98

SENSORLESS DTC OF INDUCTION MACHINES WITH COMPLETE CORE LOSS COMPENSATION

T.Pham-Dinh, E.Levi

Liverpool John Moores University, UK

INTRODUCTlON
Principles of operation of direct torque control scheme
(Takahashi and Noguchi, (I)) are based on the
assumption that the existence of the core loss in an
induction machine may be neglected. The same applies
to all the standard vector control schemes. Among the
phenomena that lead to detuned operation of a vector
controlled induction machine, core loss w a s the last to
attract attention. Initial proposals for compensation of
the core loss within rotor flux orientation schemes were
given by Mizuno et al (2) and Hintze and Schroder (3).
This was followed by detailed investigations of the core
loss induced detuning effects in rotor flux oriented,
stator flux oriented and air gap flux oriented induction
motor drives (Levi (4) and Levi et al(5)). In general, the
existence of the core loss in the machine leads to an
orientation angle error and discrepancies between the
reference and actual values of the relevant flux and
motor torque in steady state operation. As far as
dynamics are concerned, the net consequence of the core
loss is a slower acceleration (since core loss reduces the
output torque in motoring) and faster deceleration (since
shaft torque is higher in braking than estimated).
The most frequently used approach to model the core
loss is based on representation of the core loss with an
equivalent core loss resistance, placed in parallel to the
magnetising branch of the motor within the dynamic
equivalent circuit (3-5). Alternatively, equivalent core
loss resistance can be placed immediately after the stator
resistance (Dinrich (6)) or an equivalent series core loss
model can be used (Jung and Nam (7)). On the basis of
such models numerous schemes for compensation of
core loss induced detuning have been developed for
vector control ((3-5),(7), Choi et al (E), Matsuse et a1
(9)).
A common feature of all these studies is that they apply
to vector control schemes with speed (position) sensor.
Hence the process of core loss Compensation requires
modifications of the vector controller and the flux
position estimation algorithm. Recent emphasis on
sensorless control has initiated an interest in evaluation
of the core loss impact on sensorless vector control and
investigation of the means for compensation (Wang et al
(IO), Matsuse et al ( I I ) , Levi et al (U),Namiki et al
(13)). If speed estimation is performed using a model
based approach, compensation of core loss within the
vector controller and flux position estimator does not
Power Electronics, Machines and Drives, 16-18 April 2002.
Conference Publication No.487.0 IEE 2002

provide satisfactory drive operation (10,lZ). The

accuracy of speed estimation is affected by the core loss
as well (12) and therefore it becomes necessary to
additionally compensate for the core loss within the
speed estimator.
On the basis of the survey presented above one can
conclude that the impact of core loss on operation of a
vector controlled induction motor drive is nowadays
well understood and that the appropriate means for
compensation are readily available. .The same however
cannot be said for DTC. The only available studies
related to the core loss impact on DTC of an induction
motor are those reported recently by the authors of this
paper (Pham-Dinh and Levi (14), Levi and Pham-Dinh
(15)). Speed mode of operation of a sensored drive (i.e.
closed loop speed control using actual speed feedback)
and torque mode of operation were analysed in (14) and
(IS), respectively. The studies have shown that, when
stator flux estimation is performed by measuring stator
voltages and currents, existence of the core loss affects
only the correctness of the torque estimate, while stator
flux magnitude and position estimates automatically
account for the core loss. Torque estimate is higher than
the actual motor useful torque in motoring (and vice
versa during braking), so that the acceleration transients
last longer than necessary. A toque error compensation
method was proposed and verified in (14,lS).
The purpose of this paper is to further extend the
findings of (14,15) by looking at a sensorless direct
torque controlled induction machine operated with
closed loop speed control. The applied method of speed
estimation is model-based and relies on the MRAC
technique. It is therefore necessary to compensate for
the existence of the core loss in both the torque
estimation and the speed estimation algorithms. A
scheme with full core loss compensation is developed
and verified by simulation.

