To unlock a massive deployment of proton exchange membrane (PEM) fuel cells in electrified transportation applications it is essential to improve the durability of the technology. Achieving this requires accurate diagnosis of all the faulty operating conditions occurring in the system, which limit the lifespan. In this work, we propose a PEM fuel cell fault diagnosis deep learning (DL) method that integrates physics-based simulations as prior information. This alleviates the need for massive experimental datasets while enhancing the accuracy and the generalization of the algorithm. In the framework of domain adaptation, a supervised adversarial neural network is used to align the features extracted from the simulated data and the experimental data, while making them discriminative to the PEM fuel cell faults. We evaluate the performance of the method using a real experimental dataset collected on a PEM fuel cell operated during 1000 h. Our results demonstrate that integrating physics information in the training data of a fault diagnosis algorithm is highly beneficial for several DL architectures, particularly when limited data are available. A high accuracy level is reached in diagnosing the faults, even with aged data.

In this paper, we propose a novel hybrid method for fault diagnosis of Proton Exchange Membrane Fuel Cells (PEMFCs) based on the combination of a physics-based model and a long short-term memory (LSTM) neural network. By incorporating the physics-based model in the fault diagnosis algorithm, we can access to several process variables not directly measured through sensors but related to the state of the PEMFC stack. The model estimates are subsequently combined with signals measured on the PEMFC stack and inputted to a LSTM neural network. The performance of the physics-guided LSTM is evaluated on an extensive dataset comprising a thousand hours of operation of a PEMFC stack under dynamic load profiles, proving the enhanced capability of the proposed fault diagnosis method in capturing complex fault patterns. Furthermore, the effectiveness of the proposed method in dealing with PEMFC aging is tested by using data from the initial phase of stack operation for algorithm training and reserving the data from the aged stack operation for the testing phase. The experimental results reveal that the proposed physics-guided LSTM method allows for a significant amelioration over purely data-driven LSTM.

To pave the way to a large industrial deployment of proton exchange membrane fuel cells (PEMFCs) it is essential to improve the durability of the technology. Achieving this requires accurate diagnosis of abnormal operating conditions. To automate the monitoring process in PEMFC, data-driven fault diagnosis models have shown great potential. However, in practice, the performance of data-driven models can be compromised by the availability of the data for training these algorithms. This article proposes a simulation-driven domain adaptation method to circumvent the data scarcity issue using a physics-based model. Through adversarial training, we focus on extracting relevant, high level features for the fault diagnosis task from the simulated dataset, while simultaneously aligning the features extracted from a real dataset acquired on a PEMFC stack. Introducing supervision on the real dataset, we are able to extract features that are discriminative for the fault diagnosis task as well as invariant to the domain. The experimental results validate the effectiveness of the proposed method in dealing with scenarios of insufficient data availability for PEMFC fault diagnosis.

Hybrid electrical systems are complex to size because there is a strong dependence between components design and power management control law. In order to pre-size a hybrid rally race car, which uses hydrogen and batteries as energy sources, an offline power management method, which is formulated in a combinatorial form, is developed to be readily incorporated in a bi-level optimization problem. The exponential growth of memory space as a function of the size of the problem, in particular for long power profiles, is a well-known problem of this type of approach. This paper proposes solving techniques, which consist in using battery state trajectory following and constraints relaxation. Obtained results show satisfying improvements compared to the state of art. Sensitivity analyses show the influence of resolution parameters on the optimization problem solution and its required computation resources.

Hybrid system components pre-sizing and energy management strategy methods are important to ensure successful product development. These methods rely on optimization algorithms, which use mathematical models and are built on a set of assumptions. For hybrid vehicle's architecture where energy storage system is passively connected to the DC bus, one common assumption is to consider its voltage as constant hence ignored. In this context, is first presented the interest of not neglecting that aspect for some high power consuming hybrid vehicle applications. As their management system needs to be fed with reliable information of the DC bus voltage, a method is proposed for its evaluation during preliminary design by optimization.

Improving the cathode catalyst layer design requires understanding the sources of transport limitations in proton-exchange membrane fuel cells. For the purpose, a framework consisting on an electron microscopy characterization setup in couple with a numerical modeling software is proposed. The latter integrates highperforming geometry building capabilities which ensure full phase discretization (carbon, platinum, ionomer and pore phase) and freedom when designing the structure morphology and meshing. The 3D structure of the carbon phase is extracted from a focused ion beam scanning electron microscopy analysis having a 2 nm isotropic resolution. The platinum phase is built according to a nanoparticle size histogram determined from high angle annular dark field scanning transmission electron microscopy images. To add the ionomer phase, the thickness of the layer is measured on high resolution transmission electron microscopy images. The multi-physics model includes gas transport in the pores, and gas and ionic transport in the Nafion. A 4-step reaction mechanism is used to solve the electrochemistry. Numerical simulations are performed on two catalyst layer portions. The results show that structural heterogeneities can deeply impact performance. Such impact is mainly linked to oxygen diffusion limitations through the Nafion film and, to a certain extent, to interparticle competition effects.

