Validation of a three dimensional PEM fuel cell CFD model using local liquid water distributions measured with neutron imaging

Depósito de Investigación de la Universidad de Sevilla https://idus.us.es/ This is an Accepted Manuscript of an article published by Elsevier in International Journal of Hydrogen Energy, Vol. 39, Issue 13 on April 2014, available at: https://doi.org/10.1016/j.ijhydene.2014.02.115 © 2014 Elsevier. En idUS Licencia Creative Commons CC BY-NC-ND 1 Validation of a three dimensional PEM fuel cell CFD model using local liquid water distributions measured with Neutron Imaging Alfredo Iranzo a*, Pierre Boillat b, Felipe Rosa c a AICIA, School of Engineering, Sevilla, Spain b Electrochemistry Laboratory (ECL), Paul Scherrer Institut (PSI), Switzerland c Thermal Engineering Group, Energy Engineering Department, School of Engineering, University of Sevilla, Spain (*) corresponding author: email: airanz[email protected]; Tel. (+34) 954 487471; Fax (+34) 954 463153; Camino de los Descubrimientos, s/n, 41092, Sevilla, Spain Abstract This work presents the validation carried out for a three dimensional CFD 50 cm2 PEM fuel cell model, particularly focus on the prediction of liquid water distributions within the cell. The CFD model was previously validated against a set of experimental polarization curves, where model results adequately matched the experimental curves. An extension of the validation is presented in this work, by performing a comparison of the local liquid water distributions predicted by the model with the liquid water distributions of the real cell. The experimental measurements were obtained by means of Neutron Imaging, where a set of different cell operating conditions was tested. Although the exact quantitative results are not directly comparable due to differences in the cell setup, qualitative results show a very good agreement between the model results and the water distributions observed in the neutron radiographs. A model validation approach using local variable distributions (such as liquid water in this case) in addition to the integral quantities (i.e. polarization curves) is necessary to ensure the validity of models. Keywords: PEM Fuel Cell, CFD, numerical model, Neutron imaging, liquid water, model validation *Manuscript Click here to view linked References 2 1 Introduction The mass transport losses in Polymer Electrolyte Membrane Fuel Cells (PEMFCs) are the major restriction for the operation at high current densities. This is limiting the required reduction of stack size, weight, and cost of current PEMFCs. In addition to the oxygen diffusion limitations in dry GDLs, water produced by the cathode electrode at high current densities is known to block the GDL pores and electrode reaction sites, causing the flooding of the cell and preventing oxygen from effectively reaching the catalyst active sites [1-4]. Also an excess of liquid water or a dry-out membrane will result in significant durability issues [5,6]. Therefore, in order to ensure a successful operation of the cell an appropriate water balance must be achieved, which has been the objective of many research efforts in PEMFCs, mainly by means of experimental research and development of computational models. The modelling and simulation of PEMFCs constitutes a major working field in the research activities, with many research groups and modelling approaches being developed [7-10]. Some of the conclusions that can be drawn from a literature review are the following: -Multiphase flow is one of the physical phenomena more difficult to correctly model [7]. The effects of water transport and distribution are also of major importance for the cell operation and performance, and therefore the lack of accurate models, valid for all regimes of flow and cell operation, is a concern in the modelling activities. -Model validation is absolutely fundamental for the development and confident use of models, and the derivation of conclusions based on model results. It is known that the validation using only the polarization IV curves is typically not sufficient as stated by Wang [7], as the polarization curve is an integral output of the cell and the coupling of the different phenomena occurring within the cell can provide the same or similar polarization curves under certain circumstances, without being each of the phenomena correctly modelled. Therefore the validation against local quantities is necessary [7]. These quantities can be species distributions, current density distributions, or liquid water distributions measured by neutron imaging as reported by Weber and Hickner [11], Wand and Chen [12] or Quan and Lai [13]. This work presents this type of validation for a previously