DIRECT TORQUE CONTROL SCHEME

The considered DTC scheme is the one of ( I ) . The
inputs into the control system are the stator flux
reference (that is here kept at constant rated value since
the base speed region is analysed) and the speed
reference. Speed controller is of PI type. The DTC
scheme is shown in Fig. I. Estimation of the stator flux

99

and torque, as well as stator flux position estimation, are

achieved using the well known method based on stator
voltage and current measurement. It is described with
the following equations in the stationary frame of
reference:

The selected speed estimator is of MRAC type and is

illustrated in Fig. 2. As the control scheme requires
stator flux estimation, outputs of the reference and the
adjustable. model are selected as stator flux space
vectors. The speed estinktor with neglected core loss is
described with the following well-known space vector
equations in the stationary reference frame:

Since the simulation model includes the voltage source

inverter, stator voltages and currents are filtered using
the 2nd order Butterworth filter with a cut-off frequency
of 500 radis.

Here a,L, and T, stand for the total leakage coefficient,

stator self-inductance and the rotor time constant,
respectively. Error quantity in Fig. 2 is

Stator flux comparator is a two-level hysteresis

comparator, while torque comparator is a three-level
hysteresis comparator. The switching table applied here
is a modified switching table proposed by Casadei et al
(16), that enables an improved quality of performance in
the low speed region, compared to the Takahashi's
optimum switching table of (1). The table does not
utilise zero voltage space vectors in. the low speed
region. Speed at which the change is executed from low
speed region to high speed region is set to +20% of the
rated speed, (16). The switching table is reproduced in
Table I, where index k stands for the sector in which
stator flux is positioned in any given instant.

qfk
model

7
1-1-

Adjustable
model

Error
calculation

PI
contr.

- 1

11 a'''I

Fig2. MRAC based speed estimator.

Flux
Ays
comparator

'

Inverter
switching +
+ table

A Statorflux

Measured

vector sector
Torque
ATc
comparator

High negative speed region

+U" < 0 < -0.2%
Tt 1

T.

vI.f

YO

w. .1

YO

Low speed region

< 0.2%
vk+l

%-2

%+2

os

-0.2y, < w
Te7

%-I

signals

T, 1

High positive speed region

0.2% < w < %
Tc 1

vk-1

VHI

VO

%-2

vk+2

YO

100
CORE LOSS MODELLING

The induction motor model with core loss representation

is formulated on the basis of the dynamic space-vector
equivalent circuit illustrated in Fig. 3, (4). This model is
used for representation of the machine in all simulations
(stationary reference frame, 4 = 0, is utilised).
Equivalent core loss resistance RFc models the
fundamental core loss component and is placed in
parallel to the magnetising branch. It is a function of the
operating stator frequency, as indicated in Fig. 3. Such
an approach to core loss modelling has been used
extensively in discussions related to vector controlled
induction machines. However, due to inherent
differences between the vector control and DTC, the
circuit of Fig. 3 is, strictly mathematically, incorrect for
representation of an induction machine subjected to
DTC. The issue has been discussed in detail in (14,15)
and it suffices to say here that application of the model
of Fig. 3 will yield correct results with respect to the
fundamental harmonic behaviour only, while behaviour
for higher harmonics (that is, torque and stator flux
ripple) will be distorted.
Fundamental core loss component was determined
experimentally, using the procedure of (5). Base speed
region, relevant for the analysis in this paper,
encompasses frequencies up to 50 Hz. Once when the
fundamental core loss component as function of the
operating frequency is known, the equivalent core loss
resistance of Fig. 3 can be calculated. Figure 4a presents
the measurement results related to the fundamental core
loss component, its analytical approximation and the
corresponding equivalent core loss resistance
(calculated values and the analytical approximation).

a.

IO

20

30

40

Jn

Frqucnn/(Hz)

--malybcal sppmxm"n of RC
r.lr"lstsdqu,"alm~ re~lJI."Cc R C

b.
Fig. 4. Results of the experimental identification:
a. Fundamental core loss and calculated equivalent
a
core loss resistance Rfl for the circuit of Fig. 3.
b.Equivalent core loss resistance R, for the
circuit of Fig. 5.

