Universit Paris Sud, 11

Laboratoire des Signaux et Systmes (CNRS-SUPELEC)

Dossier de Demande dInscription pour le Diplme dHabilitation Diriger les Recherches

Prsent par

Godpromesse KENNE
Charg de Cours, IUT FOTSO Victor, Universit de Dschang, Cameroun Contribution l'identification et commande adaptative des systmes non linaires avec paramtres variant dans le temps

1. Formulaire de demande dincription

2. Fiche synthtique

Universit Paris-Sud (11)

Fiche Synthtique de candidature lHDR

remplir en Times 10 Pt

NomKENNEPrnomGODPROMESSEAge : 44 ans Tlphone(237) 77 59 52 19e-mail : gokenne@yahoo.com Eventuelle(s) inscription(s) antrieure(s) lHDR (Universit, anne) : ............................................................................. ............................................................................................................................................................................................... SITUATION ACTUELLE Statut Charg de Cours (Organisme : Dpartement Gnie Electrique, IUT-FV Bandjoun, Universit de Dschang, Cameroun). ............................................................................................................ Date dentre en fonction : 28 Juin 1996 Nom et Adresse du laboratoire daccueil : Laboratoire des Signaux et Systmes (LSS), CNRS, SUPELEC, Plateau de Moulon, 91192 Gif-Sur-Yvette, France.................................................................................................................................... Master (ou quivalent : DEA, validation dacquis, ...) Date dobtention .Fvrier 1998... . Universit : Universit de Yaound (Cameroun) Responsable de stage FOGUE Mdard Adresse du laboratoire Laboratoire de Mcanique des Solides, Ecole Nationale Suprieure Polytechnique, B.P. 8390 Yaound, Cameroun ............................................................................................................................................................... Titre du Mmoire Voiles minces, Etude numrique par lments finis : Application aux rservoirs cylindriques sous chargement
axisymtrique

DIPLOME

DE

DOCTORAT D'UNIVERSITE

Date de soutenance 24 Novembre 2003. Universit : Universit Paris Sud (11) Directeur de thse : FRANCOISE LAMNABHI-LAGARRIGUE Nom et Adresse du laboratoire de thse : Laboratoire des Signaux et Systmes (LSS), CNRS, SUPELEC, Plateau de Moulon,
91192 Gif-Sur-Yvette, France

Production scientifique associe (titre, journal, etc.): ............................ .

1. Parameter estimation methodology for nonlinear systems: Application to induction motor, Journal of System Science and System Engineering, vol. 14, Issue 2, pp.240-254, 2005. Impact Factor: 0.364 2. Nonlinear system parameters estimation using radial basis function network, Control Engineering Practice, vol. 14, Issue 7, pp.819-832, 2006. Impact Factor: 1.406.

SEJOURS DE RECHERCHE

Dure totale des sjours de recherche (post-doc, sabbatique) et laboratoires concerns

- France
Septembre 2005 Fvrier 2006 (5 mois): Laboratoire des Signaux et Systmes LSS et Dpartement Energie, SUPELEC, GifSur-Yvette. Juin-Juillet 2006 (1 mois) : Laboratoire des Signaux et Systmes LSS et Dpartement Energie, SUPELEC, Gif-Sur-Yvette. Septembre 2008- Dcembre 2008 (4 mois): Laboratoire des Signaux et Systmes LSS et Dpartement Energie, SUPELEC, Gif-Sur-Yvette.

- Etranger ................................................................................................................................................................. Production scientifique associe (titre, journal, etc.):

1. Nonlinear systems parameters estimation using neural network: Application to synchronous machine, Journal of Mathematical and Computer Modelling of Dynamical Systems, Vol.13, Issue 4, pp.365-382, 2007. Impact Factor: 0.513. 2. Nonlinear systems time-varying parameter estimation: Application to induction motors, Electric Power Systems Research, Vol.78, pp.1881-1888, 2008. Impact Factor: 1.396. 3. Real-time speed and flux adaptive control of induction motors using unknown time-varying rotor resistance and load torque, IEEE Trans. On Energy Conversion, Vol. 24, pp.375-387, 2009. Impact Factor: 2.26. 4. Identification of nonlinear systems with time-varying parameters using a sliding-neural network observer, Neurocomputing, Vol. 72, pp.1611-1620, 2009. Impact Factor: 1.429. 5. Real-time transient stabilization and voltage regulation of power generators with unknown mechanical power input, Energy Conversion and Management, Vol. 51, pp.218-224, 2010. Impact Factor: 2.054. 6. An improved direct feedback linearization technique for transient stability enhancement and voltage regulation of power generators, Electrical Power and Energy Systems, Vol. 32, pp.809-816, 2010. Impact Factor: 2.073. 7. An online simplified rotor resistance estimator for induction motors, IEEE Trans. On Control Systems Technology, Vol. 18, pp.1188-1194, 2010. Impact Factor: 1.419. 8. An improved rotor resistance estimator for induction motors adaptive control, Electric Power Systems Research, Vol. 81, pp.930-941, 2011. Impact Factor: 1.396

TRAVAUX DE RECHERCHE Rsum de la thmatique actuelle (en 10 lignes maximum)

Nos recherches sont axes sur des problmes de recherche et dveloppement couvrant les domaines de l'Automatique des Systmes non Linaires avec des applications dans lidentification et la commande adaptative des systmes lectromcaniques et dans le domaine de lEnergie Electrique. Nous utilisons la technique des modes glissants et des rseaux de neurones, pour tudier la dynamique des systmes non linaires et dvelopper des mthodes d'identification des tats et paramtres variant dans le temps ainsi que des algorithmes de contrle adaptatif des systmes. Au terme de l'tude, nous mettons en oeuvre des outils d'analyse et d'estimation, de diagnostic et dtection des anomalies, et des outils de contrle adaptatif robustes par rapport aux variations des paramtres et aux incertitudes de modle pour des systmes lectromcaniques. Nous proposons aussi des solutions pour amliorer les performances transitoires des rseaux lectriques par la commande des gnrateurs synchrones et le pilotage des FACTS.

Production scientifique : Total Nombre de publications dans des revues avec comit de lecture Nombre de publications dans des actes de congrs avec comit de lecture : Nombre de livres ou de chapitres de livre Nombre de confrences invites dans des congrs internationaux Nombre de brevets
11 12 1 1 0

3 dernires annes
7 4 0 1 0

Principaux contrats de recherche obtenus

..... . ENCADREMENTS (nom / niveau / % encadrement / production scientifique associe) CHEUKEM Andr Doctorat -% Encadrement: 50 Production scientifique:-Transmission capacity enhancement and economic power generation with UPFC for the future power system of Cameroon, Journe Scientifique LAIA, 8 Janvier 2010, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. -Amlioration de la stabilit des rseaux lectriques par les dispositifs FACTS : Cas du rseau interconnect Sud du Cameroun, Journes Scientifiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. -An improved online nonlinear controller for power systems transient stability enhancement using SSSC Device, European Transactions on Electrical Power, under revision, Sep. 2011. Impact Factor : 0.371. Wissam Dib Doctorat -% Encadrement: 30 Production scientifique: An Application of immersion and invariance to transient stability and voltage regulation of power systems with unknown mechanical power, Proc. of Joint 48th IEEE CDC and 28th Chinese Control Conference, Shanghai, P.R. China, December 16-18, 2009. SIMO Rostand Sorel Master -% Encadrement: 100 Production scientifique: An online simplified rotor resistance estimator for induction motors, IEEE Trans. On Control Systems Technology, Vol. 18, pp.1188-1194, 2010. Impact Factor: 1.419. NGUIMFACK NDONGMO Jean de Dieu Master -% Encadrement: 100 Production scientifique: -Une stratgie de commande de l'UPFC pour l'amlioration des performances des rseaux lectriques, Journe Scientifique LAIA, 8 Janvier 2010, IUT FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. -Contrleur non linaire simplifi du SSSC pour l'amlioration de la stabilit transitoire des systmes de puissance, Journes Scientifiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. - An improved online nonlinear controller for power systems transient stability enhancement using SSSC Device, European Transactions on Electrical Power, under revision, Sep. 2011. Impact Factor : 0.371. DOUANLA Marc Rostand Master -% Encadrement: 100 Production scientifique: Etude comparative des rgulateurs de vitesse bass sur les modes glissants des moteurs synchrones aimants permanents, Journes Scientifiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. KUATE RENE FOCHIE Master -% Encadrement: 100 Production scientifique: Application de la mthode Watt-Var dcouple dans le contrle du rseau lectrique par STATCOM, Journes Scientifiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang.

AVIS DU CORRESPONDANT HDR (APRES CONSULTATION EVENTUELLE DUNE COMMISSION AD HOC)

Nom du correspondant HDR : ........................................................................................................................................... UFR dinscription : ............................................................................................................................................................. Nom des membres de lventuelle commission ad hoc : .................................................................................................. ............................................................................................................................................................................................... Avis motiv lintention du Conseil Scientifique ............................................................................................................ ............................................................................................................................................................................................... ............................................................................................................................................................................................... ............................................................................................................................................................................................... ............................................................................................................................................................................................... ............................................................................................................................................................................................... ............................................................................................................................................................................................... Date : Signature du correspondant HDR :

3. CV Dtaill

Table des matires

1.1 1.2 tat civil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Titres Universitaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Thse de Doctorat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Diplme dtudes Approfondies . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Mmoire de n dtudes dIngnieur de Conception . . . . . . . . . . . . . 1.2.4 Projet de n dtudes dIngnieur des Travaux . . . . . . . . . . . . . . . . Parcours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Postes occups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Sjours ltranger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Activits dEnseignements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Prsentation gnrale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Liste de mes enseignements . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.3 Liste de mes Cours numriss . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.4 Liste des innovations pdagogiques . . . . . . . . . . . . . . . . . . . . . . . Activits lies lAdministration . . . . . . . . . . . . . . . . . . . . . . . . . . . . Activits lies la Recherche . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.1 Objectif gnral et domaine couvert . . . . . . . . . . . . . . . . . . . . . . 1.6.2 Prix dj reu pour un article . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.3 Participation des comits, Editorial boards, organisation de colloques, seminaires, ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.4 Actions de valorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.5 Administration lie la Recherche . . . . . . . . . . . . . . . . . . . . . . . Encadrements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7.0.1 Thses de Doctorat Encadres . . . . . . . . . . . . . . . . . . . . 1.7.0.2 Thses de Doctorat en cours . . . . . . . . . . . . . . . . . . . . . 1.7.0.3 Mmoires Bac+5 soutenus . . . . . . . . . . . . . . . . . . . . . . 1.7.0.4 Mmoires Bac+5 en cours . . . . . . . . . . . . . . . . . . . . . . . 1.7.0.5 Mmoires Bac+3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . Synthse des travaux et description des principaux rsultats, projets et perspectives 1.8.1 Synthse des travaux et description des principaux rsultats . . . . . . . . . 1.8.2 Projets et perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Liste des Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9.1 Publications dans des revues internationales avec comit de lecture . . . . . 1.9.2 Chapitre de Livre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9.3 Articles dans des confrences internationales avec comit de lecture . . . . . 1.9.4 Articles dans des confrences nationales avec comit de lecture . . . . . . . 1.9.5 Article en cours de revision : . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9.6 Article soumis : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 1 1 2 2 2 2 2 3 3 4 4 4 4 5 5 6 6 6 6 7 7 7 8 8 9 10 10 11 13 13 14 14 15 16 16

1.3

1.4

1.5 1.6

1.7

1.8

1.9

Curriculum vitae dtaill

1.1 tat civil
N le 22 Novembre 1967 Balatchi Ag de 44 ans Mari 04 enfants Camerounais

Godpromesse KENNE Institut Universitaire de Technologie FOTSO Victor (IUT-FV) Universit de Dschang (UDS) BP 134 Bandjoun Cameroun Tel : +237 77 59 52 19 Email : gokenne@yahoo.com godpromesse@gmail.com

1.2
1.2.1

Titres Universitaires
Thse de Doctorat

Doctorat de lUniversit Paris XI Orsay - de 2000 2003 Spcialit : Automatique et traitement du signal - Soutenue le 24 Novembre 2003 Titre : Mthodes didentication pour des systmes non linaires avec paramtres variant dans le temps : Application aux machines tournantes induction Sous la Direction de Mme Franoise Lamnabhi-Lagarrigue, Directeur de Recherche, LSS, CNRS-Suplec

1.2.2

Diplme dtudes Approfondies

Diplme dtudes Approfondies de lUniversit de Yaound I - de 1996 1997 Ecole Nationale Suprieure Polytechnique (ENSP) Spcialit Mcanique des solides - Mmoire soutenu en Fvrier 1998 Titre : Voiles minces, Etude numrique par lments nis : Application aux rservoirs cylindriques sous chargement axisymtrique. Sous la Direction du Professeur FOGUE Mdard, Directeur du Laboratoire de Mcanique des Solides, Directeur de lIUT-FV de Bandjoun.

1.3. PARCOURS

1.2.3

Mmoire de n dtudes dIngnieur de Conception

Spcialit : Gnie Mcanique Universit de Yaound I - Ecole Nationale Suprieure Polytechnique - de 1991 1994 Mmoire soutenu en Juin 1994 Titre : Expertise et dimensionnement dune machine de briqueterie. Sous la Direction du Professeur FOGUE Mdard, Directeur du Laboratoire de Mcanique des Solides, Directeur de lIUT-FV de Bandjoun

1.2.4

Projet de n dtudes dIngnieur des Travaux

Spcialit : lectromcanique Universit de Yaound I - Ecole Nationale Suprieure Polytechnique - de 1988 1991 Projet soutenu en Juin 1991 Titre : Complments dtudes et implantation dune cellule exible. Sous la Direction de MM. JOUVENCEAU et TISSIER

1.3
1.3.1

Parcours
Postes occups
Depuis 2007 Depuis 2005 Depuis 2009 De 1996 2005 De 2005 2007

Chef de Dpartement de Gnie Electrique Charg de Cours lIUT-FV Indice 870 Assistant lIUT-FV Coordonnateur du Dpartement de Gnie Electrique (Ex Dpartements dElectrotechnique et dElectronique) Membre Fondateur, Coordonnateur Scientique et Directeur-Adjoint du Laboratoire LAIA (Automatique et Informatique Applique) Chef dquipe Automatique et Traitement du Signal au LAIA Coordonnateur par Intrim du Dpartement GTR (Tlcommunications et Rseaux) Coordonnateur par Intrim du Dpartement dlectronique Responsable du Laboratoire dEssais et Mesures lectriques Coordonnateur du Dpartement dlectrotechnique

Depuis 2006

Du 05/02/2001 au 02/04/2001 Du 25/04/2000 au 20/07/2000 De 1999 2002 De 1998 2005.

1.3.2

Sjours ltranger

Septembre 2005 Janvier 2006 : Jai sjourn au Laboratoire des signaux et systmes et Dpartement nergie de lcole Suprieure dlectricit (Suplec) sur la nalisation des articles commencs au Cameroun et la validation exprimentale des mthodes didentication et de commande adaptative du ux et de la vitesse dune machine asynchrone (MAS). Nous avons amlior les conditions de convergence de lobservateur neuronale base radiale propos dans les travaux de thse de Doctorat et nous avons commenc les travaux sur lidentication des systmes non linaires en combinant les rseaux de neurones base radiale et la technique des modes glissants. Juin-Juillet 2006 : Jai pass nouveau un autre sjour de recherche au Laboratoire des signaux et systmes (CNRSSuplec) et Dpartement nergie de lcole Suprieure dlectricit (Suplec) sur la validation exprimentale du contrleur adaptatif de ux et de la vitesse dune machine asynchrone (MAS). Les travaux de recherche ont commenc sur les commandes non linaires des gnrateurs synchrones

1.4. ACTIVITS DENSEIGNEMENTS

pour la stabilisation des rgimes transitoires des rseaux lectriques avec adaptation de puissance mcanique. Septembre 2008 Dcembre 2008 : Jai pass nouveau un autre sjour de recherche au Laboratoire des signaux et systmes (CNRSSuplec) et Dpartement nergie de lcole Suprieure dlectricit (Suplec). Au cours de ce sjour, les travaux ont continu sur les commandes non linaires des gnrateurs synchrones pour la stabilisation des rseaux lectriques avec adaptation de puissance mcanique et la validation exprimentale des algorithmes proposs. Jai particip galement lencadrement des travaux de Thse de Doctorat de DIB Wissam sur le thme " Contribution la Stabilit Transitoire des Systmes de Puissance Multi-Machine ".

1.4
1.4.1

Activits dEnseignements
Prsentation gnrale

Jai commenc mes activits denseignements en 1996. Jai acquis une exprience denseignements dans plusieurs disciplines et niveaux dtudes de lenseignement suprieur au Cameroun. Ainsi cette exprience a t acquise en tant que Assistant et Charg de Cours. Je dispense rgulirement les enseignements dans les cursus DUT, BTS et Licence de Technologie de lIUT FV de Bandjoun de lUniversit de Dschang (UDS), au niveau V options Electrotechnique et Electronique de lEcole Normale Suprieure de lEnseignement Technique (ENSET) de BAMBILI (Dpartement de Gnie Electrique) de lUniversit de BAMENDA et au cycle de Master de Recherche option Electronique du Dpartement de Physique de la Facult (FAC) des Sciences de lUDS. Les nombreux projets dirigs des tudiants (Bac+2 et Bac+3) mont permis damliorer signicativement mes enseignements en termes dexemples applicatifs et de nouvelles maquettes de travaux pratiques. En eet, les notions de correcteurs sont dicilement apprhendes par les tudiants des cycles DUT et BTS qui suivent les cours dAsservissements pour la premire fois dans leur cursus scolaire. Cest un enseignement qui a besoin des bases mathmatiques et qui a des applications dans tous les autres domaines scientiques. Ltude dune application de correction dun processus en temps rel permettrait aux tudiants de mieux apprhender ce cours. A dfaut dune application relle, une simulation de la commande dun processus avec des animations serait trs souhaitable compte tenu du fait que nos laboratoires ne sont pas susamment quips. Pour apporter un dbut de solution ce problme, jai dvelopp en 2006/2007, avec les tudiants de Licence de Technologie Parcours Gnie Electrique, un logiciel de simulation de la correction des systmes dynamiques en sappuyant sur le cas du pendule simple. Le projet COMETES (Coordination et Modernisation des Etablissements Technologiques de lEnseignement Suprieur au Cameroun) a appuy nancirement le dveloppement de cet outil pdagogique. Pour des raisons purement pdagogiques, jai choisi lexemple dun pendule simple oscillant en boucle ouverte et qui doit tre stabilis une position angulaire quelconque en boucle ferme travers un correcteur avance de phase (AP). En 2009, des outils pdagogiques de synthse des correcteurs dans le plan de Bode et dans le plan Complexe, ainsi quune plate-forme de stabilisation des pendules simple et inversible monts dans un chariot ont t dvelopps pour complter le projet de 2006/2007. En 2010, un module de variation de vitesse en temps rel dun moteur courant continu de puissance 1.5KW base de microcontrleur a t ralis dans le cadre des projets de n dtudes de Licence de Technologie en Gnie Electrique avec les tudiants HAMO TCHOUMDAM Blerio Decimus, LIENOU Salomon Gervasoni, MOUKALA Sbastien Briange. Ce module sera complt pour servir de maquette de Travaux Pratiques des tudiants de Niveaux BAC+2 et BAC+3. La majorit de mes cours est soutenue par des polycopis rassemblant toutes les notes de cours, tous les travaux dirigs et les rfrences bibliographiques an de respecter les exigences du systme

1.5. ACTIVITS LIES LADMINISTRATION LMD.

1.4.2

Liste de mes enseignements

-Asservissements Linaires (Cours de Niveau Bac+2 : 60H) -Electricit Gnrale (Cours de Niveau Bac+2 : 60H) -Travaux Pratiques dElectrotechnique (Cours de Niveau Bac+2 : 60H) -Machines Asynchrones (Cours de Niveau Bac+2 : 60H) -Commande Numrique (Cours de Niveau Bac+2 : 60H et Bac+3 : 60H) -Commande des Systmes Continus (Cours de Niveau Bac+3 : 60H) -Travaux Pratiques Commande des Systmes (Cours de Niveau Bac+3 : 60H) -Automatique (Cours de Niveau Bac+4 : 60H) -Rseaux de Neurones et Filtres Adaptatifs (Cours de Niveau Bac+5 : 60H) -Systmes Multivariables (Cours de Niveau Bac+5 : 60H).

1.4.3

Liste de mes Cours numriss

-Systmes Multivariables (Cours de Niveau Bac+5) -Rseaux de Neurones et Filtres Adaptatifs (Cours de Niveau Bac+5) -Commande des Systmes Continus (Cours de Niveau Bac+3) -Commande Numrique (Cours de Niveau Bac+2 et Bac+3). Les informations dtailles relatives ces cours numriss sont donnes lAnnexe 1.

1.4.4

Liste des innovations pdagogiques

-Logiciel multimdia de simulation et de rgulation des systmes dynamiques : Cas du pendule simple (Travaux Pratiques dAsservissements : Niveau Bac+2 et Bac+3) -Etude et ralisation dun hacheur pour variation de vitesse des moteurs courant continu de puissance nominale 1.5 kW (Travaux Pratiques dAsservissements : Niveau Bac+2 et Bac+3) -A virtual platform for automation practical : case of study of the simple and inverted pendulum mounted on a cart. Les informations dtailles relatives ces innovations pdagogiques sont donnes lAnnexe 2.

1.5

Activits lies lAdministration

En ma qualit de Chef de Dpartement de Gnie Electrique, jai pilot plusieurs programmes denseignements dans les cursus DUT, BTS et Licence de Technologie. Par ailleurs, je suis Prsident de la cellule technique LMD du Dpartement de Gnie Electrique, et Membre de la commission LMD de lIUT-FV Bandjoun. Jai coordonn lexpertise de plusieurs programmes denseignement en Licence Professionnelle et en Master Professionnel dans les lires des Institutions Prives dEnseignement Suprieur sous la tutelle acadmique de lUniversit de Dschang. Parmi ces lires, on peut citer : -Automatique et Informatique Industrielle, INSAM Douala -Gnie Electrique et Informatique Industrielle, ISTDI Douala -Gnie Electrique et Informatique Industrielle, ISS Yaound De plus, jai t Ingnieur de suivi de trois marchs nancs par le Budget dInvestissement Public (BIP) et le Projet PROACTP (Projet dAppui la Composante Technologique et Professionnelle) lUniversit de Dschang de 2008 2010 : -2008/2009 : March N 012/MUDS/CPM/2008 : Rhabilitation de lIUT-FV de Bandjoun par DIAGNOSIS MAINTENANCE & TRADE (DMT) SARL, BP 9171 Douala. -2009/2010 : Lettre commande N 005/LC/DC/CPM-UDS/2009 : Fourniture et Installation dun Groupe Electrogne lIUT-FV de Bandjoun par les Ets WINSOFT INFORMATIQUE B.P 296

1.6. ACTIVITS LIES LA RECHERCHE

Bafoussam. -2009-2010 : Lettre commande N 004/LC/DC/CPM-UDS/2009 : Fourniture et Installation des lampadaires en vue de lEclairage externe avec Energie Solaire (Phase I) par lEntreprise 2L TECHNOLOGIES SARL, BP 15164 Douala.

1.6
1.6.1

Activits lies la Recherche

Objectif gnral et domaine couvert

Nos recherches sont axes sur des problmes de recherche et dveloppement couvrant les domaines de lAutomatique des Systmes non Linaires avec des applications lidentication et commande adaptative des systmes lectromcaniques et rseaux lectriques. Ces travaux se droulent depuis 2006 au Laboratoire dAutomatique et dInformatique Applique (LAIA) en collaboration avec lUnit de Formation Doctorale Sciences Fondamentales et Technologiques, le Laboratoire des Signaux et Systmes (LSS) de lEcole Doctorale STIC de lUniversit Paris XI et le Dpartement Energie de lEcole Suprieure dElectricit, Campus de Gif, France. Au LAIA, janime les travaux de lEquipe Automatique et traitement du Signal (ATS). Nous travaillons sur un ensemble de projets focaliss sur deux axes : Axe 1 : Automatique des Systmes Non Linaires : Nous utilisons la technique des modes glissants et des rseaux de neurones, pour tudier la dynamique des systmes non linaires et dvelopper des mthodes didentication des tats et paramtres ainsi que des algorithmes de contrle adaptatif des systmes. La plupart des processus comportent des paramtres qui varient au cours de leur fonctionnement. Mais les commandes utilises dans lindustrie actuellement sont des commandes linaires bases sur lhypothse des paramtres constants. Il est donc impratif didentier en ligne les paramtres qui varient beaucoup pendant le fonctionnement du systme pour deux raisons principales : -Amliorer le rendement du processus travers le rendement de la commande qui exploite ce type de paramtre. Par consquent raliser des conomies de lnergie lectrique consomme par le processus. -Dtecter les anomalies dues la variation anormale dun tel paramtre et arrter le processus pour viter sa dtrioration. Au terme de ltude, nous mettons en oeuvre des outils danalyse et destimation, de diagnostic et dtection des anomalies, et des outils de contrle adaptatif robustes par rapport aux variations des paramtres et aux incertitudes de modle pour des systmes lectromcaniques. Mots Cls : Modlisation, Identication, Systmes non linaires, Diagnostic, Systmes lectromcaniques. Axe 2 : Automatique applique au domaine de lEnergie Electrique. Lore en nergie lectrique est trs insusante dans le monde et en particulier au Cameroun par rapport la demande en dpit du grand potentiel hydrolectrique du Cameroun. Ce dcit nergtique est d essentiellement linsusance des moyens de production et de transport, la vtust du rseau lectrique, laugmentation rapide de la demande nergtique (en moyenne 6% par an) cause de la croissance dmographique et de la reprise progressive de la croissance conomique, et les changements climatiques. Ce dcit se manifeste au quotidien par des coupures intempestives, des dlestages et des baisses de tensions. Linsusance des moyens de transport et la vtust du rseau lectrique ont conduit la dgradation des performances transitoires. Ainsi, on peut noter une frquence anormale des indisponibilits des lignes et des transformateurs. Les moyens classiques de contrle des rseaux (transformateurs prise rglable, transformateurs dphaseurs, etc), sont alors devenus insusants et lents pour rpondre ecacement aux perturbations et aux demandes en nergie de plus en plus croissantes. Le dveloppement rcent des dispositifs dlectronique de puissance appels FACTS ouvre de trs

1.6. ACTIVITS LIES LA RECHERCHE

bonnes perspectives en ce sens quils permettent daugmenter la capacit de transit des rseaux lectriques existants et laugmentation des marges de stabilit. A la n de nos investigations, nous proposons des solutions pour amliorer les performances transitoires, les moyens de transport, et nous proposons lutilisation de nouvelles formes de production dnergie lectrique (olienne, solaire). Mots Cls : nergie lectrique, FACTS, Performances, Nouvelles formes dnergie, Diagnostic, Commande Adaptative Robuste, Rseaux Electriques.

1.6.2

Prix dj reu pour un article

Jai reu une distinction honorique en tant que jeune chercheur nomin au prix de meilleure prsentation " at the Joint CTS-HYCON Workshop on Nonlinear and Hybrid Control, Paris 1012 July 2006 ". La copie de cette distinction honorique est prsente lAnnexe 3.

1.6.3

Participation des comits, Editorial boards, organisation de colloques, seminaires, ...

Je suis membre du comit scientique de lecture de plusieurs revues internationales et Confrences internationales avec comit de lecture. On peut citer entre autres : Electric Power Systems Research : Journal Website : www.elsevier.com/epsr, International Journal of Control tandf.co.uk/journals/tf/00207179.html, : Journal Home page :

International Journal of Energy Optimization and Engineering : Journal Website : www.irmainternational.org/journal/international-journal-energy-optimization-engineering/47187/ IEEE Trans. on Industrial Electronics, Special Issue of International Journal of Systems Science on Advances in Sliding Mode Observation and Estimation, 47th IEEE Conference on Decision and Control 2008, Cancun, Mexico. Jai t membre organisateur de trois (3) sminaires scientiques du LAIA (2007, 2010 et 2011).

1.6.4

Actions de valorisation

Les thmatiques abordes sont galement orientes vers le dveloppement. En eet, de nombreux projets de n dtudes dirigs ou co-dirigs ont permis la mise en place des modules ayant des applications sur le plan industriel avec pour corollaire lamlioration des conditions de vie des populations. Parmi ces projets dvelopps, on peut citer : -lAlimentation de secours pour llectrication en zones rurales -lExtension de la base de donnes et amlioration du module dasservissement de position dans une station complte dlectricit solaire -lAuto-excitation dune gnratrice asynchrone triphase vitesse variable et ralisation dun convertisseur " alternatif-continu " : cas dune application olienne. Les informations dtailles relatives aux actions de valorisation sont prsentes lAnnexe 4.

1.6.5

Administration lie la Recherche

Jai coordonn le montage du dossier douverture dun Laboratoire de recherche (dnomm : Laboratoire dAutomatique et dInformatique Applique-LAIA) lIUT-FV de Bandjoun. Lautorisation de fonctionnement de ce Laboratoire a t accorde par le Conseil Scientique de lUniversit en

1.7. ENCADREMENTS

2006. Ce laboratoire comporte actuellement trois quipes de recherche : -lquipe ATS (Automatique et Traitement du Signal), -lquipe PDER (Planication et Dveloppement des nergies Renouvelables) et -lquipe STIC (Sciences et Techniques de lInformation et de la Communication). Je suis Coordonnateur Scientique, Directeur Adjoint de ce Laboratoire et Chef de lEquipe Automatique et Traitement du Signal. Le laboratoire a dj organis trois sminaires nationaux (2007, 2010 et 2011).

1.7

Encadrements

A des degrs et niveaux divers, je me suis impliqu dans lencadrement des tudiants. Le nombre dencadrements volue de manire pyramidale. La base plus large, est consacre pour lessentiel lencadrement des tudiants des cursus DUT, BTS et Licence de Technologie (plus de cinquante projets de n dtudes dirigs). Le sommet plus rtrci concerne les encadrements de niveaux Bac+5. Ceci se justie par le fait que jenseigne dans un IUT qui a pour vocation la base de former des techniciens prs lemploi au bout de deux annes de formation. Nous avons compte tenu de la demande des industriels, commenc former au courant de lanne acadmique 2003/2004, les techniciens de niveau Bac+3 (Licence de Technologie) pour rpondre au manque de cadre moyens entre lingnieur et le technicien de niveau Bac+2. Nous avons aussi dvelopp des collaborations de recherche avec la Facult des Sciences de lUniversit de Dschang avec louverture du Master en lectronique au Dpartement de Physique. Tout rcemment nous avons aussi commenc appuyer le Dpartement de Gnie Electrique de lEcole Normale Suprieure de lEnseignement Technique de Bambili-Bamenda dans la direction des mmoires de n dtudes des futurs Professeurs de Lyce dEnseignement Technique. Pendant mes sjours au LSS et Dpartement Energie de Suplec, jai contribu lencadrement de certaines thses de Doctorat supervises par Mme Francoise Lamnabhi-Lagarrigue. 1.7.0.1 Thses de Doctorat Encadres

2006 2010 : Thse de Doctorat de CHEUKEM Andr - sujet : Contribution ltude de lapport des dispositifs FACTS lamlioration de la capacit de transport dun rseau lectrique. Thse co-encadre 50% avec John Mucho NGUNDAM, Professeur Emrite, cole Nationale Suprieure Polytechnique de Yaound, Universit de Yaound I. Thse soutenue en Juillet 2010 lEcole Nationale Suprieure Polytechnique, Universit de Yaound I. Situation actuelle : Charg de Cours au Dpartement de Gnie lectrique de lIUT FOTSO Victor, Universit de Dschang, Cameroun. Mes contributions relatives cette thse ont t publies dans ([1.9.5.1], [1.9.4.3], [1.9.4.4], [1.9.4.8], [1.9.4.9], [1.9.4.10]). 2006 2009 : Thse de Doctorat de Dib WISSAM - sujet : Contribution la Stabilit Transitoire des Systmes de Puissance Multi-Machine. Thse co-encadre 30% avec Mme Lamnabhi-Lagarrigue 10% et Romo ORTEGA 60%. Thse soutenue en Septembre 2009 au Laboratoire des Signaux et Systmes (LSS), Universit Paris Sud. Situation actuelle : Chercheur lInstitut Franais de Ptrole sur le contrle des machines lectriques. Mes contributions relatives cette thse ont t publies dans ([1.9.3.11]). 1.7.0.2 Thses de Doctorat en cours

2010 2011 : Thse de Doctorat en cours de NGUIMFACK NDONGMO Jean de Dieu - sujet : Contribution la stabilit transitoire des systmes de puissance par les disposi-

1.7. ENCADREMENTS

tifs FACTS. Thse dirige sous la supervision de Hilaire B. FOTSIN, Matre de Confrences, Dpartement de Physique, Facult des Sciences, Universit de Dschang, Cameroun. 1.7.0.3 Mmoires Bac+5 soutenus

2008 2009 : Mmoire de Master : SIMO Rostand Sorel - sujet : Mthodes didentication des tats et paramtres dune classe de systmes non linaires. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. Une partie des travaux mens dans le cadre de ce memoire a t publie dans la revue internationale IEEE Transactions on Control systems Technology [1.9.1.9]. 2009 2010 : Mmoire de Master : NGUIMFACK-NDONGMO Jean de Dieu sujet : Etude des dispositifs FACTS : Application lamlioration des performances du RIS. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. Une partie des travaux relatives cette thse ont fait lobjet de communications scientiques dans ([1.9.4.3], [1.9.4.9]). Actuellement un papier a t soumis la revue internationale European Transactions On Electrical Power [1.9.6.1] et ce manuscript est en cours de rvision. 2009 2010 : Mmoire de Master : KENMOGNE TCHUENKAM Guy Justin - sujet : Modlisation et simulation du fonctionnement en gnratrice dune machine asynchrone. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. 2010 2011 : Mmoire de Master de : DOUANLA Marc Rostand - sujet : Rgulation de vitesse dune machine synchrone aimants permanents avec adaptation en ligne de linductance statorique, du ux des aimants et de la charge mcanique. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. Une partie des travaux relatives cette thse a fait lobjet dune communication scientique [1.9.4.7]. 2010 2011 : Mmoire de Master de : KUATE RENE FOCHIE - sujet : Contrle des puissances ractives et tensions par le STATCOM dans le rseau lectrique. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. Une partie des travaux relatives cette thse a fait lobjet dune communication scientique [1.9.4.8]. 2010 2011 : Mmoire de n dtude ENSET BAMBILI de : DOUANGUIM DOUNANG Duclair, DEMANOU TAZING Appolinaire, et AKUMBOM MAKOKIYU Godlove - sujet : Etude et ralisation dun rgulateur de charge pour panneau solaire base de microcontrleur. Mmoire de n dEtudes des Professeurs dEnseignements Techniques, Ecole Normale dEnseignement Technique, Bambili, Universit de Bamenda, Cameroun.

