IET Power Electronics

Research Article

Multilevel inverter with space vector ISSN 1755-4535

Received on 6th May 2016
Revised 13th February 2017
modulation: intelligence direct torque control Accepted on 24th March 2017
E-First on 27th June 2017
of induction motor doi: 10.1049/iet-pel.2016.0287
www.ietdl.org

B. Kirankumar1 , Y.V. Siva Reddy2, M. Vijayakumar3

1JNT University Anantapur, Saradha Nagar, Anantapur 515002, India
2Electrical and Electronics Engineering Department, G. Pulla Reddy Engineering College, Kurnool, Andhra Pradesh, India
3Electical and Electronics Engineering Department, JNT University, Anantapur, Andhra Pradesh, India

E-mail: [email protected]

Abstract: An Intelligence direct torque control (DTC) of induction motor using multilevel inverter with space vector modulation
(SVM) is proposed in this study. The novelty of the proposed intelligence technique is an artificial neural network (ANN), which is
utilised for the DTC of the induction motor. Here, ANN has been trained by using the target parameters such as reference
quadrature and direct axis voltage with the corresponding input parameters change in motor torque and flux. The change in
motor torque and flux parameters are identified by using the multilevel inverter output voltage and current. The back propagation
process is used for the training process of the ANN. By using the ANN outputs, the SVM technique develops the control pulses
for the multilevel inverter. Then the proposed intelligence technique is implemented in the MATLAB/simulink platform and the
effectiveness is analysed by comparing with the different case studies and different methods. The comparison results
demonstrate the superiority of the proposed approach and confirm its potential to solve the problem.

1 Introduction ripple and slow transient response to the step changes in torque
during start-up. Therefore, intelligent methods are used such as
In Industrial, commercial and domestic applications, induction artificial neural network (ANN), fuzzy logic (FL) and sliding mode
motors (IMs) are commonly used because they are simple, Shaggy, control (SMC) theory [19]. Majority of them are concerned with
economic in cost and easy to maintain [1]. The constant speed enhancement of the flux and torque estimator and combined
drives are used in most of the induction machines [2]. Whenever operation of DTC with a SVM technique [20].
the load is fluctuates Induction motor is complex to sustain a This paper illustrated an intelligence DTC of induction motor
constant speed [3]. The beginning of new semiconductor devices using multilevel inverter with space vector modulation (SVM).
has made variable speed drives with induction machines accessible Moreover, for proper selection of sector, a smart technique ANN is
during the last decades [4]. The high performance electric speed used, which makes better the work of induction motor in way of
drives needs decoupled torque and flux control [5]. This method is speed and torque. ANN is relatively simple learning algorithm and
generally afforded through field oriented control (FOC) or vector it can approximate any function, regardless of its linearity. The
control method [6]. The excitation current and load current can be speed reference is rapidly achieved by induction motor without
controlled individually by using FOC method. Hence, flux and overshoot and with small steady state error as simulation results
torque also be separately controlled correspondingly in DC motor also support this and also the load disturbances is rejected very
[7]. To control pulse-width-modulation in the inverter system, the fast. When conventional DTC is compared, the ANN-based DTC
FOC method needs current controller, coordinate transformation gives improved torque response in terms of decreased torque
and current regulator [8]. However, FOC needs efficient and ripple. Here, ANN has been trained using the target parameters
complex calculation of the decoupling, hence it is complex to such as reference quadrature and direct axis voltage with their
perform and easily towards by load disturbance and parameter subsequent input parameters altered in motor torque and flux. The
uncertainties [9]. change in motor torque and flux parameters are recognised by
At present, a control method known as DTC was established using the multilevel inverter output voltage and current. The back
with various method to control the induction motor torque and flux propagation method is used for the training process of the ANN.
[10]. DTC is layout to afford fast and robust reactions of the drive By using the ANN outputs the SVM technique improves the
even below low-frequency operation of power electronics [11]. It is control pulses for the multilevel inverter. The recent research
a combination of the FOC and direct self-control methods [12]. By works are presented in Section 2 and the detailed description of the
using a switching vector look-up table, the basic concept of DTC proposed technique is presented in Section 3. The experimental
of induction-motor drives is to control both stator flux-linkage and results and discussion are given in Section 4. Finally, the Section 5
electromagnetic torque of the machine concurrently [13, 14]. It is concludes the paper.
described by the deficiency of PI regulators, coordinate
transformations, current regulators and PWM signals generators
[15]. The DTC has been put into operation by using either variable 2 Recent research works: a brief review
switching frequency or constant sampling techniques [16]. Several associated works are already subsisted in literature which
Compared with a vector control scheme, DTC provides identical is in terms of DTC technique of induction motor with various
dynamic performance with simpler controller architecture [17]. intelligence methods. Few of them examined here.
The classical DTC technique is in terms of hysteresis-loop A FL hysteresis comparator-based DTC scheme of an IM under
controller with single vector switching table. Its switching different dynamic conditions is proposed by Uddin et al. [21]. The
frequency differs with speed and load torque, which can bring out FL controller (FLC) was used to vary the bandwidth of the torque
high torque pulsation particularly in low speed due to the low hysteresis controller so as to decrease the torque and flux ripples
switching frequency, which greatly restricts its application [18]. and, hence, to enhance the motor dynamic response. The effects of
Common disadvantages of conventional DTC are high torque torque hysteresis bandwidth on the amplitude of torque ripples of

IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137 1129
© The Institution of Engineering and Technology 2017
an IM were also illustrated. Based on the slopes of motor-estimated runs on the predictive switching table to develop the entire
torque and stator current, an FLC was constructed to choose the performance of the motor unit. A kind of PC is implemented for
possible bandwidth of the torque hysteresis controller. Their paper speed rules by using lots torque observer, the torque being
was implied an easier algorithm than the usual trigonometric considered an unfamiliar perturbation. A Kalman filtration (KF) is
function-based algorithm to calculate the sector number (required employed for reliable flux estimation. The validity of the proposed
for DTC scheme) of the stator flux-linkage space vector. Their controller was proved on an instant control prototyping train station
presented algorithm was decline the computational burden on the experimentally. The obtained results have proven superiority of the
microprocessor. proposed control with regards to the speed trajectory tracking, flux
Brahim Metidji et al. [22] have proposed a low-cost and simple and torque dynamic responses, and disturbance rejection. Also, a
phase-current reconstruction algorithm for three-phase IM under lesser current distortion was witnessed with PDTC, in comparison
DTC using the information acquired from only one shunt resistor. to the standard DTC, scheduled to increased mean inverter turning
Their proposed paper was used for improving the low-cost high- frequency.
performance IM drive. Their proposed algorithm was robust and A fault-tolerant DTC system have offered by Bermudez et al.
very easy. It has used the dc current to modernise the stator [29]. The performance of the technique is experimentally validated
currents required to estimate the motor flux and the in a five-phase IM drive considering an open-phase problem
electromagnetic torque. A theoretical conception was enhanced, the condition. Provided lab tests examine regular and transient claims,
adapted look-up table was presented, and current-access tables including the changeover from pre- to post-fault procedure.
were designed and used in the phase-current reconstruction. Obtained results verify the eye of the proposal, which ensures the
An integration algorithm for stator flux estimation of DTC was open-phase fault-tolerant capacity for DTC-controlled five-phase
proposed by Wang et al. [23]. Their proposed algorithm contains a IM drives. The DTC has been trusted instead of traditional FOC
fifth-order low-pass filter (LPF), a high-pass (HP) filter, and a options for three-phase drives. The traditional DTC system has
simple logical calculation part, where the fifth-order LPF and the been long to multiphase drives recently efficiently, using hysteresis
HP filter can efficiently filter out HF harmonics and the dc drift in regulators to trail the required torque and flux in symmetrical five-
the back electromotive force, respectively. At any synchronous phase IM independently.
angular frequency, the amplitude and phase-frequency A DTC IM drive that employs feedback linearisation and
distinctiveness of their algorithm were similar as those of the pure sliding-mode control (SMC) have described by Lascu et al. [30]. A
integrator, and in the interim its dc gain was zero. feedback linearisation approach yields a decoupled linear IM
A sensor less three-level neutral-point-clamped inverter-fed model with two state variables: torque and stator flux magnitude.
induction motor drive was presented by Zhang et al. [24]. The Robust, fast, and ripple-free control is achieved by using SMC with
usual direct torque control (DTC) switching table failed to consider proportional control in the vicinity of the sliding surface. SMC
the circuit limitations, such as neutral-point-balance and smooth assures robustness as in DTC, while the proportional component
vector switching, basis by the topology of a three-level inverter. eliminates the torque and flux ripple. The torque time response is
Two types of adapted schemes for three-level DTC were presented similar to conventional DTC and the proposed solution is flexible
to solve these problems. FL control and the speed-adaptive flux and highly tunable due to the P component. The controller design
observer were introduced to develop their performance of the is presented and its robust stability is analysed in simulations.
