Professional Documents
Culture Documents
DOI 10.1007/s00202-003-0214-2
O R I GI N A L P A P E R
Received: 11 August 2003 / Accepted: 3 November 2003 / Published online: 23 December 2003
Ó Springer-Verlag 2003
Abstract This paper presents a new torque control The concept of direct torque control (DTC) [1, 2, 3]
algorithm for induction motors, based on the stator flux algorithms for induction motors was introduced, sig-
vector control. For each sampling period, the value of the nificantly improving the torque control performance
stator voltage is calculated to keep the stator flux equal to with algorithms that achieve fast and robust torque
the reference vector, while the stator flux reference vector control of IM without use of the current controllers and
is calculated to keep the rotor flux amplitude constant at shaft position sensors. The DTC control strategies em-
all operating conditions. An improved stator and rotor body algorithms in which the voltage source inverter
flux estimation algorithm is proposed, enabling robust (VSI) switching patterns are generated directly as a
and stable operation of the drive, even at low speeds. The function of the torque and flux errors (with the corre-
induction motor torque is manipulated by variations of lation between the torque, flux error, and switching
the flux angular velocity, enabling drive operation with states defined by means of the switching table combined
fixed switching frequency and ripple-free torque in the with the sliding mode control strategy). The basic DTC
steady state. The performance of the proposed algorithm concept with a hysteresis regulator [1, 2] results in a very
is tested through various experimental runs, proving fast torque response, unforeseen in conventional drives
good behavior of the drive in both transient and steady- with a current controller in the minor loop. Nonlinear
state operating conditions. hysteresis-based control results in a high level of torque
ripple and in irregular inverter switching, with the
Keywords Sensorless torque control Æ Flux vector commutation frequency dependent upon the modulation
control Æ Induction motor index, speed, flux, and current level. Various attempts to
minimize the torque ripple and switching frequency
variations where done by using the switching tables with
adaptive sliding mode control [3].
1 Introduction Another attempt to minimize the torque chattering
and switching frequency variations was introduced by
The efforts in developing novel induction motor (IM) the stator flux vector control (SFVC) schemes [4, 5, 6].
control algorithms are focused on drives with the mini- The aim of SFVC is to drive the estimated stator flux
mum number of sensors required for operation. In re- vector toward the reference value by indirect control of
cent years, a wide variety of speed sensorless solutions the inverter switching states, through pulse-width mod-
has been proposed, contributing performance increase at ulation (PWM). The SFVC algorithms are based on
higher speeds, but failing to improve the drive behavior various control strategies (‘‘dead-beat’’, feed-forward,
below 1% of the rated speed. PI, sliding mode controllers, etc.), with the stator flux
error used as the input variable. The output of SFVC
scheme is fed into the space vector modulator (SVM),
D. Stojic (&) achieving the DTC with constant switching frequency
Electrical Institute INT, and smooth torque and flux waveforms.
Golsvortijeva 13, 11000 Belgrade,
Serbia and Montenegro
Papers [4, 5, 6] present three different SFVC schemes,
E-mail: djs@ieent.org proving that the direct linear control of torque and
stator flux vectors enables fast torque dynamics, im-
S. Vukosavic
Electrical Engineering Faculty of Belgrade, proved stator flux estimation at low speeds, and ripple-
Bulevar Kralja Aleksandra 73, 11000 Belgrade, free drive operation. The common drawback of these
Serbia and Montenegro SFVC algorithms is that they rely on the calculation of
24
the field vector position from the estimated stator flux torque control loop derives the torque error and in-
vector and on the estimation of field velocity by differ- creases/decreases the speed of the field rotation. Varia-
entiation of the calculated field vector angle. This leads tions of the field rotation velocity have an impact on the
to an increase of drive sensitivity to the measurement slip frequency that will, in turn, result in a desired torque
noise and to a decrease in drive performance, especially change and steer the torque error toward zero.