1.7.0.4

Mmoires Bac+5 en cours

2011 2012 : Mmoire de Master de : FOTSO SIMO Armel - sujet : Contrle dune machine asynchrone pour conversion dnergie olienne avac adaptation de la constante de temps rotorique. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun.

1.7. ENCADREMENTS

2011 2012 : Mmoire de Master de : LIENOU DZUKOU Christian - sujet : Etude comparative des mthodes de stabilisation par FACTS des rseaux lectriques : Cas du rseau interconnect sud Cameroun. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. 2011 2012 : Mmoire de Master de : DJADJOU Thierry Landry - sujet : Mthode dynamique destimation des paramtres dune classe de systmes non linaires par combinaison des rseaux de neurones et de la logique oue. Mmoire de Master de Physique, option Electronique, Facult des Sciences, Universit de Dschang, Cameroun. 1.7.0.5 Mmoires Bac+3

Jai encadr une cinquantaine de projets de n dtude des tudiants du niveau Licence de Technologie en Gnie Electrique. Ces encadrements ont conduits la ralisation de plusieurs maquettes de travaux pratiques.

1.8. SYNTHSE DES TRAVAUX ET DESCRIPTION DES PRINCIPAUX RSULTATS, PROJETS ET PERSPECTIVES

10

1.8
1.8.1

Synthse des travaux et description des principaux rsultats, projets et perspectives

Synthse des travaux et description des principaux rsultats

Lensemble des thmes de recherches auxquels je me suis intress depuis mon travail de thse de Doctorat fait appel des mthodes destimation des systmes non linaires avec paramtres variant dans le temps et leurs applications en entrainements lectriques et nergies lectriques (identication et commande adaptive). Le grand challenge de mes travaux depuis la thse de Doctorat est de mettre en oeuvre des mthodes de commande adaptative robustes et simples implmenter en temps rel. Pendant mes travaux de thse de Doctorat, jai commenc travailler sur des mthodes didentication des paramtres variant dans le temps mais dans un contexte o le systme non linaire fonctionne en boucle ouverte. Nous avons propos deux mthodes destimation des paramtres variant dans le temps. La premire approche est base sur la thorie de la structure variable et la deuxime utilise les rseaux de neurone articiel base radiale. Les travaux mens pendant la thse de Doctorat ont concern beaucoup plus les mthodes statiques didentication des paramtres variant dans le temps ([1.9.1.1] et [1.9.1.2]). Par la suite, les conditions de convergence ont t amliores [1.9.1.3]. Aussi, lextension de cette approche des classes de systmes non linaires dont le vecteur dtat est partiellement connu ainsi que lamlioration des proprits de robustesse en utilisant un observateur neuronal glissant sont introduites [1.9.1.6]. Dans ce cas, Les rsultats de simulation montrent une forte robustesse par rapport aux bruits de mesure, une convergence assez rapide des tats et paramtres estims vers leurs valeurs nominales et une trs bonne prcision des estims. Le nombre trs faible des neurones de la couche cache (N = 4) et le fait que lapprentissage des poids est ralis en ligne avec un temps trs court permettent de conclure que lalgorithme est facilement implmentable en temps rel. Par ailleurs, compte-tenu des dicults relatives linversion des systmes par rapport aux paramtres pour les mthodes didentication proposes et prsentes ci-dessus, je me suis interess au dveloppement de lapproche par mode glissant base sur lestimation dynamique des paramtres variant dans le temps avec application en commande adaptative. Cest ainsi que : - Une mthode destimation des paramtres variant dans le temps qui peut tre applique sur une classe assez large de systmes non linaires a t propose [1.9.1.4]. Ladaptation en ligne des paramtres des commandes ainsi que la dtection et lestimation des anomalies sont quelques applications principales de la mthode propose. - Une autre mthode destimation de la rsistance rotorique dune machine asynchrone [1.9.1.9] a t propose en exploitant les travaux de [1.9.1.6]. Cette mthode est base sur les concepts de mode glissant. La convergence en temps ni de lestime de la rsistance vers sa valeur nominale est garantie lorsque les conditions assez faibles (facilement ralisables) de persistence dexcitation sont satisfaites. Les rsultats exprimentaux montrent une grande robustesse par rapport aux variations de la rsistance statorique (variation jusqu 100% de sa valeur nominale), aux incertitudes de modle, et bruits de mesure. - En entranements lectriques, les techniques didentication paramtriques prsentes ci-dessus ([1.9.1.6], [1.9.1.9]) et la technique des modes glisssants de 2nd ordre ont t exploites pour proposer un contrleur adaptatif de vitesse et de ux rotorique dune machine asynchrone [1.9.1.11]. Ce qui a permis damliorer les rsultats des travaux prcdents [1.9.1.5]. Les paramtres estims en ligne et utiliss dans le contrleur sont la rsistance rotorique et le couple de charge mcanique. Les rsultats obtenus montrent que la commande adaptative propose prsente de trs bonnes performances dans la poursuite des signaux de rfrence de la vitesse angulaire et du ux rotorique dun MAS lorsque la rsistance rotorique et le couple de charge ne sont pas connus. Cette commande a galement prsent de performances remarquables de dcouplage et de robustesse par rapport

1.8. SYNTHSE DES TRAVAUX ET DESCRIPTION DES PRINCIPAUX RSULTATS, PROJETS ET PERSPECTIVES

11

aux variations en ligne de la rsistance statorique (jusqu 100%) et de la rsistance rotorique (jusqu 87%), aux bruits de mesure, aux incertitudes paramtriques (inductances) et de modle. La comparaison des rsultats avec ceux obtenus avec la mthode vectorielle ux orient a dailleurs rvl que la mthode propose dcouple mieux la poursuite des signaux de rfrence de la vitesse et du ux (en particuler quand on varie lamplitude du ux). - Dans le domaine de lnergie lectrique, un contrleur adaptatif non linaire du module dexcitation, relativement facile mettre en oeuvre, a t propos pour amliorer les performances transitoires et rguler la tension dun gnrateur synchrone connect un rseau lectrique [1.9.1.7]. Les paramtres estims en ligne et utiliss dans le contrleur sont langle de puissance et la puissance mcanique. La mthode propose est base sur le modle standard de troisime ordre dun gnrateur synchrone et les informations de sortie sont les tats facilement mesurables (vitesse angulaire relative, puissance lectrique active, tension terminale du gnrateur et tension du bus inni). Les points de fonctionnement sont calculs en temps rel laide des informations fournies par les tats mesurables. La technique de commande quivalente des modes glissants est exploite pour synthtiser un contrleur adaptatif robuste et dcentralis, qui assure une convergence exponentielle des sorties vers celles dsires. La mthode propose est valide par les rsultats dimplantation en temps rel. La comparaison de ces rsultats avec ceux obtenus avec le contrleur bien connu AVR/PSS montre une nette supriorit de la mthode propose pour les types de perturbations suivantes : variation temporaire de limpdance de ligne, variation permanente de limpdance de ligne (y compris une simulation permanente de fonctionnement en regime dsquilibr) et la simulation dun dfaut temporaire de la turbine dentrainement. - Une extension de la mthode classique DFL (Direct Feedback Linearization Technique) la rgulation de la tension dun gnrateur synchrone connect un rseau lectrique, a t galement propose [1.9.1.8]. La commande propose est galement adaptative puisque langle de puissance et la puissance mcanique sont supposs non disponibles dans llaboration du contrleur. La mthode propose est base sur le modle standard de troisime ordre dun gnrateur synchrone et les informations de sortie sont les tats facilement mesurables (vitesse angulaire relative, puissance lectrique active, tension terminale du gnrateur et tension du bus inni). Les rsultats dimplantation en temps rel des montrent que les oscillations de vitesse relative sont vites ammorties, et que la tension terminale est bien rgule en prsence des perturbations svres.

1.8.2

Projets et perspectives

Des pistes sont donnes dans le domaine de lnergie lectrique (production et transport, systme hybride, biomasse), tlmdecine (diagnostic et dtection des patients malades par examen des ECG). Les perspectives de mes activits de recherche peuvent tre classes suivant plusieurs plans et orientations. En terme dorientation des thmes de recherche, je vais continuer developper les mthodes didentication pour des commandes adaptatives des systmes non linaires et embrasser de nouveaux domaines dapplication. Le premier grand domaine dans lequel les recherches vont tre intensies est celui de lnergie lectrique en terme de production et transport. Au niveau de la production, on sintressera particulirement la modlisation et commande des nouvelles formes de production dnergie lectrique (olienne, solaire, biomasse). Au niveau du transport, il sera question doptimiser la capacit des lignes de transport existantes par lutilisation des FACTS et leurs pilotages/commandes. En eet, le problme dnergie lectrique est dactualit un peu partout dans le monde et en particulier au Cameroun o nous connaisons rgulirement les coupures dlectricit et les dlestages parceque la socit charge de la production, transport et distribution de lnergie Electrique au Cameroun (AES-Sonel), a toujours dnormes dicults satisfaire la demande nationale. tant donn la faible couverture des zones rurales de notre pays en lectricit, un accent sera mis sur laccs llectricit des populations rurales, faible pouvoir dachat, grce ltude, analyse et la mise en oeuvre des systmes dalimentation hybrides qui intgrent des gnrateurs dnergies renouvelables

1.8. SYNTHSE DES TRAVAUX ET DESCRIPTION DES PRINCIPAUX RSULTATS, PROJETS ET PERSPECTIVES

12

(solaires, oliens, pico hydrauliques) et des gnrateurs diesel. Tous ces aspects seront bien en accord avec les objectifs de lutte contre la pauvret dans notre pays. Une premire tude [1.9.1.10] a dailleurs permis de conclure quune rduction de 30% sur les taxes dimportation des plaques photovoltaiques permettrait dalimenter partir dun mini-rseau certaines localits du Nord Cameroun o lensoleillement minimal annuel est de 5.55kW h/m2 /jour avec un cot nergtique plus faible que lutilisation dun groupe lectrogne. Dans la modlisation et commande des nouvelles formes de production dnergie lectrique, les travaux ont commenc avec le Master de Monsieur KENMOGNE TCHUENKAM Guy Justin sur la Modlisation et simulation du fonctionnement en gnratrice dune machine asynchrone. Le deuxime grand domaine concerne les problmes de diagnostic, dtection/estimation et isolation des anomalies dans les processus industriels. Ce thme trs important dans la thmatique didentication des systmes non linaires bien nest pas encore bien developp. Il mrite des attentions particulires dans nos prochaines investigations. Un autre aspect de notre recherche sera dassocier les techniques dintelligence articielle aux mthodes de traitement des signaux physiologiques (thories des ondelettes, algorithmes gntiques, etc.) en vue de mettre au point des outils daide aux diagnostics automatiques, utilisables par des mdecins gnralistes an de combler partiellement la carence des spcialistes. Dans un tel contexte, lintroduction des nouvelles techniques de linformation et de la communication (NTIC) dans les consultations et des examens mdicaux spcialiss (tlmdecine) ou le dveloppement des systmes dinstruments de mesure adapts lenvironnement tropical seront au centre de nos proccupations. Cette approche a commenc avec les travaux de Thse en cours de Monsieur MBOUPDA PONE Justin Roger et le premier travail porte sur "Golay coding for penetration depth improvement in ultrasonography" [1.9.6.1]. Sur le plan de la validation exprimentale des direntes contributions thoriques qui seront developpes, en attendant la mise en place progressive des bancs dessais au LAIA, nous comptons continuer collaborer avec le Laboratoire LSS et le Dpartement nergie de Suplec pour limplantation en temps rel de : - Lalgorithme de rgulation de vitesse dune machine synchrone aimants permanents avec adaptation en ligne de linductance statorique, du ux des aimants, de la charge mcanique et du moment dinertie (perspectives du Mmoire de Master de Monsieur DOUANLA Marc Rostand). Lobservateur neuronal glissant [1.9.1.6]. Les axes de recherche qui peuvent faire lobjet dautres de thmatiques de stage de Master ou de Thse de Doctorat sont : - Lapplication de la mthode utilise dans la commande adaptative du MAS en ux et vitesse la commande en ux, vitesse et couple. Lavantage principal de cette dernire approche est la rgulation simultane du ux, de la vitesse et du couple. - Lapplication des techniques utilises dans [1.9.1.7] et [1.9.1.11] pour developper des commandes adaptatives des machines asynchrones fonctionnant en gnratrices automomes ou connectes un rseau lectrique. Cette tude a commenc avec la thse de Master en cours de FOTSO SIMO Armel sur le thme "Contrle dune machine asynchrone pour conversion dnergie olienne avec adaptation de la constante de temps rotorique" et la thse de Doctorat en cours de Mamadou Mamby KEITA. - Dveloppement des commandes adaptatives des systmes non linaires en combinant les modes glissants et les rseaux de neurones. Le travail prliminaire ce sujet peut tre lextension de lobservateur neuronal glissant [1.9.1.6] des classes de systmes non linaires dont la fonction inverse par rapport aux paramtres nest pas analytiquement connue. Il sagira donc de dvelopper une mthode dynamique destimation des paramtres base dobservateur neuronal glissant. Nous avons dailleurs commenc avec les travaux de Master de DJADJOU Thierry Landry. - Dveloppement des commandes des dispositifs de puissance FACTS pour loptimisation du transit de puissance dans les rseaux lectriques par lamlioration des performances transitoires. Cest dailleurs lobjet de la thse de Doctorat de Monsieur NGUIMFACK Jean De Dieu dont le thme est "Contribution la stabilit transitoire des systmes de puissance par les dispositifs FACTS".

1.9. LISTE DES PUBLICATIONS

13

1.9
1.9.1

Liste des Publications

Publications dans des revues internationales avec comit de lecture

[1.9.1.1] G. Kenn, F. Floret, F. Lamnabhi-Lagarrigue and H. Nkwawo, Parameter estimation methodology for nonlinear systems : Application to induction motor, in Journal of System Science and System Engineering, Vol. 14, Issue 2, pp. 240-254, June 2005. Impact Factor : 0.364 [1.9.1.2] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue and H. Nkwawo, Nonlinear system parameters estimation using radial basis function network, in Control Engineering Practice, Vol. 14, Issue 7, pp. 819-832, July 2006. Impact Factor : 1.406. [1.9.1.3] T. Ahmed-Ali, G. Kenn, and F. Lamnabhi-Lagarrigue, Nonlinear systems parameters estimation using neural network : Application to synchronous machine, in Journal of Mathematical and Computer Modelling of Dynamical Systems, Vol.13, Issue 4, pp. 365-382, Aug. 2007. Impact Factor : 0.513. [1.9.1.4] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Nonlinear systems time-varying parameter estimation : Application to induction motors, in Electric Power Systems Research, Vol.78, pp. 1881-1888, Nov. 2008. Impact Factor : 1.396. [1.9.1.5] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Real-time speed and ux adaptive control of induction motors using unknown time-varying rotor resistance and load torque, IEEE Trans. On Energy Conversion, Vol. 24, pp. 375-387, 2009. Impact Factor : 2.26. [1.9.1.6] T. Ahmed-Ali, G. Kenn, and F. Lamnabhi-Lagarrigue, Identication of nonlinear systems with time-varying parameters using a sliding-neural network observer, Neurocomputing, Vol. 72, pp. 1611-1620, 2009. Impact Factor : 1.429. [1.9.1.7] G. Kenn, R. Goma, H. Nkwawo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, Real-time transient stabilization and voltage regulation of power generators with unknown mechanical power input, Energy Conversion and Management, Vol. 51, pp. 218-224, 2010. Impact Factor : 2.054. [1.9.1.8] G. Kenn, R. Goma, H. Nkwawo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An improved direct feedback linearization technique for transient stability enhancement and voltage regulation of power generators, Electrical Power and Energy Systems, Vol. 32, pp. 809-816, 2010. Impact Factor : 2.073. [1.9.1.9] G. Kenn, R. S. Simo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An online simplied rotor resistance estimator for induction motors, IEEE Trans. On Control Systems Technology, Vol. 18, pp. 1188-1194, 2010. Impact Factor : 1.419. [1.9.1.10] E. Mbaka Nfah, J. Mucho Ngundam, G. Kenn, Economic evaluation of small-scale photovoltaic hybrid systems for mini-grid applications in far north Cameroon, Renewable Energy, Vol.35 , pp. 2391-2398, 2010. Impact Factor : 2.554. [1.9.1.11] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An improved rotor resistance estimator for induction motors adaptive control, Electric Power Systems Research, Vol. 81, pp. 930-941, 2011. Impact Factor : 1.396.

1.9. LISTE DES PUBLICATIONS

14

1.9.2

Chapitre de Livre

[1.9.2.1] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Real-time implementation of rotor ux and speed control of induction motors using on-line rotor resistance and load torque adaptation, in Book of the selected papers of the joint CTS-HYCON Workshop, Paris, France, July 2006, Ed. by ISTE Publishing Knowledge, ISBN : 978 1 905209 65 1, pp. 341-362, 2007.

1.9.3

Articles dans des confrences internationales avec comit de lecture

[1.9.3.1] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue and H. Nkwawo, Nonlinear system parameters identication using radial basis function neuronal predictor, in Proc. of the joint Int. Conf. on Articial Neural Networks and Neural Information ProcessingICANN/ICONIP2003, Istanbul, Turkey, pp. 49-53, June 2003. [1.9.3.2] G. Kenn, T. Ahmed-Ali, H. Nkwawo and F. Lamnabhi-Lagarrigue, Identication of time-varying rotor and stator resistances of induction motor, in Proc. of the European Control Conference-ECC03, Cambridge University, UK, paper number 594, Sep. 2003. [1.9.3.3] G. Kenn, T. Ahmed-Ali, H. Nkwawo et F. Lamnabhi-Lagarrigue, Algorithmes didentication des paramtres lectriques dun moteur asynchrone, Acte des Journes Doctorales dAutomatique-JDA03, Universit de Valencienne, France, pp. 203-208, Juin 2003. [1.9.3.4] G. Kenn, F. Floret, H. Nkwawo and F. Lamnabhi-Lagarrigue, Real-Time electrical parameters and rotor ux estimation of induction motors using sliding mode observer approach, in Proc. of the 16th Int. Conf. On Syst. Engineering, ICSE 2003, Conventry University, UK, Vol. 1, pp. 355-361, Sep.2003. [1.9.3.5] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and H. Nkwawo, Identication of electrical parameters and rotor speed of induction motor using radial basis neuronal network, in Proc. of IEEE International Symposium on Industry Electronics-ISIE04, Ajaccio France, pp. 483-490, Mai 2004. [1.9.3.6] G. Kenn, T. Ahmed-Ali, H. Nkwawo and F. Lamnabhi-Lagarrigue, Robust rotor ux and speed control of induction motors using on-line time-varying rotor resistance adaptation, in Proc. of the 44th IEEE Conference on Decision and Control, and European Control Conference, CDC-ECC05, Seville, Spain, pp. 7768-7774, Dec. 2005. [1.9.3.7] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Time-varying parameter identication of a class of nonlinear systems with application to online rotor resistance estimation of induction motors, in Proc. of IEEE International Symposium on Industry Electronics-ISIE 2006, Montral, Qubec, Canada, vol.1, pp. 301-306, July 2006. [1.9.3.8] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Real-time implementation of rotor ux and speed control of induction motors using on-line rotor resistance and load torque adaptation, in Joint CTS-HYCON Workshop, Paris, France, July 2006. [1.9.3.9] Godpromesse Kenn, Homre Nkwawo, Raphal Goma, Tarek Ahmed-Ali, Franoise Lamnabhi-Lagarrigue, and Jean-Claude Vannier, Transient stabilization and voltage regulation of power generators with parameters adaptation, in Proc. of 15th IFAC Symposium on System Identication, SYSID 2009, July 6-8, Saint-Malo, France, Vol. 15, Part 1, paper identier : 10.3182/20090706-3-FR-2004.00177.

1.9. LISTE DES PUBLICATIONS

15

[1.9.3.10] Godpromesse Kenn, Tarek Ahmed-Ali, Franoise Lamnabhi-Lagarrigue, Amir Arzande, and Jean-Claude Vannier, An improved rotor resistance estimator for induction motors, in Proc. of 15th IFAC Symposium on System Identication, SYSID 2009, July 6-8, Saint-Malo, France, Vol. 15, Part 1, paper identier : 10.3182/20090706-3-FR-2004.00050. [1.9.3.11] Wissam Dib, Godpromesse Kenn, Franoise Lamnabhi-Lagarrigue, An Application of immersion and invariance to transient stability and voltage regulation of power systems with unknown mechanical power, in Proc. of Joint 48th IEEE Conference on Decision and Control and 28th Chinese Control Conference, Shanghai, P.R. China, December 16-18, 2009. [1.9.3.12] E. M. Nfah, J. M. Ngundam, G. Kenn, PV hybrid technology and economics for remote electrication in Far North Cameroon, in Proc. of the International Workshop on Water Supply and Renewable Energy Systems and management, National Advanced School of Engineering, University of Yaounde I, Cameroon, 27-29 April 2009, pp. 75-85. [1.9.3.13] Godpromesse Kenn, Franoise Lamnabhi-Lagarrigue, An online rotor resistance estimator for induction machine adaptive control, Special session paper to be presented in the IEEE International Conference on Industrial Technology, Island of Kos Greece, March 19-21, 2012.

1.9.4

Articles dans des confrences nationales avec comit de lecture

[1.9.4.1] G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Contribution to nonlinear systems time-varying parameter identication and adaptive control : Sliding mode observer approach, in Acte du sminaire LAIA sur la Modlisation et le Traitement du Signal, IUT FOTSO Victor de Bandjoun, Universit de Dschang, Cameroun, 15-16 Juin 2007, pp. 1-10. [1.9.4.2] G. Kenn, R. Goma, H. Nkwawo, F. Lamnabhi-Lagarrigue, A. Arzand, and J. C. Vannier, Transient stabilization and voltage regulation of power generator using direct feedback linearization technique, Journe Scientique LAIA, 8 Janvier 2010, IUT FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.3] J. D. Nguimfack, G. Kenn, A. Cheukem, J. M. Ngundam, Une stratgie de commande de lUPFC pour lamlioration des performances des rseaux lectriques, Journe Scientique LAIA, 8 Janvier 2010, IUT FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.4] A. Cheukem, G. Kenn, J. M. Ngundam, Transmission capacity enhancement and economic power generation with UPFC for the future power system of Cameroon, Journe Scientique LAIA, 8 Janvier 2010, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.5] J. R. Mboupda Pone, M. Kom, A. Tiedeu, G. Kenn, Sur lexploration de limage chographique des organes internes par la mthode des missions codes complmentaires Golay (ECCG), in Proc. of Cameroon Biosciences Society (CBS), 17th Annual Conference Nov. 30 -Dec. 04, 2010, Universit des Montagnes, Bangangt, Cameroon, www.udesmontagnes.org. [1.9.4.6] J. R. Mboupda Pone, A. Tiedeu, M. Kom, G. Kenn, Modeling improvement of probes capabilities using Golay coding, Journes Scientiques LAIA, 15-16 Juin 2011, IUTFOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.7] R. M. Douanla, G. Kenn, H. B. Fotsin, Etude comparative des rgulateurs de vitesse bass sur les modes glissants des moteurs synchrones aimants permanents, Journes

1.9. LISTE DES PUBLICATIONS

16

Scientiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.8] R. F. Kuat, G. Kenn, A. Cheukem, H. B. Fotsin, Application de la mthode WattVar dcouple dans le contrle du rseau lectrique par STATCOM, Journes Scientiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.9] J. D. D. Nguimfack, G. Kenn, A. Cheukem, H. B. Fotsin, Contrleur non linaire simpli du SSSC pour lamlioration de la stabilit transitoire des systmes de puissance, Journes Scientiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun. [1.9.4.10] A. Cheukem, J.M. Ngundam, G. Kenn, Amlioration de la stabilit des rseaux lectriques par les dispositifs FACTS : Cas du rseau interconnect Sud du Cameroun, Journes Scientiques LAIA, 15-16 Juin 2011, IUT-FOTSO Victor, Bandjoun, Universit de Dschang, Cameroun.

1.9.5

Article en cours de revision :

[1.9.5.1] J. D. D. Nguimfack, G. Kenn, A. Cheukem, H. B. Fotsin, and F. LamnabhiLagarrigue, An improved online nonlinear controller for power systems transient stability enhancement using SSSC Device, draft manuscript submitted to European Transactions on Electrical Power, under revision, Sep. 2011. Impact Factor : 0.371.

1.9.6

Article soumis :

[1.9.6.1]. J. R. Mboupda Pone, M. Kom, A. Tiedeu, G. Kenn, Golay coding for penetration depth improvement in ultrasonography, draft manuscript submitted to Biomedical Signal Processing and Control, Sep. 2011.

ANNEXE 1: Cours Numeriss

UNIVERSITY OF BAMENDA -----------HIGHER TECHNICAL TEACHERS TRAINAING COLLEGE (H.T.T.T.C.) BAMBILI-BAMENDA

UNIVERSITE DE BAMENDA -----------ECOLE NORMALE SUPERIEURE DE LENSEIGNEMENT TECHNIQUE DE BAMBILI-BAMENDA

-----------DEPARTMENT OF ELECTRICAL AND POWER ENGINEERING

-----------DEPARTEMENT DE GENIE ELECTRIQUE

------------

------------

u(t) Entes ou commandes

Systme dynamique

y(t) Sorties ou mesures

Support de
Cours de systmes multivariables
Niveau V : Options Electrotechnique & Electronique Anne Acadmique 2010/2011 Par Godpromesse KENNE Charg de Cours

SYSTEMES

MULTI

VARIABLES,

NIVEAU

V,

OPTIONS

ELECTROTECHNIQUE

&

ELECTRONIQUE, DEPARTEMENT DE GENIE ELECTRIQUE, ENSET BAMBILI-BAMENDA, UNIVERSITE DE BAMENDA

Objectifs pdagogiques :
Dans ce cours, nous prsentons dans la premire partie des notions de base permettant daborder ltude des systmes linaires multi variables asservir en utilisant la reprsentation dtat. Cest en fait une reprsentation interne du systme par opposition la reprsentation externe par fonctions de transfert des systmes linaires monovariables gnralement prsente dans les cours dasservissements linaires continus et en temps discret. Nous supposerons que le systme asservir peut tre dcrit par un systme dquations diffrentielles linaires. Dans ce cas la reprsentation dtat utilise quatre matrices (se rduisant dans la plupart de cas 3) qui sont constantes pour des systmes temps invariant. Cette criture matricielle permet de dvelopper un formalisme puissant permettant par son caractre gnral de traiter de faon efficace des systmes trs divers. Le formalisme prsent utilise des rsultats classiques dalgbre linaire et permettant daborder la synthse dune commande partir de la reprsentation dtat. A la fin de la premire partie, ltudiant doit tre capable de reprsenter les diffrentes formes canoniques (pour lobservation et la commande), de vrifier la commandabilit et lobservabilit des systmes, et de synthtiser une commande par retour dtat ou commande modale. Dans la deuxime partie, nous prsentons des notions de base permettant daborder ltude des systmes linaires multivariables asservir en utilisant lapproche commande optimale LQR (Linear Quadratic Regulator). Cest en fait une commande qui sappuie sur le modle dtat dun systme pour trouver un retour dtat ou de sortie stabilisant, optimal au sens du compromis rapidit-nergie. A la fin de la deuxime partie, ltudiant doit tre capable de synthtiser une commande optimale LQ par application de la mthodologie dcrite dans le cours pour la plupart des processus simples supposs linaires. Pour des dveloppements plus approfondis, nous conseillons au lecteur les documents donns dans la section Rfrences bibliographiques .

Table des matires

PREMIERE PARTIE : ANALYSE ET COMMANDE DES SYSTEMES PAR REPRESENTATION DETAT .............. 2 CHAPITRE 1: RAPPELS SUR LES SYSTEMES .................................................................................................................... 2 1.1 REPRESENTATION INTERNE DES SYSTEMES DETERMINISTES: NOTION DETAT ...................................................................... 2 1.2 EQUATIONS DETAT ET DE MESURE POUR LES SYSTEMES CONTINUS ..................................................................................... 3 1.3 CAS DES SYSTEMES LINEAIRES ............................................................................................................................................. 4 CHAPITRE 2:RESOLUTION DE LEQUATION DETAT ................................................................................................... 5 2.1 REGIME LIBRE ...................................................................................................................................................................... 5 2.2 REGIME FORCE ..................................................................................................................................................................... 5 2.3 CAS PARTICULIER DES SYSTEMES LINEAIRES ET INVARIANTS ............................................................................................... 6 2.4 METHODE DE CALCUL DE LA MATRICE DE TRANSITION e A(t t0 ) DANS LE CAS PARTICULIER DES SYSTEMES LINEAIRES ET INVARIANTS ............................................................................................................................................................................... 6 2.4.1 Utilisation du dveloppement en srie ......................................................................................................................... 6 2.4.2 Formule de Sylvester: valeurs propres de A distinctes ................................................................................................ 7 2.4.3 Mthode dinterpolation de Sylvester .......................................................................................................................... 7 2.4.4 Mthode des modes ...................................................................................................................................................... 8

2.5: MODELE DETAT DES SYSTEMES DISCRETS LINEAIRES ......................................................................................................... 9 2.6. RAPPEL SUR LA STABILITE DES SYSTEMES LINEAIRES ........................................................................................................ 10 CHAPITRE 3: REPRESENTATION FREQUENTIELLE DES SYSTEMES LINEAIRES ET INVARIANTS ............. 12 3.1 PASSAGE ETAT - TRANSFERT............................................................................................................................................... 12 3.2. CHANGEMENT DE BASE DANS LESPACE DETAT................................................................................................................ 15 3.3 PASSAGE TRANSFERT ETAT .............................................................................................................................................. 16 3.3.1 Forme compagnon pour la commande ...................................................................................................................... 16 3.3.2 Forme compagnon pour lobservation ...................................................................................................................... 18 CHAPITRE 4: COMMANDABILITE OU GOUVERNABILITE DES SYSTEMES ......................................................... 21 4.1 DEFINITIONS GENERALES ................................................................................................................................................... 21 4.2 GOUVERNABILITE DES SYSTEMES CONTINUS LINEAIRES ET INVARIANTS ............................................................................ 21 4.3 GOUVERNABILITE DES SYSTEMES DISCRETS LINEAIRES ET INVARIANTS ............................................................................. 22 4.4 REPRESENTATION DES SYSTEMES PARTIELLEMENT COMMANDABLES................................................................................. 24 4.4.1 Structure canonique ................................................................................................................................................... 24 4.4.2 Reprsentation frquentielle ...................................................................................................................................... 28 CHAPITRE 5: OBSERVABILITE DES SYSTEMES ............................................................................................................ 30 5.1 DEFINITIONS ....................................................................................................................................................................... 30 5.2 OBSERVABILITE DES SYSTEMES LINEAIRES ET INVARIANTS ................................................................................................ 30 5.3 OBSERVABILITE DES SYSTEMES DISCRETS LINEAIRES ET INVARIANTS (CAS MONOVARIABLE) ............................................ 32 5.4 CAS DES SYSTEMES PARTIELLEMENT OBSERVABLES .......................................................................................................... 33 CHAPITRE 6: COMMANDE MODALE DES SYSTEMES MONOVARIABLES INVARIANTS .................................. 34 6.1 DEFINITION......................................................................................................................................................................... 34 6.2 CAS DES SYSTEMES MONOVARIABLES INVARIANTS TOTALEMENT COMMANDABLES : ........................................................ 34 6.3 COMMANDE MODALE DES SYSTEMES PARTIELLEMENT COMMANDABLES: .......................................................................... 37 6.4 CONCLUSION ...................................................................................................................................................................... 38 DEUXIEME PARTIE : INTRODUCTION A LA COMMANDE OPTIMALE .................................................................. 39 CHAPITRE 7 INTRODUCTION A LA COMMANDE OPTIMALE LQ ........................................................................ 39

7.1 INTRODUCTION ............................................................................................................................................................... 39 7.2 CONFIGURATION EN BOUCLE FERMEE DU PROBLEME DE COMMANDE LQ ....................................................................... 39 7.3 CRITERE DOPTIMALITE LQ ............................................................................................................................................ 40 7.3.1 Vitesse de rejet de perturbation ................................................................................................................................. 40 7.3.2 Energie de commande................................................................................................................................................ 41 7.3.3 Critre de compromis ................................................................................................................................................ 41 7.4 GESTION DES OBJECTIFS ET SPECIFICATION DES MATRICES DE PONDERATION .................................................................... 41 7.5 SOLUTION DU PROBLEME LQ STATIONNAIRE...................................................................................................................... 43 7.6 MATRICE HAMILTONIENNE ET SOLUTION DE LEQUATION DE RICCATI .............................................................................. 46 7.7 STABILITE DE LA BOUCLE FERMEE ...................................................................................................................................... 47 7.8 EXERCICES DAPPLICATION : .............................................................................................................................................. 48 CHAPITRE 8 8.1 8.2 8.3 8.4 8.5 INTRODUCTION A LA COMMANDE LINEAIRE QUADRATIQUE GAUSSIENNE ....................... 50

PROBLEME DE COMMANDE STOCHASTIQUE A RETOUR DE SORTIE ................................................................................... 50 OBSERVATEUR DETAT ET PRINCIPE DE SEPARATION ...................................................................................................... 50 OBSERVATEUR OPTIMAL DE KALMAN ............................................................................................................................ 53 SOLUTION DU PROBLEME LQG ....................................................................................................................................... 56 ANNEXE .......................................................................................................................................................................... 58

AUTRES EXERCICES ............................................................................................................................................................. 58 REFERENCES UTILISEES ..................................................................................................................................................... 61 [ALBSAL04] P. ALBERTOS PEREZ, A. SALA. MULTIVARIABLE CONTROL SYSTEMS (ADVANCED TEXTBOOKS IN CONTROL AND SIGNAL PROCESSING), SPRINGER, 2004. ............................................................................................................................ 61 [KHALIL96] H. KHALIL. NONLINEAR SYSTEMS. PRENTICE-HALL, 2ND EDITION, 1996. ....................................................... 61 J. J. DAZZO, C. H. HOUPIS. LINEAR CONTROL SYSTEM ANALYSIS & DESIGN (CONVENTIONAL AND [AZZHOU88] RD MODERN), MCGRAW-H ILL INC., 3 EDITION, 1988. ................................................................................................................ 61 OKKO H. BOSGRA, HUIBERT KWAKERNAAK, GJERRIT MEINSMA. DESIGN METHODS FOR CONTROL [OKKBOS08] SYSTEMS, NOTES FOR A COURSE OF THE DUTCH INSTITUTE OF SYSTEMS AND CONTROL WINTER TERM 20072008. ............. 61 B.D.O. ANDERSON, J.B. MOORE. LINEAR OPTIMAL CONTROL, PRENTICE-HALL, 1990. .............................. 61 [ANDMOO90] H. KWAKERNAAK, R. SIVAN. LINEAR OPTIMAL CONTROL SYSTEMS, WILEY INTERSCIENCE, 1972. ............. 61 [KWASIV72]

[KAILATH80] T. KAILATH. LINEAR SYSTEMS, PRENTICE-HALL, 1980. .............................................................................. 61 B. FRIEDLAND. CONTROL SYSTEM DESIGN, MC GRAW-HILL, 1986. ....................................................... 62 [FRIEDLAND86.] ASTROM-WITTENMARK. COMPUTER CONTROLLED SYSTEMS, ENGLEWOOD CLIFFS, NJ, PRENTICE-HALL, [ASTROM84] 1984. 62

UNIVERSITE DE DSCHANG FACULTE DES SCIENCES DEPARTEMENT DE PHYSIQUE

Support de
Cours de Rseaux de Neurones et Filtres Adaptatifs

Master II Physique Option Electronique

Anne Acadmique 2010/2011 Par Godpromesse KENNE Charg de Cours

RESEAUX DE NEURONES ET FILTRES ADAPTATIFS, MASTER II PHYSIQUE, OPTION ELECTRONIQUE, DEPARTEMENT DE PHYSIQUE, FACULTE DES SCIENCES, UNIVERSITE DE DSCHANG

Objectifs pdagogiques :
Ce cours prsente des notions de base permettant de dmystifier le concept de rseau de neurones artificiels. Cest ainsi que des notions darchitecture et dapprentissage des rseaux de neurones artificiels sont abordes et agrmentes par des exemples illustratifs. A la fin de ce cours ltudiant doit tre capable daborder la mise en uvre dun rseau de neurone artificiel pour la rsolution de quelques problmes non linaires simples. Ltudiant dsirant approfondir ses connaissances en rseaux de neurones artificiels peut se reporter aux documents fournis dans la partie Rfrences Bibliographiques.

