svpwm using ann

8/16/2019 Svpwm Using Ann

1/6

ISSN:2277-3754International Journal of Engineering and Innovative Technology (IJEIT)

Volume 1, Issue 2, February 2012

157

Abstract — Space vector pu lse width modulati on (SVPWM ) i san optimum pul se width m odulation techni que for an inverterused in a variabl e frequency dri ve application s. I t iscomputational ly ri gorous and hence li mi ts the inverter switchi ngfr equency. A neur al n etwork has the advantage of very f astim plementation of an SVPWM algori thm th at can incr ease theinverter switching frequency. This paper proposes a neuralNetwork based SVPWM techni que for a thr ee-phase voltagesource in verter in under modul ation r egion. T he scheme has

been simu lated and impl emented on a V/ H z control led 5-hp,50-Hz, and 230V inducti on motor dri ve. The sim ul ation r esul tsare given to validate the perf ormances of th e dri ve with ar tif icialneural network based SVPWM .

I ndex Terms — Space vector pulse width modulation, neuralnetwork, voltage source inverter.

I. INTRODUCTIONVariable frequency ac drives are increasingly replacing dc

drives in a number of industrial applications due toadvantages in size, reliability and efficiency. One of the maincomponents of an ac drive is power electronic converter in theform of voltage source inverter that takes dc voltage input(may be from a rectifier) and produces a sinusoidal acwaveform. This in turn is fed to the ac electric motor. Thefundamental frequency of this waveform is adjusted to

produce the desired speed. In modern electric drives themodulation of switching is carried out to achieve the requiredac wave shape. This is met by incorporating the duration of‗ON‘ & ‗OFF‘ methods [1, 2].

The techniques to get the duration of ON interval for a particular switch depend upon the control logic or PWMtechnique to be adopted. In a PWM scheme the output voltage

and frequency can be controlled with the help of theswitching inside the inverter. The switch may be MOSFET,IGBT etc with anti parallel connected diodes. Different PWMschemes such as sinusoidal PWM compares a high frequencytriangular carrier with three sinusoidal reference signals,known as the modulating signals, to generate the gatingsignals for the inverter switches but having a disadvantagethat it contains third harmonic in output[3]. To thecancellation of the third-harmonic components and betterutilization of the dc supply, the third harmonic injectionPWM scheme is preferred in three-phase applications. Spacevector modulation technique has advantage of an optimal

output and also reduces harmonic content of the outputvoltage/current [4]. Space vector PWM (SVPWM) has theadvantages of lower harmonics and a higher modulationindex in addition to the features of complete digital

implementation by a single chip microprocessor, because ofits flexibility of manipulation, SVPWM has increasingapplication in power converters and motor control.

The application of artificial neural network (ANN) isrecently growing in power electronic systems. A feed forwardANN implements nonlinear input – output mapping. Thecomputational delay of this mapping becomes negligible, if

parallel architecture of the network is implemented by anapplication-specific integrated circuit (ASIC) chip [5]. A feed

forward carrier-based PWM technique, Such as SVM, canalso be looked upon as a nonlinear mapping phenomenonwhere the command phase voltages are sampled at the inputand the corresponding pulse width patterns are established atthe output. Therefore, it appears logical that a back

propagation-type ANN that has high computationalcapability can implement an SVM algorithm. The ANN can

be conveniently trained offline with the data generated bycalculation of the SVM algorithm. ANN has inherentlearning capability that can give improved precision byinterpolation unlike the standard lookup table method [5].

The ANN-based SVPWM of a voltage-fed inverter canalso use competitive neural network architecture to identifythe inverter switching states and the corresponding voltagemagnitudes for the impressed command voltage. This paperdescribes feed- forward ANN-based SVPWM that coverslinear modulation region. In the beginning, the SVPWMtheory has been briefly reviewed with mathematical analysisfor the linear modulation range. Then equations foralgorithms have been developed in detail. A back

propagation type feed forward ANN is trained offline with thedata generated by this simple algorithm [5].

II.

SPACE VECTOR PWM –

A REVIEWThis section is devoted to the development of Space vector

PWM for a two-level voltage source inverter in linear regionof operation [6]. As seen from Fig 1, there are six switchingdevices and only three of them are independent as theoperation of two power switches of the same leg arecomplimentary. The combination of these three switchingstates gives out eight possible space voltage vectors. Thespace vectors forms a hexagon with 6 distinct sectors, eachspanning 60 degrees in space. At any instant of time, theinverter can produce only one space vector. In space vectorPWM a set of three vectors (two active and a zero) can beselected to synthesize the desired voltage in each switching

period. All of the eight modes are shown in Table.1.Out of eight topologies six (states 1-6) produce a non-zero

output voltage and are known as active voltage vectors and

Space vector PWM Technique for 3phase voltage

source inverter using Artificial Neural NetworkBandana, K.Banu priya, JBV Subrahmanyam, Ch.Sikanth, M.Ayyub


2/6



158

the remaining two topologies (states 0 and 7) produce zerooutput voltage (when the motor is shorted through the upperor lower transistors) and are known as zero voltage vectors,various possible switching states are shown in Fig 2.