Another possibility is to place the equivalent core loss

resistance immediately after the stator resistance, instead

of placing it in parallel to the magnetising branch (Fig.
5). Such placement is more convenient for development
of a modified speed estimator, as shown shortly. The
corresponding values of the equivalent core loss
resistance in Fig. 5 can be calculated from those in Fig.
4a. The equivalent core loss resistance of the circuit of
Fig. 5 is illustrated in Fig. 4b. Data of the 4 kW
induction motor, used in the study, are given in (10,12).

Fig. 5. Altemative placement of the core loss resistor

for modified speed estimator development.

COMPENSATION OF CORE LOSS

Core loss compensation in torque estimation

Fig 3. Space vector equivalent circuit of an induction

machine with core loss representation.

Estimation of stator flux magnitude and position using

equations (I), (2), (4) automatically accounts for the
existence of the core losses (14.15). It is therefore
necessary to compensate for the core loss in the torque
and speed estimation only. Torque estimate (3), used for
the closed loop torque control, does not recognise the
existence of the core loss and therefore delivers a too

101
high value of the estimated torque in motoring. Hence
the torque estimate needs to be modified, by deducting a
core loss related torque component, so that the new
torque estimate for closed loop torque control is given
with (l4,15):

purposes. The required torque increment is under these

conditions constant and equal to

T'=(3/2)pbmia, - w @ i m ) - ~ T F e

where index n stands for rated values and PF,. is the

fundamental core loss at rated (50 Hz here) frequency.
As the fundamental core loss at 50 Hz equals 173.4 W,
while rated motor power and torque'are 4 kW and 26.5
Nm, respectively, the required torque increment is 1.15
Nm. This constant value is used in simulations that are
discussed in the next section.

(7)

The torque increment of ( 7 ) that will provide

compensation of the fundamental core loss component
can he obtained as

Tpr' = T,'m,'
PF, = A T & ! '
AT, = P , / m r

+ PF,

AT, =

LT_
= const.

(10)

pn

(8)

Core loss compensation in speed estimation

It should be noted that (8) is valid for both motoring and

braking and that the sign of the torque increment is
always the same (that is, core loss reduces the output
torque available in motoring, while the shaft torque in
braking is higher than predicted by (3)). Symbol a&*' in
(8) stands for estimated mechanical angular speed of
rotor rotation, while PF, is the fundamental component
of the core loss.
The torque error compensation method given with (8) is
universally valid for any induction machine. It requires
that variation of the fundamental core loss with
frequency is known (graph shown in Fig. 4a) and that
the stator frequency and rotor speed of rotation are
available within the control scheme, so that calculations
according to (7)-(8) can -be done. However, stator
frequency is normally not readily available within the
DTC. Another possibility is then to use instead of stator
frequency electrical rotor speed of rotation, since the
speed value is anyway estimated for the control system
of Fig. I . Thus the two possible methods of calculating
the required toque increment are the exact one, based
on the knowledge of the stator fundamental frequency,
and the approximate one in which speed estimate is used
instead of the stator frequency:

(9)

Depending on whether the information regarding stator

frequency or the rotor speed is available, one or the
other of the two possibilities given in (9) may be
selected. However, in a number of cases the proposed
compensation method can be considerably simplified, as
discussedin (14,15).
Fundamental core loss component in this particular
machine exhibits almost linear variation with frequency,
as is evident from Fig. 4a. By approximating the core
loss in the base speed region with a linear dependence
on frequency, one arrives at the simplest possible core
loss compensation strategy for the torque estimation

A modified speed estimation algorithm, that accounts

for the core loss, can be developed using the circuit of
Fig. 3 and equations ( 5 ) as the starting point. However,
a considerably simpler algorithm results if the circuit of
Fig. 5 is used instead of the one of Fig. 3. Reference
model equation remains identical to the one in (5), while
adjustable model equation remains of the same form as
in (5). Speed estimator with core loss compensation is
described with the following equations:

yP'=jC,-R,i.)dt

Here &' is the modified stator current after the core loss
resistance in Fig. 5, calculated as

Function R, ( f) is the equivalent core loss resistance of

Fig. 4b. Since stator frequency is not readily available in
the DTC scheme, frequency corresponding to the speed
estimate ( P a r / 2 n ) is used instead. Error quantity of
the modified speed estimator remains to be given with
( 6 ) and the structure of the speed estimator is the same
as in Fig. 2.

