SPEED CONTROL OF DTC-SVM FOR INDUCTION MOTOR DRIVE USING SELF-TUNING FUZZY PI CONTROLLER ANN ANFIS

: This paper deals with the performance analysis of three phase Induction Motor (IM) with Direct Torque Control based Space Vector Modulation (DTC-SVM). The DTC-SVM scheme is a kind of high-performance control of IM drives to improve the ripples of torque and flux in steady state, which one drawback of conventional DTC. DTC-SVM has three Proportional-Integral (PI) controllers, one used as the PI speed controller and other PI flux controller and PI torque controller, which are utilized to produce the stator voltage references ( 𝑉 𝑑𝑠 and 𝑉 𝑞𝑠 ) respectively. To improve the performance of motor we used the Self-Tuning Fuzzy PI (STFPI) controller and Adaptive Neuro Fuzzy Inference System (ANFIS) controller instead of PI controllers in DTC-SVM. The simulation results carried out using MATLAB/Simulink software package show that the response of speed motor with ANFIS is improved, more robustness, and reaches to the reference value faster than a conventional PI controller and STFPI controller and also reduced the torque ripple and less steady state error.


INTRODUCTION
This Induction Motors (IMs) are broadly utilized, particularly poly phase IMs, which are commonly utilized in industrial drives.IMs are now the favorite option for industrial motors in accordance with their robust building, nonexistence of brushes and thanks to modern power electronics that are able to the control of the motor speed [1].The IM control was treated with different techniques.This contains simple linear techniques for example Direct Torque Control (DTC), Field Oriented Control (FOC) and more complicated nonlinear techniques for example input-output linearization back stepping, sliding mode and passivity.Most of the suggested controllers utilize the speed sensor and require the constant the motor parameters.This hypothesis leads to poor control performances, particularly through transient operation.The common of these controls require utilization of coordinate transformation, whereas DTC approach can be utilized to directily control for flux and torque without current controllers [2].The DTC method was first suggested for IM.DTC method for medium and low power applications is presented by Takahashi and Noguchiand, while DTC method for high power applications is presented by Depenbrock and it is commonly utilized in industry [3].Conventional DTC usages the magnitude errors of the stator flux and torque as the inputs for hysteresis comparators to determine the voltages of stator vector ( V qs and V ds ) are applied to terminals of the motor.The complex plane is divided into 6-sectors, and a switching table is purposed to get the required vector based on the outputs of the hysteresis comparators [4].The most common drawbacks of conventional DTC are a high ripple of torque and slow transient response to torque changes during startup.The performance of torque is improved by developed several techniques.One of the developed techniques utilized for reducing the ripples is DTC with Space Vector Modulation (SVM).SVM was first suggested by a researchers group from Germany in 1980s.SVM techniques offer several benefits, including the torque ripple is less, better DC bus utilization, less losses of switching, minimum Total Harmonic Distortion (THD) in the AC motor current and easier implementation in digital systems [3].In DTC-SVM approach, the PI controllers are utilized instead of the hysteresis comparators for torque and flux in conventional DTC.However, the main disadvantages of DTC-SVM with a simple PI controllers are the performance sensitivity to variations of the system parameters and the insufficient rejection load changing and external disturbances [5].The disadvantages of the conventional PI controllers are performance sensitivity to covariance in parameters of the system, and the actuality that when utilizing fixed of the controller gains may not offer the required performance of speed under covariance in the operating conditions and parameters of the motor [6].FLC can be utilized to eliminate the disadvantages of conventional PI controller [7].FLC has demonstrated its effectiveness in non-linear, complex and inaccurate processes which it is impossible to utilize the standard model based control techniques impractically [8].FLC based intelligent controller can be utilized instead of the PI controller.FLC has some advantages over conventional controllers such as simplicity of control, low cost, and the capability to design without knowledge the precise mathematical model of the plant [9].FLC is utilized in the controller design autonomously or hybrid with PI controller.FLC provides faster control and superior but the major problem of design is to specify a consistent and complete set of rules and in the form of Membership Functions (MFs).A lot of trials and errors must be made to get the desired response that takes a long time [10].The Artificial Neural Networks (ANNs) have the advantages of very fast parallel computation, impunity from input harmonic ripples and fault tolerance characteristics [11].However, ANN only is inadequate if the training data are not adequate to deal with all the operating modes.The disadvantages of FLC and ANN can be eliminated by the usage of ANFIS [10].ANFIS is a fuzzy scheme that usages a learning procedure resulting from the principle of ANN to know its parameters (fuzzy sets and rules) by dealing with data examples.By mingling both FLC and ANN, it is probable to obtain the benefits of together controls in one implementation [12].There are several studies related to DTC_SVM and PI controller ,STFPI controller and ANFIS such as: In 2010 Fayez G. Areed, e.t al., suggested a direct torque of the IM drive controlled by ANFIS.The suggested control scheme employs amplitude of the stator flux and the torque errors by using an ANFIS to change the angle and the amplitude of the required reference voltage.The simulation outcomes by utilization ANFIS are compared with the conventional DTC.The comparison results of DTC with ANFIS showed a reduction in the ripples of torque and flux in transient and steady-state response and fast stator flux response in the transient state [13].In 2013, Arun Kumar R, and Febin Daya J. L., compared between conventional PID, Fuzzy, and STFPID controllers based speed control system for a current source PWM inverter fed IFOC of IM Drives.In this study, the PI control system was substituted by STFPID based intelligent controller.The FLC uses various forms of Membership Functions (MFs) for all parameters in order to obtain an efficient controller.By comparing the results with the PID and FLC, these results showed that the suggested STFPID controller recognizes a good performance at the sudden variations with a quick settling time.As well, it has no overshoot and better behavior than FLC and PID controller [14].In 2015, A. Durga Bhavani, et. al., presented an intelligent control scheme with ANFIS controller to regulate the speed of the DTC of IM drive.The performance of conventional DTC in IM drives containing PI controller has a complex tuning and overshoot.Therefore, this controller employs the error that obtained from the reference speed and estimated speed which produces the estimated torque and compared with the actual torque.The suggested ANFIS based IM drive is providing better performance and provide less THD in steady state current ripple when compared with conventional DTC algorithm [15].

