InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013

DOI:10.14355/ijes.2013.0305.01

www.ijesci.org

RealTimeDiagnosisinWindEnergy
ConversionSystem
OuadieBENNOUNA*1
IRSEEMEA4353(InstitutdeRechercheenSystmesElectroniquesEmbarqus),TechnopleduMadrillet,avenue
Galile,BP10024,76801SaintEtienneduRouvrayCedexFrance
*1

bennouna@esigelec.fr

Abstract

(Clment, McCullen, Falcao, Fiorentino,

Gardner & Hammarlund, 2002) and
(McCormick,1981).

The current paper presents an FDI (Fault Detection and

Isolation) technique applied to the Doubly Fed Induction
Generator (DFIG) of a Wind Turbine (WT). The method is
based on the use of a Generalized Observer Scheme (GOS)
usingaKalmanfilter.Thefaultlocalizationisperformedby
the PageHinkleys test. The whole procedure is validated
usinganumericsimulator,thenarealbenchmarkemulating
a wind turbine. Finally, real time tests are performed to
allowrealtimevalidation.
Keywords

Solar photovoltaic, the direct generation of

electricityfromlightassociatedwithenormous
solar energy resources is one of the most
promising principles of power generation.
Photovoltaic conversion is static, offers total
quietoperationandallowsahighreliability.

Solar thermodynamic, legacy solutions to fuel

power plants, solar power plants can operate
directly heat radiated by the sun to transform
waterintosteam.Anintermediateheatstorage
allows smooth production. Even with the low
thermodynamic efficiencies of steam turbines
(about 30%), an annual production of 4.106
kWhwithapeakelectricalpowerof3MWcan
beconsidered.

Geothermal power, in some favorable areas,

deep water at high temperature is found by
drilling. In the range of 150C to 350C, it is
called high geothermal energy. Water is
pumpedtothesurfaceandpassesthroughheat
exchangers,steamproducedisthenusedasin
conventionalthermalpowerplants.

Windpowerwhosepotentialisquiteimportant
nottoreplaceexistingenergybuttoanswerto
the growing demand. After centuries of
evolution and further research in recent
decades, several countries are now resolutely
turned towards the wind energy. However,
severalreasonscanaffecttheperformanceofa
wind turbine (Hau, 2006) (Trujillo, Bingl,
Larsen, Mann, Khn, 2011). It can be wind
speed,windturbinelocation,blades(Srensen,
Schreck, 2011). The performance optimization
hasattractedtheattentionofmanyresearchers
(Mirecki, Roboam, Richardeau, 2007) (Jauch,
Matevosyan, Ackermann, Bolik, 2005). This

Diagnosis;WindTurbine(WT);DoublyFedInductionGenerator
(DFIG); Generalized Observer Scheme (GOS); PageHinkleys
Test;RealTimeTest

Introduction
The exhaustion of fossil energy resources and
environmental problems caused by gas emissions,
allow the development of renewable energy sources,
andofferthepossibilitytoproduceelectricityproperly
on the condition to accept its random fluctuations.
Someoftheserenewableenergysourcesare:

Hydraulicenergy,thefirstsourceofrenewable
energy for electricity production. The world
power installed reached 740 GW, when the
annualproductionis2.7.1012kWh.TheSmall
HydroPower(SHP)plantsareattractivedueto
its distributed production. It is, by definition,
powerslowerthan10MW(rangegoesfrom0.5
MWinLuxembourgto50MWinBrazil).They
are considered as new solutions to produce
energy from renewable sources. In France,
even iflargehydropoweris saturated,there is
still potential for development of SHP, 3500
units with maximal power of 1700MW and
about 7.109 kWh per year (Multon, Robin,
Ruellan & Ben Ahmed, 2004). In this context,
we can also talk about wave energy which is
also a renewable energy source studied by

297

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01

(Beltran,AhmedAli,Benbouzid,2008).

involves finding an adequate model of the

wind turbine (WT), to detect errors that may
impair its proper operation in order to
guarantee its control (Hazra, Sensarma , 2010)
(Todeschini,Emanuel,2009).

Inthewindpowerindustry,theDoublyFedInduction
Generator (DFIG), with the Permanent Magnet
Synchronous Generator (PMSG), is one of the most
usedelectricalmachineinWTbecauseofitslowcost,
its simplicity of maintenance, and its reliability.
Nevertheless, the DFIG is not a faultfree system. A
defective bearings, windings or electrical insulation
canaffectthegenerator.

