Multiphysics modeling of liquid-feed direct methanol fuel cells and characterization of diffusive transport properties of gas diffusion layers

Resumen

Polymer Electrolyte Membrane (PEM) fuel cells are leading candidates to replace today’s fossil-based energy economy, providing efficient and clean electric energy generation for the 21st century. The study of PEM fuel cells represent a multidisciplinary and dynamic field in which mechanical, chemical, and electrical engineering, as well as material design, converge and collaborate with each other, making research on this topic a continuous multiphysics challenge. Numerical modeling plays a crucial role for the analysis of the complex mass, charge, and heat transport phenomena that take place at the micrometric scales of the porous layers that make up the Membrane Electrode Assembly (MEA) of PEM fuel cells, and constitutes an essential tool for the optimization of fuel cell performance. The most promising PEM fuel cell technologies are hydrogen Polymer Electrolyte Membrane Fuel Cells (PEMFCs), and liquid-feed Direct Methanol Fuel Cells (DMFCs). Although major interest has nowadays shifted to high-performance PEMFCs for the automotive industry, liquid-feed DMFCs are attractive power sources for portable electronic devices due to the higher energy density and ease of handling and storage of liquid methanol. The aim of this thesis is to contribute to the understanding of both technologies using multiphysics and multiscale modeling techniques. A multiphysics macroscopic model of a liquid-feed DMFC that accounts for the effects of the inhomogeneous assembly compression of the Gas Diffusion Layer (GDL) is first presented. Then, the effective gas diffusive properties of GDLs under dry and partially water-saturated conditions are characterized by combining the Lattice Boltzmann Method (LBM) with X-ray Computed Tomography (XCT) images of carbon-paper GDLs. The achievement of these two objectives is divided into three tasks: 1. A Finite Element Method (FEM) model is developed to simulate the inhomogeneous assembly process of the GDL associated with the repetitive rib/channel pattern of the Bipolar Plate (BPP). The model fully accounts for the nonlinear orthotropic mechanical properties of carbon-paper GDLs, thereby providing a more realistic characterization compared to isotropic models extensively used in the literature. The proposed model, conveniently validated against previous experimental data, enables the calculation of the GDL porosity distribution, GDL intrusion into the channel, and the contact pressure profiles at the interfaces of the GDL with the BPP and the catalyst-coated membrane. This analysis constitutes a necessary first step towards the development of multiphysics Computational Fluid Dynamics (CFD) models of either PEMFCs or DMFCs aiming to explore the effects of assembly compression. For a given GDL compression ratio, the results show that a combination of channel width, GDL thickness, and shear modulus dominates the transmission of stresses from the rib to the unloaded region below the channel, whereas an accurate description of the nonlinear through-plane Young’s modulus is needed to properly capture the GDL compressive response under the rib. 2. A multiphysics multiphase isothermal CFD model of a liquid-feed DMFC is then developed and presented. The model is progressively sophisticated in three steps: i. The first studies are conducted on a 2D/1D across-the-channel model that includes a 2D two-phase description for the anode GDL and a local 1D single-phase description for the remaining components of the MEA, i.e., catalyst layers, membrane, and cathode GDL. The model incorporates the effect of non-uniform mass and charge transport properties of the anode GDL induced by the cell assembly process simulated with the FEM model (Task 1). The effective mass and charge transport properties of the GDL are correlated as a function of porosity using experimental data from anisotropic carbon paper. ii. The above 2D/1D across-the-channel model is upgraded to account for the effect of electrical contact resistances at the GDL/BPP interface, the diffusive resistance of thin anode and cathode Microporous Layers (MPLs), and the effect of assembly compression on the 1D single-phase description of the cathode GDL. iii. The 2D/1D across-the-channel model is further improved by including a fully 2D two-phase description for the cathode GDL, instead of the 1D single-phase formulation adopted in the two previous steps. The model also accounts for the effect of electrical contact resistances between the GDL-MPL diffusion medium and the membrane, and includes hydrogen evolution kinetics at the anode to give a realistic description of the electrochemical processes that occur under oxygen-starved conditions. The proposed 2D/1D across-thechannel model is also extended to a 3D/1D model and combined with 1D two-phase models for the anode and cathode channels, leading to an advanced 3D/1D + 1D model that is successfully validated against previous experimental data. Among other conclusions, the results show that fully hydrophobic relationships widely used in the literature to model capillary transport of carbon dioxide in the anode GDL lead to unrealistic results when inhomogeneous GDL compression effects are taken into account. By contrast, more realistic results are obtained when GDLspecific capillary pressure data including the mixed-wettability characteristics of GDLs are considered. The results also show that, in agreement with previous experimental data, there is an optimum assembly compression level that maximizes the cell performance due to the trade-off between ohmic and mass transport losses; the optimal compression level being strongly dependent on BPP material and, more weakly, on the actual working conditions. Beyond the GDL compression, there is an optimum methanol concentration that maximizes the power output due to the trade-off between anode polarization losses and cathode mixed overpotential caused by methanol crossover. For a given methanol solution, DMFC performance is largely affected by the oxygen supply rate, operating temperature, and gas (liquid) saturation level at the anode (cathode) GDL/channel interface. 3. The effective diffusivity of GDLs is characterized under both dry and wet conditions by performing pore-scale LBM simulations on XCT images of carbon-paper GDLs undergoing water-invasion experiments. Under dry conditions, the results show a good agreement with previous experimental data reported for morphologically similar GDLs. Under wet conditions, it is shown that the spatial distribution of water across the porous medium has a major effect that is not accounted for by the average (or total) amount of water contained in the porous medium. Specifically, it is found that the existence of local bottleneck regions near the invasion face drastically limits the diffusive flux through the porous medium. This finding, traditionally ignored in previous studies, has an important repercussion for two-phase macroscopic continuum models. In a subsequent step in the investigation, it is shown that macroscopic models require effective properties determined under uniform porosity and saturation conditions to provide a physically-consistent macroscopic formulation. Constitutive relationships for the effective diffusivity suitable for use in macroscopic models are determined from a massive computational campaign (?2,500 simulations) considering GDL representative subdomains with locally homogeneous porosity and saturation as a proxy for representative elementary volumes. Using arithmetic and harmonic upscaling rules, it is confirmed that the correlations determined on the local scale are able to recover the global data obtained on the inhomogeneous full GDL domain. Moreover, good agreement is found for the under-the-channel region when the local correlations are upscaled to previous global data from running fuel cells. The results indicate, however, that the blockage of local diffusive transport in the under-the-rib region is larger, presumably due to water condensation and interferences of water fingers with the rib walls. Both effects were not present in the X-ray tomography data, which only considered capillary invasion, and should be addressed in future work.