Journal of Power Sources 299 (2015) 202e211

Contents lists available at ScienceDirect

Journal of Power Sources

journal homepage: www.elsevier.com/locate/jpowsour

A transient electrochemical model incorporating the Donnan effect for

all-vanadium redox flow batteries
Y. Lei a, B.W. Zhang a, B.F. Bai a, T.S. Zhao b, *
a
State Key Laboratory of Multiphase Flow in Power Engineering, Xi'an Jiaotong University, Xi'an 710049, China
b
Department of Mechanical and Aerospace Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong, China

h i g h l i g h t s

A transient model with incorporation of the Donnan effect is developed in VRFB.

The model allows the exhibition of membrane selective adsorption.

The model enables an accurate prediction of the battery performance.

Effect of the fixed charge concentration on species crossover is studied.

Donnan effect is found to deteriorate vanadium crossover in Nafion membranes.

a r t i c l e i n f o a b s t r a c t

Article history: In a typical all-vanadium redox flow battery (VRFB), the ion exchange membrane is directly exposed in
Received 29 June 2015 the bulk electrolyte. Consequently, the Donnan effect occurs at the membrane/electrolyte (M/E) in-
Received in revised form terfaces, which is critical for modeling of ion transport through the membrane and the prediction of cell
10 August 2015
performance. However, unrealistic assumptions in previous VRFB models, such as electroneutrality and
Accepted 29 August 2015
Available online 10 September 2015
discontinuities of ionic potential and ion concentrations at the M/E interfaces, lead to simulated results
inconsistent with the theoretical analysis of ion adsorption in the membrane. To address this issue, this
work proposes a continuous-Donnan effect-model using the Poisson equation coupled with the Nernst
Keywords:
Flow battery
ePlanck equation to describe variable distributions at the M/E interfaces. A one-dimensional transient
Vanadium crossover VRFB model incorporating the Donnan effect is developed. It is demonstrated that the present model
Donnan effect enables (i) a more realistic simulation of continuous distributions of ion concentrations and ionic po-
Numerical modeling tential throughout the membrane and (ii) a more comprehensive estimation for the effect of the fixed
charge concentration on species crossover across the membrane and cell performance.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction
All-vanadium redox flow battery (VRFB) is a large-scale energy
storage device for intermittent renewable energy sources. With the
advantages of a long cycle life, capacity independence with power,
quick response and high security, this type of battery has received
increasing focus [1e3]. Although the VRFB possesses commercial
promise, a key limiting factor in the form of ion crossover via the
membrane must be resolved before development. This limitation
negatively affects both the performance and the lifetime of the
VRFB. Much effort has recently been devoted to prevention of ion
* Corresponding author.
E-mail address: metzhao@ust.hk (T.S. Zhao).
crossover. The study of novel materials for the ion exchange
membrane aims to bring forth merits of lowered cost, high stability,
low vanadium permeability and high proton conductivity [4e10].
Chen et al. synthesized a series of membranes based on sulfonated
poly with various additives [7e10], which feature lowered costs but
make little improvement to the overall quality. Other efforts have
attempted to design experiments to determine the transport
properties for interpretation of the transport behaviors of ions
through the ion exchange membrane [11e13].
Aside from experimental investigations, numerical simulation is
useful in analyzing the transport mechanism of ions through the
membrane and the optimization of cell performance [14e24]. So
far, a mathematical model incorporating a complete description of
species transport through the membrane for VRFBs has been
established, in which the NernstePlanck equation containing the

http://dx.doi.org/10.1016/j.jpowsour.2015.08.100
0378-7753/© 2015 Elsevier B.V. All rights reserved.
 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211 203

diffusion, migration and convection terms is used for all ions.

However, the Donnan effect occurring at the M/E interfaces under
the presence of fixed charge on the membrane is neglected in major
models, which exhibits the ion adsorption and repulsion of an ion
exchange membrane. From adsorption experiments with a cation
exchange membrane (CEM) exposed in the vanadium solution,
cation concentrations in the membrane could be dramatically
higher than ones in the solution [11]. Thus, the Donnan effect must
be incorporated in numerical models. To date, only two published
models have ever accounted for the Donnan effect. Knehr et al. [16]
treated the Donnan effect with an unrealistic discontinuity of
HSO4  concentration at the M/E interfaces to maintain electro-
neutrality, leading to simulated results inconsistent with the trends
of membrane selective adsorption. Yang et al. [21] captured the
Donnan potential and concentration jumps at the M/E interfaces
without specific mathematical descriptions or detailed illumination
on the mechanisms of this behavior, and estimated anion concen-
trations in membrane from their data did not exhibit anion repul-
sion in a CEM. In fact, a highly localized space charge region, where
the electroneutrality is not satisfied, occurs at the M/E interfaces
when the fixed charge on the membrane is exposed in the elec-
trolytes [25,26]. As the charge density in the space charge region is
composed of all ion charges including the fixed charge if any, the
Donnan effect, or in other words, the dramatic differences of the
ionic potential and ion concentrations across the M/E interfacial
region, possesses a complex relationship with the characteristics of
the space charge region. Accordingly, neglecting the space charge Fig. 1. (a) Schematic of a VRFB system; (b) modeling domain.
region lowers the accuracy of species crossover and cell perfor-
mance predicted in VRFB models. Thus, a detailed description of the
space charge region accounting for the complex correlation be- Charge
tween ionic potential and ion concentrations is required to shed Negative electrode : V3þ þe