at low speeds (with poor field angle estimation). Also, The proposed flux estimator and associated control-
SFVC drives [4] and [6] exhibit high sensitivity to the IM ler are located in the stationary reference frame. The
parameter variations, since these algorithms are de- explicit flux control deals with the stator flux in order to
signed in the dead-beat fashion. achieve a faster response and boost the robustness
In this paper, a novel SFVC algorithm is proposed against changes of IM parameters. On the other hand,
for controlling the flux and torque of induction motors the stator flux reference is calculated so as to take into
operating without shaft sensors. The improvements account the feed-forward component, related to the
introduced in this SFVC algorithm include decreased leakage flux. In turn, the rotor flux amplitude is
sensitivity to IM parameter variations and improved guaranteed to be constant in the steady-state and
stability of drive operation, even at speeds close to zero. constant-torque operating mode, contributing to the
The increase in the drive performance is achieved by the ease of torque control by making the slip-torque relation
following changes in the SFVC algorithm: implementa- linear. By operating in the constant-torque mode, the
tion of a simplified feed-forward stator flux regulator proposed solution ensures a low torque ripple and a
with a reduced set of control parameters, the direct constant switching frequency. The algorithm requires
manipulation of the field velocity (as opposed to solu- setting a relatively small number of parameters and is
tions that rely on the field position and field velocity simple to implement.
estimation), and by using the closed-loop estimator of Figure 1 shows the block diagram of the proposed
the stator and rotor flux vectors. DTC algorithm. The algorithm inputs are the reference
The proposed solution is verified through a set of rotor flux vector w*DQ, reference torque T*e, and stator
experimental tests on a setup having a 10-HP 4-pole current iabs. The main control loop consists of the torque
industrial motor. The results obtained confirm the controller and block for the stator flux reference calcu-
ability of the proposed controller to ensure both torque lation. The amplitude of the stator flux reference is
and speed control in any practical operation condition, calculated from the rotor flux and torque references,
including standstill. ensuring a constant rotor flux amplitude in the constant-
torque operating mode. The rotational frequency xe of
reference vector w*abs is determined by using the torque
2 Basic drive operating principle PI controller, in order to drive the torque error toward
zero. Consequently, the angle of the stator flux reference
In the proposed algorithm, the torque control relies on vector w*abs is calculated by integration of xe.
the fundamental behavior of the squirrel cage induction The local inner loop, within the dotted rectangle,
motor. Namely, the torque generation is proportional to represents the linear stator flux controller. The flux
the square of the rotor flux and slip frequency, and is controller is realized in the stationary reference frame,
inversely proportional to the rotor resistance. Hence, the with zero error of the stator flux vector in the stationary
The observer gain G is chosen according to the cri- references, it is necessary to design a control structure
terion that stable drive operation is to be maintained for that guarantees zero error signals in the stationary state.
variations of Rs within the range of 25% around its To achieve zero steady-state error, the feed-forward
nominal value, where this region may be extended if control structure is included in the stator flux regulation.
necessary. Since the analytical evaluation of G that Figure 2 shows the stator flux regulator, together with
matches this criterion is rather complicated, a more the space vector pulse-width modulator and dead-time
convenient procedure for setting G is to tune its value on compensation.
the experimental setup in the following manner. First, The flux regulator equation is expressed as
the value of Rs is intentionally detuned for 25% of its
nominal value in Eqs. (6) and (10). Then, for the de- vabs ðk Þ ¼ Rs is ðk Þ þ jxe ðk Þwabs ðk Þ
h i
tuned value of Rs, the value of G is increased until stable þ Kp wabs ðk Þ w ^ ðk Þ : ð10Þ
abs
drive operation is obtained. Finally, the value of Rs in
Eqs. (6) and (10) is returned to its nominal value, after The first two terms on the right-hand side in Eq. (10)
the proper value of parameter G is chosen in the second represent the feed-forward control actions, aimed at
step. By using the above three-step procedure, the compensating the voltage drops across the stator resis-
appropriate value of observer gain G is achieved that tance and stator back-EMF. Notice, the feed-forward
guarantees stable drive operation with respect to an action for the back-EMF is proportional to reference
undesirable influence of stator resistance variations. stator flux vector w*abs, opposite to the existing SFVC
The flux observer, Eqs. (6) and (7), enables stable algorithm [5], which includes feed-forward action for the
drive operation, even at very low speeds. It makes the back-EMF proportional to the estimated stator flux vec-
drive less sensitive to variations of the drive parameters tor. Consequently, the back-EMF compensation pro-
and nonlinearities of the voltage inverter. posed in [5] introduces a flux feedback signal with variable
gain (flux velocity xe), which causes undesirable varia-
tions in the controller dynamics. In Eq. (10), the variable
4 The stator flux regulator flux feedback signal is avoided by introducing back-EMF
compensation proportional to vector w*abs.
