Direct Torque Control of Induction Motor Fed by Three-Level
NPC Inverter Using Fuzzy Logic and NeuralNetwork
F. Kadri*, S. Drid**, and F. Djeffal***
*Université Kasdi Merbah Ouargla, LaboratoiredeGènie Electrique. Faculté desSciences et dela
Technologie etdesSciencesde la Matière,Ouargla,30000,Algérie : Email:
**Department of Electrical Engineering, Batna University, Email:
***Department of ElectricalEngineering ,Batna University, Email:
Abstract—multilevelvoltageinverter hasbeen receiving wideattentionin researchandhighpower applications.Compared with thetraditionaltwo-level voltage inverter,themainadvantagesofthemultilevel inverter are a smaller output voltage step, lower harmonic components, a better electromagnetic compatibilityandlower switchinglosses.Inthispaper direct torque control (DTC) is applied for three-level NPCinverterfedinductionmachinedrives.Two control approaches using FuzzyDTC,and Neural Network DTC areproposed and compared.Thevalidityof theproposed control scheme is verified bysimulation tests of an induction motor drive system. The stator current and voltage,flux,andtorquearedeterminedandcompared in the abovetechniques using MATLAB-SIMULINK environment.
Keywords— DTC,Inductionmotor, Multi-levelinverter, PWM-inverter,Fuzzylogic,NeuralNetwork, Neural-point clamped,Switchingalgorithm,PIregulator.
I.INTRODUCTION
Multilevel inverters have drawntremendous interest in thefieldofhigh-voltage high-powerapplications.Its concept isbasedonproducingsmalloutput voltage steps,resulting in betterpowerquality.Despitethe needformore powertransistors,theyoperateatlow voltagelevels andalsoatlowswitchingfrequencyso thattheswitchinglosses arealsoreduced.Other advantages include better electromagnetic compatibility due to the low dv/dt transitions [1]. Someofthe fundamentalmultileveltopologiesinclude thediode-clamped [2].
DirectTorque Control(DTC) wasfirstintroducedby Takahashiin1986[3].The objective ofourworkisto implement aswitching pattern by fuzzy, and neural networkthat describethe selectionstrategiesofthe inverter state exploiting the simplicity of implementationandrobustness offeredby artificial intelligent control.
II.THETHREELEVELNEUTRALPOINTCLAMPED
INVERTER
A multilevel voltage source inverter is a converter structure that can provide more than twolevels of line to groundvoltage inthe output ofeach leg ofthe inverter.
Multilevel powerconversiontechnology isa very fastgrowing area of powerelectronics with good potentialfor further development.Themostattractive featuresofthistechnologyareinthemedium tohigh- voltage application range (2-13 kV), which include motor drives, power distribution, power qualityand power conditioning applications. Different circuit topologies have beenimplementedinmultilevel inverters.One ofthemostusedofthesetopologiesis theNeutral PointClamped(NPC)topology[2]-[4].In Fig.1, thescheme for thethree-level NPC inverteris presented.
Fig.1.Three-level NPC inverter
InFig.2,thedifferent vectorsorinverterstates available, ina three-levelinverter,areshowninthe statorfluxlocus. Ascan be seen,there are 4 different kindsof vectors:
Zero vectors:V0,V7, and V14.
Large vectors:VI5,V16,V17, V18,V19,V20. Medium vectors: V21, V22, V23, V24, V24, V25, V26.
Smallvectors:V1,V2,V3,V4,V5,V6,V8,V9,V10, V11, V12, V13.
Althoughinathree-level inverterthere are 27 possiblestates,someofthemapplythesamevoltage
vector.Therearetwopossibleconfigurationsforeach
smallvectorandthreeforthezerovectors.Therefore,
19 different vectors are available in a three-level inverter [5]-[6].
⎧if T2
⎪
Then KT =+11.
⎪if
-ε ≤ΔT
≤ε and dΔT/dt0ThenK =+1.
⎪if
⎪
-ε ≤ΔT
≤ε and dΔT/dt<0
ThenKT =+11.
⎪if
0≤ΔT T1
and dΔT/dt0ThenKT =0.