developed CFD model for a 50 cm2 PEMFC, where model results and validation against polarization curves (for a set of different operating conditions and bipolar plate designs) were presented previously in Iranzo et al. [1415]. The new validation has been performed using local liquid water distributions obtained with neutron imaging. This is therefore a more exhaustive validation and is presented in this work. 2 Experimental 2.1 Cell description The cell hardware used was a 50 cm2 cell from ElectroChem Inc., where a five-channel serpentine flow field was used. The original graphite Bipolar Plate could not be used in the neutron imaging experiments given its high neutron absorptivity. Therefore the original plate design (graphite, 9 mm thickness) was replicated and machined in aluminium (4 mm thickness) which is in practice transparent to the neutron beam. The plates were coated in order to protect them against corrosion during the cell operation. Figure 1 show the Bipolar Plate design, where 3 the details of the flow field are presented: cross-flow layout between anode and cathode (with the cathode channels in vertical direction), nine blocks in serpentine flow field configuration, with 5 parallel channels per block. Channel width is 0.71 mm and rib width is 0.86 mm. Channel depth is 1.1 mm. Each set of ten channels is connected to a common manifold (with 5 incoming channels and five outgoing channels), that has a depth of 1.7 mm and a width of 1.5 mm. It must be noted that given the differences in thickness and electrical conductivity of graphite and aluminium coated plates, the cell voltage will not be comparable between the two sets of experiments (i.e. the original experiments with graphite BPs for IV validation in [14,16], and the present neutron imaging experiments with aluminium coated BPs). Therefore the cell voltage will not be used for comparison or validation. Instead, the liquid water distributions at equal current densities will be used. It must be noted that in addition to the difference in BP electrical resistivity, the surface wettability of the coating material is also different from graphite, as observed in Figure 2. This will influence the capability of the gas flow to remove the liquid water present along the channels, with clearly more resistance to be flushed out for the coated aluminium plate used in the neutron imaging experiments. This effect will modify the liquid water content in the channels for the two sets of experiments (original experiments with graphite BPs for IV validation in [14, 16] and neutron imaging experiments with aluminium coated BPs). A Catalyst Coated Membrane (CCM) from Baltic Fuel Cells was used, with catalyst loading at anode and cathode electrodes 0.3 mg Pt/cm2 and 0.6 mg Pt/cm2 respectively. The Ionomer/Carbon ratio (I/C ratio) is 1.2/2, with 70% Pt over carbon and Ionomer/Catalyst ratio 1/4. The membrane material is Nafion-117. A set of Gas Diffusion Layers from SGL Group (Sigracet 24 BC) 0.24 mm thick were used for the neutron imaging experiments. The original modelling work and experimental validation reported in [14-16] were performed with Sigracet 10 CC (0.42 mm thick) GDLs. In addition to the thickness, other differences are the porosity (of 0.76 vs. 0.82) and the PTFE content (5% vs. 10%). This is also expected to be a source of differences in the cell performance and quantitative liquid water content, but again the overall and qualitative liquid water distributions are expected to be similar between all experiments. The cell reactants are supplied through the ports in the upper right corner of the active area (Figure 3), whereas the anode and cathode outlets are located in the lower left corner of the active area. Anode and cathode Bipolar Plates are arranged in a cross-flow layout (Figure 1), where the anode gas flows in horizontal channels (in a serpentine-like pattern) and the cathode gas flows in vertical channels (also in a serpentine-like pattern). The cell temperature was measured in four different locations of the bipolar plates (outer sides of the bipolar plates, two measuring points for the anode plate and two measuring points for the cathode plate). The outer sides of the bipolar plates are external boundary conditions of the CFD model, where a temperature of 60 ºC was defined for all the simulations. Therefore, the experimental setup included an external cooling air flow that was adjusted to ensure that the temperature measurement was at 60 ºC during all the experiments (60 ºC +/- 2ºC were achieved). 