Toapproachtheirtruedynamics,severalwindturbine
models have been developed (Soliman, Malik,
Westwick,2009).Therearetwoclasses:analyticaland
numerical models (Bertagnolio,Rasmussen, Srensen,
Johansen, Madsen, 2010). The modeling steps and
differentmodelsoftheWTaredetailedin(Ackermann,
2005).

In previous work (Bennouna, Hraud, Rodriguez,

Camblong, 2007), a technique has been developed
allowing the diagnosis of the DFIG. This method is
based on the polynomial representation of the
different variables of the system. Thus, the data are
validated when no error is present; otherwise errors
aredetected,localized,identifiedthen,estimated.This
paper presents a Fault Detection and Isolation (FDI)
technique dealing with wind turbine. This method is
based on the use of Generalized Observer Scheme
(GOS) by a bank of Kalman filter, which allows
detectingandisolatingmultiplesensorfaults.

In addition, WTs are generally inaccessible (several

meters in height), control and diagnoses are not easy
tasks(Amirat,Benbouzid,AlAhmar,Bensaker,Turri,
2009). Different signatures can be identified on the
operationoftheWT,inparticularvibration,whichcan
lead to defects such as cracking the motor shaft.
Atmosphericconditionscanalsopowerdeficits,hence
the need to stabilize turbulence (Mirecki, Roboam,
Richardeau, 2007). These defects must be detected as
soon as possible to minimize damage (Hansen,
Barthelmie,Jensen,Sommer,2012)(Margaris,Hansen,
Sorensen, Hatziargyriou, 2010). However, the
diagnosisprocedureisjustthefirststeptowardanew
synthesis of the control law in which presence of
defect is taken into account. The second step is a
reconfigurationapproachbasedoninformationofthe
FDI module (Fault Detection and Isolation) able to
give a new control law and to ensure the stability of
the closed loop system, and to guarantee the system
performances or their minimal degradation in fault
presence:ItistheFaultTolerantControl(FTC)(Tastu,
Pinson,Kotwa,Madsen,Nielsen,2011)(Patton,Putra,
Klinkhieo,2008).A system is calledfault tolerant one
if it can maintain the design objectives even when
thereisfault.

The layout of the article is as follows. Section 2

explains the principle of the method GOS for the
diagnosis with the use of the PageHinkleys test.
Section 3 presents the model of the Doubly Fed
InductionGenerator(D.F.I.G).Testsandresultswillbe
investigated in section 4. Section 5 is devoted to the
realtime validation using Matlab RTW (RealTime
Workshop)anddSAPCE.Finally,concludingremarks
arepresentedinsection6.
Diagnosis Procedure: GOS Approach
The GOS approach is one of the method to solve the
problem of state observer for Fault Detection and
Isolation.
In this paper, FDI is done using a GOS approach
concerningsensorsfaults.Thestatevectorisassumed
completely observable and there is no uncertainty
aboutthemodelparameters.Theithobserverisdriven
by all inputs and all outputs except the ith. Thus, the
output of this observer will be sensitive to defects in
alloutputsexcepttheith.

The FTC is dividedinto two approaches: passive and

active. In the passive one, regulators are synthesized
toberobusttosomedefects.Themainideaistomake
theclosedloopsystemrobusttoincertitudeandsome
specificdefects.Theweaknessofthisapproachisthat
its capacity to tolerate defaults remains restricted to
some faults only. The active approach has to react to
various systems failures in order to preserve the
stability and the performance of the system. A fault
tolerant control approach, in multimodel, was
developed in (Odgaard, and Stoustrup, 2009). Other
control laws have been developed using neural
networks (Jurado, Saenz, 2002) and other techniques

298

The observer used in this study is the Kalman filter

whichallowsestimatingthestateofthesystemusing
themodelandtheinputs.
Thedynamicmodelofdiscretesystemsisgivenby:
(1)

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01

www.ijesci.org

where
isthestate,
ismeasurement,andA,
B,Harematricesofappropriateddimension.wandv
representtheprocessandmeasurementnoises.

fields; has permitted implementation of efficient

controllawofthismachine.However,manyproblems
remaineitheratthemachineorsensors.