!
V2þ





(2)
Discharge
light into the mechanism of the Donnan effect and species cross-
over through an ion exchange membrane. As the axis x in Fig. 1a is the dominate direction of the current
In this paper, a continuous-Donnan effect-model is proposed by density composed of all ion fluxes in electrolytes, the variations of
using the Poisson equation coupled with the NernstePlanck ion concentrations, electric and ionic potentials are able to be
equation to describe the space charge region [25e27]. The Poisson neglected in the other two axes. Therefore, the present model is
equation is used to characterize the correlation between ionic po- simplified to be one dimension through the membrane. Further-
tential distribution and space charge density at the M/E interfacial more, the lump model is utilized to describe the variables in elec-
regions, which is suitable in dilute solution theory. In addition, the trodes, since the variations of ion concentrations here are much
NernstePlanck equation is used to describe the transport processes smaller than those in the membrane [18]. As shown in Fig. 1b, the
of all ions in the solution. Consequently, continuous variations of model is one-dimensional, where the two electrodes are zero-
ionic potential and ion concentrations at the M/E interfacial regions dimensional. Each interfacial region contains a space charge re-
are obtained. A one-dimensional transient VRFB model incorpo- gion described in introduction with the width set to be 2 nm, which
rating the Donnan effect is then developed. In addition, with the will be verified in Section 3.1.2. The membrane thickness Lm is
Donnan effect considered, the effect of fixed charge concentration 183 mm. Other assumptions are as follows:
on ion distributions, species crossover and cell performance are
obtained and analyzed for CEMs. In the next section, the detail 1. The battery system is isothermal;
mathematical model is presented, followed by the analysis and 2. The model satisfies the dilute solution approximation;
conclusion. 3. The side reactions, gravity, water transport, as well as the pump
power are ignored;
4. Sulfuric acid is dissociated to be Hþ and HSO4  ;
2. Model description 5. The positions where self-reactions occur are assumed at x ¼ x2
and x ¼ x3;
The schematic of a typical VRFB system is shown in Fig. 1a. The 6. The ion diffusivities in the membrane are set to be constant with
system consists of two electrolyte tanks, two pumps and a single different fixed charge concentrations, as variations in the porous
cell, which includes an ion-exchange membrane, two graphite felts structure are neglected.
as positive and negative electrodes, as well as two current collec-
tors. As the system is operating, electrolyte in each half-cell circu-
lates between the tank and electrode, in which the redox reactions 2.1. Governing equations
occur simultaneously. The reactions are as follows:
2.1.1. Membrane region
Charge
Positive electrode : VO2þ þH2 O  e






! In the membrane region, the mathematical model is formulated

VOþ





2 þ 2H
þ
(1)
Discharge by the conservation of species and charge, as well as electrically
204 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211

Table 1
Properties and parameters in membrane.

Parameter Value

Membrane thickness, Lm 183e-6 m

Valence of the fixed charge [14], zM 1
Hydraulic permeability [16], kp 1.58e-18 m2
Electrokinetic permeability [16], kf 1.13e-20 m2
Water viscosity, mw 0.001 Pa$s
Dielectric constant, ε 6.947e-10 F$m1
V2þ membrane diffusivity [16], D2,m 3.125e-12 m2$s1
V3þ membrane diffusivity [16], D3,m 5.93e-12 m2$s1
VO2þ membrane diffusivity [16], D4,m 5e-12 m2$s1
VO2 þ membrane diffusivity [16], D5,m 1.17e-12 m2$s1
Hþ membrane diffusivity [16], DH,m 3.35e-9 m2$s1
HSO4  membrane diffusivity [16], DHS,m 4e-11 m2$s1
Fig. 2. Charge schematic at the M/E interfacial region.

