Enhanced Direct Torque Control for Doubly Fed InductionMachine by Active Learning Method Using Indirect MatrixConverter

Hamid Behnia, Davood Arab Khaburi, Majid Aghasi & Vahid Faraji
International Journal of Engineering (IJE), Volume (6) : Issue (4) : 2012 201
Enhanced Direct Torque Control for Doubly Fed Induction
Machine by Active Learning Method Using Indirect Matrix
Converter
Hamid Behnia hbehnia@elec.iust.ac.ir
Iran University of Science & Technology (IUST)
Tehran, Iran
Davood Arab Khaburi Khaburi@iust.ac.irIran
Tehran, Iran
Majid Aghasi maghasi@elec.iust.ac.ir
Tehran, Iran
Vahid Faraji vfaraji@elec.iust.ac.ir
Tehran, Iran
Abstract
The term Direct Torque Control (DTC) originally is referred to a strategy which provides good
transient and steady-state performance but it has also some negative aspects, such as non
accuracy of flux, torque estimator, torque and flux ripple caused by non-optimality of switching
and imprecision in motor model which are known as an inherent characteristic of DTC. This paper
explores reducing of flux and torque ripple with using trial and error actively as a method called
Active Learning Method (ALM) in DTC for Doubly Fed Induction Machine (DFIM) which are the
motors or generators having twist on both stator and rotor subsequence power is transferred
between shaft and system. DFIM is linked to the grid within the stator and the rotor is fed by an
Indirect Matrix Converter (IMC). The function of IMC is similar to the direct one, although it has
the line and load bridges separated. We analysis the usage of four-step commutation in rectifier
stage of IMC to achieve the object of the losses’ reduction which are caused by snubber circuit.
ALM adopts itself with torque and flux estimators and estimates the outputs with regards to the
errors in torque and flux estimation by repetition therefore achieves the object of omitting
inaccuracies in control system hence confirming the effectiveness. Another concept in ALM called
Ink Drop Spread (IDS) handles different modeling target to predict on the data consequensing a
behavior curve in DTC. According to the simulation results, it is proved that a significant torque
and stator flux ripple reduction are obtained.
Keywords: Active Learning Method; Direct Torque Control; Doubly Fed Induction Machine;
Indirect Matrix Converter.
1. INTRODUCTION
The conventional energy sources are limit and have pollution for the environment, so more
attention and interest have been paid to the utilization of renewable energy sources such as wind
energy, fuel cell and solar energy etc. Among them wind energy is the fastest growing and most
promising renewable energy source due to economically viable [1].
In the field of wind energy generation systems, the wind turbine development shows a tendency
to increase the generation power level. According to the variation wind speed, Doubly Fed
Induction Machine (DFIM) is a common solution for variable speed wind turbines [2].

