Predictive system-level modeling framework for transient ...
Transcript of Predictive system-level modeling framework for transient ...
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
1
Predictive system-level modeling framework for transient
operation and cathode platinum degradation of high temperature
proton exchange membrane fuel cells
Ambrož Kregara,b,*, Gregor Tavčara, Andraž Kravosa, Tomaž Katrašnika
aUniversity of Ljubljana, Faculty of Mechanical Engineering, Aškerčeva 6, SI-1000 Ljubljana, Slovenia
bUniversity of Ljubljana, Faculty of Education, Kardeljeva ploščad 16, SI-1000 Ljubljana, Slovenia
*Corresponding author. E-mail address: [email protected]
Abstract
High temperature proton exchange membrane fuel cells (HT-PEMFCs) are a promising and emerging
technology, which enable highly efficient, low-emission, small-scale electricity and heat generation.
The simultaneous reduction in production costs and prolongation of service life are considered as major
challenges toward their wider market adoption, which calls for the application of predictive virtual tools
during their development process. To present significant progress in the addressed area, this paper
introduces an innovative real-time capable system-level modeling framework based on the following:
a) a mechanistic spatially and temporally resolved model of HT-PEMFC operation, and b) a degradation
modeling framework based on interacting individual cathode platinum degradation mechanisms.
Additional innovative contributions arise from a consistent consideration of the varying particle size
distribution in the transient fuel cell operating regime. The degradation modeling framework
interactively considers the carbon and platinum oxidation phenomena, and platinum dissolution,
redeposition, detachment, and agglomeration; hence, covering the entire causal chain of these
phenomena. Presented results confirm capability of the modeling framework to accurately simulate the
platinum particle size redistribution. Results clearly indicate more pronounced platinum particle growth
towards the end of the channel since humidity is the main precursor of oxidation reactions. In addition,
innovative modeling framework elucidate contributions of agglomeration, which is more pronounced at
voltage cycling, and Ostwald ripening, which is more pronounced at higher voltages, to the platinum
particles growth. These functionalities position the proposed modeling framework as a beyond state-of-
the-art tool for model-supported development of the advanced clean energy conversion technologies.
Keywords: Fuel cell; Proton-exchange membrane; High temperature; Modeling; Mechanistically
based; Platinum degradation
1. Introduction
Global concerns on sustainable energy use and environmental protection call for innovative energy
conversion technologies. Among the alternatives, high temperature proton exchange membrane fuel
cells (HT-PEMFCs) are a viable solution for highly efficient, low-emission, small-scale electricity and
heat generation. The advantages of HT-PEMFCs in small-scale stationary power generation arise mainly
from their ability to use reformed hydrocarbons directly (in particular the widely available natural gas
or bio-methane), as well as in their relatively simple design compared to non-PEM based technologies.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
2
However, to achieve a higher market share and large market penetration (even without subsidies),
significant cost reduction needs to be achieved while simultaneously ensuring a long service life [1]. To
attain this objective, further optimization is required on the component and system levels. This study
focuses on the system level, because the optimization of an HT-PEMFC system—comprising
component sizing and devising control logic—crucially influences its performance and service life.
To address the requirements of shorter product development cycles and reduced development cost
efficiently, while striving to approach engineering limits in power density and service life, it is necessary
to rely on modeling and simulation tools in the development process of HT-PEMFC systems. While
three-dimensional (3D) computational fluid dynamics (CFD) models of PEMFCs, e.g., AVL FIRE [2],
Ansys Fluent [3], CFX [4], CFD-ACE+ [5], OpenFOAM [6], COMSOL Multiphysics [7], and STAR-
CD [8], have already reached a relatively high level of maturity, there are still major challenges to be
solved in the area of system-level PEMFC models. These models are specifically applied to designing
the entire system. For HT-PEMFCs, this includes the stack, and the balance-of-plant (BoP) components
Nomenclature
𝛼 electron transfer coefficient
𝜎 surface tension
Γ surface density of sites available for oxidation
𝜀 volume fraction
Θ fraction of surface covered by oxide species
𝜌 density
𝑎 activity
𝑏 Tafel slope reduced for transfer coefficient
𝑐 concentration of species
�� average conc. in computational domain
𝑑 width
𝐷 diffusion coefficient matrix
𝐸 equilibrium potential
𝑓𝑁 continuous particle size distribution
𝐹 Faraday constant
𝐻 Heaviside theta function
𝑘 reaction rate constant
𝑘𝑟𝑒𝑣 reverse reaction deceleration
𝑙 depth of computational slice
𝑚 mass loading
𝑀 molar mass
𝑁𝑖 particle size class population
𝑟 reaction rate
𝑅 universal gas constant
𝑅𝑖 particle radius
𝑆 surface
𝑣 velocity
𝑉 scalar velocity potential
𝑉 volume
𝑈𝐹𝐶 fuel cell voltage
𝑈𝑐𝑎𝑡 local voltage in cathode catalyst layer
𝑋𝑆 renormalization factor
𝑧𝑒𝑛𝑡𝑒𝑟 coordinates of channel entrance
𝑧𝑒𝑥𝑖𝑡 coordinates of channel exit
Abbreviations
𝐶 carbon
𝐶𝑂 stable 𝐶 = 𝑂 carbon surface group
𝐶𝑂2 carbon dioxide
𝐶𝑂𝐻 𝐶 − 𝑂𝐻 carbon surface group
𝑃𝑡 platinum
𝑃𝑡2+ platinum ion (2+)
𝑃𝑡𝑂 stable 𝑃𝑡 = 𝑂 platinum surface group
𝑃𝑡𝑂𝐻 𝑃𝑡 − 𝑂𝐻 platinum surface group
𝐻+ proton
𝐻2𝑂 water
𝑎𝑔𝑔 agglomeration
𝑎𝑛 anode
𝑎𝑡𝑡 attachment
𝑐𝑎𝑡 cathode
𝑐𝑜𝑟𝑟 corrosion
𝑑𝑒𝑡 detachment
𝑑𝑖𝑠𝑠 dissolution
𝑖𝑜𝑛 ionomer
𝑚𝑎𝑥 maximum
𝑚𝑒𝑎𝑛 mean
𝑚𝑒𝑟 merging
𝑜𝑥 oxidation
𝑡𝑜𝑡𝑎𝑙 total
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
3
(the thermal management component and the control unit). The ultimate aim is to design a system that
meets the required performance and service life. Furthermore, in the common development process, the
system-level models are applied in the early development stages for exploration of the design space,
designing the system, and subsequently in the validation and calibration phases, where they serve as
hardware-in-the-loop (HiL) applications and are used for the development of control functionalities and
strategies.
To support tasks in the early development stages, a system-level model must feature a high level of
predictiveness and generality, in addition to short computational times that enables the assessment of
numerous system configurations. In the validation and calibration phases (and especially in HiL
applications), the real-time capability of the model and its realistic response to variations in actuator
parameters are required. A high level of predictiveness of the models applied in the early development
stages is clearly motivated by the fact that in those development phases, hardware components are not
available. In addition, in both early and late development stages, the high level of predictiveness (and
thus realistic response to variations in actuator parameters) are required when approaching engineering
limits to support efficient exploration of the very large design space including selection and control of
the BoP components, which greatly influence the optimal parameters of the fuel cell (FC) stack.
Therefore, applicability of the data driven models, e.g., [9–13], which are commonly applied in the
system level, is limited to studies aimed at combined performance and service life optimizations. This
is because the accuracy of the data-driven model does not reach beyond the trained variation space of
parameters, and that it is generally not practically possible to train the model over a variation space of
parameters that features a very large dimensionality.
A further challenge motivating high level of predictiveness of the models arises from the objective of
ensuring prolonged service life, which requires coupling of the model of FC operation with FC
degradation models. For the case of the FC catalyst layer degradation, it is well known [14–16] that the
local rates of platinum (Pt) dissolution, redeposition, migration, and agglomeration, which will
subsequently be denoted as individual degradation mechanisms (Fig. 2 a)), strongly depend on the local
values and temporal dynamics of temperature, reactant concentrations, and electric potential field
(subsequently denoted as degradation stimuli). This in turn means that the degradation processes are
only modeled in a sufficient level of detail if spatial and temporal variations in degradation stimuli,
provided by the FC operation model, are considered. Such an approach is not possible with the use of
lumped degradation models, in which the entire chain of individual degradation mechanisms is
substituted with a simple empirical model, for example, in [17–20]. Because the causal chain of
degradation processes is poorly modeled in such an approach, its usefulness in supporting the FC
performance and degradation optimization in the early development stages and under a wide range of
parameter variations is very limited.
Models addressing individual degradation processes in the FC catalyst layer have already been
established. The carbon corrosion was modelled either by simple electrochemical model [21], by
coupling between carbon and Pt surface groups [14,22] or by more complex microscopic models [23].
A wide variety of approaches was proposed for modelling Pt surface oxidation, describing it by single
[16,24] or multi-step [14,25] electrochemical process, or coupled with other electrochemical reactions
[25,26], leading to its dissolution [16,26–28], redeposition [15] and migration in the membrane [15,27].
The effects of Pt detachment and agglomeration [29], and Ostwald ripening [28,30] on changes in the
Pt particle size distribution in the catalyst layer and the consequent loss of active surface have also been
studied using simplified models of these processes [31,32]. The degradation of alloyed Pt catalysts was
also studied by Franco et al. in [33]. The models of these individual processes have also been combined
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
4
into a more extensive multi-scale models, taking into account the interplay between degradation
processes. This was done by modelling the detailed dynamics of individual degradation processes on
different scales using methods such as coarse-grained molecular dynamics and kinetic Monte Carlo
methods, and combining the results into a fully coupled degradation model [23,31,34,35] The feedback
effects of degradation of individual fuel cell components on the degradation stimuli were also modeled
[36,37]. The reader is referred to the review article, recently published by Jahnke et al., for an extended
overview of the state-of-the-art in catalyst degradation modelling of PEM-FC [38].
In addition to the degradation of the Pt catalyst, membrane degradation is another important factor in
determining the FC performance and lifetime [38]. Several membrane degradation models have been
proposed in the literature, mostly focused on degradation of Nafion or similar membranes in low-
temperature PEMFCs (LT-PEMFCs) [39–42], but some also address HT-PEMFC membrane
degradation [43]. It is well proven that catalyst dissolution, diffusion, and redistribution in the membrane
play an important role as a precursor of membrane degradation processes [15,38,44]. Therefore, proper
modeling of catalyst degradation plays a crucial role in determining the degradation stimuli used in the
membrane degradation model and in fully developed system-level degradation model, the degradation
of the membrane should be included. However, the degradation stimuli for the catalyst degradation
model are mostly independent of the direct back-influences of membrane degradation, and can therefore,
for the testing purposes of catalyst degradation model, be considered on the level of FC operation model
via changing the macroscopic membrane parameters, as will be explained in Section 4.2 [42].
Despite a wide variety of approaches to the degradation modelling, the existing models, covering the
entire chain of degradation mechanism, are too complex to be applied as the system-level tools for FC
simulations. To the best of the authors’ knowledge, the models of individual degradation mechanism,
have not yet: a) been coupled to a system-level model of HT-PEMFC operation providing spatially and
temporally resolved traces of degradation stimuli; and b) achieved sufficient computational efficiency,
i.e., real-time capability, which enables direct modeling of HT-PEMFCs over longer time scales. To
achieve significant progress in the addressed area, this paper presents an innovative real-time capable
system-level modeling framework of the HT-PEMFC stack comprising a) a mechanistic spatially and
temporally resolved model of HT-PEMFC operation serving as the provider of degradation stimuli and
b) a degradation modeling framework based on interacting individual degradation mechanisms of Pt in
the cathode catalyst layer.
The paper is structured as follows. In Section 2, the key features of proposed modeling framework are
outlined, explaining the requirements the model need to fulfill to serve as a system-level tool, useful in
process of FC design. In Section 3, the framework components are discussed in more detail. In case of
already developed FC operation model, the basic ideas are outlined in Section 3.1 and the references are
provided where additional explanation of model features can be found. The novel physically based and
computationally efficient degradation model is presented in detail in Section 3.2 with the explanation of
all relevant degradation mechanisms and the equations used to model their effects. The coupling
between both components is explained in Section 3.3. The validation of the modeling framework is
presented in Section 4, with validation of operation model being presented in Section 4.1. Section 4.2
provides the analysis of experimental data, used in degradation model validation, with the simulation
results being presented in Section 4.3 and additional capabilities of the model, not used in validation,
being presented in Section 4.4. Concluding remarks are given is Section 5.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
5
2. Model requirements and key features of the modeling framework
The main objective of the model is to ensure high level of predictiveness of the entire causal chain from
1) the input in the form of FC operation and control parameters over to the 2) FC operational conditions
determining the rates of individual degradation mechanisms and 3) of the Pt degradation at the cathode
through modeling the Pt size redistribution in a transient operating regime. This requires innovatively
combining different nanoscale phenomena driven by transient degradation stimuli being determined by
the FC design and operating conditions.
The requirements of the FC operation model can thus be summarized as mechanistically based, spatially
resolved, transient, and computationally fast. On the other hand, the requirements of the degradation
modeling framework can be summarized as based on (the most influential) individual degradation
mechanisms, capable of considering (the most influential) interactions between different individual
degradation mechanisms, transient, and computationally fast.
The core principle of the physically based spatially and temporally resolved model of HT-PEMFC
operation is a novel computationally efficient approach that combines a one-dimensional (1D) numerical
and a two-dimensional (2D) analytic solution, denoted as hybrid 3D analytic-numerical (HAN) [45–48].
A real-time capable HAN modeling approach thus, on one hand, allows for achieving high level of
predictiveness in the FC performance modeling crucial for an adequate virtual integration of FC in the
plant model, and on the other hand, provides spatially and temporally resolved data of degradation
stimuli, which are prerequisite input parameters for the degradation modeling framework.
