Available online at www.sciencedirect.

com

Journal of Power Sources 180 (2008) 309–321

Computationally efficient modeling of the dynamic behavior

of a portable PEM fuel cell stack
S.P. Philipps, C. Ziegler ∗
Fraunhofer Institute for Solar Energy Systems ISE, Heidenhofstr. 2, D-79110 Freiburg, Germany
Received 13 April 2007; received in revised form 19 January 2008; accepted 25 January 2008
Available online 16 February 2008

Abstract
A numerically efficient mathematical model of a proton exchange membrane fuel cell (PEMFC) stack is presented. The aim of this model is
to study the dynamic response of a PEMFC stack subjected to load changes under the restriction of short computing time. This restriction was
imposed in order for the model to be applicable for nonlinear model predictive control (NMPC). The dynamic, non-isothermal model is based on
mass and energy balance equations, which are reduced to ordinary differential equations in time. The reduced equations are solved for a single cell
and the results are upscaled to describe the fuel cell stack. This approach makes our calculations computationally efficient. We study the feasibility
of capturing water balance effects with such a reduced model. Mass balance equations for water vapor and liquid water including the phase change
as well as a steady-state membrane model accounting for the electro-osmotic drag and diffusion of water through the membrane are included.
Based on this approach the model is successfully used to predict critical operating conditions by monitoring the amount of liquid water in the stack
and the stack impedance. The model and the overall calculation method are validated using two different load profiles on realistic time scales of up
to 30 min. The simulation results are used to clarify the measured characteristics of the stack temperature and the stack voltage, which has rarely
been done on such long time scales. In addition, a discussion of the influence of flooding and dry-out on the stack voltage is included. The modeling
approach proves to be computationally efficient: an operating time of 0.5 h is simulated in less than 1 s, while still showing sufficient accuracy.
© 2008 Elsevier B.V. All rights reserved.

Keywords: PEM fuel cell; Modeling; Dynamic; Stack; NMPC

1. Introduction tion [1]. The dynamic behavior of a fuel cell is a highly complex
phenomenon, as it involves different length and time scales. The
In recent years the interest in using hydrogen fuel cells as power of a PEMFC also strongly depends on operating condi-
power supply for portable electronics has grown substantially. tions such as flow rates, relative humidity and temperature of
Compared to batteries fuel cell systems can provide a higher the gases as well as ambient temperature. Mathematical mod-
energy density and instantaneous refilling while avoiding the eling is a powerful tool for understanding and handling this
problem of self-discharge. However, the use of fuel cells as complexity. Nonlinear model predictive control (NMPC) and
power supply for electronic products is challenging because the online optimization of dynamic processes have attracted increas-
power demand of these applications fluctuates. Due to the lim- ing attention over the past decade, see e.g. Ref. [2]. In contrast
ited space in portable electronics the stack can in many cases to empirical control strategies based on experimental observa-
not be buffered by a battery. Thus, the fuel cell does not usually tions and extensive testing, a model-based control allows faster
operate at steady-state. A solid understanding of the dynamic system development and optimal system operation over a wide
response of a proton exchange membrane fuel cell (PEMFC) range of operating conditions. As a prerequisite, NMPC requires
under load changes is crucial for reliable and optimized opera- detailed nonlinear process models.
A considerable amount of work has been done thus far to
model PEMFCs [3,4]. Most of the models are steady-state,
∗ see for example Refs. [5–7]. Less work has been published
Corresponding author.
E-mail address: christoph.ziegler@ise.fraunhofer.de (C. Ziegler). on dynamic fuel cell modeling. Amphlett et al. [8] modeled
URL: http://www.ise.fraunhofer.de (C. Ziegler). the behavior of the stack temperature and the voltage during

0378-7753/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jpowsour.2008.01.089
310 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

start-up, shut-down and load-step. In their model only the energy rithms. To ease the transfer to different stacks we describe the
balance of the solid is modeled dynamically whereas all other methods of parameter identification in detail. In order to meet the
equations are assumed to be at quasi-steady state for a given challenge of realizing sufficiently exact modeling results with
solid temperature. Lee et al. [9] used an object-oriented approach short calculation time some strong simplifications are made,
based on stationary equations. Dynamic profiles are created by which are justified by the good agreement between simulation
calculating the quasi-stationary solution variables for each time- and experimental results.
step. Ceraolo et al. [10] used an isothermal model to simulate
the dynamic behavior of the cell voltage to a load change on a 2. Model description
time-scale of seconds. Their model was extended to account for
non-isothermal conditions by Shan et al. [11]. In recent years 2.1. Modeling approach
several authors have presented dynamic models using a similar
approach as Amphlett et al. [8]. Golbert et al. [12] developed a The PEM fuel cell stack model presented here is dynamic and
transient along-the-channel model for control purposes, which non-isothermal. The model is based on transient energy and mass
includes mass balances of liquid water and water vapor. Yu et balance equations, a membrane model and an electrical model
al. [13] presented another extension of the model of Amphlett, based on the tafel equation. Convective heat and mass transfer
which accounts for the influence of latent heat on the energy within the stack are accounted for dynamically. A mass balance
balance. Pathapati et al. [14] included dynamic mass balance of water in the liquid and vapor phase is included. Condensation
equations and energy balance equations for the gases. In addi- and evaporation in the channels as well as water generation at
tion, a term to account for the double layer capacity is presented. the cathode are accounted for. The membrane model is steady
The influence of flooding on the dynamic behavior of the stack state and accounts for the electro-osmotic drag and back dif-
voltage under isothermal conditions was modeled by McKay et fusion of water. The steady-state electrical model incorporates
al. [15]. All of these models are reduced in terms of dimension- the influence of pressure and temperature changes as well as the
ality and comprehensiveness. In recent years several research voltage drop due to activation and ohmic losses.
groups have published valuable in-depth analyses on the tran- Fuel cell stacks are commonly characterized by measuring
sient behavior of PEM fuel cells using commercial software tools the time evolution of the stack voltage and the stack tempera-
either based on computational fluid dynamics, e.g. Refs. [16–21] ture subject to specific operating conditions like power demand,
or on finite-element, multiphysics simulation approaches, e.g. ambient and gas temperature and gas humidity. In integrated fuel
Refs. [22–25]. These studies help in understanding the funda- cell systems the stack voltage, the stack temperature and the gas
mental physical processes and interactions within the fuel cell. flow rates are usually monitored. The stack model presented here
However, due to the massive computational effort required they allows the simulation of the most important parameters for the
are not suitable for online control. operation of a PEMFC stack. The model considers four mon-
This work presents a dynamic model approach for portable itoring points for the mass and heat balance of the gases: the
fuel cell stacks. The aim of our work is to model the dynamic inlet and outlet of the stack on the anode and cathode side. At
behavior of a portable PEM fuel cell stack on relevant time these points, gas temperatures and molar fluxes of the different
scales for technical applications under the restriction of keeping species are considered. The operating conditions at the stack
the computing time short. Therefore, a reasonable compromise inlets are used as input values of the model. Furthermore, the
between physical accuracy and numerical efficiency is found current density is an input variable. Based on the values of the
which makes the model suitable for NMPC. Despite of the operating conditions, the model predicts the molar fluxes and gas
number of very valuable contributions to the dynamic fuel cell temperatures at the stack outlets, the average temperature of the
modeling, a validated model approach, which meets these needs solid material, the stack voltage and the average concentration
was not found in the literature. Many of the existing models of liquid water in the stack. Fig. 1 shows the solution variables
require massive computational effort either in terms of memory and operating conditions and illustrates that these parameters
usage or computing time or both, e.g. Refs. [16–21,23–25]. Most are accessible even in a fully integrated fuel cell system. For the
of the reduced models in the literature are either not designed for stack temperature, four monitoring points are indicated from
NMPC purposes, e.g. Refs. [9,10] or are not validated against which the average temperature is calculated.
experimental data of a PEMFC stack on realistic time scales, The derivation of the model equations is done in a bottom-up
e.g. Refs. [11–13]. The reduced model presented here is vali- approach that can be split up into three steps. First, we con-
dated against experimental data of a PEMFC stack for different sider one particular cell of the stack. Balance equations are set
load profiles on realistic time scales of up to 30 min. The model up for a representative elementary volume of the cell (REV1 )
validation study does also include an analysis of the character- and for a representative elementary channel (REC) in the cell
istics of the stack in critical states of operation. The model is (REV2 ). The model is reduced to one geometrical dimension, the
non-isothermal and considers the mass transfer and the electro- direction along the channel length. Second, the time-dependent
chemical reactions. Moreover, the model accounts for both water balance equations for the REVs are integrated along the channel
vapor and liquid water and for the phase transition. In contrast to length. For a complete description of one cell all gas channels
previous modeling studies the average liquid water concentra- are assumed to behave like the REC. Third, the stack is modeled
tion is used to predict flooding of the fuel cell based on the liquid as several coupled cell modules. The approach and the model
water balance. This can be useful for improved NMPC algo- geometry is illustrated in Figs. 1–3. The model assumptions are
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 311