system. The concern of large starting current was explored and The speed of the motor is affected by load disturbance,
solved by introducing the method of pre excitation. parameter variations, and model uncertainties in high performance
A brief review of DTC of induction motors IM as well as its applications. To maintain the constant speed, the overcome of
implementation for electric vehicle (EV) applications was proposed factors that affects the motor speed is essential and it is one most
by Sutikno et al. [25]. First, the basic DTC method is based on significant and challenging tasks. Conventionally, the speed of IM
hysteresis controllers would be introduced, and then a summary of is often controlled by fixed gain closed loop feedback control
the foremost problems in a basic DTC scheme would be presented mechanism. In fixed gain controller, the controller gains are very
and elucidated, as well as few efforts for enhancing the method. sensitive to the speed affecting factors. Hence, it is unable to
The main section presents a critical review of DTC for EV control the entire speed of the IM. This problem can be solved by
applications, taking into deliberation the vehicle mechanics and various kinds of methods such as SMC, variable structure control,
aerodynamics of EVs. The review was significant to afford self-tuning techniques, adaptive controller etc. Moreover, the above
guidelines and insights for future research and development on the existing controller depends on the accurate mathematical form of
DTC of IM drives for sustainable reliability and energy efficient the system. However, there is a complex in improving the
EV applications. mathematical model of the system because the system model
A non-linear DTC strategy with SVM for an IM have presented cannot be adaptive for the unknown load and parameter variations.
by Ammar et al. [26]. A non-linear input–output feedback Hence, the efficient of the model and the speed of the system are
linearisation (IOFL) was carried out to accomplish a decoupled affected. The vector control method has found broad acceptance in
torque and flux control and the SVM was utilised to lessen high the industry. However, this control technique needs complex
torque and flux ripples. Furthermore, the control system coordinate transformation, inner current control loop, and accurate
performance was upgraded by inserting a brilliant twisting speed system parameters. The presented control model is described in the
controller in the outside loop and lots torque observer to improve following.
the speed rules. The merging of dual non-linear strategies ensures a
good robustness and energetic against variables deviation and 3 Proposed control model
disruption.
A proportional vital immediate torque control (PIDTC) and In high-performance applications, the speed of the motor is
fuzzy-2 DTC of the IM drive using space vector pulse width affected by load disturbance, parameter variations and model
modulation for three-level diode clamped inverter (TDCI) have uncertainties. In order to maintain the constant speed, the
provided by Naik et al. [27]. The control strategies have the compensation of factors that affects the motor speed is
capability to balance the dc-link capacitor voltages of TDCI by indispensable and it is one of the important and challenging tasks.
selecting a most effective voltage vector without needing a Traditionally, the speed of IM is often controlled by various
supplementary controller. However, the three-level PIDTC techniques. The available techniques cannot be adaptive for
associated with an IM provides poor transient performance during unknown load and parameter variations. Hence, the accuracy of the
starting, launching, and speed reversal with appreciable current and model and the speed of the system are affected. In the section, we
voltage total harmonic distortion (THD). Furthermore, it takes have to explain about the ANN-based DTC of an induction motor.
additional time to attain the steady express with large spikes on Here, the control model has induction motor with the supply of
capacitor voltages through the load or velocity variation. MLI. The MLI eradicated the disturbance present in the induction
A predictive DTC for IM have suggested by Ouhrouche et al. motor supply voltage. The induction motor input parameters such
[28]. It combines the DTC and the predictive control (PC), and