TheKalmanfilterprocedureconsistsoftwosteps:

DoublyFedInductionGenerator(DFIG)

Predictionstep:

The mathematical model of the D.F.I.G in the (, )

basisisgivenby(Bennouna,Hraud,2011):

Rs Lr mLh2
is
LhRr
mLr Lh is

2
mLh Rs Lr mLr Lh LhRr is
d is
1

dt ir (Ls Lr Lh2) LhRs mLs Lh Rs Lr mLs Lr ir

i
L L
mLs Lr Rs Lr ir
r
m s h LhRs
Lr 0 Lh 0 us

1
0 Lr 0 Lh us

(Ls Lr Lh2) Lh 0 Ls 0 ur

0 Lh 0 Ls ur
(4)
Where:

(2)
Correctionstep:

(3)
More details about the Kalman filter can be found in
(Zarchan,Musoff,2005).
ThestateestimationgivenbytheKalmanfilterallows
defining residual vectors. The PageHinkleys test
(Basseville, 1986) is subsequently applied in order to
detect and isolate the sensors failures. Isolation
structureofsensorserrorsis:

i s , s i r , r
,
are respectively the currents of the stator
andtherotoronthephasesalphaandbeta.
u s , s u r , r
,
are respectively the voltages of the stator
andtherotoronthephasesalphaandbeta

m istherotationalspeedofthegenerator.
Ls , Lr , Lh arerespectivelytheinductanceofthestator,

therotorandthemutualinductance.

Rs Rr
, arerespectivelytheresistanceofthestatorand
therotor.
The(,)basisisobtainedfromtheclassicbasis(a,b,c)
using the Park transformation which requires the
knowledgeofthetetaangle.

FIG.1ISOLATIONSTRUCTUREOFSENSORSERRORS

Themodelgivenabovecanberewrittenintheformof
staterepresentation,where:

Application

i s
u s

i s
u s

x
u
i r
u r

i r
u r
0 Lh
Lr

0
L
0
1
r

B
2
0
Ls
( Ls Lr Lh ) Lh

0
0 Lh

Introduction
In this paper, the application chosen deals with the
Doubly Fed Induction Generator (DFIG) of a wind
turbine.Indeed,electricalmachinesarewidelyusedin
industryandespeciallytheinductiononewhichisthe
most robust and cheapest in the market. Since its
maintenance is insignificant, its field of use has been
extended to advanced fields such as aerospace,
chemical or wind turbines. However, despite its
robustness, defects could appear, during its lifetime,
becauseofstress.

Rs Lr

2
1
m Lh
A
2
( Ls Lr Lh ) Lh Rs

m Ls Lh

Progress in power electronics and microelectronics

m Lh 2

Lh
0

Ls

Rs Lr
m Ls Lh

Lh Rr
m Lr Lh
Rs Lr

Lh Rs

m Ls Lr

m Lr Lh

m Ls Lr

Rs Lr

Lh Rr

299

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01

Tests & Results

This scheme consists of several blocks: the system

model as a discrete state representation, the fault
simulation,thebankofKalmanfilterandthedecision.

TestsStructure
Firsttestswereachievedusingthenumericsimulator
(Matlab/Simulink). A full representation of a wind
turbinewithaD.F.I.Gwascreated,asshownnext:

Anexperimentalbenchmarkwasalsousedcomposed
ofaDCmachineof25kWemulatingtheaerodynamic
and mechanical behaviour of the wind, and a D.F.I.G
of15kWplayingtheroleoftheelectricmachine.
Thissystemconstitutesanintermediatestagebetween
testsonthenumericsimulator(Matlab/Simulink)and
arealwindturbine.

FIG.2GENERALREPRESENTATIONOFAWINDTURBINE
USINGMATLAB/SIMULINK

FIG.5THEEXPERIMENTALBENCHMARK

The diagnosis procedure is given in the following

figure:

The control of the electric machines is done by using

Digital Signal Processors (DSP). Digital to analogue
and analogue to digital converters assure the
communication between these DSP and the electronic
components of the system. The DSP of the DC
machine sends a reference of the electromagnetic
torquetotheDSPoftheD.F.I.Gbyusingtheirinputs
andoutputsandconverters.

FIG.3DIAGNOSISPROCEDURE

The generalized observer scheme for fault detection

andisolationbyabankofKalmanfilteris:

FIG.6COMPOSITIONOFEXPERIMENTALBENCHMARK

Figure 6 shows the composition of the experimental

system.Allthecomponentsare:

FIG.4THEGOSFORFDIBYABANKOFKAMANFILTER

300

DCM: is the DC machine, whose role is to

emulate the aerodynamic and mechanical
behaviourofthewind.

PCDCMandNSPDCM:theNSPoftheDCM
contains the simulations model and the
equationsoftheDCM.