neutral equation. The properties and parameters are listed in Fig. 2 shows the charge schematic of the highly localized space
Table 1. charge region at the M/E interfacial region mentioned above.
The species conservation is expressed as: Farther away from the junction between the membrane and elec-
vci . trolyte, the space is observed to be electrically neutral. As a
þ V$N i ¼ 0 (3) consequence, the ionic charge in the bulk electrolyte is zero, while
vt
that in the bulk membrane is equal to ezMcM with the presence of
.
where ci and N i are respectively the concentration and the flux of fixed charge on the membrane, where zM and cM are the valence
the subscript i denotes all ions except HSO4  , and t is time.
species i, . and concentration of the fixed charge, respectively. Under the
The flux N i is given by NernstePlanck equation: assumption that all ion concentrations are continuous, the ionic
charge changes continuously from zero in the electrolyte to ezMcM
. zi ci Di;m . in the membrane. Consequently, with the fixed charge concentra-
Ni ¼ Di;m Vci  FVfl þ ci v (4)
RT tion constant, the distribution of the space charge at the M/E
interfacial region appears to be the diffuse double layer shown in
where Di,m and zi are the membrane diffusivity and the valence of Fig. 2. Here, the Poisson equation is introduced to relate the ionic
species i, respectively, R ¼ 8.314 J/(mol K) and F ¼ 96485 C/mol are potential with the space charge [25e27], which is expressed as:
respectively the universal gas constant and the Faraday constant, T
is the operating temperature, and fl is the ionic potential in the V$ðεVfl Þ ¼ re (9)
.
solution. The velocity v described by an alternate form of Schlogl's
equation [10] satisfies the continuity equation: where ε is the dielectric constant. re is the space charge density,
defined as:
.
V$ v ¼ 0 (5) 8 P
> F zi ci 0  x < x1 ; x4 < x  x5
>
< i
P ! !
. kp kf F zi Di;m Vci re ¼ P (10)
v ¼  Vp  ðzM cM ÞF Vfl þ F 2 P 2 (6) >
> zi ci þ zM cM x1  x  x2 ; x3  x  x4
mw mw z ci Di;m :F
RT i i

where kp and kf are respectively the hydraulic and the electroki- The ionic potential difference across the interfacial region is the
netic permeability, mw and p are the viscosity and the pressure of Donnan potential.
fluid. From equation (6), the velocity is not constant during a
chargeedischarge cycle.
The electroneutrality satisfies: 2.1.3. Self-reactions in VRFBs
Self-reactions occur at the two sides of the membrane region as
X
zi ci þ zM cM ¼ 0 (7) a result of species crossover. When VO2þ and VO2 þ transport
i through the membrane to the negative electrode, they are imme-
diately deoxidized by V2þ to be V3. Similarly, V2þ and V3þ transport
where the subscript i contains all ions. to the positive electrode and are oxidized by VO2 þ to be VO2þ. The
The charge conservation is expressed as: chemical equations are expressed as:
! 
X . V2þ þ VO2þ þ 2Hþ /2V3þ þ H2 O
V$il ¼ V$ F zi N i ¼0 (8) x ¼ x2 ; (11)
2V2þ þ VOþ þ
2 þ 4H /3V
3þ
þ 2H2 O
i
(
where il is the ionic current density and composed of all fluxes. V2þ þ 2VOþ þ 2þ
x ¼ x3 ; 2 þ 2H /3VO þ H2 O
(12)
3þ þ 2þ
V þ VO2 /2VO
2.1.2. Membrane/electrolyte interfacial regions
The species conservation is modified to be
In the interfacial regions, the mathematical model, including
governing equations, properties and parameters, is the same as that vci .
in the membrane, except that the electrically neutral equation is þ V$N i ¼ Si (13)
vt
unsuitable due to the existence of space charge. Note that the
subscript i for species conservation in this section contains all ions. where Si is the reaction source term obtained from Eqs. (11) and
 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211 205

(12). The formula of Si is expressed as: Table 2

Properties and parameters in electrodes and current collectors.
   
8 .  . 
> S2 Dx ¼  N 4   2 N 5  Parameter Value
>
<     Electrode length, Lh 0.02 m
  .  .
x ¼ x2 ; S3 Dx ¼ 2 N 4  þ 3 N 5  (14) Electrode width, Lw 0.025 m
>
>     Electrode thickness, Lt 0.003 m
: .  .  Porosity [16], є 0.93
SH Dx ¼ 2 N 4   4 N 5  Specific surface area [16], a 35,000 m1
V2þ diffusivity [14], D2,e 2.4e-10 m2$s1
    V3þ diffusivity [14], D3,e 2.4e-10 m2$s1
8 .  .  VO2þ diffusivity [14], D4,e 3.9e-10 m2$s1
> S4 Dx ¼ 3 N 2  þ 2 N 3 
>
<    
VO2 þ diffusivity [14], D5,e
Hþ diffusivity [14], DH,e
3.9e-10 m2$s1
9.312e-9 m2$s1
   
. .
x ¼ x3 ; S5 Dx ¼ 2 N 2    N 3  (15) HSO4  diffusivity [16], DHS,e 1.33e-9 m2$s1
>
>   Current collector thickness, Lc 0.01 m
: .  Electronic conductivity of current collector [18], sc 91,000 S$m1
SH Dx ¼ 2 N 2 

where the subscripts 2,3,4,5 and H represent V2þ, V3þ, VO2þ, VO2 þ The ionic potential boundary is derived from the given applied
and Hþ respectively, Dx is the step length. current density and is expressed as
P
IþF zi Di;m Vci
2.2. Boundary conditions and initial values .
x ¼ 0;  n $Vfl ¼  F 2 Pi 2 (21)
RT zi Di;m ci
As the description of two electrodes is treated with the lump i
model, the electrolyte concentrations are boundary conditions for
P
the modeling domain. By neglecting the concentration difference IþF zi Di;m Vci
between the electrode and the tank, the concentrations of all va- . i
x ¼ x5 ;  n $Vfl ¼ P (22)
nadium ions and Hþ are calculated by species conservations, and F2
RT z2i Di;m ci
the HSO4  concentration is obtained from the electrically neutral i

