IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 33, NO.

2, MARCH/APRIL 1997

333

Direct Torque Control, Induction Motor

Vector Control Without an Encoder
James N. Nash, Member, IEEE

Abstract The basic evolution of direct torque control from

other drive types is explained. Qualitative comparisons with
other drives are included. The basic concepts behind direct
torque control are clarified. An explanation of direct self control
and the field-orientation concepts implemented in the adaptive
motor model block is presented. The reliance of the control
method on fast processing techniques is stressed. The theoretical
foundations for the control concept are provided in summary
format. Information on the ancillary control blocks outside the
basic direct torque control is given. The implementation of special
functions directly related to the control approach is described.
Finally, performance data from an actual system are presented.
Index Terms AC adjustable-speed drives, direct torque control, motor control, sensorless vector control.

I. INTRODUCTION

HE EVOLUTION from dc drives to various forms of

ac drives has been motivated by the continuing need
for simultaneous performance, simplicity, and reliability. In
practice, these needs have often proved to encompass mutually
exclusive goals. Therefore, new developments often strive to
improve the available mix of advantages versus disadvantages. Direct torque control represents a recent step in mix
improvement. Fig. 1 shows drive evolution as a basic four-step
process. This viewpoint is simplistic, but useful for comparison
purposes.

Fig. 1. Evolution of drive control techniques.

B. Scalar Frequency Control

A. DC Drive
Torque is directly proportional to armature current in the
dc motor. By using an inner-current control loop, the dc drive
can directly control torque. Likewise, the constant magneticfield orientation, which is achieved mechanically through
commutator action, makes direct flux control a given. Thus,
two primary factors toward insuring responsive control (i.e.,
direct torque control and direct flux control) are both present
in the dc drive. The relatively simple electronics required to
implement the dc drive represents another advantage. On the
negative side, both the initial and maintenance costs of dc
motors are high, and high performance speed accuracy can
only be achieved if an encoder is included for feedback.

All of the ac drives compared here allow the use of economical, robust ac induction motors. Scalar frequency control also
offers the advantage of operation without an encoder. On the
negative side, torque and flux are neither directly nor indirectly
controlled. Control is instead provided by a frequency and
voltage reference generator with constant volts per hertz
output, which then drives a pulsewidth-modulated (PWM)
modulator. Although simple, this arrangement provides limited
speed accuracy and poor torque response. Flux and torque
levels are dictated by the response of the motor to the applied
frequency and voltage and are not under the control of the
drive.
C. Flux Vector Control

Paper PID 9636, approved by the Pulp and Paper Industry Committee of
the IEEE Industry Applications Society for presentation at the 1996 IEEE Pulp
and Paper Industry Conference, Birmingham, AL, June 1014. Manuscript
released for publication October 6, 1996.
The author is with ABB Industrial Systems Inc., Milwaukee, WI 532010372 USA.
Publisher Item Identifier S 0093-9994(97)01659-9.

Flux vector control reestablishes one of the advantages of

the dc drive through implementation of direct flux control.
In this case, field orientation is controlled electronically.
The spatial angular position of rotor flux is calculated and
controlled by the drive, based on a relative comparison of
the known stator field vector to feedback of rotor angular

00939994/97$10.00 1997 IEEE

334

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 33, NO. 2, MARCH/APRIL 1997