1130 IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137
© The Institution of Engineering and Technology 2017
Fig. 1  Structure of the proposed control technique

as voltage V abc and current I abc are utilised for the calculation of bridge consists of four insulated gate bipolar transistors and one dc
the flux ϕact and the torque T act estimation (Fig. 1). source. The number of output phase voltage levels of cascade
The motor torque and the flux can be estimated by the multilevel inverter is 2h + 1, where h is the number of dc
following equations generators. The output phase voltage of the five-level inverter is
five-level staircase waveform. The fundamental output voltage of
3 the multilevel inverter can be described in the following equation
T act = P ϕs ϕr sin θsr (1)
2 ∞

2 2
V(t) = ∑ ansin nαn + bncos nαn (5)
ϕact = ϕsα + ϕsβ (2) n=1

where, ϕs and ϕr are the stator and rotor flux, P is the input power, where, V(t) is the output voltage, an is the odd harmonics, bn is the
θsr is the angle between stator and rotor flux, ϕsα and ϕsβ are the even harmonics, which is zero because of the quarter wave
components of the stator flux and which can be determined by the symmetry of the output voltage, α is the switching angle. The odd
following relations harmonics at n order is described in the following equation.
∞
4
∫
t
nπ n∑
ϕsα = V sα − RsI sα dt (3) an = V ndc[cos(nα1) + cos(nα2) + ⋯ + cos(nαs)]sin(nωt) (6)
0 =1

and 0 < α1 < α2…αk < π/2

∫
t
ϕsβ = V sβ − RsI sβ dt (4)
0
∑kn = 2 V(t) 2

THD = (7)
where, Rs is the stator resistance, I sα and I sβ are the components of V1
the stator current, which is derived from the three phase induction
motor input current I abc, V sα and V sβ are the components of the where, V dc is the input dc voltage. Here, we consider the five level
stator voltage, which is derived from the three phase induction inverter; it consists of h = 2 number of bridges. The output phase
motor voltage V abc [20]. The estimated torque and flux is verified voltage is the sum of all the individual bridge outputs. The five
with the reference values, i.e. reference torque T ref and reference level inverter output voltage variation and THD is identified from
flux ϕref . The error is determined for both the torque ΔT and flux the above mentioned equations. The five-level inverter output
voltage in d−q form can be expressed in the research paper [31].
Δϕ. Then both the error values are allowed as the input of the
proposed artificial intelligence (AI) technique ANN. From the For five-level of multilevel inverter voltage, there are 53 = 125
proposed technique, the flux equivalent reference direct axis phase voltage combinations and 3N (N − 1) + 1 = 61 different
voltage vectors, where N is the level of the inverter. The higher the
voltage V ref
d and the torque equivalent reference quadrature axis level of the multilevel inverter voltage, the larger the number of
voltage V ref
q are evaluated. Depending on the reference direct axis voltage vectors generated, hence giving more degrees of freedom
and quadrature axis voltage, the SVM generates the switching in choosing suitable voltage vectors for control purposes. The
pulses of the multilevel inverter. It is used to provide the supply of multilevel inverter normal and abnormal condition outputs are the
the induction motor. required dataset to develop the motor torque and flux. Depending
on the flux and torque variation, the ANN has been trained. During
3.1 Multilevel inverter the testing time the ANN develops the reference direct and
quadrature axis voltage, which are converted into controlling
The five level inverter contains two number of bridges connected pulses to control the five-level inverter switches.
with the two number of dc generators, which is known by a single-
phase topology. Three-phase topology can be realised by star or
delta connection of three single phase topologies. Generally the H-

IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137 1131
© The Institution of Engineering and Technology 2017
 N
r1NN (out) = α1 + ∑wi 21
r1NN (n)
n=1
N
r2NN (out) = α2 + ∑wi 22
r2NN (n) (9)
n=1
N
rkNN (out) = αk + ∑ wknk rkNN(n)
n=1

where, α1, α2 and αk are the bias function of the node 1, 2 and k
respectively

1
Fig. 2  Structure of feed forward neural network r1NN (n) =
1 + exp( − w1n1r1 − w2n1r2)
3.2 Artificial neural network 1
r2NN (n) = (10)
1 + exp( − w2n2r2 − wkn2rk)
The neural network is one of the AI techniques [32] which
workings are based on the training and testing process. Here the 1
rkNN (n) =
reference quadratic and direct axis voltage can be identified by the 1 + exp( − wknkrk − w1nkr1)
ANN technique, which works on the basis of a machine learning
approach that models a human brain and consists of a number of Step 5: The new weights of the each neurons of the network are
artificial neurons. The presented neurons have the interior updated by wnew = wold + Δw. Here, wnew is the new weight, wold is
connections and each neuron in ANN receives a number of inputs, the previous weight and Δw is the change of weight of each output.
depending on the activation functions of the ANN results in the The change of weight is determined as follows:
output level of the neuron. The learning task is given in the form of
examples, which is known as training examples. Here, the ANN's 1
Δw1 = δ ⋅ r1 ⋅ BPerror
performance is divided into two stages, which is explained as
follows. Δw2 = δ ⋅ r2 ⋅ BP2error (11)