4 2.2 Experimental procedure The cell operation was performed in galvanostatic mode at the desired current densities. Cell temperature and pressure was 60 C and 2 bar(a) respectively. Table 1 shows the operating conditions tested. Neutron images were continuously recorded during the operation. Radiograms were performed at the NEUTRA beam line [17-19] of PSI, with an Andor Neo camera. Exposure time was 10s. The sensitivity/dynamic range was 16-bit (low noise and high well capacity). After performing the necessary corrections (detector background, change of beam intensity, neutrons scattered by the setup), the radiograms were referenced pixel-wise by dividing the obtained image by a reference image of the dry cell before operation (depicted in Figure 3). With this operation, the neutron attenuation corresponding exclusively to water can be obtained. The thickness of water δw is calculated from the relative neutron transmission (I/I0) by inverting the Lambert-Beer law:    ⁄ /Σ (1) where ∑ is the attenuation coefficient of neutrons in liquid water, with a value of 3.5 cm-1 for the given setup. The obtained liquid water thickness δw can be converted into volume fraction by dividing δw by the depth of the active area in the beam direction (through-plane direction). During the initial examination of the neutron radiographs it was observed that in general a significant amount of liquid water was present in the channels, especially in the cathode side. In order to allow for the determination of the liquid water content distributions in GDL/MEA only (where the CFD model is able to better predict liquid water as discussed in section 4 and in [14]), it was necessary to flush out the liquid water in the channels. This is possible by performing a sudden decompression/purging in the anode and cathode sides. This has been described in Owejan et al [20], and has been adapted as a tool for neutron imaging visualization by Boillat et al [21]. The method consists of decompressing the anode and cathode channels, in order to flush out the liquid water in the channels. When this operation is performed in a short period of time, the water content in GDL/MEA is not significantly affected (it was verified that the High Frequency Resistance of the cell was not modified) and therefore quantitative analysis can also be done with confidence [21]. 2.3 Experimental results The neutron images recorded for the different operating conditions tested (Table 1) are presented in Figure 4, where the images on the left correspond to the cell in steady state conditions, and the images on the right correspond to the same cell after the decompression/purge operation. The images on the right therefore represent the state of the cell without liquid water in anode/cathode channels. It can be observed that liquid water was present in the cathode channels. The original neutron radiographs (grey scale) were coloured with a yellow/blue mask for visual presentation purposes (where blue colour indicate liquid water). The curved distortion observed in the upper right area corresponds to the neutron absorption of the inlet pipes insulation, which could not be removed from the images with the referencing 5 procedure (as the position of the insulating material slightly varied during the set of experiments due to thermal expansion). 3 CFD Model description The 3D CFD model was developed using ANSYS-FLUENT software [22]. The complete description of the cell model, parameters and experimental validation were fully reported in Iranzo et al. [14-16]. CFD Best Practice Guidelines were used for ensuring appropriate mesh and solver parameters. The model parameters were identified either from the cell supplier data sheets, values and correlations found in the open literature, and values obtained from dedicated experiments. The comprehensive list of parameters used in the CFD model is reported in Iranzo et al. [14], and are reproduced in Table 2 for the sake of completeness. Only the reference exchange current density of the electrode was unknown and was used as a model parameter to fit the experimental IV data. The proposed value of the reference exchange current density for the electrode was consistent with the typical values reported in the literature, and provided accurate results for the full set of operation conditions and bipolar plate designs modelled and tested for the real cell in the test bench [15]. The model was considered to be validated, although a more exhaustive validation work using local distributions of liquid water is presented in the next section. 