This study proposes a fault detection and isolation tool for proton exchange membrane fuel cell (PEMFC) operating in embedded applications. A model-based approach, taking partially into account degradation phenomena, is proposed in order to increase the robustness of the tool regarding transient operations and stack ageing. The considered faults are the abnormal operating conditions that can decrease the fuel cell lifetime. The fault detection approach is based on residual generation using both voltage and high frequency resistance measurements and thus combining the advantages of knowledge-based model and electrochemical impedance spectroscopy (EIS) diagnosis approaches. To that end, a multi-physics fuel cell model has been used. This model computes not only the stack voltage but also the high frequency resistance in dynamic conditions. Additionally, the model is modified to take into account the ageing of the fuel cell. Validation is carried out on experimental characterizations during 1,000 h ageing. The results on a new fuel cell stack show a score of 91% for fault isolation. However, without any adaptation, this score drops dramatically as the stack ages. Finally, thanks to ageing modeling and to the proposed adaptation of the detection/isolation procedure, the diagnosis performance remains reliable during fuel cell stack ageing.

This paper presents a study of the carbon support corrosion and mitigation strategies through the use of a pseudo-3D model. This model consists in coupling a 2D model along the channel with another model perpendicular to the flow at the rib/channel scale. Simulations offer a deeper understanding of the corrosion through the analysis of the local conditions. Rib/channel heterogeneities show the higher degradation in the zones facing the anodic rib. These results are validated qualitatively on literature data by analysis of SEM images and carbon dioxide concentration at the cathode outlet. Three mitigation strategies are studied using the model. The first one consists in speeding up the hydrogen filling of the cell. The second strategy involves an external electrical resistance to create a current leak during the startup. Third, a design study of the rib/channel is performed to minimize the cathode degradation. Whatever the mitigation strategy, it consists in reducing either the duration or the magnitude of the high cathode electrode potential.

This study presents a review of prognostic methods applied to automotive proton exchange membrane fuel cell (PEMFC). PEMFC durability is strongly affected when it is subjected to automotive load cycling (ALC). ALC is normally composed of four operation modes such as start-up, idle, transient high-current demand and shutdown. All of these operation modes drastically change the internal variables of the system like temperature, pressure, relative humidity etc. causing degradation of the fuel cell components in a short time. Prognostic methods could be a possible solution to tackle the PEMFC's low durability issue because they allow predicting the remaining useful life of the system in order to apply preventive maintenance plans. Therefore, the objective of this study is to review the prognostic techniques applied to PEMFC under ALC. In the first part of this study, a summary of PEMFC degradation mechanisms caused by ALC is realised based on literature review. In the second part, the prognostic methods review for automotive PEMFCs is carried out and a general synthesis and future challenges are given in the third part of the study.