TABLE I
COMPARISONS OF CONTROL TYPES

position and speed. The motors electrical characteristics are

mathematically modeled with microprocessor techniques to
enable processing of the data. Torque control is indirect
because of its position in the control algorithm prior to the
vector control process, however, good torque response is
achieved nonetheless. Inclusion of the pulse encoder insures
high-performance speed and torque accuracy.
The biggest disadvantage of flux vector control is the
mandated inclusion of the pulse encoder. Another minor
disadvantage is that torque is indirectly, rather than directly,
controlled. Finally, the inclusion of the PWM modulator,
which processes the voltage and frequency reference outputs
of the vector control stage, creates a signal delay between
the input references and the resulting stator voltage vector
produced. These last two factors limit the ultimate ability
of flux vector control to achieve very rapid flux and torque
control.
D. Direct Torque Control
Direct torque control also reestablishes direct flux control. In
addition, direct torque control is implemented. Both flux and
torque are controlled by a hysteresis controller. The delays
associated with the PWM modulator stage are removed, since
the PWM modulator is replaced by optimal switching logic.
The original benefits associated with the dc drive of direct
torque control, direct flux control, and high responsiveness
are, thus, all reestablished. Torque response is better than that
available with either dc or flux vector control. In addition,
assuming moderate speed accuracy is acceptable (typically
0.1%0.3%, or 10% of motor slip) the need for a pulse encoder
is eliminated. The theory and operation of direct torque control
is described in subsequent sections of this paper.
Table I provides a summary comparison of the four drives
described above.
II. DIRECT TORQUE CONTROL CONCEPT
Direct torque control has its roots in field-oriented control
and direct self control. Field-oriented control uses spatial vector theory to optimally control magnetic field orientation. It has
been successfully applied to the design of flux vector controls

and is well documented. Direct self-control theory is less well

known. It is a patented concept developed in Germany by
Manfred Depenbrock and has been described in several papers
[2], [3] which he has published. The fundamental premise of
direct self control is as follows.
Given a specific dc-link voltage
and a specific stator
flux level
a unique frequency of inverter operation
is established.
This is true because the time
required by the time
integral of the voltage
to integrate up to the fieldflux level
is unique and represents the half-period time
of the frequency of operation
Since the operational
frequency
is established without a frequency reference,
this operational mode is referred to as direct self control.
Output frequency is, thus, not requested, but rather, is self
controlled via the actual frequencies present. Once sensed,
whether the frequency increases or decreases depends on
what the torque reference from the speed regulator requests.
Differential changes to operational frequency are determined
by the torque request. Nominal operating frequency is self
determined via the sensed feedbacks.
Direct torque control combines field-oriented control theory,
direct self-control theory, and recent advances in digital signal
processor (DSP) and application specific integrated circuit
(ASIC) technology to achieve a practical sensorless variablefrequency drive.
Fig. 2 shows the basic functional blocks used to implement
the core of the direct torque-control scheme. The relationship
of this core to the complete control will be described in
a subsequent section. Three key blocks interact to provide
the primary control required: 1) torque/flux comparators; 2)
optimal switching logic; and 3) adaptive motor model.
A. Torque/Flux Comparators
The torque comparator and the flux comparator are both
contained in the hysteresis control block. These function to
compare the torque reference with actual torque and the flux
reference with actual flux. Actual levels are calculated by
the adaptive motor model. When actual torque drops below
its differential hysteresis limit, the torque status output goes
high. Likewise, when actual torque rises above its differential

NASH: DIRECT TORQUE CONTROL, INDUCTION MOTOR VECTOR CONTROL

335

Fig. 2. Basic direct torque-control scheme.

hysteresis limit, the torque status output goes low. Similarly,

when actual flux drops below its differential hysteresis limit,
the flux status output goes high, and when actual flux rises
above its differential hysteresis limit, the flux status output
goes low. The upper and lower differential limit switching
points for both torque and flux are determined by the hysteresis
window input. This input is used to vary the differential
hysteresis limit windows, such that the switching frequencies
of the power output devices are maintained within the range
of 1.53.5 kHz.
B. Optimal Switching Logic
Processing of the torque status output and the flux status
output is handled by the optimal switching logic contained in
the ASIC block. The function of the optimal switching logic is
to select the appropriate stator voltage vector that will satisfy
both the torque status output and the flux status output. In
reality, there are only six voltage vectors and two zero-voltage
vectors that a voltage-source inverter can produce. These are
shown in Fig. 3.
The analysis performed by the optimal switching logic
is based on the mathematical spatial vector relationships of
stator flux, rotor flux, stator current, and stator voltage. These
relationships are shown as a vector diagram in Fig. 4.
The torque developed by the motor is proportional to the
cross product of the stator flux vector
and the rotor flux
vector
The magnitude of stator flux is normally kept
as constant as possible, and torque is controlled by varying
the angle
between the stator flux vector and the rotor flux
vector. This method is feasible because the rotor time constant
is much larger than the stator time constant. Thus, rotor flux is
relatively stable and changes quite slowly, compared to stator
flux.
When an increase in torque is required, the optimal switching logic selects a stator voltage vector
that develops
a tangential pull on the stator flux vector
, tending to