Often, the wind energy generation demands good torque dynamic performance as well [2].
Among all methods of torque control developed for the induction machine, the most widely used
technique may be classified within the Field Oriented Control (FOC) techniques and the direct
control techniques [2]. It seems to be accepted that the direct control techniques first introduced
such as Direct Torque Control (DTC) [3] and Direct Self Control (DSC) [4] achieve better steady-
state and transient torque control conditions rather than FOC techniques [5],[2].
Direct Torque Control (DTC) accompanies by some problems such as non accuracy of flux,
torque estimator, torque and flux ripple caused by non-optimality of switching and imprecision in
motor model which are all the inherent characteristics of DTC. To overcome these difficulties lots
of papers have been published on solving DTC drawbacks. Some of these papers fuzzified the
DTC system inputs and improve its characteristics [6],[7], some else tried to improve the torque
and flux estimators [8]-[10].
Active Learning Method (ALM) is a powerful recursive fuzzy modeling without computational
complexity. ALM has been proposed as a new approach to soft computing. The concept of the
ALM is based on the hypothesis that humans interpret information in the form of pattern-like
images rather than in numerical or logical forms. The ALM is modeled algorithmically on the
intelligent information-handling processes of the human brain, and it is characterized by
computing on the basis of intuitive pattern information [10]-[13].
In the Active Learning Method to model the information, a method called as Ink Drop Spread
(IDS) is used. The IDS method is able to deal with various modeling targets, ranging from logic
operations to complex nonlinear systems [10]. The IDS method possesses stable fast
convergence, and its modeling process, which is based on computing that uses pattern
information instead of complex formulas, is simple and efficient [10],[14].
Matrix converters (MCs) have been studied widely since their principle was introduced in
1970[20]. A MC is an AC-AC converter, with m×n bidirectional switch, which connects an m-
phase voltage source to an n-phase load . The matrix converter Compared with the conventional
AC/DC/AC converter, has some merits such as : eliminating bulky DC link capacitor in it,
straightforward Four-quadrant operation, also by controlling the switching devices appropriately,
both output voltage and input current are sinusoidal with only harmonics around or above
switching frequency [21].
Matrix-converter topologies can be divided into two types, one of them is direct matrix converters
(DMC) and another one is indirect matrix converters (IMC) which consists of separated line and
load bridges as presented in section [22].
Commutation problem of DMC is reduced considerably by utilize specific current commutation
methods. Typically two types of commutations methods have been proposed which don’t require
snubber circuits for a PWM rectifier of AC-to-AC converters without DC link components. The first
method is named rectifier zero current commutation and the second is named rectifier four-step
commutation. In these methods, although the losses in snubber circuits and the switching losses
in the PWM rectifier can be reduced but a complicated control circuit must be added to
synchronize the switching of both the PWM rectifier and the PWM inverter [22].
This research used ALM for DFIM to overcome the problems that were presented to DTC. ALM
can adapt itself with torque and flux estimators and estimate the outputs regards to the errors in
torque and flux estimations. Also proposed method avoids mathematical complexities of fuzzy like
methods so it is faster than conventional methods [10],[14]. From another side for feeding DFIM’s
rotor indirect matrix converter is used. The benefit of four-step commutation is analyzed in
rectifier stage of IMC to achieve the object of the losses’ reduction which are caused by snubber
circuit.

2. DIRECT TORQUE ONTROL PRINIPLE
Fig.1 shows the schematic of Direct Torque Control of DFIM. Stator winding of induction machine
is connected directly to the grid and the rotor is fed by converter that is also connected to the grid.
The main goal of the DTC is directly control the rotor flux and the electromagnetic torque of the
DFIM with Choosing the best voltage vector.
As shown in fig.2, the position of the rotor flux is divided into six sectors. There are also 8 voltage
vectors which-correspond to possible inverter states. These vectors are shown in fig.2. There are
also six active vectors 1 6V V− and two zero vectors 0 7,V V .The torque rotor flux equation of
doubly fed induction machine is as follows:
ܶ௘௠ ൌ
3
2
‫݌‬
‫ܮ‬௛
ߪ‫ܮ‬௦‫ܮ‬௥
|ߖ௥||ߖ௦| sin
(1)
refr−Ψ
refeT−
eT
rΨ
aS
bS
cS
ri
si
mω
rΨ
eT
mω
FIGURE 1: The diagram of the DFIM direct torque control system
Where ߪ ൌ 1 െ
௅೓
మ
௅ೞ௅ೝ
is the leakage coefficient. ‫ܮ‬௦ and ‫ܮ‬௥ are the stator and rotor inductance, ‫ܮ‬௛
is mutual inductance, ܴ௥ is rotor resistance and is phase angle difference between ߖ௥ and ߖ௦.
As the stator winding of DFIM is connected to power grid, by ignoring the voltage drop of stator
winding resistance and the fluctuation of supply voltage, one can appropriately consider the
magnitude of the stator flux to be constant and rotate at synchronous speed [15]. Therefore,
according to equation (1), we know that the torque control of wound rotor doubly-fed machines
can be realized through adjusting the rotor flux vector. Furthermore, in the case that the rotor flux
rψ
uuv
has a circular trajectory, T becomes the function of phase angleθ .T increases as θ
increases. Conversely, T decreases as θ decreases. Therefore, the control of the torque/speed
can be realized through adjusting the phase angleθ .