Space vector is defined as [6],

* 223 a b c s

v v av a v (1)

Where exp 2 / 3a j .The space vector is a simultaneous representation of all the

three-phase quantities [14]. It is a complex variable and isfunction of time in contrast to the phasors. Phase-to-neutralvoltages of a star-connected load are most easily found bydefining a voltage difference between the star point n of theload and the negative rail of the dc bus N . The followingcorrelation then holds true:

A a nN

B b nN

C c nN

v v v

v v v

v v v

(2)

Since the phase voltages in a start connected load sum tozero, summation of equation (2) yields

1/ 3nN A B C v v v v (3)Substitution of (3) into (2) yields phase-to-neutral voltages

of the load in the following form:

2 / 3 1/ 3

2 / 3 1/ 3

2 / 3 1/ 3

a A B C

b B A C

c C B A

v v v v

v v v v

v v v v

(4)

Phase voltages are summarized and their correspondingspace vectors are listed in Table 1. The eight vectorsincluding the zero voltage vectors can be expressedgeometrically as shown in fig.3. Each of the space vectors, inthe diagram represent the six voltage steps developed by theinverter with the zero voltages V 0 (0 0 0) and V 7 (1 1 1)located at the origin. Space Vector PWM require to averagingof the adjacent vectors in each sector. Two adjacent vectorsand zero vectors are used to synthesis the input referencedetermined from Fig.4 for sector I. Using the appropriatePWM signals a vector is produced that transitions smoothly

between sectors and thus provide sinusoidal line to linevoltages to the motor.

In order to generate the PWM signals that produces therotating vector. The PWM time intervals for each sector isdetermined from Fig. 4 for sector I as

Along real axis:

cos2)60cos2(1

1 S mV sT

T V

sT

T V (5)

Along imaginary axis:

sin)60sin(0 22 S s

mV T T

V (6)

Solving equations (5) and (6)

60sin

)60sin(

11 V

mV T T S S

60sin32

)60sin(21

1

DC

DC

s

V

V m

T T

60sin23

1 S mT T (7)

60sin

sin

22 V

mV T T S S

)sin(23

2 S mT T (8)

)( 2170 T T T T T s

(9)Generalizing the time expressions gives

3sin

23

1k

mTsT (10)

3)1(

sin23

2k

mTsT ; k=1, 2, 3... (11)

The time of application of active and zero space vectors forall six sectors are given in Table 2 [13]. The periods

1, 2 0T T and T depends only on the reference vector

amplitude Vs and the angle ‗ ‘. This shows that the period

1, 2 0T T and T are the same in all sectors for the same V S and‗ ‘ position. In the under modulation region, the vector V S always remains within the hexagon. The mode ends in theupper limit when V S describes the inscribed circle of thehexagon. Modulation Index MI (m) is given by

SixstepV Vs

m1

*

Where, V S = input reference vector magnitude

V 1Six step = fundamental peak value DC V 2 of the six step

output.The maximum value of input reference is the radius of

largest circle inscribed in the hexagon given by

DC DC s V V V 577.0)30cos(32 0*

Therefore, maximum modulation index

907.02577.0

1

*

DC

DC

SW V

V V Vs

m

(12)

This means that 90.7% of the fundamental of the six stepwave is available in the linear region, compared to 78.55% inthe sinusoidal PWM [7].


3/6



159

III. NEURAL NETWORK BASED SPACE VECTOR PWM

The first step that involves in the implementation of neuralnetwork based space vector PWM of a voltage source inverter

is generation of training data. The training data is generated by using eqs. (1), (2), (3).the an gle subnet is trained with an

angle interval of 2 in the range of

3600 .The phase-A turn ON time can be expressed as

Equation (4) can be written in the general form

ON AT **4 A s g V f T Where *V f is the voltage amplitude scale factor and

Fig.(3) shows schematic of the ANN based SVPWMinverter. The input signal to the neural network is angle .Fig.(4) shows the neural network model. The model uses amultilayer function in the first and second layer respectively.The Neural network uses one neuron at the input 25 neuronsin the hidden layer and three output neurons. The digitalwords corresponding to turn on time are generated by

multiplying the output of neural network with S T V *

and thenadding 4 sT as shown in figure. The PWM signals are then

generated by comparing turn-On time with a triangularreference having time period of sT and amplitude 2 sT .the

PWM signals are then applied to the inverter. The back propagation algorithm in the Matlab toolbox is used for thetraining. The angle subnet takes 101650 epochs for trainingwith an error 0.53%.The value of size and the correspondingtraining time are, thus reasonably small.