INDUCTION MOTOR MODELING
The equivalent circuit of the Squirrel Cage Induction Motor (SCIM) in (d − q) axis is shown in Fig. 1.The three phase voltage stator of an IM can be given by the following equations [16].
) These three phase balance voltages are transferred into the two phase stationary reference frame of (d − q) axis.This can be given by Eq. ( 4) and Eq. ( 5).
Then, the voltage in the (d − q) axis Figure 1 Equivalent Circuit of SCIM in (d − q) The mathematical model of three-phase SCIM in the stationary reference frame (d − q) can be given by the following Eqs.[17].
V ds = (R s + pL s )i ds + pL m i dr − ωψ ds V qs = (R s + pL s )i qs + ωψ qs + pL m i qr V dr = 0 = (R r + pL r )i dr + pL m i ds − (ω − ω r )ψ dr (8) ) Where ψ ds , ψ qs are the stator flux linkages in the (d − q) frame in Eq.( 11) and ψ dr , ψ dr are rotor flux linkages in the(d − q) frame in Eq.( 12) , while p = d/dt.ψ qs = L s i qs + L m i qr & ψ ds = L s i ds + L m i dr (11) The torque and speed can be determined as following equations [17,5] Where V ds , V qs , and V dr , V qr are the voltages for stator and rotor in the (d − q) frame respectively, while R s , R r , ω r , ω are stator resistance, rotor resistance ,rotor angular speed, and the electrical angular speed respectively.The stator, rotor and mutual inductance are L s, L r and L m , respectively.DTC usas the hysteresis band to directly control the torque and flux of the machine.The main benefits of DTC are the rapid response for torque and robust, no requisites for coordinate transformation no requisites for current regulators and PWM pulse generation.The main drawback of the DTC drive is the ripples of the flux and the torque in steady state.The torque and flux pulses impact the accuracy of the speed estimation.There are two approaches to reduce the ripples for flux and torque of the DTC drives.The first approach is multilevel inverter and the second approach is SVM.In the first approach, the complication and the cost will be rised while in the second approach the ripples of flux and torque can be decreased [4].The suggested DTC-SVM system that is shown in the Fig. 2 has used three PI controllers for processing errors of speed, flux and torque.The torque reference (T e * ) is generated by speed error, that it is processed through the PI speed controller ,while the PI flux controller and PI torque controller are producing the command voltages V ds and V qs respectively [18].
Figure 2 The suggested DTC_SVM with PI controller The flux estimator model is dependent on stator resistance (R s ) only, the stator flux linkages ψ ds and ψ qs can be determined from Eq.( 16) and Eq.( 17) respectively, while the magnitude of the stator flux linkage ψ s and the phase angle of the stator flux linkage θ s are given in Eq.( 18) and Eq.( 19) respectively [19].