ANN stage 1: Evaluation of reference quadrature axis voltage V q ∗ Δwk = δ ⋅ rk ⋅ BPkerror

with the help of input parameter change in torque ΔT .
ANN stage 2: Evaluation of reference direct axis voltage V ∗d with where, δ is the learning rate (0.2–0.5).
Step 6: Repeat the above steps till the BPerror gets minimised
the help of input parameter change in flux Δϕ.
BPerror < 0.1.
This section describes the combined structure of both the stages of Once the neural network training process is completed, the
ANN. In general ANN consists of three layers like input layer, network is trained well for the identifying r(k) of the input. Back
hidden layer and output layer. The change in motor torque ΔT and propagation is a systematic method for training multilayer artificial
change in flux Δϕ is given as the input layer (Fig. 2). The target networks. It is a multilayer forward network using extend gradient-
output is known as the reference quadrature and direct axis voltage descent based delta-learning rule, commonly known as back
r(k). propagation rule. The aim of this network is to train the net to
The inputs of the network are denoted as achieve a balance between the ability to respond correctly to the
[(ΔT 1, Δϕ1), (ΔT 2, Δϕ2)…(ΔT n, Δϕn)], the outputs of the network input patterns that are used for training and the ability to provide
good responses. Based on the output of the network, the control
are denoted as r(k) = [(V ∗q, V ∗d)1, (V q∗ , V d∗ )2…(V ∗q, V ∗d)n]. This network pulses of the five-level inverter have been decided by using the
is trained by back propagation training algorithm. The back- SVM technique. Then the proposed method is implemented in the
propagation algorithm is one of the most famous algorithm to train MATLAB platform and the results were analysed in Section 4.
a feed forward network [33]. The back propagation training
algorithm is divided into two phases that are named as propagation
4 Results and discussions
and weight update. The network training steps are described as
following. The proposed mutual method is implemented in MATLAB/
simulink 7.10.0 (R2012a) platform, 4 GB RAM and Intel(R)
3.2.1 Back propagation learning algorithm steps: Step 1: core(TM) i5. Then the proposed method effectiveness is analysed
Initialisation of the input layer, hidden layer, and output layer by comparing the different technique and different case studies, i.e.
weights of the neural network, i.e. change in motor torque ΔT , case 1 and case 2. In the proposed method, the ANN is utilised for
change in flux Δϕ and residual r(k). the DTC of the induction motor. The input parameters of the ANN
Step 2: Learning the network according to the input and the like change in flux and torque has been identified by using the
corresponding target. motor input voltage and current. Depending on the dataset the
Step 3: Calculate the back propagation error of the target ANN was trained effectively with the target of reference
r1, r2 and rk quadrature axis and direct axis voltage. The structure of the
proposed method in the MATLAB platform is described in Fig. 3.
The components ratings present in the model are explained in
1
BPerror = r1NN (tar) − r1NN (out) Table 1. Then the effectiveness of the proposed method is analysed
2
BPerror = r2NN (tar) − r2NN (out) by the different cases, which are explained as follows.
(8)
Case 1: This section describes the testing of the proposed
BPkerror = rkNN (tar) − rkNN (out) method's effectiveness. Here, the torque controller has been
continuing with the proposed ANN stage 1 controller but the flux
controller is dropped the ANN stage 2 controllers. Depending on
where, rkNN (tar) is the network target of the kth node and rkNN (out) is
these changes the performance variations of the system has been
the current output of the network. noted. The stator current during the above mentioned scenario is
Step 4: The current output of the network is determined as explained in Fig. 4a. It was seen that the stator current gets
follows, disturbed and it is not fully controlled at the operation period. The

1132 IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137
© The Institution of Engineering and Technology 2017
Fig. 3  Structure of the proposed method

