### SlideShare for iOS

by Linkedin Corporation

FREE - On the App Store

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

- Total Views
- 481
- Views on SlideShare
- 481
- Embed Views

- Likes
- 0
- Downloads
- 11
- Comments
- 0

No embeds

Uploaded via SlideShare as Adobe PDF

© All Rights Reserved

- 1. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 Neural Network Based Space Vector Pwm Control Of Induction Motor Neelima.A,Hari krishna.Ch *(Department of electrical and electronics engineering, jntu University, Hyderabad-10 ** (Department o electrical and electronics engineering, jntu University, Hyderabad-10.ABSTRACT The demand for high performance parameter variations, and the difficulties ofinduction motor drives have been increasing at a processing feedback signals in the presence ofsteady rise in the industry and control techniques harmonics.also have become sophisticated and it is a scope To limit the above problems, different types controlfor implement controllers for ac drives. The strategies are introduced in recent years. In that, theconventional control methods need modeling and control strategies are likelimited for linear systems. Most of controllers Scalar controlthose are designed for drives are linearized near Vector control etc.[3]their operating points. Therefore there is scope to Direct Vector Controlimprove these using artificial intelligence Indirect Vector Controltechniques. Artificial intelligent system contains This project emphasis indirect vectorhard computation and soft computation. control strategy.The inverter place a vital role inArtificial intelligent system has found high drives system. For inverter the switching pulseapplication in most nonlinear systems same as pattern are generated by using Pulse Widthinduction motors drive. Because artificial Modulation (PWM) controller.intelligent techniques can use as controller for In a PWM control in order to synthesis theany system without requirement to system inverter switches a desired reference stator a voltagemathematical model, it has been used in space vector need to fulfill few requirements like aelectrical drive control. Further, the necessity of constant switching frequency , and smallestintelligent controllers is to achieve the instantaneous deviation from its reference value.satisfactory dynamic performance of Minimum switching losses in the inverter and theconventional methods. lowest ripple current are also to be achieved. To This project considered two objectives: meet these requirements an average voltage vector isTo understand and develop conventional synthesized by means of the two instantaneous basicSVPWM technique and the same is applied to non-zero voltage vector that form the sector (ininduction motor drive; and To develop ANN which the average voltage vectors to be synthesizedmodel to realize the functionality of SVPWM. lies) and both the zero voltage vectors, such thatThe performances of these methods are verified each transition causes change of only one switchon induction motor drive in MATLAB/Simulink status to minimize the inverter switching loss.environment. Pulse Width Modulation variable speed drives are increasingly applied in many newKeywords-artifical neuralnetwork,field oriented industrial applications that require superiorcontrol,induction motor,neural based space vector performance. Recently, developments in powerpulse widh modulation, space vector pulse width electronics and semiconductor technology have leadmodulation improvements in power electronic systems. Hence, different circuit configurations namely multilevel1.1 INTRODUCTION inverters have become popular and considerable The control and estimation of induction interest by researcher are given on them. Variablemotor drives constitute a vast subject, and the voltage and frequency supply to a.c drives istechnology has further advanced in recent years. invariably obtained from a three-phase voltageThe control and estimation of ac drives in general source inverter. A number of Pulse widthare considerably more complex than those of dc modulation (PWM) schemes are used to obtaindrives, and this complexity increases substantially if variable voltage and frequency supply. The mosthigh performances are demanded. The main reasons widely used PWM schemes for three-phase voltagefor this complexity are the need of variable- source inverters are carrier-based sinusoidal PWMfrequency, harmonically optimum converter power and space vector PWM (SVPWM). There is ansupplies, the complex dynamics of ac machines, increasing trend of using space vector PWMmachine (SVPWM) because of their easier digital realization and better dc bus utilization. Space vector 104 | P a g e