Fig. 3. Sideview of a cell along the cut indicated in Fig. 2, illustrating the direc-
tion of the molar flow Na,i of species i in the anode gas channel and the molar
flow Nc,j of species j in the cathode gas channel from the channel inlets to the
channel outlets. The representative elementary volume REV2 is exemplified for
the anode gas channel.

a complete and instantaneous mixing perpendicular to the

direction of the gas flow is assumed. Thus, the gas velocity
and concentrations within a gas channel are uniform in the y-
and z-direction. The underlying assumption of laminar flow
Fig. 1. 3D-View of a PEMFC stack consisting of six cells. The is widely used in fuel cell modeling, e.g. Refs. [16,21,31–33],
membrane–electrode assemblies between the bipolar plates are indicated in dark and was also satisfied by a Reynolds number calculation.
gray. The measurement points that were used for the characterization of the stack • The stack temperature is constant in the y- and z-direction.
with respect to thermal and mass balance are indicated. The values obtained at • Liquid water is assumed to exist in the form of small droplets
these measurement points correspond to the input and solution variables of the
on the surface of the gas channels [5]. As heat exchange with
model. The cut for Fig. 2 is indicated in light gray.
the solid material is much faster than with the gas, liquid water
is further assumed to have stack temperature.
motivated by the preservation of computational efficiency which
• The volume of the liquid water is assumed negligible. Hence,
is required for the simulation of realistic load cycles in system
it has no influence on the gas transport in the channels.
simulation and for the application in NMPC. In contrast, more
• Cross-over of nitrogen, hydrogen and oxygen through the
detailed model approaches, e.g. Refs. [26,27] require massive
membrane is neglected.
computational effort either in terms of memory usage or com-
• The electrochemical reactions and the transport processes are
puting time or both. The validity of our approach is shown by a
assumed to be homogeneous throughout the stack.
comparison of simulation and experimental results for different
• The mass transport resistance of the gas diffusion layer is
realistic load profiles.
neglected.
• The electrical and thermal contact resistances are neglected.
2.2. Model assumptions
The model assumptions are listed below. The axis system is
illustrated in Fig. 2. Computational efficiency is mainly achieved by the following
model characteristics: in order to reduce the number of mod-
• The gas channels are treated as plug-flow reactors. This ules to be calculated only one cell is modeled. The results are
approach is used by several other authors, e.g. Refs. upscaled according to the number of cells in the stack. As an
[6,10,12,28–30] and is generally accepted. A good overview adequate modeling of the highly complex phenomena in the
of approaches used for the modeling of gas flow in the chan- GDL and the membrane would require massive computational
nels is given in Refs. [3,4]. A uniform velocity profile and effort, the GDL is neglected and the membrane is assumed to
be at steady-state. The spatially resolved characteristics of the
fuel cell stack are neglected through the reduction of the model
equations to ordinary differential equations in time.

3. Mathematical model

In the first step of the model derivation equations for energy

and mass balance are formulated for a representative elementary
volume of the solid material (REV1 ) and of a representative gas
channel (REV2 ). As illustrated in Figs. 2 and 3, REV1 corre-
sponds to a section of a representative cell, whereas REV2 is
a channel section of a representative elementary gas channel.
Fig. 2. 3D-View of the cell model corresponding to the cut in Fig. 1, indicating Mass balance equations are set up for hydrogen H2 , oxygen O2 ,
the cross-sectional area of the cell As and of the gas channels Aa and Ac , the nitrogen N2 , water vapor H2 Ov , and liquid water H2 Ol in REV2 .
representative elementary volume REV1 as well as the cut for Fig. 3. The structure of the mass balance equations is identical for all
312 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

species: where Ak is the cross-sectional area of REV2 and Nk,i is the

molar flux of species i along the REV. Analogously, Eq. (2) for
∂t ck,i = −(∇ · Mk,i ) + Rk,i , (1) the energy balance of both REVs is reduced to

where the change in concentration ck,i of species i in volume ∂t Uk (x, t)= − ∂x (Uk (x, t)vk,i (x, t))−κ∂x2 T (x, t)+Sk (x, t). (4)
element k is given by the divergence of the molar flow rate Mk,i As a consequence of this reduction, the mathematical model
and by appropriate rates of production and consumption Rk,i . contains several effective parameters that include spatially dis-
Energy balance equations are formulated for the gas channels of tributed effects in the y–z-plane. In Section 6 it is described how
anode and cathode as well as for the solid material. Their basic these parameters are obtained. Below, the different balance equa-
structure is tions are presented in the concise form corresponding to Eqs. (3)
and (4), which is the form after step one of the derivation. Steps
∂t Uk = −(∇ · Uk vk,i ) − ∇ · (κ∇T ) + Sk , (2) two and three are illustrated in Section 4. The parameters and
symbols used in the model equations are listed in Tables 1 and 2.
where the change of internal energy Uk in volume element k is Table 3 contains a list of subscripts and superscripts.
given by convective transport of energy associated with the mass
average velocity vk,i , transfer by heat conduction with the heat 3.1. Energy balance
conduction coefficient κ, and energy production or consumption
with the rate Sk . 3.1.1. Energy balance, gas channels
As mentioned previously, we assume plug-flow conditions The following continuity equation describes the energy bal-
for the gas channels. Hence, Eq. (1) can easily be reduced to the ance of the gaseous species in REV2 . The considered species are
following equation: i = H2 , H2 Ov for the anode and i = O2 , N2 , H2 Ov for the cath-
ode. Liquid water is assumed to exist in form of small droplets
∂t ck,i (x, t)Ak = −∂x Nk,i (x, t) + Rk,i (x, t)Ak , (3) at the surface of the channels. The liquid water is assumed to be

Table 1
List of parameters and constants
Symbol Explanation Value Reference

Aa Cross-sectional area of anode gas channel (9 ± 2) × 10−7 m2 Meas.