)001(1V
)011(2V)010(3V
)110(4V
)100(5V )101(6V
)111(7V
)000(0V
rΨ
sΨ
δ
rΨ
sΨ
)101(6V)100(5V
)011(2V)010(3V
)110(4V
)111(7V
)000(0V )001(1V
δ
FIGURE 2: Flux space vectors in the rotor reference frame, in motor and generator modes. (a) Motor mode,
(b) Generator mode
Considering to Fig.1 to find out which vector is appropriate for DTC drive, first by comparing the
flux and torque reference values with the calculated ones, Error values of them are determined.
Second the error values are applied into two-level hysteresis comparators. The output of both
two-level hysteresis comparators that show flux and torque should be increase or decrease.
According to the operation condition that can be generator or motor mode, sector that rotor flux is
located and based on analysis above the selection of rotor voltage vector selected by table I and
II.
‫ܪ‬்௘ 1 0 -1 1 0 -1
Rotor
Flux
Sector
1 V6 ܸ଴ ܸଶ ܸହ ܸ଻ ܸଷ
2 ܸଵ ܸ଻ ܸଷ ܸ଺ ܸ଴ ܸସ
3 ܸଶ ܸ଴ ܸସ ܸଵ ܸ଻ ܸହ
4 ܸଷ ܸ଻ ܸହ ܸଶ ܸ଴ ܸ଺
5 ܸସ ܸ଴ ܸ଺ ܸଷ ܸ଻ ܸଵ
6 ܸହ ܸ଻ ܸଵ ܸସ ܸ଴ ܸଶ
TABLE 1: CLASSICAL DTC LOOK-UP TABLE FOR MOTOR MODE[15]
‫ܪ‬்௘ 1 0 -1 1 0 -1
Rotor
Flux
Sector
1 V6 ܸ଴ ܸଶ ܸହ ܸ଻ ܸଷ
2 ܸଵ ܸ଻ ܸଷ ܸ଺ ܸ଴ ܸସ
3 ܸଶ ܸ଴ ܸସ ܸଵ ܸ଻ ܸହ
4 ܸଷ ܸ଻ ܸହ ܸଶ ܸ଴ ܸ଺
5 ܸସ ܸ଴ ܸ଺ ܸଷ ܸ଻ ܸଵ
6 ܸହ ܸ଻ ܸଵ ܸସ ܸ଴ ܸଶ
TABLE 2: All CLASSICAL DTC LOOK-UP TABLE FOR GANARATOR MODE
3. INDIRECT MATRIX CONVENTOR
Indirect matrix converter (IMC), which consists of separated line and load bridges as presented in
Fig.4. IMC is similar to the traditional AC/DC/AC converter system and to previous proposed
capacitorless DC link circuits. On the load side, the arrangement has the same conventional
inverter as for the AC/DC/AC converter. As a consequence, traditional PWM methods may be
used to generate the output voltage waveform.

