Subject Ph. D.

Introduction and context:

Proton exchange membrane fuel cells (PEMFCs) are a preferred technology for decarbonising the transport sector. However, their high cost and limited lifespan, currently around 20,000 hours for transport applications, are still hindering their large-scale deployment. In order to limit and anticipate system degradation, it is essential to operate the fuel cell (FC) under appropriate operating conditions. This involves, in particular, controlling the supply of reactants (air and hydrogen) and managing water at the membrane level, while taking into account the thermodynamic state of the system (pressure, temperature, humidity, etc.).

Water management in the membrane is essential in order to prevent performance degradation or even the appearance of irreversible defects [YHMC11]. An insufficiently hydrated membrane leads to an increase in its ohmic resistance and the appearance of drying defects, while excess water leads to flooding defects at the cathode [AJB+23], which limits the diffusion of reactants at the catalytic layer.

The FC system is generally equipped with sensors located at the reagent supply, the cooling circuit and the electrical section, which are essential for controlling the system. However, certain internal states of the FC, which are nevertheless decisive for the development of control strategies, are not directly accessible due to technical limitations or even the impossibility of measuring them. This is particularly the case for the water content of the membrane [WMGH21]. It is therefore necessary to estimate this state in order to ensure the proper functioning of the FC.

In the literature, two main approaches are generally used to estimate the water content of the membrane: model-based methods and data-based methods. Data-based approaches generally rely on artificial intelligence techniques, such as neural networks. However, these techniques require a large amount of data and their real-time implementation for transport applications remains difficult. In contrast, model-based methods do not require high computational costs and are suitable for real-time embedded applications [CCJ24]. Nevertheless, the quality of the estimation depends heavily on the quality of the FC model, which is highly non-linear, as well as on the design of the observer [LLS24].

Thesis objectives:

The objective of this thesis is to develop a model-based non-linear observer in order to estimate the water content of the membrane in real time for a transport application. More specifically, the doctoral work will include the following tasks:

- Establish a non-linear model oriented towards FC control. This model must reproduce acceptable behaviour of the fuel cell under study. Consequently, a validation stage for this model is necessary. - Develop and synthesise a non-linear observer. - The observer will be validated in real time on a test bench. The results obtained will be confirmed by EIS tests, measuring the resistance of the membrane. - Once the observer has been validated, a membrane water management strategy will be developed to ensure better system performance.

The experimental tests will be carried out at GreenGT.

The subject covered in this project is multidisciplinary in nature, encompassing both electrical engineering and automatic control.

References :

[AJB+23] M. Ait Ziane, C. Join, M. Benne, C. Damour, N. Yousfi Steiner, and M.C. Péra. A new concept of water management diagnosis for a PEM fuel cell system. Energy Conversion and Management, 285:ID 116986, 2023.

[CCJ24] X. Chi, F. Chen, and J. Jiao. Model-based observer for vehicle proton exchange membrane fuel cell humidity based on adaptive sliding mode estimation technique. Int. J. of Hydrogen Energy, 52:750-766, 2024.

[LLS24] G. Lance, T. Leroy, and J. Sery. Adaptive extended Kalman filter for PEMFC membrane water content estimation. Int. J. of Hydrogen Energy, 71:1164-1173, 2024.

[WMGH21] H.Wang, S.Morando, A.Gaillard, andD.Hissel. Sensordevelopmentandoptimization for a proton exchange membrane fuel cell system in automotive applications. J. of Power Sources, 487:ID 229415, 2021.

[YHMC11] N. Yousfi-Steiner, D. Hissel, P. Moçotéguy, and D. Cantusso. Diagnosis of polymer electrolyte fuel cells failure modes (flooding & dryingout) by neural networks modeling. Int. J. of Hydrogen Energy, 36:3067-3075, 2011.