VERIFICATION
OF
THE
COMPENSATION SCHEME

PROPOSED

The control scheme of Fig. 1 and the speed estimator of

Fig. 2 are implemented in Simulink. The model of the
inverter is included. A 4 kW, 26.5 Nm, 50 Hz, four-pole
motor, with the core loss characteristics given in Fig. 4 ,
is used. Torque is limited to 1.5 times the rated value
(39.75 Nm). Hysteresis bands are set to +I% of the rated
torque and rated stator flux (peak) values. Stator flux
reference is equal to rated (peak) stator flux (0.9889

102
Wb). The examined sequence of transients consists of
acceleration from standstill to one half of the rated speed
(which is 301.8 rad/s electrical), followed by reversing
to negative one half of the rated speed. The test applies
to no-load conditions. Figure 6 illustrates simulation
results obtained when core loss existence is ignored in
both the torque and the speed estimation. Figure 7 shows
results for the same transients, arrived at with
compensated core loss in both the torque and the speed
estimation (stator flux trace is the same as in Fig. 6 and
is therefore not repeated). Stator flux is kept, on
average, at the required level. Stator flux ripple appears
to somewhat exceed the prescribed hysteresis band
during reversing transient, in vicinity of zero speed
crossover. The excursion of the stator flux outside the
hysteresis hand is nevertheless significantly smaller than
when Takahashi's table is used.

40
30

tan

80

'z
7
E
I

T
i

-40

E,/,",.I.b

1 .we1

-0.6

s2

i
;-60
2

0.6

0.4

ArlYa1 V..d

-120

0.2~

-IS0
0

0.3

06

0.9

I .*

l i m e (I,

0.3

0.6

0.9

1.2

,.5

Tme(s)

Fig. 7. Acceleration and reversing transient with full

compensation of core loss: motor torque, and
actual and estimated electrical speed.
Torque developed by the motor is smaller than the
reference during acceleration and reversing, due to the
core loss. The difference is approximately I . I5 Nm, or
the value that is further used for torque error
compensation. Transients therefore last longer. The
speed estimation error is relatively small, of the order of
2.to 3 rpm, as shown by the zoomed steady-state at one
half of the rated speed, shown in Fig. 8.

Fig. 6. Acceleration and reversing transient without any

core loss compensation: stator flux, motor output
torque, and actual and estimated speed.

Once when the core loss is compensated in torque and

speed estimation, performance improves (Fig. 7). Motor
torque is essentially kept at the limited value (i.e. it is
equal to the reference torque value) and the speed
estimation error is practically eliminated, as witnessed
by Fig. 8..Residual speed estimation error is below 0.3
rpm and is the consequence of the .use of frequency
related to rotor speed instead of stator frequency in (12).
Some very small deviation of the motor torque from the
desired reference value in Fig. 7 is explained as follows.
Characteristic of motor core loss in Fig. 4a deviates
from linear relationship against frequency at very low
frequencies. Constant torque increment of (10) is
therefore too high at low frequencies and torque
compensation using constant torque increment of (IO)
leads to slight over-compensation at low frequency in
motoring and slight under-compensation at low
frequency during braking. Consequently, motor torque is

103
slightly higher than it should be at the start of the
acceleration and slightly lower (by absolute value) than
it should be during braking at low frequency. It should
be noted that the sign of the torque increment in (7)-(IO)
is determined with the sign of the speed and therefore it
changes once when the reverse motoring commences.
This explains a small jump, evident in the torque
response, that happens at zero speed crossing. It was
verified during the course of investigation that, when
torque increment is calculated using (9), developed
motor torque is essentially constant during rapid
transients and equal to the reference torque in the limit.