However, in order to ensure proper operation of this converter, the DC side voltage should always
be positive. On the line side, the converter has a rectifier which is similar to traditional one except
that the switches are all bidirectional [22]. Typically two types of commutations methods have
been proposed which don’t require snubber circuits for a IMC.
The first method named rectifier zero current commutation and the second method named
rectifier four-step commutation. In these methods, although a complicated control circuit must be
added to synchronize the switching of both the PWM rectifier and the PWM inverter but the
losses in snubber circuits and the switching losses in the PWM rectifier can be reduced [21]. In
this paper, four-step commutation method in the rectifier stage is used. In four-step commutation
method, direction of output current and value of input voltage determines the sequence of
switching and the commutation reliability.
The process of commutation between phase A and B is explained with Fig.3. phase A connects to
rectifier output through IGBT of switch S11 and diode of switch S12 . At this point, as it is shown
(dotted lines in fig. 3.a) current does not pass from other transistors and diodes. When 0DC
i >
the following four-step switching sequence is: 1) turn off S12 ; 2) turn on S31 ; 3) turn off S11 ; 4)
turn on S32 . When 0DC
i < , the following four- step switching sequence is: 1)turn off S11; 2)turn
on S32 ; 3)turn off S12 ; 4)turn on S31 [22].
FIGURE 3: Four-Step Commutation Methode Block Diagram[15]
FIGURE 4: Indirect Matrix Converter Structure[15]
4. ACTIVE LEARNING METHOD (ALM)
Active Learning Method (ALM) is a new fuzzy modeling method which has been introduced by
Bagheri Shouraki and Honda (1997a).

In contrast to the humans, when human learns an object, in the first step, he grasps its
characteristics from much information which apparently looks disorderly, and finds out its
tendency, then he finds out the connection with the knowledge formerly learned and he stores it
together with the relationship in his brain. The same is true with ALM. When learning the action of
a system, it starts from grasping the input-output relations. The input-output data of the object
system are collected and the system is modeled. And while memorizing the knowledge, the input-
output data are further collected by trial and error, and the system is modeled using the past
knowledge and the data. This process is repeated[18].
Active Learning Method is the learning mode in which the learner improves the performance by
acquiring information from the behavior of his own [18]. Actually the concept of the ALM is based
on the hypothesis that humans interpret information in the form of pattern-like images rather than
in numerical or logical forms, in fact it is algorithmically modeled on the intelligent information-
handling processes of the human brain, and it is characterized by computing on the basis of
intuitive pattern information [10],[16],[17].
ALM considers the behavior of complicated Multi input Multi output (MIMO) systems as collection
of simple systems which are single input single output (SISO) systems and the system is
expressed by combining them (Fig.5). In the case of two inputs and one output, for example, the
input-output relation is plotted on a three-dimensional space [18].
This modeling method not only is similar to human logical thinking but also avoids mathematical
complexity. In this method, the learning is done by mutual action with the environment (Fig.6) and
promoted by reinforcement learning. The reinforcement learning originated from animal learning
psychology and the optimization method like dynamic programming. In this method, the action is
FIGURE 5: Division and Combination of System[14]
reinforced by giving reward or punishment according to the behavior taken in a certain state.
ALM starts with gathering data and projecting them on different data planes. The horizontal axis
of each data plane is one of the inputs and the vertical axis is the output. The method called IDS
(Ink Drop Spread) which is a processing engine is used to look for a behavior curve, hereafter
narrow line, on each data plane. The heart of this learning algorithm is a fuzzy interpolation
method which is used to derive a smooth curve among data points [16], [17].
As a matter of fact IDS method is a modeling technique used in the active learning method
(ALM), which is a new approach to soft computing. It is characterized by a modeling process
which is based on computing that uses intuitive pattern information instead of complex formulas
[10].