Aa, sg Heat exchange area per unit length between solid and gas, anode side (4.2 ± 0.4) × 10−3 m Meas.
Ac Cross-sectional area of cathode gas channel (12 ± 2) × 10−7 m2 Meas.
Ac, sg Heat exchange area per unit length between solid and gas, cathode side (4.6 ± 0.4) × 10−3 m Meas.
As Cross-sectional area of cell (5.7 ± 0.9) × 10−5 m2 Meas.
Ass Heat exchange area between solid and surroundings per unit length (1.9 ± 0.2) × 10−2 m Meas.
CH2 Heat capacity of H2 28.8 J mol−1 K−1 [37]
CH2 Ol Heat capacity of H2 Ol 75.3 J mol−1 K−1 [37]
CH2 Ov Heat capacity of H2 Ov 33.6 J mol−1 K−1 [37]
CN2 Heat capacity of N2 29.1 J mol−1 K−1 [37]
CO2 Heat capacity of O2 29.3 J mol−1 K−1 [37]
Cs Specific heat capacity of stack solid material (770 ± 40) J kg−1 K−1 Meas.
ref
Dm,H 2O
Diffusion coefficient of water in membrane for standard conditions 5.5 × 10−11 m2 s−1 [36]
dy Scaling factor for the channel width 2.2 ± 0.2 Meas.
F Faraday constant 96, 485 C mol−1 [38]
ha Height of anode gas channel (6 ± 1) × 10−4 m Meas.
hc Height of cathode gas channel (8 ± 1) × 10−4 m Meas.
Hvap Enthalpy of phase transition of water 44, 000 J mol−1 [37]
kphase Condensation rate constant 100 s−1 [36]
km,p Permeability of water in membrane 1.58 × 10−18 m2 [36]
Leff Effective channel length (465 ± 2) × 10−3 m Meas.
ncells Number of cells 6 Meas.
nchan Number of channels in each flow field 2 Meas.
R Ideal gas constant 8.314 J K−1 mol−1 [38]
Sa Molar entropy of reaction at anode 0.104 J K−1 mol−1 [39]
Sc Molar entropy of reaction at cathode −326.36 J K−1 mol−1 [39]
T0 Reference temperature 298 K Meas.
tm Thickness of membrane (250 ± 1) × 10−7 m Meas.
0
Uoc Open-circuit voltage for standard conditions 1.23 V Meas.
Usg Heat transfer coefficient between solid material and gas 25 W m−2 K−1 [5]
Uss Heat transfer coefficient between solid material and surroundings 4.7 ± 0.9 W m−2 K−1 Meas.
wq Width of gas channel (1.5 ± 0.1) × 10−3 m Meas.
μH2 Ol Viscosity of H2 Ol 3.56 × 10−4 Pa s [37]
ρs Average density of stack material 2500 ± 100 kg m−3 Meas.
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 313

Table 2 Table 3
List of symbols Subscripts and superscripts
Symbol Explanation Unit Symbol Explanation

Ak Cross-sectional area of REV2 m2 0 Reference conditions

Aq Cross-sectional area of gas channel in electrode q m2 a Anode side
aH2 Ov Activity of water vapor – act Active area
ck Constant with index k – av Average
ck,i Concentration of species i in volume element k mol m−3 c Cathode side
cq,i Concentration of species i in gas channel q mol m−3 cell One cell in the stack
Dm,H2 O Diffusion coefficient of H2 O in the membrane m2 s−1 chan Channel
I Current A drag Electro-osmotic drag
j Current density A m−2 eff Effective
j0,c Cathodic exchange current density A m−2 H2 Ol Liquid water
Mk,i Molar flow rate of species i in volume element k mol s−1 m−2 H2 Ov Water vapor
Nq,i Molar flow of species i in gas channel q mol s−1 i Species index
Nk,i Molar flow of species i in REV2 mol s−1 in Gas inlet
mol s−1
in/out,rec
Nq,i Molar flow of species i into/out of the REC of j Species index
electrode q k Volume element
in,cell
Nq,i Molar flow of species i into the gas inlet of the mol s−1 l Liquid phase
cell on the q side m Membrane
out,cell
Nq,i Molar flow of species i out of the gas outlet of the mol s−1 oc Open-circuit
cell on the q side out Gas outlet
in,stack
Nq,i Molar flow of species i into the gas inlet of the mol s−1 phase Phase transition
stack on the q side q Anode or cathode
rec Representative elementary channel
out,stack
Nq,i Molar flow of species i out of the gas outlet of the mol s−1
ref Reference
stack on the q side
s, stack Of the stack
Nq,phase Rate of phase change in electrode q mol s−1 m−1
sg Exchange between solid and gas
ndrag Electro-osmotic drag coefficient –
ss Exchange between solid and surroundings
Pq Average pressure in gas channel q Pa
sur Surroundings
Pq,sat Saturation pressure in gas channel q Pa
v Vapor phase
pq,i Partial pressure of species i in electrode q Pa
p0i Partial pressure of species i at reference Pa
conditions
Rk,i Rate of consumption or production of species i in mol m−3 s−1 at stack temperature. The energy balance of the gas reads
volume element k
Sk Rate of energy conversion in volume element k W m−3 
T Temperature K ∂t [cq,i (x, t)Ci Tq (x, t)Aq ]
Ta,in Gas temperature at anode gas inlet K i
Ta,out Gas temperature at anode gas outlet K 
Tc,in Gas temperature at cathode gas inlet K =− ∂x [Nq,i (x, t)Ci Tq (x, t)]+Aq,sg Usg [Ts (x, t)−Tq (x, t)]
Tc,out Gas temperature at cathode gas outlet K i
rec
Tq,in/out Gas temperature at gas outlet/inlet of REC K
−Hvap Nq,phase (x, t), (5)
Tq Gas temperature in gas channel q K
Ts Stack temperature K
where q = a, c denotes the anode and the cathode side, respec-
Ts,av Average stack temperature K
Tsur Temperature of surroundings K tively. i is the species index, cq,i is the concentration of species
t Time s i in the channel q, Ci is the heat capacity, Tq is the gas tem-
Ucell Cell voltage V perature and Aq is the cross-sectional area of the REC, which
Uk,i Internal energy of species i in volume element k J m−3 equals Ak . Nq,i is the molar flux of species i along the channel
Uoc Open-circuit voltage V
q. Aq,sg is the heat exchange area between gas and solid mate-
Ustack Stack voltage V
vk,i Velocity of species i in volume element k m s−1 rial, Usg is the corresponding heat exchange coefficient, Ts is
vq,i Velocity of species i in gas channel q m s−1 the stack temperature. Hvap is the enthalpy of evaporation and
z Number of electrons exchanged in reaction – Nq,phase is the rate of the phase change per unit length. The term
α Symmetry factor – on the left-hand side of the equation describes the change of
αnet Net water migration coefficient –
internal energy in a volume element of unit length. The terms
ηcD Activation losses in cathodic reaction V
ηoc Losses in open-circuit voltage V on the right-hand side describe from left to right: the transfer of
ηohm Ohmic losses V internal energy by convection; the heat transfer between gas and
κs Heat conduction coefficient of solid material W m−1 K−1 solid stack material and the heat consumption or production due
λm Membrane water content – to evaporation and condensation of water.
νi Stoichiometry factor of species i –
σm Membrane conductivity S m−1
3.1.2. Energy balance, solid material
The energy balance for the solid material in REV1 links
the processes in the gas channels of the anode and cathode.
314 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