This ensures a plausible response of the degradation model to varying design and control parameters or
functionalities and thus allow it to reflect the degradation dynamics in FCs consistently. To attain high
computation speeds, the presented model relies on a discreet particle size distribution, which is
consistently applied to all relevant individual degradation mechanisms. This treatment of particle size
distribution is crucial for the correct treatment of two mechanisms:
- platinum and carbon oxidation reaction models that feature the Kelvin term addressing the
surface tension, which depends on the particle size, and
- calculation of detachment rate of Pt particles considering both the electrochemical kinetics of
carbon corrosion and the particle size.
Thereby, the proposed modeling framework enables full modeling of the coupled catalyst degradation
phenomena during the FC’s entire operating lifespan by continuously simulating the varying particle
size distribution through a consistent application of the conservation laws. Note that since the listed
degradation processes are not unique to pure Pt catalyst, very similar framework can be used also for
binary catalyst, provided that appropriately modified degradation rates are used. The degradation
mechanisms, specific for binary catalysts, such as component-dependent dissolution and leeching from
particles ([33]), however, obviously cannot be explained by the proposed model. The membrane
degradation is not modeled directly in the proposed framework; however, its influence on the catalyst
degradation stimuli is considered as a slow temporal change of membrane thickness and ionic
conductivity in the FC operation model.
It is important to note that the proposed innovative modeling framework is intended for a system-level
simulation over longer time scales. To comply with the listed model requirements, care was devoted to
meticulously balancing between the level of detail and the computational expense to model and replicate
the most significant performance and degradation phenomena and their interactions adequately. The
significant contribution of the study, therefore, originates from the fact that it presents a system-level
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
6
mechanistic modeling framework for performance and degradation modeling bringing the FC system-
level modeling to a new level. The proposed innovative modeling framework enables—compared to the
current state of the art—an exploration of the design space not only in terms of the FC performance, but
also its longevity. Additionally, it offers a higher fidelity model-supported design of FC control
functionalities tailored to both performance and longevity objectives.
3. Modeling framework
To preserve the focus of the study, the modeling framework of the HT-PEMFC stack, being the
innovative contribution of this study, is presented in this section. A description of the models of BoP
components is omitted, as they are already widely reported in the literature [49–52].
3.1. Model of FC operation
The physically based spatially and temporally resolved model of HT-PEMFC operation (HAN) is based
on a modeling approach of the nonlinear electrochemical phenomena fully coupled to the species
transport framework that was presented in [45,46]. A computationally optimized HAN modeling
approach that complies with the real-time constraints denoted as HAN-RT (real-time) was presented in
[47] and the extensions of the HAN-RT modeling approach to high temperature FC was presented in
[48].
For completeness and brevity of this paper, the main principles of the model of FC operation are outlined
subsequently (please refer to [45–48] for more information). The detailed description of hydrogen
oxidation and oxygen reduction electrochemical reaction kinetics, membrane water transport, and proton
conductivity are omitted, as they follow the conventional expressions found in the literature (the exact
formulations used can be found in [45–48]). However, the electrochemical reaction kinetics relevant to
the degradation processes are given in detail in Section 3.2. Thus, the focus of this HAN model principles
summary is on the innovative modeling of species transport to and from the catalyst surface.
In the presented approach, the modeled FC is of a straight parallel channel geometry, which is
numerically discretized in the direction along the channels shown in Fig. 1 a) and b). This 1D numerical
discretization yields shallow 2D slices (Fig. 1 c)), which are subjected to the analytical solution. A 2D
slice consists of three parts: a cathode feed part, a membrane electrode assembly (MEA) part, and an
anode feed part, where each feed part is divided into one channel and two gas diffusion layer (GDL)
domains (under the channel and under the rib) yielding seven domains (Fig. 1 d)).
The complete solution is obtained by coupling the individual 2D analytic slice solutions to one another
via the perpendicular numerically resolved 1D gas flow. This gives the approach its name “hybrid 3D
analytic-numerical.” The HAN model thus yields a full 3D result on species concentration distribution.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
7
The basis of the HAN modeling approach is solving a steady-state species transport equation [45–47]:
𝜕𝑐
𝜕𝑡= −𝛁 ⋅ 𝑱 = 𝐷𝛁2𝑐 − 𝛁 ⋅ (𝒖 𝑐) = 0, (1)
with 𝑱 representing the 3D species fluxes, 𝛁 the 3D nabla operator, 𝐷 the diffusion coefficient matrix, 𝑐
the species concentrations vector, and 𝒖 the 3D gas velocity. This hybrid 3D approach introduces a
distinction that splits the general 3D notation into a 2D+1D notation reflecting a different treatment of
physical phenomena in the dimensions perpendicular to the channel gas flow (𝑥 − 𝑦 plane) and in the
dimension along the channel gas flow (𝑧 axis) yielding
𝜕𝑐
𝜕𝑡= ∇2(𝐷 𝑐 − �� 𝑈)⏟
2D
−1
𝑙((𝑣 𝑐)|𝑧=𝑧𝑒𝑥𝑖𝑡 − (𝑣 𝑐)|𝑧=𝑧𝑒𝑛𝑡𝑒𝑟)⏟
1𝐷
= 0, (2)
with 𝑈 being the scalar velocity potential as defined in [25,26], 𝑙 = 𝑧𝑒𝑥𝑖𝑡 − 𝑧𝑒𝑛𝑡𝑒𝑟, which is the depth
of the slice with 𝑧𝑒𝑛𝑡𝑒𝑟 and 𝑧𝑒𝑥𝑖𝑡 denoting the domain entry and exit coordinates of species flow,
respectively. Eq. (2) can be interpreted as a simple diffusion equation on a 2D plane with a source term
(1
𝑙(𝑣 𝑐)|𝑧=𝑧𝑒𝑛𝑡𝑒𝑟) and a sink term (−
1
𝑙(𝑣 𝑐)|𝑧=𝑧𝑒𝑥𝑖𝑡).
The analytic part of the HAN modeling approach is obtained by the following: a) first, devising a general
analytic solution of the 2D diffusion problem in each of the seven computational domain types (Fig. 1
d)) by finding the domain-specific particular integral and eigenfunctions (also called harmonics) of the
Laplace (∇2) operator; and b) second, expressing this specific solution in the form of a linear expansion
comprising the particular integral and a Fourier series of eigenfunctions in such a way that
- the continuity of the solution and its derivative is ensured at the interfaces between
computational domains, and
- the boundary conditions are met.
The smallest number of terms considered in such an expansion, which still leads to satisfactory results,
is used in the real-time capable HAN model (HAN-RT):
Fig. 1: Schematic fuel cell geometry broken down into elementary units used in the HAN modeling approach. Blue regions represent the membrane and spotted translucent the GDLs; the green surfaces represent the rib symmetry plane and the yellow surfaces the symmetry plane between two ribs. a) A parallel channel co- or counter-flow fuel cell geometry. b) Representative unit with a highlighted slice (red). c) A sliced-out section of the representative unit. d) Computational domains of a slice (from top down: cathode channel domain, two cathode GDL domains, MEA domain, two anode GDL domains, and anode channel domain).
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
8
𝑐(𝑥, 𝑦) = 𝐴0𝑥2
ℎ2+ 𝐵0,0 + 𝐵0,1 𝑐𝑜𝑠 (
𝜋
ℎ𝑥) . (3)
The physical significance of the three terms in (3) is explained in [47].
With this number of harmonics taken into consideration, the solution distinguishes two distinct species
concentration values pertaining to the two regions of catalyst per feed part per slice: the one under the
channel and the one under the rib (Fig. 1 d)).
In addition to distinct concentration values (reactant and product concentration), the solution of the
diffusion problem coupled to the electrochemical kinetics yields also distinct values of temperature,
electrode potential, reaction overvoltage, and electric current density. This offers not only spatially
resolved performance results, but also spatially resolved physical conditions that represent the
degradation stimuli.
3.2. Degradation modeling framework
The degradation modeling framework comprises interacting models of six individual degradation
mechanisms relevant for Pt degradation in the cathode catalyst layer:
1. carbon support surface oxidation,
2. carbon corrosion,
3. platinum particle detachment and agglomeration,
4. platinum particle surface oxidation,
5. platinum dissolution and redeposition/Ostwald ripening, and
6. platinum diffusion into membrane.
These are presented in the schematic of interacting individual degradation mechanisms in Fig. 2 a). To
provide a more intuitive insight, Fig. 2 b) shows a schematic of the intertwined individual degradation
mechanisms.
All individual degradation mechanisms are modeled as 0D reactors, which obtain spatially and
temporally resolved data of degradation stimuli as input parameters from the HAN-RT model of the FC
operation. The interaction between the model of FC operation and the degradation model is presented
in Section 3.3.
Fig. 2: Schematics of a) interacting individual degradation mechanisms and b) intertwined degradation mechanisms. The schematics aim at highlighting the importance of coupling the models of several degradation mechanisms due to their entanglement.
a) b)
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
9
The oxidation dynamics on the surface of platinum particles depend strongly on the particle size due to
the effects of surface tension. The modeling approach is, therefore, based on the particle size
distribution, as explained in Section 3.2.1, where the particle size is considered in all relevant
electrochemical equations (see Sections 3.2.2 and 3.2.3), resulting in different oxidation dynamics for
each size class and a consequent change in the particle population of each class, which is explained in
Sections 3.2.4 and 3.2.5.
The platinum diffusion into the ionomer was modeled by assuming a linear concentration profile
between the catalyst and the position of the platinum band in the membrane [44] considering the Fickian
diffusion [27]. The diffusion of platinum into the membrane causes a net loss of platinum mass in the
catalyst, resulting in the loss of an active area; however, using the parameters taken from [27] and [44],
this contribution turned out to be very small compared to other degradation mechanisms. Nevertheless,
the modeling of this process is an important step in providing a physically plausible coupling to
membrane degradation models, where the platinum diffusion is a crucial initiator of further degradation
processes [44].
Particle size distribution model
The particles are sorted into 𝑀 size classes, indexed by integer 𝑖, with each class being populated with
𝑁𝑖 Pt particles featuring radii between 𝑅𝑖 −1
2Δ𝑅 and 𝑅𝑖 +
1
2Δ𝑅, where 𝑅𝑖 is the mean particle radius in
the 𝑖-th size class. The main goal of the degradation model is to calculate the rate of change of class
populations 𝑁𝑖(𝑡) over time caused by Ostwald ripening and particle agglomeration. This in turn allows
to track the changes of several important catalyst parameters, such as catalyst platinum mass loading
and catalyst surface area. Note, however, that the catalyst surface area, calculated from the particle size
distribution, does not directly relate to the electrochemically active surface area (see [53]), which should
be considered when the degradation model is coupled back to the FC operation model.
This approach is similar to the approach of Rinaldo et al. [54–56], where the particle size distribution is
expressed as a continuous function. However, the use of discreet particle size distribution with relatively
small number of size classes is essential for obtaining the high computation speeds required to simulate
long-term degradation effects on a system level. However, the use of such an approach requires various
adaptations in existing models of particle agglomeration and Ostwald ripening, as will be explained in
Sections 3.2.4 and 3.2.5.
Platinum and carbon oxidation and corrosion model
The precursors of catalyst degradation processes, addressed in our model, are oxidation processes on
the surface of Pt particles and carbon support in the catalyst layer, which are initiated by a combination
of high electric potential, temperature, and the presence of water vapor. On the one hand, oxidation leads
to carbon corrosion and consequent particle detachment and agglomeration; on the other hand, the part
of the surface of platinum particles, which is free of oxides, is prone to dissolution into the ionomer [16].
The carbon and Pt oxidation model is based on the oxidation mechanisms proposed by Pandy [14].
However, to improve the performance of the proposed model and enable its use in long-term degradation
simulations, the oxidation model was reduced to a minimal number of reactions required to describe the
transient phenomena such as the 𝐶𝑂2 emission peaks after voltage change [57], which will be discussed
in Section 4.4.2. The mechanism was originally developed for LT-PEMFCs; however, because the
carbon support and catalyst materials used in HT-PEMFCs are similar [58], a similar mechanism
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
10
comprising the most significant reaction pathways was applied in the proposed model. However, the
difference in operating temperature of LT- and HT-PEMFCs results in different values of equilibrium
potentials and reaction constants of the proposed electrochemical reactions, as will be further discussed
in this section.
The core of the reduced mechanism consists of a set of six electrochemical reactions (4), describing the
oxidation processes of the carbon support surface and platinum particles in the catalyst layer. The
equilibrium potential 𝐸 of each equation is given in reference to the reversible hydrogen electrode
(RHE).
𝐶1: 𝐶∗ +𝐻2𝑂 ⇌ 𝐶 − 𝑂𝐻 + 𝐻+ + 𝑒−; 𝐸𝐶1,0 = 0.29 V RHE
𝐶2: 𝐶 − 𝑂𝐻 ⇌ 𝐶 = 𝑂 + 𝐻+ + 𝑒−; 𝐸𝐶2,0 = 0.80 V RHE
𝐶3: 𝐶 − 𝑂𝐻 + 𝐻2𝑂 → 𝐶∗ + 𝐶𝑂2 + 3𝐻+ + 3𝑒−; 𝐸𝐶2,0 = 0.96 V RHE
𝑃𝑡1: 𝑃𝑡∗ +𝐻2𝑂 ⇌ 𝑃𝑡 − 𝑂𝐻 +𝐻+ + 𝑒−, 𝐸𝑃𝑡1,0 = 0.79 V RHE
𝑃𝑡2: 𝑃𝑡 − 𝑂𝐻 ⇌ 𝑃𝑡 = 𝑂 + 𝐻+ + 𝑒−; 𝐸𝑃𝑡2,0 = 0.80 V RHE
𝑃𝑡𝐶: 𝑃𝑡 − 𝑂𝐻 + 𝐶 − 𝑂𝐻 → 𝐶∗ + 𝑃𝑡∗ + 𝐶𝑂2 + 2𝐻+ + 2𝑒−; 𝐸𝑃𝑡𝐶,0 = 0.62 V RHE
(4)
Oxidation of both the carbon support surface (reactions 𝐶1 and 𝐶2 in (4)) and the platinum particles
(reactions 𝑃𝑡1 and 𝑃𝑡2 in (4)) occurs in two stages. Hydroxide groups (𝐶 − 𝑂𝐻 and 𝑃𝑡 − 𝑂𝐻) are
formed on suitable defect sites on the carbon and Pt surface (𝐶∗ and 𝑃𝑡∗) in the presence of water
(reactions 𝐶1 and 𝑃𝑡1) at sufficiently high electric potential and are further oxidized into stable oxide
groups (𝐶 = 𝑂 and 𝑃𝑡 = 𝑂) at even higher potentials (reactions 𝐶2 and 𝑃𝑡2).