rotate it counterclockwise with respect to the rotor flux vector

The enlarged angle ( ) created effectively increases the
torque produced. When a decrease in torque is required, the
optimal switching logic selects a zero-voltage vector, which
allows both stator flux and produced torque to decay naturally.
If stator flux decays below its normal lower limit the flux
status output will again request an increase in stator flux.
If the torque status output is still low, a new stator voltage
vector
is selected that tends to increase stator flux while
simultaneously reducing the angle
between the stator and
rotor flux vectors.
Note that the combination of the hysteresis control block
(torque and flux comparators) and the ASIC control block
(optimal switching logic) eliminate the need for a traditional
PWM modulator. This provides two benefits. First, small
signal delays associated with the modulator are eliminated;
and second, the discrete constant carrier frequencies used by
the modulator are no longer present. The latter benefit can be
readily discerned when observing a running drive, since the
characteristic gear-shifting sounds associated with a PWM
inverter no longer exist. The sound remaining is a white noise
that is less obtrusive to the listener.

C. Adaptive Motor Model

With reference to Fig. 2, it can be seen that the adaptive
motor model is responsible for generating four internal feedback signals: 1) actual flux (stator); 2) actual torque; 3) actual
speed; and 4) actual frequency. The first two values, which are
critical to proper direct torque control operation, are calculated
every 25 s. The latter two values, which are used by outer
loop controllers, are calculated once per millisecond.
Dynamic inputs to the adaptive motor model include: 1)
motor current from two stator phases; 2) link voltage; and 3)
power switch positions. Static motor data is also utilized in
making calculations. The static data come from two sources:

336

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 33, NO. 2, MARCH/APRIL 1997

Fig. 3. Available stator voltage vectors.

Fig. 5. Estimated air-gap torque with hysteresis control.

D. Torque Production
Fig. 4. Stator spatial vector relationships.

1) user input data and 2) information determined automatically

from a motor identification run that occurs during commissioning. The user input data include motor nominal voltage, motor
nominal current, motor nominal frequency, motor nominal
speed, and motor nominal power. The data collected during
the motor identification run include motor inductances, stator
resistances, and stator saturation effects.
The exact mathematical details of how the adaptive motor
model calculates its outputs are beyond the scope of this paper;
however, Section III is included to provide an overview of
the mathematical foundations upon which the calculations are
based.

The estimated air-gap torque

produced by the drive
is shown in Fig. 5. The periods during which torque has a
in Fig. 5,
positive slope, such as the region indicated as
indicate intervals where an active stator voltage vector is
causing torque to increase. The periods during which torque
in
has a negative slope, such as the region indicated as
Fig. 5, indicate intervals where a zero-voltage vector has been
selected. During most periods, either a simple positive slope
or simple negative slope is indicated. During these periods, a
single stator voltage vector selection has been able to satisfy
both the torque status and the flux status output. However,
identifier,
during some periods, such as the region near the
a dual slope is present. This is indicative of an interval during
which, after one voltage vector has been chosen, and although

NASH: DIRECT TORQUE CONTROL, INDUCTION MOTOR VECTOR CONTROL

additional torque still needs to be developed (i.e., the torque

status output is not satisfied), a change in the flux status output
has occurred that requires selection of another stator voltage
vector.
Normally, switching occurs whenever torque drops below
or exceeds
In the special case where
torque exceeds
, the optimal switching logic selects
a stator voltage vector
that forces a decrease in torque
by causing a reduction in the angle
between the stator flux
vector and the rotor flux vector.
Due to the hysteresis control, the torque appears to have a
choppy waveform. In practice, this is not a concern, since the
average period of the waveform is under 1 ms and substantially
less than the motors mechanical time constant.
The fact that all of the described spatial vector comparisons
occur once every 25 s is a key factor in the operation of the
drive. Since the maximum power device switching frequency
is 3.5 kHz, ten or more torque and flux comparisons are
possible per switching decision. Without this kind of datato-decision ratio, stable operation wouldnt be possible. It is
only due to recent advances in DSP technology that this type
of operation has become practical.