condition: .
where n is the outward normal unit vector.
8 The state of charge SOC is defined to be
>
> dc2 j .
> ðVt þ εVe Þ
> ¼ Ve a  N 2  Lh  Lw
>
> dt F !
>
> cn2 c5 p
>
>
< ðV þ εV Þ dc3 ¼ V a j  . N 3  Lh  Lw SOC ¼ min ; (23)
x ¼ 0;
t e
dt
e
F (16) c2 þ cn3 cp4 þ cp5
n
>
>
>
> dc .
>
> ðVt þ εVe Þ H ¼ N H  Lh  Lw
>
> where the superscripts n and p represent the negative and positive
>
> dt
: electrodes.
2  c2 þ 3  c3 þ cH  cHS ¼ 0 The initial electrolyte concentrations are defined to be
8 8
>
> dc j . >
> c ¼ c0  SOC0
>
> ðVt þ εVe Þ 4 ¼ Ve a þ N 4  Lh  Lw < 2
>
> dt F c3 ¼ c0  ð1  SOC0 Þ
>
> x ¼ 0; t ¼ 0; (24)
>
> >
> c ¼ cD þ c0  SOC0
< ðV þ εV Þ dc5 ¼ V a j þ .N 5  Lh  Lw : H
t e e cHS ¼ 2  c2 þ 3  c3 þ cH
x ¼ x5 ; dt F (17)
>
>
>
> dc j . 8
>
> ðVt þ εVe Þ H ¼ 2Ve a þ N H  Lh  Lw
>
> > c ¼ c0  ð1  SOC0 Þ
>
> dt F >
< 4
: c5 ¼ c0  SOC0
2  c4 þ c5 þ cH  cHS ¼ 0 x ¼ x5 ; t ¼ 0; (25)
>
> c ¼ cD þ c0 þ c0  SOC0
: H
where Vt and Ve are the volume of the tank and electrode. ε, a, Lh
cHS ¼ 2  c4 þ c5 þ cH
and Lw are the porosity, the specific surface area, the height and
where SOC0 ¼ 0.15 is the initial state of charge. c0 ¼ 1000 mol/m3 is
width of the electrode, respectively.
the initial total vanadium concentration in each half-cell.
Local reaction current density j is obtained by:
cD ¼ 1000 mol/m3 is the tuning concentration for Hþ. The initial
I distributions of all variables in the modeling domain are obtained
j¼ (18) from the steady-state model with the above boundary conditions.
aLt
The operating conditions and initial values are listed in Table 3.
where I is the applied current density, Lt is the thickness of elec-
trode. The properties and parameters are listed in Table 2. Table 3
The constant concentration boundaries are Operating conditions and parameters.

Parameter Value
x ¼ x2 ; c4 ¼ 0; c5 ¼ 0
(19) Tank volume, Vt 30 ml
x ¼ x3 ; c2 ¼ 0; c3 ¼ 0
Operating temperature, T 298 K
The pressure boundary condition is set as Flow rate, Q 60 ml min1
Initial state of charge, SOC0 0.15
Initial total vanadium concentration in each electrolyte, c0 1000 mol m3
Dp ¼ pðx ¼ x5 Þ  pðx ¼ 0Þ ¼ 0 (20) Tuning concentration, cD 1000 mol m3
206 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211