The surface groups induce the formation of carbon dioxide and consequential carbon corrosion via two
reactions: 𝐶3 and 𝑃𝑡𝐶 in (4). The carbon hydroxide groups 𝐶 − 𝑂𝐻 , on one hand, result in 𝐶𝑂2
formation in the presence of water at a high potential of 0.96 V (reaction 𝐶3), leaving behind new defect
site 𝐶∗ , and on the other hand, they also interact with 𝑃𝑡 − 𝑂𝐻 groups on the platinum surface at
potentials above 0.62 V (reaction 𝑃𝑡𝐶), again resulting in 𝐶𝑂2 formation and leaving behind both free
platinum and carbon defect sites 𝐶∗ and 𝑃𝑡∗ [14].
The temperature dependence of equilibrium potentials in reactions (4) is based on
𝑛 𝑒 𝐸0 = Δ𝐺 = Δ𝐻 − 𝑇 Δ𝑆, (5)
relating the equilibrium potential to the difference in Gibbs potential between reactants and products
[59]. To calculate the equilibrium potentials at HT-PEMFC conditions, as indicated in (4), from the
values given in [14] for LT-PEMFC, the entropy changes Δ𝑆 of the given reactions need to be evaluated.
Because the concentrations of reactants at equilibrium potential are equal to their reference values, only
the entropy changes due to the phase transitions of reactants were considered. Due to high temperature,
both water and 𝐶𝑂2 in reactions (4) can be assumed to be in gaseous state, and therefore, the entropy
change of condensation of both substances [60,61] was used to approximate the shifts of equilibrium
potentials in reactions 𝐶1, 𝐶3, 𝑃𝑡1, and 𝑃𝑡𝐶 in (4), which were calculated to be 0.09, 0.01, 0.09, and
−0.03 V, respectively, compared to the values given in [14].
In addition to the equilibrium potential change, an important difference between LT- and HT-PEMFCs
also lies in the origin of water and protons participating in oxidation reactions. In LT-PEMFCs, the
concentration of both water and protons depends strongly on the membrane humidity. By contrast, in
HT-PEMFCs, water is present in the gaseous phase; thus, its concentration is more directly related to
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
11
feed gas properties, while the protons participating in the reactions originate from phosphoric acid,
which is used as the membrane ionic conductor. Hence, their concentration is directly related to pH
values of the acid. In the proposed degradation model, a constant value of pH = 1.25 (related to 85%
phosphoric acid used for doping of polybenzimidazole membrane) is assumed [62,63]. Time-dependent
proton concentration could be obtained from a dedicated membrane degradation model describing the
acid leakage out of the membrane, but this effect was assumed to be negligible in the proposed model.
The state of surface oxidation of platinum and carbon is defined by the fraction of the surface covered
by each oxide species, i.e., Θ𝑃𝑡𝑂𝐻 , Θ𝑃𝑡𝑂 , Θ𝐶𝑂𝐻 , and Θ𝐶𝑂 , and the fraction of the surface free of
oxides, Θ𝑃𝑡 and Θ𝐶. It is also assumed that not more than one monolayer of oxides can cover the carbon
or platinum surface, Θ𝑋 < 1, (𝑋 = 𝑃𝑡, 𝑃𝑡𝑂𝐻, 𝑃𝑡𝑂, 𝐶, 𝐶𝑂𝐻, 𝐶𝑂), and that the sum of coverage fractions
on either carbon or platinum is constant over time:
Θ𝑃𝑡 + Θ𝑃𝑡𝑂𝐻 + Θ𝑃𝑡𝑂 = 1,
Θ𝐶 + Θ𝐶𝑂𝐻 + Θ𝐶𝑂 = 1. (6)
To obtain realistic results, it is important to consider how the particle size affects the equilibrium electric
potential of reactions 𝑃𝑡1, 𝑃𝑡2, and 𝑃𝑡𝐶 due to the surface tension. This is addressed by the introduction
of the Kelvin term, shifting the equilibrium potential from that of a flat Pt surface (denoted 𝐸0) by a
value that is inversely proportional to the particle radius 𝑅𝑖 [16,27]:
𝐸𝑃𝑡1(𝑅𝑖) = 𝐸𝑃𝑡1,0 +1
2𝐹𝑅𝑖(𝜎𝑃𝑡𝑂𝐻𝑀𝑃𝑡𝑂𝐻
𝜌𝑃𝑡𝑂𝐻 −𝜎𝑃𝑡𝑀𝑃𝑡
𝜌𝑃𝑡 ),
𝐸𝑃𝑡2(𝑅𝑖) = 𝐸𝑃𝑡2,0 +1
2𝐹𝑅𝑖(𝜎𝑃𝑡𝑂𝑀𝑃𝑡𝑂
𝜌𝑃𝑡𝑂 −𝜎𝑃𝑡𝑂𝐻𝑀𝑃𝑡𝑂𝐻
𝜌𝑃𝑡𝑂𝐻 ),
𝐸𝑃𝑡𝐶(𝑅𝑖) = 𝐸𝑃𝑡𝐶,0 −1
2𝐹𝑅𝑖
𝜎𝑃𝑡𝑂𝐻𝑀𝑃𝑡𝑂𝐻
𝜌𝑃𝑡𝑂𝐻 .
(7)
The magnitude of the Kelvin shift depends on the surface tensions 𝜎, molar mass 𝑀, and density 𝜌 of
the surface groups involved in the reaction (𝑃𝑡, 𝑃𝑡𝑂𝐻, and 𝑃𝑡𝑂).
The inclusion of the Kelvin term into the rates of reactions (4) is an important improvement over the
oxidation dynamics found in [14]. The rates of reactions (4) are [14,16,27]
a) 𝑟𝐶1,𝑖 = 𝛾𝐻𝑇𝑘𝐶1(𝑎𝐻2𝑂Θ𝐶,𝑖𝑒𝛼𝐶1(𝑈𝑐𝑎𝑡−𝐸𝐶1)/𝑏 − 𝑎𝐻+Θ𝐶𝑂𝐻,𝑖𝑒
−(1−𝛼𝐶1)(𝑈𝑐𝑎𝑡−𝐸𝐶1)/𝑏),
b) 𝑟𝐶2,𝑖 = 𝛾𝐻𝑇𝑘𝐶2(Θ𝐶𝑂𝐻,𝑖𝑒𝛼𝐶2(𝑈𝑐𝑎𝑡−𝐸𝐶2)/𝑏 − 𝑎𝐻+Θ𝐶𝑂,𝑖𝑒
−(1−𝛼𝐶2)(𝑈𝑐𝑎𝑡−𝐸𝐶2)/𝑏),
c) 𝑟𝐶3,𝑖 = 𝛾𝐻𝑇𝑘𝐶3𝑎𝐻2𝑂Θ𝐶𝑂𝐻,𝑖𝑒3𝛼𝐶3(𝑈𝑐𝑎𝑡−𝐸𝐶3)/𝑏,
d) 𝑟𝑃𝑡1,𝑖 = 𝛾𝐻𝑇𝑘𝑃𝑡1(𝑎𝐻2𝑂Θ𝑃𝑡,𝑖𝑒𝛼𝑃𝑡1(𝑈𝑐𝑎𝑡−𝐸𝑃𝑡1(𝑅𝑖)−𝑟𝑂𝑥ΘOx)/𝑏 −
𝑘𝑟𝑒𝑣𝑎𝐻+Θ𝑃𝑡𝑂𝐻,𝑖𝑒−(1−𝛼𝑃𝑡1)(𝑈𝑐𝑎𝑡−𝐸𝑃𝑡1(𝑅𝑖))/𝑏),
e) 𝑟𝑃𝑡2,𝑖 = 𝛾𝐻𝑇𝑘𝑃𝑡2(Θ𝑃𝑡𝑂𝐻,𝑖𝑒𝛼𝑃𝑡2(𝑈𝑐𝑎𝑡−𝐸𝑃𝑡2(𝑅𝑖)−𝑟𝑂𝑥ΘOx)/𝑏 − 𝑘𝑟𝑒𝑣𝑎𝐻+Θ𝑃𝑡𝑂,𝑖𝑒
−(1−𝛼𝑃𝑡2)(𝑈𝑐𝑎𝑡−𝐸𝑃𝑡2(𝑅𝑖))/𝑏),
f) 𝑟𝑃𝑡𝐶,𝑖 = 𝛾𝐻𝑇𝑘𝑃𝑡𝐶Θ𝐶𝑂𝐻,𝑖Θ𝑃𝑡𝑂𝐻,𝑖𝑒2𝛼𝑃𝑡𝐶(𝑈𝑐𝑎𝑡−𝐸𝑃𝑡𝐶(𝑅𝑖))/𝑏,
(8)
where 𝑟𝑗,𝑖 denotes the rate of the 𝑗-th reaction in (4) (𝑗 = 𝐶1, 𝐶2, 𝐶3, 𝑃𝑡1, 𝑃𝑡2, 𝑃𝑡𝐶) for particles in the
𝑖-th size class with mean radius 𝑅𝑖.
General scaling of the reaction rates is given by the reaction rate constants 𝑘𝑗, given for LT-PEMFC in
[14,15]. The reaction rates are additionally affected by temperature via the change in Gibbs free energy
of the transition states of each electrochemical reaction. As a detailed analysis of transition states
required to completely describe these effects requires insights on the atomistic scale [59], which is not
considered in the scope of this study, the reaction rate constants for the case of HT-PEMFCs were
modeled by multiplying all LT-PEMFC reaction rates, reported in [14], by the same temperature
dependent factor 𝛾𝐻𝑇 in (8), determined by the calibration of the model using experimental data.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
12
Because the reaction rates change exponentially with the enthalpy of transition states Δ𝐻++, the change
in temperature from 80℃, typical for LT-PEMFCs, to 160℃ for HT-PEMFCs, can result in a relatively
large reaction rate acceleration 𝛾𝐻𝑇 of several thousands, assuming that the enthalpies of transition state
are of the order of 1 eV [59]. Another effect of temperature 𝑇 on the reaction rates (8) is captured in the
Tafel slope, 𝑏 = 𝑅𝑇/𝐹, with the gas constant 𝑅 and the Faraday constant 𝐹.
The balance between forward and backward rates of reactions is affected by two factors: first, the rate
and direction are determined by the concentration of reactants and products, namely the surface
coverages of different oxide species (Θ𝑋), water activity (𝑎𝐻2𝑂 = 𝑐𝐻2𝑂/𝑐𝐻2𝑂,𝑟𝑒𝑓), and proton activity
(𝑎𝐻+ = 𝑐𝐻+/𝑐𝐻+,𝑟𝑒𝑓). Second, the rate changes exponentially with the difference between the local
electric potential in the catalyst layer 𝑈𝑐𝑎𝑡, which is defined as the potential difference between catalyst
particles and the adjacent ionomer, and the equilibrium potential of each reaction (𝐸𝑗), which is affected
by the particle size via the Kelvin term (7) for the reactions featuring platinum surface oxides (𝐸𝑗(𝑅𝑖)
in (8) d)–f)). Note that the potential 𝑈𝑐𝑎𝑡 is in general different from the FC potential 𝑈𝐹𝐶 (see Section
4.2). The reaction rates are further shifted due to the presence of oxide layers [64,65], expressed as an
additional exponential term 𝑟𝑂𝑥ΘOx = 𝑟𝑂𝑥(ΘPtOH + ΘPtO) (Eq. (8) d)–e)). The slower rates of
reduction, in comparison to oxidation, are addressed by an additional parameter 𝑘𝑟𝑒𝑣 in (8) d)–e)
introduced in [16,27].
The reaction rates (8) determine the temporal dynamics of surface groups’ production and
consequentially their surface coverage Θ𝑋,𝑖 for each particle size class 𝑅𝑖:
a) 𝑑ΘPt,i
𝑑𝑡=
𝐹
ΓPt(−𝑟𝑃𝑡1,𝑖 +
𝑑𝑃𝑡𝐶
𝑅𝑖𝑟𝑃𝑡𝐶,𝑖),
b) 𝑑ΘPtOH,i
𝑑𝑡=
𝐹
ΓPt(𝑟𝑃𝑡1,𝑖 − 𝑟𝑃𝑡2,𝑖 −
𝑑𝑃𝑡𝐶
𝑅𝑖𝑟𝑃𝑡𝐶,𝑖),
c) 𝑑ΘPtO,i
𝑑𝑡=
𝐹
ΓPt𝑟𝑃𝑡2,𝑖,
d) 𝑑ΘC,i
𝑑𝑡=
𝐹
ΓC(−𝑟𝐶1,𝑖 + 𝑟𝐶3,𝑖 + 𝑟𝑃𝑡𝐶,𝑖),
e) 𝑑ΘCOH,i
𝑑𝑡=
𝐹
ΓC(𝑟𝐶1,𝑖 − 𝑟𝐶2,𝑖 − 𝑟𝐶3,𝑖 − 𝑟𝑃𝑡𝐶,𝑖),
f) 𝑑ΘCO,i
𝑑𝑡=
𝐹
ΓC𝑟𝐶2,𝑖.
