (EPE), London/UK (1993),Vol. 5, pp. 408-412
[7] Matsuo, F.; and Lipo, T.A.: Rotor parameter identification scheme for vector-controlled
induction motor drives. IEEE Transactions on Industry Applications. (1985) no. 21, pp. 624-
632
[8] Ohm, D.Y.; Khersonsky, Y.; and Kimzey, J.R.: Operating characteristics of indirect field
oriented induction motor. Conf. Rec. Applied Power Electronics Conference (APEC). (1989),
pp. 91-100.

19
[9] Saitoh, T; Okuyama, T.; and Matsui, T.: An automated secondary resistence identification
sheme in vector controlled induction motor drives. Conf. Rec. IEEE Industry Applications
Society Annual Meeting (IAS). 1989, pp. 594-600.
[10] Schumaker, W.; Leonhard, W.: Transistor-fed ac-servo drive with microprocessor control.
Conf. Rec. IEEE Int. Power Electronics Conference (IPEC). 1983, pp. 1465.

20