Table 1 Implementation parameters condition, i.e. 1.8 Nm. The induction motor stator flux is analysed
Parameters Values in Fig. 5d, which describes the flux variation during the induction
nominal power 40 MVA motor operating condition. Here, the stator flux experiences the
distortion at the −0.015 to 0.015 wb. The mentioned stator flux,
rated voltage 400 V
rotor speed and the motor torque is affected by the supply
rated frequency 50 Hz induction motor stator current. In the case 2 scenario the induction
pole pairs 2 motor stator current has 6.77% of THD, which is illustrated in
stator resistance 3.67 Ω Fig. 5e. Then the proposed method is applied for the induction
stator inductance 0.0269 H motor DTC scheme, i.e. both ANN stages are involved for flux and
rotor resistance 2.1 Ω torque controlling. During the operation condition the performance
rotor inductance 0.0269 H of the proposed method is evaluated by using the following
analysis.
mutual inductance 0.0354 H
The induction motor stator current by using proposed
intelligence DTC scheme is described in Fig. 6a. In the figure
single phase stator current out of three-phase is illustrated, which
induction motor speed is measured and analysed in Fig. 4b. From experienced reduced distortion in the whole period 0.2 s. Then the
the figure we obtained that the induction motor needed 0.9 s to induction motor speed is analysed with the mentioned supply stator
achieve the maximum overshoot value; it requires high rise time current, which is illustrated in Fig. 6b. It was seen that the rotor
and it does not make constant speed at the end of the operating speed of the induction motor is measured for 5 s. The speed curve
condition. The induction motor speed is experienced in stability at settles down early; it also contains minimum rise time and peak
the presence of the case 1 condition. overshoot time. Hence it makes constant speed at 1.1 s. The
During the absence of flux controller ANN, the induction motor induction motor torque using the proposed method is explained in
torque is described in Fig. 4c. Here, the torque is experienced a Fig. 6c. It illustrates that the motor torque is smaller than the other
negative polarity during the starting condition, which is going to conditions. Here, the torque reaches negative polarity at the initial
minimum value at −3 Nm at 0.9 s. Then it enters the positive time; it reaches the positive polarity at the time and increased the
polarity and makes increasing values at the end of the operating torque value at 1 Nm at the end of the operation. The induction
condition. The induction motor stator flux is analysed in Fig. 4d, motor stator flux using the proposed method is explained in
which describes the flux variation during the induction motor Fig. 6d. The stator flux contains reduced disturbance at the
operating condition. Here, the stator flux experiences the distortion operation period. Here the flux value is takes distortion by −0.005
at the −0.015–0.009 wb. The mentioned stator flux, rotor speed and to 0.005 wb. The stator current THD by using the proposed method
the motor torque is affected by the supply induction motor stator is analysed in Fig. 6e. From the figure we obtained, the stator
current. In case 1 scenario the induction motor stator current has current at the proposed DTC scheme has 5.27% of THD, which is
6.40% of THD, which is illustrated in Fig. 4e. better than the other techniques. Then, the obtained result was
Case 2: This section describes the evaluation about the compared to prove the effectiveness of the proposed method. The
proposed method's effectiveness, during which the torque induction motor speeds by using different techniques are explained
controller has been dropped at the ANN stage 1 controller but the in Fig. 7a. Here, the induction motor DTC with PI controller, case
flux controller is allowed with the ANN stage 2 controllers. 1, case 2 and proposed method is explained. From that we know
Depending on these ultimate changes the performance variations of that the proposed method contains minimum rise time, peak over
the system has been noted. The stator current during the above shoot and settling time compared to other techniques. The motor
mentioned scenario is explained in Fig. 5a. It is also seen that there torque and stator flux comparison is given in Figs. 7b and c. From
is not effective supply to the induction motor, because the stator the comparison analysis, it is shown that the proposed method has
current gets much more disturbance and it is not fully controlled at minimum torque and less stator flux distortion compared to the
the operation period. The induction motor speed at the case 2 other techniques.
scenario is measured and analysed in Fig 5b. From the figure we
obtained that the induction motor have high amount of time to
achieve the maximum overshoot value; it requires high rise time 5 Conclusion
and it does not make constant speed at the end of the operating This paper explains an intelligence DTC of induction motor using
condition. The induction motor speed is experienced in stability at five-level inverter with SVM. Here the ANN is utilised for the
the presence of the mentioned operating condition. DTC of the induction motor. The reference quadrature axis and
During the case 2 operating condition, the induction motor direct axis voltage is assumed as the target of the ANN with the
torque is described in Fig. 5c. It clearly describes the torque which corresponding input parameters like the change in torque and
experiences a negative polarity during the starting condition, so it change in flux. Depending on these dataset, the ANN has been
reaches the value at −3 Nm at 0.9 s. Then it enters the positive trained using the back propagation algorithm. Then the proposed
polarity and makes the increasing values at the end of the operating system performance has been validated through the simulations.
IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137 1133
© The Institution of Engineering and Technology 2017
Fig. 4  Induction motor speed is measured and analysed
(a) Induction motor stator current, (b) Induction motor rotor speed, (c) Induction motor torque, (d) Induction motor stator flux THD analysis of stator current (e) THD

The effectiveness of the proposed method has been analysed by

different case studies and different techniques. From the results
analysis, we could justify that the proposed intelligence technique
has an improved dynamic performance, smaller motor torque,
reduced flux ripple and lower THD, which is competent over the
other techniques.

