DIRECT TORQUE CONTROL OF INDUCTION MACHINES WITH

CONSTANT SWITCHING FREQUENCY AND IMPROVED STATOR FLUX

ESTIMATION
Nik Rumzi Nik Idris Abdul Halim Mohd Yatim Naziha Ahmad Azli
Department of Energy Conversion
Faculty of Electrical Engineering,
University Teknologi Malaysia
Johor, MALAYSIA
nikrumzi@ieee.org http://encon.fke.utm.my


Abstract
Direct torque control (DTC) of induction machines is
known to have a simple control structure yet has
comparable performance to that of the field oriented
control (FOC). To maintain this simple control
structure while at the same time improving the
performance of the DTC drive in terms of switching
frequency and stator flux estimation, two simple
techniques are proposed. The effectiveness of the
proposed techniques are verified experimentally
which is centered around TMS320C31 digital signal
processor.


I. INTRODUCTION
More than a decade ago, direct torque control
(DTC) was introduced to give a fast and good
dynamic torque response and can be considered as an
alternative to the field oriented control (FOC)
technique[1,2]. The DTC scheme is very simple; in
its basic configuration it consists of DTC controller,
torque and flux calculator, and VSI. The
configuration is much simpler than the FOC system
due to the absence of frame transformer. It also does
not need pulse width modulator and position encoder,
which introduce delays and requires mechanical
transducers respectively.
Since it was introduced in 1986 [1], a large
number of technical papers appear in the literature
mainly to improve the performance of DTC of
induction machines Two problems usually associated
with DTC drives which are based on hysteresis
comparators are: (i) variable switching frequency due
to the hysteresis comparators used for the torque and
flux comparators and (ii) inaccurate stator flux
estimations which can degrade the drive
performance. Some schemes have managed to
maintain an average constant switching frequency by
utilising space vector modulation[3], predictive
control[4,5], dead beat control[3] and artificial
intelligence.[6,7,8]. All of these techniques however
increases the complexity of the drive systems. Thus,
the inherent simple control structure of the DTC
drive is lost. In the case of stator flux estimation, the
problems associated with DTC drive are similar to
that of the field oriented control (FOC) drives.
Generally, the stator flux is estimated based on the
variation of the two types of estimators, namely the
voltage and current models [9]. To take advantage of
both type of estimators, some researchers have
combined both types of estimators, namely the
voltage and current models. However, the use of
current model will require a speed sensor or speed
estimator which in both cases are undesirable. To
maintain a speed sensor less operation, only voltage
model has to be used. In voltage model technique,
low pass (LP) estimator is normally used in place of a
pure integrator to avoid integration drift problem[10].
The LP filter however introduce magnitude and phase
errors in the estimated flux.
This paper proposed a simple yet effective
methods to overcome these two problems. To
maintain a constant switching frequency, a simple
torque controller is introduced to replace the
conventional 3-level hysteresis comparator. The
magnitude and phase errors associated with stator
flux estimation based on voltage model is
compensated by using simple compensator which is
based on steady state operations. The rest of the
paper is organised as follows. Sections II and III
presents the proposed torque controller and improved
stator flux estimation methods, respectively. Section
IV gives the simulation and experimental results of
the proposed methods. Finally, conclusions are given
in section V.

II. FIXED SWITCHING FREQUENCY CONTROLLER
A. Torque equation
A DTC drive based on fixed torque hysteresis
band, suffer from a varying rate of change of torque
which is simply referred to as torque slope and
hence torque controller switching frequency, as a
2
function of motor speed, stator and rotor fluxes , and
stator voltage. This in turn results in a varying device
switching frequency. Variable switching frequency is
undesirable since the switching capabilities of the
inverter is not fully utilised. For fixed torque band,
the choice of the bands width is normally based on
the worst condition which is limited by the thermal
condition of the switching devices, thus operation for
other conditions are not optimised.