337

TABLE II
FUNDAMENTAL EQUATIONS FOR CALCULATING MOTOR VARIABLES

E. Initial Starting
The statement that the operational frequency
is established without using a frequency reference, or otherwise
directly controlling frequency, was made at the beginning of
this section. This followed from the fact that direct torque
control utilizes direct self-control as a part of its control
scheme. Subsequent discussion has described how this is
accomplished in a running system. The issue of system initialization, however, has not been addressed. How the drive
can establish a correct start mode, when feedback data is not
available for the adaptive motor model, is a valid question.
In reality, the drive does not initially know what frequency
to operate at. Only after the adaptive motor model analyzes
the current and switch position feedbacks can a determination
of flux, torque, speed, and frequency be made. To obtain the
initial feedbacks, a low-level dc flux is established in the
motor. This level is low enough that overcurrent tripping and
inappropriate torques are not created and high enough that
meaningful feedback data is received. Once valid feedback
data is available, operation proceeds as previously described.

III. THEORETICAL FOUNDATIONS

The DSP is responsible for calculating all motor variables
required by other portions of the direct torque control circuitry.
These calculations are based on basic motor theory and fieldoriented control theory. While not intending to provide a
complete understanding of the mathematics and theory involved, Table II is included to provide a general feel for
the basis of operation. Calculated variables are listed in the
order that they are typically determined. Note that many of the
variables used are vector quantities. These quantities are shown
in the fundamental equations with a bar above the appropriate
variable name.

Determination of actual flux (stator) is always of first

and primary concern. Note that two equations are listed.
Both equations are utilized by the control. The first equation
provides a first-order estimate of the flux level present. The
second equation is used to effectively fine tune the value
provided by the first. The next equation calculates actual
torque. Since both actual flux and actual torque are required
directly by the hysteresis control comparators, these equations
are solved every 25 s.
The remaining four equations are used to calculate actual
frequency and actual speed, which are used by the outer loops
of the control. Since the outer loops are not as time critical,

338

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 33, NO. 2, MARCH/APRIL 1997

Fig. 6. Complete direct torque control drive.

these values are determined only once per millisecond. The

rotor flux vector and rotor flux angle variables provide values
needed to calculate actual frequency and actual speed, but are
not used directly, elsewhere in the control scheme.

IV. THE COMPLETE DRIVE

Circuitry, in addition to that shown in Fig. 2 and described
in Section II, is required to build a complete drive. Fig. 6
shows the block diagram for a complete direct torque control
drive. Blocks for speed control, torque reference control, flux
reference control, and switching frequency control have been
added.
The speed control block is implemented with a traditional
proportional integral and differential (PID) controller. The
speed reference input is compared to the actual speed feedback
from the adaptive motor model. The resulting output signal,
torque (speed) reference, becomes the reference for the torque
reference control. An inertia compensator is also included,
although not shown. The compensator helps reduce control
deviation due to acceleration or deceleration of system inertia.
It also allows the speed control to be primarily tuned based
on changes in static load. Automatic tuning is provided.
The torque reference control has two inputs, the torque
(speed) reference and absolute torque reference. When the
drive functions as a speed control, only the torque (speed)
reference is used. When the drive functions as a torque control,
only the absolute torque reference is used. These references are
never used simultaneously. The output, torque reference, is
processed by the direct torque control as described in Section
II.
The output of the torque reference control provides automatic limiting based on any of four independent system
occurrences: 1) if during deceleration the dc-link voltage starts
to exceed the set maximum level; 2) if maximum permitted