2.3. Battery voltage parameters are listed in Table 4.

By balancing the electrochemical reaction rate with the rate of 3. Results and discussion
mass transfer at the electrode surface, concentration csi at the liq-
uidesolid interface in the porous electrode can be achieved from The governing equations were solved using the finite-element
software COMSOL Multiphysics 3.5a. The uniform grid was con-
.s   j
N2 ¼ kmn c2  c2s ¼  structed in both interfacial regions, while the proportional grid was
F utilized to connect the membrane with two interfacial regions. The
(26)
.s   j relative tolerance was set to 1  106.
N 3 ¼ kmn c3  c3s ¼
F
3.1. Model validation
.s   j
N4 ¼ kmp c4  c4s ¼
F 3.1.1. Cell performance
(27)
.s   j The comparison of cell performance between simulation and
N 5 ¼ kmp c5  c5s ¼  experiment is shown in Fig. 3a with all properties and parameters
F
.s set as reference [16]. The numerical result is validated against the
where N i is the flux at the electrode surface. kmp and kmn are the experimental data with a margin of error under 3.6%. The voltage
mass transfer coefficients nearby the positive and negative elec- discrepancy may be caused by neglecting variable distributions in
trode surfaces, expressed as: the electrodes, using estimated and fitted parameters, as well as
 0:4 taking no account of side reactions.
Q
kmn ¼ kmp ¼ 1:6  104 (28)
Lt  Lw 3.1.2. Donnan effect at the M/E interface
The present model is validated against an analytical model with
Relationship of the reaction rate with overpotential is given by
the Donnan effect [25] in Fig. 3b. It is seen that the present model is
the ButlereVolmer equation:
in good agreement with the analytical solution at both binary so-
s    
lution side (x < 0) and the charged membrane side (x > 0).
c3 an;c Fhn cs an;a Fhn
j ¼ i0n exp   2 exp (29) As the interfacial region has no clear boundary, the given width
c3 RT c2 RT
should not affect the result. Two cases are compared with only
s    
interfacial width difference. The operating conditions are
c5 ap;c Fhp cs ap;a Fhp
j ¼ i0p exp   4 exp (30) cM ¼ 1200 mol/m3, I ¼ 400 A/m2 and SOC ¼ 0.3 of the charge
c5 RT c4 RT process. In Fig. 3c, the ionic potential distributions at the M/E
interfacial region of negative side show a high level of consistency
where ac, aa are the cathodic and anodic transfer coefficients, h is so that the interfacial width set in this paper is appropriate.
the overpotential, the subscripts n and p represent the negative and
positive electrodes, respectively. The exchange current densities i0n
3.2. Variable distributions
and i0p for the negative and positive electrodes are expressed as
In the following sections, we discuss the effect of fixed charge
i0n ¼ Fkn ðc2 Þan;c ðc3 Þan;a (31)
concentration on the species crossover and cell performance taking
the Donnan effect into consideration.
i0p ¼ Fkp ðc4 Þap;c ðc5 Þap;a (32) Distributions of ion concentrations and the ionic potential with
cM ¼ 1200 and 2400 mol/m3 are shown in Fig. 4, where the
where kn and kp are the standard rate constants for the negative and enlarged views of variable distributions at the M/E interfacial re-
positive reactions. gions are also exhibited. The operating conditions are I ¼ 400 A/m2
The cell voltage Ecell is calculated as and SOC ¼ 0.5 of the charge process. In Fig. 4a, the comparison of
V2þ concentration between the models with and without the
 2  
RT cp cn cp Donnan effect is presented. For the model without the Donnan
Ecell ¼ E0p  E0n þ ln p5 n2 H þ fpl  fnl þ hp  hn þ Vohm effect, HSO4  concentration at the membrane/electrolyte junction
F c4 c3  10 6
is discontinuous, the difference of which is equal to zMcM to satisfy
(33) electroneutrality. The distribution of V2þ concentration in the
present model shows a steep jump at the M/E interfacial region,
where En0 and Ep0 are the equilibrium potentials, the superscripts n
known as the Donnan effect, which has not been captured in the
and p represent the negative and positive electrodes, respectively.
compared one. In addition, the jump is enhanced proportionally
Under the assumption that the conductivity in the electrode matrix
is consistent with that in the electrolyte, the ohmic loss Vohm
associated with the electrodes and current collectors is formulated Table 4
to be: Kinetic parameters.

0 1 0 1 Parameter Value

B ILt C B ILt C 2I Reaction rate constant: positive [14], kp 6.8e-7 m$s1
Vohm ¼@ P 2 A þ @ 3=2 F 2 P 2 A þ Lc 1.7e-7 m$s1
3 3=2
F2 z c D 3 z c D sc Reaction rate constant: negative [14], kn
RT i i i;e RT i i i;e Cathodic transfer coefficient: positive, ap,c 0.5
i n i p
Anodic transfer coefficient: positive, ap,a 0.5
(34) Cathodic transfer coefficient: negative, an,c 0.5
Anodic transfer coefficient: negative, an,a 0.5
where sc and Lc are the conductivity and the width of current Equilibrium potential: positive, Ep0 1.004 V
Equilibrium potential: negative, En0 0.255 V
collectors, Di,e is the diffusivity of species i in electrolytes. All kinetic
Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211 207

with an increase in cM. For other ion concentrations and the ionic
potential shown in Fig. 4, the situations are similar to that of the V2þ
concentration. It is found that for a CEM, the Donnan effect leads to
an increase in the cation concentrations in the membrane, and a
decrease in the anion concentrations and the ionic potential. These
results are consistent with the theoretical analysis of ion adsorption
and repulsion in the membrane. Thus, a more complete effect of the
fixed charge concentration on the variable distributions in the
membrane is mathematically achieved. Furthermore, Fig. 4c shows
that the fixed charge concentration does not have a large effect on
the ionic potential gradient in the membrane.
Another factor that influences the jump is the electrolyte con-
centrations. As shown in Fig. 5a, the jump is enhanced with an
increase of V2þ concentration in electrolyte, which is evaluated to
be similar for all other ions. The operating conditions are I ¼ 400 A/
m2 and cM ¼ 1200 mol/m3.
Fig. 5b and c shows the distributions of V2þ concentration with
different pressure boundary conditions and different applied cur-
rent densities, respectively. The pressure difference across the
membrane is set to be zero and 1500 Pa, respectively in Fig. 5b.
With the same cM ¼ 1200 mol/m3, I ¼ 400 A/m2 and SOC ¼ 0.393 of
the charge process, the jump does not vary with the pressure
boundary conditions, while the slopes in the membrane differ due
to the effect of the osmotic flow. It reveals that the pressure
boundary condition has no influence on the Donnan effect. With
SOC ¼ 0.5 and cM ¼ 1200 mol/m3 in Fig. 5c, it reveals that the
Donnan effect does not vary with current density, while the slope in
the membrane changes due to the variation of the velocity, which is
related to the current density and derived from the combination of
Eqs. (4), (6) and (8):

. kp kf I
vz Vp  ðzM cM ÞF P (35)
mw mw F2 z2i Di;m ci
RT
i

In the above equation, it shows clearly that the electro-osmotic

term in the velocity is proportional to the current density.