4 Validation results and discussion The CFD model was already validated in terms of polarization curves, for different operating conditions and two different bipolar plate designs. The details of this validation can be found elsewhere [14-16]. Despite the fact that the broad validation conditions tested are an indication that the model predictions are valid, it is true that a comprehensive validation work must include local variable distributions as stated by Wang [7]. In this work, the validation using locally resolved liquid water distributions (through-plane direction) is presented. The main facts to be accounted for during the validation are the following: -Bipolar Plates are different in terms of materials and thickness, as stated in section 2.1. This affects the voltage and the water removal capability of the cell (given the different surface wettability). -The GDLs used were similar but not identical, as discussed in section 2.1. -A direct quantitative validation will not be completely possible given the differences in the GDL type, Bipolar Plate, and also given the differences in the cell pressure (4.0 bara for the experiments and CFD model vs. 2.0 bara for the neutron imaging experiments). However, despite the facts analysed, the analysis of the qualitative trends of the CFD model results when compared to the neutron imaging results is of high value for the model validation. -The CFD software used (ANSYS-FLUENT [22]) features a multiphase model to account for the liquid water content in the channels. It is the so-called saturation model [22], where the liquid phase in the channels is modelled as a fine mist with the same velocity field as the gas phase. The model is therefore accurate and suitable only for mist-type flow regimes, which are 6 achieved only for superficial gas velocities over 10 m/s [23, 24]. However this is not the case in the cell analysed where the typical velocity in the cathode side is 1-2 m/s. Therefore the multiphase model is not able to model the liquid water accumulation in the channels. In this sense, the main and straightforward conclusion is that the CFD model completely fails when compared to the neutron imaging results, where severe water accumulation in the cathode side channels is observed in Figure 4 for most of the conditions tested. Having considered this, it is however possible to perform the validation of the CFD model with the neutron radiographs where the channels liquid water is removed via the decompression/purging method, introduced by Boillat et al [21] and briefly described in section 2.2. This enables the possibility of using the liquid water distribution in GDL/MEA only, to be compared between experimental neutron radiographs and CFD results for the purpose of model validation. 4.1 Effect of current density In order to verify the validity of the model for the prediction of the water distributions at different current densities, the distributions obtained with the experimental setup and the numerical model are presented in Figure 5 for 10A (top) and 25A (bottom) for the case with H2 λa = 1.5, air λc = 5.0, RHa = 60%, RHc = 60%. Liquid water distributions are shown in Figure 5, where the images on the left correspond to the neutron radiographs of the cell, first in steady state operation and secondly after applying the decompression/purge method, i.e. without liquid water in channels (as it would be predicted by the CFD model). The images on the right show the water saturation (liquid volume fraction) results of the CFD model, first at the cathode channel-GDL interface and secondly at the cathode GDL-electrode interface. As observed in Figure 5, the liquid water distributions predicted by the CFD model agree well with the ones observed in the experimental setup with neutron imaging. In the MEA/GDL region influenced by the inlet the amount of liquid water is low, with a progressive accumulation of water towards the outlet. These results are consistent as well with the typical behaviour known for serpentine flow fields [25], featuring a relative dry-out in the entrance region (upper-right corner in Figure 5) and a progressive water accumulation towards the outlet region (lower-left corner in Figure 5). When the cell is set at 25A instead of 10A, less water is observed in the cell both in the CFD model and in the neutron radiographs. The reason is that increasing the current density increases as well the water generated in the cathode, but as the cell is operated with constant stoichiometric factors, the anode and cathode flows are also increased. For the particular case analysed, the increase of flow especially at the cathode is strong enough for inducing a higher degree of water removal from the channels and GDL/MEA In addition, a higher current density also increases the heat generation at the cathode and therefore its local temperature, leading to more water being transported as vapour phase and therefore favouring a relative drying of the MEA. The liquid water distribution in the channels observed in Figure 5 (far left) is not predicted in the CFD model, given its particular multiphase model formulation (saturation model), where water is treated as a very fine mist with same velocity field as the gas phase, and therefore not prone to accumulation. In addition to the local distributions, a quantitative analysis of the liquid water content is presented in Table 3. In the table, NI δw represents the liquid water thickness (average over the active area) measured in the neutron radiographs and calculated according to Eq.