It can be shown that the torque slope is given
by:

( ) [ ]
g
n , r
g
n , s r n , s
r s
m
r s
n , e
n , e 1 n , e
j j v
L L
L
2
p
2
3
1 1
T
t
T T

,
`

.
|

+

(1)
The symbols carry their usual meaning with the
subscript g denotes the general reference frame.
n , s
v denotes the voltage vector selected for a
particular sector. The negative slope can be obtained
by setting the active voltage,
n , s
v to zero, thus

( ) [ ]
r s r
r s
m
sr
n , e n , e n , e
j j
L L
L p T
t
T T
slope

+
4
3
1
(2)

On the other hand, positive slope is given by the non-
zero voltage vectors, thus:

( ) [ ]
r s r s
r s
m
sr
n , e n , e n , e
j j v
L L
L p T
t
T T
slope

+ +
4
3
1

(3)
B. Proposed torque controller
The proposed torque controller consists of
two triangular waveform generators, two
comparators and a PI controller as shown in Fig. 1.
The two triangular waveforms (which will be
referred to as upper and lower carriers ), are 180
o

out of phase with each other. The absolute values of
the DC offsets for upper and lower carriers are set
to half of its peak-peak value; the upper dc offset is
positive while the lower is negative. In principle,
the output of the proposed torque controller is
similar to that of the three level hysteresis
comparator [1], which can be either of three states:
1 , 0 or 1.


Fig. 1 Proposed torque controller

The value of the instantaneous output of the
torque controller named as torque error status
and designated by q(t) is given by :

'

< <

lower pi
u pper pi lower
u pper pi
C T for 1
C T C for 0
C T for 1
) t ( q (4)

The average torque error status, ) (t q , is
defined as continuous duty ratio and is denoted by
d(t). The average is taken over an interval T
tri
,
which is the period of the trangular carrier
waveform. It is defined by

t r i
T t
t
t r i
dt ) t ( q
T
1
) t ( d (5)

The values of the PI controller parameters are
chosen based on the linear analysis of the controller
in Fig. 1. The linear model is obtained by averaging
and linearising the torque equations (2) and (3). It can
be shown that [11] the positive and negative torque
slope equations are given by (6) and (7) respectively:

,
`

.
|

+ +

+
r
e
1
s
s e
e
d
K Bv AT
dt
dT
(6)
r 1 e
e
K AT
dt
dT

(7)

where
s r
1
A

s
r s
m
L L
L
4
p 3
B

) (
4
3
1
s
r s
r s
m
L L
L p
K

3
Averaging and linearising equations (6) and
(7), and with further simplifications, it can be shown
that [11],

( )
s lip 1
s
s e
e
K d Bv AT
dt
dT
+ +

(8)

Based on (8), the transfer function between d
and T
e
can be obtained by setting the slip frequency
to zero. It can also be shown that the transfer function
between the output of the PI controller, T
pi
and the
averaged duty ratio d is given by a reciprocal of the
peak-peak triangular carrier, C
pp
. Thus the overall
small signal block diagram of the proposed torque
controller of Fig 1 with averaged, linearised torque
equation is shown in Fig. 2


Fig. 2 Averaged and linearised torque loop

The values of the PI controllers parameters
are chosen based on the following constraints:

The absolute slope of equation (5) and (6)
cannot exceed the triangular carrier slope

The cross-over frequency of the loop gain of
Fig.2 (bandwidth) cannot exceed half of the
carrier triangular frequency.

With these constraints and the motor
parameters given in appendix, the parameters for the
PI controller are chosen as:

K
p
= 330 and K
i
= 112826


III. IMPROVED STATOR FLUX ESTIMATION

Stator flux estimation based on voltage model
is estimated by using equation (9)

dt ) R i v (
s s s
(9)

To avoid integration drift problem due to the
dc off-set or measurement noise, a low-pass filter is
normally used in place of the pure integrator in stator
flux estimation. With a low-pass filter, sinusoidal
steady state form, it can be shown that:

c e
s s
s
j
R I V
'
+

(10)

Where
c
is the cut-off frequency of the low-pass
filter in rad/s and
s
is the estimated stator based on
LP filter. For a synchronous frequency larger than the
cut-off, equation (10) can be graphically visualized
using a phasor diagram as shown in fig. 3.

Fig. 3. Phasor diagram for steady state operation of
induction machine showing the actual and estimated
stator flux based on LP filter.