The stator flux regulator determines the stator voltage The first two feed-forward terms in Eq. (10) deter-
reference value, which is fed into the PWM block. The mine the steady-state value of the voltage command,
voltage value is calculated in order to achieve the stator which guarantees zero error for a given stator flux ref-
flux equal to reference vector w*abs within a finite set- erence vector w*abs. The third term in Eq. (10) represents
tling time. In doing so, the control is designed to achieve the feedback control action, which is proportional to the
zero flux error in less than three sampling periods (set- stator flux error. In Eq. (10), K p is the gain of the flux
tling time is tst=2.5 T s, where Ts is the digital controller regulator and xe ( k) represents the instantaneous stator
sampling period, Ts=200 ls). flux velocity ( xe ( k) is determined by the outer torque
Since the flux regulator is designed in the stationary control loop). Notice that in Eq. (10) only one param-
reference ab frame, with sinusoidal variable frequency eter, Kp, is to be set, which makes the proposed design
procedure simple to implement.
The feed-forward term in Eq. (10), which compensates
the voltage drop across the stator resistance, can be
Fig. 2 Block diagram of stator flux regulator
27
found in various SFVC algorithms [4, 5]. This feed- torque control by introduction of a linear controller that
forward action represents the positive feedback path for manipulates the torque by using the field velocity as a
the stator current, thus causing an increased sensitivity of command variable.
the drive to the nonlinearities of the VSI and to variations The proposed torque regulation relies on the oper-
of R s. As discussed in the previous section, the problems ating principle of IM that the torque is directly pro-
related to the positive current feedback, inherent for basic portional to the slip frequency and to the squared
DTC algorithms, are resolved by introducing a negative amplitude of the rotor flux. This statement is proven by
current feedback indirectly through the closed-loop stator the following set of equations. Namely, flux velocity xe
flux estimation algorithm. represents the rate of change of rotor flux angle he.
Since one of the objectives of the proposed DTC Angle h e is defined by
algorithm is to enable drive operation with constant
rotor flux amplitude, it is necessary to calculate the wbr
he ¼ arctan ; ð15Þ
adequate stator flux reference for each operating con- war
dition. By using the rotational reference frame model in
and the flux angular velocity can be calculated as
Eqs. (1)–(5) of IM, under the condition that the rotor
flux vector is constant, the stator currents in the sta- dhe war w_ br wbr w_ ar
tionary state can be derived from the electromagnetic xe ¼ ¼ 2 : ð16Þ
torque. Hence, for the rotor flux and torque references dt
wabr
w* Q =0, T* e, and w* D, the reference drive stator
currents are calculated by Using Eq. (2) the following relation is obtained:
2Lr 2 2
iqs ¼ T ; ð11Þ
w
3p abr 3p abr
w
3pLm wD e Te ¼ ðx e x r Þ ¼ xs ; ð17Þ
2 Rr 2 Rr
wD
ids ¼ : ð12Þ where x s is slip frequency.
Lm
The estimated torque signal, used as the input to the
From Eqs. (7), (11), and (12), the stator flux reference torque regulator, is calculated by
vector is expressed as
^ 3 _ ^
Te ¼ p w as ibs wbs ias : ð18Þ
M 2
wabs ¼ w þ rLs iabs ; ð13Þ
Lr abr
The torque controller is designed to generate both the
where vector w*abr is given by Eq. (8), and vector i*abs is transient and steady-state components of the slip
calculated from i* dqs by using the following inverse velocity. The transient component determines the set-
rotational transformation: tling time of torque, while the steady-state component
determines the torque steady-state value. The PI torque
ias cos ðhe Þ sin ðhe Þ ids controller applied is defined by the following expression:
¼¼ : ð14Þ
ibs sin ðhe Þ cos ðhe Þ iqs
xe ðk Þ ¼ xe ðk 1Þ þ KT 1 DT^e ðk Þ KT 2 DT^e ðk 1Þ ;
In Eq. (14), the value of angle h e is given by Eq. (9).
