Thesis of Mohd Faizan

Power Loss Estimation using Bond Graph for Resilience and Prognosis. Application to Green Hydrogen Production

The increasing complexity of modern energy systems requires advanced monitoring and prognostic approaches to ensure reliable, efficient, and resilient operation. Traditional Fault Detection and Isolation (FDI) methods based on Analytical Redundancy Relations (ARRs) are effective for fault detection but provide limited information regarding the progressive degradation of system performance. To address this limitation, this thesis proposes a Bond Graph (BG)-based framework for online power loss estimation, resilience assessment, and prognosis. The proposed approach exploits the energy-based representation of bond graphs to estimate power losses and monitor their evolution over time. Since faults and degradation phenomena are generally associated with abnormal energy dissipation, power loss is used as a key indicator linking fault occurrence to performance degradation. An augmented filtering stage is employed to improve the quality of the estimated power loss, while a machine learning layer learns its evolution and predicts future degradation trends. The methodology is first applied to a two-tank hydraulic system, where power loss tracking is used to detect performance degradation and assess system resilience. To anticipate critical operating conditions, a new indicator called the Remaining Time To Recovery (RTTR) is introduced. RTTR estimates the time interval between the detection of abnormal power losses and the occurrence of actuator saturation if no corrective action is taken, thereby enabling anticipatory recovery strategies before significant performance degradation occurs. The proposed framework is then extended to Proton Exchange Membrane (PEM) electrolyzers used for green hydrogen production. A hybrid prognostic approach combining a diagnostic Bond Graph (DBG) model and a Support Vector Regression (SVR) model is developed foronline Remaining Useful Life (RUL) estimation. The DBG model interacts with sensor measurements to estimate degradation effects in terms of power loss, while the SVR model predicts the future evolution of degradation. By combining these two sources of information, the RUL of the PEM electrolyzer can be estimated online, supporting predictive maintenance decisions. The bond graph models are developed in 20-sim, while the monitoring, resilience, and prognostic frameworks are implemented in MATLAB/Simulink. The obtained results demonstrate that power loss tracking provides earlier and more informative indicators of degradation than conventional residual-based methods, while the proposed RTTR and RUL approaches contribute to improving the reliability, resilience, and operational performance of complex energy systems.

Jury

- Mme Anne-Lise GEHIN Maîtresse de conférences Université de Lille Directrice de thèse, - M. Belkacem OULD BOUAMAMA Professeur des universités Université de Lille Co-encadrant de thèse, - M. Mahdi BOUKERDJA Maître de conférences Université de Lille Co-encadrant de thèse, - M. Ahmed MEBARKI Professeur des universités Université Gustave Eiffel Examinateur, - M. Noureddine ZERHOUNI Professeur des universités École Nationale Supérieure de Mécanique et des Microtechniques, Besançon Examinateur, - M. Kamal MEDJAHER Professeur des universités École Nationale d’Ingénieurs de Tarbes Rapporteur, - Mme Nadia STEINER YOUSFI Professeure des universités Université de Franche Comté Rapporteure.

Thesis of the team PERSI defended on 30/11/2026