Majlesi Journal of Mechatronic Systems Vol. 8, No. 1, March 2019
11
ANFIS Speed Controller of IM Drives with Threelevel DTCBased Neural Network
enbouhenniHabib B
Laboratoire d’Automatique et d’Analyse des Systèmes (LAAS), Departement de Génie Électrique, Ecole Nationale Polytechnique d’Oran Maurice Audin, Oran,
Algeria. Email: habib0264@gmail.com
Received: October 2018 Revised: December 2018 Accepted: January 2019
ABSTRACT:
Direct torque control (DTC) is a control technique in AC drive systems to obtain highperformance torque control. In this paper, we use a technique based on the neural network and ANFIS controller in order to reduce the torque ripple, stator flux ripple and THD value of stator current in the induction machine (IM). The effect is highlighted by considering the machine controlled by the 24 sectors DTC. In this paper, we propose to replace conventional selector states of switches inverter by a neural selector able to generate the same signals to control this inverter, on the other hand, the classic PI controller of speed is proposed based on Adaptive Neuro Fuzzy Inference System (ANFIS) controller. Simulation results are presented and show the effectiveness of the proposed control scheme.
KEYWORDS:
DTC, ANFIS, Neural Network, Induction Machine, 24 sectors.
1. INTRODUCTION
Induction machine (IM) is an important class of electric machines which finds wide applicability as a motor in industry. An IM is an AC motor, where power is supplied to the rotor by means of electromagnetic induction [1]. However, the control of IM is complex due to the nonlinear nature and the parameters changes with operating conditions [2]. In 1970, F. Blaschke presented the first paper on the fieldoriented control (FOC) for induction motors. In recent years, commercial applications of vector controlled induction motor drives have greatly increased. The disadvantage of this control scheme is the inclusion of the pulse encoder, PWM modulator and the indirect torque control [3]. However, the complexity of field oriented algorithms led to the development in recent years of many studies to find out different solutions for the induction motor control having the features of precise and quick torque response [4]. The direct torque control strategies were
proposed amidst the 1980’s for inducti
on motor [5] has been recognised to be a variable solution to achieve these requirements [4]. The main advantages of DTC are the simplicity of the control scheme and its unresponsiveness to parameters variations (except stator resistor) [6]. The DTC drive consists of DTC controller, torque, and flux calculator, and a voltage source inverter (VSI) as depicted in Fig. 1 [7]. The configuration is much simpler than the FOC system. The technique can be easily implemented with using hysteresis controllers that is one for torque and for stator flux. The implementation of DTC of IM, although simple but requires a fast processor to perform realtime simulation of electromagnetic torque and stator flux based on sampled terminal variables [8]. Conventional DTC has also some disadvantages such as possible problems during starting, lowspeed operation, high requirements upon flux and torque estimation and variable switching frequency. These are disadvantages that we want to remove by using and implementing modern resources of artificial intelligence [9]. In this paper, we present the performance of the sensorless speed control of induction motor using a speed proportional integral (PI) ANFIS controller. The artificial neural network (ANN) the replaces the switching table of the conventional DTC control. An intelligent technique is used to improve dynamic response performance and decrease the torque ripple and stator flux ripple.
2. THREELEVEL NPC INVERTER
Fig. 2 show the circuit of a threelevel diode clamped inverter and the switching states of each leg of the inverter. Each leg is composed of two upper and lower switches with antiparallel diodes. Two series DClink capacitors split the DCbus voltage in half, and six
clamping diodes confine the voltage across the
Majlesi Journal of Mechatronic Systems Vol. 8, No. 1, March 2019
12 switches within the voltage of the capacitors, each leg of the inverter can have three possible switching states, 2, 1 or 0 [10]. The representation of the space voltage vectors of a threelevel inverter for all switching states is given by Fig. 3 [11, 12]. According to the magnitude of the voltage vectors, the voltage vectors can be partitioned into four groups: the zero voltage vectors V
0
, the large voltage vectors, the middle voltage vectors and the small voltage vectors [10].
Fig. 1.
Conventional direct torque control scheme.
Fig. 2.
Schematic diagram of a threelevel inverter.
3. 24 SECTORS DTC
The wellknown DTC strategy is based on flux and torque control using hysteresis comparators. These controllers use the estimated errors of the control variables at each sampling time of operation. The considered flux and torque controllers ensure the separate control of these two variables, as for the DC drives. When the level of torque or stator flux passes to the high or low hysteresis limit, a suitable voltage vector is applied to bring back each variable in its corresponding band [13].
Fig. 3.
Space vector diagram of threelevel inverter.
DTC is the first technique to control the “real”
motor control variable of torque and flux [14]. The stator flux can be evaluated by integrating the estimated stator voltage equation: DC bus
Switching Table
S
abc
φ
s
*
+ +  
I
s,ab
V
s,ab
φ
s
Flux, Zone and Torque Estimation
T
em
Converter
w
W*
+
PI controller

T
em
*
Cflx Ccpl
IM
Majlesi Journal of Mechatronic Systems Vol. 8, No. 1, March 2019
13
dt sst ssdt sst ss
i Rvi Rv
)0()0(
(1) The electromagnetic torque can be estimated by:
)(23
ississ pTe
(2) The relation between stator flux magnitude and its
components in plane αβ is given by equation (3):
22
sss
(3) The actual position of the stator flux can be determined by equation (4), from the orthogonal flux components:
)ss(
arctg
(4) The hysteresis comparator for electromagnetic torque is shown in Fig. 4. The stator flux hysteresis comparator is shown in Fig. 5.