1134 IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137
© The Institution of Engineering and Technology 2017
Fig. 5  There is not effective supply to the induction motor
(a) Induction motor stator current, (b) Induction motor rotor speed, (c) Induction motor torque, (d) Induction motor stator flux (e) THD

IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137 1135
© The Institution of Engineering and Technology 2017
Fig. 6  Induction motor stator current by using proposed intelligence DTC
(a) Induction motor performance using proposed technique (a) Stator current, (b) rotor speed, (c) Torque, (d) Stator flux, (e) THD

1136 IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137
© The Institution of Engineering and Technology 2017
 [6] Wei, X., Chen, D., Zhao, C.: ‘Minimization of torque ripple of direct-torque
controlled induction machines by improved discrete space vector
modulation’, J. Electr. Power Syst. Res., 2004, 72, (2), pp. 103–112
[7] Chen, C.Y.: ‘Sliding mode controller design of induction motor based on
space-vector pulse width modulation method’, Int. J. Innov. Comput. Inf.
Control, 2009, 5, (10-B), pp. 3603–3614
[8] Abbou, A., Mahmoudi, H.: ‘Performance of a sensorless speed control for
induction motor using DTFC strategy and intelligent techniques’, J. Electr.
Syst., 2009, 5, (3), pp. 64–81
[9] Gendy, E.E., Ibrahim, A.F., Saraya, S.F., et al.: ‘A sliding mode controller for
a three phase induction motor’, Int. J. Comput. Appl., 2013, 64, (11), pp. 33–
39
[10] Jalluri, S.R., Ram, B.V.S.: ‘Direct torque control method using fuzzy logic for
IM drives’, Int. Int. J. Adv. Res. Electr. Electron. Instrum. Eng., 2012, 1, (4),
pp. 275–281
[11] Kennel, R., Kholy, E.E.E., Mahmoud, S., et al.: ‘Improved direct torque
control for induction motor drives with rapid prototyping system’, J. Energy
Convers. Manage., 2006, 47, (13–14), pp. 1999–2010
[12] Faiz, J., Sharifian, M.B.B.: ‘Comparison of different switching patterns in
direct torque control technique of induction motors’, J. Electr. Power Syst.
Res., 2001, 60, (2), pp. 63–75
[13] Ryu, J.H., Lee, K.W., Lee, J.S.: ‘A unified flux and torque control method for
DTC-based induction-motor drives’, IEEE Trans. Power Electron., 2006, 21,
(1), pp. 234–242
[14] Gopi, A., Kiran, T.V.: ‘Direct torque control of induction motor employing
differential evolution algorithm’, Int. J. Adv. Res. Electr. Electron. Instrum.
Eng., 2013, 2, (9), pp. 4581–4587
[15] Casadei, D., Serra, G., Tani, A., et al.: ‘Assessment of direct torque control
for induction motor drives’, J. Bull. Pol.Acad. Sci. Tech. Sci., 2006, 54, (3),
pp. 237–254
[16] Zaid, S.A., Mahgoub, O.A., Metwally, K.A.E.: ‘Implementation of a new fast
direct torque control algorithm for induction motor drives’, J. IET Electr.
Power Appl., 2010, 4, (5), pp. 305–313
[17] Gadoue, S.M., Giaouris, D., Finch, J.W.: ‘Tuning of PI speed controller in
DTC of induction motor based on genetic algorithms and fuzzy logic
schemes’. Proc. of 5th Int. Conf. on Technology and Automation, 2005, pp.
85–90
[18] Lee, K.B., Song, J.H., Choy, I., et al.: ‘Torque ripple reduction in DTC of
induction motor driven by three-level inverter with low switching frequency’,
IEEE Trans. Power Electron., 2002, 17, (2), pp. 255–264
[19] Vasudevan, M., Arumugam, R., Paramasivam, S.: ‘Real time implementation
of viable torque and flux controllers and torque ripple minimization algorithm
for induction motor drive’, J. Energy Convers. Manage., 2006, 47, pp. 1359–
1371
[20] Toufouti, R., Meziane, S., Benalla, H.: ‘Direct torque control for induction
motor using intelligent techniques’, J. Theor. Appl. Inf. Technol., 2007, 3, (3),
pp. 35–44
[21] Uddin, M.N., Hafeez, M.: ‘FLC-based DTC scheme to improve the dynamic
performance of an IM drive’, IEEE Trans. Ind. Appl., 2012, 48, (2), pp. 823–