FIGURE 6: Mutual Action with the Environment[14]
5. IV. Ink Drop Spread (IDS)
The basic concept of IDS is to extract the system properties from the input-output data by using
fuzzy process. This method searches for continuous possible paths on the interpolated data
points on each plane. In this method, we assume that each data point on each data plane is a
light source (Fig.7), which has a cone shape illumination pattern. As the distance from these light
sources increases, their illumination pattern will interfere and generate new bright areas. The
lights interfere with each other and the illuminated pattern appears to show light and darkness.
That is, the part where many lights fall is lighter than other part. By combining the light parts
continuously, a kind of narrow path expressing the input-output relations can be obtained [18].
By applying IDS method on each data plane, two different types of information would be
extracted. One is the narrow path and the other is the deviation of the data points around each
narrow path (Fig.8). Each narrow path shows the behavior of output relative to an input and
spread of the data points around this path shows the importance degree of that input in overall
system behavior. Less deviation of data points around the path presents higher degree of
importance and vice versa [18].
Fig.9 illustrates the architecture of an IDS model with two-input, two-partition structure. The IDS
model comprises three processing layers. The bottom input layer breaks down input-output data
into SISO data, and transfers them to the upper modeling layer. The top inferential layer
computes the prediction with the learning data transferred from IDS units. With the exception of
the case where a particularly high accuracy is required for an IDS model, the upper layer does
not intervene the learning process of IDS units [10],[16]-[19].
Also the spread functions, which show the amount of spread of data on each plane resulting from
the effects of other variables, can be calculated using a method presented in [12] by S.B.Shouraki
and N.Honda. Then the output of the system can be calculated by (2) [24].








+++








+++
=
aaa
xf
a
xf
a
xf
a
n
nn
n
y
1
...
11
)(
1
...)(
1
)(
1
21
22
2
11
1
(2)

FIGURE 7: Concept of IDS (Ink Drop Spread)[24]
FIGURE 8: Irradiation Pyramid
where
y = the output of system (function)
x1, x2,…, xn = inputs of the system (variables)
f1, f2,… fn = the narrow path functions for plane x-y for each variable
a1, a2,…, an = spread values
6. DTC MODEL BY ALM WITH USING IMC
This research presents and analyses a recommended model of enhanced DTC with the help of
ALM, shown in fig10. it uses IMC to feed DFIM’s rotor that block control determines unitedly
about the appropriate voltage vector in the inverter and rectifier section of IMC. According to the
explanations were given about ALM, in this method, with the help of the database, which is
obtained by the method of trial and error, the input-output information is collected from the control
object and the controller is constructed by the fuzzy-like processing of these data. In the other
word some trial inputs are applied to control object and this action is reinforced by giving reward
or punishment according to the result. It should be mentioned that trial and error inputs should be

2( , )x y
1 2Training data ( , , )x x y
1( , )x y
1( , )x y 2( , )x y
11Ψ
M odel output y
)
11Ψ11Ψ11Ψ
' '
1 2( , )x x
FIGURE 9: Structure of IDS Model[24]
selected so that covers all possible system inputs. On the other hand, by increasing number of
trial and error actions, the better model of system can be obtained [10],[14].
FIGURE 10: DTC using IMC
Regarding the DTC system diagram, shown in Fig10, This research is used IMC to feed DFIM’s
rotor and the DTC system is a multi input-single output system. Fig. 11 shows the input-output
structure of the DTC system control block. According to the ALM basis, this multi input-single
output system should be divided into systems with single input and single output (SISO) systems.
The DTC control block inputs consist of Torque error, Flux error and rotor Flux angle (position).
The rotor flux angle is not considered as an independent input in SISO systems because there is
not any direct relationship between rotor flux angle and inverter state (Fig.1). So as mentioned
above, the rotor flux is divided into six sectors and this modeling procedure is done for each
sector. In the other word, there are 6 couple SISO systems and every couple is for a sector [10].