It reads equation of water vapor in REV2 on the cathode side reads

ρs Cs As ∂t Ts (x, t) ∂t [cc,H2 Ov (x, t)Ac ] = −∂x [Nc,H2 Ov (x, t)] − Nc,phase (x, t)
= +nchan Hvap [Na,phase (x, t)+Nc,phase (x, t)] dy wq j(x, t) dy wq αnet j(x, t)
   + + , (8)
Sa Sc 2F F
−nchan dy wq + Ts (x, t)−ηoc −|ηD |−ηohm j(x, t)
c
2F 4F where cc,H2 Ov is the concentration and Nc,H2 Ov is the molar flow
−nchan Usg [Aa, sg (Ts (x, t) − Ta (x, t)) + Ac, sg (Ts (x, t) of water vapor in REV2 . The net water migration coefficient αnet
describes the net-number of water molecules carried through the
−Tc (x, t))] + Ass Uss [Tsur (x) − Ts (x, t)] + As κs ∂x2 Ts (x, t) membrane per proton (Eq. (15)). The change in concentration
−nchan ∂x ([Na,H2 Ol (x, t) + Nc,H2 Ol (x, t)]CH2 Ol Ts (x, t)), (6) of water vapor is balanced by: the convective transport of water
vapor; the condensation or evaporation of water; the generation
where ρs is the density of the solid material, Cs is its heat capac- of water in the electrochemical reaction and the transport of
ity, As is the cross-sectional area of the cell and nchan is the water through the membrane. The mass balance equation for
number of channels in each flow-field. dy is a scaling coeffi- liquid water in REV2 on the cathode side is
cient, which takes into account the enlargement of the contact
area between gas and membrane due to diffusion under land ∂t [cc,H2 Ol (x, t)Ac ] = −∂x [Nc,H2 Ol (x, t)] + Nc,phase (x, t), (9)
areas of the flow-field. wq is the channel width, Sq is the
entropy of reaction in electrode q and F is the Faraday con- where cc,H2 Ol is the concentration and Nc,H2 Ol is the molar flow
stant. The open-circuit losses are ηoc , the activation losses in of liquid water. The concentration of liquid water changes due
the cathode are ηcD and the ohmic losses are ηohm . j denotes the to convective transport of liquid water along the channel and
current density, Ass is the heat exchange area between stack and condensation or evaporation of water. The amount of water
surroundings, Uss is the heat transfer coefficient, Tsur denotes that condenses or evaporates in a volume element is modeled
the temperature of the surroundings and κs is the heat conduc- according to Golbert and Lewin [12]:
tion coefficient. The change of internal energy in the solid stack
material in REV1 , which is described by the term on the left-hand kphase wq hc
side, is given by the following source and sink terms on the right- Nc,phase (x, t)= (pc,H2 Ov (x, t)−Pc,sat (Tc (x, t))), (10)
RTc (x, t)
hand side of the equation: (a) heat generation and consumption
due to condensation or evaporation of water in anode or cath- where kphase is the condensation rate constant, hc is the channel
ode; (b) heat generation due to activation energy and irreversible height, R is the ideal gas constant, pc,H2 Ov denotes the partial
losses; (c) heat transfer between bulk material and gases of pressure of water vapor in the cathode gas channel and Pc,sat is
anode and cathode side; (d) heat transfer between solid material the saturation pressure of water vapor, respectively. It is assumed
and surroundings; (e) heat conduction driven by a temperature that water condenses when pc,H2 Ov > Pc,sat and that existing
gradient within the stack; (f) heat transferred by convection liquid water evaporates when pc,H2 Ov < Pc,sat .
of liquid water. As a cell contains several channels the terms
(a)–(c) and (f) need to be multiplied with the number of channels
nchan . 3.3. Electrical model

The cell potential Ucell is calculated from the following equa-

3.2. Mass balance tion [12]:

For the mass balances of REV2 the mass transport along the Ucell = Uoc − ηoc − |ηcD | − ηohm
channel by convection, the fuel consumption of H2 and O2 , the  
production of H2 Ov in the cathode side reaction, the evapora- = Uoc − ηoc −
RTs
ln
j(t)
−
j(t)tm
, (11)
tion and condensation of water as well as the transport of water αzF j0,c σm (x, T, λm (t))
vapor through the membrane are taken into account. Below only
the mass balance equations for the species on the cathode side where Uoc denotes the open-circuit voltage, which is calculated
are given. The corresponding equations for the anode follow by by the Nernst-equation given in Appendix A. α is the symme-
analogy. The mass balance equation of oxygen in REV2 on the try factor and z is the number of electrons exchanged in the
cathode side is reaction. j0,c is the cathodic exchange current density, tm is the
thickness of the membrane and σm is the membrane conductiv-
dy wq I(x, t)
∂t [cc,O2 (x, t)Ac ] = −∂x [Nc,O2 (x, t)] − , (7) ity. The membrane conductivity σm depends on the membrane
4F water content λm . To account for this, λm is calculated for each
where cc,O2 is the concentration and Nc,O2 is the molar flow of time-step. σm is calculated according to Springer et al. [34]:
oxygen in REV2 . The concentration of oxygen changes due to   
1 1
convective transport along the channel and due to consumption σm =(0.00514λm −0.00326) exp 1268 − . (12)
of oxygen in the electrochemical reaction. The mass balance 303 Ts
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 315

3.4. Membrane model Integration under use of the mean value theorem leads to

The membrane water content λm is defined as Aq Leff Ci dt (cq,i Tq )
i
nm,H2 O 
λm = , (13) out,rec rec in,rec rec
nm,SO3 =− Ci (Nq,i Tq,out −Nq,i Tq,in )+Leff Aq,sg Usg (Ts −Tq )
i
where nm,H2 O is the number of water molecules and nm,SO3 is
the number of SO3 -groups in the membrane. The dependency −Leff Hvap Nq,phase . (17)
of λm on the water vapor activity aH2 Ov is modeled according to
[34]: Rearranging Eq. (17) results in