- 2. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116modulation is based on representation of the three method, was invented by Hasse. The methods arephase voltages as space vectors. Most space vector different essentially by how the unit vectormodulation schemes generate the same required ( , ) is generated for the control. Theoutput voltage but differ in their performance with indirect vector control method is essentially therespect to THD, peak-to-peak ripple and switching same as direct vector control, except the unit vectorlosses. signals ( , ) are generated in feed Space Vector Modulation (SVM) wasoriginally developed as vector approach to Pulse forward manner [3].Width Modulation (PWM) for three phase inverters. Inverter requires the pulse pattern for theIt is a more sophisticated technique for generating switching the switches according to the indirectsine wave that provides a higher voltage to the vector control signal. The Pulse Width Modulationmotor with lower total harmonic distortion. The (PWM) techniques have given advantages likemain aim of any modulation technique is to obtain harmonic reduction and lower switching losses etcvariable output having a maximum fundamental for the inverter. One of the most efficient techniquescomponent with minimum harmonics. Space Vector is Space Vector Pulse Width Modulation (SVPWM)PWM (SVPWM) method is an advanced; technique [4]computation intensive PWM method and possibly The switching frequency of the inverter isthe best techniques for variable frequency drive limited in Space Vector Pulse Width Modulationapplication.In SVPWM methods, the voltage (SVPWM). By using the intelligent techniques likereference is provided using a revolving reference Neural Networks [5], the switching frequency isvector. In this case magnitude and frequency of the easily extended to higher levels.In existing literaturefundamental component in the line side are on Neural Network based SVPWM, the Spacecontrolled by the magnitude and frequency, Vector PWM inverter is studied along with therespectively, of the reference voltage vector. Space modeling of SVPWM based inverter, vector controlvector modulation utilizes dc bus voltage more of induction motor, indirect vector control ofefficiently and generates less harmonic distortion in induction motor and Kohonen’s competitive layer ina three phase voltage source inverter. Neural Network is studied along with its modeling The application of artificial network networks [6].(ANNs) is recently growing in the areas of powerelectronics and drives. Remarkable features of 1.3 PROBLEM FORMULATIONneural networks are both fast processing speed and In aspect of drive systems, the conventionalfault tolerance to some overlook connections in the control methods need modeling and limited fornetwork system. Because of the parallel processing linear systems. Most of controllers those aremechanism of neural networks, it is expected that designed for drives are linearized near theirthe neural networks can execute the non-linear data operating points. Therefore there is scope tomapping in short time and of the distributed network improve these using artificial intelligencestructure, the performance of the neural network techniques. The necessity of intelligent controllers ismay not be influenced by some miss connections in to achieve the satisfactory dynamic performance ofthe network itself. A control system with multi input conventional methods.neural network may not be affected by partial faultin the system, because some other correct input data 1.4 OBJECTIVES OF THE THESISmay compensate the influence of wrong input data. This project considered two objectives: (i) To understand and develop1.2 LITERATURE SURVEY conventional SVPWM technique and In indirect vector control scheme applied for the the same iscontrol the induction machine to acquire desired applied to induction motor drive;performance. The dynamic modeling of induction andmachine [1] in synchronously rotating frame is (ii) To develop ANN model to realize theapplied to indirect vector control scheme. The functionality of SVPWM. Thefundamentals of vector control implementation are performances of these methods areexplained. verified on induction motor drive in The machine model is represented in a MATLAB/Simulink environment.synchronously rotating reference frame. The II VECTOR CONTROL OF INDUCTIONmachine terminal parameters are transformed from MOTOR3- to 2- according to the principle of vector 2.1 INTRODUCTIONcontrol [2]. There are essentially two general An Induction motor (IM) is a type ofmethods of vector control. One, called the direct or asynchronous AC motor where power is supplied tofeedback method, was invented by Blaschke, and the rotating device by means of electromagneticthe other, known as the indirect or feed forward induction. Other commonly used name is squirrel cage motor due to the fact that the rotor bars with 105 | P a g e