3.3. Vanadium crossover

With an increase in cM, the variation of vanadium crossover

becomes more complex and is affected by both the Donnan effect
and the electro-osmotic flow. Fluxes of V2þ and VO2þ via the
membrane of the charge process with different cM between the
models with and without the Donnan effect are separately pre-
sented in Fig. 6. The flux of V2þ is chosen at x ¼ x3, and that of VO2þ
is at x ¼ x2. It is found that compared to the model without the
Donnan effect at I ¼ 400 A/m2, fluxes of V2þ and VO2þ in the pre-
sent model are enhanced with the same cM, resulting from an in-
crease in the diffusion term deduced from Fig. 4a, i.e. the Donnan
effect promotes vanadium crossover for a VRFB with a CEM, even
when the convection term imposes very little difference.
With an increase in cM, only the electro-osmotic flow affects
vanadium crossover in the model without the Donnan effect.
Therefore, the flux of every vanadium ion obeys the variation of the
velocity in Eq. (35). Moreover, the difference in flux with an in-
crease in cM is compared between the models with and without the
Donnan effect to explore the role of the Donnan effect on the va-
nadium fluxes. The results imply that for all vanadium ions, the
Donnan effect leads to an impetus for flux to transport through the

Fig. 3. (a) Comparison of the simulated voltage with experimental data [16]; (b) model
validation of the Donnan effect at the M/E interface; (c) the ionic potential distribu-
tions with different negative-side M/E interfacial widths at cM ¼ 1200 mol/m3,
I ¼ 400 A/m2 and SOC ¼ 0.3 of the charge process.
208 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211

Fig. 5. Distributions of V2þ concentration with different (a) time, (b) pressure
Fig. 4. Distributions of (a) V2þ, (b) HSO4  concentrations and (c) the ionic potential boundary conditions at SOC ¼ 0.393, and (c) applied currents at SOC ¼ 0.5. For all cases
with different cM at I ¼ 400 A/m2 and SOC ¼ 0.5 of the charge process. DE denotes the with no specific mention, the conditions are I ¼ 400 A/m2 and cM ¼ 1200 mol/m3 of the
Donnan effect. charge process.
 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211
Fig. 6. Fluxes of (a) V2þ and (b) VO2þ via the membrane with different cM between the
models with and without the Donnan effect (DE) at I ¼ 400 A/m2 in the charge process.
membrane. This phenomenon occurs due to an improvement in
average diffusion with an increase in cM.
3.4. Self-reactions and the coulombic efficiency
As the number of self-reactions is determined by vanadium
crossover, the coulombic efficiency is also affected by the fixed
charge concentration. Fig. 7 shows the effect of cM on the coulombic
efficiency between the models with and without the Donnan effect
at different current densities. The charge and discharge times are
determined by the charge and discharge periods between SOC ¼ 0.2
and SOC ¼ 0.8, respectively. It demonstrates that the coulombic
efficiency in the present model is lower at the same cM when
compared to that of the model without the Donnan effect. This is
particularly pronounced at a large cM. This is due to the enhance-
ment of all vanadium ion fluxes, illustrated in Fig. 6 and further
discussed in Section 3.3.
With an increase in cM, the coulombic efficiency decreases in the
model without the Donnan effect, caused by the promotion of the
electro-osmotic flow. And the present model exhibits a larger
decrease in the coulombic efficiency due to the consideration of the
Donnan effect. Moreover, it is revealed in Fig. 7 that with an in-
crease in cM, the electro-osmotic flow dominates the variation of
209
Fig. 7. Effect of cM on the coulombic efficiency between the models with and without
the Donnan effect (DE) at different applied currents. The charge and discharge periods
are set between SOC ¼ 0.2 and SOC ¼ 0.8.
the coulombic efficiency at large current densities, while the
Donnan effect dominates at small current densities. It is explained
that the electro-osmotic flow and its variation with cM both decline
with current density, analyzed by Eq. (35).
Finally, the electro-osmotic flow virtually vanishes when the cell
is open circuit and the effect of the fixed charge concentration on
the species crossover is totally owed to the Donnan effect.
3.5. Electrolyte imbalance in a VRFB
Species crossover also influences the electrolyte imbalance be-
tween two half-cells, the variation of which is determined by the
sum of the vanadium ion fluxes through the membrane. Fig. 8
shows the effect of cM on the total vanadium concentration in
each half-cell. The applied current density is 400 A/m2. The results
show that the total vanadium concentration between two half-cells
is at an imbalance after an arbitrary operating period, due to the
parameter differences of ions such as the valence and diffusivity.
With an increase in cM in Fig. 8a, the model without the Donnan
effect shows that the total vanadium concentration (dash dot line)
at the end of a cycle appears to be the same in each half-cell. This is
interpreted with an analysis of the electro-osmotic flow below. In
the charge and discharge processes, the magnitudes of electro-
osmotic flow are equivalent, while the directions are opposite,
derived from Eq. (35). Therefore, the effect of the electro-osmotic
flow on electrolyte imbalance is largely counteracted after a char-
geedischarge cycle, and thus the electrolyte imbalance does not
vary with cM in the model without the Donnan effect. However, the
electrolyte imbalance (solid line) after a cycle is enhanced with an
increase in cM in the present model, due to an improvement in
average diffusion caused by the Donnan effect, which leads to a
greater impact of diffusivity difference on the sum of vanadium ion
fluxes through the membrane.
In Fig. 8b, it can be seen that the pressure boundary condition is
an important factor in controlling the extent of electrolyte imbal-
ance, while its variation with an increase in cM is almost the same
under different pressure boundary conditions. Thus, the Donnan
effect dominates the variation of electrolyte imbalance with an
increase in cM.
4. Conclusion
In this work, a one-dimensional transient model incorporating
210 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211

