ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February Space vector PWM echnique for phae voltage ource inverter uing Artificial Neural Network Bandana, K.Banu priya, JBV Subrahmanyam, Ch.Sikanth, M.Ayyub Abtract Space vector pule width modulation (SVPWM) i an optimum pule width modulation technique for an inverter ued in a variable frequency drive application. It i computationally rigorou and hence limit the inverter witching frequency. A neural network ha the advantage of very fat implementation of an SVPWM algorithm that can increae the inverter witching frequency. hi paper propoe a neural Network baed SVPWM technique for a three-phae voltage ource inverter in under modulation region. he cheme ha been imulated and implemented on a V/Hz controlled 5-hp, 5-Hz, and V induction motor drive. he imulation reult are given to validate the performance of the drive with artificial neural network baed SVPWM. Index erm Space vector pule width modulation, neural network, voltage ource inverter. I. INRODUCION Variable frequency ac drive are increaingly replacing dc drive in a number of indutrial application due to advantage in ize, reliability and efficiency. One of the main component of an ac drive i power electronic converter in the form of voltage ource inverter that take dc voltage input (may be from a rectifier) and produce a inuoidal ac waveform. hi in turn i fed to the ac electric motor. he fundamental frequency of thi waveform i adjuted to produce the deired peed. In modern electric drive the modulation of witching i carried out to achieve the required ac wave hape. hi i met by incorporating the duration of ON & OFF method [, ]. he technique to get the duration of ON interval for a particular witch depend upon the control logic or PWM technique to be adopted. In a PWM cheme the output voltage and frequency can be controlled with the help of the witching inide the inverter. he witch may be MOSFE, IGB etc with anti parallel connected diode. Different PWM cheme uch a inuoidal PWM compare a high frequency triangular carrier with three inuoidal reference ignal, known a the modulating ignal, to generate the gating ignal for the inverter witche but having a diadvantage that it contain third harmonic in output[]. o the cancellation of the third-harmonic component and better utilization of the dc upply, the third harmonic injection PWM cheme i preferred in three-phae application. Space vector modulation technique ha advantage of an optimal output and alo reduce harmonic content of the output voltage/current [4]. Space vector PWM (SVPWM) ha the advantage of lower harmonic and a higher modulation index in addition to the feature of complete digital implementation by a ingle chip microproceor, becaue of it flexibility of manipulation, SVPWM ha increaing application in power converter and motor control. he application of artificial neural network (ANN) i recently growing in power electronic ytem. A feed forward ANN implement nonlinear input output mapping. he computational delay of thi mapping become negligible, if parallel architecture of the network i implemented by an application-pecific integrated circuit (ASIC) chip [5]. A feed forward carrier-baed PWM technique, Such a SVM, can alo be looked upon a a nonlinear mapping phenomenon where the command phae voltage are ampled at the input and the correponding pule width pattern are etablihed at the output. herefore, it appear logical that a back propagation-type ANN that ha high computational capability can implement an SVM algorithm. he ANN can be conveniently trained offline with the data generated by calculation of the SVM algorithm. ANN ha inherent learning capability that can give improved preciion by interpolation unlike the tandard lookup table method [5]. he ANN-baed SVPWM of a voltage-fed inverter can alo ue competitive neural network architecture to identify the inverter witching tate and the correponding voltage magnitude for the impreed command voltage. hi paper decribe feed- forward ANN-baed SVPWM that cover linear modulation region. In the beginning, the