electrical frequency is reached; 3) if requested torque reaches a

level near motor pullout torque (which is dependent on motor
characteristics); and 4) if current overload is detected. The last
of these limits, current overload, is a function of time where
very high current (typically 200%) is permitted for a short time
period ( 2 s), and high current (typically 150%) is permitted
for a moderate time period ( 1 min).
The flux level reference and actual frequency feedback are
the primary inputs to the flux reference control. The flux
level reference is a parameter which can be set by the user.
The actual frequency feedback is used to provide frequencysensitive flux manipulation for field-weakening control, flux
braking, and flux optimization. The latter two functions are
described in the next section. The output, flux reference, is
processed by the direct torque control, as described in Section
II.
The last additional block is the switching frequency control.
The power output devices have limitations on their maximum
switching frequency. Likewise, the desired output waveform
places restrictions on the minimum desired switching frequency. The function of the switching frequency control is
to vary the size of the hysteresis windows utilized by the flux
comparator and torque comparator, such that power switching
frequencies are maintained between 1.53.5 kHz. The output,
hysteresis window, is processed by the direct torque control
as described in Section II.

V. IMPLEMENTATION OF SPECIAL FUNCTIONS

The implementation of several special functions and control
properties is made easier because direct torque control works
directly to control stator flux. The included special functions
are flying start, flux braking, flux optimization, and power loss
ride-through.

NASH: DIRECT TORQUE CONTROL, INDUCTION MOTOR VECTOR CONTROL

339

Fig. 7. 70% torque reference step at 25 Hz. Estimated air-gap torque. No encoder.

Fig. 8. Full speed fast reversing with constant torque load (20%). Measured phase current. No encoder.

TABLE III
TEST SYSTEM EQUIPMENT DATA

dc current, which is injected into the stator. Once the status

of stator flux is known, full synchronization of the inverter
with the motor is immediate. This makes direct torque control
ideal for flying-start applications. Proper operation is possible
regardless of motor speed or direction prior to synchronization.
An alternate mode of starting, called dc magnetization is
also available. In this mode, an optimized value of dc flux of
relatively high level is immediately impressed on the stator.
This minimizes the time required by the adaptive motor model
to gather all needed data from a stationary motor. Full torque
can typically be available within 150400 ms, dependent on
motor size.

A. Flying Start

B. Flux Braking

The adaptive motor model is capable of determining the

status of stator flux and all other required motor variables
within 200500 ms, assuming that all residual flux in the motor
is absent. If residual flux exists, the determination occurs much
faster. This resolution is accomplished based on a low-level

It is common to inject dc into one or more stator windings

to provide braking of an ac drive. This is effective, but
is accompanied by a required delay, to allow the flux to
decay both before the dc can be applied and afterwards,
before normal ac can be reapplied. Direct torque control uses

340

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 33, NO. 2, MARCH/APRIL 1997

Fig. 9. Slow reversing with constant torque load (80%). Measured shaft torque and speed. No encoder.

Fig. 10.

Torque ramp from

Tnom

to

0Tnom at zero speed. Measured shaft torque. No encoder.

TYPICAL TORQUE

TABLE IV
SPEED PERFORMANCE DATA

AND

a different method to achieve similar results. The stator is

overexcited in a controlled manner, to allow the braking energy
losses. Since direct torque
to dissipate in the stator as
control directly controls stator flux, this is a straightforward
approach. In addition, because the flux continues to be applied
at the appropriate excitation frequency, there is no delay
required to either initiate this method or to reinitiate the

normal powering mode. Thus, this method of braking can be

used dynamically to slow the motor between any two normal
operating points with immediate transfer back to normal
powering mode. It should be noted, however, that this method
is primarily useful at lower speeds, since the necessary voltage
is not available to appreciably overexcite the stator at higher
frequencies.