(1). From this 7 value, the total water volume in the cell can be calculated and compared against the value provided by the CFD model. For each condition (10 and 25 A) two different values of water volume are provided: total water within the cell (full cell) and water within GDL/MEA (excluding the channels). These two conditions correspond to the pairs of images in Figure 4 and Figure 5. Additionally, the % of variation in the liquid water in GDL/MEA when the conditions are switched from 10 to 25 A is indicated in Table 3. A first observation in the Table 3 is that the experiments show that a significant amount of water accumulated in the channels, which cannot be adequately predicted by the CFD model as already discussed. When comparing the quantitative values of liquid water in GDL/MEA only, it is observed that the CFD model does not provide exactly the same amount of water as measured in the experiments (probably due to the differences in both “setups” as discussed above). The total amount of liquid water is higher in the experiment than in the CFD model. Presumably the higher amount of water is caused by the water accumulated in the channels in the real cell (a phenomenon which is amplified by the non-hydrophobic nature of the channels walls as observed in Figure 2) and not adequately modelled in the CFD model. It will be shown that in fact all experiments carried out with the cell set up for neutron imaging, showed higher water content than the corresponding CFD model. However, it is observed that the relative variation in the liquid water content when switching to 10 to 25 A is very similar in both the experiment and the CFD model result (28% - 29%). In both cases the amount of water is reduced despite the fact that current is increased and therefore the water generation. The reason for the reduction in the water content has been already discussed above. 4.2 Effect of Relative Humidity In order to verify the validity of the model for the prediction of the water distributions at different reactants relative humidity, the distributions obtained with the experimental setup and the CFD model are presented in Figure 6 for 60%/60% (top) and 35%/35% (bottom) for the case with H2 λa = 1.5, air λc = 5.0, and 10 A. The same general trends are observed as in the previous case, where the case with less relative humidity results in less water content both in the experiment and in the CFD results. The quantitative analysis of the liquid water content is presented in Table 4, where again the trends observed before are reproduced in the analysis. The exact quantitative amount of liquid water is not the same in the experiment and CFD results, but the relative variation of water content when switching from 60% to 35% is very similar in both cases. 4.3 Effect of stoichiometric factors In order to verify the validity of the model for the prediction of the water distributions at different stoichiometric factors, the distributions obtained with the experimental setup and the numerical model are presented in Figure 7 for λc = 5.0 (top) and λc = 3.5 (bottom) for the case with H2 λa = 1.5, 60% RH, and 10 A. In the figure it is observed that reducing the cathode stoichiometric factor increases the water content in the cell, as the water removal capability of the cathode air flow is being reduced. This trend is observed also in the quantitative analysis of the liquid water content presented in Table 5, where again the relative variation in the liquid water content when changing the operating conditions is similar in both experimental results and CFD model results. 8 4.4 Effect of oxygen feed in the cathode side In order to verify the validity of the model for the prediction of the water distributions when operating with different oxidant concentrations (pure oxygen), the distributions obtained with the experimental setup and the numerical model are presented in Figure 8 for the case with H2 λa = 1.5, λc = 5.0 with air (top), λc = 10.0 with O2 (bottom) 60% RHa, 60% RHc (top), 90% RHc, (bottom), and 10 A. The comparison of the distributions shows that more liquid water is accumulated when the cell is fed with oxygen at the indicated conditions. This is also qualitatively reproduced by the CFD model (note the different scales in the CFD legend of Figure 5). However, the quantitative analysis for this case shows that the relative variation is not similar in the experiments and CFD model results. The latest predicts a significantly more influence than actually observed in the neutron radiographs. This is observed in the quantitative analysis provided in Table 6. 