Te
eΨ
Te
eΨ
FIGURE 11: Inputs and Outputs of DTC and DTC SISO Models
To apply trial and error inputs Sampling frequency and inverter switching frequency is justified on
8 kHz. To experience different possible errors, torque set value is a random function in proposed
model. Some of the sample data obtained by 20000 repetitions are presented in Table 3. This
table shows that some inputs lead to improvement in result and some of them worsen the results.
So as mentioned above, the inputs which lead to decrease in torque or flux errors should be
rewarded and the others should be punished.
These 20000 samples are plotted in a three dimensional space (Fig.12) and the following formula
is used to determine the efficiency of each trial action:
1i i iT T Td e e +
= − If 0≥eTi
(3)
eed TTT iii
−=
+1
If 0
iTe ≤ (4)
)(dT i
MAXM = (5)
M
Td
E i
i
= (6)
The flux equations and torque equations are same. By equation (6), the efficiency of each applied
inverter vector is calculated and Ei determine the magnitude (reward) of each trial and error and
its popularity. Any inverter vector with bigger Ei, is reinforced because leads to more
improvement in decreasing error. So its relevant vector magnitude in Fig.12 will be bigger [10].
In Fig.12 the plane of trial actions with reward and punishment of data is shown from four different
angle of view. The horizontal plane expresses the inverter voltage state and the vertical axis
determines the error. The respective correct inverter state of any torque error can be calculated
based on these three dimensional plots and the control rules are acquired by these plots, also
rules obtained by ALM are similar to classic ones with some minor differences. The rules format
is as:
If eT is eT1 and flux sector is α then inverter state is Vm
If eΨ is eΨ1 and flux sector is β then inverter state is Vn
The ALM output in each sector will be achieved by combining SISO models of sector torque and
flux errors. DTC total system model is achieved by combining 12 SISO models of six sectors and
this combination is based on the sum of adaptability of each SISO model. Equation (7) is used for
calculating output by combining SISO outputs [10].
(7)yyTTy ψϕαα ×+×=
Where α is the adaptability of each SISO, determined by the efficiency (Ei) of each case. By this
equation the output of a case in sector one will be a compromise of sector one torque and flux
error SISO systems outputs [10].

eT (NM) eΨ (wb) Flux Sector
Inverter
vector
3.5918541 -0.023571 1 2
0.839314 -0.03453 1 2
-1.42841 -0.18675463 1 2
0.7784831 -0.1834733 1 3
-0.512115 -0.0021443 1 3
1.63797 0.0523732 1 3
-1.346901 0.03453232 1 3
-2.19589 0.0737643 1 5
- 3.219250 -0.0348734 1 5
2.432973 -0.20013 1 5
1.646507 -0.19714 1 5
3.2917926 -0.032432 1 6
1.4556623 -0.0347334 1 6
-0.3716126 -0.042329 1 6
TABLE 3: SOME SAMPLE TRIALS
FIGURE 12: IDS irradiation pattern for SISO system with torque error as input and inverter vector as output
7. SIMULATION RESULT AND COMPARISON WITH CLASSIC METHODS
The result of torque control and rotor flux for both DTC based on IMC and the proposed method
is presented in Fig.13-fig.16. Good dynamic behavior of torque responses when reference torque
suddenly changes from 100Nm to -100Nm at time t=0.3 is shown. As can be seen, the ALM leads
to less deviation from the desired value of torque and rotor flux (reference value) rather than the

conventional DTC, this is due to its adaptability with motor model and the total system, also
obviously switching frequency is decreased. According to the torque figures, ripple reduction of
about 25% during using this method is created. Also by applying this method to the DTC, ripple
rotor flux is reduced around 15%. Fig.17 and fig.18 show the flux response, flux circular trajectory
and rotor flux sector. DC-link voltage of IMC is shown in Fig.19, Also Fig.20 and fig.21 show the
3-phase stator and rotor currents. These currents are sinusoidal and demonstrates that there are
no low order harmonics.
FIGURE 13: Output torque of DTC based on IMC
FIGURE 14: Output torque of enhanced DTC based on IMC by ALM
FIGURE 15: Flux response of DTC based on IMC

FIGURE 16: Flux response enhanced DTC based on IMC by ALM
FIGURE 17: Flux circular trajectory
FIGURE 18: rotor flux sector

FIGURE 19: DC voltage of IMC
FIGURE 20: Three phase stator current
FIGURE 21: Three phase rotor current
FIGURE 22: Current harmonic spectra of stator