- 3. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116short circuit rings resemble a squirrel cage. An When a three-phase supply is connected to the statorelectric motor converts electrical power to windings, a rotating magnetic field is produced. Asmechanical power in its rotor. the magnetic flux cuts a bar on the rotor, an E.M.F. There are several ways to supply power to is induced in it and since it is joined, via the endthe rotor. In a DC motor this power is supplied to conducting rings, to another bar one pole pitchthe armature directly from a DC source, while in an away, current flows in the bars.induction motor this power is induced in the rotating The magnetic field associated with thisdevice. An induction motor is sometimes called a current flowing in the bars interacts with the rotatingrotating transformer because the stator (stationary magnetic field and a force is produced, tending topart) is essentially the primary side of the turn the rotor in the same direction as the rotatingtransformer and the rotor (rotating part) is the magnetic field. Similar forces are applied to all thesecondary side. Induction motors are widely used, conductors on the rotor, so that a torque is producedespecially poly phase induction motors, which are causing the rotor to rotate.frequently used in industrial drives. The Induction motor is a three phase ACmotor and is the most widely used machine. Itscharacteristic features are- Simple and rugged construction Low cost and minimum maintenance High reliability and sufficiently high efficiency Needs no extra starting motor and need not be synchronized Fig 2.1 Production of Magnetic Field An Induction motor has basically two parts They are widely used for different – Stator and Rotor applications ranging from small induction motors in The Stator is made up of a number of stampings washing machines, household fans etc to vary largewith slots to carry three phase windings. It is wound induction motors which are capable of tens offor a definite number of poles. The windings are thousands of kW in output, for pipelinegeometrically spaced 120 degrees apart. Two types compressors, wind-tunnel drives and overlandof rotors are used in Induction motors - Squirrel- conveyor systems. Through electromagneticcage rotor and Wound rotor. induction, the rotating magnetic field induces a The following sections describe the dynamicmodeling of induction motor and different types of current in the conductors in the rotor, which in turn sets up a counterbalancing magnetic field thatcontrol strategies of induction motor drives. The causes the rotor to turn in the direction the field isvector control is one of the most important control rotating.The rotor must always rotate slower thanstrategies. Direct vector control and indirect vector the rotating magnetic field produced by the polycontrol techniques are given in detail. phase electrical supply; otherwise, no counterbalancing field will be produced in the rotor.2.2 WORKING PRINCIPLE OF INDUCTION MOTOR 2.3 MODELLING OF INDUCTION As a general rule, conversion of electrical MOTORpower into mechanical power takes place in the The dynamic model of the induction motorrotating parts of an electrical motor. In dc motor, the is derived by using a two phase motor in direct andelectrical power is conducted directly in armature quadrature axes. This approach is desirable becausethe rotating part of the motor through brush or of the conceptual simplicity obtained with two setscommutates and hence dc motor called as of windings, one on stator and the other on the rotor.conduction motor but in case of induction motor the The equivalence between the three phase and twomotor does not receive the electrical power by phase machine models is derived from simpleconduction but by induction in exactly same way as observation, and this approach is suitable forthe secondary of a 2-winding transformer receives extending it to model an n-phase machine by meansits power from the primary. That is why such motor of a two phase machine. The concept of powerknown as induction motor.In fact, an induction invariance is introduced: the power must be equal inmotor can be treated as a rotating transformer i.e. the three phase machine and its equivalent twoone in which primary winding is stationary but the phase model. The modeling of induction motorsecondary is free to rotate. Of all the a.c. motors, the considers the equivalent circuit as shown in Fig 2.2.poly phase induction motor is the one which isextensively used for various kinds of industrialdrives. 106 | P a g e
- 4. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-1162.3.1 EQUIVALENT CIRCUIT The induction motor D-Q axes equivalentcircuit plays key role in design of Dynamical model.These circuits are shown in Fig.2.2 2.5 2.6 2.7 2.8 2.9 2.10 2.11Fig 2.2 Dynamic or d-q equivalent circuit of an 2.12induction machine The one of the popular induction motor Wheremodels derived from Kron’s Model. According tohis model, the modeling equations in flux linkage d: direct axis,form are as follows: 2.1 q: quadrature axis, s: stator variable, 2.2 r: rotor variable, Fij: is the flux linkage (i=q or d and j=s or r) , : q and d-axis stator voltages, 2.3 , : q and d-axis rotor voltages, , : q and d axis magnetizing flux linkages, : Rotor resistance,2.4 : Stator resistance, : Stator leakage reactance , 107 | P a g e