NASH: DIRECT TORQUE CONTROL, INDUCTION MOTOR VECTOR CONTROL

C. Flux Optimization
A lightly loaded motor does not need full stator flux
to produce the required torque. Direct torque control takes
advantage of this by selecting an optimal magnetizing level
based on load. When full torque is required, full stator flux
is requested. At reduced load levels, a reduced level of stator
flux is developed. An unloaded motor may run with as little
as 50% of its nominal magnetizing current. Dependent on the
application, this may lead to significant reductions in motor
heating and improvements in overall efficiency.
D. Power Loss Ride-Through
If the power feed to a direct torque control drive is lost,
the system automatically enters a regenerative mode that
maintains the dc-link voltage at its minimal operating level
as long as possible. The energy to accomplish this comes
from the mechanical inertia stored in the load and motor.
As long as rotational energy continues to exist, the link
voltage is maintained. If the power feed comes back on line
before rotation ceases, the drive will immediately return to the
operating point at which it was running prior to the loss of
power. The time that this power loss ride-through capability
can be maintained is extremely load dependent. A high-inertia
system which is lightly loaded can ride through a power loss of
a minute or more. On the other hand, a low-inertia system that
is fully loaded may only be able to stay on line for 200300
ms. This feature can be disabled if safety concerns dictate
doing so.

VI. PERFORMANCE DATA

Performance data was taken with the test system listed
in Table III. Figs. 710 show actual results. Note that all
results were taken without a shaft encoder. Fig. 7 shows the
extremely fast torque response possible with direct torque
control. Torque responses below 2 ms are typical, even in large
motors. Speed response is also excellent, as shown in Fig. 8.
Note the stability of the current throughout the entire reversing
process. Fig. 9 demonstrates the smooth speed response right
through zero speed, with a slow reversing reference and 80%
constant torque loading. Regenerative energy was dissipated
using a braking chopper connected across the dc link. Fig. 10
demonstrates the smooth torque control possible, even at zero
speed. Torque linearity error is below 4%.
Table IV shows typical torque and speed performance data
for direct torque control with and without an encoder. Typical

341

values for a dc drive, scalar frequency control, and flux vector

control are also included for comparison.
VII. CONCLUSION
Direct torque control combines the benefits of direct flux and
direct torque control into a sensorless variable-frequency drive
that does not require a PWM modulator. Recent advances in
DSP and ASIC technology, plus the theoretical concepts developed for direct self control, make this possible. Implementation
of special functions such as flying start, flux braking, flux
optimization, and power loss ride-through are all made easier,
due to the control approach utilized. For many applications,
exceptional performance can be realized without an encoder.
For those applications that require high-performance speed
accuracy, an encoder can be added to achieve speed accuracy
equivalent to competitive approaches. In addition, with the
encoder, dynamic speed and torque response are better than
any alternate approach. Direct torque control represents a
definite step forward in the mix of drive controls to choose
from.
ACKNOWLEDGMENT
Major portions of this paper are based upon the research
of Tiitinen, Pohjalainen, and Lalu [1]. The author gratefully
acknowledges the importance of their work.
REFERENCES
[1] P. Tiitinen, P. Pohjalainen, and J. Lalu, The next generation motor
control methodDirect torque control, DTC, in Proc. EPE Chapter
Symp., Lausanne, Switzerland, 1994.
[2] M. Depenbrock, Direct self-control (DSC) of inverter-fed induction
machine, IEEE Trans. Power Electron., vol. 3, pp. 420429, Oct. 1988.
[3] U. Baader and M. Depenbrock, Direct self control (DSC) of inverterfed induction machine: A basis for speed control without speed measurement, IEEE Trans. Ind. Applicat., vol. 28, pp. 581588, May/June
1992.

James N. Nash (M91) received the B.S.E.E. degree from Marquette University, Milwaukee, WI, in
1970.
From 1970 to 1983, he worked first as a Development Engineer and later as Application Engineering
Manager in the Electric Products Division, Harnischfeger Corporation. In 1983, he joined what is
now ABB Industrial Systems Inc., Milwaukee, WI,
where he has served as Systems Engineering Manager and in various engineering staff positions. He
currently holds the position of Principal Engineer.
His primary areas of interest include ac and dc motor drives, power electronics,
and control systems.