ð19Þ
The ‘‘dead-time’’ compensation is achieved by using
the algorithm proposed in [8]. The compensation is where KT 1 and KT2 represent the control parameters,
based on the calculation of the volt-seconds lost in the and DT^e ðk Þ represents the torque error, DT^e ðk Þ ¼
blanking period, averaged over the switching cycle. Te T^e ðk Þ. The torque controller works with a sampling
Since the voltage distortion introduced by the blanking period of 200 ls. The conventional PI regulator of
time is of opposite sign to the phase current, the com- Eq. (19) enables a fast transient response of torque and
pensation voltage added to the commanded voltage is of determines the appropriate steady-state value of the flux
the same sign as the appropriate phase current. angular velocity, which varies with the torque reference
The proposed linear controller enables fast and robust and operating speed of IM.
stator flux regulation, which, together with the torque
regulator, yields zero steady-state errors of the rotor flux
and torque for given rotor flux and torque references. 6 Experimental tests
estimation, torque estimation and control, anti-windup distortions, while the torque matches the set value.
integrators, limiters, and other diagnostic facilities), with Since, in this experiment, the drive operates at very a low
sampling period Ts=200 ls. frequency, the presented results show that the proposed
The induction machine under test was coupled to a control algorithm keeps the machine magnetized, even
separately controlled DC machine, used as a dynamic at zero speed.
break. The 4-pole squirrel cage induction motor was Figure 5 presents the results for the mid-speed region,
characterized by the following data: P=7.5 kW, with torque set to 0.4 pu and rotor speed set to 0.5 pu by
n =1500 rpm, V=220 V, f=50 Hz. The steady-state a DC motor coupled with the IM. Figure 6 shows tor-
and transient behavior of the drive was investigated by que and stator current behaviour for torque set to 0.2 pu
various sets of tests. and speed set to 0.8 pu. The experimental results in
Figs. 5 and 6 show that in the mid- and high-speed re-
gions stator current is sinusoidal with small distortions,
6.1 The steady-state operating conditions while torque matches the reference value.
The results presented in this section show that the
The steady-state behaviour was investigated in different proposed torque control technique enables the zero
operating conditions, i.e. with locked rotor, and for mid- torque error operation in the stationary state in a wide
and high-speed regions. The drive behaviour in operat- range of operating speeds. This proves that zero torque
ing conditions involving low flux angular velocities was error operation can be achieved by using the proposed
tested by using the locked rotor tests. Figure 3 shows the PI control strategy, which does not require precise
current and torque behaviour for the locked rotor, with knowledge of motor parameters. This is an improvement
the rotor flux and torque set to their rated values. The compared to the existing SFVC solutions [4, 6], which
stator current is sinusoidal, while the estimated torque require accurate knowledge of motor parameters in or-
matches the reference value. The experimental results in der to generate adeqate command values for zero torque
Fig. 3 prove that the proposed PI torque control algo- error operation.
rithm enables zero torque error signal in the stationary
state. This shows that precise torque regulation can be
achieved with simple PI control strategy, contrary to the 6.2 Transient operating conditions
‘‘dead-beat’’ structure based on solutions that require
knowledge of exact IM parameter values to calculate the The transient behavior of the proposed DTC algorithm
adequate regulation commands (cf. [4] and [6]). was investigated for locked rotor and the low- and mid-
Figure 4 shows the test results for the torque refer- speed regions. Figure 7 shows torque and stator current
ence set to 0.4 pu. The current waveform exhibits some responses for locked rotor and for torque reference step
change from 0.1 to 0.2 pu. The results show that the prove that achieved torque dynamic is faster when
response time of torque for the step excitation equals compared to the existing SFVC strategies [4, 5, 6] with
4–5 sampling periods ( T s=200 ls), with zero steady- torque response times above 3 ms. Figure 8 represents
state error signal. Moreover, the presented experimental the torque response for a step excitation for rotor speed
results show that the proposed DTC algorithm has re- set to 0.2 pu, while the results in Fig. 9 correspond to
sponse times comparable with the ones achieved with the speed set to 0.4 pu. The measurements presented in
basic DTC algorithm [1, 2]. Also, the presented results Figs. 7, 8, and 9 show that the torque regulator retains
the same response times in a wide range of operating cant distortions in the speed waveform. This experiment
speeds. These test results prove that the proposed DTC was performed in order to enable further comparison of
algorithm is robust in relation to rotor speed variations. the proposed algorithm with existing SFVC drives [4, 5, 6].
The results in Fig. 10 represent the rotor speed Namely, the torque and speed measurements in Fig. 10
behavior for torque reversal, showing that rotor speed prove that the stability of drive operation in the low-
has the typical ‘‘saw-tooth’’ waveform, without signifi- speed region is improved, since the drives in [4, 5, 6]