If ' ' '
s s
and
s s
, then it can be
shown that the relation between the estimated stator
flux based on LP filter and pure integrator is given by
(11):


+

c
2
e
2
e
s
s
'
'
(11)
where ) ( t a n 2
c e
1




The core of the proposed improvement is to
provide a magnitude and phase compensations for the
estimated flux, under steady state condition, only at
the operating frequency thus improving the steady
state performance of the DTC drive. In other words,
the LP filter action is effective at all frequency except
at the operating frequency, that is,

'

e
c
e
'
s
for
j
iR v
for
j
iR v
(12)

The LP filter action is therefore valid or
effective for the DC offsets and low frequency
4
components present in the sensed currents or
voltages. This avoids the integration drift problem
while maintaining good system stability. This is due
to the fact that the phase and magnitude errors are
compensated at the operating frequency. The stator
flux is compensated at the operating frequency for
the d and q components by determining the
expressions for the actual stator flux in terms of
estimated stator flux in the stationary reference
frame. Fig. 4 shows the relationship of the estimated
and actual stator flux with their d and q components
in the stationary reference frame, denoted by the
direct and quadrature axis , d
s
and q
s
.












Fig. 4 The d and q components of the estimated and
actual stator flux in stationary reference frame

It can be shown that the compensated flux is
given by (12) and (13) and can be incorporated in the
flux estimator block diagram as shown by Fig 5. The
compensator can be initiated by the flag manually or
based on steady state speed error.

,
`

.
|

e
c
d q s q
' ' (12)

,
`

.
|

+
e
c
q d s d
' ' (13)
Fig 5. Block diagram of the improved stator flux
estimator

The synchronous speed is
e
is obtained from
the measured terminal variables using (14).

( )
s
2
s
s s s
e
j
R i v

(14)

IV. SIMULATION AND EXPERIMENTAL RESULTS

Simulation as well as experiments were
carried out to verify the viability of the proposed
methods. The simulation was carried out using
Matlab/Simulink while the experiment is carried out
consisting of dSPACE controller card based on
TMS320C31 DSP, IGBT VSI, standard HP
induction machine coupled to DC machine, and
Xilinx FPGA to implement the voltage selection
table and blanking time of the VSI. The block
diagram of the experiment set-up is shown in Fig.6
Fig. 6 Experiment Set-up

A. Constant switching frequency
The controller of Fig. 2 is implemented within the
software with the sampling period of 55s. The
frequency spectrum of the phase currents at rotor
speed of 10 and 70 rad/s for both the conventional
hysteresis and proposed torque controller are shown
in Fig. 7. It can be seen that the proposed controller
has managed to maintain the dominant harmonics of
the phase current at the triangular frequency
regardless of the rotor speed., i.e. at about 2.3 kHz.
To look at the dynamic response of the proposed
controller, a square wave speed command is applied
and the waveforms of the rotor speed, torque and d-
axis stator flux is as shown in Fig. 8(a). For
comparison, the same command is applied to the
hysteresis-based torque controller with the results
shown in Fig. 8(b). From the figure it can be seen that
with the proposed controller, the dynamic torque
controller is the same as with the hysteresis based
controller, with an added advantage of a reduced
torque ripple [12].

5



(a)



(b)
Fig. 7 Frequency spectrum of phase current at (a) 10
rad/s, (b) 70 rad/s



(a)


(b)

Fig. 8. Response for a square wave speed reference
for (a) proposed torque controller and (b) a
hysteresis-based controller. Top: speed (63 rads
-
1
/div), middle: torque (1 Nm/div), bottom: stator
flux d-axis (0.5 Wb/div)

B. Improved stator flux
The block digram of the compensation
scheme is shown in Fig 9. All of the compensation
and estimations were done within the software with
sampling frequency of 55s. Fig. 10 shows the
simulation and experimental results of the stator flux
before and after the compensation is applied with LP
cut-off frequency of 5 rad/s and synchronous
frequency of 31 rad/s. The waveforms of the stator
flux from the simulation and experimental results
clearly indicate the improvement in the stator flux
after the compensation is applied. Fig 11 shows the
waveforms of the referenece torque and actual
torque obtained from simulation and experimental
results before and after the compensation is applied.
Due to the phase error present in the estimated stator
flux, the actual torque before the compensation
contain ripple at six times the fundamental
frequency.