- 5. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 : Rotor leakage reactance , Base frequency = 100 Stator inductance = 5.25/(2*pi* ) , Rotor inductance = 4.57/(2*pi* ) Magnetizing inductance =139/(2*pi* ) , : q and d-axis stator currents, Number of poles P=4 , : q and d-axis rotor currents, Moment of inertia j = 0.025 = +p: number of poles,J: moment of inertia, =Te: electrical output torque, Base speed = 2*pi*TL: load torque, Stator impedance = *ωe :stator angular electrical frequency, Rotor impedance = *ωb: motor angular electrical base frequency, and Magnetizing impedance = *ωr: rotor angular electrical speed. = 300VFor a squirrel cage induction machine and inequations (3) and (4) are set to zero. An induction 2.5 MODELING EQUATIONS IN STATEmachine model can be represented with five SPACEdifferential equations as seen above.To solve these equations, they have to be rearrangedin the state-space form, X = Ax + B 2.14Where X= [Fqs Fds Fqr Fdr ωr]T is the state vector. = *ωb, 2.13Where “ is the flux linkage and is the flux. 2.15In this case, state space form can be achieved byinserting (2.5)and(2.6) in (2.1-2.4) and collectingthe similar terms together so that each statederivative is a function of only other state variablesand model inputs.2.4 MOTOR PARAMETERS 2.16Machine Rating 2.2KWSwitching Frequency 10KHZRotor resistance = 1.34 2.17Stator resistance = 1.77 108 | P a g e
- 6. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 control. By contrast, ac induction motor drives require a coordinated control of stator current 2.18 magnitudes, frequencies, and their phases, making it a complex control. The requirement of phase,2.5 CONTROL STRATEGIES OF frequency, and magnitude control of the currents INDUCTION MACHINE and hence of the flux phasor is made possible by The control and estimation of ac drives in inverter control. The control is achieved in fieldgeneral are considerably more complex than those coordinates hence it is called “field orientedof dc drives, and this complexity increases control” or “vector control”.substantially if high performances are demanded.The main reasons for this complexity are the need ofvariable-frequency, harmonically optimum 2.7.1 PRINCIPLE OF VECTOR CONTROLconverter power supplies, the complex dynamics of The fundamental principle vector control ofac machines, machine parameter variations, and the the vector controlled induction machine can bedifficulties of processing feedback signals in the explained with the d-q model which is as shown inpresence of harmonics. Fig. 2.2. To limit the above problems, different Control Machinetypes control strategies are introduced in recent s syears. Induction motor has 3 control strategies. Volts/Hz control of Induction motor i ds de e i ds i a i a i ds i ds -q ds-qs a-b-c ds-qs Machine Vector control or Field Oriented i Control(FOC) to a-b-c s to a-b-c i b b to ds-qs s to de-qe de-qe model Direct Torque Control(DTC) i qs i qs i c i c i qs i qs In these, vector control is used as a controlstrategy because in this the induction motor can becontrolled like a separately excited motor, brought a i qs i renaissance in the high performance control of ac dsdrives. Because of dc machine like performance, cos e sin e cos e sin e evector control is known as decoupling, orthogonal, Inverse transformation Transformationor Trans vector control. Machine terminal2.7 VECTOR CONTROL CONCEPT Fig 2.3 Vector Control Implementation Principle The various control strategies for the with Machine de–qe Modelcontrol of the inverter fed induction motor haveprovided good steady-state but poor dynamic In diagram the machine models representedresponse. From the traces of the dynamic responses, in a synchronously rotating reference frame. Thethe cause of such poor dynamic response is found to inverter is omitted, assuming that it has unity currentbe that the air gap flux linkages deviate from their gain i.e. it generates current ia ,ib ,ic .by theset values. The deviation is not only in magnitude corresponding command currents i*a ,i*b ,i*c from thebut also in phase. The variations in the flux linkages controller. The machine terminal phase currents i a ,ib s shave to be controlled by the magnitude and ,ic .are converted into i qs, i ds components by 3-phasefrequency of the stator and rotor phase currents and to 2-phase transformation. These are then rotatingtheir instantaneous phases. frame by the unit vector components and before applying them to the de - qe machineThe oscillations in the air gap flux linkages result in model.oscillations in electromagnetic torque and, if left To explain the principle of vector control,unchecked, reflect as speed oscillations. This is an assumption is made that the position of the rotorundesirable in many high-performance applications. flux linkage phasor, , is known. is at from a Separately excited dc drives are simpler in stationary reference, is referred to as field angle,control because they independently control flux, and the three stator currents can be transformed intowhich, when maintained constant contributes to an q and d axes currents in the synchronous referenceindependent control of torque. This is made possible frames by using the transformationwith separate control of field and armature currentswhich, in turn, control the field flux and torqueindependently. The dc motor control requires only thecontrol of the field or armature current magnitudes,providing simplicity not possible with ac machine 109 | P a g e