821
[22] Metidji, B., Taib, N., Baghli, L., et al.: ‘Low-cost direct torque control
algorithm for induction motor without AC phase current sensors’, IEEE
Trans. Power Electron., 2012, 27, (9), pp. 4132–4139
[23] Wang, Y., Deng, Z.: ‘An integration algorithm for stator flux estimation of a
direct-torque-controlled electrical excitation flux-switching generator’, IEEE
Trans. Energy Convers., 2012, 27, (2), pp. 411–420
[24] Zhang, Y., Zhu, J., Zhao, Z., et al.: ‘An improved direct torque control for
three-level inverter-fed induction motor sensorless drive’, IEEE Trans. Power
Electron., 2012, 27, (3), pp. 1502–1513
[25] Sutikno, T., Idris, N.R.N., Jidin, A.: ‘A review of direct torque control of
induction motors for sustainable reliability and energy efficient drives’,
Renew. Sustain. Energy Rev., 2014, 32, pp. 548–558
[26] Ammar, A., Bourek, A., Benakcha, A.: ‘Nonlinear SVM-DTC for induction
motor drive using input-output feedback linearization and high order sliding
mode control’, ISA Trans., 2017, 67, pp. 428–442
[27] Naik, V., Panda, A., Singh, S.P.: ‘A three-level fuzzy-2 DTC of induction
motor drive using SVPWM’, IEEE Trans. Ind. Electron., 2016, 63, (3), pp.
1467–1479
Fig. 7  Induction motor speeds by using different techniques
[28] Ouhrouche, M., Errouissi, R., Trzynadlowski, A.M., et al.: ‘A novel
(a) Induction motor rotor speed under different conditions, (b) Induction motor torque predictive direct torque controller for induction motor drives’, IEEE Trans.
under different conditions, (c) Induction motor flux under different conditions Ind. Electron., 2016, 63, (8), pp. 5221–5230
[29] Bermudez, M., Prieto, I.G., Barrero, F., et al.: ‘Open-phase fault-tolerant
direct torque control technique for five-phase induction motor drives’, IEEE
6 References Trans. Ind. Electron., 2017, 64, (2), pp. 902–911
[1] Manuel, A., Francis, J.: ‘Simulation of direct torque controlled induction [30] Lascu, C., Jafarzadeh, S., Fadali, M.S., et al.: ‘Direct torque control with
motor drive by using space vector pulse width modulation for torque ripple feedback linearization for induction motor drives’, IEEE Trans. Power
reduction’, Int. Int. J. Adv. Res. Electr. Electron. Instrum. Eng.., 2013, 2, (9), Electron., 2017, 32, (3), pp. 2072–2080
pp. 4471–4478 [31] Micusik, D., Stopjakova, V., Benuskova, L.: ‘Application of feed-forward
[2] Dabbeti, S., Lakshmi, K.V.: ‘Sensorless speed control of an induction motor artificial neural networks to the identification of defective analog integrated
drive using predictive current and torque controllers’, Adv. Electron. Electr. circuits’, Neural Comput. Appl., 2002, 11, pp. 71–79
Eng., 2013, 3, (9), pp. 1177–1196 [32] Nandy, S., Sarkar, P.P., Das, A.: ‘Training a feed-forward neural network with
[3] Purwahyudi, B., Suryoatmojo, H., Ashari, S.M., et al.: ‘Feed-forward neural artificial bee colony based back propagation method’, Int. J. Comput. Sci. Inf.
network for direct torque control of induction motor’, Int. J. Innov. Comput. Tech. (IJCSIT), 2012, 4, (4), pp. 33–46
Inf. Control, 2011, 7, (11), pp. 6135–6145 [33] Park, H., Han, J., Song, J.: ‘Control strategy for shunt active power filter
[4] Kumar, N., Singh, J.: ‘Direct torque control of three phase induction motor using hybrid multi-level inverter in AC electrified railway systems’, IETE J.
using MATLAB’, Int. Journal of research in applied science and engineering Res., 2012, 58, (6), pp. 451–458
technology, 2013, 1, (2), pp. 16–22
[5] Matic, P.R., Blanusa, B.D., Vukosavic, S.N.: ‘A novel direct torque and flux
control algorithm for the induction motor drive’. Proc. of IEEE Conf. on
Electric Machines and Drives, 2003, vol. 2, pp. 965–970

IET Power Electron., 2017, Vol. 10 Iss. 10, pp. 1129-1137 1137
© The Institution of Engineering and Technology 2017