Fig. 9 Implementation of the compensation
Proposed
controller
Hysteresis
controller
Proposed
controller
Hysteresis
controller
6

4.5 4.6 4.7 4.8 4.9 5 5.1 5.2 5.3 5.4 5.5
0
0.2
0.4
0.6
4.5 4.6 4.7 4.8 4.9 5 5.1 5.2 5.3 5.4 5.5
-0.4
-0.2
0
0.2
0.4
0.6

(a)


(b)
Fig. 10 Magnitude of the stator flux (top) d and q
axis of stator flux (bottom), (a) simulation, (b)
experiment

4.75 4.8 4.85 4.9 4.95 5 5.05 5.1 5.15 5.2 5.25
0
0.2
0.4
0.6
4.75 4.8 4.85 4.9 4.95 5 5.05 5.1 5.15 5.2 5.25
0
0.2
0.4
0.6

(a)

(b)
Fig. 11 Reference torque (upper trace) and
actual torque (lower trace) obtained from (a)
Simulation, (b) experiment

V. CONCLUSIONS

A new torque controller for constant switching
frequency and a compensation scheme to eliminate
the phase and magnitude errors for the stator flux
estimation under steady state condition is proposed..
The validity of the proposed compensation scheme is
supported by simulation and experimental results.
Theses simple schemes have significantly improved
the performance of the DTC drive system while at the
same time maintaining the simple control structure of
the DTC drive.



APPENDIX

Induction machine parameters:
HP, 240V, 2-pole standard IM,
L
r
=L
s
=0.859H, L
m
=0.828H,
R
s
=10.9 , R
r
=9.5



REFERENCES

[1] I. Takahashi , T. Noguchi, A new quick-
response and high-efficiency control strategy of
an induction motor, IEEE Trans. Ind. Appl.,
vol IA-22, No 5 Sept/Oct 1986

[2]. P. Tiitinen, The next generation motor control
method, DTC direct torque control, Proc. of
Int. Conf on Power Electronics, Drives and
Energy System for Industrial Growth, N. Delhi,
India, pp. 37-43, 1996.
[3] T. G. Habetler,. F. Profumo, M. Pastorelli and
L. M. Tolbert, Direct torque control of
induction machines using space vector
modulation, IEEE Trans. Ind. Appl., Vol. 28,
No. 5, pp. 1045-1053, 1992.
[4] Y. Li, J. Shao,. and B. Si, Direct torque control
of induction motors for low speed drives
considering discrete effect of control and dead-
time timing of inverters, in Conf. Rec. IEEE-
IAS Annual Meeting, pp. 781-788, 1997.
[5] J. K. Kang. and S. K. Sul, Torque ripple
minimisation strategy for direct torque control
of induction motor, in Conf. Rec. IEEE-IAS
Annual Meeting, pp. 438-443, 1998
[6] S. Mir, M. E. Elbuluk and D. S. Zinger, Fuzzy
implementation of direct self control of
induction motors, IEEE Trans Ind. Appl., Vol.
30, No. 3, pp. 729-735, 1994.
[7] I. G. Bird. and H. Zelaya De La Parra, Fuzzy
logic torque ripple reduction for DTC based AC

0.1 N-m/div
7
drives, Electronic Letters, Vol. 33, No.17, pp.
1501-1502, 1997.
[8] S. Mir. and M. E. Elbuluk, Precision torque
control in inverter-fed induction machines using
fuzzy logic, In Conf. Rec. IEEE-IAS Annual
Meeting, pp. 396-401, 1995.
[9] P.L. Jansen and R.D. Lorenz, A physically
insightful approach to the design and accuracy
assessment of flux observers for field oriented
induction machine drives, IEEE Trans Ind
Appl., vol 30, no. 1, Jan/Feb, 1994
[10] Jun Hu and Bin Hu, New integration
algorithms for estimating motor flux over a
wide speed range, IEEE Trans Power
Electronics., vol 13, no. 5, September, 1998.
[11] N.R.N. Idris, Improved direct torque control of
induction machines, PhD Thesis, Universiti
Teknologi Malaysia, 2000.

[12] N.R.N. Idris and A.H.M. Yatim, Reduced
Torque Ripple And Constant Torque Switching
Frequency Strategy For Direct Torque Control
Of Induction Machine, In Conf Rec. IEEE-
APEC, pp. 154-161, vol .1, 2000.