−1
(0.043+17.8aH2 Ov −39.85aH2 Ov +36aH2 Ov ) for aH2 Ov ≤1
2 3
 
λm = dt Tq = Ci cq,i · −Tq Ci cq,i
(14 + 1.4(aH2 Ov − 1)) for aH2 Ov >1.
i i
(14)
1  out,rec rec in,rec rec
The net water migration coefficient is given by [34]: − Ci (Nq,i Tq,out − Nq,i Tq,in )
Aq Leff
i
F cc,H2 Ov − ca,H2 Ov 
αnet = ndrag − Dm,H2 O Aq,sg Usg Hvap
j tm + (Ts − Tq ) + Nq,phase . (18)
Aq Aq
cc,H2 Ov + ca,H2 Ov km,p F pc,H2 Ov − pa,H2 Ov
− , (15)
2 μ H2 Ol j tm Integrating Eq. (7) along the channel length yields
 Leff  Leff
where ndrag is the electro-osmotic drag coefficient. Dm,H2 O is Ac ∂t [cc,O2 (x, t)] dx = − ∂x [Nc,O2 (x, t)] dx
the diffusion coefficient of water in the membrane, km,p is the 0 0
permeability and μH2 Ol the viscosity of liquid water. Hence, the  Leff
dy wq
transport of water through the membrane is given by electro- − j(x, t) dx. (19)
4F 0
osmotic drag and diffusion due to a concentration and pressure
gradient across the membrane. Using the mean value theorem integration leads to
1 dy wq j(t)
4. Discretization dt cc,O2 (t) = [N in,rec (t) − Nc,O
out,rec
(t)] − . (20)
Ac Leff c,O2 2 Ac 4F
The stack model presented here simulates the most important Thus, balance equations are derived for the REC which are
parameters for the operation of a PEMFC stack, which are illus- reduced to net flows at the channel inlet and outlet. Under the
trated in Fig. 1. To derive appropriate and rapidly computable assumption that all gas channels in a cell behave like the REC
equations, the system of partial differential equations for the net flows at the cell inlet and outlet are calculated through mul-
REVs described in Section 3 is integrated along the effective tiplication of the channel net flows with the number of channels
channel length Leff . This corresponds to the second step of the in each electrode. Integration of Eq. (6) directly yields a net
model derivation as described in Section 2. The model equations energy balance equation for the solid material of one complete
after step one of the model derivation are formulated along the cell in the stack. Hence, a system of equations is derived, which
channel, that is, in the x-direction. This corresponds to assum- describes a complete cell module.
ing straight channels. Thus, Leff is the geometrical length of In step three of the model derivation the stack is modeled as
the channel measured along the channel, even in a serpentine several coupled cell modules. To preserve computational effi-
flow-field. With the aid of the Gaussian law and the mean value ciency we assume that each cell in the stack works identically
theorem, ordinary differential equations (ODEs) in time were and that the fuel is distributed equally among the cells. Fig. 4
derived which are discretized to the inlet and outlet of the gas gives an overview of the input and solution variables of the model
channels or the cell’s solid material, respectively. In the follow- and illustrates how the variables are adapted from cell to stack
ing the approach is exemplified for Eqs. (5) and (7). Integration level.
of Eq. (5) along the channel length yields
  Leff
5. Numerical solution method
Ci Aq ∂t (cq,i (x, t)Tq (x, t)) dx
i 0
 The discretized model equations form a linear-implicit sys-
 Leff
tem of differential-algebraic-equations (DAEs), which can be
= − Ci ∂x (Nq,i (x, t)Tq (x, t)) dx
0 written as
i
 Leff M(y, t)ẏ = f (y, t) (21)
+Aq,sg Usg (Ts (x, t)
0 with a singular mass matrix M(y, t). The system is of index 1,
 Leff
that is, one derivation is necessary to transform the DAE-system
−Tq (x, t)) dx − Hvap Nq,phase (x, t) dx. (16)
0 into a system of ODEs. Numerically solving a DAE-system is
316 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

parameters need to be adjusted accordingly. The stack used

for model validation has a serpentine flow-field. As the model
assumes straight channels the effective channel length Leff cor-
responds to the geometrical channel length measured along the
channel. The heat exchange area between solid and gas Aq,sg
equals the surface of one channel in flow-field q standardized
to one unit length. The heat exchange area between solid and
surroundings Ass takes into account the surface of the cells, the
endplates and the cooling fins. The surface area is divided by the
number of cells and standardized to one unit length. The scaling
coefficient dy takes into account the enlargement of the contact
area between gas and membrane due to diffusion under the land
Fig. 4. Overview of the boundary conditions and the solution variables of the areas of the flow-field. For a well-designed flow-field it can be
model. The conversion of the extensive variables from the stack scale to the cell assumed that the whole active area is supplied with gases. Thus,
scale is done by division through the number of cells ncells or by multiplication dy equals the ratio of active area and direct contact area between
with ncells , respectively. gas channels and GDL.
The physical and reaction parameters are determined by mea-
surements and literature research as indicated in Table 1. The
generally more complex than solving ODEs, as DAE-systems
cathodic exchange current density j0,c , the symmetry factor α
require consistent initial conditions. The model was imple-
and the losses in open-circuit voltage ηoc are identified through
mented in MATLABTM using the built-in solver ode15s of the
least-square-fits of Eq. (11) to IV-curves. The heat transfer coef-
ODE Solver environment which demonstrated good stability.
ficient between solid material and surroundings Uss is identified
Ode15s is a variable-order solver based on the numerical dif-
on the basis of cooling curves monitoring the average stack tem-
ferentiation formulas capable of solving stiff DAE-systems of
perature Ts,av in a setting without gas and current flow. In this
index 1 [40].
case the following equation can be deduced from Eq. (6):
6. Parameter identification dTs,av Uss Ass
= (Tsur (t) − Ts,av (t)). (22)
dt ρs Cs As
The mathematical model described contains a high num- Integration leads to
ber of constants and parameters, which can be grouped  
into stack-dependent geometrical, electrochemical and physi- Ass Uss
Ts,av (t) = (Ts,av (t0 ) − Tsur (t0 )) exp − t + Tsur . (23)
cal parameters, as well as stack-independent physical constants. ρs Cs As
Finding suitable values for these parameters is one of the crucial The parameter Uss is determined through least-square-fits of Eq.
points for the success of the model. The identification methods (23) to measured cooling curves.
exemplified for the stack shown in Fig. 5 are easily transferable
to different stack designs. 7. Experiment
Most of the geometrical stack and flow-field parameters can
be taken from direct measurements as illustrated in Fig. 2. How- In order to identify the dynamic behavior of a portable PEM
ever, due to the simplified model structure several geometrical fuel cell stack an experimental investigation was carried out.
The portable stack used for the measurements was developed
at the Fraunhofer ISE. It consists of six cells with an active
cell area of 30.2 cm2 each. A Gore Primea 5510 MEA with a
platinum loading of 0.4 mg cm−2 is used. Cooling fins on the
outer cells allow operation with passive cooling only. Due to
its small geometrical dimensions the stack shows a pronounced
dynamic behavior with respect to stack temperature and water
management. Therefore, it is very well suited for a validation
study of the dynamic stack model. In addition, the quick response
of the stack to changes in the operating conditions allows short,
controlled operation under critical conditions such as dry-out
or flooding without damaging the stack. The particular stack is
used for the implementation of NMPC in an ongoing project.
Results of this work will be published elsewhere.
The measurements were carried out at a computer-controlled
test stand, which controls the operating conditions of the stack as
well as the data acquisition. Gas flow controllers manage the flow
Fig. 5. Fuel cell stack used for parameter identification and model validation. of hydrogen into the anode and oxygen or air into the cathode.
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 317