- 7. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 The perpendicular component is the torque = producing component. 2.22 2.232.19 and have only dc components in steady state,The stator current phasor, , is derived as because the relative speed with respect to that of the rotor field is zero; the rotor flux linkages phasor has a speed equal to the sum of the rotor and slip speeds, = which is equal to the synchronous speed.2.20 The field angle can be written as = + 2.24 Where the rotor is position and is the slip angle. In terms of the speeds and time, the field angle is written as = dt= dt 2.25 The controller makes two stages of inverse transformation both at the control currents to corresponds to the machine model and respectively. The unit vector assures correct alignment of current with the flux vector (ψr) and perpendicular to it. The transformation and inverse transformation including the inverter ideally do not incorporate any dynamics and therefore, the response to and is instantaneous. The orientation of with rotor flux (Ψr) air gap flux (Ψm) or stator flux (Ψs) is possible in the vector control. Rotor flux orientation gives natural decoupling control where as air gap or stator flux orientation gives coupling effect, which has to be Fig 2.4 Phasor diagram of the Vector Controller compensated by a decoupling compensator current.And the stator phasor angle is 2.8 TYPES OF FIELD ORIENTATED CONTROL (FOC) METHODS = In the induction motor torque control is performed by the quadrature component of the stator2.21 current space phasor iqs, whereas in the DC motor, it is performed by armature current ia. The field Where and are the q and d axes control is performed by the direct component of thecurrents in the synchronous reference frames that stator current space phasor ids, whereas in the DCare obtained by projecting the stator current phasor motor, it is performed by the field current i f , in theon the q and d axes, respectively. separately excited winding. One should note that in The current phasor produces the rotor the control portion of the drive for the inductionflux and the torque . The component of current motor, all the variables are in the rotor fluxproducing the rotor flux phasor has to be in phase reference frame, which is rotating synchronouslywith . Therefore, resolving the stator current with the rotor flux linkage space phasor. Therefore,phasor along reveals that the component is the all the variables are dc quantities as explained earlier. Thus, a vector controlled induction motorfield producing component, as shown in Fig 2.4. drive system is similar to a DC motor drive system. 110 | P a g e
- 8. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116Further, the vector controlled induction motor drive III NSVPWM BASED INDIRECT VECTORsystem will give dynamic performances similar to CONTROL OF INDUCTION MOTOR DRIVEthat of DC motor drive system. For vector control operation of the induction 3.1 INTRODUCTIONmotor, the arbitrary reference frame must be aligned The conventional Space Vector Pulsealong the rotor flux linkage space phasor at every Width Modulation (SVPWM) for indirect vectorinstant. It is therefore essential that the position of controlled induction motor drive has somethe rotor flux linkage space phasor “”, be limitations to obtain the desired performance of theaccurately known at every instant. This knowledge system. The neural network methodology isof rotor flux linkage space phasor position can be introduced to give better performance.acquired either by measuring the flux directly or by A neural network is a powerful data-modelingestimating the flux from terminal variables i.e. by tool that is able to capture and represent complexindirect means. This leads to two possible control input/output relationships. The motivation for thetechniques of induction motor. development of neural network technology stemmed Direct field oriented control from the desire to develop an artificial system that Indirect field oriented control could perform "intelligent" tasks similar to those performed by the human brain. Neural networks2.8.1 DIRECT FIELD ORIENTED resemble the human brain in the following twoCONTROL (DFOC) ways: In this mode of control the flux A neural network acquires knowledgemeasurement can be made using either the hall through learning.sensors or the stator search (sense) coils. If the stator A neural networks knowledge is storedcoils are used, then the voltage sensed from the coils within inter-neuron connection strengthswill have to be integrated to obtain the air gap flux known as synaptic weights.linkages. The