IV. DRIVE SYSTEM PARAMETERS

DC-link voltage 300 VSampling time (T s) 200 s Induction motor 5 hp 230 V four pole, squirrel cage

Frequency range: 0-50 Hz

Stator resistance (R s): 0.5814 Rotor resistance (R r ): 0.4165 Stator leakage inductance (L ls): 3.479 mHRotor leakage inductance (L lr ): 4.15 mHMutual inductance (L

m): 78.25mH

Rotor inertia (J): 0.1 kg.m 2

V. SIMULINK MODEL AND RESULTSThe Simulink model used to simulate model results on

computer is shown in fig. (7).Fig. (8) Shows thecharacteristics of speed of induction motor. In the startingspeed of motor increases very sharply and reaches to 1700rpm after 0.1 seconds. Some oscillations are followed tillthe motor catches the reference speed (i.e. 1500 rpm) in0.3 seconds. At 0.75 seconds when load torque is changedfrom 0 Nm to 10 Nm it attains a value slightly lower than

the reference speed as shown in the figure. This lowervalue of speed is due to increase in torque and the system is

operating in open loop. Fig.(9) shows torque characteristicsfor load torque (Tl) and developed torque (Te). In the startingthe applied torque is zero and developed torque is very highfor few seconds (till 0.125 second). After some fluctuation itattains a value of zero in 0.25 seconds, as Tl=0. When loadtorque is increased to a value 10 Nm at 0.75 seconds. Thedeveloped torque also attains the value of 10 Nm followingsome transients as depicted in figure. The stator and rotorcurrent for dynamic conditions are shown in fig.(10) &(11)respectively. The current shows the expected trend and aresinusoidal in nature. These currents have high initial valuesduring starting due to starting transients, as the induced emftakes time to develop its rated value. After these transientsare over, the current settles at its steady state values after 0.2

sec. The currents (stator and rotor) get higher values whenload torque of 10 Nm is applied at 0.75 sec. These can be seenin the figure.

Fig. 1. Power circuit of a three-phase voltage source inverter

[12]

5,si n3

si n.42

4,3,si n3

sin.42

2,si n3

si n.42

6,1,sin3

sin.44

***0

***0

0

***0

***0

S V K T

t t

S V K T

t t t

S V K T

t t

S V K T t

T

s

a

sb

sb

s

ON A

5,sin3

sin

4,3,sin3

sin

2,sin3

sin

6,1,sin3

sin

**

**

**

**

*

S K

S K

S K

S K

g A


4/6



160

V1 ( 1 0 0 )

V2 ( 1 1 0 )V3 ( 0 1 0 )

V4( 0 1 1 )

V5 ( 0 0 1) V6 ( 1 0 1 )

SECTOR 1

SECTOR 2

SECTOR 3

SECTOR 4

SECTOR 5

SECTOR 6

t1 / ts *V1

t 2 / t s

* V 2

Vs

aV7 ( 1 1 1)

V0 ( 0 0 0)

Table-1 possible modes of operation of a three-phaseVSI[12]

Fig. 2 -The switches position during eight topologies [12]

Fig. 3 - Space Vector representation of Line to Neutral Voltages

[13]

V 2

V 1

mV S 60 0

T 1/T s

T 2/T s

Real

Imag.

Fig. 4- Principle of time calculation for SVPWM in sector I[13]

Table 2 – Application of time [13]

Fig.5 Turn – on pulse width function of phase A, B, C as afunction of Angle in different sectors [13]

S T V * 4S T

NeuralNetwork

comparator

Comparator

Comparator

Triangular reference

+ -

LoadInverter

dcV

AS

BS

cS

S T V *

S T V *

4S T

4S T

ON BT

ON C T

O AT

Fig.6 Schematic of the ANN based SVPWM inverter [5]

Fig.7 Model of Ann based space vector PWM controller forvoltage Fed inverter Induction motor drive


5/6



161

0 0.2 0.4 0.6 0.8 1 1.2 1.4-200

0

200

400

600

800

1000

1200

1400

1600

1800

Time (s)

w m

( r p m )

Fig.8 Speed characteristics

0 0.2 0.4 0.6 0.8 1 1.2 1.4-30

-20

-10

0

10

20

30

40

50

60

70

Time (s)

T e

& T

l ( N m )

Te

T l

Fig.9 Torque characteristics

0 0.2 0.4 0.6 0.8 1 1.2 1.4-100

-80

-60

-40

-20

0

20

40

60

80

100

Time (s)

