Type d'Évènement
Soutenance de Gregory AHUNGWA pour une thèse de DOCTORAT de l’université de Grenoble, spécialité Signal Image Parole Télécoms
Intitulé
Contribution au développement de nouveaux capteurs pour le monitorage des turbines hydrauliques
Ecole Doctorale
Électronique, Électrotechnique, Automatique, Traitement du Signal (EEATS)
Lieu de l'Évènement
GIPSA-lab, 11 Rue des Mathématiques, 38400 Saint-Martin-d’Hères, Salle Jean-Marc
CHASSERY (Rez-de-chaussée)
Dirigée par :
- Cornel IOANA, Grenoble INP, GIPSA-lab
Description de l'Évènement
Thèse préparée au sein du laboratoire Gipsa-lab en co-tutelle avec General Electric (GE) Vernova Hydro Power Solutions
ABSTRACT
The increasing integration of renewable energy sources into modern power systems has intensified the need for flexible hydropower operation, resulting in more frequent start–stop cycles, rapid load variations, and prolonged operation under off-design conditions. While these operating patterns support grid stability, they also accelerate fatigue accumulation in critical turbine components, creating new challenges for condition monitoring and asset management. This thesis investigates the use of data-driven approaches to improve hydropower monitoring through two complementary use-cases. The first concerns the automatic identification of turbine operating modes using routinely available electrical, hydraulic, and mechanical measurements from industrial DCS/SCADA systems. A machine-learning framework is developed to infer both steady-state and transient operating conditions and to generate operational statistics relevant to fatigue assessment. The second use-case addresses the indirect reconstruction of runner stress signals associated with vortex rope excitation using deep-learning models and alternative sensing configurations. Particular emphasis is placed on understanding how sensing topology and temporal context influence reconstruction performance. To evaluate the reconstructed signals, a physically informed assessment framework combining conventional time-domain metrics with spectral-domain indicators is proposed. The results highlight the importance of integrating domain knowledge, spectral analysis, and machine learning to develop scalable, interpretable, and physics-aware monitoring strategies. Overall, this work contributes to the development of a new generation of digital monitoring solutions capable of supporting the reliable and flexible operation of hydropower assets in an increasingly renewable-dominated energy landscape.
Membres du jury :
- Mme Diana BUCUR, FULL PROFESSOR, Université POLITEHNICA de Bucarest, Rapporteure
- M. Ali MANSOUR, PROFESSEUR DES UNIVERSITÉS, ENSTA IP Paris, Rapporteur
- M. Philippe RAVIER, PROFESSEUR DES UNIVERSITÉS, Université d’Orléans, Examinateur
- M. Srdjan STANKOVIC, FULL PROFESSOR, Université de Monténégro, Examinateur
- M. Michel CERVANTES, PROFESSEUR, Université technologique de Luleå, Examinateur
- M. Vincent BOUILLET, INGÉNIEUR EXPERT, Institut SuperGrid, Invité
- M. Antoine BOMBENGER, INGÉNIEUR, GE Vernova Hydro Power Solutions, Invité
- M. Pedro DIAS DE VERAS SOUSA, INGENIEUR DOCTEUR, GE Vernova Hydro Power Solutions, Invité
- M. Cornel IOANA, MAITRE DE CONFERENCES, Grenoble INP – UGA, Directeur de thèse