(9)
The set of differential equations (9) by definition conserves the sum of surface coverages on the platinum
and carbon support (6). Because the reaction rates are defined per unit area ([𝑚𝑜𝑙 ∙ 𝑚−2 ∙ 𝑠−1]), the rate
of change of oxide coverages is calculated via the surface density of sites available for oxidation (ΓPt, ΓC)
[15,66,67]. Note that the reaction rates, given by Pandy [14] already consider the factor 𝐹/Γ and are,
therefore, given as dimensionless numbers. Additionally, it is important to consider that the reaction
between Pt and carbon OH groups (reaction 𝑃𝑡𝐶 in Eq. (5)) only takes place on a narrow band of width
𝑑𝑃𝑡𝐶 at the junction between the Pt and carbon surface (see Fig. 3) and the oxide coverage of carbon
therefore only relates to this region.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
13
Electrochemical processes, described in this section, affect the particle size distribution 𝑁𝑖 in two ways:
First, the part of platinum surface not covered by oxides, is prone to dissolution and redeposition of
𝑃𝑡2+ions, as explained in Section 3.2.3, causing some particles to grow and others to shrink, leading to
changes in particle size distribution, as explained in Section 3.2.4. Second, the carbon corrosion,
addressed by reactions 𝐶3 and 𝑃𝑡𝐶, results in the detachment of Pt particles from the carbon support
and their agglomeration into larger particles, as discussed in Section 3.2.5. Thus, a relatively small set
of reactions, describing basic oxidation phenomena on carbon and Pt surface, provides a physically
grounded and computationally efficient method to distinguish between underlying causes of different Pt
particles growth mechanism.
Platinum dissolution and redeposition model
The dissolution of platinum is described by the electrochemical reaction [16,27]
𝑃𝑡 ⇌ 𝑃𝑡2+ + 2𝑒−; 𝐸𝑑𝑖𝑠𝑠,0 = 1.155 V RHE. (10)
Similar to the case of carbon and Pt oxidation, the equilibrium potential of dissolution, obtained from
[16,27], is shifted by the temperature due to the entropy of hydration of 𝑃𝑡2+ ions [68], causing a
decrease of −0.033 V (Eq. (5)) compared to that of LT-PEMFC, resulting in a value 𝐸𝑑𝑖𝑠𝑠,0 = 1.155 V.
In the proposed model of platinum dissolution, it is assumed that the dissolution and redeposition of
𝑃𝑡2+ ions can only take place on the parts of the platinum surface, which are not covered by oxides
[16]. Similar to the rates of oxidation (8), the dynamics of platinum dissolution are given by the reaction
rate, consisting of forward and backward terms:
𝑎) 𝑟𝑑𝑖𝑠𝑠,𝑖 = 𝛾𝐻𝑇𝑘𝑑𝑖𝑠𝑠Θ𝑃𝑡 (𝑒𝛼𝑑𝑖𝑠𝑠(𝑈𝑐𝑎𝑡−𝐸𝑑𝑖𝑠𝑠(𝑅𝑖))
𝑏𝑑𝑖𝑠𝑠 −𝑐𝑃𝑡2+
𝑐𝑃𝑡2+,𝑟𝑒𝑓𝑒−(1−𝛼𝑑𝑖𝑠𝑠)(𝑈𝑐𝑎𝑡−𝐸𝑑𝑖𝑠𝑠(𝑅𝑖))
𝑏𝑑𝑖𝑠𝑠 ),
𝑏) 𝐸𝑑𝑖𝑠𝑠(𝑅𝑖) = 𝐸𝑑𝑖𝑠𝑠,0 −1
2𝐹
𝜎𝑃𝑡𝑀𝑃𝑡𝜌𝑃𝑡𝑅𝑖
.
(11)
The dissolution rate is proportional to the reaction rate constant 𝑘𝑑𝑖𝑠𝑠 , multiplied by the oxide-free
surface fraction ΘPt. The increase in reaction rate due to temperature is described by the same factor
𝛾𝐻𝑇 as for oxidation reactions (8). The balance between the forward and backward reactions is again
determined by the difference between the local catalyst electric potential 𝑈𝑐𝑎𝑡 and the equilibrium
Fig. 3: Schematic of microscopic geometric parameters of degradation model. Carbon corrosion, relevant for particle detachment, only takes place in a narrow region of width 𝑑𝑃𝑡𝐶 on the junction of platinum and carbon; therefore, the oxide coverage of carbon is tracked only in this region.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
14
dissolution potential 𝐸𝑑𝑖𝑠𝑠(𝑅𝑖), affected by the Kelvin term (11) b), as well as the concentration of Pt
ions dissolved in the ionomer 𝑐𝑃𝑡2+. It is assumed in the proposed model that platinum particles of
various sizes are randomly spread throughout the catalyst and, therefore, all particles, regardless of size,
interact with a single reservoir of 𝑃𝑡2+ ions with a uniform concentration. Therefore, the time dynamics
of 𝑃𝑡2+ ion concentration 𝑐𝑃𝑡2+ is expressed as a sum of contributions of particles from all size classes:
𝑑𝑐𝑃𝑡2+
𝑑𝑡=
1
𝑉𝑖𝑜𝑛∑ 4𝜋𝑅𝑖
2𝑀𝑖=1 𝑁𝑖𝑟𝑑𝑖𝑠𝑠,𝑖. (12)
The model also considers the total volume of the ionomer in the catalyst layer, 𝑉𝑖𝑜𝑛 = 𝑆𝐹𝐶𝑑𝑐𝜀𝑖𝑜𝑛 ,
calculated as a product of the FC surface 𝑆𝐹𝐶, the catalyst layer thickness 𝑑𝑐, and the volume fraction
of ionomer 𝜀𝑖𝑜𝑛.
The coupling of particles of all sizes to a shared reservoir of 𝑃𝑡2+ ions is crucial for modeling the so-
called Ostwald ripening [16]. The equilibrium concentration 𝑐𝑃𝑡2+, at which the dissolution rate (11) a)
is zero, depends on the particle size due to the Kelvin term: it is larger for small particles and smaller
for large particles. The global stationary concentration 𝑐𝑃𝑡2+ can, therefore, only be achieved when
small particles are dissolving and the platinum is redepositing on large particles [26]. This is an
important factor in the degradation of the FC catalyst layer, which leads to an increase in mean particle
size as will be explained in Section 3.2.4.
Particle size redistribution: Ostwald ripening
The dissolution or redeposition of platinum on a particle of typical size 𝑅𝑖 in the 𝑖-th size distribution
class leads to its growth or shrinkage depending on dissolution rate (11):
𝑑𝑅𝑖
𝑑𝑡= −
𝑀𝑃𝑡
𝜌𝑃𝑡𝑟𝑑𝑖𝑠𝑠,𝑖. (13)
The growth or shrinkage of particles in a particular size class leads to a decrease in its population unless
it is outweighed by new entries to the class coming in as shrinking particles from a larger size class, or
growing particles from a smaller size class. The number of particles transferred between classes depends
also on the width Δ𝑅𝑖 of each size class and the number of particles 𝑁𝑖 it contains. The governing
equations written below are given for an equidistant distribution with a uniform class width Δ𝑅:
𝑑𝑁𝑑𝑖𝑠𝑠,1
𝑑𝑡=
1
Δ𝑅[−𝑁1|��1| + 𝑁2|��2|𝐻(−��2)],
𝑑𝑁𝑑𝑖𝑠𝑠,𝑖
𝑑𝑡=
1
Δ𝑅[𝑁𝑖−1|��𝑖−1|𝐻(��𝑖−1) − 𝑁𝑖|��𝑖| + 𝑁𝑖+1|��𝑖+1|𝐻(−��𝑖+1)], 2 ≤ 𝑖 ≤ 𝑀 − 1,
𝑑𝑁𝑑𝑖𝑠𝑠,𝑀
𝑑𝑡=
1
Δ𝑅[𝑁𝑀−1��𝑀−1𝐻(��𝑀−1) + 𝑁𝑀��𝑀𝐻(−��𝑀)],
(14)
where the Heaviside theta function 𝐻(. ) is used to properly distinguish the contributions of neighboring
size classes in case of growth or shrinkage of particles populating them. Note that (14) do not conserve
the total number of particles because the shrinking particles from size class 𝑖 = 1 are completely
dissolved and vanished from the distribution. The growing particles in largest size class 𝑖 = 𝑀, are on
the other hand considered as staying in the same size class.
In the limit of infinite number of classes (𝑀 → ∞,Δ𝑅 → 0), equations (14) transform into the expression
for continuous particle size distribution used in [54–56]:
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
15
𝑑𝑓𝑁(𝑟, 𝑡)
𝑑𝑡=𝜕
𝜕𝑟[𝑓𝑁(𝑟, 𝑡)
𝑑𝑟
𝑑𝑡]. (15)
For long-term degradation modeling, however, when a small number of particle size classes are desired,
as is the case in computationally efficient system-level models, aiming at short simulation times, the
expression (14) is more physically plausible than one that would be obtained from (15) with the use of
the finite difference method.
As explained in Section 3.2.3, dissolution and redeposition lead to shrinkage of smaller particles and
growth of larger ones. Applied to (14), this leads to a decrease in population of classes with small
particles, and an increase in population of classes with large particles. Even in the case of negligible
platinum diffusion into the membrane and consequent mass conservation, this leads to a reduction in
catalyst surface area. The effect of this process will be demonstrated in Section 4.4.1, where the results
of degradation modeling are presented.
Particle size redistribution: particle detachment and agglomeration
The platinum and carbon surface oxidation processes in the presence of water at elevated voltage and
temperature lead to carbon corrosion, as explained in Section 3.2.2. Because corrosion is most prominent
at the junction of the Pt particle and its carbon support, it inevitably leads to a decrease in contact surface
between the particle and its carbon support and therefore to the eventual detachment of the particle. In
this section, a model of platinum particle detachment from its carbon support, caused by corrosion, and
the consequent agglomeration is presented.
The detailed mechanisms of the agglomeration processes are not yet completely understood [69];
therefore, some assumptions have been used to model these effects on a system level. First, it is assumed
that at each moment, the number of particles attached to the carbon support is much larger than the
number of detached particles. This means that the detached particles will always merge with particles
attached to the support. This is closely related to the second assumption that a detached particle instantly
merges with another particle. This eliminates the need for tracking the number of detached particles. It
is also assumed that the probability of two particles merging is independent of their size: a detached
particle will merge with an attached particle of a random size.
To calculate the change in the particle size distribution, three variables need to be calculated for each
size class 𝑖: the rate of particle detachment ��𝑑𝑒𝑡,𝑖, the rate at which the detached particles are attached
to the particles in the respective size class ��𝑎𝑡𝑡,𝑖, and the rate of new particles appearing in the respective
class due to merging ��𝑚𝑒𝑟,𝑖. The net dynamics of particle population in a class due to agglomeration
depend on all three contributions:
𝑑𝑁𝑎𝑔𝑔,𝑖
𝑑𝑡= ��𝑑𝑒𝑡,𝑖 + ��𝑎𝑡𝑡,𝑖 + ��𝑚𝑒𝑟,𝑖, (16)
where ��𝑑𝑒𝑡,𝑖 and ��𝑎𝑡𝑡,𝑖 are defined as negative quantities. A similar model was proposed in [56] for the
case of a continuous particle size distribution.
The rate of particle detachment is directly linked to the carbon corrosion rate ((8) c) and f)): a particle
will detach when a sufficient amount of carbon underneath it is corroded. The mass of corroded carbon,
required for detachment, depends on the size of the Pt particle and on the strength of its attachment to
the support. In the proposed model, it is assumed that even for particles of similar size, the strength of
attachment to the carbon support can vary, and thus, even for a small amount of corroded carbon, some
particles will already detach. As another limit, the amount of corroded carbon required to detach all
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
16
particles of size 𝑅𝑖 will be calculated by assuming that the most strongly attached particles have half the
surface submerged into the carbon support. The mass of corroded carbon required to detach such a
particle is
𝑚𝐶,𝑚𝑎𝑥,𝑖 = 2𝜋𝑅𝑖2𝑑𝑃𝑡𝐶𝜌𝐶 , (17)
assuming that the corrosion takes place mostly at the Pt–carbon junction of width 𝑑𝑃𝑡𝐶 (Fig. 3). The
assumption that the number of detached particles increases linearly with the amount of corroded carbon,
yields the result
𝑑𝑁𝑑𝑒𝑡,𝑖𝑑𝑡
= −𝑘𝑑𝑒𝑡��𝐶,𝑐𝑜𝑟𝑟,𝑖𝑚𝐶,𝑚𝑎𝑥,𝑖
𝑁𝑖 = −𝑘𝑑𝑒𝑡𝑀𝐶𝜌𝐶(𝑟𝑃𝑡𝐶,𝑖 + 𝑟𝐶3,𝑖)
𝑁𝑖2𝜋𝑅𝑖
, (18)
which indicates that smaller particles are more likely to detach than larger ones. The model includes one
physically motivated calibration parameter 𝑘𝑑𝑒𝑡, which covers all other influences not accounted for
specifically in the simple model, as for example, the effects of possible surface defects in carbon, which
may decrease the strength of particle attachment and result in higher detachment rates, or exact topology
of interaction between the Pt particle and its support. Eq. (18) is an important improvement over the
existing models of particle detachment [31,32,56] because it considers the effects of particle size on the
detachment rate as well as the detailed electrochemical background of carbon corrosion via reaction
rates 𝑟𝑃𝑡𝐶 and 𝑟𝐶3 (8).