Table des matires

CHAPITRE 1 GENERALITES SUR LES RESEAUX DE NEURONES ARTIFICIELS ........................................... 2 1.1 INTRODUCTION ET HISTORIQUE ................................................................................................................................ 2 1.1.1 Introduction .................................................................................................................................................... 2 1.1.2 Historique ........................................................................................................................................................ 3 1.2 STRUCTURE DUN NEURONE BIOLOGIQUE ET SON FONCTIONNEMENT ..................................................................... 3 1.2.1 Structure d'un neurone .............................................................................................................................. 3 1.2. 2 Fonctionnement ........................................................................................................................................... 5 1.3. NEURONE FORMEL ...................................................................................................................................................... 6 1.3.1 Prsentation .................................................................................................................................................... 6 1.3.2. Interprtation mathmatique ................................................................................................................. 6 1.3.3 Fonctions d'activation:................................................................................................................................ 7 1.4. ARCHITECTURE DUN RESEAU DE NEURONE ............................................................................................................. 7 1.4.1. Prsentation .................................................................................................................................................. 7 1.4.2 Architecture .................................................................................................................................................... 8 1.5 RECAPITULATION ....................................................................................................................................................... 10 1.6 APPLICATIONS ........................................................................................................................................................... 10 1.7 CONCLUSION ............................................................................................................................................................. 10 CHAPITRE 2 ALGORITHMES DAPPRENTISSAGES DES RESEAUX DE NEURONES ARTIFICIELS ....... 12 2.1 GENERALITES SUR LAPPRENTISSAGE DES RNA : COMMENT FONCTIONNENT LES RNA ? ................................ 12 2.1.1 Principe de lapprentissage des RNA ................................................................................................... 12 2.1.2 Types dapprentissages ............................................................................................................................ 12 ELEMENTS DUN ALGORITHME DAPPRENTISSAGE ............................................................................................ 13 2.2 2.2.1 Un ensemble dapprentissage ................................................................................................................ 13 2.2.2 Une fonction de cot. ................................................................................................................................ 14 2.3 Un ensemble de validation .......................................................................................................................... 16 2.4 CLASSIFICATION/MODELISATION ............................................................................................................................ 17 QUELQUES EXEMPLES DAPPRENTISSAGE DES RNA ........................................................................................ 18 2.5 2.5.1 Rseau de Hopfield .................................................................................................................................... 18 2.5.2 Classifieur 2 classes ............................................................................................................................... 20 2.5.3 Le calcul des moindres carrs ................................................................................................................ 21 2.5.4 La rgle Delta ............................................................................................................................................... 21 2.5.5 Algorithme du Perceptron (Rosenblat) ............................................................................................... 22 2.5.6 Perceptron multicouches (Rumlhart, Lecun, 1986) ....................................................................... 24 2.5.7 Droulement de lalgorithme :............................................................................................................... 27 2.5.8 Minima de la fonction cot ...................................................................................................................... 28 QUELQUES NOTIONS SUR LE RESEAU A BASE RADIALE.................................................................................... 28 2.6

2.6.1 Prsentation .................................................................................................................................................. 28 2.6.2 Apprentissage des rseaux de neurones RBF .................................................................................. 30 CHAPITRE 3 PRINCIPE DE MISE EN UVRE ET VALIDATION DES RESEAUX DE NEURONES ARTIFICIELS 33 3.1 PRINCIPE DE MISE EN UVRE .................................................................................................................................. 33 3.1.1 Constitution dune base de donnes ................................................................................................... 33 3.1.2 Pr-traitement des donnes dentres ............................................................................................... 33 3.1.3 Choix des entres pertinentes (descripteurs).................................................................................. 33 3.1.4 Choix de larchitecture du rseau ......................................................................................................... 33 3.1.5 Phase dapprentissage .............................................................................................................................. 34 3.1.6 Phase de gnralisation ........................................................................................................................... 34 3.2 VALIDATION DUN RESEAU DE NEURONES ARTIFICIELS ......................................................................................... 34 3.2.1 Critres de performance : EQMA et EQMT ........................................................................................ 34 3.2.2 Sur-apprentissage ...................................................................................................................................... 35 3.2.3 Compromis biais-variance ....................................................................................................................... 36 3.2.4 Notion de Early Stopping .................................................................................................................... 37 3.2.5 Validation croise (Cross-Validation) .............................................................................................. 37 3.2.6 Mthode du Leave-One-Out (LOO) ............................................................................................... 39 3.2.7 Mthode du Leave-One-Out Virtuel (LOOV) .............................................................................. 39 EXERCICES ......................................................................................................................................................................... 42 TP PERCEPTRON ............................................................................................................................................................... 43 PROJET RESEAUX DE NEURONES MLP ..................................................................................................................... 44 RFRENCES BIBLIOGRAPHIQUES............................................................................................................................ 46 [1] ALVAREZ J. AN INTERNAL MODEL CONTROLLER FOR NONLINEAR SYSTEMS, PROCEEDINGS OF THE EUROPEAN CONTROL CONFERENCE, ROME, SEPT. 1995, PP. 301-306. .................................................................................... 46 [2] AKAKE H. A NEW LOOK AT THE STATISTICAL MODEL IDENTIFICATION, IEEE TRANS. ON AUTOMATIC CONTROL, N19, 1974, PP. 716-723. ...................................................................................................................... 46 [3] BISHOP C. REGULARIZATION AND COMPLEXITY CONTROL IN FEED FORWARD NETWORK, PROCEEDINGS OF THE ICANN 95, PARIS, OCT. 1995, PP. 141-148. ......................................................................................................... 46 [4] BISHOP C., NEURAL NETWORK FOR PATTERN RECOGNITION, 3RD EDITION, CLARENDON PRESS, OXFORD 1997. ................................................................................................................................................................................ 46 [5] BILINGS S. A., CHEN S., KORENBERG M. IDENTIFICATION OF MIMO NONLINEAR SYSTEMS USING A FORWARD REGRESSION ORTHOGONAL ESTIMATOR, INT. J. CONTROL, VOL. 49, N5, 1989, PP. 2157-2189. 46 [6] CHEN S., BILINGS S. A. ORTHOGONAL LEAST SQUARE METHODS AND THEIR APPLICATION TO NONLINEAR SYSTEMS IDENTIFICATION, INT. J. CONTROL, VOL. 50, N5, 1989, PP. 1873-1896. ........................................ 46 [7] CHEN S., BILINGS S. A., GRANT P. M. NONLINEAR SYSTEM IDENTIFICATION USING NEURAL NETWORKS, INT. J. CONTROL, VOL. 51, N6, 1990, PP. 1191-1214. ...................................................................................... 46 [8] DREYFUS G., MARTINEZ J. M. RESEAUX DE NEURONES: METHODOLOGIE ET APPLICATIONS, COLLECTION ALGORITHMES, EYROLLES EDITION, 2002. .................................................................................................................. 46 [9] HORNIK K., STINCHCOMBE M., WHITE H., AUER P. DEGREE OF APPROXIMATION RESULTS FOR FEEDFORWARD NETWORKS APPROXIMATING UNKNOWING MAPPING AND THEIR DERIVATIVES, NEURAL COMPUTATION, VOL. 6, 1994, PP. 1262-1275. ....................................................................................................... 46 [10] HORNIK K., STINCHCOMBE M., WHITE H. MULTILAYER FEED FORWARD NETWORKS ARE UNIVERSAL APPROXIMATES, NEURAL NETWORKS, VOL. 2, 1989, PP. 359-366. ....................................................................... 46 [11] HUNT K., SBARBARO D. NEURAL NETWORKS FOR NONLINEAR INTERNAL MODEL CONTROL, IEE PROCEEDINGS, VOL. 138, 1991, N 5, PP. 431-438. ............................................................................................ 46 [12] IDAN Y, DREYFUS D. THE CANONICAL FORM OF NONLINEAR DISCRETE-TIME MODELS, NEURAL COMPUTATION, VOL. 10, N1. ...................................................................................................................................... 46 [13] LECUN Y., DENKER J., SOLLA S. OPTIMAL BRAIN DAMAGE, NIPS, 1990, TOURETSZY ED. MORGAN KAUFMANN, PP. 598-605. ............................................................................................................................................. 46 [14] LEVIN A.U., NARENDRA K. CONTROL OF NONLINEAR DYNAMICAL SYSTEMS USING NEURAL NETWORKS: CONTROLLABILITY AND STABILIZATION, IEEE TRANS. ON NEURAL NETWORKS, VOL. 4, N2, PP. 1011-1020. ........................................................................................................................................................................................... 46 [15] LJUNG L. SYSTEMS IDENTIFICATION: THEORY FOR THE USER, PRENTICE HALL, 1987.................................. 46 [16] MONARI G. SELECTION DE MODELES NON LINEAIRES PAR LEAVE-ONE-OUT, THESE DE DOCTORAT DE LUNIVERSITE DE PARIS VI, 1999. ............................................................................................................................... 46

[18] MORARI M., ZAFIROU E. ROBUST PROCESS CONTROL, PRENTICE HALL, INTERNATIONAL EDITIONS, 1989. ........................................................................................................................................................................................... 47 [19] NARENDRA K., PARTHSARATHY K. IDENTIFICATION AND CONTROL OF DYNAMICAL SYSTEMS USING NEURAL NETWORKS, IEEE TRANS. ON NEURAL NETWORKS, VOL. 1, N1, PP. 4-27. ......................................................... 47 [20] NERRAND O. RESEAUX DE NEURONES POUR LE FILTRAGE ADAPTATIF, LIDENTIFICATION ET LA COMMANDE DES PROCESSUS, THESE DE DOCTORAT DE LUNIVERSITE DE PARIS VI, 1993. ...................................................... 47 [21] NERRAND O., ROUSSEL-RAGOT P., PERSONNAZ L., DREYFUS G. NONLINEAR RECURSIVE IDENTIFICATION AND CONTROL BY NEURAL NETWORKS : A GENERAL FRAMEWORK, PROCEEDINGS OF THE EUROPEAN CONTROL CONFERENCE, 1993, GRNINGEN................................................................................................................................. 47 [22] NERRAND O., PERSONNAZ L., DREYFUS G. NEURAL NETWORKS AND NONLINEAR ADAPTIVE FILTRING: UNIFIED CONCEPTS AND NEW ALGORITHMS, NEURAL COMPUTATION, VOL. 5, N2, 1993, PP. 165-199. ......... 47 [23] OUSSAR Y. TRAINING WAVELET NETWORKS FOR NONLINEAR DYNAMIC INPUT-OUTPUT MODELING, NEUROCOMPUTING, VOL. 20, 1989, PP. 173-188. .................................................................................................. 47 [24] OUSSAR Y., DREYFUS D. HOW TO BE A GRAY BOX: DYNAMIC SEMI-PHYSICAL MODELING, NEURAL NETWORKS, N14, 2001, PP. 1161-1172. .............................................................................................................. 47 [25] PLOIS J. I., DREYFUS D., GORRIOU J. P., PASCAL D. FROM KNOWLEDGE BASED MODEL TO RECURRENT NETWORKS: AN APPLICATION TO AN INDUSTRIAL DISTILLATION PROCESS, NEURONIMES 94. ................................ 47 [26] RENDERS J. METAPHORES BIOLOGIQUES APPLIQUEES A LA COMMANDE DES PROCESSUS, THESE DE DOCTORAT DE LUNIVERSITE LIBRE DE BRUXELLES, 1994. ........................................................................................ 47 [27] RIVALS I., CANAS D., PERSONNAZ L., DREYFUS D. MODELING AND CONTROL OF MOBILE ROBOTS AND INTELLIGENT VEHICLES BY NEURAL NETWORKS, IEEE CONFERENCE ON INTELLIGENT VEHICLES, PARIS OCT. 1994, PP.137-142. ....................................................................................................................................................... 47 [28] RIVALS I. MODELISATION ET COMMANDE DE PROCESSUS PAR RESEAUX DE NEURONES : APPLICATION AU PILOTAGE DUN VEHICULE AUTONOME, THESE DE DOCTORAT DE LUNIVERSITE PARIS VI, 1995. ......................... 47 [29] RIVALS I., PERSONNAZ L. INTERNAL MODEL CONTROL USING NEURAL NETWORKS, PROCEEDINGS OF THE IEEE CONFERENCE ON INT. SYMPOSIUM ON INDUSTRIAL ELECTRONICS, VARSOVIE JUNE 1996. ........................ 47 [30] SANNER R., SLOTINE J. J. E. STABLE ADAPTIVE CONTROL OF ROBOT MANIPULATORS USING NEURAL NETWORKS, NEURAL COMPUTATION, VOL. 7, N4, 1995, PP. 753-790. .............................................................. 47 [31] SJBERG J., ZHANG Q., LJUNG L., BENVENISTE A., DELYON B. NONLINEAR BLACK-BOX MODELING IN SYSTEM IDENTIFICATION: A UNIFIED OVERVIEWS, AUTOMATICA, VOL. 31, N12, 1995, PP. 1691-1724. ...... 47 [32] SONTAG E.D. NEURAL NETWORKS FOR CONTROL, IN ESSAYS ON CONTROL: PERSPECTIVES IN THE THEORY AND ITS APPLICATION, TRENTELMAN HL AND WILLEMS J. C. ED., BIRKHASER, BOSTON, 1993, PP. 339-380. ........................................................................................................................................................................................... 47 [33] STOPPIGLIA H. METHODES STATISTIQUES DE SELECTION DARCHITECTURES NEURONALES : APPLICATIONS FINANCIERES ET BANCAIRES, THESE DE DOCTORAT DE LUNIVERSITE PARIS VI, 1998. ........................................ 47 [34] STRICKER M., ET AL. TRAINING CONTEXT SENSITIVE NEURAL NETWORKS WITH FEW RELEVANT EXAMPLES FOR THE TREC-9 ROUTINE, PROCEEDINGS OF THE 9-TH TEXT RETRIEVERAL CONFERENCE (TREC-9), NIST SPECIAL PUBLICATION, 2001. ........................................................................................................................................ 48 [35] URBANI D. MTHODES STATISTIQUES DE SLECTION DARCHITECTURES NEURONALES : APPLICATIONS LA CONCEPTION DE MODELES DE PROCESSUS DYNAMIQUES, THESE DE DOCTORAT DE LUNIVERSITE PARIS VI, 1995. ................................................................................................................................................................................ 48 [36] URBANI D., ROUSSEL-RAGOT, P., PERSONNAZ L., DREYFUS G. THE SELECTION OF NEURAL MODELS OF NONLINEAR DYNAMICAL SYSTEMS BY STATISTICAL TESTS, PROCEEDINGS OF THE IEEE WORKSHOP ON NEURAL NETWORKS FOR SIGNAL PROCESSING, 1994, PP. 229-237. .................................................................................... 48

[17] MONARI G., DREYFUS D. WITHDRAWING AN EXAMPLE FROM THE TRAINING SET: AN ANALYTIC ESTIMATION OF ITS EFFECT ON A NONLINEAR PARAMETRIZED MODEL, NEUROCOMPUTING, VOL. 35, 2000, PP. 195-201..... 46

UNIVERSITE DE DSCHANG INSTITUT UNIVERSITAIRE DE TECHNOLOGIE FOTSO VICTOR DE BANDJOUN DEPARTEMENT DE GENIE ELECTRIQUE

E(p)

(p)

C(p)

U(p)

S(p)

G(p)

Support de
Cours de commande des systmes continus
Licence de Technologie en Gnie Electrique Anne Acadmique 2010/2011 Par Godpromesse KENNE Charg de Cours

COMMANDE DES SYSTEMES CONTINUS, LICENCE GE, DEPARTEMENT DE GENIE ELECTRIQUE, IUT-FV BANDJOUN, UNIVERSITE DE DSCHANG

Objectifs pdagogiques :
Le cours comporte sept chapitres. Dans le premier chapitre, nous prsentons des notions de base permettant daborder la correction des systmes continus en exploitant les reprsentations graphiques dans le plan de Bode (Amplitude et Phase). Dans chaque type de correction srie prsente, une mthode quasi-systmatique est dcrite. A la fin de ce cours ltudiant doit tre capable de calculer un correcteur analogique type P, PI, PD ou PID partir de la mthode classique propose dans ce chapitre en sappuyant sur les diagrammes de Bode. Dans le deuxime chapitre, on prsente deux mthodes empiriques de synthse du rgulateur PID lorsquune approximation des paramtres du rgulateur est suffisante pour certaines applications industrielles. Au terme de ce chapitre, ltudiant doit tre capable de synthse un correcteur PID laide des mthodes empiriques tudies. Le troisime chapitre prsente une mthode de synthse directe des correcteurs. Le quatrime chapitre est consacr la description dune mthode graphique qui a lavantage de permettre la fois ltude de la stabilit et la synthse dun correcteur une fois que le lieu dEvans ou lieu des racines est trac. Au terme de cette partie ltudiant doit tre capable de tracer le lieu dEvans en suivant la mthodologie dcrite. Le cinquime chapitre prsente une mthode de synthse des correcteurs base sur le lieu dEvans tudi au chapitre prcdent. Elle permet donc de dduire partir du lieu dEvans les paramtres du correcteur utiliser pour obtenir un comportement dsir et dfini dans un cahier des charges. A la fin de ce chapitre, ltudiant doit tre capable de synthtiser un correcteur PID en exploitant la mthodologie du plan complexe dcrite et base sur le lieu dEvans. Le sixime chapitre prsente des mthodes qui permettent didentifier un processus partir des modles mathmatiques afin dutiliser les modles identifis pour laborer les commandes automatiques. A la fin ce chapitre, ltudiant doit tre capable de rsoudre un problme didentification des systmes linaires temps continu par les mthodes classiques dcrites. Enfin, le dernier chapitre est consacr aux notions de base permettant daborder ltude des systmes linaires asservir en utilisant la reprsentation dtat. Cest en fait une reprsentation interne du systme par opposition la reprsentation externe par fonctions de transfert des systmes linaires monovariables gnralement prsente dans les cours dasservissements linaires. Nous supposerons que le systme asservir peut tre dcrit par un systme dquations diffrentielles linaires. Dans ce cas la reprsentation dtat utilise quatre matrices (se rduisant dans la plupart de cas 3) qui sont constantes pour des systmes temps invariant. Cette criture matricielle permet de dvelopper un formalisme puissant permettant par son caractre gnral de traiter de faon efficace des systmes trs divers. Le formalisme prsent utilise des rsultats classiques dalgbre linaire et permettant daborder la synthse dune commande partir de la reprsentation dtat. Pour des dveloppements plus approfondis, nous conseillons au lecteur les ouvrages [AndMoo90, KwaSiv72, Kail80, Frie86]. A la fin de ce dernier chapitre, ltudiant doit tre capable de reprsenter les diffrentes formes canoniques (pour lobservation et la commande), de vrifier la commandabilit et lobservabilit des systmes, et de synthtiser une commande par retour dtat.

Table des matires

CHAPITRE 1 SYNTHESE DES CORRECTEURS DANS LE PLAN DE BODE........................................................ 2

1.1 CAHIER DES CHARGES ................................................................................................................................................. 2 1.2 MODES DE CORRECTION PAR ACTION PURE ................................................................................................................. 2 Correction par action proportionnelle.............................................................................................................. 2 1.2.1 Correction par action proportionnelle et intgrale .......................................................................................... 4 2.2.2 Correction par action proportionnelle et drive ............................................................................................. 6 2.2.3 Correction par action proportionnelle, intgrale et drive ............................................................................. 7 2.2.4 MODES DE CORRECTION PAR ACTION APPROCHEE .................................................................................................. 9 2.3. Correction par action sur le dphasage : avance de phase .............................................................................. 9 2.3.1 Correction par action sur le gain dynamique : retard de phase ..................................................................... 14 2.3.2 Correction par action mixte avance-retard de phase ..................................................................................... 17 2.3.3 CHAPITRE 2 METHODES EMPIRIQUES DE SYNTHESE DU PID ...................................................................... 20

2.1 INTRODUCTION ............................................................................................................................................................. 20 2.2 METHODE DE ZIEGLER-NICHOLS ............................................................................................................................ 20 2.2.1 Essai en boucle ouverte ........................................................................................................................................ 20 2.2.1 Essai en boucle ferme ......................................................................................................................................... 20 METHODE DE CHIEN-HRONES-RESWICK ...................................................................................................... 21 2.3. CHAPITRE 3 METHODE DE SYNTHESE DIRECTE DUN CORRECTEUR ANALOGIQUE .......................... 22

3.1 INTRODUCTION ............................................................................................................................................................ 22 3.2 SYNTHESE DU CORRECTEUR ......................................................................................................................................... 22 3.3 CHOIX DU POLYNOME DE REFERENCE .......................................................................................................................... 23 3.3.1 Comportement souhait quivalent celui dun premier ordre ........................................................................... 23 3.3.2 Comportement souhait quivalent celui dun 2nd ordre ................................................................................ 23 3.3.3 Comportement oscillant amorti ........................................................................................................................... 23 3.3.4 Comportement sans oscillation ........................................................................................................................... 23 3.4 EXEMPLE DE CAS DE SYSTEME ..................................................................................................................................... 23 3.4.1 Systme du 1er ordre ........................................................................................................................................... 23 3.4.2 Systme du 2er ordre ............................................................................................................................................ 24 3.4.3 Systme avec retard............................................................................................................................................. 24 CHAPITRE 4 ETUDE DES SYSTEMES ASSERVIS PAR LA METHODE DES LIEUX DEVANS .................... 25

4.1 DEFINITION DU LIEU DEVANS .................................................................................................................................. 25 4.2 PROPRIETES DES LIEUX DES RACINES ........................................................................................................................ 27 4.2.1 Premire proprit : conditions des angles et des modules ................................................................................. 27 4.2.2 Deuxime proprit : nombre de branches ......................................................................................................... 28 4.2.3 Troisime proprit : symtrie des lieux .............................................................................................................. 28 4.2.4 Quatrime proprit : asymptotes ........................................................................................................................ 28 4.2.5 Cinquime proprit : Lieux sur l'axe rel ........................................................................................................... 29 4.2.6 Sixime proprit : angles d'arrive ou de dpart des ples ou zros complexes ................................................ 33 4.2.7 Septime proprit : intersection avec l'axe imaginaire ...................................................................................... 33 4.3 SYNTHESE DE LA METHODE POUR TRACER LE LIEU DEVANS .................................................................................... 34 CHAPITRE 5 SYNTHESE DES CORRECTEURS DANS LE PLAN COMPLEXE ................................................ 36

5.1 INTRODUCTION.......................................................................................................................................................... 36 5.2 CAHIER DES CHARGES ............................................................................................................................................... 36 5.3 INFLUENCE DES POLES ET DES ZEROS ADDITIONNELS ................................................................................................ 36 5.3 SYSTEMES A POLES DOMINANTS ................................................................................................................................ 37 5.4 MODES DE CORRECTIONS PAR ACTION PURE ............................................................................................................. 38 5.4.1 Correction par action proportionnelle ................................................................................................................. 38 5.4.2 Correction par action proportionnelle et intgrale.............................................................................................. 39 5.4.3 Correction par action proportionnelle et drive ................................................................................................ 40 5.4.4 Correction par action proportionnelle, intgrale et drive ................................................................................ 41 5.5 MODES DE CORRECTIONS PAR ACTION APPROCHEE ....................................................................................................... 43 5.5.1 Correction par avance de phase (AP) .................................................................................................................. 44 5.5.2 Correction par retard de phase (RP) ................................................................................................................... 45 5.5.3 Correction par avance- retard de phase (ARP) ................................................................................................... 47 CHAPITRE 6 METHODES DIDENTIFICATION DES SYSTEMES LINEAIRES ................................................ 51

6.1 INTRODUCTION.......................................................................................................................................................... 51 6.2 MODELE DE CONNAISSANCE ......................................................................................................................................... 51 6.3 MODELE DE CONDUITE.................................................................................................................................................. 52 Simplification du modle ............................................................................................................................................... 52

6.4 IDENTIFICATION PAR ANALYSE INDICIELLE ................................................................................................................... 52 6.4.1 Exprimentation ................................................................................................................................................... 52 6.4.2 Processus avec intgration (ple p=0).............................................................................................................. 53 6.4.3 Rponse des systmes de premier ordre et identification ..................................................................................... 54 6.4.4. Rponse des systmes du second ordre et identification ..................................................................................... 56 6.5 Rponses indicielles de modles mathmatiques..................................................................................................... 62 6.6 ANALYSE FREQUENTIELLE ............................................................................................................................................ 65 6.6.1 Principe ................................................................................................................................................................ 65 6.6.2 Analyse harmonique ............................................................................................................................................. 65 6.6.3 Mise en vidence d'un retard T ............................................................................................................................ 66 6.6.4 Ordre du systme.................................................................................................................................................. 67 6.6.5 Ples p = 0 {intgration} .................................................................................................................................. 68 6.6.6 Processus une constante de temps..................................................................................................................... 69 6.6.7 Processus deux constantes de temps ................................................................................................................. 69 6.7 METHODE DES POLES ET ZEROS .................................................................................................................................... 70 CHAPITRE 7 ANALYSE ET COMMANDE DES SYSTEMES PAR REPRESENTATION DETAT ................... 72

7.1 RAPPELS SUR LES SYSTEMES ......................................................................................................................................... 72 7.2 REPRESENTATION INTERNE DES SYSTEMES DETERMINISTES : NOTION DETAT .............................................................. 72 7.3 EQUATION DETAT ET DE MESURE POUR LES SYSTEMES CONTINUS ............................................................................... 73 7.4 CAS DES SYSTEMES LINEAIRES ...................................................................................................................................... 73 7.5 RESOLUTION DE LEQUATION DETAT ........................................................................................................................... 74 7.5.1 Rgime libre ......................................................................................................................................................... 74 7.5.2 Rgime forc......................................................................................................................................................... 74 7.5.3 Cas particulier des systmes linaires et invariants ............................................................................................ 75 7.5.4 Mthode de calcul de la matrice de transition e A(t t0 ) dans le cas particulier des systmes linaires et invariants ...................................................................................................................................................................... 75 7.6 Application la modlisation des systmes discrets linaires ................................................................................ 78 7.7 REPRESENTATION FREQUENTIELLE DES SYSTEMES LINEAIRES ET INVARIANTS ......................................................... 79 7.7.1 Passage tat - transfert ........................................................................................................................................ 79 7.7.2 Passage transfert tat ........................................................................................................................................ 80 7.8 COMMANDABILITE OU GOUVERNABILITE DES SYSTEMES.......................................................................................... 83 7.8.1 Dfinitions ............................................................................................................................................................ 83 7.8.2 Gouvernabilit des systmes linaires et invariants............................................................................................. 83 7.9 COMMANDE MODALE DES SYSTEMES MONOVARIABLES INVARIANTS TOTALEMENT COMMANDABLES.......................... 84 7.10 OBSERVABILITE DES SYSTEMES .................................................................................................................................. 87 7.10.1 Dfinitions .......................................................................................................................................................... 87 7.10.2 Observabilit des systmes linaires et invariants ............................................................................................. 87 TRAVAUX DIRIGES ............................................................................................................................................................. 89 REFERENCES UTILISEES ............................................................................................................................................... 94 P. BORNE, G. DAUPHIN-TANGUY, J.P. RICHARD, F. ROTELLA, AND I. ZAMBETTAKIS. ANALYSE ET [BORDAU93] REGULATION DES PROCESSUS INDUSTRIELS, TOME 1: REGULATION CONTINUE, TECHNIP, FRANCE, 1993. ........................ 94 M. RIVOIRE, J. L. FERRIER, J. GROLEAU. ASSERVISSEMENT, REGULATION, COMMANDE [RIVFERGRO95] ANALOGIQUE, EYROLLES, 1995. ......................................................................................................................................... 95 F. DE CARFORT, C. FOULARD. ASSERVISSEMENTS LINEAIRES CONTINUS, C3-AUTOMATIQUE, DUNOD [CARFOU71] PARIS, 1971. 95 K. DUTTON, S. THOMPSON, BILL BARRACLOUGH. THE ART OF CONTROL ENGINEERING, PRENTICE [DUTTHO97] HALL, 1997. 95 J. J. DAZZO, C. H. HOUPIS. LINEAR CONTROL SYSTEM ANALYSIS & DESIGN (CONVENTIONAL AND [AZZHOU88] RD MODERN), MCGRAW-H ILL INC., 3 EDITION, 1988. .......................................................................................................... 95 B.D.O. ANDERSON, J.B. MOORE. LINEAR OPTIMAL CONTROL, PRENTICE-HALL, 1990. ......................... 95 [ANDMOO90] H. KWAKERNAAK, R. SIVAN. LINEAR OPTIMAL CONTROL SYSTEMS, WILEY INTERSCIENCE, 1972. ....... 95 [KWASIV72] T. KAILATH. LINEAR SYSTEMS, PRENTICE-HALL, 1980. ........................................................................ 95 [KAILATH80] B. FRIEDLAND. CONTROL SYSTEM DESIGN, MC GRAW-HILL, 1986. .................................................. 95 [FRIEDLAND86] ASTROM-WITTENMARK. COMPUTER CONTROLLED SYSTEMS, ENGLEWOOD CLIFFS, NJ, PRENTICE-HALL, [ASTROM84] 1984. 95

UNIVERSITE DE DSCHANG INSTITUT UNIVERSITAIRE DE TECHNOLOGIE FOTSO VICTOR DE BANDJOUN DEPARTEMENT DE GENIE ELECTRIQUE

E(z)

(z)

C(z)

U(z)

S(z)

G(z)

Support de
Cours de commande numrique
DUT EL2 + EN2 Anne Acadmique 2010/2011 Par Godpromesse KENNE Charg de Cours

COMMANDE NUMERIQUE, DUT EL2 et EN2, DEPARTEMENT DE GENIE ELECTRIQUE, IUT-FV BANDJOUN, UNIVERSITE DE DSCHANG

Objectifs pdagogiques :
Dans ce cours, nous prsentons des notions de base permettant daborder lanalyse des performances en termes de rapidit, prcision et stabilit des systmes asservis discrets. Les notions que nous allons abordes sont quasi-identiques celles dveloppes aux cours de systmes asservis mais avec de nouveaux outils mathmatiques mieux adapts au formalisme de modlisation des systmes discrets. Le cours est illustr par des exemples simples et pratiques ainsi que des fiches de Travaux Dirigs. A la fin de cours e chapitre, ltudiant devra tre capable de calculer la transforme en Z dune fonction discrte, de dterminer loriginale dune transforme en Z donne, de calculer lerreur permanente correspondant une entre type donne pour un systme asservi discret et synthtiser un correcteur numrique par transposition dun correcteur analogique.