(4)
⎪if
⎨
0≤ΔT T1
and dΔT/dt0
ThenKT =+1.
⎪if
-ε ≤ΔT0
and dΔT/dt0ThenKT =−1.
⎪if
⎪
-ε ≤ΔT0
and dΔT/dt0
ThenKT =0.
⎪if -εT2 ≤ΔT≤-εT1
⎪
and dΔT/dt0ThenKT =−11.
if
⎪
⎪⎩if
-ε ≤ΔT
ΔT-ε
≤-ε and dΔT/dt<0
ThenKT =−1.
ThenKT =−11.
Fig.2. Voltagevectorsof athree-levelNPC inverter
III. DEVELOPMENTOFTWODTCAPPROACHES
A.BasicDTC
DTCtechniqueisaconvenient,relatively complete and easy applicablemethod. Inthismethod,the requiredstatorfluxandtorqueare determinedanda properswitching pattern of theinverteris thenapplied “Fig.4”.Sincethemotorinput voltage contributesin bothfluxand requiredtorque,aproper switching pattern forinverterfiringisessential.In ordertodefine the inverter switching patternusing flux and torque errors,two hysteresiscontrollers, one forthefluxand otherfortorqueareemployed[7]-[8] “Fig.3”. At any instant,theinverterisswitched on usingtheseerrors andpositionofstatorfluxovertwelve-regioncontrol insucha waythattheinverteroutputvoltage vector minimizesthefluxandtorqueerrorsand definesthe directionofthefluxrotation.The outputs ofthese controllersareSi(i=a,b,c)wheretheirvalues(-1,0 or1)are usedtodeterminetheinverter output voltage vector asfollows “Fig.1”.
WhereεT1,εT2, andεΦare the predefinederrorlevels ofthetorqueandfluxrespectively.Finally,basedon thevaluesofconstantsKT andKΦ andthepositionof the stator (twelve-region control), the inverter switchingalgorithmis as shown in Tab.1.
Fig.3.Torqueand Fluxcomparator hysteresis
2 ⎛ i4π i4π⎞
v = ⋅U0 ⋅⎜S
+S ⋅e 3 +S
⋅e 3 ⎟
(1)
s 3 2 ⎜ a b c ⎟
(i= a,b, c)
⎧Si =−1⇒(Si1,Si2,Si3,Si4)=(0,0,1,1)
Fig.4. Inductionmotor directtorquecontrolblock diagram(BasicDTC).
⎪ ⇒(S
⎪ ⇒(S
,Si2
,Si2
,Si3
,Si3
,Si4
,Si4
)=(0,1,1,0)
)=(1,1,0,0)
(2)
Thestatorfluxcanbeevaluatedby integratingfrom thestatorvoltageequation:
The control algorithm of this technique can be summarizedas follows[9].
φs =∫(vs
−Rs
⋅is)⋅dt
(5)
⎧if Δϕ
⎪
εϕ
ThenKϕ =+1.
Theelectromagnetictorqueis estimatedfromthefluxand currentinformationas:
⎪if
0≤Δϕ ≤
εϕ anddΔϕ/dt0ThenKϕ =0.
T=p
⋅(φ
⋅iqs
+φqs
⋅ids )
(6)
⎪if
⎨
0≤Δϕ ≤
εϕ anddΔϕ/dt0
ThenKϕ =+1.
(3)
⎪if -εϕ ≤Δϕ ≤ 0
anddΔϕ/dt0ThenKϕ =−1.
Thestatorfluxangleθsiscalculatedby:
⎪if -ε ≤Δϕ ≤ 0
anddΔϕ/dt0
ThenKϕ =0.
⎛φqs ⎞
⎪ θs =arctg⎜ ⎟
(7)
if Δϕ
-εϕ
ThenKϕ =−1.
⎝φds ⎠
TABLEI
SWITCHINGTABLEFORBASIC DTC.
B.FuzzyDTC
Theproposed fuzzy controller is Sugeno model, usingtriangular and trapezoidalmembership for inputs, andconstant (Vi,i=0,…26.)for outputs.
(a) (b)
(c)
Fig.5.Membership functionsof inputs ofFuzzy DTC.