measured air flux linkage components Artificial Neural Networks are being counted as theare used to calculate the required (rotor, stator or air wave of the future in computing. ANN has severalgap) flux linkage space phasor magnitude and types of methods are introduced in recentposition “”. The value of thus computed is used researches. Here in this project adapted Kohonen’sto align the arbitrary axis along the flux linkage Neural Network. The following section explains thespace phasor to achieve decoupled control of the description of neural based Space Vector Pulsetorque and flux producing components of the stator Width Modulation using Kohonen’s Competitivecurrent and space phasor. net. The flux sensing devices are placed in theair gap of the machine, which will determine the air 3.2 KOHONEN’S NEURAL NETWORKgap flux space phasor. Any other flux space phasor The Kohonen neural network differscan be calculated as it has an algebraic relationship considerably from the feed forward backwith the air gap flux space phasor. The air gap flux propagation neural network. The Kohonen neuralsensed by either hall-effect devices or stator search network differs both in how it is trained and how itcoils suffer from the disadvantage that a specially recalls a pattern. The Kohonen neural network doesconstructed induction motor is required. Further, not use any sort of activation function. Further, thehall sensors are very sensitive to temperature and Kohonen neural network does not use any sort of amechanical vibrations and the flux signal is distorted bias weight. Output from the Kohonen neuralby large slot harmonics that cannot be filtered network does not consist of the output of severaleffectively because their frequency varies with neurons. When a pattern is presented to a Kohonenmotor speed. network one of the output neurons is selected as a In the case of stator Search (sense) coils, "winner". This "winning" neuron is the output fromthey are placed in the wedges close to the stator the Kohonen network. Often these "winning"slots to sense the rate of change of air flux. The neurons representinduced voltage in the search coil is proportional to groups in the data that is presented to thethe rate of change of flux. This induced voltage has Kohonen network. The most significant differenceto be integrated to obtain the air gap flux. At low between the Kohonen neural network and the feedspeeds below about 1HZ, the induced voltage will forward back propagation neural network is that thebe significantly low which would give rise to in Kohonen network trained in an unsupervised mode.accurate flux sensing due to presence of comparable This means that the Kohonen network is presentedamplitudes of noise and disturbances in a practical with data, but the correct output that corresponds tosystem. As an alternative, indirect flux estimation that data is not specified. Using the Kohonentechniques are preferred as explained in the next network this data can be classified into groups.sub-section. It is also important to understand the limitations of the Kohonen neural network. Kohonen neural networks are used because they are 111 | P a g e
- 9. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116a relatively simple network to construct that can be different patterns, their ability to recognize thattrained very rapidly. particular pattern will be increased. Kohonen’s neural network is trained by3.2.1 STRUCTURE OF THE KOHONEN repeating epochs until one of two things happens. IfNEURAL NETWORK they calculated error is below acceptable level The Kohonen neural network works business at block will complete the training process.differently than the feed forward neural network. On the other hand, if the error rate has all onlyThe Kohonen neural network contains only an input changed by a very marginal amount this individualand output layer of neurons. There is no hidden cycle will be aborted with tile any additional epochslayer in a Kohonen neural network. The input to a taking place. If it is determined that the cycle is toKohonen neural network is given to the neural be aborted the weights will be initialized randomnetwork using the input neurons. These input values and a new training cycle began. This trainingneurons are each given the floating point numbers cycle will continue the previous training cycle andthat make up the input pattern to the network. A that it will analyze epochs on to solve get the two isKohonen neural network requires that these inputs either abandoned or produces a set of weights thatbe normalized to the range between -1 and 1. produces an acceptable error level. The mostPresenting an input pattern to the network will cause important part in the networks training cycles is thea reaction from the output neurons. individual epochs. The output of a Kohonen neural network is The learning rate is a constant that will bevery different from the output of a feed forward used by the learning algorithm. The learning rateneural network. In a Kohonen neural network