An external consumer is simulated by a galvanostat, which can

subject the stack to arbitrary current profiles. The gas inlet flows
as well as the voltages of each individual cell and of the whole
stack are measured. The stack impedance is measured at 1 kHz.
The stack temperature was measured with four thermocouples
at the external surface of the stack as usually done in integrated
fuel cell systems. As indicated in Fig. 1 the thermocouples were
placed on the long and short sides of the top and of one of
the middle cells in order to obtain a reasonable average value.
It is assumed that due to the high thermal conductivity of the
bipolar plates the temperature distribution throughout the stack
balances quickly. This assumption is supported by the quick
reaction of the measured stack temperature to changes in the
load. In addition, a comparison of the measured temperatures of Fig. 6. Comparison between simulated and measured stack temperature for a
the four thermocouples showed a maximum deviation of only current step profile. Simulation and experiment show good correlation. The break
on step 4 is caused by the phase change enthalpy of water.
2 K.
The stack was subjected to different load profiles, e.g. step
and jump profiles during which it was supplied with pure hydro- cooling of the stack is dominated by heat exchange between
gen on the anode side and dry air on the cathode side. For solid material and surroundings (term (d) in Eq. (6)), whereas
each measurement the gas flow was kept at a constant rate, only a small amount of heat is dissipated by the exhaust gases.
which corresponds to a stoichiometry of 2 for the highest cur- It is noticeable that current changes affect the stack temper-
rent in the profile. The measurements were carried out under ature almost without time-delay, even though the temperature
ambient pressure. The temperature of the surroundings was was measured at the outside of the stack as indicated in Fig. 1.
Tsur = (298 ± 1) K throughout the experiment. Each profile was This is explained by the small thermal mass and the high heat
measured several times to ensure reproducibility. conductivity of the solid stack material, hence validating the
assumption of a homogeneous stack temperature in the y- and
8. Results and discussion z-direction. Although the simulated and measured stack tem-
peratures agree well, there are also slight deviations. It can be
For the validation study several profiles were simulated. Sta- noted, that the deviations are mostly due to a slower increase
ble numerical convergence behavior was observed in all cases. and a faster decrease of the simulated stack temperature at low
The parameters used for the simulations are listed in Table 1. In current (Fig. 6, steps 1 and 2; Fig. 7, steps with odd numbers),
the following, a step and a jump profile are discussed. For the whereas at high current the slopes agree well. As the water pro-
step profile stack currents of I = 1, 3, 6 A were applied. The duction and therefore the degree of humidification of membrane
time-spans for each step were 3 min on the ascent and 6 min on and electrode is correlated with the current density, it can be
the descent. For the later a longer interval was chosen to fur- concluded that a humidification-dependent heat source is not
ther investigate the cooling of the stack. The flow rates at the modeled comprehensively. This can be the heat dissipation due
gas inlets were set to 0.7 l min−1 for hydrogen and 1.4 l min−1 to insufficient humidification of the membrane–electrode assem-
for air. For the jump profile the current was alternated between bly as well as inhomogeneous current distributions and losses
1 A and 5 A with a hydrogen flow rate of 0.6 l min−1 and an air throughout the stack. The break on step 4 in Fig. 6 is caused
flow rate of 1.2 l min−1 . The simulations were carried out on an
AMD Athlon 1533 MHz. Short computing times of 0.6 s for the
step profile with a time of operation of 23 min and 0.8 s for the
jump profile with a time of operation of 30 min are proofs of the
numerical efficiency of the model.

8.1. Model validation

Validation of the model is performed by comparison of mea-

sured stack temperatures and voltages to the model predictions.
Figs. 6 and 7 show a comparison between measured and sim-
ulated stack temperature for the step and the jump profile,
respectively. The measured value is an average of the measure-
ments of the four thermocouples. Simulation and experiment
show good correlation. At high current the stack temperature
Fig. 7. Comparison between simulated and measured stack temperature for a
increases due to the different loss mechanisms corresponding current jump profile. Simulation and experiment agree well. Deviations can be
to term (b) in Eq. (6). Parametric studies showed that activa- explained by heat dissipation due to insufficient humidification of the electrode
tion losses at the cathode are the dominant loss mechanism. The and inhomogeneous current density distribution.
318 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

Fig. 10. Simulation results for the step profile indicate a high concentration of
liquid water on the cathode side for steps 2–4. This coincides with low values of
Fig. 8. Comparison between simulated and measured stack voltage for a current simulated and measured impedance on these steps. Observation of the average
step profile. A good correlation is obtained. Deviations between model predic- concentration of liquid water and the stack impedance can be used to predict
tions and experiment are connected with critical operating conditions of the critical operating conditions like flooding or dry-out.
stack.

by phase change enthalpy of water. In the simulation all liquid convergence behavior was also achieved for the other load pro-
water has evaporated at this point as shown in Fig. 11. Heat con- files in the validation study. Thus, the model predictions for the
sumption due to evaporation stops, which leads to an increase of stack temperature and the voltage are validated.
the stack temperature. In the model this is reflected by a change
of sign of term (a) in Eq. (6) (Nq,phase is negative, when evapo-
8.2. Water balance and critical operating conditions
ration takes place and positive elsewise). In reality a remainder
of water will longer exist in the GDL or the membrane leading
Water management plays an important role for the opera-
to a milder change-over between the two regimes. A detailed
tion of a PEMFC. Low concentrations of liquid water lead to a
discussion on water balance is given below. Similar breaks indi-
decrease in membrane conductivity whereas an accumulation of
cating the switch between the two regimes are visible on steps
liquid water can block the gas flow due to flooding of electrode
5, 7 and 9 in Fig. 7. As the temperature level increases with the
and gas channels. The stack model presented here incorporates
step number the amount of accumulated water throughout the
mass balances of water vapor and liquid water (Eqs. (8) and (9)),
preceding high current step gets smaller and evaporation gets
which are linked through a phase change term (Eq. (10)). To
faster. Therefore, the breaks on higher numbered steps occur
allow prediction of critical operating conditions in an integrated
closer to the preceding load change. As the stack was dry at
fuel cell system, the stack model predicts the average concen-
the beginning of the profile no breaks occur on steps 1 and 3.
trations of liquid water at the anode and cathode side and the
The validity of this interpretation is shown after the discussion
impedance of the stack. Fig. 10 shows corresponding simulation
of water balance below. A comparison between measured and
results and experimental values of the stack impedance for the
simulated stack voltage is shown in Figs. 8 and 9. Simulated and
step profile. Simulation results indicate a high concentration of
measured stack voltage agree well. Deviations between model
liquid water at the cathode side for steps 2–4. This coincides with
predictions and experiment are connected with the water bal-
low values of simulated and measured impedance on steps 2–4.
ance of the stack as discussed below and in Fig. 12. A satisfying
As the membrane model is steady-state, the simulated values
change rapidly upon a load change, whereas the experimen-
tal results indicate slower variations of membrane conductivity.
However, the simulated impedance reflects the trend of the mea-
surement well. In order to cover the dynamic behavior of the
water impregnated into the membrane, a dynamic membrane
model is necessary. Yet, models that describe the dynamic water
transport through the membrane accurately are usually com-
putationally expensive, see e.g. Ref. [35]. The development of
a computationally efficient, dynamic membrane model is very
challenging, but is seen as one of the key points to improve the
predictive capability of dynamic PEMFC models. To illustrate
the evolution of the average concentration of liquid water in the
cathode, Fig. 11 shows the relation between phase transition of
water, its partial pressure and the saturation pressure for the step
Fig. 9. Comparison between simulated and measured stack voltage for a current profile. Water condenses when pH2 Ov > Pc, sat and liquid water
jump profile. Measurement and experiment agree well. The model parameters evaporates when pH2 Ov < Pc, sat (Eq. (10)). It is assumed that
remain unchanged for each case of the validation study. the stack does not contain liquid water at the beginning of the
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 319