I s ( A )

a-phaseb-phasec-phase

Fig.10 Stator current waveform

0 0.2 0.4 0.6 0.8 1 1.2 1.4-100

-80

-60

-40

-20

0

20

40

60

80

100

Time (s)

I r ( A )

a-phaseb-phasec-phase

Fig.11 Rotor current waveform

VI. CONCLUSION

A neural-network-based space-vector modulator has beendescribed that operates very well in under modulation region.The digital words corresponding to turn-on time aregenerated by the ANN and then converted to pulse widthsthrough a single timer. The scheme uses a back

propagation-type feed forward network. The training dataand training time are reasonably small. The method canoperate from dc (zero frequency). The scheme has been fullyimplemented and evaluated with a V/Hz-controlled inductionmotor drive, and gives excellent performance. The PWMcontroller is currently being used in a stator-flux-orientedvector-controlled induction motor drive. The ANN-basedSVM can give higher switching frequency, which is not

possible by conventional DSP-based SVM. The switchingfrequency can be easily extended up to 50 kHz if the ANN isimplemented by a dedicated hardware ASIC chip.

VII. FUTURE ENHANCEMENT OF WORK

The ANN-based SVM can give higher switchingfrequency, which is not possible by conventional DSP-basedSVM. The switching frequency can be easily extended up to

50 kHz if the ANN is implemented by a dedicated hardwareASIC chip.

REFERENCES

[1] W. vander Broek, H. C. Skudelny, and G. V. Stanke, ―Analysisand realization of PWM based on voltage space vectors , IEEETrans. Ind. Applicat., vol. 24, no. 1, pp. 142 – 150, 1988.

[2] Grahame Holmes and Thomas A. Lipo , ― Pulse WidthModulation For Power Converters – Principle s and Practices ,IEE Press, Wiley Publications2000.

[3] Blasko, ―A hybrid PWM strategy combining modified spacevector and triangle comparison methods , in IEEE PESC Conf.

Rec., 1996, pp.1872 – 1878.[4] S. Patel and R. G. Hoft, ―Generalized techniques of harmo nic

elimination and voltage control in thyristor inverters: PartI —Harmonic elimination, IEEE Trans. Ind. Applicat., vol. 9,

pp. 310 – 317, May/June 1973.

[5] J.O.P. Pinto, B.K. Bose, L.E. Borges da Silva and M.P.Kazmierkowski, ―A Neural Network based space ve ctor pwmcontroller for voltage fed inverter induction motor drive , IEEETrans, Ind. Appl. Vol. 36, No.6, pp.1628-1636

November/December 2000.

[6] J. Kerkman, B. J. Seibel, D. M. Brod, T. M. Rowan, and D.Leggate, ―A simplified inverter model for on -line control and

simulation , IEEE Trans. Ind. Applicat., vol. 27, no. 3, pp.567 – 573, 1991.

[7] Simon Haykin, ―Neural Networks , ND prent ice Hall, 2004.

[8] Muthuramalingam and S.Himavathi. Performance Evaluationof a Neural Network based General Purpose Space VectorModulator , IJECSE Vol.1, No.1, pp.19 -26, April 2007.

[9] K. Zhou and D. Wang, ―Relationship between space vectormodulation and three phase carrier base PWM – Acomprehensive analysis , IEEE Trans, Ind. Electr, Vol 49,

No.1,pp 186-196, Feb. 2002.

[10] akhshai, J. E spinoza, G. Joos, and H. Jin, ―A combinedartificial neural network and DSP approach to theimplementation of space vector modulation techniques, ―InConf, Rec. IEEE-IAS Annu, Meeting,. 1996 pp.934-940.

[11] H.W. Van Der Brock, H.C. Skundelny and G.V. Stanke,―Analysis and realization of a pulse width modulator based on


6/6



162

voltage space vectors‘, IEEE Trans Ind. Appl., 24. pp. 140 -150,Jan/Feb 1998.

[12] O.Ogasawara, H. Akagi, and Nabel, ― A novel PWM scheme ofvoltage source inverters based on space vector theory, in proc.EPE European conf. Power Electronics and Applications pp.

1197-1202, 1989.

[13] S.R. Bowes and Y.S. Lai, ―The relationship between spacevector modulation and regular- sampled PWM, ―IEE Trans.Power Electron, Vol. 14, pp. 670-679, Sept. 1997.

[14] Reichmann‘s, Bernet.S, ―A Comparison of Three -LevelConverters versus Two-Level Converters for Low-VoltageDrives, Traction, and Utility Applications, IEEE Transactionon Industry Applications, Vol.41, No. 3, pp: 855-865,May/June, 2005.

svpwm using ann

Documents