only must be a positive number less than 1. Typically theone of the output neurons actually produces a value. learning rate is a number such as 0.4 or 0.5.Additionally, this single value is either true or false. Generally setting the learning rate to a larger valueWhen the pattern is presented to the Kohonen neural will cause the training to progress faster. Thoughnetwork, one single output neuron is chosen as the setting the learning rate to too large a number couldoutput neuron. Therefore, the output from the cause the network to never converge. This isKohonen neural network is usually the index of the because the oscillations of the weight vectors will beneuron (i.e. Neuron #5) that fired. The structure of a too great for the classification patterns to evertypical Kohonen neural network is shown in Fig5.1. emerge. 3.3 NSVPWM FOR INDIRECT VECTOR CONTROL (IVCIM) DRIVE In a voltage source inverter the space vector modulation technique requires the use of the adjacent switching vectors to the reference voltage vector and the pulse times of these vectors. For this purpose, the sector where the reference voltage vector is positioned must be determined. This sector number is then used to calculate the position θ of the reference voltage vector with respect to the closest clockwise switching vector (Fig 3.2). The pulse time can then be determined by using the trigonometric function Sin (θ) and Sin (60 −θ) as in (5.5) and (5.6). Fig 3.1 Kohonen Neural Network3.2.2 KOHONEN NETWORK LEARNING There several steps involved in this trainingprocess. Overall the process for training a Kohonenneural network involves stepping through severalepochs until the error of the Kohonen neuralnetwork is below acceptable level. The trainingprocess for the Kohonen neural network iscompetitive. For each training set one neuron will"win". This winning neuron will have its weightadjusted so that it will react even more strongly tothe input the next time. As different neurons win for Fig 3.2 Two winner neurons of the competitive layer closest to 112 | P a g e
- 10. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116However, it is also possible to determine the two Wherenon-zero switching vectors which are adjacent to thereference voltage vector by computing the cosine of Wangles between the reference voltage vector and sixswitching vector and then by finding those two =angles whose cosine values are the largest. Mathematically this can be obtained by 1 1 1 1 1 1 2computing the real parts of the products of thereference voltage space vector and the six non-zero 2 2 2switching vectors and selecting the two largest 1 1 1 1 values. These are proportional to Cos (θ) and Cos(60−θ) respectively, Where θ and (60−θ) are the 1 1angles between the reference voltage vector and the 2 2 2 2 ; V =adjacent switching vectors. It is also possible to use an ANN based on 1 1 1 1 1 1 refKohonen’s competitive layers. It has two winnerneurons and the outputs of the winner neurons are 2 2 2 2set to their net inputs. If normalized values of the V input vectors are used, then the six outputs (six net Aref values ( , ,….., ) will be proportional to the V Bref cosine of the angle between the reference voltage vector and one of the six switching vectors. V Cref The two largest net values are thenselected. These are and , proportional to Cos(θ) and Cos (60 −θ). Since the space vectormodulation is a deterministic problem and all Assuming is applied to the competitive layerclasses are known in advance, there is no need to and and are the neurons who win thetrain the competitive layer. competition. Then from (5.1) we have Net = .W= 5.1 = Since the input vector and the weightvector [6] are normalized, the instars net input gives 5.3the cosine of the angles between the input vectorand the weight vectors that represent the classes. AlsoThe largest in star net input wins the competitionand the input vector is then classified in that class. =The winner of the competition is the closest vectorto the reference vector. The six net values can be written in a 5.4matrix form for all neurons as: 1 1 1 Substituting (5.4) in (5.3) we get 2 2 1 1 = n1 2 1 V 5.5 2 n2 1 1 Aref V Bref 1 Equation (3.5) is the on duration of then 2 2 consecutive adjacent switching state vector and 3 1 V Cref = 1 n4 1 , which is same as (4.12) and (4.13).Therefore 2 2 we haven5 1 1n6 1 2 2 = 1 1 1 5.6 2 2 The implementation of this method is5.2 depicted in fig5.3 first for k=1………6 are calculated. 113 | P a g e