Following the assumption that the detached particles merge with the attached particles independently of
their size, the rate of attachment to particles of size 𝑅𝑖 is proportional to the number of such particles 𝑁𝑖
and the total rate of particle detachment, summed over all classes, ��𝑑𝑒𝑡,𝑡𝑜𝑡 = ∑ ��𝑑𝑒𝑡,𝑖𝑀𝑖=1 :
𝑑𝑁𝑎𝑡𝑡,𝑖𝑑𝑡
=��𝑑𝑒𝑡,𝑡𝑜𝑡∑ 𝑁𝑖𝑀𝑖=1
𝑁𝑖 . (19)
The rate of particle production due to their merging, ��𝑚𝑒𝑟,𝑖, is based on the conservation of mass of
platinum particles during merging. When a detached particle of size 𝑅𝑗 attaches to a particle of size 𝑅𝑘,
the size of the newly formed particle is
𝑅𝑖 = √𝑅𝑗3 + 𝑅𝑘
33. (20)
The rate at which the merged particles appear in size class 𝑅𝑖 is therefore obtained by the summation of
the contributions of all detachment rates in class 𝑅𝑗 and attachment rates in class 𝑅𝑘, such that 𝑅𝑗 and
𝑅𝑘 fulfill condition (20):
𝑑𝑁𝑚𝑒𝑟,𝑖𝑑𝑡
=𝑅𝑖𝑋𝑆𝑛𝑖Δ𝑅
∑��𝑑𝑒𝑡,𝑗��𝑎𝑡𝑡,𝑘
��𝑑𝑒𝑡,𝑡𝑜𝑡
√𝑅𝑗3+𝑅𝑘
33≈𝑅𝑖
. (21)
Note that Eq. (21) is equivalent to the expression for continuous particle size distributions, given in [56],
but much more suitable for numerical calculations with a limited number of particle size classes.
However, to properly account for different numbers of possible pairs, 𝑅𝑗 and 𝑅𝑘, contributing to the new
particles in class 𝑅𝑖, the sum needs to be properly renormalized by comparing the number of terms in
the sum in (21), denoted as 𝑛𝑖, to the area of integration in the continuous case, expressed by 𝑋𝑆 =𝑑
𝑑𝑅∫ √𝑅3 − 𝑥3
3𝑅
0𝑑𝑥 ≈ 1.766. This approach ensures the conservation of Pt mass in agglomeration
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
17
model even at small number of particle size classes desired to ensure sufficient computational efficiency
of system-level degradation model.
Because the newly formed particles in the distribution (Eq. (21)) will always be larger than either of
both particles involved in the merging, the total change in distribution due to agglomeration (Eq. (16))
will lead to a decrease in the number of small particles and an increase in the number of large ones. The
net effect is similar to the Ostwald ripening (14), which will be further discussed in Section 4.4.
Interaction of individual degradation mechanisms
The full time dynamics of the particle size redistribution is calculated by combining the effects of the
Ostwald ripening (14) and the particle agglomeration (16):
𝑑𝑁𝑖𝑑𝑡
= 𝑑𝑁𝑑𝑖𝑠𝑠,𝑖𝑑𝑡
+𝑑𝑁𝑎𝑔𝑔,𝑖
𝑑𝑡. (22)
This set of 𝑀 differential equations for all size classes in the particle size distribution determines the
changes in surface area of the FC catalyst, which is a standard way of tracking the degradation processes
in FCs. However, as both terms in (22) depend on oxidation rates and surface oxide coverages, these
also need to be calculated using Eq. (9). Two equations for each size class can be eliminated based on
the constant sum of surface coverages (6), resulting in 5 ×𝑀 differential equations. One additional
equation is required to track the 𝑃𝑡2+ ion concentration (12), resulting in a set of 5 × 𝑀 + 1 differential
equations. What is important is that the number of size classes 𝑀 can be used to balance between the
level of details in the modeled particle size distribution and the performance efficiency of the model
based on the intended use.
3.3. Interaction of models
As depicted in Fig. 4, the spatial resolution of the degradation model consisting of one dedicated 0D
reactor for each respective computational domain of the FC operation model is defined by the resolution
of the FC operation model. This approach is justified by the fact that the model of FC operation has a
sufficiently coarse discretization to support the assumption that the short-range Pt phenomena of
detachment and agglomeration as well as the Pt redeposition due to dissolution occur well within one
computational domain. Since the degradation study in the anode catalyst layer are very similar, the
proposed model could in principle also be used to model the anode degradation by coupling it to the
degradation stimuli outputs of FC operation model, calculated for the anode catalyst layer. The
experimental data on the anode degradation in HT PEMFC are scarce and the proposed model was
therefore tested only for cathode catalyst layer.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
18
The proposed modeling framework also allows for modeling the back influences of the degradation
phenomena (Fig. 4) on the parameters of the operational model via the reduced catalyst surface area
(Section 3.2.1), which in turn affects the catalyst’s exchange current density. However, to simplify the
analysis of the degradation mechanisms, this sort of back-coupling of the degradation to operation model
is a task deserving its own dedicated study and was therefore omitted.
4. Results and discussion
4.1. Validating the model of FC operation
The model of HT-PEMFC operation was extensively validated on three levels by benchmarking its
simulation results with the following:
1. spatially resolved results generated by a validated [2,70] 3D multiphase CFD tool [47],
2. experimental results of a steady-state FC operation at various boundary conditions [70], and
3. experimental results of a transient FC operation [48].
Therefore, its validation is only briefly addressed in this study as more emphasis is placed on the
validation and the results of modeling the degradation phenomena, which are being reported for the first
time.
The spatial variations in degradation stimuli lead to important differences in degradation rates at
different locations in the FC [71,72]; therefore, it is instructive to demonstrate that the model of HT-
PEMFC operation produces high-reliability results, which serve as the degradation stimuli used in the
degradation modeling.
The performance of the HAN-RT model in simulating the steady-state operation of the HT-PEMFC is
validated by comparing the results of the spatial distribution of important physical quantities obtained
with the HAN-RT simulation to those obtained by the 3D multiphase CFD tool.
Fig. 4: Schematic of the interaction between the performance model (Fig. 1) and the degradation modeling framework (Fig. 2).
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
19
The spatial distributions of four variables were compared as follows:
1) distribution of oxygen concentration at cathode catalyst surface,
2) distribution of water vapor concentration at cathode catalyst surface to validate gas flow,
diffusion, convection, and general species transport modeling,
3) distribution of electric current density, and
4) distribution of catalyst electric potential to validate the modeling of electrochemical reaction
kinetics and the thermodynamic equilibrium potential.
All plots in Fig. 5 show close agreement between the results obtained by the high-resolution high-
accuracy 3D CFD model and the results obtained by the computationally optimized HAN-RT model.
The HAN-RT model distinguishes average values of variables at the (cathode) catalyst surface under
the channel and values under the rib [47]. For easier comparison, the detailed 3D CFD results are also
gathered and averaged to provide corresponding “under channel” and “under rib” values and plotted
alongside the HAN-RT results in Fig. 5.
To avoid the effects of too many physical phenomena overshadowing one another, a case with a high
hydrogen concentration in the anode feed gas and a close to uniform temperature has been simulated.
Thus, predominantly, the effects of variation in oxygen and water vapor concentration are reflected in
Fig. 5: Plots of distribution of values of various variables over cathode catalyst surface and in the cathode channel of the representative unit at the output current level of the steady-state operating point. The surface under channel and surface under rib are distinguished. The x-axis runs along the channel gas flow direction with the anode inlet and the cathode outlet at x = 0 mm. Solid lines refer to 3D CFD results and dots to HAN results with blue color denoting values pertaining to “under channel” location and orange to “under rib” location.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
20
the variation in the plotted variables. As the oxygen is being consumed on the way from the inlet (𝑥 =
0.018 m) to the outlet (𝑥 = 0.000 m), its concentration at the catalyst surface correspondingly decreases
as shown in Fig. 5 a). Additionally, the difference between the conditions at the catalyst surface under
the channel and under the rib is clearly visible with oxygen levels under the rib being lower due to the
longer pathway from the channel. Inversely, water is being produced at the catalyst surface; thus, on the
way from the inlet to the outlet, its concentration at the catalyst surface increases and also the
concentrations “under rib” are higher than those “under channel” with the same rationale (Fig. 5 b)). As
shown in Fig. 5 c), the variation in the distribution of current density closely follows that of the oxygen
catalyst surface concentration. This is an exemplary effect of the reactant concentration term in the
Butler–Volmer equation. The effect of oxygen concentration levels also dominates the cathode potential
as plotted in Fig. 5 d), overshadowing the effect of higher current density, which, on its own, would
assume the opposite trend, i.e., lower electrode potentials at higher current densities.
Within the scope of this study, it is important to present the capability of the HT-PEMFC operation
model to demonstrate the transient operation of an FC adequately. Thus, a challenging simulation case
of a start-up sequence, featuring temporal variations in volume flows of feed gases, FC temperature, and
electric current, is shown in Fig. 6. The FC is started up according to a specific electric current ramp-up
sequence and the accompanying reactant supply and cooling adjustment. The effects of these three
controlled processes are recorded as time variations in electric current (green), air (orange), and
hydrogen (purple) volume flows, and stack temperatures (black) are used as the time-varying boundary
conditions for the simulation.
The simulation objective and the main result are then the temporal variation in the stack voltage (Fig.
6). The voltage trace predicted by the simulation (red) and plotted alongside the experimentally
measured voltage data (blue) shown in Fig. 6 validates the model’s accuracy and predictive capabilities
in transient operation, which inherently depend on the accurate prediction of the reactant concentrations
being validated in [48].
The average real-time factor (simulation time divided by physical time) over the entire simulation was
less than 0.2 and it never exceeded 0.5, confirming the computational efficiency of the HAN-RT model.
Based on the presented results, it can be concluded that the model of FC operation fulfills all four
requirements listed in the introduction and thus represents an adequate basis for coupling to the
degradation model.
Fig. 6: Temporal variation in boundary conditions (normalized output current I, volume flows of reactant gases, and temperature) and comparative evaluation of the temporal variation in the simulated voltage output against the measured output stack voltage (output voltage 𝑈𝐹𝐶 plotted relative to the stack open circuit voltage OCV) during fuel cell start-up procedure. Close agreement between measured and modeled voltages during the entire simulation time validates the transient capabilities of the HAN-RT model.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
21
4.2. Analysis of measured degradation data
The degradation model was calibrated and validated based on experimental data obtained by long-term
degradation tests performed on a real FC system. In [53], two HT-PEMFCs having an initial catalyst
particle size distribution with mean radius of 𝑅𝑚𝑒𝑎𝑛 = 2.06 nm and standard deviation of 𝜎 =
0.58 nm (MEA_t0) were operated at identical steady-state conditions at a temperature of 160℃ and
current density 0.2 A/m2, one for 2700 h (MEA_t1) and the other for 4800 h (MEA_t2). The input
variables of the degradation model were generated using the HAN-RT model, replicating these operating
conditions. As no information is provided in [53] on the position in the FC where the aged MEA was
analyzed, it was assumed that the samples used to determine the particle size distribution were taken
halfway between the inlet and outlet of the feed gas channels.
To distinguish the effects of catalyst and membrane degradation, allowing for a correct calculation of
local catalyst potential, the polarization curves of the fresh and the two aged MEA samples given in [53]
were reproduced using the HAN-RT model. The changes in electrochemical surface area and membrane
thickness during degradation, measured in [53], were used as input parameters of the HAN-RT model,
while the membrane ionic conductivity was used as a fitting parameter. The polarization curves for all
three samples, generated by the HAN-RT model, are compared to experimental data from [53] in Fig.
7. The fitted values of the membrane ionic conductivity show a decrease of 24% for MEA_t1 and 52%
for MEA_t2 compared to the fresh sample MEA_t0, indicating the importance of membrane
degradation.
However, although membrane degradation is not covered in the proposed degradation model, the
estimation of membrane degradation based on polarization curves enables the generation of correct
degradation stimuli for the catalyst degradation model, even in the absence of a dedicated membrane
degradation model. Most notably, the local catalyst potential, defined as the potential difference between
the Pt particles and the adjacent ionomer, can be estimated as
𝑈𝑐𝑎𝑡 = 𝑈𝐹𝐶 + 𝜂𝑎𝑛 + 𝜂𝑚𝑒𝑚, (23)
where 𝑈𝐹𝐶 is the operational voltage of the FC, 𝜂𝑎𝑛 is the anode activation voltage, and 𝜂𝑚𝑒𝑚 is the
membrane voltage drop, all of which are calculated using the HAN-RT model. Based on the polarization
curves for samples MEA_t0, MEA_t1, and MEA_t2, the degradation stimuli were calculated at 0, 2700,
and 4800 h of operation and the values at intermediate times were obtained by quadratic interpolation.
Fig. 7: Comparison between experimentally measured [53] and HAN-RT generated polarization curves for fresh and two
aged MEAs. Excellent agreement between modeled and measured values was achieved by using the measured values of
exchange current density and membrane thickness for all three samples and determining membrane conductivity as a
fitting parameter.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
22
4.3. Degradation model calibration and validation
The calibration of the degradation model was aimed at achieving the best agreement of the modeled
particle size distribution 𝑁𝑖, calculated as explained in Section 3.2.6, for both aged MEA samples with
the ones given in the supplementary materials of [53]. To ensure a sufficiently smooth particle size
distribution and to avoid boundary effects of the largest particles in the distribution, M = 15 size classes
were used between 0 and 15 nm to cover the entire spectra of particle sizes reported in [53]. Numerical
integration of the set of 76 governing equations (see Section 3.2.6) was performed on a personal
computer using the Bulirsch–Stoer semi-implicit extrapolation method with an adaptive time step,
implemented in programming language C [73]. The computational time needed to simulate 4800 h of
FC degradation in steady state operation regime was ~10 s, indicating high computational efficiency of
the proposed model.
The model was calibrated by adapting only two parameters, 𝛾𝐻𝑇 and 𝑘𝑑𝑒𝑡 , which were then kept
constant during the entire degradation simulation. The first adaptive parameter was reaction rate
acceleration due to high temperature 𝛾𝐻𝑇 (introduced in Section 3.2.2). This factor also covers possible
differences in reaction rates originating from the fact that the catalyst material used in [53] is not pure
platinum but Pt/Ru nanoparticles, which might lead to different oxidation kinetics. The second adaptive
parameter was particle detachment factor 𝑘𝑑𝑒𝑡 (introduced in Section 3.2.5), addressing the detailed
mechanism of particle detachment not covered in the proposed model. Other parameters used in the
model were determined based on the established literature. To ensure the consistency between the rates
of electrochemical reactions, most values were taken from the paper by Pandy [14], covering most of
the relevant processes. The values of parameters used in the present study are given in Appendix 1.