Table des matires

CHAPITRE 1 1. ELEMENTS MATHEMATIQUES DE LA THEORIE DES SYSTEMES ECHANTILLONNES .... 2 ETUDE DES SYSTEMES DISCRETS ................................................................................................................................. 2 1.1 Introduction ............................................................................................................................................................... 2 1.2 Systme discret .......................................................................................................................................................... 2 1.3 LES FONCTIONS DISCRETES - CONVENTIONS ............................................................................................................... 2 1.3.1 Impulsion de Dirac discrte ou impulsion de Kronecker ....................................................................................... 2 1.3.2 Fonction unit discrte ........................................................................................................................................... 3 1.3.3 Fonction dHeaviside discrte

{ } ....................................................................................................................3
n

1.3.4 Fonction chantillonne ......................................................................................................................................... 3 1.4 TRANSFORMATION EN Z .............................................................................................................................................. 5 1.4.1 Dfinition de la transforme en Z d'une suite......................................................................................................... 5 1.4.2 Proprits de la transforme en z........................................................................................................................... 6 1.5 RECHERCHE DUN ORIGINAL (TRANSFORMEE INVERSE) ................................................................................................ 13 1.5.1 Dcomposition en lments simples ..................................................................................................................... 13 1.5.2 Gnralisation de la division polynomiale ........................................................................................................... 18 1.5.3 Mthode des rsidus : Intgration dans le plan complexe ................................................................................... 18 1.6 TRANSFORMEE EN Z PAR LA METHODE DES RESIDUS .................................................................................................... 19 TABLE DES TRANSFORMEES ...................................................................................................................................... 21 CHAPITRE 2 REPONSE DES SYSTEMES A TEMPS DISCRET ................................................................................ 22 2.1 INTRODUCTION ............................................................................................................................................................. 22 2.2 CALCUL DE LA REPONSE ............................................................................................................................................... 22 2.3 REPONSE ECHANTILLONNEE ...................................................................................................................................... 23 EXERCICES.......................................................................................................................................................................... 31 2.4 REPONSES ECHANTILLONNEES DES SYSTEMES .............................................................................................................. 32 CHAPITRE 3 ANALYSE DES SYSTEMES ASSERVIS DISCRETS : STABILITE, PRECISION, RAPIDITE .. 35

3.1 MODELISATION DES SYSTEMES ASSERVIS PILOTES PAR CALCULATEUR ..................................................................... 35 3.1.1 Impact de lensemble chantillonneur-bloqueur .................................................................................................. 35 3.1.2 Modle discret quivalent .................................................................................................................................... 36 3.2 Application une structure boucle ........................................................................................................................ 38 3.3 Cas dun retour unitaire.......................................................................................................................................... 39 3.4 STABILITE ................................................................................................................................................................. 39 3.4.1 Critre de Nyquist ................................................................................................................................................ 40 3.4.2 Diagramme de Bode ............................................................................................................................................. 45 3.4.3 Rgime transitoire des systmes discrets ............................................................................................................. 46

3.5 PRECISION DES SYSTEMES ASSERVIS DISCRETS.......................................................................................................... 49 3.5.1 Prcision vis--vis de lentre .............................................................................................................................. 49 3.5.2 Prcision vis--vis de la perturbation .................................................................................................................. 52 CHAPITRE 4 SYNTHESE DES CORRECTEURS NUMERIQUES .......................................................................... 56

4.1 INTRODUCTION.......................................................................................................................................................... 56 4.2 RAPPEL DE QUELQUES METHODES DAPPROXIMATION NUMERIQUE DES SYSTEMES CONTINUS.................................. 56 4.2.1 Transformation dEuler........................................................................................................................................ 56 4.2.2 Transformation homographique - Mthode des trapzes ..................................................................................... 58 4.2.3 Approximations temporelles du drivateur numrique ........................................................................................ 58 4.3 REPONSES FREQUENTIELLES DES DERIVATEURS NUMERIQUES .................................................................................. 59 4.4 TRANSPOSITION DUN CORRECTEUR CONTINU .............................................................................................................. 60 TRAVAUX DIRIGES .......................................................................................................................................................... 66 TRANSFORMATIONS EN Z.................................................................................................................................................... 66 ANALYSE DES SYSTEMES LINEAIRES ECHANTILLONNES .................................................................................................... 67 SYNTHESE DES CORRECTEURS NUMERIQUES...................................................................................................................... 69 REFERENCES UTILISEES ............................................................................................................................................... 69 M. RIVOIRE, J. L. FERRIER, J. GROLEAU. COMMANDE PAR CALCULATEUR, INDENTIFICATION, [RIVFERGRO93] EYROLLES, 1993. 69 P. BORNE, G. DAUPHIN-TANGUY, J.P. RICHARD, F. ROTELLA, AND I. ZAMBETTAKIS. ANALYSE ET [BORDAU93] REGULATION DES PROCESSUS INDUSTRIELS, TOME 1: REGULATION NUMERIQUE, TECHNIP, FRANCE, 1993. ..................... 69 B. DANDREA NOVEL AND M. COHEN DE LARA. COMMANDE LINEAIRE DES SYSTEMES DYNAMIQUES. [ANDCOH94] MASSON, FRANCE, 1994. .................................................................................................................................................... 69 E. DIEULESAINT AND D. ROYER. AUTOMATIQUE APPLIQUEE: TOME 2. SYSTEMES LINEAIRES DE [DIEROY90] COMMANDE A SIGNAUX ECHANTILLONNES, MASSON, FRANCE, 1990. ................................................................................ 70 K. DUTTON, S. THOMPSON, BILL BARRACLOUGH. THE ART OF CONTROL ENGINEERING, PRENTICE[DUTTHO97] 70 HALL, 1997. J. J. DAZZO, C. H. HOUPIS. LINEAR CONTROL SYSTEM ANALYSIS & DESIGN (CONVENTIONAL AND [AZZHOU88] MODERN), MCGRAW-H ILL INC., 3RD EDITION, 1988. ......................................................................................................... 70 T. KAILATH. LINEAR SYSTEMS, PRENTICE-HALL, 1980. ........................................................................ 70 [KAILATH80] B. FRIEDLAND. CONTROL SYSTEM DESIGN, MC GRAW-HILL, 1986. .................................................. 70 [FRIEDLAND86] ASTROM-WITTENMARK. COMPUTER CONTROLLED SYSTEMS, ENGLEWOOD CLIFFS, NJ, PRENTICE-HALL, [ASTROM84] 1984. 70 [JAUTHE91] D. JAUME, S. THELLIEZ, AND M. VERGE. COMMANDE DES SYSTEMES DYNAMIQUES PAR CALCULATEUR, EYROLLES, FRANCE, 1991. ................................................................................................................................................. 70

ANNEXE 2: Innovations pdagogiques

ANNEXE 3: Copie Distinction Honorique

ANNEXE :

4. Copie des cinq Publications Principales

T. Ahmed-Ali, G., Kenn, F. Lamnabhi-Lagarrigue, Identification of nonlinear systems with time-varying parameters using a sliding-neural network observer, in Neurocomputing, Vol. 72, pp. 1611-1620, 2009. Editor: Elsevier. Impact Factor: 1.429.

ARTICLE IN PRESS
Neurocomputing 72 (2009) 16111620

Contents lists available at ScienceDirect

Neurocomputing
journal homepage: www.elsevier.com/locate/neucom

Identication of nonlinear systems with time-varying parameters using a sliding-neural network observer
Tarek Ahmed-Ali a,, Godpromesse Kenne b, Franc oise Lamnabhi-Lagarrigue c a b c

Ecole Nationale Superieure des Ingenieurs des Etudes et Techniques dArmement (ENSIETA), 2 Rue Franc ois Verny, 29806 Brest Cedex 9, France Laboratoire dAutomatique et dInformatique Appliquee (LAIA), Departement de Genie Electrique, IUT FOTSO Victor, Universite de Dschang, B.P. 134 Bandjoun, Cameroun Laboratoire des Signaux et Systemes (L2S), C.N.R.SSUPELEC, Universite Paris XI, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France

a r t i c l e in fo
Article history: Received 2 January 2008 Received in revised form 8 September 2008 Accepted 17 September 2008 Communicated by T. Heskes Available online 1 October 2008 Keywords: State/parameter estimation Time-varying parameter Output equivalent injection Radial basis function Learning algorithms

abstract
In this paper, a new method for the identication of nonlinear systems with time-varying parameters using a sliding-neural network observer is investigated. The proof of the nite-time convergence of the estimates to their true values is achieved using Lyapunov arguments and sliding mode theories. An application example illustrated the effectiveness of the approach and the obtained results show high convergence rate and very satisfactory parameter estimation accuracy. The computing results under noisy condition also demonstrate that good state and parameter estimation can be achieved despite the disturbance (noise) in the system. The reduced number of hidden units and the small transient period demonstrate that the proposed method can be easily implementable in real-time. & 2008 Elsevier B.V. All rights reserved.

1. Introduction This paper is devoted to parameter identication for a large class of nonlinear systems using a sliding mode neural observer. In many practical cases, system parameters are unknown and are time varying. In linear systems, parameter estimation is often achieved using the least square method [17]. The application of the least square algorithm to the estimation of nonlinear system parameters, usually requires the nonlinear model outputs to be expressed linearly in terms of the unknown parameters. Unfortunately, some nonlinear plants cannot be parameterized linearly. The identication of nonlinear system parameters is also often studied by using the combination of the least square technique and the passivity approach [9]. However, in the latter approach, the linearization of the nonlinear model around an equilibrium point is required. Thus, this method may not provide successful results for a wide range of operating points of the plant. Recently, a robust identication and control algorithm with time-varying parameter perturbations has been proposed in [1]. But the application of this method requires the nonlinear model outputs to be expressed linearly in terms of the unknown

Corresponding author. Tel./fax: +33 2 98 34 88 56.

E-mail addresses: ahmedali@ensieta.fr (T. Ahmed-Ali), gokenne@yahoo.com (G. Kenne), lamnabhi@lss.supelec.fr (F. Lamnabhi-Lagarrigue). 0925-2312/$ - see front matter & 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.neucom.2008.09.001

parameters. This motivated our previous works [7,2] where parameter estimation methods based on the radial basis function (RBF) neuronal predictor have been introduced. Although different approaches have been developed [35,8,13] for nonlinear system states estimation, only partial and quite weak results have been obtained in terms of time-varying function approximation and time-varying parameter estimation. The state of the art concerning the estimation of the nonmeasurable states using articial neural network (ANN) has been presented in [7]. The main contribution of this paper is the extension of the works proposed in [7,2]. In the latter approaches, the asymptotic convergence was proved under some conditions. In this paper, the convergence and the robustness properties of the previous results are improved by using a sliding-neural observer and the extension of the previous works to the cases where the state vector of the nonlinear system is partially known is achieved. A new adaptive law is also used to estimate the unknown bound of the error between the neural network output and the output of the nonlinear system. The proposed method is dedicated to parameter estimation in the continuous time-domain for a large class of nonlinear systems whose model outputs can be expressed as linear or nonlinear combination of the unknown parameters. The method proposed in this paper can be viewed as a step towards the design of nonlinear systems time-varying parameter identication using sliding-neural network observer without using the assumption that the inverse dynamics of the system is known.

ARTICLE IN PRESS
1612 T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620

The remainder of the paper is organized as follows. In Section 2, a sliding-neural observer based on the radial basis function network (RBFN) is designed and the proof of the convergence in nite time is achieved using Lyapunov and sliding modes theories. The extension of the approach to the cases where the state vector is not completely known is presented in Section 3 and an application example with some numerical simulations are reported in Section 4. Finally some concluding remarks are given in Section 5.

one can write f x; yn ; u Cw; w ef w or f x; yt; u Cw; w ef w Df x; u; t with

Ci w; w

N X j1

w fkw C j k; nj , ij

(6)

2. Sliding-neural observer for parameter identication In this section, a modied version of the methods proposed in [7,2] is presented and the convergence and robustness properties are improved using a sliding-neural observer. The extension of the previous works to the cases where the state vector of the nonlinear system is partially known is also investigated. 2.1. Problem setting Let us consider the following class of nonlinear systems: _ x f x; yt; u, (1)
n

where f denotes a nonlinear function; C j and nj , j 1; . . . ; N are the center and the width of the j-th hidden unit, respectively; N is the number of the hidden nodes or RBF units; w is the optimal weight vector and satises kw kpRo ; w x; uT is the input vector of the RBFN; ef w is the optimal approximation error tolerance, which is unknown and bounded 8w 2 X. The term Df x; u; t is time-varying and cannot be approximated by a static neural network. In the following analysis, sliding robust terms will be used in the identication scheme to compensate the effect of the uncertainty Df x; u; t. The aim is to realize or to approximate the underlying dynamics f x; yt; u using ANNs assuming that the terms ef w and Df x; u; t are bounded by unknown positive constants. 2.2. Sliding-neural observer In order to identify the time-varying parameters yt, let us now consider the following observer: _ ^ ^ ^ x Cw; w bx; x; t or _ ^ ^ xi bi x; x; t
N X j1

where f is a continuous function on a compact Of 2 R; x 2 R is the state vector; u 2 Rm , mpn is the vector of the measurable inputs and yt 2 Rp is the vector of the unknown time-varying parameters which can be expressed as follows:

yi t yi;n Dyi t with jDyi tjpmi ; i 1; . . . ; p.

(2)

(7)

In (2), yi;n is the nominal value of yi t and mi are unknown positive constants. The following assumptions will be made until further notice. (i) The components of the state vector x are measurable. Therefore the state vector x is assumed to be available and can be considered as the output vector of the nonlinear system (1). (ii) yt 2 Xy which is a compact set of Rp . (iii) the system described by (1) is invertible in terms of the unknown parameters yt in the sense of the work of [10]. Remark 1. Assumption (i) is not a restriction for the method proposed in Section 2.2. It has been used to simplify the study in this section. The cases where the state vector is not completely known are discussed in Section 3. Assumption (iii) means that, there exists a function g continuous and bounded such that _ yt gx; x; u for 1pppn. (3)

^ wij fkw C j k; nj ,

(8)

^ where the term bi x; x; t; i 1; . . . ; n are introduced in order to improve the convergence of the neural network in the presence of the uncertainty term Df x; u; t. The RBF f has the following form:   kZk2 . fZ; n exp 2n2 The center C j and the width nj of the j-th hidden unit are chosen as follows [6]:

nij

wimax wimin
N

(9)

C ij wimin

2j 1 nij , 2

(10)

where wimin and wimax are the lower and upper bounds of the i-th element of the RBF input vector w, respectively. Remark 2. The selection of the center C j and the width nj for RBFN has signicant effect on the performance of the algorithm. In literature several strategies exist but one can distinguish two main categories. The rst one consists in simultaneously optimizing the center C j, the width nj and the weights wij , by using for example the well-known backpropagation algorithm. Unfortunately, this approach may lead to the convergence of the algorithm to the local minimum due to the slowly convergence of the hidden layer (with nonlinear neurons) and the fast convergence of the output layer (with linear neurons). The second approach consists in optimizing or selecting initially the center and the width before carrying out the adjustment of the weights. Therefore, once the optimization or selection of the center and the width is achieved, the learning of the weights is completely a linear problem and is thus more easier to perform as compared to

In the general case, the function g contains the higher order _ time-derivatives of the terms x and u. Note that it is not possible to use immediately Eq. (3) to estimate the unknown parameter _ yt since the derivative state x is not available. By using Taylor formula, the system described by (1) can be rewritten as follows: _ x f x; yn ; u Df x; u; t, where Z
0

(4)

Df x; u; t

qf   Dyt dx qaayn xDyt

(5)

is the uncertainty term due to the variation of the parameter yt. The term f x; yn ; u is continuous on a compact X; thus it can be approximated by an RBF neural network [4,5]. Therefore,

ARTICLE IN PRESS
T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620 1613

the rst approach. This is the main reason that motivated the choice of the second approach in this paper. The proposed RBFN identier is based on the structure depicted in Fig. 1 which is the same structure used in [7,2]. This structure has three layers: an input layer, a hidden layer and an output layer. The weights in the output layers are continuously adjusted by a learning rule which will be derived later on. ^ By dening e x x, the vector of the neural observer error and taking into consideration (6) and (7), it follows that the dynamic equation of the observer error is given by ^ ^ _ e Cw; w Cw; w bx; x; t ef w Df x; u; t. (11)

Using (12), it follows that _ ^ V eT bx; x; t

T

 n N XX qCi  _  ^ ^ wij w wij ei ij qwij w i1 j1

^ ij wij

e ef w Df x; u; t.

(16)

To prove the nite-time convergence of the sliding-neural observer (7) and derive the estimate of the derivative of the state vector and the time-varying unknown parameters, let the terms ^ bi x; x; t; i 1; . . . ; n, the learning rule of the weight w, the adaptive law for l be chosen as follows: ^ ^ bi x; x; t li sgnei , (17)

Using Taylor series expansion and the fact that the neural network ^ is a linear function of w, (11) can be rewritten as
N P qC1   ^ d w1j w 7  6 1j ^ 7 6 j1 qw1j ww1j 7 6 7 6 . 7 ef w Df x; u; t, 6 . _ ^ e bx; x; t 6 . 7  7 6 N 7 6 P qCn   4 ^ wnj w 5  nj j1 qwnj ww ^
nj

" #  qCi  _  ^ wij Proj ei ; i 1; . . . ; n; j 1; . . . ; N  qwij wij wij ^ " # 8   qCi  > > e qCi   ^ ^ > if jwij joRw or jwij j Rw and ei 40 ; > i qw   > qwij wij wij < ij wij wij ^ ^  > qCi  > >0  ^ > o0; if wij jRw j and ei > : qwij w w ^
ij ij

(12) where ef w and Df x; u; t are assumed to be bounded as follows: jef i w Df i x; u; tjpli ; i 1; . . . ; n (13) _ ^ li ( 0 if ei 0;

(18)

ai if ei a0;

(19)

and li ; i 1; . . . ; n are unknown positive constants. ^ In order to derive the expression of the term bx; x; t, the learning rule of the weight w and the adaptation law for l, the following Lyapunov candidate function is considered: V
n N 1 T 1XX ^ w w 2 . e e ij 2 2 i1 j1 ij

^ with ai 40, li 0 0, and Proj the projection function [12] on the compact set Oo fo : kokpRo g. This function guarantees ^ that the estimates wij remain bounded. By taking into account the above considerations, the following inequalities are derived: _ Vp or
n X i1

(14)

^ ei bi x; x; t

n X i1

jei jli

(20)

Its time-derivative is _ V
n X i1

_ ei ei

n N XX i1 j1

_ ^ ^ wij wij w . ij

(15)

_ Vp

n X i1

^ li jei j

n X i1

li jei j.

(21)

Weight updating algorithm b1 X1 X3 1 2 b2 Xn u1 u2 um 3 4 5 N bn 1 s Weight updating algorithm 1 s

+ y1 + y2

Weight updating algorithm 1 s

+ yn

Input layer

Radial basis units

output layer

Fig. 1. Basic structure of the proposed RBFN identier.

ARTICLE IN PRESS
1614 T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620

Inequality (21) can also be rewritten as _ eT ep

n X i1

^ li jei j

n X i1

li jei j

n N XX i1 j1

  qC    i ^ jwij w j ij  qwij o

^ ij wij

   jei j.  (22)

^ From the fact that the projection function is used for oij and Ci ; i 1; . . . ; n are RBFs, one can conclude that all the terms ^ joij o jjqCi =qwij jwij wij j are bounded. ^ ij From (19), one can deduce that ^ li ai t bi if ei a0 with 0pbi oli . (23)

^ Therefore, the gains li , i 1; . . . ; n will always increase while the errors ei , i 1; . . . ; n are not null. This means that there exists a ^ nite time T f such that the gains li satisfy the following inequality:      n n n N X X XX    qCi  ^4 ^ (24) li li jwij wij j . qwij o w  ^ i1 i1 i1 j1
ij ij

width D, the switching frequency f of the control should be increased, otherwise the amplitude of oscillations would exceed D since D % 1=f . To eliminate the high frequency component of the control in sliding mode, the frequency should be much higher than 1=t, or 1=f 5t. Thus, the condition D5t should be fullled. In other words, the time constant of the lowpass lter should be made to tend to zero because the lter should not distort the slow component of the control. Finally, the conditions t ! 0 and D=t ! 0 (e.g., D ! 0) should be satised to extract the slow component equal to the equivalent control and to lter out the high frequency component of the control. ^ Assuming that the equivalent controls bi eq x; x; t; i 1; . . . ; n _ are approximated using (26), the estimation of the derivative x can be deduced as _ ^ ^ ^ ^ x Cw; w beq x; x; t; 8t4T f . (27)

2.3. Parameter identication scheme Using the property of the identiability of the nonlinear system described by (1), the estimation of the unknown timevarying parameter yt can be treated separately by utilizing the convergence in nite time of the above sliding-neural network combined with a static parameter estimation law or simultaneously using an adaptive state observer combined with a dynamic parameter estimation law. In this paper the rst approach is discussed. Two cases will be considered. The method proposed in this paper can be viewed as a step towards the design of nonlinear systems time-varying parameter identication using a sliding-neural network observer without using the assumption that the inverse dynamics of the system is known. 2.3.1. Case of nonlinearly parameterized system with respect to unknown parameters In this case, it is assumed that the nonlinear system described by (1) cannot be expressed linearly in term of the unknown timevarying parameter yt but is inversible with respect to yt and the derivative of the state x is not available. Hence, the parameter identication scheme can be computed by replacing in (3) the ^ ^ ^ _ derivative x with the estimated term Cx; w beq x; x; t. The following expression is obtained: ^ ^ ^ ^ yt gx; Cx; w beq x; x; t; u. (28)

Remark 3. The selection of b may inuence the convergence of the proposed approach. But the inuence of this parameter has not been investigated because there is no guideline for the selection of the value of this parameter. However, since l is completely unknown, it is convenient to choose b 0. As a consequence of the above considerations, a sliding mode _ regime will occur on the manifolds ei ei 0, i 1; . . . ; n in nite time T f [14,16]. This means that Eq. (11) can be rewritten as ^ ^ _ beq x; x; t Cw; w x; 8t4T f , (25)

^ where the terms bi eq x; x; t; i 1; . . . ; n represent the equivalent controls. Remark 4. From a practical point of view, it is not possible to ^ _ implement beq x; x; t because the derivative state x is not available. To overcome this problem one can use the average control vector as an approximation for the equivalent control ^ beq x; x; t [15]: ^ ^ bi eq x; x; t 1 ^ ^ b x; x; t, 1 tp i (26)

where t is tuning small positive constant. In fact, the motion in sliding mode has been considered with a certain idealization. It was assumed that the control changes at high (theoretically innite) frequency such that the state velocity vector is oriented precisely along the intersection of discontinuity surfaces. However, in reality, various imperfections make the state oscillate in some vicinity (trajectories are conned in the boundary layer of with D40) of the intersection and components of control are switched at nite frequency. These oscillations have high frequency and slow components. It is reasonable to assume that the equivalent control is close to the slow component of the real control which may be derived by ltering out the high frequency component using a lowpass lter. Its time constant should be sufciently small to preserve the slow component undistorted but large enough to eliminate the high frequency component. As shown in [15], the output of the lowpass lter _ ^ ^ tbi bi bi x; x; t tends to the equivalent control
t!0;D=t!0

The main result of this section can be summarized by the following theorem. Theorem 1. Consider the class of nonlinear systems described by (1) ^ and the sliding-neural network predictor (7), with the terms bi x; x; t, the learning rule of the weights wij and the adaptive law of the unknown positive constants li given by (17)(19), respectively. ^ The estimated parameters yt converge to their true values yt in nite time under the assumptions that conditions (2) and (13) are satised. Proof. The expression of the parameter estimation error is given by _ ^ ^ _ ey yt y gx; x; u gx; x; u. (29)

lim

^ bi bi eq x; x; t.

One can also note that the vicinity of a discontinuity manifold of width D, where the state oscillates, should be reduced to make the real motion close to ideal sliding mode. In order to reduce the

Following the development of Section 2.2, a sliding mode regime _ occurs in a nite time T f on the manifolds ei ei 0; i 1; . . . ; n: _ ^ _ Consequently, the convergence of x to x in nite time implies that ^ the estimated parameters yt converge to their true values yt in nite time. & Remark 5. Although in the proof of the convergence of the ^ estimated parameters yt to their true values yt, the persistency

ARTICLE IN PRESS
T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620 1615

of excitation condition is not explicitly required, this condition remains necessary since it is physical property of the plant. 2.3.2. Case of linearly parameterized system with respect to unknown parameters In this case it is assumed that the nonlinear system described by (1) can be expressed linearly in terms of the unknown timevarying parameter yt as follows: _ x FxT yt hx; u, (30)

^ ^ Let e1 x1 x1 and e2 x2 x2 be the observer errors. Using (33), (34) and the fact that f x1 ; yt; u Cw; w ef w Df x1 ; u; t or f x1 ; yt; u
N X j1

w fkw C j k; nj w ij

with w ef w Df x1 ; u; t, the dynamic equations of the observer errors can be computed as _ e 1 e2

N X j1

where F is an n p matrix with elements continuous and bounded on a compact OF & R and h is a continuous function on a compact Oh 2 R. Let us consider a weighting matrix P symmetric and non_ negative. If the derivative x is computed using (27) and the matrix FPF T is inversible, then the estimation of the unknown timevarying parameter yt can be obtained as follows: _ ^ ^ y FPF T 1 FPx FPh. (31)

^ wij fkw C j k; nj lt sgne1 w,

(35)

_ e2 k2 e1 dt, ^ where wij wij w . ij

(36)

^ The proof of the convergence in nite time of yt to yt is similar to that of Theorem 1 by considering _ _ gx; x; u FxPFxT 1 FxP x FxPhx; u. 3. Case with unavailable states In most of the cases, all the elements of the state vector x are not available. In this section two forms of nonlinear systems with partial vector state are investigated. 3.1. First form of the nonlinear system In this case, the following form of nonlinear systems is considered: _ x1 x2 f x1 ; yt; u, _ x2 dt, y x1 , (33) where the state x2 , the functions f and d and the timevarying parameter yt are bounded but unknown; u is the control input and the measured output is x1 . The rst equation of (33) has the form of (4) in which x2 is unknown but can be estimated. In order to derive the parameter estimation law for the class of nonlinear systems (33), let us consider the following modied version of the above sliding-neural observer: _ ^ ^ ^ ^ ^ x1 x2 Cw; o lt sgnx1 x1 ^ with Cw; o
N X j1

The following theorem summarizes the main result of this section. Theorem 2. Given the nonlinear system described by (33) and the observer (34). Assuming that the state x2 , the functions f and d and the time-varying parameter yt are bounded but unknown, and that the learning rule of the weights wij and the adaptive law of the unknown positive constant l are given by (18) and (19), respectively, then there exists a positive constant Z and a nite time t f such that     N  X ^  wij fkw C j k; nj lt sgne1 w je2 jpZ; 8t4tf .     j1 (37) Proof. To prove the above theorem, let us consider the following change of variables: z1 e1 , z2 e2 k1 e1 , (38) where k1 is a small positive constant (e.g., 0ok1 o1). With the above change of variables, the observer errors dynamics (35) and (36) become _ z1 k1 z1 z2
N X j1

(32)

^ wij fkw C j k; nj lt sgnz1 w, 2

(39)

_ z2 k2 z1 dt k1 4k1 z1 z2 3 ^ lt sgnz1 w5.

N X j1

wij fkw C j k; nj

^ wij fkw C j k; nj ,

(40)

_ ^ ^ x2 k2 x1 x1 .

(34)

^ ^ ^ ^ In (34), x2 is the estimate of x2 ; Cw; o lt sgnx1 x1 is the estimate of f x1 ; yt; u using the sliding-neural observer ^ ^ presented in Section 2.2; o and l are computed as in Section 2.2 (see Eqs. (18) and (19)) and wT x1 ; u is the input of the RBFN. Remark 6. Note that, in this case, the input vector w of the RBFN does not contain the unknown state x2 . Therefore the center C j and the width nj of the j-th hidden unit can be computed as in Section 2.2 using (9) and (10). Also note that, in this case only the computation of the unknown parameter needs the estimated value of the unavailable state (e.g., x2 in this case).

By taking into account the results obtained in Section 2.2, there _ exists a nite time tf such that 8t4t f , one has z1 z1 0. Consequently, the above observer errors dynamics become (41) 2 3 N X k2 14 ^ t sgnz w5, _ z2 z2 dt w fkw C j k; nj l 1 k1 k1 j1 ij 8t4t f . (42) _ z1 0,

To achieve the proof, the following Lyapunov candidate function can be considered. V 1z2 . 2 2 (43)

ARTICLE IN PRESS
1616 T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620

Its time-derivative is 2 3 N X ^ t sgnz w5. _ k2 z2 z2 dt z2 4 V w fkw C j k; nj l 1 k1 2 k1 j1 ij (44) Thus, there exist positive constants z1 and z2 such that the derivative of the Lyapunov candidate function V can verify the following inequality: _ Vp z1 z2 z2 jz2 j, 2 where z1 k2 =k1 and z2 is dened as follows:  2 3   N   1 4X ^ t sgnz w5pz . dt wij fkw C j k; nj l 1 2   k1 j1   Inequality (45) can be rewritten as   z _ Vp z1 jz2 j 2 jz2 j. (45)

z1

(46)

_ If jz2 j4z2 =z1 then V will be negative-denite. This means that z2 converges in the ball with radius Z z2 =z1 which can be reduced to the neighborhood of zero by chosen z1 bz2 . Therefore, the estimation of the time-varying parameter yt is achieved as in Section 2.3. & 3.2. Second form of the nonlinear system The other more general class of nonlinear systems is given by _ x1 x2 , _ x2 x3 , . . . _ xn1 xn , _ xn f x1 ; x2 ; . . . ; xn1 ; xn ; ut; yt, (47) where only x1 is measured or available. In this case, the unavailable states x2 ; . . . ; xn can be estimated step-by-step using model-free observers (e.g., sliding mode observers or second-

order sliding mode observers [11]) and assuming that their estimates converge to their true values in nite time, the above nonlinear system (47) can be rewritten in the form given by (1). Step-by-step estimation of the unavailable states x2 ; . . . ; xn means that the estimate of x2 is computed by using the measured state ^ x1 . Assuming that x2 converges in nite time to the true value x2 , then the estimation of x3 ; . . . ; xn can be achieved using the same procedure. Theorem 1 can then be applied since the unavailable states of the input vector wT x1 ; x2 ; . . . ; xn1 ; xn ; ut of the RBFN can be replaced by their estimated values such that the bounds wimin and wimax can be computed in order to evaluate the centers and the widths of the RBFs. In other words, two main steps are required: (a) First step: Computation/estimation of the unavailable states using model-free observers (e.g., sliding mode observers or second-order sliding mode observers [11]). (b) Second step: After the nite-time convergence of the unavailable states to their nominal values, the centers and the spreads of the RBFs can then be computed as in Section 2.2 since the unavailable states are known at this step and the system described by (47) is similar to that represented by (1) and Theorem 1 can then be applied. Note that Theorem 2 is not valid for systems represented by (47) since the algorithm that provides the estimates of the unavailable states in this case is different from the one used in Section 3.1.

4. An application example To illustrate the effectiveness of the proposed technique, let us consider the following nonlinear system: _ x1 x2 y1 x2 y2 x3 u1 , 1 _ x3 x4 y2 x2 u2 , 3 _ x2 2 x2 sin x2 d2 t, _ x4 3 x4 sin x4 d4 t,

(48)

where the states x2 , x4 and the time-varying parameters y1 , y2 are bounded but unknown; u1 and u2 are the control inputs and

Reference values of the unknown timevarying parameters 1 2 3 1ref 4 5 6 7 8 0 1.5 2 2.5 2ref 3 3.5 4 4.5 0 1 2 3 4 Time (s) 5 6 7 8 1 2 3 4 5 6 7 8

Fig. 2. Reference values of the time-varying parameters.

ARTICLE IN PRESS
T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620 1617

yT x1 ; x3 is the vector of the measurable outputs. It is also assumed that u1 , u2 , x1 and x3 are bounded. By dening X T x1 ; x3 ; 1 U T u1 ; u2 X T x2 ; x4 ; 2 and

The application of the method proposed in Sections 2 and 3 yields X _ ^ ^ ^ ^ ^ X 1i X 2i wij fkw C ij k; nij li tsgnX 1i X 1i
N j1

YT y1 ; y2 ,

with wT x1 ; x2 ; u1 ; u2 , _ ^ ^ X 2i k2i X 1i X 1i . (50)

DT d1 ; d2 ,

the above nonlinear system can be rewritten as the system described by (33) as follows: _ X 1 X 2 f X 1 ; Yt; U, _ 2 Dt, X Y X1. (49)

To derive the estimation law of the unknown time-varying _ _ parameters y1 and y2 , assuming that the unavailable state x1 , x3 , x2 and x4 are computed using the modied sliding-neural observer (50), we rewrite the system described by (48) as follows: _ ^ ^ ^ X 1 FX 1 T Yt hX 2 ; U, (51)

x1 and its estimate 2 1.5 x1 1 0.5 0 0.5 0 0.2 0 0.2 e1 0.4 0.6 0.8 0 1 2 3 4 Time (s) 5 6 7 8 e2 1 2 3 4 5 6 7 8 Observer error e1 0.2 0 0.2 0.4 0.6 0.8 0 1 2 x2 3.5 3 2.5 2 1.5 1 0.5 0 0 1 2

x2 and its estimate

Observer error e2

4 Time (s)

^ Fig. 3. Performance of the proposed sliding-neural estimation algorithm. (a) x1 (dotted line) and its estimate x1 (solid line) and the observer error e1. (b) Unknown state x2 ^ (dotted line) and its estimate x2 (solid line) and the observer error e2.

x3 and its estimate 3.5 3 2.5 2 1.5 1 0.5 0 0 0.5 0 e4 e3 0.5 1 1.5 0.4 0 1 2 3 4 Time (s) 5 6 7 8 0 1 2 1 2 3 4 5 6 7 8 0.4 0.2 0 0.2 Observer error e3 4 3 x4 2 1 0 0 1 2

x4 and its estimate

x3

Observer error e4

4 Time (s)

^ Fig. 4. Performance of the proposed sliding-neural estimation algorithm. (a) x3 (dotted line) and its estimate x3 (solid line) and the observer error e3. (b) Unknown state x4 ^ (dotted line) and its estimate x4 (solid line) and the observer error e4.