efficiency decreases, and its variation is dominated by the

Donnan effect at small applied current densities.
(4) For a VRFB with a CEM, the extent of the electrolyte imbal-
ance after a cycle is enhanced with an increase in cM. This
phenomenon is almost entirely influenced by the Donnan
effect.

Acknowledgments

This work was supported by the Open Projects of State Key

Laboratory of Multiphase Flow in Power Engineering of China.

Nomenclature

c Concentration, mol m3

t Time, s
.
N Species flux, mol m2 s1
D Diffusivity, m2 s1
z Valence
R Universal gas constant, J mol1 K1
T Operating temperature, K
F Faraday constant, C mol1
.
v Velocity, m s1
p Pressure, Pa
i Current density, A m2
x Position at x axis, m
S Reaction source term
V Volume
j Local reaction current density, A m2
a Specific surface area, m1
I Applied current density, A m2
L Length, m
SOC State of charge
cD Tuning concentration, mol m3
km Mass transfer coefficient nearby the electrode surface, s1
i0 Exchange current density, A m2
k Standard reaction rate constant, m s1
E0 Equilibrium potential, V
Vohm Ohmic loss, V
Fig. 8. Effect of cM on vanadium imbalance during a cycle (a) between the models with
and without the Donnan effect (DE) at Dp ¼ 0 Pa and (b) between different pressure Greek
difference across the membrane in the present model. The current is 400 A/m2. f Potential, V
kp Hydraulic permeability, m2
kf Electrokinetic permeability, m2
the Donnan effect is developed for all-vanadium redox flow bat- mw Viscosity, Pa s
teries. Unlike previous models with unrealistic assumptions ε Dielectric constant, F m1
including electroneutrality and discontinuities of ionic potential re Space charge density, C m3
and ion concentrations at the M/E interfaces, a continuous-Donnan є Porosity
effect-model is proposed in this work to simulate the space charge aa Anodic transfer coefficient
region at the M/E interfaces. It is demonstrated that the present ac Cathodic transfer coefficient
model predicts (i) a realistic continuous variation for all the vari- h Overpotential, V
ables across the entire cell, and (ii) a more complicated effect of the s Conductivity, S m1
fixed charge concentration on species crossover and cell perfor-
mance. The findings are as follows: Superscripts and subscripts
i Species
(1) The jump in the vanadium ions at the M/E interfacial regions m,M Membrane
is enhanced with an increase in the fixed charge concentra- 2,3,4,5 V2þ, V3þ, VO2þ, VO2 þ
tion or electrolyte concentrations, while it does not vary with t Tank
pressure boundary conditions or applied current densities. e Electrolyte
(2) For a VRFB with a CEM, the Donnan effect promotes vana- h Height of electrode
dium crossover. The promotion is more pronounced with an w Width of electrode
increase in cM. 0 Initial condition
(3) The Donnan effect leads to a decline in coulombic efficiency s Solid
for a VRFB with a CEM. With an increase in cM, coulombic n Negative side
 Y. Lei et al. / Journal of Power Sources 299 (2015) 202e211
p Positive side
l Liquid
c Current collector
H Hþ
HS HSO4 
References
[1] G. Kear, A.A. Shah, F.C. Walsh, Development of the all-vanadium redox flow
battery for energy storage: a review of technological, financial and policy
aspects, Int. J. Energy Res. 36 (2012) 1105e1120.
[2] A.Z. Weber, M.M. Mench, J.P. Meyers, et al., Redox flow batteries: a review,
J. Appl. Electrochem. 41 (2011) 1137e1164.
[3] A. Parasuraman, T.M. Lim, C. Menictas, et al., Review of material research and
development for vanadium redox flow battery applications, Electrochim. Acta
101 (2013) 27e40.
[4] X. Li, H. Zhang, Z. Mai, et al., Ion exchange membranes for vanadium redox
flow battery (VRB) applications, Energy Environ. Sci. 4 (2011) 1147e1160.
[5] P.K. Leung, Q. Xu, T.S. Zhao, et al., Preparation of silica nanocomposite anion-