Good agreement between experimental and modeled particle size distribution, as shown in Fig. 8, was
obtained at parameter values of 𝛾𝐻𝑇 ≈ 1300, consistent with the predictions made in Section 3.2.2, and
𝑘𝑑𝑒𝑡 ≈ 8900. Note that values of both calibration parameters directly depend on the values of other
degradation model parameters, obtained from the literature and listed in Appendix 1, which differ from
source to source [14,16,22,27]. Both values, as well as other model parameters, should therefore be
further verified by additional experiments, as proposed in Section 4.4. The deviations between both
distributions could (to some extent) also be attributed to the limited precision of the experimentally
Fig. 8: Comparison between experimentally measured particle size distribution (white-filled histogram) [53] and one produced by degradation model (colored histogram) for fresh and aged MEA. The measured particle distribution on a fresh sample MEA_t0 (black) was used as the model input. The modeled distribution of particles in both aged samples MEA_t1 (red) and MEA_t2 (blue) shows good agreement with experimentally measured data.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
23
measured distribution, as measurement uncertainty is not reported in the source. The result clearly shows
that the proposed degradation model is capable of correctly reproducing the changes in particle size
distribution for two different stages of catalyst aging using the same calibration parameters, indicating
a valid approach to degradation modeling. The capability of the model to reproduce the degradation
phenomena mechanistically over an extended period of 2700 and 4800 h of FC operation can be
considered as a significant progress in the field of system-level FC degradation modeling.
4.4. Analysis of individual degradation mechanisms
This section presents a more detailed analysis of the degradation model results, elucidating the model
capabilities in distinguishing different degradation mechanisms. Furthermore, acknowledging that the
degradation tests, given in [53], were performed at a single FC operating point (i.e., at fixed operating
conditions) additional experiments supporting more precise model calibration are proposed.
Detailed analysis of Oswald ripening and particle agglomeration
To elucidate the details of the Oswald ripening modeling and particle agglomeration, as well as the
individual contributions of these effects to the overall particle growth, Fig. 9Fig. shows the rates of
change of particle size distribution due to the platinum particle dissolution (Oswald ripening, (14)) and
agglomeration (Eq. (16)) as a function of particle size for two FC operation modes and two particle size
distributions with M = 30 size classes. Fig. 9 a) shows the degradation rates after 48 h of long-term
degradation test, used in the model calibration (see Section 4.2), i.e., at a current density of 0.2 A/cm2
[53], resulting in a catalyst voltage (as calculated by HAN-RT) of 𝑈𝑐𝑎𝑡 ≈ 0.6 V. Fig. 9 b) shows the
same quantities at a higher FC voltage of 0.96 V on a partially aged sample (MEA_t1 in [53]) with mean
particle radius of 𝑅𝑚𝑒𝑎𝑛 = 5.49 nm and a standard deviation of 𝜎 = 2.28 nm. As for the case of model
calibration, the degradation stimuli were generated by the HAN-RT model.
The shape of blue curves in Fig. 9 shows the rates of change in particle size distribution due to the
Ostwald ripening (��𝑑𝑖𝑠𝑠, Eq. (14)). The Kelvin term initiates a rapid dissolution of smaller particles and
redeposition of platinum on larger particles, as explained in Section 3.2.4. The sharp peak in the
dissolution rate of 5.7 × 10−5 at 0.75 nm in Fig. 9 b), not completely shown on plot, results from the
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
24
competition between the rapid dissolution rate of small particles due to the Kelvin term (11) and the
small number of available small particles in the size distribution.
The detachment of particles from the carbon support (orange dashed line, ��𝑑𝑒𝑡 in Eq. (18)), and their
attachment to the remaining particles (green dashed line, ��𝑎𝑡𝑡 in Eq. (19)) are both proportional to the
total number of particles. The detachment is more prominent for smaller particles due to the faster carbon
corrosion (Kelvin term in reaction rate Eq. (8) f)) and smaller contact surface between the particle and
its carbon support (Eq. (18)). The merged particles (black dashed line, ��𝑚𝑒𝑟 in Eq. (21)) appear at larger
sizes (Eq. (20)), resulting in the combined effect of particle agglomeration (��𝑎𝑔𝑔, red line, Eq. (16))
being similar to the Ostwald ripening, initiating a net particle growth. Note that the left (dissolution rate)
and the right (agglomeration rate) axes in Fig. 9 are scaled differently. At a high voltage (Fig. 9 b)), the
effects of dissolution seem to be orders of magnitude more pronounced than the particle agglomeration,
while at a lower voltage during long-term operation (Fig. 9 a)), the contributions of both effects are
comparable, which can be explained by the decreased platinum dissolution at lower voltages (Eq. (11)).
The presented results indicate that both phenomena are modeled in a physically/chemically plausible
way. Because the proposed models of Ostwald ripening and particle agglomeration are based on
different mechanisms, they yield different size-dependence redistribution rates, as shown in Fig. 9, and
different relative rates depending on the FC voltage. These phenomena have already been reported as
experimental observation in some studies [74], which further confirms the validity of the proposed
degradation model and suggests that the relative contribution of Oswald ripening and particle
agglomeration could be determined from more precise measurements of the shape of particle size
distribution of aged samples obtained from degradation tests performed at different voltages.
Transient degradation processes
To highlight the importance of the impact of transient FC conditions on the relative rate of different
degradation phenomena, the model was tested in one of the standard DOE voltage cycling regimes
between 𝑈𝐹𝐶 = 0.6 V and 0.96 V with a cycle period of 6 s [75]. The model was tested on a partially
aged sample (MEA_t1 in [53]) with a mean particle radius of 𝑅𝑚𝑒𝑎𝑛 = 5.49 nm and a standard
deviation of 𝜎 = 2.28 nm, with the time traces of degradation stimuli being generated by the HAN-RT
model. The effects of this cycling were analyzed based on simulated 𝐶𝑂2 emissions, indicating carbon
corrosion, 𝑃𝑡2+ ion concentration in the ionomer, Pt particles dissolution, and surface oxide
concentrations, revealing detailed microscopic processes in the catalyst layer.
Fig. 9: Relative rates of particle redistribution ��/𝑁𝑡𝑜𝑡𝑎𝑙 due to Pt dissolution (blue) and agglomeration (red), split down to individual contributions due to detachment (orange dashed), attachment (green dashed), and merging (black dashed). a) Operation at current density of 0.2 A/cm2 (𝑈𝑐𝑎𝑡 ≈ 0.6 V) on 2.06 nm particles (MEA_t0). b) High voltage operation at 𝑈𝐹𝐶 =0.96 V on 5.49 nm particles (MEA_t1). Both contributions to particle redistribution show clear differences in particle size dependence as well as strong dependence on fuel cell potential.
b) a)
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
25
The black dotted line in Fig. 10 b) shows that the electric potential in the catalyst layer 𝑈𝑐𝑎𝑡, calculated
by the HAN-RT model, is slightly higher than the FC potential 𝑈𝐹𝐶, as expected based on Eq. (23). Fig.
10 a) shows the rate of carbon mass loss (orange) and the 𝑃𝑡2+ion concentration in the ionomer (blue),
resulting from voltage cycling as a function of time. The platinum ion concentration shows a sharp
increase when the voltage is increased and a rapid decline when it is decreased. This is in agreement
with the prediction that platinum dissolution is caused by high voltage, whereas this figure also shows
that the reaction rates, determined during model calibration, result in a very fast establishment of steady-
state concentrations. Such fast transitions suggest that the inclusion of diffusion effects to describe the
redistribution of 𝑃𝑡2+ions in the ionomer might improve the quality of the proposed degradation model.
The rate of carbon corrosion, expressed as a mass loss of carbon per second per catalyst geometric area,
shows typical peaks after the increase in voltage, similar to the ones measured by Borup [57] and
explained by Pandy [14] for LT-PEMFCs. The peaks are much narrower that the ones measured in [57],
which is the result of a much faster reaction kinetics in the HT-PEMFC compared to that in LT-PEMFC.
The appearance of carbon corrosion peaks is explained by the interaction of the surface oxide groups on
platinum and carbon. As shown in Fig. 10 b), after the change in voltage (black dotted line), the surface
oxide coverages are converging toward a steady-state value, which depends on both the voltage and the
particle size (Eq. (8)). The 𝐶 − 𝑂𝐻 surface groups accumulate on the carbon surface during the low
voltage part of the cycle [14], as shown in Fig. 10Fig. b). Afterward, when the voltage is increased far
above the equilibrium potential of the carbon corrosion reaction 𝑈𝑃𝑡𝐶 = 0.62 V, the 𝐶 − 𝑂𝐻 groups
rapidly react with the accumulated 𝑃𝑡 − 𝑂𝐻 groups via the 𝑃𝑡𝐶 reaction (8) f), resulting in intense
carbon corrosion. This can be observed in Fig. 10 b) as a rapid drop in Θ𝑃𝑡𝑂𝐻 and Θ𝐶𝑂𝐻 coverages after
voltage increase and as a peak in carbon corrosion rate in Fig. 10 a).
With the relative values of reaction rates taken from [14], the model predicts higher carbon corrosion
rates at low voltage compared to that at high voltage (Fig. 10Fig. a)). This is explained by the larger
availability of the 𝐶 − 𝑂𝐻 groups, initiating carbon corrosion at low voltage, as shown in Fig. 10 b). At
high voltage, although the exponential part of the reaction rate for the reaction 𝑃𝑡𝐶 (Eq. (8) f)) is being
increased, the concentration of the available 𝐶 − 𝑂𝐻 and 𝑃𝑡 − 𝑂𝐻 groups is decreased due to their
concurrent transformation to stable 𝐶 = 𝑂 and 𝑃𝑡 = 𝑂 groups, which leads to reduced rates of carbon
corrosion at these operating conditions.
These results highlight the capability of the proposed model to plausibly address the transient
degradation phenomena and suggest that the time resolved measurement of 𝐶𝑂2 emissions and 𝑃𝑡2+
Fig. 10: Results of degradation simulation during accelerated stress test of HT-PEMFC by fuel cell voltage cycling between𝑈𝐹𝐶 = 0.6 and 0.96 V every 3 s. a) Rate of corrosion of carbon support (orange) and concentration of dissolved platinum ions 𝑃𝑡2+ in catalyst ionomer (blue). The carbon corrosion rate shows typical peaks, reported by experiments in voltage cycling regime. b) Electric potential in catalyst layer (𝑈𝑐𝑎𝑡, black dotted) and surface oxide coverage of Pt particles and carbon support for two different particle sizes: 2 nm (solid lines) and 5 nm (dashed lines). The coverages change with time due to changes in voltage (Eq. (8)), but also depend on particle size due to the Kelvin term (7).
b) a)
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
26
ion concentration in the catalyst layer could be used to improve the model calibration and further
distinguish between the contribution of the Oswald ripening and the particle agglomeration to the overall
catalyst degradation.
Position dependence of catalyst degradation
Several experiments show that catalyst degradation is more pronounced toward the outlet side of cathode
gas channels compared to the inlet side [71,72]. Because the HAN-RT model is fully capable of
comprehending the differences in degradation stimuli along the gas paths, it was used for simulating
long-term degradation tests, described in [53], for regions of catalyst at the cathode inlet, in the middle
of the FC, and at the cathode outlet. The resulting changes in mean radius and standard deviations of
particle size distribution over time, shown in Fig. 11, indicate that the degradation model indeed
reproduces the trends observed in experiments. It is discernible from Fig. 11, that the particle growth is
more prominent at the cathode outlet (blue line), which can be explained by the increase in humidity of
cathode gas toward the outlet, as shown in Fig. 5 d). Because humidity is the main precursor of oxidation
reactions (4), the regions of the catalyst closer to the outlet will experience faster carbon corrosion and
consequently faster particle growth, which is in agreement with the results of experimental carbon
corrosion analysis on segmented FCs [71,72]. This result indicates the importance of spatially resolved
fields of degradation stimuli for accurate modeling of catalyst aging and the importance of knowledge
on sample positions for correct degradation model calibrations.
The results presented in this section show that all considered catalyst degradation processes are modeled
in a plausible way, and are compatible with experimental results given in the established literature. In
addition, the results highlight additional capabilities of the model to track individual degradation
processes, which are valuable for the efficient model calibration and crucial for the analysis of aging
processes in real FC systems and development of suitable mitigation strategies.
5. Conclusion
This paper presents an innovative predictive real-time capable system-level modeling framework for the
performance and cathode platinum degradation modeling of high temperature proton exchange
membrane fuel cells. The unique features of this computationally efficient model on the system level
Fig. 11: Increase in mean radius (solid line) and standard deviation (dashed line) of Pt particle size distribution during fuel cell operation for inlet, middle, and outlet of cathode channel, compared to experimental data (black dots with error bars) [53]. Particle agglomeration is more severe toward the end of the channel where water concentration is higher, resulting in faster carbon corrosion.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
27
arise from the interaction of the mechanistic spatially and temporally resolved model of the high
temperature proton exchange membrane fuel cells operation and the degradation modeling framework,
based on interacting individual cathode platinum degradation mechanisms. The degradation modeling
framework considers various size distributions of platinum particles and comprises carbon and platinum
oxidation phenomena as well as platinum dissolution, redeposition, detachment, and agglomeration
phenomena, combined in physically consistent and computationally efficient model, enabling consistent
treatment of the entire causal chain while simulating the degradation phenomena over longer time scales
up to the fuel cell’s operating lifespan.
The presented results confirm the credibility of the proposed framework in modeling the FC
performance and platinum degradation at the cathode as well as its computational speed and efficiency.