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01

DFIG: plays the role of the generator of the

windturbine.

PCDFIG:theDFIGisconnectedtothenetwork
byamagnetizing,thenasynchronisationofthe
electricvariablesofthenetworkandthestator
oftheDFIG.

www.ijesci.org

Severaltestswerecarriedoutaccordingtothreekinds
ofdefects:additive,multiplicativeandlooseofsensor.
Results
An error of 20% is multiplied to the rotor current on
i
alpha phase r from the sample 250. Figure 7
representsthefourcurrents.

FIG.9RESIDUESOFTHEGOSSTRUCTURE

Using a logical decision, we can easily find the

suspected variable. In this case, it is indeed the rotor
i
currentonalphaphase r .Theoccurrencetimeofthe
defect can also be known, as shown in the following
figure:

FIG.7THEFOURCURRENTSOFTHEDFIG

Fromtheabovefigure,itisunabletodetectifthereisa
problemontheDFIG.Nevertheless,theresiduesofthe
fourstatevariables,giveninfigure8,showthatthere
isaprobleminthegeneratorwithoutlocalizingit.

FIG.10FAULTDETECTIONOFTHESUSPECTEDVARIABLE

RealTimeTests
The work presented in this paper was validated on a
test bench, which is based on a dSPACE DS1006
processor board. This board dispose of an AMD
Opteronprocessorclockedat2.6Ghz.

FIG.11REALTIMETESTBENCH

FIG.8RESIDUESOFTHEFOURCURRENTSOFTHEDFIG

The Simulinks model of the DFIG and the fault

detection procedure were translated in C code using
Matlab RealTime Workshop then downloaded to the
test bench. The result of the FDI processus was
displayed on LEDs of the dSPACE DS4002 card. This

TheadvantageofusingaGOSstructureisnotonlyto
detectbutalsotolocalizethedefect.Thus,inourcase,
the results corresponding to residues of the four
observersaregiveninfigure9.

301

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01

validationindicatedthatourapproachcanbeusedin
arealtimeenvironment.

(4):323338.DOI:10.1002/we.342
Blanke M, Kinnaert M, Lunze J, Staroswiecki M. Diagnosis
and FaultTolerant Control. SpringerVerlag: New York,

Conclusions

2003,ch.7.

A technique of fault detection has been presented in

thispaperbasedontheuseofaGeneralizedObserver
Scheme. This method is applied to a Doubly Fed
Induction Generator of a wind turbine using an
experimental benchmark. Results show that the
proposed fault diagnosis method can detect and
localize three kinds of errors: additive, multiplicative
and loose of sensor. However, this technique cannot
beusedinthecaseofpresenceofmultiplefaults.

ClmentA,McCullenP,FalcaoA,FiorentinoA,GardnerF,
Hammarlund K, Lemonis G, Lewis T, Nielsen K,
PetronciniS,PontesM,SchildP,SjstrmB,ChristianH,
Thorpe T. Wave energy in Europe: current status and
perspectives. Renewable and Sustainable Energy Reviews
2002;6(5):405431.
Hansen KS, Barthelmie RJ, Jensen LE, Sommer A. The
impact of turbulence intensity and atmospheric stability

Our next work will focus on use of the Dedicated

ObserverSchemewhichallowsdetectingandisolating
the simultaneous sensor faults as well as the
comparisonofproposedobserverschemeperformance
withotherones.OtherFDIproblemsdealingwiththe
mechanicalcouplingwillbestudied.

on power deficits due to wind turbine wakes at Horns

Revwindfarm.WindEnergy2012;15(1):183196.DOI:
10.1002/we.512
HauE.Windturbines:Fundamentals,Technologies,Application,
Economics.SpringerVerlag:NewYork,2006.
HazraS,SensarmaP.Vectorapproachforselfexcitationand

REFERENCES

controlofinductionmachineinstandalonewindpower
Ackermann T. Wind Power in Power Systems. John Wiley &

generation. IET Renew. Power Gener.; 5 (5) : 397405.

Sons,2005,ch.27.

http://dx.doi.org/10.1049/ietrpg.2010.0168

AmiratY,BenbouzidMEH,AlAhmarE,BensakerB,TurriS.

JauchC,MatevosyanJ,AckermannT,BolikS.International

A brief status on condition monitoring and fault

comparison of requirements for connection of wind

diagnosisinwindenergyconversionsystems.Renewable

turbinestopowersystems.WindEnergy2005;8(3):295

andSustainableEnergyReviews2009;13(9):26292636.