4.5 Liquid water profile along the cell In addition to the through-plane water content distributions presented in the previous sections, a comparison of the liquid water profiles was also carried out so that a more complete verification of the results is made. For the base case (10 A, H2 λa = 1.5, air λc = 5.0, RHa = 60%, RHc = 60%), the liquid water profile along the cell has been represented in Figure 9, both for the corresponding neutron radiograph and the CFD model results. The exact location where the profile is represented is shown in the top of Figure 9 (black line along the centre of the cell). As the profile obtained from the neutron radiograph was not smooth (the image post-processor provides the profile pixel-wise), a third profile is presented in the plot (MidPlane NI smoothed), which is an artificially smoothed profile obtained from the original neutron radiograph, to allow for a better visual interpretation. The resulting profiles shown in Figure 9 present essentially the same trend. Although the absolute values are different, with more liquid water for the neutron imaging experiments as already discussed, the evolution of the profile along the coordinate is essentially the same: less water is observed in the right side of the cell where the inlet is located providing non-saturated air, and progressively more water is accumulated towards the left side of the cell where the outlet is located, as air becomes saturated and is not able to effectively removes the water being generated. 5 Conclusions This work presents a validation of a previously developed CFD model for a 50 cm2 PEM fuel cell. The validation was performed by comparing the local liquid water distributions obtained from the CFD model with experimental measurements where neutron imaging was used. The model had been previously validated in terms of the polarization curve for a set of different operating conditions and bipolar plate designs, but a more in-depth validation is presented in this work, comparing the local distributions of liquid water content within the cell in the through-plane direction. A major conclusion is that the CFD model used is not able to reproduce the liquid water accumulated in the channels, clearly observed in the neutron radiographs but 15 Table 3. Quantitative liquid water content for model validation (H2 λa = 1.5, air λc = 5.0, RHa = 60%, RHc = 60%). 10A vs. 25A. NI δ w (cm) NI water (ml) % 10-25A (GDL/MEA) CFD water (ml) in GDL CFD water (ml) in GDL/MEA % 10-25A (GDL/MEA) 10A a full cell 0.022 1.079 0.162 0.487 10A b GDL/MEA 0.012 0.607 0.162 0.487 25A a full cell 0.013 0.667 0.153 0.344 25A b GDL/MEA 0.009 0.437 28.1 0.152 0.344 29.3 Table 4. Quantitative liquid water content for model validation (H2 λa = 1.5, air λc = 5.0, 10A). RHa = 60%, RHc = 60% vs. RHa = 35%, RHc = 35%. NI δ w (cm) NI water (ml) % 60-35% (GDL/MEA) CFD water (ml) in GDL CFD water (ml) in GDL/MEA % 60-35% (GDL/MEA) 60/60 a full cell 0.022 1.079 0.162 0.487 60/60 b GDL/MEA 0.012 0.607 0.162 0.487 35/35 a full cell 0.015 0.728 0.079 0.321 35/35 b GDL/MEA 0.008 0.394 35.2 0.079 0.321 34.1 16 Table 5. Quantitative liquid water content for model validation (10A, RHa = 60%, RHc = 60%, H2 λa = 1.5, air). λc = 5.0 vs. λc = 3.5. NI δ w (cm) NI water (ml) % 5.0-3.5 (GDL/MEA) CFD water (ml) in GDL CFD water (ml) in GDL/MEA % 5.0-3.5 (GDL/MEA) 1.5/5.0 a full cell 0.022 1.079 0.162 0.487 1.5/5.0 b GDL/MEA 0.012 0.607 0.162 0.487 1.5/3.5 a full cell 0.023 1.160 0.194 0.526 1.5/3.5 b GDL/MEA 0.013 0.636 -4.7 0.193 0.526 -7.9 Table 6. Quantitative liquid water content for model validation. 10A, H2 λa = 1.5, λc = 5.0 (air), λc = 10.0 (O2) 60% RHa, 60% RHc (air), 90% RHc, (O2). NI δ w (cm) NI water (ml) % 60-35% (GDL/MEA) CFD water (ml) in GDL CFD water (ml) in GDL/MEA % 60-35% (GDL/MEA) Air 5.0 60% full cell 0.022 1.079 0.162 0.487 Air 5.0 60% GDL/MEA 0.012 0.607 0.162 0.487 O2 10.0 90% full cell 0.027 1.361 0.303 0.646 O2 10.0 90% GDL/MEA 0.013 0.630 -3.7 0.300 0.642 -31.9 Figure1 Click here to download high resolution image Figure2 Click here to download high resolution image Figure3 Click here to download high resolution image Figure4 Click here to download high resolution image Figure5 Click here to download high resolution image Figure6 Click here to download high resolution image Figure7 Click here to download high resolution image Figure8 Click here to download high resolution image