- 11. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 Fig 3.5 Block diagram representation of NSVPWM based (IVCIM) drive Figure captions appear below the figure,Fig 3.3 Kohonen’s Competitive Layer based are flush left, and are in lower case letters. WhenImplementation of the Space Vector Modulation referring to a figure in the body of the text, theTechnique for VSI abbreviation "Fig." is used. Figures should be numbered in the order they appear in the text.TableTwo largest , and their corresponding indexes captions appear centered above the table in upper(i.e. i and i+1) are selected by Kohonen’s and lower case letters. When referring to a table incompetitive network. The on duration ( and ) of the text, no abbreviation is used and "Table" isthe two adjacent space vectors are computed. The capitalized.Space Vector and are selected according to The speed response is reached to thethe value of i and i+1. When adjacent vectors and reference speed gradually as compared to theon times are determined the procedure for defining conventional SVPWM. From the Figthe sequence for implementing the chosen 3.9 it is observed that the speed response of thecombination is identical to that used in conventional NSVPWM is improved as compared to the SVPWMspace vector modulation as depicted in Fig. 3.4. at load condition. Fig.3.4. Pulse patterns generated byspace vector modulation in sector13.4 BLOCK DIAGRAMREPRESENTATION OF NSVPWM FOR(IVCIM) DRIVE Indirect vector controlled drive hasenhance the performance using neural network. The Fig 3.10 Speed response of IVCIM usingproposed scheme is implementing by replacing the NSVPWMconventional SVPWM with Neural based SVPWM.According to the Neural Network theory the At initial conditions of the load i.e. at t=0 theKohonen Competitive layer is adapted. The Fig 3.5 magnitudes of the currents are shown, when loadshows the block diagram representation of applied the variations in the currents are shown inNSVPWM based indirect vector control for Fig 3.10. The response of the SVPWM has beeninduction motor drive. improved as compared to the NSVPWM. 114 | P a g e
- 12. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 out in this project is aimed and focused to implement Neural based IVCIM drive. The simulation has been carried out for different operating conditions. In developing of drive, the switching pulses are generated by using SVPWM and NSVPWM. Use of ANN based technique avoids the direct computation of non-linear function as in conventional space vector modulation implementation. The ANN based SVPWM can give higher switching frequency which is not possible by conventional SVPWM. The performance of the NSVPWM based IVCIM drive is better than SVPWM based IVCIM drive. .References [1]. Burak Ozpineci, Leon M. Tolbert, “Simulink Implementation of Induction Machine Model- A Modular Approach”. [2]. Bimal K. Bose, “Modern Power Electronics and AC Drives”- PearsonFig 3.11 Stator Currents of IVCIM using Education, Inc, 2002.NSVPWM [3]. R Krishnan- Electric Motor Drives- The step command of the load torque isapplied to the induction motor; the output torque of Modeling, Analysis, and Control- Virginiathe motor is improved as compared to the SVPWM. Tech, Blacksburge. VA @ 2001 Printice Hall, Inc.The improved torque response is shown in Fig 5.11. [4]. Jin-Woo Jung, “Space Vector PWM Inverter”, Feb, 2005. [5]. K. Vinoth Kumar, Prawin Angel Michael, Joseph P.John and Dr. S. Suresh Kumar, “Simulation and Comparison of SPWM and SVPWM Control for Three Phase Inverter” [6]. Rajesh Kumar, R.A Gupta, Rajesh S. Surjuse, “A Vector Controlled Induction Motor drive with Neural Network based Space Vector Pulse Width Modulator”. [7]. alexandru onea, vasile horga, marcel ratoi, “ Indirect Vector Control of Induction Motor”, Proceedings of the 6th international conference on simulation, modelling and optimization, Libson, Portugal, September 22-24, 2006. [8]. Joao O.P.Pinto, IEEE, Luiz Eduardo Borges da Silva, IEEE, and Marian P. Kazmierkowski, IEEE, “A Neural NetworkFig 3.12 Electromagnetic Torque response of Based Space Vector PWM Controller forIVCIM using NSVPWM Voltage Fed Inverter Induction Motor Drive”.I. CONCLUSIONIn the present work, indirect vector control (IVC)technique is employed to a 3-Ф induction motor andcomplete model has been designed inMATLAB/SIMULINK package. The work carried 115 | P a g e
- 13. Neelima.A,Hari krishna.Ch / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue6, November- December 2012, pp.104-116 Chalasani Hari Krishna was born in 1982. He graduated from Jawaharlal Nehre Technological University, Hyderabad in the year 2003. He received M.E degree from Satyabama University, Chennai in the year 2005. Hepresently Associate Professor in the Department ofElectrical and Electronics Engineering at MotherTeresa Institute of Science and Technology, India.His research area includes DTC and Drives. Hepresented 11 research papers in various nationaland international conferences and journals.His research areas include PWM techniques, DC toAC converters and control of electrical drives Aitham Neelima was born in 1988. She graduated from sree kavitha engineering college,karepally in the year 2010.She pursuing mtech degree from Mother Teresa Institute ofScience and Technology,india in the year 2012. Currently, she is assistant Professor of ElectricalEngineering at SVIT. Her areas of interests includethe application of computer technology to powersystem monitoring, protection, control, motorcontrol, and power electronics. 116 | P a g e

Full NameComment goes here.