SVPWM theory ha been briefly reviewed with mathematical analyi for the linear modulation range. hen equation for algorithm have been developed in detail. A back propagation type feed forward ANN i trained offline with the data generated by thi imple algorithm [5]. II. SPACE VECOR PWM A REVIEW hi ection i devoted to the development of Space vector PWM for a two-level voltage ource inverter in linear region of operation [6]. A een from Fig, there are ix witching device and only three of them are independent a the operation of two power witche of the ame leg are complimentary. he combination of thee three witching tate give out eight poible pace voltage vector. he pace vector form a hexagon with 6 ditinct ector, each panning 6 degree in pace. At any intant of time, the inverter can produce only one pace vector. In pace vector PWM a et of three vector (two active and a zero) can be elected to yntheize the deired voltage in each witching period. All of the eight mode are hown in able.. Out of eight topologie ix (tate -6) produce a non-zero output voltage and are known a active voltage vector and 57
ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February the remaining two topologie (tate and 7) produce zero mvs in(6 ) output voltage (when the motor i horted through the upper S V in6 or lower tranitor) and are known a zero voltage vector, variou poible witching tate are hown in Fig. m V in(6 ) Space vector i defined a [6], v v av a v () V a b c in 6 Where a exp j /. he pace vector i a imultaneou repreentation of all the ms in6 (7) three-phae quantitie [4]. It i a complex variable and i function of time in contrat to the phaor. Phae-to-neutral mvs in S voltage of a tar-connected load are mot eaily found by V in 6 defining a voltage difference between the tar point n of the load and the negative rail of the dc bu N. he following m S in( ) (8) correlation then hold true: va va vnn 7 ( ) (9) vb vb vnn () Generalizing the time expreion give vc vc vnn k m in () Since the phae voltage in a tart connected load um to zero, ummation of equation () yield vnn /va vb vc () Subtitution of () into () yield phae-to-neutral voltage of the load in the following form: v / v / v v / / / / a A B C vb vb va vc vc vc vb va Phae voltage are ummarized and their correponding pace vector are lited in able. he eight vector including the zero voltage vector can be expreed geometrically a hown in fig.. Each of the pace vector, in the diagram repreent the ix voltage tep developed by the inverter with the zero voltage V ( ) and V 7 ( ) located at the origin. Space Vector PWM require to averaging of the adjacent vector in each ector. wo adjacent vector and zero vector are ued to ynthei the input reference determined from Fig.4 for ector I. Uing the appropriate PWM ignal a vector i produced that tranition moothly between ector and thu provide inuoidal line to line voltage to the motor. In order to generate the PWM ignal that produce the rotating vector. he PWM time interval for each ector i determined from Fig. 4 for ector I a Along real axi: V ( co 6) V mv S co (5) Along imaginary axi: ( V in 6) mvs in (6) Solving equation (5) and (6) (4) ( k ) m in ; k=,,... () he time of application of active and zero pace vector for all ix ector are given in able []. he period and depend only on the reference vector, amplitude V and the angle. hi how that the period and are the ame in all ector for the ame V S and, poition. In the under modulation region, the vector V S alway remain within the hexagon. he mode end in the upper limit when V S decribe the incribed circle of the hexagon. Modulation Index MI (m) i given by m V V Sixtep Where, V S = input reference vector magnitude V Six tep = fundamental peak value of the ix tep output. he maximum value of input reference i the radiu of larget circle incribed in the hexagon given by V co( V V ). 577V herefore, maximum modulation index V m V SW.577V V.97 () hi mean that 9.7% of the fundamental of the ix tep wave i available in the linear region, compared to 78.55% in the inuoidal PWM [7]. 58
ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February III. NEURAL NEWORK BASED SPACE VECOR Stator reitance (R ):.584 PWM Rotor reitance (R r ):.465 he firt tep that involve in the implementation of neural Stator leakage inductance (L l ):.479 mh network baed pace vector PWM of a voltage ource inverter Rotor leakage inductance (L lr ): 4.5 mh i generation of training data. he training data i generated Mutual inductance (L m ): 78.5mH by uing eq. (), (), ().the angle ubnet i trained with an Rotor inertia (J):. kg.m angle interval of in the range of 6. he phae-a turn ON time can be expreed a t K. V in in, S,6 4 4 t tb K. V in in, S 4 AON t t tb K. V in in, S,4 4 t ta K. V in in, S 5 4 g A Equation (4) can be written in the general form A ON 4 f V g A Where V f i the voltage amplitude cale factor and K in in, S K in in, S K in in, S K in in, S 5 Fig.() how chematic of the ANN baed SVPWM inverter. he input ignal to the neural network i angle. Fig.(4) how the neural network model. he model ue a multilayer function in the firt and econd layer repectively. he Neural network ue one neuron at the input 5 neuron in the hidden layer and three output neuron. he digital word correponding to turn on time are generated by multiplying the output of neural network with V S and then adding 4 a hown in figure. he PWM ignal are then generated by comparing turn-on time with a triangular reference having time period of and amplitude.the PWM ignal are then applied to the inverter. he back propagation algorithm in the Matlab toolbox i ued for the training. he angle ubnet take 65 epoch for training with an error.5%.he value of ize and the correponding training time are, thu reaonably mall.,6,4 V. SIMULINK MODEL AND RESULS he Simulink model ued to imulate model reult on computer i hown in fig. (7).Fig. (8) Show the characteritic of peed of induction motor. In the tarting peed of motor increae very harply and reache to 7 rpm after. econd. Some ocillation are followed till the motor catche the reference peed (i.e. 5 rpm) in. econd. At.75 econd when load torque i changed from Nm to Nm it attain a value lightly lower than the reference peed a hown in the figure. hi lower value of peed i due to increae in torque and the ytem i operating in open loop. Fig.(9) how torque characteritic for load torque (l) and developed torque (e). In the tarting the applied torque i zero and developed torque i very high for few econd (till.5 econd). After ome fluctuation it attain a value of zero in.5 econd, a l=. When load torque i increaed to a value Nm at.75 econd. he developed torque alo attain the value of Nm following ome tranient a depicted in figure. he tator and rotor current for dynamic condition are hown in fig.() &() repectively. he current how the expected trend and are inuoidal in nature. hee current have high initial value during tarting due to tarting tranient, a the induced emf take time to develop it rated value. After thee tranient are over, the current ettle at it teady tate value after. ec. he current (tator and rotor) get higher value when load torque of Nm i applied at.75 ec. hee can be een in the figure. IV. DRIVE SYSEM PARAMEERS -link voltage V Sampling time ( ) Induction motor 5 hp V four pole, quirrel cage Frequency range: -5 Hz Fig.. Power circuit of a three-phae voltage ource inverter [] 59
ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February able- poible mode of operation of a three-phaevsi[] able Application of time [] Fig. -he witche poition during eight topologie [] SECOR V ( ) V ( ) Fig.5 urn on pule width function of phae A, B, C a a function of Angle in different ector [] V S S riangular reference 4 + - V dc AON comparator S A SECOR V4( ) V ( ) V7 ( ) t / t V a V t / t V SECOR V ( ) Neural Network V S S 4 B ON C ON Comparato r Comparato r S B S c Inverter Loa d SECOR 4 V5 ( ) V6 ( ) SECOR 5 SECOR 6 V S S 4 Fig.6 Schematic of the ANN baed SVPWM inverter [5] Fig. - Space Vector repreentation of Line to Neutral Voltage [] Imag. V mv S 6 / / V Real Fig.7 Model of Ann baed pace vector PWM controller for voltage Fed inverter Induction motor drive Fig. 4- Principle of time calculation for SVPWM in ector I[] 6