E.lll.<

V.81

=,

P 720

IClUSl

.m.1

Xtlh Llll

,,on l o l l ro.p.n$.lion

119
0.42

0.4282

0.1204
Time

04288

0.4208

0.421

(I)

Fig. 8. Zoomed extracts from Figs. 6 and 7, comparing

the estimated speed to the actual speeds in steady
state operation at one half of the rated speed.

CONCLUSION
The paper discusses impact of core loss on sensorless
DTC of an induction motor and develops a modified
DTC scheme with.full core loss compensation. Core loss
is compensated in both the process of torque estimation
and speed estimation. A novel modified speed estimator
of MRAC type is developed for that purpose. The
proposed compensated DTC scheme is verified by
simulation.

REFERENCES
1. Takahashi I and Noguchi N, 1986, A new quick
response and high efficiency control strategy of an
induction motor:
22, 820-827.
Mizuno T, Takayama J, Ichioka T and Terashima
M, 1990, Decoupling control method of induction
motors taking stator core loss into consideration,
Tokyo, Japan, 69-74.
3. Hintze D and Schroder D, 1992, Induction motor
drive with intelligent controller and parameter
adaptation, IEEE Ind. Aopl. Soc. Ann. Meeting IAS,
Houston, Texas, 970-977.
4. Levi E, 1995, Impact of iron loss on behaviour of
vector controlled induction machines, IEEE Trans. on

31, 1287-1296.
5. Levi E, Sokola M, Boglietti A and Pastorelli M,
1996, Iron loss in rotor flux oriented induction
machines: identification, assessment of detuning and
compensation, IEEE Trans. on Power Electronics, 11,
698-709.
6. Dittrich A, 1998, Model based identification of
the iron loss resistance of an induction machine, proC.
7 Int Conf. Power Electronics and Variable Soeed
Drives PEVD98, London, UK, IEE Conf. Pub. No. 456,
500-503.
7. lung J and Nam K, 1998, A vector control scheme
for EV induction motors with a series iron loss model.
IEEE Trans. on lndusmal Electronics, 45,617-624.
8. Choi JW, Chung DW and SUI SK, 1996,
Implementation of field oriented induction machine
considering iron losses, Proc. 1 Ith IEEE Aoulied
San Hose, California, 375379.
9. Matsuse K, Yoshizumi T, Katsusa S and Taniguchi
oriented induction motor with high efficiency taking
core loss into account, IEEE Trans. on Industry
Applications, 35,62-69.
10. Wang M, Levi E and Jovanovic M, 1999,
Compensation of parameter variation effects in
sensorless indirect vector controlled induction machines
using model based approach, Electric Machines and
Power Systems, 27, 1001-1027.
11. Matsuse K, Taniguchi S, Yoshizumi T and Namiki
K, 2001, A speed sensorless vector control of induction
motor operating at high efficiency taking core loss into
account,
37,
548-558.
12. Levi E, Wang M and Williams D, 1999,
Evaluation of iron loss influence on speed estimation in
sensorless MRAC-based field-oriented induction
machines, European Trans. on Electrical Power ETEP.
9, 77-84.
13. Namiki K, Matsuse K, Kishimoto T and Yokomizo
S, 2001, Characteristics of speed sensorless vector
controlled induction motor drive operating at high
efficiency taking core loss into account, Proc. 16Ih
IEEE Applied Power Electronics Conf. APEC,
Anaheim, California, 25-3 I .
14. Pham-Dinh T and Levi E, 2001, Core loss in
direct torque controlled induction motor drives:
detuning and compensation, Proc. IEEE Power
Electronics Suecialists Conference PESC, Vancouver,
Canada, 1429-1434.
15. Levi E and Pham-Dinh T, 2001, Impact of iron
loss on direct torque control of induction machines and
its compensation, Proc. of the 9th European Conf. on
Power Electronics and Aoolications EPE, Graz, Austria,
CD-ROM.
16. Casadei D, Grandi G, Serra G and Tani A, 1994;
Switching strategies in direct torque control of
induction machines, Proc. Int. Conf. on Electrical
Paris, France, 204-209.