exchange membranes with low vanadium-ion crossover for vanadium redox
flow batteries, Electrochim. Acta 105 (2013) 584e592.
[6] X.L. Zhou, T.S. Zhao, L. An, et al., The use of polybenzimidazole membranes in
vanadium redox flow batteries leading to increased coulombic efficiency and
cycling performance, Electrochim. Acta 153 (2015) 492e498.
[7] D. Chen, S. Kim, V. Sprenkle, et al., Composite blend polymer membranes with
increased proton selectivity and lifetime for vanadium redox flow batteries,
J. Power Sources 231 (2013) 301e306.
[8] D. Chen, M.A. Hickner, S. Wang, et al., Directly fluorinated polyaromatic
composite membranes for vanadium redox flow batteries, J. Membr. Sci.
415e416 (2012) 139e144.
[9] D. Chen, S. Wang, M. Xiao, et al., Preparation and properties of sulfonated
poly(fluorenyl ether ketone) membrane for vanadium redox flow battery
application, J. Power Sources 195 (2010) 2089e2095.
[10] D. Chen, S. Wang, M. Xiao, et al., Synthesis of sulfonated poly(fluorenyl ether
thioether ketone)s with bulky-block structure and its application in vanadium
redox flow battery, Polymer 52 (2011) 5312e5319.
[11] E. Wiedemann, A. Heintz, R.N. Lichtenthaler, Sorption isotherms of vanadium
with H3Oþ ions in cation exchange membranes, J. Membr. Sci. 141 (1998)
207e213.
211
[12] J.S. Lawton, A. Jones, T. Zawodzinski, Concentration dependence of VO2þ
crossover of Nafion for vanadium redox flow batteries, J. Electrochem. Soc.
160 (2013) A697eA702.
[13] J.Q. Chen, B.G. Wang, J.C. Yang, Adsorption and diffusion of VO2þ and VOþ 2
across cation membrane for all-vanadium redox flow battery, Solv. Extr. Ion
Exch. 27 (2009) 312e327.
[14] D. You, H. Zhang, J. Chen, A simple model for the vanadium redox battery,
Electrochim. Acta 54 (2009) 6827e6836.
[15] A.A. Shah, M.J. Watt-Smith, F.C. Walsh, A dynamic performance model for
redox-flow batteries involving soluble species, Electrochim. Acta 53 (2008)
8087e8100.
[16] K.W. Knehr, E. Agar, C.R. Dennison, et al., A transient vanadium flow battery
model incorporating vanadium crossover and water transport through the
membrane, J. Electrochem. Soc. 159 (2012) A1446eA1459.
[17] S. Won, K. Oh, H. Ju, Numerical analysis of vanadium crossover effects in all-
vanadium redox flow batteries, Electrochim. Acta (2015), http://dx.doi.org/
10.1016/j.electacta.2015.01.166.
[18] A.A. Shah, R. Tangirala, R. Singh, et al., A dynamic unit cell model for the all-
vanadium flow battery, J. Electrochem. Soc. 158 (2011) A671eA677.
[19] M. Skyllas-Kazacos, L. Goh, Modeling of vanadium ion diffusion across the ion
exchange membrane in the vanadium redox battery, J. Membr. Sci. 399e400
(2012) 43e48.
[20] K.W. Knehr, E.C. Kumbur, Role of convection and related effects on species
crossover and capacity loss in vanadium redox flow batteries, Electrochem.
Commun. 23 (2012) 76e79.
[21] X.G. Yang, Q. Ye, P. Cheng, et al., Effects of the electric field on ion crossover in
vanadium redox flow batteries, Appl. Energy 145 (2015) 306e319.
[22] D. You, H. Zhang, C. Sun, et al., Simulation of the self-discharge process in
vanadium redox flow battery, J. Power Sources 196 (2011) 1578e1585.
[23] D. Chen, M.A. Hickner, E. Agar, et al., Optimizing membrane thickness for
vanadium redox flow batteries, J. Membr. Sci. 437 (2013) 108e113.
[24] E. Agar, K.W. Knehr, D. Chen, et al., Species transport mechanisms governing
capacity loss in vanadium flow batteries: comparing Nafion® and sulfonated
Radel membranes, Electrochim. Acta 98 (2013) 66e74.
[25] H. Ohshima, S. Ohki, Donnan potential and surface potential of a charged
membrane, Biophys. J. 47 (1985) 673e678.
[26] A. Mauro, Space charge regions in fixed charge membranes and the associated
property of capacitance, Biophys. J. 2 (1962) 179e198.
[27] Q. Zheng, G.W. Wei, Poisson-boltzmann-nernst-planck model, J. Chem. Phys.
134 (2011) 194101.