The model is capable to reproduce and explain several phenomena, observed in fuel cell degradation
experiments. More pronounced cathode catalyst degradation towards the fuel cell gas channel outlet
compared to inlet is explained in the model by the increase in humidity along the channel. Furthermore,
the innovative modeling framework elucidates the differences in contributions to platinum particle
growth: the agglomeration, caused by carbon corrosion, is more pronounced at voltage cycling, while
the Ostwald ripening, caused by platinum dissolution, is more pronounced at higher voltages,
The proposed modeling framework, therefore, represents a significant progress in the area of model-
supported design of fuel cells and in-depth understanding of the cause and effect chain from the fuel cell
operation to its degradation. Hence, it efficiently contributes to the model-supported development of
advanced clean energy conversion technologies. Application areas of the proposed modeling framework
comprise the development and optimization of balance-of-plant design and control functionalities as
well as the optimization of fuel stack parameters to simultaneously prolong the service life and high fuel
cell performance and to contribute to the reduced production costs.
Acknowledgments
The research is partially funded by Slovenian Research Agency (research core funding No. P2-0401)
and by Austrian Research Promotion Agency (research project no. 848810: MEA Power and research
project no. 854867: SoH4PEM). The authors are also grateful to Freudenberg Sealing Technologies
GmbH & Co. KG for sharing the experimental data for model validation.
References
[1] Najafi B, Haghighat Mamaghani A, Rinaldi F, Casalegno A. Long-term performance analysis
of an HT-PEM fuel cell based micro-CHP system: Operational strategies. Appl Energy
2015;147:582–92. doi:10.1016/j.apenergy.2015.03.043.
[2] Fink C, Fouquet N. Three-dimensional simulation of polymer electrolyte membrane fuel cells
with experimental validation. Electrochim Acta 2011;56:10820–31.
doi:10.1016/j.electacta.2011.05.041.
[3] Sivertsen BR, Djilali N. CFD-based modelling of proton exchange membrane fuel cells. J
Power Sources 2005;141:65–78.
[4] Berning T, Djilali N. A 3D, Multiphase, Multicomponent Model of the Cathode and Anode of
a PEM Fuel Cell. J Electrochem Soc 2003;150:A1589. doi:10.1149/1.1621412.
[5] Cheng C-H, Lin H-H. Numerical analysis of effects of flow channel size on reactant transport
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
28
in a proton exchange membrane fuel cell stack. J Power Sources 2009;194:349–59.
doi:10.1016/j.jpowsour.2009.04.075.
[6] Kone J-P, Zhang X, Yan Y, Hu G, Ahmadi G. CFD modeling and simulation of PEM fuel cell
using OpenFOAM. Energy Procedia 2018;145:64–9. doi:10.1016/j.egypro.2018.04.011.
[7] Lobato J, Cañizares P, Rodrigo MA, Pinar FJ, Mena E, Úbeda D. Three-dimensional model of
a 50 cm2 high temperature PEM fuel cell. Study of the flow channel geometry influence. Int J
Hydrogen Energy 2010;35:5510–20. doi:10.1016/j.ijhydene.2010.02.089.
[8] Shimpalee S, Spuckler D, Van Zee JW. Prediction of transient response for a 25-cm2 PEM fuel
cell. J Power Sources 2007;167:130–8. doi:10.1016/j.jpowsour.2007.02.004.
[9] Lee T-W, Tseng AA, Bae K-S, Do YH. Simulation of the Proton-Exchange Membrane (PEM)
Fuel Cell Life-Cycle Performance with Data-Driven Parameter Estimation. Energy & Fuels
2010;24:1882–8. doi:10.1021/ef901519f.
[10] Javed K, Gouriveau R, Zerhouni N, Hissel D. Prognostics of Proton Exchange Membrane Fuel
Cells stack using an ensemble of constraints based connectionist networks. J Power Sources
2016;324:745–57. doi:10.1016/j.jpowsour.2016.05.092.
[11] Ma R, Yang T, Breaz E, Li Z, Briois P, Gao F. Data-driven proton exchange membrane fuel
cell degradation predication through deep learning method. Appl Energy 2018;231:102–15.
doi:10.1016/j.apenergy.2018.09.111.
[12] Silva RE, Gouriveau R, Jemeï S, Hissel D, Boulon L, Agbossou K, et al. Proton exchange
membrane fuel cell degradation prediction based on Adaptive Neuro-Fuzzy Inference Systems.
Int J Hydrogen Energy 2014;39:11128–44. doi:10.1016/j.ijhydene.2014.05.005.
[13] Chen K, Laghrouche S, Djerdir A. Degradation model of proton exchange membrane fuel cell
based on a novel hybrid method. Appl Energy 2019;252:113439.
doi:10.1016/j.apenergy.2019.113439.
[14] Pandy A, Yang Z, Gummalla M, Atrazhev V V., Kuzminyh NY, Sultanov VI, et al. A Carbon
Corrosion Model to Evaluate the Effect of Steady State and Transient Operation of a Polymer
Electrolyte Membrane Fuel Cell. J Electrochem Soc 2013;160:F972–9.
doi:10.1149/2.036309jes.
[15] Bi W, Fuller TF. Modeling of PEM fuel cell Pt/C catalyst degradation. J Power Sources
2008;178:188–96. doi:10.1016/j.jpowsour.2007.12.007.
[16] Darling RM, Meyers JP. Kinetic Model of Platinum Dissolution in PEMFCs. J Electrochem
Soc 2003;150:A1523. doi:10.1149/1.1613669.
[17] Motobayashi K, Árnadóttir L, Matsumoto C, Stuve EM, Jónsson H, Kim Y, et al. Adsorption
of water dimer on platinum(111): Identification of the -OH⋯Pt hydrogen bond. ACS Nano
2014;8:11583–90. doi:10.1021/nn504824z.
[18] Pohl E, Maximini M, Bauschulte A, Vom Schloß J, Hermanns RTE. Degradation modeling of
high temperature proton exchange membrane fuel cells using dual time scale simulation. J
Power Sources 2015;275:777–84. doi:10.1016/j.jpowsour.2014.11.054.
[19] Haghighat Mamaghani A, Najafi B, Casalegno A, Rinaldi F. Predictive modelling and adaptive
long-term performance optimization of an HT-PEM fuel cell based micro combined heat and
power (CHP) plant. Appl Energy 2017;192:519–29. doi:10.1016/j.apenergy.2016.08.050.
[20] Chen H, Pei P, Song M. Lifetime prediction and the economic lifetime of proton exchange
membrane fuel cells. Appl Energy 2015;142:154–63. doi:10.1016/j.apenergy.2014.12.062.
[21] Meyers JP, Darling RM. Model of Carbon Corrosion in PEM Fuel Cells. J Electrochem Soc
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
29
2006;153:A1432. doi:10.1149/1.2203811.
[22] Macauley N, Papadias DD, Fairweather J, Spernjak D, Langlois D, Ahluwalia R, et al. Carbon
Corrosion in PEM Fuel Cells and the Development of Accelerated Stress Tests. J Electrochem
Soc 2018;165:F3148–60. doi:10.1149/2.0061806jes.
[23] Malek K, Franco AA. Microstructure-based modeling of aging mechanisms in catalyst layers
of polymer electrolyte fuel cells. J Phys Chem B 2011;115:8088–101. doi:10.1021/jp111400k.
[24] Ahluwalia RK, Arisetty S, Wang X, Wang X, Subbaraman R, Ball SC, et al. Thermodynamics
and Kinetics of Platinum Dissolution from Carbon-Supported Electrocatalysts in Aqueous
Media under Potentiostatic and Potentiodynamic Conditions. J Electrochem Soc
2013;160:F447–55. doi:10.1149/2.018306jes.
[25] Jayasankar B, Karan K. O2 electrochemistry on Pt: A unified multi-step model for oxygen
reduction and oxide growth. Electrochim Acta 2018;273:367–78.
doi:10.1016/j.electacta.2018.03.191.
[26] Franco AA. Transient Multi-Scale Modelling of Ageing Mechanisms in a Polymer Electrolyte
Fuel Cell: An Irreversible Thermodynamics Approach. ECS Trans 2007;6:1–23.
doi:10.1002/j.1556-6676.1991.tb01597.x.
[27] Darling RM, Meyers JP. Mathematical Model of Platinum Movement in PEM Fuel Cells. J
Electrochem Soc 2005;152:A242. doi:10.1149/1.1836156.
[28] Li Y, Moriyama K, Gu W, Arisetty S, Wang CY. A One-Dimensional Pt Degradation Model
for Polymer Electrolyte Fuel Cells. J Electrochem Soc 2015;162:F834–42.
doi:10.1149/2.0101508jes.
[29] Ruckenstein E. Growth kinetics and the size distributions of supported metal crystallites. J
Catal 1973;29:224–45. doi:10.1016/0021-9517(73)90226-1.
[30] Holby EF, Sheng W, Shao-Horn Y, Morgan D. Pt nanoparticle stability in PEM fuel cells:
influence of particle size distribution and crossover hydrogen. Energy Environ Sci 2009;2:865.
doi:10.1039/b821622n.
[31] Baroody HA, Stolar DB, Eikerling MH. Modelling-based data treatment and analytics of
catalyst degradation in polymer electrolyte fuel cells. Electrochim Acta 2018;283:1006–16.
doi:10.1016/j.electacta.2018.06.108.
[32] Urchaga P, Kadyk T, Rinaldo SG, Pistono AO, Hu J, Lee W, et al. Catalyst Degradation in
Fuel Cell Electrodes : Accelerated Stress Tests and Model-based Analysis. Electrochim Acta
2015;176:1500–10. doi:10.1016/j.electacta.2015.03.152.
[33] Franco AA, Passot S, Fugier P, Anglade C, Billy E, Guetaz L, et al. Pt[sub x]Co[sub y]
Catalysts Degradation in PEFC Environments: Mechanistic Insights. J Electrochem Soc
2009;156:B410. doi:10.1149/1.3056048.
[34] Strahl S, Husar A, Franco AA. Electrode structure effects on the performance of open-cathode
proton exchange membrane fuel cells: A multiscale modeling approach. Int J Hydrogen Energy
2014;39:9752–67. doi:10.1016/j.ijhydene.2014.03.218.
[35] Franco AA, Coulon R, Ferreira de Morais R, Cheah SK, Kachmar A, Gabriel MA. Multi-scale
Modeling-based Prediction of PEM Fuel Cells MEA Durability under Automotive Operating
Conditions. ECS Trans., vol. 25, ECS; 2009, p. 65–79. doi:10.1149/1.3210560.
[36] Franco AA, Tembely M. Transient Multiscale Modeling of Aging Mechanisms in a PEFC
Cathode. J Electrochem Soc 2007;154:B712. doi:10.1149/1.2731040.
[37] Franco AA, Guinard M, Barthe B, Lemaire O. Impact of carbon monoxide on PEFC catalyst
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
30
carbon support degradation under current-cycled operating conditions. Electrochim Acta
2009;54:5267–79. doi:10.1016/j.electacta.2009.04.001.
[38] Jahnke T, Futter G, Latz A, Malkow T, Papakonstantinou G, Tsotridis G, et al. Performance
and degradation of Proton Exchange Membrane Fuel Cells: State of the art in modeling from
atomistic to system scale. J Power Sources 2016;304:207–33.
doi:10.1016/j.jpowsour.2015.11.041.
[39] Wong KH, Kjeang E. Macroscopic In-Situ Modeling of Chemical Membrane Degradation in
Polymer Electrolyte Fuel Cells. J Electrochem Soc 2014;161:F823–32.
doi:10.1149/2.0031409jes.
[40] Wong KH, Kjeang E. Mitigation of chemical membrane degradation in fuel cells:
Understanding the effect of cell voltage and iron ion redox cycle. ChemSusChem
2015;8:1072–82. doi:10.1002/cssc.201402957.
[41] Kjeang E, Singh R, Knights S, Djilali N, Wong KH, Sui PC. Modeling the Effect of Chemical
Membrane Degradation on PEMFC Performance. J Electrochem Soc 2018;165:F3328–36.
doi:10.1149/2.0351806jes.
[42] Quiroga MA, Malek K, Franco AA. A Multiparadigm Modeling Investigation of Membrane
Chemical Degradation in PEM Fuel Cells. J Electrochem Soc 2015;163:F59–70.
doi:10.1149/2.0931514jes.
[43] Reimer U, Schumacher B, Lehnert W. Accelerated Degradation of High-Temperature Polymer
Electrolyte Fuel Cells: Discussion and Empirical Modeling. J Electrochem Soc
2015;162:F153–64. doi:10.1149/2.0961501jes.
[44] Bi W, Gray GE, Fuller TF. PEM Fuel Cell Pt∕C Dissolution and Deposition in Nafion
Electrolyte. Electrochem Solid-State Lett 2007;10:B101. doi:10.1149/1.2712796.
[45] Tavčar G, Katrašnik T. An Innovative Hybrid 3D Analytic-Numerical Approach for System
Level Modelling of PEM Fuel Cells. Energies 2013;6:5426–85. doi:10.3390/en6105426.
[46] Tavčar G, Katrašnik T. An Innovative Hybrid 3D Analytic-numerical Model for Air Breathing
Parallel Channel Counter-flow PEM Fuel Cells. Acta Chim Slov 2014;61:284–301.
[47] Tavčar G, Katrašnik T. A computationally efficient hybrid 3D analytic-numerical approach for
system level modelling of PEM fuel cells. Proc. 5th Eur. PEFC H2 Forum 2015, vol. 2015,
Luzerne: 2015.
[48] Tavčar G, Urthaler P, Heinzl C, Locher T, Kregar A, Katrašnik T, et al. Real time startup
simulation of a high temperature PEM fuel cell for combined heat and power generation. Proc.
6th Eur. PEFC Electrolyser Forum 2017, vol. 2017, Luzern: 2017, p. 127–35.
[49] Rocheleau DN, Sagona JF. Modeling Balance of Plant Components for a PEM Fuel Cell
2006:319–28. doi:10.1115/FUELCELL2006-97182.
[50] Pukrushpan JT. Modeling and control of fuel cell systems and fuel processors. Mech Eng
2003:133. doi:10.1016/S0096-3003(03)00207-8.