Optimizing proton exchange membrane fuel cells (PEMFC) performance is crucial in order to attain sustainable commercialization. A major part of the issue involves understanding the role of Nafion in the catalyst layer. Nafion’s behavior on bulk mode has been extensively reported in the literature, though on ultra-thin layer mode (<20nm) its structural arrangement which dictates the transport properties highly depends on the type of substrate interfacing [1]. In PEMFC catalyst layers, Nafion covers carbon-supported platinum aggregates on ultra-thin layer mode. In these composite layers, Nafion works as a binder for the aggregates, at the same time as it is meant to ensure pathways for protons and let arrive to the catalyst surface. At high current densities, a steep performance drop is usually observed in PEMFC operation. A part of this drop is reported to be due to activation losses as a doubling of the Tafel slope is observed [2]. The origin of this Tafel slope doubling has been subject of multiple studies. Understanding the causes requires understanding the phenomena occurring at the electrode/electrolyte interface. A technique which is commonly employed is cyclic voltammetry (CV) as it can provide insights regarding these interactions. Comparative CV studies on monocrystalline platinum Pt (111) in a PFSI solution and solution exhibit distinct features, where oxide formation appears to be inhibited at the beginning of this range for PFSI [3]. These CV studies coupled with electrochemical quartz crystal microbalance (EQCM) measurements showed that when extending the potential range to 1.4V and holding the potential at 1.1V (place-exchanged oxide formation range) a large mass gain at 0.5V emerges. The origin of these features is found to be due to strongly adsorbed sulfonate groups on the platinum surface. Other works validate this behavior [4], where sulfonate groups are found to be adsorbed in both the double layer region (0.4-0.5V) and in the hydroxyl adsorption region (0.6-0.85V). In this work, the continuum model proposed by Huang et al. [5] is coupled with a reaction framework comprising multistep mechanism [6]. The adsorption of SO3- is expected to follow a Langmuirian behavior [7], indicating that the oxygen reduction reaction (ORR) is hindered through site blocking. For the adsorption of on Pt(111) a single electron transfer mechanism is assumed [8]. Kinetic parameters are obtained by fitting the model’s response to experimental data as in [6]. The simulations allow quantifying the adsorption of sulfonate groups on the platinum surface and estimating its impact on the ORR. References: 1. A. Kusoglu, A.Z. Weber, Chemical Reviews, 117 : 987-1104 (2017). 2. P. Subramanian, T.A. Greszler, J. Zhang, W. Gu, R. Makharia, Journal of The Electrochemical Society, 159(5) : B531-B540 (2012). 3. T. Masuda, F. Sonsudin, P.R. Singh, H. Naohara, K. Uosaki, The Journal of Physical Chemistry C, 117 : 15704-15709 (2013). 4. K. Kodama, R. Jinnouchi, T. Suzuki, H. Murata, T. Halanka, Y. Morimoto, Electrochemistry Communications, 36 : 26-28 (2013). 5. J. Huang, A. Malek, J. Zhang, M.H. Eikerling, The Journal of Physical Chemistry C, 120: 13587-13595 (2016). 6. B. Jayasankar, K. Karan, Electrochimica Acta, 273 : 367-378 (2018). 7. S.M. Andersen, Applied Catalysis B : Environmental, 181: 146-155 (2016). 8. K. Kodama, K. Motobayashi, A. Shinohara, N. Hasegawa, K. Kudo, R. Jinnouchi, M. Osawa, Y. Morimoto, ACS Catalysis, 8 : 694-700 (2018).

The heterogeneous nature of the cathode catalyst layer has been a major obstacle toward the comprehension of the mechanisms hindering the PEMFC performance at high current densities. To deconvolute these, an approach coupling multiscale modeling and multiscale electron microscopy characterization-allowing to move from the local to the cathode catalyst layer scale-is adopted. Here, an agglomerate scale model is developed and coupled with a MEA-scale model. Different structures are analyzed and the effects of certain structural parameters (Pt particle size, agglomerate structure and Nafion layer thickness) on the electrode performance are quantified and discussed. Reducing the Pt particle size is found to improve performance in most cases, though the improvement margin is highly dependent on the structure, and to a certain extent, on the particle spatial distribution. It is found that thickening the Nafion film is detrimental for performance only when the porosity is not sufficiently large. Performance gains upon structure rearrangement into an ideal-type structure (Nafion and pore tortuosities minimized to unity) are also quantified. These analyses have unvealed the main mechanism limiting performance when shifting to moderate/high current densities and allowed quantifying and ranking the physical phenomena hampering performance by order of importance.