ARTICLE IN PRESS
1618 T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620

where " FX 1 T x2 1 0 x3 x2 3 # ; ^ hX 2 ; U ^ x 2 u1 ^ x 4 u2 ! and ^ Y ^ y

1

! .

^ y2

In this case, the matrix P can be chosen as an identity matrix since the determinant detFF T x4 x4 1 3 is always different to zero if x1 and x3 are different to zero. Therefore, the estimation law of the unknown time-varying parameters y1 and y2 are given by ^ ^ ^ Y FX 1 FX 1 T 1 FX 1 X 1 FX 1 hX 2 ; U. _ (52)

To verify the efciency of this method, simulations have been carried out in Matlab/Simulink environment software with a

sampling time of 500 ms. The values of the control inputs u1 and u2 ^ are 10 and 15, respectively, with x1 0 x1 0 0:75 and ^ ^ x3 0 x3 0 1. The initial values of wij are uniform random ^ numbers while zero initial values have been chosen for li t. The values of the tuning parameters used in the estimation scheme have been obtained by trial and error during multiple simulation tests and are given by: a1 0:75, k21 k2 2:15, a3 2:26, k22 k4 2, t 0:01 and N 4. The reference values of the unknown parameters used in all simulations are shown in Fig. 2. The magnitude of the variation of these time-varying parameters is 40% of their nominal values. The performance of the proposed sliding-neural estimation algorithm is shown in Figs. 35. These results show that the state restoring errors e1 , e2 , e2 , e3 and the parameter estimation errors ey1 and ey2 converge to zero in a very short nite time.

10 5 0 1 5 10 15 0 15 10 e1 1 2

Estimate of 1 (t) 0 2 5 10 3 4 5 6 7 8 0 1 2

Estimate of 2 (t)

Parameter estimation error e1 1 0 1 2 3 4 5 0 1

Parameter estimation error e2

0 5 0 1 2 3 4 5 Time (s) 6 7 8

e2

4 5 Time (s)

^ ^ Fig. 5. Estimated parameters and their estimation errors. (a) y1 and ey1 . (b) y2 and ey2 .

Observer errors under noisy condition 0.2 0 0.2 e1 0.4 0.6 0.8 0 0.5 0 e3 e4 0.5 1 1.5 0.4 0 1 2 3 4 Time (s) 5 6 7 8 1 2 3 4 5 6 7 8 0.4 0.2 0 0.2 e2 0.2 0 0.2 0.4 0.6 0.8 1

Observer errors of the unknown states under noisy condition

4 Time (s)

Fig. 6. Performance of the proposed sliding-neural estimation algorithm under noisy condition (when the magnitude of the noise reaches about 3.5% of the maximum ^ ^ ^ ^ values of the measurable outputs x1 and x3 ). (a) Observer errors e1 x1 x1 and e3 x3 x3 . (b) Observer errors e2 x2 x2 and e4 x4 x4 .

ARTICLE IN PRESS
T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620 1619

Estimate of 1(t) under noisy condition 5 0 1 5 10 0 1 2 3 4 5 6 7 8 2

2 0 2 4 6 8 10 12 0 1

Estimate of 2 under noisy condition

Parameter estimation error e1 under noisy condition 5 e1 e2 0 5 0 1 2 3 4 Time (s) 5 6 7 8 2 0 2 4 6 8 10 0

Parameter estimation error e2 under noisy condition

4 Time (s)

Fig. 7. Estimated parameters and their estimation errors under noisy condition (when the magnitude of the noise reaches about 3.5% of the maximum values of the ^ ^ measurable outputs x1 and x3 ). (a) y1 and ey1 . (b) y2 and ey2 .

Furthermore, the robustness of the algorithm under noise condition has been veried and the results obtained are reported in Figs. 6 and 7 when the magnitude of the noise reaches about 3.5% of the maximum values of the measurable outputs x1 and x3 . It can be seen that good state and parameter estimation can be achieved despite the disturbance (noise) in the system. The reduced number of hidden units (N 4) and the small transient period demonstrate that the proposed method can be easily implementable in real-time. Figs. 3b and 4b show that there exists a time interval during the transient period where the estimates of x2 and x4 are much faster than the real states. However, it is well known that the response of the observer depends upon the real state. This phenomenon is due to the fact that the states x2 and x4 are _ assumed to be completely unknown as well as their dynamics x2 _ and x4 . Therefore, it is not possible to compare the estimates and the real states in practice since this is only realizable in simulation environments.

system is yet to be done. This new formulation would permit the identication of more general class of nonlinear systems. References
[1] T. Ahmed-Ali, F. Floret, F. Lamnabhi-Lagarrigue, Robust identication and control with time-varying parameters perturbations, J. Math. Comput. Modelling Dynamical Syst. 10 (34) (2004) 201216. [2] T. Ahmed-Ali, G. Kenne, F. Lamnabhi-Lagarrigue, Nonlinear systems parameter estimation using neural network: application to synchronous machine, J. Math. Comput. Modelling Dynamical Syst. 13 (4) (2007) 365382. [3] M. Antonio, W. Yu, Non-linear benchmark system identication with partial states measurement, in: Proceedings of the 38-th Conference on Decision and Control, Phoenix, Arizona, USA, December 1999, pp. 10771082. [5] S. Chen, A. Billings, Neural networks for nonlinear dynamic system modelling and identication, Int. J. Control 56 (1992) 319346. [4] S. Chen, S.A. Billings, C.F.N. Cowan, P.M. Grant, Practical identication of NARMAX models using radial basis functions, Int. J. Control 52 (1990) 13271350. [6] A.K. Jain, R.C. Dubes, Algorithms for Clustering Data, Prentice-Hall, Englewood Cliffs, NJ, 1988. [7] G. Kenne, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, H. Nkwawo, Nonlinear systems parameters estimation using radial basis function network, Control Eng. Pract. 14 (7) (2006) 819832. [8] J.G. Kuschewski, S. Hui, S.H. Zak, Application of feedforward neural networks to dynamical system identication and control, IEEE Trans. Control Syst. Technol. 1 (1) (1993) 3747. [9] I.D. Landau, B.D.O. Anderson, F. De Bruyne, Algorithms for identication of continuous time nonlinear systems: a passivity approach, in: A. Isidori, F. Lamnabhi-Lagarrigue, W. Respondek (Eds.), Nonlinear Control in the Year 2000, vol. 2, Springer, Paris, 2000, pp. 1344. [10] Y. Lecourtier, F. Lamnabhi-Lagarrigue, E. Walter, Volterra and generating power series approaches to identiability testing, in: E. Walter (Ed.), Identiability of Parametric Models, Pergamon Press, Oxford, 1987, pp. 5066. [11] A. Levant, Robust exact differentiation via sliding mode technique, Automatica 34 (1998) 379384. [12] D.G. Luenberger, Linear and Nonlinear Programming, Addison-Wesley Publishing Company, MA, 1984. [13] R.J. Schilling, J.J. Carroll, A.F. Al-Ajlouni, Approximation of nonlinear systems with radial basis function neural networks, IEEE Trans. Neural Networks 12 (1) (2001) 115. [14] J.J. Slotine, W. Li, Applied Nonlinear Control, Prentice-Hall, New York, 1991. [15] V.I. Utkin, Sliding Modes in Optimization and Control, Springer, Berlin, 1992. [16] V.I. Utkin, Sliding mode control design principles and applications to electric drives, IEEE Trans. Ind. Electron. 40 (1993) 2636. ` [17] E. Walter, L. Pronzato, Identication de modeles parametriques, Masson Publishing, Paris, 1994.

5. Conclusion In this paper, a new scheme for time-varying parameter identication using sliding-neural network observer has been designed. The convergence of the proposed algorithm has been achieved using the Lyapunov arguments and sliding mode theories. An application example illustrated the effectiveness of the approach and the simulation results obtained show high convergence rate and very satisfactory parameter estimation accuracy. Furthermore, the robustness of the algorithm under noise condition has been veried and the computing results show that good state and parameter estimation can be achieved despite the disturbance (noise) in the system. The reduced number of hidden units (N 4) and the small transient period demonstrate that the proposed method can be easily implementable in realtime. This feature distinguishes the proposed identier from the known ones. The design of nonlinear systems time-varying parameter identication using a sliding-neural network observer without using the assumption that the inverse dynamics of the

ARTICLE IN PRESS
1620 T. Ahmed-Ali et al. / Neurocomputing 72 (2009) 16111620

Tarek Ahmed-Ali was born in Algiers, Algeria in 1972. He received the B.Sc. degree in electrical engineering from the Ecole Nationale Polytechnique of Algiers, Algiers, Algeria in 1994. He received the M.Sc. and Ph.D. degrees both in electrical and computer engineering, from the University of Paris VI and the University of Paris XI, Paris, France in 1995 and 1998, respectively. In 1998, he joined the University of Paris XIII, Paris, France and Ecole Centrale de Lille, Lille, France, as a Teaching and Research Assistant. In 2000, he was appointed Research and Development Engineer at SNCF (the French Railway Corporation). Since 2002, he is Senior Lecturer in Control Engineering at Ecole nieurs des Etudes et Techniques de lArmement (ENSIETA), Nationale des Inge Brest, France. His main research interests are sliding mode control, nonlinear observers and fault-tolerant control and diagnosis in the eld of ac drives. Godpromesse Kenne was born in Balatchi, Mbouda, Cameroon in 1967. He received the B.S. degree in electro-mechanical engineering in 1991, the M.S. degree in mechanical engineering in 1994, both from Ecole Nationale Superieure Polytechnique, Universite de Yaounde I and the Ph.D. degree in Control Theory from University of Paris XI, France in 2003. Since 1996, he has been with the University of Dschang, Cameroon, where he is currently Senior Lecturer and Head of the Department of Electrical Engineering at the FOTSO Victor University Institute of Technology. He was Guest ` Researcher at the Laboratoire des Signaux et Systemes (L2S), CNRSSUPELEC, Universite Paris XI, and the Departement Energie, Ecole Superieure dElectricite (SUPELEC) from September 2005 to January 2006 and from June to July 2006. Since April 2006, he is founding member of Automation and Applied Computer Engineering Laboratory (LAIA). His research interests include identication and control of nonlinear systems using variable structure and neural network with applications in electro-mechanical systems.

Francoise Lamnabhi-Lagarrigue is Research Director at CNRS since 1993 at the Laboratoire des Signaux et ` Systemes (L2S), Gif-sur-Yvette, France. She obtained the Matrise es sciences mathematiques pures degree at the Universite Paul Sabatier (Toulouse) in 1976 and she held a CNRS position in 1980. She obtained her Doctorat dEtat es Sciences Physiques from University of Paris Sud in 1985. Some recent activities include: Associate Editor of International Journal of Control; Director of the Marie Curie Control Training SiteCTS www.mc-cts.org; Elected Member of the Board of Governors of the IEEE Control Systems Society for 20032005; Scientic Manager of the HYCON Network of Excellencewww.ist-hycon.organd of the European Embedded Control Institute EECIwww.eeci-institute.eu. Her main research interests include performance and robustness issues in nonlinear and/or hybrid control, identication of nonlinear systems, and the application of these techniques, in particular to power systems, electro-mechanical systems and to vehicle global chassis control, and in general to networked and embedded systems.

G. Kenn, R. S. Simo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An online simplified rotor resistance estimator for induction motors, in IEEE Trans. On Control Systems Technology, Vol. 18, pp. 1188-1194, 2010, Impact Factor: 1.419.

1188

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 18, NO. 5, SEPTEMBER 2010

An Online Simplied Rotor Resistance Estimator for Induction Motors

Godpromesse Kenn, Rostand Sorel Simo, Franoise Lamnabhi-Lagarrigue, Amir Arzand, and Jean Claude Vannier

AbstractThis brief presents an adaptive variable structure identier that provides nite time convergent estimate of the induction motor rotor resistance under feasible persistent of excitation condition. The proposed rotor resistance scheme is based on the standard dynamic model of induction motor expressed in a xed reference frame attached to the stator. The available variables are the rotor speed, the stator currents and voltages. Experiments show that the proposed method achieved very good estimation of the rotor resistance which is subjected to online large variation during operation of the induction motor. Also, the proposed online simplied rotor resistance estimator is robust with respect to the variation of the stator resistance, measurement noise, modeling errors, discretization effects and parameter uncertainties. Important advantages of the proposed can algorithm include that it is an online method (the value of be continuously updated) and it is very simple to implement in real-time (this feature distinguishes the proposed identier from the known ones). Index TermsEquivalent injection term, nonlinear observer, online parameter estimation.

I. INTRODUCTION HE FACT that the induction motor (IM) is a multivariable, nonlinear and highly coupled process with time-varying parameters, has motivated a lot of work in the control community during the last decade [1][14]. The popular alternative method in many drive applications is the eld-oriented control (F.O.C.) which provides a means to obtain high-performance control of an IM. But this F.O.C. methodology requires knowledge of the rotor uxes which are not usually measured [13]. Traditionally, observers are used to estimate the rotor uxes. However, the ux observers used in the currently IM control rely on a good knowledge of the rotor resistance. It is well known in literature (e.g., see [15], [16], [1], [2], [4], [5]) that the rotor resistance and the stator resistance may vary up to 100% and 50% of their nominal values, respectively, during operation of the IM due to rotor heating. Standard methods for the estimation of IM parameters include the blocked rotor test, the no-load test and the standstill frequency response test. But, these methods
Manuscript received June 12, 2009; accepted September 14, 2009. Manuscript received in nal form September 29, 2009. First published November 03, 2009; current version published August 25, 2010. Recommended by Associate Editor L. Dessaint. G. Kenn and R. Sorel Simo are with the Laboratoire dAutomatique et dInformatique Applique (LAIA), Dpartement de Gnie lectrique, IUT FOTSO Victor Bandjoun, Universit de Dschang, B.P. 134 Bandjoun, Cameroun (e-mail: gokenne@yahoo.com; ssimo81@yahoo.fr). F. Lamnabhi-Lagarrigue is with the Laboratoire des Signaux et Systmes (L2S), CNRS-SUPELEC, Universit Paris XI, 91192 Gif-sur-Yvette, France (e-mail: lamnabhi@lss.supelec.fr). A. Arzand and J. Claude Vannier are with the Dpartement nergie, cole Suprieure dlectricit (SUPELEC), 91192 Gif-sur-Yvette, France (e-mail: amir.arzande@supelec.fr; jean-claude.vannier@supelec.fr). Digital Object Identier 10.1109/TCST.2009.2033790

cannot be used online during normal operation of the machine. The most natural solution is to online identify the time-varying parameters. Several papers addressed the problem of the online IM parameter estimation [1][13], [10], [14]. In [3], [6], [7], [10], [14], the problem of IM time-varying parameter has been studied but experiments have not been carried out to verify the effectiveness of the approaches. In [8], [9], and [13], interesting algorithms for IM parameter estimation are proposed using least square technique but more sophisticated lters are required when PWM inverter is used. Moreover, online variation of the IM parameters was not investigated. In [12], a method for rotor resistance estimation for indirect eld oriented control of IM based on reactive power reference model is presented under motoring and generating modes. Sensitivity of the algorithm to errors in other machines parameters is investigated but without variation of the rotor resistance. In [11], a robust nested sliding mode regulation with application to rotor ux modulus and rotor speed of IM with unknown load torque has been introduced. The variation of the stator/rotor resistance has been investigated but the estimation of these parameters was not achieved. Remarkable results have been obtained by the authors of [17] in deriving rotor resistance and load torque estimators suitable for online rotor speed and ux adaptive control. The main drawback of this approach is that the rotor resistance estimator is based on a simplify model of IM which requires the rotor speed to vary slowly. In this brief, the rotor resistance is estimated and its online implementation does not require the assumption of slowly variation of the rotor speed. The effect of the stator resistance variation on the estimation of is also investigated. This brief is organized as follows. In Section II, the IM mathematical model is recalled. The design procedure of the proposed rotor resistance identier is described in Section III and the proof of the nite time convergent estimate to its nominal value is achieved under feasible persistent of excitation (P.E.) condition. Experimental results of online implementation are reported in Section IV and some concluding remarks are given in Section V. II. INDUCTION MOTOR MODEL According to the classical - axes transformation with a xed reference frame attached to the stator, and assuming linear magnetic behavior, the dynamic of a balanced IM is given by the following fth-order nonlinear system [1], [18]:

(1)

1063-6536/$26.00 2009 IEEE

KENN et al.: ONLINE SIMPLIFIED ROTOR RESISTANCE ESTIMATOR FOR INDUCTION MOTORS

1189

(2) (3) (4) (5) In the above equations, the state variables are rotor speed , rotor ux , and stator currents ; the control inputs are stator voltages ; the measured variables are while are not measured; the parameters are the external load torque , total motor, and load moment of inertia , rotor and stator winding resistances , inductances and mutual inductance is the electromagnetic torque and is the number of pole pairs. To simplify the notations, we use (leakage parameter) and the constant . The following assumptions will be considered until further notice: (i) stator current and voltage are bounded signals; (ii) rotor resistance , where is a compact set of . Our goal is to design a rotor resistance estimation algorithm assuming that there exists a control input which can stabilizes the motor in a wide range of operating points. III. ROTOR RESISTANCE ADAPTATION ALGORITHM To derive an online estimate of the rotor resistance, let us consider the following observer ( is a constant designed parameter): (12) (11)

(13) The above associated error dynamics can be rewritten as

(14)

(15) (16) (17) To achieve the design of the rotor resistance identier the following additive assumption is required. Assumption (iii): It is assumed that the following rotor resistance identiability condition holds: (18) Remark 1: The identiability condition (18) can be replaced by the following persistency of excitation (P.E.) condition. There exists such that (19)

(6)

(7) (8) (9) where and are additional signals yet to be designed and sign is the well known sign function. The estimated quantities are shown as while the error quantities are shown as (e.g., ). The dynamics of the observer error can be computed using (1)(4) and (6)(9) as Remark 2: The persistency of excitation condition (19) is often satised when the IM is fed by PWM power inverter. This is the case of the control system considered in this work. By considering the following Lyapunov candidate function: (20) and computing its time-derivative along the trajectories of (14) and (15), we obtain

(10)

(21)

1190

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 18, NO. 5, SEPTEMBER 2010

From (21), by taking into account assumptions (i) and (ii), the following inequalities hold:

We now consider the following quadratic function of the rotor ux observer error and rotor resistance estimation error:

(25) Its time-derivative along the trajectories of (16) and (17) yields

(26) If we choose parameter): , and as follows ( is a designed

(27) (22) and are bounded,1 let posAssuming that the estimate itive constants and be available such that becomes (28) Consequently, under P.E. (19) and if the are auxiliaries vari, and are chosen as in (27), will be negative ables denite and . Thus, and converge in nite time to their nominal values and with the convergence rate and , respectively. Remark 4: If , the rotor resistance convergence will be faster than that of the rotor ux. In contrary, if , the rotor ux convergence will be faster than that of the rotor resistance. The case is difcult to implement in practice since is assumed to be unknown and is time-varying but veried in normal operation of the IM . To achieve the design of the rotor resistance estimator, implementable expression for is required. Under condition (24), a sliding-mode occurs in nite time on the 2-D manifold (29) The equivalent injection terms [19] can be computed by solving the equation (30) Consequently, (14) and (15) can be rewritten as

(23) where denotes the maximum value of . Remark 3: The values of the constants and can be evaluated for any given operating condition on the IM by using the nominal values of the rotor resistance and inductance (to compute the value of ) and the maximum admissible values of the rotor resistance and rotor ux estimation errors in transient period. Other ofine methods can be exploited to evaluate the nominal value of the rotor time constant without using the nominal values of and in the case of squirrel IM. By choosing (24) will be negative denite and the derivative of . Therefore, the observer errors and converge to 0 in nite time if is chosen such that condition (24) is satised.
1The

(31)

proof of the boundness and convergence will given later.

(32)

KENN et al.: ONLINE SIMPLIFIED ROTOR RESISTANCE ESTIMATOR FOR INDUCTION MOTORS

1191

constant, the rotor ux convergence will be faster than that of the rotor resistance. Other available ofine methods can be exploited to evaluate the nominal value of the rotor time constant without using the nominal values of and in the case of squirrel IM. Under this assumption and (P.E.) condition (19), the implementable expression of the rotor resistance estimation error can be derived from (31) and (32) by neglecting the terms containing the rotor ux estimation error. We then obtain

with (34) Remark 5: The denominator of (34) can become zero in transient periods since the identiability condition (18) or (P.E.) condition (19) is based on the real value of the ux and not on the estimated value. However, this singularity cannot affect signicantly the estimate value of the rotor resistance since the adaptation law (27) uses the sign function. A singularity detector can also be used and such algorithm can provide as output the nominal value of the rotor resistance when the singularity is detected. Finally, the overall simplied rotor resistance estimator can be summarized as follows:

Fig. 1. Block diagram of the experimental setup.

Fig. 2. Speed and ux reference signals. (i) Speed reference in experiments 1 and 3. (ii) Speed reference in experiment 2. (iii) Rotor ux reference.

where and . The expressions of the equivalent injection terms and can be deduced from (31) and (32) but these expressions and are not cannot be implemented in practice since available.2 To overcome this problem, we approximated the equivalent injection terms and by using rst order low-pass lters as in [19]. If the design parameter is chosen such that with (33)

(35) IV. EXPERIMENTAL RESULTS The effectiveness of the proposed algorithm combined with a nonlinear controller which stabilizes the rotor ux magnitude and the rotor speed to references values with adaptation of the rotor resistance and load torque (see [17] for more details) has been veried experimentally in various operating conditions. Remark 6: The combination of both estimation algorithms (rotor resistance and load torque estimators) still converges since it has been proved in [17] that the proposed nonlinear controller can stabilize the IM to reference trajectories when the estimated values of and are bounded in the operational domain and the (P.E.) condition satised.

and are the nominal values of the rotor inducwhere tance and rotor resistance and is the nominal rotor time2

is assumed to be unknown and  is often not measurable.

1192

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 18, NO. 5, SEPTEMBER 2010

Fig. 3. First Experiment: Tracking performance of the proposed method with respect to online variation of the rotor resistance. (a) Control voltage, stator current, observer error and applied load torque: (i) control voltage; (ii) stator current; (iii) observer error; (iv) load torque. (b) Rotor speed, rotor ux magnitude, and estimate of the rotor resistance: (i) rotor speed; (ii) estimated rotor ux magnitude; (iii) estimate of the rotor resistance.

The experimental setup is illustrated by the block diagram of Fig. 1 which includes a development system DSP1103, an input/output electronics board (for analog/digital conversions) and a Personal Computer (PC). A 5-kW induction motor whose data are reported in the Appendix has been used. A PWM power converter with switching frequency of 10 kHz is controlled by a DSP. The external load torque is produced by a loaded dc generator. The motor instantaneous speed is measured by an optical incremental encoder with 1024 lines per revolution. The stator currents are measured by Hall-type sensors. All measured electrical parameters are converted by 16-b analog-todigital (A/D) converter channels with 1 s conversion time. A DSP1103 performs data acquisition and implements in real-time within the MATLAB/Simulink environment software with sampling time of 150 s. Three sets of experiments have been carried out. In all cases, experiments have been performed during motor startup and after the motor is operated under load torque. After the motor startup and in all experiments, the applied external unknown load torque is estimated by using the method described in [17]. In all experiments, the parameters of the rotor resistance identier (35) were chosen as follows. . The equivalent injection terms and has been approximated using rst order lowpass lter with time-constant of 5 ms. Note that the value of veries condition (33) since 10.08 s . Using expression (23), the value of the constant or is approximately 5500. Therefore, the value of also veries condition (24). Both speed and ux reference signals used in all experiments are given in Fig. 2. In the rst experiment, the performance of the algorithm to track the variation of the rotor resistance has been investigated. In this case, the online variation of the rotor resistance has been carried out using a three-phase variable rheostat and the value of the corresponding additional resistance was 0.36 . The results

obtained in this case are reported in Fig. 3. These results demonstrated that the proposed algorithm has a powerful approach to track the variation of the rotor resistance. The second experiment has been performed to verify the robustness property of the proposed method with respect to the variation of the stator resistance when the motor operates at relatively low speed. The performance of the proposed method in this case is given in Fig. 4. These results show that there is no signicant effect on the rotor resistance estimate for a wide range of variation of the stator resistance (up to 100%). Note that the assumption of non-saturated condition is often made in the literature on induction motor control. But under nominal operating condition, the induction machine will generally enter the saturation region. Therefore, the assumption that mutual inductance is constant is good only if the ux level of the machine is maintained constant and the machine operating condition is non-saturated. But in certain cases, the ux level of the machine can be varied to get better performance such as efciency improvement or eld-weakening control for higher speed operation. By taking into account this remark, the variation effect of the machine mutual inductance has been investigated. From the fact that the leakage parameter is a strictly positive constant, the mutual inductance variation should be chosen such that

or % assuming that the inductances of the stator and rotor circuits and are constant parameters (see the Appendix). The results obtained in this case are depicted in Fig. 5. As it can be seen, the proposed rotor resistance estimation algorithm is more sensitive to the variation of the machine mutual inductance than that of the stator resistance.

KENN et al.: ONLINE SIMPLIFIED ROTOR RESISTANCE ESTIMATOR FOR INDUCTION MOTORS

1193

Fig. 4. Second Experiment: Performance of the proposed method at relatively low speed (63 rad/s) with 100% variation of the stator resistance R . (a) Control voltage, stator current, observer error, and load torque: (i) control voltage; (ii) stator current; (iii) observer error; (iv) load torque. (b) Rotor speed, rotor ux magnitude, stator resistance, and estimate of the rotor resistance: (i) rotor speed; (ii) estimated rotor ux magnitude; (iii) stator resistance; (iv) estimated rotor resistance.

Fig. 5. Third Experiment: Investigation of the variation effect of the machine mutual inductance when the loaded motor is in the steady-state period. (a) Control voltage, stator current, observer error, and applied load torque: (i) control voltage; (ii) stator current; (iii) observer error; (iv) load torque (b) Rotor speed, rotor ux magnitude, mutual inductance, and estimate of the rotor resistance: (i) rotor speed; (ii) estimated rotor ux magnitude; (iii) mutual inductance; (iv) estimate of the rotor resistance.

In all cases, the estimate of the rotor resistance is very accurate and exhibits a short convergence transient. The steady-state error between the estimated rotor resistance and its nominal value is due to the measurement noise, mismatching between the motor and the model parameters, ohmic heating during experiments, and unmodeled dynamics. V. CONCLUSION In this brief, a simple structure has been designed to estimate the rotor resistance of induction motors. The proposed method

has been tested in closed-loop conguration by using a nonlinear controller which has been made adaptive with respect to the rotor resistance (35). The nite time convergence of the rotor resistance estimate to its nominal value has been achieved under mild P.E. requirements which can be fullled easily during normal operating conditions of the IM. Experimental results with online variation of the rotor resistance show that the proposed algorithm gives very satisfactory performance. The proposed online simplied rotor resistance estimator has also presented very interesting robustness properties with

1194

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 18, NO. 5, SEPTEMBER 2010

respect to the variation of the stator resistance, measurement noise, modeling errors, discretization effects and parameter uncertainties. Important advantages of the proposed algorithm include that it is an online method (the value of can be continuously updated) and it is very simple to implement in real-time (this feature distinguishes the proposed identier from the known ones). The extension of the proposed technique in speed sensorless adaptive control of IM is yet to be done.

APPENDIX A INDUCTION MOTOR DATA Rated power Rated torque Rated frequency Rated current Stator resistance Rotor resistance Stator inductance Rotor inductance Mutual inductance Number of pole pairs Motor-load inertia 5 kW 32 Nm 50 Hz 22.9 A 0.22 0.52 0.052 H 0.0516 H 0.0495 H 0.12 kg m

ACKNOWLEDGMENT The main part of the experimental setup used in this work has been supported by the Dpartement Energie, Ecole Suprieure dElectricit, Gif-sur-Yvette, Paris, France. REFERENCES
[1] R. Marino, S. Peresada, and P. Tomei, On-line stator and rotor resistance estimation for induction motors, IEEE Trans. Control Syst. Technol., vol. 8, no. 3, pp. 570579, May 2000. [2] K. Akatsu and A. Kawamura, On-line rotor resistance estimation using the transient state under the speed sensorless control of induction motor, IEEE Trans. Power Electron., vol. 15, no. 3, pp. 553560, May 2000.

[3] G. Bartolini, A. Pisano, and P. Pisu, Simplied exponentially convergent rotor resistance estimation for induction motors, IEEE Trans. Autom. Control, vol. 48, no. 2, pp. 325330, Feb. 2003. [4] P. Castaldi, W. Geri, M. Montanari, and A. Tilli, A new adaptive approach for on-line parameter and state estimation of induction motors, Control Eng. Practice, vol. 13, pp. 8194, 2005. [5] R. Marino, S. Peresada, and C. M. Verrelli, Adaptive control for speedsensorless induction motors with uncertain load torque and rotor resistance, Int. J. Adapt. Control Signal Process., vol. 19, pp. 661685, 2005. [6] M. Barut, S. Bogosyan, and M. Gokasan, Speed sensorless direct torque control of induction motors with rotor resistance estimation, Energy Conv. Manage., vol. 46, pp. 335349, 2005. [7] C. Picardi and F. Scibilia, Sliding-mode observer with resistances or speed adaptation for eld-oriented induction motor drives, in Proc. 32nd Ann. Conf. IECON, 2006, pp. 14811486. [8] Y. Koubaa, Application of least-squares techniques for induction motor parameters estimation, Math. Comput. Model. Dyn. Syst., vol. 12, pp. 363375, 2006. [9] Y. Koubaa, Asynchronous machine parameters estimation using recursive method, Simulation Model. Practice Theory, vol. 14, pp. 10101021, 2006. [10] A. Mezouar, M. K. Fellah, S. Hadjeri, and Y. Sahali, Adaptive speed sensorless vector control of induction motor using singularly perturbed sliding mode observer, in Proc. 32nd Ann. Conf. IECON, 2006, pp. 932939. [11] B. Castillo, S. D. Gennaro, A. Loukianov, and J. Rivera, Robust nested sliding mode regulation with application to induction motors, in Proc. Amer. Control Conf., New York, 2007, pp. 52425247. [12] P. Roncero-Snchez, A. Garca-Cerraba, and V. Feliu-Batlle, Rotor resistance estimation for induction machines with indirect eldorientation, Control Eng. Practice, vol. 15, pp. 11191133, 2007. [13] K. Wang, J. Chiasson, M. Bodson, and L. M. Tolbert, An online rotor time constant estimator for the induction machine, IEEE Trans. Control Syst. Technol., vol. 15, no. 5, pp. 339348, Sep. 2007. [14] A. Mezouar, M. K. Fellah, and S. Hadjeri, Adaptive sliding-mode-observer for sensorless induction motor drive using two-time-scale approach, Simulation Modelling Practice and Theory, 2008. [15] R. Marino, S. Peresada, and P. Tomei, Exponentially convergent rotor resistance estimation for induction motors, IEEE Trans. Ind. Electron., vol. 5, no. 5, pp. 508515, Oct. 1995. [16] T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and R. Ortega, A globallystable adaptive indirect eld-oriented controller for current-fed induction motors, Int. J. Control, vol. 72, pp. 9961005, 1999. [17] G. Kenne, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, and A. Arzand, Real-time speed and ux adaptive control of induction motors using unknown time-varying rotor resistance and load torque, IEEE Trans. Energy Conv., vol. 24, no. 2, pp. 375387, Jun. 2009. [18] W. Leonhard, Control of Electric Drives. New York: SpringerVerlag, 1984. [19] V. I. Utkin, Sliding Modes in Optimization and Control. New York: Springer-Verlag, 1992.

G. Kenn, R. Goma, H. Nkwawo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, Real-time transient stabilization and voltage regulation of power generators with unknown mechanical power input, in Energy Conversion and Management, Vol. 51, pp. 218224, 2010. Editor: Elsevier. Impact Factor: 2.054.

Author's personal copy

Energy Conversion and Management 51 (2010) 218224

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Real-time transient stabilization and voltage regulation of power generators with unknown mechanical power input
Godpromesse Kenn a,*, Raphal Goma b, Homre Nkwawo c, Franoise Lamnabhi-Lagarrigue b, Amir Arzand d, Jean Claude Vannier d
a

Laboratoire dAutomatique et dInformatique Applique (LAIA), Dpartement de Gnie Electrique, Universit de Dschang, B.P. 134 Bandjoun, Cameroon Laboratoire des Signaux et Systmes (L2S), CNRSSUPELEC, Universit Paris XI, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France Dpartement GEII, Universit Paris XIII, 99 Avenue Jean Baptiste Clment, 93430 Villetaneuse, France d Dpartement Energie, Ecole Suprieure dElectricitSUPELEC, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France
b c

a r t i c l e

i n f o

a b s t r a c t
A nonlinear adaptive excitation controller is proposed to enhance the transient stability and voltage regulation of synchronous generators with unknown power angle and mechanical power input. The proposed method is based on a standard third-order model of a synchronous generator which requires only information about the physical available measurements of relative angular speed, active electric power, innite bus and generator terminal voltages. The operating conditions are computed online using the above physical available measurements, the terminal voltage reference value and the estimate of the mechanical power input. The proposed design is therefore capable of providing satisfactory voltage in the presence of unknown variations of the power system operating conditions. Using the concept of sliding mode equivalent control techniques, a robust decentralized adaptive controller which insures the exponential convergence of the outputs to the desired ones, is obtained. Real-time experimental results are reported, comparing the performance of the proposed adaptive nonlinear control scheme to one of the conventional AVR/PSS controller. The high simplicity of the overall adaptive control scheme and its robustness with respect to line impedance variation including critical unbalanced operating condition and temporary turbine fault, constitute the main positive features of the proposed approach. 2009 Elsevier Ltd. All rights reserved.