(a)Torqueerror(ΔT);(b)Fluxerror(Δφ);(c)FluxSector(S)
Inthisapproach, we replacedthe two hysteresis controllers (basicDTC)and theswitching tablebya single fuzzy controller withtorqueerror, flux errorand Flux sector as inputs. The vector voltage of the inverterwill bethe controloutput.
Thecontrol strategy dependsmostly oninference rules. The inference is based onthe method ofsum- productinference.Tosimplify thedescription of inferences,we useatableinference"TableII".For defuzzification,the center ofgravity method was used
Fig.6.Induction MotorFuzzyDirectTorqueControl
BlockDiagram(FuzzyDTC)
TABLEII
INFERENCE RULES
C.NeuralNetworkDTC
The neuron network with backpropagation learning algorithm consistsofaninputlayer,outputlayeranda hiddenlayer. The weighted connectionbetweentwo successivelylayershavetheirvaluesdeterminedby the iterative learning stage.
Duetothe paralleldataprocessing,the learning capacity,thegeneralizing, andsynthesizing properties; the neuron network can extract the adequate
mathematicalmodelfromthepresentedinputpattern
and provide the good approximation of the desired target in the irrelevant case too [10].
Inthisapproach,we replacedtheswitchingtable of basicDTC by aneural networkcontrollerwithtorque errorεT, fluxerrorεΦand numberofsector Sasinputs. The vectorvoltageoftheinverter will bethecontrol outputVi,i=0, …26.
IV.SIMULATIONRESULTS
Specifications of the motor under control are summarized as follow:
TABLEIII
CHARACTERISTICSOFTHEINDUCTIONMOTOR
F (Hz) / P (KW) / p / V (V) / RsΩ / Rr
Ω / Ls=Lr
(H) / M (H) / J (kgm2)
50 / 1.5 / 2 / 280 / 0.5 / 1 / 0.105 / 0.1 / 0.02
Fig.7Induction MotorFuzzyDirectTorqueControlBlock
Diagram(NeuralnetworkDTC)
Thetrainingofthe neuron networktoobtainthe wished outputs wasobtained withanerror of0.1%.
Therearetwo hiddenlayers with40neuronsinthe first hidden layer and 50 neurons in the second hidden layer.Thenumberoftheoff-linetrainingepochsis
163 toreach the 0.001-imposed error “Fig 8”.
Fig.8 Training ErroroftheDevelopedNeuronNetwork
DirectTorque Controlmodeling ofinductionmotor was performed in the environment of Matlab-Simulink simulation. The DTC command structure is made with various blocks.The realizationofneuralnetwork controllerwas throughprogramminginMatlab(*.m files).Thetrainingwas performedoffline,andthenthe transition to the neural network control Simulink model will be called by Matlab function. Concerning the fuzzycontroller, it is through theuse ofthe toolbox of fuzzylogic whichiscalledafile(*.Fis). Inpractice, therearetwoimportant factorsthat affectthe performance of the DTC: the error bands, and samplingfrequency.Concerningtheerror bands,the practice has shown that the choice to 4% of the nominalvaluesgivesthe bestoperatingrange.This justifies the choice of this value in our simulation work.
A. General Performance of the System
“Fig.9,”showsthesystem start-upstabilizationin ordertoobserveits dynamicbehaviorundertransient state,untilitreachesitssteadystate. We observethat thecontrolsystem reachesitssteadystatebyshowing good performance and great stability for the two approaches. The flux and torque stabilize around their referencevalues.(Fs =0.75Wb.),(Ts =10N.m.).
(a) (b) Fig.9.Systemresponseinstart-upstate(Ts=10N.m,Fs=0.75Wb.)
(a)NeuralDTC, (b)FuzzyDTC
We notealso,thatthe torque ripplesare greaterthan the flux ripples. This is due to the stator phase currents,whichdirectlyaffectthetorque. Thestability ofthe system canbe observedoverthestatorphase voltageVab,whichpresentsaperiodicPWMform. Thecurves obtainedbythecontrollersforthetwo approaches havethesamelook,except aslight difference, which is observedat the start-up state:
•In the neural network DTC, the flux reaches its
referenceandstabilizesinsteadyslightlyfasterbut
it gave largeripplesthan theFuzzyDTC.