Decreasing the cost of proton exchange membrane fuel cells (PEMFC) is vital for the realization of the fuel cell vehicle market. For this challenge, it is essential not only to still reduce the electrode platinum loading but also to maintain a high performance at high current density. However, even for electrodes made with the most performant catalysts, a large performance loss is observed at high current density and this loss becomes larger as the Pt loading is lower. In operating conditions, recent work has shown that this performance loss was predominantly limited by oxygen [1] and proton transport to the catalytic surface. In order to quantify and link both effects to real active layer structural aspects a coupled modeling/imaging approach is performed. The microstructure of a real electrode is reconstructed in 3D using FIB-SEM (Fig.1a), STEM (Fig.1b) and HRTEM (Fig.1c). The first technique provides the carbon grain arrangement, whereas from the second one platinum nanoparticle size distribution is extracted. With HRTEM, an averaged-thickness is extracted for the ionomer thin layer. The raw imaging data acquired (Fig.1a) is then post-treated using a defined procedure. From the post-treated stack, certain portions that can be representative are selected according to [2]. An example is illustrated in Fig.1e with the respective raw portion in Fig.1d. From this portion, a statistically/imaging-based structure is then built in COMSOL ® Multiphysics software (Figs.1f-1h). The agglomerate 3D model is coupled with a 2D MEA model [3] – a model that allows computing local operating conditions. Studies on volume elements [2] of the catalyst layer are performed with physical input parameters as relative humidity, O2 concentration on the pore, among others, taken from the 2D MEA model. The following set of physics are included in the 3D agglomerate model: the Fickian diffusion for oxygen transport in the ionomer film – accounting for deviations of ionomer’s thin layer behaviour from bulk [4] – and in the pore phase – accounting for Knudsen effects; the ionic transport, through charge conservation equation; a kinetic model assuming a 4-step ORR reaction [5]. From the studies performed, simulation results show that the catalyst layer performance is mainly limited by oxygen diffusion in the ionomer phase and ionic transport in the primary pores [6]. Figs. 1a to 1k – Process from a raw image to numerical modeling : raw image stack (Fig.1a), platinum particle distribution from STEM acquisition (Fig.1b), HRTEM where Nafion layer can be observed (Fig.1c), portion of the image stack (Fig.1d) and the respective segmented portion (Fig.1e) ; carbon phase (brown), platinum phase (yellow), ionomer and pore phase (green and gray respectively) (Figs.1f-1h). Simulated oxygen concentration profile on Nafion (Fig.1i) and pore (Fig.1j) phases, ionic potential distribution (Fig.1k). References: 1. J.P. Owejan, J. E. Owejan, W. Gu, Journal of The Electrochemical Society, 160-(8): 824-833 (2013). 2. H.R. Sanei, R. S. Fertig III, Composites Science and Technology, 117 : 191-198 (2015). 3. B. Randrianarizafy, P. Schott, M. Chandesris, M. Gerard, Y. Bultel, International Journal Of Hydrogen Energy 43: 8907-8926 (2018). 4. Y. Kurihara, T. Mabuchi, T. Tokumasu, ECS Transactions, 75-(14): 129-137 (2016). 5. M. Moore, A. Putz, M. Secanell, Journal of the Electrochemical Society, 160-(6) : 670-681 (2013). 6. T. Mashio, K. Sato, A. Ohma, Electrochimica Acta, 140 : 238-249 (2014). Figure 1

An estimation of the actual State of health (SoH) of the Proton Exchange Membrane Fuel Cell (PEMFC) based on a Bayesian observer model is presented. The observer model considers a degradation model which describes the electrodes platinum dissolution to characterize the energy source deterioration under dynamic operation conditions in real time. This observer model is carried out using an Unscented Kalman Filter to correct and update the SoH estimation. The proposed method is evaluated using a 1000h durability test under a dynamic highly load profile, similar to an automotive load cycling, from a 20 cells PEMFC stack.

The estimation and increase of the lifetime of PEM fuel cell under dynamic conditions is one of a major challenge. Increasing the durability of the fuel cell must be treated by both the development of new material and design but also by optimal strategies and management of the operating conditions of the fuel cell. The startup and shut down phase are important to optimize to reduce the carbon support corrosion [1].x Based on previous development of a multi-physics modeling framework, combining complex transport such as multicomponent transport in porous media and electrochemistry with the use of local conditions for MEA and channel design optimization [2], two 2D multi-physics models (2D model along the channel and 2D model at the rib/channel scale) are updated. The different reactions (hydrogen oxidation reaction, oxygen reduction reaction and the oxidation of the carbon support) are written in the general form [3-4]. The linking of the two models allows to simulate the transient potentials during startup and shutdown phase in two direction (along the channel and in the section of the MEA). In particular, during the injection of hydrogen in the channel, the reverse current mechanisms that accelerate the carbon support corrosion, is directly simulated without hypothesis. The validated models provide in-silico characterization to better explain the reverse current mechanisms and the interactions between the operating conditions of the cell and the local conditions in the catalyst layer. The CO2 concentration at the outlet of the channel is used as an observer to quantify the degradation of the carbon support. In a second step, different mitigation strategies are proposed. In particular, some strategies are studied to limit the high potential during the startup and shutdown phase (influence of the catalyst loading in the anode, external electrical resistance). Other strategies decrease the time during the reverse current mechanism (sensitivity of the hydrogen flow rate during the startup (Fig. sensitivity study of the hydrogen flow rate during startup on the cathodic potential and the CO2 production), design optimization of the rib/channel patern). These different strategies are explained and compared. An improvement of the carbon support corrosion is quantified and can be decreased by 50%. [1] Qiang Shen, Ming Hou, Dong Liang, Zhimin Zhou, Xiaojin Li, Zhigang Shao, and Baolian Yi. Study on the processes of start-up and shutdown in proton exchange membrane fuel cells. Journal of Power Sources, 189(2):1114–1119, 2009. [2] Bolahaga Randrianarizafy, Pascal Schott, Marion Chandesris, Mathias Gerard, and Yann Bultel. Design optimization of rib/channel patterns in a pemfc through performance heterogeneities modelling. International Journal of Hydrogen Energy, 43(18):8907 – 8926, 2018. [3] G. Maranzana, A. Lamibrac, J. Dillet, S. Abbou, S. Didierjean, and O. Lottin. Startup (and shutdown) model for polymer electrolyte membrane fuel cells. Journal of the Electrochemical Society, 162(7):F694–F706, 2015. [4] B. Randrianarizafy. Multi-physics modeling of startup and shutdown of a PEM fuel cell and study of the carbon support degradation: mitigation strategies and design optimization. PhD thesis, Communauté Université Grenoble Alpes, 2018. Figure 1