PI CONTROLLER
Proportional-Integral (PI) controller is easy to design, simple structure and low cost.It is most broadly utilized in industrial application because of these benefits for PI controller.The output signal of PI controller consists of a sum of errors and the integral of that errors [7].
The transfer function for PI controller is given by Eq.( 20) Torque and Flux estimator The proportional controller will decrease the steady-state error and rise time.But, will never remove the error.The integral control will decrease the error close to zero.But, it has a negative effect on the response of the speed and overall stability of the system [20].

SELF-TUNING FUZZY PI (STFPI) CONTROLLER
The STFPI controller is a mixture from the concept of fuzzy logic and the conventional PI controller [20].FLC utilized to tune the parameters of the PI controller; it gets scaled values of the error (E) and change of error (dE/dt).Its output is updating the parameters of PI controller based on a set of rules to keep good control performance even in the existence of drive non-linearity and variation of parameters.[13,14].The FLC is comprised of four main stages: Fuzzification, knowledge base, decision making and defuzzification as demonstrated in Fig.
3 [21].Fuzzy model is a collection of ifthen rules with indistinguishable predicates that use a fuzzy reasoning such as Mamdani and Sugeno models.Mamdani type systems can be utilized to model any inference system in which the output Membership Functions (MFs) are either linear or nonlinear whereas Sugeno type produces either linear or constant output [14].
The performance of STFPI controller can be shown as following [4].
1-Dynamically adjusts the PI controller parameters to guarantee the stability of system over the wide speed range.
2-Fast response of speed.
3-Small errors for steady state and less overshoot.The equations for the STFPI controller process are given below [21].
Where: ∝ is the scaling factor is obtained from the FLC.  &   are obtained from the FLC, while   &   are the new gains of PI controller

ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM (ANFIS)
ANFIS put forward by Jang in the early 90s, mixing the concepts of FLC and ANNs to form a hybrid intelligent system that improves the capability to learn and adapt automatically [8].ANFIS is one of the best reciprocating among ANN and FLC, which provides the adaptability and interpolation in accordance with the ANN and smooth control in accordance with the FLC.Some of the benefits of ANFIS are compactness of model, they require a faster convergence and less size of training set than typical feed forward neural network.The nonlinear cartographical in a ANFIS is obtained by utilizing a fuzzy MF based ANN [10].The ANFIS is an adaptive network multilayer in Fig. 5.There are two different types of nodes according to their features: square (adaptive) nodes and circular nodes.The first type contains parameters, while the second type has no paramters [22].The Fuzzy Inference System (FIS) usually utilized in ANFIS is first order Sugeno fuzzy model due to its easiness, efficiency of computational and built-in optimization techniques and adaptation [23].The Sugeno FIS is like to Mamadani format but that the output MFs are singleton spikes instead of a distributed fuzzy set.Utilization singleton output simplifies the defuzzifIation step [24].
ANFIS system uses the following rule base [23].
Where A i and B i are linguistic variables, while a i , b i and r i are constants The ANFIS comprises five layers; the functions of these layers can be listed below [25].
Layer1: Every node of this layer transforms the input signal x or y utilizing MF (fuzzification).The most commonly utilized MF is a bell-shaped or Gaussian function.Layer2 : Each node of this layer, discriminated as П, performs the intersection of the input signals, simulates a logical AND operation, and sends a value to the output: Essentially, every node specifies an activating rule force w i .In actuality, any operator of T-norm, which summarizes the AND operation can be utilized in these nodes.Layer3: Every node in this layer determines the weight of the normalized rules.
Layer4: In this layer, nodes produce the output variables values: ) Layer 5.In this layer is obtained the neural network output and carried out the defuzzification: Figure 5 Input MF and outputs of proposed ANFIS controller