[51] Sadler M, Stapleton AJ, Heath RPG, Jackson NS. Application of Modeling Techniques to the
Design and Development of Fuel Cell Vehicle Systems, SAE International; 2001.
doi:10.4271/2001-01-0542.
[52] Cruz Rojas A, Lopez Lopez G, Gomez-Aguilar FJ, Alvarado MV, Sandoval Torres LC.
Control of the Air Supply Subsystem in a PEMFC with Balance of Plant Simulation.
Sustainability 2017;9. doi:10.3390/su9010073.
[53] Hengge K, Heinzl C, Perchthaler M, Varley D, Lochner T, Scheu C. Unraveling micro- and
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
31
nanoscale degradation processes during operation of high-temperature polymer-electrolyte-
membrane fuel cells. J Power Sources 2017;364:437–48. doi:10.1016/j.jpowsour.2017.08.042.
[54] Rinaldo SG, Stumper J, Eikerling M. Physical Theory of Platinum Nanoparticle Dissolution in
Polymer Electrolyte Fuel Cells. J Phys Chem C 2010;114:5773–85. doi:10.1021/jp9101509.
[55] Rinaldo SG, Lee W, Stumper J, Eikerling M. Model- and Theory-Based Evaluation of Pt
Dissolution for Supported Pt Nanoparticle Distributions under Potential Cycling. Electrochem
Solid-State Lett 2011;14:B47. doi:10.1149/1.3548504.
[56] Rinaldo SG, Urchaga P, Hu J, Lee W, Stumper J, Rice C, et al. Theoretical analysis of
electrochemical surface-area loss in supported nanoparticle catalysts. Phys Chem Chem Phys
2014;16:26876–86. doi:10.1039/c4cp03349c.
[57] Borup RL, Papadias DD, Mukundan R, Spernjak D, Langlois DA, Ahluwalia R, et al. Carbon
Corrosion in PEM Fuel Cells during Drive Cycle Operation. ECS Trans 2015;69:1029–38.
doi:10.1149/06917.1029ecst.
[58] Bujalski W, Chandan A, El-kharouf A, Dhir A, Hattenberger M, Ingram A, et al. High
temperature (HT) polymer electrolyte membrane fuel cells (PEMFC) – A review. J Power
Sources 2013;231:264–78. doi:10.1016/j.jpowsour.2012.11.126.
[59] House JE. Principles of chemical kinetics. 2nd ed. Academic Press; 2007.
[60] Lemmon EW, McLinden MO, Friend DG. Thermophysical Properties of Fluid Systems. In:
Linstrom PJ, Mallard WG, editors. NIST Chem. WebBook, NIST Stand. Ref. Database
Number 69, National Institute of Standards and Technology; 2019. doi:10.18434/T4D303.
[61] Harris JG, Yung KH. Carbon Dioxide’s Liquid-Vapor Coexistence Curve And Critical
Properties as Predicted by a Simple Molecular Model. J Phys Chem 1995;99:12021–4.
doi:10.1021/j100031a034.
[62] Rosli RE, Sulong AB, Daud WRW, Zulkifley MA, Husaini T, Rosli MI, et al. A review of
high-temperature proton exchange membrane fuel cell (HT-PEMFC) system. Int J Hydrogen
Energy 2017;42:9293–314. doi:10.1016/j.ijhydene.2016.06.211.
[63] Bevilacqua N, George MG, Galbiati S, Bazylak A, Zeis R. Phosphoric Acid Invasion in High
Temperature PEM Fuel Cell Gas Diffusion Layers. Electrochim Acta 2017;257:89–98.
doi:10.1016/j.electacta.2017.10.054.
[64] Conway BE, Barnett B, Angerstein-Kozlowska H, Tilak B V., Angersteinkozlowska H, Tilak B
V. A surface-electrochemical basis for the direct logarithmic growth law for initial stages of
extension of anodic oxide films formed at noble metals. J Chem Phys 1990;93:8361–73.
doi:10.1063/1.459319.
[65] Heyd D V, Harrington DA. Platinum oxide growth kinetics for cyclic voltammetry. J
Electroanal Chem 1992;335:19–31. doi:10.1016/0022-0728(92)80229-W.
[66] Lázaro MJ, Calvillo L, Celorrio V, Pardo J, Perathoner S, Moliner R. Study and application of
Vulcan XC-72 in low temperature fuel cells, 2011, p. 41–68.
[67] Cherstiouk OV V., Simonov ANN, Moseva NSS, Cherepanova SV V., Simonov PAA,
Zaikovskii VII, et al. Microstructure effects on the electrochemical corrosion of carbon
materials and carbon-supported Pt catalysts. Electrochim Acta 2010;55:8453–60.
doi:10.1016/j.electacta.2010.07.047.
[68] Marcus Y, Loewenschuss A. Chapter 4. Standard entropies of hydration of ions. Annu Reports
Sect “C” (Physical Chem 1984;81:81. doi:10.1039/pc9848100081.
[69] Zorko M, Jozinović B, Bele M, Hodnik N, Gaberšček M. SEM method for direct visual
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
32
tracking of nanoscale morphological changes of platinum based electrocatalysts on fixed
locations upon electrochemical or thermal treatments. Ultramicroscopy 2014;140:44–50.
doi:10.1016/j.ultramic.2014.02.006.
[70] Tatschl R, Fink C, Tavčar G, Urthaler P, Katrašnik T. A Scalable PEM Fuel Cell Modelling
Approach to Support FCEV Component and System Development. Eur. Batter. Hybrid Fuel
Cell Veh. Congr., Geneva: 2017.
[71] Engl T, Gubler L, Schmidt TJ. Think Different! Carbon Corrosion Mitigation Strategy in High
Temperature PEFC: A Rapid Aging Study. J Electrochem Soc 2015;162:F291–7.
doi:10.1149/2.0681503jes.
[72] Robin C, Gerard M, Quinaud M, d’Arbigny J, Bultel Y. Proton exchange membrane fuel cell
model for aging predictions: Simulated equivalent active surface area loss and comparisons
with durability tests. J Power Sources 2016;326:417–27. doi:10.1016/j.jpowsour.2016.07.018.
[73] Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numerical Recipes in C (2nd Ed.):
The Art of Scientific Computing. New York, NY, USA: Cambridge University Press; 1992.
[74] Vion-Dury B, Chatenet M, Guetaz L, Maillard F. Determination of Aging Markers and their
Use as a Tool to Characterize Pt/C Nanoparticles Degradation Mechanism in Model PEMFC
Cathode Environment. ECS Trans., vol. 41, 2011, p. 697–708. doi:10.1149/1.3635604.
[75] US Department of Energy (DOE). Multi-Year Research, Development, and Demonstration
Plan: 3.4 Fuel Cells. Fuel Cell Technol Off 2017;2015:3.4.1-3.4.58. doi:Department of Energy.
[76] Passalacqua E, Lufrano F, Squadrito G, Patti A, Giorgi L. Nafion content in the catalyst layer
of polymer electrolyte fuel cells: effects on structure and performance. Electrochim Acta
2001;46:799–805. doi:10.1016/S0013-4686(00)00679-4.
[77] Zlotorowicz A, Jayasayee K, Dahl PII, Thomassen MSS, Kjelstrup S. Tailored porosities of the
cathode layer for improved polymer electrolyte fuel cell performance. J Power Sources
2015;287:472–7. doi:10.1016/j.jpowsour.2015.04.079.
[78] Zhou F, Simon Araya S, Florentina Grigoras I, Juhl Andreasen S, Knudsen Kær S.
Performance Degradation Tests of Phosphoric Acid Doped Polybenzimidazole Membrane
Based High Temperature Polymer Electrolyte Membrane Fuel Cells. J Fuel Cell Sci Technol
2015;12:021002. doi:10.1115/1.4029081.
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
33
Appendix 1
The input parameters of the degradation model were divided into three groups. The rates and equilibrium
potentials of degradation processes and various additional degradation parameters, introduced in Section
3.2, are listed in Table A1 as “Degradation model parameters.” These parameters were either extracted
directly from the literature or estimated based on the data available in the literature [14–16,53,66,67].
Table A1: Degradation model parameters
Parameter Value Units Description Source
𝐸𝑃𝑡1 0.79 𝑉 Equilibrium potential for reaction 𝑃𝑡1 Calculated from Ref. [14]
𝐸𝑃𝑡2 0.80 𝑉 Equilibrium potential for reaction 𝑃𝑡2 Calculated from Ref. [14]
𝐸𝑃𝑡𝐷𝑖𝑠 1.155 𝑉 Equilibrium potential for Pt dissolution Calculated from Ref. [16]
𝐸𝐶1 0.29 𝑉 Equilibrium potential for reaction 𝐶1 Calculated from Ref. [14]
𝐸𝐶2 0.80 𝑉 Equilibrium potential for reaction 𝐶2 Calculated from Ref. [14]
𝐸𝐶3 0.96 𝑉 Equilibrium potential for reaction 𝐶3 Calculated from Ref. [14]
𝐸𝑃𝑡𝐶 0.62 𝑉 Equilibrium potential for reaction 𝑃𝑡𝐶 Calculated from Ref. [14]
𝑘𝑃𝑡1,𝐿𝑇 2.2 × 10−5 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝑃𝑡1 Calculated from Ref. [14] and [15]
𝑘𝑃𝑡2,𝐿𝑇 2.2 × 10−5 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝑃𝑡2 Calculated from Ref. [14] and [15]
𝑘𝑃𝑡𝐷𝑖𝑠𝑠,𝐿𝑇 3.4 × 10−9 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of Pt dissolution Ref. [15]
𝑘𝐶1,𝐿𝑇 4.6 × 10−8 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝐶1 Calculated from Ref. [14] and [66]
𝑘𝐶2,𝐿𝑇 2.3 × 10−7 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝐶2 Calculated from Ref. [14] and [66]
𝑘𝐶3,𝐿𝑇 1.4 × 10−10 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝐶3 Calculated from Ref. [14] and [66]
𝑘𝑃𝑡𝐶,𝐿𝑇 1.1 × 10−5 𝑚𝑜𝑙/𝑚2𝑠 Reaction rate of reaction 𝑃𝑡𝐶 Calculated from Ref. [14] and [66]
𝑟𝑂𝑥 30 𝑘𝐽/𝑚𝑜𝑙 Pt oxide interaction Ref. [15]
𝑘𝑟𝑒𝑣 0.3 − Pt oxidation reversibility Ref. [16]
Γ𝑃𝑡 2.15 𝐴𝑠/𝑚2 Pt oxidation surface area Ref. [15]
Γ𝐶 4.6 𝐴𝑠/𝑚2 Carbon oxidation surface area Ref. [66], [67]
𝑑𝑃𝑡𝐶 1 𝑛𝑚 Carbon corrosion depth Assumed
𝑐𝑃𝑡2+,𝑟𝑒𝑓 1 𝑚𝑜𝑙/𝑙 Reference 𝑃𝑡2+concentration Ref. [15]
𝑘𝑑𝑒𝑡,𝐶𝑉 8900 − Particle detachment factor Fitted on [53]
γ𝐻𝑇 1300 − Particle detachment factor Fitted on [53]
The parameters, listed as “Material constants” in Table A2, define the physical properties of carbon,
platinum, and water and are independent of detailed FC and MEA properties.
Table A2: Material constants
Parameter Value Units Description Source
𝜌𝑃𝑡 21090 𝑘𝑔/𝑚3 Pt density Ref. [15]
𝜌𝑃𝑡𝑂𝐻 14170 𝑘𝑔/𝑚3 Pt-OH density Calculated from Ref. [15]
𝜌𝑃𝑡𝑂 14100 𝑘𝑔/𝑚3 Pt=O density Ref. [15]
𝜌𝐶 2000 𝑘𝑔/𝑚3 Carbon density -
𝜎𝑃𝑡 2.73 𝐽/𝑚2 Pt surface tension Ref. [15]
𝜎𝑃𝑡𝑂𝐻 1.34 𝐽/𝑚2 Pt-OH surface tension Calculated from Ref. [15]
𝜎𝑃𝑡𝑂 1 𝐽/𝑚2 Pt=O surface tension Ref. [15]
𝑀𝑃𝑡 195 𝑔/𝑚𝑜𝑙 Pt molar mass Ref. [15]
𝑀𝑃𝑡𝑂𝐻 212 𝑔/𝑚𝑜𝑙 Pt-OH molar mass Calculated from Ref. [15]
𝑀𝑃𝑡𝑂 211 𝑔/𝑚𝑜𝑙 Pt=O molar mass Ref. [15]
𝑀𝐶 12 𝑔/𝑚𝑜𝑙 Carbon molar mass -
𝑀𝐻2𝑂 18 𝑔/𝑚𝑜𝑙 Water molar mass -
The parameters listed as “MEA properties” in Table A3 depend on the detailed structure and materials
used in the membrane and catalyst layer of the FC to be modeled. As it was not possible to extract all
parameters for the analysis from the source providing the experimental data used for validation [53],
some estimations were made based on other sources [76–78].
The short version of the paper was presented at the 10th International Conference on Applied Energy (ICAE2018) on August
22–25, 2018, Hong Kong. This paper is a substantial extension of the short version of the conference paper.
34
Table A3: MEA properties
Parameter Value Units Description Source
𝑆0 50 𝑐𝑚2 Fuel cell surface Ref. [53]
𝑑𝑐 48 𝜇𝑚 Catalyst layer thickness Ref. [53]
𝑑𝑚𝑒𝑚
𝑁𝑡𝑜𝑡𝑎𝑙,0
85
1.0 × 1017
𝜇𝑚
−
Membrane thickness
Initial number of Pt particles
Ref. [53]
Calculated from Ref. [78]
𝜀𝑖𝑜𝑛 0.3 − Ionomer volume ratio in catalyst layer Ref. [76]
𝜀𝑃𝑡𝐶 0.2 − Carbon volume ratio in catalyst layer Ref. [76]
𝜀0 0.5 − Void volume ratio in catalyst layer Ref. [77]