Numerical simulations for PEMFC (Proton Exchange Membrane Fuel Cell) for understanding and co-optimization of designs (bipolar plate with MEA (Membrane Electrode Assembly)) play a key role to achieve the objective of cost reduction for PEMFC. To improve the overall simulation of PEMFC used for the design validation step of the bipolar plates, very detailed physical mechanisms are included in the PEMFC models in order to describe the main electrochemical and transport mechanisms (local 2D models) [1]. Present limitation of these approaches is that they are difficult to integrate in higher scale simulations, in terms of geometry. Pseudo 3D models allow to integrate the real geometry of the bipolar plate, but with reduction in term of meshing and physics [2]. To go further, the objectives are: i/ to perform reference simulations with no compromise (in term of geometry and physics); ii/ to promote multi-physics model with open-source code, to the international community. TRUST platform is used. The platform TRUST-FC is built on the CEA/DEN thermohydraulic TRUST framework. This C++ framework is an open-source software package of Computational Fluid Dynamics (CFD) which supports massively parallel computations with a distributed memory model (MPI) [3]. TRUST-FC contains physical models for gas, liquid, electron, ion transport and heat transfer. A model of anisotropic heat conduction is developed in TRUST-FC for taking into account the in-plane and through-plane anisotropic conductivity according to the real deformation of the gas diffusion layer (GDL) by the bipolar plate in the mechanical assembly. The multi-scale modelling framework for solving PEMFC specific physics are also developed: electrochemical reactions, multi-components gas transport in porous media for GDL and catalyst layer, coupling of free flow with porous media flow [2]. The used physical models are validated on the CEA-LITEN multi-physics and multi-scale simulation platform MUSES built on Comsol Multiphysic [1-2] and the experimental measurements [4]. TRUST-FC advantages are the robust numerical methods and the massive parallelism that allows to simulate coupled multi-physics phenomena on large scale domains. TRUST-FC can currently simulate a real CAD design containing tens millions of elements and numerous state variables in a few hours on cluster. SALOME is used as the meshing and visualization tool [5]. A full simulation on a design of bipolar plate with the MEA is presented (Fig.: temperature profile on the bipolar plate, with the coupling of flow cooling and heat production of the reactions) and discussed and compared to pseudo 3D simulations on Comsol Multiphysics. [1] Randrianarizafy B., Schott P., Chandesris M., Gerard M. and Bultel Y. Design optimization of rib/channel patterns in a PEMFC through performance heterogeneities modelling. Int. J. Hydrogen Energy, 43(18):8907 8926, (2018) [2] Nandjou F.,Poirot-Crouvezier J.-P.,Chandesris M. and Bultel Y.A pseudo-3D model to investigate heat and water transport in large area fPEMg fuel cells - Part 1: Model development and validation. Int. J. Hydrogen Energy, 41(34):15545-15561, (2016) [3] https://sourceforge.net/projects/trust-platform/ [4] Robin C., Gerard M., d'Arbigny J., Schott P., Jabbour L. and Bultel Y. Development and experimental validation of a PEM fuel cell 2D-model to study heterogeneities effects along large-area cell surface. Int. J. Hydrogen Energy, 40(32):10211-10230, (2015) [5] https://www.salome-platform.org/ Figure 1

This work proposes a state observer as a tool to manage cost and durability issues for PEMFC (Proton Exchange Membrane Fuel Cell) in automotive applications. Based on a dead-end anode architecture, the observer estimates the nitrogen build-up in the anode side, as well as relative humidities in the channels. These estimated parameters can then be used at fuel cell management level to enhance the durability of the stack. This observer is based on transport equations through the membrane and it reconstructs the behavior of the water and nitrogen inside the channels without the need of additional humidity sensors to correct the estimate. The convergence of the output variables is proved with Lyapunov theory for dynamic operating conditions. The validation is made with a high-fidelity model running a WLTC (Worldwide harmonized Light vehicles Test Cycle). This observer provides the average values of nitrogen and relative humidities with sufficient precision to be used in a global real-time control scheme.