SIMULATION AND RESULT
Simulink implementation of the DTC-SVM for IM drive is shown in Fig. 6, it was carried out using MATLAB/SIMULINK software.The speed reference and stator flux reference value are 350 rad/sec and 0.476 Wb. respectively.The motor is unloaded at startup and the torque load (T l ) 8 Nm is applied at t 2s.The switching frequency of SVM is 10 kHz, while the paramters of IM are used in this paper as shown in table-1.
The results of simulation show the comparison between the conventional PI controller (trial and error), STFPI controller and ANFIS based on DTC-SVM of IM drive.The Simulink model of DTC-SVM for IM drive in Fig. 6 is replacing all PI controllers by STFPI controllers.Three STFPI controllers are observing the pattern of the errors for speed , flux and the torque and correspondingly updates the outputs K p and K i .The structure of STFPI controller for speed, torque and flux is the same and shown in Fig. 7. MATLAB/Fuzzy Logic Toolbox is employed to simulate MF of FLC is choosing the triangular shape and No. of MFs are five for inputs and outputs as shown in Fig. 8, while the value of K pf and K if are calculated from a fuzzy rule base defined in function of the control ( E and dE/dt) as shown in the table-2.The Centroid defuzzification method is employed here.The Simulink model of DTC-SVM for IM drive in Fig. 6 is replacing all PI controllers by ANFIS controllers to achieve precision speed control.The ANFIS controller produces one output such as the reference torque (T e * ), V ds or V qs based on (E and dE/dt) for speed, torque and flux respectively.The ANFIS designed through the ANFIS editor, the grid partition is utilized to produce FIS when the hybrid method is used for optimization.In ANFIS controller utilized 25 if -then rules, Gaussian MF type and the number of training Epochs are used 2000 epochs.Result ANFIS turning is shown in Fig. 9.While Fig. 10 and Fig. 11 show that input MFs and ANFIS model structure respectively.Figure 12 to Fig. 15 show the response of speed and torque at various loaded when using conventional PI controller (trial and error), While Fig. 16 to Fig. 19 illustrate the simulation results of speed and torque under same load variation.The simulation results of the speed and torque at various load using ANFIS are given in Fig. 20 to Fig. 23.The results of simulation show that the ANFIS controller realizes a good response of motor speed to sudden changes without overshoot, a fast settling time, the settling time is 0.166 s while the settling time for STFPI controller and conventional PI controller are 0.33s and 0.46s respectively, less steady state error, less ripple in torque response and improved the performance of motor when compared with both PI and STFPI controllers.

Torque (N.m)
Figure 21 The torque response for used ANFIS at no-load Figure 22 The speed response for used ANFIS controller with T l =8 Nm at t 2s Figure 23 The torque response for used ANFIS controller with T l =8 Nm at t 2s i ds R s − ωψ ds + L ls L lr + (ω − ω r )ψ dr − R r i dr i qs R s + ωψ qs − L ls L lr − (ω − ω r )ψ qr + R r i qr + 2. DIRECT TORQUE CONTROL WITH SPACE VECTOR MODULATION (DTC-SVM) where: K p & K i are the proportional and the integral gains respectively, U(s) is the output control signal which represents torque reference (T e * ) in DTC-SVM drive and E(s) is the error which represents the difference among the desired and the actual of motor speed and it is specified by Eq.(21) E = ω r * − ω r

Figure 3 Figure 4
Figure 3 Block diagram of FLC The fuzzy tuner has two inputs (E & dE/dt) and two outputs: K pf a& K if .The structure of STFPI controller is shown in the Fig.4.

Figure 6 Figure 7 Figure 8 Figure 9 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 18 Figure 19 Figure 20
Figure 6 The Simulink model of the DTC_SVM for IM drive Table-1 Parameters of IM Parameter Value The power rating of IM 3 hp Voltage applied 220 V Frequency 60 Hz No of poles(P) 4