FIGURE 23: Current harmonic spectra of rotor
8. CONCLUSION
This paper tries to explore reducing of flux and torque ripple which are the inherent negative
characteristics of DTC by an actively trial and error method called Active Learning Method for
Doubly Fed Induction Machine based on Indirect Matrix Converter. Using DTC strategy with IMC
result that the benefits of both method simultaneously obtained. The advantages are: fast
response in torque control, regeneration capability, near sinusoidal stator and rotor current. Also
usage of four-step commutation in rectifier stage of IMC which is the source of feeding the rotor is
analysed to reduce the losses’ caused by snubber circuit. In addition the concept Ink Drop
Spread (IDS) is applied to handle different modeling target to predict on the data and get a
behavior curve in DTC. Finally the simulation results confirm that a significant torque and stator
flux ripple reduction are obtained.
9. REFERENCES
[1] B.Chitti Babu , K.B.Mohanty “Doubly-Fed Induction Generator for Variable Speed Wind
Energy Conversion Systems- Modeling & Simulation”, International Journal of Computer
and Electrical Engineering, Vol. 2, No. 1, February, 2010 1793-8163.
[2] Abad, G.; Rodriguez, M.A.; Poza, J.; , "Two-Level VSC Based Predictive Direct Torque
Control of the Doubly Fed Induction Machine With Reduced Torque and Flux Ripples at
Low Constant Switching Frequency," Power Electronics, IEEE Transactions on , vol.23,
no.3, pp.1050-1061, May 2008.
[3] I. Takahashi and Y. Ohmori, “High-performance direct torque control of an induction motor,”
IEEE Trans. Ind. Appl., vol. IA-25, no. 2, pp. 257–264, Mar./Apr. 1989.
[4] M. Depenbrock, “Direct self-control (DSC) of inverter-fed induction machine,” IEEE Trans.
Power Electron., vol. PE-3, no. 4, pp. 420–429, Oct. 1988.
[5] D. Casadei, F. Profumo, G. Serra, and A. Tani, “FOC and DTC: Two viable schemes for
induction motors torque control,” IEEE Trans. Power Electron., vol. 17, no. 5, pp. 779–787,
Sep. 2002.
[6] Jiang Zhijun1, Hu Shimiao, Cao Wenhui, “A New Fuzzy Logic Torque Control Scheme Based
on Vector Control and Direct Torque Control for Induction Machine”, the 3rd International
Conference on Innovative Computing Information and Control (ICICIC'08), june.2008.
[7] YING-PEI LIU, JIAN-RU WAN, CHEN-HU YUAN, GUANG-YE LI, “fuzzy based direct torque
control of pmsm drive using an extended kalman filter”, Proceedings of the Eighth
International Conference on Machine Learning and Cybernetics, Baoding, 12-15 July 2009.
[8] L.Zhong, M.F.Rahman, K.W. Lim, “a fuzzy observer for induction motor stator resistance for
application in direct torque control”, May 1997.