Article history: Received 20 February 2009 Accepted 22 September 2009 Available online 24 October 2009 Keywords: Nonlinear control Time-varying parameter estimation Adaptive control Equivalent control Power system stabilization Power generators

1. Introduction In recent years, signicant advances have been made by both control and power systems communities to enhance the dynamic stability and voltage regulation of power systems. Whatever method is used, an important problem is to achieve both voltage regulation and system stability (damping torque) simultaneously. Over the past few decades, conventional automatic voltage regulation and power system stabilizers (AVR/PSS) have been extensively studied and successfully used in the industry. These AVR/ PSS approaches are designed using the theory of phase compensation in the frequency domain and are introduced as a lead lag compensator [14]. The parameters of AVR/PSS are determined based on the linearized model of the power system. Therefore, to provide good damping over a wide operating range, the parameters of the

* Corresponding author. Tel.: +237 77 59 52 19. E-mail addresses: gokenne@yahoo.com (G. Kenn), raphael.goma@lss.supelec.fr (R. Goma), homere.nkwawo@iutv.univ-paris13.fr (H. Nkwawo), lamnabhi@lss. supelec.fr (F. Lamnabhi-Lagarrigue), Amir.arzande@supelec.fr (A. Arzand), Jeanclaude.vannier@supelec.fr (J.C. Vannier). 0196-8904/$ - see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.enconman.2009.09.018

AVR/PSS need to be ne tuned in response to both intra-area and inter-area modes oscillations. Since power systems are highly nonlinear systems, with congurations and parameters that change with time, the AVR/PSS design based on the linearized model of the power system cannot guarantee its performance in a practical operating environment [5]. After decades of theoretical studies and eld experiments, AVR/PSS techniques have made a great contribution in enhancing the operating quality of the power system. But, with the development of power systems and increasing demand for quality electricity, it is worthwhile looking into the possibility of using modern control techniques. The linear optimal control strategy and the direct feedback linearization technique are some alternatives that has been proposed for supplementary excitation controllers [6,7,5,8]. The main drawback of optimal controller and direct feedback linearization technique is that the computation of the controllers parameters are based on a good knowledge of the linear model and the measurement of some variables that are difcult in practice, are required. In addition, currently direct feedback linearization technique cannot achieves both voltage regulation and system stability simultaneously [8]. Some recent results can be found in [914]. In [9], an new

Author's personal copy

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

219

passivity-based controller design approach for the excitation control of synchronous generators is investigated and it is shown that the proposed method can enlarge both the estimates and the actual domain of attraction, thus increases critical clearing time for faults. In [10], an intelligent based LQR controller design has been developed for power system stabilization but the voltage regulation has not been considered. In [11], an interesting method based on nonlinear feedback linearization is proposed for transient stabilization and voltage regulation of power generators with unknown parameters. The main drawback of the latter method is that, it requires the power angle reference signal to be at least C 3 and this requirement may be difcult to fulll in practice. In addition, a method/guideline for the selection of the controller parameters has not been described. In [12], a generalized neuron-based PSS and adaptive PSS have been used for power systems transient stabilization but the problem of voltage regulation has not been addressed. In [13], an output feedback controller is proposed to enhance the transient stability of nonlinear multimachine power systems using high-order sliding-mode technique and robust highorder sliding-mode differentiator. But, in this latter approach, the voltage regulation problem has not been investigated and the implementation of the control law may be difcult in practice since the switching of the control law may cause a high-frequency disturbance to power systems. In [14], Immersion and Invariance control strategy has been applied to design a nonlinear controller that provides asymptotic stabilization of a single machine innite bus system using a controllable series capacitor. But in this approach, the problems of voltage regulation and real-time experiment have not been addressed. In addition most of the above control algorithms assumes that the mechanical power input and power angle are available. But these parameters or variables are physically not available for measurement in practice. On the other hand, most of the control systems developed consider that the operating points are exactly known and are more complicated and may be difcult to implement in real-time. These are the main motivations of the proposed nonlinear control algorithm in this paper. Our main goal is to propose a new adaptive control algorithm which can achieves both voltage regulation and transient stability simultaneously and can be easily implementable in real-time. In Section 2, the power system model is described as well as the problem statement. The design procedure of the adaptive control is presented in Section 3. Experimental results are presented in Section 4 to illustrate the performance of the proposed control scheme and its robustness properties compared to those of conventional AVR/PSS. Finally, the conclusion of this paper is given in Section 5. 2. Plant model and problem statement Consider a single-machine innite-bus (SMIB) power system as shown in Fig. 1. The dynamics of this SMIB power system are presented by the following third order model [11,15]

Fig. 1. Single-machine innite-bus power system.

which is valid over the region dened by 0 < d < p. The rst two equations represent the mechanical dynamics of the power generator and the third equation gives the electrical dynamics of the power system. In (1), d is the power angle of the generator; xs is the synchronous machine speed; x is the relative rotor speed of the generator (x xg xs with xg being the generator angular speed); H is the inertia constant; D is the damping constant; xds xd xT 1 xL is 2 the total reactance which takes into account xd , the generator direct axis reactance, xL , the transmission line reactance, and xT , the reactance of the transformer; x0ds x0d xT 1 xL is the total 2 reactance which takes into account x0d , the direct axis transient reactance of the generator, xL , the transmission line reactance, and xT , the reactance of the transformer; Ef kc uf is the equivalent EMF in the excitation coil; kc is the gain of the excitation amplier; uf is the input to the SCR amplier of the generator; T 0d0 is the direct axis transient open circuit time constant; V s is the innite bus voltage; P e is the active electrical power and Pm is the mechanical power input. The generator terminal voltage modulus is given by [11]

Vt

x2 V 2 2xs xd Pe cotd d2 s 2 xds xds V 2 sin d s x2 P 2 s e

!1 2 ; 2

where xs xT 1 xL is the total line reactance which takes into ac2 count xT , the reactance of the transformer and xL , the transmission line reactance. The following assumptions will be considered until further notice: (i) It is assumed that the outputs/states of the system are continuous and bounded. (ii) It is also assumed that the value of the mechanical power Pm _ varies slowly such that P m can be negligible or small with respect to other existing dynamics. Remark 1. In the model described by (1) the power angle d and the mechanical power input P m are assumed to be not available for measurement. In the following analysis (Section 3) a new adaptive control for transient stabilization and voltage regulation of power generators when the mechanical power input is not available is presented. A nite time convergence algorithm for the estimation of P m is described (Section 3) and the power angle d estimation is given by the expression derived in [11] as follows:

_ dx Pe Pm & ! xd x0d xds xds V s sind _ Pe 0 0 Pe 0 0 kc uf T 0d0 V s x sind xds xds xds T d0 xds T d0 ' x0ds T 0d0 P e x cotd xds & ! xds x0ds xds xds V s sind 0 0 Pe 0 0 kc uf T 0d0 V s x sind xds xds xds T d0 xds T d0 ' x0ds T 0d0 P e x cotd ; 1 xds H _ x x D H

xs

d arccot

Vs xd V s xs P e xds

s!! x2 : V 2 s2 P2 t e Vs

The objective of the control is to achieve both transient stability and voltage regulation. These objectives may be summarized as

2
t!1

6 7 6 7 lim 4 Pe 5 4 Pm 5; V tref Vt

3 with 0 < d < p: 4

Author's personal copy

220

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

A reduction from Pm to P mf of the mechanical power generated by the turbine [11], changes the operating condition (e.g. from ds to ds f ) and the new value of the power angle is a function of Pm and Pmf . Since Pm and P mf are typically unknown, the corresponding new stable power angle ds f is also unknown. As a consequence, the control system must recover this new operation point, generate a trajectory towards it, and drive the system to this trajectory. To achieve these control goals using the third order model (1), the following control objectives is considered

_ The time derivative P , computed using (12) and taking into ace count assumption (ii) is
D 1 k0 H a2 H _ P S Dx xs Pe Pm : e k 0 xs k0 xs

15

_ _ Substituting P e and P using (7) and (15) in (14), yields e

2
t!1

6 7 6 7 lim 4 x 5 4 0; 5; with 0 < d < p Pe Pm ~ or lim x 0; with 0 < d < p;

t!1

dref

3 5 6

xds xds a2 H _ V P e P 0 0 P e 0 0 v f S e k 0 xs xds T d0 xds T d0 ! D 1 k0 H Dx xs Pe Pm k0 xs

By chosen

vf

as

where dref is computed using (3) by substituting Pe and V t by P m and V tref , respectively. Let ~T d dref ; x x; Pe Pm be the vector of the tracking error. 3. Description of the adaptive control First, we suppose that the mechanical power input in system (1) is known. In order to derive the control law, we recall Eq. (1) and rewrite it as follows

D 1 k0 H a2 Hx0ds T 0d0 v f Pe S x0ds T 0d0 Dx xs P e Pm kP e P e xds k0 xs xds k0 xs 17 & x0ds T 0d0 aH ~ xds or kc uf d k0 xa k Pe V s sind xds k0 xs   H D D x k x s Pe Pm kx 0 xs H H #) 2 V 2 sin d kP e Pm x s xds x0ds P e x cotd 18 xds x0ds

_ d x; D x _ x x s Pe Pm ; H H xds xds _ Pe 0 0 Pe 0 0 v f ; xds T d0 xds T d0

where

the time-derivative of V can be rewritten as

_ V kP e P 2 : e

19

vf

! xds x0ds V s sind kc uf T 0d0 V s x sind xds xds 0 0 xds P e x cotd; T d0 xds

From (19) we can conclude that P e converges exponentially to P . Consequently, S converges exponentially to 0, and the power e angle d, the relative rotor speed x also converge exponentially to dref and 0, respectively. By virtue of the relations (3) and (10), the generator terminal voltage V t and the active electric power Pe converge to their reference value V tref and P m . Remark 2. In the above controller (18), it has been assumed that the mechanical power input P m is available. Since P m is typically unknown, on-line adaptation law for Pm is required to complete the design. To this end, we propose a nite time convergence estimation algorithm of the mechanical power input. The third order model of the power generator is rewritten as follows

has been used to simplify the notation. Consider the following coordinate transformation

S d dref k0 x ~ k0 x; d

where k0 > 0 is a design parameter and ~ is the power angle trackd ing error. The trajectory dref is given by

_ x

xs
H

D x x s Pe ; H H

h Pm :

20

dref

Vs xd V s arccot xs P m xds

s!! x2 : V 2 s2 P2 tref m Vs

Next, we consider the following adaptive observer

10

_ ^ x

xs ^
H

To ensure the boundedness of the trajectory in the new coordinate S and also that lim St 0, we take
t!1

D x x s Pe v ; H H

21

_ S aS;

a > 0:
P e

11
which is the unique

^ where v ksignx x is the input of the tuning model (21) and k > 0 is a design parameter. ^ Dening ex x x and eh ^ h, the dynamic of the obserh ver error can be computed using (20) and (21) as

Next, we calculate the reference signal solution of (11) as follows

_ ex 12

xs
H

 ! H D P aS x 1 k0 Pm ; e k0 xs H

eh v :

22

To achieve the design of the estimator, the following additive assumptions are required. Assumption (iii). It is assumed that the following parameter identiability condition holds.

assuming that assumption (ii) is satised. Our goal now is to nd the control input kc uf such that the closed-loop system is exponentially stable at the operating point ~ 0. To this end, consider the followx ing Lyapunov candidate function

xs
H

P n > 0:

23

2 1 Pe P ; e 2

13

whose time derivative is

Assumption (iv). It is assumed that the gain of the adaptive observer (21) is chosen such that

  _ _ _ V Pe P Pe P : e e

14

kP

xs
H

jeh jmax :

24

Author's personal copy

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

221

Under normal operating conditions of the system, the identiability condition (23) is always satised. Under assumption (iv), a sliding mode regime will occur on the manifold ex 0 in nite time. In this case, the dynamic equation of the adaptive observer error can be rewritten as

xs
H

eh v eq ;

 The tuning parameters a and k should not have the same value by virtue of the presence of the factor a k in the relation (18).  In the case where a k, the global convergence and stabilization of the system can not be achieved since the control input in this case will be independent of the power angle tracking error ~ d. 4. Experimental results

25

where v eq is the equivalent control which cannot be computed since the estimation error eh is not available. Using the average control as the approximation of the equivalent control v eq [16,17], we consider the following adaptation law for h

  _ ^ kh sign Hv eq ; h

xs

26

where kh > 0 is the tuning parameter of the above adaptation law. Under assumption (ii), the dynamic of the parameter estimation error can be computed as

_ eh kh sign

  Hv eq :

xs

27

The estimator synthesis can be summarized by the following theorem. Theorem 1. Given the reference model (20), and the adaptive observer (21); assuming that the parameter of the observer (21) is selected according to (24) with assumptions (i)(iii) and (23) satised, then the estimate ^ given by (26) converges in nite time to its h, nominal value h. The proof of the above theorem can be found in [17]. The structure of the proposed adaptive nonlinear controller (ANLC) is depicted in Fig. 2. 3.1. Selection of the controller parameters It is worth observing that, in order to increase the tracking performance of the above controller a rigorous methodology for the optimal tuning of the controller design parameters is required. However, it is difcult to dene such a methodology, since the overall adaptive control is nonlinear with parameter that may vary with time (i.e., Pm ). In the following, some guidelines for the selection of the controller parameter values are given using some considerations.  The values of k0 ; a and k should be chosen such that the control system always presents acceptable output behaviour. During experiment, it has been noticed that the number of oscillations is proportional to 1=k0 . Thus the choice of k0 is very important for damping oscillations.

The effectiveness of the proposed adaptive nonlinear control algorithm has been veried by experiments in various operating conditions. The experimental setup is illustrated by the block diagram of Fig. 3 which includes a development system DSP1103, an input/output electronics board (for analog/digital conversions) and a Personal Computer (PC). This PC is used to program the DSP, to store and display experimental data. The three phase electrical network used as innite bus is the local Electricit De France (EDF) low voltage network (136 V/ 50 Hz). The synchronous generator connected to EDF network whose data are reported in the Appendix A, has been used. The physical limit of the excitation voltage is jkc uf j 6 2 with kc 4. A voltage controlled dc motor provides the mechanical power input. The angular speed is measured by a tachometer with speed to voltage ratio constant of 1000 rpm/60 V. The speed measured noise is properly ltered by a very low-pass rst-order lter. The stator currents and voltages are measured by Hall-type sensors and isolated voltage sensors, respectively. All measured electrical parameters are converted by 16-b A/D converter channels with 1 ls conversion time. A DSP1103 performs data acquisition, implements control algorithms, and generates reference signals in realtime within the Matlab/Simulink environment software with sampling time of 0.5 ms. Three sets of experiments have been performed. The operating points in all experiments are computed online with V tref 1:01 p:u: The control parameters have been selected according to the procedure described in Section 3.1 and a ner tuning has been performed and nally the following tuning parameters were used

k0 1:5;

k 15 and a 10:

In all experiments the controller (18) has been made adaptive with respect to the mechanical power input by means of the adaptive identier (26). The parameters of the estimator of P m were chosen as follows. k 10 and kh 0:5. The equivalent control has been approximated using rst order low-pass lter with time-constant of 1 ms.

Fig. 2. Structure of the proposed adaptive nonlinear controller (ANLC).

Fig. 3. Schematic of the experimental setup.

Author's personal copy

222

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

The rst experiment has been performed to verify the robustness property with respect to temporary turbine fault. This experiment has been carried out according to the following sequence. 1. At t 0, the system is in pre-fault state. 2. At t 0:5 s, a temporary turbine fault occurs. 3. At t 1:5 s, the turbine fault is removed. In the second experiment, the effect of temporary fault on the transmission line is veried using the following sequence. 1. At t 0, the system is in pre-fault state. 2. At t 1 s, a fault (open-circuit) occurs in one of the three transmission lines.

3. At t 1:3 s, the transmission line is restored and the system is in a post-fault state. In the third experiment, the robustness property with respect to a permanent fault is tested. This experiment has been carried out according to the following sequence. 1. At t 0, the system is in pre-fault state. 2. At t 1 s, a fault (open-circuit) occurs in one of the three transmission lines and the system operates under a critical unbalanced condition. The results are shown in Figs. 47.

i Control voltage (degree)

1.6 k u
c f

i Power angle
24 22 20 18 16 14 12 0

k u (p.u.)

1.5 1.4 1.3 1.2 1.1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

ref 1 2 3 4 5 6 7 8 9 10

ii Terminal and infinite bus voltage

1.02

V & V (p.u.)

1 0

1.01 Vs 1 0 1
3

ii Active electric power and mechanical power input

0.4

V 2 3 4 5 6 7 8 9

Pe and Pm (p.u.)

10

0.35

6 Pe Pm Pm estimate

x 10

iii Relative speed

(p.u.)
14 15

0.3

10

11

12

13

2 0

10

Time (s)

Time (s)

(a)

(b)

Fig. 4. Test of the robustness with respect to turbine temporary fault (experiment 1). (a) i Excitation control input; ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; iii Relative speed.

i Control voltage
1.5

i Power angle
22

kcuf

(degree)

1.6

20

kcuf (p.u.)

1.4 1.3 1.2 1.1 1 0 1 2 3 4 5 6 7 8 9 10

ref

18 0

10

11

12

13

14

ii Terminal and infinite bus voltage

1.01

V (p.u.)

Vt V
s

ii Active electric power and mechanical power input

0.4

1 0.99 0

Pe and P (p.u.)

0.389

10

11

12

13

14

6
Pm Pm estimate 0.355 0 1 2 3 4 5 6 7 8 9 10

x 10

iii Relative speed

m 0.372

(p.u.)

4 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Time (s)

(a)

Time (s)

(b)

Fig. 5. Test of the robustness with respect to the transmission line temporary fault (experiment 2). (a) i Excitation control input; ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; iii Relative speed.

Author's personal copy

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

223

i Power angle
21

(degree)

20 19 18

i Control voltage
1.6 1.5

ref

kcuf

kcuf (p.u.)

1.4 1.3 1.2 1.1 1 0 1 2 3 4 5 6 7 8

ii Terminal and infinite bus voltage

1.01

V (p.u.)

1 0.99 0 1 2 3 4 5 6 7

Vt V
s

ii Active electric power and mechanical power input

0.4

P and P (p.u.)

0.38 6 0.36 0.34 0.32 0.3 0 1 2 3 4 5 6 P 5 4 3 2 1

x 10

iii Relative speed

Pm Pm estimate 7 8

( p.u.)

Time (s)

(a)

Time (s)

(b)

Fig. 6. Test of the robustness with respect to the transmission line permanent fault (experiment 3: critical unbalanced operation of the power generator). (a) i Excitation control input; ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; iii Relative speed.

i Control voltage
k u (p.u.) c f

i Control voltage
2

1.6 1.4 1.2 1 0.8 0 1 2 3 4 5 6 k u AVR/PSS

c f c f

kc u (p.u.)

1.5 1 0.5 0 1 2 3 4

k u ANLC 7 8

kcuf AVR/PSS kcuf ANLC

5

ii Terminal voltage
1.013
V (p.u.)

ii Terminal voltage
Vt AVR/PSS

V (p.u.) t

1.008

Vt ANLC

1.01 1 0.99

V AVR/PSS
t

1.003 0
3

Vt ANLC
0
3

6
(p.u.)

x 10

iii Relative speed

6

x 10

iii Relative speed

AVR/PSS ANLC

4 AVR/PSS ANLC 1 2 3

(p.u.)

4 2 0 1 2 3 4

2 0

Time (s)

Time (s)
Fig. 8. Test of the robustness with respect to the transmission line temporary fault (experiment 2): Comparative results. ANLC (solid line). AVR/PSS (dashed line).

Fig. 7. Test of the robustness with respect to temporary turbine fault (experiment 1): Comparative results. ANLC (solid line). AVR/PSS (dashed line).

In all cases, the proposed nonlinear adaptive controller (ANLC) quickly regulates the generator terminal voltage to its pre-fault value (experiments 1 and 2) and to the neighborhood of the pre-fault value (experiment 3: permanent fault and critical unbalanced operating condition). In addition, the control system presents admissible peak overshoots values and permissible number of oscillations to settling time. For comparison, the above three sets of experiments have been performed by using the conventional AVR/PSS method. The comparative results are reported in Figs. 79.

From these comparative results, we can notice that the proposed ANLC achieves very good tracking performance without sustained oscillations while AVR/PSS damps oscillations in the rst and second experiments and exhibits sustained undesirable oscillations in the third experiment. 5. Conclusion In this paper, a nonlinear adaptive excitation controller has been proposed to stabilize and regulate in nite time the voltage

Author's personal copy

224

G. Kenn et al. / Energy Conversion and Management 51 (2010) 218224

i Control voltage
3

Appendix A. System parameters

Synchronous speed Damping constant Inertia constant Direct axis transient open circuit time constant Generator direct axis reactance Generator direct axis transient reactance Transmission line reactance

k u (p.u.) c

2 1 0 0 1 2 3 4 5 6 7

kcuf AVR/PSS kcuf ANLC

8 9 10

xs 314:159 rad s1 D 0:1 p:u: H 0:576 s

T 0d0 0:44 s xd 0:894 p:u: x0d 0:620 p:u: xL 0:294 p:u:

ii Terminal voltage
V (p.u.)
1.01 1 0.99

V AVR/PSS
t

Vt ANLC
0 x 10
3

10

References
[1] Yu YN. Electric power system dynamics. New York: Academic Press; 1983. [2] Anderson PM, Fouad AA. Power system control and stability. NJ: IEEE Press; 1994. [3] Kundur P. Power systems stability and control. New York: McGraw-Hill Inc.; 1994. [4] Law KT, Hill DJ, Godfrey NR. Robust controllers structure for coordinated power system voltage regulator and stabilizer design. IEEE Trans Control Sys Technol 1994;2:22032. [5] Cao Y, Jiang L, Chen S, Chen D, Malik OP, Hope GS. A nonlinear variable structure stabilizer for power system stability. IEEE Trans Energy Convers 1994;9:48995. [6] Gao L, Chen L, Fan Y, Ma H. A nonlinear control design for power systems. Automatica 1992;28:9759. [7] Wang Y, Hill DJ, Gao L, Middleton RH. Transient stability enhancement and voltage regulation of power system. IEEE Trans Power Syst 1993;8:6207. [8] Zhu C, Zhou R, Wang Y. A new nonlinear voltage controller for power systems. Electr Power Energy Syst 1997;19:1927. [9] Galaz M, Ortega R, Bazanella AS, Stankovic AM. An energy-shaping approach to the design of excitation control of synchronous generators. Automatica 2003;39:1119. [10] Ko HS, Lee KY, Kimc HC. An intelligent based lqr controller design to power system stabilization. Electr Power Syst Res 2004;71:19. [11] Damm G, Marino R, Lamnabhi-Lagarrigue F. Adaptive nonlinear output feedback for transient stabilization and voltage regulation of power generators with unknown parameters. Int J Robust Nonlinear Control 2004;14:83355. [12] Chaturvedia DK, Malikb OP. Generalized neuron-based pss and adaptive pss. Control Eng Practice 2005;13:150714. [13] Colbia-Vega A, De Len-Morales J, Fridman L, Salas-Pea O, Mata-Jimnez MT. Robust excitation control design using sliding-mode technique for multimachine power systems. Electr Power Syst Res 2008;78:162734. [14] Manjarekar NS, Banavar RN, Ortega R. Nonlinear control synthesis for asymptotic stabilization of the swing equation using a controllable series capacitor via immersion and invariance. In: Proceedings of the 47th IEEE conference on decision and control, Cancun, Mexico, December 911, 2008. p. 249398. [15] Bergen AR, Vittal V. Power systems analysis. Englewood Cliffs (NJ): PrenticeHall; 2000. [16] Utkin VI. Sliding modes in optimization and control. Springer-Verlag; 1992. [17] Kenn G, Ahmed-Ali T, Lamnabhi-Lagarrigue F, Arzand A. Nonlinear systems time-varying parameter estimation: application to induction motors. Electr Power Syst Res 2008;78:18818.

iii Relative speed

AVR/PSS ANLC

(p.u.)

4 2 0 1 2 3 4 5 6 7 8

10

Time (s)
Fig. 9. Test of the robustness with respect to the transmission line permanent fault (experiment 3: critical unbalanced operation of the power generator): Comparative results. ANLC (solid line). AVR/PSS (dashed line).

of synchronous generators assuming that the power angle and mechanical power input are not available for measurement. The controller recovers by adaptation the unknown values (power angle and mechanical power) and simultaneously generates trajectories to be followed by the states, that converge exponentially to the new equilibrium point. Real-time experimental results have been performed to illustrate the performance of the proposed adaptive controller. Three experiments have been performed. In all cases, the proposed ANLC has shown very good tracking performance without sustained oscillations while AVR/PSS damps oscillations in the rst and second experiments (temporary turbine and impedance line perturbations) and exhibits undesirable sustained oscillations in the third experiment. The high simplicity of the overall ANLC and its robustness with respect to line impedance variation (including critical unbalanced operating condition and temporary turbine fault) constitute the main positive features of the proposed approach. Acknowledgements The main part of the experimental setup used in this work has been supported by the Dpartement Energie, Ecole Suprieure dElectricit, Gif-sur-Yvette, Paris, France.

G. Kenn, R. Goma, H. Nkwawo, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An improved direct feedback linearization technique for transient stability enhancement and voltage regulation of power generators, in International Journal of Electrical Power and Energy Systems, Vol. 32, pp. 809-816, 2010. Editor: Elsevier. Impact Factor: 2.073.

Author's personal copy

Electrical Power and Energy Systems 32 (2010) 809816

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

An improved direct feedback linearization technique for transient stability enhancement and voltage regulation of power generators
Godpromesse Kenn a,*, Raphal Goma c, Homre Nkwawo b, Franoise Lamnabhi-Lagarrigue c, Amir Arzand d, Jean Claude Vannier d
a

Laboratoire dAutomatique et dInformatique Applique (LAIA), Dpartement de Gnie Electrique, Universit de Dschang, B.P. 134 Bandjoun, Cameroun Dpartement GEII, Universit Paris XIII, IUT Villetaneuse, 99 Avenue Jean Baptiste Clment, 93430 Villetaneuse, France Laboratoire des Signaux et Systmes (L2S), CNRSSUPELEC, Universit Paris XI, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France d Dpartement Energie, Ecole Suprieure dElectricitSUPELEC, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France
b c

a r t i c l e

i n f o

a b s t r a c t
In this paper, a simple improved direct feedback linearization design method for transient stability and voltage regulation of power systems is discussed. Starting with the classical direct feedback linearization technique currently applied to power systems, an adaptive nonlinear excitation control of synchronous generators is proposed, which is new and effective for engineering. The power angle and mechanical power input are not assumed to be available. The proposed method is based on a standard third-order model of a synchronous generator which requires only information about the physical available measurements of angular speed, active electric power and generator terminal voltage. Experimental results of a practical power system show that fast response, robustness, damping, steady-state and transient stability as well as voltage regulation are all achieved satisfactorily. 2010 Elsevier Ltd. All rights reserved.

Article history: Received 24 February 2009 Received in revised form 8 December 2009 Accepted 28 January 2010

Keywords: Feedback control Linearization technique Parameter estimation Nonlinear adaptive control Power system stabilization Voltage regulation

1. Introduction Operating conditions of modern large-scale power systems are continuously varying in order to satisfy different load demands. Power system stabilization has been dealt with for many years by both control and power systems communities. The goal is to design controllers for synchronous generators such that they provide within stringent tolerances, an uninterrupted supply of power at a specic frequency and voltage to the loads present in power systems. On the other hand, control system teams develop quite more complicated algorithms which may be difcult to implement in the industry. This justies the fact that automatic voltage regulation and power system stabilizer (AVR/PSS) have been extensively used in modern power systems to enhance system damping and to improve the overall system stability [14]. But since the AVR/PSS are designed separately, both voltage regulation and system stability enhancement is difcult to achieve simultaneously [4]. Moreover, the controllers are designed according to approximately linearized

* Corresponding author. Tel./fax: +237 77 59 52 19. E-mail addresses: gokenne@yahoo.com (G. Kenn), raphael.goma@lss.supelec.fr (R. Goma), homere.nkwawo@iutv.univ-paris13.fr (H. Nkwawo), lamnabhi@lss. supelec.fr (F. Lamnabhi-Lagarrigue), Amir.arzande@supelec.fr (A. Arzand), Jeanclaude.vannier@supelec.fr (J.C. Vannier). 0142-0615/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijepes.2010.01.018

power systems models which are dependent on certain operating conditions and will suffer from performance degeneracy when the operating condition changes [5]. In [4], internal model control (IMC) theory is used to propose a robust coordinated AVR/PSS which allows an effective trade-off between voltage regulation and damping improvement but the problem of linearized models remains. In order to improve power system stability and performance, nonlinear excitation controllers using feedback linearization technique have been proposed [58]. The problem of the direct feedback linearization technique (DFL) controller is that it cannot achieve voltage regulation and system stability simultaneously [8]. To overcome this problem, switching of different feedback control laws are proposed in [7]. But the switching of different feedback control laws may cause a high-frequency disturbance to power systems. Another solution to this voltage regulation problem is to use the DFL technique and robust control theory [8]. Some recent results can be found in [916]. In [9], a nite-element model of a turbine-generator innite-busbar system is used to assess the robustness and performance capabilities of a PSS/ AVR scheme. The numerical model incorporates simulation of rotor motion, iron magnetic nonlinearity and eddy currents in the solid rotors of turbine generators. In order to circumvent the lack of real power systems, nite-element models are suitable tools of

Author's personal copy

810

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

validation of new control designs under nearly true conditions that exist in real plants. It is shown that robust control designs, obtained with small order models, keep their good performance characteristics when applied to the actual machine. In [10], an new passivity-based controller design approach for the excitation control of synchronous generators is investigated and it is shown that the proposed method can enlarge both the estimates and the actual domain of attraction, thus increases critical clearing time for faults. Unfortunately, the passive output is a nonlinear function of the system states and the commonly unknown operating point makes unfeasible the direct implementation of passive controllers [13]. In [11], an intelligent based LQR controller design has been developed for power system stabilization but the voltage regulation has not been considered. In [12], an interesting method based on nonlinear feedback linearization is proposed for transient stabilization and voltage regulation of power generators with unknown parameters. The main drawback of the latter method is that, it requires the power angle reference signal to be at least C 3 and this requirement may be difcult to fulll in practice. In addition, a method/guideline for the selection of the controller parameters has not been described. In [13], an excitation controller for a single generator based on modern multi-loop design methodology is presented. It has been shown that the proposed two-loop excitation control scheme enlarges the potential region of attraction of the operating point. Therefore, this approach is able to provide local damping of swing oscillations and asymptotic terminal voltage regulation about a given setpoint. In [14], a generalized neuronbased PSS and adaptive PSS have been used for power systems transient stabilization but the problem of voltage regulation has not been addressed. In [15], an output feedback controller is proposed to enhance the transient stability of nonlinear multimachine power systems using high-order sliding-mode technique and robust high-order sliding-mode differentiator. But in this latter approach, the voltage regulation problem has not been investigated and the implementation of the control law may be difcult in practice since the switching of the control law may cause a high-frequency disturbance to power systems. In [16], Immersion and Invariance control strategy has been applied to design a nonlinear controller that provides asymptotic stabilization of a single machine innite bus system using a controllable series capacitor. But in this approach, the problems of voltage regulation and realtime experiment have not been addressed. Moreover, most of control algorithms assume that the mechanical power and power angle are available by measurements. For the measurement of power angle, the main problem is the detection of the rotor position. Various techniques have been used [1721]. The power angle instrument presented in [17,18], uses a toothed gear mounted on the rotor shaft, with a magnetic type toothed gear pickup to detect the rotor position. The technique described in [19], uses an optical encoder to detect the rotor position. But in some cases, mounting a toothed gear on the rotor shaft can be very complicated, and coupling an encoder is almost impossible [21]. An approach to power angle measurement discussed in [20] is based on a photoelectric sensor installed on the stator to detect the rotor position and has been tested on a small laboratory machine. A new approach discussed in [21], was developed for power angle determination of the salient pole synchronous machine and is based on air gap measurements. However, the value obtained under steady-state condition using this technique, is corrupted with periodical oscillations which can affect the accuracy of the power angle measurement. The zero phase lter [22] can be used to eliminate these oscillations in order to improve the accuracy but this technique is only convenient for off-line applications. Moreover, the presence of these sensors increases the cost and complexity of the control system and reduces the robustness of the overall system.

In addition, most of the control systems developed consider that the operating points are exactly known and are more complicated and may be difcult to implement in real-time. While most of the above control algorithms have been successfully proven in numerical simulation, there is still a lack of experimental results which are the only way to verify the effectiveness of those methods. These are the main motivations of the proposed nonlinear control algorithm in this paper. Our main goal is to propose an improved DFL adaptive control algorithm which can achieve both voltage regulation and transient stability simultaneously and can be easily implementable in realtime by using the works of [12,23]. The third-order nonlinear model for power generator [24] is used and the power angle and mechanical power input are not assumed to be available. The main advantage of the proposed method is that when an equivalent linear system is obtained, a variety of proven linear-control design techniques can be applied to complete the control design. In Section 2, the dynamic model of the power system is described as well as the problem statement. The design procedure of the improved direct feedback linearization technique is presented in Section 3. Experimental results are presented in Section 4 to illustrate the performance of the proposed control scheme and its robustness properties compared to those of AVR/PSS scheme. Finally, in Section 5, some concluding remarks end the paper. 2. Dynamic model description Consider a single-machine innite-bus (SMIB) power system as shown in Fig. 1. The dynamics of this single-machine innite-bus power system are given by the following third-order model [2,25,24]

_ d x; P e Pm ; & xds xds V s sind _ Pe 0 0 Pe 0 0 kc uf : xds xds T d0 xds T d0 ! ' xds x0ds x0 T 0d0 V s x sind T 0d0 ds Pe xcotd ; xds xds H _ x x D H

xs

which is valid over the region dened by 0 < dt < p. The rst two equations represent the mechanical dynamics of the power generator and the third equation gives the electrical dynamics of the power system. In (1), d is the power angle of the generator; xs is the synchronous machine speed; x is the relative rotor speed of the generator (x xg xs with xg being the generator angular speed); H is the inertia constant; D is the damping constant; xds xd xT 1 xL is 2 the total reactance which takes into account xd , the generator direct axis reactance, xL , the transmission line reactance, and xT , the reactance of the transformer; x0ds x0d xT 1 xL is the total 2 reactance which takes into account x0d , the direct axis transient reactance of the generator, xL , the transmission line reactance,

Fig. 1. Single-machine innite-bus power system.