and electrode take up water. The opposite effect is observed dur-

ing steps 4 and 5. The simulated average concentration of liquid
water in Fig. 10 as well as the impedance indicate a low concen-
tration of water in the stack. In step 3 the stack voltage shows
flooding effects, which are not captured by the model. How-
ever, the simulated concentration of liquid water indicates the
breakdown of the stack voltage due to flooding. Thus, observ-
ing the characteristics of the voltage allows a conclusion on the
water balance of the stack. This also supports the validity of
the interpretation of Fig. 7. During step 1 in Fig. 9 the voltage
increases rapidly indicating a low, but increasing degree of mem-
brane and electrode humidification, whereas throughout steps 3,
5, 7 and 9 the voltage decreases due to drying-out of the mem-
Fig. 11. Simulation results visualizing the relation between phase transition of brane. For higher numbered steps the decrease becomes much
water, its partial pressure pc (H2 Ov ) and the saturation pressure Pc, sat of water faster. Although the model does not capture the voltage decrease
vapor on the cathode for the step profile. Condensation occurs between point a quantitatively, the position of the breaks in the simulated stack
and b as pc (H2 Ov ) is higher than Pc, sat . Liquid water evaporates between points
b and c as pc (H2 Ov ) < Pc, sat holds.
temperature in Fig. 7 shows that the regimes of high and low
water concentration are modeled appropriately.
As Fig. 12 indicates the predictive capability of the model
measurement. Condensation occurs during steps 2 and 3 starting
shows limitations when the stack is operated in a high cur-
at point (a), where pH2 Ov is higher than Pc, sat due to the high
rent range. This is due to the negligence of the GDL and the
production of water in the electrochemical reaction. This coin-
simple model approach used for the membrane. Therefore, the
cides with an increasing concentration of liquid water in these
highly complex two-phase effects in the membrane and the GDL
steps in Fig. 10. Condensation ends as pH2 Ov decreases due to
are not covered. To develop a computationally efficient model,
lower water production after the load reduction (b). Existing liq-
that includes the corresponding phenomena goes far beyond the
uid water evaporates in the course of several minutes (c). After
scope of the present paper. In addition, the predictive capability
this point due to the higher temperatures and the low load water
of the model might show limitations for PEMFC stacks with a
exists in vapor form only and condensation or evaporation does
strong spatial inhomogeneity of temperature or electric poten-
not take place.
tial. This is due to the fact that the gradients in temperature and
Predicting the impedance and the average concentration of
electric potential across the stack need to be small to ensure the
liquid water, the model allows the early detection of critical oper-
validity of the model approach.
ating conditions such as flooding or dry-out. However, the model
does not yet predict the corresponding quantitative decrease in
9. Conclusions
the stack voltage as the two-phase effects require detailed mod-
els on the micro-scale, which are not agreeable to the demand
In this work, a numerically efficient model was developed to
of limited computational effort. Fig. 12 illustrates the effects of
study the dynamic response of a PEMFC stack subjected to load
the water balance on the stack voltage. The measured voltage
changes. The attribute of numerical efficiency was imposed in
curve shows three characteristics: humidification, flooding and
order for the model to be applicable for NMPC. Therefore, a rea-
dry-out. At step 2 the stack voltage increases as the membrane
sonable compromise was found between computational effort
and physical accuracy. Mainly three model properties make
our calculations computationally efficient: firstly, only one cell
is simulated and the results are then upscaled to account for
the whole stack. Secondly, the spatial dependency of the fuel
cell parameters is neglected as the energy and mass balance
equations on which the model is based are reduced to ordi-
nary differential equations in time. And thirdly, the GDL is not
modeled and the membrane is assumed to be at steady-state as
a detailed modeling of these cell parts would require massive
computational effort. As a result of this approach the model
proves to be computationally efficient: an operating time of 0.5
h can be simulated in less than 1 s. Despite the relatively simple
model approach, the model and the calculation method could be
validated against experimental data for different realistic load
Fig. 12. Comparison of measured and simulated voltage for the current step
profiles through a comparison of measured and simulated stack
profile indicating the effects of the water balance. The measured voltage curve
shows three characteristics: humidification, flooding and dry-out. The simulated voltage and stack temperature on time scales of up to 30 min.
concentration of liquid water in Fig. 10 indicates the breakdown of the stack The model was used to study the run of the stack voltage and the
voltage due to flooding. stack temperature curves. Parametric studies showed that activa-
320 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321

tion losses at the cathode cause the main mechanism of heating, Appendix A
whereas the heat exchange between solid and surroundings is
the main mechanism for cooling. Slight deviations of the slope A.1. Supporting equations
of the simulated stack temperature at low current indicate that
a humidification-dependent heat source is not modeled com- The internal energy Uk,i of species i in volume element k is
prehensively, which can be heat dissipation due to insufficient given by
humidification of the membrane–electrode assembly as well as
inhomogeneous current distributions throughout the stack. Fur- Uk,i = ck,i Ci Tk . (24)
thermore, it is mentionable that current changes affect the stack The open-circuit voltage Uoc is calculated using the Nernst-
temperature almost without time-delay, although the tempera- equation:
ture was measured at the outside of the stack. This is explained  
by the small thermal mass and the high heat conductivity of the Smol RT  pi
Uoc = Uoc0
+ (T − T 0 ) − νi ln 0
, (25)
solid material. 2F 2F p i
i
In order to ensure reliable operation of a PEMFC under load
where Uoc 0 is the open-circuit voltage at reference conditions,
changes it is essential to prevent critical operating conditions of
flooding and dry-out, which depend on the highly complex phe- Smol is the molar entropy of the overall reaction and T 0 is the
nomena of water balance of the stack. Therefore, we put special reference temperature. νi denotes the stoichiometry factor, pi is
emphasis on the study of possibilities and limitations of the water the partial pressure and p0i is the partial pressure of species i at
balance description with respect to critical operating conditions reference conditions. The partial pressure pi of the gas species
within such a reduced model. Our model incorporates mass bal- i is calculated using Dalton’s law:
ance equations for water vapor and liquid water including a term Nq,i
for the phase change. For the purpose of studying the capabil- pi =  Pq , (26)
Nq,j
ity of this approach the stack was operated in critical states of
j
flooding and dry-out. Through a comparison of measured and
simulated stack voltage three characteristics of the stack voltage where Pq is the average pressure in gas channel q. Assuming that
indicating humidification, flooding and dry-out of the stack are the gases in the channels can be treated as ideal, the following
identified. Thereby indicating how the observation of the stack equation is used to calculate the concentration cq,i from the molar
voltage allows a conclusion on the water balance of the stack. flow Nq,i :
The model was successfully applied to predict the regimes of
Nq,i
high and low water concentrations in the stack. It is shown, that cq,i = , (27)
by monitoring the simulated stack impedance and the average Aq vq,i
concentration of liquid water in the stack critical states of oper- where vq,i is the velocity of species i in electrode q.
ation can be predicted. However, the model does not yet capture The electro-osmotic drag coefficient ndrag is calculated
the corresponding quantitative change of the stack voltage as according to Springer et al. [34]:
the underlying two-phase effects require detailed descriptions
on the micro-scale which are not compatible with the demand ndrag = (5/44)λm . (28)
for limited computing time.
The following empirical equation is used to calculate the satu-
With our model we have demonstrated that the dynamic
ration pressure of water vapor [34]:
response of a PEMFC stack under load changes can be sim-
Pq,sat =cP1 x10(cp2 +cp3 (Tq −273)+cP4 (Tq −273)
2 +c (T −273)3 )
ulated in short computing times with sufficient accuracy. Efforts P5 q . (29)
are currently underway to implement NMPC algorithms based
on this model. In order to further improve the predictive capabil- with cP1 = 101, 325, cP2 = −2.18, cP3 = 2.95 × 10−2 , cP4
ities of the model future work could include the development of = −9.18 × 10−5 and cP5 = 1.44 × 10−7 . The diffusion coef-
a computationally efficient, dynamic membrane model as well ficient of water in the membrane Dm,H2 O as a function of λm
as the integration of a GDL model to improve simulation results and Ts was described by Golbert and Lewin [36] as
when the stack is operated in a high current range.   
1 1
Dm,H2 O = ndrag Dm,H
ref
2O
exp cD1 − (30)
Acknowledgements Tσ,ref Ts
with the empirical values cD1 = 2416 and Tσ,ref = 303 K [36].
The authors would like to thank J.O. Schumacher for his valu- ref
Dm,H is the diffusion coefficient of water in the membrane at
2O
able contribution to the work presented here. Part of the author’s reference conditions.
research was funded by the German Federal Ministry of Educa- The activity of water in channel q is calculated according to
tion and Research (BMBF) via the project ’Model based design Golbert and Lewin [36]:
of fuel cells and fuel cell systems: PEM-Design (03SF0310A)’.
Nq,H Ov Pq
Part of the work took also benefit from the project ’Optimal a H2 Ov =  2 . (31)
control of small fuel cells (MBZ6)’ commissioned by the Lan- Nq,i Pq,sat
desstiftung Baden-Wuerttemberg foundation. i
 S.P. Philipps, C. Ziegler / Journal of Power Sources 180 (2008) 309–321 321