306.DOI:10.1002/we.160

Basseville M. Detection of abrupt changes in signals and

Jurado F, Saenz JR. Neurofuzzy control in biomassbased

dynamical systems. ControlandInformationSciences1986;

winddieselsystems.InProceedingsofthe14thPSCC2002,

77:926.

Sevilla,Spain.

Beltran B, AhmedAli T, Benbouzid MEH. Sliding mode

Margaris D, Hansen D, Sorensen P, Hatziargyriou D.

powercontrolofvariablespeedwindenergyconversion

Illustration of modern wind turbine ancillary services.

systems. IEEETransl. onEnergy Conversion 2008; 23 (2) :

Energies2010;3(6);12901302.

551558.

McCormick ME. Ocean wave energy conversion. Wiley

Bennouna O, Hraud N, Rodriguez M, Camblong H. Data

Interscience:NewYork,1981.

reconciliation and gross error detection applied to wind

Mirecki A, Roboam X, Richardeau F. Architecture

power. Proceedings of the Institution of Mechanical

complexityandenergyefficiencyofsmallwindturbines.

Engineers,JournalofSystemsandControlEngineering2007;

IEEETransl.onIndustrialElectronics2007;54(1).

221(3):497506.

Multon B, Robin G, Ruellan M, Ben Ahmed H. Situation

Bennouna O, Hraud N. Diagnosis and fault detection in

nergtique mondiale laube du 3me millnaire.

wind energy conversion system. Proceedings of the IEEE

Perspectives offertes par les ressources renouvelables.

International Conference on Environment and Electrical

3EI2004;36.

EngineeringEEEIC,May2011,Roma,Italy.

Odgaard PF, Stoustrup J. Unknown input observer based

Bertagnolio F, Rasmussen F, Srensen NN, Johansen J,

scheme for detecting faults in a wind turbine converter.

Madsen HA. A stochastic model for the simulation of

InProceedingsofthe7thIFACSymposiumonFaultDetection,

wind turbine blades in static stall. WindEnergy 2010; 13

302

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.01
SupervisionandSafetyofTechnicalProcesses2009:161166,

www.ijesci.org

http://dx.doi.org/10.1049/ietrpg.2009.0147

Barcelona,Spain.

Trujillo JJ, Bingl F, Larsen GC, Mann J, Khn M. Light

Patton RJ, Putra D, Klinkhieo S. Friction compensation as a

detectionandrangingmeasurementsofwakedynamics.

faulttolerantcontrolproblem.Int.J.Sys.Sci.2008;41(8):

Part II: twodimensional scanning. WindEnergy 2011; 14

9871001.

(1):6175.DOI:10.1002/we.402

Soliman M, Malik OP, Westwick DT. Multiple model

Zarchan P, Musoff H. Funfamentals of Kalman filetring: A

multipleinput multipleoutput predictive control for

practicalapproach,SecondEdition.AmercicanInstituteof

variable speed variable pitch wind energy conversion

AeronauticsandAstronautics,Inc.,2005.

systems. IET Renew. Power Gener.; 5 (2): 124136.

http://dx.doi.org/10.1049/ietrpg.2009.0137

Ouadie BENNOUNA received the Ph. D.

degree from the University of Corsica,
France,in2006,andtheDipl.Ing.degreein
mechanical engineering from the ENSAM
(Ecole Nationale Suprieure dArts &
Mtiers), in 2003. He joined the IRSEEM
(Institut de Recherche en Systmes
Electroniques EMbraqus) in February 2008. Since
September 2009, he has been an associate professor at the
ESIGELEC, Rouen, France. Currently, he is the head of
Mechatronics major at the same school. His research
interests include diagnosis and fault detection, signal
processing,andfaulttolerantcontrol.Themainapplications
concern aircraft control, and wind energy conversion
systems.

Srensen NN, Schreck S. Computation of the National

Renewable Energy Laboratory PhaseVI rotor in pitch
motion during standstill. Wind Energy 2011. DOI:
10.1002/we.480
TastuJ,PinsonP,KotwaE,MadsenH,NielsenHA.Spatio
temporal analysis and modeling of shortterm wind
power forecast errors. Wind Energy 2011; 14 (1) : 4360.
DOI:10.1002/we.401
Todeschini G, Emanuel AE. Wind energy conversion
systemsasactivefilters:designandcomparisonofthree
controlmethods.IETRenew.PowerGener.;4(4):341353.

303