Author's personal copy

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

811

and xT , the reactance of the transformer; Ef kc uf is the equivalent EMF in the excitation coil; kc is the gain of the excitation amplier; uf is the input to the SCR amplier of the generator; T 0d0 is the direct axis transient open-circuit time-constant; V s is the innite bus voltage; Pe is the active electrical power and P m is the mechanical power input. The generator terminal voltage modulus is given by [12]

kc uf

x0 xds v f T 0d0 ds Pe Dxcotd Pm V s sind xds 0 0 xds xds V s Dx sind: T d0 xds

By taking into account Eq. (5) and using (6), the nonlinear power system model (1) can be re-rewritten as follows:

Vt

!1 2 x2 V 2 2xs xd Pe cotd ; d2 s 2 2 xds xds V s sin d x2 P 2 s e

where xs xT 1 xL is the total line reactance which takes into ac2 count xT , the reactance of the transformer and xL , the transmission line reactance. The following assumptions will be considered until further notice: Assumption (i). It is assumed that the outputs/states of the system are continuous and bounded. Assumption (ii). It is also assumed that the value of the mechan_ ical power P m varies slowly such that Pm can be negligible or small with respect to other existing dynamics. Remark 1. In the model described by (1) the power angle d and the input mechanical power Pm are assumed to be not available for measurement. In the following analysis (Section 3) a new adaptive direct feedback linearization technique control scheme for transient stabilization and voltage regulation of power generators when the power angle and mechanical power input are not assumed to be available for measurement, is presented. A nite time convergence algorithm for the estimation of Pm will be used (Section 3) while the power angle d calculation will be obtained as in [12]. According to this reference, the value of d can be computed by using physical available states as (see Appendix A for the derivation of Eqs. (2) and (3))

_ Dd Dx; D xs _ Dx Dx DP e ; H H xds xds _ DPe 0 0 DP e 0 0 v f ; xds T d0 xds T d0

where with:

vf

is the new input for the above DFL compensated model

Dd d dref ; Dx x g x s

and DPe P e Pm :

After linearization, any robust linear control theory can be applied to the model (8) to design the feedback control law v f . Remark 2. The DFL nonlinear controller (7) with v f given by any robust linear control theory can achieve effective transient stability enhancement but cannot achieve voltage regulation. The problems of this DFL nonlinear controller have been studied in [8]. To solve the problem, we propose an improved DFL technique with adaptation of the mechanical power input by using the works of [12,23]. 3.2. Improved DFL strategy The above DFL nonlinear controller cannot achieve voltage regulation since V t t is a nonlinear function of dt; Pe t and the system structure. To improve the DFL nonlinear controller such that it can achieve both transient stability and voltage regulation, we dene an equivalent control objective as follows:

Vs xd V s d arccot xs P e xds

s!! x2 : V 2 s2 P2 t e Vs

t!1

The objective of the control is to achieve both transient stability and voltage regulation. These objectives can be summarized as

6 7 6 7 lim 4 Dx 5 4 0 5; Pe Pm
t!1

dref

with 0 < dt < p;

10 11

0 x 6 7 6 7 lim 4 Pe 5 4 Pm 5; t!1 V tref Vt

or

lim Dx 0;

with 0 < dt < p;

with 0 < dt < p:

where dref is computed using (3) by substituting Pe ; V t by Pm ; V tref , respectively, and DxT d dref ; Dx; P e Pm is the vector of the tracking error. We then obtain:

3. Adaptive DFL nonlinear controller 3.1. Actual DFL technique In this section, we describe the DFL technique under the assumption that the power angle and the mechanical power input are available. If we let

dref arccot

Vs xd V s xs P m xds

s!! x2 : V 2 s2 P2 tref m Vs

12

vf

! xds x0ds V s sind kc uf T 0d0 V s Dx sind T 0d0 xds xds x0ds P e Dxcotd Pm ; xds

where Dx x xg xs , the second equation of (1) becomes

Therefore, a reduction from Pm1 to P m2 of the mechanical power generated by the turbine changes the operating condition from dref 1 to dref 2 but with the generator terminal voltage reference value v tref being constant in both operating conditions. As a consequence, the control system must recover the new operating point, generate a trajectory towards it, and drive the system to this trajectory. The new DFL nonlinear controller is computed using (7) with v f given by any robust linear control theory and the power angle d and its reference value being computed online using (3) and (12). Therefore, the new DFL nonlinear controller which is valid over the range 0 < dt < p, can achieve both effective transient stability enhancement and voltage regulation. Remark 3. In the above control design, it has been assumed that the mechanical power input P m is available. Since P m is typically unknown, on-line adaptation law for Pm is required to complete the design. In [24], an estimation method of P m is proposed but this technique is only valid under steady-state conditions.

xds xds _ Pe 0 0 Pe Pm 0 0 v f : xds T d0 xds T d0

Thus the third model (1) has been linearized. The mapping (5) is invertible over the normal working region 0 < dt < p. The DFL compensation law can be deduced from (5) as follows:

Author's personal copy

812

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

To this end, we propose a nite time convergence estimation algorithm of the mechanical power input by using the work of [23]. The third-order model of the power generator is rewritten as follows:

In this paper, we choose the adaptive feedback control law u as

u kd Ddt kx Dxt kPe DPe t

21

D x _ x h x s Pe ; H H H

xs

h Pm :

13

Next, we consider the following adaptive observer

and select the feedback gains ki ; i d; x; Pe by using the poles placement technique. The new DFL compensated law is simple and practically realizable since only information about measured variables xg , V t and Pe are required. 4. Experimental results The experimental setup is illustrated by the block diagram of Fig. 2 which includes a development system DSP1103, an input/ output electronics board (for analog/digital conversions) and a Personal Computer (PC). This PC is used to program the DSP, to store and display experimental data. The three phase electrical network is the local Electrict De France (EDF) low voltage network (136 V/50 Hz). The synchronous generator connected to EDF network whose data are reported in Appendix B has been used. The machine parameters have been obtained by using off-line methods (e.g. no load test, slip test, open-circuit test, and step response test). The physical limit of the excitation voltage is jkc uf j 6 2 with kc 4. A voltage controlled dc motor provides the mechanical power input. The angular speed is measured by a tachometer with speed to voltage ratio constant of 1000 rpm=60 V. The speed measured noise is properly ltered by a rst order low-pass lter with cutoff frequency of 1 rad=s. The stator currents and voltages are measured by Hall-type sensors and isolated voltage sensors, respectively. All measured electrical parameters are converted by 16-b A/D converter channels with 1 ls conversion time. A DSP1103 performs data acquisition, implements control algorithms, and generates reference signals in real-time within the Matlab/Simulink environment software with sampling time of 0.5 ms. Three experiments have been carried out in order to verify the effectiveness of the proposed approach. In all experiments, the operating points are computed online with V tref 1 p:u: The parameters of the controller have been selected by using poles placement technique and are given as follows: kd 6:4539, kx 7:4151 and kPe 2:8340. In all experiments the nonlinear controller (7) has been made adaptive by using the mechanical power identier (19) and the parameters of the estimator of P m were chosen as follows: k 10 and kh 0:5. The equivalent control has been approximated using rst order low-pass lter with time-constant of 1ms. In the rst experiment, the effect of temporary turbine fault has been tested by using the following sequence:

_ ^ x

xs ^
H

D x x s Pe v ; H H

14

^ where v k signx x is the input of the tuning model (14) and k > 0 is a design parameter. ^ Dening ex x x and eh ^ h, the dynamic of the obserh ver error can be computed using (13) and (14) as

_ ex

xs
H

eh v :

15

To achieve the design of the estimator, the following additive assumptions are required: Assumption (iii). It is assumed that the following parameter identiability condition holds.

xs
H

P n > 0:

16

Assumption (iv). It is assumed that the gain of the adaptive observer (14) is chosen such that

kP

xs
H

jeh jmax ;

17

where jeh jmax is the maximum value of the mechanical power input estimation error. Under normal operating conditions of the system, the identiability condition (16) is always satised. Under Assumption (iv), a sliding-mode regime will occur on the manifold ex 0 in nite time. In this case, the dynamic equation of the adaptive observer error can be rewritten as

xs
H

eh v eq ;

18

where v eq is the equivalent control which cannot be computed since the estimation error eh is not available. Using the average control as the approximation of the equivalent control v eq [26,23], we consider the following adaptation law for h

  _ ^ kh sign Hv eq kh signeh ; h

xs

19

where kh > 0 is the tuning parameter of the above adaptation law. The estimator synthesis can be summarized by the following theorem: Theorem 1. Given the reference model (13), and the adaptive observer (14); assuming that the parameter of the observer (14) is selected according to (17) with Assumptions (i)(iii) and (16) satised, then the estimate ^ given by (19) converges in nite time h, to its nominal value h. For the sake of simplicity, we leave the proof of the above theorem for Kenn et al. [23]. The DFL compensated model (8) can then be written as follows:

_ x Ax Bu;
where u v f

20 3 7 5; 2 3

1
D H

0
xs H xds x0 T 0 d0
ds

0
xds x0 T 0 d0
ds

6 A 40 0

6 7 B 4 0 5:
Fig. 2. Block diagram of the experimental setup.

Author's personal copy

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

813

i Excitation control voltage

2

i Power angle
kc uf
22

1.5

(degree)

20 18 16 0 1.01 1 2 3 4 5 6 7 8 9

(p.u.)

ref

1 0.5 0 0

10

ii Terminal and infinite bus voltage (p.u.)

1 2 3 4 5 6 7 8 9 10 1 0.99 0.98 0.97 0 0.4 1
3

ii Active electric power and mechanical power input

0.45

Vs V
2 3 4 5 6 7 8 9
t

10

(p.u.)

2 0.35

x 10

iii Relative rotor speed

P
0.3 0.25 0

(p.u.)
9 10

e m m

1 0 1 0

P estimate
1 2 3 4 5 6 7 8

10

Time (s)

Time (s)

(a)

(b)

Fig. 3. Experiment 1. Temporary turbine fault test. (a) i Excitation control input and ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; and iii Relative rotor speed.

1. At t 0, the system is in pre-fault state. 2. At t 1:3 s, a temporary turbine fault occurs. 3. At t 2:3 s, the turbine fault is removed. The results obtained are shown in Fig. 3. The second experiment has been performed to verify the effect of temporary line fault by using the following sequence: 1. At t 0, the system is in pre-fault state. 2. At t 0:52 s, a fault (open-circuit) occurs in one of the three transmission lines. 3. At t 0:8 s, the transmission line is restored and the system is in a post-fault state.

In this case, the results are depicted in Fig. 4. In the third experiment, the robustness property with respect to a permanent fault (critical large disturbance) is tested. This experiment has been carried out according to the following sequence: 1. At t 0, the system is in pre-fault state. 2. At t 0:6 s, a fault (open-circuit) occurs in one of the three transmission lines and the system operates under this condition. The results in this case are shown in Fig. 5. In all cases, the improved direct feedback linearization controller (IDFL) regulates with a reasonably short time the generator ter-

i Excitation control voltage

1.6 1.4

i Power angle
kc uf
22

(degree)

(p.u.)

1.2 1 0.8 0.6 0 1 2 3 4 5 6 7 8 9 10

20 18 0

ref
1 2 3 4 5 6 7 8 9 10

ii Terminal and infinite bus voltage

1

(p.u.)

ii Active electric power and mechanical power input

0.4 0.39 0.38 0.37 0.36 0 Pe Pm Pm estimate

0.99 0.98 0 4 x 10
3

Vs Vt
1 2 3 4 5 6 7 8 9 10

iii Relative rotor speed

(p.u.)

(p.u.)
1 2 3 4 5 6 7 8 9 10

2 0 2 0

10

Time (s)

Time (s)

(a)

(b)

Fig. 4. Experiment 2. Temporary transmission line fault. (a) i Excitation control input and ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; and iii Relative rotor speed.

Author's personal copy

814

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

i Excitation control voltage (degree)

1.6 1.4

i Power angle
kc uf
22 21 20 19 0 1

ref

(p.u.)

1.2 1 0.8 0.6

10

ii Terminal and infinite bus voltage (p.u.)

1 2 3 4 5 6 7 8 9 10

Vs
0.99

Vt

Active electric power and mechanical power input

0.4 0.39 Pe Pm Pm estimate

0.98 0 3 x 10
3

10

(p.u.)

iii Relative rotor speed

0.38

(p.u.)
1 2 3 4 5 6 7 8 9 10

0.37 0.36 0.35 0

2 1 0 1 2 3 4 5 6 7 8 9 10

Time (s)

Time (s)

(a)

(b)

Fig. 5. Experiment 3. Test of the robustness with respect to the transmission line fault. (a) i Excitation control input and ii Active electric power, mechanical power input and its estimate. (b) i Power angle; ii Terminal and innite bus voltage; and iii Relative rotor speed.

i Excitation control voltage

3
1.5

i Excitation control voltage

k u AVR/PSS
c c f f

(p.u.)

(p.u.)

2 1

k u IDFL

k u AVR/PSS
c c f

k u IDFL
f

0.5 0

10

ii Terminal voltage
1
1

ii Terminal and infinite bus voltage

(p.u.)

(p.u.)

0.99

V AVR/PSS
t t

0.99

V AVR/PSS
t t

V IDFL
0.98 0 x 10 2 1 0 1 0
3

V IDFL
10

0.98 0 x 10 4 2 0 2 0
3

iii Relative rotor speed

iii Relative rotor speed

(p.u.)

AVR/PSS IDFL
1 2 3 4 5 6 7 8 9 10

(p.u.)

Time (s)
Fig. 6. Experiment 1. Temporary turbine fault test comparative results. IDFL (solid line). AVR/PSS (dashed line).

Time (s)
Fig. 7. Experiment 2. Test of the robustness with respect to the transmission line temporary fault comparative results. IDFL (solid line). AVR/PSS (dashed line).

minal voltage to its pre-fault value (experiments 1 and 2) and to the neighborhood of the pre-fault value (experiment 3: permanent line fault) despite the fact that the assumed well-known structures may not accurately model the system at all operating conditions1 [27]. It is also meaningful to note that the results obtained can be signicantly improved if modern methods for power synchronous generators parameter identication are used (e.g. [24]). Remark 4. Although the model of the power generator used in this work assumes balanced operation, the test of the permanent fault
1

(open-circuit under load operating condition) simulated in one of the three transmission lines forces the loaded machine to operate with only two transmission lines. This situation can be viewed as a sort of unbalanced operating condition. For comparison, the above experiments have been performed using the conventional AVR/PSS scheme. The comparative results are reported in Figs. 68. In all cases, the proposed IDFL achieves very good tracking performance while AVR/PSS scheme exhibits sustained and undesirable oscillations in the third experiment (case of critical large disturbance).

Effect of the parameter uncertainties due to off-line identication methods.

Author's personal copy

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816

815

i Excitation control voltage

3 kc uf AVR/PSS

V t eju
kc uf IDFL

jxs Eq ejp=2d jxd V s ejp=2 : jxds

22

(p.u.)

By using (22), the modulus of this terminal voltage can be deduced as follows:

Vt
0 1 2 3 4 5 6 7 8 9 10

1 xds

q x2 E2 V 2 x2 2xs xd Eq V s cosd: s q s d

23

ii Terminal voltage
1 Vt AVR/PSS

The active power is given by

Pe

(p.u.)

0.99

Vt IDFL

V s Eq sind: xds

24

Substituting in (23) Eq with the expression derived from (24) yields

1
3

0.98 0 x 10 3 2

10

iii Relative rotor speed

AVR/PSS IDFL

Vt

x2 P 2 s e V 2 sin d s
2

x2 V 2 2xs xd d s Pe cotd xds x2 ds

!1 2 : 25

Since 1= sin d 1 cot2 d, Eq. (25) can be rewritten as

(p.u.)

1 0 1 0

V2 t
1 2 3 4 5 6 7 8 9 10

Time (s)
Fig. 8. Experiment 3. Test of the robustness with respect to the transmission line permanent fault comparative results. IDFL (solid line). AVR/PSS (dashed line).

x2 V 2 x2 P 2 2xs xd d s s 2e cot2 d Pe cotd xds x2 Vs ds  2 xs P e cotd xd V s x2 P 2 s 2e : Vs xds Vs x2 P 2 s e V2 s

26

Finally the expression of the power angle d can be derived from Eq. (26) as follows:

The main feature of the proposed IDFL is that any linear control theory can be applied to design the control law for v f and derive the effective control kc uf of the power generator. 5. Conclusion In this paper, a simple improved direct feedback linearization design method for transient stability and voltage regulation of power systems has been described. Starting with the classical direct feedback linearization technique currently applied to power systems, an adaptive nonlinear excitation control of synchronous generators has been proposed. This method is based on a standard third-order model of a synchronous generator which requires only information about the physical available measurements of angular speed, active electric power and generator terminal voltage. Experimental results of a practical power system show that fast response, robustness, damping, steady-state and transient stability as well as voltage regulation are all achieved satisfactorily. The comparative results have shown that the proposed IDFL achieves very good tracking performance while AVR/PSS scheme exhibits sustained and undesirable oscillations in the third experiment (case of permanent transmission line fault). Another important issue of the proposed method is that when an equivalent linear system is obtained, a variety of proven linear-control design techniques can be applied to complete the control design. The scheme proposed in this paper could be viewed as a starting point in the design of the decentralized controllers. Acknowledgment The main part of the experimental setup used in this work has been supported by the Dpartement nergie, cole Suprieure dlectricit, Gif-sur-Yvette, Paris, France. Appendix A. Derivation of Eqs. (2) and (3) The generator terminal voltage is given by

Vs xd V s d arc cot xs P e xds

s!! x2 P 2 : V 2 s 2e t Vs

27

Appendix B. System parameters The rated characteristics of the synchronous generator considered in this work are: 4 kVA, 120 V, 50 Hz, four poles, 0.8 PF (lag). The parameters of the power system are listed below: Damping constant (p.u.) Inertia constant (s) Direct axis transient open-circuit time-constant (s) Generator direct axis reactance (p.u.) Generator direct axis transient reactance (p.u.) Transmission line reactance (p.u.) D 0:1 H 0:576 T 0d0 0:44 xd 0:894 x0d 0:620 xL 0:294

References
[1] Yu YN. Electric power system dynamics. New York: Academic Press; 1983. [2] Anderson PM, Fouad AA. Power system control and stability. NJ: IEEE Press; 1994. [3] Kundur P. Power systems stability and control. New York: McGraw-Hill, Inc.; 1994. [4] Law KT, Hill DJ, Godfrey NR. controllers structure for coordinated power system voltage regulator and stabilizer design. IEEE Trans Contr Syst Technol 1994;2:22032. [5] Cao Y, Jiang L, Chen S, Chen D, Malik OP, Hope GS. A nonlinear variable structure stabilizer for power system stability. IEEE Trans Energy Convers 1994;9:48995. [6] Gao L, Chen L, Fan Y, Ma H. A nonlinear control design for power systems. Automatica 1992;28:9759. [7] Wang Y, Hill DJ, Gao L, Middleton RH. Transient stability enhancement and voltage regulation of power system. IEEE Trans Power Syst 1993;8:6207. [8] Zhu C, Zhou R, Wang Y. A new nonlinear voltage controller for power systems. Electr Power Energy Syst 1997;19:1927. [9] Escarela-Perez R, Alvarez-Ramirez J, Cervantes I, Campero-Littlewood E. Testing robustness and performance of PSSAVR schemes for synchronous generators using nite-element models. Int J Electr Power Energy Syst 2003;25:55165. [10] Galaz M, Ortega R, Bazanella AS, M SA. An energy-shaping approach to the design of excitation control of synchronous generators. Automatica 2003;39:1119.

Author's personal copy

816

G. Kenn et al. / Electrical Power and Energy Systems 32 (2010) 809816 [19] Barrera-Cardiel E, Pastor-Gomez N. Microcontroller-based power-angle instrument for a power system laboratory. IEEE Power Eng Soc 1999;2(Summer Meeting):100812. [20] Chen Y, Zhang C, Hu Z, Wang X. A new approach to real time measurement of power angles of generators at different locations for stability control. IEEE Power Eng Soc 2000;2(Winter Meeting):123742. [21] Despalatovic M, Jadric M, Terzic B. Real-time power angle determination of salient-pole synchronous machine based on air gap measurements. Electr Power Syst Res 2008;78:187380. [22] Signal processing toolbox users guide. Natick: The MathworksWiley Inc.; 2006. [23] Kenne G, Ahmed-Ali T, Lamnabhi-Lagarrigue F, Arzand A. Nonlinear systems time-varying parameter estimation: application to induction motors. Electr Power Syst Res 2008;78:18818. [24] Arjona MA, Escarela-Perez R, Espinosa-Perez G, Alvarez-Ramirez J. Validity testing of third-order nonlinear models for synchronous generators. Electr Power Syst Res 2009;79:9538. [25] Bergen AR, Vittal V. Power systems analysis. Englewood Cliffs (NJ): PrenticeHall; 2000. [26] Utkin VI. Sliding modes in optimization and control. Springer-Verlag; 1992. [27] Saied SA, Bathaee SM, Karrari M, Rosehart W, Malik OP. Identication of electric parameters of synchronous generator using inputoutput data set. In: Proc of the fth IFAC symposium on power plants and power systems control, vol. 5; 2006. p. 2536.

[11] Ko HS, Lee KY, Kimc HC. An intelligent based LQR controller design to power system stabilization. Electr Power Syst Res 2004;71:19. [12] Damm G, Marino R, Lamnabhi-Lagarrigue F. Adaptive nonlinear output feedback for transient stabilization and voltage regulation of power generators with unknown parameters. Int J Robust Nonlin Contr 2004;14:83355. [13] Alvarez-Ramirez J, Cervantes I, Escarela-Perez R, Espinosa-Perez G. A two loop excitation control system for synchronous generators. Int J Electr Power Energy Syst 2005;27:55666. [14] Chaturvedia DK, Malikb OP. Generalized neuron-based pss and adaptive pss. Control Eng Pract 2005;13:150714. [15] Colbia-Vega A, De Len-Morales J, Fridman L, Salas-Pea O, Mata-Jimnez MT. Robust excitation control design using sliding-mode technique for multimachine power systems. Electr Power Syst Res 2008;78:162734. [16] Manjarekar NS, Banavar RN, Ortega R. Nonlinear control synthesis for asymptotic stabilization of the swing equation using a controllable series capacitor via immersion and invariance. In: Proceedings of the 47th IEEE conference on decision and control, Cancun, Mexico, December 911 2008. p. 24938. [17] Tsai H, Keyhani A, Demcko JA, Farmer RG. On-line synchronous machine parameter estimation from small disturbance operating data. IEEE Trans Energy Convers 1995;10:2536. [18] Demcko JA, Pillutla S, Keyhani A. Measurement of synchronous generator data from digital fault recorders for tracking of parameters and eld degradation detection. Electr Power Syst Res 1996;39:20513.

G. Kenn, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, A. Arzand, J. C. Vannier, An improved rotor resistance estimator for induction motors adaptive control, Electric Power Systems Research, Vol. 81, pp. 930-941, 2011. Editor: Elsevier. Impact Factor: 1.396.

Author's personal copy

Electric Power Systems Research 81 (2011) 930941

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

An improved rotor resistance estimator for induction motors adaptive control

Godpromesse Kenn a, , Tarek Ahmed-Ali b , Francoise Lamnabhi-Lagarrigue c , d d Amir Arzand , Jean Claude Vannier
a

Dpartement de Gnie Electrique, Laboratoire dAutomatique et dInformatique Applique (LAIA), Universit de Dschang, B.P. 134 Bandjoun, Cameroun Groupe de Recherche en Informatique, Image, Automatique et Instrumentation de Caen (GREYC), Universit de Caen, France c Laboratoire des Signaux et Systmes (L2S), C.N.R.SSUPELEC, Universit Paris XI, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France d Dpartement Energie, Ecole Suprieure dElectricitSUPELEC, 3 Rue Joliot Curie, 91192 Gif-sur-Yvette, France
b

a r t i c l e

i n f o

a b s t r a c t
An online rotor resistance estimator for induction motor adaptive control is presented. The rotor resistance scheme uses the rotor speed, the stator current, voltage provided by the controller, time-derivatives of the stator current and voltage. The time-derivatives of the stator current and voltage are estimated using 2nd order sliding mode observer. The nite time convergence of the estimated rotor resistance to the nominal value is achieved when the classical persistent excitation condition is satised by the input signal. The method is suitable for online operation to continuously update the rotor resistance value. Experimental results with online variation of the rotor resistance show that the proposed algorithm gives better performance compared to the previous work. The other interesting feature of the proposed method is the robustness properties with respect to the variation of the stator resistance both at relatively high speed and low speed. 2010 Elsevier B.V. All rights reserved.

Article history: Received 22 July 2009 Received in revised form 6 August 2010 Accepted 28 November 2010 Available online 5 January 2011 Keywords: Time-varying parameter estimation Nonlinear observer Second order sliding mode Equivalent control

1. Introduction The problem of the rotor resistance (Rr ) estimation is still recognized as a crucial one in induction motor (IM) control. It is well known in literature (e.g. see [17] to cite a few) that this parameter may vary up to 100% during operation of the IM and that due to this large variation, the online estimated value of this parameter is required to design an efcient feedback control of high performance induction motors using the rotor speed, stator current and stator voltage signals measurements. In the area of IM parameter estimation and adaptive control of IM, various approaches have been recently developed [317], but only partial and quite weak results have been obtained in term of the investigation of the online variation of these parameters in real-time experiments. The problem of IM time-varying parameter estimation has been studied in [5,8,10,16,17] but there is a lack of experimental results which are the only way to verify the effectiveness of those methods. In [9], the method of eld-oriented control is revisited and a modication of it known as inputoutput linearization control of IM is presented. These controllers are based on the nonlinear differen-

Corresponding author. Tel.: +237 77 59 52 19; fax: +237 77 59 52 19. E-mail addresses: gokenne@yahoo.com (G. Kenn), tarek.ahmed-ali@greyc.ensicaen.fr (T. Ahmed-Ali), lamnabhi@lss.supelec.fr (F. Lamnabhi-Lagarrigue), amir.arzande@supelec.fr (A. Arzand), jean-claude.vannier@supelec.fr (J. Claude Vannier). 0378-7796/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.epsr.2010.11.020

tial equation model of the IM and require full state feedback. Some techniques to achieve the estimation of the non-available rotor ux linkages are also described as well as a nonlinear least-squares approach to identication of the induction motor parameters. In [11,12,15], interesting algorithms for IM parameter estimation are proposed using least squares technique but online variation of the IM parameters was not investigated. In [14], a method for rotor resistance estimation for indirect eld oriented control of IM based on reactive power reference model is presented under motoring and generating modes. Sensitivity of the algorithm to errors in other machines parameters is investigated but without variation of the rotor resistance. In [13], a robust nested sliding mode regulation with application to rotor ux modulus and rotor speed of IM with unknown load torque has been introduced. The variation of the stator/rotor resistance has been investigated but the estimation of these parameters was not achieved. Therefore, the problem of combining parameter adaptation with rotor ux observer and rotor speed/ux tracking remains open in general and particularly when the unknown parameters are assumed to be time-varying. This paper revisits the problem of IM time-varying parameter estimation and establishes an online rotor resistance estimator suitable for online operation to continuously update the rotor resistance value. The present contribution is an improvement of the previous work [18] in terms of the rotor resistance estimation accuracy and rotor ux and speed tracking performance. Moreover, a global convergence and stability analysis, taking into account the interconnection of the different estimators, have been investigated.

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941 931

In Section 2, the description of IM model and problem statement are presented. The design procedure of the rotor resistance identier is revisited in Section 3 and a global convergence and stability analysis is given. Experimental results of online implementation are presented in Section 4 to verify that the previous adaptive control [18] can be effectively improved using the new rotor resistance estimator. Finally, Section 5 is devoted to the conclusion of the paper. 2. Model description and problem statement Assuming linear magnetic circuits, the dynamic model of a balanced IM in a xed reference frame attached to the stator, is given by the following fth order model (see [3,9,19,20] for derivations, modelling assumptions and general machine theory) d Te D TL = J J J dt d r = dt dis dt Rr I + np K Lr M Ls Lr 1 Ls
r

of the IM used for the rotor resistance identier is obtained by differentiating (3) combined with (2) and by taking into account that the terms containing d/dt, dRs /dt and dRr /dt are negligible with respect to other existing terms.1 This transformation gives the following differential equation d2 is = f0 + Rr f1 dt 2 where f0 and f1 are given as follows f0 = f1 =
1

(4)

d vs dis Rs dt dt
3 is

+ K dis dt
3, 4

2 is + 3 vs + 1

dis dt

2 vs

1,

and 1 , 2 , 3 ,

2,

are IM parameters expressed as

(1) + Rr Mis Lr
r

(2)

np Rs np 1 = np , 2 = , 3 = Ls Ls 1 1 Rs , , 1 = 2 = 3 = Ls Ls Lr Ls Lr

and

1 . Lr

Rr I + np K Lr M 2 Rr
2 Lr

The model given by (4) is an adequate representation of the motor dynamics if (3) d dt 2 || . Tr (5)

Rs +

is +

1 vs Ls

where I= is = 1 0 0 1 , , K= 0 1 1 0 , , =
ra rb

Introducing the variables x1 = is , x2 = dis /dt, u1 = vs and u2 = dvs /dt, the model described by (4) can be rewritten as x 1 = x2 , u1 = u2 , x2 = f (t, x1 , x2 , u1 , u2 ) + (t, x1 , x2 , u1 , u2 ) y1 = x1 , y2 = u1

isa isb

vs =

vsa vsb

(6)

In (1)(3), the state variables are rotor speed , rotor ux ( ra , and stator currents (isa , isb ); the control inputs are stator voltages (vsa , vsb ); the measured variables are (, isa , isb ) while ( ra , rb ) are not measured; the parameters are the external load torque TL , total motor and load moment of inertia J, rotor and stator winding resistances (Rr , Rs ), inductances (Lr , Ls ) and mutual inductance M; the number of pole pairs is np and the damping gain is D. To simplify the notations in (1)(3), we use = 1 (M2 /Ls Lr ) (leakage T parameter), Te = is K r (electromagnetic torque) with the constant = (3np M/2Lr ). Note that the parameters Rr and TL are typically uncertain since Rr may vary up to 100% of its nominal value due to rotor heating and TL depends on applications. The following assumptions will be considered until further notice:
rb )

where the nominal part of the system is represented by the function f, e.g. f = MN LsN LrN N RsN + 1
N LsN

RrN I + np K LrN x1 RrN

2 LrN

N LsN LrN

MN + 1
N LsN

x2 +

1 N LsN

2 MN RrN 2 LrN 2 RsN + MN

u1 N LsN x2 +

RrN MN x1 LrN u2 .

(i) The stator current and voltage are continuous and bounded with time-derivatives bounded piecewise-continuous. (ii) The rotor resistance Rr Rr , where Rr is a compact set of R. (iii) It is also assumed that the rate of variation of the rotor resistance Rr , stator resistance Rs and rotor speed are negligible compared to other dynamics of the induction motor states = = variables (e.g., Rr 0, Rs 0 and 0) and that the other = parameters are constants. Remark 1. The stator voltage usually can be discontinuous. This is the case when the IM is fed by PWM voltage-source-inverter. Assumption (i) can be satised if inverter reference voltages are used in the estimation scheme instead of the measured voltages. Our goal is to design a rotor resistance estimation algorithm using second-order sliding mode observer when the motor is driven in closed loop with a nonlinear controller which stabilizes the rotor ux magnitude and the rotor speed to references values with adaptation of the rotor resistance and load torque. The dynamic equation

and while the uncertainties are concentrated in the term (t, x1 , x2 , u1 , u2 ). Note that the term f(t, x1 , x2 , u1 , u2 ) is expressed as function of the derivatives of the stator current and voltage which cannot be obtained directly using numerical differentiation due to the presence of noise. 3. Design procedure of the rotor resistance estimate 3.1. State observer design To estimate the unavailable stator current and voltage derivatives let us consider the following second-order sliding mode observer [21]
1 x1i = x2i + zxi , 2 x2i = zxi , 1 u1i = u2i + zui , 2 u2i = zui ,

i = a, b

(7)

i = a, b

(8)

See Appendix A for more details.

Author's personal copy

932 G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941

Fig. 1. Structure of the proposed adaptive nonlinear controller.

1 2 1 2 where the variables zxi , zxi , zui and zui are given by the following expressions 1 zxi = xi | 1i x1i |1/2 sign( 1i x1i ) x x 2 zxi = xi sign( 1i x1i ) x 1 zui = ui |u1i u1i |1/2 sign(u1i u1i ) 2 zui = ui sign(u1i u1i )

with F(t, x1 , x2 , u1 , u2 ) = f (t, x1 , x2 , u1 , u2 ) (t, x1 , x2 , u1 , u2 ). Assuming that the states of the system is bounded as reported in Assumption (i), then there exists positive constants F and u such that the inequalities |F(t, x1 , x2 , u1 , u2 )| < |u2i | <
u F

(9)

(10)

(13) (14)

and xi , xi , ui and ui are the positive tuning parameters. The solutions of the above observer are understood in the Filippov sense [22]. Taking ex1i = x1i x1i , ex2i = x2i x2i , eu1i = u1i u1i , and eu2i = x2i x2i , the following dynamics errors equations are obtained ex1i = ex2i xi |ex1i |1/2 sign(ex1i ) ex2i = F(t, x1 , x2 , u1 , u2 ) xi sign(ex1i ) eu1i = eu2i ui |eu1i |1/2 sign(eu1i ) eu2i = u2i ui sign(eu1i ) (11)

hold t, x1 , x2 , u1 and u2 . Theorem 1. Assuming that Assumptions (i)(iii) are satised and conditions (5), (13) and (14) hold, then the variables of the observer (7) and (8) converge to the states of the nonlinear model of the IM (6).
i Speed reference in experiments 1 and 2
150 100 50 0 0 100 2 4 6 8 10 12 14 16 18 20

(12)

(rad/s)

ii Speed reference in experiment 3

(rad/s)

50 0 0

10

12

14

16

18

20

iii Speed reference in experiment 4

(rad/s)

0 0 1

10

12

14

16

18

20

iv Flux reference

(Wb)

0.5 0 0

10

12

14

16

18

20

Time (s)

Fig. 2. Block diagram of the experimental setup.