[9] Yanping Xu, Yanru Zhong, Jie Li, “Fuzzy Stator Resistance Estimator for a Direct Torque
Controlled Interior Permanent Magnet Synchronous Motor”, Department of Electrical
Engineering, Xi'an University of Technology, Xi'an 710048, China, 2006.
[10] M. J. Ghorbani; M. Akhbari; H. Mokhtari; “Direct torque control of induction motor by Active
Learning Method,’’ 1st Conference on Power Electronic & Drive Systems & Technologies
(PEDSTC), pp: 267 – 272, 2010.
[11] G.Yuasa, S.B.Shouraki, N.Honda, Y.Sakurai "Applying an active learning method to control
problems", Asian Fuzzy System Symposium 2000(AFSS2000), tukuba Japan, pp572-577,
2000.
[12] S. B. Shouraki and N. Honda, “Recursive fuzzy modeling based on fuzzy Interpolation,”
Journal of Advanced Computational Intelligence, Vol. 3, No. 2, pp. 114–125, April 1999.
[13] S. B. Shouraki, G.Yuasa, and N. Honda, “Fuzzy interpretation of human intelligence,”
International Journal of Uncertainty, Fuzziness and Knowledgebased Systems, Vol. 7, No.
4, pp. 407–414, Aug. 1999.
[14] Faraji, V.; Aghasi, M.; Khaburi, D.A.; Ghorbani, M.J.; , "A modified DTC for induction motor
drive system fed by Indirect Matrix Converter using Active Learning Method," Power
Electronics, Drive Systems and Technologies Conference (PEDSTC), 2011 2nd , vol., no.,
pp.356-361, 16-17 Feb. 2011.
[15] Aghasi, M.; Faraji, V.; Khaburi, D.A.; Kalantar, M.; , "A novel Direct Torque Control for
Doubly Fed Induction Machine based on Indirect Matrix Converter," Electrical, Electronics
and Computer Engineering (ELECO), 2010 National Conference on , vol., no., pp.303-308,
2-5 Dec. 2010.
[16] S. B. Shouraki, “A novel fuzzy approach to modeling and control and its hardware
implementation based on brain functionality and specifications,” Ph.D. dissertation, The
Univ. of Electro-Communications, Chofu, Japan, March 2000.
[17] S.Bagheri, G.Yuasa, N.Honda, “Fuzzy Controller Design by an active learning method”,31st
Symposium of Intelligent Control, SIC 98.
[18] Sakurai, Y.; Honda, N.; Nishino, J.; , "Acquisition of control knowledge of nonholonomic
system by Active Learning Method," Systems, Man and Cybernetics, 2003. IEEE
International Conference on , vol.3, no., pp. 2400- 2405 vol.3, 5-8 Oct. 2003.
[19] Murakami, M.; Honda, N.; , "Classification performance of the IDS method based on the
two-spiral benchmark," Systems, Man and Cybernetics, 2005 IEEE International
Conference on , vol.4, no., pp. 3797- 3803 Vol. 4, 10-12 Oct. 2005.
[20] Jussila, M.; Tuusa, H.; , "Comparison of Simple Control Strategies of Space-Vector
Modulated Indirect Matrix Converter Under Distorted Supply Voltage," Power Electronics,
IEEE Transactions on , vol.22, no.1, pp.139-148, Jan. 2007
[21] Wei, L.; Lipo, T.A.; , "A novel matrix converter topology with simple commutation," Industry
Applications Conference, 2001. Thirty-Sixth IAS Annual Meeting. Conference Record of the
2001 IEEE , vol.3, no., pp.1749-1754 vol.3, 30 Sep-4 Oct 2001

[22] Aghasi, M.; Faraji, V.; Khaburi, D.A.; Kalantar, M.; , "A novel Direct Torque Control for
Doubly Fed Induction Machine based on Indirect Matrix Converter," Electrical, Electronics
and Computer Engineering (ELECO), 2010 National Conference on , vol., no., pp.303-308,
2-5 Dec. 2010
[23] Muroya, M.; Shinohara, K.; Iimori, K.; Sako, H.; “Four-step commutation strategy of PWM
rectifier of converter without DC link components for induction motor drive” Electric
Machines and Drives Conference, 2001. IEMDC 2001. IEEE International page(s): 770 –
772
[24] H. Behnia, D. A. Khaburi, and A. Ejlali, “Enhanced Direct Torque Control for Doubly Fed
Induction Machine by Active Learning Method,” in Universities’ Power Engineering
Conference (UPEC), Proceedings of 2011 46th International, 2011, pp. 1–6.

Enhanced Direct Torque Control for Doubly Fed InductionMachine by Active Learning Method Using Indirect MatrixConverter

More Related Content

What's hot (18)

Viewers also liked (13)

Similar to Enhanced Direct Torque Control for Doubly Fed InductionMachine by Active Learning Method Using Indirect MatrixConverter (20)

Recently uploaded (20)

Enhanced Direct Torque Control for Doubly Fed InductionMachine by Active Learning Method Using Indirect MatrixConverter