Fig.4.
Torque hysteresis comparator
.
Fig. 5.
Stator flux hysteresis comparator. Table 1 illustrate of proposed table of threelevel DTC with 24 sectors.
Table 1.
Switching table for threelevel DTC control
N Cflx 10Ccpl1011011
16820171119
2
16820171119
3
22 9 26 23 12 25
4
22 9 26 23 12 25
5
17 9 15 18 12 20
6
17 9 15 18 12 20
7
23 10 21 24 13 26
8
23 10 21 24 13 26
9
18 10 16 19 13 15
10
18 10 16 19 13 15
11
24 11 22 25 8 21
12
24 11 22 25 8 21
13
19 11 17 20 8 16
14
19 11 17 20 8 16
15
25 12 23 26 9 22
16
25 12 23 26 9 22
17
20 12 18 15 9 17
18
20 12 18 15 9 17
19
26 13 24 21 10 23
20
26 13 24 21 10 23
21
15 13 19 16 10 18
22
15 13 19 16 10 18
23
21 8 25 22 11 24
24
21 8 25 22 11 24
4. ANFIS SPEED CONTROLLER OF DTCANN
The principle of neural direct torque control (DTCANN) is similar to traditional DTC with threelevel NPC inverter. However, the classical PI controller of speed is replaced by the ANFIS controller, and the switching table is replaced by neural controller. The general structure of DTCANN with ANFIS controller is represented by Fig. 6.
4.1. Design of Speed Controller based on ANFIS
Adaptive neuro fuzzy inference system (ANFIS) is the combination of neural network (NN) and fuzzy logic controller (FLC). This hybrid combination enables to reduce the complexity of power intelligent system. In ANFIS, the fuzzy inference system is implied through the structure and neurons of the feedforward adaptive neural network [1]. The block diagram for the ANFIS based classic PI controller of speed is shown in Fig. 7. Then the designed ANFIS has two inputs namely, the reference speed and estimated speed while the output is the reference torque (Te
ref
). The structure of the ANFIS speed controller is shown in Fig. 8.
Majlesi Journal of Mechatronic Systems Vol. 8, No. 1, March 2019
14
Fig. 6.
DTCANN with ANFIS controller.
Fig.7.
ANFIS control of classic PI controller.
1Te refdu/dtNNETInputOutput 1W ref  W
Fig. 8.
Architecture of ANFIS speed controller.
4.2. Design of ANN based Switching Table
ANN is part of the family of statistical learning methods inspired by biological nervous system and are used to estimate and approximate functions that depends only on a large number of inputs [15]. A group of artificial neurons, which work in parallel, their inputs and outputs have the same destination from a layer. Each neural network must contain at least one layer of neurons but can join as many as someone projects [16]. The layer gathering the neurons which give the output of the neural networks is called the output layer. layers which contain the neurons interposed between the global inputs of the neural network and the inputs of the neurons from the output layer are called hidden layers. Usually, there are used feedforward NN which contains a hidden layer and an output layer [17]. The ANN has many models, but the usual model is the multilayer feedforward network using the error backpropagation algorithm. Such a neural network contains three layers: input layers, hidden layers and output layers [10]. The structure of the neural to perform the threelevel DTC with 24 sectors applied to IM satisfactorily was a neural network with 3 linear input node, 30 neurones in the hidden layer, and 3 neurons in the output layer, as shown in Fig. 9.
Outputa{1}Process Output 1Process Input 1Layer 2Layer 1 a{1} Input
Fig. 9.
Neural network structure for threelevel DTC with 24 sectors. The structure of layer 1 is shown in Fig. 10, and the layer 2 of ANN as it is shown in Fig. 11.
ANN Controller
S
abc
E
Tem
φ
s
*
DC bus
IM
+ +  
I
s,ab
V
s,ab
φ
s
E
φ
s
Flux, Zone and Torque Estimation
T
em
Converter
W*
+
ANFIS controller

w T
em
*
+  Te ref W ref W
ANFIS
Majlesi Journal of Mechatronic Systems Vol. 8, No. 1, March 2019
15
a{1}tansig+netsumbb{1}WIW{1,1}0Delays 1p{1}
Fig. 10.
Architecture of Layer 1.
a{2}purelin+netsumbb{2}WLW{2,1}0Delays 1a{1}
Fig. 11.
Architecture of Layer 2.
5. SIMULATIONS RESULTS
The simulations of the threelevel DTCANN with ANFIS speed controller for IM drive are compared with conventional threelevel DTC. The performance analysis is done with stator current, torque ripple and stator flux. The dynamic performance of threelevel DTC control of IM is shown in Fig. 12. The dynamic performance of the threelevel DTCANN control with ANFIS controller is shown in Fig. 13.
Fig. 12.
Dynamic responses of conventional threelevel DTC with 24 sectors for IM.