•VoltagedynamicgivenbyFuzzyDTCwasbetter
thanthosegivenby theneuralnetwork DTC.
B.Steady Performanceof theSystem
Toobservetherobustnessandsystemperformance in steady state, we conducted thesimulationin order to observethe influenceof references (torque and flux) withatest ofaninversionoftorque,anda variation of fluxsimultaneously.Thistestisdonetoillustrateand to demonstrate the decoupling control between the torqueandflux for the two approaches. “Fig10”.
Itcanthereforebe notedthattheinversion of the couple does not affect the stability of flux and the variation of the flux does not also affecting the stabilityofthetorque.Weconcludethattheproblem of coupling characterizing the induction motor has been solved by our control system.
V. CONCLUSION
DTC withaNeutralPointClamped(NPC) Inverter hasbeenstudiedinthepresentwork. Duetothe enhancedpossibilitiesforthe inverterstateselectionin a three-level inverter an intelligent controller has been designedinordertoreplace theconventionalmethod basedonaswitchingtable. Thisapproachprovidesa moreaccurateselectionof theinverterstateby introducingfuzzy and neural network regulators.
Webelieve thatweappliedastructure of identificationofnonlinearsystems.Theeffectiveness ofthisstructurewasshownbysimulationby giving satisfactoryresults
VI.REFERENCES
[1] A.K. Al-Othman, T.H. Abdelhamid, “Elimination of harmonics inmultilevelinverterswithnon-equal dcsources usingPSO”,ElectricPowerSystemsResearch78,pp1736–
1746,2008.
[2] Y. Xiaoming, I. Barbi, “Fundamentals of a new diode clampingmultilevelinverter”,IEEETransPowerElectron,
15(4):711–8,2000.
[3] I.Takahashi,T.Noguchi,"Anewquick-responseandhigh efficiencycontrolstrategyofaninductionmotor",IEEETrans onIndustrialApplications,Vol.IA-22,N°.5,pp820-827,1986.
[4] V.Chitra,R.S.Prabhakar,“InductionMotorSpeedControl usingFuzzyLogicController“,Proceedingsofworldacademy ofscience,Engineeringandtechnology,Volume 17,December
2006.
[5] Z. Yongchang, Z. Jianguo, Z. Zhengming, “An Improved
DirectTorqueControlforThree-LevelInverter-Fed Induction
Motor Sensor less Drive“, IEEE Transactions on Power
Electronics,Vol.21,no.5,November2010.
[6] K.Jinsong,W.Ying,X.Yonggan,“DirectTorqueControl
TechniqueforDriveSystemFedonThree-LevelInverter",
International ConferenceonControl,AutomationandSystems
Engineering,IITA,pp105-107,2009.
[7] H.Liming,L.Yunfeng,F.Xiaoyun,C.Hengbin,“Improved
Virtual Vector Direct Torque Control of Induction Motor BasedonThree-level NPCInverter”,TheNinthInternational Conference on Electronic Measurement Instruments, pp
668-672,2009.
[8] I.Messaif, E.M.Berkouk, N.Saadia,"Nouvelle tablede commandepourlecontrôle directducoupledunemachine
asynchronealimentée parunonduleurà3niveaux detension",
CGE07,EMP,Algérie12-13avril.2007.
[9] X.D.T.Garcia,A.Arias,M.G.Jayne,P.A.Witting,“Direct
TorqueControlofInduction MotorsUtilizing Three-Level Voltage Source Inverters", IEEE transactionson industrial electronics,Vol.55,no.2,February2008.
[10] R .Toufouti, S .Meziane , and H. Benalla, “Direct torque controlforinductionmotorusingintelligent techniques”, Journalof Theoretical and Applied Information
Technology,2007
(a) (b) Fig.10.Inversionoftorque(+10 →−10 →+10N.m.)andavariationoftheflux.(0.75→
(a) NeuralDTC, (b)FuzzyDTC
0.5→
0.75Wb.)