Fig. 3. Speed and ux reference signals.

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941 933

a
200 (v) 0 200 0 40 20 0 20 40 0 40 2 4 6 2 4 6

Control voltage component

b
10 5

Stator voltage observer error

(V)

0 5

8 10 12 Stator current component

14

16

18

20

10 0 1 0.5

10

12

14

16

18

20

Stator current observer error

(A)

(A)

0 0.5

8 10 12 Applied load torque

14

16

18

20

1 0

10

12

14

16

18

20

Adaptive error of the derivative of the stator current

2000

20

( A/s )

(Nm)

0 2000

0 0

10 Time (s)

12

14

16

18

20

10

12

14

16

18

20

Time (s)

Fig. 4. 1st Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 torque. (b) Observer/adaptive errors.

to 0.44

). (a) Control voltage, stator current and applied load

The proof of the convergence of the estimated states to the real states can be found in [21]. The variables of the observer (7) and (8) converge in nite time to the states of the nonlinear model of the IM (6) in the ideal case (when the available states are not contaminated with measurement noise N(t)). Otherwise, if supN(t) = , the above observer provides estimation of x2i and u2i with accuracy proportional to ( )1/2 ( F )1/2 and ( )1/2 ( u )1/2 , respectively. Remark 2. The rst attempt to choose the observer parameter can be achieved by applying the guidelines given in [21]. e.g. xi and xi can be chosen as follows. xi = a1 F and xi = a2 ( F )1/2 with a1 = 1.1 and a2 = 1.5. 3.2. Rotor resistance adaptation algorithm In this section the method used in [18] is revisited and more details are given concerning the convergence conditions. Assuming that the unavailable states x2i and u2i are provided in nite time by the observer (7) and (8), the induction motor model (4) can be rewritten as x2i = f0i + Rr f1i , i = a, b (15)
i Rotor speed

with x2i = x2i f0a = 1 u2a Rs x2a 2 x1b + 3 u1b + 1 x2b 0b = 1 u2b Rs x2b + 2 x1a + 3 u1a + 1 x2a f f1i = 2 u1i 3 x1i 4 x2i . Estimation of the parameter Rr can be achieved using (15) as the reference model of the induction motor. This estimation can be treated separately using a state observer combined with a static parameter estimation law or simultaneous using an adaptive state observer combined with a dynamic parameter estimation law. The rst approach required the inversion of the system (15). However, the results obtained using this approach always exhibit peak values. These peak values are due to the singularity of the denominators of the static parameter estimation law. Therefore, it is more convenient to choose a dynamic parameter estimation law combined with an adaptive state observer. In order to design the rotor resistance estimate using the later approach, let us consider the following adaptive observer x2i = Rr f1i + f0i kxi sign(x2i x2i ), i = a, b. (16)

a
150 100

b
0.45 0.4

i Stator resistance with 100% variation (from 0.22 to 0.44)

(rad/s)

(Ohm)
ref : actual value
2 4 6 8 10 12 14 16 18 20

0.35 0.3 0.25 0.2 0

50

0 0

10

12

14

16

18

20

ii Rotor flux magnitude

1 0.8
1.4 1.2

ii Rotor resistance
R
rN

estimated value

(Ohm)
Reference value Estimated value
2 4 6 8 10 12 14 16 18 20

(Wb)

0.6 0.4 0.2 0 0

1 0.8 0.6 0.4 0.2 0 2 4 6 8 10 12 14 16 18 20

Time (s)

Time (s)
to 0.44 ). (a) Rotor speed and rotor ux magnitude. (b) Stator

Fig. 5. 1st Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 resistance and rotor resistance estimate.

Author's personal copy

934 G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941
i Rotor speed tracking error with 100% variation in R
4

i Rotor speed
150 100 50 0 0

(rad/s)

0 2 4 0 2 4 6 8 10 12 14 16
s

(rad/s)

ref

: actual value
18 20

10

12

14

16

18

20

ii Rotor flux tracking error with 100% variation in R

ii Rotor flux magnitude

1

0.2

(Wb)

(Wb)

0 0.2 0.4 0 2 4 6 8 10 12 14 16
s

0.5

Reference value Estimated value

0 0

18

20

10

12

14

16

18

20

iii Rotor resistance estimation error with 100% in R

1

iii Rotor resistance estimate with online variation

(Ohm)

0.5

(Ohm)

1 0.5 0 0

0 0 2 4 6 8 10 12 14 16 18 20

10

12

14

16

18

20

Time (s)

Time (s)

Fig. 6. Speed tracking error (i), rotor ux magnitude tracking error (ii) and rotor resistance estimation error (iii) with 100% variation of the stator resistance (Experiment 1).

Fig. 8. 2nd Experiment: Tracking performance of the proposed method with respect to online variation of the rotor resistance.

Dening exi = x2i x2i and eRr = Rr Rr , the dynamics of the observer error can be computed using (15) as exi = eRr f1i kxi sign(exi ), i = a, b. (17)
t

There exists > 0, T > 0, t0 > 0 such that for all t 0

t+T

f1 (s)f1 (s)T ds I > 0.

(20)

To achieve the design of the rotor resistance identier the following additive assumptions are required. Assumption (iv). It is assumed that the following rotor resistance identiability condition holds. There exists > 0, such that for all t 0 |f1i (t)| > 0, i = a, b. (18)

The physical circumstances for which condition (18) or (20) holds, are that the rotor currents of the IM are not zero. Taking into consideration Assumption (v), a sliding mode regime will occur on the manifold exi = 0 in nite time. Therefore the dynamic equation of the adaptive observer error can be rewritten as exi = eRr f1i kxi signeq (exi ) = 0, T f1 uxeq or eRr = ||f1 ||2 where uT x
eq

i = a, b (21)

Assumption (v). It is assumed that the gain of the adaptive observer is chosen such that kxi |eRr f1i |max , i = a, b. (19)

= (kxa signeq (exa ), kxb signeq (exb )) is the equivalent con-

Remark 3. The identiability condition (18) can be substituted by the following persistency of excitation condition [23].

trol which cannot be computed since the rotor resistance estimation error is not available. Using the average control as the approximation of the equivalent control uxeq [24], we considered

a
(V)

Control voltage component

400 200 0 200 400 0 40 20 2 4 6 8 10 12 14 16 18 20

b
10 5

Stator voltage observer error

(V)

0 5 10 0 1 0.5 2 4 6 8 10 12 14 16 18 20

Stator current component

Stator current observer error

(A)

0 20 40 0 60 2 4 6 8 10 12 14 16 18 20

(A)

0 0.5 1 0 2 4 6 8 10 12 14 16 18 20

Applied load torque

2000

Adaptive error of the derivative of the stator current

( Nm )

( A/s )
2 4 6 8 10 12 14 16 18 20

40 20 0 0

0 2000 0 2 4 6 8 10 12 14 16 18 20

Time (s)

Time (s)

Fig. 7. 2nd Experiment: Tracking performance of the proposed method with respect to online variation of the rotor resistance. (a) Control voltage, stator current, applied load torque. (b) Observer/adaptive errors.

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941
i Rotor speed tracking error with online variation of the rotor resistance
4

935

Its time-derivative is: W = eRr eRr . (24)

Under Assumptions (iii), (iv) and (v), (24) can be rewritten as W = kRr |eRr |. (25)

(rad/s)

4 0

10

12

14

16

18

20

ii Rotor flux tracking error with online variation of the rotor resistance
0.4

As a consequence of (25), the rotor resistance estimation error con verges in nite time to zero (i.e., the rotor resistance estimate Rr converges to the nominal value Rr ) with convergence rate kRr > 0, if the identiability condition (18) or the persistency of excitation condition (20) is satised. 3.3. Global convergence and stability analysis

0.2

To achieve the global convergence and stability analysis, let us recall the nonlinear controller [18] (c > 0, c r > 0, kvq and kvd are
s

(Wb)

positive design parameters).

0.2

0.4 0

vs =
2 4 6 8 10 12 14 16 18 20

||

r ||

vd + s

K r q v || r || s s = s = d + c e , dt d +c dt
r

(26) e = ref e = ||
2 r ||

Time (s)

Fig. 9. Speed tracking error (i) and rotor ux magnitude tracking error (ii) with online variation of the rotor resistance (Experiment 2).

vq = vq eq kvq s , s s
s

(27) Fref (28)

vd = vd eq kvd s r , s s
s

e r,

the following rotor resistance adaptation law (kRr > 0 is the tuning parameter of the adaptation law). Rr = kRr sign T f1 uxeq ||f1 ||2 . (22)

where

vq eq = s vd eq = s

1 ( f c e ) b ref 1 (F f b r ref
r

(29) (30)

c re r)

The main result can be summarized by the following theorem. Theorem 2. Given the reference model (15) and the adaptive observer (16). Suppose that the parameter of the observer (16) is selected according to (19) with Assumptions (i)(iv) satised, then the rotor resistance estimate Rr given by (22) converges in nite time to the nominal value Rr . Proof. In order to prove Theorem 2, the following Lyapunov candidate function is considered. W= 1 2 e . 2 Rr (23) f = np kTe M || r ||2 + T is r J Ls Lr J D dTL , 2 (Te D TL ) Jdt J 2Rr 2Rr 2Rr T f r = || r ||2 Mis r Lr Lr Lr Rr M 2 M Rr T + M kis r + ||is || + || Lr Ls Lr Lr 1 b = || r || and b r = || r ||. Ls Ls
i Stator voltage observer error
4 2

2 r ||

np

Te

a
200

i Control voltage component

b
(V)
0 2

(V)

200 0 50

10

12

14

16

18

20

4 0 1

10

12

14

16

18

20

ii Stator current component

ii Stator current observer error

(A)

(A)
2 4 6 8 10 12 14 16 18 20

50 0 40

1 0 500

10

12

14

16

18

20

iii Applied load torque

iii Adaptive error of the derivative of the stator current

(Nm)

20

(A/s)
2 4 6 8 10 12 14 16 18 20

0 0

500 0

10

12

14

16

18

20

Time (s)

Time (s)

Fig. 10. 3rd Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 voltage, stator current and applied load torque. (b) Observer/adaptive errors.

to 0.44

) at relatively low speed (63 rad/s). (a) Control

Author's personal copy

936 G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941

Fig. 11. 3rd Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 and rotor ux magnitude. (b) Stator resistance and rotor resistance estimate.

to 0.44

) at relatively low speed (63 rad/s). (a) Rotor speed

In (27) and (28), ref (t) and Fref (t) are the smooth bounded reference signals for the rotor speed and the squared norm of the rotor ux || r | | 2 to be controlled, respectively. Summary of basic equations used for the rotor speed and ux control is given in Appendix B. It has been proved in [18] that the above non-adaptive controller achieves exponential convergence of the speed and ux tracking errors to zero. In practice, the adaptive controller is used. In this case, the rotor resistance, the rotor ux, the electromagnetic and the load torque are substituted in the above non-adaptive controller by their estimates. Therefore, the adaptive controller is given by

vd = vd eq kvd s s s
s

(33)

with

vq eq = s vd eq = s

1 (ref f c e ) b 1 (Fref f b r
r

(34)

c re r)

(35)

np kTe f = J J f
r

M || r ||2 + T is r Ls Lr

D dTL , (Te DTL ) 2 Jdt J

vs =

r || r ||

vd + s

K r q vs || r ||

(31)

2Rr Lr

2Rr 2 2Rr T || r || Mis r Lr Lr

T kis r +

s vq = vq eq kvq s s
s

(32) b =

+M

Rr M 2 ||is || + Lr = 1 || r || Ls

M Rr 2 np || r || + Te Ls Lr Lr

(36)

i Rotor speed tracking error with 100% variation in R

2

Ls

|| r ||,

Rr = kRr sign(eRr ) = kRr sign dr = dt

2 4 6 8 10 12 14 16
s

T f1 uxeq ||f1 ||2

(rad/s)

2 0

Rr I + np K Lr

r + Rr Mis Lr (37) (38)

18

20

ii Rotor flux tracking error with 100% variation in R

0.2

Te =

T is K r

(Wb)

0 0.2 0.4 0

T L = kTL sign(eTL ) = kTL sign(Jueq ).

10

12

14

16
s

18

20

iii Rotor resistance estimation error with 100% variation in R

1

0.5

0 0 2 4 6 8 10 12 14 16 18 20

Time (s)

Fig. 12. Speed tracking error (i), rotor ux magnitude tracking error (ii) and rotor resistance estimation error (iii) with 100% variation of the stator resistance (Experiment 3) at relatively low speed (63 rad/s).

The global convergence and stability analysis taking into account the interconnections between the different estimators are based on the separation principle theorem. The nite-time convergence of the stator current and voltage time-derivatives observers and the rotor resistance estimator allows to design the observer/estimator and the control law separately. i.e., the separation principle is satised [21,25]. The only requirement for its implementation is the boundedness of the states of the system (e.g., boundedness of f , f r , b and b r ) in the operational domain and the persistency of excitation condition. The condition of persistency of excitation is always satised under normal operating conditions of the IM when voltage-source-converter is used to supply the IM. Therefore, if the controller is known to stabilize the IM (as it has been proved in [18] for the non-adaptive controller), one of

(Ohm)

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941 937

Fig. 13. 4th Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 stator current and applied load torque. (b) Observer/adaptive errors.

to 0.44

) at very low speed (5 rad/s). (a) Control voltage,

a
6 5 4

i Rotor speed

b
0.45 0.4

i Stator resistance with 100% variation (from 0.22 to 0.44)

(rad/s)

3 2 1 0 0 2 4 6 8 10 12 14 16
ref

(Ohm)
: actual value 18 20

0.35 0.3 0.25 0.2 0

10

12

14

16

18

20

ii Rotor flux magnitude

1 0.8 Reference value Estimated value
1.2 1

ii Rotor resistance
R
rN

Estimated value

(Ohm)
2 4 6 8 10 12 14 16 18 20

(Wb)

0.6 0.4 0.2 0 0

0.8 0.6 0.4 0.2 0 2 4 6 8 10 12 14 16 18 20

Fig. 14. 4th Experiment: Robustness test with respect to 100% variation in the stator resistance Rs (from 0.22 rotor ux magnitude. (b) Stator resistance and rotor resistance estimate.

to 0.44

) at very low speed (5 rad/s). (a) Rotor speed and

a
0.1 0.05 0 0.05 13 14

i Rotor resistance estimation error with 100% in Rs

b
0.05 0 0.05 13

i Rotor resistance estimation error with 100% variation in Rs

(Ohm)

15

16

17

18
s

19

20

(Ohm)

14

15

16

17

18
s

19

20

ii Rotor flux tracking error with 100% variation in R

0.05 0 0.05 13 14 15 16 17 18

ii Rotor flux tracking error with 100% variation in R (Wb)

19 20 0.01 0 0.01 13 14 15 16 17 18

(Wb)

19
s

20

iii Rotor speed tracking error with 100% variation in Rs

0.5 0 0.5 13 14 15 16 17 18 19 20 0.5 0 0.5 13 14

iii Rotor speed tracking error with 100% variation in R (rad/s)

(rad/s)

15

16

17

18

19

20

iv Stator resistance with 100% variation (from 0.22 to 0.44) (Ohm) (Ohm)
0.5 0 13 0.5

iv Stator resistance with 100% variation (from 0.22 to 0.44)

14

15

16

17

18

19

20

0 13

14

15

16

17

18

19

20

Time (s)

Time (s)

Fig. 15. Comparative results in the case of 100% variation of the stator resistance (Experiment 1). (a) Previous method. (b) Actual method.

Author's personal copy

938 G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941

a
0.1 0.05 0 0.05 10

i Rotor resistance estimation error with 100% variation in R

b
(Ohm)
0.05 0 0.05 10 11

i Rotor resistance estimation error with 100% variation in R

(Ohm)

11

12

13

14

15

16

17

18
s

19

20

12

13

14

15

16

17

18
s

19

20

ii Rotor flux tracking error with 100% variation in R (Wb)

0.01 0 0.01 10 11 12 13 14 15 16 17

ii Rotor flux tracking error with 100% variation in R

0.01 0 0.01 10 11 12 13 14 15 16 17

18
s

19

20

(Wb)

18
s

19

20

iii Rotor speed tracking error with 100% variation in R (rad/s)

0.5 0 0.5 10 11 12 13 14 15 16 17 18

iii Rotor speed tracking error with 100% variation in R

0.5 0 0.5 10 11 12 13 14 15 16 17 18

19

20

(rad/s)

19

20

iv stator resistance with 100% variation (from 0.22 to 0.44) (Ohm) (Ohm)
0.5 0 10 0.5

iv Stator resistance with 100% variation (from 0.22 to 0.44)

11

12

13

14

15

16

17

18

19

20

0 10

11

12

13

14

15

16

17

18

19

20

Time (s)

Time (s)

Fig. 16. Comparative results in the case of 100% variation of the stator resistance at relatively low speed (63 rad/s). (a) Previous method. (b) Actual method.

the admissible and recommended ways is to choose the observer dynamics fast enough to provide for the exact evaluation of the stator current and voltage time-derivatives to the rotor resistance estimator which should maintain the estimate Rr in an acceptable range and achieve its convergence before leaving some operational domain, where the stabilization is assured. The structure of the proposed adaptive nonlinear controller is depicted in Fig. 1. 4. Experimental results The effectiveness of the proposed algorithm combined with a nonlinear controller which stabilizes the rotor ux magnitude and the rotor speed to references values with adaptation of the rotor resistance and load torque has been veried experimentally in various operating conditions. The advantage of using adaptive algorithm is that it achieves online updating of the time-varying rotor resistance without using an additional algorithm which would compare the value used in the

Fig. 17. Comparative results in the case of 100% variation of the stator resistance (Experiment 1). Adaptive error of the derivative of the stator current: previous method (i) and actual method (ii).

control algorithm with that of the estimator every given nite time interval and update this value if the difference is outside some given acceptable value. Another control objective is that, if the value of the rotor resistance is outside some acceptable known bounds for normal operation, then a fault should be identied and the IM has to be stopped. The experimental setup is illustrated by the block diagram of Fig. 2 which includes a development system DSP1103, an input/output electronics board (for analog/digital conversions) and a Personal Computer (PC). A 5 kW induction motor whose data are reported in Appendix C has been used. A PWM power converter with switching frequency of 10 kHz is controlled by a DSP. The external load torque TL is produced by a loaded dc generator. The motor instantaneous speed is measured by an optical incremental encoder with 1024 lines per revolution. The stator currents are measured by Hall-type sensors. All measured electrical parameters are converted by 16b A/D converter channels with 1 s conversion time. A DSP1103 performs data acquisition and implements in real-time within the Matlab/Simulink environment software with sampling time of 105.8 s. In all experiments the measured stator currents used in the estimation scheme were passed through a low-pass lter with cut-off frequency of 1000 rad/s. We performed four experiments. In all experiments the magnitude of the rotor ux is required to reach the reference value of 0.7 Wb in 0.75 s. In the rst and second experiment the steady state speed reference value is 136.30 rad/s. In the third experiment the steady state speed reference value is 63 rad/s (relatively low speed) and this value is 5 rad/s in the fourth experiment. Both speed and ux reference signals used in all experiments are given in Fig. 3. The parameters and of the second-order sliding-mode observer (7) and (8) have been selected according to the procedure described in Remark 2 and a ner tuning has been performed to reduce the observer errors and nally the following tuning parameters were used. xi = 3 107 , xi = 3000, ui = 108 , and ui = 5000. In all experiments, the parameters of the rotor resistance identier (22) were chosen as follows. kxi = 675,000, kRr = 0.65. The equivalent control has been approximated using rst order lowpass lter with time-constant of 1 ms. After the motor start-up and in all cases, the external unknown load torque TL estimated using the adaptation law (38) is applied. In all experiments the controller parameters and the load torque estimator parameters are the same as in [18] (i.e., c = 200, c r = 200, kvd = 7, kvq = 1, k = 130 and kTL = 65). All experiments were
s s

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941 939

a
1.2 1 0.8 0.6 0.4 0.2 19

i Rotor resistance estimate with online variation

b
(Ohm)
1 0.8 0.6 0.4 0.2 10 11

i Rotor resistance estimate with online variation

(Ohm)

20

21

22

23

24

25

26

27

28

29

12

13

14

15

16

17

18

19

20

ii Rotor flux tracking error with online variation of Rr

0.1 0.1

ii Rotor flux tracking error with online variation of Rr

(Wb)

(Wb)
20 21 22 23 24 25 26 27 28 29

0.05 0 19

0.05 0 10 11 12 13 14 15 16 17 18 19 20

iii Rotor speed tracking error with online variation of Rr

1 1

iii Rotor speed tracking error with online variation of Rr

(rad/s)

0 1 19 20 21 22 23 24 25 26 27 28 29

(rad/s)

0 1 10 11 12 13 14 15 16 17 18 19 20

Time (s)

Time (s)

Fig. 18. Comparative results in the case of online variation of the rotor resistance (Experiment 2). (a) Previous method. (b) Actual method.

carried out using the estimate Rr given by (22) in the adaptive controller and over-estimated initial values have been chosen for the rotor resistance estimation dynamics. To verify the effectiveness of the proposed method, all experiments have been carried out during motor start-up and after the motor is operated with load torque. In the rst experiment, the robustness property with respect to stator resistance uncertainty is tested with 100% variation (from 0.22 up to 0.44 ). The results obtained in this case are reported in Figs. 46. In the second experiment, the performance of the algorithm to track the variation of the rotor resistance has been investigated. In this case, the online variation of the rotor resistance has been carried out using a 3-phase variable rheostat. The value of the known additional resistance is 0.47 . The experimental results are shown in Figs. 79. From Fig. 8, one may conclude that the proposed estimator is slow. From the fact that the value of the additional resistance (0.47 ) has been introduced progressively/manually and not instantaneously by using an external 3-phase variable rheostat, the convergence time is less than 6 s. The convergence time can be reduced by increasing the value of the rotor resistance estimation gain2 kRr . The third and fourth experiments have been performed to verify the robustness property of the proposed method with respect to the variation of the stator resistance when the motor operates at relatively low speed (63 rad/s) and at very low speed (5 rad/s). The closed-loop performance of the adaptive controller in this case is given in Figs. 1014. Finally, the results of the proposed method are compared to those of the previous work [18] and the comparative results are presented in Figs. 1518. In all cases, the comparative results show that the rotor resistance estimation error and the tracking errors of the adaptive controller are globally smaller than those of the previous work. These results show that there is no signicant effect on the rotor resistance estimate and the adaptive control algorithm for a wide range of variation of the stator resistance (up to 100%). The rotor resistance estimation error due to the stator resistance variation at relatively high speed operation is smaller than the error at low speed operation of the IM. This robustness property can be justied theoretically by the fact that the rotor resistance estimator (22) is based on the sliding mode theory which is a known tech-

nique for improving robustness under parameter uncertainties and exogenous disturbances. Note that the residual parameter estimation error is due to the measurement noise, mismatching between the motor and the model parameters, ohmic heating during experiments and unmodeled dynamics. The very low steady-state residual parameter error is indicative of a very good match of the data with the simplied model of IM (4) compared to the parameter estimation error obtained in [26,9,15] where least square technique has been used to estimate the parameters of the IM. We can also note that in all experiments, the adaptive observer error ex1 = x1 x1 of the pro posed method is particularly smaller than the value obtained in the previous work. This can justify the accuracy of the proposed rotor resistance estimation scheme and the good tracking performance of the adaptive controller. The very good behaviour of the proposed algorithm to track the variation of the parameter with very good accuracy demonstrates that this scheme can be used to detect and estimate the rotor resistance fault if its value is outside some acceptable known bounds. Note that the undesired rotor resistance estimation problem shown in Fig. 5(b), in the time interval [4, 5s], is due to the fact that, in this time interval, the rotor ux modulus is constant and the value of the motor torque tends to zero, which implies that the identiability condition is not satised. This problem can be solved by using an online test algorithm which will let Rr = RrN in any time interval where the rotor ux modulus is constant and the motor torque value tends to zero. 5. Conclusion An online rotor resistance estimator for induction motor adaptive control has been investigated. The nite time convergence of the estimated rotor resistance to the nominal value is achieved when the classical persistent excitation condition is satised by the input signal. A global convergence and stability analysis has also been investigated. Experimental results with online variation of the rotor resistance show that the proposed algorithm gives better performance compared to the previous work. The rotor resistance estimator presented in this paper is dedicated to improve the previous work in order to make adaptive control more benecial for motor efciency since very high tracking performance of the rotor resistance variation is achieved. This high tracking performance of Rr demonstrates that this scheme can also be used to detect

See equation (22) and the proof of Theorem 2.

Author's personal copy

940 G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941

and estimate the rotor resistance fault if its value is outside some acceptable known bounds. The other interesting feature of the proposed method is that it is very simple and easily implementable in real-time.

Appendix B. Summary of basic equations used for the rotor speed and ux control In order to derive the dynamic equation for the rotor speed control, let us consider the derivative of d/dt 1 dTe D d 1 dTL d2 = . J dt J dt J dt dt 2

Acknowledgments The main part of the experimental setup used in this work has been supported by the Dpartement Energie, Ecole Suprieure dElectricit, Gif-sur-Yvette, Paris, France.

(45)

In (45) the derivative of the electromagnetic torque Te can be expressed as dTe = dt

T is K

Appendix A. Details for IM model (4) and validity condition (5) Let us recall (1)(3) d Te D TL = J J J dt d r = dt Rr I + np K Lr
r

d r dt

T rK

dis dt

(46)

T Multiplying both sides of (2) and (3) by is K and T K respectively, r and then combining the results with (45) and (46), it follows that

(39) Rr Mis Lr

np d2 k = Te J J dt 2 1 dTL J dt

M || Ls Lr

2 r ||

T r is

Ls

T r K vs

D d J dt (47)

(40)

M dis = Ls Lr dt

Rr I + np K Lr

1 Ls

Rs +

M 2 Rr
2 Lr

is +

1 vs . Ls (41)

with k =

Rs Rr + . Ls Lr

By differentiating both sides of (41), yields M d2 is = Ls Lr dt 2 1 Ls Rr I + np K Lr Rr

2 Lr

d r dt

d M np K Ls Lr dt

To derive the dynamic equation for the rotor ux control let us multiply both sides of (2) by T . Therefore by using the property3 of r the matrix operator K, the following expression for the derivative of || r | | 2 is obtained d(|| (42)
2 r || )

Rs + M 2

1 dvs dis + . Ls dt dt

dt

= 2

Rr || Lr

2 r ||

+2

Rr T Mis Lr

r.

(48)

In order to determine when the simplied model (4) represents adequately the motor dynamics, let us re-write (42) as follows M Ls Lr = Rr I + np K Lr d r dt M d K np Ls Lr dt Rr
2 Lr r

1 dvs d2 is 1 + + Ls dt Ls dt 2

Rs + M 2

dis . dt

Eq. (48) does not contain the stator voltage as an input and thus cannot be used to derive the control scheme for the rotor ux. The following transformations are used to make the voltage appear explicitly as an input in this equation. First, we compute T the dynamic equation for the ctitious variable is r by multiplying T both sides of (2) and (3) by is and T respectively, and summing r them up. We then obtain:
T d(is r ) T = kis dt

(43)
r

Then using (40), the left-hand side of (43) can be re-written as: d r d M Rr M K I + np K np Ls Lr Lr Ls Lr dt dt M Rr Rr = I + np K I + np K Ls Lr Lr Lr M d np K r Ls Lr dt = M Ls Lr np
r

Rr M ||is ||2 + Lr

M Rr || Ls Lr Lr

2 r ||

np

Te (49)

+
r

1 Ls

T r vs .

Rr Mis Lr

Secondly, under Assumption (iii), we differentiate (48) and use (49) to derive the dynamic equation for the rotor ux containing the stator voltage vs as a control input. This leads to d2 (|| r ||2 ) Rr = 2 Lr dt 2 + d(||
2 r || ) T + M kis

2 d K Tr dt 1 np MKis , Tr

1
2 Tr

Mis )

dt
2 r ||

Rr M 2 ||is || Lr . (50)

(np )2

M Rr || Ls Lr Lr

np

Te +

1 Ls

T r vs

where Tr = Lr /Rr is the rotor time-constant. We can then deduce that (4) adequately represents the motor dynamics if: d dt 2 || . Tr (44)

T rK r

= 0.

Author's personal copy

G. Kenn et al. / Electric Power Systems Research 81 (2011) 930941 941

Appendix C. Induction motor data

Rated power Rated speed Rated torque Rated frequency Rated current Stator resistance Rotor resistance Stator inductance Rotor inductance Mutual inductance Number of pole pairs Motor-load inertia 5 kW 1500 rpm 32 N m 50 Hz 22.9 A RsN = 0.22 RrN = 0.52 LsN = 0.052 H LrN = 0.0516 H MN = 0.0495 H np = 2 J = 0.12 kg m2

References
[1] R. Marino, S. Peresada, P. Tomei, Exponentially convergent rotor resistance estimation for induction motors, IEEE Trans. Ind. Electron. 5 (1995) 508515. [2] T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, R. Ortega, A globally-stable adaptive indirect eld-oriented controller for current-fed induction motors, Int. J. Control 72 (1999) 9961005. [3] R. Marino, S. Peresada, P. Tomei, On-line stator and rotor resistance estimation for induction motors, IEEE Trans. Control Syst. Technol. 8 (2000) 570 579. [4] K. Akatsu, A. Kawamura, On-line rotor resistance estimation using the transient state under the speed sensorless control of induction motor, IEEE Trans. Power Electron. 15 (2000) 553560. [5] G. Bartolini, A. Pisano, P. Pisu, Simplied exponentially convergent rotor resistance estimation for induction motors, IEEE Trans. Autom. Control 48 (2003) 325330. [6] P. Castaldi, W. Geri, M. Montanari, A. Tilli, A new adaptive approach for online parameter and state estimation of induction motors, Control Eng. Pract. 13 (2005) 8194. [7] R. Marino, S. Peresada, C.M. Verrelli, Adaptive control for speed-sensorless induction motors with uncertain load torque and rotor resistance, Int. J. Adapt. Control Signal Process. 19 (2005) 661685. [8] M. Barut, S. Bogosyan, M. Gokasan, Speed sensorless direct torque control of induction motors with rotor resistance estimation, Energy Convers. Manage. 46 (2005) 335349.

[9] J. Chiasson, Modelling and High-performance Control of Electric Machines, Wiley Interscience, John Wiley & Sons, NY, 2005. [10] C. Picardi, F. Scibilia, Sliding-mode observer with resistances or speed adaptation for eld-oriented induction motor drives, in: IEEE Industrial Electronics-IECON 2006, 2006, pp. 14811486. [11] Y. Koubaa, Application of least-squares techniques for induction motor parameters estimation, Math. Comput. Model. Dyn. Syst. 12 (2006) 363375. [12] Y. Koubaa, Asynchronous machine parameters estimation using recursive method, Simul. Model. Pract. Theory 14 (2006) 10101021. [13] B. Castillo, S.D. Gennaro, A. Loukianov, J. Rivera, Robust nested sliding mode regulation with application to induction motors, in: Proceedings of the 2007 American Control Conference, New York City, USA, 2007, pp. 52425247. [14] P. Roncero-Snchez, A. Garca-Cerraba, V. Feliu-Batlle, Rotor resistance estimation for induction machines with indirect eld-orientation, Control Eng. Pract. 15 (2007) 11191133. [15] K. Wang, J. Chiasson, M. Bodson, L.M. Tolbert, An online rotor time constant estimator for the induction machine, IEEE Trans. Control Syst. Technol. 15 (2007) 339348. [16] A. Mezouar, M.K. Fellah, S. Hadjeri, Y. Sahali, Adaptive speed sensorless vector control of induction motor using singularly perturbed sliding mode observer, in: IEEE Industrial Electronics-IECON 2006, 2006, pp. 932939. [17] A. Mezouar, M.K. Fellah, S. Hadjeri, Adaptive sliding-mode-observer for sensorless induction motor drive using two-time-scale approach, Simul. Model. Pract. Theory 16 (2008) 13231336. [18] G. Kenne, T. Ahmed-Ali, F. Lamnabhi-Lagarrigue, A. Arzand, Real-time speed and ux adaptive control of induction motors using unknown time-varying rotor resistance and load torque, IEEE Trans. Energy Convers. 24 (2009) 375387. [19] W. Leonhard, Control of Electric Drives, Springer Verlag, 1984. [20] P.C. Krause, O. Wasynczuk, S.D. Sudhoff, Analysis of Electric Machinery and Drive Systems, Wiley Interscience, John Wiley & Sons, NY, 2002. [21] A. Levant, Robust exact differentiation via sliding mode technique, Automatica 34 (1998) 379384. [22] A.F. Filippov, Differential Equations with Discontinuous Right-hand Sides, Kluwer, Dordrecht, The Netherlands, 1988. [23] A.L. Fradkov, L.V. Miroshnik, V.O. Nikiforov, Nonlinear and Adaptive Control of Complex Systems, Kluwer Academic Publishers, 1999. [24] V.I. Utkin, Sliding Modes in Optimization and Control, Springer-Verlag, 1992. [25] A. Levant, L.V. Levantovsky, Sliding order and sliding accuracy in sliding mode control, Int. J. Control 58 (1993) 12471263. [26] J. Stephan, M. Bodson, J. Chiasson, Real-time estimation of the parameters and uxes of induction motors, IEEE Trans. Ind. Appl. 30 (1994) 746759.