The reaction entropy of the overall reaction Smol for the con- [18] S. Shimpalee, W.-K. Lee, J. Van Zee, H. Naseri-Neshat, J. Power Sources
sumption of 1 mol H2 is calculated from the reaction entropy at 156 (2006) 355–368.
[19] S. Shimpalee, W.-K. Lee, J. Van Zee, H. Naseri-Neshat, J. Power Sources
the anode Sa and the cathode Sc :
156 (2006) 369–374.
Sc [20] Y. Wang, C.-Y. Wang, Electrochim. Acta 51 (2006) 3924–3933.
Smol = Sa + . (32) [21] H. Guilin, F. Jianren, J. Power Sources 165 (2007) 171–184.
2
[22] C. Ziegler, H. Yu, J. Schumacher, J. Electrochem. Soc. 152 (2005)
A1555–A1567.
References [23] H. Wu, X. Li, P. Berg, Int. J. Hydrogen Energy 32 (2007) 2022–2031.
[24] A. Shah, G.-S. Kim, P. Sui, D. Harvey, J. Power Sources 163 (2007)
793–806.
[1] Q. Yan, H. Toghiani, H. Causey, J. Power Sources 161 (2006) 492–502.
[25] A. Shah, P. Sui, G.-S. Kim, S. Ye, J. Power Sources 166 (2007) 1–21.
[2] F. Allgöwer, T. Bodgwell, J. Qin, J. Rawlings, S. Wright, in: P. Frank (Ed.),
[26] T. Berning, N. Djilali, J. Electrochem. Soc. 150 (2003) A1589–A1598.
Advances in Control, Highlights of ECC’99, Springer, London, 1999, pp.
[27] Y. Wang, C.-Y. Wang, J. Electrochem. Soc. 153 (2006) A1193–A1200.
391–449.
[28] T. Nguyen, R. White, J. Electrochem. Soc. 140 (1993) 2178–2186.
[3] K. Yao, K. Karan, K. McAuley, P. Oosthuizen, B. Peppley, T. Xie, Fuel
[29] K. Dannenberg, P. Ekdunge, G. Lindbergh, J. Appl. Electrochem. 30 (2000)
Cells 4 (2004) 3–29.
1377–1387.
[4] C.-Y. Wang, Chem. Rev. 104 (2004) 4727–4766.
[30] S.-H. Ge, B.-L. Yi, J. Power Sources 124 (2003) 1–11.
[5] J. Yi, T. Nguyen, J. Electrochem. Soc. 145 (1998) 1149–1159.
[31] P. Nguyen, T. Berning, N. Djilali, J. Power Sources 130 (2004) 149–157.
[6] T. Fuller, J. Newman, J. Electrochem. Soc. 140 (1993) 1218–1225.
[32] M. Hu, A. Gu, M. Wang, X. Zhu, L. Yu, Energy Convers. Manage. 45
[7] T. Springer, M. Wilson, S. Gottesfeld, J. Electrochem. Soc 140 (1993)
(2004) 1861–1882.
3513–3526.
[33] S. Maharudrayya, S. Jayanti, A. Deshpande, J. Power Sources 138 (2004)
[8] J. Amphlett, R. Mann, B. Peppley, P. Roberge, A. Rodrigues, J. Power
1–13.
Sources 61 (1996) 183–188.
[34] T. Springer, T. Zawodzinski, S. Gottesfeld, J. Electrochem. Soc 138 (1991)
[9] J. Lee, T. Lalk, J. Power Sources 73 (1998) 229–241.
2334–2341.
[10] M. Ceraolo, C. Miulli, A. Pozio, J. Power Sources 113 (2003) 131–144.
[35] A. Weber, J. Newman, J. Electrochem. Soc. 151 (2004) A326–A339.
[11] Y. Shan, S.-Y. Choe, J. Power Sources 145 (2005) 30–39.
[36] J. Golbert, D. Lewin, J. Power Sources 135 (2004) 135–151.
[12] J. Golbert, D. Lewin, J. Power Sources 135 (2004) 135–151.
[37] P. Atkins, Physical Chemistry, Oxford University Press, Oxford, 1994.
[13] X. Yu, B. Zhou, A. Sobiesiak, J. Power Sources 147 (2005) 184–195.
[38] R. Bird, W. Stewart, E. Lightfoot, Transport Phenomena, 2nd ed., Wiley,
[14] P. Pathapati, X. Xue, J. Tang, Renew. Energy 30 (2005) 1–22.
New York, 2001.
[15] D. McKay, W. Ott, A. Stefanopoulou, Proceedings of IMECE, 2005, pp.
[39] M. Lampinen, M. Fomino, J. Electrochem. Soc. 140 (1993) 3537–3546.
1–10.
[40] L.F. Shampine, M. Reichelt, J. Kierzenka, SIAM Review 41 (1999)
[16] S. Um, C.-Y. Wang, K. Chen, J. Electrochem. Soc. 147 (2000) 4485–4493.
538–552.
[17] Y. Wang, C.-Y. Wang, Electrochim. Acta 50 (2005) 1307–1315.