I r (A) I (A) e & l (Nm) w m (rpm) 8 6 4 8 6 4 ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February propagation-type feed forward network. he training data and training time are reaonably mall. he method can operate from dc (zero frequency). he cheme ha been fully implemented and evaluated with a V/Hz-controlled induction motor drive, and give excellent performance. he PWM controller i currently being ued in a tator-flux-oriented vector-controlled induction motor drive. he ANN-baed SVM can give higher witching frequency, which i not poible by conventional DSP-baed SVM. he witching frequency can be eaily extended up to 5 khz if the ANN i implemented by a dedicated hardware ASIC chip. -..4.6.8..4 ime () 7 6 5 4 e Fig.8 Speed characteritic l VII. FUURE ENHANCEMEN OF WORK he ANN-baed SVM can give higher witching frequency, which i not poible by conventional DSP-baed SVM. he witching frequency can be eaily extended up to 5 khz if the ANN i implemented by a dedicated hardware ASIC chip. REFERENCES - - -..4.6.8..4 ime () 8 6 4 - -4-6 -8 8 6 4 - -4-6 -8 Fig.9 orque characteritic a-phae b-phae c-phae -..4.6.8..4 ime () Fig. Stator current waveform a-phae b-phae c-phae -..4.6.8..4 ime () Fig. Rotor current waveform VI. CONCLUSION A neural-network-baed pace-vector modulator ha been decribed that operate very well in under modulation region. he digital word correponding to turn-on time are generated by the ANN and then converted to pule width through a ingle timer. he cheme ue a back [] W. vander Broek, H. C. Skudelny, and G. V. Stanke, Analyi and realization of PWM baed on voltage pace vector, IEEE ran. Ind. Applicat., vol. 4, no., pp. 4 5, 988. [] Grahame Holme and homa A. Lipo, Pule Width Modulation For Power Converter Principle and Practice, IEE Pre, Wiley Publication. [] Blako, A hybrid PWM trategy combining modified pace vector and triangle comparion method, in IEEE PESC Conf. Rec., 996, pp.87 878. [4] S. Patel and R. G. Hoft, Generalized technique of harmonic elimination and voltage control in thyritor inverter: Part I Harmonic elimination, IEEE ran. Ind. Applicat., vol. 9, pp. 7, May/June 97. [5] J.O.P. Pinto, B.K. Boe, L.E. Borge da Silva and M.P. Kazmierkowki, A Neural Network baed pace vector pwm controller for voltage fed inverter induction motor drive, IEEE ran, Ind. Appl. Vol. 6, No.6, pp.68-66 November/December. [6] J. Kerkman, B. J. Seibel, D. M. Brod,. M. Rowan, and D. Leggate, A implified inverter model for on-line control and imulation, IEEE ran. Ind. Applicat., vol. 7, no., pp. 567 57, 99. [7] Simon Haykin, Neural Network, ND prentice Hall, 4. [8] Muthuramalingam and S.Himavathi. Performance Evaluation of a Neural Network baed General Purpoe Space Vector Modulator, IJECSE Vol., No., pp.9-6, April 7. [9] K. Zhou and D. Wang, Relationhip between pace vector modulation and three phae carrier bae PWM A comprehenive analyi, IEEE ran, Ind. Electr, Vol 49, No.,pp 86-96, Feb.. [] akhhai, J. Epinoza, G. Joo, and H. Jin, A combined artificial neural network and DSP approach to the implementation of pace vector modulation technique, In Conf, Rec. IEEE-IAS Annu, Meeting,. 996 pp.94-94. [] H.W. Van Der Brock, H.C. Skundelny and G.V. Stanke, Analyi and realization of a pule width modulator baed on 6
ISSN:77-754 International Journal of Engineering and Innovative echnology (IJEI) Volume, Iue, February voltage pace vector, IEEE ran Ind. Appl., 4. pp. 4-5, Jan/Feb 998. [] O.Ogaawara, H. Akagi, and Nabel, A novel PWM cheme of voltage ource inverter baed on pace vector theory, in proc. EPE European conf. Power Electronic and Application pp. 97-, 989. [] S.R. Bowe and Y.S. Lai, he relationhip between pace vector modulation and regular-ampled PWM, IEE ran. Power Electron, Vol. 4, pp. 67-679, Sept. 997. [4] Reichmann, Bernet.S, A Comparion of hree-level Converter veru wo-level Converter for Low-Voltage Drive, raction, and Utility Application, IEEE ranaction on Indutry Application, Vol.4, No., pp: 855-865, May/June, 5. 6